Re: [PATCH RESEND v2 12/18] xfs: parent pointer attribute creation

["Darrick J. Wong" <djwong@kernel.org>]

On Tue, Aug 09, 2022 at 11:01:01AM -0700, Darrick J. Wong wrote:
> On Thu, Aug 04, 2022 at 12:40:07PM -0700, Allison Henderson wrote:
> > Add parent pointer attribute during xfs_create, and subroutines to
> > initialize attributes
> > 
> > [bfoster: rebase, use VFS inode generation]
> > [achender: rebased, changed __unint32_t to xfs_dir2_dataptr_t,
> 
> Nit: uint32_t, not unint32_t.
> 
> >            fixed some null pointer bugs,
> >            merged error handling patch,
> >            remove unnecessary ENOSPC handling in xfs_attr_set_first_parent]
> > 
> > Signed-off-by: Dave Chinner <dchinner@redhat.com>
> > Signed-off-by: Allison Henderson <allison.henderson@oracle.com>
> > ---
> >  fs/xfs/Makefile            |   1 +
> >  fs/xfs/libxfs/xfs_attr.c   |   4 +-
> >  fs/xfs/libxfs/xfs_attr.h   |   4 +-
> >  fs/xfs/libxfs/xfs_parent.c | 134 +++++++++++++++++++++++++++++++++++++
> >  fs/xfs/libxfs/xfs_parent.h |  34 ++++++++++
> >  fs/xfs/xfs_inode.c         |  37 ++++++++--
> >  fs/xfs/xfs_xattr.c         |   2 +-
> >  fs/xfs/xfs_xattr.h         |   1 +
> >  8 files changed, 208 insertions(+), 9 deletions(-)
> > 
> > diff --git a/fs/xfs/Makefile b/fs/xfs/Makefile
> > index 1131dd01e4fe..caeea8d968ba 100644
> > --- a/fs/xfs/Makefile
> > +++ b/fs/xfs/Makefile
> > @@ -40,6 +40,7 @@ xfs-y				+= $(addprefix libxfs/, \
> >  				   xfs_inode_fork.o \
> >  				   xfs_inode_buf.o \
> >  				   xfs_log_rlimit.o \
> > +				   xfs_parent.o \
> >  				   xfs_ag_resv.o \
> >  				   xfs_rmap.o \
> >  				   xfs_rmap_btree.o \
> > diff --git a/fs/xfs/libxfs/xfs_attr.c b/fs/xfs/libxfs/xfs_attr.c
> > index 2ef3262f21e8..0a458ea7051f 100644
> > --- a/fs/xfs/libxfs/xfs_attr.c
> > +++ b/fs/xfs/libxfs/xfs_attr.c
> > @@ -880,7 +880,7 @@ xfs_attr_lookup(
> >  	return error;
> >  }
> >  
> > -static int
> > +int
> >  xfs_attr_intent_init(
> >  	struct xfs_da_args	*args,
> >  	unsigned int		op_flags,	/* op flag (set or remove) */
> > @@ -898,7 +898,7 @@ xfs_attr_intent_init(
> >  }
> >  
> >  /* Sets an attribute for an inode as a deferred operation */
> > -static int
> > +int
> >  xfs_attr_defer_add(
> >  	struct xfs_da_args	*args)
> >  {
> > diff --git a/fs/xfs/libxfs/xfs_attr.h b/fs/xfs/libxfs/xfs_attr.h
> > index af92cc57e7d8..b47417b5172f 100644
> > --- a/fs/xfs/libxfs/xfs_attr.h
> > +++ b/fs/xfs/libxfs/xfs_attr.h
> > @@ -544,6 +544,7 @@ int xfs_inode_hasattr(struct xfs_inode *ip);
> >  bool xfs_attr_is_leaf(struct xfs_inode *ip);
> >  int xfs_attr_get_ilocked(struct xfs_da_args *args);
> >  int xfs_attr_get(struct xfs_da_args *args);
> > +int xfs_attr_defer_add(struct xfs_da_args *args);
> >  int xfs_attr_set(struct xfs_da_args *args);
> >  int xfs_attr_set_iter(struct xfs_attr_intent *attr);
> >  int xfs_attr_remove_iter(struct xfs_attr_intent *attr);
> > @@ -552,7 +553,8 @@ bool xfs_attr_namecheck(struct xfs_mount *mp, const void *name, size_t length,
> >  int xfs_attr_calc_size(struct xfs_da_args *args, int *local);
> >  void xfs_init_attr_trans(struct xfs_da_args *args, struct xfs_trans_res *tres,
> >  			 unsigned int *total);
> > -
> > +int xfs_attr_intent_init(struct xfs_da_args *args, unsigned int op_flags,
> > +			 struct xfs_attr_intent  **attr);
> >  /*
> >   * Check to see if the attr should be upgraded from non-existent or shortform to
> >   * single-leaf-block attribute list.
> > diff --git a/fs/xfs/libxfs/xfs_parent.c b/fs/xfs/libxfs/xfs_parent.c
> > new file mode 100644
> > index 000000000000..4ab531c77d7d
> > --- /dev/null
> > +++ b/fs/xfs/libxfs/xfs_parent.c
> > @@ -0,0 +1,134 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * Copyright (c) 2022 Oracle, Inc.
> > + * All rights reserved.
> > + */
> > +#include "xfs.h"
> > +#include "xfs_fs.h"
> > +#include "xfs_format.h"
> > +#include "xfs_da_format.h"
> > +#include "xfs_log_format.h"
> > +#include "xfs_shared.h"
> > +#include "xfs_trans_resv.h"
> > +#include "xfs_mount.h"
> > +#include "xfs_bmap_btree.h"
> > +#include "xfs_inode.h"
> > +#include "xfs_error.h"
> > +#include "xfs_trace.h"
> > +#include "xfs_trans.h"
> > +#include "xfs_da_btree.h"
> > +#include "xfs_attr.h"
> > +#include "xfs_da_btree.h"
> > +#include "xfs_attr_sf.h"
> > +#include "xfs_bmap.h"
> > +#include "xfs_defer.h"
> > +#include "xfs_log.h"
> > +#include "xfs_xattr.h"
> > +#include "xfs_parent.h"
> > +
> > +/*
> > + * Parent pointer attribute handling.
> > + *
> > + * Because the attribute value is a filename component, it will never be longer
> > + * than 255 bytes. This means the attribute will always be a local format
> > + * attribute as it is xfs_attr_leaf_entsize_local_max() for v5 filesystems will
> > + * always be larger than this (max is 75% of block size).
> > + *
> > + * Creating a new parent attribute will always create a new attribute - there
> > + * should never, ever be an existing attribute in the tree for a new inode.
> > + * ENOSPC behavior is problematic - creating the inode without the parent
> > + * pointer is effectively a corruption, so we allow parent attribute creation
> > + * to dip into the reserve block pool to avoid unexpected ENOSPC errors from
> > + * occurring.
> 
> Shouldn't we increase XFS_LINK_SPACE_RES to avoid this?  The reserve
> pool isn't terribly large (8192 blocks) and was really only supposed to
> save us from an ENOSPC shutdown if an unwritten extent conversion in the
> writeback endio handler needs a few more blocks.
> 
> IOWs, we really ought to ENOSPC at transaction reservation time instead
> of draining the reserve pool.
> 
> > + */
> > +
> > +
> > +/* Initializes a xfs_parent_name_rec to be stored as an attribute name */
> > +void
> > +xfs_init_parent_name_rec(
> > +	struct xfs_parent_name_rec	*rec,
> > +	struct xfs_inode		*ip,
> > +	uint32_t			p_diroffset)
> > +{
> > +	xfs_ino_t			p_ino = ip->i_ino;
> > +	uint32_t			p_gen = VFS_I(ip)->i_generation;
> > +
> > +	rec->p_ino = cpu_to_be64(p_ino);
> > +	rec->p_gen = cpu_to_be32(p_gen);
> > +	rec->p_diroffset = cpu_to_be32(p_diroffset);
> > +}
> > +
> > +/* Initializes a xfs_parent_name_irec from an xfs_parent_name_rec */
> > +void
> > +xfs_init_parent_name_irec(
> > +	struct xfs_parent_name_irec	*irec,
> > +	struct xfs_parent_name_rec	*rec)
> > +{
> > +	irec->p_ino = be64_to_cpu(rec->p_ino);
> > +	irec->p_gen = be32_to_cpu(rec->p_gen);
> > +	irec->p_diroffset = be32_to_cpu(rec->p_diroffset);
> > +}
> > +
> > +int
> > +xfs_parent_init(
> > +	xfs_mount_t                     *mp,
> > +	xfs_inode_t			*ip,

More nits: Please don't use struct typedefs here.

> > +	struct xfs_name			*target_name,
> > +	struct xfs_parent_defer		**parentp)
> > +{
> > +	struct xfs_parent_defer		*parent;
> > +	int				error;
> > +
> > +	if (!xfs_has_parent(mp))
> > +		return 0;
> > +
> > +	error = xfs_attr_grab_log_assist(mp);
> 
> At some point we might want to consider boosting performance by setting
> XFS_SB_FEAT_INCOMPAT_LOG_XATTRS permanently when parent pointers are
> turned on, since adding the feature requires a synchronous bwrite of the
> primary superblock.
> 
> I /think/ this could be accomplished by setting the feature bit in mkfs
> and teaching xlog_clear_incompat to exit if xfs_has_parent()==true.
> Then we can skip the xfs_attr_grab_log_assist calls.
> 
> But, let's focus on getting this patchset into good enough shape that
> we can be confident that we don't need any ondisk format changes, and
> worry about speed later.
> 
> > +	if (error)
> > +		return error;
> > +
> > +	parent = kzalloc(sizeof(*parent), GFP_KERNEL);
> 
> These objects are going to be created and freed fairly frequently; could
> you please convert these to a kmem cache?  (That can be a cleanup at the
> end.)
> 
> > +	if (!parent)
> > +		return -ENOMEM;
> > +
> > +	/* init parent da_args */
> > +	parent->args.dp = ip;
> > +	parent->args.geo = mp->m_attr_geo;
> > +	parent->args.whichfork = XFS_ATTR_FORK;
> > +	parent->args.attr_filter = XFS_ATTR_PARENT;
> > +	parent->args.op_flags = XFS_DA_OP_OKNOENT | XFS_DA_OP_LOGGED;
> > +	parent->args.name = (const uint8_t *)&parent->rec;
> > +	parent->args.namelen = sizeof(struct xfs_parent_name_rec);
> > +
> > +	if (target_name) {
> > +		parent->args.value = (void *)target_name->name;
> > +		parent->args.valuelen = target_name->len;
> > +	}
> > +
> > +	*parentp = parent;
> > +	return 0;
> > +}
> > +
> > +int
> > +xfs_parent_defer_add(
> > +	struct xfs_trans	*tp,
> > +	struct xfs_inode	*ip,
> > +	struct xfs_parent_defer	*parent,
> > +	xfs_dir2_dataptr_t	diroffset)
> > +{
> > +	struct xfs_da_args	*args = &parent->args;
> > +
> > +	xfs_init_parent_name_rec(&parent->rec, ip, diroffset);
> > +	args->trans = tp;
> > +	args->hashval = xfs_da_hashname(args->name, args->namelen);
> > +	return xfs_attr_defer_add(args);
> > +}
> > +
> > +void
> > +xfs_parent_cancel(
> > +	xfs_mount_t		*mp,
> > +	struct xfs_parent_defer *parent)
> > +{
> > +	xlog_drop_incompat_feat(mp->m_log);
> > +	kfree(parent);
> > +}
> > +
> > diff --git a/fs/xfs/libxfs/xfs_parent.h b/fs/xfs/libxfs/xfs_parent.h
> > new file mode 100644
> > index 000000000000..21a350b97ed5
> > --- /dev/null
> > +++ b/fs/xfs/libxfs/xfs_parent.h
> > @@ -0,0 +1,34 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * Copyright (c) 2022 Oracle, Inc.
> > + * All Rights Reserved.
> > + */
> > +#ifndef	__XFS_PARENT_H__
> > +#define	__XFS_PARENT_H__
> > +
> > +/*
> > + * Dynamically allocd structure used to wrap the needed data to pass around
> > + * the defer ops machinery
> > + */
> > +struct xfs_parent_defer {
> > +	struct xfs_parent_name_rec	rec;
> > +	struct xfs_da_args		args;
> > +};
> > +
> > +/*
> > + * Parent pointer attribute prototypes
> > + */
> > +void xfs_init_parent_name_rec(struct xfs_parent_name_rec *rec,
> > +			      struct xfs_inode *ip,
> > +			      uint32_t p_diroffset);
> > +void xfs_init_parent_name_irec(struct xfs_parent_name_irec *irec,
> > +			       struct xfs_parent_name_rec *rec);
> > +int xfs_parent_init(xfs_mount_t *mp, xfs_inode_t *ip,
> > +		    struct xfs_name *target_name,
> > +		    struct xfs_parent_defer **parentp);
> > +int xfs_parent_defer_add(struct xfs_trans *tp, struct xfs_inode *ip,
> > +			 struct xfs_parent_defer *parent,
> > +			 xfs_dir2_dataptr_t diroffset);
> > +void xfs_parent_cancel(xfs_mount_t *mp, struct xfs_parent_defer *parent);
> > +
> > +#endif	/* __XFS_PARENT_H__ */
> > diff --git a/fs/xfs/xfs_inode.c b/fs/xfs/xfs_inode.c
> > index 09876ba10a42..ef993c3a8963 100644
> > --- a/fs/xfs/xfs_inode.c
> > +++ b/fs/xfs/xfs_inode.c
> > @@ -37,6 +37,8 @@
> >  #include "xfs_reflink.h"
> >  #include "xfs_ag.h"
> >  #include "xfs_log_priv.h"
> > +#include "xfs_parent.h"
> > +#include "xfs_xattr.h"
> >  
> >  struct kmem_cache *xfs_inode_cache;
> >  
> > @@ -950,7 +952,7 @@ xfs_bumplink(
> >  int
> >  xfs_create(
> >  	struct user_namespace	*mnt_userns,
> > -	xfs_inode_t		*dp,
> > +	struct xfs_inode	*dp,
> >  	struct xfs_name		*name,
> >  	umode_t			mode,
> >  	dev_t			rdev,
> > @@ -962,7 +964,7 @@ xfs_create(
> >  	struct xfs_inode	*ip = NULL;
> >  	struct xfs_trans	*tp = NULL;
> >  	int			error;
> > -	bool                    unlock_dp_on_error = false;
> > +	bool			unlock_dp_on_error = false;
> >  	prid_t			prid;
> >  	struct xfs_dquot	*udqp = NULL;
> >  	struct xfs_dquot	*gdqp = NULL;
> > @@ -970,6 +972,8 @@ xfs_create(
> >  	struct xfs_trans_res	*tres;
> >  	uint			resblks;
> >  	xfs_ino_t		ino;
> > +	xfs_dir2_dataptr_t	diroffset;
> > +	struct xfs_parent_defer	*parent = NULL;
> >  
> >  	trace_xfs_create(dp, name);
> >  
> > @@ -996,6 +1000,12 @@ xfs_create(
> >  		tres = &M_RES(mp)->tr_create;
> >  	}
> >  
> > +	if (xfs_has_parent(mp)) {
> > +		error = xfs_parent_init(mp, dp, name, &parent);
> > +		if (error)
> > +			goto out_release_dquots;
> > +	}
> > +
> >  	/*
> >  	 * Initially assume that the file does not exist and
> >  	 * reserve the resources for that case.  If that is not
> > @@ -1011,7 +1021,7 @@ xfs_create(
> >  				resblks, &tp);
> >  	}
> >  	if (error)
> > -		goto out_release_dquots;
> > +		goto drop_incompat;
> >  
> >  	xfs_ilock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
> >  	unlock_dp_on_error = true;
> > @@ -1021,6 +1031,7 @@ xfs_create(
> >  	 * entry pointing to them, but a directory also the "." entry
> >  	 * pointing to itself.
> >  	 */
> > +	init_xattrs |= xfs_has_parent(mp);
> >  	error = xfs_dialloc(&tp, dp->i_ino, mode, &ino);
> >  	if (!error)
> >  		error = xfs_init_new_inode(mnt_userns, tp, dp, ino, mode,
> > @@ -1035,11 +1046,12 @@ xfs_create(
> >  	 * the transaction cancel unlocking dp so don't do it explicitly in the
> >  	 * error path.
> >  	 */
> > -	xfs_trans_ijoin(tp, dp, XFS_ILOCK_EXCL);
> > +	xfs_trans_ijoin(tp, dp, 0);
> >  	unlock_dp_on_error = false;
> >  
> >  	error = xfs_dir_createname(tp, dp, name, ip->i_ino,
> > -				   resblks - XFS_IALLOC_SPACE_RES(mp), NULL);
> > +				   resblks - XFS_IALLOC_SPACE_RES(mp),
> > +				   &diroffset);
> >  	if (error) {
> >  		ASSERT(error != -ENOSPC);
> >  		goto out_trans_cancel;
> > @@ -1055,6 +1067,17 @@ xfs_create(
> >  		xfs_bumplink(tp, dp);
> >  	}
> >  
> > +	/*
> > +	 * If we have parent pointers, we need to add the attribute containing
> > +	 * the parent information now.
> > +	 */
> > +	if (parent) {
> > +		parent->args.dp	= ip;

...and on second thought, it seems a little odd that you pass @dp to
xfs_parent_init only to override parent->args.dp here.  Given that this
doesn't do anything with @parent until here, why not pass NULL to the
init function above?

--D

> > +		error = xfs_parent_defer_add(tp, dp, parent, diroffset);
> > +		if (error)
> > +			goto out_trans_cancel;
> > +	}
> > +
> >  	/*
> >  	 * If this is a synchronous mount, make sure that the
> >  	 * create transaction goes to disk before returning to
> > @@ -1080,6 +1103,7 @@ xfs_create(
> >  
> >  	*ipp = ip;
> >  	xfs_iunlock(ip, XFS_ILOCK_EXCL);
> > +	xfs_iunlock(dp, XFS_ILOCK_EXCL | XFS_ILOCK_PARENT);
> 
> I don't think we need the ILOCK class annotations for unlocks.
> 
> Other than the two things I asked about, this is looking good.
> 
> --D
> 
> >  	return 0;
> >  
> >   out_trans_cancel:
> > @@ -1094,6 +1118,9 @@ xfs_create(
> >  		xfs_finish_inode_setup(ip);
> >  		xfs_irele(ip);
> >  	}
> > + drop_incompat:
> > +	if (parent)
> > +		xfs_parent_cancel(mp, parent);
> >   out_release_dquots:
> >  	xfs_qm_dqrele(udqp);
> >  	xfs_qm_dqrele(gdqp);
> > diff --git a/fs/xfs/xfs_xattr.c b/fs/xfs/xfs_xattr.c
> > index c325a28b89a8..d9067c5f6bd6 100644
> > --- a/fs/xfs/xfs_xattr.c
> > +++ b/fs/xfs/xfs_xattr.c
> > @@ -27,7 +27,7 @@
> >   * they must release the permission by calling xlog_drop_incompat_feat
> >   * when they're done.
> >   */
> > -static inline int
> > +int
> >  xfs_attr_grab_log_assist(
> >  	struct xfs_mount	*mp)
> >  {
> > diff --git a/fs/xfs/xfs_xattr.h b/fs/xfs/xfs_xattr.h
> > index 2b09133b1b9b..3fd6520a4d69 100644
> > --- a/fs/xfs/xfs_xattr.h
> > +++ b/fs/xfs/xfs_xattr.h
> > @@ -7,6 +7,7 @@
> >  #define __XFS_XATTR_H__
> >  
> >  int xfs_attr_change(struct xfs_da_args *args);
> > +int xfs_attr_grab_log_assist(struct xfs_mount *mp);
> >  
> >  extern const struct xattr_handler *xfs_xattr_handlers[];
> >  
> > -- 
> > 2.25.1
> >