From: "Darrick J. Wong" <djwong@kernel.org>
To: Christoph Hellwig <hch@lst.de>
Cc: Chandan Babu R <chandan.babu@oracle.com>, linux-xfs@vger.kernel.org
Subject: Re: [PATCH 10/16] xfs: optimize removing the last 8-byte inode from a shortform directory
Date: Wed, 1 May 2024 14:25:19 -0700 [thread overview]
Message-ID: <20240501212519.GY360919@frogsfrogsfrogs> (raw)
In-Reply-To: <20240430124926.1775355-11-hch@lst.de>
On Tue, Apr 30, 2024 at 02:49:20PM +0200, Christoph Hellwig wrote:
> When removing the last 8-byte inode, xfs_dir2_sf_removename calls
> xfs_dir2_sf_toino4 after removing the entry. This is rather inefficient
> as it causes two buffer realloacations. Instead of that pass a bool
> argument to xfs_dir2_sf_toino4 so that it can remove the entry pointed
> to by args as part of the conversion and use that to shortcut the
> process.
>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
> fs/xfs/libxfs/xfs_dir2_sf.c | 41 ++++++++++++++++++++++++++++---------
> 1 file changed, 31 insertions(+), 10 deletions(-)
>
> diff --git a/fs/xfs/libxfs/xfs_dir2_sf.c b/fs/xfs/libxfs/xfs_dir2_sf.c
> index 3b6d6dda92f29f..21e04594606b89 100644
> --- a/fs/xfs/libxfs/xfs_dir2_sf.c
> +++ b/fs/xfs/libxfs/xfs_dir2_sf.c
> @@ -34,7 +34,7 @@ static void xfs_dir2_sf_check(xfs_da_args_t *args);
> #define xfs_dir2_sf_check(args)
> #endif /* DEBUG */
>
> -static void xfs_dir2_sf_toino4(xfs_da_args_t *args);
> +static void xfs_dir2_sf_toino4(struct xfs_da_args *args, bool remove);
> static void xfs_dir2_sf_toino8(xfs_da_args_t *args);
>
> int
> @@ -935,6 +935,15 @@ xfs_dir2_sf_removename(
> ASSERT(dp->i_df.if_bytes == oldsize);
> ASSERT(oldsize >= xfs_dir2_sf_hdr_size(sfp->i8count));
>
> + /*
> + * If this is the last 8-byte, directly convert to the 4-byte format
> + * and just skip the removed entry when building the new fork.
> + */
> + if (args->inumber > XFS_DIR2_MAX_SHORT_INUM && sfp->i8count == 1) {
> + xfs_dir2_sf_toino4(args, true);
> + return 0;
> + }
> +
> /*
> * Loop over the old directory entries.
> * Find the one we're deleting.
> @@ -980,10 +989,8 @@ xfs_dir2_sf_removename(
> * Are we changing inode number size?
> */
> if (args->inumber > XFS_DIR2_MAX_SHORT_INUM) {
> - if (sfp->i8count == 1)
> - xfs_dir2_sf_toino4(args);
> - else
> - sfp->i8count--;
> + ASSERT(sfp->i8count > 1);
> + sfp->i8count--;
> }
> xfs_dir2_sf_check(args);
> xfs_trans_log_inode(args->trans, dp, XFS_ILOG_CORE | XFS_ILOG_DDATA);
> @@ -1087,7 +1094,7 @@ xfs_dir2_sf_replace(
> if (i == sfp->count) {
> ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
> if (i8elevated)
> - xfs_dir2_sf_toino4(args);
> + xfs_dir2_sf_toino4(args, false);
> return -ENOENT;
> }
> }
> @@ -1100,7 +1107,7 @@ xfs_dir2_sf_replace(
> * And the old count was one, so need to convert to small.
> */
> if (sfp->i8count == 1)
> - xfs_dir2_sf_toino4(args);
> + xfs_dir2_sf_toino4(args, false);
> else
> sfp->i8count--;
> }
> @@ -1128,7 +1135,8 @@ xfs_dir2_sf_replace(
> */
> static void
> xfs_dir2_sf_toino4(
> - xfs_da_args_t *args) /* operation arguments */
> + struct xfs_da_args *args,
> + bool remove)
> {
> struct xfs_inode *dp = args->dp;
> struct xfs_mount *mp = dp->i_mount;
> @@ -1148,6 +1156,8 @@ xfs_dir2_sf_toino4(
> * Compute the new inode size.
> */
> newsize = oldsize - (oldsfp->count + 1) * XFS_INO64_DIFF;
> + if (remove)
> + newsize -= xfs_dir2_sf_entsize(mp, oldsfp, args->namelen);
>
> dp->i_df.if_data = sfp = kmalloc(newsize, GFP_KERNEL | __GFP_NOFAIL);
> dp->i_df.if_bytes = newsize;
> @@ -1166,11 +1176,22 @@ xfs_dir2_sf_toino4(
> i < sfp->count;
> i++, sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep),
> oldsfep = xfs_dir2_sf_nextentry(mp, oldsfp, oldsfep)) {
> + xfs_ino_t ino = xfs_dir2_sf_get_ino(mp, oldsfp, oldsfep);
> +
> + /*
> + * Just skip over the entry that is removed if there is one.
> + */
> + if (remove && args->inumber == ino) {
> + oldsfep = xfs_dir2_sf_nextentry(mp, oldsfp, oldsfep);
> + sfp->count--;
> + if (++i == sfp->count)
> + break;
> + }
What happens if a shortform directory contains two entries to the same
file? I think @remove really means that the caller has verified that
the sf directory only contains one link @args->inumber? If that's so,
then the comment for this function should say that.
--D
> +
> sfep->namelen = oldsfep->namelen;
> memcpy(sfep->offset, oldsfep->offset, sizeof(sfep->offset));
> memcpy(sfep->name, oldsfep->name, sfep->namelen);
> - xfs_dir2_sf_put_ino(mp, sfp, sfep,
> - xfs_dir2_sf_get_ino(mp, oldsfp, oldsfep));
> + xfs_dir2_sf_put_ino(mp, sfp, sfep, ino);
> xfs_dir2_sf_put_ftype(mp, sfep,
> xfs_dir2_sf_get_ftype(mp, oldsfep));
> }
> --
> 2.39.2
>
>
next prev parent reply other threads:[~2024-05-01 21:25 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-04-30 12:49 optimize local for and shortform directory handling Christoph Hellwig
2024-04-30 12:49 ` [PATCH 01/16] xfs: allow non-empty forks in xfs_bmap_local_to_extents_empty Christoph Hellwig
2024-04-30 15:51 ` Darrick J. Wong
2024-05-01 4:37 ` Christoph Hellwig
2024-05-01 21:04 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 02/16] xfs: remove an extra buffer allocation in xfs_attr_shortform_to_leaf Christoph Hellwig
2024-05-01 21:11 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 03/16] xfs: rationalize dir2_sf entry condition asserts Christoph Hellwig
2024-05-01 21:13 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 04/16] xfs: remove an extra buffer allocation in xfs_dir2_sf_to_block Christoph Hellwig
2024-05-01 21:15 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 05/16] xfs: move the "does it fit" check into xfs_dir2_block_to_sf Christoph Hellwig
2024-05-01 21:16 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 06/16] xfs: remove the buffer allocation size in xfs_dir2_try_block_to_sf Christoph Hellwig
2024-05-01 21:17 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 07/16] xfs: remove a superfluous memory allocation in xfs_dir2_block_to_sf Christoph Hellwig
2024-05-01 21:18 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 08/16] xfs: remove a superfluous memory allocation in xfs_dir2_sf_toino8 Christoph Hellwig
2024-05-01 21:20 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 09/16] xfs: remove a superfluous memory allocation in xfs_dir2_sf_toino4 Christoph Hellwig
2024-05-01 21:20 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 10/16] xfs: optimize removing the last 8-byte inode from a shortform directory Christoph Hellwig
2024-05-01 21:25 ` Darrick J. Wong [this message]
2024-05-02 4:13 ` Christoph Hellwig
2024-04-30 12:49 ` [PATCH 11/16] xfs: add xfs_dir2_block_overhead helper Christoph Hellwig
2024-05-01 21:27 ` Darrick J. Wong
2024-05-02 4:14 ` Christoph Hellwig
2024-04-30 12:49 ` [PATCH 12/16] xfs: factor out a xfs_dir2_sf_addname_common helper Christoph Hellwig
2024-05-01 21:31 ` Darrick J. Wong
2024-05-02 4:15 ` Christoph Hellwig
2024-04-30 12:49 ` [PATCH 13/16] xfs: move common code into xfs_dir2_sf_addname Christoph Hellwig
2024-05-01 21:32 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 14/16] xfs: optimize adding the first 8-byte inode to a shortform directory Christoph Hellwig
2024-05-01 21:50 ` Darrick J. Wong
2024-05-02 4:25 ` Christoph Hellwig
2024-05-02 14:43 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 15/16] xfs: move the block format conversion out of line in xfs_dir2_sf_addname Christoph Hellwig
2024-05-01 21:33 ` Darrick J. Wong
2024-04-30 12:49 ` [PATCH 16/16] xfs: make the hard case in xfs_dir2_sf_addname less hard Christoph Hellwig
2024-05-01 22:10 ` Darrick J. Wong
2024-05-10 6:29 ` kernel test robot
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240501212519.GY360919@frogsfrogsfrogs \
--to=djwong@kernel.org \
--cc=chandan.babu@oracle.com \
--cc=hch@lst.de \
--cc=linux-xfs@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).