linux-xfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Brian Foster <bfoster@redhat.com>
To: Jeff Liu <jeff.liu@oracle.com>, "xfs@oss.sgi.com" <xfs@oss.sgi.com>
Subject: Re: [PATCH 3/10] xfs: consolidate xfs_inumbers
Date: Fri, 03 Jan 2014 15:53:54 -0500	[thread overview]
Message-ID: <52C72362.2000005@redhat.com> (raw)
In-Reply-To: <52BEB3FA.5080402@oracle.com>

On 12/28/2013 06:20 AM, Jeff Liu wrote:
> From: Jie Liu <jeff.liu@oracle.com>
> 
> To fetch the file system number tables, we currently just ignore the
> errors and proceed to loop over the next AG or bump agino to the next
> chunk in case of btree operations failed, that is not properly because
> those errors might hint us potential file system problems.
> 
> This patch rework xfs_inumbers() to handle the btree operation errors
> as well as the loop conditions.  Also, add pre-checkups for the given
> inode, we can save alloc/free the format buffer once against an invalid
> inode number.
> 
> Signed-off-by: Jie Liu <jeff.liu@oracle.com>
> ---
>  fs/xfs/xfs_itable.c | 163 +++++++++++++++++++++++-----------------------------
>  1 file changed, 72 insertions(+), 91 deletions(-)
> 
> diff --git a/fs/xfs/xfs_itable.c b/fs/xfs/xfs_itable.c
> index 692671c..4d262f6 100644
> --- a/fs/xfs/xfs_itable.c
> +++ b/fs/xfs/xfs_itable.c
> @@ -558,12 +558,12 @@ xfs_bulkstat_single(
>  int
>  xfs_inumbers_fmt(
>  	void			__user *ubuffer, /* buffer to write to */
> -	const xfs_inogrp_t	*buffer,	/* buffer to read from */
> +	const struct xfs_inogrp	*buffer,	/* buffer to read from */
>  	long			count,		/* # of elements to read */
>  	long			*written)	/* # of bytes written */
>  {
>  	if (copy_to_user(ubuffer, buffer, count * sizeof(*buffer)))
> -		return -EFAULT;
> +		return XFS_ERROR(EFAULT);
>  	*written = count * sizeof(*buffer);
>  	return 0;
>  }
> @@ -573,121 +573,102 @@ xfs_inumbers_fmt(
>   */
>  int					/* error status */
>  xfs_inumbers(
> -	xfs_mount_t	*mp,		/* mount point for filesystem */
> -	xfs_ino_t	*lastino,	/* last inode returned */
> -	int		*count,		/* size of buffer/count returned */
> -	void		__user *ubuffer,/* buffer with inode descriptions */
> -	inumbers_fmt_pf	formatter)
> +	struct xfs_mount	*mp,/* mount point for filesystem */
> +	xfs_ino_t		*lastino,/* last inode returned */
> +	int			*count,/* size of buffer/count returned */
> +	void			__user *ubuffer,/* buffer with inode desc */
> +	inumbers_fmt_pf		formatter)
>  {
> -	xfs_buf_t	*agbp;
> -	xfs_agino_t	agino;
> -	xfs_agnumber_t	agno;
> -	int		bcount;
> -	xfs_inogrp_t	*buffer;
> -	int		bufidx;
> -	xfs_btree_cur_t	*cur;
> -	int		error;
> -	xfs_inobt_rec_incore_t r;
> -	int		i;
> -	xfs_ino_t	ino;
> -	int		left;
> -	int		tmp;
> -
> -	ino = (xfs_ino_t)*lastino;
> -	agno = XFS_INO_TO_AGNO(mp, ino);
> -	agino = XFS_INO_TO_AGINO(mp, ino);
> -	left = *count;
> +	xfs_agnumber_t		agno = XFS_INO_TO_AGNO(mp, *lastino);
> +	xfs_agino_t		agino = XFS_INO_TO_AGINO(mp, *lastino);
> +	int			left = *count;
> +	struct xfs_btree_cur	*cur = NULL;
> +	struct xfs_buf		*agbp = NULL;
> +	struct xfs_inogrp	*buffer;
> +	int			bcount;
> +	int			bufidx;
> +	int			error;
> +
>  	*count = 0;
> +	if (agno >= mp->m_sb.sb_agcount ||
> +	    *lastino != XFS_AGINO_TO_INO(mp, agno, agino))
> +		return 0;
> +
>  	bcount = MIN(left, (int)(PAGE_SIZE / sizeof(*buffer)));
>  	buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
> -	error = bufidx = 0;
> -	cur = NULL;
> -	agbp = NULL;
> -	while (left > 0 && agno < mp->m_sb.sb_agcount) {
> -		if (agbp == NULL) {
> +	bufidx = error = 0;
> +	do {
> +		struct xfs_inobt_rec_incore	r;
> +		int				stat;
> +
> +		if (!agbp) {
>  			error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
> -			if (error) {
> -				/*
> -				 * If we can't read the AGI of this ag,
> -				 * then just skip to the next one.
> -				 */
> -				ASSERT(cur == NULL);
> -				agbp = NULL;
> -				agno++;
> -				agino = 0;
> -				continue;
> -			}
> +			if (error)
> +				break;
>  			cur = xfs_inobt_init_cursor(mp, NULL, agbp, agno);
> -			error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE,
> -						 &tmp);
> -			if (error) {
> -				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
> -				cur = NULL;
> -				xfs_buf_relse(agbp);
> -				agbp = NULL;
> -				/*
> -				 * Move up the last inode in the current
> -				 * chunk.  The lookup_ge will always get
> -				 * us the first inode in the next chunk.
> -				 */
> -				agino += XFS_INODES_PER_CHUNK - 1;
> -				continue;
> -			}
>  		}
> -		error = xfs_inobt_get_rec(cur, &r, &i);
> -		if (error || i == 0) {
> -			xfs_buf_relse(agbp);
> -			agbp = NULL;
> -			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
> -			cur = NULL;
> -			agno++;
> -			agino = 0;
> -			continue;
> +		error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_GE, &stat);
> +		if (error)
> +			break;

Isn't this lookup only needed after cursor initialization? i.e., we
lookup once and increment through the records via xfs_btree_increment()
below.

> +		if (!stat) {
> +			/* Done, proceed to look up the next AG */
> +			goto next_ag;
>  		}
> +
> +		error = xfs_inobt_get_rec(cur, &r, &stat);
> +		if (error)
> +			break;
> +		XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
> +
>  		agino = r.ir_startino + XFS_INODES_PER_CHUNK - 1;
>  		buffer[bufidx].xi_startino =
>  			XFS_AGINO_TO_INO(mp, agno, r.ir_startino);
>  		buffer[bufidx].xi_alloccount =
>  			XFS_INODES_PER_CHUNK - r.ir_freecount;
>  		buffer[bufidx].xi_allocmask = ~r.ir_free;
> -		bufidx++;
> -		left--;
> -		if (bufidx == bcount) {
> -			long written;
> -			if (formatter(ubuffer, buffer, bufidx, &written)) {
> -				error = XFS_ERROR(EFAULT);
> +		if (++bufidx == bcount) {
> +			long	written;
> +			error = formatter(ubuffer, buffer, bufidx, &written);
> +			if (error)
>  				break;
> -			}
>  			ubuffer += written;
>  			*count += bufidx;
>  			bufidx = 0;
>  		}
> -		if (left) {
> -			error = xfs_btree_increment(cur, 0, &tmp);
> -			if (error) {
> -				xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
> -				cur = NULL;
> -				xfs_buf_relse(agbp);
> -				agbp = NULL;
> -				/*
> -				 * The agino value has already been bumped.
> -				 * Just try to skip up to it.
> -				 */
> -				agino += XFS_INODES_PER_CHUNK;
> -				continue;
> -			}
> +		if (!--left)
> +			break;
> +
> +		error = xfs_btree_increment(cur, 0, &stat);
> +		if (error)
> +			break;
> +		if (stat) {
> +			/*
> +			 * The agino value has already been bumped, just try
> +			 * to skip up to it.
> +			 */
> +			agino += XFS_INODES_PER_CHUNK;
> +			continue;
>  		}

Maybe it's just me, but this reads a little funny to me. In particular
because we only get here if stat == 1. I wonder if this would look a bit
cleaner if we pulled the next_ag labeled block below up into the goto,
since that appears to be the only reference. Then just let the loop fall
through.

Also, I think the agino addition here becomes unnecessary when the
lookup issue above is addressed.

> -	}
> +
> +next_ag:
> +		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
> +		cur = NULL;
> +		xfs_buf_relse(agbp);
> +		agbp = NULL;
> +		agino = 0;
> +	} while (++agno < mp->m_sb.sb_agcount);
> +

... and just thinking about the logic that way highlights the bug here,
where we bump agno due to the continue above (where IIUC, we intend to
only move forward within the ag). Perhaps the ++agno should be part of
the broken off 'next_ag' logic as well.

Brian

>  	if (!error) {
>  		if (bufidx) {
> -			long written;
> -			if (formatter(ubuffer, buffer, bufidx, &written))
> -				error = XFS_ERROR(EFAULT);
> -			else
> +			long	written;
> +			error = formatter(ubuffer, buffer, bufidx, &written);
> +			if (!error)
>  				*count += bufidx;
>  		}
>  		*lastino = XFS_AGINO_TO_INO(mp, agno, agino);
>  	}
> +
> +error0:
>  	kmem_free(buffer);
>  	if (cur)
>  		xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
> 

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

  reply	other threads:[~2014-01-03 20:53 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-12-28 11:20 [PATCH 3/10] xfs: consolidate xfs_inumbers Jeff Liu
2014-01-03 20:53 ` Brian Foster [this message]
2014-01-06  6:23   ` Jeff Liu
2014-01-06 15:05     ` Brian Foster
2014-01-07  6:58       ` Jeff Liu

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=52C72362.2000005@redhat.com \
    --to=bfoster@redhat.com \
    --cc=jeff.liu@oracle.com \
    --cc=xfs@oss.sgi.com \
    /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).