xfs
[Top] [All Lists]

Re: [PATCH V2] xfs: be honest about used inodes in statfs

To: Eric Sandeen <sandeen@xxxxxxxxxxx>
Subject: Re: [PATCH V2] xfs: be honest about used inodes in statfs
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Tue, 25 Feb 2014 10:56:14 +1100
Cc: Eric Sandeen <sandeen@xxxxxxxxxx>, xfs-oss <xfs@xxxxxxxxxxx>
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <530BD167.2020600@xxxxxxxxxxx>
References: <53067DC0.9040800@xxxxxxxxxx> <530BD167.2020600@xxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Mon, Feb 24, 2014 at 05:10:31PM -0600, Eric Sandeen wrote:
> Because we have lazy counters, it's possible that we over-allocate
> inodes past the maxicount (imaxpct) limit.  
> 
> A previous commit,
> 
>  2fe3366 xfs: ensure f_ffree returned by statfs() is non-negative
> 
> stopped statfs from underflowing f_ffree in this case, but that
> only happened when we mis-reported f_files, capped at maxicount.
> 
> Change statfs to report the actual number of inodes allocated,
> even if it is greater than maxicount.  It's reality.
> Deal with it.  
> 
> (New clearer code flow thanks to Brian!)
> 
> Logic-made-readable-by: Brian Foster <bfoster@xxxxxxxxxx>
> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>
> ---
> 
> V2: Use Brian's suggested logic for working out the numbers
> 
> diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
> index f317488..0dbcc17 100644
> --- a/fs/xfs/xfs_super.c
> +++ b/fs/xfs/xfs_super.c
> @@ -1083,7 +1083,6 @@ xfs_fs_statfs(
>       struct xfs_inode        *ip = XFS_I(dentry->d_inode);
>       __uint64_t              fakeinos, id;
>       xfs_extlen_t            lsize;
> -     __int64_t               ffree;
>  
>       statp->f_type = XFS_SB_MAGIC;
>       statp->f_namelen = MAXNAMELEN - 1;
> @@ -1100,17 +1099,19 @@ xfs_fs_statfs(
>       statp->f_blocks = sbp->sb_dblocks - lsize;
>       statp->f_bfree = statp->f_bavail =
>                               sbp->sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
> +     /*
> +      * Potential number of new inodes in free blocks, limited by maxicount.
> +      */
>       fakeinos = statp->f_bfree << sbp->sb_inopblog;

Can we rename "fakeinos" to something like "free_inodes" so that
the code reads a little bit better while we are touching this
code?

> -     statp->f_files =
> -         MIN(sbp->sb_icount + fakeinos, (__uint64_t)XFS_MAXINUMBER);
>       if (mp->m_maxicount)
> -             statp->f_files = min_t(typeof(statp->f_files),
> -                                     statp->f_files,
> -                                     mp->m_maxicount);
> +             fakeinos = mp->m_maxicount > sbp->sb_icount ?
> +                        MIN(mp->m_maxicount - sbp->sb_icount, fakeinos) : 0;

Get rid of MIN - it should be min() or min_t().

Also the mix of if() and ternary operations makes this difficult to
follow the logic. Better, IMO, is this:

        free_inodes = statp->f_bfree << sbp->sb_inopblog;
        if (mp->m_maxicount > sbp->sb_icount)
                free_inodes = min(mp->m_maxicount - sbp->sb_icount,
                                  free_inodes);
        else if (mp->m_maxicount)
                free_inodes = 0;


> +
> +     /* Total possible files is current inodes + potential new inodes */
> +     statp->f_files = MIN(sbp->sb_icount + fakeinos,
> +                          (__uint64_t) XFS_MAXINUMBER);

        statp->f_files = min_t(u64, sbp->sb_icount + free_inodes,
                                    XFS_MAXINUMBER);

And for bonus points: while we are looking at maxicount, the setting
on maxicount in the growfs code should call xfs_set_maxicount()
rather than open coding it, and xfs_set_maxicount() needs to be
reworked to prevent overflow when sbp->sb_dblocks * sbp->sb_imax_pct
is greater than 64 bits....

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

<Prev in Thread] Current Thread [Next in Thread>