xfs
[Top] [All Lists]

Re: [PATCH V5] xfs_db: add crc manipulation commands

To: Eric Sandeen <sandeen@xxxxxxxxxxx>
Subject: Re: [PATCH V5] xfs_db: add crc manipulation commands
From: Dave Chinner <david@xxxxxxxxxxxxx>
Date: Mon, 22 Aug 2016 10:54:41 +1000
Cc: xfs@xxxxxxxxxxx
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <0bb88897-3dfc-de4e-05fb-b737221f0d80@xxxxxxxxxxx>
References: <8cb34b7c-5154-ebb4-7706-1375e94abd74@xxxxxxxxxx> <0bb88897-3dfc-de4e-05fb-b737221f0d80@xxxxxxxxxxx>
User-agent: Mutt/1.5.21 (2010-09-15)
On Sun, Aug 14, 2016 at 07:30:51PM -0500, Eric Sandeen wrote:
> This adds a new "crc" command to xfs_db for CRC-enabled filesystems.
> 
> If a structure has a CRC field, we can validate it, invalidate/corrupt
> it, or revalidate/rewrite it:
> 
> xfs_db> sb 0
> xfs_db> crc -v
> crc = 0x796c814f (correct)
> xfs_db> crc -i
> Metadata CRC error detected at block 0x0/0x200
> crc = 0x796c8150 (bad)
> xfs_db> crc -r
> crc = 0x796c814f (correct)
> 
> (-i and -r require "expert" write-capable mode)
> 
> This requires temporarily replacing the write verifier with
> a dummy which won't recalculate the CRC on the way to disk.
> 
> It also required me to write a new flist function, which is
> totally foreign to me, so hopefully done right - but it seems
> to work here.
> 
> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>

Actually, this need more help....

> @@ -473,12 +474,15 @@ xfs_verify_recalc_crc(
>  void
>  write_cur(void)
>  {
> +     int skip_crc = (iocur_top->bp->b_ops->verify_write == xfs_dummy_verify);
> +

bool, and we can't use the iocur until we've checked if it's valid.
i.e. after this first check:

>       if (iocur_sp < 0) {
>               dbprintf(_("nothing to write\n"));
>               return;
>       }
>  
> -     if (xfs_sb_version_hascrc(&mp->m_sb) && iocur_top->ino_buf) {
> +     if (xfs_sb_version_hascrc(&mp->m_sb) && iocur_top->ino_buf &&
> +         !skip_crc) {
>               libxfs_dinode_calc_crc(mp, iocur_top->data);
>               iocur_top->ino_crc_ok = 1;
>       }

The line of code below this calculates CRCs for dquot buffers, so it
seems kind of broken to be skipping crc update for an inode but not
a dquot given they are updated in the same way....

Shouldn't this "skip_crc" condition also be applied to dquot
buffers, too?

Cheers,

Dave.
-- 
Dave Chinner
david@xxxxxxxxxxxxx

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