Re: [PATCH] xfs_logprint: Fix super block buffer interpretation issue

[Chi Zhiling]

On 2024/10/14 06:50, Dave Chinner wrote:
> On Sun, Oct 13, 2024 at 12:00:22PM +0800, Chi Zhiling wrote:
> > On 2024/10/13 06:10, Dave Chinner wrote:
> > > On Fri, Oct 11, 2024 at 11:54:08AM +0800, Chi Zhiling wrote:
> > > > On 2024/10/11 11:24, Darrick J. Wong wrote:
> > > > > On Fri, Oct 11, 2024 at 11:08:10AM +0800, Chi Zhiling wrote:
> > > > > > From: chizhiling<chizhiling@xxxxxxxxxx>
> > > > > >
> > > > > > When using xfs_logprint to interpret the buffer of the super block, the
> > > > > > icount will always be 6360863066640355328 (0x5846534200001000). This is
> > > > > > because the offset of icount is incorrect, causing xfs_logprint to
> > > > > > misinterpret the MAGIC number as icount.
> > > > > > This patch fixes the offset value of the SB counters in xfs_logprint.
> > > > > >
> > > > > > Before this patch:
> > > > > > icount: 6360863066640355328  ifree: 5242880  fdblks: 0  frext: 0
> > > > > >
> > > > > > After this patch:
> > > > > > icount: 10240  ifree: 4906  fdblks: 37  frext: 0
> > > > > >
> > > > > > Signed-off-by: chizhiling<chizhiling@xxxxxxxxxx>
> > > > > > ---
> > > > > >     logprint/log_misc.c | 8 ++++----
> > > > > >     1 file changed, 4 insertions(+), 4 deletions(-)
> > > > > >
> > > > > > diff --git a/logprint/log_misc.c b/logprint/log_misc.c
> > > > > > index 8e86ac34..21da5b8b 100644
> > > > > > --- a/logprint/log_misc.c
> > > > > > +++ b/logprint/log_misc.c
> > > > > > @@ -288,13 +288,13 @@ xlog_print_trans_buffer(char **ptr, int len, int *i, int num_ops)
> > > > > >     			/*
> > > > > >     			 * memmove because *ptr may not be 8-byte aligned
> > >                                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > >
> > > This is important. I'll come back to it.
> > >
> > > > > >     			 */
> > > > > > -			memmove(&a, *ptr, sizeof(__be64));
> > > > > > -			memmove(&b, *ptr+8, sizeof(__be64));
> > > > > How did this ever work??  This even looks wrong in "Release_1.0.0".
> > > > Yes, I was surprised when I find this issue
> > > I"ve never cared about these values when doing diagnosis because
> > > lazy-count means they aren't guaranteed to be correct except at
> > > unmount. At which point, the correct values are generally found
> > > in the superblock. IOWs, the values are largely meaningless whether
> > > they are correct or not, so nobody has really cared enough about
> > > this to bother fixing it...
> > Because I got a log which shows that the fdblocks was (-8),   it caused
> > the filesystem to fail mounting again. 'SB summary counter sanity check failed'
> What kernel? Because AFAIK, that was fixed in commit 58f880711f2b
> ("xfs: make sure sb_fdblocks is non-negative") in 6.10...

It's a 4.19 kernel. As you said, the fdblocks is meaningless, I think 
that patch (commit 58f880711f2b) is enough to fix the issue.

Thank you for your reminding.


chi