Re: df command reports incorrect usage

[Rob Landley]

On 3/22/22 9:12 AM, Fariya F wrote:
> Hi,
> 
> The version of df command is 1.85.
> Kernel version 4.9.31

A 2016 release, plus bugfixes.

$ git log --no-merges v4.9..master fs/ext4 | grep ^commit | wc -l
1213

Bit of water under that bridge in the past 8 years.

> The fsck check on that partition succeeds. I am able to use the partition,
> create files etc. However df reports incorrectly the size and used
> percentage.
> 
> Output from strace command is as follows;
> 
> statfs64("/data", 88, {f_type="EXT2_SUPER_MAGIC",
> f_bsize=1024, f_blocks=18446744073659310077, f_bfree=87628, f_bavail=80460,
> f_files=25688, f_ffree=25189, f_fsid={-1446355608, 1063639410},
> f_namelen=255, f_frsize=1024, f_flags=4128}) = 0
> 
> As can be seen, value of f_blocks is a huge one. How can we get this
> corrected and where is the corruption because of which this value goes
> wrong?

In the kernel's ext4 driver. 0xfffffffffd015ffdL looks like a negative number
mapped to positive (around -50241539). Possibly something returned an error
value that didn't get checked, or there was an underflow, or...?

Is this 32 bit hardware? That gets less testing these days than 64 bit does,
could be a "long" where a "long long" is needed? (In the ext4 driver.)

Probably fixed in one of the thousand-plus commits to fs/ext4 since then. I'd
start with:

$ git log --no-merges v4.9..master fs/ext4 | egrep '(over|under)flow' | wc -l
40

But it could also be "error return used as a result" which is "track it down
with a bunch of printk()" territory for me.

Another thing you could do is copy your image to file (cat /dev/sdx > file.img)
and use a VM image to read it (I use https://landley.net/toybox/faq.html#mkroot
for this sort of thing) and then see if current kernels can read it properly (if
not, report it to lkml), and if it WAS fixed git bisect between v4.9 and master
to find the commit that fixed it. (Remember the bad-is-good swap in that case.)

Hmmm, I'm making tutorial videos now... How big is this storage device, and does
it have anything private on it?

Rob