On 03/11/2013 08:39 PM, Tyler Hobbs wrote:
> If I run stat --printf='%D', the result is "10ca70", which is correct.
> However, if I run stat --printf='%t %T' /mountpoint, the result is
> erroneously "0 0".  If I instead run stat against the device directly (stat
> --printf='%t %T' /dev/xvdx), I get the correct result of "ca 170".
> 
> I believe the proper fix is to replace (in stat.c):
> 
>   out_uint_x (pformat, prefix_len, major (statbuf->st_rdev));
> 
> with:
> 
>   out_uint_x (pformat, prefix_len, major (statbuf->st_dev));
> 
> That is, use statbuf->st_dev instead of st_rdev, which is what the %d and
> %D directives use.
> 
> 
> I'm using coreutils 8.9, compiled from source, and this is the output of
> uname -a:
> 
> Linux ip-10-39-122-238 2.6.32-276.el6.x86_64 #1 SMP Tue May 29 17:38:19 EDT
> 2012 x86_64 x86_64 x86_64 GNU/Linux
> 
> Thanks for your time.
> - Tyler
> 

So %t and %T are only currently defined for device special files,
allowing one to distinguish between the represented device
and the device of the inode storing the representation (the special file).

$ stat -c '%T%t %D' /dev/sda /dev
08 5
00 5

Now %t and %T returning 0 for non special device nodes is not that useful.
I suppose in this case we might use st_rdev only for block and char specials,
and switch to st_dev otherwise.

However...

st_rdev is not always 0
I notice that st_rdev is not 0 on FreeBSD 9 for files.
I don't know what it represents, but it's non random so may be significant.
Also the man page I have here says that on XENIX
named special file subtypes are distinguished by st_rdev values

Also is it useful to get low level access like this
to the backing device major and minor for normal files
from a shell script?

So I think we'll just improve this through documentation.

Hopefully the attached clarifies things.

thanks,
Pádraig.