bug#21018: Support for "libnvdimm BLK" devices

[Brian C. Lane]

On Thu, Jul 09, 2015 at 10:00:10AM -0700, Dan Williams wrote:
> BLK mode devices emitted by the libnvdimm sub-system [1] have a "." in
> their name.  When parted calls _probe_sys_block() it trips over this
> fact and unconditionally re-writes the block device name.
> 
>                 /* in /sys/block, '/'s are replaced with '!' or '.' */
>                 for (ptr = dev_name; *ptr != '\0'; ptr++) {
>                         if (*ptr == '!' || *ptr == '.')
>                                 *ptr = '/';
>                 }
> 
> This is confirmed by strace where parted emits calls to
> 
> lstat("/dev/ndblk2", 0x7ffec6b33e50)    = -1 ENOENT (No such file or 
> directory)
> stat("/dev/ndblk2/0", 0x7ffec6b33dc0)   = -1 ENOENT (No such file or 
> directory)
> 
> The original BLK device name is /dev/ndblk2.0.
> 
> This can be reproduced with a 4.2-rc1+ kernel and the nfit_test
> module.  See the libndctl readme for instructions on getting the test
> environment running [2].
> 
> [1]: 
> https://git.kernel.org/cgit/linux/kernel/git/djbw/nvdimm.git/tree/Documentation/nvdimm/nvdimm.txt?h=libnvdimm-for-next
> 
> [2]: https://github.com/pmem/ndctl

Thanks for the report. I think we can reduce this to just replacing '!',
as far as I can tell cciss is the only device that has subdirectories
and uses the ! in the name. Looking at the current kernel code it looks
like the substitution is done in drivers/base/core.c ->
device_get_devnode()

I'm no kernel expert though. What does everyone think about dropping the
'.'? Do you know of anything that this would break?

In blivet (our storage library which handles all the devices we've ever
seen) it only looks for ! so that's reassuring.

-- 
Brian C. Lane | Anaconda Team | IRC: bcl #anaconda | Port Orchard, WA (PST8PDT)