Possible patch attached. This maps all the "numbered" device filenames that were previously recognized to the same grub device as before:

• /dev/md_dNN -> mdNN
• /dev/md_dNNpMM -> mdNN,MM
• /dev/md/dNN -> mdNN
• /dev/md/dNNpMM -> mdNN,MM
• /dev/md/NN -> mdNN
• /dev/md/NNpMM -> mdNNpMM
• /dev/mdNN -> mdNN
• /dev/mdNNpMM -> mdNN,MM

And it allows the following "named" device filenames:

• /dev/md/NAME -> md/NAME (this is unchanged)
• /dev/md_NAME -> md/NAME (this is new)

In addition, the entire device filename is checked, instead of relying only on the prefix, allowing the name in the "named" filenames to begin with a 'd' or a number. This is still not perfect -- naming an array "d1" will still give md1 instead of md/d1, but giving an array such a misleading name is probably a bad idea anyway.

The device is still rejected even if an mduuid/UUID style device could be found. This is unchanged behavior.

Finally, named devices are now always treated as unpartitioned. This is because it's pretty much impossible to figure out the layout from the name alone: /dev/md/ap1p2 could refer to either

1. an array named ap1p2,
2. partition 2 of an array named ap1p, or
3. partition 2 of an array named ap1.

This patch always chooses option 1. The previous behavior was inconsistent:

• if the UUID could be obtained:
  • /dev/md/a2 was mapped to mduuid/UUID
  • /dev/md/a1p2 was mapped to mduuid/UUID,2
  • /dev/md/ap1p2 was mapped to mduuid/UUID,2
• if the UUID could not be obtained:
  • /dev/md/a2 was mapped to md/a2
  • /dev/md/a1p2 was mapped to md/a1,2
  • /dev/md/ap1p2 was mapped to md/a,1p2

(file #53423)