[FixeQD]

I've found them while writing the finix keventd's VM tests, only the keventd backend failed to create virtio-<serial>-part1 after hotplugging a GPT disk.

Two bugs:

1. $attr{} / %s{} read only the device's own sysfs directory, while the matching key ATTRS{} walks parents. Partitions have no serial of their own, so ID_SERIAL came out empty and the shipped rule published virtio--part1. Affects the nvme, ieee1394 and mmc partition rules too.
   The fix adds sysfs_read_attr_chain(), which falls back to sysfs_parent_walk().
2. disk_serial_id() falls back to <disk>/../serial, so a partition got a by-id link named after its parent disk's serial, overwriting the whole disk's link.
   The fix skips for DEVTYPE=partition.

[troglobit]

@copilot resolve the merge conflicts in this pull request

[FixeQD]

Uhhh... Is it going to start? Or I should resolve them, they look easy

[troglobit]

Uhhh... Is it going to start? Or I should resolve them, they look easy

Good question, I'll have a look. Would be neat to not have to do such stuff ourselves :-)

[FixeQD]

Alr, btw welcome back to finit :D

[troglobit]

Alr, btw welcome back to finit :D

Heh, yeah I've been in the wars lately. A bit too much going on at $DAYJOB atm. 😅

[troglobit]

Copilot failed miserably at this simple task, so I've cherry-picked the partition fix (your second commit) onto master as c03059a. The first commit conflicted with #516, which fixes the same thing the way udev does it: $attr{} falls back to the parent matched by ATTRS{serial}=="?*". Your virtio -part1 case is covered by that, so I've left the first commit out.

[FixeQD]

Alr, thanks