RFE: Parse relevant messages from `/dev/kmsg` for mount calls

[tbzatek]

trafficstars

Syscalls (e.g. mount) or writing to certain device files (e.g. /dev/nvme-fabrics) only tend to return errno with no further context. This is too generic and makes users often confused.

The idea here is to open a side channel to /dev/kmsg and read relevant messages that are structured and do provide device context.

6,888,4307125,-;nvme nvme0: creating 6 I/O queues.
 SUBSYSTEM=nvme
 DEVICE=c241:0
6,889,4313950,-;nvme nvme0: mapped 6/0/0 default/read/poll queues.
 SUBSYSTEM=nvme
 DEVICE=c241:0
6,890,4315079,-;nvme nvme0: new ctrl: NQN "nqn.2014-08.org.nvmexpress.boot.poc:test-target", addr 192.168.122.1:4420
 SUBSYSTEM=nvme
 DEVICE=c241:0

3,892,7382617,-;nvme nvme1: failed to connect socket: -110
 SUBSYSTEM=nvme
 DEVICE=c241:1

However, not so many drivers actually provide context.

5,1035,31137642,-;XFS (nvme0n1p2): Mounting V5 Filesystem
6,1036,31157902,-;XFS (nvme0n1p2): Ending clean mount

Since /dev/kmsg is likely a ring buffer, messages need to be filtered out by timestamps. Either read it full (by a non-blocking read) before spawning the syscall or retrieve sequence number from the last record.

There's a high chance of mixing messages from multiple syscalls on the same device. Reliability of such filtering is a question.

[tbzatek]

Actually, as of kernel-6.6.2, there's not a single filesystem driver that provides structured messages. Tested btrfs, ext4, f2fs, nilfs, ntfs3, udf and xfs.

Until the situation improves, there's no point of making an attempt to implement this.