The 2.5 device model
Posted Aug 30, 2002 3:20 UTC (Fri) by
otterley (guest, #3500)
Parent article:
The 2.5 device model
If driverfs is actually implemented, devfs should simply map /dev shortcuts to drivefs leaf nodes, which in turn should be the actual device files.
Whatever driverfs does, the tree model should be bound as tightly as possible to the devices' physical layout. One example I'd like us to avoid repeating is having Fibre Channel devices named like SCSI devices, because Fibre Channel devices have unique World Wide Names (WWNs) for addressing purposes. I want to be capable of assuring myself that the disk I'm trying to access is the correct disk; I'm tired of having my disks haphazardly assigned to device nodes and ending up accessing the wrong disk because the kernel decided that disk 0x210100e08b245b51 that was exported as /dev/sda before I rebooted is now at /dev/sdb.
devfs can still serve a purpose here, to help map user-friendly names (/dev/sdX) to physical pathnames -- similar to Solaris' path_to_inst system. But it should be entirely optional to use.
Solaris has had this right since the early '90s; it's about time Linux got with the program.
(
Log in to post comments)