More sysfs symlinks [LWN.net]

June 21, 2006

This article was contributed by Greg Kroah-Hartman.

Another new change in sysfs that will be in the 2.6.18 kernel release, is the addition of another symlink in all device and class device directories. Kay Sievers has written a patch that adds the symlink "subsystem" to these directories. This symlink points back to either the class that the device is associated with, or the bus that the device is associated with.

This symlink is identical to the information that the kernel has always been emitting to userspace through the hotplug interface whenever a device was created or removed from the system. Userspace uses the subsystem information in order to determine what to do with the device.

If you look at the older hotplug package, it is broken down into a set of different scripts that run depending on the subsystem that is being addressed:

$ ls /etc/hotplug/*.rc
/etc/hotplug/input.rc
/etc/hotplug/isapnp.rc
/etc/hotplug/pci.rc
/etc/hotplug/pnp.rc
/etc/hotplug/usb.rc

And udev rules also act on the subsystem type in order to determine what to do with the device:

$ head -n 3 /etc/udev/rules.d/05-udev-early.rules
# ignore these events until someone needs them
SUBSYSTEM=="drivers",   OPTIONS="ignore_device"
SUBSYSTEM=="module",    OPTIONS="ignore_device"

But before this kernel patch, if a program wanted to walk through sysfs and try to determine the subsystem that a specific device was associated with, they had to do the following steps:

If this is a device, look for the bus symlink and follow it.
If this is a class device, go up a directory and see if this is a class directory. If not, go up another directory, until the class is found.

Now, with the subsystem symlink, this logic can be greatly simplified, as only this symlink needs to be followed in order to determine the subsystem that the device is associated with:

$ tree /sys/class/tty/ttyS0/
/sys/class/tty/ttyS0/
|-- dev
|-- device -> ../../../devices/platform/serial8250
|-- subsystem -> ../../../class/tty
`-- uevent

$ tree /sys/devices/pci0000:00/0000:00:00.0/
/sys/devices/pci0000:00/0000:00:00.0/
|-- broken_parity_status
|-- bus -> ../../../bus/pci
|-- class
|-- config
|-- device
|-- driver -> ../../../bus/pci/drivers/e752x_edac
|-- enable
|-- irq
|-- local_cpus
|-- modalias
|-- power
|   |-- state
|   `-- wakeup
|-- resource
|-- subsystem -> ../../../bus/pci
|-- subsystem_device
|-- subsystem_vendor
|-- uevent
`-- vendor

Index entries for this article
GuestArticles	Kroah-Hartman, Greg

to post comments

More sysfs symlinks

Posted Jun 22, 2006 7:36 UTC (Thu) by nix (subscriber, #2304) [Link] (2 responses)

I'm not really sure what this buys you. If you could rip out the old code it would reduce complexity a bit --- but if you do that, you have *another* version of udev that won't work on older kernels. So you'd need to maintain the old and new code in parallel, and given the not-terribly-great complexity of the old code, is the result worth it?

More sysfs symlinks

Posted Jun 23, 2006 0:00 UTC (Fri) by Tobu (subscriber, #24111) [Link] (1 responses)

Change must happen "sometime". Think of the future generations :)

More sysfs symlinks

Posted Jun 23, 2006 9:03 UTC (Fri) by nix (subscriber, #2304) [Link]

Yeah, but this change doesn't let you simplify the userspace consumers of sysfs at all, at least not until they drop kernel <2.6.18 compatibility, which is hopefully not going to be for a while!