[arch-dev-public] udev changes - and the comeback of a long lost feature
tpowa did some work today updating udev to 139, cleaning up the rules and fixing some small bugs. With it comes the implementation of a recommendation of the udev team: We now have a directory /lib/udev/devices/ which contains a number of static devices nodes which are always present. A udev rule also makes sure that none of these devices is deleted even if udev normally would do so (the stock udev rule for that is broken, but I fixed it). This simplifies our rc.sysinit a bit: http://projects.archlinux.org/?p=initscripts.git;a=commitdiff;h=623e3fac060b... In that context, I thought of an old feature: "module autoloading" as Linux 2.4 understood it: Imagine you have a static /dev, and you try to open a device: 1) /dev/XYZ is a block/char device with major X and minor Y 2) the kernel looks it up, but doesn't find it 3) the kernel issues "modprobe block-major-X-Y" or "modprobe char-major-X-Y" 4) the kernel looks up the device again, it can now find it Note that step 1 requires that /dev/XYZ exists! This feature completely broke with the introduction of udev. This feature is obsolete for most devices, but there are those devices that are not bound to physical hardware - best examples here are tun and loop. I am going to add the loop and tun (any maybe fuse) devices to the static device list that we now support. That means that you can use loop mounts again without running "modprobe loop" manually, the module will be loaded as soon as you access /dev/loopX. Same goes for /dev/net/tun or /dev/fuse. This basically works with all modules that have a block-major-X-* or char-major-X-* alias, so if you think of any more modules that do this, we could add them too.
On Sat, Mar 7, 2009 at 5:55 PM, Thomas Bächler <thomas@archlinux.org> wrote:
tpowa did some work today updating udev to 139, cleaning up the rules and fixing some small bugs.
With it comes the implementation of a recommendation of the udev team: We now have a directory /lib/udev/devices/ which contains a number of static devices nodes which are always present. A udev rule also makes sure that none of these devices is deleted even if udev normally would do so (the stock udev rule for that is broken, but I fixed it).
This simplifies our rc.sysinit a bit: http://projects.archlinux.org/?p=initscripts.git;a=commitdiff;h=623e3fac060b...
In that context, I thought of an old feature: "module autoloading" as Linux 2.4 understood it: Imagine you have a static /dev, and you try to open a device:
1) /dev/XYZ is a block/char device with major X and minor Y 2) the kernel looks it up, but doesn't find it 3) the kernel issues "modprobe block-major-X-Y" or "modprobe char-major-X-Y" 4) the kernel looks up the device again, it can now find it Note that step 1 requires that /dev/XYZ exists! This feature completely broke with the introduction of udev. This feature is obsolete for most devices, but there are those devices that are not bound to physical hardware - best examples here are tun and loop.
I am going to add the loop and tun (any maybe fuse) devices to the static device list that we now support. That means that you can use loop mounts again without running "modprobe loop" manually, the module will be loaded as soon as you access /dev/loopX. Same goes for /dev/net/tun or /dev/fuse. This basically works with all modules that have a block-major-X-* or char-major-X-* alias, so if you think of any more modules that do this, we could add them too.
Nice, all these changes look good. Thanks for doing this. -Dan
participants (2)
-
Dan McGee
-
Thomas Bächler