On Fri, Aug 31, 2018 at 09:13:31AM -0700, Greg KH wrote: > On Thu, Aug 30, 2018 at 05:10:53PM -0400, Alan Stern wrote: > > On Thu, 30 Aug 2018, Faisal Mehmood wrote: > > > > > Based on my (limited) understanding if I were to disable udev, the > > > userspace should not be able to enumerate/interact with any newly > > > connected device since udev handles uevents generated by kernel. > > > (right?) > > > > That's not quite right. Yes, udev does handle uevents. But a newly > > connected device may be usable from userspace, to some extent, without > > any handling of uevents. > > > > > So as a test, I disabled systemd-udevd and then plugged in a flash > > > drive. I disabled support for usb mass storage in the kernel. So 'lsblk' > > > didn't show the device, just as expected. However, 'lsusb' still > > > dynamically updates the list as I repeatedly plugin and remove a flash > > > drive. > > > > An example of what I described above. > > > > > Then I studied libusb/lsusb code for a bit. It seems that lsusb makes use > > > of 'libusb_get_device_list()' to find the connected usb devices. But I > > > couldn't find any detail regarding implementation of > > > libusb_get_device_list. > > > > That's an internal aspect of libusb, so it might not be described in > > the documentation. You can always check the source code if you really > > want to know how it works. I think it may look at the files under > > /sys/bus/usb/devices/ -- that's certainly one possibility which would > > always work even without udev. > > > > > I am sure I'm missing something. Could someone please help me understand > > > what is happening behind the scenes? > > > > udev does things like loading drivers into the kernel (modprobe'ing > > them) and creating nodes under /dev. But a driver that is already > > present in the kernel doesn't need to be modprobe'd, and the files > > under /sys are created automatically by the kernel rather than by udev. > > Also, udev has not created /dev nodes for probably a decade now. That's > what devtmpfs does. udev can create symlinks to existing dev nodes, and > some "special-case" device nodes where the kernel name is not what > userspace wants to use. > > But really, why would anyone _want_ to remove udev, it's such a nice > program and solves so many problems :) > I wanted to learn how Linux handles plug and play devices. Thanks for pointing out devtmpfs. I guess making and breaking things is a good way of learning. > thanks, > > greg k-h BTW great work with Linux Kernel in a Nutshell. It's still a fantastic resource for kernel modders/newbies despite being over a decade old. Thanks for the helpful comments guys. Regards, Faisal
