Re: [PATCH] modules: add modalias file to sysfs for modules.

From: Allen Webb
Date: Wed Nov 16 2022 - 15:42:25 EST


I think I am misunderstanding something because the aliases visible
through kmod_config_get_aliases don't look anything like the
modaliases used by udev to match devices to modules.

I have some concern that making the changes necessary to use kmod
would result in the files/indexes being much larger to include all the
extra values that are being left out, so I think the sysfs approach
might be the better way to go.

On Tue, Nov 15, 2022 at 12:40 PM Allen Webb <allenwebb@xxxxxxxxxx> wrote:
>
> When i get the modinfo for usbhid, there are no aliases listed:
> ```
> localhost ~ # modinfo usbhid
> name: usbhid
> filename: (builtin)
> author: Andreas Gal
> author: Vojtech Pavlik
> author: Jiri Kosina
> description: USB HID core driver
> file: drivers/hid/usbhid/usbhid
> license: GPL
> parm: quirks:Add/modify USB HID quirks by specifying
> quirks=vendorID:productID:quirks where vendorID, productID, and quirks
> are all in 0x-prefixed hex (array of charp)
> parm: ignoreled:Autosuspend with active leds (uint)
> parm: kbpoll:Polling interval of keyboards (uint)
> parm: jspoll:Polling interval of joysticks (uint)
> parm: mousepoll:Polling interval of mice (uint)
> ```
>
> bluetooth however has an alias listed:
> ```
> localhost ~ # modinfo bluetooth
> filename:
> /lib/modules/5.10.154-20424-gea7532c123d8/kernel/net/bluetooth/bluetooth.ko.gz
> author: Marcel Holtmann <marcel@xxxxxxxxxxxx>
> description: Bluetooth Core ver 2.22
> version: 2.22
> license: GPL
> alias: net-pf-31
> vermagic: 5.10.154-20424-gea7532c123d8 SMP preempt mod_unload
> name: bluetooth
> intree: Y
> retpoline: Y
> depends: ecdh_generic
> srcversion: F8E46CD048C50B0AA1CD471
> parm: disable_esco:Disable eSCO connection creation (bool)
> parm: enable_ecred:Enable enhanced credit flow control mode (bool)
> parm: disable_ertm:Disable enhanced retransmission mode (bool)
> ```
>
> I believe the reason for this is many modules use `#define
> MODULE_DEVICE_TABLE(type, name)` which is a noop for buildin modules.
> I have a local patch that resolves that issue here:
> See: https://chromium-review.googlesource.com/c/chromiumos/third_party/kernel/+/3840672/1/include/linux/module.h#b246
>
> However, I probably ought to rework that patch to create the
> MODULE_ALIAS defines instead of the buildin.alias file.
>
> On Tue, Nov 15, 2022 at 11:35 AM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> >
> > On Tue, Nov 15, 2022 at 10:05:35AM -0600, Allen Webb wrote:
> > > On Mon, Nov 14, 2022 at 11:22 AM Lucas De Marchi
> > > <lucas.demarchi@xxxxxxxxx> wrote:
> > > >
> > > > On Mon, Nov 14, 2022 at 10:42:50AM -0600, Allen Webb wrote:
> > > > >On Fri, Nov 11, 2022 at 12:29 PM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> > > > >>
> > > > >> On Fri, Nov 11, 2022 at 09:28:52AM -0600, Allen Webb wrote:
> > > > >> > USB devices support the authorized attribute which can be used by
> > > > >> > user-space to implement trust-based systems for enabling USB devices. It
> > > > >> > would be helpful when building these systems to be able to know in
> > > > >> > advance which kernel drivers (or modules) are reachable from a
> > > > >> > particular USB device.
> > > > >> >
> > > > >> > This information is readily available for external modules in
> > > > >> > modules.alias. However, builtin kernel modules are not covered. This
> > > > >> > patch adds a sys-fs attribute to both builtin and loaded modules
> > > > >> > exposing the matching rules in the modalias format for integration
> > > > >> > with tools like USBGuard.
> > > > >> >
> > > > >> > Signed-off-by: Allen Webb <allenwebb@xxxxxxxxxx>
> > > > >>
> > > > >> Thanks for the patch Allen!
> > > > >>
> > > > >> I'd rather have something generic though, and it would seem kmod [0] already
> > > > >> does this, have you seen the kmod support for builtin.alias.bin
> > > > >>
> > > > >> Can't that be used?
> > > > >
> > > > >Probably, but I don't see the builtin.alias.bin in my build. Is it experimental?
> > > >
> > > > no. That is generated by depmod since v27 using modules.builtin.modinfo
> > > > generated by the kernel build system. Highly recommend v30 though
> > > > as there were fixes in v28 and v29 and some changes to speed up its
> > > > generation/use in v30: See entries mentioning
> > > > builtin.alias and bultin.modinfo in
> > > > https://git.kernel.org/pub/scm/utils/kernel/kmod/kmod.git/tree/NEWS
> > > >
> > > > libkmod/modprobe/modinfo also have the corresponding changes to lookup that
> > > > index when resolving aliases.
> > >
> > > I see the file but it is largely missing the aliases I am interested
> > > in, so it looks like I might need to modify my patch that creates
> > > buildin.alias to add the missing alias defines in the header along
> > > with the other module metadata for builtin modules. Does this sound
> > > right to you?
> >
> > Can you clarify what is missing and why? And an RFC is welcomed if it
> > helps demonstrates what you mean.
> >
> > Luis