In Chapter 8, we installed the udev daemon when systemd was built. Before we go into the details regarding how udev works, a brief history of previous methods of handling devices is in order.
Linux systems in general traditionally used a static device creation
method, whereby a great many device nodes were created under
/dev
(sometimes literally thousands of
nodes), regardless of whether the corresponding hardware devices
actually existed. This was typically done via a MAKEDEV script, which contained a
number of calls to the mknod program with the relevant
major and minor device numbers for every possible device that might
exist in the world.
Using the udev method, device nodes are only created for those
devices which are detected by the kernel. These device nodes are
created each time the system boots; they are stored in a devtmpfs
file system (a virtual file system that
resides entirely in system memory). Device nodes do not require much
space, so the memory that is used is negligible.
In February 2000, a new filesystem called devfs
was merged into the 2.3.46 kernel and was
made available during the 2.4 series of stable kernels. Although it
was present in the kernel source itself, this method of creating
devices dynamically never received overwhelming support from the
core kernel developers.
The main problem with the approach adopted by devfs
was the way it handled device detection,
creation, and naming. The latter issue, that of device node naming,
was perhaps the most critical. It is generally accepted that if
device names are configurable, the device naming policy should be
chosen by system administrators, and not imposed on them by the
developer(s). The devfs
file system
also suffered from race conditions that were inherent in its
design; these could not be fixed without a substantial revision of
the kernel. devfs
was marked as
deprecated for a long time, and was finally removed from the kernel
in June, 2006.
With the development of the unstable 2.5 kernel tree, later
released as the 2.6 series of stable kernels, a new virtual
filesystem called sysfs
came to be.
The job of sysfs
is to provide
information about the system's hardware configuration to userspace
processes. With this userspace-visible representation, it became
possible to develop a userspace replacement for devfs
.
The sysfs
filesystem was
mentioned briefly above. One may wonder how sysfs
knows about the devices present on a
system and what device numbers should be used for them. Drivers
that have been compiled into the kernel register their objects in
sysfs
(devtmpfs internally) as
they are detected by the kernel. For drivers compiled as modules,
registration happens when the module is loaded. Once the
sysfs
filesystem is mounted (on
/sys
), data which the drivers have
registered with sysfs
are
available to userspace processes and to udevd for processing
(including modifications to device nodes).
Device files are created by the kernel in the devtmpfs
file system. Any driver that wishes
to register a device node will use the devtmpfs
(via the driver core) to do it. When
a devtmpfs
instance is mounted on
/dev
, the device node will
initially be exposed to userspace with a fixed name, permissions,
and owner.
A short time later, the kernel will send a uevent to udevd. Based on the rules
specified in the files within the /etc/udev/rules.d
, /usr/lib/udev/rules.d
, and /run/udev/rules.d
directories, udevd will create additional
symlinks to the device node, or change its permissions, owner, or
group, or modify the internal udevd database entry (name) for
that object.
The rules in these three directories are numbered and all three
directories are merged together. If udevd can't find a rule for the
device it is creating, it will leave the permissions and
ownership at whatever devtmpfs
used initially.
Device drivers compiled as modules may have aliases built into
them. Aliases are visible in the output of the modinfo program and are usually
related to the bus-specific identifiers of devices supported by a
module. For example, the snd-fm801 driver supports PCI devices
with vendor ID 0x1319 and device ID 0x0801, and has an alias of
“pci:v00001319d00000801sv*sd*bc04sc01i*”.
For most devices, the bus driver exports the alias of the driver
that would handle the device via sysfs
. E.g., the /sys/bus/pci/devices/0000:00:0d.0/modalias
file
might contain the string “pci:v00001319d00000801sv00001319sd00001319bc04sc01i00”.
The default rules provided with udev will cause udevd to call out to
/sbin/modprobe with
the contents of the MODALIAS
uevent
environment variable (which should be the same as the contents of
the modalias
file in sysfs), thus
loading all modules whose aliases match this string after
wildcard expansion.
In this example, this means that, in addition to snd-fm801, the obsolete (and unwanted) forte driver will be loaded if it is available. See below for ways in which the loading of unwanted drivers can be prevented.
The kernel itself is also able to load modules for network protocols, filesystems, and NLS support on demand.
When you plug in a device, such as a Universal Serial Bus (USB) MP3 player, the kernel recognizes that the device is now connected and generates a uevent. This uevent is then handled by udevd as described above.
There are a few possible problems when it comes to automatically creating device nodes.
Udev will only load a module if it has a bus-specific alias and
the bus driver properly exports the necessary aliases to
sysfs
. In other cases, one should
arrange module loading by other means. With Linux-6.4.12, udev is
known to load properly-written drivers for INPUT, IDE, PCI, USB,
SCSI, SERIO, and FireWire devices.
To determine if the device driver you require has the necessary
support for udev, run modinfo with the module name as
the argument. Now try locating the device directory under
/sys/bus
and check whether there is
a modalias
file there.
If the modalias
file exists in
sysfs
, the driver supports the
device and can talk to it directly, but doesn't have the alias,
it is a bug in the driver. Load the driver without the help from
udev and expect the issue to be fixed later.
If there is no modalias
file in the
relevant directory under /sys/bus
,
this means that the kernel developers have not yet added modalias
support to this bus type. With Linux-6.4.12, this is the case
with ISA busses. Expect this issue to be fixed in later kernel
versions.
Udev is not intended to load “wrapper” drivers such as snd-pcm-oss and non-hardware drivers such as loop at all.
If the “wrapper” module only enhances the
functionality provided by some other module (e.g., snd-pcm-oss enhances the functionality
of snd-pcm by making the
sound cards available to OSS applications), configure
modprobe to load
the wrapper after udev loads the wrapped module. To do this, add
a “softdep” line to the corresponding
/etc/modprobe.d/
file. For example:
<filename>
.conf
softdep snd-pcm post: snd-pcm-oss
Note that the “softdep” command also allows
pre:
dependencies, or a mixture of
both pre:
and post:
dependencies. See the modprobe.d(5)
manual page for more information
on “softdep” syntax and capabilities.
Either don't build the module, or blacklist it in a /etc/modprobe.d/blacklist.conf
file as done
with the forte module in
the example below:
blacklist forte
Blacklisted modules can still be loaded manually with the explicit modprobe command.
This usually happens if a rule unexpectedly matches a device. For example, a poorly-written rule can match both a SCSI disk (as desired) and the corresponding SCSI generic device (incorrectly) by vendor. Find the offending rule and make it more specific, with the help of the udevadm info command.
This may be another manifestation of the previous problem. If
not, and your rule uses sysfs
attributes, it may be a kernel timing issue, to be fixed in later
kernels. For now, you can work around it by creating a rule that
waits for the used sysfs
attribute and appending it to the /etc/udev/rules.d/10-wait_for_sysfs.rules
file
(create this file if it does not exist). Please notify the LFS
Development list if you do so and it helps.
First, be certain that the driver is built into the kernel or already loaded as a module, and that udev isn't creating a misnamed device.
If a kernel driver does not export its data to sysfs
, udev lacks the information needed to
create a device node. This is most likely to happen with third
party drivers from outside the kernel tree. Create a static
device node in /usr/lib/udev/devices
with the appropriate
major/minor numbers (see the file devices.txt
inside the kernel documentation or
the documentation provided by the third party driver vendor). The
static device node will be copied to /dev
by udev.
This is due to the fact that udev, by design, handles uevents and loads modules in parallel, and thus in an unpredictable order. This will never be “fixed”. You should not rely upon the kernel device names being stable. Instead, create your own rules that make symlinks with stable names based on some stable attributes of the device, such as a serial number or the output of various *_id utilities installed by udev. See Section 9.4, “Managing Devices” and Section 9.2, “General Network Configuration” for examples.
Additional helpful documentation is available at the following sites:
A Userspace Implementation of devfs
http://www.kroah.com/linux/talks/ols_2003_udev_paper/Reprint-Kroah-Hartman-OLS2003.pdf
The sysfs
Filesystem
https://www.kernel.org/pub/linux/kernel/people/mochel/doc/papers/ols-2005/mochel.pdf