Note - this was a nice tutorial, written by Igor Medvyedyev. It was hidden away in his usbloger files on Github. I placed it here so it will be properly rendered and readable. It was written back in 2011, so it is doubtlessly out of date with respect to the current linux kernel.
On Unix and Unix-like systems, hardware devices are accessed through
special files (also called device files or nodes) located in the
/dev
directory. These files are read from and written to just
like normal files, but instead of writing and reading data on a disk, they
communicate directly with a kernel driver which then communicates with the
hardware. There are many online resources describing /dev
files in
more detail. Traditonally, these special files were created at install time
by the distribution, using the mknod
command. In recent years,
Linux systems began using udev
to manage these
/dev
files at runtime. For example, udev
will
create nodes when devices are detected and delete them when devices are removed
(including hotplug devices at runtime). This way,
the /dev
directory contains (for the most part) only entries
for devices which actually exist on the system at the current time, as
opposed to devices which could exist.
Udev also has a powerful scripting interface (with files commonly located
in /etc/udev/rules.d
) which distributors (and end users) often
use to customize the way device nodes are created. Customizable properties include file permissions,
location within
the filesystem, and symbolic links. As could be imagined, this customization
can make it difficult for application writers to locate specific device
files (or
types of devices), because they could be easily moved by modifying the
udev
rules. For example, in recent years, the js
(joystick) nodes
were moved from /dev
to /dev/input
. Many older
programs explicitly opened devices in /dev
(for example
/dev/js0
). When these older programs are run today, and
try to open /dev/js0
, they will simply not work
since /dev/js0
has been moved.
Another problem is that when using multiple devices of the same type, the
order in which they appear in /dev
is not guaranteed to be the
same every time. This often happens with USB devices. Some USB devices will
show up in a different order after a reboot even when plugged into the same
USB ports. I've
observed this directly with FTDI USB serial ports. For example, with two of
these ports plugged in, udev will create /dev/ttyUSB0
and
/dev/ttyUSB1
, but the order is undefined. (This particular
problem can be worked around by creating udev
rules which create
symlinks based on something like a device serial number).
Another issue is that when dealing with things like HID devices, simply knowing
that an entry such as /dev/hidraw0
exists tells you nothing about
what kind of device it is. It could be any type of HID device.
Sysfs is a virtual filesystem exported by the kernel, similar to
/proc
. The files in Sysfs contain information about devices and
drivers. Some files in Sysfs are even writable, for configuration and
control of devices attached to the system. Sysfs is always mounted on /sys
.
The directories in Sysfs contain the heirarchy of devices, as they are
attached to the computer. For example, on my computer, the
hidraw0
device is located under:
/sys/devices/pci0000:00/0000:00:12.2/usb1/1-5/1-5.4/1-5.4:1.0/0003:04D8:003F.0001/hidraw/hidraw0
Based on the path, the device is attached to (roughly, starting from
the end) configuration 1 (:1.0)
of the device attached to port
number 4 of device 1-5, connected to USB controller 1 (usb1), connected to
the PCI bus. While interesting, this directory path doesn't do us very much
good, since it's dependent on how the hardware is physically connected to
the computer.
Fortunately, Sysfs also provides a large number of symlinks, for easy access
to devices
without having to know which PCI and USB ports they are connected to. In
/sys/class
there is a directory for each different class of
device. My /sys/class
directory looks like this:
alan@ato:/sys/class$ ls atm graphics ieee1394_protocol printer thermal backlight hidraw input rfkill tty bdi hwmon mem scsi_device usb block i2c-adapter misc scsi_disk vc bluetooth ide_port net scsi_generic video_output dma ieee1394 pci_bus scsi_host vtconsole dmi ieee1394_host power_supply sound firmware ieee1394_node ppdev spi_master
Following our example of using hidraw
, one can see that there is a
hidraw
directory here. Inside it is a symbolic link named
hidraw0
which points to
../../devices/pci0000:00/0000:00:12.2/usb1/1-5/1-5.4/1-5.4:1.0/0003:04D8:003F.0001/hidraw/hidraw0
This way, hidraw devices can easily be found under /sys/class/hidraw
without knowing anything about their USB or PCI heirarchy. It would be
a good exercise to examine the contents of the /sys
directory,
especially /sys/bus
, /sys/class
, and
/sys/subsystem
. Since much of /sys
is symbolic
links, it may also benefit you to use the utility realpath
to show
physical directory paths, as opposed to symbolic link paths. This is useful when
trying to find the actual parent directories of device directories. For
example, to find the containing USB device entry for hidraw0
,
one could use realpath
to do something like the following:
alan@ato:/sys$ cd /sys/class/hidraw/hidraw0/ alan@ato:/sys/class/hidraw/hidraw0$ ls dev device power subsystem uevent alan@ato:/sys/class/hidraw/hidraw0$ cd `realpath $PWD` alan@ato:/sys/devices/pci0000:00/0000:00:12.2/usb2/2-5/2-5.4/2-5.4:1.0/0003:04D8:003F.0001/hidraw/hidraw0$ ls dev device power subsystem uevent alan@ato:/sys/devices/pci0000:00/0000:00:12.2/usb2/2-5/2-5.4/2-5.4:1.0/0003:04D8:003F.0001/hidraw/hidraw0$ cd ../../../../ alan@ato:/sys/devices/pci0000:00/0000:00:12.2/usb2/2-5/2-5.4$ ls 2-5.4:1.0 bDeviceSubClass configuration idProduct remove authorized bmAttributes descriptors idVendor serial avoid_reset_quirk bMaxPacketSize0 dev manufacturer speed bcdDevice bMaxPower devnum maxchild subsystem bConfigurationValue bNumConfigurations devpath power uevent bDeviceClass bNumInterfaces driver product urbnum bDeviceProtocol busnum ep_00 quirks version alan@ato:/sys/devices/pci0000:00/0000:00:12.2/usb2/2-5/2-5.4$
Because it's cumbersome and error-prone to walk the Sysfs tree from within
an application's code,
there's a convenient library called libudev
to do this task for
us. Currently, the closest thing to a manual for libudev
is the
gtk-doc-genereated API reference located here:
http://www.kernel.org/pub/linux/utils/kernel/hotplug/libudev/
The documentation there is not really enough for the average developer to get started, so hopefully this guide and its example will make it a bit easier.
For the remainder of this guide, we'll be using libudev
to
access hidraw
devices. Using libudev, we'll be able to inspect
the devices, including their Vendor ID (VID), Product ID (PID), serial
number, and device strings, without ever opening the device. Further,
libudev
will tell us exactly where inside /dev
the
device's node is located, giving the application a robust and
distribution-independent way of accessing the device.
Building with libudev is as simple as including libudev.h
and
passing -ludev
to the compiler to link with the
libudev
library.
The first example gets a list of the hidraw
objects connected
to the system, and
prints out their device node path, manufacturer strings, and serial number.
To do this, a udev_enumerate
object is created, and the text
string "hidraw"
is specified to be used as its filter.
libudev
will then return a list of udev_device
objects which match the filter. In our example, this will be a list of all
the hidraw
devices attached to the system. The example code performs
the following tasks:
struct udev
./dev/hidraw0
).#include <libudev.h> #include <stdio.h> #include <stdlib.h> #include <locale.h> #include <unistd.h> int main (void) { struct udev *udev; struct udev_enumerate *enumerate; struct udev_list_entry *devices, *dev_list_entry; struct udev_device *dev; /* Create the udev object */ udev = udev_new(); if (!udev) { printf("Can't create udev\n"); exit(1); } /* Create a list of the devices in the 'hidraw' subsystem. */ enumerate = udev_enumerate_new(udev); udev_enumerate_add_match_subsystem(enumerate, "hidraw"); udev_enumerate_scan_devices(enumerate); devices = udev_enumerate_get_list_entry(enumerate); /* For each item enumerated, print out its information. udev_list_entry_foreach is a macro which expands to a loop. The loop will be executed for each member in devices, setting dev_list_entry to a list entry which contains the device's path in /sys. */ udev_list_entry_foreach(dev_list_entry, devices) { const char *path; /* Get the filename of the /sys entry for the device and create a udev_device object (dev) representing it */ path = udev_list_entry_get_name(dev_list_entry); dev = udev_device_new_from_syspath(udev, path); /* usb_device_get_devnode() returns the path to the device node itself in /dev. */ printf("Device Node Path: %s\n", udev_device_get_devnode(dev)); /* The device pointed to by dev contains information about the hidraw device. In order to get information about the USB device, get the parent device with the subsystem/devtype pair of "usb"/"usb_device". This will be several levels up the tree, but the function will find it.*/ dev = udev_device_get_parent_with_subsystem_devtype( dev, "usb", "usb_device"); if (!dev) { printf("Unable to find parent usb device."); exit(1); } /* From here, we can call get_sysattr_value() for each file in the device's /sys entry. The strings passed into these functions (idProduct, idVendor, serial, etc.) correspond directly to the files in the directory which represents the USB device. Note that USB strings are Unicode, UCS2 encoded, but the strings returned from udev_device_get_sysattr_value() are UTF-8 encoded. */ printf(" VID/PID: %s %s\n", udev_device_get_sysattr_value(dev,"idVendor"), udev_device_get_sysattr_value(dev, "idProduct")); printf(" %s\n %s\n", udev_device_get_sysattr_value(dev,"manufacturer"), udev_device_get_sysattr_value(dev,"product")); printf(" serial: %s\n", udev_device_get_sysattr_value(dev, "serial")); udev_device_unref(dev); } /* Free the enumerator object */ udev_enumerate_unref(enumerate); udev_unref(udev); return 0; }
libudev
programs can be compiled using the following:
gcc -Wall -g -o udev_example udev_example.c -ludev
On my system, I have a Microchip Application Demo connected, so my output is the following (notice how the non-ASCII, Unicode character from the USB serial number is propagated through to userspace as UTF-8):
alan@alan-desktop:~/tmp$ ./test_udev Device Node Path: /dev/hidraw0 VID/PID: 04d8 003f Microchip Technology Inc. Simple HID Device Demo serial: 1234Å
Some Notes on libudev
Before moving on, it seems appropriate to mention some important things about libudev.
libudev
's functions are string-based. Since the data is
coming directly from sysfs (which contains (virtual) files with text in
them), all the data which comes from libudev
is in text string
format.
This means that the user will have to manually convert strings to integer types
if desired. udev_device_get_sysattr_value()
correspond to file names in the sysfs tree. In this example,
idVendor
corresponds to
/sys/devices/pci0000:00/0000:00:12.2/usb1/1-5/1-5.4/idVendor
which can also be reached as (using the sysfs symlinks)
/sys/bus/usb/devices/1-5.4/idVendor
In order to find out which properties are available for a device, simply
find the device in sysfs, and look at which files exist. On my system, my
/sys/bus/usb/devices/1-5.4/
directory looks like the following:
Any non-directory file or link in that directory can be queried with1-5.4:1.0 bDeviceSubClass configuration idProduct remove authorized bmAttributes descriptors idVendor serial avoid_reset_quirk bMaxPacketSize0 dev manufacturer speed bcdDevice bMaxPower devnum maxchild subsystem bConfigurationValue bNumConfigurations devpath power uevent bDeviceClass bNumInterfaces driver product urbnum bDeviceProtocol busnum ep_00 quirks version
udev_device_get_sysattr_value()
to determine the properties of
the device.
libudev
is reference counted. Instead of specifically
allocating and freeing objects, ref() and unref() functions (such as
udev_ref()
and udev_unref()
) are used for keeping track of how many
references to an object exist. When the reference count drops to zero, the
object is freed. Functions which return a new object return it with a
reference count of 1, so calling its unref() function will effectively free
it. See the libudev
documentation (referenced above). libudev
- Monitoring Interface
libudev
also provides a monitoring interface. The monitoring
interface will report events to the application when the status of a device
changes. This is useful for receiving notification when devices are connected
or disconnected from the system. Like the enumeration interface described above,
the monitoring interface also provides a filtering mechanisn, so that an
application can subscribe to only events with which it is concerned. For
example, if an application added "hidraw"
to the filter,
only events concerning hidraw
devices would be delivered
to the application. When a device changes state, the
udev_monitor_receive_device()
function will return a handle to
a udev_device
which represents the object which changed. The returned object
can then be queried with the udev_device_get_action()
function
to determine which action occurred. The actions are returned as the
following strings:
add
- Device is connected to the system
remove
- Device is disconnected from the system
change
- Something about the device changed
move
- Device node was moved, renamed, or re-parented
The udev_monitor_receive_device()
function is blocking. That is,
when called, program execution will stop until there is an event to be
returned. This use case does not seem to be very useful. Fortunately, the
udev_monitor
object can provide a file descriptor, suitable for
use with the select()
system call. select()
can be
used to determine if a call to udev_monitor_receive_device()
will block, providing a way to receive events in a non-blocking way.
The following code shows an example of the libudev
monitor
interface. The example runs a loop which executes select()
to
determine if there has been an event. If there has, it calls
udev_monitor_receive_device()
to receive the event and prints
it out. At the end of the loop it
sleep()
's for 250 milliseconds. In real life, a simple program
like this would be just fine to not use select()
and just let
udev_monitor_receive_device()
block, but it is written this way
to show an example of how to get non-blocking behavior from the
libudev
monitoring interface.
/* Set up a monitor to monitor hidraw devices */ mon = udev_monitor_new_from_netlink(udev, "udev"); udev_monitor_filter_add_match_subsystem_devtype(mon, "hidraw", NULL); udev_monitor_enable_receiving(mon); /* Get the file descriptor (fd) for the monitor. This fd will get passed to select() */ fd = udev_monitor_get_fd(mon); /* This section will run continuously, calling usleep() at the end of each pass. This is to demonstrate how to use a udev_monitor in a non-blocking way. */ while (1) { /* Set up the call to select(). In this case, select() will only operate on a single file descriptor, the one associated with our udev_monitor. Note that the timeval object is set to 0, which will cause select() to not block. */ fd_set fds; struct timeval tv; int ret; FD_ZERO(&fds); FD_SET(fd, &fds); tv.tv_sec = 0; tv.tv_usec = 0; ret = select(fd+1, &fds, NULL, NULL, &tv); /* Check if our file descriptor has received data. */ if (ret > 0 && FD_ISSET(fd, &fds)) { printf("\nselect() says there should be data\n"); /* Make the call to receive the device. select() ensured that this will not block. */ dev = udev_monitor_receive_device(mon); if (dev) { printf("Got Device\n"); printf(" Node: %s\n", udev_device_get_devnode(dev)); printf(" Subsystem: %s\n", udev_device_get_subsystem(dev)); printf(" Devtype: %s\n", udev_device_get_devtype(dev)); printf(" Action: %s\n",udev_device_get_action(dev)); udev_device_unref(dev); } else { printf("No Device from receive_device(). An error occured.\n"); } } usleep(250*1000); printf("."); fflush(stdout); }
It's important to note that when using monitoring and enumeration together, that monitoring should be enabled before enumeration. This way, any events (for example devices being attached to the system) which happen during enumeration will not be lost. If enumeration is done before monitoring is enabled, any device attached between the time the enumeration happens and when monitoring starts will be missed. The algorithm should be:
The example file (linked at the end of this document) uses enumeration and monitoring together, and follows this algorithm. The code above shows only monitoring, for simplicity.
Output
The code above will run forever. (Terminate it with Ctrl-C
). With
the above section of code running, the following data is printed out as I
disconnect and reconnect my HID device (note that a . character is printed
every 250 milliseconds):
........................... select() says there should be data Got Device Node: /dev/hidraw0 Subsystem: hidraw Devtype: (null) Action: remove ............. select() says there should be data Got Device Node: /dev/hidraw0 Subsystem: hidraw Devtype: (null) Action: add ......^C
The libudev interface is very useful for creating robust software which needs to access a specific hardware device or monitor the real-time connection and disconnection status of hot-pluggable hardware. I hope you find this document useful. The full source code of the demo is available through the following link:
udev_example.c
Alan Ott
[email protected]
Signal 11 Software
2010-05-23