/[gentoo]/xml/htdocs/doc/en/udev-guide.xml
Gentoo

Contents of /xml/htdocs/doc/en/udev-guide.xml

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1.34 - (hide annotations) (download) (as text)
Wed Nov 29 15:48:57 2006 UTC (11 years ago) by nightmorph
Branch: MAIN
Changes since 1.33: +3 -3 lines
File MIME type: application/xml
revert previous change

1 swift 1.1 <?xml version='1.0' encoding="UTF-8"?>
2     <!DOCTYPE guide SYSTEM "/dtd/guide.dtd">
3 neysx 1.4
4 nightmorph 1.33 <!-- $Header: /var/cvsroot/gentoo/xml/htdocs/doc/en/udev-guide.xml,v 1.32 2006/11/27 00:08:22 nightmorph Exp $ -->
5 neysx 1.4
6 swift 1.1 <guide link="/doc/en/udev-guide.xml">
7     <title>Gentoo udev Guide</title>
8    
9     <author title="Author">
10 nightmorph 1.34 <mail link="swift@gentoo.org">Sven Vermeulen</mail>
11 swift 1.8 </author>
12     <author title="Contributor">
13 neysx 1.24 <mail link="greg_g@gentoo.org">Gregorio Guidi</mail>
14 swift 1.1 </author>
15    
16     <abstract>
17     This document explains what udev is and how you can use udev to fit your needs.
18     </abstract>
19    
20     <license/>
21    
22 nightmorph 1.34 <version>0.25</version>
23     <date>2006-11-26</date>
24 swift 1.1
25     <chapter>
26     <title>What is udev?</title>
27     <section>
28     <title>The /dev Directory</title>
29     <body>
30    
31     <p>
32     When Linux-users talk about the hardware on their system in the vicinity of
33 swift 1.3 people who believe Linux is some sort of virus or brand of coffee, the use of
34 swift 1.1 "slash dev slash foo" will return a strange look for sure. But for the fortunate
35     user (and that includes you) using <path>/dev/hda1</path> is just a fast way of
36     explaining that we are talking about the primary master IDE, first partition. Or
37     aren't we?
38     </p>
39    
40     <p>
41     We all know what a device file is. Some even know why device files have special
42     numbers when we take a closer look at them when we issue <c>ls -l</c> in
43     <path>/dev</path>. But what we always take for granted is that the primary
44     master IDE disk is referred to as <path>/dev/hda</path>. You might not see it
45     this way, but this is a flaw by design.
46     </p>
47    
48     <p>
49 swift 1.3 Think about hotpluggable devices like USB, IEEE1394, hot-swappable PCI, ... What
50     is the first device? And for how long? What will the other devices be named when
51 swift 1.1 the first one disappears? How will that affect ongoing transactions? Wouldn't it
52 swift 1.3 be fun that a printing job is suddenly moved from your supernew laserprinter to
53 swift 1.1 your almost-dead matrix printer because your mom decided to pull the plug of the
54 swift 1.20 laserprinter which happened to be the first printer?
55 swift 1.1 </p>
56    
57     <p>
58     Enter <e>udev</e>. The goals of the udev project are both interesting and
59     needed:
60     </p>
61    
62     <ul>
63 swift 1.3 <li>Runs in userspace</li>
64     <li>Dynamically creates/removes device files</li>
65     <li>Provides consistent naming</li>
66     <li>Provides a user-space API</li>
67 swift 1.1 </ul>
68    
69     <p>
70     To provide these features, udev is developed in three separate projects:
71     <e>namedev</e>, <e>libsysfs</e> and, of course, <e>udev</e>.
72     </p>
73    
74     </body>
75     </section>
76     <section>
77     <title>namedev</title>
78     <body>
79    
80     <p>
81     Namedev allows you to define the device naming separately from the udev program.
82     This allows for flexible naming policies and naming schemes developed by
83     separate entities. This device naming subsystem provides a standard interface
84     that udev can use.
85     </p>
86    
87     <p>
88     Currently only a single naming scheme is provided by namedev; the one provided
89 swift 1.3 by LANANA, used by the majority of Linux systems currently and therefore very
90 swift 1.1 suitable for the majority of Linux users.
91     </p>
92    
93     <p>
94     Namedev uses a 5-step procedure to find out the name of a given device. If the
95     device name is found in one of the given steps, that name is used. The steps
96     are:
97     </p>
98    
99     <ul>
100     <li>label or serial number</li>
101     <li>bus device number</li>
102     <li>bus topology</li>
103     <li>statically given name</li>
104     <li>kernel provided name</li>
105     </ul>
106    
107     <p>
108     The <e>label or serial number</e> step checks if the device has a unique
109     identifier. For instance USB devices have a unique USB serial number; SCSI
110     devices have a unique UUID. If namedev finds a match between this unique number
111     and a given configuration file, the name provided in the configuration file is
112     used.
113     </p>
114    
115     <p>
116     The <e>bus device number</e> step checks the device bus number. For
117     non-hot-swappable environments this procedure is sufficient to
118     identify a hardware device. For instance PCI bus numbers rarely change in the
119     lifetime of a system. Again, if namedev finds a match between this position and
120     a given configuration file, the name provided in that configuration file is
121     used.
122     </p>
123    
124     <p>
125     Likewise the <e>bus topology</e> is a rather static way of defining devices as
126     long as the user doesn't switch devices. When the position of the device matches
127 swift 1.3 a given setting provided by the user, the accompanying name is used.
128 swift 1.1 </p>
129    
130     <p>
131     The fourth step, <e>statically given name</e>, is a simple string replacement.
132     When the kernel name (the default name) matches a given replacement string, the
133     substitute name will be used.
134     </p>
135    
136     <p>
137     The final step (<e>kernel provided name</e>) is a catch-all: this one takes
138     the default name provided by the kernel. In the majority of cases this is
139     sufficient as it matches the device naming used on current Linux systems.
140     </p>
141    
142     </body>
143     </section>
144     <section>
145     <title>libsysfs</title>
146     <body>
147    
148     <p>
149     udev interacts with the kernel through the sysfs pseudo filesystem. The libsysfs
150     project provides a common API to access the information given by the sysfs
151     filesystem in a generic way. This allows for querying all kinds of hardware
152     without having to make assumptions on the kind of hardware.
153     </p>
154    
155     </body>
156     </section>
157     <section>
158     <title>udev</title>
159     <body>
160    
161     <p>
162     Every time the kernel notices an update in the device structure, it calls the
163     <path>/sbin/hotplug</path> program. Hotplug runs the applications linked in the
164     <path>/etc/hotplug.d/default</path> directory where you will also find a symlink
165     to the udev application. Hotplug directs the information given by the kernel to
166     the udev application which performs the necessary actions on the
167     <path>/dev</path> structure (creating or deleting device files).
168     </p>
169    
170     </body>
171     </section>
172     </chapter>
173    
174     <chapter>
175     <title>Using udev on Gentoo</title>
176     <section>
177     <title>Requirements</title>
178     <body>
179    
180     <p>
181 swift 1.8 udev is meant to be used in combination with a 2.6 kernel (like
182 swift 1.25 <c>vanilla-sources</c> or <c>gentoo-sources</c> with the default 2005.0
183     profile). If you're using such a kernel then you just have to make sure that
184     you have a recent <c>sys-apps/baselayout</c> version. That's all you need.
185 swift 1.1 </p>
186    
187 swift 1.8 <pre caption="Installing udev">
188     # <i>emerge udev</i>
189 swift 1.1 </pre>
190    
191     <p>
192 nightmorph 1.30 udev will install <c>hotplug-base</c> as one of its dependencies. You do not
193     need to install <c>hotplug</c> unless you want your modules automatically
194     loaded when you plug devices in. <c>hotplug</c> also handles the automated
195     bringup of network devices and firmware downloading.
196 swift 1.1 </p>
197    
198 swift 1.8 <pre caption="Installing optional hotplug scripts">
199     # <i>emerge hotplug</i>
200 swift 1.1 </pre>
201    
202     <p>
203 swift 1.19 If you want modules loaded for devices that have been plugged in before you
204     boot, use the coldplug package:
205     </p>
206    
207     <pre caption="Installing the coldplug package">
208     # <i>emerge coldplug</i>
209     </pre>
210    
211     <p>
212 swift 1.26 Don't forget to add <c>coldplug</c> to the boot runlevel:
213     </p>
214    
215     <pre caption="Adding coldplug to the boot runlevel">
216     # <i>rc-update add coldplug boot</i>
217     </pre>
218    
219 nightmorph 1.32 <note>
220     You no longer need to emerge <c>coldplug</c> if you're using udev version 103
221     and higher, as udev itself will handle module loading. If you are upgrading to
222     version 103, you must unmerge coldplug (<c>emerge -aC coldplug</c>) before
223     upgrading udev. Be sure to remove the coldplug init script from all runlevels
224     (<c>rc-update del coldplug</c>), and then delete it from
225     <path>/etc/init.d/</path> afterward.
226     </note>
227    
228 swift 1.26 <p>
229 swift 1.21 Kernelwise, be sure to activate the following options:
230 swift 1.1 </p>
231    
232     <pre caption="Required kernel options">
233 swift 1.6 General setup ---&gt;
234 swift 1.1 [*] Support for hot-pluggable devices
235    
236     File systems ---&gt;
237     Pseudo filesystems ---&gt;
238     [*] /proc file system support
239     [*] Virtual memory file system support (former shm fs)
240     </pre>
241    
242     <p>
243     You can leave the <c>/dev file system support (OBSOLETE)</c> active if you
244 swift 1.16 wish but you have to make sure that "Automatically mount at boot" is disabled:
245 swift 1.1 </p>
246    
247 swift 1.16 <pre caption="Don't automatically mount devfsd">
248     File systems ---&gt;
249     Pseudo Filesystems ---&gt;
250     [*] /dev file system support (OBSOLETE)
251     [ ] Automatically mount at boot
252     </pre>
253    
254 swift 1.21 <p>
255     If you use <c>genkernel</c>, don't forget to run it with the <c>--udev</c>
256     option to enable all the required kernel configuration directives. The default
257     configuration given by this <c>genkernel</c> invocation is sufficient.
258     </p>
259    
260 swift 1.1 </body>
261     </section>
262     <section>
263     <title>Configuration</title>
264     <body>
265    
266     <p>
267 neysx 1.5 If you want to use the udev-tweaks Gentoo added to make your life
268 swift 1.15 comfortable, then read no more. Gentoo will use udev but keep a static
269     <path>/dev</path> so that you will never have any missing device nodes.
270     The Gentoo init scripts won't run the devfsd daemon and will deactivate devfs
271     when you boot up.
272 swift 1.1 </p>
273    
274     <p>
275     But if you are a die-hard and want to run a udev-only, no-tweaked system as is
276 swift 1.2 intended by the udev development (including the difficulties of missing device
277     nodes because udev doesn't support them yet), by all means, read on :)
278 swift 1.1 </p>
279    
280     <p>
281 bennyc 1.13 We'll deactivate the rules that save the device file nodes: edit the
282 swift 1.2 <c>RC_DEVICE_TARBALL</c> variable in <path>/etc/conf.d/rc</path> and set it to
283     <c>no</c>:
284     </p>
285 swift 1.1
286 swift 1.2 <pre caption="/etc/conf.d/rc">
287     RC_DEVICE_TARBALL="no"
288 swift 1.1 </pre>
289    
290     <p>
291 swift 1.8 If you have included devfs support in your kernel, you can deactivate it in
292 swift 1.14 the bootloader configuration: add <c>gentoo=nodevfs</c> as a kernel parameter.
293     If you want to use devfs and deactivate udev, add <c>gentoo=noudev</c> as kernel
294 swift 1.8 parameter.
295     </p>
296    
297     </body>
298     </section>
299     </chapter>
300    
301     <chapter>
302     <title>Known Issues</title>
303     <section>
304     <title>Missing device node files at boot</title>
305     <body>
306    
307     <p>
308 neysx 1.9 If you can't boot successfully because you get an error about
309 swift 1.8 <path>/dev/null</path> not found, or because the initial console is missing, the
310     problem is that you lack some device files that must be available <e>before</e>
311     <path>/dev</path> is mounted and handled by udev. This is common on Gentoo
312     machines installed from old media.
313     </p>
314    
315     <p>
316     If you run <c>sys-apps/baselayout-1.8.12</c> or later, this problem is
317     alleviated since the boot process should still manage to complete. However, to
318 swift 1.10 get rid of those annoying warnings, you should create the missing device nodes
319 swift 1.8 as described below.
320     </p>
321    
322     <p>
323     To see which devices nodes are present before the <path>/dev</path> filesystem
324     is mounted, run the following commands:
325     </p>
326    
327     <pre caption="Listing device nodes available at boot">
328     # <i>mkdir test</i>
329     # <i>mount --bind / test</i>
330     # <i>cd test/dev</i>
331     # <i>ls</i>
332     </pre>
333    
334     <p>
335 neysx 1.9 The devices needed for a successful boot are <path>/dev/null</path> and
336 swift 1.8 <path>/dev/console</path>. If they didn't show up in the previous test, you have
337 cam 1.12 to create them manually. Issue the following commands in the
338     <path>test/dev/</path> directory:
339 swift 1.1 </p>
340    
341     <pre caption="Creating necessary device node files">
342 cam 1.11 # <i>mknod -m 660 console c 5 1</i>
343     # <i>mknod -m 660 null c 1 3</i>
344 swift 1.1 </pre>
345    
346     <p>
347 swift 1.8 When you're finished, don't forget to unmount the <path>test/</path> directory:
348 swift 1.1 </p>
349    
350 swift 1.8 <pre caption="Unmounting the test/ directory">
351 cam 1.11 # <i>cd ../..</i>
352 swift 1.8 # <i>umount test</i>
353 cam 1.11 # <i>rmdir test</i>
354 swift 1.8 </pre>
355    
356     </body>
357     </section>
358     <section>
359     <title>udev and nvidia</title>
360     <body>
361    
362 swift 1.1 <p>
363 swift 1.8 If you use the proprietary driver from nVidia and the X server fails to start on
364     a udev-only system, then make sure you have:
365 swift 1.1 </p>
366    
367 swift 1.8 <ul>
368     <li>
369     the <c>nvidia</c> module listed in
370     <path>/etc/modules.autoload.d/kernel-2.6</path>
371     </li>
372     <li>
373 swift 1.10 a version of <c>nvidia-kernel</c> equal to or greater than
374 swift 1.8 <c>media-video/nvidia-kernel-1.0.5336-r2</c>
375     </li>
376     <li>
377     a version of baselayout equal to or greater than
378     <c>sys-apps/baselayout-1.8.12</c>
379     </li>
380     </ul>
381    
382 swift 1.28 <p>
383     If <c>xorg-x11</c> refuses to start, it might be because the
384     <path>/dev/nvidia</path> device file is missing. If that is the case, run
385     <path>/sbin/NVmakedevices.sh</path> to (re)create it.
386     </p>
387    
388 swift 1.1 </body>
389     </section>
390     <section>
391 swift 1.17 <title>LVM2 Names Disappear</title>
392     <body>
393    
394     <p>
395     When you use <c>udev</c> and LVM2 together, you might notice that your created
396     volume groups and logical volumes have disappeared. Well, they haven't, but they
397     are unfortunately named <path>/dev/dm-#</path> with # being 0, 1, ...
398     </p>
399    
400     <p>
401     To fix this, edit <path>/etc/udev/rules.d/50-udev.rules</path> and uncomment the
402     following line:
403     </p>
404    
405     <pre caption="Uncomment this line from /etc/udev/rules.d/50-udev.rules">
406     KERNEL="dm-[0-9]*", PROGRAM="/sbin/devmap_name %M %m", NAME="%k", SYMLINK="%c"
407     </pre>
408    
409 swift 1.27 <p>
410     Next, install <c>sys-fs/multipath-tools</c> which contains the
411     <c>devmap_name</c> application.
412     </p>
413    
414     <pre caption="Installing multipath-tools">
415     <comment>(At the moment of writing, multipath-tools is only available in the testing branch:)</comment>
416     # <i>echo "=sys-fs/multipath-tools-0.4.2 ~x86" &gt;&gt; /etc/portage/package.keywords</i>
417     # <i>emerge multipath-tools</i>
418     </pre>
419    
420 swift 1.17 </body>
421     </section>
422     <section>
423 swift 1.18 <title>No Consistent Naming between DevFS and udev</title>
424     <body>
425    
426     <p>
427     Even though our intention is to have a consistent naming scheme between both
428     dynamical device management solutions, sometimes naming differences do occur.
429 swift 1.22 </p>
430    
431     <p>
432 swift 1.18 One reported clash is with a HP Smart Array 5i RAID controller (more precisely
433     the <c>cciss</c> kernel module). With udev, the devices are named
434     <path>/dev/cciss/cXdYpZ</path> with X, Y and Z regular numbers. With devfs, the
435     devices are <path>/dev/hostX/targetY/partZ</path> or symlinked from
436     <path>/dev/cciss/cXdY</path>.
437     </p>
438    
439     <p>
440     If this is the case, don't forget to update your <path>/etc/fstab</path> and
441     bootloader configuration files accordingly.
442     </p>
443    
444 swift 1.22 <p>
445     The same happens with all-round symlinks that used to exist in
446     <path>/dev</path>, such as <path>/dev/mouse</path>, which <c>udev</c> doesn't
447     create anymore. Be certain to check your X configuration file and see if the
448 swift 1.23 Device rule for your mouse points to an existing device file.
449 swift 1.22 </p>
450    
451 fox2mike 1.29 <p>
452     Another issue is the difference in naming of terminals between devfs and udev.
453 nightmorph 1.31 While devfs calls its terminals <c>tty</c>, udev calls them <c>vc</c> and
454     <c>tty</c>. This could lead to a problem in case you are restricting root
455     logins from consoles using <path>/etc/securetty</path>. You will need to make
456     sure that both <c>tty1</c> and <c>vc/1</c> are listed in
457     <path>/etc/securetty</path> to ensure that root can login using the console.
458 fox2mike 1.29 </p>
459    
460 swift 1.18 </body>
461     </section>
462     <section>
463 swift 1.8 <title>Other issues</title>
464 swift 1.1 <body>
465    
466 swift 1.8 <p>
467     If device nodes are not created when a module is loaded from
468     <path>/etc/modules.autoload.d/kernel-2.6</path> but they appear when you load
469     the module manually with modprobe then you should try upgrading to
470     <c>sys-apps/baselayout-1.8.12</c> or later.
471     </p>
472 swift 1.7
473     <p>
474 swift 1.8 Support for the framebuffer devices (<path>/dev/fb/*</path>) comes with the
475     kernel starting from version 2.6.6-rc2.
476 swift 1.7 </p>
477    
478     <p>
479 swift 1.8 For kernels older than 2.6.4 you have to explicitly include support for the
480     <path>/dev/pts</path> filesystem.
481 swift 1.1 </p>
482    
483 swift 1.8 <pre caption="Enabling the /dev/pts filesystem">
484     File systems ---&gt;
485     Pseudo filesystems ---&gt;
486     [*] /dev/pts file system for Unix98 PTYs
487     </pre>
488    
489 swift 1.1 </body>
490     </section>
491     </chapter>
492    
493     <chapter>
494     <title>Resources &amp; Acknowledgements</title>
495     <section>
496     <body>
497    
498     <p>
499     The udev talk on the Linux Symposium (Ottawa, Ontario Canada - 2003) given by
500     Greg Kroah-Hartman (IBM Corporation) provided a solid understanding on the udev
501     application.
502     </p>
503    
504     <p>
505     <uri link="http://webpages.charter.net/decibelshelp/LinuxHelp_UDEVPrimer.html">Decibel's
506     UDEV Primer</uri> is an in-depth document about udev and Gentoo.
507     </p>
508    
509 swift 1.8 <p>
510     <uri link="http://www.reactivated.net/udevrules.php">Writing udev rules</uri> by
511     fellow Gentoo developer Daniel Drake is an excellent document to learn how to
512     customize your udev installation.
513     </p>
514    
515 swift 1.1 </body>
516     </section>
517     </chapter>
518    
519     </guide>

  ViewVC Help
Powered by ViewVC 1.1.20