Commit | Line | Data |
---|---|---|
46336009 KS |
1 | Rules on how to access information in the Linux kernel sysfs |
2 | ||
3 | The kernel exported sysfs exports internal kernel implementation-details | |
4 | and depends on internal kernel structures and layout. It is agreed upon | |
5 | by the kernel developers that the Linux kernel does not provide a stable | |
6 | internal API. As sysfs is a direct export of kernel internal | |
7 | structures, the sysfs interface can not provide a stable interface eighter, | |
8 | it may always change along with internal kernel changes. | |
9 | ||
10 | To minimize the risk of breaking users of sysfs, which are in most cases | |
11 | low-level userspace applications, with a new kernel release, the users | |
12 | of sysfs must follow some rules to use an as abstract-as-possible way to | |
13 | access this filesystem. The current udev and HAL programs already | |
14 | implement this and users are encouraged to plug, if possible, into the | |
15 | abstractions these programs provide instead of accessing sysfs | |
16 | directly. | |
17 | ||
18 | But if you really do want or need to access sysfs directly, please follow | |
19 | the following rules and then your programs should work with future | |
20 | versions of the sysfs interface. | |
21 | ||
22 | - Do not use libsysfs | |
23 | It makes assumptions about sysfs which are not true. Its API does not | |
24 | offer any abstraction, it exposes all the kernel driver-core | |
25 | implementation details in its own API. Therefore it is not better than | |
26 | reading directories and opening the files yourself. | |
27 | Also, it is not actively maintained, in the sense of reflecting the | |
28 | current kernel-development. The goal of providing a stable interface | |
29 | to sysfs has failed, it causes more problems, than it solves. It | |
30 | violates many of the rules in this document. | |
31 | ||
32 | - sysfs is always at /sys | |
33 | Parsing /proc/mounts is a waste of time. Other mount points are a | |
34 | system configuration bug you should not try to solve. For test cases, | |
35 | possibly support a SYSFS_PATH environment variable to overwrite the | |
36 | applications behavior, but never try to search for sysfs. Never try | |
37 | to mount it, if you are not an early boot script. | |
38 | ||
39 | - devices are only "devices" | |
40 | There is no such thing like class-, bus-, physical devices, | |
41 | interfaces, and such that you can rely on in userspace. Everything is | |
42 | just simply a "device". Class-, bus-, physical, ... types are just | |
43 | kernel implementation details, which should not be expected by | |
44 | applications that look for devices in sysfs. | |
45 | ||
46 | The properties of a device are: | |
47 | o devpath (/devices/pci0000:00/0000:00:1d.1/usb2/2-2/2-2:1.0) | |
48 | - identical to the DEVPATH value in the event sent from the kernel | |
49 | at device creation and removal | |
50 | - the unique key to the device at that point in time | |
51 | - the kernels path to the device-directory without the leading | |
52 | /sys, and always starting with with a slash | |
53 | - all elements of a devpath must be real directories. Symlinks | |
54 | pointing to /sys/devices must always be resolved to their real | |
55 | target, and the target path must be used to access the device. | |
56 | That way the devpath to the device matches the devpath of the | |
57 | kernel used at event time. | |
58 | - using or exposing symlink values as elements in a devpath string | |
59 | is a bug in the application | |
60 | ||
61 | o kernel name (sda, tty, 0000:00:1f.2, ...) | |
62 | - a directory name, identical to the last element of the devpath | |
63 | - applications need to handle spaces and characters like '!' in | |
64 | the name | |
65 | ||
66 | o subsystem (block, tty, pci, ...) | |
67 | - simple string, never a path or a link | |
68 | - retrieved by reading the "subsystem"-link and using only the | |
69 | last element of the target path | |
70 | ||
71 | o driver (tg3, ata_piix, uhci_hcd) | |
72 | - a simple string, which may contain spaces, never a path or a | |
73 | link | |
74 | - it is retrieved by reading the "driver"-link and using only the | |
75 | last element of the target path | |
76 | - devices which do not have "driver"-link, just do not have a | |
77 | driver; copying the driver value in a child device context, is a | |
78 | bug in the application | |
79 | ||
80 | o attributes | |
81 | - the files in the device directory or files below a subdirectories | |
82 | of the same device directory | |
83 | - accessing attributes reached by a symlink pointing to another device, | |
84 | like the "device"-link, is a bug in the application | |
85 | ||
86 | Everything else is just a kernel driver-core implementation detail, | |
87 | that should not be assumed to be stable across kernel releases. | |
88 | ||
89 | - Properties of parent devices never belong into a child device. | |
90 | Always look at the parent devices themselves for determining device | |
91 | context properties. If the device 'eth0' or 'sda' does not have a | |
92 | "driver"-link, then this device does not have a driver. Its value is empty. | |
93 | Never copy any property of the parent-device into a child-device. Parent | |
94 | device-properties may change dynamically without any notice to the | |
95 | child device. | |
96 | ||
97 | - Hierarchy in a single device-tree | |
98 | There is only one valid place in sysfs where hierarchy can be examined | |
99 | and this is below: /sys/devices. | |
100 | It is planned, that all device directories will end up in the tree | |
101 | below this directory. | |
102 | ||
103 | - Classification by subsystem | |
104 | There are currently three places for classification of devices: | |
105 | /sys/block, /sys/class and /sys/bus. It is planned that these will | |
106 | not contain any device-directories themselves, but only flat lists of | |
107 | symlinks pointing to the unified /sys/devices tree. | |
108 | All three places have completely different rules on how to access | |
109 | device information. It is planned to merge all three | |
110 | classification-directories into one place at /sys/subsystem, | |
111 | following the layout of the bus-directories. All buses and | |
112 | classes, including the converted block-subsystem, will show up | |
113 | there. | |
114 | The devices belonging to a subsystem will create a symlink in the | |
115 | "devices" directory at /sys/subsystem/<name>/devices. | |
116 | ||
117 | If /sys/subsystem exists, /sys/bus, /sys/class and /sys/block can be | |
118 | ignored. If it does not exist, you have always to scan all three | |
119 | places, as the kernel is free to move a subsystem from one place to | |
120 | the other, as long as the devices are still reachable by the same | |
121 | subsystem name. | |
122 | ||
123 | Assuming /sys/class/<subsystem> and /sys/bus/<subsystem>, or | |
124 | /sys/block and /sys/class/block are not interchangeable, is a bug in | |
125 | the application. | |
126 | ||
127 | - Block | |
128 | The converted block-subsystem at /sys/class/block, or | |
129 | /sys/subsystem/block will contain the links for disks and partitions | |
130 | at the same level, never in a hierarchy. Assuming the block-subsytem to | |
131 | contain only disks and not partition-devices in the same flat list is | |
132 | a bug in the application. | |
133 | ||
134 | - "device"-link and <subsystem>:<kernel name>-links | |
135 | Never depend on the "device"-link. The "device"-link is a workaround | |
136 | for the old layout, where class-devices are not created in | |
137 | /sys/devices/ like the bus-devices. If the link-resolving of a | |
138 | device-directory does not end in /sys/devices/, you can use the | |
139 | "device"-link to find the parent devices in /sys/devices/. That is the | |
140 | single valid use of the "device"-link, it must never appear in any | |
141 | path as an element. Assuming the existence of the "device"-link for | |
142 | a device in /sys/devices/ is a bug in the application. | |
143 | Accessing /sys/class/net/eth0/device is a bug in the application. | |
144 | ||
145 | Never depend on the class-specific links back to the /sys/class | |
146 | directory. These links are also a workaround for the design mistake | |
147 | that class-devices are not created in /sys/devices. If a device | |
148 | directory does not contain directories for child devices, these links | |
149 | may be used to find the child devices in /sys/class. That is the single | |
150 | valid use of these links, they must never appear in any path as an | |
151 | element. Assuming the existence of these links for devices which are | |
152 | real child device directories in the /sys/devices tree, is a bug in | |
153 | the application. | |
154 | ||
155 | It is planned to remove all these links when when all class-device | |
156 | directories live in /sys/devices. | |
157 | ||
158 | - Position of devices along device chain can change. | |
159 | Never depend on a specific parent device position in the devpath, | |
160 | or the chain of parent devices. The kernel is free to insert devices into | |
161 | the chain. You must always request the parent device you are looking for | |
162 | by its subsystem value. You need to walk up the chain until you find | |
163 | the device that matches the expected subsystem. Depending on a specific | |
164 | position of a parent device, or exposing relative paths, using "../" to | |
165 | access the chain of parents, is a bug in the application. | |
166 |