| blob | 3828e85345ae3f8a41e56adaddc442e648c9be7b |
2 configfs - Userspace-driven kernel object configuration.
4 Joel Becker <joel.becker@oracle.com>
6 Updated: 31 March 2005
8 Copyright (c) 2005 Oracle Corporation,
9 Joel Becker <joel.becker@oracle.com>
12 [What is configfs?]
14 configfs is a ram-based filesystem that provides the converse of
15 sysfs's functionality. Where sysfs is a filesystem-based view of
16 kernel objects, configfs is a filesystem-based manager of kernel
17 objects, or config_items.
19 With sysfs, an object is created in kernel (for example, when a device
20 is discovered) and it is registered with sysfs. Its attributes then
21 appear in sysfs, allowing userspace to read the attributes via
22 readdir(3)/read(2). It may allow some attributes to be modified via
23 write(2). The important point is that the object is created and
24 destroyed in kernel, the kernel controls the lifecycle of the sysfs
25 representation, and sysfs is merely a window on all this.
27 A configfs config_item is created via an explicit userspace operation:
28 mkdir(2). It is destroyed via rmdir(2). The attributes appear at
29 mkdir(2) time, and can be read or modified via read(2) and write(2).
30 As with sysfs, readdir(3) queries the list of items and/or attributes.
31 symlink(2) can be used to group items together. Unlike sysfs, the
32 lifetime of the representation is completely driven by userspace. The
33 kernel modules backing the items must respond to this.
35 Both sysfs and configfs can and should exist together on the same
36 system. One is not a replacement for the other.
38 [Using configfs]
40 configfs can be compiled as a module or into the kernel. You can access
41 it by doing
43 mount -t configfs none /config
45 The configfs tree will be empty unless client modules are also loaded.
46 These are modules that register their item types with configfs as
47 subsystems. Once a client subsystem is loaded, it will appear as a
48 subdirectory (or more than one) under /config. Like sysfs, the
49 configfs tree is always there, whether mounted on /config or not.
51 An item is created via mkdir(2). The item's attributes will also
52 appear at this time. readdir(3) can determine what the attributes are,
53 read(2) can query their default values, and write(2) can store new
54 values. Don't mix more than one attribute in one attribute file.
56 There are two types of configfs attributes:
58 * Normal attributes, which similar to sysfs attributes, are small ASCII text
59 files, with a maximum size of one page (PAGE_SIZE, 4096 on i386). Preferably
60 only one value per file should be used, and the same caveats from sysfs apply.
61 Configfs expects write(2) to store the entire buffer at once. When writing to
62 normal configfs attributes, userspace processes should first read the entire
63 file, modify the portions they wish to change, and then write the entire
64 buffer back.
66 * Binary attributes, which are somewhat similar to sysfs binary attributes,
67 but with a few slight changes to semantics. The PAGE_SIZE limitation does not
68 apply, but the whole binary item must fit in single kernel vmalloc'ed buffer.
69 The write(2) calls from user space are buffered, and the attributes'
70 write_bin_attribute method will be invoked on the final close, therefore it is
71 imperative for user-space to check the return code of close(2) in order to
72 verify that the operation finished successfully.
73 To avoid a malicious user OOMing the kernel, there's a per-binary attribute
74 maximum buffer value.
76 When an item needs to be destroyed, remove it with rmdir(2). An
77 item cannot be destroyed if any other item has a link to it (via
78 symlink(2)). Links can be removed via unlink(2).
80 [Configuring FakeNBD: an Example]
82 Imagine there's a Network Block Device (NBD) driver that allows you to
83 access remote block devices. Call it FakeNBD. FakeNBD uses configfs
84 for its configuration. Obviously, there will be a nice program that
85 sysadmins use to configure FakeNBD, but somehow that program has to tell
86 the driver about it. Here's where configfs comes in.
88 When the FakeNBD driver is loaded, it registers itself with configfs.
89 readdir(3) sees this just fine:
91 # ls /config
92 fakenbd
94 A fakenbd connection can be created with mkdir(2). The name is
95 arbitrary, but likely the tool will make some use of the name. Perhaps
96 it is a uuid or a disk name:
98 # mkdir /config/fakenbd/disk1
99 # ls /config/fakenbd/disk1
100 target device rw
102 The target attribute contains the IP address of the server FakeNBD will
103 connect to. The device attribute is the device on the server.
104 Predictably, the rw attribute determines whether the connection is
105 read-only or read-write.
107 # echo 10.0.0.1 > /config/fakenbd/disk1/target
108 # echo /dev/sda1 > /config/fakenbd/disk1/device
109 # echo 1 > /config/fakenbd/disk1/rw
111 That's it. That's all there is. Now the device is configured, via the
112 shell no less.
114 [Coding With configfs]
116 Every object in configfs is a config_item. A config_item reflects an
117 object in the subsystem. It has attributes that match values on that
118 object. configfs handles the filesystem representation of that object
119 and its attributes, allowing the subsystem to ignore all but the
120 basic show/store interaction.
122 Items are created and destroyed inside a config_group. A group is a
123 collection of items that share the same attributes and operations.
124 Items are created by mkdir(2) and removed by rmdir(2), but configfs
125 handles that. The group has a set of operations to perform these tasks
127 A subsystem is the top level of a client module. During initialization,
128 the client module registers the subsystem with configfs, the subsystem
129 appears as a directory at the top of the configfs filesystem. A
130 subsystem is also a config_group, and can do everything a config_group
131 can.
133 [struct config_item]
135 struct config_item {
136 char *ci_name;
137 char ci_namebuf[UOBJ_NAME_LEN];
138 struct kref ci_kref;
139 struct list_head ci_entry;
140 struct config_item *ci_parent;
141 struct config_group *ci_group;
142 struct config_item_type *ci_type;
143 struct dentry *ci_dentry;
144 };
146 void config_item_init(struct config_item *);
147 void config_item_init_type_name(struct config_item *,
148 const char *name,
149 struct config_item_type *type);
150 struct config_item *config_item_get(struct config_item *);
151 void config_item_put(struct config_item *);
153 Generally, struct config_item is embedded in a container structure, a
154 structure that actually represents what the subsystem is doing. The
155 config_item portion of that structure is how the object interacts with
156 configfs.
158 Whether statically defined in a source file or created by a parent
159 config_group, a config_item must have one of the _init() functions
160 called on it. This initializes the reference count and sets up the
161 appropriate fields.
163 All users of a config_item should have a reference on it via
164 config_item_get(), and drop the reference when they are done via
165 config_item_put().
167 By itself, a config_item cannot do much more than appear in configfs.
168 Usually a subsystem wants the item to display and/or store attributes,
169 among other things. For that, it needs a type.
171 [struct config_item_type]
173 struct configfs_item_operations {
174 void (*release)(struct config_item *);
175 int (*allow_link)(struct config_item *src,
176 struct config_item *target);
177 void (*drop_link)(struct config_item *src,
178 struct config_item *target);
179 };
181 struct config_item_type {
182 struct module *ct_owner;
183 struct configfs_item_operations *ct_item_ops;
184 struct configfs_group_operations *ct_group_ops;
185 struct configfs_attribute **ct_attrs;
186 struct configfs_bin_attribute **ct_bin_attrs;
187 };
189 The most basic function of a config_item_type is to define what
190 operations can be performed on a config_item. All items that have been
191 allocated dynamically will need to provide the ct_item_ops->release()
192 method. This method is called when the config_item's reference count
193 reaches zero.
195 [struct configfs_attribute]
197 struct configfs_attribute {
198 char *ca_name;
199 struct module *ca_owner;
200 umode_t ca_mode;
201 ssize_t (*show)(struct config_item *, char *);
202 ssize_t (*store)(struct config_item *, const char *, size_t);
203 };
205 When a config_item wants an attribute to appear as a file in the item's
206 configfs directory, it must define a configfs_attribute describing it.
207 It then adds the attribute to the NULL-terminated array
208 config_item_type->ct_attrs. When the item appears in configfs, the
209 attribute file will appear with the configfs_attribute->ca_name
210 filename. configfs_attribute->ca_mode specifies the file permissions.
212 If an attribute is readable and provides a ->show method, that method will
213 be called whenever userspace asks for a read(2) on the attribute. If an
214 attribute is writable and provides a ->store method, that method will be
215 be called whenever userspace asks for a write(2) on the attribute.
217 [struct configfs_bin_attribute]
219 struct configfs_attribute {
220 struct configfs_attribute cb_attr;
221 void *cb_private;
222 size_t cb_max_size;
223 };
225 The binary attribute is used when the one needs to use binary blob to
226 appear as the contents of a file in the item's configfs directory.
227 To do so add the binary attribute to the NULL-terminated array
228 config_item_type->ct_bin_attrs, and the item appears in configfs, the
229 attribute file will appear with the configfs_bin_attribute->cb_attr.ca_name
230 filename. configfs_bin_attribute->cb_attr.ca_mode specifies the file
231 permissions.
232 The cb_private member is provided for use by the driver, while the
233 cb_max_size member specifies the maximum amount of vmalloc buffer
234 to be used.
236 If binary attribute is readable and the config_item provides a
237 ct_item_ops->read_bin_attribute() method, that method will be called
238 whenever userspace asks for a read(2) on the attribute. The converse
239 will happen for write(2). The reads/writes are bufferred so only a
240 single read/write will occur; the attributes' need not concern itself
241 with it.
243 [struct config_group]
245 A config_item cannot live in a vacuum. The only way one can be created
246 is via mkdir(2) on a config_group. This will trigger creation of a
247 child item.
249 struct config_group {
250 struct config_item cg_item;
251 struct list_head cg_children;
252 struct configfs_subsystem *cg_subsys;
253 struct list_head default_groups;
254 struct list_head group_entry;
255 };
257 void config_group_init(struct config_group *group);
258 void config_group_init_type_name(struct config_group *group,
259 const char *name,
260 struct config_item_type *type);
263 The config_group structure contains a config_item. Properly configuring
264 that item means that a group can behave as an item in its own right.
265 However, it can do more: it can create child items or groups. This is
266 accomplished via the group operations specified on the group's
267 config_item_type.
269 struct configfs_group_operations {
270 struct config_item *(*make_item)(struct config_group *group,
271 const char *name);
272 struct config_group *(*make_group)(struct config_group *group,
273 const char *name);
274 int (*commit_item)(struct config_item *item);
275 void (*disconnect_notify)(struct config_group *group,
276 struct config_item *item);
277 void (*drop_item)(struct config_group *group,
278 struct config_item *item);
279 };
281 A group creates child items by providing the
282 ct_group_ops->make_item() method. If provided, this method is called from mkdir(2) in the group's directory. The subsystem allocates a new
283 config_item (or more likely, its container structure), initializes it,
284 and returns it to configfs. Configfs will then populate the filesystem
285 tree to reflect the new item.
287 If the subsystem wants the child to be a group itself, the subsystem
288 provides ct_group_ops->make_group(). Everything else behaves the same,
289 using the group _init() functions on the group.
291 Finally, when userspace calls rmdir(2) on the item or group,
292 ct_group_ops->drop_item() is called. As a config_group is also a
293 config_item, it is not necessary for a separate drop_group() method.
294 The subsystem must config_item_put() the reference that was initialized
295 upon item allocation. If a subsystem has no work to do, it may omit
296 the ct_group_ops->drop_item() method, and configfs will call
297 config_item_put() on the item on behalf of the subsystem.
299 IMPORTANT: drop_item() is void, and as such cannot fail. When rmdir(2)
300 is called, configfs WILL remove the item from the filesystem tree
301 (assuming that it has no children to keep it busy). The subsystem is
302 responsible for responding to this. If the subsystem has references to
303 the item in other threads, the memory is safe. It may take some time
304 for the item to actually disappear from the subsystem's usage. But it
305 is gone from configfs.
307 When drop_item() is called, the item's linkage has already been torn
308 down. It no longer has a reference on its parent and has no place in
309 the item hierarchy. If a client needs to do some cleanup before this
310 teardown happens, the subsystem can implement the
311 ct_group_ops->disconnect_notify() method. The method is called after
312 configfs has removed the item from the filesystem view but before the
313 item is removed from its parent group. Like drop_item(),
314 disconnect_notify() is void and cannot fail. Client subsystems should
315 not drop any references here, as they still must do it in drop_item().
317 A config_group cannot be removed while it still has child items. This
318 is implemented in the configfs rmdir(2) code. ->drop_item() will not be
319 called, as the item has not been dropped. rmdir(2) will fail, as the
320 directory is not empty.
322 [struct configfs_subsystem]
324 A subsystem must register itself, usually at module_init time. This
325 tells configfs to make the subsystem appear in the file tree.
327 struct configfs_subsystem {
328 struct config_group su_group;
329 struct mutex su_mutex;
330 };
332 int configfs_register_subsystem(struct configfs_subsystem *subsys);
333 void configfs_unregister_subsystem(struct configfs_subsystem *subsys);
335 A subsystem consists of a toplevel config_group and a mutex.
336 The group is where child config_items are created. For a subsystem,
337 this group is usually defined statically. Before calling
338 configfs_register_subsystem(), the subsystem must have initialized the
339 group via the usual group _init() functions, and it must also have
340 initialized the mutex.
341 When the register call returns, the subsystem is live, and it
342 will be visible via configfs. At that point, mkdir(2) can be called and
343 the subsystem must be ready for it.
345 [An Example]
347 The best example of these basic concepts is the simple_children
348 subsystem/group and the simple_child item in
349 samples/configfs/configfs_sample.c. It shows a trivial object displaying
350 and storing an attribute, and a simple group creating and destroying
351 these children.
353 [Hierarchy Navigation and the Subsystem Mutex]
355 There is an extra bonus that configfs provides. The config_groups and
356 config_items are arranged in a hierarchy due to the fact that they
357 appear in a filesystem. A subsystem is NEVER to touch the filesystem
358 parts, but the subsystem might be interested in this hierarchy. For
359 this reason, the hierarchy is mirrored via the config_group->cg_children
360 and config_item->ci_parent structure members.
362 A subsystem can navigate the cg_children list and the ci_parent pointer
363 to see the tree created by the subsystem. This can race with configfs'
364 management of the hierarchy, so configfs uses the subsystem mutex to
365 protect modifications. Whenever a subsystem wants to navigate the
366 hierarchy, it must do so under the protection of the subsystem
367 mutex.
369 A subsystem will be prevented from acquiring the mutex while a newly
370 allocated item has not been linked into this hierarchy. Similarly, it
371 will not be able to acquire the mutex while a dropping item has not
372 yet been unlinked. This means that an item's ci_parent pointer will
373 never be NULL while the item is in configfs, and that an item will only
374 be in its parent's cg_children list for the same duration. This allows
375 a subsystem to trust ci_parent and cg_children while they hold the
376 mutex.
378 [Item Aggregation Via symlink(2)]
380 configfs provides a simple group via the group->item parent/child
381 relationship. Often, however, a larger environment requires aggregation
382 outside of the parent/child connection. This is implemented via
383 symlink(2).
385 A config_item may provide the ct_item_ops->allow_link() and
386 ct_item_ops->drop_link() methods. If the ->allow_link() method exists,
387 symlink(2) may be called with the config_item as the source of the link.
388 These links are only allowed between configfs config_items. Any
389 symlink(2) attempt outside the configfs filesystem will be denied.
391 When symlink(2) is called, the source config_item's ->allow_link()
392 method is called with itself and a target item. If the source item
393 allows linking to target item, it returns 0. A source item may wish to
394 reject a link if it only wants links to a certain type of object (say,
395 in its own subsystem).
397 When unlink(2) is called on the symbolic link, the source item is
398 notified via the ->drop_link() method. Like the ->drop_item() method,
399 this is a void function and cannot return failure. The subsystem is
400 responsible for responding to the change.
402 A config_item cannot be removed while it links to any other item, nor
403 can it be removed while an item links to it. Dangling symlinks are not
404 allowed in configfs.
406 [Automatically Created Subgroups]
408 A new config_group may want to have two types of child config_items.
409 While this could be codified by magic names in ->make_item(), it is much
410 more explicit to have a method whereby userspace sees this divergence.
412 Rather than have a group where some items behave differently than
413 others, configfs provides a method whereby one or many subgroups are
414 automatically created inside the parent at its creation. Thus,
415 mkdir("parent") results in "parent", "parent/subgroup1", up through
416 "parent/subgroupN". Items of type 1 can now be created in
417 "parent/subgroup1", and items of type N can be created in
418 "parent/subgroupN".
420 These automatic subgroups, or default groups, do not preclude other
421 children of the parent group. If ct_group_ops->make_group() exists,
422 other child groups can be created on the parent group directly.
424 A configfs subsystem specifies default groups by adding them using the
425 configfs_add_default_group() function to the parent config_group
426 structure. Each added group is populated in the configfs tree at the same
427 time as the parent group. Similarly, they are removed at the same time
428 as the parent. No extra notification is provided. When a ->drop_item()
429 method call notifies the subsystem the parent group is going away, it
430 also means every default group child associated with that parent group.
432 As a consequence of this, default groups cannot be removed directly via
433 rmdir(2). They also are not considered when rmdir(2) on the parent
434 group is checking for children.
436 [Dependent Subsystems]
438 Sometimes other drivers depend on particular configfs items. For
439 example, ocfs2 mounts depend on a heartbeat region item. If that
440 region item is removed with rmdir(2), the ocfs2 mount must BUG or go
441 readonly. Not happy.
443 configfs provides two additional API calls: configfs_depend_item() and
444 configfs_undepend_item(). A client driver can call
445 configfs_depend_item() on an existing item to tell configfs that it is
446 depended on. configfs will then return -EBUSY from rmdir(2) for that
447 item. When the item is no longer depended on, the client driver calls
448 configfs_undepend_item() on it.
450 These API cannot be called underneath any configfs callbacks, as
451 they will conflict. They can block and allocate. A client driver
452 probably shouldn't calling them of its own gumption. Rather it should
453 be providing an API that external subsystems call.
455 How does this work? Imagine the ocfs2 mount process. When it mounts,
456 it asks for a heartbeat region item. This is done via a call into the
457 heartbeat code. Inside the heartbeat code, the region item is looked
458 up. Here, the heartbeat code calls configfs_depend_item(). If it
459 succeeds, then heartbeat knows the region is safe to give to ocfs2.
460 If it fails, it was being torn down anyway, and heartbeat can gracefully
461 pass up an error.
463 [Committable Items]
465 NOTE: Committable items are currently unimplemented.
467 Some config_items cannot have a valid initial state. That is, no
468 default values can be specified for the item's attributes such that the
469 item can do its work. Userspace must configure one or more attributes,
470 after which the subsystem can start whatever entity this item
471 represents.
473 Consider the FakeNBD device from above. Without a target address *and*
474 a target device, the subsystem has no idea what block device to import.
475 The simple example assumes that the subsystem merely waits until all the
476 appropriate attributes are configured, and then connects. This will,
477 indeed, work, but now every attribute store must check if the attributes
478 are initialized. Every attribute store must fire off the connection if
479 that condition is met.
481 Far better would be an explicit action notifying the subsystem that the
482 config_item is ready to go. More importantly, an explicit action allows
483 the subsystem to provide feedback as to whether the attributes are
484 initialized in a way that makes sense. configfs provides this as
485 committable items.
487 configfs still uses only normal filesystem operations. An item is
488 committed via rename(2). The item is moved from a directory where it
489 can be modified to a directory where it cannot.
491 Any group that provides the ct_group_ops->commit_item() method has
492 committable items. When this group appears in configfs, mkdir(2) will
493 not work directly in the group. Instead, the group will have two
494 subdirectories: "live" and "pending". The "live" directory does not
495 support mkdir(2) or rmdir(2) either. It only allows rename(2). The
496 "pending" directory does allow mkdir(2) and rmdir(2). An item is
497 created in the "pending" directory. Its attributes can be modified at
498 will. Userspace commits the item by renaming it into the "live"
499 directory. At this point, the subsystem receives the ->commit_item()
500 callback. If all required attributes are filled to satisfaction, the
501 method returns zero and the item is moved to the "live" directory.
503 As rmdir(2) does not work in the "live" directory, an item must be
504 shutdown, or "uncommitted". Again, this is done via rename(2), this
505 time from the "live" directory back to the "pending" one. The subsystem
506 is notified by the ct_group_ops->uncommit_object() method.