2 .\" Copyright (c) 2002 Poul-Henning Kamp
3 .\" Copyright (c) 2002 Networks Associates Technology, Inc.
4 .\" All rights reserved.
6 .\" This software was developed for the FreeBSD Project by Poul-Henning Kamp
7 .\" and NAI Labs, the Security Research Division of Network Associates, Inc.
8 .\" under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9 .\" DARPA CHATS research program.
11 .\" Redistribution and use in source and binary forms, with or without
12 .\" modification, are permitted provided that the following conditions
14 .\" 1. Redistributions of source code must retain the above copyright
15 .\" notice, this list of conditions and the following disclaimer.
16 .\" 2. Redistributions in binary form must reproduce the above copyright
17 .\" notice, this list of conditions and the following disclaimer in the
18 .\" documentation and/or other materials provided with the distribution.
19 .\" 3. The names of the authors may not be used to endorse or promote
20 .\" products derived from this software without specific prior written
23 .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24 .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27 .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42 .Nd "modular disk I/O request transformation framework"
45 .Cd options GEOM_CACHE
46 .Cd options GEOM_CONCAT
49 .Cd options GEOM_JOURNAL
50 .Cd options GEOM_LABEL
51 .Cd options GEOM_LINUX_LVM
53 .Cd options GEOM_MIRROR
54 .Cd options GEOM_MOUNTVER
55 .Cd options GEOM_MULTIPATH
57 .Cd options GEOM_PART_APM
58 .Cd options GEOM_PART_BSD
59 .Cd options GEOM_PART_BSD64
60 .Cd options GEOM_PART_EBR
61 .Cd options GEOM_PART_EBR_COMPAT
62 .Cd options GEOM_PART_GPT
63 .Cd options GEOM_PART_LDM
64 .Cd options GEOM_PART_MBR
66 .Cd options GEOM_RAID3
67 .Cd options GEOM_SHSEC
68 .Cd options GEOM_STRIPE
70 .Cd options GEOM_VIRSTOR
75 framework provides an infrastructure in which
77 can perform transformations on disk I/O requests on their path from
78 the upper kernel to the device drivers and back.
82 context range from the simple geometric
83 displacement performed in typical disk partitioning modules over RAID
84 algorithms and device multipath resolution to full blown cryptographic
85 protection of the stored data.
87 Compared to traditional
88 .Dq "volume management" ,
91 and in some cases all previous implementations in the following ways:
96 It is trivially simple to write a new class
97 of transformation and it will not be given stepchild treatment.
99 someone for some reason wanted to mount IBM MVS diskpacks, a class
100 recognizing and configuring their VTOC information would be a trivial
104 is topologically agnostic.
105 Most volume management implementations
106 have very strict notions of how classes can fit together, very often
107 one fixed hierarchy is provided, for instance, subdisk - plex -
111 Being extensible means that new transformations are treated no differently
112 than existing transformations.
114 Fixed hierarchies are bad because they make it impossible to express
115 the intent efficiently.
116 In the fixed hierarchy above, it is not possible to mirror two
117 physical disks and then partition the mirror into subdisks, instead
118 one is forced to make subdisks on the physical volumes and to mirror
119 these two and two, resulting in a much more complex configuration.
121 on the other hand does not care in which order things are done,
122 the only restriction is that cycles in the graph will not be allowed.
123 .Sh "TERMINOLOGY AND TOPOLOGY"
125 is quite object oriented and consequently the terminology
126 borrows a lot of context and semantics from the OO vocabulary:
130 represented by the data structure
133 particular kind of transformation.
134 Typical examples are MBR disk
135 partition, BSD disklabel, and RAID5 classes.
137 An instance of a class is called a
139 and represented by the data structure
144 will be one geom of class MBR for each disk.
148 represented by the data structure
150 is the front gate at which a geom offers service.
153 a disk-like thing which appears in
157 All providers have three main properties:
165 is the backdoor through which a geom connects to another
166 geom provider and through which I/O requests are sent.
168 The topological relationship between these entities are as follows:
171 A class has zero or more geom instances.
173 A geom has exactly one class it is derived from.
175 A geom has zero or more consumers.
177 A geom has zero or more providers.
179 A consumer can be attached to zero or one providers.
181 A provider can have zero or more consumers attached.
184 All geoms have a rank-number assigned, which is used to detect and
185 prevent loops in the acyclic directed graph.
190 A geom with no attached consumers has rank=1.
192 A geom with attached consumers has a rank one higher than the
193 highest rank of the geoms of the providers its consumers are
196 .Sh "SPECIAL TOPOLOGICAL MANEUVERS"
197 In addition to the straightforward attach, which attaches a consumer
198 to a provider, and detach, which breaks the bond, a number of special
199 topological maneuvers exists to facilitate configuration and to
200 improve the overall flexibility.
203 is a process that happens whenever a new class or new provider
204 is created, and it provides the class a chance to automatically configure an
205 instance on providers which it recognizes as its own.
206 A typical example is the MBR disk-partition class which will look for
207 the MBR table in the first sector and, if found and validated, will
208 instantiate a geom to multiplex according to the contents of the MBR.
210 A new class will be offered to all existing providers in turn and a new
211 provider will be offered to all classes in turn.
213 Exactly what a class does to recognize if it should accept the offered
214 provider is not defined by
216 but the sensible set of options are:
219 Examine specific data structures on the disk.
221 Examine properties like
227 Examine the rank number of the provider's geom.
229 Examine the method name of the provider's geom.
232 is the process by which a provider is removed while
233 it potentially is still being used.
235 When a geom orphans a provider, all future I/O requests will
237 on the provider with an error code set by the geom.
239 consumers attached to the provider will receive notification about
240 the orphanization when the event loop gets around to it, and they
241 can take appropriate action at that time.
243 A geom which came into being as a result of a normal taste operation
244 should self-destruct unless it has a way to keep functioning whilst
245 lacking the orphaned provider.
246 Geoms like disk slicers should therefore self-destruct whereas
247 RAID5 or mirror geoms will be able to continue as long as they do
250 When a provider is orphaned, this does not necessarily result in any
251 immediate change in the topology: any attached consumers are still
252 attached, any opened paths are still open, any outstanding I/O
253 requests are still outstanding.
255 The typical scenario is:
257 .Bl -bullet -offset indent -compact
259 A device driver detects a disk has departed and orphans the provider for it.
261 The geoms on top of the disk receive the orphanization event and
262 orphan all their providers in turn.
263 Providers which are not attached to will typically self-destruct
265 This process continues in a quasi-recursive fashion until all
266 relevant pieces of the tree have heard the bad news.
268 Eventually the buck stops when it reaches geom_dev at the top
273 to stop any more requests from
275 It will sleep until any and all outstanding I/O requests have
277 It will explicitly close (i.e.: zero the access counts), a change
278 which will propagate all the way down through the mesh.
279 It will then detach and destroy its geom.
281 The geom whose provider is now detached will destroy the provider,
282 detach and destroy its consumer and destroy its geom.
284 This process percolates all the way down through the mesh, until
285 the cleanup is complete.
288 While this approach seems byzantine, it does provide the maximum
289 flexibility and robustness in handling disappearing devices.
291 The one absolutely crucial detail to be aware of is that if the
292 device driver does not return all I/O requests, the tree will
295 is a special case of orphanization used to protect
296 against stale metadata.
297 It is probably easiest to understand spoiling by going through
302 on top of which an MBR geom provides
312 and that both the MBR and BSD geoms have
313 autoconfigured based on data structures on the disk media.
314 Now imagine the case where
316 is opened for writing and those
317 data structures are modified or overwritten: now the geoms would
318 be operating on stale metadata unless some notification system
319 can inform them otherwise.
321 To avoid this situation, when the open of
324 all attached consumers are told about this and geoms like
325 MBR and BSD will self-destruct as a result.
328 is closed, it will be offered for tasting again
329 and, if the data structures for MBR and BSD are still there, new
330 geoms will instantiate themselves anew.
332 Now for the fine print:
334 If any of the paths through the MBR or BSD module were open, they
335 would have opened downwards with an exclusive bit thus rendering it
338 for writing in that case.
340 the requested exclusive bit would render it impossible to open a
341 path through the MBR geom while
345 From this it also follows that changing the size of open geoms can
346 only be done with their cooperation.
348 Finally: the spoiling only happens when the write count goes from
349 zero to non-zero and the retasting happens only when the write count goes
350 from non-zero to zero.
352 is the process where the administrator issues instructions
353 for a particular class to instantiate itself.
355 ways to express intent in this case - a particular provider may be
356 specified with a level of override forcing, for instance, a BSD
357 disklabel module to attach to a provider which was not found palatable
358 during the TASTE operation.
360 Finally, I/O is the reason we even do this: it concerns itself with
361 sending I/O requests through the graph.
362 .It Em "I/O REQUESTS" ,
365 originate at a consumer,
366 are scheduled on its attached provider and, when processed, are returned
368 It is important to realize that the
370 which enters through the provider of a particular geom does not
372 come out on the other side
374 Even simple transformations like MBR and BSD will clone the
376 modify the clone, and schedule the clone on their
378 Note that cloning the
380 does not involve cloning the
381 actual data area specified in the I/O request.
383 In total, four different I/O requests exist in
385 read, write, delete, and
388 Read and write are self explanatory.
390 Delete indicates that a certain range of data is no longer used
391 and that it can be erased or freed as the underlying technology
393 Technologies like flash adaptation layers can arrange to erase
394 the relevant blocks before they will become reassigned and
395 cryptographic devices may want to fill random bits into the
396 range to reduce the amount of data available for attack.
398 It is important to recognize that a delete indication is not a
399 request and consequently there is no guarantee that the data actually
400 will be erased or made unavailable unless guaranteed by specific
404 semantics are required, a
405 geom should be pushed which converts delete indications into (a
406 sequence of) write requests.
409 supports inspection and manipulation
410 of out-of-band attributes on a particular provider or path.
411 Attributes are named by
413 strings and they will be discussed in
414 a separate section below.
417 (Stay tuned while the author rests his brain and fingers: more to come.)
419 Several flags are provided for tracing
421 operations and unlocking
422 protection mechanisms via the
423 .Va kern.geom.debugflags
425 All of these flags are off by default, and great care should be taken in
427 .Bl -tag -width indent
428 .It 0x01 Pq Dv G_T_TOPOLOGY
429 Provide tracing of topology change events.
430 .It 0x02 Pq Dv G_T_BIO
431 Provide tracing of buffer I/O requests.
432 .It 0x04 Pq Dv G_T_ACCESS
433 Provide tracing of access check controls.
435 .It 0x10 (allow foot shooting)
436 Allow writing to Rank 1 providers.
437 This would, for example, allow the super-user to overwrite the MBR on the root
438 disk or write random sectors elsewhere to a mounted disk.
439 The implications are obvious.
440 .It 0x40 Pq Dv G_F_DISKIOCTL
441 This is unused at this time.
442 .It 0x80 Pq Dv G_F_CTLDUMP
443 Dump contents of gctl requests.
448 .Xr DECLARE_GEOM_CLASS 9 ,
458 .Xr g_provider_by_name 9
460 This software was initially developed for the
463 .An Poul-Henning Kamp
464 and NAI Labs, the Security Research Division of Network Associates, Inc.\&
465 under DARPA/SPAWAR contract N66001-01-C-8035
468 DARPA CHATS research program.
470 The following obsolete
472 components were removed in
474 .Bl -bullet -offset indent -compact
489 .Bl -bullet -offset indent -compact
500 options, respectively, instead.
502 .An Poul-Henning Kamp Aq Mt phk@FreeBSD.org