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"
46 .Cd options GEOM_CACHE
47 .Cd options GEOM_CONCAT
50 .Cd options GEOM_JOURNAL
51 .Cd options GEOM_LABEL
52 .Cd options GEOM_LINUX_LVM
54 .Cd options GEOM_MIRROR
55 .Cd options GEOM_MOUNTVER
56 .Cd options GEOM_MULTIPATH
58 .Cd options GEOM_PART_APM
59 .Cd options GEOM_PART_BSD
60 .Cd options GEOM_PART_BSD64
61 .Cd options GEOM_PART_EBR
62 .Cd options GEOM_PART_EBR_COMPAT
63 .Cd options GEOM_PART_GPT
64 .Cd options GEOM_PART_LDM
65 .Cd options GEOM_PART_MBR
66 .Cd options GEOM_PART_VTOC8
68 .Cd options GEOM_RAID3
69 .Cd options GEOM_SHSEC
70 .Cd options GEOM_STRIPE
72 .Cd options GEOM_VIRSTOR
77 framework provides an infrastructure in which
79 can perform transformations on disk I/O requests on their path from
80 the upper kernel to the device drivers and back.
84 context range from the simple geometric
85 displacement performed in typical disk partitioning modules over RAID
86 algorithms and device multipath resolution to full blown cryptographic
87 protection of the stored data.
89 Compared to traditional
90 .Dq "volume management" ,
93 and in some cases all previous implementations in the following ways:
98 It is trivially simple to write a new class
99 of transformation and it will not be given stepchild treatment.
101 someone for some reason wanted to mount IBM MVS diskpacks, a class
102 recognizing and configuring their VTOC information would be a trivial
106 is topologically agnostic.
107 Most volume management implementations
108 have very strict notions of how classes can fit together, very often
109 one fixed hierarchy is provided, for instance, subdisk - plex -
113 Being extensible means that new transformations are treated no differently
114 than existing transformations.
116 Fixed hierarchies are bad because they make it impossible to express
117 the intent efficiently.
118 In the fixed hierarchy above, it is not possible to mirror two
119 physical disks and then partition the mirror into subdisks, instead
120 one is forced to make subdisks on the physical volumes and to mirror
121 these two and two, resulting in a much more complex configuration.
123 on the other hand does not care in which order things are done,
124 the only restriction is that cycles in the graph will not be allowed.
125 .Sh "TERMINOLOGY AND TOPOLOGY"
127 is quite object oriented and consequently the terminology
128 borrows a lot of context and semantics from the OO vocabulary:
132 represented by the data structure
135 particular kind of transformation.
136 Typical examples are MBR disk
137 partition, BSD disklabel, and RAID5 classes.
139 An instance of a class is called a
141 and represented by the data structure
146 will be one geom of class MBR for each disk.
150 represented by the data structure
152 is the front gate at which a geom offers service.
155 a disk-like thing which appears in
159 All providers have three main properties:
167 is the backdoor through which a geom connects to another
168 geom provider and through which I/O requests are sent.
170 The topological relationship between these entities are as follows:
173 A class has zero or more geom instances.
175 A geom has exactly one class it is derived from.
177 A geom has zero or more consumers.
179 A geom has zero or more providers.
181 A consumer can be attached to zero or one providers.
183 A provider can have zero or more consumers attached.
186 All geoms have a rank-number assigned, which is used to detect and
187 prevent loops in the acyclic directed graph.
192 A geom with no attached consumers has rank=1.
194 A geom with attached consumers has a rank one higher than the
195 highest rank of the geoms of the providers its consumers are
198 .Sh "SPECIAL TOPOLOGICAL MANEUVERS"
199 In addition to the straightforward attach, which attaches a consumer
200 to a provider, and detach, which breaks the bond, a number of special
201 topological maneuvers exists to facilitate configuration and to
202 improve the overall flexibility.
205 is a process that happens whenever a new class or new provider
206 is created, and it provides the class a chance to automatically configure an
207 instance on providers which it recognizes as its own.
208 A typical example is the MBR disk-partition class which will look for
209 the MBR table in the first sector and, if found and validated, will
210 instantiate a geom to multiplex according to the contents of the MBR.
212 A new class will be offered to all existing providers in turn and a new
213 provider will be offered to all classes in turn.
215 Exactly what a class does to recognize if it should accept the offered
216 provider is not defined by
218 but the sensible set of options are:
221 Examine specific data structures on the disk.
223 Examine properties like
229 Examine the rank number of the provider's geom.
231 Examine the method name of the provider's geom.
234 is the process by which a provider is removed while
235 it potentially is still being used.
237 When a geom orphans a provider, all future I/O requests will
239 on the provider with an error code set by the geom.
241 consumers attached to the provider will receive notification about
242 the orphanization when the event loop gets around to it, and they
243 can take appropriate action at that time.
245 A geom which came into being as a result of a normal taste operation
246 should self-destruct unless it has a way to keep functioning whilst
247 lacking the orphaned provider.
248 Geoms like disk slicers should therefore self-destruct whereas
249 RAID5 or mirror geoms will be able to continue as long as they do
252 When a provider is orphaned, this does not necessarily result in any
253 immediate change in the topology: any attached consumers are still
254 attached, any opened paths are still open, any outstanding I/O
255 requests are still outstanding.
257 The typical scenario is:
259 .Bl -bullet -offset indent -compact
261 A device driver detects a disk has departed and orphans the provider for it.
263 The geoms on top of the disk receive the orphanization event and
264 orphan all their providers in turn.
265 Providers which are not attached to will typically self-destruct
267 This process continues in a quasi-recursive fashion until all
268 relevant pieces of the tree have heard the bad news.
270 Eventually the buck stops when it reaches geom_dev at the top
275 to stop any more requests from
277 It will sleep until any and all outstanding I/O requests have
279 It will explicitly close (i.e.: zero the access counts), a change
280 which will propagate all the way down through the mesh.
281 It will then detach and destroy its geom.
283 The geom whose provider is now detached will destroy the provider,
284 detach and destroy its consumer and destroy its geom.
286 This process percolates all the way down through the mesh, until
287 the cleanup is complete.
290 While this approach seems byzantine, it does provide the maximum
291 flexibility and robustness in handling disappearing devices.
293 The one absolutely crucial detail to be aware of is that if the
294 device driver does not return all I/O requests, the tree will
297 is a special case of orphanization used to protect
298 against stale metadata.
299 It is probably easiest to understand spoiling by going through
304 on top of which an MBR geom provides
314 and that both the MBR and BSD geoms have
315 autoconfigured based on data structures on the disk media.
316 Now imagine the case where
318 is opened for writing and those
319 data structures are modified or overwritten: now the geoms would
320 be operating on stale metadata unless some notification system
321 can inform them otherwise.
323 To avoid this situation, when the open of
326 all attached consumers are told about this and geoms like
327 MBR and BSD will self-destruct as a result.
330 is closed, it will be offered for tasting again
331 and, if the data structures for MBR and BSD are still there, new
332 geoms will instantiate themselves anew.
334 Now for the fine print:
336 If any of the paths through the MBR or BSD module were open, they
337 would have opened downwards with an exclusive bit thus rendering it
340 for writing in that case.
342 the requested exclusive bit would render it impossible to open a
343 path through the MBR geom while
347 From this it also follows that changing the size of open geoms can
348 only be done with their cooperation.
350 Finally: the spoiling only happens when the write count goes from
351 zero to non-zero and the retasting happens only when the write count goes
352 from non-zero to zero.
354 is the process where the administrator issues instructions
355 for a particular class to instantiate itself.
357 ways to express intent in this case - a particular provider may be
358 specified with a level of override forcing, for instance, a BSD
359 disklabel module to attach to a provider which was not found palatable
360 during the TASTE operation.
362 Finally, I/O is the reason we even do this: it concerns itself with
363 sending I/O requests through the graph.
364 .It Em "I/O REQUESTS" ,
367 originate at a consumer,
368 are scheduled on its attached provider and, when processed, are returned
370 It is important to realize that the
372 which enters through the provider of a particular geom does not
374 come out on the other side
376 Even simple transformations like MBR and BSD will clone the
378 modify the clone, and schedule the clone on their
380 Note that cloning the
382 does not involve cloning the
383 actual data area specified in the I/O request.
385 In total, four different I/O requests exist in
387 read, write, delete, and
390 Read and write are self explanatory.
392 Delete indicates that a certain range of data is no longer used
393 and that it can be erased or freed as the underlying technology
395 Technologies like flash adaptation layers can arrange to erase
396 the relevant blocks before they will become reassigned and
397 cryptographic devices may want to fill random bits into the
398 range to reduce the amount of data available for attack.
400 It is important to recognize that a delete indication is not a
401 request and consequently there is no guarantee that the data actually
402 will be erased or made unavailable unless guaranteed by specific
406 semantics are required, a
407 geom should be pushed which converts delete indications into (a
408 sequence of) write requests.
411 supports inspection and manipulation
412 of out-of-band attributes on a particular provider or path.
413 Attributes are named by
415 strings and they will be discussed in
416 a separate section below.
419 (Stay tuned while the author rests his brain and fingers: more to come.)
421 Several flags are provided for tracing
423 operations and unlocking
424 protection mechanisms via the
425 .Va kern.geom.debugflags
427 All of these flags are off by default, and great care should be taken in
429 .Bl -tag -width indent
430 .It 0x01 Pq Dv G_T_TOPOLOGY
431 Provide tracing of topology change events.
432 .It 0x02 Pq Dv G_T_BIO
433 Provide tracing of buffer I/O requests.
434 .It 0x04 Pq Dv G_T_ACCESS
435 Provide tracing of access check controls.
437 .It 0x10 (allow foot shooting)
438 Allow writing to Rank 1 providers.
439 This would, for example, allow the super-user to overwrite the MBR on the root
440 disk or write random sectors elsewhere to a mounted disk.
441 The implications are obvious.
442 .It 0x40 Pq Dv G_F_DISKIOCTL
443 This is unused at this time.
444 .It 0x80 Pq Dv G_F_CTLDUMP
445 Dump contents of gctl requests.
449 The following options have been deprecated and will be removed in
462 .Cd GEOM_PART_VTOC8 ,
464 options, respectively, instead.
467 .Xr DECLARE_GEOM_CLASS 9 ,
477 .Xr g_provider_by_name 9
479 This software was developed for the
482 .An Poul-Henning Kamp
483 and NAI Labs, the Security Research Division of Network Associates, Inc.\&
484 under DARPA/SPAWAR contract N66001-01-C-8035
487 DARPA CHATS research program.
489 The first precursor for
491 was a gruesome hack to Minix 1.2 and was
493 An earlier attempt to implement a less general scheme
498 .An Poul-Henning Kamp Aq Mt phk@FreeBSD.org