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"
47 .Cd options GEOM_CACHE
48 .Cd options GEOM_CONCAT
52 .Cd options GEOM_JOURNAL
53 .Cd options GEOM_LABEL
54 .Cd options GEOM_LINUX_LVM
57 .Cd options GEOM_MIRROR
58 .Cd options GEOM_MOUNTVER
59 .Cd options GEOM_MULTIPATH
61 .Cd options GEOM_PART_APM
62 .Cd options GEOM_PART_BSD
63 .Cd options GEOM_PART_BSD64
64 .Cd options GEOM_PART_EBR
65 .Cd options GEOM_PART_EBR_COMPAT
66 .Cd options GEOM_PART_GPT
67 .Cd options GEOM_PART_LDM
68 .Cd options GEOM_PART_MBR
69 .Cd options GEOM_PART_VTOC8
71 .Cd options GEOM_RAID3
72 .Cd options GEOM_SHSEC
73 .Cd options GEOM_STRIPE
74 .Cd options GEOM_SUNLABEL
76 .Cd options GEOM_VIRSTOR
82 framework provides an infrastructure in which
84 can perform transformations on disk I/O requests on their path from
85 the upper kernel to the device drivers and back.
89 context range from the simple geometric
90 displacement performed in typical disk partitioning modules over RAID
91 algorithms and device multipath resolution to full blown cryptographic
92 protection of the stored data.
94 Compared to traditional
95 .Dq "volume management" ,
98 and in some cases all previous implementations in the following ways:
103 It is trivially simple to write a new class
104 of transformation and it will not be given stepchild treatment.
106 someone for some reason wanted to mount IBM MVS diskpacks, a class
107 recognizing and configuring their VTOC information would be a trivial
111 is topologically agnostic.
112 Most volume management implementations
113 have very strict notions of how classes can fit together, very often
114 one fixed hierarchy is provided, for instance, subdisk - plex -
118 Being extensible means that new transformations are treated no differently
119 than existing transformations.
121 Fixed hierarchies are bad because they make it impossible to express
122 the intent efficiently.
123 In the fixed hierarchy above, it is not possible to mirror two
124 physical disks and then partition the mirror into subdisks, instead
125 one is forced to make subdisks on the physical volumes and to mirror
126 these two and two, resulting in a much more complex configuration.
128 on the other hand does not care in which order things are done,
129 the only restriction is that cycles in the graph will not be allowed.
130 .Sh "TERMINOLOGY AND TOPOLOGY"
132 is quite object oriented and consequently the terminology
133 borrows a lot of context and semantics from the OO vocabulary:
137 represented by the data structure
140 particular kind of transformation.
141 Typical examples are MBR disk
142 partition, BSD disklabel, and RAID5 classes.
144 An instance of a class is called a
146 and represented by the data structure
151 will be one geom of class MBR for each disk.
155 represented by the data structure
157 is the front gate at which a geom offers service.
160 a disk-like thing which appears in
164 All providers have three main properties:
172 is the backdoor through which a geom connects to another
173 geom provider and through which I/O requests are sent.
175 The topological relationship between these entities are as follows:
178 A class has zero or more geom instances.
180 A geom has exactly one class it is derived from.
182 A geom has zero or more consumers.
184 A geom has zero or more providers.
186 A consumer can be attached to zero or one providers.
188 A provider can have zero or more consumers attached.
191 All geoms have a rank-number assigned, which is used to detect and
192 prevent loops in the acyclic directed graph.
197 A geom with no attached consumers has rank=1.
199 A geom with attached consumers has a rank one higher than the
200 highest rank of the geoms of the providers its consumers are
203 .Sh "SPECIAL TOPOLOGICAL MANEUVERS"
204 In addition to the straightforward attach, which attaches a consumer
205 to a provider, and detach, which breaks the bond, a number of special
206 topological maneuvers exists to facilitate configuration and to
207 improve the overall flexibility.
210 is a process that happens whenever a new class or new provider
211 is created, and it provides the class a chance to automatically configure an
212 instance on providers which it recognizes as its own.
213 A typical example is the MBR disk-partition class which will look for
214 the MBR table in the first sector and, if found and validated, will
215 instantiate a geom to multiplex according to the contents of the MBR.
217 A new class will be offered to all existing providers in turn and a new
218 provider will be offered to all classes in turn.
220 Exactly what a class does to recognize if it should accept the offered
221 provider is not defined by
223 but the sensible set of options are:
226 Examine specific data structures on the disk.
228 Examine properties like
234 Examine the rank number of the provider's geom.
236 Examine the method name of the provider's geom.
239 is the process by which a provider is removed while
240 it potentially is still being used.
242 When a geom orphans a provider, all future I/O requests will
244 on the provider with an error code set by the geom.
246 consumers attached to the provider will receive notification about
247 the orphanization when the event loop gets around to it, and they
248 can take appropriate action at that time.
250 A geom which came into being as a result of a normal taste operation
251 should self-destruct unless it has a way to keep functioning whilst
252 lacking the orphaned provider.
253 Geoms like disk slicers should therefore self-destruct whereas
254 RAID5 or mirror geoms will be able to continue as long as they do
257 When a provider is orphaned, this does not necessarily result in any
258 immediate change in the topology: any attached consumers are still
259 attached, any opened paths are still open, any outstanding I/O
260 requests are still outstanding.
262 The typical scenario is:
264 .Bl -bullet -offset indent -compact
266 A device driver detects a disk has departed and orphans the provider for it.
268 The geoms on top of the disk receive the orphanization event and
269 orphan all their providers in turn.
270 Providers which are not attached to will typically self-destruct
272 This process continues in a quasi-recursive fashion until all
273 relevant pieces of the tree have heard the bad news.
275 Eventually the buck stops when it reaches geom_dev at the top
280 to stop any more requests from
282 It will sleep until any and all outstanding I/O requests have
284 It will explicitly close (i.e.: zero the access counts), a change
285 which will propagate all the way down through the mesh.
286 It will then detach and destroy its geom.
288 The geom whose provider is now detached will destroy the provider,
289 detach and destroy its consumer and destroy its geom.
291 This process percolates all the way down through the mesh, until
292 the cleanup is complete.
295 While this approach seems byzantine, it does provide the maximum
296 flexibility and robustness in handling disappearing devices.
298 The one absolutely crucial detail to be aware of is that if the
299 device driver does not return all I/O requests, the tree will
302 is a special case of orphanization used to protect
303 against stale metadata.
304 It is probably easiest to understand spoiling by going through
309 on top of which an MBR geom provides
319 and that both the MBR and BSD geoms have
320 autoconfigured based on data structures on the disk media.
321 Now imagine the case where
323 is opened for writing and those
324 data structures are modified or overwritten: now the geoms would
325 be operating on stale metadata unless some notification system
326 can inform them otherwise.
328 To avoid this situation, when the open of
331 all attached consumers are told about this and geoms like
332 MBR and BSD will self-destruct as a result.
335 is closed, it will be offered for tasting again
336 and, if the data structures for MBR and BSD are still there, new
337 geoms will instantiate themselves anew.
339 Now for the fine print:
341 If any of the paths through the MBR or BSD module were open, they
342 would have opened downwards with an exclusive bit thus rendering it
345 for writing in that case.
347 the requested exclusive bit would render it impossible to open a
348 path through the MBR geom while
352 From this it also follows that changing the size of open geoms can
353 only be done with their cooperation.
355 Finally: the spoiling only happens when the write count goes from
356 zero to non-zero and the retasting happens only when the write count goes
357 from non-zero to zero.
359 is the process where the administrator issues instructions
360 for a particular class to instantiate itself.
362 ways to express intent in this case - a particular provider may be
363 specified with a level of override forcing, for instance, a BSD
364 disklabel module to attach to a provider which was not found palatable
365 during the TASTE operation.
367 Finally, I/O is the reason we even do this: it concerns itself with
368 sending I/O requests through the graph.
369 .It Em "I/O REQUESTS" ,
372 originate at a consumer,
373 are scheduled on its attached provider and, when processed, are returned
375 It is important to realize that the
377 which enters through the provider of a particular geom does not
379 come out on the other side
381 Even simple transformations like MBR and BSD will clone the
383 modify the clone, and schedule the clone on their
385 Note that cloning the
387 does not involve cloning the
388 actual data area specified in the I/O request.
390 In total, four different I/O requests exist in
392 read, write, delete, and
395 Read and write are self explanatory.
397 Delete indicates that a certain range of data is no longer used
398 and that it can be erased or freed as the underlying technology
400 Technologies like flash adaptation layers can arrange to erase
401 the relevant blocks before they will become reassigned and
402 cryptographic devices may want to fill random bits into the
403 range to reduce the amount of data available for attack.
405 It is important to recognize that a delete indication is not a
406 request and consequently there is no guarantee that the data actually
407 will be erased or made unavailable unless guaranteed by specific
411 semantics are required, a
412 geom should be pushed which converts delete indications into (a
413 sequence of) write requests.
416 supports inspection and manipulation
417 of out-of-band attributes on a particular provider or path.
418 Attributes are named by
420 strings and they will be discussed in
421 a separate section below.
424 (Stay tuned while the author rests his brain and fingers: more to come.)
426 Several flags are provided for tracing
428 operations and unlocking
429 protection mechanisms via the
430 .Va kern.geom.debugflags
432 All of these flags are off by default, and great care should be taken in
434 .Bl -tag -width indent
435 .It 0x01 Pq Dv G_T_TOPOLOGY
436 Provide tracing of topology change events.
437 .It 0x02 Pq Dv G_T_BIO
438 Provide tracing of buffer I/O requests.
439 .It 0x04 Pq Dv G_T_ACCESS
440 Provide tracing of access check controls.
442 .It 0x10 (allow foot shooting)
443 Allow writing to Rank 1 providers.
444 This would, for example, allow the super-user to overwrite the MBR on the root
445 disk or write random sectors elsewhere to a mounted disk.
446 The implications are obvious.
447 .It 0x40 Pq Dv G_F_DISKIOCTL
448 This is unused at this time.
449 .It 0x80 Pq Dv G_F_CTLDUMP
450 Dump contents of gctl requests.
454 .Xr DECLARE_GEOM_CLASS 9 ,
464 .Xr g_provider_by_name 9
466 This software was developed for the
469 .An Poul-Henning Kamp
470 and NAI Labs, the Security Research Division of Network Associates, Inc.\&
471 under DARPA/SPAWAR contract N66001-01-C-8035
474 DARPA CHATS research program.
476 The first precursor for
478 was a gruesome hack to Minix 1.2 and was
480 An earlier attempt to implement a less general scheme
485 .An Poul-Henning Kamp Aq Mt phk@FreeBSD.org