1.\" 2.\" Copyright (c) 2002 Poul-Henning Kamp 3.\" Copyright (c) 2002 Networks Associates Technology, Inc. 4.\" All rights reserved. 5.\" 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. 10.\" 11.\" Redistribution and use in source and binary forms, with or without 12.\" modification, are permitted provided that the following conditions 13.\" are met: 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 21.\" permission. 22.\" 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 33.\" SUCH DAMAGE. 34.\" 35.\" $FreeBSD$ 36.\" 37.Dd March 27, 2002 38.Os 39.Dt GEOM 4 40.Sh NAME 41.Nm GEOM 42.Nd modular disk I/O request transformation framework. 43.Sh DESCRIPTION 44The GEOM framework provides an infrastructure in which "classes" 45can perform transformations on disk I/O requests on their path from 46the upper kernel to the device drivers and back. 47.Pp 48Transformations in a GEOM context range from the simple geometric 49displacement performed in typical disk partitioning modules over RAID 50algorithms and device multipath resolution to full blown cryptographic 51protection of the stored data. 52.Pp 53Compared to traditional "volume management", GEOM differs from most 54and in some cases all previous implementations in the following ways: 55.Bl -bullet 56.It 57GEOM is extensible. It is trivially simple to write a new class 58of transformation and it will not be given stepchild treatment. If 59someone for some reason wanted to mount IBM MVS diskpacks, a class 60recognizing and configuring their VTOC information would be a trivial 61matter. 62.It 63GEOM is topologically agnostic. Most volume management implementations 64have very strict notions of how classes can fit together, very often 65one fixed hierarchy is provided for instance subdisk - plex - 66volume. 67.El 68.Pp 69Being extensible means that new transformations are treated no differently 70than existing transformations. 71.Pp 72Fixed hierarchies are bad because they make it impossible to express 73the intent efficiently. 74In the fixed hierarchy above it is not possible to mirror two 75physical disks and then partition the mirror into subdisks, instead 76one is forced to make subdisks on the physical volumes and to mirror 77these two and two resulting in a much more complex configuration. 78GEOM on the other hand does not care in which order things are done, 79the only restriction is that cycles in the graph will not be allowed. 80.Pp 81.Sh "TERMINOLOGY and TOPOLOGY" 82Geom is quite object oriented and consequently the terminology 83borrows a lot of context and semantics from the OO vocabulary: 84.Pp 85A "class", represented by the data structure g_class implements one 86particular kind of transformation. Typical examples are MBR disk 87partition, BSD disklabel, and RAID5 classes. 88.Pp 89An instance of a class is called a "geom" and represented by the 90data structure "g_geom". In a typical i386 FreeBSD system, there 91will be one geom of class MBR for each disk. 92.Pp 93A "provider", represented by the data structure "g_provider", is 94the front gate at which a geom offers service. 95A provider is "a disk-like thing which appears in /dev" - a logical 96disk in other words. 97All providers have three main properties: name, sectorsize and size. 98.Pp 99A "consumer" is the backdoor through which a geom connects to another 100geom provider and through which I/O requests are sent. 101.Pp 102The topological relationship between these entities are as follows: 103.Bl -bullet 104.It 105A class has zero or more geom instances. 106.It 107A geom has exactly one class it is derived from. 108.It 109A geom has zero or more consumers. 110.It 111A geom has zero or more providers. 112.It 113A consumer can be attached to zero or one providers. 114.It 115A provider can have zero or more consumers attached. 116.El 117.Pp 118All geoms have a rank-number assigned, which is used to detect and 119prevent loops in the acyclic directed graph. This rank number is 120assigned as follows: 121.Bl -enum 122.It 123A geom with no attached consumers has rank=1 124.It 125A geom with attached consumers has a rank one higher than the 126highest rank of the geoms of the providers its consumers are 127attached to. 128.El 129.Sh "SPECIAL TOPOLOGICAL MANEUVERS" 130In addition to the straightforward attach, which attaches a consumer 131to a provider, and detach, which breaks the bond, a number of special 132topological maneuvers exists to facilitate configuration and to 133improve the overall flexibility. 134.Pp 135.Em TASTING 136is a process that happens whenever a new class or new provider 137is created and it provides the class a chance to automatically configure an 138instance on providers, which it recognize as its own. 139A typical example is the MBR disk-partition class which will look for 140the MBR table in the first sector and if found and validated it will 141instantiate a geom to multiplex according to the contents of the MBR. 142.Pp 143A new class will be offered to all existing providers in turn and a new 144provider will be offered to all classes in turn. 145.Pp 146Exactly what a class does to recognize if it should accept the offered 147provider is not defined by GEOM, but the sensible set of options are: 148.Bl -bullet 149.It 150Examine specific data structures on the disk. 151.It 152Examine properties like sectorsize or mediasize for the provider. 153.It 154Examine the rank number of the provider's geom. 155.It 156Examine the method name of the provider's geom. 157.El 158.Pp 159.Em ORPHANIZATION 160is the process by which a provider is removed while 161it potentially is still being used. 162.Pp 163When a geom orphans a provider, all future I/O requests will 164"bounce" on the provider with an error code set by the geom. Any 165consumers attached to the provider will receive notification about 166the orphanization when the eventloop gets around to it, and they 167can take appropriate action at that time. 168.Pp 169A geom which came into being as a result of a normal taste operation 170should selfdestruct unless it has a way to keep functioning lacking 171the orphaned provider. 172Geoms like diskslicers should therefore selfdestruct whereas 173RAID5 or mirror geoms will be able to continue, as long as they do 174not loose quorum. 175.Pp 176When a provider is orphaned, this does not necessarily result in any 177immediate change in the topology: any attached consumers are still 178attached, any opened paths are still open, any outstanding I/O 179requests are still outstanding. 180.Pp 181The typical scenario is 182.Bl -bullet -offset indent -compact 183.It 184A device driver detects a disk has departed and orphans the provider for it. 185.It 186The geoms on top of the disk receive the orphanization event and 187orphans all their providers in turn. 188Providers, which are not attached to, will typically self-destruct 189right away. 190This process continues in a quasi-recursive fashion until all 191relevant pieces of the tree has heard the bad news. 192.It 193Eventually the buck stops when it reaches geom_dev at the top 194of the stack. 195.It 196Geom_dev will call destroy_dev(9) to stop any more request from 197coming in. 198It will sleep until all (if any) outstanding I/O requests have 199been returned. 200It will explicitly close (ie: zero the access counts), a change 201which will propagate all the way down through the mesh. 202It will then detach and destroy its geom. 203.It 204The geom whose provider is now attached will destroy the provider, 205detach and destroy its consumer and destroy its geom. 206.It 207This process percolates all the way down through the mesh, until 208the cleanup is complete. 209.El 210.Pp 211While this approach seems byzantine, it does provide the maximum 212flexibility and robustness in handling disappearing devices. 213.Pp 214The one absolutely crucial detail to be aware is that if the 215device driver does not return all I/O requests, the tree will 216not unravel. 217.Pp 218.Em SPOILING 219is a special case of orphanization used to protect 220against stale metadata. 221It is probably easiest to understand spoiling by going through 222an example. 223.Pp 224Imagine a disk, "da0" on top of which a MBR geom provides 225"da0s1" and "da0s2" and on top of "da0s1" a BSD geom provides 226"da0s1a" through "da0s1e", both the MBR and BSD geoms have 227autoconfigured based on data structures on the disk media. 228Now imagine the case where "da0" is opened for writing and those 229data structures are modified or overwritten: Now the geoms would 230be operating on stale metadata unless some notification system 231can inform them otherwise. 232.Pp 233To avoid this situation, when the open of "da0" for write happens, 234all attached consumers are told about this, and geoms like 235MBR and BSD will selfdestruct as a result. 236When "da0" is closed again, it will be offered for tasting again 237and if the data structures for MBR and BSD are still there, new 238geoms will instantiate themselves anew. 239.Pp 240Now for the fine print: 241.Pp 242If any of the paths through the MBR or BSD module were open, they 243would have opened downwards with an exclusive bit rendering it 244impossible to open "da0" for writing in that case and conversely 245the requested exclusive bit would render it impossible to open a 246path through the MBR geom while "da0" is open for writing. 247.Pp 248From this it also follows that changing the size of open geoms can 249only be done with their cooperation. 250.Pp 251Finally: the spoiling only happens when the write count goes from 252zero to non-zero and the retasting only when the write count goes 253from non-zero to zero. 254.Pp 255.Em INSERT/DELETE 256are a very special operation which allows a new geom 257to be instantiated between a consumer and a provider attached to 258each other and to remove it again. 259.Pp 260To understand the utility of this, imagine a provider with 261being mounted as a file system. 262Between the DEVFS geoms consumer and its provider we insert 263a mirror module which configures itself with one mirror 264copy and consequently is transparent to the I/O requests 265on the path. 266We can now configure yet a mirror copy on the mirror geom, 267request a synchronization, and finally drop the first mirror 268copy. 269We have now in essence moved a mounted file system from one 270disk to another while it was being used. 271At this point the mirror geom can be deleted from the path 272again, it has served its purpose. 273.Pp 274.Em CONFIGURE 275is the process where the administrator issues instructions 276for a particular class to instantiate itself. There are multiple 277ways to express intent in this case, a particular provider can be 278specified with a level of override forcing for instance a BSD 279disklabel module to attach to a provider which was not found palatable 280during the TASTE operation. 281.Pp 282Finally IO is the reason we even do this: it concerns itself with 283sending I/O requests through the graph. 284.Pp 285.Em "I/O REQUESTS 286represented by struct bio, originate at a consumer, 287are scheduled on its attached provider, and when processed, returned 288to the consumer. 289It is important to realize that the struct bio which 290enters through the provider of a particular geom does not "come 291out on the other side". 292Even simple transformations like MBR and BSD will clone the 293struct bio, modify the clone, and schedule the clone on their 294own consumer. 295Note that cloning the struct bio does not involve cloning the 296actual data area specified in the IO request. 297.Pp 298In total four different IO requests exist in GEOM: read, write, 299delete, and get attribute. 300.Pp 301Read and write are self explanatory. 302.Pp 303Delete indicates that a certain range of data is no longer used 304and that it can be erased or freed as the underlying technology 305supports. 306Technologies like flash adaptation layers can arrange to erase 307the relevant blocks before they will become reassigned and 308cryptographic devices may want to fill random bits into the 309range to reduce the amount of data available for attack. 310.Pp 311It is important to recognize that a delete indication is not a 312request and consequently there is no guarantee that the data actually 313will be erased or made unavailable unless guaranteed by specific 314geoms in the graph. If "secure delete" semantics are required, a 315geom should be pushed which converts delete indications into (a 316sequence of) write requests. 317.Pp 318Get attribute supports inspection and manipulation 319of out-of-band attributes on a particular provider or path. 320Attributes are named by ascii strings and they will be discussed in 321a separate section below. 322.Pp 323(stay tuned while the author rests his brain and fingers: more to come.) 324.Sh HISTORY 325This software was developed for the FreeBSD Project by Poul-Henning Kamp 326and NAI Labs, the Security Research Division of Network Associates, Inc. 327under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the 328DARPA CHATS research program. 329.Pp 330The first precursor for GEOM was a gruesome hack to Minix 1.2 and was 331never distributed. An earlier attempt to implement a less general scheme 332in FreeBSD never succeeded. 333.Sh AUTHORS 334.An "Poul-Henning Kamp" Aq phk@FreeBSD.org 335