tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1============
2dm-integrity
3============
4
5The dm-integrity target emulates a block device that has additional
6per-sector tags that can be used for storing integrity information.
7
8A general problem with storing integrity tags with every sector is that
9writing the sector and the integrity tag must be atomic - i.e. in case of
10crash, either both sector and integrity tag or none of them is written.
11
12To guarantee write atomicity, the dm-integrity target uses journal, it
13writes sector data and integrity tags into a journal, commits the journal
14and then copies the data and integrity tags to their respective location.
15
16The dm-integrity target can be used with the dm-crypt target - in this
17situation the dm-crypt target creates the integrity data and passes them
18to the dm-integrity target via bio_integrity_payload attached to the bio.
19In this mode, the dm-crypt and dm-integrity targets provide authenticated
20disk encryption - if the attacker modifies the encrypted device, an I/O
21error is returned instead of random data.
22
23The dm-integrity target can also be used as a standalone target, in this
24mode it calculates and verifies the integrity tag internally. In this
25mode, the dm-integrity target can be used to detect silent data
26corruption on the disk or in the I/O path.
27
28There's an alternate mode of operation where dm-integrity uses a bitmap
29instead of a journal. If a bit in the bitmap is 1, the corresponding
30region's data and integrity tags are not synchronized - if the machine
31crashes, the unsynchronized regions will be recalculated. The bitmap mode
32is faster than the journal mode, because we don't have to write the data
33twice, but it is also less reliable, because if data corruption happens
34when the machine crashes, it may not be detected.
35
36When loading the target for the first time, the kernel driver will format
37the device. But it will only format the device if the superblock contains
38zeroes. If the superblock is neither valid nor zeroed, the dm-integrity
39target can't be loaded.
40
41Accesses to the on-disk metadata area containing checksums (aka tags) are
42buffered using dm-bufio. When an access to any given metadata area
43occurs, each unique metadata area gets its own buffer(s). The buffer size
44is capped at the size of the metadata area, but may be smaller, thereby
45requiring multiple buffers to represent the full metadata area. A smaller
46buffer size will produce a smaller resulting read/write operation to the
47metadata area for small reads/writes. The metadata is still read even in
48a full write to the data covered by a single buffer.
49
50To use the target for the first time:
51
521. overwrite the superblock with zeroes
532. load the dm-integrity target with one-sector size, the kernel driver
54 will format the device
553. unload the dm-integrity target
564. read the "provided_data_sectors" value from the superblock
575. load the dm-integrity target with the target size
58 "provided_data_sectors"
596. if you want to use dm-integrity with dm-crypt, load the dm-crypt target
60 with the size "provided_data_sectors"
61
62
63Target arguments:
64
651. the underlying block device
66
672. the number of reserved sector at the beginning of the device - the
68 dm-integrity won't read of write these sectors
69
703. the size of the integrity tag (if "-" is used, the size is taken from
71 the internal-hash algorithm)
72
734. mode:
74
75 D - direct writes (without journal)
76 in this mode, journaling is
77 not used and data sectors and integrity tags are written
78 separately. In case of crash, it is possible that the data
79 and integrity tag doesn't match.
80 J - journaled writes
81 data and integrity tags are written to the
82 journal and atomicity is guaranteed. In case of crash,
83 either both data and tag or none of them are written. The
84 journaled mode degrades write throughput twice because the
85 data have to be written twice.
86 B - bitmap mode - data and metadata are written without any
87 synchronization, the driver maintains a bitmap of dirty
88 regions where data and metadata don't match. This mode can
89 only be used with internal hash.
90 R - recovery mode - in this mode, journal is not replayed,
91 checksums are not checked and writes to the device are not
92 allowed. This mode is useful for data recovery if the
93 device cannot be activated in any of the other standard
94 modes.
95 I - inline mode - in this mode, dm-integrity will store integrity
96 data directly in the underlying device sectors.
97 The underlying device must have an integrity profile that
98 allows storing user integrity data and provides enough
99 space for the selected integrity tag.
100
1015. the number of additional arguments
102
103Additional arguments:
104
105journal_sectors:number
106 The size of journal, this argument is used only if formatting the
107 device. If the device is already formatted, the value from the
108 superblock is used.
109
110interleave_sectors:number (default 32768)
111 The number of interleaved sectors. This values is rounded down to
112 a power of two. If the device is already formatted, the value from
113 the superblock is used.
114
115meta_device:device
116 Don't interleave the data and metadata on the device. Use a
117 separate device for metadata.
118
119buffer_sectors:number (default 128)
120 The number of sectors in one metadata buffer. The value is rounded
121 down to a power of two.
122
123journal_watermark:number (default 50)
124 The journal watermark in percents. When the size of the journal
125 exceeds this watermark, the thread that flushes the journal will
126 be started.
127
128commit_time:number (default 10000)
129 Commit time in milliseconds. When this time passes, the journal is
130 written. The journal is also written immediately if the FLUSH
131 request is received.
132
133internal_hash:algorithm(:key) (the key is optional)
134 Use internal hash or crc.
135 When this argument is used, the dm-integrity target won't accept
136 integrity tags from the upper target, but it will automatically
137 generate and verify the integrity tags.
138
139 You can use a crc algorithm (such as crc32), then integrity target
140 will protect the data against accidental corruption.
141 You can also use a hmac algorithm (for example
142 "hmac(sha256):0123456789abcdef"), in this mode it will provide
143 cryptographic authentication of the data without encryption.
144
145 When this argument is not used, the integrity tags are accepted
146 from an upper layer target, such as dm-crypt. The upper layer
147 target should check the validity of the integrity tags.
148
149recalculate
150 Recalculate the integrity tags automatically. It is only valid
151 when using internal hash.
152
153journal_crypt:algorithm(:key) (the key is optional)
154 Encrypt the journal using given algorithm to make sure that the
155 attacker can't read the journal. You can use a block cipher here
156 (such as "cbc(aes)") or a stream cipher (for example "chacha20"
157 or "ctr(aes)").
158
159 The journal contains history of last writes to the block device,
160 an attacker reading the journal could see the last sector numbers
161 that were written. From the sector numbers, the attacker can infer
162 the size of files that were written. To protect against this
163 situation, you can encrypt the journal.
164
165journal_mac:algorithm(:key) (the key is optional)
166 Protect sector numbers in the journal from accidental or malicious
167 modification. To protect against accidental modification, use a
168 crc algorithm, to protect against malicious modification, use a
169 hmac algorithm with a key.
170
171 This option is not needed when using internal-hash because in this
172 mode, the integrity of journal entries is checked when replaying
173 the journal. Thus, modified sector number would be detected at
174 this stage.
175
176block_size:number (default 512)
177 The size of a data block in bytes. The larger the block size the
178 less overhead there is for per-block integrity metadata.
179 Supported values are 512, 1024, 2048 and 4096 bytes.
180
181sectors_per_bit:number
182 In the bitmap mode, this parameter specifies the number of
183 512-byte sectors that corresponds to one bitmap bit.
184
185bitmap_flush_interval:number
186 The bitmap flush interval in milliseconds. The metadata buffers
187 are synchronized when this interval expires.
188
189allow_discards
190 Allow block discard requests (a.k.a. TRIM) for the integrity device.
191 Discards are only allowed to devices using internal hash.
192
193fix_padding
194 Use a smaller padding of the tag area that is more
195 space-efficient. If this option is not present, large padding is
196 used - that is for compatibility with older kernels.
197
198fix_hmac
199 Improve security of internal_hash and journal_mac:
200
201 - the section number is mixed to the mac, so that an attacker can't
202 copy sectors from one journal section to another journal section
203 - the superblock is protected by journal_mac
204 - a 16-byte salt stored in the superblock is mixed to the mac, so
205 that the attacker can't detect that two disks have the same hmac
206 key and also to disallow the attacker to move sectors from one
207 disk to another
208
209legacy_recalculate
210 Allow recalculating of volumes with HMAC keys. This is disabled by
211 default for security reasons - an attacker could modify the volume,
212 set recalc_sector to zero, and the kernel would not detect the
213 modification.
214
215The journal mode (D/J), buffer_sectors, journal_watermark, commit_time and
216allow_discards can be changed when reloading the target (load an inactive
217table and swap the tables with suspend and resume). The other arguments
218should not be changed when reloading the target because the layout of disk
219data depend on them and the reloaded target would be non-functional.
220
221For example, on a device using the default interleave_sectors of 32768, a
222block_size of 512, and an internal_hash of crc32c with a tag size of 4
223bytes, it will take 128 KiB of tags to track a full data area, requiring
224256 sectors of metadata per data area. With the default buffer_sectors of
225128, that means there will be 2 buffers per metadata area, or 2 buffers
226per 16 MiB of data.
227
228Status line:
229
2301. the number of integrity mismatches
2312. provided data sectors - that is the number of sectors that the user
232 could use
2333. the current recalculating position (or '-' if we didn't recalculate)
234
235
236The layout of the formatted block device:
237
238* reserved sectors
239 (they are not used by this target, they can be used for
240 storing LUKS metadata or for other purpose), the size of the reserved
241 area is specified in the target arguments
242
243* superblock (4kiB)
244 * magic string - identifies that the device was formatted
245 * version
246 * log2(interleave sectors)
247 * integrity tag size
248 * the number of journal sections
249 * provided data sectors - the number of sectors that this target
250 provides (i.e. the size of the device minus the size of all
251 metadata and padding). The user of this target should not send
252 bios that access data beyond the "provided data sectors" limit.
253 * flags
254 SB_FLAG_HAVE_JOURNAL_MAC
255 - a flag is set if journal_mac is used
256 SB_FLAG_RECALCULATING
257 - recalculating is in progress
258 SB_FLAG_DIRTY_BITMAP
259 - journal area contains the bitmap of dirty
260 blocks
261 * log2(sectors per block)
262 * a position where recalculating finished
263* journal
264 The journal is divided into sections, each section contains:
265
266 * metadata area (4kiB), it contains journal entries
267
268 - every journal entry contains:
269
270 * logical sector (specifies where the data and tag should
271 be written)
272 * last 8 bytes of data
273 * integrity tag (the size is specified in the superblock)
274
275 - every metadata sector ends with
276
277 * mac (8-bytes), all the macs in 8 metadata sectors form a
278 64-byte value. It is used to store hmac of sector
279 numbers in the journal section, to protect against a
280 possibility that the attacker tampers with sector
281 numbers in the journal.
282 * commit id
283
284 * data area (the size is variable; it depends on how many journal
285 entries fit into the metadata area)
286
287 - every sector in the data area contains:
288
289 * data (504 bytes of data, the last 8 bytes are stored in
290 the journal entry)
291 * commit id
292
293 To test if the whole journal section was written correctly, every
294 512-byte sector of the journal ends with 8-byte commit id. If the
295 commit id matches on all sectors in a journal section, then it is
296 assumed that the section was written correctly. If the commit id
297 doesn't match, the section was written partially and it should not
298 be replayed.
299
300* one or more runs of interleaved tags and data.
301 Each run contains:
302
303 * tag area - it contains integrity tags. There is one tag for each
304 sector in the data area. The size of this area is always 4KiB or
305 greater.
306 * data area - it contains data sectors. The number of data sectors
307 in one run must be a power of two. log2 of this value is stored
308 in the superblock.