1 |
Hi, everyone. |
2 |
|
3 |
Here's the updated version of GLEP 74 taking into consideration |
4 |
the points made during the Council pre-review. |
5 |
|
6 |
ReST: https://dev.gentoo.org/~mgorny/tmp/glep-0074.rst |
7 |
HTML: https://dev.gentoo.org/~mgorny/tmp/glep-0074.html |
8 |
|
9 |
Changes: |
10 |
|
11 |
09ed01f glep-0074: Explain combining multiple Manifest trees |
12 |
9de0840 glep-0074: Clarify timestamp handling of sub-Manifests |
13 |
516c2ec glep-0074: Forbid compressing top-level Manifest |
14 |
b01783e glep-0074: Clarify sub-Manifest signing paragraph |
15 |
|
16 |
|
17 |
--- |
18 |
GLEP: 74 |
19 |
Title: Full-tree verification using Manifest files |
20 |
Author: Michał Górny <mgorny@g.o>, |
21 |
Robin Hugh Johnson <robbat2@g.o>, |
22 |
Ulrich Müller <ulm@g.o> |
23 |
Type: Standards Track |
24 |
Status: Draft |
25 |
Version: 1 |
26 |
Created: 2017-10-21 |
27 |
Last-Modified: 2017-11-16 |
28 |
Post-History: 2017-10-26, 2017-11-16 |
29 |
Content-Type: text/x-rst |
30 |
Requires: 59, 61 |
31 |
Replaces: 44, 58, 60 |
32 |
--- |
33 |
|
34 |
Abstract |
35 |
======== |
36 |
|
37 |
This GLEP extends the Manifest file format to cover full-tree file |
38 |
integrity and authenticity checks.The format aims to be future-proof, |
39 |
efficient and provide means of backwards compatibility. |
40 |
|
41 |
|
42 |
Motivation |
43 |
========== |
44 |
|
45 |
The Manifest files as defined by GLEP 44 [#GLEP44]_ provide the current |
46 |
means of verifying the integrity of distfiles and package files |
47 |
in Gentoo. Combined with OpenPGP signatures, they provide means to |
48 |
ensure the authenticity of the covered files. However, as noted |
49 |
in GLEP 57 [#GLEP57]_ they lack the ability to provide full-tree |
50 |
authenticity verification as they do not cover any files outside |
51 |
the package directory. In particular, they provide multiple ways |
52 |
for a third party to inject malicious code into the ebuild environment. |
53 |
|
54 |
Historically, the topic of providing authenticity coverage for the whole |
55 |
repository has been mentioned multiple times. The most noteworthy effort |
56 |
are GLEPs 58 [#GLEP58]_ and 60 [#GLEP60]_ by Robin H. Johnson from 2008. |
57 |
They were accepted by the Council in 2010 but have never been |
58 |
implemented. When potential implementation work started in 2017, a new |
59 |
discussion about the specification arose. It prompted the creation |
60 |
of a competing GLEP that would provide a redesigned alternative to |
61 |
the old GLEPs. |
62 |
|
63 |
This specification is designed with the following goals in mind: |
64 |
|
65 |
1. It should provide means to ensure the authenticity of the complete |
66 |
repository, including preventing the injection of additional files. |
67 |
|
68 |
2. The format should be universal enough to work both for the Gentoo |
69 |
repository and third-party repositories of different characteristics. |
70 |
|
71 |
3. The Manifest files should be verifiable stand-alone, that is without |
72 |
knowing any details about the underlying repository format. |
73 |
|
74 |
|
75 |
Specification |
76 |
============= |
77 |
|
78 |
Manifest file format |
79 |
-------------------- |
80 |
|
81 |
This specification reuses and extends the Manifest file format defined |
82 |
in GLEP 44 [#GLEP44]_. For the purpose of it, the *file type* field is |
83 |
repurposed as a generic *tag* that could also indicate additional |
84 |
(non-checksum) metadata. Appropriately, those tags can be followed by |
85 |
other space-separated values. |
86 |
|
87 |
Unless specified otherwise, the paths used in the Manifest files |
88 |
are relative to the directory containing the Manifest file. The paths |
89 |
must not reference the parent directory (``..``). |
90 |
|
91 |
|
92 |
Manifest file locations and nesting |
93 |
----------------------------------- |
94 |
|
95 |
The ``Manifest`` file located in the root directory of the repository |
96 |
is called top-level Manifest, and it is used to perform the full-tree |
97 |
verification. In order to verify the authenticity, it must be signed |
98 |
using OpenPGP, using the armored cleartext format. |
99 |
|
100 |
The top-level Manifest may reference sub-Manifests contained |
101 |
in subdirectories of the repository. The sub-Manifests are traditionally |
102 |
named ``Manifest``; however, the implementation must support arbitrary |
103 |
names, including the possibility of multiple (split) Manifests |
104 |
for a single directory. The sub-Manifest can only cover the files inside |
105 |
the directory tree where it resides. |
106 |
|
107 |
The sub-Manifest can also be signed using OpenPGP armored cleartext |
108 |
format. However, the signature verification can be omitted since it |
109 |
already is covered by the signed top-level Manifest. |
110 |
|
111 |
|
112 |
Directory tree coverage |
113 |
----------------------- |
114 |
|
115 |
The specification provides three ways of skipping Manifest verification |
116 |
of specific files and directories (recursively): |
117 |
|
118 |
1. explicit ``IGNORE`` entries in Manifest files, |
119 |
|
120 |
2. injected ignore paths via package manager configuration, |
121 |
|
122 |
3. using names starting with a dot (``.``) which are always skipped. |
123 |
|
124 |
All files that are not ignored must be covered by at least one |
125 |
of the Manifests. |
126 |
|
127 |
A single file may be matched by multiple identical or equivalent |
128 |
Manifest entries, if and only if the entries have the same semantics, |
129 |
specify the same size and the checksums common to both entries match. |
130 |
It is an error for a single file to be matched by multiple entries |
131 |
of different semantics, file size or checksum values. It is an error |
132 |
to specify another entry for a file matching ``IGNORE``, or one of its |
133 |
subdirectories. |
134 |
|
135 |
The file entries (except for ``IGNORE``) can be specified for regular |
136 |
files only. Symbolic links are followed when opening files |
137 |
and traversing directories. It is an error to specify an entry for |
138 |
a different file type. If the tree contain files of other types |
139 |
that are not otherwise ignored, they need to be covered by an explicit |
140 |
``IGNORE``. |
141 |
|
142 |
All the local (non-``DIST``) files covered by a Manifest tree must |
143 |
reside on the same filesystem. It is an error to specify entries |
144 |
applying to files on another filesystem. If files or directories that |
145 |
are not otherwise ignored reside on a different filesystem, or symbolic |
146 |
links point to targets on a different filesystem, they must |
147 |
be explicitly excluded via ``IGNORE``. |
148 |
|
149 |
|
150 |
File verification |
151 |
----------------- |
152 |
|
153 |
When verifying a file against the Manifest, the following rules are |
154 |
used: |
155 |
|
156 |
1. If the file is covered directly or indirectly by an entry |
157 |
of the ``IGNORE`` type, the verification always succeeds. |
158 |
|
159 |
2. If the file is covered by an entry of the ``MANIFEST``, ``DATA``, |
160 |
``MISC``, ``EBUILD`` or ``AUX`` type: |
161 |
|
162 |
a. if the file is not present, then the verification fails, |
163 |
|
164 |
b. if the file is present but has a different size or one |
165 |
of the checksums does not match, the verification fails, |
166 |
|
167 |
c. otherwise, the verification succeeds. |
168 |
|
169 |
3. If the file is present but not listed in Manifest, the verification |
170 |
fails. |
171 |
|
172 |
Unless specified otherwise, the package manager must not allow using |
173 |
any files for which the verification failed. The package manager may |
174 |
reject any package or even the whole repository if it may refer to files |
175 |
for which the verification failed. |
176 |
|
177 |
|
178 |
Timestamp verification |
179 |
---------------------- |
180 |
|
181 |
The top-level Manifest file can contain a ``TIMESTAMP`` entry to account |
182 |
for attacks against tree update distribution. If such an entry |
183 |
is present, it should be updated every time at least one |
184 |
of the Manifests changes. Every unique timestamp value must correspond |
185 |
to a single tree state. |
186 |
|
187 |
During the verification process, the client should compare the timestamp |
188 |
against the update time obtained from a local clock or a trusted time |
189 |
source. If the comparison result indicates that the Manifest at the time |
190 |
of receiving was already significantly outdated, the client should |
191 |
either fail the verification or require manual confirmation from user. |
192 |
|
193 |
Furthermore, the Manifest provider may employ additional methods |
194 |
of distributing the timestamps of recently generated Manifests |
195 |
using a secure channel from a trusted source for exact comparison. |
196 |
The exact details of such a solution are outside the scope of this |
197 |
specification. |
198 |
|
199 |
``TIMESTAMP`` entries may also be present in sub-Manifests. Those |
200 |
timestamps must not be newer than the timestamp of the top-level |
201 |
Manifest (if present). This specification does not define any specific |
202 |
use for them. |
203 |
|
204 |
|
205 |
Modern Manifest tags |
206 |
-------------------- |
207 |
|
208 |
The Manifest files can specify the following tags: |
209 |
|
210 |
``TIMESTAMP <iso8601>`` |
211 |
Specifies a timestamp of when the Manifest file was last updated. |
212 |
The timestamp must be a valid second-precision ISO8601 extended format |
213 |
combined date and time in UTC timezone, i.e. using the following |
214 |
``strftime()`` format string: ``%Y-%m-%dT%H:%M:%SZ``. Optional. |
215 |
The package manager can use it to detect an outdated repository |
216 |
checkout as described in `Timestamp verification`_. |
217 |
|
218 |
``MANIFEST <path> <size> <checksums>...`` |
219 |
Specifies a sub-Manifest. The sub-Manifest must be verified like |
220 |
a regular file. If the verification succeeds, the entries from |
221 |
the sub-Manifest are included for verification as described |
222 |
in `Manifest file locations and nesting`_. |
223 |
|
224 |
``IGNORE <path>`` |
225 |
Ignores a subdirectory or file from Manifest checks. If the specified |
226 |
path is present, it and its contents are omitted from the Manifest |
227 |
verification (always pass). *Path* must be a plain file or directory |
228 |
path without a trailing slash, and must not contain wildcards. |
229 |
|
230 |
``DATA <path> <size> <checksums>...`` |
231 |
Specifies a regular file subject to Manifest verification. The file |
232 |
is required to pass verification. Used for all files that do not match |
233 |
any other type. |
234 |
|
235 |
``DIST <filename> <size> <checksums>...`` |
236 |
Specifies a distfile entry used to verify files fetched as part |
237 |
of ``SRC_URI``. The filename must match the filename used to store |
238 |
the fetched file as specified in the PMS [#PMS-FETCH]_. The package |
239 |
manager must reject the fetched file if it fails verification. |
240 |
``DIST`` entries apply to all packages below the Manifest file |
241 |
specifying them. |
242 |
|
243 |
|
244 |
Deprecated Manifest tags |
245 |
------------------------ |
246 |
|
247 |
For backwards compatibility, the following tags are additionally |
248 |
allowed at the package directory level: |
249 |
|
250 |
``EBUILD <filename> <size> <checksums>...`` |
251 |
Equivalent to the ``DATA`` type. |
252 |
|
253 |
``MISC <path> <size> <checksums>...`` |
254 |
Equivalent to the ``DATA`` type. Historically indicated that |
255 |
the package manager may ignore a verification failure if operating |
256 |
in non-strict mode. However, that behavior is deprecated. |
257 |
|
258 |
``AUX <filename> <size> <checksums>...`` |
259 |
Equivalent to the ``DATA`` type, except that the filename is relative |
260 |
to ``files/`` subdirectory. |
261 |
|
262 |
|
263 |
Algorithm for full-tree verification |
264 |
------------------------------------ |
265 |
|
266 |
In order to perform full-tree verification, the following algorithm |
267 |
can be used: |
268 |
|
269 |
1. Collect all files present in the repository into *present* set. |
270 |
|
271 |
2. Start at the top-level Manifest file. Verify its OpenPGP signature. |
272 |
Optionally verify the ``TIMESTAMP`` entry if present as specified |
273 |
in `timestamp verification`. Remove the top-level Manifest |
274 |
from the *present* set. |
275 |
|
276 |
3. Process all ``MANIFEST`` entries, recursively. Verify the Manifest |
277 |
files according to `file verification`_ section, and include their |
278 |
entries in the current Manifest entry list (using paths relative |
279 |
to directories containing the Manifests). |
280 |
|
281 |
4. Process all ``IGNORE`` entries. Remove any paths matching them |
282 |
from the *present* set. |
283 |
|
284 |
5. Collect all files covered by ``DATA``, ``MISC``, ``EBUILD`` |
285 |
and ``AUX`` entries into the *covered* set. |
286 |
|
287 |
6. Verify the entries in *covered* set for incompatible duplicates |
288 |
and collisions with ignored files as explained in `Manifest file |
289 |
locations and nesting`_. |
290 |
|
291 |
7. Verify all the files in the union of the *present* and *covered* |
292 |
sets, according to `file verification`_ section. |
293 |
|
294 |
|
295 |
Algorithm for finding parent Manifests |
296 |
-------------------------------------- |
297 |
|
298 |
In order to find the top-level Manifest from the current directory |
299 |
the following algorithm can be used: |
300 |
|
301 |
1. Store the current directory as *original* and the device ID |
302 |
of the containing filesystem (``st_dev``) as *startdev*, |
303 |
|
304 |
2. If the device ID of the containing filesystem (``st_dev``) |
305 |
of the current directory is different than *startdev*, stop. |
306 |
|
307 |
3. If the current directory contains a ``Manifest`` file: |
308 |
|
309 |
a. If a ``IGNORE`` entry in the ``Manifest`` file covers |
310 |
the *original* directory (or one of the parent directories), stop. |
311 |
|
312 |
b. Otherwise, store the current directory as *last_found*. |
313 |
|
314 |
4. If the current directory is the root system directory (``/``), stop. |
315 |
|
316 |
5. Otherwise, enter the parent directory and jump to step 2. |
317 |
|
318 |
Once the algorithm stops, *last_found* will contain the relevant |
319 |
top-level Manifest. If *last_found* is null, then the directory tree |
320 |
does not contain any valid top-level Manifest candidates and one should |
321 |
be created in the *original* directory. |
322 |
|
323 |
Once the top-level Manifest is found, its ``MANIFEST`` entries should |
324 |
be used to find any sub-Manifests below the top-level Manifest, |
325 |
up to and including the *original* directory. Note that those |
326 |
sub-Manifests can use different filenames than ``Manifest``. |
327 |
|
328 |
|
329 |
Checksum algorithms |
330 |
------------------- |
331 |
|
332 |
This section is informational only. Specifying the exact set |
333 |
of supported algorithms is outside the scope of this specification. |
334 |
|
335 |
The algorithm names reserved at the time of writing are: |
336 |
|
337 |
- ``MD5`` [#MD5]_, |
338 |
- ``RMD160`` -- RIPEMD-160 [#RIPEMD160]_, |
339 |
- ``SHA1`` [#SHS]_, |
340 |
- ``SHA256`` and ``SHA512`` -- SHA-2 family of hashes [#SHS]_, |
341 |
- ``WHIRLPOOL`` [#WHIRLPOOL]_, |
342 |
- ``BLAKE2B`` and ``BLAKE2S`` -- BLAKE2 family of hashes [#BLAKE2]_, |
343 |
- ``SHA3_256`` and ``SHA3_512`` -- SHA-3 family of hashes [#SHA3]_, |
344 |
- ``STREEBOG256`` and ``STREEBOG512`` -- Streebog family of hashes |
345 |
[#STREEBOG]_. |
346 |
|
347 |
The method of introducing new hashes is defined by GLEP 59 [#GLEP59]_. |
348 |
It is recommended that any new hashes are named after the Python |
349 |
``hashlib`` module algorithm names, transformed into uppercase. |
350 |
|
351 |
|
352 |
Manifest compression |
353 |
-------------------- |
354 |
|
355 |
The topic of Manifest file compression is covered by GLEP 61 [#GLEP61]_. |
356 |
This section merely addresses interoperability issues between Manifest |
357 |
compression and this specification. |
358 |
|
359 |
The compressed Manifest files are required to be suffixed for their |
360 |
compression algorithm. This suffix should be used to recognize |
361 |
the compression and decompress Manifests transparently. The exact list |
362 |
of algorithms and their corresponding suffixes are outside the scope |
363 |
of this specification. |
364 |
|
365 |
The top-level Manifest file must not be compressed. Since the OpenPGP |
366 |
signature covers the uncompressed text and is compressed itself, |
367 |
the data would have to be decompressed without any prior verification. |
368 |
This could expose users e.g. to zip bombs or exploits on decompressor |
369 |
vulnerabilities. |
370 |
|
371 |
Whenever this specification refers to sub-Manifests, they can use any |
372 |
names but are also required to use a specific compression suffix. |
373 |
The ``MANIFEST`` entries are required to specify the full name including |
374 |
compression suffix, and the verification is performed on the compressed |
375 |
file. |
376 |
|
377 |
The specification permits uncompressed Manifests to exist alongside |
378 |
their compressed counterparts, and multiple compressed formats |
379 |
to coexist. If that is the case, the files must have the same |
380 |
uncompressed content and the specification is free to choose either |
381 |
of the files using the same base name. |
382 |
|
383 |
|
384 |
Combining multiple Manifest trees (informational) |
385 |
------------------------------------------------- |
386 |
|
387 |
This specification permits nesting multiple hierarchical Manifest trees. |
388 |
In this layout, the specific directories of the Manifest tree can |
389 |
be verified both as a part of another top-level Manifest, |
390 |
and as an independent Manifest tree (when obtained without the parent |
391 |
directory). |
392 |
|
393 |
For this to work, the sub-Manifest file in the directory must also |
394 |
satisfy the requirements for the top-level Manifest file. That is: |
395 |
|
396 |
- it must be named ``Manifest`` and not compressed, |
397 |
|
398 |
- it must cover all the files in this directory and its subdirectories |
399 |
(i.e. no files from the directory tree can be covered by parent |
400 |
Manifest), |
401 |
|
402 |
- if authenticity verification is desired, it must be OpenPGP-signed. |
403 |
|
404 |
It should be noted that if such a directory is a subdirectory of a valid |
405 |
Manifest tree, the sub-Manifest needs to be valid according |
406 |
to the top-level Manifest and the OpenPGP signature is disregarded |
407 |
as detailed in `Manifest file locations and nesting`_. The top-level |
408 |
behavior is exhibited only when the directory is obtained without parent |
409 |
directories. |
410 |
|
411 |
|
412 |
An example Manifest file (informational) |
413 |
---------------------------------------- |
414 |
|
415 |
An example top-level Manifest file for the Gentoo repository would have |
416 |
the following content:: |
417 |
|
418 |
TIMESTAMP 2017-10-30T10:11:12Z |
419 |
IGNORE distfiles |
420 |
IGNORE local |
421 |
IGNORE lost+found |
422 |
IGNORE packages |
423 |
MANIFEST app-accessibility/Manifest 14821 SHA256 1b5f.. SHA512 f7eb.. |
424 |
... |
425 |
MANIFEST eclass/Manifest.gz 50812 SHA256 8c55.. SHA512 2915.. |
426 |
... |
427 |
|
428 |
An example modern Manifest (disregarding backwards compatibility) |
429 |
for a package directory would have the following content:: |
430 |
|
431 |
DATA SphinxTrain-0.9.1-r1.ebuild 932 SHA256 3d3b.. SHA512 be4d.. |
432 |
DATA SphinxTrain-1.0.8.ebuild 912 SHA256 f681.. SHA512 0749.. |
433 |
DATA metadata.xml 664 SHA256 97c6.. SHA512 1175.. |
434 |
DATA files/gcc.patch 816 SHA256 b56e.. SHA512 2468.. |
435 |
DATA files/gcc34.patch 333 SHA256 c107.. SHA512 9919.. |
436 |
DIST SphinxTrain-0.9.1-beta.tar.gz 469617 SHA256 c1a4.. SHA512 1b33.. |
437 |
DIST sphinxtrain-1.0.8.tar.gz 8925803 SHA256 548e.. SHA512 465d.. |
438 |
|
439 |
|
440 |
Rationale |
441 |
========= |
442 |
|
443 |
Stand-alone format |
444 |
------------------ |
445 |
|
446 |
The first question that needed to be asked before proceeding with |
447 |
the design was whether the Manifest file format was supposed to be |
448 |
stand-alone, or tightly bound to the repository format. |
449 |
|
450 |
The stand-alone format has been selected because of its three |
451 |
advantages: |
452 |
|
453 |
1. It is more future-proof. If an incompatible change to the repository |
454 |
format is introduced, only developers need to be upgrade the tools |
455 |
they use to generate the Manifests. The tools used to verify |
456 |
the updated Manifests will continue to work. |
457 |
|
458 |
2. It is more flexible and universal. With a dedicated tool, |
459 |
the Manifest files can be used to sign and verify arbitrary file |
460 |
sets. |
461 |
|
462 |
3. It keeps the verification tool simpler. In particular, we can easily |
463 |
write an independent verification tool that could work on any |
464 |
distribution without needing to depend on a package manager |
465 |
implementation or rewrite parts of it. |
466 |
|
467 |
Designing a stand-alone format requires that the Manifest carries enough |
468 |
information to perform the verification following all the rules specific |
469 |
to the Gentoo repository. |
470 |
|
471 |
|
472 |
Tree design |
473 |
----------- |
474 |
|
475 |
The second important point of the design was determining whether |
476 |
the Manifest files should be structured hierarchically, or independent. |
477 |
Both options have their advantages. |
478 |
|
479 |
In the hierarchical model, each sub-Manifest file is covered by a higher |
480 |
level Manifest. As a result, only the top-level Manifest has to be |
481 |
OpenPGP-signed, and subsequent Manifests need to be only verified by |
482 |
checksum stored in the parent Manifest. This has the following |
483 |
implications: |
484 |
|
485 |
- Verifying any set of files in the repository requires using checksums |
486 |
from the most relevant Manifests and the parent Manifests. |
487 |
|
488 |
- The OpenPGP signature of the top-level Manifest needs to be verified |
489 |
only once per process. |
490 |
|
491 |
- Altering any set of files requires updating the relevant Manifests, |
492 |
and their parent Manifests up to the top-level Manifest, and signing |
493 |
the last one. |
494 |
|
495 |
- As a result, the top-level Manifest changes on every commit, |
496 |
and various middle-level Manifests change (and need to be transferred) |
497 |
frequently. |
498 |
|
499 |
In the independent model, each sub-Manifest file is independent |
500 |
of the parent Manifests. As a result, each of them needs to be signed |
501 |
and verified independently. However, the parent Manifests still need |
502 |
to list sub-Manifests (albeit without verification data) in order |
503 |
to detect removal or replacement of subdirectories. This has |
504 |
the following implications: |
505 |
|
506 |
- Verifying any set of files in the repository requires using checksums |
507 |
and verifying signatures of the most relevant Manifest files. |
508 |
|
509 |
- Altering any set of files requires updating the relevant Manifests |
510 |
and signing them again. |
511 |
|
512 |
- Parent Manifests are updated only when Manifests are added or removed |
513 |
from subdirectories. As a result, they change infrequently. |
514 |
|
515 |
While both models have their advantages, the hierarchical model was |
516 |
selected because it reduces the number of OpenPGP operations |
517 |
which are comparatively costly to the minimum. |
518 |
|
519 |
|
520 |
Tree layout restrictions |
521 |
------------------------ |
522 |
|
523 |
The algorithm is meant to work primarily with ebuild repositories which |
524 |
normally contain only files and directories. Directories provide |
525 |
no useful metadata for verification, and specifying special entries |
526 |
for additional file types is purposeless. Therefore, the specification |
527 |
is restricted to dealing with regular files. |
528 |
|
529 |
The Gentoo repository does not use symbolic links. Some Gentoo |
530 |
repositories do, however. To provide a simple solution for dealing with |
531 |
symlinks without having to take care to implement special handling for |
532 |
them, the common behavior of implicitly resolving them is used. |
533 |
Therefore, symbolic links to files are stored as if they were regular |
534 |
files, and symbolic links to directories are followed as if they were |
535 |
regular directories. |
536 |
|
537 |
Dotfiles are implicitly ignored as that is a common notion used |
538 |
in software written for POSIX systems. All other filenames require |
539 |
explicit ``IGNORE`` lines. |
540 |
|
541 |
An ability to inject additional ignore entries is provided to account |
542 |
for site configuration affecting the repository tree -- placing |
543 |
additional files in it, skipping some of the categories from syncing. |
544 |
This configuration can extend beyond the limits of this GLEP, |
545 |
e.g. by allowing wildcards or regular expressions. |
546 |
|
547 |
The algorithm is restricted to work on a single filesystem. This is |
548 |
mostly relevant when scanning for top-level Manifest -- we do not want |
549 |
to cross filesystem boundaries then. However, to ensure consistent |
550 |
bidirectional behavior we need to also ban them when operating downwards |
551 |
the tree. |
552 |
|
553 |
The directories and files on different filesystems need to be ignored |
554 |
explicitly as implicitly skipping them would cause confusion. |
555 |
In particular, tools might then claim that a file does not exist when |
556 |
it clearly does because it was skipped due to filesystem boundaries. |
557 |
|
558 |
|
559 |
File verification model |
560 |
----------------------- |
561 |
|
562 |
The verification model aims to provide full coverage against different |
563 |
forms of attack. In particular, three different kinds of manipulation |
564 |
are considered: |
565 |
|
566 |
1. Alteration of the file content. |
567 |
|
568 |
2. Removal of a file. |
569 |
|
570 |
3. Addition of a new file. |
571 |
|
572 |
In order to prevent against all three, the system requires that all |
573 |
files in the repository are listed in Manifests and verified against |
574 |
them. |
575 |
|
576 |
As a special case, ignores are allowed to account for directories |
577 |
that are not part of the repository but were traditionally placed inside |
578 |
it. Those directories were ``distfiles``, ``local`` and ``packages``. It |
579 |
could be also used to ignore VCS directories such as ``CVS``. |
580 |
|
581 |
|
582 |
Non-strict Manifest verification |
583 |
-------------------------------- |
584 |
|
585 |
Originally the Manifest2 format provided a special ``MISC`` tag that |
586 |
was used for ``metadata.xml`` and ``ChangeLog`` files. This tag |
587 |
indicated that the Manifest verification failures could be ignored for |
588 |
those files unless the package manager was working in strict mode. |
589 |
|
590 |
The first versions of this specification continued the use of this tag. |
591 |
However, after a long debate it was decided to deprecate it along with |
592 |
the non-strict behavior, and require all files to strictly match. |
593 |
|
594 |
Two arguments were mentioned for the usefulness of a ``MISC`` type: |
595 |
|
596 |
1. being able to reduce the checkout size by stripping unnecessary |
597 |
files out, and |
598 |
|
599 |
2. being able to run update automatically generated files locally |
600 |
without causing unnecessary verification failures. |
601 |
|
602 |
However, the usefulness of ``MISC`` in both cases is doubtful. |
603 |
|
604 |
The cases for stripping unnecessary files mostly focused around space |
605 |
savings. For this purpose, stripping ``metadata.xml`` and similar files |
606 |
has little value. It is much more common for users to strip whole |
607 |
packages or categories. The ``MISC`` type is not suitable for that, |
608 |
and so a dedicated package manager mechanism needs to be developed |
609 |
instead. The same mechanism can also handle files that historically used |
610 |
the ``MISC`` type. As an example, the package manager may choose |
611 |
to generate both the rsync exclusion list and Manifest ignore list |
612 |
using a single source list. |
613 |
|
614 |
The cases for autogenerated files involve such cache files |
615 |
as ``use.local.desc``. However, we can not include ``md5-cache`` there |
616 |
due to security concerns which results in inconsistent cache handling. |
617 |
Furthermore, the tools were historically modified to provide stable |
618 |
output which means that their content can not change without |
619 |
a non-``MISC`` content being changed first. This practically defeats |
620 |
the purpose of using ``MISC``. |
621 |
|
622 |
Finally, the non-strict mode could be used as means to an attack. |
623 |
The allowance of missing or modified documentation file could be used |
624 |
to spread misinformation, resulting in bad decisions made by the user. |
625 |
A modified file could also be used e.g. to exploit vulnerabilities |
626 |
of an XML parser. |
627 |
|
628 |
|
629 |
Timestamp field |
630 |
--------------- |
631 |
|
632 |
The top-level Manifests optionally allows using a ``TIMESTAMP`` tag |
633 |
to include a generation timestamp in the Manifest. A similar feature |
634 |
was originally proposed in GLEP 58 [#GLEP58]_. |
635 |
|
636 |
A malicious third-party may use the principles of exclusion or replay |
637 |
[#C08]_ to deny an update to clients, while at the same time recording |
638 |
the identity of clients to attack. The timestamp field can be used to |
639 |
detect that. |
640 |
|
641 |
In order to provide a more complete protection, the Gentoo |
642 |
Infrastructure should provide an ability to obtain the timestamps |
643 |
of all Manifests from a recent timeframe over a secure channel |
644 |
from a trusted source for comparison. |
645 |
|
646 |
Strictly speaking, this information is already provided by the various |
647 |
``metadata/timestamp*`` files that are already present. However, |
648 |
including the value in the Manifest itself has a little cost |
649 |
and provides the ability to perform the verification stand-alone. |
650 |
|
651 |
Furthermore, some of the timestamp files are added very late |
652 |
in the distribution process, past the Manifest generation phase. Those |
653 |
files will most likely receive ``IGNORE`` entries and therefore |
654 |
be not suitable to safe use. |
655 |
|
656 |
The specification permits additional timestamps in sub-Manifest files |
657 |
for local use. A generic testing tool should ignore them. |
658 |
|
659 |
|
660 |
New vs deprecated tags |
661 |
---------------------- |
662 |
|
663 |
Out of the four types defined by Manifest2, only one is reused |
664 |
and the remaining three is replaced by a single, universal ``DATA`` |
665 |
type. |
666 |
|
667 |
The ``DIST`` tag is reused since the specification does not change |
668 |
anything with regard to distfile handling. |
669 |
|
670 |
The ``EBUILD`` tag could potentially be reused for generic file |
671 |
verification data. However, it would be confusing if all the different |
672 |
data files were marked as ``EBUILD``. Therefore, an equivalent ``DATA`` |
673 |
type was introduced as a replacement. |
674 |
|
675 |
The ``MISC`` tag and the relevant non-strict mode has been removed |
676 |
as being of little value, as detailed in the `Non-strict Manifest |
677 |
verification`_ section. |
678 |
|
679 |
The ``AUX`` tag is deprecated as it is redundant to ``DATA``, and has |
680 |
the limiting property of implicit ``files/`` path prefix. |
681 |
|
682 |
|
683 |
Finding top-level Manifest |
684 |
-------------------------- |
685 |
|
686 |
The development of a reference implementation for this GLEP has brought |
687 |
the following problem: how to find all the relevant Manifests when |
688 |
the Manifest tool is run inside a subdirectory of the repository? |
689 |
|
690 |
One of the options would be to provide a bi-directional linking |
691 |
of Manifests via a ``PARENT`` tag. However, that would not solve |
692 |
the problem when a new Manifest file is being created. |
693 |
|
694 |
Instead, an algorithm for iterating over parent directories is proposed. |
695 |
Since there is no obligatory explicit indicator for the top-level |
696 |
Manifest, the algorithm assumes that the top-level Manifest |
697 |
is the highest ``Manifest`` in the directory hierarchy that can cover |
698 |
the current directory. This generally makes sense since the Manifest |
699 |
files are required to provide coverage for all subdirectories, so all |
700 |
Manifests starting from that one need to be updated. |
701 |
|
702 |
If independent Manifest trees are nested in the directory structure, |
703 |
then an ``IGNORE`` entry needs to be used to separate them. |
704 |
|
705 |
Since sub-Manifests can use any filenames, the Manifest finding |
706 |
algorithm must not short-cut the procedure by storing all ``Manifest`` |
707 |
files along the parent directories. Instead, it needs to retrace |
708 |
the relevant sub-Manifest files along ``MANIFEST`` entries |
709 |
in the top-level Manifest. |
710 |
|
711 |
|
712 |
Injecting ChangeLogs into the checkout |
713 |
-------------------------------------- |
714 |
|
715 |
One of the problems considered in the new Manifest format was that |
716 |
of injecting historical and autogenerated ChangeLog into the repository. |
717 |
Normally we are not including those files to reduce the checkout size. |
718 |
However, some users have shown interest in them and Infra is working |
719 |
on providing them via an additional rsync module. |
720 |
|
721 |
If such files were injected into the repository, they would cause |
722 |
verification failures of Manifests. To account for this, Infra could |
723 |
provide ``IGNORE`` entries to allow them to exist. |
724 |
|
725 |
|
726 |
Splitting distfile checksums from file checksums |
727 |
------------------------------------------------ |
728 |
|
729 |
Another problem with the current Manifest format is that the checksums |
730 |
for fetched files are combined with checksums for local files |
731 |
in a single file inside the package directory. It has been specifically |
732 |
pointed out that: |
733 |
|
734 |
- since distfiles are sometimes reused across different packages, |
735 |
the repeating checksums are redundant [#DIST]_. |
736 |
|
737 |
- mirror admins were interested in the possibility of verifying all |
738 |
the distfiles with a single tool. |
739 |
|
740 |
This specification does not provide a clean solution to this problem. |
741 |
It technically permits moving ``DIST`` entries to higher-level Manifests |
742 |
but the usefulness of such a solution is doubtful. |
743 |
|
744 |
However, for the second problem we will probably deliver a dedicated |
745 |
tool working with this Manifest format. |
746 |
|
747 |
|
748 |
Hash algorithms |
749 |
--------------- |
750 |
|
751 |
While maintaining a consistent supported hash set is important |
752 |
for interoperability, it is no good fit for the generic layout of this |
753 |
GLEP. Furthermore, it would require updating the GLEP in the future |
754 |
every time the used algorithms change. |
755 |
|
756 |
Instead, the specification focuses on listing the currently used |
757 |
algorithm names for interoperability, and sets a recommendation |
758 |
for consistent naming of algorithms in the future. The Python |
759 |
``hashlib`` module is used as a reference since it is used |
760 |
as the provider of hash functions for most of the Python software, |
761 |
including Portage and PkgCore. |
762 |
|
763 |
The basic rules for changing hash algorithms are defined in GLEP 59 |
764 |
[#GLEP59]_. The implementations can focus only on those algorithms |
765 |
that are actually used or planned on being used. It may be feasible |
766 |
to devise a new GLEP that specifies the currently used hashes (or update |
767 |
GLEP 59 accordingly). |
768 |
|
769 |
|
770 |
Manifest compression |
771 |
-------------------- |
772 |
|
773 |
The support for Manifest compression is introduced with minimal changes |
774 |
to the file format. The ``MANIFEST`` entries are required to provide |
775 |
the real (compressed) file path for compatibility with other file |
776 |
entries and to avoid confusion. |
777 |
|
778 |
The compression of top-level Manifest file has been prohibited |
779 |
as the specification currently does not provide any means of verifying |
780 |
the file prior to decompression. This would make it possibly for |
781 |
a malicious third party to provide a compressed Manifest exposing |
782 |
decompressor vulnerabilities, or being a zip bomb, and the tooling |
783 |
would have to unpack it before being able to verify the contents. |
784 |
|
785 |
The OpenPGP cleartext signature covers the contents of the Manifest, |
786 |
and is therefore compressed along with them. The possibility of using |
787 |
detached signature has been considered but it was rejected as |
788 |
unnecessary complexity for minor gain. |
789 |
|
790 |
Technically, a similar result could be effected via moving all the data |
791 |
into a compressed sub-Manifest in the top directory (e.g. |
792 |
``Manifest.sub.gz``), and including a ``MANIFEST`` entry for this file |
793 |
in a signed, uncompressed top-level Manifest. |
794 |
|
795 |
The existence of additional entries for uncompressed Manifest checksums |
796 |
was debated. However, plain entries for the uncompressed file would |
797 |
be confusing if only compressed file existed, and conflicting if both |
798 |
uncompressed and compressed variants existed. Furthermore, it has been |
799 |
pointed out that ``DIST`` entries do not have uncompressed variant |
800 |
either. |
801 |
|
802 |
|
803 |
Performance considerations |
804 |
-------------------------- |
805 |
|
806 |
Performing a full-tree verification on every sync raises some |
807 |
performance concerns for end-user systems. The initial testing has shown |
808 |
that a cold-cache verification on a btrfs file system can take up around |
809 |
4 minutes, with the process being mostly I/O bound. On the other hand, |
810 |
it can be expected that the verification will be performed directly |
811 |
after syncing, taking advantage of warm filesystem cache. |
812 |
|
813 |
To improve speed on I/O and/or CPU-restrained systems even further, |
814 |
the algorithms can be easily extended to perform incremental |
815 |
verification. Given that rsync does not preserve mtimes by default, |
816 |
the tool can take advantage of mtime and Manifest comparisons to recheck |
817 |
only the parts of the repository that have changed. |
818 |
|
819 |
Furthermore, the package manager implementations can restrict checking |
820 |
only to the parts of the repository that are actually being used. |
821 |
|
822 |
|
823 |
Backwards Compatibility |
824 |
======================= |
825 |
|
826 |
This GLEP provides optional means of preserving backwards compatibility. |
827 |
To preserve the backwards compatibility, the following needs to hold |
828 |
for the ``Manifest`` file in every package directory: |
829 |
|
830 |
- all files must be covered by the single ``Manifest`` file, |
831 |
|
832 |
- all distfiles used by the package must be included, |
833 |
|
834 |
- all files inside the ``files/`` subdirectory need to use |
835 |
the ``AUX`` tag (rather than ``DATA``), |
836 |
|
837 |
- all ``.ebuild`` files need to use the ``EBUILD`` tag, |
838 |
|
839 |
- the ``metadata.xml`` and ``ChangeLog`` files need to use |
840 |
the ``MISC`` tag, |
841 |
|
842 |
- the Manifest can be signed to provide authenticity verification, |
843 |
|
844 |
- an uncompressed Manifest must always exist, and a compressed Manifest |
845 |
of identical content may be present. |
846 |
|
847 |
Once the backwards compatibility is no longer a concern, the above |
848 |
no longer needs to hold and the deprecated tags can be removed. |
849 |
|
850 |
|
851 |
Reference Implementation |
852 |
======================== |
853 |
|
854 |
The reference implementation for this GLEP is being developed |
855 |
as the gemato project [#GEMATO]_. |
856 |
|
857 |
|
858 |
Credits |
859 |
======= |
860 |
|
861 |
Thanks to all the people whose contributions were invaluable |
862 |
to the creation of this GLEP. This includes but is not limited to: |
863 |
|
864 |
- Robin Hugh Johnson, |
865 |
- Ulrich Müller. |
866 |
|
867 |
Additionally, thanks to Robin Hugh Johnson for the original |
868 |
MataManifest GLEP series which served both as inspiration and source |
869 |
of many concepts used in this GLEP. Recursively, also thanks to all |
870 |
the people who contributed to the original GLEPs. |
871 |
|
872 |
|
873 |
References |
874 |
========== |
875 |
|
876 |
.. [#GLEP44] GLEP 44: Manifest2 format |
877 |
(https://www.gentoo.org/glep/glep-0044.html) |
878 |
|
879 |
.. [#GLEP57] GLEP 57: Security of distribution of Gentoo software |
880 |
- Overview |
881 |
(https://www.gentoo.org/glep/glep-0057.html) |
882 |
|
883 |
.. [#GLEP58] GLEP 58: Security of distribution of Gentoo software |
884 |
- Infrastructure to User distribution - MetaManifest |
885 |
(https://www.gentoo.org/glep/glep-0058.html) |
886 |
|
887 |
.. [#GLEP59] GLEP 59: Manifest2 hash policies and security implications |
888 |
(https://www.gentoo.org/glep/glep-0059.html) |
889 |
|
890 |
.. [#GLEP60] GLEP 60: Manifest2 filetypes |
891 |
(https://www.gentoo.org/glep/glep-0060.html) |
892 |
|
893 |
.. [#GLEP61] GLEP 61: Manifest2 compression |
894 |
(https://www.gentoo.org/glep/glep-0061.html) |
895 |
|
896 |
.. [#PMS-FETCH] Package Manager Specification: Dependency Specification |
897 |
Format - SRC_URI |
898 |
(https://projects.gentoo.org/pms/6/pms.html#x1-940008.2.10) |
899 |
|
900 |
.. [#MD5] RFC1321: The MD5 Message-Digest Algorithm |
901 |
(https://www.ietf.org/rfc/rfc1321.txt) |
902 |
|
903 |
.. [#RIPEMD160] The hash function RIPEMD-160 |
904 |
(https://homes.esat.kuleuven.be/~bosselae/ripemd160.html) |
905 |
|
906 |
.. [#SHS] FIPS PUB 180-4: Secure Hash Standard (SHS) |
907 |
(http://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.180-4.pdf) |
908 |
|
909 |
.. [#WHIRLPOOL] The WHIRLPOOL Hash Function |
910 |
(http://www.larc.usp.br/~pbarreto/WhirlpoolPage.html) |
911 |
|
912 |
.. [#BLAKE2] BLAKE2 -- fast secure hashing |
913 |
(https://blake2.net/) |
914 |
|
915 |
.. [#SHA3] FIPS PUB 202: SHA-3 Standard: Permutation-Based Hash |
916 |
and Extendable-Output Functions |
917 |
(http://nvlpubs.nist.gov/nistpubs/FIPS/NIST.FIPS.202.pdf) |
918 |
|
919 |
.. [#STREEBOG] GOST R 34.11-2012: Streebog Hash Function |
920 |
(https://www.streebog.net/) |
921 |
|
922 |
.. [#C08] Cappos, J et al. (2008). "Attacks on Package Managers" |
923 |
(https://www2.cs.arizona.edu/stork/packagemanagersecurity/attacks-on-package-managers.html) |
924 |
|
925 |
.. [#DIST] According to Robin H. Johnson, 8.4% of all DIST entries |
926 |
at the time of writing are duplicate, representing a 2 MiB |
927 |
out of 25 MiB of DIST entries altogether. |
928 |
|
929 |
.. [#GEMATO] gemato: Gentoo Manifest Tool |
930 |
(https://github.com/mgorny/gemato/) |
931 |
|
932 |
Copyright |
933 |
========= |
934 |
This work is licensed under the Creative Commons Attribution-ShareAlike 3.0 |
935 |
Unported License. To view a copy of this license, visit |
936 |
http://creativecommons.org/licenses/by-sa/3.0/. |
937 |
|
938 |
|
939 |
-- |
940 |
Best regards, |
941 |
Michał Górny |