The inventory of the sha256:* images and the manifest index that
reference them is incomplete because it does not take into account any
image older than the expiration limit. As a result some sha256:* will
be considered orphaned although they are referenced from a manifest
index that was created more recently than the expiration limit.
There must not be any filtering based on the creation time when
building the inventory. The expiration limit must only be taken into
account when deleting orphaned images: those that are more recent than
the expiration limit must not be deleted.
This limit is specially important because it protects against a race
between a cleanup task and an ongoing mirroring task. A mirroring
task (such as skopeo sync) will first upload sha256:* images and then
create the corresponding manifest index. If a cleanup races against
it, the sha256:* images that are not yet referenced will be deleted
without skopeo noticing and the published index manifest that happens
at a later time will contain references to non-existent images.
- [x] add architecture-specific removal support
- [x] Fix upload competition
- [x] Fix not checking input when downloading
docs: https://codeberg.org/forgejo/docs/pulls/874
### Release notes
- [ ] I do not want this change to show in the release notes.
Reviewed-on: https://codeberg.org/forgejo/forgejo/pulls/5351
Reviewed-by: Earl Warren <earl-warren@noreply.codeberg.org>
Co-authored-by: Exploding Dragon <explodingfkl@gmail.com>
Co-committed-by: Exploding Dragon <explodingfkl@gmail.com>
- The Conan and Container packages use a different type of
authentication. It first authenticates via the regular way (api tokens
or user:password, handled via `auth.Basic`) and then generates a JWT
token that is used by the package software (such as Docker) to do the
action they wanted to do. This JWT token didn't properly propagate the
API scopes that the token was generated for, and thus could lead to a
'scope escalation' within the Conan and Container packages, read
access to write access.
- Store the API scope in the JWT token, so it can be propagated on
subsequent calls that uses that JWT token.
- Integration test added.
- Resolves #5128
Previous arch package grouping was not well-suited for complex or multi-architecture environments. It now supports the following content:
- Support grouping by any path.
- New support for packages in `xz` format.
- Fix clean up rules
<!--start release-notes-assistant-->
## Draft release notes
<!--URL:https://codeberg.org/forgejo/forgejo-->
- Features
- [PR](https://codeberg.org/forgejo/forgejo/pulls/4903): <!--number 4903 --><!--line 0 --><!--description c3VwcG9ydCBncm91cGluZyBieSBhbnkgcGF0aCBmb3IgYXJjaCBwYWNrYWdl-->support grouping by any path for arch package<!--description-->
<!--end release-notes-assistant-->
Reviewed-on: https://codeberg.org/forgejo/forgejo/pulls/4903
Reviewed-by: Earl Warren <earl-warren@noreply.codeberg.org>
Co-authored-by: Exploding Dragon <explodingfkl@gmail.com>
Co-committed-by: Exploding Dragon <explodingfkl@gmail.com>
Replace a double select with a simple select.
The complication originates from the initial implementation which
deleted packages instead of selecting them. It was justified to
workaround a problem in MySQL. But it is just a waste of resources
when collecting a list of IDs.
- We were previously using `github.com/keybase/go-crypto`, because the
package for openpgp by Go itself is deprecated and no longer
maintained. This library provided a maintained version of the openpgp
package. However, it hasn't seen any activity for the last five years,
and I would therefore consider this also unmaintained.
- This patch switches the package to `github.com/ProtonMail/go-crypto`
which provides a maintained version of the openpgp package and was
already being used in the tests.
- Adds unit tests, I've carefully checked the callstacks to ensure the
OpenPGP-related code was covered under either a unit test or integration
tests to avoid regression, as this can easily turn into security
vulnerabilities if a regression happens here.
- Small behavior update, revocations are now checked correctly instead
of checking if they merely exist and the expiry time of a subkey is used
if one is provided (this is just cosmetic and doesn't impact security).
- One more dependency eliminated :D
- Currently in the Cargo section of the packages setting menu two
buttons are always shown, "Initalize index" and "Rebuild index", however
only of these should be shown depending on the state of the index, if
there's no index the "Initalize index" button should be shown and if
there's an index the "Rebuild index" button should be shown. This patch
does exactly that.
- Resolves #2628
Port of https://github.com/go-gitea/gitea/pull/29205
Use a clearly defined "signing secret" for token signing.
(cherry picked from commit 8be198cdef0a486f417663b1fd6878458d7e5d92)
There is a missing newline when generating the debian apt repo InRelease
file, which results in output like:
```
[...]
Date: Wed, 14 Feb 2024 05:03:01 UTC
Acquire-By-Hash: yesMD5Sum:
51a518dbddcd569ac3e0cebf330c800a 3018 main-dev/binary-amd64/Packages
[...]
```
It appears this would probably result in apt ignoring the
Acquire-By-Hash setting and not using the by-hash functionality,
although I'm not sure how to confirm it.
(cherry picked from commit 33400a02d4eb35a0656fd6d20fc56801de09b959)
Resolves https://github.com/go-gitea/gitea/issues/28704
Example of an entry in the generated `APKINDEX` file:
```
C:Q1xCO3H9LTTEbhKt9G1alSC87I56c=
P:hello
V:2.12-r1
A:x86_64
T:The GNU Hello program produces a familiar, friendly greeting
U:https://www.gnu.org/software/hello/
L:GPL-3.0-or-later
S:15403
I:36864
o:hello
m:
t:1705934118
D:so:libc.musl-x86_64.so.1
p:cmd:hello=2.12-r1
i:foobar=1.0 !baz
k:42
```
the `i:` and `k:` entries are new.
---------
Co-authored-by: KN4CK3R <admin@oldschoolhack.me>
Fixes #22236
---
Error occurring currently while trying to revert commit using read-tree
-m approach:
> 2022/12/26 16:04:43 ...rvices/pull/patch.go:240:AttemptThreeWayMerge()
[E] [63a9c61a] Unable to run read-tree -m! Error: exit status 128 -
fatal: this operation must be run in a work tree
> - fatal: this operation must be run in a work tree
We need to clone a non-bare repository for `git read-tree -m` to work.
bb371aee6e
adds support to create a non-bare cloned temporary upload repository.
After cloning a non-bare temporary upload repository, we [set default
index](https://github.com/go-gitea/gitea/blob/main/services/repository/files/cherry_pick.go#L37)
(`git read-tree HEAD`).
This operation ends up resetting the git index file (see investigation
details below), due to which, we need to call `git update-index
--refresh` afterward.
Here's the diff of the index file before and after we execute
SetDefaultIndex: https://www.diffchecker.com/hyOP3eJy/
Notice the **ctime**, **mtime** are set to 0 after SetDefaultIndex.
You can reproduce the same behavior using these steps:
```bash
$ git clone https://try.gitea.io/me-heer/test.git -s -b main
$ cd test
$ git read-tree HEAD
$ git read-tree -m 1f085d7ed8 1f085d7ed8 9933caed00
error: Entry '1' not uptodate. Cannot merge.
```
After which, we can fix like this:
```
$ git update-index --refresh
$ git read-tree -m 1f085d7ed8 1f085d7ed8 9933caed00
```
I noticed the `BuildAllRepositoryFiles` function under the Alpine folder
is unused and I thought it was a bug.
But I'm not sure about this. Was it on purpose?
- Remove `ObjectFormatID`
- Remove function `ObjectFormatFromID`.
- Use `Sha1ObjectFormat` directly but not a pointer because it's an
empty struct.
- Store `ObjectFormatName` in `repository` struct
Refactor Hash interfaces and centralize hash function. This will allow
easier introduction of different hash function later on.
This forms the "no-op" part of the SHA256 enablement patch.
Changed behavior to calculate package quota limit using package `creator
ID` instead of `owner ID`.
Currently, users are allowed to create an unlimited number of
organizations, each of which has its own package limit quota, resulting
in the ability for users to have unlimited package space in different
organization scopes. This fix will calculate package quota based on
`package version creator ID` instead of `package version owner ID`
(which might be organization), so that users are not allowed to take
more space than configured package settings.
Also, there is a side case in which users can publish packages to a
specific package version, initially published by different user, taking
that user package size quota. Version in fix should be better because
the total amount of space is limited to the quota for users sharing the
same organization scope.
Hello there,
Cargo Index over HTTP is now prefered over git for package updates: we
should not force users who do not need the GIT repo to have the repo
created/updated on each publish (it can still be created in the packages
settings).
The current behavior when publishing is to check if the repo exist and
create it on the fly if not, then update it's content.
Cargo HTTP Index does not rely on the repo itself so this will be
useless for everyone not using the git protocol for cargo registry.
This PR only disable the creation on the fly of the repo when publishing
a crate.
This is linked to #26844 (error 500 when trying to publish a crate if
user is missing write access to the repo) because it's now optional.
---------
Co-authored-by: KN4CK3R <admin@oldschoolhack.me>
Cargo registry-auth feature requires config.json to have a property
auth-required set to true in order to send token to all registry
requests.
This is ok for git index because you can manually edit the config.json
file to add the auth-required, but when using sparse
(setting index url to
"sparse+https://git.example.com/api/packages/{owner}/cargo/"), the
config.json is dynamically rendered, and does not reflect changes to the
config.json file in the repo.
I see two approaches:
- Serve the real config.json file when fetching the config.json on the
cargo service.
- Automatically detect if the registry requires authorization. (This is
what I implemented in this PR).
What the PR does:
- When a cargo index repository is created, on the config.json, set
auth-required to wether or not the repository is private.
- When the cargo/config.json endpoint is called, set auth-required to
wether or not the request was authorized using an API token.
Until now expired package data gets deleted daily by a cronjob. The
admin page shows the size of all packages and the size of unreferenced
data. The users (#25035, #20631) expect the deletion of this data if
they run the cronjob from the admin page but the job only deletes data
older than 24h.
This PR adds a new button which deletes all expired data.
---------
Co-authored-by: silverwind <me@silverwind.io>
The version listed in rpm repodata should only contain the rpm version
(1.0.0) and not the combination of version and release (1.0.0-2). We
correct this behaviour in primary.xml.gz, filelists.xml.gz and
others.xml.gz.
Signed-off-by: Peter Verraedt <peter@verraedt.be>
Bumping `github.com/golang-jwt/jwt` from v4 to v5.
`github.com/golang-jwt/jwt` v5 is bringing some breaking changes:
- standard `Valid()` method on claims is removed. It's replaced by
`ClaimsValidator` interface implementing `Validator()` method instead,
which is called after standard validation. Gitea doesn't seem to be
using this logic.
- `jwt.Token` has a field `Valid`, so it's checked in `ParseToken`
function in `services/auth/source/oauth2/token.go`
---------
Co-authored-by: Giteabot <teabot@gitea.io>
Fixes #24723
Direct serving of content aka HTTP redirect is not mentioned in any of
the package registry specs but lots of official registries do that so it
should be supported by the usual clients.
