Backport #38406 by @bircni
Addresses a batch of privately reported security issues, grouped by
area:
- **SSRF** - migration PR-patch/asset fetches, OAuth2 avatar & OpenID
discovery, pull-mirror URL re-validation, and the outbound proxy path.
- **Access-token scope** - prevent scope escalation on token creation;
keep public-only tokens confined (feeds, packages, Actions listings,
star/watch lists, limited/private owners).
- **Access control / disclosure** - go-get default-branch leak, webhook
authorization-header leak, watch clearing on private transitions,
label/attachment scoping.
- **Denial of service** - input bounds for npm dist-tags, Debian control
files, Arch file lists, and SSH keys.
### 📌 Attention for site admins
Not breaking - existing configs keep working - but two changes are worth
a look:
- **New SSRF protection** Outbound requests (migrations, OAuth2 avatars,
OpenID discovery, pull mirrors, proxy path) are now validated against
the allow/block host lists. If your instance legitimately reaches
internal hosts, you may need to add them to
`[security].ALLOWED_HOST_LIST` (and the relevant `ALLOW_LOCALNETWORKS`
settings).
- **Deprecation** `[webhook].ALLOWED_HOST_LIST` is deprecated and will
be removed in a future release. Use `[security].ALLOWED_HOST_LIST`
instead; the old key still works for now.
Co-authored-by: bircni <bircni@icloud.com>
Co-authored-by: TheFox0x7 <thefox0x7@gmail.com>
Co-authored-by: techknowlogick <techknowlogick@gitea.io>
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Zettat123 <zettat123@gmail.com>
Backport #38274 by @afahey03
Fixes HTTP 500 when OIDC auto account linking (`ACCOUNT_LINKING=auto`)
requires local 2FA. `oauth2LinkAccount` set `linkAccount` in the session
before redirecting to 2FA but did not persist `linkAccountData`, so
`TwoFactorPost` failed with `not in LinkAccount session`. The manual
linking flow already stored both, this aligns auto-link with that
behavior.
Created the test, `TestOAuth2AutoLinkWithTwoFactor`, which verifies that
automatic account linking completes after the user passes local 2FA when
an OIDC identity matches an existing account.
DISCLAIMER: I used AI to create the test
Closes#38171
Co-authored-by: Aidan Fahey <afahey2003@yahoo.com>
Co-authored-by: bircni <bircni@icloud.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
This PR replaces a set of struct-based `Get` lookups with explicit
`db.Get` / `db.Exist` conditions in places where zero-value fields can
lead to ambiguous matches or incorrect records being returned.
The main goal is to make read paths deterministic and avoid accidentally
matching the wrong row when only part of a struct is populated.
### What changed
- replace many `db.GetEngine(ctx).Get(bean)` calls with explicit
`builder.Eq` conditions across models such as actions, admin tasks,
issues, pull requests, repositories, users, packages, redirects,
watches, stars, and follows
- use quoted column names where needed for reserved fields like `index`,
`type`, and `name`
- add dedicated user lookup helpers for:
- primary email
- OAuth login source / login name
- update sign-in and OAuth-related flows to use explicit individual-user
lookups instead of partially populated `User` structs
- tighten package property and Terraform lock lookups to avoid ambiguous
reads and updates
- keep existing fallback behavior where needed, while removing reliance
on zero-value struct matching
### User-facing impact
These changes primarily affect authentication and account lookup paths:
- email/username sign-in now re-fetches users through explicit keys
- OAuth2 auto-linking now resolves users by name or primary email
explicitly
- OAuth2 login/sync now looks up users by login source, login type, and
login name explicitly
- non-individual accounts are no longer implicitly matched through
partial user lookups in these flows
This should reduce the risk of incorrect account matches and make query
behavior more predictable across the codebase.
---------
Co-authored-by: bircni <bircni@icloud.com>
The OAuth2 sign-in callback unconditionally set IsActive=true on the
local user row whenever the IdP authenticated them, silently undoing an
administrator's "Disable Account" action and granting the user a fresh
session in the same response. Treat the local IsActive flag as an
authoritative admin override: inactive users get a session and are
routed through the existing activate / prohibit-login pages by
verifyAuthWithOptions, matching the local-credentials sign-in path.
Adds an integration regression test that disables a linked local user
and asserts the row stays IsActive=false after a full OIDC callback.
---------
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
fixes defect where claims where only applies on login but not during
account linking making only the second login take them into account
fixes: https://github.com/go-gitea/gitea/issues/32566
## Summary
This fixes an OIDC sign-in edge case where a stale `external_login_user`
record can still point to an organization or a deleted user.
In that situation, Gitea may keep resolving the external login to the
wrong account during sign-in. For affected instances, this matches the
behavior reported in #36439 and #37812, where a user signing in with
OIDC/Entra ID could appear as an organization, or hit a 404 after that
organization was removed.
## What changed
- validate the user resolved from `external_login_user` during
OAuth2/OIDC login
- ignore stale links when the linked user no longer exists
- ignore stale links when the linked user is not an individual user
- remove the stale external login row so the sign-in flow can relink the
external account to the correct user
## Related
- Fixes#37812
- Related to #36439
---------
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: Claude (Opus 4.8) <noreply@anthropic.com>
Using the standard OpenID Connect OAuth2 provider type doesn't work well
for AWS Cognito. Most of the functionality works absolutely fine,
however the query parameter `post_logout_redirect_uri` is not understood
by Cognito and results in a bad experience when logging out.
To combat this i've added a new `AWS Cognito` provider which is almost
identical to the `Open ID Connect` type except it overrides the query
parameter to `logout_uri` which is what Cognito expects.
<img width="647" height="272" alt="image"
src="https://github.com/user-attachments/assets/d4bb30e2-f25e-41a1-91cb-4efa67137c57"
/>
This then results in a nice experience logging out with no errors seen -
even though the logout does succeed. Why AWS thought they would deviate
from the OAuth spec in this particular area is beyond me...
---------
Co-authored-by: Tom Thornton <tom.thornton@sony.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Nicolas <bircni@icloud.com>
## Summary
Fixes
[go-gitea/gitea#37564](https://github.com/go-gitea/gitea/issues/37564):
when an OIDC provider returns a `picture` claim, Gitea is supposed to
download that image as the user's avatar (if `[oauth2_client]
UPDATE_AVATAR = true`). Two latent bugs prevented this from working
consistently:
1. **Default Go User-Agent rejected by some image hosts.**
`oauth2UpdateAvatarIfNeed` used `http.Get`, which sends `User-Agent:
Go-http-client/1.1`. Hosts like `upload.wikimedia.org` reject that UA
with `403`, and every error path silently returned, so the user was left
with an identicon and **no log line** to diagnose the issue.
2. **Link-account *register* path skipped avatar sync.** First-time OIDC
sign-ins where auto-registration is disabled (or required a
username/password retype) go through `LinkAccountPostRegister`, which
created the user but never called `oauth2SignInSync`. So the avatar /
full name / SSH keys from the IdP were dropped on the floor for those
users, even though the existing-account-link path (`oauth2LinkAccount`)
and the auto-register path (`handleOAuth2SignIn`) both already did the
sync.
## Changes
- `routers/web/auth/oauth.go` — `oauth2UpdateAvatarIfNeed` now uses
`http.NewRequest` + `http.DefaultClient.Do`, sets `User-Agent: Gitea
<version>`, and logs every failure path at `Warn` (invalid URL, fetch
error, non-200, body read error, oversize body, upload error). No silent
failures.
- `routers/web/auth/linkaccount.go` — `LinkAccountPostRegister` now
calls `oauth2SignInSync` after a successful user creation, mirroring the
auto-register and link-existing-account flows.
- `tests/integration/oauth_avatar_test.go` — new
`TestOAuth2AvatarFromPicture` integration test with five sub-cases:
- `AutoRegister_FetchesAvatarFromPictureWithGiteaUA` — happy path,
asserts `use_custom_avatar=true`, an avatar hash is set, exactly one
HTTP request was made, and the request carried a `Gitea ` UA. The mock
server enforces the UA prefix to mirror real-world hosts that reject
Go's default UA.
- `AutoRegister_NonOK_DoesNotUpdateAvatar` — server returns 403; user's
avatar must remain unset.
- `AutoRegister_EmptyPicture_NoFetch` — empty `picture` claim must not
trigger any HTTP request.
- `AutoRegister_UpdateAvatarFalse_NoFetch` — `UPDATE_AVATAR=false` must
not trigger any HTTP request.
- `LinkAccountRegister_FetchesAvatarFromPicture` — guards the
`linkaccount.go` fix; without the new `oauth2SignInSync` call this
assertion fails.
## Related
- Upstream issue: go-gitea/gitea#37564
--------------------------------------------
AI Editor was used in this PR
---------
Signed-off-by: silverwind <me@silverwind.io>
Co-authored-by: silverwind <me@silverwind.io>
Co-authored-by: Claude (Opus 4.7) <noreply@anthropic.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: Nicolas <bircni@icloud.com>
Follow up #37327. See the comments.
* Root problem: the design of OAuth2 providers is a mess, the display
name is used as provider's name and used in the URL directly
* The regressions:
* When trying to fix https://github.com/go-gitea/gitea/issues/36409 , it
introduced inconsistent URL escaping for the "path" part.
* This fix: always use "path escaping" for the path part, add more tests
to cover all escaping cases.
Now, frontend "pathEscape" and "pathEscapeSegments" generate exactly the
same result as backend.
The link to authentication sources is now escaped with the QueryEscape.
This commit fixes that by unescaping the provider name in the URL.
---------
Signed-off-by: prettysunflower <me@prettysunflower.moe>
Signed-off-by: wxiaoguang <wxiaoguang@gmail.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
1. Fix the "flash message" layout problem for different cases
* I am sure most of the users should have ever seen the ugly
center-aligned error message with multiple lines.
2. Fix inconsistent "Details" flash message EOL handling, sometimes
`\n`, sometimes `<br>`
* Now, always use "\n" and use `<pre>` to render
3. Remove SanitizeHTML template func because it is not useful and can be
easily abused.
* But it is still kept for mail templates, for example:
https://github.com/go-gitea/gitea/issues/36049
4. Clarify PostProcessCommitMessage's behavior and add FIXME comment
By the way: cleaned up some devtest pages, move embedded style block to
CSS file
At logout time, if the user authenticated via OIDC, we look up the
provider's `end_session_endpoint` (already discovered by Goth from the
OIDC metadata) and redirect there with `client_id` and
`post_logout_redirect_uri`.
Non-OIDC OAuth2 providers (GitHub, GitLab, etc.) are unaffected — they
fall back to local-only logout.
Fix#14270
---------
Signed-off-by: Nikita Vakula <nikita.vakula@alpsalpine.com>
Co-authored-by: Nikita Vakula <nikita.vakula@alpsalpine.com>
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Fix#35998
1. Fix `<a rel>` :
* "_blank" already means "noopener"
* "noreferrer" is already provided by page's `<meta name="referrer">`
2. Fix "redirect_to" mechisam
* Use "referer" header to determine the redirect link for a successful
login
3. Simplify code and merge duplicate logic
Fix#880
Design:
1. A global setting `security.TWO_FACTOR_AUTH`.
* To support org-level config, we need to introduce a better "owner
setting" system first (in the future)
2. A user without 2FA can login and may explore, but can NOT read or
write to any repositories via API/web.
3. Keep things as simple as possible.
* This option only aggressively suggest users to enable their 2FA at the
moment, it does NOT guarantee that users must have 2FA before all other
operations, it should be good enough for real world use cases.
* Some details and tests could be improved in the future since this
change only adds a check and seems won't affect too much.
---------
Co-authored-by: Lunny Xiao <xiaolunwen@gmail.com>
In history (from some legacy frameworks), both `:name` and `name` are
supported as path path name, `:name` is an alias to `name`.
To make code consistent, now we should only use `name` but not `:name`.
Also added panic check in related functions to make sure the name won't
be abused in case some downstreams still use them.
See discussion on #31561 for some background.
The introspect endpoint was using the OIDC token itself for
authentication. This fixes it to use basic authentication with the
client ID and secret instead:
* Applications with a valid client ID and secret should be able to
successfully introspect an invalid token, receiving a 200 response
with JSON data that indicates the token is invalid
* Requests with an invalid client ID and secret should not be able
to introspect, even if the token itself is valid
Unlike #31561 (which just future-proofed the current behavior against
future changes to `DISABLE_QUERY_AUTH_TOKEN`), this is a potential
compatibility break (some introspection requests without valid client
IDs that would previously succeed will now fail). Affected deployments
must begin sending a valid HTTP basic authentication header with their
introspection requests, with the username set to a valid client ID and
the password set to the corresponding client secret.
This leverages the existing `sync_external_users` cron job to
synchronize the `IsActive` flag on users who use an OAuth2 provider set
to synchronize. This synchronization is done by checking for expired
access tokens, and using the stored refresh token to request a new
access token. If the response back from the OAuth2 provider is the
`invalid_grant` error code, the user is marked as inactive. However, the
user is able to reactivate their account by logging in the web browser
through their OAuth2 flow.
Also changed to support this is that a linked `ExternalLoginUser` is
always created upon a login or signup via OAuth2.
### Notes on updating permissions
Ideally, we would also refresh permissions from the configured OAuth
provider (e.g., admin, restricted and group mappings) to match the
implementation of LDAP. However, the OAuth library used for this `goth`,
doesn't seem to support issuing a session via refresh tokens. The
interface provides a [`RefreshToken`
method](https://github.com/markbates/goth/blob/master/provider.go#L20),
but the returned `oauth.Token` doesn't implement the `goth.Session` we
would need to call `FetchUser`. Due to specific implementations, we
would need to build a compatibility function for every provider, since
they cast to concrete types (e.g.
[Azure](https://github.com/markbates/goth/blob/master/providers/azureadv2/azureadv2.go#L132))
---------
Co-authored-by: Kyle D <kdumontnu@gmail.com>
This PR only does "renaming":
* `Route` should be `Router` (and chi router is also called "router")
* `Params` should be `PathParam` (to distingush it from URL query param, and to match `FormString`)
* Use lower case for private functions to avoid exposing or abusing
This commit forces the resource owner (user) to always approve OAuth 2.0
authorization requests if the client is public (e.g. native
applications).
As detailed in [RFC 6749 Section 10.2](https://www.rfc-editor.org/rfc/rfc6749.html#section-10.2),
> The authorization server SHOULD NOT process repeated authorization
requests automatically (without active resource owner interaction)
without authenticating the client or relying on other measures to ensure
that the repeated request comes from the original client and not an
impersonator.
With the implementation prior to this patch, attackers with access to
the redirect URI (e.g., the loopback interface for
`git-credential-oauth`) can get access to the user account without any
user interaction if they can redirect the user to the
`/login/oauth/authorize` endpoint somehow (e.g., with `xdg-open` on
Linux).
Fixes#25061.
Co-authored-by: wxiaoguang <wxiaoguang@gmail.com>
Follow #30454
And fix#24957
When using "preferred_username", if no such field,
`extractUserNameFromOAuth2` (old `getUserName`) shouldn't return an
error. All other USERNAME options do not return such error.
And fine tune some logic and error messages, make code more stable and
more friendly to end users.
Since `modules/context` has to depend on `models` and many other
packages, it should be moved from `modules/context` to
`services/context` according to design principles. There is no logic
code change on this PR, only move packages.
- Move `code.gitea.io/gitea/modules/context` to
`code.gitea.io/gitea/services/context`
- Move `code.gitea.io/gitea/modules/contexttest` to
`code.gitea.io/gitea/services/contexttest` because of depending on
context
- Move `code.gitea.io/gitea/modules/upload` to
`code.gitea.io/gitea/services/context/upload` because of depending on
context
Follow #29165.
* Introduce JSONTemplate to help to render JSON templates
* Introduce JSEscapeSafe for templates. Now only use `{{ ... |
JSEscape}}` instead of `{{ ... | JSEscape | Safe}}`
* Simplify "UserLocationMapURL" useage
Fixes#28660
Fixes an admin api bug related to `user.LoginSource`
Fixed `/user/emails` response not identical to GitHub api
This PR unifies the user update methods. The goal is to keep the logic
only at one place (having audit logs in mind). For example, do the
password checks only in one method not everywhere a password is updated.
After that PR is merged, the user creation should be next.
Giteabot