Compare commits

...

27 Commits

Author SHA1 Message Date
Eduard Gert
b9247201e9 Merge branch 'main' into fix-browser-dialog-not-closing 2026-07-20 14:53:21 +02:00
Maycon Santos
d64e9542eb [proxy] Bedrock cost-allocation metadata + per-provider metadata_disabled (#6791) 2026-07-20 14:45:50 +02:00
Viktor Liu
3fb26d458e [relay] Remove deprecated Hello handshake and gob token decode (#6783) 2026-07-20 14:13:40 +02:00
s-shimizu-clpl
a411fd300c [client] Add Japanese (ja) UI translation (#6790)
## Describe your changes
Adds a Japanese (ja) locale for the desktop UI, following the procedure
in `client/ui/i18n/TRANSLATING.md`.

- New `client/ui/i18n/locales/ja/common.json` — all 441 keys from the
`en` source bundle, `message` only, same key order.
- Registered in `client/ui/i18n/locales/_index.json`: `{"code": "ja",
"displayName": "日本語", "englishName": "Japanese"}`.

Translation notes:
- Placeholders (`{version}`, `{count}`, `{name}`, `{remaining}`, ...),
`\n`, trailing `…`/`...`, the leading space in
`notify.update.enforcedSuffix`, and the `<docs>…</docs>` inline-link
tags are all preserved verbatim.
- Brands kept as-is: NetBird, WireGuard® (® preserved), Rosenpass,
GitHub, NetBird Cloud. Acronyms kept: SSO, DNS, IP/IPv6, ACL, SSH, JWT,
TTL, SFTP, MTU, PSK, LAN, P2P, ICE, IdP.
- Formal register (です・ます); short labels for buttons/tray; quoted UI
labels use Japanese brackets 「」.

No code changes are required: the React frontend auto-loads every
`locales/*/common.json` via `import.meta.glob`, the tray/Go side embeds
the tree via `//go:embed all:i18n/locales`, and the language picker
lists whatever `_index.json` declares.

### Verification
- `go test ./client/ui/i18n/` passes.
- JSON valid; key set and order identical to `en`; no `description`
fields; no empty messages; placeholder/`\n` counts match `en`.
- Built the Windows UI and confirmed the General, Network, Security, and
Troubleshooting settings tabs plus the main window render correctly in
Japanese (language picker shows "日本語 (Japanese)"), with no text
truncation, overflow, or leaked placeholders.

## Issue ticket number and link
N/A

## Stack

<!-- branch-stack -->

### Checklist
- [ ] Is it a bug fix
- [ ] Is a typo/documentation fix
- [x] Is a feature enhancement
- [ ] It is a refactor
- [ ] Created tests that fail without the change (if possible)
- [x] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— adding a shipped UI locale is the documented extension path in
`client/ui/i18n/TRANSLATING.md`.

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation
Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change — UI translation
strings only; no user-facing product docs are affected.

### Docs PR URL (required if "docs added" is checked)
N/A

🤖 Generated with [Claude Code](https://claude.com/claude-code)

<!-- codesmith:footer -->
---
<a
href="https://app.blacksmith.sh/netbirdio/codesmith/netbird/pr/6790"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-light-v2.svg"><img
alt="View with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"></picture></a>
<a
href="https://backend.blacksmith.sh/track/enable-autofix?expires=1786742881&installation_id=146802194&pr_number=6790&repository=netbirdio%2Fnetbird&return_to=https%3A%2F%2Fgithub.com%2Fnetbirdio%2Fnetbird%2Fpull%2F6790&signature=a1dcd7faf63ae10071ca2da3639b4daf6505eeb0258790e622222e742a8ebcb6"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-light.svg"><img
alt="Autofix with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"></picture></a>
<sup>Need help on this PR? Tag <code>/codesmith</code> with what you
need. Autofix is disabled.</sup>

<!-- codesmith:autofix:disabled -->
<!-- /codesmith:footer -->

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
  * Added Japanese as a supported interface language.
* **Localization**
* Provided full Japanese translations for the UI, covering navigation,
settings, onboarding, desktop/tray notifications, connection and status
messaging, troubleshooting, update/about screens, and
authentication/error dialogs.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Co-authored-by: Eduard Gert <kontakt@eduardgert.de>
2026-07-20 14:12:13 +02:00
Eduard Gert
2af9d0e72e Merge branch 'main' into fix-browser-dialog-not-closing 2026-07-20 10:21:25 +02:00
Viktor Liu
92a5ed19d3 [client] Merge 0.74.7 to main (#6819)
## Describe your changes

Merge 0.74.7 to main

## Issue ticket number and link

## Stack

<!-- branch-stack -->

### Checklist
- [ ] Is it a bug fix
- [ ] Is a typo/documentation fix
- [x] Is a feature enhancement
- [ ] It is a refactor
- [ ] Created tests that fail without the change (if possible)
- [ ] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation
Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

### Docs PR URL (required if "docs added" is checked)
Paste the PR link from https://github.com/netbirdio/docs here:

https://github.com/netbirdio/docs/pull/__

<!-- codesmith:footer -->
---
<a
href="https://app.blacksmith.sh/netbirdio/codesmith/netbird/pr/6819"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-light-v2.svg"><img
alt="View with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"></picture></a>
<a
href="https://backend.blacksmith.sh/track/enable-autofix?expires=1786890471&installation_id=146802194&pr_number=6819&repository=netbirdio%2Fnetbird&return_to=https%3A%2F%2Fgithub.com%2Fnetbirdio%2Fnetbird%2Fpull%2F6819&signature=6078105908fabf6520f504b594cc394dc2a2d749a709c88c22f2d187301f6c71"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-light.svg"><img
alt="Autofix with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"></picture></a>
<sup>Need help on this PR? Tag <code>/codesmith</code> with what you
need. Autofix is disabled.</sup>

<!-- codesmith:autofix:disabled -->
<!-- /codesmith:footer -->

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* Firewall now safely evaluates fragmented inbound traffic while
blocking unsupported, malformed, overlapping, or unapproved fragments.
* SOCKS5 listener address and port can be configured through environment
variables, defaulting safely to loopback.
* **Bug Fixes**
* Strengthened token validation for malformed authentication responses.
* Prevented invalid hostnames from injecting settings into generated SSH
configurations.
* Improved handling of command-line-like account names and concurrent
QUIC connections.
* **Tests**
* Added coverage and performance benchmarks for fragmentation, listener
configuration, and security validation.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-07-18 00:07:44 +09:00
Viktor Liu
be6777427d Merge remote-tracking branch 'origin/main' into 0.74.7-branch-sync
# Conflicts:
#	client/internal/peer/conn.go
#	client/internal/peer/conn_test.go
#	client/internal/peer/wg_watcher.go
#	client/internal/peer/wg_watcher_test.go
#	go.mod
#	go.sum
2026-07-17 16:34:58 +02:00
Viktor Liu
a1c9427d80 [client] Evaluate IP fragments against firewall ACLs (#6781) 2026-07-17 16:05:32 +02:00
Pascal Fischer
a59d7fba95 [management] propagate auth grant types for combined server (#6817) 2026-07-17 15:41:57 +02:00
Zoltan Papp
41d7bf4bbd [client] Diagnose empty vs corrupt state (#6816)
## Describe your changes
When loadStateFile fails to unmarshal the state file, log whether the
file is empty (0 bytes) or has malformed content, including the byte
size.

## Issue ticket number and link

## Stack

<!-- branch-stack -->

### Checklist
- [ ] Is it a bug fix
- [ ] Is a typo/documentation fix
- [x] Is a feature enhancement
- [ ] It is a refactor
- [ ] Created tests that fail without the change (if possible)
- [ ] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation
Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

### Docs PR URL (required if "docs added" is checked)
aste the PR link from https://github.com/netbirdio/docs here:

https://github.com/netbirdio/docs/pull/__


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Bug Fixes**
* Improved state-file loading warnings by distinguishing empty files
from files containing malformed content.
  * Preserved existing recovery behavior for corrupted state files.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-07-17 15:31:07 +02:00
Riccardo Manfrin
b7b0d5796e [client] Bind netstack SOCKS5 proxy to 127.0.0.1 by default (#6812)
## Describe your changes

In netstack mode the SOCKS5 proxy bridges local host applications into
the
userspace WireGuard stack (`client/iface/netstack/proxy.go`), so it only
needs
to be reachable from the same machine. It was binding to `0.0.0.0`,
making an
unauthenticated proxy reachable from the network — any host able to
reach the
port could relay traffic through the client into its NetBird overlay.

Bind to `127.0.0.1` by default. Add `NB_SOCKS5_LISTENER_ADDRESS` to
override the
bind host for the rare case the proxy must be reachable from other hosts
(e.g. a
container gateway); it is validated as an IP and falls back to loopback.
`ListenAddr` is split into `listenHost`/`listenPort` helpers, with
tests.

Behavior change: setups that relied on reaching the netstack SOCKS5
proxy from
another host must now set `NB_SOCKS5_LISTENER_ADDRESS=0.0.0.0`
explicitly.

## Issue ticket number and link

Internal security hardening of the netstack SOCKS5 listener bind address

([client/iface/netstack/env.go](https://github.com/netbirdio/netbird/blob/main/client/iface/netstack/env.go)).

## Stack

- `0.74.7-branch` - ⚠️ No PR associated with branch <!--
branch-stack -->
  - \#6812 :point\_left:

### Checklist

- [x] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [x] Created tests that fail without the change (if possible)
- [ ] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation

Select exactly one:

- [x] I added/updated documentation for this change
- [ ] Documentation is **not needed** for this change (explain why)

### Docs PR URL (required if "docs added" is checked)

Paste the PR link from <https://github.com/netbirdio/docs> here:

<https://github.com/netbirdio/docs/pull/860>

<!-- codesmith:footer -->

***

<a
href="https://app.blacksmith.sh/netbirdio/codesmith/netbird/pr/6812"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-light-v2.svg"><img
alt="View with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"></picture></a>
<a
href="https://backend.blacksmith.sh/track/enable-autofix?expires=1786872199&installation_id=146802194&pr_number=6812&repository=netbirdio%2Fnetbird&return_to=https%3A%2F%2Fgithub.com%2Fnetbirdio%2Fnetbird%2Fpull%2F6812&signature=6e78df77b784915baee9647b555483a8f7550a604d5393edcc501dd8f58b1395"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-light.svg"><img
alt="Autofix with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"></picture></a>
<sup>Need help on this PR? Tag <code>/codesmith</code> with what you
need. Autofix is disabled.</sup>

<!-- codesmith:autofix:disabled -->

<!-- /codesmith:footer -->

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **New Features**
- Added configuration options for the SOCKS5 listener’s bind address and
port.
- SOCKS5 now defaults to listening only on the local machine for
improved security.
- Valid address and port overrides are supported, with safe defaults
used for invalid values.

- **Tests**
- Added coverage for default settings and valid or invalid address and
port configurations.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-07-17 15:09:09 +02:00
Riccardo Manfrin
3f8c447378 [client] Rename isValidAccessToken to reflect audience-only check (#6806)
## Describe your changes

`isValidAccessToken` only decodes the JWT payload and checks the
audience
claim, but its name suggested full token validation. Rename it to
`validateTokenAudience` and document what it does: a client-side
audience/shape
check on a token just obtained from the IdP over TLS. Token authenticity
is
enforced server-side by the management server, which verifies the
signature
against the IdP's JWKS (`shared/auth/jwt/validator.go`) on every
request.

Also harden the parser: a non-empty token lacking the three-part JWT
structure
caused an index-out-of-range panic (`strings.Split(token, ".")[1]`); the
shape
is now validated first. `parseEmailFromIDToken` is documented as
best-effort UX
data (login hint/display), never used for authorization. Added tests for
audience matching, malformed tokens, and the panic regression.

Changes:

- Rename `isValidAccessToken` → `validateTokenAudience`; document that
it does
not verify the signature and that authenticity is enforced server-side.
- Fix an index-out-of-range panic on a non-empty token lacking JWT
structure
(`strings.Split(token, ".")[1]`) by validating the three-part shape
first.
- Document `parseEmailFromIDToken` as best-effort/unverified, used only
for the
  login-hint/display UX, never for an authorization decision.
- Add `util_test.go` covering audience matching (string and array),
missing
  audience, malformed payloads, and the panic regression.

## Issue ticket number and link

Internal cleanup: rename a misleadingly-named client-side helper and
harden JWT
parsing against malformed input (`client/internal/auth/util.go`).

## Stack

- `0.74.7-branch` - ⚠️ No PR associated with branch <!--
branch-stack -->
  - \#6806 :point\_left:

### Checklist

- [x] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [x] Created tests that fail without the change (if possible)

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation

Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

Internal client-side helper rename plus a panic hardening fix. No public
API,
gRPC, CLI/service flag, or configuration change; token authenticity
enforcement
(server-side JWKS verification) is unchanged.

### Docs PR URL (required if "docs added" is checked)

Paste the PR link from <https://github.com/netbirdio/docs> here:

N/A

<!-- codesmith:footer -->

***

<a
href="https://app.blacksmith.sh/netbirdio/codesmith/netbird/pr/6806"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-light-v2.svg"><img
alt="View with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/view-with-codesmith-dark-v2.svg"></picture></a>
<a
href="https://backend.blacksmith.sh/track/enable-autofix?expires=1786811124&installation_id=146802194&pr_number=6806&repository=netbirdio%2Fnetbird&return_to=https%3A%2F%2Fgithub.com%2Fnetbirdio%2Fnetbird%2Fpull%2F6806&signature=fe45ce9f6df46a609594f84037aaab2a21893621f29d2be48d7f8b347c3c8776"><picture><source
media="(prefers-color-scheme: dark)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"><source
media="(prefers-color-scheme: light)"
srcset="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-light.svg"><img
alt="Autofix with Codesmith"
src="https://pr-comments-assets.blacksmith.sh/codesmith/autofix-with-codesmith-dark.svg"></picture></a>
<sup>Need help on this PR? Tag <code>/codesmith</code> with what you
need. Autofix is disabled.</sup>

<!-- codesmith:autofix:disabled -->

<!-- /codesmith:footer -->

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **Bug Fixes**
- Improved access-token audience validation during device and PKCE
authentication flows.
- Malformed tokens now return clear validation errors instead of risking
runtime failures.
  - Added support for validating both string and array audience claims.

- **Tests**
- Added coverage for malformed tokens, invalid claims, missing
audiences, and panic prevention.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-07-17 11:00:55 +02:00
Viktor Liu
6e3f4d8722 [client] Disable gVisor TCP RACK loss detection on Windows (#6808) 2026-07-17 10:45:30 +02:00
Riccardo Manfrin
099ae4bc6c [client] Sanitize peer FQDN/hostname in generated SSH config (#6805)
## Describe your changes

Validate peer-supplied FQDN and hostname before they are written into
the
generated NetBird SSH client config (`client/ssh/config/manager.go`).
These
values originate from remote peers and were previously written verbatim
into
the config; malformed values (e.g. containing unexpected characters)
could
produce a broken or unintended config. FQDN/hostname are now checked
with
`domain.IsValidDomainNoWildcard`, and invalid, non-empty values are
dropped
with a warning. IPs are unaffected (already validated `netip.Addr`).
Added a
test covering malformed hostnames.

## Issue ticket number and link

Internal input-validation hardening for peer-supplied hostnames in the
generated SSH client config (`client/ssh/config/manager.go`).

## Stack

- \#6726 <!-- branch-stack -->
  - \#6805 :point\_left:

### Checklist

- [x] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [x] Created tests that fail without the change (if possible)

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation

Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

Internal client SSH config generation. No public API, gRPC, CLI/service
flag,
or configuration change — only input validation on peer-supplied
hostnames
before they are written to the generated ssh\_config.

### Docs PR URL (required if "docs added" is checked)

Paste the PR link from <https://github.com/netbirdio/docs> here:

N/A
2026-07-17 10:10:22 +02:00
Viktor Liu
63d60ba490 [client] Reject leading hyphen in getent input to prevent flag injection (#6787) 2026-07-17 09:28:33 +02:00
Viktor Liu
62fc8d254e [relay] Handle QUIC connections concurrently to prevent handshake head-of-line blocking (#6784) 2026-07-15 18:18:47 +02:00
Zoltan Papp
3a2f773d65 [client] preserve WireGuard key on interactive re-login (#6777)
NewAuth built a fresh in-memory config on every call via
CreateInMemoryConfig, which generates a new WireGuard private key when
none is set. The iOS Swift layer calls this on interactive re-login and
writes the resulting config back to the profile's netbird.cfg, so each
re-auth replaced the peer's persisted private key with a new one. A new
key means a new public key, so the management server registered a
brand-new peer on every re-authentication — named after the fallback
hostname.

Load the existing config with DirectUpdateOrCreateConfig when a config
file is already present so re-login reuses the peer's persisted private
key (and its identity). Only fall back to a fresh in-memory config for
the first-time login when no config file exists yet (or after logout,
which deletes the file). DirectUpdateOrCreateConfig uses non-atomic
writes so it also works inside the tvOS App Group sandbox. This matches
what Run() and LoginForMobile() already do.

## Describe your changes

## Issue ticket number and link

## Stack

<!-- branch-stack -->

### Checklist
- [x] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [ ] Created tests that fail without the change (if possible)
- [ ] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation
Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

### Docs PR URL (required if "docs added" is checked)
Paste the PR link from https://github.com/netbirdio/docs here:

https://github.com/netbirdio/docs/pull/__


<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **New Features**
* Added support for loading or creating persistent configuration when a
configuration file path is provided.
* Continued support for in-memory configuration for temporary or
first-time use.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-07-15 14:12:12 +02:00
Maycon Santos
f0eed7564f [management] Remove proxy peer stale deduplication logic (#6768)
## Describe your changes

Removing a leftover from an initial implementation. We ended up
resolving it on the client with status checks on the DNS response

## Issue ticket number and link

## Stack

- \#6726 <!-- branch-stack -->
  - \#6768 :point\_left:

### Checklist

- [x] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [ ] Created tests that fail without the change (if possible)
- [ ] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation

Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

### Docs PR URL (required if "docs added" is checked)

Paste the PR link from <https://github.com/netbirdio/docs> here:

<https://github.com/netbirdio/docs/pull/>\_\_

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

- **New Features**
- Added support for Bedrock-native request routing in agent network
scenarios.
- Added guardrail management capabilities for creating and removing
model allowlists.

- **Bug Fixes**
- Model allowlists now reject requests when the model is missing or
blank.
- Improved Rosenpass and WireGuard recovery after repeated handshake
failures.
- Improved relay connection handling so status and cleanup operations
remain responsive during stalled connections.
- Updated private service DNS zones to avoid unintended search-domain
behavior.

- **Tests**
- Expanded coverage for model allowlists, handshake recovery, relay
concurrency, and Bedrock routing.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->
2026-07-14 20:13:00 +02:00
Maycon Santos
277d8e4c53 [proxy] enforce model allowlist for URL-routed providers (Bedrock/Vertex) (#6764)
## Describe your changes

The Agent Network policy Guardrail "Model Allowlist" was not enforced
for providers whose model travels in the URL/path rather than the JSON
body — most visibly AWS Bedrock (reported in netbirdio/netbird#6751),
and the same class applies to Google Vertex.

Root cause: the `llm_guardrail` allowlist check **failed open**.
`evaluateAllowlist` returned allow whenever the request model was absent
from the metadata bag (`middleware.go`, `if !modelPresent { return nil
}`). The model is stamped upstream by `llm_request_parser`; for
body-routed providers (OpenAI/Anthropic) it comes from the JSON body,
but for path-routed providers the model is recovered only when the
request matches a recognized path shape (Bedrock
`/model/{id}/{invoke|converse|...}`, Vertex
`/v1/projects/.../publishers/.../models/...`). Any shape the parser did
not recognize reached the guardrail with no model and was allowed
regardless of the allowlist.

Fix (provider-agnostic): **fail closed**. When an allowlist is
configured and the model cannot be determined (absent or empty), the
request is denied `403` with a distinct `llm_policy.model_unknown`
reason. This closes the bypass for Bedrock, Vertex, and any future
URL-routed provider in one place. When no allowlist is configured,
behavior is unchanged.

The model allowlist is enforced solely in the proxy `llm_guardrail`;
management's `CheckLLMPolicyLimits` handles only token/budget caps, so
no management change is required.

## Issue ticket number and link

<https://github.com/netbirdio/netbird/discussions/6751>

## Stack

- \#6726 <!-- branch-stack -->
  - \#6764 :point\_left:

### Checklist

- [x] Is it a bug fix
- [ ] Is a typo/documentation fix
- [ ] Is a feature enhancement
- [ ] It is a refactor
- [x] Created tests that fail without the change (if possible)
- [x] This change does **not** modify the public API, gRPC protocols,
functionality behavior, CLI / service flags, or introduce a new feature
— **OR** I have discussed it with the NetBird team beforehand (link the
issue / Slack thread in the description). See
[CONTRIBUTING.md](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTING.md#discuss-changes-with-the-netbird-team-first).

> By submitting this pull request, you confirm that you have read and
agree to the terms of the [Contributor License
Agreement](https://github.com/netbirdio/netbird/blob/main/CONTRIBUTOR_LICENSE_AGREEMENT.md).

## Documentation

Select exactly one:

- [ ] I added/updated documentation for this change
- [x] Documentation is **not needed** for this change (explain why)

Bug fix that restores the documented allowlist behavior; no user-facing
surface changes.

### Docs PR URL (required if "docs added" is checked)

Paste the PR link from <https://github.com/netbirdio/docs> here:

<https://github.com/netbirdio/docs/pull/>\_\_

## Tests

- `llm_guardrail`: absent/empty model under a configured allowlist now
denies (`model_unknown`); empty allowlist still allows a missing model
(fail-closed only applies when a list is set); existing
allow/deny/case-insensitive cases retained.
- `llm_request_parser`: new parser→guardrail integration test drives
real **Bedrock** (`/model/{id}/invoke`) and **Vertex**
(`/v1/projects/.../models/...`) URL shapes and asserts allowed→200,
disallowed→403 (`model_blocked`), and an unrecognized Bedrock action→403
(`model_unknown`, the #6751 regression guard).

Note: a full through-tunnel e2e for the allowlist is intentionally
deferred — the agent-network e2e (`WaitProxyPeer`) is currently red on
`main`/`0.74.x` for an unrelated lazy-connection reason; it will be
added once that harness gate is fixed.
2026-07-14 19:03:01 +02:00
Eduard Gert
8e387b5dc5 [client] Guard browser-login cleanup on user close 2026-07-13 11:54:56 +02:00
Eduard Gert
abd7401a78 [client] Suppress browser-login:cancel on programmatic close 2026-07-13 11:36:57 +02:00
Eduard Gert
ecfb7ccf9c [client] Fix browser dialog not closing on renew session flow 2026-07-13 11:01:54 +02:00
Viktor Liu
3d87547d95 [client] Bump golang.org/x/crypto to v0.54.0 and Go toolchain to 1.25.12 (#6709) 2026-07-10 17:42:06 +02:00
Viktor Liu
4d4cc551fd [client] Recover from rosenpass key desync (#6714) 2026-07-10 17:38:29 +02:00
blaugrau90
08e46aa62f [management] fix: prevent reverse proxy domain from being pushed as DNS search domain (#6498)
SynthesizePrivateServiceZones created CustomZones for private services
without setting SearchDomainDisabled, causing the reverse proxy domain
to be injected as a search domain suffix on all connected peers.

This broke local hostname resolution: short names like 'myserver' were
expanded to 'myserver.app.example.com' (matching the reverse proxy
domain) before local DNS search domains were tried.

Fix: set SearchDomainDisabled: true so the zone is registered as a
match-only supplemental resolver, consistent with the NonAuthoritative
intent already expressed on the same zone.
2026-07-10 12:20:57 +02:00
Theodor Midtlien
7cd5c1732b [client] Fix hanging status command during relay dial (#6694)
* Add regression test for relay state lock
* Make connect not hold a lock in openConnVia
2026-07-08 14:36:42 +02:00
Maycon Santos
816d80602f [client] Update gopsutil to v4 (#6688) 2026-07-08 10:15:31 +02:00
43 changed files with 3153 additions and 393 deletions

View File

@@ -121,6 +121,7 @@ type Manager struct {
udpTracker *conntrack.UDPTracker
icmpTracker *conntrack.ICMPTracker
tcpTracker *conntrack.TCPTracker
fragments *fragmentTracker
forwarder atomic.Pointer[forwarder.Forwarder]
pendingCapture atomic.Pointer[forwarder.PacketCapture]
logger *nblog.Logger
@@ -183,6 +184,41 @@ func (d *decoder) decodePacket(data []byte) error {
}
}
// decodeTransport decodes the transport header of a first fragment (which
// gopacket leaves undecoded) into the decoder and appends its layer type to
// decoded, so the ACL pipeline can evaluate it like a normal packet. It returns
// false if the protocol is unsupported or the header is truncated.
func (d *decoder) decodeTransport(proto layers.IPProtocol, payload []byte) bool {
var l4 gopacket.DecodingLayer
var layerType gopacket.LayerType
var minLen int
switch proto {
case layers.IPProtocolTCP:
l4, layerType, minLen = &d.tcp, layers.LayerTypeTCP, 20
case layers.IPProtocolUDP:
l4, layerType, minLen = &d.udp, layers.LayerTypeUDP, 8
case layers.IPProtocolICMPv4:
l4, layerType, minLen = &d.icmp4, layers.LayerTypeICMPv4, 8
case layers.IPProtocolICMPv6:
l4, layerType, minLen = &d.icmp6, layers.LayerTypeICMPv6, 8
default:
return false
}
// Reject a fragment too small to hold the full transport header before
// decoding: it can't be ACL-evaluated (tiny-fragment attack), and skipping
// the decode avoids gopacket allocating an error on the drop path.
if len(payload) < minLen {
return false
}
if err := l4.DecodeFromBytes(payload, gopacket.NilDecodeFeedback); err != nil {
return false
}
d.decoded = append(d.decoded, layerType)
return true
}
// Create userspace firewall manager constructor
func Create(iface common.IFaceMapper, disableServerRoutes bool, flowLogger nftypes.FlowLogger, mtu uint16) (*Manager, error) {
return create(iface, nil, disableServerRoutes, flowLogger, mtu)
@@ -286,6 +322,8 @@ func create(iface common.IFaceMapper, nativeFirewall firewall.Manager, disableSe
if err := m.localipmanager.UpdateLocalIPs(iface); err != nil {
return nil, fmt.Errorf("update local IPs: %w", err)
}
m.fragments = newFragmentTracker(m.logger)
if disableConntrack {
log.Info("conntrack is disabled")
} else {
@@ -299,6 +337,7 @@ func create(iface common.IFaceMapper, nativeFirewall firewall.Manager, disableSe
}
}
if err := iface.SetFilter(m); err != nil {
m.fragments.Close()
return nil, fmt.Errorf("set filter: %w", err)
}
return m, nil
@@ -694,6 +733,10 @@ func (m *Manager) resetState() {
m.tcpTracker.Close()
}
if m.fragments != nil {
m.fragments.Close()
}
if fwder := m.forwarder.Load(); fwder != nil {
fwder.SetCapture(nil)
fwder.Stop()
@@ -1046,19 +1089,20 @@ func (m *Manager) filterInbound(packetData []byte, size int) bool {
return true
}
// TODO: pass fragments of routed packets to forwarder
// gopacket does not decode the transport header of any IP fragment, so
// fragments take a dedicated path: the first fragment's header is decoded
// and ACL-evaluated here, and the remaining fragments inherit its verdict.
if fragment {
if m.logger.Enabled(nblog.LevelTrace) {
if d.decoded[0] == layers.LayerTypeIPv4 {
m.logger.Trace4("packet is a fragment: src=%v dst=%v id=%v flags=%v",
srcIP, dstIP, d.ip4.Id, d.ip4.Flags)
} else {
m.logger.Trace2("packet is an IPv6 fragment: src=%v dst=%v", srcIP, dstIP)
}
}
return false
return m.filterInboundFragment(d, srcIP, dstIP, size)
}
return m.filterInboundDecoded(d, srcIP, dstIP, packetData, size)
}
// filterInboundDecoded runs the ACL, DNAT and conntrack pipeline on a fully
// decoded (non-fragment) inbound packet. It returns true if the packet should
// be dropped.
func (m *Manager) filterInboundDecoded(d *decoder, srcIP, dstIP netip.Addr, packetData []byte, size int) bool {
// TODO: optimize port DNAT by caching matched rules in conntrack
if translated := m.translateInboundPortDNAT(packetData, d, srcIP, dstIP); translated {
// Re-decode after port DNAT translation to update port information
@@ -1089,33 +1133,226 @@ func (m *Manager) filterInbound(packetData []byte, size int) bool {
return m.handleRoutedTraffic(d, srcIP, dstIP, packetData, size)
}
// fragmentMeta holds the reassembly identity and layout of an IP fragment,
// extracted uniformly for IPv4 and IPv6.
type fragmentMeta struct {
key fragmentKey
// offset is the fragment offset in 8-byte units (zero for the first
// fragment).
offset uint16
// moreFragments is the More Fragments bit. A first fragment with it unset is
// an IPv6 atomic fragment (a complete datagram, RFC 6946): it has no trailing
// fragments to inherit a verdict, so it must not be recorded.
moreFragments bool
proto layers.IPProtocol
// l4payload is the fragmentable payload of this fragment. For the first
// fragment it starts with the transport header.
l4payload []byte
// headerEndOctets is the first fragment's payload length in 8-byte units:
// the smallest offset a trailing fragment may start at without overlapping
// the inspected transport header.
headerEndOctets uint16
}
// fragmentMetadata extracts the fragment identity and layout from a decoded IP
// fragment. It returns false for fragments it can't interpret (e.g. an IPv6
// fragment header shorter than 8 bytes), which are then dropped.
func fragmentMetadata(d *decoder, srcIP, dstIP netip.Addr) (fragmentMeta, bool) {
switch d.decoded[0] {
case layers.LayerTypeIPv4:
payload := d.ip4.Payload
return fragmentMeta{
key: fragmentKey{srcIP: srcIP, dstIP: dstIP, id: uint32(d.ip4.Id), proto: uint8(d.ip4.Protocol)},
offset: d.ip4.FragOffset,
moreFragments: d.ip4.Flags&layers.IPv4MoreFragments != 0,
proto: d.ip4.Protocol,
l4payload: payload,
headerEndOctets: octets(len(payload)),
}, true
case layers.LayerTypeIPv6:
// IPv6 fragment extension header: 8 bytes, followed by the fragmentable
// payload. Layout: next header (1), reserved (1), offset+flags (2), id (4).
payload := d.ip6.Payload
if len(payload) < 8 {
return fragmentMeta{}, false
}
nextHeader := layers.IPProtocol(payload[0])
offsetFlags := binary.BigEndian.Uint16(payload[2:4])
id := binary.BigEndian.Uint32(payload[4:8])
l4 := payload[8:]
return fragmentMeta{
key: fragmentKey{srcIP: srcIP, dstIP: dstIP, id: id, proto: uint8(nextHeader)},
offset: offsetFlags >> 3,
moreFragments: offsetFlags&1 != 0,
proto: nextHeader,
l4payload: l4,
headerEndOctets: octets(len(l4)),
}, true
default:
return fragmentMeta{}, false
}
}
// octets rounds a byte length up to whole 8-byte units, the granularity of the
// IP fragment offset field.
func octets(nbytes int) uint16 {
return uint16((nbytes + 7) / 8)
}
// filterInboundFragment decides the fate of an IP fragment. gopacket stops
// decoding at the network layer for every fragment, so the first fragment's
// transport header is decoded and ACL-evaluated here and its verdict recorded;
// the remaining (headerless) fragments inherit that verdict. Anything that
// cannot be tied to an allowed, non-overlapping first fragment is dropped.
func (m *Manager) filterInboundFragment(d *decoder, srcIP, dstIP netip.Addr, size int) bool {
meta, ok := fragmentMetadata(d, srcIP, dstIP)
if !ok {
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace2("dropping unsupported fragment: src=%v dst=%v", srcIP, dstIP)
}
return true
}
if meta.offset != 0 {
return m.filterTrailingFragment(meta, srcIP, dstIP)
}
// A new first fragment supersedes any recorded verdict for this datagram, so
// a re-sent or overlapping offset-zero fragment can't inherit the old one.
m.fragments.poison(meta.key)
// First fragment: decode its transport header so the ACL can evaluate it. A
// decode failure means the fragment is too small to hold the full transport
// header (RFC 1858 §3 tiny-fragment attack); it can't be evaluated, so drop it.
if !d.decodeTransport(meta.proto, meta.l4payload) {
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace3("dropping first fragment without full L4 header: src=%v dst=%v id=%v",
srcIP, dstIP, meta.key.id)
}
return true
}
return m.filterFirstFragment(d, meta, srcIP, dstIP, size)
}
// filterTrailingFragment applies a recorded first-fragment verdict to a
// non-first fragment.
func (m *Manager) filterTrailingFragment(meta fragmentMeta, srcIP, dstIP netip.Addr) bool {
switch m.fragments.verdict(meta.key, meta.offset) {
case fragmentAllow:
return false
case fragmentOverlap:
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace3("dropping overlapping fragment rewriting inspected header: src=%v dst=%v id=%v",
srcIP, dstIP, meta.key.id)
}
return true
default:
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace3("dropping fragment with no allowed first fragment: src=%v dst=%v id=%v",
srcIP, dstIP, meta.key.id)
}
return true
}
}
// filterFirstFragment runs the verdict part of the inbound pipeline on a first
// fragment with its transport header decoded. It mirrors filterInboundDecoded
// but skips DNAT (port rewriting on fragments is unsupported) and forwarder
// injection (fragments are left to the stack to reassemble, not forwarded).
// Allowed fragments have their verdict recorded so the datagram's trailing
// fragments inherit it.
func (m *Manager) filterFirstFragment(d *decoder, meta fragmentMeta, srcIP, dstIP netip.Addr, size int) bool {
if m.stateful && m.isValidTrackedConnection(d, srcIP, dstIP, size) {
m.recordFirstFragment(meta)
return false
}
if m.localipmanager.IsLocalIP(dstIP) {
ruleID, blocked := m.peerACLsBlock(srcIP, d, nil)
if blocked {
m.storeDropFlow("Dropping local first fragment (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
d, srcIP, dstIP, ruleID, size)
return true
}
m.trackInbound(d, srcIP, dstIP, ruleID, size)
m.recordFirstFragment(meta)
return false
}
if !m.routingEnabled.Load() {
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace2("Dropping routed fragment (routing disabled): src=%s dst=%s", srcIP, dstIP)
}
return true
}
if m.nativeRouter.Load() {
m.trackInbound(d, srcIP, dstIP, nil, size)
m.recordFirstFragment(meta)
return false
}
// TODO: pass fragments of routed packets to the forwarder; until then
// allowed routed fragments go to the native stack.
srcPort, dstPort := getPortsFromPacket(d)
ruleID, pass := m.routeACLsPass(srcIP, dstIP, d.decoded[1], srcPort, dstPort)
if !pass {
m.storeDropFlow("Dropping routed first fragment (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
d, srcIP, dstIP, ruleID, size)
return true
}
m.recordFirstFragment(meta)
return false
}
// recordFirstFragment caches an allowed first fragment's verdict for its
// trailing fragments to inherit. Atomic fragments (no More Fragments bit) are
// complete datagrams with no trailing fragments, so they are not cached and
// cannot exhaust the verdict table.
func (m *Manager) recordFirstFragment(meta fragmentMeta) {
if !meta.moreFragments {
return
}
m.fragments.recordAllowed(meta.key, meta.headerEndOctets)
}
// storeDropFlow logs and records a netflow drop event for an inbound packet
// denied by the ACLs. msg is the trace format taking rule id, protocol, source
// and destination.
func (m *Manager) storeDropFlow(msg string, d *decoder, srcIP, dstIP netip.Addr, ruleID []byte, size int) {
pnum := getProtocolFromPacket(d)
srcPort, dstPort := getPortsFromPacket(d)
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace6(msg, ruleID, pnum, srcIP, srcPort, dstIP, dstPort)
}
m.flowLogger.StoreEvent(nftypes.EventFields{
FlowID: uuid.New(),
Type: nftypes.TypeDrop,
RuleID: ruleID,
Direction: nftypes.Ingress,
Protocol: pnum,
SourceIP: srcIP,
DestIP: dstIP,
SourcePort: srcPort,
DestPort: dstPort,
// TODO: icmp type/code
RxPackets: 1,
RxBytes: uint64(size),
})
}
// handleLocalTraffic handles local traffic.
// If it returns true, the packet should be dropped.
func (m *Manager) handleLocalTraffic(d *decoder, srcIP, dstIP netip.Addr, packetData []byte, size int) bool {
ruleID, blocked := m.peerACLsBlock(srcIP, d, packetData)
if blocked {
pnum := getProtocolFromPacket(d)
srcPort, dstPort := getPortsFromPacket(d)
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace6("Dropping local packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
ruleID, pnum, srcIP, srcPort, dstIP, dstPort)
}
m.flowLogger.StoreEvent(nftypes.EventFields{
FlowID: uuid.New(),
Type: nftypes.TypeDrop,
RuleID: ruleID,
Direction: nftypes.Ingress,
Protocol: pnum,
SourceIP: srcIP,
DestIP: dstIP,
SourcePort: srcPort,
DestPort: dstPort,
// TODO: icmp type/code
RxPackets: 1,
RxBytes: uint64(size),
})
m.storeDropFlow("Dropping local packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
d, srcIP, dstIP, ruleID, size)
return true
}
@@ -1168,27 +1405,8 @@ func (m *Manager) handleRoutedTraffic(d *decoder, srcIP, dstIP netip.Addr, packe
ruleID, pass := m.routeACLsPass(srcIP, dstIP, protoLayer, srcPort, dstPort)
if !pass {
proto := getProtocolFromPacket(d)
if m.logger.Enabled(nblog.LevelTrace) {
m.logger.Trace6("Dropping routed packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
ruleID, proto, srcIP, srcPort, dstIP, dstPort)
}
m.flowLogger.StoreEvent(nftypes.EventFields{
FlowID: uuid.New(),
Type: nftypes.TypeDrop,
RuleID: ruleID,
Direction: nftypes.Ingress,
Protocol: proto,
SourceIP: srcIP,
DestIP: dstIP,
SourcePort: srcPort,
DestPort: dstPort,
// TODO: icmp type/code
RxPackets: 1,
RxBytes: uint64(size),
})
m.storeDropFlow("Dropping routed packet (ACL denied): rule_id=%s proto=%v src=%s:%d dst=%s:%d",
d, srcIP, dstIP, ruleID, size)
return true
}

View File

@@ -5,7 +5,9 @@ import (
"fmt"
"net"
"net/netip"
"os"
"runtime"
"strconv"
"sync"
"time"
@@ -31,6 +33,11 @@ const (
defaultMaxInFlight = 1024
iosReceiveWindow = 16384
iosMaxInFlight = 256
// envForceTCPRACK overrides the platform default for gVisor's RACK loss
// detection. Set to a truthy value to force RACK on, or a falsy value to
// force it off, on any platform.
envForceTCPRACK = "NB_FORCE_TCP_RACK"
)
type Forwarder struct {
@@ -152,6 +159,8 @@ func New(iface common.IFaceMapper, logger *nblog.Logger, flowLogger nftypes.Flow
maxInFlight = iosMaxInFlight
}
configureTCPRecovery(s)
tcpForwarder := tcp.NewForwarder(s, receiveWindow, maxInFlight, f.handleTCP)
s.SetTransportProtocolHandler(tcp.ProtocolNumber, tcpForwarder.HandlePacket)
@@ -466,3 +475,31 @@ func probeRawICMP(network, addr string, logger *nblog.Logger) bool {
logger.Debug1("forwarder: raw %s socket access available", network)
return true
}
// configureTCPRecovery disables gVisor's RACK loss detection on Windows, where
// it interacts poorly with the host and collapses throughput on routed TCP
// connections (gVisor issue #9778). Other platforms keep the default. The
// EnvForceTCPRACK environment variable overrides the platform default.
func configureTCPRecovery(s *stack.Stack) {
disableRACK := runtime.GOOS == "windows"
if val := os.Getenv(envForceTCPRACK); val != "" {
force, err := strconv.ParseBool(val)
if err != nil {
log.Warnf("parse %s: %v", envForceTCPRACK, err)
} else {
disableRACK = !force
}
}
if !disableRACK {
return
}
opt := tcpip.TCPRecovery(0)
if err := s.SetTransportProtocolOption(tcp.ProtocolNumber, &opt); err != nil {
log.Warnf("disable TCP RACK loss detection: %v", err)
return
}
log.Info("forwarder: TCP RACK loss detection disabled")
}

View File

@@ -0,0 +1,204 @@
package uspfilter
import (
"context"
"net/netip"
"os"
"strconv"
"sync"
"time"
nblog "github.com/netbirdio/netbird/client/firewall/uspfilter/log"
)
const (
// defaultFragmentTimeout bounds how long a first-fragment verdict is kept
// while the remaining fragments arrive. It mirrors the Linux IP reassembly
// timeout (net.ipv4.ipfrag_time).
defaultFragmentTimeout = 30 * time.Second
// fragmentCleanupInterval is how often expired verdicts are purged.
fragmentCleanupInterval = 10 * time.Second
// defaultMaxFragmentEntries caps the number of concurrently tracked
// fragmented datagrams. The table stays bounded because each datagram is a
// single small entry regardless of how many fragments it is split into, and
// the 13-bit IPv4 fragment-offset field limits any datagram to 64 KiB.
defaultMaxFragmentEntries = 16384
// EnvFragmentMaxEntries overrides defaultMaxFragmentEntries.
EnvFragmentMaxEntries = "NB_FRAGMENT_MAX_ENTRIES"
)
// fragmentVerdict is the decision for a trailing (headerless) fragment.
type fragmentVerdict int
const (
// fragmentDeny drops the fragment: no allowed first fragment is on record.
fragmentDeny fragmentVerdict = iota
// fragmentAllow passes the fragment: it belongs to an allowed datagram and
// does not overlap the already-inspected transport header.
fragmentAllow
// fragmentOverlap drops the fragment and poisons its datagram: it overlaps
// the transport header the ACL inspected (RFC 1858 §4, RFC 3128; RFC 5722
// requires discarding the whole datagram on overlap for IPv6).
fragmentOverlap
)
// fragmentKey identifies a fragmented datagram. It matches the RFC 791 / RFC
// 8200 reassembly key: source, destination, protocol and identification. The id
// is 32-bit to hold both the IPv4 (16-bit) and IPv6 (32-bit) identification.
type fragmentKey struct {
srcIP netip.Addr
dstIP netip.Addr
id uint32
proto uint8
}
// fragmentEntry records the verdict of an allowed first fragment.
type fragmentEntry struct {
// headerEndOctets is the offset, in 8-byte units, at which the first
// fragment's payload ended. A trailing fragment starting before this
// overlaps bytes the ACL already inspected and is rejected.
headerEndOctets uint16
// recordedAt is when the first fragment was accepted. The verdict expires a
// fixed timeout later and is not refreshed, mirroring the kernel reassembly
// timer so a trailing-fragment flood can't keep a datagram alive.
recordedAt time.Time
}
// fragmentTracker records the ACL verdict of a datagram's first fragment so the
// remaining fragments, which carry no L4 header, can inherit the decision
// without reassembling the datagram. Only allowed first fragments are stored;
// anything that cannot be tied to an allowed, non-overlapping first fragment is
// dropped (fail closed).
type fragmentTracker struct {
logger *nblog.Logger
mutex sync.Mutex
entries map[fragmentKey]fragmentEntry
timeout time.Duration
// maxEntries caps the table; atCapacity dedups the capacity warning until
// the table drains below the cap again.
maxEntries int
atCapacity bool
cleanupTicker *time.Ticker
cancel context.CancelFunc
}
func newFragmentTracker(logger *nblog.Logger) *fragmentTracker {
ctx, cancel := context.WithCancel(context.Background())
t := &fragmentTracker{
logger: logger,
entries: make(map[fragmentKey]fragmentEntry),
timeout: defaultFragmentTimeout,
maxEntries: fragmentMaxEntries(logger),
cleanupTicker: time.NewTicker(fragmentCleanupInterval),
cancel: cancel,
}
go t.cleanupRoutine(ctx)
return t
}
func fragmentMaxEntries(logger *nblog.Logger) int {
v := os.Getenv(EnvFragmentMaxEntries)
if v == "" {
return defaultMaxFragmentEntries
}
n, err := strconv.Atoi(v)
if err != nil || n <= 0 {
logger.Warn2("invalid %s=%q, using default", EnvFragmentMaxEntries, v)
return defaultMaxFragmentEntries
}
return n
}
// recordAllowed stores the verdict of an allowed first fragment. headerEndOctets
// is the first fragment's payload length in 8-byte units. When the table is full
// the record is dropped, which fails closed: the datagram's trailing fragments
// will be denied.
func (t *fragmentTracker) recordAllowed(key fragmentKey, headerEndOctets uint16) {
t.mutex.Lock()
defer t.mutex.Unlock()
if t.entries == nil {
return
}
if _, ok := t.entries[key]; !ok && len(t.entries) >= t.maxEntries {
if !t.atCapacity {
t.atCapacity = true
t.logger.Warn2("fragment verdict table at capacity (%d/%d): trailing fragments of new datagrams will be dropped",
len(t.entries), t.maxEntries)
}
return
}
t.entries[key] = fragmentEntry{
headerEndOctets: headerEndOctets,
recordedAt: time.Now(),
}
}
// poison drops any recorded verdict for a datagram, so its later fragments are
// denied until a new allowed first fragment is recorded. Called on every
// offset-zero fragment to defeat offset-zero overlap rewrites (RFC 3128).
func (t *fragmentTracker) poison(key fragmentKey) {
t.mutex.Lock()
defer t.mutex.Unlock()
delete(t.entries, key)
}
// verdict decides the fate of a trailing fragment at fragOffsetOctets (the IPv4
// fragment offset, in 8-byte units). A fragment overlapping the inspected
// header poisons the datagram: the entry is removed so all further fragments of
// that datagram are denied too.
func (t *fragmentTracker) verdict(key fragmentKey, fragOffsetOctets uint16) fragmentVerdict {
t.mutex.Lock()
defer t.mutex.Unlock()
entry, ok := t.entries[key]
if !ok {
return fragmentDeny
}
if time.Since(entry.recordedAt) > t.timeout {
delete(t.entries, key)
return fragmentDeny
}
if fragOffsetOctets < entry.headerEndOctets {
delete(t.entries, key)
return fragmentOverlap
}
return fragmentAllow
}
func (t *fragmentTracker) cleanupRoutine(ctx context.Context) {
defer t.cleanupTicker.Stop()
for {
select {
case <-t.cleanupTicker.C:
t.cleanup()
case <-ctx.Done():
return
}
}
}
func (t *fragmentTracker) cleanup() {
t.mutex.Lock()
defer t.mutex.Unlock()
for key, entry := range t.entries {
if time.Since(entry.recordedAt) > t.timeout {
delete(t.entries, key)
}
}
if len(t.entries) < t.maxEntries {
t.atCapacity = false
}
}
// Close stops the cleanup routine and releases resources.
func (t *fragmentTracker) Close() {
t.cancel()
t.mutex.Lock()
t.entries = nil
t.mutex.Unlock()
}

View File

@@ -0,0 +1,115 @@
package uspfilter
import (
"encoding/binary"
"testing"
)
// benchFilterInbound drives filterInbound over a fixed packet in a tight loop.
// Packets are built once, outside the timed region, so the benchmark measures
// only pipeline cost, which is what an attacker can amplify.
func benchFilterInbound(b *testing.B, pkt []byte) {
b.Helper()
b.ReportAllocs()
b.SetBytes(int64(len(pkt)))
b.ResetTimer()
for i := 0; i < b.N; i++ {
m := benchManager
m.filterInbound(pkt, len(pkt))
}
}
// benchManager is a package-level manager reused across fragment benchmarks so
// setup cost stays out of the timed region.
var benchManager *Manager
func setupBenchManager(b *testing.B) *Manager {
b.Helper()
m := newFragmentTestManager(b)
allowUDP(b, m, 8080)
// Disable conntrack so the allowed-first-fragment path measures transport
// decode + ACL every iteration instead of matching the connection tracked
// on the first iteration.
m.stateful = false
benchManager = m
return m
}
// BenchmarkInbound_NormalPacket is the baseline: a full, non-fragmented UDP
// packet that passes the ACL. Fragment paths should stay comparable to this.
func BenchmarkInbound_NormalPacket(b *testing.B) {
setupBenchManager(b)
pkt := normalUDPPacket(b, 8080, 32)
benchFilterInbound(b, pkt)
}
// BenchmarkInbound_FirstFragmentAllowed measures the first-fragment path:
// transport decode + ACL evaluation + verdict record.
func BenchmarkInbound_FirstFragmentAllowed(b *testing.B) {
setupBenchManager(b)
pkt := firstFragmentUDP(b, 0x2000, 8080, 32)
benchFilterInbound(b, pkt)
}
// BenchmarkInbound_TrailingFragmentAllowed measures the common trailing-fragment
// path: a single map lookup after the first fragment is on record.
func BenchmarkInbound_TrailingFragmentAllowed(b *testing.B) {
m := setupBenchManager(b)
first := firstFragmentUDP(b, 0x3000, 8080, 32)
m.filterInbound(first, len(first))
pkt := trailingFragment(b, 0x3000, 5, false, 24)
benchFilterInbound(b, pkt)
}
// BenchmarkInbound_TrailingFragmentNoFirst is the primary DoS vector: an
// attacker floods trailing fragments with no first fragment on record. Each is
// a map miss and must be cheap.
func BenchmarkInbound_TrailingFragmentNoFirst(b *testing.B) {
setupBenchManager(b)
pkt := trailingFragment(b, 0x4000, 185, false, 40)
benchFilterInbound(b, pkt)
}
// BenchmarkInbound_TinyFirstFragment measures the tiny-fragment drop path: a
// first fragment too small to decode a transport header.
func BenchmarkInbound_TinyFirstFragment(b *testing.B) {
setupBenchManager(b)
pkt := trailingFragment(b, 0x5000, 0, true, 4)
benchFilterInbound(b, pkt)
}
// BenchmarkInbound_TrailingFragmentDistinctIDs is the worst case for the
// verdict table: an attacker varies the datagram id on every packet so no first
// fragment ever matches. Verdict lookups always miss and nothing is recorded,
// so the table cannot grow. Each iteration rewrites the id field in place.
func BenchmarkInbound_TrailingFragmentDistinctIDs(b *testing.B) {
setupBenchManager(b)
pkt := trailingFragment(b, 0x6000, 185, false, 40)
m := benchManager
b.ReportAllocs()
b.SetBytes(int64(len(pkt)))
b.ResetTimer()
for i := 0; i < b.N; i++ {
// IPv4 identification field is at bytes 4:6.
binary.BigEndian.PutUint16(pkt[4:6], uint16(i))
m.filterInbound(pkt, len(pkt))
}
}
// BenchmarkInbound_FirstFragmentDistinctIDs measures sustained first-fragment
// pressure with distinct ids: transport decode + ACL + verdict insert until the
// table caps, exercising the map growth and capacity guard.
func BenchmarkInbound_FirstFragmentDistinctIDs(b *testing.B) {
setupBenchManager(b)
pkt := firstFragmentUDP(b, 0x7000, 8080, 32)
m := benchManager
b.ReportAllocs()
b.SetBytes(int64(len(pkt)))
b.ResetTimer()
for i := 0; i < b.N; i++ {
binary.BigEndian.PutUint16(pkt[4:6], uint16(i))
m.filterInbound(pkt, len(pkt))
}
}

View File

@@ -0,0 +1,554 @@
package uspfilter
import (
"encoding/binary"
"net"
"net/netip"
"testing"
"time"
"github.com/google/gopacket"
"github.com/google/gopacket/layers"
"github.com/stretchr/testify/require"
fw "github.com/netbirdio/netbird/client/firewall/manager"
nbiface "github.com/netbirdio/netbird/client/iface"
"github.com/netbirdio/netbird/client/iface/device"
"github.com/netbirdio/netbird/client/iface/wgaddr"
)
const (
fragTestSrc = "100.10.0.1"
fragTestDst = "100.10.0.100"
fragTestSrcV6 = "fd00::1"
fragTestDstV6 = "fd00::100"
)
func newFragmentTestManager(tb testing.TB) *Manager {
tb.Helper()
ifaceMock := &IFaceMock{
SetFilterFunc: func(device.PacketFilter) error { return nil },
AddressFunc: func() wgaddr.Address {
return wgaddr.Address{
IP: netip.MustParseAddr(fragTestDst),
Network: netip.MustParsePrefix("100.10.0.0/16"),
IPv6: netip.MustParseAddr(fragTestDstV6),
IPv6Net: netip.MustParsePrefix("fd00::/64"),
}
},
}
m, err := Create(ifaceMock, false, flowLogger, nbiface.DefaultMTU)
require.NoError(tb, err)
require.NoError(tb, m.UpdateLocalIPs())
tb.Cleanup(func() { require.NoError(tb, m.Close(nil)) })
return m
}
// firstFragmentUDPTo builds the first fragment of a fragmented UDP datagram to
// the given destination: it carries the full UDP header plus payloadLen bytes
// of data, with the More Fragments flag set and offset zero.
func firstFragmentUDPTo(tb testing.TB, dst string, id uint16, dstPort uint16, payloadLen int) []byte {
tb.Helper()
ip := &layers.IPv4{
Version: 4,
TTL: 64,
Id: id,
Protocol: layers.IPProtocolUDP,
SrcIP: net.ParseIP(fragTestSrc),
DstIP: net.ParseIP(dst),
Flags: layers.IPv4MoreFragments,
}
udp := &layers.UDP{SrcPort: 40000, DstPort: layers.UDPPort(dstPort)}
require.NoError(tb, udp.SetNetworkLayerForChecksum(ip))
buf := gopacket.NewSerializeBuffer()
opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, udp, gopacket.Payload(make([]byte, payloadLen))))
return buf.Bytes()
}
func firstFragmentUDP(tb testing.TB, id uint16, dstPort uint16, payloadLen int) []byte {
tb.Helper()
return firstFragmentUDPTo(tb, fragTestDst, id, dstPort, payloadLen)
}
// firstFragmentTCP builds the first fragment of a fragmented TCP datagram: the
// full 20-byte TCP header plus 12 bytes of data, with the More Fragments flag
// set and offset zero.
func firstFragmentTCP(tb testing.TB, id uint16, dstPort uint16) []byte {
tb.Helper()
ip := &layers.IPv4{
Version: 4,
TTL: 64,
Id: id,
Protocol: layers.IPProtocolTCP,
SrcIP: net.ParseIP(fragTestSrc),
DstIP: net.ParseIP(fragTestDst),
Flags: layers.IPv4MoreFragments,
}
tcp := &layers.TCP{SrcPort: 40000, DstPort: layers.TCPPort(dstPort), SYN: true, Window: 64240}
require.NoError(tb, tcp.SetNetworkLayerForChecksum(ip))
buf := gopacket.NewSerializeBuffer()
opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, tcp, gopacket.Payload(make([]byte, 12))))
return buf.Bytes()
}
// trailingFragmentTo builds a non-first fragment to the given destination: an
// IPv4 header at the given fragment offset (in 8-byte units) carrying raw
// payload and no L4 header.
func trailingFragmentTo(tb testing.TB, dst string, proto layers.IPProtocol, id uint16, fragOffsetOctets uint16, moreFragments bool, payloadLen int) []byte {
tb.Helper()
ip := &layers.IPv4{
Version: 4,
TTL: 64,
Id: id,
Protocol: proto,
SrcIP: net.ParseIP(fragTestSrc),
DstIP: net.ParseIP(dst),
FragOffset: fragOffsetOctets,
}
if moreFragments {
ip.Flags = layers.IPv4MoreFragments
}
buf := gopacket.NewSerializeBuffer()
opts := gopacket.SerializeOptions{FixLengths: true}
require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, gopacket.Payload(make([]byte, payloadLen))))
return buf.Bytes()
}
func trailingFragment(tb testing.TB, id uint16, fragOffsetOctets uint16, moreFragments bool, payloadLen int) []byte {
tb.Helper()
return trailingFragmentTo(tb, fragTestDst, layers.IPProtocolUDP, id, fragOffsetOctets, moreFragments, payloadLen)
}
// outboundUDPPacket builds a complete outbound UDP packet from the local
// address, used to establish conntrack state for reply-direction tests.
func outboundUDPPacket(tb testing.TB, srcPort, dstPort uint16) []byte {
tb.Helper()
ip := &layers.IPv4{
Version: 4,
TTL: 64,
Id: 1,
Protocol: layers.IPProtocolUDP,
SrcIP: net.ParseIP(fragTestDst),
DstIP: net.ParseIP(fragTestSrc),
}
udp := &layers.UDP{SrcPort: layers.UDPPort(srcPort), DstPort: layers.UDPPort(dstPort)}
require.NoError(tb, udp.SetNetworkLayerForChecksum(ip))
buf := gopacket.NewSerializeBuffer()
opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, udp, gopacket.Payload(make([]byte, 16))))
return buf.Bytes()
}
// normalUDPPacket builds a complete, non-fragmented UDP packet for baseline
// comparisons against the fragment paths.
func normalUDPPacket(tb testing.TB, dstPort uint16, payloadLen int) []byte {
tb.Helper()
ip := &layers.IPv4{
Version: 4,
TTL: 64,
Id: 1,
Protocol: layers.IPProtocolUDP,
SrcIP: net.ParseIP(fragTestSrc),
DstIP: net.ParseIP(fragTestDst),
}
udp := &layers.UDP{SrcPort: 40000, DstPort: layers.UDPPort(dstPort)}
require.NoError(tb, udp.SetNetworkLayerForChecksum(ip))
buf := gopacket.NewSerializeBuffer()
opts := gopacket.SerializeOptions{ComputeChecksums: true, FixLengths: true}
require.NoError(tb, gopacket.SerializeLayers(buf, opts, ip, udp, gopacket.Payload(make([]byte, payloadLen))))
return buf.Bytes()
}
func allowUDP(tb testing.TB, m *Manager, dstPort uint16) {
tb.Helper()
_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolUDP, nil,
&fw.Port{Values: []uint16{dstPort}}, fw.ActionAccept, "")
require.NoError(tb, err)
}
// TestFragment_TrailingWithoutFirstDropped is the core bypass repro: a trailing
// fragment with no allowed first fragment on record must be dropped. Before the
// fix, filterInbound returned false (allow) for any fragment.
func TestFragment_TrailingWithoutFirstDropped(t *testing.T) {
m := newFragmentTestManager(t)
frag := trailingFragment(t, 0x1234, 185, false, 40)
require.True(t, m.filterInbound(frag, len(frag)),
"trailing fragment without an allowed first fragment must be dropped")
}
// TestFragment_AllowedFirstPassesTrailing verifies that once a first fragment
// passes the ACL, its trailing fragments inherit the allow verdict.
func TestFragment_AllowedFirstPassesTrailing(t *testing.T) {
m := newFragmentTestManager(t)
allowUDP(t, m, 8080)
// First fragment: UDP header (8) + 32 payload = 40 octets -> headerEnd = 5.
first := firstFragmentUDP(t, 0x2222, 8080, 32)
require.False(t, m.filterInbound(first, len(first)),
"allowed first fragment should pass and be recorded")
trailing := trailingFragment(t, 0x2222, 5, false, 24)
require.False(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment of an allowed datagram should pass")
}
// TestFragment_DeniedFirstDropsTrailing verifies that a first fragment blocked
// by the ACL leaves no verdict, so its trailing fragments are dropped.
func TestFragment_DeniedFirstDropsTrailing(t *testing.T) {
m := newFragmentTestManager(t)
// No accept rule: local traffic defaults to deny.
first := firstFragmentUDP(t, 0x3333, 9999, 32)
require.True(t, m.filterInbound(first, len(first)),
"first fragment to a blocked port should be dropped by the ACL")
trailing := trailingFragment(t, 0x3333, 5, false, 24)
require.True(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment of a denied datagram must be dropped")
}
// TestFragment_OverlappingHeaderDropped covers the RFC 1858 §4 / RFC 3128
// overlapping-fragment rewrite: a trailing fragment starting inside the range
// the ACL already inspected is dropped and poisons the datagram. TCP is used so
// the overlap lands on real header bytes (the flags at byte 13).
func TestFragment_OverlappingHeaderDropped(t *testing.T) {
m := newFragmentTestManager(t)
_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolTCP, nil,
&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
require.NoError(t, err)
// First fragment: TCP header (20) + 12 data = 32 bytes -> headerEnd = 4 octets.
first := firstFragmentTCP(t, 0x4444, 8080)
require.False(t, m.filterInbound(first, len(first)))
// Overlapping fragment at offset 1 (byte 8) falls inside the inspected TCP
// header, so it could rewrite the flags or port on reassembly.
overlap := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x4444, 1, true, 32)
require.True(t, m.filterInbound(overlap, len(overlap)),
"fragment overlapping the inspected header must be dropped")
// The datagram is now poisoned: a later, non-overlapping fragment is also
// dropped because the verdict was removed.
later := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x4444, 4, false, 24)
require.True(t, m.filterInbound(later, len(later)),
"fragments after an overlap must be dropped (datagram poisoned)")
}
// TestFragment_OffsetZeroOverlapPoisons covers the RFC 3128 offset-zero rewrite:
// an allowed first fragment followed by a denied offset-zero fragment for the
// same datagram must not leave the earlier allow verdict in place.
func TestFragment_OffsetZeroOverlapPoisons(t *testing.T) {
m := newFragmentTestManager(t)
allowUDP(t, m, 8080)
allowed := firstFragmentUDP(t, 0x5A5A, 8080, 32)
require.False(t, m.filterInbound(allowed, len(allowed)),
"allowed first fragment should pass and be recorded")
// A second offset-zero fragment to a denied port supersedes the datagram's
// verdict; it is dropped and must not leave the allow in place.
denied := firstFragmentUDP(t, 0x5A5A, 9999, 32)
require.True(t, m.filterInbound(denied, len(denied)),
"denied offset-zero fragment must be dropped")
trailing := trailingFragment(t, 0x5A5A, 5, false, 24)
require.True(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment must be denied after the datagram was poisoned")
}
// TestFragment_TinyFirstDropped covers the tiny-fragment attack: a first
// fragment too small to contain the full transport header can't be
// ACL-evaluated and must be dropped.
func TestFragment_TinyFirstDropped(t *testing.T) {
m := newFragmentTestManager(t)
allowUDP(t, m, 8080)
// IPv4 header + 4 raw bytes, MF set, offset 0: too small for the 8-byte UDP
// header, so it decodes to L3 only.
tiny := trailingFragment(t, 0x5555, 0, true, 4)
require.True(t, m.filterInbound(tiny, len(tiny)),
"tiny first fragment without a full L4 header must be dropped")
}
// TestFragment_TCPFirstFragment verifies the TCP arm of the transport decode: a
// first fragment carrying the full 20-byte TCP header is ACL-evaluated and its
// trailing fragments inherit the verdict.
func TestFragment_TCPFirstFragment(t *testing.T) {
m := newFragmentTestManager(t)
_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolTCP, nil,
&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
require.NoError(t, err)
// TCP header (20) + 12 data = 32 bytes -> headerEnd = 4 octets.
first := firstFragmentTCP(t, 0x6666, 8080)
require.False(t, m.filterInbound(first, len(first)),
"allowed TCP first fragment should pass and be recorded")
trailing := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x6666, 4, false, 24)
require.False(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment of an allowed TCP datagram should pass")
}
// TestFragment_TCPTinyFirstDropped verifies the TCP minimum header length: 12
// bytes would satisfy a UDP header but falls short of the 20-byte TCP header.
func TestFragment_TCPTinyFirstDropped(t *testing.T) {
m := newFragmentTestManager(t)
_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrc), fw.ProtocolTCP, nil,
&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
require.NoError(t, err)
tiny := trailingFragmentTo(t, fragTestDst, layers.IPProtocolTCP, 0x7777, 0, true, 12)
require.True(t, m.filterInbound(tiny, len(tiny)),
"first fragment shorter than the TCP header must be dropped")
}
// TestFragment_ConntrackAllowsFirstFragment verifies the conntrack branch: reply
// fragments of an outbound-established UDP flow pass without any inbound rule.
func TestFragment_ConntrackAllowsFirstFragment(t *testing.T) {
m := newFragmentTestManager(t)
out := outboundUDPPacket(t, 12345, 40000)
require.False(t, m.filterOutbound(out, len(out)))
first := firstFragmentUDP(t, 0x8888, 12345, 32)
require.False(t, m.filterInbound(first, len(first)),
"reply first fragment should pass via conntrack")
trailing := trailingFragment(t, 0x8888, 5, false, 24)
require.False(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment of a tracked flow should pass")
}
// TestFragment_RoutingDisabledDropsFragment verifies routed first fragments are
// dropped when routing is disabled.
func TestFragment_RoutingDisabledDropsFragment(t *testing.T) {
m := newFragmentTestManager(t)
m.routingEnabled.Store(false)
first := firstFragmentUDPTo(t, "198.51.100.10", 0x9999, 8080, 32)
require.True(t, m.filterInbound(first, len(first)),
"routed first fragment must be dropped when routing is disabled")
}
// TestFragment_RouteACL verifies the route-ACL branch: fragments to a non-local
// destination follow the route rules, allowed datagrams pass their trailing
// fragments and denied ones don't.
func TestFragment_RouteACL(t *testing.T) {
m := newFragmentTestManager(t)
m.routingEnabled.Store(true)
m.nativeRouter.Store(false)
_, err := m.AddRouteFiltering(
[]byte("rt-1"),
[]netip.Prefix{netip.MustParsePrefix("100.10.0.0/16")},
fw.Network{Prefix: netip.MustParsePrefix("198.51.100.0/24")},
fw.ProtocolUDP,
nil,
&fw.Port{Values: []uint16{8080}},
fw.ActionAccept,
)
require.NoError(t, err)
first := firstFragmentUDPTo(t, "198.51.100.10", 0xAAAA, 8080, 32)
require.False(t, m.filterInbound(first, len(first)),
"route-ACL-allowed first fragment should pass")
trailing := trailingFragmentTo(t, "198.51.100.10", layers.IPProtocolUDP, 0xAAAA, 5, false, 24)
require.False(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment of an allowed routed datagram should pass")
denied := firstFragmentUDPTo(t, "198.51.100.10", 0xBBBB, 9999, 32)
require.True(t, m.filterInbound(denied, len(denied)),
"route-ACL-denied first fragment must be dropped")
deniedTrailing := trailingFragmentTo(t, "198.51.100.10", layers.IPProtocolUDP, 0xBBBB, 5, false, 24)
require.True(t, m.filterInbound(deniedTrailing, len(deniedTrailing)),
"trailing fragment of a denied routed datagram must be dropped")
}
// TestFragment_ExpiredVerdictDropsTrailing verifies a verdict older than the
// tracker timeout no longer admits trailing fragments.
func TestFragment_ExpiredVerdictDropsTrailing(t *testing.T) {
m := newFragmentTestManager(t)
allowUDP(t, m, 8080)
first := firstFragmentUDP(t, 0xCCCC, 8080, 32)
require.False(t, m.filterInbound(first, len(first)))
m.fragments.mutex.Lock()
for key, entry := range m.fragments.entries {
entry.recordedAt = time.Now().Add(-defaultFragmentTimeout - time.Second)
m.fragments.entries[key] = entry
}
m.fragments.mutex.Unlock()
trailing := trailingFragment(t, 0xCCCC, 5, false, 24)
require.True(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment after verdict expiry must be dropped")
}
// TestFragment_CapacityFailsClosed verifies the table cap: at capacity, new
// datagram verdicts are not recorded (their trailing fragments are dropped)
// while already-recorded datagrams keep working.
func TestFragment_CapacityFailsClosed(t *testing.T) {
m := newFragmentTestManager(t)
allowUDP(t, m, 8080)
m.fragments.mutex.Lock()
m.fragments.maxEntries = 1
m.fragments.mutex.Unlock()
first1 := firstFragmentUDP(t, 0x0101, 8080, 32)
require.False(t, m.filterInbound(first1, len(first1)))
first2 := firstFragmentUDP(t, 0x0202, 8080, 32)
require.False(t, m.filterInbound(first2, len(first2)),
"first fragment itself still passes at capacity")
trailing2 := trailingFragment(t, 0x0202, 5, false, 24)
require.True(t, m.filterInbound(trailing2, len(trailing2)),
"trailing fragment of an unrecorded datagram must be dropped at capacity")
trailing1 := trailingFragment(t, 0x0101, 5, false, 24)
require.False(t, m.filterInbound(trailing1, len(trailing1)),
"already-recorded datagram should keep passing at capacity")
}
// v6FragmentHeader builds the 8-byte IPv6 fragment extension header for the
// given inner protocol, offset (8-byte units), More Fragments bit and id.
func v6FragmentHeader(proto layers.IPProtocol, offsetOctets uint16, moreFragments bool, id uint32) []byte {
offsetFlags := offsetOctets << 3
if moreFragments {
offsetFlags |= 1
}
hdr := make([]byte, 8)
hdr[0] = uint8(proto)
binary.BigEndian.PutUint16(hdr[2:4], offsetFlags)
binary.BigEndian.PutUint32(hdr[4:8], id)
return hdr
}
func v6UDPHeader(dstPort uint16, dataLen int) []byte {
hdr := make([]byte, 8)
binary.BigEndian.PutUint16(hdr[0:2], 40000)
binary.BigEndian.PutUint16(hdr[2:4], dstPort)
binary.BigEndian.PutUint16(hdr[4:6], uint16(8+dataLen))
return hdr
}
// firstFragmentUDPv6 builds the first fragment of a fragmented IPv6 UDP
// datagram: fragment header (offset 0, More Fragments set) + full UDP header +
// data.
func firstFragmentUDPv6(tb testing.TB, id uint32, dstPort uint16, dataLen int) []byte {
tb.Helper()
return fragmentUDPv6(tb, id, dstPort, dataLen, true)
}
// fragmentUDPv6 builds an offset-zero IPv6 UDP fragment. With moreFragments
// false it is an atomic fragment (a complete datagram, RFC 6946).
func fragmentUDPv6(tb testing.TB, id uint32, dstPort uint16, dataLen int, moreFragments bool) []byte {
tb.Helper()
ip := &layers.IPv6{
Version: 6,
NextHeader: layers.IPProtocolIPv6Fragment,
HopLimit: 64,
SrcIP: net.ParseIP(fragTestSrcV6),
DstIP: net.ParseIP(fragTestDstV6),
}
payload := append(v6FragmentHeader(layers.IPProtocolUDP, 0, moreFragments, id), v6UDPHeader(dstPort, dataLen)...)
payload = append(payload, make([]byte, dataLen)...)
buf := gopacket.NewSerializeBuffer()
require.NoError(tb, gopacket.SerializeLayers(buf, gopacket.SerializeOptions{FixLengths: true}, ip, gopacket.Payload(payload)))
return buf.Bytes()
}
// trailingFragmentV6 builds a non-first IPv6 fragment: fragment header at the
// given offset carrying raw data and no transport header.
func trailingFragmentV6(tb testing.TB, id uint32, offsetOctets uint16, moreFragments bool, dataLen int) []byte {
tb.Helper()
ip := &layers.IPv6{
Version: 6,
NextHeader: layers.IPProtocolIPv6Fragment,
HopLimit: 64,
SrcIP: net.ParseIP(fragTestSrcV6),
DstIP: net.ParseIP(fragTestDstV6),
}
payload := append(v6FragmentHeader(layers.IPProtocolUDP, offsetOctets, moreFragments, id), make([]byte, dataLen)...)
buf := gopacket.NewSerializeBuffer()
require.NoError(tb, gopacket.SerializeLayers(buf, gopacket.SerializeOptions{FixLengths: true}, ip, gopacket.Payload(payload)))
return buf.Bytes()
}
// TestFragmentV6_TrailingWithoutFirstDropped verifies the IPv6 bypass is closed:
// a trailing fragment with no allowed first fragment is dropped.
func TestFragmentV6_TrailingWithoutFirstDropped(t *testing.T) {
m := newFragmentTestManager(t)
frag := trailingFragmentV6(t, 0xAABBCCDD, 100, false, 40)
require.True(t, m.filterInbound(frag, len(frag)),
"IPv6 trailing fragment without an allowed first fragment must be dropped")
}
// TestFragmentV6_AllowedFirstPassesTrailing verifies IPv6 fragments are
// evaluated like IPv4: an allowed first fragment lets its trailing fragments
// through.
func TestFragmentV6_AllowedFirstPassesTrailing(t *testing.T) {
m := newFragmentTestManager(t)
_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrcV6), fw.ProtocolUDP, nil,
&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
require.NoError(t, err)
// First fragment: UDP header (8) + 32 data = 40 octets -> headerEnd = 5.
first := firstFragmentUDPv6(t, 0xAABBCCDD, 8080, 32)
require.False(t, m.filterInbound(first, len(first)),
"allowed IPv6 first fragment should pass and be recorded")
trailing := trailingFragmentV6(t, 0xAABBCCDD, 5, false, 24)
require.False(t, m.filterInbound(trailing, len(trailing)),
"trailing fragment of an allowed IPv6 datagram should pass")
}
// TestFragmentV6_AtomicNotCached verifies an IPv6 atomic fragment (fragment
// header with offset 0 and no More Fragments, a complete datagram per RFC 6946)
// is evaluated but not recorded, so a flood of allowed atomic fragments can't
// exhaust the verdict table.
func TestFragmentV6_AtomicNotCached(t *testing.T) {
m := newFragmentTestManager(t)
_, err := m.AddPeerFiltering(nil, net.ParseIP(fragTestSrcV6), fw.ProtocolUDP, nil,
&fw.Port{Values: []uint16{8080}}, fw.ActionAccept, "")
require.NoError(t, err)
atomic := fragmentUDPv6(t, 0xA70301C, 8080, 16, false)
require.False(t, m.filterInbound(atomic, len(atomic)),
"allowed IPv6 atomic fragment should pass")
m.fragments.mutex.Lock()
n := len(m.fragments.entries)
m.fragments.mutex.Unlock()
require.Zero(t, n, "atomic fragment must not create a verdict entry")
// A genuine fragmented datagram (More Fragments set) is still recorded.
first := fragmentUDPv6(t, 0xBEEF, 8080, 32, true)
require.False(t, m.filterInbound(first, len(first)))
m.fragments.mutex.Lock()
n = len(m.fragments.entries)
m.fragments.mutex.Unlock()
require.Equal(t, 1, n, "genuine first fragment must record a verdict")
}

View File

@@ -3,14 +3,31 @@
package netstack
import (
"fmt"
"net"
"os"
"strconv"
log "github.com/sirupsen/logrus"
)
const EnvUseNetstackMode = "NB_USE_NETSTACK_MODE"
const (
EnvUseNetstackMode = "NB_USE_NETSTACK_MODE"
// EnvSocks5ListenerPort overrides the port the SOCKS5 proxy listens on.
EnvSocks5ListenerPort = "NB_SOCKS5_LISTENER_PORT"
// EnvSocks5ListenerAddress overrides the host/IP the SOCKS5 proxy binds to.
// The proxy is a bridge for local host applications into the userspace
// WireGuard netstack, so it binds to loopback by default. Override this only
// when the proxy must be reachable from other hosts (e.g. a container
// gateway); doing so exposes an unauthenticated SOCKS5 proxy on that
// address.
EnvSocks5ListenerAddress = "NB_SOCKS5_LISTENER_ADDRESS"
// defaultSocks5Host is the loopback address the SOCKS5 proxy binds to unless
// overridden via EnvSocks5ListenerAddress.
defaultSocks5Host = "127.0.0.1"
)
// IsEnabled todo: move these function to cmd layer
func IsEnabled() bool {
@@ -18,24 +35,40 @@ func IsEnabled() bool {
}
func ListenAddr() string {
sPort := os.Getenv("NB_SOCKS5_LISTENER_PORT")
return net.JoinHostPort(listenHost(), strconv.Itoa(listenPort()))
}
// listenHost returns the host/IP the SOCKS5 proxy binds to. It defaults to
// loopback and only honors EnvSocks5ListenerAddress when it holds a valid IP.
func listenHost() string {
addr := os.Getenv(EnvSocks5ListenerAddress)
if addr == "" {
return defaultSocks5Host
}
if net.ParseIP(addr) == nil {
log.Warnf("invalid socks5 listener address %q, falling back to default: %s", addr, defaultSocks5Host)
return defaultSocks5Host
}
return addr
}
// listenPort returns the port the SOCKS5 proxy binds to, defaulting to
// DefaultSocks5Port when EnvSocks5ListenerPort is unset or invalid.
func listenPort() int {
sPort := os.Getenv(EnvSocks5ListenerPort)
if sPort == "" {
return listenAddr(DefaultSocks5Port)
return DefaultSocks5Port
}
port, err := strconv.Atoi(sPort)
if err != nil {
log.Warnf("invalid socks5 listener port, unable to convert it to int, falling back to default: %d", DefaultSocks5Port)
return listenAddr(DefaultSocks5Port)
return DefaultSocks5Port
}
if port < 1 || port > 65535 {
log.Warnf("invalid socks5 listener port, it should be in the range 1-65535, falling back to default: %d", DefaultSocks5Port)
return listenAddr(DefaultSocks5Port)
return DefaultSocks5Port
}
return listenAddr(port)
}
func listenAddr(port int) string {
return fmt.Sprintf("0.0.0.0:%d", port)
return port
}

View File

@@ -0,0 +1,63 @@
//go:build !js
package netstack
import (
"net"
"strconv"
"testing"
)
func TestListenAddr_DefaultsToLoopback(t *testing.T) {
// No env overrides: must bind loopback, never all interfaces.
got := ListenAddr()
want := net.JoinHostPort("127.0.0.1", strconv.Itoa(DefaultSocks5Port))
if got != want {
t.Fatalf("ListenAddr() = %q, want %q", got, want)
}
}
func TestListenAddr_AddressOverride(t *testing.T) {
tests := []struct {
name string
env string
want string
}{
{name: "valid override honored", env: "0.0.0.0", want: "0.0.0.0"},
{name: "valid specific ip honored", env: "10.0.0.5", want: "10.0.0.5"},
{name: "ipv6 loopback bracketed", env: "::1", want: "::1"},
{name: "invalid falls back to loopback", env: "not-an-ip", want: "127.0.0.1"},
{name: "empty falls back to loopback", env: "", want: "127.0.0.1"},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
t.Setenv(EnvSocks5ListenerAddress, tc.env)
want := net.JoinHostPort(tc.want, strconv.Itoa(DefaultSocks5Port))
if got := ListenAddr(); got != want {
t.Fatalf("ListenAddr() = %q, want %q", got, want)
}
})
}
}
func TestListenAddr_PortOverride(t *testing.T) {
tests := []struct {
name string
env string
want int
}{
{name: "valid port honored", env: "1081", want: 1081},
{name: "non-numeric falls back", env: "abc", want: DefaultSocks5Port},
{name: "out of range falls back", env: "70000", want: DefaultSocks5Port},
{name: "zero falls back", env: "0", want: DefaultSocks5Port},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
t.Setenv(EnvSocks5ListenerPort, tc.env)
want := net.JoinHostPort("127.0.0.1", strconv.Itoa(tc.want))
if got := ListenAddr(); got != want {
t.Fatalf("ListenAddr() = %q, want %q", got, want)
}
})
}
}

View File

@@ -299,7 +299,7 @@ func (d *DeviceAuthorizationFlow) WaitToken(ctx context.Context, info AuthFlowIn
UseIDToken: d.providerConfig.UseIDToken,
}
err = isValidAccessToken(tokenInfo.GetTokenToUse(), d.providerConfig.Audience)
err = validateTokenAudience(tokenInfo.GetTokenToUse(), d.providerConfig.Audience)
if err != nil {
return TokenInfo{}, fmt.Errorf("validate access token failed with error: %v", err)
}

View File

@@ -306,7 +306,7 @@ func (p *PKCEAuthorizationFlow) parseOAuthToken(token *oauth2.Token) (TokenInfo,
audience = p.providerConfig.ClientID
}
if err := isValidAccessToken(tokenInfo.GetTokenToUse(), audience); err != nil {
if err := validateTokenAudience(tokenInfo.GetTokenToUse(), audience); err != nil {
return TokenInfo{}, fmt.Errorf("authentication failed: invalid access token - %w", err)
}
@@ -320,6 +320,11 @@ func (p *PKCEAuthorizationFlow) parseOAuthToken(token *oauth2.Token) (TokenInfo,
return tokenInfo, nil
}
// parseEmailFromIDToken extracts the email (or name) claim from an ID token
// without verifying its signature. The value is best-effort and used only as a
// UX convenience (login hint prefill and display); it never drives an
// authorization decision. The authoritative identity is established server-side
// from the signature-verified token.
func parseEmailFromIDToken(token string) (string, error) {
parts := strings.Split(token, ".")
if len(parts) < 2 {

View File

@@ -20,14 +20,26 @@ func randomBytesInHex(count int) (string, error) {
return hex.EncodeToString(buf), nil
}
// isValidAccessToken is a simple validation of the access token
func isValidAccessToken(token string, audience string) error {
// validateTokenAudience checks that the token is a well-formed JWT whose
// audience claim matches the expected audience.
//
// It does NOT verify the token's cryptographic signature and therefore must not
// be treated as an authenticity check. The token is obtained by the client
// directly from the IdP token endpoint over TLS, and its signature is verified
// server-side by the management server against the IdP's JWKS
// (see shared/auth/jwt/validator.go). This function is only a client-side
// sanity check that the returned token targets the expected audience.
func validateTokenAudience(token string, audience string) error {
if token == "" {
return fmt.Errorf("token received is empty")
}
encodedClaims := strings.Split(token, ".")[1]
claimsString, err := base64.RawURLEncoding.DecodeString(encodedClaims)
parts := strings.Split(token, ".")
if len(parts) != 3 {
return fmt.Errorf("token is not a well-formed JWT")
}
claimsString, err := base64.RawURLEncoding.DecodeString(parts[1])
if err != nil {
return err
}

View File

@@ -0,0 +1,108 @@
package auth
import (
"encoding/base64"
"encoding/json"
"testing"
)
// makeJWT builds an unsigned JWT-shaped string (header.payload.signature) with
// the given claims payload. The signature part is arbitrary because
// validateTokenAudience intentionally does not verify it.
func makeJWT(t *testing.T, claims map[string]interface{}) string {
t.Helper()
header := base64.RawURLEncoding.EncodeToString([]byte(`{"alg":"RS256","typ":"JWT"}`))
payloadBytes, err := json.Marshal(claims)
if err != nil {
t.Fatalf("marshal claims: %v", err)
}
payload := base64.RawURLEncoding.EncodeToString(payloadBytes)
return header + "." + payload + ".unverified-signature"
}
func TestValidateTokenAudience(t *testing.T) {
tests := []struct {
name string
token string
audience string
wantErr bool
}{
{
name: "empty token",
token: "",
audience: "netbird",
wantErr: true,
},
{
name: "not a JWT - no dots",
token: "notajwt",
audience: "netbird",
wantErr: true,
},
{
name: "not a JWT - two parts only",
token: "header.payload",
audience: "netbird",
wantErr: true,
},
{
name: "matching string audience",
token: makeJWT(t, map[string]interface{}{"aud": "netbird"}),
audience: "netbird",
wantErr: false,
},
{
name: "mismatching string audience",
token: makeJWT(t, map[string]interface{}{"aud": "other"}),
audience: "netbird",
wantErr: true,
},
{
name: "matching audience in array",
token: makeJWT(t, map[string]interface{}{"aud": []interface{}{"other", "netbird"}}),
audience: "netbird",
wantErr: false,
},
{
name: "mismatching audience array",
token: makeJWT(t, map[string]interface{}{"aud": []interface{}{"a", "b"}}),
audience: "netbird",
wantErr: true,
},
{
name: "missing audience claim",
token: makeJWT(t, map[string]interface{}{"sub": "user"}),
audience: "netbird",
wantErr: true,
},
{
name: "invalid base64 payload",
token: "header.!!!not-base64!!!.sig",
audience: "netbird",
wantErr: true,
},
}
for _, tc := range tests {
t.Run(tc.name, func(t *testing.T) {
err := validateTokenAudience(tc.token, tc.audience)
if tc.wantErr && err == nil {
t.Fatalf("expected error, got nil")
}
if !tc.wantErr && err != nil {
t.Fatalf("expected no error, got %v", err)
}
})
}
}
// TestValidateTokenAudienceNoPanic guards the regression where a non-empty
// token without the JWT dot structure caused an index-out-of-range panic.
func TestValidateTokenAudienceNoPanic(t *testing.T) {
inputs := []string{"a", ".", "a.", "aaaa", "no-dots-here"}
for _, in := range inputs {
if err := validateTokenAudience(in, "netbird"); err == nil {
t.Fatalf("expected error for malformed token %q, got nil", in)
}
}
}

View File

@@ -1,6 +1,7 @@
package statemanager
import (
"bytes"
"context"
"encoding/json"
"errors"
@@ -305,6 +306,11 @@ func (m *Manager) loadStateFile(deleteCorrupt bool) (map[string]json.RawMessage,
var rawStates map[string]json.RawMessage
if err := json.Unmarshal(data, &rawStates); err != nil {
if len(bytes.TrimSpace(data)) == 0 {
log.Warnf("state file %s is empty (%d bytes)", m.filePath, len(data))
} else {
log.Warnf("state file %s has malformed content (%d bytes)", m.filePath, len(data))
}
m.handleCorruptedState(deleteCorrupt)
return nil, fmt.Errorf("unmarshal states: %w", err)
}

View File

@@ -14,6 +14,7 @@ import (
log "github.com/sirupsen/logrus"
nbssh "github.com/netbirdio/netbird/client/ssh"
"github.com/netbirdio/netbird/shared/management/domain"
)
const (
@@ -218,11 +219,20 @@ func (m *Manager) buildHostPatterns(peer PeerSSHInfo) []string {
if peer.IPv6.IsValid() {
hostPatterns = append(hostPatterns, peer.IPv6.String())
}
if peer.FQDN != "" {
// Peer FQDNs and hostnames originate from remote peers, so they must be
// validated as plain DNS names before being embedded in the ssh_config
// "Match host" pattern list. This prevents injection of arbitrary
// ssh_config directives via embedded quotes, whitespace, newlines, the
// comma pattern separator, or the "*"/"?" pattern metacharacters.
if domain.IsValidDomainNoWildcard(peer.FQDN) {
hostPatterns = append(hostPatterns, peer.FQDN)
} else if peer.FQDN != "" {
log.Warnf("skipping peer FQDN with invalid characters in SSH config: %q", peer.FQDN)
}
if peer.Hostname != "" && peer.Hostname != peer.FQDN {
if peer.Hostname != peer.FQDN && domain.IsValidDomainNoWildcard(peer.Hostname) {
hostPatterns = append(hostPatterns, peer.Hostname)
} else if peer.Hostname != "" && peer.Hostname != peer.FQDN {
log.Warnf("skipping peer hostname with invalid characters in SSH config: %q", peer.Hostname)
}
return hostPatterns
}

View File

@@ -148,6 +148,45 @@ func TestManager_MatchHostFormat(t *testing.T) {
"should use Match host with comma-separated patterns")
}
func TestManager_HostPatternInjection(t *testing.T) {
tempDir, err := os.MkdirTemp("", "netbird-ssh-config-test")
require.NoError(t, err)
defer func() { assert.NoError(t, os.RemoveAll(tempDir)) }()
manager := &Manager{
sshConfigDir: filepath.Join(tempDir, "ssh_config.d"),
sshConfigFile: "99-netbird.conf",
}
// A malicious peer FQDN/hostname attempts to break out of the Match host
// directive and inject arbitrary ssh_config (a ProxyCommand executing a
// command). It must be rejected, not written to the config.
peers := []PeerSSHInfo{
{
Hostname: "evil\"\n ProxyCommand touch /tmp/pwned\nHost x",
IP: netip.MustParseAddr("100.125.1.1"),
FQDN: "evil\"\n ProxyCommand touch /tmp/pwned\nHost x.nb.internal",
},
{Hostname: "peer2", IP: netip.MustParseAddr("100.125.1.2"), FQDN: "peer2.nb.internal"},
}
err = manager.SetupSSHClientConfig(peers)
require.NoError(t, err)
configPath := filepath.Join(manager.sshConfigDir, manager.sshConfigFile)
content, err := os.ReadFile(configPath)
require.NoError(t, err)
configStr := string(content)
assert.NotContains(t, configStr, "ProxyCommand touch /tmp/pwned",
"injected directive must not appear in generated config")
assert.NotContains(t, configStr, "evil",
"malicious pattern must be dropped entirely")
// The valid peer must still be present, on a single Match host line.
assert.Contains(t, configStr, "Match host \"100.125.1.1,100.125.1.2,peer2.nb.internal,peer2\"",
"valid peers must survive, injected patterns dropped")
}
func TestManager_ForcedSSHConfig(t *testing.T) {
// Set force environment variable
t.Setenv(EnvForceSSHConfig, "true")

View File

@@ -69,7 +69,8 @@ func parseGetentPasswd(output string) (*user.User, string, error) {
// validateGetentInput checks that the input is safe to pass to getent or id.
// Allows POSIX usernames, numeric UIDs, and common NSS extensions
// (@ for Kerberos, $ for Samba, + for NIS compat).
// (@ for Kerberos, $ for Samba, + for NIS compat). A leading hyphen is
// rejected so the input can never be parsed as a command-line flag.
func validateGetentInput(input string) bool {
maxLen := 32
if runtime.GOOS == "linux" {
@@ -80,6 +81,10 @@ func validateGetentInput(input string) bool {
return false
}
if input[0] == '-' {
return false
}
for _, r := range input {
if isAllowedGetentChar(r) {
continue

View File

@@ -157,6 +157,9 @@ func TestValidateGetentInput(t *testing.T) {
{"numeric UID", "1001", true},
{"dots and underscores", "alice.bob_test", true},
{"hyphen", "alice-bob", true},
{"leading hyphen rejected", "-i", false},
{"leading double hyphen rejected", "--no-idn", false},
{"lone hyphen rejected", "-", false},
{"kerberos principal", "user@REALM", true},
{"samba machine account", "MACHINE$", true},
{"NIS compat", "+user", true},

View File

@@ -73,6 +73,13 @@ export default function SessionExpirationDialog() {
let offCancel: (() => void) | undefined;
// Return the dialog to its interactive state and dismiss the browser popup
const resetDialog = () => {
offCancel?.();
WindowManager.CloseBrowserLogin().catch(console.error);
setBusy(false);
};
try {
const start = await Session.RequestExtend({ hint: "" });
const uri = start.verificationUriComplete || start.verificationUri;
@@ -105,25 +112,22 @@ export default function SessionExpirationDialog() {
if (outcome.kind === "cancel") {
waitPromise.cancel?.();
waitPromise.catch(() => {});
resetDialog();
return;
}
// Another surface owns this flow; keep the dialog open to retry.
if (outcome.result.preempted) {
resetDialog();
return;
}
// Close before the popup so the restore can't flash this window back.
WindowManager.CloseSessionExpiration().catch(console.error);
WindowManager.CloseRenewFlow().catch(console.error);
} catch (e) {
resetDialog();
await errorDialog({
Title: t("sessionExpiration.extendFailedTitle"),
Message: formatErrorMessage(e),
});
} finally {
offCancel?.();
WindowManager.CloseBrowserLogin().catch(console.error);
setBusy(false);
}
}, [busy, t]);
@@ -139,12 +143,11 @@ export default function SessionExpirationDialog() {
});
WindowManager.CloseSessionExpiration().catch(console.error);
} catch (e) {
setBusy(false);
await errorDialog({
Title: t("sessionExpiration.logoutFailedTitle"),
Message: formatErrorMessage(e),
});
} finally {
setBusy(false);
}
}, [busy, t]);

View File

@@ -8,6 +8,7 @@
{"code": "fr", "displayName": "Français", "englishName": "French"},
{"code": "it", "displayName": "Italiano", "englishName": "Italian"},
{"code": "pt", "displayName": "Português", "englishName": "Portuguese"},
{"code": "zh-CN", "displayName": "简体中文", "englishName": "Simplified Chinese"}
{"code": "zh-CN", "displayName": "简体中文", "englishName": "Simplified Chinese"},
{"code": "ja", "displayName": "日本語", "englishName": "Japanese"}
]
}

File diff suppressed because it is too large Load Diff

View File

@@ -185,37 +185,38 @@ func (s *WindowManager) OpenBrowserLogin(uri string) {
startURL = "/#/dialog/browser-login?uri=" + url.QueryEscape(uri)
}
s.hideOtherWindowsLocked("browser-login")
// Prefer the main window's screen (multi-monitor); falls back to OS-default centering.
var screen *application.Screen
if s.mainWindow != nil {
if sc, err := s.mainWindow.GetScreen(); err == nil {
screen = sc
}
}
opts := DialogWindowOptions("browser-login", s.title("window.title.signIn"), startURL, s.linuxIcon)
// Not always-on-top: it would obscure the browser tab the user logs in through.
opts.AlwaysOnTop = false
opts.InitialPosition = application.WindowCentered
opts.Screen = screen
// Open on the active (where users cursor is) display, like the session-expiration dialog.
opts.Screen = s.getScreenBasedOnCursorPosition()
s.browserLogin = s.app.Window.NewWithOptions(opts)
bl := s.browserLogin
// Red-X close means cancel: emit the event so startLogin() tears down the SSO wait.
bl.OnWindowEvent(events.Common.WindowClosing, func(_ *application.WindowEvent) {
s.app.Event.Emit(EventBrowserLoginCancel)
s.mu.Lock()
s.browserLogin = nil
s.restoreHiddenWindowsLocked()
// Only a live user red-X still has this registered; programmatic closers
// nil s.browserLogin first and clean up themselves. Guarding here stops a
// stale close event from wiping a replacement popup's state.
userClosed := s.browserLogin == bl
if userClosed {
s.browserLogin = nil
s.restoreHiddenWindowsLocked()
}
s.mu.Unlock()
if userClosed {
s.app.Event.Emit(EventBrowserLoginCancel)
}
})
s.centerWhenReady(s.browserLogin)
s.centerOnCursorScreen(s.browserLogin)
return
}
if uri != "" {
s.browserLogin.SetURL("/#/dialog/browser-login?uri=" + url.QueryEscape(uri))
}
s.centerOnCursorScreen(s.browserLogin)
s.browserLogin.Show()
s.browserLogin.Focus()
s.centerWhenReady(s.browserLogin)
}
// BrowserLoginWindow returns the live SSO popup, or nil. While non-nil it is the
@@ -238,6 +239,8 @@ func (s *WindowManager) CloseBrowserLogin() {
s.mu.Lock()
w := s.browserLogin
s.browserLogin = nil
// The WindowClosing hook no-ops on a programmatic close, so restore here.
s.restoreHiddenWindowsLocked()
s.mu.Unlock()
if w != nil {
w.Close()
@@ -279,6 +282,35 @@ func (s *WindowManager) CloseSessionExpiration() {
}
}
// CloseRenewFlow tears down the SSO session-renewal UI in a single call: it
// closes the browser-login popup and the session-expiration window together.
func (s *WindowManager) CloseRenewFlow() {
s.mu.Lock()
bl := s.browserLogin
se := s.sessionExpiration
s.browserLogin = nil
s.sessionExpiration = nil
if se != nil {
kept := s.hiddenForLogin[:0]
for _, w := range s.hiddenForLogin {
if w != se {
kept = append(kept, w)
}
}
s.hiddenForLogin = kept
}
s.restoreHiddenWindowsLocked()
s.mu.Unlock()
// Close after unlock so the re-entrant handlers can take s.mu.
if bl != nil {
bl.Close()
}
if se != nil {
se.Close()
}
}
// OpenInstallProgress shows the install-progress window and hides the rest for the duration
// (restored on close). It owns its own result polling since the daemon restarts mid-install.
func (s *WindowManager) OpenInstallProgress(version string) {

View File

@@ -145,6 +145,7 @@ type AuthConfig struct {
CLIRedirectURIs []string `yaml:"cliRedirectURIs"`
Owner *AuthOwnerConfig `yaml:"owner,omitempty"`
DashboardPostLogoutRedirectURIs []string `yaml:"dashboardPostLogoutRedirectURIs"`
GrantTypes []string `yaml:"grantTypes"`
}
// AuthStorageConfig contains auth storage settings
@@ -604,6 +605,7 @@ func (c *CombinedConfig) buildEmbeddedIdPConfig(mgmt ManagementConfig) (*idp.Emb
DashboardRedirectURIs: mgmt.Auth.DashboardRedirectURIs,
CLIRedirectURIs: mgmt.Auth.CLIRedirectURIs,
DashboardPostLogoutRedirectURIs: mgmt.Auth.DashboardPostLogoutRedirectURIs,
GrantTypes: mgmt.Auth.GrantTypes,
}
if mgmt.Auth.Owner != nil && mgmt.Auth.Owner.Email != "" {

View File

@@ -226,7 +226,7 @@ func (s *serverInstances) createRelayServer(cfg *CombinedConfig, tlsSupport bool
}
hashedSecret := sha256.Sum256([]byte(cfg.Relay.AuthSecret))
authenticator := auth.NewTimedHMACValidator(hashedSecret[:], 24*time.Hour)
authenticator := auth.NewTimedHMACValidator(hashedSecret[:])
relayCfg := relayServer.Config{
Meter: s.metricsServer.Meter,

View File

@@ -197,6 +197,12 @@ type JSONMetadataInjection struct {
// enforces a 128-char limit per value; oversized values are
// truncated rather than failing the request. 0 disables the cap.
MaxValueLength int
// Sanitize, when true, replaces characters outside the destination's
// accepted set with '_' before emitting each value. AWS Bedrock's
// X-Amzn-Bedrock-Request-Metadata restricts values to a limited character
// class, so unsanitized group display names (e.g. containing spaces) would
// make Bedrock reject the request with 400.
Sanitize bool
}
// providers is the canonical list of supported Agent Network providers.
@@ -329,6 +335,18 @@ var providers = []Provider{
{ID: "amazon.nova-lite", Label: "Amazon Nova Lite (Bedrock)", InputPer1k: 0.00006, OutputPer1k: 0.00024, ContextWindow: 300000},
{ID: "amazon.nova-micro", Label: "Amazon Nova Micro (Bedrock)", InputPer1k: 0.000035, OutputPer1k: 0.00014, ContextWindow: 128000},
},
// Bedrock accepts a cost-allocation metadata header; stamp the caller's
// user + authorizing group so spend can be attributed in AWS Cost
// Management. Sanitized because Bedrock restricts the value character set.
IdentityInjection: &IdentityInjection{
JSONMetadata: &JSONMetadataInjection{
Header: "X-Amzn-Bedrock-Request-Metadata",
UserKey: "user",
GroupsKey: "group",
MaxValueLength: 256,
Sanitize: true,
},
},
},
{
ID: "vertex_ai_api",

View File

@@ -540,6 +540,7 @@ type identityInjectJSONMetadata struct {
UserKey string `json:"user_key,omitempty"`
GroupsKey string `json:"groups_key,omitempty"`
MaxValueLength int `json:"max_value_length,omitempty"`
Sanitize bool `json:"sanitize,omitempty"`
}
// buildIdentityInjectConfigJSON walks the enabled providers and emits
@@ -583,9 +584,11 @@ func buildIdentityInjectConfigJSON(providers []*types.Provider, groupIndex map[s
func buildIdentityInjectRule(p *types.Provider, entry catalog.Provider) (identityInjectProvider, bool) {
rule := identityInjectProvider{ProviderID: p.ID}
// Identity-stamping shape (one of HeaderPair / JSONMetadata). Skip the
// shape silently when the catalog entry doesn't declare one — extras
// can still apply, see below.
if entry.IdentityInjection != nil {
// shape silently when the catalog entry doesn't declare one, or when the
// operator disabled metadata for this provider — extras can still apply,
// see below. MetadataDisabled suppresses only the identity dimensions
// (user + authorizing group), not the catalog's routing ExtraHeaders.
if !p.MetadataDisabled && entry.IdentityInjection != nil {
switch {
case entry.IdentityInjection.HeaderPair != nil:
rule.HeaderPair = buildIdentityHeaderPair(p, entry.IdentityInjection.HeaderPair)
@@ -651,6 +654,7 @@ func buildIdentityJSONMetadata(p *types.Provider, jm *catalog.JSONMetadataInject
UserKey: userKey,
GroupsKey: groupsKey,
MaxValueLength: jm.MaxValueLength,
Sanitize: jm.Sanitize,
}
}

View File

@@ -698,6 +698,94 @@ func TestSynthesizeServices_IdentityInject_Portkey_NotCustomizable(t *testing.T)
"same fixed-schema guarantee for the groups dimension")
}
// TestSynthesizeServices_IdentityInject_Bedrock pins Bedrock's cost-allocation
// metadata: a JSONMetadata shape emitting X-Amzn-Bedrock-Request-Metadata with
// the reserved user/group keys, sanitized to Bedrock's accepted charset.
func TestSynthesizeServices_IdentityInject_Bedrock(t *testing.T) {
ctx := context.Background()
ctrl := gomock.NewController(t)
defer ctrl.Finish()
mockStore := store.NewMockStore(ctrl)
br := newSynthTestProvider()
br.ID = "prov-bedrock"
br.ProviderID = "bedrock_api"
br.UpstreamURL = "https://bedrock-runtime.us-east-1.amazonaws.com"
br.APIKey = "bedrock-bearer"
br.CreatedAt = time.Date(2026, 4, 2, 0, 0, 0, 0, time.UTC)
policy := newSynthTestPolicy(br.ID, "grp-eng", "")
policy.ID = "pol-bedrock"
expectSynthBaseInputs(mockStore, ctx, newSynthTestSettings(),
[]*types.Provider{br},
[]*types.Policy{policy},
[]*types.Guardrail{})
services, err := SynthesizeServices(ctx, mockStore, testAccountID)
require.NoError(t, err)
require.Len(t, services, 1)
var injectCfg identityInjectConfig
for _, m := range services[0].Targets[0].Options.Middlewares {
if m.ID == middlewareIDLLMIdentityInject {
require.NoError(t, json.Unmarshal(m.ConfigJSON, &injectCfg))
break
}
}
require.Len(t, injectCfg.Providers, 1)
entry := injectCfg.Providers[0]
require.NotNil(t, entry.JSONMetadata, "Bedrock uses the JSONMetadata shape for cost-allocation metadata")
assert.Nil(t, entry.HeaderPair, "shapes are mutually exclusive")
assert.Equal(t, "X-Amzn-Bedrock-Request-Metadata", entry.JSONMetadata.Header,
"the caller identity lands in Bedrock's cost-allocation metadata header")
assert.Equal(t, "user", entry.JSONMetadata.UserKey)
assert.Equal(t, "group", entry.JSONMetadata.GroupsKey)
assert.True(t, entry.JSONMetadata.Sanitize,
"Bedrock restricts the metadata value charset, so values must be sanitized")
}
// TestSynthesizeServices_MetadataDisabled_SuppressesInjection verifies the
// per-provider opt-out: a provider with MetadataDisabled set emits no
// identity-inject entry (Bedrock has no catalog ExtraHeaders, so the whole
// entry is dropped).
func TestSynthesizeServices_MetadataDisabled_SuppressesInjection(t *testing.T) {
ctx := context.Background()
ctrl := gomock.NewController(t)
defer ctrl.Finish()
mockStore := store.NewMockStore(ctrl)
br := newSynthTestProvider()
br.ID = "prov-bedrock"
br.ProviderID = "bedrock_api"
br.UpstreamURL = "https://bedrock-runtime.us-east-1.amazonaws.com"
br.APIKey = "bedrock-bearer"
br.MetadataDisabled = true
br.CreatedAt = time.Date(2026, 4, 2, 0, 0, 0, 0, time.UTC)
policy := newSynthTestPolicy(br.ID, "grp-eng", "")
policy.ID = "pol-bedrock"
expectSynthBaseInputs(mockStore, ctx, newSynthTestSettings(),
[]*types.Provider{br},
[]*types.Policy{policy},
[]*types.Guardrail{})
services, err := SynthesizeServices(ctx, mockStore, testAccountID)
require.NoError(t, err)
require.Len(t, services, 1)
var injectCfg identityInjectConfig
for _, m := range services[0].Targets[0].Options.Middlewares {
if m.ID == middlewareIDLLMIdentityInject {
require.NoError(t, json.Unmarshal(m.ConfigJSON, &injectCfg))
break
}
}
assert.Empty(t, injectCfg.Providers,
"metadata_disabled must drop the provider's identity-inject entry")
}
// TestSynthesizeServices_IdentityInject_Vercel pins Vercel AI
// Gateway's wiring: HeaderPair shape with fixed wire names dictated
// by Vercel's Custom Reporting API (ai-reporting-user /

View File

@@ -51,6 +51,12 @@ type Provider struct {
// private or self-signed certificate. The synthesiser propagates it into
// the router route so the proxy dials that provider's upstream insecurely.
SkipTLSVerification bool `gorm:"column:skip_tls_verification"`
// MetadataDisabled suppresses identity metadata injection for this provider.
// Metadata (the caller's user + authorizing group) is injected by default;
// when true the synthesiser omits the provider's identity-inject shape, so no
// user/group headers (e.g. Bedrock's X-Amzn-Bedrock-Request-Metadata) are
// stamped. Catalog ExtraHeaders (routing config) are unaffected.
MetadataDisabled bool `gorm:"column:metadata_disabled"`
// SessionPrivateKey + SessionPublicKey are the ed25519 keypair the
// synthesised reverse-proxy service uses to sign / verify session
// JWTs after a successful OIDC handshake. Generated once on
@@ -137,6 +143,9 @@ func (p *Provider) FromAPIRequest(req *api.AgentNetworkProviderRequest) {
if req.SkipTlsVerification != nil {
p.SkipTLSVerification = *req.SkipTlsVerification
}
if req.MetadataDisabled != nil {
p.MetadataDisabled = *req.MetadataDisabled
}
// Identity-header overrides for catalogs flagged Customizable.
// nil pointer = "field omitted on the wire" → leave the stored
// value untouched (per the openapi description). Empty string is
@@ -170,6 +179,7 @@ func (p *Provider) ToAPIResponse() *api.AgentNetworkProvider {
Models: models,
Enabled: p.Enabled,
SkipTlsVerification: p.SkipTLSVerification,
MetadataDisabled: p.MetadataDisabled,
CreatedAt: &created,
UpdatedAt: &updated,
}

View File

@@ -42,3 +42,38 @@ func TestProvider_SkipTLSVerification_RoundTrip(t *testing.T) {
assert.False(t, p.SkipTLSVerification, "explicit false must clear skip_tls_verification")
assert.False(t, p.ToAPIResponse().SkipTlsVerification, "response must reflect the cleared value")
}
// TestProvider_MetadataDisabled_RoundTrip covers the request→provider→response
// mapping of metadata_disabled, with the same update semantics: nil preserves,
// explicit false clears.
func TestProvider_MetadataDisabled_RoundTrip(t *testing.T) {
enable := true
disable := false
base := func() *api.AgentNetworkProviderRequest {
return &api.AgentNetworkProviderRequest{
ProviderId: "bedrock_api",
Name: "bedrock",
UpstreamUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
}
}
p := NewProvider("acc-1")
req := base()
req.MetadataDisabled = &enable
p.FromAPIRequest(req)
assert.True(t, p.MetadataDisabled, "create with metadata_disabled=true must set the field")
assert.True(t, p.ToAPIResponse().MetadataDisabled, "response must surface metadata_disabled")
// Omitting the field on update leaves the stored value untouched.
p.FromAPIRequest(base())
assert.True(t, p.MetadataDisabled, "omitting metadata_disabled on update must preserve it")
// Explicit false clears it (re-enables metadata).
req = base()
req.MetadataDisabled = &disable
p.FromAPIRequest(req)
assert.False(t, p.MetadataDisabled, "explicit false must clear metadata_disabled")
assert.False(t, p.ToAPIResponse().MetadataDisabled, "response must reflect the cleared value")
}

View File

@@ -64,6 +64,11 @@ type JSONMetadataRule struct {
UserKey string `json:"user_key,omitempty"`
GroupsKey string `json:"groups_key,omitempty"`
MaxValueLength int `json:"max_value_length,omitempty"`
// Sanitize replaces characters outside the destination provider's accepted
// set with '_' before emitting each value. AWS Bedrock's
// X-Amzn-Bedrock-Request-Metadata restricts values to [A-Za-z0-9 +-=._:/@];
// group display names with other characters would otherwise 400.
Sanitize bool `json:"sanitize,omitempty"`
}
// Config is the on-wire configuration accepted by the factory. An

View File

@@ -292,15 +292,21 @@ func applyJSONMetadata(rule *JSONMetadataRule, in *middleware.Input) *middleware
mutations := &middleware.Mutations{}
mutations.HeadersRemove = append(mutations.HeadersRemove, rule.Header)
emit := func(v string) string {
if rule.Sanitize {
v = sanitizeMetadataValue(v)
}
return truncate(v, rule.MaxValueLength)
}
payload := map[string]string{}
if rule.UserKey != "" {
if identity := identityFor(in); identity != "" {
payload[rule.UserKey] = truncate(identity, rule.MaxValueLength)
payload[rule.UserKey] = emit(identity)
}
}
if rule.GroupsKey != "" {
if csv := authorisingTagsCSV(in); csv != "" {
payload[rule.GroupsKey] = truncate(csv, rule.MaxValueLength)
payload[rule.GroupsKey] = emit(csv)
}
}
if len(payload) == 0 {
@@ -359,6 +365,36 @@ func truncate(s string, maxBytes int) string {
return s[:maxBytes]
}
// sanitizeMetadataValue replaces any character outside AWS Bedrock's accepted
// request-metadata class — letters, digits, space, and + - = . _ : / @ — with
// '_'. This keeps values (notably the groups CSV, whose commas are rejected, and
// group display names with arbitrary characters) from making Bedrock reject the
// request with 400. The result stays opaque to the gateway.
func sanitizeMetadataValue(s string) string {
var b strings.Builder
b.Grow(len(s))
for _, r := range s {
if metadataCharAllowed(r) {
b.WriteRune(r)
} else {
b.WriteByte('_')
}
}
return b.String()
}
func metadataCharAllowed(r rune) bool {
switch {
case r >= 'a' && r <= 'z', r >= 'A' && r <= 'Z', r >= '0' && r <= '9':
return true
}
switch r {
case ' ', '+', '-', '=', '.', '_', ':', '/', '@':
return true
}
return false
}
// tagsIDsFromAuthorising reads llm_router's authorising-groups metadata
// (a CSV of group ids) and returns the parsed slice. Returns nil when
// the key is absent or empty so the caller can fall back to the full

View File

@@ -304,6 +304,46 @@ func TestInject_JSONMetadata_TruncatesValues(t *testing.T) {
"per-value byte length must be capped at MaxValueLength")
}
// TestInject_JSONMetadata_Sanitize pins the AWS-Bedrock sanitization path: when
// Sanitize is set, characters outside Bedrock's accepted metadata class
// (notably the groups CSV comma and arbitrary characters in group display
// names) are replaced with '_' so Bedrock doesn't reject the request. Allowed
// characters (letters, digits, spaces, and @ . _ : / + - =) pass through.
func TestInject_JSONMetadata_Sanitize(t *testing.T) {
rule := ProviderInjection{
ProviderID: portkeyProvider,
JSONMetadata: &JSONMetadataRule{
Header: "X-Amzn-Bedrock-Request-Metadata",
UserKey: "user",
GroupsKey: "group",
MaxValueLength: 256,
Sanitize: true,
},
}
mw := New(Config{Providers: []ProviderInjection{rule}})
in := newInput(portkeyProvider, "alice", []string{"g1", "g2"})
in.UserEmail = "alice@example.com"
// Group display names carry characters Bedrock rejects (comma, '#'); the CSV
// join adds another comma between the two groups.
in.UserGroupNames = []string{"Eng,Team", "Ops#1"}
out, err := mw.Invoke(context.Background(), in)
require.NoError(t, err)
require.NotNil(t, out.Mutations)
require.Len(t, out.Mutations.HeadersAdd, 1)
added := out.Mutations.HeadersAdd[0]
assert.Equal(t, "X-Amzn-Bedrock-Request-Metadata", added.Key,
"the Bedrock cost-allocation header carries the metadata JSON")
var payload map[string]string
require.NoError(t, json.Unmarshal([]byte(added.Value), &payload))
assert.Equal(t, "alice@example.com", payload["user"],
"'@' and '.' are in Bedrock's accepted set and must be preserved")
assert.NotContains(t, payload["group"], ",", "commas must be sanitized — Bedrock rejects them")
assert.NotContains(t, payload["group"], "#", "disallowed characters must be sanitized")
assert.Contains(t, payload["group"], "Eng", "allowed characters must be preserved")
}
// TestInject_JSONMetadata_EmptyIdentity_StripsButDoesNotAdd verifies the
// anti-spoof Remove still fires when there's nothing to stamp.
func TestInject_JSONMetadata_EmptyIdentity_StripsButDoesNotAdd(t *testing.T) {

View File

@@ -173,7 +173,7 @@ func execute(cmd *cobra.Command, args []string) error {
}
hashedSecret := sha256.Sum256([]byte(cobraConfig.AuthSecret))
authenticator := auth.NewTimedHMACValidator(hashedSecret[:], 24*time.Hour)
authenticator := auth.NewTimedHMACValidator(hashedSecret[:])
cfg := server.Config{
Meter: metricsServer.Meter,

View File

@@ -5,14 +5,8 @@ import (
"fmt"
"time"
log "github.com/sirupsen/logrus"
"github.com/netbirdio/netbird/relay/server/listener"
"github.com/netbirdio/netbird/shared/relay/messages"
//nolint:staticcheck
"github.com/netbirdio/netbird/shared/relay/messages/address"
//nolint:staticcheck
authmsg "github.com/netbirdio/netbird/shared/relay/messages/auth"
)
const (
@@ -23,55 +17,30 @@ const (
type Validator interface {
Validate(any) error
// Deprecated: Use Validate instead.
ValidateHelloMsgType(any) error
}
// preparedMsg contains the marshalled success response messages
// preparedMsg contains the marshalled success response message
type preparedMsg struct {
responseHelloMsg []byte
responseAuthMsg []byte
responseAuthMsg []byte
}
func newPreparedMsg(instanceURL string) (*preparedMsg, error) {
rhm, err := marshalResponseHelloMsg(instanceURL)
if err != nil {
return nil, err
}
ram, err := messages.MarshalAuthResponse(instanceURL)
if err != nil {
return nil, fmt.Errorf("failed to marshal auth response msg: %w", err)
}
return &preparedMsg{
responseHelloMsg: rhm,
responseAuthMsg: ram,
responseAuthMsg: ram,
}, nil
}
func marshalResponseHelloMsg(instanceURL string) ([]byte, error) {
addr := &address.Address{URL: instanceURL}
addrData, err := addr.Marshal()
if err != nil {
return nil, fmt.Errorf("failed to marshal response address: %w", err)
}
//nolint:staticcheck
responseMsg, err := messages.MarshalHelloResponse(addrData)
if err != nil {
return nil, fmt.Errorf("failed to marshal hello response: %w", err)
}
return responseMsg, nil
}
type handshake struct {
conn listener.Conn
validator Validator
preparedMsg *preparedMsg
handshakeMethodAuth bool
peerID *messages.PeerID
peerID *messages.PeerID
}
func (h *handshake) handshakeReceive(ctx context.Context) (*messages.PeerID, error) {
@@ -93,17 +62,11 @@ func (h *handshake) handshakeReceive(ctx context.Context) (*messages.PeerID, err
return nil, fmt.Errorf("determine message type from %s: %w", h.conn.RemoteAddr(), err)
}
var peerID *messages.PeerID
switch msgType {
//nolint:staticcheck
case messages.MsgTypeHello:
peerID, err = h.handleHelloMsg(buf)
case messages.MsgTypeAuth:
h.handshakeMethodAuth = true
peerID, err = h.handleAuthMsg(buf)
default:
if msgType != messages.MsgTypeAuth {
return nil, fmt.Errorf("invalid message type %d from %s", msgType, h.conn.RemoteAddr())
}
peerID, err := h.handleAuthMsg(buf)
if err != nil {
return peerID, err
}
@@ -112,46 +75,17 @@ func (h *handshake) handshakeReceive(ctx context.Context) (*messages.PeerID, err
}
func (h *handshake) handshakeResponse(ctx context.Context) error {
var responseMsg []byte
if h.handshakeMethodAuth {
responseMsg = h.preparedMsg.responseAuthMsg
} else {
responseMsg = h.preparedMsg.responseHelloMsg
}
if _, err := h.conn.Write(ctx, responseMsg); err != nil {
if _, err := h.conn.Write(ctx, h.preparedMsg.responseAuthMsg); err != nil {
return fmt.Errorf("handshake response write to %s (%s): %w", h.peerID, h.conn.RemoteAddr(), err)
}
return nil
}
func (h *handshake) handleHelloMsg(buf []byte) (*messages.PeerID, error) {
//nolint:staticcheck
peerID, authData, err := messages.UnmarshalHelloMsg(buf)
if err != nil {
return nil, fmt.Errorf("unmarshal hello message: %w", err)
}
log.Warnf("peer %s (%s) is using deprecated initial message type", peerID, h.conn.RemoteAddr())
authMsg, err := authmsg.UnmarshalMsg(authData)
if err != nil {
return nil, fmt.Errorf("unmarshal auth message: %w", err)
}
//nolint:staticcheck
if err := h.validator.ValidateHelloMsgType(authMsg.AdditionalData); err != nil {
return nil, fmt.Errorf("validate %s (%s): %w", peerID, h.conn.RemoteAddr(), err)
}
return peerID, nil
}
func (h *handshake) handleAuthMsg(buf []byte) (*messages.PeerID, error) {
rawPeerID, authPayload, err := messages.UnmarshalAuthMsg(buf)
if err != nil {
return nil, fmt.Errorf("unmarshal hello message: %w", err)
return nil, fmt.Errorf("unmarshal auth message: %w", err)
}
if err := h.validator.Validate(authPayload); err != nil {

View File

@@ -51,7 +51,10 @@ func (l *Listener) Listen(acceptFn func(conn relaylistener.Conn)) error {
log.Infof("QUIC client connected from: %s", session.RemoteAddr())
conn := NewConn(session)
acceptFn(conn)
// Run the accept handler (which performs the pre-auth handshake) in its
// own goroutine so a slow or stalled handshake cannot block accepting
// further connections.
go acceptFn(conn)
}
}

View File

@@ -5164,6 +5164,10 @@ components:
type: boolean
description: Whether upstream TLS certificate verification is skipped when the proxy dials this provider's URL. Intended for self-hosted / internal gateways behind a private or self-signed certificate.
example: false
metadata_disabled:
type: boolean
description: Whether identity metadata injection is disabled for this provider. When enabled (the default), the proxy stamps the caller's user and authorizing group onto upstream requests as provider-specific metadata (e.g. AWS Bedrock's X-Amzn-Bedrock-Request-Metadata header). Set true to suppress it.
example: false
created_at:
type: string
format: date-time
@@ -5184,6 +5188,7 @@ components:
- models
- enabled
- skip_tls_verification
- metadata_disabled
- created_at
- updated_at
AgentNetworkProviderRequest:
@@ -5240,6 +5245,10 @@ components:
type: boolean
description: Skip upstream TLS certificate verification when the proxy dials this provider's URL. For self-hosted / internal gateways behind a private or self-signed certificate. Defaults to false. When omitted on update, the stored value is left unchanged.
example: false
metadata_disabled:
type: boolean
description: Disable identity metadata injection (the caller's user + authorizing group) for this provider. Defaults to false (metadata is injected). When omitted on update, the stored value is left unchanged.
example: false
required:
- provider_id
- name

View File

@@ -2227,6 +2227,9 @@ type AgentNetworkProvider struct {
// IdentityHeaderUserId Wire header name the proxy stamps with the caller's display identity (user email or peer name) when the catalog entry's HeaderPair is `customizable`. Empty disables stamping for this dimension. Ignored when the catalog entry has a fixed HeaderPair (e.g. LiteLLM, Portkey). Used today by Bifrost: typical values are `x-bf-lh-netbird_user_id` (always-on log metadata) or `x-bf-dim-netbird_user_id` (Prometheus / OTEL — requires the label to be pre-declared in the gateway's `client.prometheus_labels` config).
IdentityHeaderUserId *string `json:"identity_header_user_id,omitempty"`
// MetadataDisabled Whether identity metadata injection is disabled for this provider. When enabled (the default), the proxy stamps the caller's user and authorizing group onto upstream requests as provider-specific metadata (e.g. AWS Bedrock's X-Amzn-Bedrock-Request-Metadata header). Set true to suppress it.
MetadataDisabled bool `json:"metadata_disabled"`
// Models Models exposed through this endpoint, with the operator's per-1k input/output prices. Empty means all catalog models are allowed at catalog prices.
Models []AgentNetworkProviderModel `json:"models"`
@@ -2278,6 +2281,9 @@ type AgentNetworkProviderRequest struct {
// IdentityHeaderUserId Wire header name for the caller's display identity. See AgentNetworkProvider.identity_header_user_id. When omitted on a request, the stored value is left unchanged; pass an empty string explicitly to clear it (which disables stamping for this dimension).
IdentityHeaderUserId *string `json:"identity_header_user_id,omitempty"`
// MetadataDisabled Disable identity metadata injection (the caller's user + authorizing group) for this provider. Defaults to false (metadata is injected). When omitted on update, the stored value is left unchanged.
MetadataDisabled *bool `json:"metadata_disabled,omitempty"`
// Models Models exposed through this endpoint, with the operator's per-1k input/output prices. Empty means all catalog models are allowed at catalog prices.
Models *[]AgentNetworkProviderModel `json:"models,omitempty"`

View File

@@ -8,7 +8,3 @@ type Auth struct {
func (a *Auth) Validate(any) error {
return nil
}
func (a *Auth) ValidateHelloMsgType(any) error {
return nil
}

View File

@@ -1,10 +1,8 @@
package hmac
import (
"bytes"
"crypto/hmac"
"encoding/base64"
"encoding/gob"
"fmt"
"hash"
"strconv"
@@ -18,14 +16,6 @@ type Token struct {
Signature string
}
func unmarshalToken(payload []byte) (Token, error) {
var creds Token
buffer := bytes.NewBuffer(payload)
decoder := gob.NewDecoder(buffer)
err := decoder.Decode(&creds)
return creds, err
}
// TimedHMAC generates a token with TTL and uses a pre-shared secret known to the relay server
type TimedHMAC struct {
secret string

View File

@@ -1,33 +0,0 @@
package hmac
import (
"crypto/sha256"
"fmt"
"time"
log "github.com/sirupsen/logrus"
)
type TimedHMACValidator struct {
*TimedHMAC
}
func NewTimedHMACValidator(secret string, duration time.Duration) *TimedHMACValidator {
ta := NewTimedHMAC(secret, duration)
return &TimedHMACValidator{
ta,
}
}
func (a *TimedHMACValidator) Validate(credentials any) error {
b, ok := credentials.([]byte)
if !ok {
return fmt.Errorf("invalid credentials type")
}
c, err := unmarshalToken(b)
if err != nil {
log.Debugf("failed to unmarshal token: %s", err)
return err
}
return a.TimedHMAC.Validate(sha256.New, c)
}

View File

@@ -1,28 +1,19 @@
package auth
import (
"time"
auth "github.com/netbirdio/netbird/shared/relay/auth/hmac"
authv2 "github.com/netbirdio/netbird/shared/relay/auth/hmac/v2"
)
type TimedHMACValidator struct {
authenticatorV2 *authv2.Validator
authenticator *auth.TimedHMACValidator
}
func NewTimedHMACValidator(secret []byte, duration time.Duration) *TimedHMACValidator {
func NewTimedHMACValidator(secret []byte) *TimedHMACValidator {
return &TimedHMACValidator{
authenticatorV2: authv2.NewValidator(secret),
authenticator: auth.NewTimedHMACValidator(string(secret), duration),
}
}
func (a *TimedHMACValidator) Validate(credentials any) error {
return a.authenticatorV2.Validate(credentials)
}
func (a *TimedHMACValidator) ValidateHelloMsgType(credentials any) error {
return a.authenticator.Validate(credentials)
}

View File

@@ -1,21 +0,0 @@
// Deprecated: This package is deprecated and will be removed in a future release.
package address
import (
"bytes"
"encoding/gob"
"fmt"
)
type Address struct {
URL string
}
func (addr *Address) Marshal() ([]byte, error) {
var buf bytes.Buffer
enc := gob.NewEncoder(&buf)
if err := enc.Encode(addr); err != nil {
return nil, fmt.Errorf("encode Address: %w", err)
}
return buf.Bytes(), nil
}

View File

@@ -1,43 +0,0 @@
// Deprecated: This package is deprecated and will be removed in a future release.
package auth
import (
"bytes"
"encoding/gob"
"fmt"
)
type Algorithm int
const (
AlgoUnknown Algorithm = iota
AlgoHMACSHA256
AlgoHMACSHA512
)
func (a Algorithm) String() string {
switch a {
case AlgoHMACSHA256:
return "HMAC-SHA256"
case AlgoHMACSHA512:
return "HMAC-SHA512"
default:
return "Unknown"
}
}
type Msg struct {
AuthAlgorithm Algorithm
AdditionalData []byte
}
func UnmarshalMsg(data []byte) (*Msg, error) {
var msg *Msg
buf := bytes.NewBuffer(data)
dec := gob.NewDecoder(buf)
if err := dec.Decode(&msg); err != nil {
return nil, fmt.Errorf("decode Msg: %w", err)
}
return msg, nil
}

View File

@@ -14,9 +14,10 @@ const (
CurrentProtocolVersion = 1
MsgTypeUnknown MsgType = 0
// Deprecated: Use MsgTypeAuth instead.
MsgTypeHello = 1
// Deprecated: Use MsgTypeAuthResponse instead.
// MsgTypeHello and MsgTypeHelloResponse are the removed legacy handshake
// message types. They are retained only to reserve wire values 1 and 2 so
// the values are never reused; the server rejects both.
MsgTypeHello = 1
MsgTypeHelloResponse = 2
MsgTypeTransport = 3
MsgTypeClose = 4
@@ -42,10 +43,6 @@ const (
offsetAuthPeerID = sizeOfProtoHeader + sizeOfMagicByte
headerTotalSizeAuth = sizeOfProtoHeader + headerSizeAuth
// hello message
headerSizeHello = sizeOfMagicByte + peerIDSize
headerSizeHelloResp = 0
// transport
headerSizeTransport = peerIDSize
offsetTransportID = sizeOfProtoHeader
@@ -113,7 +110,6 @@ func DetermineClientMessageType(msg []byte) (MsgType, error) {
msgType := MsgType(msg[1])
switch msgType {
case
MsgTypeHello,
MsgTypeAuth,
MsgTypeTransport,
MsgTypeClose,
@@ -135,7 +131,6 @@ func DetermineServerMessageType(msg []byte) (MsgType, error) {
msgType := MsgType(msg[1])
switch msgType {
case
MsgTypeHelloResponse,
MsgTypeAuthResponse,
MsgTypeTransport,
MsgTypeClose,
@@ -148,67 +143,6 @@ func DetermineServerMessageType(msg []byte) (MsgType, error) {
}
}
// Deprecated: Use MarshalAuthMsg instead.
// MarshalHelloMsg initial hello message
// The Hello message is the first message sent by a client after establishing a connection with the Relay server. This
// message is used to authenticate the client with the server. The authentication is done using an HMAC method.
// The protocol does not limit to use HMAC, it can be any other method. If the authentication failed the server will
// close the network connection without any response.
func MarshalHelloMsg(peerID PeerID, additions []byte) ([]byte, error) {
msg := make([]byte, sizeOfProtoHeader+sizeOfMagicByte, sizeOfProtoHeader+headerSizeHello+len(additions))
msg[0] = byte(CurrentProtocolVersion)
msg[1] = byte(MsgTypeHello)
copy(msg[sizeOfProtoHeader:sizeOfProtoHeader+sizeOfMagicByte], magicHeader)
msg = append(msg, peerID[:]...)
msg = append(msg, additions...)
return msg, nil
}
// Deprecated: Use UnmarshalAuthMsg instead.
// UnmarshalHelloMsg extracts peerID and the additional data from the hello message. The Additional data is used to
// authenticate the client with the server.
func UnmarshalHelloMsg(msg []byte) (*PeerID, []byte, error) {
if len(msg) < sizeOfProtoHeader+headerSizeHello {
return nil, nil, ErrInvalidMessageLength
}
if !bytes.Equal(msg[sizeOfProtoHeader:sizeOfProtoHeader+sizeOfMagicByte], magicHeader) {
return nil, nil, errors.New("invalid magic header")
}
peerID := PeerID(msg[sizeOfProtoHeader+sizeOfMagicByte : sizeOfProtoHeader+headerSizeHello])
return &peerID, msg[headerSizeHello:], nil
}
// Deprecated: Use MarshalAuthResponse instead.
// MarshalHelloResponse creates a response message to the hello message.
// In case of success connection the server response with a Hello Response message. This message contains the server's
// instance URL. This URL will be used by choose the common Relay server in case if the peers are in different Relay
// servers.
func MarshalHelloResponse(additionalData []byte) ([]byte, error) {
msg := make([]byte, sizeOfProtoHeader, sizeOfProtoHeader+headerSizeHelloResp+len(additionalData))
msg[0] = byte(CurrentProtocolVersion)
msg[1] = byte(MsgTypeHelloResponse)
msg = append(msg, additionalData...)
return msg, nil
}
// Deprecated: Use UnmarshalAuthResponse instead.
// UnmarshalHelloResponse extracts the additional data from the hello response message.
func UnmarshalHelloResponse(msg []byte) ([]byte, error) {
if len(msg) < sizeOfProtoHeader+headerSizeHelloResp {
return nil, ErrInvalidMessageLength
}
return msg, nil
}
// MarshalAuthMsg initial authentication message
// The Auth message is the first message sent by a client after establishing a connection with the Relay server. This
// message is used to authenticate the client with the server. The authentication is done using an HMAC method.

View File

@@ -4,28 +4,11 @@ import (
"testing"
)
func TestMarshalHelloMsg(t *testing.T) {
peerID := HashID("abdFAaBcawquEiCMzAabYosuUaGLtSNhKxz+")
msg, err := MarshalHelloMsg(peerID, nil)
if err != nil {
t.Fatalf("error: %v", err)
}
msgType, err := DetermineClientMessageType(msg)
if err != nil {
t.Fatalf("error: %v", err)
}
if msgType != MsgTypeHello {
t.Errorf("expected %d, got %d", MsgTypeHello, msgType)
}
receivedPeerID, _, err := UnmarshalHelloMsg(msg)
if err != nil {
t.Fatalf("error: %v", err)
}
if receivedPeerID.String() != peerID.String() {
t.Errorf("expected %s, got %s", peerID, receivedPeerID)
func TestDetermineClientMessageTypeRejectsHello(t *testing.T) {
// The reserved legacy Hello message (type 1) must be rejected by the server.
msg := []byte{byte(CurrentProtocolVersion), byte(MsgTypeHello)}
if _, err := DetermineClientMessageType(msg); err == nil {
t.Fatalf("expected hello message type to be rejected")
}
}