The SSRF blocklist has no escape hatch, so a self-hosted proxy cannot forward to your own network #204

New Issue

clawbot · 2026-08-20T05:47:37+02:00

clawbot commented

2026-08-20 05:47:37 +02:00

internal/delivery/ssrf.go:38-72 hardcodes the blocked ranges with no configuration field. RFC1918, loopback and CGNAT are permanently unreachable as delivery destinations.

That guard is correct as a default and must stay the default — it is what #36, #68 and #69 bought. But webhooker is self-hosted software whose main use is receiving a public webhook and forwarding it to an internal service: a container on the same Docker network, a box on 10.x, a service on 127.0.0.1. All of those are refused today, with no way to permit one.

Concrete evidence of the bite: during the deployability run I could not point a target at a sink on the same host, because every address on it is in the blocklist. The end-to-end test had to use a public internet echo service instead.

Definition of done:

a config value (e.g. ALLOWED_EGRESS_CIDRS, defaulting empty) allows specific CIDRs through the guard
it is a CIDR allowlist, not a boolean off switch: no setting disables SSRF protection wholesale
the link-local metadata range stays blocked unconditionally, even if a supplied CIDR would cover it, since that is credential theft rather than an internal target
set-but-unparseable values abort startup, per #80 and repo policy
the allowlist applies at BOTH target-creation validation and delivery time, so the two paths cannot disagree (that split was the #68 and #69 bug)
startup logs the effective allowlist when non-empty, so an operator can see the hole they opened
tests: a permitted CIDR delivers, a non-permitted private address is still refused, metadata stays refused even when covered by a supplied CIDR
README documents it with the risk stated plainly

`internal/delivery/ssrf.go:38-72` hardcodes the blocked ranges with no configuration field. RFC1918, loopback and CGNAT are permanently unreachable as delivery destinations. That guard is correct as a default and must stay the default — it is what #36, #68 and #69 bought. But webhooker is self-hosted software whose main use is receiving a public webhook and forwarding it to an internal service: a container on the same Docker network, a box on `10.x`, a service on `127.0.0.1`. All of those are refused today, with no way to permit one. Concrete evidence of the bite: during the deployability run I could not point a target at a sink on the same host, because every address on it is in the blocklist. The end-to-end test had to use a public internet echo service instead. Definition of done: - a config value (e.g. `ALLOWED_EGRESS_CIDRS`, defaulting empty) allows specific CIDRs through the guard - it is a CIDR allowlist, not a boolean off switch: no setting disables SSRF protection wholesale - the link-local metadata range stays blocked unconditionally, even if a supplied CIDR would cover it, since that is credential theft rather than an internal target - set-but-unparseable values abort startup, per #80 and repo policy - the allowlist applies at BOTH target-creation validation and delivery time, so the two paths cannot disagree (that split was the #68 and #69 bug) - startup logs the effective allowlist when non-empty, so an operator can see the hole they opened - tests: a permitted CIDR delivers, a non-permitted private address is still refused, metadata stays refused even when covered by a supplied CIDR - README documents it with the risk stated plainly

clawbot added this to the 1.0.0 milestone 2026-08-20 05:47:37 +02:00

clawbot referenced this issue

2026-08-20 05:49:42 +02:00

1.0/mvp #33

clawbot referenced this issue

2026-08-20 05:56:52 +02:00

Milestone 1.0.0: internet-facing readiness #111

clawbot referenced this issue from a commit

2026-08-20 06:20:09 +02:00

Add an egress CIDR allowlist to the SSRF guard (closes #204)

clawbot referenced a pull request that will close this issue

2026-08-20 06:20:36 +02:00