Repository navigation
Conversation
The Container group of the session create and restart forms gets an "Allow all network access" checkbox above Extra allowed domains. It defaults to Enclave's resolved allow_all_network and is locked on under the QEMU backend. Like hostConfig, it is recorded and passed only when it differs from the config (--allow-all-network, or =false), and Restart and Continue replay it. CLI-started sessions pick it up from ps --json. While all network is allowed, --allow-domain is not passed: nothing enforces it on Docker, and QEMU rejects it. The form greys the domains out through a generic daemon-sent schema keyword, x-disabled-when, and keeps their value. Refs eclipse-enclave#15
Only the Add input greyed out, so a list disabled by x-disabled-when, or by a form in flight, kept its enabled look in a screenshot.
EclipseSourceAI
left a comment
There was a problem hiding this comment.
Note
Autonomous AI review. May contain mistakes.
Useful comment? React 👍. Disagree? React 👎 to reject it for follow-up reviews. A reply explaining why helps too.
This agent never approves PRs. A maintainer must sign off.
Only a maintainer can re-request a review from this agent.
Powered by Eclipse Enclave via review-guard-mcp
This PR adds an Allow all network access checkbox to the Enclave Container settings. It passes --allow-all-network, or --allow-all-network=false when the session should be stricter than the config. The checkbox defaults to Enclave's resolved allow_all_network, and Restart, Continue and CLI-adopted sessions replay it. It also adds a generic x-disabled-when schema keyword: the daemon states the rule and the form only compares values. The form uses it to grey out Extra allowed domains while all network is allowed, and the daemon stops passing --allow-domain then. The daemon/UI split matches the rest of the schema form, and both unit and e2e tests are included.
What a maintainer should look at:
- QEMU special case: HomeShell always starts sessions with
--background, and Enclave's QEMU backend can't run detached.config --jsonalso already reports the forced value. So the QEMU branch and the read-only lock can't be reached or tested. - Network policy mode: the known limitation is documented as something HomeShell can't see.
enclave network print --jsondoes report the effectivemode. - Drift after upgrade: sessions started under an unrestricted config with extra domains will show as out of date on
allow_domainsonce this lands.
UI check in make dev-demo (Playwright):
- The checkbox sits right above Extra allowed domains, unticked by default, with the token warning.
- Ticking it greys out the domains box. The chips can't be removed and nothing can be added. The values survive unticking, and they are recorded in the launch config.
- The only visual issue is the double-dimmed chip × (inline comment).
- Drop the QEMU case: HomeShell always starts sessions with --background, which Enclave's QEMU backend does not support, so it could never run. - Say HomeShell does not yet ask `enclave network print --json` for the network policy mode, rather than that it cannot see it, and that --allow-all-network=false cannot override an unrestricted mode. - An unrestricted session still runs an observing gateway; the domains do not apply because unrestricted mode has no allowlist. - Drift skips allow_domains while both the session and the config allow all network, so sessions started that way before are not flagged. - Keep the chip × at full opacity inside a disabled list box, so the box's dimming is the only one. Refs eclipse-enclave#15
EclipseSourceAI
left a comment
There was a problem hiding this comment.
Note
Autonomous AI review. May contain mistakes.
Useful comment? React 👍. Disagree? React 👎 to reject it for follow-up reviews. A reply explaining why helps too.
This agent never approves PRs. A maintainer must sign off.
Only a maintainer can re-request a review from this agent.
Powered by Eclipse Enclave via review-guard-mcp
Follow-up review of 37f3799, which addresses my earlier review from 8ecca53.
All five points I raised are genuinely fixed, not just argued against:
- The QEMU branch, its
readOnlylock, and the related tests are fully removed (confirmed noqemureferences remain anywhere underdaemon/ordocs/). docs/PROVIDERS.mdand theallowAllNetworkPropcomment now say the network policy mode is "not asked for yet" instead of implying it's invisible, and note that--allow-all-network=falsecan't override an unrestricted policy mode.- The
schema.gocomment no longer claims the gateway disappears in unrestricted mode; it now correctly says unrestricted mode has no allowlist to extend. changedOptionsindrift.gonow skipsallow_domainswhen both the running session and the current config allow all network, with a table-driven test (TestChangedOptionsAllowDomainsUnderAllNetwork) that correctly still compares when the session reports the option as missing or null.- The chip
×double-dimming is fixed inschema-fields.module.cssby resetting its opacity to 1 inside a disabled box, backed by a new e2e assertion insession-settings.spec.ts.
I tried to re-verify the chip fix live via Playwright this round, but the sandbox has no Go toolchain available to run the daemon half of make dev-demo, so I couldn't exercise the UI. The CSS rule and the new e2e assertion both check out by reading the code and match the selector structure used elsewhere in the same file.
The diff since my last review (8ecca53..37f3799) is scoped entirely to these five fixes, no other changes or new issues to flag.
These previous comments can be resolved as they are now handled:
- QEMU branch, readOnly lock and its tests were unreachable dead code; now removed
- docs implied the network policy mode is invisible to HomeShell; reworded to "not asked for yet"
- comment wrongly said the gateway disappears in unrestricted mode; corrected to "no allowlist to extend"
- drift check would flag allow_domains as changed even though it has no effect under all-network; now skipped with tests
- chip × was double-dimmed inside a disabled list box; opacity reset so only the box dims
I can't resolve them myself as I would need write permission on this repository.
What and why
Closes #15. The Container settings of the session start and restart forms get an Allow all network
access checkbox, directly above Extra allowed domains. It passes Enclave's
--allow-all-network.I found no documented reason why HomeShell left this option out, so this PR adds it.
allow_all_network, with a caption naming the config layer itcame from.
--allow-all-network=false. Like Host config, the setting is recorded and passed only when itdiffers from the config, and Restart and Continue replay it. CLI-started sessions pick it up from
enclave ps --json.Enclave injects secrets with their real values.
and no
--allow-domainis passed. Unrestricted mode has no allowlist for them to extend."x-disabled-when": {"allowAllNetwork": true}. The UI only compares values. The keyword isdocumented in
docs/API.mdnext tox-sourceandx-inherited.Add input. Before, the greyed-out domains looked editable.
no longer flagged out of date on
allow_domains, as the domains have no effect there.enclave network print --jsonfor Enclave'snetwork policy mode (
enclave network set-mode unrestricted, or a project policy). A sessionunder an unrestricted mode runs unrestricted whatever the checkbox says, as
--allow-all-network=falsecannot override that mode. This is documented indocs/PROVIDERS.md.To try it, run
make dev-demo, start a containerized session and open its Container group. With areal Enclave, set
allow_all_network = truein a project config to see the box start ticked.Changelog
Checklist
AGENTS.md)AGENTS.md