From 0623fa3323a42fed8639e4f308f7747d751d7930 Mon Sep 17 00:00:00 2001 From: Zeusina Date: Tue, 13 Oct 2026 16:10:37 +0300 Subject: [PATCH] docs: record DoH outage and live validation Co-authored-by: GPT-6 --- MANAGED_TARGET.md | 62 ++++++++++++++++++++++----- PROGRESS.md | 10 ++++- deploy/managed-target/README.md | 75 ++++++++++++++++++++++++++++++--- 3 files changed, 128 insertions(+), 19 deletions(-) diff --git a/MANAGED_TARGET.md b/MANAGED_TARGET.md index f3ede0f..08e52a5 100644 --- a/MANAGED_TARGET.md +++ b/MANAGED_TARGET.md @@ -1,5 +1,42 @@ # Fresh managed-test target contract (isolated acceptance, not production approval) +## Managed DoH live acceptance (2026-10-13): DNS verified, NOT READY + +Supplied verified reports bind source `544dda7c85965fc3200de3af17fca9ef9999b2ce` +(DoH `367d956`, TCP deadline `721cad4`, public-health `socks5h` fix `544dda7`) +to image `sha256:b3085b7ceca4a629c28bd58f23bf97f02a34c2a38df08b57756e40594ba2d427`, +container `4ad54175a54c`, owned SOCKS `127.0.0.1:11082`. +The 303-second LIVE snapshot retained root/DNS22/SS59,60/monitor253 PIDs, +children PPID1, owned readiness ACK and parent FD8/9 anchors. UDP/TCP DNS +IDs/questions matched, rcode0, answers2/8; SOCKS `0500`, WWW HTTP200/SSL0 +and configured `health youtube` exit0 passed. Wire evidence: 41 marked direct +DoH connections, zero external port53. Scope `suffix`, generation9, IDs `[1,9]` +and hash preservation versus staging are measured snapshots, not universal +state; autonomous updates have occurred since the earlier generation8 snapshot. + +In isolated staging, blocking both bootstrap IPs for 61 seconds returned +UDP/TCP SERVFAIL/rcode2 at about 5000ms without changing root, DNS, SS or monitor +PIDs or sending external port53 traffic. DNS recovered immediately after removal; +WWW HTTP200 passed after 46 seconds without restart. Graceful stop: 1 second, +exit0, no survivors. Staging nft fault rules were removed; this is a tested DNS +dependency failure, not all-outage or nonconfiguration-guard acceptance. +No new runtime firewall policy beyond normal generated native rules is supplied. + +Existing shared GGC qualification is separate evidence, not an 18-transfer rerun +under DoH. Fresh TLS1.3 media: HTTP206/65536 bytes/MP4 `ftyp`/SSL0; TLS1.2 +TIMEOUT is not qualified. Avatar/thumbnail qualification remains independently +outstanding; no whole-service READY claim. Historical media results below do +not close these gates on this image. Generic service integration and maximum +strategy collection remain future work, without a new policy framework. + +Private supplied reports (not reread here): +`/home/agent/zapret2-autotune-rc-doh-721cad4/live-upgrade-private/`, then latest +`/home/agent/zapret2-autotune-rc-doh-544dda7/live-upgrade-private/`. +No deployment/probes accompany this docs update; later docs are not tested source. +State/credentials/private env were preserved, staging containers removed, old +user SS1080 and user-stopped proxy unchanged. See the deployment guide for the +four DNS settings, LIVE env location, privacy boundary and recreation lifecycle. + ## Autonomous three-CDN acceptance (2026-10-13): media qualified, NOT READY Supplied field results bind source `3fc8563995e890310580ba8f78850b92fea42136` @@ -73,7 +110,7 @@ must supply all of the following before any managed command can run: | `ZAPRET_AUTOTUNE_TARGET_DIR` | `/var/lib/zapret-autotune/target` | | `ZAPRET_CONFIG` | `/var/lib/zapret-autotune/target/config` | | `SOCKS_PORT` | Base image's actual ss-local service port (canonical decimal, 1–65535) | -| `ZAPRET_AUTOTUNE_SOCKS_PROXY` | `socks5://127.0.0.1:$SOCKS_PORT` (expanded value) | +| `ZAPRET_AUTOTUNE_SOCKS_PROXY` | `socks5h://127.0.0.1:$SOCKS_PORT` (expanded value; legacy `socks5` validated and canonicalized internally) | The fixed directory must already be a private runtime-user-owned **dedicated persistent volume mountpoint**, not just a directory under another mount. @@ -222,9 +259,9 @@ DB/safety/journal/runner failures stop nonzero; joined cancellation plus a clean failure is also fatal. A qualified frontend at startup is probed immediately, not rescanned merely because the process restarted. -Entrypoint ordering is SS children → listener → fresh managed init if needed → managed reconcile → zapret start +Entrypoint ordering is owned DNS readiness (default DoH) → SS children → listener → fresh managed init if needed → managed reconcile → zapret start → one installed `/usr/local/bin/zapret-autotune managed monitor` child. It supervises -all three children; unexpected monitor exit is not a successful shutdown. +the DNS, SS and monitor children; unexpected monitor exit is not a successful shutdown. SIGTERM/SIGINT cancels and waits the monitor before target-local zapret stop and SS child termination/wait. Existing targets and Xray remain unchanged. Deployment allows 240s Compose shutdown grace for bounded @@ -362,10 +399,11 @@ signed ordinary media host samples are required; multiple VideoIDs may yield one CDN host and leave `DEGRADED` with no authorized suffix. Sampled thumbnail/avatar routes do not prove availability of the whole thumbnail pool. The reported `8.8.8.8` DNS diagnostic corrected a tampered avatar, not all thumbnail routes; -native DNS correction is not full asset availability proof. Target DNS uses -validated native `-d` argv (nonsecret), with installed-option support checked -at image build; passwords remain private JSON only. No DoH bridge, permanent -Google IP pins, global host DNS edits or external startup manager are introduced. +native DNS correction is not full asset availability proof. Default DoH uses +SS `-d 127.0.0.1:5353`; explicit native mode uses validated literal `-d` argv, +with installed-option support checked at image build. Passwords remain private +JSON only. No permanent service IP pins, global host DNS edits or external +startup manager are introduced. ### Supervision and failure handling @@ -379,13 +417,16 @@ errors never authorize retry. No new dependency, host service installation or external process manager is required. SIGTERM/SIGINT cancels backoff or the running monitor, waits for bounded restoration and joins children before exit. -In default DoH mode PID 1 first starts the Go UDP/TCP loopback DNS forwarder, +In default DoH mode PID 1 first starts the stdlib UDP/TCP loopback DNS forwarder +inside the existing Go binary (no dependency or new container), waits at most five seconds for its owned private PID marker, and only then starts ss-server with `-d 127.0.0.1:5353`. No host DNS port or DNS/iptables changes are made. Hostname SOCKS traffic and managed `socks5h` checks use the same encrypted HTTPS resolver; clients must enable remote SOCKS DNS, not resolve locally or send IP literals. Bootstrap uses configured literal IPs with verified TLS and -marked direct sockets excluding probe/live NFQUEUE. Endpoint failures produce +mark `0x50000000` direct sockets excluding probe/live NFQUEUE, with no system DNS, proxy +or plaintext fallback. The DoH resolver sees queries directly, not SS-tunneled; +the plaintext helper exchange stays on loopback5353. Endpoint failures produce SERVFAIL without child exit or plaintext fallback. A forwarder crash, unlike an upstream outage, uses the same bounded technical retry budget. The DNS child stays alive through monitor restoration, zapret stop and SS joins; it is then @@ -393,7 +434,8 @@ terminated/joined within five seconds and only its validated readiness inode is removed using PID1-held directory/file anchors. Foreign/stale/replaced readiness is preserved and vetoes startup/retry; SIGKILL does not lose owned cleanup proof. Explicit native mode has no DNS child. See the deployment README for provider settings -and scope limits. New DoH runtime acceptance remains pending. +and scope limits. Supplied staged-outage and LIVE DNS acceptance are recorded +above; they do not establish all-outage resilience or whole-service readiness. Compose restart remains disabled: a blanket container restart could loop over a safety refusal or reset the in-container retry budget. Compose stop grace is 240s: diff --git a/PROGRESS.md b/PROGRESS.md index 94759f0..f322e17 100644 --- a/PROGRESS.md +++ b/PROGRESS.md @@ -4,12 +4,18 @@ ## Текущий статус -### Managed DoH integration (2026-10-13): local implementation, remote gate pending +### Managed DoH acceptance (2026-10-13): live DNS path and staged outage verified, NOT READY - Managed target defaults to the stdlib Go UDP/TCP loopback DoH forwarder; PID 1 validates owned readiness before ss-server `-d 127.0.0.1:5353`. SOCKS remote-hostname traffic and managed `socks5h` checks share this path. Provider URL/bootstrap IPs/timeout are private-env configurable; explicit native mode alone preserves legacy DNS, never automatic fallback. - DNS stays available during monitor restoration, zapret stop and SS joins, then is terminated/joined with a five-second watchdog and readiness cleanup. DNS child death uses existing bounded retries; upstream failures return SERVFAIL without stopping the forwarder. No host DNS port, global DNS changes, new dependency or capability. - Security follow-up: PID 1 retains directory/readiness inode descriptors for validated owned cleanup after graceful exit or SIGKILL; foreign/stale/replaced markers are preserved and fail closed. Compose passes empty bootstrap by default, allowing Google defaults only for the exact Google URL; custom URLs require explicit IPs. -- Offline shell lifecycle/static/config fixtures (real readiness CLI with inherited anchors), sh/bash syntax checks, `go test ./... -count=1 -timeout 180s` and targeted readiness/entrypoint `-race` tests passed locally. Coverage includes foreign regular/symlink preservation, SIGKILL owned cleanup/retry, custom-provider Compose validation, signal/cooldown cancellation and DNS-last cleanup. No deployment or external-network tests performed here. Remote encrypted-path, browser/check and upstream-outage acceptance remains pending; earlier PASS is not DoH qualification. +- Earlier offline shell lifecycle/static/config fixtures (real readiness CLI with inherited anchors), sh/bash syntax checks, full Go and targeted readiness/entrypoint race checks passed as previously reported. Coverage includes foreign regular/symlink preservation, SIGKILL owned cleanup/retry, custom-provider Compose validation, signal/cooldown cancellation and DNS-last cleanup. This docs-only update adds no source/runtime changes or probes and does not rerun the Go suite. +- Supplied verified reports bind DoH implementation `367d956`, TCP reply deadline fix `721cad4`, and explicit public-health `socks5h` fix **`544dda7c85965fc3200de3af17fca9ef9999b2ce`** to LIVE image **`sha256:b3085b7ceca4a629c28bd58f23bf97f02a34c2a38df08b57756e40594ba2d427`**, container **`4ad54175a54c`**, owned SOCKS **`127.0.0.1:11082`**. Private reports: `/home/agent/zapret2-autotune-rc-doh-721cad4/live-upgrade-private/`, then latest `/home/agent/zapret2-autotune-rc-doh-544dda7/live-upgrade-private/`; not independently reread here. Later documentation commits are not the tested source. +- Real isolated staging blocked **both** bootstrap IPs for **61 seconds**: UDP and TCP replies returned SERVFAIL (rcode2) at about **5000ms**; DNS, monitor, SS and root PIDs stayed unchanged, with zero external port53 traffic. After removal, DNS recovered immediately and WWW HTTP200 was verified after **46 seconds**, without process restart. Graceful stop took **1 second**, exit0, no survivors. This proves the tested DNS dependency outage, not every conceivable outage or nonconfiguration guard behavior. Staging-only nft fault rules were removed; no new runtime firewall policy was added beyond normal generated native rules. +- LIVE **303-second** measured snapshot: same root and children DNS22, SS59/60, monitor253, all children PPID1; owned readiness ACK and PID1-held FD8/9 anchors were positive. UDP/TCP DNS IDs and questions matched, rcode0, answer counts2/8. SOCKS greeting `0500`, WWW HTTP200/SSLVerifyResult0, and configured `health youtube` exit0 passed. Wire evidence counted **41 marked DoH connections**, zero external port53; direct HTTPS sockets use mark **`0x50000000`** to bypass live/probe NFQUEUE. Scope `suffix`, generation9, IDs `[1,9]`, config hash preserved versus staging are measured snapshot values, not universal constants: autonomous state has advanced since an earlier generation8 observation. +- Existing normal shared GGC qualification remains separate proof; no new 18-transfer qualification under DoH is claimed. Fresh media TLS1.3 returned HTTP206/65536 bytes/MP4 `ftyp`/SSLVerifyResult0; TLS1.2 timed out and is **not qualified**. Avatar and thumbnail qualification remain independent outstanding gates; no full-service READY claim. Maximum strategy collection and generic service integration remain future work, not a new framework. +- DNS configuration is the private mode0600 LIVE `/home/agent/zapret-autotune-cleanops-3fc8563-1791882112/.env`; the four public DNS keys and defaults are documented in the deployment guide. URL/bootstrap come from env, never argv; changing DNS requires official Compose recreation under the controlled PID1 lifecycle. Credentials/private env and state were preserved. Staging cleanup left no new running test containers; the old user SS1080 and user-stopped proxy were unchanged. +- Default `https://dns.google/dns-query` selects literal `8.8.8.8,8.8.4.4` bootstrap only when missing/empty; custom endpoints require explicit IPs. TLS is verified, with no system-DNS, proxy or plaintext fallback; native mode is manual and explicit only. The direct DoH resolver sees queries: they are **not SS-tunneled**. Plaintext UDP/TCP exists only at internal loopback5353. IP-transmitting SOCKS clients using local DNS are outside this guarantee; use remote DNS/`socks5h`. The existing Go binary supplies the stdlib forwarder, no dependency or extra container, bounded to 4096-byte queries, 65535-byte replies, 8192-byte HTTP headers and 16 workers, with no forwarder cache. ### Autonomous three-CDN field acceptance (2026-10-13): media qualified, NOT READY diff --git a/deploy/managed-target/README.md b/deploy/managed-target/README.md index 770d0f0..a14aac7 100644 --- a/deploy/managed-target/README.md +++ b/deploy/managed-target/README.md @@ -5,6 +5,44 @@ reconfigure, or stop the existing `zapret2-ss-autotune` container. The target starts from a disabled managed baseline; `managed monitor` bootstraps the frontend/API strategy, then discovers and applies an exact-host GGC strategy. +## Recorded DoH acceptance (2026-10-13): live DNS and staged outage verified + +Supplied verified reports, not remote evidence reread by this docs update: +source `544dda7c85965fc3200de3af17fca9ef9999b2ce`, managed image +`sha256:b3085b7ceca4a629c28bd58f23bf97f02a34c2a38df08b57756e40594ba2d427`, +container `4ad54175a54c`, owned loopback SOCKS11082. This source includes DoH +`367d956`, TCP timeout-reply fix `721cad4` and public-health `socks5h` fix +`544dda7`. Later docs commits are not the tested image source. + +- LIVE 303-second snapshot: unchanged root and DNS22/SS59,60/monitor253 PIDs, + children PPID1; owned readiness ACK and parent FD8/9 anchors positive. + UDP/TCP IDs/questions correct, rcode0, answers2/8; SOCKS `0500`, WWW + HTTP200/SSLVerifyResult0, configured `health youtube` exit0. Wire evidence: + 41 marked direct DoH connections, zero external port53. Scope `suffix`, + generation9, IDs `[1,9]`, hash preserved versus staging are measured snapshots, + not universal state; autonomous updates occurred since an earlier generation8. +- Staging blocked both literal bootstrap IPs for 61 seconds. UDP/TCP returned + SERVFAIL/rcode2 at about 5000ms; root, DNS, monitor and SS PIDs stayed the same, + external port53 stayed zero. Removing the block restored DNS immediately; + WWW HTTP200 passed after 46 seconds without restart. Graceful stop took + 1 second, exit0, no survivors. This tests a DNS dependency outage, not all + conceivable failures or nonconfiguration guards. Temporary nft fault rules + were staging-only and removed; runtime firewall remains normal generated + native policy, not a new fault-injection policy. +- Existing shared GGC qualification is separate proof: no new 18-transfer + qualification under DoH. Fresh TLS1.3 media returned HTTP206/65536 bytes/MP4 + `ftyp`/SSL0; TLS1.2 TIMEOUT is not qualified. Avatar and thumbnail qualification + remain independent outstanding gates. **No full-service READY claim.** +- Staging cleanup removed the test containers, leaving no new running tests; + old user SS1080 and user-stopped proxy were unchanged. State, credentials and + private env were preserved. No source/runtime changes or probes in this update. + +Private reports: `/home/agent/zapret2-autotune-rc-doh-721cad4/live-upgrade-private/`, +then latest `/home/agent/zapret2-autotune-rc-doh-544dda7/live-upgrade-private/`. +Historical acceptance below remains scoped to its own source/image and DNS mode. +Generic service integration and maximum strategy collection remain future work, +not a new framework. + ## Autonomous three-CDN field acceptance (2026-10-13): media qualified, NOT READY Supplied field snapshot: source `3fc8563995e890310580ba8f78850b92fea42136`, @@ -203,13 +241,16 @@ The current service/adaptive hardware acceptance gate is **pending**; the old clean-run PASS below does not qualify these new modes. **Managed DNS defaults to DoH**, independently of service/YouTube mode. PID 1 -starts the stdlib Go forwarder on container loopback `127.0.0.1:5353` (UDP and +starts the stdlib forwarder inside the existing Go binary, with no dependency +or new container, on container loopback `127.0.0.1:5353` (UDP and TCP), validates its private PID readiness acknowledgement, then starts ss-server with `-d 127.0.0.1:5353`. No DNS port is published. Both client hostname traffic and managed checks share this encrypted resolver path; it works for arbitrary services, not only YouTube. HTTPS uses normal certificate verification, direct -literal-IP bootstrap and a socket mark excluding live/probe NFQUEUE rules; -there is no recursive SOCKS resolution or plaintext fallback. Endpoint errors +literal-IP bootstrap and socket mark `0x50000000` excluding live/probe NFQUEUE rules; +there is no system-DNS, proxy or plaintext fallback, or recursive SOCKS resolution. +The DoH resolver sees queries directly: they are not tunneled through SS. +Only the internal loopback5353 helper exchange is plaintext UDP/TCP. Endpoint errors return SERVFAIL while the forwarder stays alive, rather than switching DNS modes. The private `.env` controls `ZAPRET_AUTOTUNE_DNS_MODE=doh|native`, @@ -224,6 +265,25 @@ or fragment. A path can contain a private provider token: do not print it. Validation reports opaque errors before any services start. The subprocess gets provider configuration through env, never endpoint argv/logging. +Default public settings in the private `.env`: + +```dotenv +ZAPRET_AUTOTUNE_DNS_MODE=doh +ZAPRET_AUTOTUNE_DOH_URL=https://dns.google/dns-query +ZAPRET_AUTOTUNE_DOH_BOOTSTRAP_IPS= +ZAPRET_AUTOTUNE_DOH_TIMEOUT_SEC=5 +``` + +LIVE configuration is +`/home/agent/zapret-autotune-cleanops-3fc8563-1791882112/.env` (mode0600). +These four key names/defaults are public; credentials, custom provider paths, +queries and signed resource URLs are not. URL/bootstrap are never child argv. +Changing DNS requires official Compose recreation using that private env under +the controlled PID1 lifecycle, not a detached helper or ad hoc child restart. +The forwarder bounds queries to 4096 bytes, replies to 65535 bytes, HTTP headers +to 8192 bytes and concurrent workers to 16; it adds no DNS cache. No claim is +made about caching inside the existing SS resolver. + Only explicit `ZAPRET_AUTOTUNE_DNS_MODE=native` restores the old resolver behavior, with no forwarder child. In that mode optional `SS_DNS_SERVER` accepts only an unbracketed literal IPv4/IPv6 address; empty retains the base resolver. @@ -242,9 +302,10 @@ availability proof**. No permanent service IP pins or global host DNS changes are supplied. Clients must request remote DNS (browser SOCKS5 remote DNS or `curl --socks5-hostname 127.0.0.1:11082` when that is the chosen host port). Client-local resolution and IP-literal requests bypass this hostname DNS path; -the bridge does not encrypt unrelated container processes' DNS. The new DoH -integration still needs remote runtime acceptance; earlier clean-run PASS does -not qualify it. PID 1 remains the sole startup/recovery owner. +the bridge does not encrypt unrelated container processes' DNS. IP-transmitting +SOCKS clients with local resolution are not covered: use remote DNS/`socks5h`. +Supplied LIVE and staged-outage DNS acceptance is recorded above; earlier +clean-run PASS is not DoH evidence. PID 1 remains the sole startup/recovery owner. Compose has no automatic restart. PID 1 restarts the entire owned runtime only for technical failures, with three starts per rolling six hours (including initial @@ -295,7 +356,7 @@ use host or shared/container PID namespaces. It does not grant inspection of broader host processes. The full service/adaptive hardware gate remains **pending**, not PASS; this ownership fix alone does not qualify it. -Startup order: ss-server/ss-local → listener 1080 → guarded managed init on +Startup order: owned DNS readiness (default DoH) → ss-server/ss-local → listener 1080 → guarded managed init on fresh state → `managed reconcile` on every startup → target-local `zapret2 start` → supervised managed monitor. Reconciliation uses the fixed private target DB and restores an interrupted prepared transaction's qualified last-good bytes