chore(release): bump to v0.14.0

Headline: ODoH (RFC 9230) — client + self-hosted relay. Set mode = "odoh" in [upstream] to seal queries before they leave the machine; run `numa relay` to add to the public ODoH ecosystem.
Merge pull request #121 from razvandimescu/feat/odoh
2026-04-20 17:07:31 +03:00 · 2026-04-20 16:26:54 +03:00 · 2026-04-20 16:03:34 +03:00 · 2026-04-20 15:44:29 +03:00 · 2026-04-20 15:44:20 +03:00 · 2026-04-20 15:44:09 +03:00
27 changed files with 3521 additions and 240 deletions
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -87,12 +87,26 @@ jobs:
          sleep 2
          curl -sf http://127.0.0.1:5380/health
          dig @127.0.0.1 example.com +short +timeout=5 | grep -q '.'
+          user=$(ps -o user= -p "$(systemctl show -p MainPID --value numa)" | tr -d ' ')
+          echo "numa running as: $user"
+          test "$user" != "root"
          sudo ./target/release/numa install
          sleep 2
          curl -sf http://127.0.0.1:5380/health
          sudo ./target/release/numa uninstall
          sleep 1
          ! curl -sf http://127.0.0.1:5380/health 2>/dev/null
+      - name: diagnostics on failure
+        if: failure()
+        run: |
+          echo "=== systemctl status numa ==="
+          sudo systemctl status numa --no-pager -l || true
+          echo "=== journalctl -u numa (last 200) ==="
+          sudo journalctl -u numa --no-pager -n 200 || true
+          echo "=== ss -tulnp on 53/80/443/853/5380 ==="
+          sudo ss -tulnp 2>/dev/null | grep -E ':(53|80|443|853|5380)\b' || true
+          echo "=== systemctl is-active systemd-resolved ==="
+          systemctl is-active systemd-resolved || true
      - name: cleanup
        if: always()
        run: |
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -8,6 +8,41 @@ version = "2.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "320119579fcad9c21884f5c4861d16174d0e06250625266f50fe6898340abefa"

+[[package]]
+name = "aead"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d122413f284cf2d62fb1b7db97e02edb8cda96d769b16e443a4f6195e35662b0"
+dependencies = [
+ "crypto-common",
+ "generic-array",
+]
+
+[[package]]
+name = "aes"
+version = "0.8.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b169f7a6d4742236a0a00c541b845991d0ac43e546831af1249753ab4c3aa3a0"
+dependencies = [
+ "cfg-if",
+ "cipher",
+ "cpufeatures",
+]
+
+[[package]]
+name = "aes-gcm"
+version = "0.10.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "831010a0f742e1209b3bcea8fab6a8e149051ba6099432c8cb2cc117dec3ead1"
+dependencies = [
+ "aead",
+ "aes",
+ "cipher",
+ "ctr",
+ "ghash",
+ "subtle",
+]
+
 [[package]]
 name = "aho-corasick"
 version = "1.1.4"
@@ -109,7 +144,7 @@ dependencies = [
 "nom",
 "num-traits",
 "rusticata-macros",
- "thiserror",
+ "thiserror 2.0.18",
 "time",
 ]

@@ -257,6 +292,15 @@ version = "2.11.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "843867be96c8daad0d758b57df9392b6d8d271134fce549de6ce169ff98a92af"

+[[package]]
+name = "block-buffer"
+version = "0.10.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3078c7629b62d3f0439517fa394996acacc5cbc91c5a20d8c658e77abd503a71"
+dependencies = [
+ "generic-array",
+]
+
 [[package]]
 name = "bumpalo"
 version = "3.20.2"
@@ -299,6 +343,30 @@ version = "0.2.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"

+[[package]]
+name = "chacha20"
+version = "0.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3613f74bd2eac03dad61bd53dbe620703d4371614fe0bc3b9f04dd36fe4e818"
+dependencies = [
+ "cfg-if",
+ "cipher",
+ "cpufeatures",
+]
+
+[[package]]
+name = "chacha20poly1305"
+version = "0.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "10cd79432192d1c0f4e1a0fef9527696cc039165d729fb41b3f4f4f354c2dc35"
+dependencies = [
+ "aead",
+ "chacha20",
+ "cipher",
+ "poly1305",
+ "zeroize",
+]
+
 [[package]]
 name = "ciborium"
 version = "0.2.2"
@@ -326,6 +394,17 @@ dependencies = [
 "half",
 ]

+[[package]]
+name = "cipher"
+version = "0.4.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "773f3b9af64447d2ce9850330c473515014aa235e6a783b02db81ff39e4a3dad"
+dependencies = [
+ "crypto-common",
+ "inout",
+ "zeroize",
+]
+
 [[package]]
 name = "clap"
 version = "4.6.0"
@@ -383,6 +462,15 @@ version = "0.4.31"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "75984efb6ed102a0d42db99afb6c1948f0380d1d91808d5529916e6c08b49d8d"

+[[package]]
+name = "cpufeatures"
+version = "0.2.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "59ed5838eebb26a2bb2e58f6d5b5316989ae9d08bab10e0e6d103e656d1b0280"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "crc32fast"
 version = "1.5.0"
@@ -473,6 +561,51 @@ version = "0.2.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "460fbee9c2c2f33933d720630a6a0bac33ba7053db5344fac858d4b8952d77d5"

+[[package]]
+name = "crypto-common"
+version = "0.1.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "78c8292055d1c1df0cce5d180393dc8cce0abec0a7102adb6c7b1eef6016d60a"
+dependencies = [
+ "generic-array",
+ "rand_core 0.6.4",
+ "typenum",
+]
+
+[[package]]
+name = "ctr"
+version = "0.9.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0369ee1ad671834580515889b80f2ea915f23b8be8d0daa4bbaf2ac5c7590835"
+dependencies = [
+ "cipher",
+]
+
+[[package]]
+name = "curve25519-dalek"
+version = "4.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "97fb8b7c4503de7d6ae7b42ab72a5a59857b4c937ec27a3d4539dba95b5ab2be"
+dependencies = [
+ "cfg-if",
+ "cpufeatures",
+ "curve25519-dalek-derive",
+ "fiat-crypto",
+ "rustc_version",
+ "subtle",
+]
+
+[[package]]
+name = "curve25519-dalek-derive"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f46882e17999c6cc590af592290432be3bce0428cb0d5f8b6715e4dc7b383eb3"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]
+
 [[package]]
 name = "data-encoding"
 version = "2.10.0"
@@ -502,6 +635,17 @@ dependencies = [
 "powerfmt",
 ]

+[[package]]
+name = "digest"
+version = "0.10.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292"
+dependencies = [
+ "block-buffer",
+ "crypto-common",
+ "subtle",
+]
+
 [[package]]
 name = "displaydoc"
 version = "0.2.5"
@@ -576,6 +720,12 @@ dependencies = [
 "windows-sys 0.61.2",
 ]

+[[package]]
+name = "fiat-crypto"
+version = "0.2.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "28dea519a9695b9977216879a3ebfddf92f1c08c05d984f8996aecd6ecdc811d"
+
 [[package]]
 name = "find-msvc-tools"
 version = "0.1.9"
@@ -707,6 +857,16 @@ dependencies = [
 "slab",
 ]

+[[package]]
+name = "generic-array"
+version = "0.14.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "85649ca51fd72272d7821adaf274ad91c288277713d9c18820d8499a7ff69e9a"
+dependencies = [
+ "typenum",
+ "version_check",
+]
+
 [[package]]
 name = "getrandom"
 version = "0.2.17"
@@ -747,6 +907,16 @@ dependencies = [
 "wasip3",
 ]

+[[package]]
+name = "ghash"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f0d8a4362ccb29cb0b265253fb0a2728f592895ee6854fd9bc13f2ffda266ff1"
+dependencies = [
+ "opaque-debug",
+ "polyval",
+]
+
 [[package]]
 name = "h2"
 version = "0.4.13"
@@ -820,7 +990,7 @@ dependencies = [
 "rand",
 "ring",
 "rustls",
- "thiserror",
+ "thiserror 2.0.18",
 "tinyvec",
 "tokio",
 "tokio-rustls",
@@ -846,13 +1016,51 @@ dependencies = [
 "resolv-conf",
 "rustls",
 "smallvec",
- "thiserror",
+ "thiserror 2.0.18",
 "tokio",
 "tokio-rustls",
 "tracing",
 "webpki-roots 0.26.11",
 ]

+[[package]]
+name = "hkdf"
+version = "0.12.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7b5f8eb2ad728638ea2c7d47a21db23b7b58a72ed6a38256b8a1849f15fbbdf7"
+dependencies = [
+ "hmac",
+]
+
+[[package]]
+name = "hmac"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c49c37c09c17a53d937dfbb742eb3a961d65a994e6bcdcf37e7399d0cc8ab5e"
+dependencies = [
+ "digest",
+]
+
+[[package]]
+name = "hpke"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f65d16b699dd1a1fa2d851c970b0c971b388eeeb40f744252b8de48860980c8f"
+dependencies = [
+ "aead",
+ "aes-gcm",
+ "chacha20poly1305",
+ "digest",
+ "generic-array",
+ "hkdf",
+ "hmac",
+ "rand_core 0.9.5",
+ "sha2",
+ "subtle",
+ "x25519-dalek",
+ "zeroize",
+]
+
 [[package]]
 name = "http"
 version = "1.4.0"
@@ -1081,6 +1289,15 @@ dependencies = [
 "serde_core",
 ]

+[[package]]
+name = "inout"
+version = "0.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "879f10e63c20629ecabbb64a8010319738c66a5cd0c29b02d63d272b03751d01"
+dependencies = [
+ "generic-array",
+]
+
 [[package]]
 name = "ipconfig"
 version = "0.3.4"
@@ -1330,7 +1547,7 @@ dependencies = [

 [[package]]
 name = "numa"
-version = "0.13.1"
+version = "0.14.0"
 dependencies = [
 "arc-swap",
 "axum",
@@ -1344,7 +1561,9 @@ dependencies = [
 "hyper",
 "hyper-util",
 "log",
+ "odoh-rs",
 "qrcode",
+ "rand_core 0.9.5",
 "rcgen",
 "reqwest",
 "ring",
@@ -1363,6 +1582,19 @@ dependencies = [
 "x509-parser",
 ]

+[[package]]
+name = "odoh-rs"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cbb89720b7dfdddc89bc7560669d41a0bb68eb64784a4aebd293308a489f3837"
+dependencies = [
+ "aes-gcm",
+ "bytes",
+ "hkdf",
+ "hpke",
+ "thiserror 1.0.69",
+]
+
 [[package]]
 name = "oid-registry"
 version = "0.8.1"
@@ -1394,6 +1626,12 @@ version = "11.1.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d6790f58c7ff633d8771f42965289203411a5e5c68388703c06e14f24770b41e"

+[[package]]
+name = "opaque-debug"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c08d65885ee38876c4f86fa503fb49d7b507c2b62552df7c70b2fce627e06381"
+
 [[package]]
 name = "page_size"
 version = "0.6.0"
@@ -1483,6 +1721,29 @@ dependencies = [
 "plotters-backend",
 ]

+[[package]]
+name = "poly1305"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8159bd90725d2df49889a078b54f4f79e87f1f8a8444194cdca81d38f5393abf"
+dependencies = [
+ "cpufeatures",
+ "opaque-debug",
+ "universal-hash",
+]
+
+[[package]]
+name = "polyval"
+version = "0.6.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9d1fe60d06143b2430aa532c94cfe9e29783047f06c0d7fd359a9a51b729fa25"
+dependencies = [
+ "cfg-if",
+ "cpufeatures",
+ "opaque-debug",
+ "universal-hash",
+]
+
 [[package]]
 name = "portable-atomic"
 version = "1.13.1"
@@ -1561,7 +1822,7 @@ dependencies = [
 "rustc-hash",
 "rustls",
 "socket2",
- "thiserror",
+ "thiserror 2.0.18",
 "tokio",
 "tracing",
 "web-time",
@@ -1582,7 +1843,7 @@ dependencies = [
 "rustls",
 "rustls-pki-types",
 "slab",
- "thiserror",
+ "thiserror 2.0.18",
 "tinyvec",
 "tracing",
 "web-time",
@@ -1630,7 +1891,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6db2770f06117d490610c7488547d543617b21bfa07796d7a12f6f1bd53850d1"
 dependencies = [
 "rand_chacha",
- "rand_core",
+ "rand_core 0.9.5",
 ]

 [[package]]
@@ -1640,7 +1901,16 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d3022b5f1df60f26e1ffddd6c66e8aa15de382ae63b3a0c1bfc0e4d3e3f325cb"
 dependencies = [
 "ppv-lite86",
- "rand_core",
+ "rand_core 0.9.5",
+]
+
+[[package]]
+name = "rand_core"
+version = "0.6.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ec0be4795e2f6a28069bec0b5ff3e2ac9bafc99e6a9a7dc3547996c5c816922c"
+dependencies = [
+ "getrandom 0.2.17",
 ]

 [[package]]
@@ -1789,6 +2059,15 @@ version = "2.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "357703d41365b4b27c590e3ed91eabb1b663f07c4c084095e60cbed4362dff0d"

+[[package]]
+name = "rustc_version"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cfcb3a22ef46e85b45de6ee7e79d063319ebb6594faafcf1c225ea92ab6e9b92"
+dependencies = [
+ "semver",
+]
+
 [[package]]
 name = "rusticata-macros"
 version = "4.1.0"
@@ -1953,6 +2232,17 @@ dependencies = [
 "serde",
 ]

+[[package]]
+name = "sha2"
+version = "0.10.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a7507d819769d01a365ab707794a4084392c824f54a7a6a7862f8c3d0892b283"
+dependencies = [
+ "cfg-if",
+ "cpufeatures",
+ "digest",
+]
+
 [[package]]
 name = "shlex"
 version = "1.3.0"
@@ -2046,13 +2336,33 @@ version = "0.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7b2093cf4c8eb1e67749a6762251bc9cd836b6fc171623bd0a9d324d37af2417"

+[[package]]
+name = "thiserror"
+version = "1.0.69"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b6aaf5339b578ea85b50e080feb250a3e8ae8cfcdff9a461c9ec2904bc923f52"
+dependencies = [
+ "thiserror-impl 1.0.69",
+]
+
 [[package]]
 name = "thiserror"
 version = "2.0.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4288b5bcbc7920c07a1149a35cf9590a2aa808e0bc1eafaade0b80947865fbc4"
 dependencies = [
- "thiserror-impl",
+ "thiserror-impl 2.0.18",
+]
+
+[[package]]
+name = "thiserror-impl"
+version = "1.0.69"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4fee6c4efc90059e10f81e6d42c60a18f76588c3d74cb83a0b242a2b6c7504c1"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
 ]

 [[package]]
@@ -2298,6 +2608,12 @@ version = "0.2.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"

+[[package]]
+name = "typenum"
+version = "1.20.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "40ce102ab67701b8526c123c1bab5cbe42d7040ccfd0f64af1a385808d2f43de"
+
 [[package]]
 name = "unicode-ident"
 version = "1.0.24"
@@ -2310,6 +2626,16 @@ version = "0.2.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ebc1c04c71510c7f702b52b7c350734c9ff1295c464a03335b00bb84fc54f853"

+[[package]]
+name = "universal-hash"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fc1de2c688dc15305988b563c3854064043356019f97a4b46276fe734c4f07ea"
+dependencies = [
+ "crypto-common",
+ "subtle",
+]
+
 [[package]]
 name = "untrusted"
 version = "0.9.0"
@@ -2351,6 +2677,12 @@ dependencies = [
 "wasm-bindgen",
 ]

+[[package]]
+name = "version_check"
+version = "0.9.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0b928f33d975fc6ad9f86c8f283853ad26bdd5b10b7f1542aa2fa15e2289105a"
+
 [[package]]
 name = "walkdir"
 version = "2.5.0"
@@ -2860,6 +3192,16 @@ version = "0.6.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9edde0db4769d2dc68579893f2306b26c6ecfbe0ef499b013d731b7b9247e0b9"

+[[package]]
+name = "x25519-dalek"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c7e468321c81fb07fa7f4c636c3972b9100f0346e5b6a9f2bd0603a52f7ed277"
+dependencies = [
+ "curve25519-dalek",
+ "rand_core 0.6.4",
+]
+
 [[package]]
 name = "x509-parser"
 version = "0.18.1"
@@ -2874,7 +3216,7 @@ dependencies = [
 "oid-registry",
 "ring",
 "rusticata-macros",
- "thiserror",
+ "thiserror 2.0.18",
 "time",
 ]

@@ -2956,6 +3298,20 @@ name = "zeroize"
 version = "1.8.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b97154e67e32c85465826e8bcc1c59429aaaf107c1e4a9e53c8d8ccd5eff88d0"
+dependencies = [
+ "zeroize_derive",
+]
+
+[[package]]
+name = "zeroize_derive"
+version = "1.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "85a5b4158499876c763cb03bc4e49185d3cccbabb15b33c627f7884f43db852e"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn",
+]

 [[package]]
 name = "zerotrie"
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "numa"
-version = "0.13.1"
+version = "0.14.0"
 authors = ["razvandimescu <razvan@dimescu.com>"]
 edition = "2021"
 description = "Portable DNS resolver in Rust — .numa local domains, ad blocking, developer overrides, DNS-over-HTTPS"
@@ -29,6 +29,10 @@ rustls = "0.23"
 tokio-rustls = "0.26"
 arc-swap = "1"
 ring = "0.17"
+odoh-rs = "1"
+# rand_core 0.9 matches the version odoh-rs (via hpke 0.13) depends on, so we
+# share one RngCore trait and OsRng impl across the dep tree.
+rand_core = { version = "0.9", features = ["os_rng"] }
 rustls-pemfile = "2.2.0"
 qrcode = { version = "0.14", default-features = false, features = ["svg"] }
 webpki-roots = "1"
--- a/numa.service
+++ b/numa.service
@@ -8,6 +8,39 @@ Type=simple
 ExecStart={{exe_path}}
 Restart=always
 RestartSec=2
+
+# Transient system user per start; no PKGBUILD/sysusers setup required.
+# systemd remaps the StateDirectory ownership to the dynamic UID on each
+# launch, including legacy root-owned trees from pre-drop installs.
+DynamicUser=yes
+
+AmbientCapabilities=CAP_NET_BIND_SERVICE
+CapabilityBoundingSet=CAP_NET_BIND_SERVICE
+
+StateDirectory=numa
+StateDirectoryMode=0750
+ConfigurationDirectory=numa
+ConfigurationDirectoryMode=0755
+
+# Sandboxing — conservative set known to work with Rust network daemons.
+# Aggressive hardening (MemoryDenyWriteExecute, SystemCallFilter, seccomp
+# allow-lists) can be layered on once tested in isolation.
+NoNewPrivileges=true
+ProtectSystem=strict
+# DynamicUser= sets ProtectHome=read-only by default — leaves /home
+# readable so systemd can exec binaries installed under it (cargo install,
+# source builds), while blocking writes to user $HOMEs. Don't set =yes:
+# that hides /home entirely and fails with status=203/EXEC.
+PrivateTmp=true
+PrivateDevices=true
+ProtectKernelTunables=true
+ProtectKernelModules=true
+ProtectControlGroups=true
+RestrictRealtime=true
+RestrictSUIDSGID=true
+# AF_NETLINK for interface enumeration on network changes
+RestrictAddressFamilies=AF_INET AF_INET6 AF_UNIX AF_NETLINK
+
 StandardOutput=journal
 StandardError=journal
 SyslogIdentifier=numa
--- a/numa.toml
+++ b/numa.toml
@@ -8,6 +8,16 @@ api_port = 5380
                              # %PROGRAMDATA%\numa on windows. Override for
                              # containerized deploys or tests that can't
                              # write to the system path.
+# filter_aaaa = true          # on IPv4-only networks, answer AAAA queries with
+                              # NODATA (NOERROR + empty answer) so Happy Eyeballs
+                              # clients don't wait on a v6 attempt that can't
+                              # succeed. Also strips `ipv6hint` from HTTPS/SVCB
+                              # records (RFC 9460) so modern browsers (Chrome
+                              # ≥103, Firefox, Safari) don't bypass the AAAA
+                              # filter via SVCB hints. Local zones, overrides,
+                              # and the .numa proxy are NOT filtered — you can
+                              # still configure v6 records for local services.
+                              # Default: false.

 # [upstream]
 # mode = "forward"                              # "forward" (default) — relay to upstream
@@ -66,6 +76,13 @@ api_port = 5380
 # [[forwarding]]                                    # DoH upstream: full https:// URL
 # suffix = "example.corp"
 # upstream = "https://dns.quad9.net/dns-query"
+#
+# [[forwarding]]                                    # array of upstreams → SRTT-aware failover
+# suffix = ["google.com", "goog"]                   # fastest-healthy first, dead one skipped
+# upstream = [
+#     "tls://9.9.9.9#dns.quad9.net",
+#     "tls://149.112.112.112#dns.quad9.net",
+# ]

 # [blocking]
 # enabled = true               # set to false to disable ad blocking
--- a/packaging/relay/Caddyfile
+++ b/packaging/relay/Caddyfile
@@ -0,0 +1,15 @@
+odoh-relay.example.com {
+    handle /relay {
+        reverse_proxy numa-relay:8443
+    }
+    handle /health {
+        reverse_proxy numa-relay:8443
+    }
+    respond 404
+
+    # Per-request access logs defeat the point of an oblivious relay.
+    # Aggregate counters are exposed at /health on the relay itself.
+    log {
+        output discard
+    }
+}
--- a/packaging/relay/README.md
+++ b/packaging/relay/README.md
@@ -0,0 +1,48 @@
+# Numa ODoH Relay — Docker deploy
+
+Two-container deploy: Caddy terminates TLS (auto-provisioning a Let's Encrypt
+cert via ACME) and reverse-proxies to a Numa relay running on an internal
+Docker network. The relay never reads sealed payloads; Caddy never logs them.
+
+## Prerequisites
+
+- A host with public 80/443 reachable from the internet.
+- A DNS record (`A` or `AAAA`) pointing your chosen hostname at the host.
+- Docker + Docker Compose v2.
+
+## Configure
+
+Edit `Caddyfile` and replace `odoh-relay.example.com` with your hostname.
+That hostname is what ACME validates against and what ODoH clients will
+configure as their relay URL: `https://<hostname>/relay`.
+
+## Deploy
+
+```sh
+docker compose up -d
+docker compose logs -f caddy   # watch ACME provisioning
+```
+
+First boot takes a few seconds while Caddy obtains the cert. Subsequent
+restarts reuse the cached cert from the `caddy_data` volume.
+
+## Verify
+
+```sh
+curl https://<hostname>/health
+# ok
+# total 0
+# forwarded_ok 0
+# forwarded_err 0
+# rejected_bad_request 0
+```
+
+Then point any ODoH client at `https://<hostname>/relay` and watch the
+counters tick.
+
+## Listing on the public ecosystem
+
+DNSCrypt's [v3/odoh-relays.md](https://github.com/DNSCrypt/dnscrypt-resolvers/blob/master/v3/odoh-relays.md)
+is the canonical list. The pruned 2025-09-16 commit shows one public ODoH
+relay survived the cull — running this compose file doubles global supply.
+Open a PR there once your relay has been up for ~24 hours.
--- a/packaging/relay/docker-compose.yml
+++ b/packaging/relay/docker-compose.yml
@@ -0,0 +1,26 @@
+services:
+  numa-relay:
+    image: ghcr.io/razvandimescu/numa:latest
+    command: ["relay", "8443", "0.0.0.0"]
+    restart: unless-stopped
+    networks: [internal]
+
+  caddy:
+    image: caddy:2
+    ports:
+      - "80:80"
+      - "443:443"
+    volumes:
+      - ./Caddyfile:/etc/caddy/Caddyfile:ro
+      - caddy_data:/data
+      - caddy_config:/config
+    restart: unless-stopped
+    depends_on: [numa-relay]
+    networks: [internal]
+
+networks:
+  internal:
+
+volumes:
+  caddy_data:
+  caddy_config:
--- a/site/dashboard.html
+++ b/site/dashboard.html
@@ -228,6 +228,7 @@ body {
 .path-bar-fill.tcp { background: var(--violet); }
 .path-bar-fill.dot { background: var(--emerald); }
 .path-bar-fill.doh { background: var(--teal); }
+.path-bar-fill.odoh { background: var(--violet-dim); }
 .path-pct {
  font-family: var(--font-mono);
  font-size: 0.75rem;
@@ -637,16 +638,26 @@ body {
    </div>
  </div>

-  <!-- Transport breakdown -->
+  <!-- Inbound wire (apps → numa) -->
  <div class="panel">
    <div class="panel-header">
-      <span class="panel-title">Transport</span>
+      <span class="panel-title">Inbound Wire <span style="color: var(--text-dim); font-weight: normal;">apps → numa</span></span>
      <span class="panel-title" id="transportEncrypted" style="color: var(--text-dim)"></span>
    </div>
    <div class="panel-body" id="transportBars">
    </div>
  </div>

+  <!-- Outbound wire (numa → internet) -->
+  <div class="panel">
+    <div class="panel-header">
+      <span class="panel-title">Outbound Wire <span style="color: var(--text-dim); font-weight: normal;">numa → internet</span></span>
+      <span class="panel-title" id="upstreamWireEncrypted" style="color: var(--text-dim)"></span>
+    </div>
+    <div class="panel-body" id="upstreamWireBars">
+    </div>
+  </div>
+
  <!-- Main grid: query log + sidebar -->
  <div class="main-grid">
    <!-- Query log -->
@@ -960,9 +971,11 @@ function renderBarChart(containerId, defs, data, total) {
    }).join('');
 }

-function encryptionPct(transport) {
-  const total = (transport.udp + transport.tcp + transport.dot + transport.doh) || 1;
-  return (((transport.dot + transport.doh) / total) * 100).toFixed(0);
+function encryptionPct(data, encryptedKeys, allKeys) {
+  const total = allKeys.reduce((s, k) => s + (data[k] || 0), 0);
+  if (total === 0) return 0;
+  const encrypted = encryptedKeys.reduce((s, k) => s + (data[k] || 0), 0);
+  return Math.round((encrypted / total) * 100);
 }

 const PATH_DEFS = [
@@ -990,9 +1003,25 @@ const TRANSPORT_DEFS = [
 function renderTransport(transport) {
  const total = (transport.udp + transport.tcp + transport.dot + transport.doh) || 1;
  renderBarChart('transportBars', TRANSPORT_DEFS, transport, total);
-  const encPct = encryptionPct(transport);
+  const encPct = encryptionPct(transport, ['dot', 'doh'], ['udp', 'tcp', 'dot', 'doh']);
  const el = document.getElementById('transportEncrypted');
-  el.textContent = `${encPct}% encrypted`;
+  el.textContent = `${encPct}% encrypted inbound`;
+  el.style.color = encPct >= 80 ? 'var(--emerald)' : encPct >= 50 ? 'var(--amber)' : 'var(--rose)';
+}
+
+const UPSTREAM_WIRE_DEFS = [
+  { key: 'udp',  label: 'UDP',  cls: 'udp' },
+  { key: 'doh',  label: 'DoH',  cls: 'doh' },
+  { key: 'dot',  label: 'DoT',  cls: 'dot' },
+  { key: 'odoh', label: 'ODoH', cls: 'odoh' },
+];
+
+function renderUpstreamWire(ut) {
+  const total = (ut.udp + ut.doh + ut.dot + ut.odoh) || 0;
+  renderBarChart('upstreamWireBars', UPSTREAM_WIRE_DEFS, ut, total || 1);
+  const encPct = encryptionPct(ut, ['doh', 'dot', 'odoh'], ['udp', 'doh', 'dot', 'odoh']);
+  const el = document.getElementById('upstreamWireEncrypted');
+  el.textContent = total > 0 ? `${encPct}% encrypted outbound` : '';
  el.style.color = encPct >= 80 ? 'var(--emerald)' : encPct >= 50 ? 'var(--amber)' : 'var(--rose)';
 }

@@ -1234,6 +1263,7 @@ async function refresh() {
    // Panels
    renderPaths(q);
    renderTransport(stats.transport);
+    renderUpstreamWire(stats.upstream_transport || { udp: 0, doh: 0, dot: 0, odoh: 0 });
    renderQueryLog(logs);
    renderOverrides(overrides);
    renderCache(cache);
--- a/src/api.rs
+++ b/src/api.rs
@@ -170,6 +170,7 @@ struct StatsResponse {
    srtt: bool,
    queries: QueriesStats,
    transport: TransportStats,
+    upstream_transport: UpstreamTransportStats,
    cache: CacheStats,
    overrides: OverrideStats,
    blocking: BlockingStatsResponse,
@@ -186,6 +187,14 @@ struct TransportStats {
    doh: u64,
 }

+#[derive(Serialize)]
+struct UpstreamTransportStats {
+    udp: u64,
+    doh: u64,
+    dot: u64,
+    odoh: u64,
+}
+
 #[derive(Serialize)]
 struct MobileStatsResponse {
    enabled: bool,
@@ -566,6 +575,12 @@ async fn stats(State(ctx): State<Arc<ServerCtx>>) -> Json<StatsResponse> {
            dot: snap.transport_dot,
            doh: snap.transport_doh,
        },
+        upstream_transport: UpstreamTransportStats {
+            udp: snap.upstream_transport_udp,
+            doh: snap.upstream_transport_doh,
+            dot: snap.upstream_transport_dot,
+            odoh: snap.upstream_transport_odoh,
+        },
        cache: CacheStats {
            entries: cache_len,
            max_entries: cache_max,
--- a/src/config.rs
+++ b/src/config.rs
@@ -1,7 +1,7 @@
 use std::collections::HashMap;
-use std::net::Ipv4Addr;
-use std::net::Ipv6Addr;
+use std::net::{IpAddr, Ipv4Addr, Ipv6Addr, SocketAddr};
 use std::path::{Path, PathBuf};
+use std::time::Duration;

 use serde::Deserialize;

@@ -41,17 +41,30 @@ pub struct Config {
 pub struct ForwardingRuleConfig {
    #[serde(deserialize_with = "string_or_vec")]
    pub suffix: Vec<String>,
-    pub upstream: String,
+    #[serde(deserialize_with = "string_or_vec")]
+    pub upstream: Vec<String>,
 }

 impl ForwardingRuleConfig {
    fn to_runtime_rules(&self) -> Result<Vec<crate::system_dns::ForwardingRule>> {
-        let upstream = crate::forward::parse_upstream(&self.upstream, 53)
-            .map_err(|e| format!("forwarding rule for upstream '{}': {}", self.upstream, e))?;
+        if self.upstream.is_empty() {
+            return Err(format!(
+                "forwarding rule for suffix {:?}: upstream must not be empty",
+                self.suffix
+            )
+            .into());
+        }
+        let mut primary = Vec::with_capacity(self.upstream.len());
+        for s in &self.upstream {
+            let u = crate::forward::parse_upstream(s, 53)
+                .map_err(|e| format!("forwarding rule for upstream '{}': {}", s, e))?;
+            primary.push(u);
+        }
+        let pool = crate::forward::UpstreamPool::new(primary, vec![]);
        Ok(self
            .suffix
            .iter()
-            .map(|s| crate::system_dns::ForwardingRule::new(s.clone(), upstream.clone()))
+            .map(|s| crate::system_dns::ForwardingRule::new(s.clone(), pool.clone()))
            .collect())
    }
 }
@@ -80,6 +93,12 @@ pub struct ServerConfig {
    /// Defaults to `crate::data_dir()` (platform-specific system path) if unset.
    #[serde(default)]
    pub data_dir: Option<PathBuf>,
+    /// Synthesize NODATA (NOERROR + empty answer) for AAAA queries, and
+    /// strip `ipv6hint` from HTTPS/SVCB responses (RFC 9460). For IPv4-only
+    /// networks where Happy Eyeballs fallback adds latency. Local zones,
+    /// overrides, and the service proxy are not affected. Default false.
+    #[serde(default)]
+    pub filter_aaaa: bool,
 }

 impl Default for ServerConfig {
@@ -89,6 +108,7 @@ impl Default for ServerConfig {
            api_port: default_api_port(),
            api_bind_addr: default_api_bind_addr(),
            data_dir: None,
+            filter_aaaa: false,
        }
    }
 }
@@ -114,6 +134,7 @@ pub enum UpstreamMode {
    #[default]
    Forward,
    Recursive,
+    Odoh,
 }

 impl UpstreamMode {
@@ -122,6 +143,20 @@ impl UpstreamMode {
            UpstreamMode::Auto => "auto",
            UpstreamMode::Forward => "forward",
            UpstreamMode::Recursive => "recursive",
+            UpstreamMode::Odoh => "odoh",
+        }
+    }
+
+    /// Hedging duplicates the in-flight query against the same upstream to
+    /// rescue tail latency. Beneficial for UDP/DoH/DoT (cheap retransmit /
+    /// h2 stream multiplexing). For ODoH it doubles the relay's HPKE
+    /// seal/unseal load and the sealed-byte footprint a passive observer
+    /// can correlate, with no latency win — the relay hop dominates either
+    /// way. Force-zero in oblivious mode regardless of `hedge_ms`.
+    pub fn hedge_delay(self, hedge_ms: u64) -> Duration {
+        match self {
+            UpstreamMode::Odoh => Duration::ZERO,
+            _ => Duration::from_millis(hedge_ms),
        }
    }
 }
@@ -134,7 +169,7 @@ pub struct UpstreamConfig {
    pub address: Vec<String>,
    #[serde(default = "default_upstream_port")]
    pub port: u16,
-    #[serde(default)]
+    #[serde(default, deserialize_with = "string_or_vec")]
    pub fallback: Vec<String>,
    #[serde(default = "default_timeout_ms")]
    pub timeout_ms: u64,
@@ -146,6 +181,30 @@ pub struct UpstreamConfig {
    pub prime_tlds: Vec<String>,
    #[serde(default = "default_srtt")]
    pub srtt: bool,
+
+    /// Only used when `mode = "odoh"`. Full https:// URL of the relay
+    /// endpoint (including path, e.g. `https://odoh-relay.numa.rs/relay`).
+    #[serde(default)]
+    pub relay: Option<String>,
+    /// Only used when `mode = "odoh"`. Full https:// URL of the target
+    /// resolver (`https://odoh.cloudflare-dns.com/dns-query`).
+    #[serde(default)]
+    pub target: Option<String>,
+    /// Only used when `mode = "odoh"`. When true (the default), relay failure
+    /// returns SERVFAIL instead of downgrading to the `fallback` upstream —
+    /// a user who configured ODoH rarely wants a silent non-oblivious path.
+    #[serde(default)]
+    pub strict: Option<bool>,
+
+    /// Bootstrap IP for the relay host, used when numa is its own system
+    /// resolver (otherwise the ODoH HTTPS client loops resolving through
+    /// itself). TLS still validates the cert against `relay`'s hostname.
+    #[serde(default)]
+    pub relay_ip: Option<IpAddr>,
+
+    /// Same as `relay_ip` but for the target host.
+    #[serde(default)]
+    pub target_ip: Option<IpAddr>,
 }

 impl Default for UpstreamConfig {
@@ -160,10 +219,90 @@ impl Default for UpstreamConfig {
            root_hints: default_root_hints(),
            prime_tlds: default_prime_tlds(),
            srtt: default_srtt(),
+            relay: None,
+            target: None,
+            strict: None,
+            relay_ip: None,
+            target_ip: None,
        }
    }
 }

+/// Parsed ODoH config fields. `mode = "odoh"` requires both URLs to be
+/// present, to parse as `https://`, and to resolve to distinct hosts.
+#[derive(Debug)]
+pub struct OdohUpstream {
+    pub relay_url: String,
+    pub relay_host: String,
+    pub target_host: String,
+    pub target_path: String,
+    pub strict: bool,
+    pub relay_bootstrap: Option<SocketAddr>,
+    pub target_bootstrap: Option<SocketAddr>,
+}
+
+impl UpstreamConfig {
+    /// Validate and extract ODoH-specific fields. Called during `load_config`
+    /// so misconfigured ODoH fails fast at startup, the same care we take
+    /// with the DNSSEC strict boot check.
+    pub fn odoh_upstream(&self) -> Result<OdohUpstream> {
+        let relay = self
+            .relay
+            .as_deref()
+            .ok_or("mode = \"odoh\" requires upstream.relay")?;
+        let target = self
+            .target
+            .as_deref()
+            .ok_or("mode = \"odoh\" requires upstream.target")?;
+
+        let relay_url = reqwest::Url::parse(relay)
+            .map_err(|e| format!("upstream.relay invalid URL '{}': {}", relay, e))?;
+        let target_url = reqwest::Url::parse(target)
+            .map_err(|e| format!("upstream.target invalid URL '{}': {}", target, e))?;
+
+        if relay_url.scheme() != "https" || target_url.scheme() != "https" {
+            return Err("upstream.relay and upstream.target must both use https://".into());
+        }
+        if relay_url.host_str().is_none() || target_url.host_str().is_none() {
+            return Err("upstream.relay and upstream.target must include a host".into());
+        }
+        if relay_url.host_str() == target_url.host_str() {
+            return Err(format!(
+                "upstream.relay and upstream.target resolve to the same host ({}); the privacy property requires distinct operators",
+                relay_url.host_str().unwrap_or("?")
+            )
+            .into());
+        }
+
+        let relay_host = relay_url
+            .host_str()
+            .ok_or("upstream.relay has no host")?
+            .to_string();
+        let target_host = target_url
+            .host_str()
+            .ok_or("upstream.target has no host")?
+            .to_string();
+        let target_path = if target_url.path().is_empty() {
+            "/".to_string()
+        } else {
+            target_url.path().to_string()
+        };
+
+        let relay_port = relay_url.port_or_known_default().unwrap_or(443);
+        let target_port = target_url.port_or_known_default().unwrap_or(443);
+
+        Ok(OdohUpstream {
+            relay_url: relay.to_string(),
+            relay_host,
+            target_host,
+            target_path,
+            strict: self.strict.unwrap_or(true),
+            relay_bootstrap: self.relay_ip.map(|ip| SocketAddr::new(ip, relay_port)),
+            target_bootstrap: self.target_ip.map(|ip| SocketAddr::new(ip, target_port)),
+        })
+    }
+}
+
 fn string_or_vec<'de, D>(deserializer: D) -> std::result::Result<Vec<String>, D::Error>
 where
    D: serde::Deserializer<'de>,
@@ -567,6 +706,17 @@ mod tests {
        assert!(config.lan.enabled);
    }

+    #[test]
+    fn filter_aaaa_defaults_false() {
+        assert!(!ServerConfig::default().filter_aaaa);
+    }
+
+    #[test]
+    fn filter_aaaa_parses_from_server_section() {
+        let config: Config = toml::from_str("[server]\nfilter_aaaa = true").unwrap();
+        assert!(config.server.filter_aaaa);
+    }
+
    #[test]
    fn custom_bind_addrs_parse() {
        let toml = r#"
@@ -612,12 +762,22 @@ mod tests {
    }

    #[test]
-    fn fallback_parses() {
+    fn fallback_array_parses() {
        let config: Config =
            toml::from_str("[upstream]\nfallback = [\"8.8.8.8\", \"1.1.1.1\"]").unwrap();
        assert_eq!(config.upstream.fallback, vec!["8.8.8.8", "1.1.1.1"]);
    }

+    #[test]
+    fn fallback_string_parses_as_singleton_vec() {
+        let config: Config =
+            toml::from_str("[upstream]\nfallback = \"tls://1.1.1.1#cloudflare-dns.com\"").unwrap();
+        assert_eq!(
+            config.upstream.fallback,
+            vec!["tls://1.1.1.1#cloudflare-dns.com"]
+        );
+    }
+
    #[test]
    fn empty_address_gives_empty_vec() {
        let config: Config = toml::from_str("").unwrap();
@@ -625,6 +785,169 @@ mod tests {
        assert!(config.upstream.fallback.is_empty());
    }

+    // ── [upstream] mode = "odoh" ────────────────────────────────────────
+
+    #[test]
+    fn odoh_config_parses_and_validates() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert!(matches!(config.upstream.mode, UpstreamMode::Odoh));
+        let odoh = config.upstream.odoh_upstream().unwrap();
+        assert_eq!(odoh.relay_url, "https://odoh-relay.numa.rs/relay");
+        assert_eq!(odoh.target_host, "odoh.cloudflare-dns.com");
+        assert_eq!(odoh.target_path, "/dns-query");
+        assert!(odoh.strict, "strict defaults to true under mode=odoh");
+    }
+
+    #[test]
+    fn odoh_strict_false_is_honoured() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+strict = false
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert!(!config.upstream.odoh_upstream().unwrap().strict);
+    }
+
+    #[test]
+    fn odoh_rejects_same_host_relay_and_target() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh.example.com/relay"
+target = "https://odoh.example.com/dns-query"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let err = config.upstream.odoh_upstream().unwrap_err().to_string();
+        assert!(err.contains("same host"), "got: {err}");
+    }
+
+    #[test]
+    fn odoh_rejects_non_https() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "http://odoh-relay.numa.rs/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let err = config.upstream.odoh_upstream().unwrap_err().to_string();
+        assert!(err.contains("https"), "got: {err}");
+    }
+
+    #[test]
+    fn odoh_missing_relay_rejected() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let err = config.upstream.odoh_upstream().unwrap_err().to_string();
+        assert!(err.contains("upstream.relay"), "got: {err}");
+    }
+
+    #[test]
+    fn odoh_bootstrap_ips_parse_into_socket_addrs() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+relay_ip = "178.104.229.30"
+target_ip = "104.16.249.249"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let odoh = config.upstream.odoh_upstream().unwrap();
+        assert_eq!(odoh.relay_host, "odoh-relay.numa.rs");
+        assert_eq!(
+            odoh.relay_bootstrap.unwrap().to_string(),
+            "178.104.229.30:443"
+        );
+        assert_eq!(
+            odoh.target_bootstrap.unwrap().to_string(),
+            "104.16.249.249:443"
+        );
+    }
+
+    #[test]
+    fn odoh_bootstrap_ips_optional() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let odoh = config.upstream.odoh_upstream().unwrap();
+        assert!(odoh.relay_bootstrap.is_none());
+        assert!(odoh.target_bootstrap.is_none());
+    }
+
+    #[test]
+    fn odoh_bootstrap_ip_rejects_garbage() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+relay_ip = "not-an-ip"
+"#;
+        let err = toml::from_str::<Config>(toml).err().unwrap().to_string();
+        assert!(err.contains("relay_ip"), "got: {err}");
+    }
+
+    #[test]
+    fn odoh_bootstrap_uses_url_port_when_non_default() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs:8443/relay"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+relay_ip = "178.104.229.30"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let odoh = config.upstream.odoh_upstream().unwrap();
+        assert_eq!(
+            odoh.relay_bootstrap.unwrap().to_string(),
+            "178.104.229.30:8443"
+        );
+    }
+
+    #[test]
+    fn hedge_delay_zeroed_for_odoh_mode() {
+        assert_eq!(
+            UpstreamMode::Odoh.hedge_delay(50),
+            Duration::ZERO,
+            "ODoH mode must zero hedge regardless of configured hedge_ms"
+        );
+        assert_eq!(
+            UpstreamMode::Forward.hedge_delay(50),
+            Duration::from_millis(50),
+            "non-ODoH modes honour configured hedge_ms"
+        );
+    }
+
+    #[test]
+    fn odoh_missing_target_rejected() {
+        let toml = r#"
+[upstream]
+mode = "odoh"
+relay = "https://odoh-relay.numa.rs/relay"
+"#;
+        let config: Config = toml::from_str(toml).unwrap();
+        let err = config.upstream.odoh_upstream().unwrap_err().to_string();
+        assert!(err.contains("upstream.target"), "got: {err}");
+    }
+
    // ── issue #82: [[forwarding]] config section ────────────────────────

    #[test]
@@ -643,7 +966,7 @@ mod tests {
        let config: Config = toml::from_str(toml).unwrap();
        assert_eq!(config.forwarding.len(), 1);
        assert_eq!(config.forwarding[0].suffix, &["home.local"]);
-        assert_eq!(config.forwarding[0].upstream, "100.90.1.63:5361");
+        assert_eq!(config.forwarding[0].upstream, vec!["100.90.1.63:5361"]);
    }

    #[test]
@@ -671,7 +994,7 @@ mod tests {
        "#;
        let config: Config = toml::from_str(toml).unwrap();
        assert_eq!(config.forwarding.len(), 2);
-        assert_eq!(config.forwarding[1].upstream, "10.0.0.1");
+        assert_eq!(config.forwarding[1].upstream, vec!["10.0.0.1"]);
    }

    #[test]
@@ -693,28 +1016,29 @@ mod tests {
    fn forwarding_suffix_array_expands_to_multiple_runtime_rules() {
        let rule = ForwardingRuleConfig {
            suffix: vec!["168.192.in-addr.arpa".to_string(), "onsite".to_string()],
-            upstream: "192.168.88.1".to_string(),
+            upstream: vec!["192.168.88.1".to_string()],
        };
        let runtime = rule.to_runtime_rules().unwrap();
        assert_eq!(runtime.len(), 2);
        assert_eq!(runtime[0].suffix, "168.192.in-addr.arpa");
        assert_eq!(runtime[1].suffix, "onsite");
-        assert_eq!(runtime[0].upstream, runtime[1].upstream);
+        assert_eq!(
+            runtime[0].upstream.preferred(),
+            runtime[1].upstream.preferred()
+        );
    }

    #[test]
    fn forwarding_upstream_with_explicit_port() {
        let rule = ForwardingRuleConfig {
            suffix: vec!["home.local".to_string()],
-            upstream: "100.90.1.63:5361".to_string(),
+            upstream: vec!["100.90.1.63:5361".to_string()],
        };
        let runtime = rule.to_runtime_rules().unwrap();
        assert_eq!(runtime.len(), 1);
-        assert!(matches!(
-            runtime[0].upstream,
-            crate::forward::Upstream::Udp(_)
-        ));
-        assert_eq!(runtime[0].upstream.to_string(), "100.90.1.63:5361");
+        let preferred = runtime[0].upstream.preferred().unwrap();
+        assert!(matches!(preferred, crate::forward::Upstream::Udp(_)));
+        assert_eq!(preferred.to_string(), "100.90.1.63:5361");
        assert_eq!(runtime[0].suffix, "home.local");
    }

@@ -722,17 +1046,20 @@ mod tests {
    fn forwarding_upstream_defaults_to_port_53() {
        let rule = ForwardingRuleConfig {
            suffix: vec!["home.local".to_string()],
-            upstream: "100.90.1.63".to_string(),
+            upstream: vec!["100.90.1.63".to_string()],
        };
        let runtime = rule.to_runtime_rules().unwrap();
-        assert_eq!(runtime[0].upstream.to_string(), "100.90.1.63:53");
+        assert_eq!(
+            runtime[0].upstream.preferred().unwrap().to_string(),
+            "100.90.1.63:53"
+        );
    }

    #[test]
    fn forwarding_invalid_upstream_returns_error() {
        let rule = ForwardingRuleConfig {
            suffix: vec!["home.local".to_string()],
-            upstream: "not-a-valid-host".to_string(),
+            upstream: vec!["not-a-valid-host".to_string()],
        };
        assert!(rule.to_runtime_rules().is_err());
    }
@@ -741,14 +1068,14 @@ mod tests {
    fn forwarding_upstream_accepts_dot_scheme() {
        let rule = ForwardingRuleConfig {
            suffix: vec!["google.com".to_string()],
-            upstream: "tls://9.9.9.9#dns.quad9.net".to_string(),
+            upstream: vec!["tls://9.9.9.9#dns.quad9.net".to_string()],
        };
        let runtime = rule
            .to_runtime_rules()
            .expect("tls:// upstream should parse");
        assert_eq!(runtime.len(), 1);
        assert_eq!(
-            runtime[0].upstream.to_string(),
+            runtime[0].upstream.preferred().unwrap().to_string(),
            "tls://9.9.9.9:853#dns.quad9.net"
        );
    }
@@ -757,14 +1084,14 @@ mod tests {
    fn forwarding_upstream_accepts_doh_scheme() {
        let rule = ForwardingRuleConfig {
            suffix: vec!["goog".to_string()],
-            upstream: "https://dns.quad9.net/dns-query".to_string(),
+            upstream: vec!["https://dns.quad9.net/dns-query".to_string()],
        };
        let runtime = rule
            .to_runtime_rules()
            .expect("https:// upstream should parse");
        assert_eq!(runtime.len(), 1);
        assert_eq!(
-            runtime[0].upstream.to_string(),
+            runtime[0].upstream.preferred().unwrap().to_string(),
            "https://dns.quad9.net/dns-query"
        );
    }
@@ -773,44 +1100,90 @@ mod tests {
    fn forwarding_config_rules_take_precedence_over_discovered() {
        let config_rules = vec![ForwardingRuleConfig {
            suffix: vec!["home.local".to_string()],
-            upstream: "10.0.0.1:53".to_string(),
+            upstream: vec!["10.0.0.1:53".to_string()],
        }];
        let discovered = vec![crate::system_dns::ForwardingRule::new(
            "home.local".to_string(),
-            crate::forward::Upstream::Udp("192.168.1.1:53".parse().unwrap()),
+            crate::forward::UpstreamPool::new(
+                vec![crate::forward::Upstream::Udp(
+                    "192.168.1.1:53".parse().unwrap(),
+                )],
+                vec![],
+            ),
        )];
        let merged = merge_forwarding_rules(&config_rules, discovered).unwrap();
        let picked = crate::system_dns::match_forwarding_rule("host.home.local", &merged)
            .expect("rule should match");
-        assert_eq!(picked.to_string(), "10.0.0.1:53");
+        assert_eq!(picked.preferred().unwrap().to_string(), "10.0.0.1:53");
    }

    #[test]
    fn forwarding_merge_preserves_non_overlapping_discovered() {
        let config_rules = vec![ForwardingRuleConfig {
            suffix: vec!["home.local".to_string()],
-            upstream: "10.0.0.1:53".to_string(),
+            upstream: vec!["10.0.0.1:53".to_string()],
        }];
        let discovered = vec![crate::system_dns::ForwardingRule::new(
            "corp.example".to_string(),
-            crate::forward::Upstream::Udp("192.168.1.1:53".parse().unwrap()),
+            crate::forward::UpstreamPool::new(
+                vec![crate::forward::Upstream::Udp(
+                    "192.168.1.1:53".parse().unwrap(),
+                )],
+                vec![],
+            ),
        )];
        let merged = merge_forwarding_rules(&config_rules, discovered).unwrap();
        assert_eq!(merged.len(), 2);
        let picked = crate::system_dns::match_forwarding_rule("host.corp.example", &merged)
            .expect("discovered rule should still match");
-        assert_eq!(picked.to_string(), "192.168.1.1:53");
+        assert_eq!(picked.preferred().unwrap().to_string(), "192.168.1.1:53");
    }

    #[test]
    fn forwarding_merge_suffix_array_expands_to_multiple_rules() {
        let config_rules = vec![ForwardingRuleConfig {
            suffix: vec!["a.local".to_string(), "b.local".to_string()],
-            upstream: "10.0.0.1:53".to_string(),
+            upstream: vec!["10.0.0.1:53".to_string()],
        }];
        let merged = merge_forwarding_rules(&config_rules, vec![]).unwrap();
        assert_eq!(merged.len(), 2);
    }
+
+    #[test]
+    fn forwarding_parses_upstream_array() {
+        let toml = r#"
+            [[forwarding]]
+            suffix = "google.com"
+            upstream = ["tls://9.9.9.9#dns.quad9.net", "tls://149.112.112.112#dns.quad9.net"]
+        "#;
+        let config: Config = toml::from_str(toml).unwrap();
+        assert_eq!(config.forwarding.len(), 1);
+        assert_eq!(config.forwarding[0].upstream.len(), 2);
+    }
+
+    #[test]
+    fn forwarding_upstream_array_builds_pool_with_multiple_primaries() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["google.com".to_string()],
+            upstream: vec![
+                "tls://9.9.9.9#dns.quad9.net".to_string(),
+                "tls://149.112.112.112#dns.quad9.net".to_string(),
+            ],
+        };
+        let runtime = rule.to_runtime_rules().unwrap();
+        assert_eq!(runtime.len(), 1);
+        let label = runtime[0].upstream.label();
+        assert!(label.contains("+1 more"), "label was: {}", label);
+    }
+
+    #[test]
+    fn forwarding_empty_upstream_array_errors() {
+        let rule = ForwardingRuleConfig {
+            suffix: vec!["home.local".to_string()],
+            upstream: vec![],
+        };
+        assert!(rule.to_runtime_rules().is_err());
+    }
 }

 pub struct ConfigLoad {
--- a/src/ctx.rs
+++ b/src/ctx.rs
@@ -16,7 +16,9 @@ use crate::blocklist::BlocklistStore;
 use crate::buffer::BytePacketBuffer;
 use crate::cache::{DnsCache, DnssecStatus};
 use crate::config::{UpstreamMode, ZoneMap};
-use crate::forward::{forward_query_raw, forward_with_failover_raw, Upstream, UpstreamPool};
+#[cfg(test)]
+use crate::forward::Upstream;
+use crate::forward::{forward_with_failover_raw, UpstreamPool};
 use crate::header::ResultCode;
 use crate::health::HealthMeta;
 use crate::lan::PeerStore;
@@ -75,6 +77,10 @@ pub struct ServerCtx {
    pub ca_pem: Option<String>,
    pub mobile_enabled: bool,
    pub mobile_port: u16,
+    /// When true, AAAA queries short-circuit with NODATA (NOERROR + empty
+    /// answer) instead of hitting cache/forwarding/upstream. Local data
+    /// (overrides, zones, .numa proxy, blocklist sinkhole) is unaffected.
+    pub filter_aaaa: bool,
 }

 /// Transport-agnostic DNS resolution. Runs the full pipeline (overrides, blocklist,
@@ -99,6 +105,7 @@ pub async fn resolve_query(
    // Pipeline: overrides -> .localhost -> local zones -> special-use (unless forwarded)
    //        -> .tld proxy -> blocklist -> cache -> forwarding -> recursive/upstream
    // Each lock is scoped to avoid holding MutexGuard across await points.
+    let mut upstream_transport: Option<crate::stats::UpstreamTransport> = None;
    let (response, path, dnssec) = {
        let override_record = ctx.overrides.read().unwrap().lookup(&qname);
        if let Some(record) = override_record {
@@ -170,6 +177,13 @@ pub async fn resolve_query(
                60,
            ));
            (resp, QueryPath::Blocked, DnssecStatus::Indeterminate)
+        } else if qtype == QueryType::AAAA && ctx.filter_aaaa {
+            // RFC 2308 NODATA: NOERROR with empty answer section. Prevents
+            // Happy Eyeballs clients from waiting on an AAAA they'll never use
+            // on IPv4-only networks. NXDOMAIN would be wrong (it'd imply the
+            // name doesn't exist for A either).
+            let resp = DnsPacket::response_from(&query, ResultCode::NOERROR);
+            (resp, QueryPath::Local, DnssecStatus::Indeterminate)
        } else {
            let cached = ctx.cache.read().unwrap().lookup_with_status(&qname, qtype);
            if let Some((cached, cached_dnssec, freshness)) = cached {
@@ -190,13 +204,32 @@ pub async fn resolve_query(
                    resp.header.authed_data = true;
                }
                (resp, QueryPath::Cached, cached_dnssec)
-            } else if let Some(upstream) =
+            } else if let Some(pool) =
                crate::system_dns::match_forwarding_rule(&qname, &ctx.forwarding_rules)
            {
                // Conditional forwarding takes priority over recursive mode
                // (e.g. Tailscale .ts.net, VPC private zones)
-                match forward_and_cache(raw_wire, upstream, ctx, &qname, qtype).await {
+                upstream_transport = pool.preferred().map(|u| u.transport());
+                match forward_with_failover_raw(
+                    raw_wire,
+                    pool,
+                    &ctx.srtt,
+                    ctx.timeout,
+                    ctx.hedge_delay,
+                )
+                .await
+                {
+                    Ok(resp_wire) => match cache_and_parse(ctx, &qname, qtype, &resp_wire) {
                        Ok(resp) => (resp, QueryPath::Forwarded, DnssecStatus::Indeterminate),
+                        Err(e) => {
+                            error!("{} | {:?} {} | PARSE ERROR | {}", src_addr, qtype, qname, e);
+                            (
+                                DnsPacket::response_from(&query, ResultCode::SERVFAIL),
+                                QueryPath::UpstreamError,
+                                DnssecStatus::Indeterminate,
+                            )
+                        }
+                    },
                    Err(e) => {
                        error!(
                            "{} | {:?} {} | FORWARD ERROR | {}",
@@ -210,6 +243,9 @@ pub async fn resolve_query(
                    }
                }
            } else if ctx.upstream_mode == UpstreamMode::Recursive {
+                // Recursive resolution makes UDP hops to roots/TLDs/auths;
+                // tag as Udp so the dashboard can aggregate plaintext-wire
+                // egress honestly. Only mark on success — errors stay None.
                let key = (qname.clone(), qtype);
                let (resp, path, err) = resolve_coalesced(&ctx.inflight, key, &query, || {
                    crate::recursive::resolve_recursive(
@@ -232,6 +268,8 @@ pub async fn resolve_query(
                        qname,
                        err.as_deref().unwrap_or("leader failed")
                    );
+                } else {
+                    upstream_transport = Some(crate::stats::UpstreamTransport::Udp);
                }
                (resp, path, DnssecStatus::Indeterminate)
            } else {
@@ -246,7 +284,10 @@ pub async fn resolve_query(
                .await
                {
                    Ok(resp_wire) => match cache_and_parse(ctx, &qname, qtype, &resp_wire) {
-                        Ok(resp) => (resp, QueryPath::Upstream, DnssecStatus::Indeterminate),
+                        Ok(resp) => {
+                            upstream_transport = pool.preferred().map(|u| u.transport());
+                            (resp, QueryPath::Upstream, DnssecStatus::Indeterminate)
+                        }
                        Err(e) => {
                            error!("{} | {:?} {} | PARSE ERROR | {}", src_addr, qtype, qname, e);
                            (
@@ -314,6 +355,15 @@ pub async fn resolve_query(
        strip_dnssec_records(&mut response);
    }

+    // filter_aaaa: also strip ipv6hint from HTTPS/SVCB answers so modern
+    // browsers (Chrome ≥103 etc.) don't receive v6 address hints via the
+    // HTTPS record path that bypasses AAAA entirely. Gated on !client_do
+    // because modifying rdata invalidates any accompanying RRSIG — a DO-bit
+    // validator downstream would reject the response as Bogus.
+    if ctx.filter_aaaa && !client_do {
+        strip_svcb_ipv6_hints(&mut response);
+    }
+
    // Echo EDNS back if client sent it
    if query.edns.is_some() {
        response.edns = Some(crate::packet::EdnsOpt {
@@ -357,7 +407,7 @@ pub async fn resolve_query(
    // Record stats and query log
    {
        let mut s = ctx.stats.lock().unwrap();
-        let total = s.record(path, transport);
+        let total = s.record(path, transport, upstream_transport);
        if total.is_multiple_of(1000) {
            s.log_summary();
        }
@@ -433,17 +483,6 @@ pub async fn refresh_entry(ctx: &ServerCtx, qname: &str, qtype: QueryType) {
    }
 }

-async fn forward_and_cache(
-    wire: &[u8],
-    upstream: &Upstream,
-    ctx: &ServerCtx,
-    qname: &str,
-    qtype: QueryType,
-) -> crate::Result<DnsPacket> {
-    let resp_wire = forward_query_raw(wire, upstream, ctx.timeout).await?;
-    cache_and_parse(ctx, qname, qtype, &resp_wire)
-}
-
 pub async fn handle_query(
    mut buffer: BytePacketBuffer,
    raw_len: usize,
@@ -482,6 +521,20 @@ fn strip_dnssec_records(pkt: &mut DnsPacket) {
    pkt.resources.retain(|r| !is_dnssec_record(r));
 }

+fn strip_svcb_ipv6_hints(pkt: &mut DnsPacket) {
+    let https_qtype = QueryType::HTTPS.to_num();
+    let svcb_qtype = QueryType::SVCB.to_num();
+    pkt.for_each_record_mut(|rec| {
+        if let DnsRecord::UNKNOWN { qtype, data, .. } = rec {
+            if *qtype == https_qtype || *qtype == svcb_qtype {
+                if let Some(new_data) = crate::svcb::strip_ipv6hint(data) {
+                    *data = new_data;
+                }
+            }
+        }
+    });
+}
+
 fn is_special_use_domain(qname: &str) -> bool {
    if qname.ends_with(".in-addr.arpa") {
        // RFC 6303: private + loopback + link-local reverse DNS
@@ -1082,7 +1135,7 @@ mod tests {
        let mut ctx = crate::testutil::test_ctx().await;
        ctx.forwarding_rules = vec![ForwardingRule::new(
            "168.192.in-addr.arpa".to_string(),
-            Upstream::Udp(upstream_addr),
+            UpstreamPool::new(vec![Upstream::Udp(upstream_addr)], vec![]),
        )];
        let ctx = Arc::new(ctx);

@@ -1178,6 +1231,201 @@ mod tests {
        }
    }

+    #[tokio::test]
+    async fn pipeline_filter_aaaa_returns_nodata() {
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.filter_aaaa = true;
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "example.com", QueryType::AAAA).await;
+        assert_eq!(path, QueryPath::Local);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        assert!(resp.answers.is_empty(), "AAAA must be filtered to NODATA");
+    }
+
+    #[tokio::test]
+    async fn pipeline_filter_aaaa_leaves_a_queries_alone() {
+        let mut upstream_resp = DnsPacket::new();
+        upstream_resp.header.response = true;
+        upstream_resp.header.rescode = ResultCode::NOERROR;
+        upstream_resp.answers.push(DnsRecord::A {
+            domain: "example.com".to_string(),
+            addr: Ipv4Addr::new(93, 184, 216, 34),
+            ttl: 300,
+        });
+        let upstream_addr = crate::testutil::mock_upstream(upstream_resp).await;
+
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.filter_aaaa = true;
+        ctx.upstream_pool
+            .lock()
+            .unwrap()
+            .set_primary(vec![Upstream::Udp(upstream_addr)]);
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "example.com", QueryType::A).await;
+        assert_eq!(path, QueryPath::Upstream);
+        assert_eq!(resp.answers.len(), 1);
+    }
+
+    #[tokio::test]
+    async fn pipeline_filter_aaaa_respects_override() {
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.filter_aaaa = true;
+        ctx.overrides
+            .write()
+            .unwrap()
+            .insert("v6.test", "2001:db8::1", 60, None)
+            .unwrap();
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "v6.test", QueryType::AAAA).await;
+        assert_eq!(path, QueryPath::Overridden);
+        assert_eq!(resp.answers.len(), 1, "override must win over filter");
+    }
+
+    #[tokio::test]
+    async fn pipeline_filter_aaaa_strips_ipv6hint_from_https_and_svcb() {
+        let rdata = crate::svcb::build_rdata(
+            1,
+            &[],
+            &[
+                (1, vec![0x02, b'h', b'3']),
+                (
+                    6,
+                    vec![
+                        0x26, 0x06, 0x47, 0x00, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x01,
+                    ],
+                ),
+            ],
+        );
+
+        let mut pkt = DnsPacket::new();
+        pkt.header.response = true;
+        pkt.header.rescode = ResultCode::NOERROR;
+        pkt.questions.push(crate::question::DnsQuestion {
+            name: "hints.test".to_string(),
+            qtype: QueryType::HTTPS,
+        });
+        pkt.answers.push(DnsRecord::UNKNOWN {
+            domain: "hints.test".to_string(),
+            qtype: 65,
+            data: rdata.clone(),
+            ttl: 300,
+        });
+
+        let mut svcb_pkt = pkt.clone();
+        svcb_pkt.questions[0].name = "svc.test".to_string();
+        svcb_pkt.questions[0].qtype = QueryType::SVCB;
+        if let DnsRecord::UNKNOWN { domain, qtype, .. } = &mut svcb_pkt.answers[0] {
+            *domain = "svc.test".to_string();
+            *qtype = 64;
+        }
+
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.filter_aaaa = true;
+        ctx.cache
+            .write()
+            .unwrap()
+            .insert("hints.test", QueryType::HTTPS, &pkt);
+        ctx.cache
+            .write()
+            .unwrap()
+            .insert("svc.test", QueryType::SVCB, &svcb_pkt);
+        let ctx = Arc::new(ctx);
+
+        for (name, qtype, label) in [
+            ("hints.test", QueryType::HTTPS, "HTTPS"),
+            ("svc.test", QueryType::SVCB, "SVCB"),
+        ] {
+            let (resp, path) = resolve_in_test(&ctx, name, qtype).await;
+            assert_eq!(path, QueryPath::Cached, "{label}");
+            assert_eq!(resp.answers.len(), 1, "{label}");
+            match &resp.answers[0] {
+                DnsRecord::UNKNOWN { data, .. } => {
+                    assert!(
+                        data.len() < rdata.len(),
+                        "{label}: ipv6hint (20 bytes) must be removed"
+                    );
+                    // Bytes for key=6 must not appear at any 4-byte boundary in the
+                    // params section — cheap structural check.
+                    assert!(
+                        !data.windows(4).any(|w| w == [0, 6, 0, 16]),
+                        "{label}: ipv6hint TLV header must be absent"
+                    );
+                }
+                other => panic!("{label}: expected UNKNOWN record, got {other:?}"),
+            }
+        }
+    }
+
+    #[tokio::test]
+    async fn pipeline_filter_aaaa_preserves_ipv6hint_for_dnssec_clients() {
+        // Regression guard for the DO-bit gate in resolve_query: modifying
+        // HTTPS rdata invalidates any accompanying RRSIG, so a DO=1 client
+        // must receive the record untouched even when filter_aaaa is on.
+        let rdata = crate::svcb::build_rdata(
+            1,
+            &[],
+            &[(
+                6,
+                vec![
+                    0x26, 0x06, 0x47, 0x00, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x01,
+                ],
+            )],
+        );
+
+        let mut pkt = DnsPacket::new();
+        pkt.header.response = true;
+        pkt.header.rescode = ResultCode::NOERROR;
+        pkt.questions.push(crate::question::DnsQuestion {
+            name: "hints.test".to_string(),
+            qtype: QueryType::HTTPS,
+        });
+        pkt.answers.push(DnsRecord::UNKNOWN {
+            domain: "hints.test".to_string(),
+            qtype: 65,
+            data: rdata.clone(),
+            ttl: 300,
+        });
+
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.filter_aaaa = true;
+        ctx.cache
+            .write()
+            .unwrap()
+            .insert("hints.test", QueryType::HTTPS, &pkt);
+        let ctx = Arc::new(ctx);
+
+        // Build a query with EDNS DO bit set — can't use resolve_in_test
+        // because it constructs a plain query without EDNS.
+        let mut query = DnsPacket::query(0xBEEF, "hints.test", QueryType::HTTPS);
+        query.edns = Some(crate::packet::EdnsOpt {
+            do_bit: true,
+            ..Default::default()
+        });
+        let mut buf = BytePacketBuffer::new();
+        query.write(&mut buf).unwrap();
+        let raw = &buf.buf[..buf.pos];
+        let src: SocketAddr = "127.0.0.1:1234".parse().unwrap();
+
+        let (resp_buf, _) = resolve_query(query, raw, src, &ctx, Transport::Udp)
+            .await
+            .unwrap();
+        let mut resp_parse_buf = BytePacketBuffer::from_bytes(resp_buf.filled());
+        let resp = DnsPacket::from_buffer(&mut resp_parse_buf).unwrap();
+
+        match &resp.answers[0] {
+            DnsRecord::UNKNOWN { data, .. } => {
+                assert_eq!(
+                    data, &rdata,
+                    "ipv6hint must be preserved for DO-bit clients"
+                );
+            }
+            other => panic!("expected UNKNOWN record, got {:?}", other),
+        }
+    }
+
    #[tokio::test]
    async fn pipeline_blocklist_sinkhole() {
        let ctx = crate::testutil::test_ctx().await;
@@ -1237,7 +1485,7 @@ mod tests {
        let mut ctx = crate::testutil::test_ctx().await;
        ctx.forwarding_rules = vec![ForwardingRule::new(
            "corp".to_string(),
-            Upstream::Udp(upstream_addr),
+            UpstreamPool::new(vec![Upstream::Udp(upstream_addr)], vec![]),
        )];
        let ctx = Arc::new(ctx);

@@ -1254,6 +1502,37 @@ mod tests {
        }
    }

+    #[tokio::test]
+    async fn pipeline_forwarding_fails_over_to_second_upstream() {
+        let dead = crate::testutil::blackhole_upstream();
+
+        let mut live_resp = DnsPacket::new();
+        live_resp.header.response = true;
+        live_resp.header.rescode = ResultCode::NOERROR;
+        live_resp.answers.push(DnsRecord::A {
+            domain: "internal.corp".to_string(),
+            addr: Ipv4Addr::new(10, 9, 9, 9),
+            ttl: 600,
+        });
+        let live = crate::testutil::mock_upstream(live_resp).await;
+
+        let mut ctx = crate::testutil::test_ctx().await;
+        ctx.forwarding_rules = vec![ForwardingRule::new(
+            "corp".to_string(),
+            UpstreamPool::new(vec![Upstream::Udp(dead), Upstream::Udp(live)], vec![]),
+        )];
+        let ctx = Arc::new(ctx);
+
+        let (resp, path) = resolve_in_test(&ctx, "internal.corp", QueryType::A).await;
+        assert_eq!(path, QueryPath::Forwarded);
+        assert_eq!(resp.header.rescode, ResultCode::NOERROR);
+        assert_eq!(resp.answers.len(), 1);
+        match &resp.answers[0] {
+            DnsRecord::A { addr, .. } => assert_eq!(*addr, Ipv4Addr::new(10, 9, 9, 9)),
+            other => panic!("expected A record, got {:?}", other),
+        }
+    }
+
    #[tokio::test]
    async fn pipeline_default_pool_reports_upstream_path() {
        let mut upstream_resp = DnsPacket::new();
--- a/src/forward.rs
+++ b/src/forward.rs
@@ -1,14 +1,16 @@
 use std::fmt;
 use std::net::{IpAddr, SocketAddr};
-use std::sync::RwLock;
+use std::sync::{Arc, RwLock};
 use std::time::{Duration, Instant};

 use tokio::net::UdpSocket;
 use tokio::time::timeout;

 use crate::buffer::BytePacketBuffer;
+use crate::odoh::{query_through_relay, OdohConfigCache};
 use crate::packet::DnsPacket;
 use crate::srtt::SrttCache;
+use crate::stats::UpstreamTransport;
 use crate::Result;

 #[derive(Clone)]
@@ -23,6 +25,36 @@ pub enum Upstream {
        tls_name: Option<String>,
        connector: tokio_rustls::TlsConnector,
    },
+    /// Oblivious DNS-over-HTTPS (RFC 9230). Queries are HPKE-sealed to the
+    /// target and forwarded through an independent relay. Target host lives
+    /// on `target_config` (single source of truth — the cache keys on it).
+    Odoh {
+        relay_url: String,
+        target_path: String,
+        client: reqwest::Client,
+        target_config: Arc<OdohConfigCache>,
+    },
+}
+
+impl Upstream {
+    /// IP address to key SRTT tracking on, if the upstream has a stable one.
+    /// `Doh` and `Odoh` route through a URL + connection pool, so there's no
+    /// single IP to track; SRTT is skipped for them.
+    pub fn tracked_ip(&self) -> Option<IpAddr> {
+        match self {
+            Upstream::Udp(addr) | Upstream::Dot { addr, .. } => Some(addr.ip()),
+            Upstream::Doh { .. } | Upstream::Odoh { .. } => None,
+        }
+    }
+
+    pub fn transport(&self) -> UpstreamTransport {
+        match self {
+            Upstream::Udp(_) => UpstreamTransport::Udp,
+            Upstream::Doh { .. } => UpstreamTransport::Doh,
+            Upstream::Dot { .. } => UpstreamTransport::Dot,
+            Upstream::Odoh { .. } => UpstreamTransport::Odoh,
+        }
+    }
 }

 impl PartialEq for Upstream {
@@ -31,6 +63,20 @@ impl PartialEq for Upstream {
            (Self::Udp(a), Self::Udp(b)) => a == b,
            (Self::Doh { url: a, .. }, Self::Doh { url: b, .. }) => a == b,
            (Self::Dot { addr: a, .. }, Self::Dot { addr: b, .. }) => a == b,
+            (
+                Self::Odoh {
+                    relay_url: ra,
+                    target_path: pa,
+                    target_config: ca,
+                    ..
+                },
+                Self::Odoh {
+                    relay_url: rb,
+                    target_path: pb,
+                    target_config: cb,
+                    ..
+                },
+            ) => ra == rb && pa == pb && ca.target_host() == cb.target_host(),
            _ => false,
        }
    }
@@ -51,6 +97,18 @@ impl fmt::Display for Upstream {
                Some(name) => write!(f, "tls://{}#{}", addr, name),
                None => write!(f, "tls://{}", addr),
            },
+            Upstream::Odoh {
+                relay_url,
+                target_path,
+                target_config,
+                ..
+            } => write!(
+                f,
+                "odoh://{}{} via {}",
+                target_config.target_host(),
+                target_path,
+                relay_url
+            ),
        }
    }
 }
@@ -70,22 +128,20 @@ pub(crate) fn parse_upstream_addr(
    Err(format!("invalid upstream address: {}", s))
 }

+/// Parse a slice of upstream address strings into `Upstream` values, failing
+/// on the first invalid entry.
+pub fn parse_upstream_list(addrs: &[String], default_port: u16) -> Result<Vec<Upstream>> {
+    addrs
+        .iter()
+        .map(|s| parse_upstream(s, default_port))
+        .collect()
+}
+
 pub fn parse_upstream(s: &str, default_port: u16) -> Result<Upstream> {
    if s.starts_with("https://") {
-        let client = reqwest::Client::builder()
-            .use_rustls_tls()
-            .http2_initial_stream_window_size(65_535)
-            .http2_initial_connection_window_size(65_535)
-            .http2_keep_alive_interval(Duration::from_secs(15))
-            .http2_keep_alive_while_idle(true)
-            .http2_keep_alive_timeout(Duration::from_secs(10))
-            .pool_idle_timeout(Duration::from_secs(300))
-            .pool_max_idle_per_host(1)
-            .build()
-            .unwrap_or_default();
        return Ok(Upstream::Doh {
            url: s.to_string(),
-            client,
+            client: build_https_client(),
        });
    }
    // tls://IP:PORT#hostname  or  tls://IP#hostname  (default port 853)
@@ -106,6 +162,50 @@ pub fn parse_upstream(s: &str, default_port: u16) -> Result<Upstream> {
    Ok(Upstream::Udp(addr))
 }

+/// HTTP/2 client tuned for DoH/ODoH: small windows for low latency, long-lived
+/// keep-alive. Shared by the DoH upstream and the ODoH config-fetcher +
+/// seal/open path. Pool defaults to one idle conn per host — good for
+/// resolvers that talk to a single upstream; relays that fan out to many
+/// targets should use [`build_https_client_with_pool`].
+pub fn build_https_client() -> reqwest::Client {
+    build_https_client_with_pool(1)
+}
+
+/// Same shape as [`build_https_client`], but caller picks
+/// `pool_max_idle_per_host`. Relay workloads hit many distinct target hosts
+/// and benefit from a larger pool so warm connections survive concurrent
+/// fan-out.
+pub fn build_https_client_with_pool(pool_max_idle_per_host: usize) -> reqwest::Client {
+    https_client_builder(pool_max_idle_per_host)
+        .build()
+        .unwrap_or_default()
+}
+
+/// HTTPS client for the ODoH upstream, with bootstrap-IP overrides applied
+/// so relay/target hostname resolution can bypass system DNS.
+pub fn build_odoh_client(odoh: &crate::config::OdohUpstream) -> reqwest::Client {
+    let mut builder = https_client_builder(1);
+    if let Some(addr) = odoh.relay_bootstrap {
+        builder = builder.resolve(&odoh.relay_host, addr);
+    }
+    if let Some(addr) = odoh.target_bootstrap {
+        builder = builder.resolve(&odoh.target_host, addr);
+    }
+    builder.build().unwrap_or_default()
+}
+
+fn https_client_builder(pool_max_idle_per_host: usize) -> reqwest::ClientBuilder {
+    reqwest::Client::builder()
+        .use_rustls_tls()
+        .http2_initial_stream_window_size(65_535)
+        .http2_initial_connection_window_size(65_535)
+        .http2_keep_alive_interval(Duration::from_secs(15))
+        .http2_keep_alive_while_idle(true)
+        .http2_keep_alive_timeout(Duration::from_secs(10))
+        .pool_idle_timeout(Duration::from_secs(300))
+        .pool_max_idle_per_host(pool_max_idle_per_host)
+}
+
 fn build_dot_connector() -> Result<tokio_rustls::TlsConnector> {
    let _ = rustls::crypto::ring::default_provider().install_default();
    let mut root_store = rustls::RootCertStore::empty();
@@ -270,6 +370,22 @@ pub async fn forward_query_raw(
            tls_name,
            connector,
        } => forward_dot_raw(wire, *addr, tls_name, connector, timeout_duration).await,
+        Upstream::Odoh {
+            relay_url,
+            target_path,
+            client,
+            target_config,
+        } => {
+            query_through_relay(
+                wire,
+                relay_url,
+                target_path,
+                client,
+                target_config,
+                timeout_duration,
+            )
+            .await
+        }
    }
 }

@@ -345,18 +461,17 @@ pub async fn forward_with_failover_raw(
    timeout_duration: Duration,
    hedge_delay: Duration,
 ) -> Result<Vec<u8>> {
-    let mut candidates: Vec<(usize, u64)> = pool
-        .primary
+    let mut candidates: Vec<(usize, u64)> = {
+        let srtt_read = srtt.read().unwrap();
+        pool.primary
            .iter()
            .enumerate()
            .map(|(i, u)| {
-            let rtt = match u {
-                Upstream::Udp(addr) => srtt.read().unwrap().get(addr.ip()),
-                _ => 0,
-            };
+                let rtt = u.tracked_ip().map(|ip| srtt_read.get(ip)).unwrap_or(0);
                (i, rtt)
            })
-        .collect();
+            .collect()
+    };
    candidates.sort_by_key(|&(_, rtt)| rtt);

    let all_upstreams: Vec<&Upstream> = candidates
@@ -380,15 +495,15 @@ pub async fn forward_with_failover_raw(
        };
        match result {
            Ok(resp) => {
-                if let Upstream::Udp(addr) = upstream {
+                if let Some(ip) = upstream.tracked_ip() {
                    let rtt_ms = start.elapsed().as_millis() as u64;
-                    srtt.write().unwrap().record_rtt(addr.ip(), rtt_ms, false);
+                    srtt.write().unwrap().record_rtt(ip, rtt_ms, false);
                }
                return Ok(resp);
            }
            Err(e) => {
-                if let Upstream::Udp(addr) = upstream {
-                    srtt.write().unwrap().record_failure(addr.ip());
+                if let Some(ip) = upstream.tracked_ip() {
+                    srtt.write().unwrap().record_failure(ip);
                }
                log::debug!("upstream {} failed: {}", upstream, e);
                last_err = Some(e);
@@ -707,4 +822,62 @@ mod tests {
        assert!(!pool.maybe_update_primary("not-an-ip", 53));
        assert_eq!(pool.preferred().unwrap().to_string(), "1.2.3.4:53");
    }
+
+    fn tcp_closed_port() -> SocketAddr {
+        // Bind a TCP listener, grab the port, drop → kernel returns RST on connect.
+        let listener = std::net::TcpListener::bind("127.0.0.1:0").unwrap();
+        let addr = listener.local_addr().unwrap();
+        drop(listener);
+        addr
+    }
+
+    #[tokio::test]
+    async fn udp_failure_records_in_srtt() {
+        let blackhole = crate::testutil::blackhole_upstream();
+        let pool = UpstreamPool::new(vec![Upstream::Udp(blackhole)], vec![]);
+        let srtt = RwLock::new(SrttCache::new(true));
+        let _ = forward_with_failover_raw(
+            &[0u8; 12],
+            &pool,
+            &srtt,
+            Duration::from_millis(100),
+            Duration::ZERO,
+        )
+        .await;
+        assert!(srtt.read().unwrap().is_known(blackhole.ip()));
+    }
+
+    #[tokio::test]
+    async fn dot_failure_records_in_srtt() {
+        let dead1 = tcp_closed_port();
+        let dead2 = tcp_closed_port();
+        let connector = build_dot_connector().unwrap();
+        let pool = UpstreamPool::new(
+            vec![
+                Upstream::Dot {
+                    addr: dead1,
+                    tls_name: Some("dns.quad9.net".to_string()),
+                    connector: connector.clone(),
+                },
+                Upstream::Dot {
+                    addr: dead2,
+                    tls_name: Some("dns.quad9.net".to_string()),
+                    connector,
+                },
+            ],
+            vec![],
+        );
+        let srtt = RwLock::new(SrttCache::new(true));
+        let _ = forward_with_failover_raw(
+            &[0u8; 12],
+            &pool,
+            &srtt,
+            Duration::from_millis(500),
+            Duration::ZERO,
+        )
+        .await;
+        let cache = srtt.read().unwrap();
+        assert!(cache.is_known(dead1.ip()));
+        assert!(cache.is_known(dead2.ip()));
+    }
 }
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -13,6 +13,7 @@ pub mod health;
 pub mod lan;
 pub mod mobile_api;
 pub mod mobileconfig;
+pub mod odoh;
 pub mod override_store;
 pub mod packet;
 pub mod proxy;
@@ -20,11 +21,13 @@ pub mod query_log;
 pub mod question;
 pub mod record;
 pub mod recursive;
+pub mod relay;
 pub mod serve;
 pub mod service_store;
 pub mod setup_phone;
 pub mod srtt;
 pub mod stats;
+pub mod svcb;
 pub mod system_dns;
 pub mod tls;
 pub mod wire;
--- a/src/main.rs
+++ b/src/main.rs
@@ -60,6 +60,32 @@ fn main() -> numa::Result<()> {
                .block_on(numa::setup_phone::run())
                .map_err(|e| e.into());
        }
+        "relay" => {
+            let port: u16 = std::env::args()
+                .nth(2)
+                .as_deref()
+                .and_then(|s| s.parse().ok())
+                .unwrap_or(8443);
+            let bind: std::net::IpAddr = std::env::args()
+                .nth(3)
+                .as_deref()
+                .map(|s| {
+                    s.parse().unwrap_or_else(|e| {
+                        eprintln!("invalid bind address '{}': {}", s, e);
+                        std::process::exit(1);
+                    })
+                })
+                .unwrap_or(std::net::IpAddr::V4(std::net::Ipv4Addr::LOCALHOST));
+            let addr = std::net::SocketAddr::new(bind, port);
+            eprintln!(
+                "\x1b[1;38;2;192;98;58mNuma\x1b[0m — ODoH relay on {}\n",
+                addr
+            );
+            let runtime = tokio::runtime::Builder::new_multi_thread()
+                .enable_all()
+                .build()?;
+            return runtime.block_on(numa::relay::run(addr));
+        }
        "lan" => {
            let sub = std::env::args().nth(2).unwrap_or_default();
            let config_path = std::env::args()
@@ -91,6 +117,8 @@ fn main() -> numa::Result<()> {
            eprintln!("  service status  Check if the service is running");
            eprintln!("  lan on          Enable LAN service discovery (mDNS)");
            eprintln!("  lan off         Disable LAN service discovery");
+            eprintln!("  relay [PORT] [BIND]");
+            eprintln!("                  Run as an ODoH relay (RFC 9230, default 127.0.0.1:8443)");
            eprintln!("  setup-phone     Generate a QR code to install Numa DoT on a phone");
            eprintln!("  help            Show this help");
            eprintln!();
--- a/src/odoh.rs
+++ b/src/odoh.rs
@@ -0,0 +1,489 @@
+//! ODoH target-config fetcher and TTL cache (RFC 9230 §6).
+//!
+//! ## Ciphersuite policy
+//! `odoh-rs` deserialization rejects any config whose KEM/KDF/AEAD triple is
+//! not the mandatory `(X25519, HKDF-SHA256, AES-128-GCM)` (see
+//! `ObliviousDoHConfigContents::deserialize`). This is stricter than the
+//! plan's "pick the mandatory suite if mixed": a response containing *any*
+//! non-mandatory config fails parse entirely. Real-world targets publish a
+//! single mandatory config, so this is fine in practice; revisit if a target
+//! that matters starts mixing suites.
+
+use std::sync::Arc;
+use std::time::{Duration, Instant};
+
+use arc_swap::ArcSwapOption;
+use odoh_rs::{
+    ObliviousDoHConfigContents, ObliviousDoHConfigs, ObliviousDoHMessage,
+    ObliviousDoHMessagePlaintext,
+};
+use rand_core::{OsRng, TryRngCore};
+use reqwest::header::HeaderMap;
+use tokio::sync::Mutex;
+use tokio::time::timeout;
+
+use crate::Result;
+
+/// MIME type used for both directions of the ODoH exchange (RFC 9230 §4).
+pub(crate) const ODOH_CONTENT_TYPE: &str = "application/oblivious-dns-message";
+
+/// Cap on the response body we read into memory when the relay returns
+/// non-success. Protects against a hostile relay streaming a huge body on
+/// the error path; keeps enough room to carry a human-readable reason.
+const ERROR_BODY_PREVIEW_BYTES: usize = 1024;
+
+/// Fallback TTL when the target's response lacks a usable `Cache-Control`
+/// directive. RFC 9230 §6.2 places no hard floor; 24 h matches what Cloudflare
+/// publishes in practice.
+const DEFAULT_CONFIG_TTL: Duration = Duration::from_secs(24 * 60 * 60);
+
+/// Cap on any TTL we'll honour, regardless of what the target advertises.
+/// Keeps a misconfigured server from pinning an old key indefinitely.
+const MAX_CONFIG_TTL: Duration = Duration::from_secs(7 * 24 * 60 * 60);
+
+/// After a failed `/.well-known/odohconfigs` fetch, refuse to refetch again
+/// within this window — a target that is genuinely broken would otherwise
+/// receive one request per query. Queries that arrive during the backoff
+/// return the cached error immediately.
+const REFRESH_BACKOFF: Duration = Duration::from_secs(60);
+
+/// Parsed ODoH target config plus the freshness metadata needed to age it out.
+#[derive(Debug)]
+pub struct OdohTargetConfig {
+    pub contents: ObliviousDoHConfigContents,
+    pub key_id: Vec<u8>,
+    expires_at: Instant,
+}
+
+impl OdohTargetConfig {
+    pub fn is_expired(&self) -> bool {
+        Instant::now() >= self.expires_at
+    }
+}
+
+struct FailedRefresh {
+    at: Instant,
+    err: String,
+}
+
+/// TTL-gated cache of a single target's HPKE config.
+///
+/// Reads go through `ArcSwapOption` (lock-free hot path). Refreshes serialize
+/// on an async mutex so a burst of simultaneous misses produces a single
+/// outbound fetch, and a failed refresh blocks subsequent refetches for
+/// [`REFRESH_BACKOFF`] to prevent hot-looping against a broken target.
+pub struct OdohConfigCache {
+    target_host: String,
+    configs_url: String,
+    client: reqwest::Client,
+    current: ArcSwapOption<OdohTargetConfig>,
+    last_failure: ArcSwapOption<FailedRefresh>,
+    refresh_lock: Mutex<()>,
+}
+
+impl OdohConfigCache {
+    pub fn new(target_host: String, client: reqwest::Client) -> Self {
+        let configs_url = format!("https://{}/.well-known/odohconfigs", target_host);
+        Self {
+            target_host,
+            configs_url,
+            client,
+            current: ArcSwapOption::from(None),
+            last_failure: ArcSwapOption::from(None),
+            refresh_lock: Mutex::new(()),
+        }
+    }
+
+    pub fn target_host(&self) -> &str {
+        &self.target_host
+    }
+
+    /// Return a valid config, refetching when the cache is cold or expired.
+    /// Within [`REFRESH_BACKOFF`] of a failed refresh, returns the cached
+    /// error without issuing another fetch.
+    pub async fn get(&self) -> Result<Arc<OdohTargetConfig>> {
+        if let Some(cfg) = self.current.load_full() {
+            if !cfg.is_expired() {
+                return Ok(cfg);
+            }
+        }
+
+        if let Some(err) = self.backoff_error() {
+            return Err(err);
+        }
+
+        let _guard = self.refresh_lock.lock().await;
+
+        // Another task may have refreshed or failed while we waited.
+        if let Some(cfg) = self.current.load_full() {
+            if !cfg.is_expired() {
+                return Ok(cfg);
+            }
+        }
+        if let Some(err) = self.backoff_error() {
+            return Err(err);
+        }
+
+        match fetch_odoh_config(&self.client, &self.configs_url).await {
+            Ok(fresh) => {
+                let fresh = Arc::new(fresh);
+                self.current.store(Some(fresh.clone()));
+                self.last_failure.store(None);
+                Ok(fresh)
+            }
+            Err(e) => {
+                let msg = format!("ODoH config fetch failed: {e}");
+                self.last_failure.store(Some(Arc::new(FailedRefresh {
+                    at: Instant::now(),
+                    err: msg.clone(),
+                })));
+                Err(msg.into())
+            }
+        }
+    }
+
+    /// Drop the cached config. Called after the target rejects ciphertext
+    /// (key rotation race) so the next `get()` refetches.
+    pub fn invalidate(&self) {
+        self.current.store(None);
+    }
+
+    fn backoff_error(&self) -> Option<crate::Error> {
+        let fail = self.last_failure.load_full()?;
+        if fail.at.elapsed() < REFRESH_BACKOFF {
+            Some(format!("{} (backoff active)", fail.err).into())
+        } else {
+            None
+        }
+    }
+}
+
+/// Fetch `/.well-known/odohconfigs` from `configs_url` and parse it into an
+/// [`OdohTargetConfig`]. The TTL is taken from the response's
+/// `Cache-Control: max-age=`, clamped to [`DEFAULT_CONFIG_TTL`,
+/// [`MAX_CONFIG_TTL`]] when absent or obviously wrong.
+pub async fn fetch_odoh_config(
+    client: &reqwest::Client,
+    configs_url: &str,
+) -> Result<OdohTargetConfig> {
+    let resp = client.get(configs_url).send().await?.error_for_status()?;
+    let ttl = cache_control_ttl(resp.headers()).unwrap_or(DEFAULT_CONFIG_TTL);
+    let body = resp.bytes().await?;
+    parse_odoh_config(&body, ttl)
+}
+
+fn parse_odoh_config(body: &[u8], ttl: Duration) -> Result<OdohTargetConfig> {
+    let mut buf = body;
+    let configs: ObliviousDoHConfigs = odoh_rs::parse(&mut buf)
+        .map_err(|e| format!("failed to parse ObliviousDoHConfigs: {e}"))?;
+    let first = configs
+        .into_iter()
+        .next()
+        .ok_or("target published no ODoH configs with a supported version + ciphersuite")?;
+    let contents: ObliviousDoHConfigContents = first.into();
+    let key_id = contents
+        .identifier()
+        .map_err(|e| format!("failed to derive key_id from ODoH config: {e}"))?;
+    Ok(OdohTargetConfig {
+        contents,
+        key_id,
+        expires_at: Instant::now() + ttl.min(MAX_CONFIG_TTL),
+    })
+}
+
+/// Send a DNS wire query through an ODoH relay to a target and return the
+/// plaintext DNS wire response.
+///
+/// Flow: fetch the target's HPKE config (cached), seal the query, POST to the
+/// relay with `Targethost`/`Targetpath` headers, then unseal the response.
+/// On seal/unseal failure we invalidate the cache and retry once — this
+/// handles the benign race where the target rotated its key between our
+/// cached config and the POST.
+pub async fn query_through_relay(
+    wire: &[u8],
+    relay_url: &str,
+    target_path: &str,
+    client: &reqwest::Client,
+    cache: &OdohConfigCache,
+    timeout_duration: Duration,
+) -> Result<Vec<u8>> {
+    let req = OdohRequest {
+        wire,
+        relay_url,
+        target_path,
+        client,
+        cache,
+        timeout: timeout_duration,
+    };
+    match attempt_query(&req).await {
+        Ok(v) => Ok(v),
+        Err(AttemptError::KeyRotation(_)) => {
+            cache.invalidate();
+            attempt_query(&req).await.map_err(AttemptError::into_error)
+        }
+        Err(e) => Err(e.into_error()),
+    }
+}
+
+struct OdohRequest<'a> {
+    wire: &'a [u8],
+    relay_url: &'a str,
+    target_path: &'a str,
+    client: &'a reqwest::Client,
+    cache: &'a OdohConfigCache,
+    timeout: Duration,
+}
+
+/// Classification used only by the retry path in [`query_through_relay`].
+enum AttemptError {
+    /// Target signalled the config we used is stale (key rotation race).
+    /// Callers should invalidate the cache and retry exactly once.
+    KeyRotation(String),
+    /// Any other failure — transport, timeout, malformed response.
+    Other(crate::Error),
+}
+
+impl AttemptError {
+    fn into_error(self) -> crate::Error {
+        match self {
+            AttemptError::KeyRotation(m) => format!("ODoH key rotation race: {m}").into(),
+            AttemptError::Other(e) => e,
+        }
+    }
+}
+
+async fn attempt_query(req: &OdohRequest<'_>) -> std::result::Result<Vec<u8>, AttemptError> {
+    let cfg = req.cache.get().await.map_err(AttemptError::Other)?;
+
+    let plaintext = ObliviousDoHMessagePlaintext::new(req.wire, 0);
+    // rand_core 0.9's OsRng is fallible-only; wrap for the infallible bound.
+    let mut os = OsRng;
+    let mut rng = os.unwrap_mut();
+    let (encrypted_query, client_secret) =
+        odoh_rs::encrypt_query(&plaintext, &cfg.contents, &mut rng)
+            .map_err(|e| AttemptError::Other(format!("ODoH encrypt failed: {e}").into()))?;
+    let body = odoh_rs::compose(&encrypted_query)
+        .map_err(|e| AttemptError::Other(format!("ODoH compose failed: {e}").into()))?
+        .freeze();
+
+    // RFC 9230 §5 and the reference client use URL query parameters, not
+    // HTTP headers, to carry the target routing. `Targethost`/`Targetpath`
+    // headers cause relays to treat the request as an unspecified-target and
+    // reject it.
+    let (status, resp_body) = timeout(req.timeout, async {
+        let resp = req
+            .client
+            .post(req.relay_url)
+            .header(reqwest::header::CONTENT_TYPE, ODOH_CONTENT_TYPE)
+            .header(reqwest::header::ACCEPT, ODOH_CONTENT_TYPE)
+            .header(reqwest::header::CACHE_CONTROL, "no-cache, no-store")
+            .query(&[
+                ("targethost", req.cache.target_host()),
+                ("targetpath", req.target_path),
+            ])
+            .body(body)
+            .send()
+            .await?;
+        let status = resp.status();
+        let body = resp.bytes().await?;
+        Ok::<_, reqwest::Error>((status, body))
+    })
+    .await
+    .map_err(|_| AttemptError::Other("ODoH relay request timed out".into()))?
+    .map_err(|e| AttemptError::Other(format!("ODoH relay request failed: {e}").into()))?;
+
+    // RFC 9230 §4.3 expects a target that can't decrypt to reply with a DNS
+    // error in a sealed 200 response; a 401 from the relay/target is the
+    // practical signal that our cached HPKE key is stale. Treat 400 as a
+    // client-side bug (malformed ODoH envelope) — retrying would loop-fail.
+    if !status.is_success() {
+        let preview_len = resp_body.len().min(ERROR_BODY_PREVIEW_BYTES);
+        let body_preview = String::from_utf8_lossy(&resp_body[..preview_len]);
+        let msg = format!("ODoH relay returned {status}: {}", body_preview.trim());
+        return Err(if status.as_u16() == 401 {
+            AttemptError::KeyRotation(msg)
+        } else {
+            AttemptError::Other(msg.into())
+        });
+    }
+
+    let mut buf = resp_body;
+    let encrypted_response: ObliviousDoHMessage = odoh_rs::parse(&mut buf)
+        .map_err(|e| AttemptError::Other(format!("ODoH response parse failed: {e}").into()))?;
+    let plaintext_response =
+        odoh_rs::decrypt_response(&plaintext, &encrypted_response, client_secret)
+            .map_err(|e| AttemptError::KeyRotation(format!("ODoH decrypt failed: {e}")))?;
+
+    Ok(plaintext_response.into_msg().to_vec())
+}
+
+fn cache_control_ttl(headers: &HeaderMap) -> Option<Duration> {
+    let cc = headers.get(reqwest::header::CACHE_CONTROL)?.to_str().ok()?;
+    for directive in cc.split(',') {
+        let directive = directive.trim();
+        if let Some(rest) = directive.strip_prefix("max-age=") {
+            if let Ok(secs) = rest.trim().parse::<u64>() {
+                if secs > 0 {
+                    return Some(Duration::from_secs(secs));
+                }
+            }
+        }
+    }
+    None
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use odoh_rs::{ObliviousDoHConfig, ObliviousDoHKeyPair};
+
+    // RFC 9180 HPKE IDs for the sole ODoH mandatory suite:
+    // KEM = X25519, KDF = HKDF-SHA256, AEAD = AES-128-GCM.
+    const KEM_X25519: u16 = 0x0020;
+    const KDF_SHA256: u16 = 0x0001;
+    const AEAD_AES128GCM: u16 = 0x0001;
+
+    fn synth_configs_bytes() -> Vec<u8> {
+        let kp = ObliviousDoHKeyPair::from_parameters(
+            KEM_X25519,
+            KDF_SHA256,
+            AEAD_AES128GCM,
+            &[0u8; 32],
+        );
+        let pk = kp.public().clone();
+        let configs: ObliviousDoHConfigs = vec![ObliviousDoHConfig::from(pk)].into();
+        odoh_rs::compose(&configs).unwrap().to_vec()
+    }
+
+    #[test]
+    fn parse_accepts_well_formed_config() {
+        let bytes = synth_configs_bytes();
+        let cfg = parse_odoh_config(&bytes, Duration::from_secs(3600)).unwrap();
+        assert!(!cfg.key_id.is_empty());
+        assert!(!cfg.is_expired());
+    }
+
+    #[test]
+    fn parse_rejects_garbage() {
+        let bytes = [0xffu8; 16];
+        assert!(parse_odoh_config(&bytes, Duration::from_secs(3600)).is_err());
+    }
+
+    #[test]
+    fn parse_rejects_empty() {
+        assert!(parse_odoh_config(&[], Duration::from_secs(3600)).is_err());
+    }
+
+    #[test]
+    fn ttl_capped_at_max() {
+        let bytes = synth_configs_bytes();
+        let cfg = parse_odoh_config(&bytes, Duration::from_secs(100 * 24 * 60 * 60)).unwrap();
+        let remaining = cfg.expires_at.saturating_duration_since(Instant::now());
+        assert!(remaining <= MAX_CONFIG_TTL);
+        assert!(remaining >= MAX_CONFIG_TTL - Duration::from_secs(1));
+    }
+
+    #[test]
+    fn cache_control_parses_max_age() {
+        let mut h = HeaderMap::new();
+        h.insert("cache-control", "public, max-age=86400".parse().unwrap());
+        assert_eq!(cache_control_ttl(&h), Some(Duration::from_secs(86400)));
+    }
+
+    #[test]
+    fn cache_control_ignores_max_age_zero() {
+        let mut h = HeaderMap::new();
+        h.insert("cache-control", "max-age=0, no-store".parse().unwrap());
+        assert_eq!(cache_control_ttl(&h), None);
+    }
+
+    #[test]
+    fn cache_control_missing_falls_back() {
+        let h = HeaderMap::new();
+        assert_eq!(cache_control_ttl(&h), None);
+    }
+
+    #[test]
+    fn is_expired_tracks_ttl() {
+        let bytes = synth_configs_bytes();
+        let mut cfg = parse_odoh_config(&bytes, Duration::from_secs(3600)).unwrap();
+        assert!(!cfg.is_expired());
+        cfg.expires_at = Instant::now() - Duration::from_secs(1);
+        assert!(cfg.is_expired());
+    }
+
+    #[tokio::test]
+    async fn cache_backoff_blocks_refetch_after_failure() {
+        // Point the cache at a host that does not exist so the fetch fails
+        // deterministically; this exercises the backoff wiring without a
+        // network round-trip succeeding.
+        let cache = OdohConfigCache::new(
+            "odoh-target.invalid".to_string(),
+            reqwest::Client::builder()
+                .timeout(Duration::from_millis(200))
+                .build()
+                .unwrap(),
+        );
+
+        let first = cache.get().await;
+        assert!(first.is_err(), "first fetch must fail against invalid host");
+
+        // Within the backoff window, the cached error is returned immediately.
+        let second = cache.get().await.unwrap_err().to_string();
+        assert!(
+            second.contains("backoff active"),
+            "expected backoff hint, got: {second}"
+        );
+
+        // Reaching past the backoff window allows a fresh attempt — simulate
+        // by rewinding the recorded failure timestamp.
+        cache.last_failure.store(Some(Arc::new(FailedRefresh {
+            at: Instant::now() - (REFRESH_BACKOFF + Duration::from_secs(1)),
+            err: "prior".to_string(),
+        })));
+        let third = cache.get().await.unwrap_err().to_string();
+        assert!(
+            !third.contains("backoff active"),
+            "expected fresh fetch attempt, got: {third}"
+        );
+    }
+
+    /// Round-trip the HPKE seal/unseal path in isolation from HTTP, using the
+    /// odoh-rs primitives that `query_through_relay` wires together. Guards
+    /// against silently breaking the crypto glue if we refactor that path.
+    #[test]
+    fn seal_unseal_round_trip() {
+        use odoh_rs::{decrypt_query, encrypt_response, ResponseNonce};
+
+        let kp = ObliviousDoHKeyPair::from_parameters(
+            KEM_X25519,
+            KDF_SHA256,
+            AEAD_AES128GCM,
+            &[0u8; 32],
+        );
+
+        let query_wire = b"\x12\x34\x01\x00\x00\x01\x00\x00\x00\x00\x00\x00\x07example\x03com\x00\x00\x01\x00\x01";
+        let query_pt = ObliviousDoHMessagePlaintext::new(query_wire, 0);
+        let mut os = OsRng;
+        let mut rng = os.unwrap_mut();
+        let (query_enc, client_secret) =
+            odoh_rs::encrypt_query(&query_pt, kp.public(), &mut rng).unwrap();
+
+        let (query_back, server_secret) = decrypt_query(&query_enc, &kp).unwrap();
+        assert_eq!(query_back.into_msg().as_ref(), query_wire);
+
+        let response_wire = b"\x12\x34\x81\x80\x00\x01\x00\x01\x00\x00\x00\x00";
+        let response_pt = ObliviousDoHMessagePlaintext::new(response_wire, 0);
+        let response_enc = encrypt_response(
+            &query_pt,
+            &response_pt,
+            server_secret,
+            ResponseNonce::default(),
+        )
+        .unwrap();
+
+        let response_back =
+            odoh_rs::decrypt_response(&query_pt, &response_enc, client_secret).unwrap();
+        assert_eq!(response_back.into_msg().as_ref(), response_wire);
+    }
+}
--- a/src/packet.rs
+++ b/src/packet.rs
@@ -85,6 +85,14 @@ impl DnsPacket {
            + self.edns.as_ref().map_or(0, |e| e.options.capacity())
    }

+    /// Apply `f` to every record in the three RR sections (answers,
+    /// authorities, resources). Does not touch questions or edns.
+    pub fn for_each_record_mut(&mut self, mut f: impl FnMut(&mut DnsRecord)) {
+        self.answers.iter_mut().for_each(&mut f);
+        self.authorities.iter_mut().for_each(&mut f);
+        self.resources.iter_mut().for_each(&mut f);
+    }
+
    pub fn response_from(query: &DnsPacket, rescode: crate::header::ResultCode) -> DnsPacket {
        let mut resp = DnsPacket::new();
        resp.header.id = query.header.id;
--- a/src/question.rs
+++ b/src/question.rs
@@ -1,115 +1,67 @@
 use crate::buffer::BytePacketBuffer;
 use crate::Result;

+macro_rules! define_qtypes {
+    ( $( $variant:ident = $num:literal, $str:literal ),* $(,)? ) => {
        #[derive(PartialEq, Eq, Debug, Clone, Hash, Copy)]
        pub enum QueryType {
            UNKNOWN(u16),
-    A,      // 1
-    NS,     // 2
-    CNAME,  // 5
-    SOA,    // 6
-    PTR,    // 12
-    MX,     // 15
-    TXT,    // 16
-    AAAA,   // 28
-    SRV,    // 33
-    DS,     // 43
-    RRSIG,  // 46
-    NSEC,   // 47
-    DNSKEY, // 48
-    NSEC3,  // 50
-    OPT,    // 41 (EDNS0 pseudo-type)
-    HTTPS,  // 65
+            $( $variant, )*
        }

        impl QueryType {
            pub fn to_num(&self) -> u16 {
                match *self {
                    QueryType::UNKNOWN(x) => x,
-            QueryType::A => 1,
-            QueryType::NS => 2,
-            QueryType::CNAME => 5,
-            QueryType::SOA => 6,
-            QueryType::PTR => 12,
-            QueryType::MX => 15,
-            QueryType::TXT => 16,
-            QueryType::AAAA => 28,
-            QueryType::SRV => 33,
-            QueryType::OPT => 41,
-            QueryType::DS => 43,
-            QueryType::RRSIG => 46,
-            QueryType::NSEC => 47,
-            QueryType::DNSKEY => 48,
-            QueryType::NSEC3 => 50,
-            QueryType::HTTPS => 65,
+                    $( QueryType::$variant => $num, )*
                }
            }

            pub fn from_num(num: u16) -> QueryType {
                match num {
-            1 => QueryType::A,
-            2 => QueryType::NS,
-            5 => QueryType::CNAME,
-            6 => QueryType::SOA,
-            12 => QueryType::PTR,
-            15 => QueryType::MX,
-            16 => QueryType::TXT,
-            28 => QueryType::AAAA,
-            33 => QueryType::SRV,
-            41 => QueryType::OPT,
-            43 => QueryType::DS,
-            46 => QueryType::RRSIG,
-            47 => QueryType::NSEC,
-            48 => QueryType::DNSKEY,
-            50 => QueryType::NSEC3,
-            65 => QueryType::HTTPS,
+                    $( $num => QueryType::$variant, )*
                    _ => QueryType::UNKNOWN(num),
                }
            }

            pub fn as_str(&self) -> &'static str {
                match self {
-            QueryType::A => "A",
-            QueryType::NS => "NS",
-            QueryType::CNAME => "CNAME",
-            QueryType::SOA => "SOA",
-            QueryType::PTR => "PTR",
-            QueryType::MX => "MX",
-            QueryType::TXT => "TXT",
-            QueryType::AAAA => "AAAA",
-            QueryType::SRV => "SRV",
-            QueryType::OPT => "OPT",
-            QueryType::DS => "DS",
-            QueryType::RRSIG => "RRSIG",
-            QueryType::NSEC => "NSEC",
-            QueryType::DNSKEY => "DNSKEY",
-            QueryType::NSEC3 => "NSEC3",
-            QueryType::HTTPS => "HTTPS",
                    QueryType::UNKNOWN(_) => "UNKNOWN",
+                    $( QueryType::$variant => $str, )*
                }
            }

            pub fn parse_str(s: &str) -> Option<QueryType> {
                match s.to_ascii_uppercase().as_str() {
-            "A" => Some(QueryType::A),
-            "NS" => Some(QueryType::NS),
-            "CNAME" => Some(QueryType::CNAME),
-            "SOA" => Some(QueryType::SOA),
-            "PTR" => Some(QueryType::PTR),
-            "MX" => Some(QueryType::MX),
-            "TXT" => Some(QueryType::TXT),
-            "AAAA" => Some(QueryType::AAAA),
-            "SRV" => Some(QueryType::SRV),
-            "DS" => Some(QueryType::DS),
-            "RRSIG" => Some(QueryType::RRSIG),
-            "DNSKEY" => Some(QueryType::DNSKEY),
-            "NSEC" => Some(QueryType::NSEC),
-            "NSEC3" => Some(QueryType::NSEC3),
-            "HTTPS" => Some(QueryType::HTTPS),
+                    $( $str => Some(QueryType::$variant), )*
                    _ => None,
                }
            }
        }
+    };
+}
+
+define_qtypes! {
+    A      = 1,  "A",
+    NS     = 2,  "NS",
+    CNAME  = 5,  "CNAME",
+    SOA    = 6,  "SOA",
+    PTR    = 12, "PTR",
+    MX     = 15, "MX",
+    TXT    = 16, "TXT",
+    AAAA   = 28, "AAAA",
+    LOC    = 29, "LOC",
+    SRV    = 33, "SRV",
+    NAPTR  = 35, "NAPTR",
+    OPT    = 41, "OPT",
+    DS     = 43, "DS",
+    RRSIG  = 46, "RRSIG",
+    NSEC   = 47, "NSEC",
+    DNSKEY = 48, "DNSKEY",
+    NSEC3  = 50, "NSEC3",
+    SVCB   = 64, "SVCB",
+    HTTPS  = 65, "HTTPS",
+}

 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct DnsQuestion {
--- a/src/relay.rs
+++ b/src/relay.rs
@@ -0,0 +1,342 @@
+//! ODoH relay (RFC 9230 §5) — the forward-without-reading half of the
+//! protocol. Runs `numa relay`; skips all resolver initialisation (no port
+//! 53, no cache, no recursion, no dashboard). The relay never reads the
+//! HPKE-sealed payload and keeps no per-request logs — only aggregate
+//! counters.
+
+use std::net::SocketAddr;
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::Arc;
+use std::time::Duration;
+
+use axum::body::Bytes;
+use axum::extract::{DefaultBodyLimit, Query, State};
+use axum::http::{header, StatusCode};
+use axum::response::{IntoResponse, Response};
+use axum::routing::{get, post};
+use axum::Router;
+use log::{error, info};
+use serde::Deserialize;
+use tokio::net::TcpListener;
+
+use crate::forward::build_https_client_with_pool;
+use crate::odoh::ODOH_CONTENT_TYPE;
+use crate::Result;
+
+/// Cap on the opaque body we accept from a client. ODoH envelopes are
+/// ~100–300 bytes in practice; anything larger is malformed or hostile.
+const MAX_BODY_BYTES: usize = 4 * 1024;
+
+/// Cap on the body we read back from the target before streaming to client.
+/// Slightly larger: target responses carry DNS answers plus HPKE overhead.
+const MAX_TARGET_RESPONSE_BYTES: usize = 8 * 1024;
+
+/// Covers the whole client-to-target round trip — not just `.send()` — so a
+/// slow-drip target can't hang a worker indefinitely after headers arrive.
+const TARGET_REQUEST_TIMEOUT: Duration = Duration::from_secs(5);
+
+/// The relay hits many distinct target hosts on behalf of clients. A
+/// per-host idle pool of 4 keeps warm TLS connections available for concurrent
+/// fan-out without blowing up memory on a small VPS.
+const RELAY_POOL_PER_HOST: usize = 4;
+
+#[derive(Deserialize)]
+struct RelayParams {
+    targethost: String,
+    targetpath: String,
+}
+
+struct RelayState {
+    client: reqwest::Client,
+    total_requests: AtomicU64,
+    forwarded_ok: AtomicU64,
+    forwarded_err: AtomicU64,
+    rejected_bad_request: AtomicU64,
+}
+
+impl RelayState {
+    fn new() -> Arc<Self> {
+        Arc::new(RelayState {
+            client: build_https_client_with_pool(RELAY_POOL_PER_HOST),
+            total_requests: AtomicU64::new(0),
+            forwarded_ok: AtomicU64::new(0),
+            forwarded_err: AtomicU64::new(0),
+            rejected_bad_request: AtomicU64::new(0),
+        })
+    }
+}
+
+/// `DefaultBodyLimit` overrides axum's 2 MiB default so hostile clients
+/// can't force the relay to buffer multi-MB bodies before our own cap.
+fn build_app(state: Arc<RelayState>) -> Router {
+    Router::new()
+        .route("/relay", post(handle_relay))
+        .layer(DefaultBodyLimit::max(MAX_BODY_BYTES))
+        .route("/health", get(handle_health))
+        .with_state(state)
+}
+
+pub async fn run(addr: SocketAddr) -> Result<()> {
+    let app = build_app(RelayState::new());
+    let listener = TcpListener::bind(addr).await?;
+    info!("ODoH relay listening on {}", addr);
+    axum::serve(listener, app).await?;
+    Ok(())
+}
+
+async fn handle_health(State(state): State<Arc<RelayState>>) -> impl IntoResponse {
+    let body = format!(
+        "ok\ntotal {}\nforwarded_ok {}\nforwarded_err {}\nrejected_bad_request {}\n",
+        state.total_requests.load(Ordering::Relaxed),
+        state.forwarded_ok.load(Ordering::Relaxed),
+        state.forwarded_err.load(Ordering::Relaxed),
+        state.rejected_bad_request.load(Ordering::Relaxed),
+    );
+    (
+        StatusCode::OK,
+        [(header::CONTENT_TYPE, "text/plain; charset=utf-8")],
+        body,
+    )
+}
+
+async fn handle_relay(
+    State(state): State<Arc<RelayState>>,
+    Query(params): Query<RelayParams>,
+    headers: axum::http::HeaderMap,
+    body: Bytes,
+) -> Response {
+    state.total_requests.fetch_add(1, Ordering::Relaxed);
+
+    if !content_type_matches(&headers, ODOH_CONTENT_TYPE) {
+        state.rejected_bad_request.fetch_add(1, Ordering::Relaxed);
+        return (
+            StatusCode::UNSUPPORTED_MEDIA_TYPE,
+            "expected application/oblivious-dns-message",
+        )
+            .into_response();
+    }
+
+    if body.len() > MAX_BODY_BYTES {
+        state.rejected_bad_request.fetch_add(1, Ordering::Relaxed);
+        return (StatusCode::PAYLOAD_TOO_LARGE, "body exceeds 4 KiB cap").into_response();
+    }
+
+    if !is_valid_hostname(&params.targethost) || !params.targetpath.starts_with('/') {
+        state.rejected_bad_request.fetch_add(1, Ordering::Relaxed);
+        return (StatusCode::BAD_REQUEST, "invalid targethost or targetpath").into_response();
+    }
+
+    let target_url = format!("https://{}{}", params.targethost, params.targetpath);
+    match forward_to_target(&state.client, &target_url, body).await {
+        Ok((status, resp_body)) => {
+            state.forwarded_ok.fetch_add(1, Ordering::Relaxed);
+            (
+                status,
+                [(header::CONTENT_TYPE, ODOH_CONTENT_TYPE)],
+                resp_body,
+            )
+                .into_response()
+        }
+        Err(e) => {
+            // Log the underlying reason for operators; don't leak reqwest
+            // internals (which can reveal the target's TLS config, IP, etc.)
+            // back to arbitrary clients.
+            error!("relay forward to {} failed: {}", target_url, e);
+            state.forwarded_err.fetch_add(1, Ordering::Relaxed);
+            (StatusCode::BAD_GATEWAY, "target unreachable").into_response()
+        }
+    }
+}
+
+async fn forward_to_target(
+    client: &reqwest::Client,
+    url: &str,
+    body: Bytes,
+) -> Result<(StatusCode, Bytes)> {
+    let response = tokio::time::timeout(TARGET_REQUEST_TIMEOUT, async {
+        let resp = client
+            .post(url)
+            .header(header::CONTENT_TYPE, ODOH_CONTENT_TYPE)
+            .header(header::ACCEPT, ODOH_CONTENT_TYPE)
+            .body(body)
+            .send()
+            .await?;
+        let status = StatusCode::from_u16(resp.status().as_u16())?;
+        let resp_body = resp.bytes().await?;
+        Ok::<_, crate::Error>((status, resp_body))
+    })
+    .await
+    .map_err(|_| "timed out talking to target")??;
+
+    if response.1.len() > MAX_TARGET_RESPONSE_BYTES {
+        return Err("target response exceeds cap".into());
+    }
+    Ok(response)
+}
+
+fn content_type_matches(headers: &axum::http::HeaderMap, expected: &str) -> bool {
+    headers
+        .get(header::CONTENT_TYPE)
+        .and_then(|v| v.to_str().ok())
+        .map(|ct| ct.split(';').next().unwrap_or("").trim() == expected)
+        .unwrap_or(false)
+}
+
+/// Strict DNS-hostname validator, aimed at closing the SSRF surface a naive
+/// `contains('.')` check leaves open (e.g. `example.com@internal.host`,
+/// `evil.com/../admin`). Requires ASCII letters/digits/dot/dash, at least
+/// one dot, no leading dot or dash, length ≤ 253 per RFC 1035.
+fn is_valid_hostname(h: &str) -> bool {
+    if h.is_empty() || h.len() > 253 || !h.contains('.') {
+        return false;
+    }
+    if h.starts_with('.') || h.starts_with('-') || h.ends_with('.') || h.ends_with('-') {
+        return false;
+    }
+    h.chars()
+        .all(|c| c.is_ascii_alphanumeric() || c == '.' || c == '-')
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    async fn spawn_relay() -> (SocketAddr, Arc<RelayState>) {
+        let listener = TcpListener::bind("127.0.0.1:0").await.unwrap();
+        let addr = listener.local_addr().unwrap();
+
+        let state = RelayState::new();
+        let app = build_app(state.clone());
+
+        tokio::spawn(async move {
+            let _ = axum::serve(listener, app).await;
+        });
+        (addr, state)
+    }
+
+    #[tokio::test]
+    async fn rejects_missing_content_type() {
+        let (addr, state) = spawn_relay().await;
+        let client = reqwest::Client::new();
+        let resp = client
+            .post(format!(
+                "http://{}/relay?targethost=odoh.example.com&targetpath=/dns-query",
+                addr
+            ))
+            .body("body")
+            .send()
+            .await
+            .unwrap();
+        assert_eq!(resp.status(), reqwest::StatusCode::UNSUPPORTED_MEDIA_TYPE);
+        assert_eq!(state.rejected_bad_request.load(Ordering::Relaxed), 1);
+    }
+
+    #[tokio::test]
+    async fn rejects_oversized_body() {
+        let (addr, _state) = spawn_relay().await;
+        let big = vec![0u8; MAX_BODY_BYTES + 1];
+        let client = reqwest::Client::new();
+        let resp = client
+            .post(format!(
+                "http://{}/relay?targethost=odoh.example.com&targetpath=/dns-query",
+                addr
+            ))
+            .header(header::CONTENT_TYPE, ODOH_CONTENT_TYPE)
+            .body(big)
+            .send()
+            .await
+            .unwrap();
+        // axum's DefaultBodyLimit rejects before our handler runs, so the
+        // counter doesn't increment — but the status code proves the layer
+        // enforced the cap. Either status is acceptable evidence.
+        assert!(matches!(
+            resp.status(),
+            reqwest::StatusCode::PAYLOAD_TOO_LARGE | reqwest::StatusCode::BAD_REQUEST
+        ));
+    }
+
+    #[tokio::test]
+    async fn rejects_targethost_without_dot() {
+        let (addr, state) = spawn_relay().await;
+        let client = reqwest::Client::new();
+        let resp = client
+            .post(format!(
+                "http://{}/relay?targethost=localhost&targetpath=/dns-query",
+                addr
+            ))
+            .header(header::CONTENT_TYPE, ODOH_CONTENT_TYPE)
+            .body("body")
+            .send()
+            .await
+            .unwrap();
+        assert_eq!(resp.status(), reqwest::StatusCode::BAD_REQUEST);
+        assert_eq!(state.rejected_bad_request.load(Ordering::Relaxed), 1);
+    }
+
+    #[tokio::test]
+    async fn rejects_userinfo_ssrf_attempt() {
+        let (addr, state) = spawn_relay().await;
+        let client = reqwest::Client::new();
+        // The naive contains('.') check would let this through and reqwest
+        // would route to `internal.host` using `evil.com` as userinfo.
+        let resp = client
+            .post(format!(
+                "http://{}/relay?targethost=evil.com@internal.host&targetpath=/dns-query",
+                addr
+            ))
+            .header(header::CONTENT_TYPE, ODOH_CONTENT_TYPE)
+            .body("body")
+            .send()
+            .await
+            .unwrap();
+        assert_eq!(resp.status(), reqwest::StatusCode::BAD_REQUEST);
+        assert_eq!(state.rejected_bad_request.load(Ordering::Relaxed), 1);
+    }
+
+    #[tokio::test]
+    async fn rejects_targetpath_without_leading_slash() {
+        let (addr, state) = spawn_relay().await;
+        let client = reqwest::Client::new();
+        let resp = client
+            .post(format!(
+                "http://{}/relay?targethost=odoh.example.com&targetpath=dns-query",
+                addr
+            ))
+            .header(header::CONTENT_TYPE, ODOH_CONTENT_TYPE)
+            .body("body")
+            .send()
+            .await
+            .unwrap();
+        assert_eq!(resp.status(), reqwest::StatusCode::BAD_REQUEST);
+        assert_eq!(state.rejected_bad_request.load(Ordering::Relaxed), 1);
+    }
+
+    #[tokio::test]
+    async fn health_endpoint_reports_counters() {
+        let (addr, _state) = spawn_relay().await;
+        let client = reqwest::Client::new();
+        let resp = client
+            .get(format!("http://{}/health", addr))
+            .send()
+            .await
+            .unwrap();
+        assert_eq!(resp.status(), reqwest::StatusCode::OK);
+        let body = resp.text().await.unwrap();
+        assert!(body.contains("ok\n"));
+        assert!(body.contains("forwarded_ok 0"));
+    }
+
+    #[test]
+    fn hostname_validator_accepts_and_rejects() {
+        assert!(is_valid_hostname("odoh.cloudflare-dns.com"));
+        assert!(is_valid_hostname("a.b"));
+        assert!(!is_valid_hostname(""));
+        assert!(!is_valid_hostname("localhost"));
+        assert!(!is_valid_hostname(".leading.dot"));
+        assert!(!is_valid_hostname("trailing.dot."));
+        assert!(!is_valid_hostname("-leading.dash"));
+        assert!(!is_valid_hostname("evil.com@internal.host"));
+        assert!(!is_valid_hostname("evil.com/../admin"));
+        assert!(!is_valid_hostname(&"a".repeat(254)));
+    }
+}
--- a/src/serve.rs
+++ b/src/serve.rs
@@ -17,7 +17,10 @@ use crate::buffer::BytePacketBuffer;
 use crate::cache::DnsCache;
 use crate::config::{build_zone_map, load_config, ConfigLoad};
 use crate::ctx::{handle_query, ServerCtx};
-use crate::forward::{parse_upstream, Upstream, UpstreamPool};
+use crate::forward::{
+    build_https_client, build_odoh_client, parse_upstream_list, Upstream, UpstreamPool,
+};
+use crate::odoh::OdohConfigCache;
 use crate::override_store::OverrideStore;
 use crate::query_log::QueryLog;
 use crate::service_store::ServiceStore;
@@ -54,10 +57,7 @@ pub async fn run(config_path: String) -> crate::Result<()> {
                (crate::config::UpstreamMode::Recursive, false, pool, label)
            } else {
                log::warn!("recursive probe failed — falling back to Quad9 DoH");
-                let client = reqwest::Client::builder()
-                    .use_rustls_tls()
-                    .build()
-                    .unwrap_or_default();
+                let client = build_https_client();
                let url = DOH_FALLBACK.to_string();
                let label = url.clone();
                let pool = UpstreamPool::new(vec![Upstream::Doh { url, client }], vec![]);
@@ -82,16 +82,8 @@ pub async fn run(config_path: String) -> crate::Result<()> {
                config.upstream.address.clone()
            };

-            let primary: Vec<Upstream> = addrs
-                .iter()
-                .map(|s| parse_upstream(s, config.upstream.port))
-                .collect::<crate::Result<Vec<_>>>()?;
-            let fallback: Vec<Upstream> = config
-                .upstream
-                .fallback
-                .iter()
-                .map(|s| parse_upstream(s, config.upstream.port))
-                .collect::<crate::Result<Vec<_>>>()?;
+            let primary = parse_upstream_list(&addrs, config.upstream.port)?;
+            let fallback = parse_upstream_list(&config.upstream.fallback, config.upstream.port)?;

            let pool = UpstreamPool::new(primary, fallback);
            let label = pool.label();
@@ -102,6 +94,28 @@ pub async fn run(config_path: String) -> crate::Result<()> {
                label,
            )
        }
+        crate::config::UpstreamMode::Odoh => {
+            let odoh = config.upstream.odoh_upstream()?;
+            let client = build_odoh_client(&odoh);
+            let target_config = Arc::new(OdohConfigCache::new(
+                odoh.target_host.clone(),
+                client.clone(),
+            ));
+            let primary = vec![Upstream::Odoh {
+                relay_url: odoh.relay_url,
+                target_path: odoh.target_path,
+                client,
+                target_config,
+            }];
+            let fallback = if odoh.strict {
+                Vec::new()
+            } else {
+                parse_upstream_list(&config.upstream.fallback, config.upstream.port)?
+            };
+            let pool = UpstreamPool::new(primary, fallback);
+            let label = pool.label();
+            (crate::config::UpstreamMode::Odoh, false, pool, label)
+        }
    };
    let api_port = config.server.api_port;

@@ -123,7 +137,11 @@ pub async fn run(config_path: String) -> crate::Result<()> {

    for fwd in &config.forwarding {
        for suffix in &fwd.suffix {
-            info!("forwarding .{} to {} (config rule)", suffix, fwd.upstream);
+            info!(
+                "forwarding .{} to {} (config rule)",
+                suffix,
+                fwd.upstream.join(", ")
+            );
        }
    }
    let forwarding_rules =
@@ -209,7 +227,7 @@ pub async fn run(config_path: String) -> crate::Result<()> {
        upstream_port: config.upstream.port,
        lan_ip: Mutex::new(crate::lan::detect_lan_ip().unwrap_or(std::net::Ipv4Addr::LOCALHOST)),
        timeout: Duration::from_millis(config.upstream.timeout_ms),
-        hedge_delay: Duration::from_millis(config.upstream.hedge_ms),
+        hedge_delay: resolved_mode.hedge_delay(config.upstream.hedge_ms),
        proxy_tld_suffix: if config.proxy.tld.is_empty() {
            String::new()
        } else {
@@ -232,6 +250,7 @@ pub async fn run(config_path: String) -> crate::Result<()> {
        ca_pem,
        mobile_enabled: config.mobile.enabled,
        mobile_port: config.mobile.port,
+        filter_aaaa: config.server.filter_aaaa,
    });

    let zone_count: usize = ctx.zone_map.values().map(|m| m.len()).sum();
--- a/src/stats.rs
+++ b/src/stats.rs
@@ -102,6 +102,10 @@ pub struct ServerStats {
    transport_tcp: u64,
    transport_dot: u64,
    transport_doh: u64,
+    upstream_transport_udp: u64,
+    upstream_transport_doh: u64,
+    upstream_transport_dot: u64,
+    upstream_transport_odoh: u64,
    started_at: Instant,
 }

@@ -124,6 +128,31 @@ impl Transport {
    }
 }

+/// Wire protocol used for a forwarded upstream call. Orthogonal to
+/// `QueryPath`: the path answers "where the answer came from"; this answers
+/// "over what wire we spoke to the forwarder." Callers pass
+/// `Option<UpstreamTransport>` — `None` for resolutions that never touched
+/// a forwarder (cache/local/blocked) or for recursive mode, which has its
+/// own counter via `QueryPath::Recursive`.
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum UpstreamTransport {
+    Udp,
+    Doh,
+    Dot,
+    Odoh,
+}
+
+impl UpstreamTransport {
+    pub fn as_str(&self) -> &'static str {
+        match self {
+            UpstreamTransport::Udp => "UDP",
+            UpstreamTransport::Doh => "DOH",
+            UpstreamTransport::Dot => "DOT",
+            UpstreamTransport::Odoh => "ODOH",
+        }
+    }
+}
+
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
 pub enum QueryPath {
    Local,
@@ -202,11 +231,20 @@ impl ServerStats {
            transport_tcp: 0,
            transport_dot: 0,
            transport_doh: 0,
+            upstream_transport_udp: 0,
+            upstream_transport_doh: 0,
+            upstream_transport_dot: 0,
+            upstream_transport_odoh: 0,
            started_at: Instant::now(),
        }
    }

-    pub fn record(&mut self, path: QueryPath, transport: Transport) -> u64 {
+    pub fn record(
+        &mut self,
+        path: QueryPath,
+        transport: Transport,
+        upstream_transport: Option<UpstreamTransport>,
+    ) -> u64 {
        self.queries_total += 1;
        match path {
            QueryPath::Local => self.queries_local += 1,
@@ -225,6 +263,14 @@ impl ServerStats {
            Transport::Dot => self.transport_dot += 1,
            Transport::Doh => self.transport_doh += 1,
        }
+        if let Some(ut) = upstream_transport {
+            match ut {
+                UpstreamTransport::Udp => self.upstream_transport_udp += 1,
+                UpstreamTransport::Doh => self.upstream_transport_doh += 1,
+                UpstreamTransport::Dot => self.upstream_transport_dot += 1,
+                UpstreamTransport::Odoh => self.upstream_transport_odoh += 1,
+            }
+        }
        self.queries_total
    }

@@ -253,6 +299,10 @@ impl ServerStats {
            transport_tcp: self.transport_tcp,
            transport_dot: self.transport_dot,
            transport_doh: self.transport_doh,
+            upstream_transport_udp: self.upstream_transport_udp,
+            upstream_transport_doh: self.upstream_transport_doh,
+            upstream_transport_dot: self.upstream_transport_dot,
+            upstream_transport_odoh: self.upstream_transport_odoh,
        }
    }

@@ -263,7 +313,7 @@ impl ServerStats {
        let secs = uptime.as_secs() % 60;

        log::info!(
-            "STATS | uptime {}h{}m{}s | total {} | fwd {} | upstream {} | recursive {} | coalesced {} | cached {} | local {} | override {} | blocked {} | errors {}",
+            "STATS | uptime {}h{}m{}s | total {} | fwd {} | upstream {} | recursive {} | coalesced {} | cached {} | local {} | override {} | blocked {} | errors {} | up-udp {} | up-doh {} | up-dot {} | up-odoh {}",
            hours, mins, secs,
            self.queries_total,
            self.queries_forwarded,
@@ -275,6 +325,10 @@ impl ServerStats {
            self.queries_overridden,
            self.queries_blocked,
            self.upstream_errors,
+            self.upstream_transport_udp,
+            self.upstream_transport_doh,
+            self.upstream_transport_dot,
+            self.upstream_transport_odoh,
        );
    }
 }
@@ -295,4 +349,8 @@ pub struct StatsSnapshot {
    pub transport_tcp: u64,
    pub transport_dot: u64,
    pub transport_doh: u64,
+    pub upstream_transport_udp: u64,
+    pub upstream_transport_doh: u64,
+    pub upstream_transport_dot: u64,
+    pub upstream_transport_odoh: u64,
 }
--- a/src/svcb.rs
+++ b/src/svcb.rs
@@ -0,0 +1,179 @@
+//! Minimal SVCB/HTTPS (RFC 9460) RDATA parser — just enough to strip
+//! the `ipv6hint` SvcParam. Used by the `filter_aaaa` feature so
+//! HTTPS-record-aware clients (Chrome ≥103, Firefox, Safari) don't
+//! receive v6 address hints on IPv4-only networks.
+
+/// SvcParamKey = 6 (RFC 9460 §14.3.2).
+const IPV6_HINT_KEY: u16 = 6;
+
+/// Strip the `ipv6hint` SvcParam from an HTTPS/SVCB RDATA blob.
+///
+/// Returns `Some(new_rdata)` if `ipv6hint` was present and removed.
+/// Returns `None` if the record had no `ipv6hint`, or if the RDATA
+/// couldn't be parsed — in both cases the caller should keep the
+/// original bytes untouched.
+///
+/// SVCB RDATA (RFC 9460 §2.2):
+///   SvcPriority (u16)
+///   TargetName  (uncompressed DNS name — labels terminated by 0 octet)
+///   SvcParams   (series of {u16 key, u16 len, opaque[len] value}, sorted by key)
+pub fn strip_ipv6hint(rdata: &[u8]) -> Option<Vec<u8>> {
+    if rdata.len() < 2 {
+        return None;
+    }
+    let mut pos = 2;
+
+    // TargetName — uncompressed per RFC 9460 §2.2
+    loop {
+        let len = *rdata.get(pos)? as usize;
+        pos += 1;
+        if len == 0 {
+            break;
+        }
+        if len & 0xC0 != 0 {
+            // Pointer: forbidden in SVCB but defend against a broken upstream.
+            return None;
+        }
+        pos = pos.checked_add(len)?;
+        if pos > rdata.len() {
+            return None;
+        }
+    }
+
+    // Scan params once to decide whether we need to rebuild.
+    let params_start = pos;
+    let mut scan = pos;
+    let mut has_ipv6hint = false;
+    while scan < rdata.len() {
+        if scan + 4 > rdata.len() {
+            return None;
+        }
+        let key = u16::from_be_bytes([rdata[scan], rdata[scan + 1]]);
+        let vlen = u16::from_be_bytes([rdata[scan + 2], rdata[scan + 3]]) as usize;
+        let end = scan.checked_add(4)?.checked_add(vlen)?;
+        if end > rdata.len() {
+            return None;
+        }
+        if key == IPV6_HINT_KEY {
+            has_ipv6hint = true;
+        }
+        scan = end;
+    }
+    if scan != rdata.len() || !has_ipv6hint {
+        return None;
+    }
+
+    // Rebuild without ipv6hint, preserving param order (RFC 9460 requires
+    // ascending key order, which we preserve by filtering in place).
+    let mut out = Vec::with_capacity(rdata.len());
+    out.extend_from_slice(&rdata[..params_start]);
+    let mut pos = params_start;
+    while pos < rdata.len() {
+        let key = u16::from_be_bytes([rdata[pos], rdata[pos + 1]]);
+        let vlen = u16::from_be_bytes([rdata[pos + 2], rdata[pos + 3]]) as usize;
+        let end = pos + 4 + vlen;
+        if key != IPV6_HINT_KEY {
+            out.extend_from_slice(&rdata[pos..end]);
+        }
+        pos = end;
+    }
+    Some(out)
+}
+
+/// Build an SVCB RDATA blob from a priority, target labels, and
+/// (key, value) param pairs. Shared by `svcb` unit tests and `ctx`
+/// pipeline tests that need to seed the cache with a synthetic HTTPS RR.
+#[cfg(test)]
+pub(crate) fn build_rdata(priority: u16, target: &[&str], params: &[(u16, Vec<u8>)]) -> Vec<u8> {
+    let mut out = Vec::new();
+    out.extend_from_slice(&priority.to_be_bytes());
+    for label in target {
+        out.push(label.len() as u8);
+        out.extend_from_slice(label.as_bytes());
+    }
+    out.push(0);
+    for (key, value) in params {
+        out.extend_from_slice(&key.to_be_bytes());
+        out.extend_from_slice(&(value.len() as u16).to_be_bytes());
+        out.extend_from_slice(value);
+    }
+    out
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn alpn_h3() -> (u16, Vec<u8>) {
+        // alpn = ["h3"]: one length-prefixed ALPN id
+        (1, vec![0x02, b'h', b'3'])
+    }
+
+    fn ipv4hint_single() -> (u16, Vec<u8>) {
+        (4, vec![93, 184, 216, 34])
+    }
+
+    fn ipv6hint_single() -> (u16, Vec<u8>) {
+        // 2606:4700::1
+        (
+            6,
+            vec![
+                0x26, 0x06, 0x47, 0x00, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0x01,
+            ],
+        )
+    }
+
+    #[test]
+    fn strips_ipv6hint_and_keeps_other_params() {
+        let rdata = build_rdata(1, &[], &[alpn_h3(), ipv4hint_single(), ipv6hint_single()]);
+        let stripped = strip_ipv6hint(&rdata).expect("ipv6hint present → stripped");
+        let expected = build_rdata(1, &[], &[alpn_h3(), ipv4hint_single()]);
+        assert_eq!(stripped, expected);
+    }
+
+    #[test]
+    fn no_ipv6hint_returns_none() {
+        let rdata = build_rdata(1, &[], &[alpn_h3(), ipv4hint_single()]);
+        assert!(strip_ipv6hint(&rdata).is_none());
+    }
+
+    #[test]
+    fn alias_mode_empty_params_returns_none() {
+        let rdata = build_rdata(0, &["example", "com"], &[]);
+        assert!(strip_ipv6hint(&rdata).is_none());
+    }
+
+    #[test]
+    fn only_ipv6hint_yields_empty_param_section() {
+        let rdata = build_rdata(1, &[], &[ipv6hint_single()]);
+        let stripped = strip_ipv6hint(&rdata).expect("ipv6hint present → stripped");
+        let expected = build_rdata(1, &[], &[]);
+        assert_eq!(stripped, expected);
+    }
+
+    #[test]
+    fn preserves_target_name() {
+        let rdata = build_rdata(1, &["svc", "example", "net"], &[ipv6hint_single()]);
+        let stripped = strip_ipv6hint(&rdata).unwrap();
+        assert!(stripped.starts_with(&[0x00, 0x01])); // priority
+        assert_eq!(&stripped[2..6], b"\x03svc");
+    }
+
+    #[test]
+    fn truncated_rdata_returns_none() {
+        // Priority only, no target terminator.
+        assert!(strip_ipv6hint(&[0, 1, 3, b'c', b'o', b'm']).is_none());
+    }
+
+    #[test]
+    fn empty_input_returns_none() {
+        assert!(strip_ipv6hint(&[]).is_none());
+    }
+
+    #[test]
+    fn param_length_overflow_returns_none() {
+        // key=6, length=0xFFFF but value is short — malformed.
+        let rdata = vec![0, 1, 0, 0, 6, 0xFF, 0xFF, 0, 1, 2];
+        assert!(strip_ipv6hint(&rdata).is_none());
+    }
+}
--- a/src/system_dns.rs
+++ b/src/system_dns.rs
@@ -2,7 +2,9 @@ use std::net::SocketAddr;

 use log::info;

+#[cfg(any(target_os = "macos", target_os = "linux"))]
 use crate::forward::Upstream;
+use crate::forward::UpstreamPool;

 fn print_recursive_hint() {
    let is_recursive = crate::config::load_config("numa.toml")
@@ -20,15 +22,15 @@ fn is_loopback_or_stub(addr: &str) -> bool {
 }

 /// A conditional forwarding rule: domains matching `suffix` are forwarded to `upstream`.
-#[derive(Debug, Clone)]
+#[derive(Clone)]
 pub struct ForwardingRule {
    pub suffix: String,
    dot_suffix: String, // pre-computed ".suffix" for zero-alloc matching
-    pub upstream: Upstream,
+    pub upstream: UpstreamPool,
 }

 impl ForwardingRule {
-    pub fn new(suffix: String, upstream: Upstream) -> Self {
+    pub fn new(suffix: String, upstream: UpstreamPool) -> Self {
        let dot_suffix = format!(".{}", suffix);
        Self {
            suffix,
@@ -216,7 +218,8 @@ fn discover_macos() -> SystemDnsInfo {
    for rule in &rules {
        info!(
            "auto-discovered forwarding: *.{} -> {}",
-            rule.suffix, rule.upstream
+            rule.suffix,
+            rule.upstream.label()
        );
    }
    if rules.is_empty() {
@@ -235,7 +238,8 @@ fn discover_macos() -> SystemDnsInfo {
 #[cfg(any(target_os = "macos", target_os = "linux"))]
 fn make_rule(domain: &str, nameserver: &str) -> Option<ForwardingRule> {
    let addr = crate::forward::parse_upstream_addr(nameserver, 53).ok()?;
-    Some(ForwardingRule::new(domain.to_string(), Upstream::Udp(addr)))
+    let pool = UpstreamPool::new(vec![Upstream::Udp(addr)], vec![]);
+    Some(ForwardingRule::new(domain.to_string(), pool))
 }

 #[cfg(target_os = "linux")]
@@ -1033,7 +1037,7 @@ fn uninstall_windows() -> Result<(), String> {
 pub fn match_forwarding_rule<'a>(
    domain: &str,
    rules: &'a [ForwardingRule],
-) -> Option<&'a Upstream> {
+) -> Option<&'a UpstreamPool> {
    for rule in rules {
        if domain == rule.suffix || domain.ends_with(&rule.dot_suffix) {
            return Some(&rule.upstream);
@@ -1412,7 +1416,7 @@ pub fn service_status() -> Result<(), String> {
    }
 }

-#[cfg(any(target_os = "macos", target_os = "linux"))]
+#[cfg(target_os = "macos")]
 fn replace_exe_path(service: &str) -> Result<String, String> {
    let exe_path =
        std::env::current_exe().map_err(|e| format!("failed to get current exe: {}", e))?;
@@ -1660,10 +1664,78 @@ fn uninstall_linux() -> Result<(), String> {
    Ok(())
 }

+/// Fallback install location when current_exe() sits on a path the
+/// dynamic user cannot traverse (e.g. `/home/<user>/` mode 0700).
+#[cfg(target_os = "linux")]
+fn linux_service_exe_path() -> std::path::PathBuf {
+    std::path::PathBuf::from("/usr/local/bin/numa")
+}
+
+/// True iff every ancestor of `p` (excluding `/`) grants world-execute —
+/// i.e. the `DynamicUser=yes` service account can traverse the path and
+/// exec the binary without being in any group. Linuxbrew's
+/// `/home/linuxbrew` is 0755 (traversable, keep brew's path, upgrades
+/// via `brew` propagate). A build tree under `/home/<user>/` (0700) or
+/// `~/.cargo/bin/` is not (copy to /usr/local/bin so systemd can reach it).
+#[cfg(target_os = "linux")]
+fn path_world_traversable_linux(p: &std::path::Path) -> bool {
+    use std::os::unix::fs::PermissionsExt;
+    let mut current = p;
+    while let Some(parent) = current.parent() {
+        if parent.as_os_str().is_empty() || parent == std::path::Path::new("/") {
+            break;
+        }
+        match std::fs::metadata(parent) {
+            Ok(m) if m.permissions().mode() & 0o001 != 0 => {}
+            _ => return false,
+        }
+        current = parent;
+    }
+    true
+}
+
+#[cfg(target_os = "linux")]
+fn install_service_binary_linux() -> Result<std::path::PathBuf, String> {
+    let src = std::env::current_exe().map_err(|e| format!("current_exe(): {}", e))?;
+    if path_world_traversable_linux(&src) {
+        return Ok(src);
+    }
+    let dst = linux_service_exe_path();
+    if src == dst {
+        return Ok(dst);
+    }
+    if let Some(parent) = dst.parent() {
+        std::fs::create_dir_all(parent)
+            .map_err(|e| format!("failed to create {}: {}", parent.display(), e))?;
+    }
+    // Atomic replace via temp + rename. Plain copy fails with ETXTBSY when
+    // re-installing while the service is running the previous binary —
+    // rename swaps the path while the running process keeps the old inode.
+    let tmp = dst.with_extension("new");
+    std::fs::copy(&src, &tmp).map_err(|e| {
+        format!(
+            "failed to copy {} -> {}: {}",
+            src.display(),
+            tmp.display(),
+            e
+        )
+    })?;
+    std::fs::rename(&tmp, &dst).map_err(|e| {
+        let _ = std::fs::remove_file(&tmp);
+        format!(
+            "failed to rename {} -> {}: {}",
+            tmp.display(),
+            dst.display(),
+            e
+        )
+    })?;
+    Ok(dst)
+}
+
 #[cfg(target_os = "linux")]
 fn install_service_linux() -> Result<(), String> {
-    let unit = include_str!("../numa.service");
-    let unit = replace_exe_path(unit)?;
+    let exe = install_service_binary_linux()?;
+    let unit = include_str!("../numa.service").replace("{{exe_path}}", &exe.to_string_lossy());
    std::fs::write(SYSTEMD_UNIT, unit)
        .map_err(|e| format!("failed to write {}: {}", SYSTEMD_UNIT, e))?;

@@ -1675,7 +1747,9 @@ fn install_service_linux() -> Result<(), String> {
        eprintln!("  warning: failed to configure system DNS: {}", e);
    }

-    run_systemctl(&["start", "numa"])?;
+    // restart, not start: on re-install the service is already running
+    // the previous binary; restart picks up the new one.
+    run_systemctl(&["restart", "numa"])?;

    eprintln!("  Service installed and started.");
    eprintln!("  Numa will auto-start on boot and restart if killed.");
@@ -1991,22 +2065,25 @@ Wireless LAN adapter Wi-Fi:
    }

    #[test]
-    #[cfg(any(target_os = "macos", target_os = "linux"))]
-    fn replace_exe_path_substitutes_template() {
+    fn install_templates_contain_exe_path_placeholder() {
+        // Both files are substituted at install time — plist via
+        // replace_exe_path on macOS, numa.service via inline .replace
+        // in install_service_linux. Catch placeholder removal early.
        let plist = include_str!("../com.numa.dns.plist");
        let unit = include_str!("../numa.service");
-
        assert!(plist.contains("{{exe_path}}"), "plist missing placeholder");
        assert!(
            unit.contains("{{exe_path}}"),
            "unit file missing placeholder"
        );
+    }

+    #[test]
+    #[cfg(target_os = "macos")]
+    fn replace_exe_path_substitutes_template() {
+        let plist = include_str!("../com.numa.dns.plist");
        let result = replace_exe_path(plist).expect("replace_exe_path failed for plist");
        assert!(!result.contains("{{exe_path}}"));
-
-        let result = replace_exe_path(unit).expect("replace_exe_path failed for unit");
-        assert!(!result.contains("{{exe_path}}"));
    }

    #[test]
--- a/src/testutil.rs
+++ b/src/testutil.rs
@@ -63,6 +63,7 @@ pub async fn test_ctx() -> ServerCtx {
        ca_pem: None,
        mobile_enabled: false,
        mobile_port: 8765,
+        filter_aaaa: false,
    }
 }

--- a/tests/docker/install-systemd.sh
+++ b/tests/docker/install-systemd.sh
@@ -0,0 +1,288 @@
+#!/usr/bin/env bash
+#
+# Systemd service install verification for the DynamicUser-based Linux
+# service unit. Stands up a privileged ubuntu:24.04 container with systemd
+# as PID 1, builds numa inside, runs three scenarios that CI does not:
+#
+#   A. Fresh install — every advertised port is not just bound but
+#      functional (DNS resolves on :53, TLS handshake validates against
+#      numa's CA on :853/:443, HTTP responds on :80, API on :5380).
+#   B. Upgrade from pre-drop layout (root-owned /var/lib/numa) preserves
+#      the CA fingerprint — users' browser-installed CA trust survives.
+#   C. Install from a 0700 source directory stages the binary under
+#      /usr/local/bin/numa and the service starts from there.
+#
+# First run is slow (~5-10 min): image pull + apt + cold cargo build.
+# Subsequent runs reuse cached docker volumes for cargo + target (~30s).
+#
+# Requirements: docker
+# Usage:        ./tests/docker/install-systemd.sh
+
+set -u
+set -o pipefail
+
+GREEN="\033[32m"; RED="\033[31m"; RESET="\033[0m"
+
+pass() { printf "  ${GREEN}PASS${RESET}: %s\n" "$*"; }
+fail() { printf "  ${RED}FAIL${RESET}: %s\n" "$*"; FAIL=1; }
+
+# ============================================================
+# Mode B: running inside the systemd container — run scenarios
+# ============================================================
+if [ "${NUMA_INSIDE:-}" = "1" ]; then
+    set +e  # assertions report pass/fail, don't abort
+    FAIL=0
+    NUMA=/work/target/release/numa
+
+    reset_state() {
+        "$NUMA" uninstall >/dev/null 2>&1 || true
+        systemctl reset-failed numa 2>/dev/null || true
+        rm -rf /var/lib/numa /var/lib/private/numa /etc/numa /home/builder /usr/local/bin/numa
+        systemctl daemon-reload 2>/dev/null || true
+    }
+
+    main_pid_user() {
+        local pid
+        pid=$(systemctl show -p MainPID --value numa)
+        [ "$pid" != "0" ] || { echo ""; return; }
+        ps -o user= -p "$pid" 2>/dev/null | tr -d ' '
+    }
+
+    # MainPID + user briefly stabilize after a fresh restart. Retry so we
+    # don't race the moment systemd flips the service to "active" vs when
+    # the forked numa process actually owns MainPID.
+    assert_nonroot() {
+        local pid user comm n=0
+        while [ $n -lt 20 ]; do
+            pid=$(systemctl show -p MainPID --value numa)
+            if [ "$pid" != "0" ]; then
+                comm=$(ps -o comm= -p "$pid" 2>/dev/null | tr -d ' ')
+                user=$(ps -o user= -p "$pid" 2>/dev/null | tr -d ' ')
+                if [ "$comm" = "numa" ]; then
+                    if [ "$user" = "root" ]; then
+                        fail "daemon runs as root (expected transient UID)"
+                    else
+                        pass "daemon runs as $user (non-root)"
+                    fi
+                    return
+                fi
+            fi
+            sleep 0.2
+            n=$((n + 1))
+        done
+        fail "numa MainPID did not settle (last: pid=${pid:-?} comm=${comm:-?} user=${user:-?})"
+    }
+
+    # Functional DNS check: just "port 53 bound" isn't enough — systemd-resolved
+    # listens on 127.0.0.53 and would satisfy a bind test. Retries for ~15s
+    # to tolerate cold-start upstream / blocklist warmup.
+    assert_dns_works() {
+        local n=0
+        while [ $n -lt 15 ]; do
+            if dig @127.0.0.1 -p 53 example.com +short +timeout=2 +tries=1 2>/dev/null \
+                 | grep -qE '^[0-9]+\.[0-9]+\.[0-9]+\.[0-9]+$'; then
+                pass "DNS resolves on :53 (A record returned)"
+                return
+            fi
+            sleep 1
+            n=$((n + 1))
+        done
+        fail "DNS did not return an A record on :53 within 15s"
+    }
+
+    # TLS handshake: cert must validate against numa's CA when connecting
+    # to a .numa SNI. Catches port-not-bound, wrong cert, missing CA file.
+    assert_tls_handshake() {
+        local port=$1 sni=${2:-numa.numa} out
+        if out=$(openssl s_client -connect "127.0.0.1:${port}" \
+                    -servername "$sni" \
+                    -CAfile /var/lib/numa/ca.pem \
+                    -verify_return_error </dev/null 2>&1); then
+            if echo "$out" | grep -q 'Verify return code: 0 (ok)'; then
+                pass "TLS handshake + cert chain verified on :${port}"
+            else
+                fail "TLS handshake on :${port} did not report 'Verify return code: 0'"
+            fi
+        else
+            fail "openssl s_client failed connecting to :${port}"
+        fi
+    }
+
+    assert_http_responds() {
+        local code
+        code=$(curl -s -o /dev/null -w "%{http_code}" --max-time 3 http://127.0.0.1/ || echo 000)
+        if [ "$code" != "000" ]; then
+            pass "HTTP responds on :80 (status $code)"
+        else
+            fail "HTTP :80 connection failed"
+        fi
+    }
+
+    assert_api_healthy() {
+        if curl -sf --max-time 3 http://127.0.0.1:5380/health >/dev/null; then
+            pass "API /health OK on :5380"
+        else
+            fail "API /health failed on :5380"
+        fi
+    }
+
+    ca_fingerprint() {
+        openssl x509 -in /var/lib/numa/ca.pem -noout -fingerprint -sha256 2>/dev/null \
+            | sed 's/.*=//'
+    }
+
+    wait_active() {
+        local n=0
+        while [ $n -lt 20 ]; do
+            systemctl is-active --quiet numa && return 0
+            sleep 0.5
+            n=$((n + 1))
+        done
+        fail "service did not become active within 10s"
+        systemctl status numa --no-pager -l 2>&1 | head -20 || true
+        return 1
+    }
+
+    # ---- Scenario A ----
+    printf "\n=== Scenario A: fresh install — every advertised port is functional ===\n"
+    reset_state
+    "$NUMA" install >/tmp/installA.log 2>&1 || { fail "install failed"; tail -20 /tmp/installA.log; }
+    wait_active || true
+    assert_nonroot
+    assert_dns_works
+    assert_tls_handshake 853
+    assert_tls_handshake 443
+    assert_http_responds
+    assert_api_healthy
+
+    # ---- Scenario B ----
+    # Pre-drop installs left /var/lib/numa as a plain root-owned tree.
+    # Flattening the current DynamicUser layout back into that shape
+    # simulates the upgrade path without needing an actual old binary.
+    printf "\n=== Scenario B: CA fingerprint survives upgrade from pre-drop layout ===\n"
+    fp_before=$(ca_fingerprint)
+    if [ -z "$fp_before" ]; then
+        fail "could not read initial CA fingerprint (skipping scenario B)"
+    else
+        echo "  CA fingerprint before: $fp_before"
+        "$NUMA" uninstall >/dev/null 2>&1 || true
+        tmp=$(mktemp -d)
+        cp -a /var/lib/private/numa/. "$tmp"/ 2>/dev/null || true
+        rm -rf /var/lib/numa /var/lib/private/numa
+        mv "$tmp" /var/lib/numa
+        chown -R root:root /var/lib/numa
+        chmod 755 /var/lib/numa
+        [ -f /var/lib/numa/ca.pem ] || fail "ca.pem missing from seeded legacy tree"
+
+        "$NUMA" install >/tmp/installB.log 2>&1 || { fail "upgrade install failed"; tail -20 /tmp/installB.log; }
+        wait_active || true
+        assert_nonroot
+        fp_after=$(ca_fingerprint)
+        if [ -z "$fp_after" ]; then
+            fail "could not read CA fingerprint after upgrade"
+        elif [ "$fp_before" = "$fp_after" ]; then
+            pass "CA fingerprint preserved across upgrade"
+        else
+            fail "CA fingerprint changed: before=$fp_before after=$fp_after"
+        fi
+        assert_dns_works
+    fi
+
+    # ---- Scenario C ----
+    printf "\n=== Scenario C: install from unreachable source stages binary to /usr/local/bin ===\n"
+    reset_state
+    mkdir -p /home/builder
+    chmod 700 /home/builder
+    cp "$NUMA" /home/builder/numa
+    chmod 755 /home/builder/numa
+    /home/builder/numa install >/tmp/installC.log 2>&1 || { fail "install failed"; tail -20 /tmp/installC.log; }
+    wait_active || true
+    if [ -x /usr/local/bin/numa ]; then
+        pass "binary staged to /usr/local/bin/numa"
+    else
+        fail "/usr/local/bin/numa missing after install from 0700 source"
+    fi
+    exec_line=$(grep '^ExecStart=' /etc/systemd/system/numa.service 2>/dev/null || echo "ExecStart=<unit missing>")
+    if echo "$exec_line" | grep -q '/usr/local/bin/numa'; then
+        pass "unit ExecStart points to staged path"
+    else
+        fail "unit ExecStart wrong: $exec_line"
+    fi
+    assert_nonroot
+    assert_dns_works
+
+    reset_state
+    rm -rf /home/builder
+    echo
+    if [ "$FAIL" -eq 0 ]; then
+        printf "${GREEN}── all scenarios passed ──${RESET}\n"
+        exit 0
+    else
+        printf "${RED}── some scenarios failed ──${RESET}\n"
+        exit 1
+    fi
+fi
+
+# ============================================================
+# Mode A: host-side bootstrap
+# ============================================================
+set -e
+cd "$(dirname "$0")/../.."
+
+IMAGE=numa-install-systemd:local
+CONTAINER="numa-install-systemd-$$"
+trap 'docker rm -f "$CONTAINER" >/dev/null 2>&1 || true' EXIT
+
+echo "── building systemd-in-container image (cached after first run) ──"
+docker build --quiet -t "$IMAGE" -f - . <<'DOCKERFILE' >/dev/null
+FROM ubuntu:24.04
+ENV DEBIAN_FRONTEND=noninteractive
+RUN apt-get update -qq && apt-get install -y -qq \
+      systemd systemd-sysv systemd-resolved \
+      ca-certificates curl build-essential \
+      pkg-config libssl-dev cmake make perl \
+      dnsutils iproute2 openssl \
+    && rm -rf /var/lib/apt/lists/* \
+    && for u in dev-hugepages.mount sys-fs-fuse-connections.mount \
+                systemd-logind.service getty.target console-getty.service; do \
+         systemctl mask $u; \
+       done
+STOPSIGNAL SIGRTMIN+3
+CMD ["/lib/systemd/systemd"]
+DOCKERFILE
+
+echo "── starting systemd container ──"
+docker run -d --name "$CONTAINER" \
+    --privileged --cgroupns=host \
+    --tmpfs /run --tmpfs /run/lock --tmpfs /tmp:exec \
+    -v "$PWD:/src:ro" \
+    -v numa-install-systemd-cargo:/root/.cargo \
+    -v numa-install-systemd-work:/work \
+    "$IMAGE" >/dev/null
+
+# Wait for systemd to be up
+for _ in $(seq 1 30); do
+    state=$(docker exec "$CONTAINER" systemctl is-system-running 2>&1 || true)
+    case "$state" in running|degraded) break ;; esac
+    sleep 0.5
+done
+
+echo "── copying source into /work (writable) ──"
+docker exec "$CONTAINER" bash -c '
+mkdir -p /work
+tar -C /src --exclude=./target --exclude=./.git --exclude=./.claude -cf - . | tar -C /work -xf -
+'
+
+echo "── rustup + cargo build --release --locked ──"
+docker exec "$CONTAINER" bash -c '
+set -e
+if ! command -v cargo &>/dev/null; then
+    curl -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --quiet
+fi
+. "$HOME/.cargo/env"
+cd /work
+cargo build --release --locked 2>&1 | tail -5
+'
+
+echo "── running scenarios ──"
+docker exec -e NUMA_INSIDE=1 "$CONTAINER" bash /src/tests/docker/install-systemd.sh
--- a/tests/integration.sh
+++ b/tests/integration.sh
@@ -1,7 +1,10 @@
 #!/usr/bin/env bash
 # Integration test suite for Numa
 # Runs a test instance on port 5354, validates all features, exits with status.
-# Usage: ./tests/integration.sh [release|debug]
+# Usage:
+#   ./tests/integration.sh [release|debug]       # all suites
+#   SUITES=7 ./tests/integration.sh              # only Suite 7
+#   SUITES=1,3,7 ./tests/integration.sh          # Suites 1, 3, and 7

 set -euo pipefail

@@ -14,6 +17,14 @@ LOG="/tmp/numa-integration-test.log"
 PASSED=0
 FAILED=0

+# Suite filter: empty runs all; comma list runs a subset.
+SUITES="${SUITES:-}"
+should_run_suite() {
+    [ -z "$SUITES" ] && return 0
+    case ",$SUITES," in *",$1,"*) return 0;; esac
+    return 1
+}
+
 # Colors
 GREEN="\033[32m"
 RED="\033[31m"
@@ -166,6 +177,7 @@ CONF
 }

 # ---- Suite 1: Recursive mode + DNSSEC ----
+if should_run_suite 1; then
 echo ""
 echo "╔══════════════════════════════════════════╗"
 echo "║  Suite 1: Recursive + DNSSEC + Blocking  ║"
@@ -234,7 +246,10 @@ kill "$NUMA_PID" 2>/dev/null || true
 wait "$NUMA_PID" 2>/dev/null || true
 sleep 1

+fi  # end Suite 1
+
 # ---- Suite 2: Forward mode (backward compat) ----
+if should_run_suite 2; then
 echo ""
 echo "╔══════════════════════════════════════════╗"
 echo "║  Suite 2: Forward (DoH) + Blocking       ║"
@@ -261,7 +276,10 @@ enabled = true
 enabled = false
 "

+fi  # end Suite 2
+
 # ---- Suite 3: Forward UDP (plain, no DoH) ----
+if should_run_suite 3; then
 echo ""
 echo "╔══════════════════════════════════════════╗"
 echo "║  Suite 3: Forward (UDP) + No Blocking    ║"
@@ -307,7 +325,10 @@ kill "$NUMA_PID" 2>/dev/null || true
 wait "$NUMA_PID" 2>/dev/null || true
 sleep 1

+fi  # end Suite 3
+
 # ---- Suite 4: Local zones + Overrides API ----
+if should_run_suite 4; then
 echo ""
 echo "╔══════════════════════════════════════════╗"
 echo "║  Suite 4: Local Zones + Overrides API    ║"
@@ -416,7 +437,10 @@ kill "$NUMA_PID" 2>/dev/null || true
 wait "$NUMA_PID" 2>/dev/null || true
 sleep 1

+fi  # end Suite 4
+
 # ---- Suite 5: DNS-over-TLS (RFC 7858) ----
+if should_run_suite 5; then
 echo ""
 echo "╔══════════════════════════════════════════╗"
 echo "║  Suite 5: DNS-over-TLS (RFC 7858)        ║"
@@ -538,7 +562,10 @@ CONF
 fi
 sleep 1

+fi  # end Suite 5
+
 # ---- Suite 6: Proxy + DoT coexistence ----
+if should_run_suite 6; then
 echo ""
 echo "╔══════════════════════════════════════════╗"
 echo "║  Suite 6: Proxy + DoT Coexistence        ║"
@@ -698,6 +725,332 @@ CONF
    rm -rf "$NUMA_DATA"
 fi

+fi  # end Suite 6
+
+# ---- Suite 7: filter_aaaa (IPv4-only networks) ----
+if should_run_suite 7; then
+echo ""
+echo "╔══════════════════════════════════════════╗"
+echo "║  Suite 7: filter_aaaa                    ║"
+echo "╚══════════════════════════════════════════╝"
+
+# Config A — filter on, with a local AAAA zone to prove local data bypass.
+cat > "$CONFIG" << 'CONF'
+[server]
+bind_addr = "127.0.0.1:5354"
+api_port = 5381
+filter_aaaa = true
+
+[upstream]
+mode = "forward"
+address = "9.9.9.9"
+port = 53
+
+[cache]
+max_entries = 10000
+
+[blocking]
+enabled = false
+
+[proxy]
+enabled = false
+
+[[zones]]
+domain = "v6.test"
+record_type = "AAAA"
+value = "2001:db8::1"
+ttl = 60
+CONF
+
+RUST_LOG=info "$BINARY" "$CONFIG" > "$LOG" 2>&1 &
+NUMA_PID=$!
+sleep 3
+
+DIG="dig @127.0.0.1 -p $PORT +time=5 +tries=1"
+
+echo ""
+echo "=== filter_aaaa = true ==="
+
+# A queries must be untouched.
+check "A record resolves under filter_aaaa" \
+    "." \
+    "$($DIG google.com A +short | head -1)"
+
+# AAAA must be NOERROR (NODATA), not NXDOMAIN, not SERVFAIL.
+check "AAAA returns NOERROR (not NXDOMAIN)" \
+    "status: NOERROR" \
+    "$($DIG google.com AAAA 2>&1 | grep 'status:')"
+
+check "AAAA returns zero answers (NODATA shape)" \
+    "ANSWER: 0" \
+    "$($DIG google.com AAAA 2>&1 | grep -oE 'ANSWER: [0-9]+' | head -1)"
+
+# Local zone AAAA must survive the filter (PR claim: local data bypasses).
+check "Local [[zones]] AAAA bypasses filter" \
+    "2001:db8::1" \
+    "$($DIG v6.test AAAA +short)"
+
+# HTTPS RR: ipv6hint (SvcParamKey 6) must be stripped. Query as `type65`
+# because dig 9.10.6 (macOS) misparses `HTTPS` as a domain name; `type65`
+# works on both 9.10.6 and 9.18. Assert on the raw rdata hex (RFC 3597
+# generic format), since dig 9.10.6 doesn't pretty-print HTTPS params.
+# cloudflare.com's ipv6hint values sit under the 2606:4700 prefix —
+# checking that `26064700` is absent from the rdata hex is a precise,
+# upstream-stable signal that the TLV was stripped.
+HTTPS_OUT=$($DIG cloudflare.com type65 2>&1)
+if echo "$HTTPS_OUT" | grep -qE "cloudflare\.com\..*IN[[:space:]]+TYPE65"; then
+    HTTPS_HEX=$(echo "$HTTPS_OUT" | grep -A5 "IN[[:space:]]*TYPE65" | tr -d " \t\n")
+    if echo "$HTTPS_HEX" | grep -qi "26064700"; then
+        check "HTTPS ipv6hint stripped (2606:4700 absent from rdata)" "absent" "present"
+    else
+        check "HTTPS ipv6hint stripped (2606:4700 absent from rdata)" "absent" "absent"
+    fi
+else
+    # Upstream didn't return an HTTPS record — skip rather than false-pass.
+    printf "  ${DIM}~ HTTPS ipv6hint stripped (skipped: no HTTPS RR returned by upstream)${RESET}\n"
+fi
+
+kill "$NUMA_PID" 2>/dev/null || true
+wait "$NUMA_PID" 2>/dev/null || true
+sleep 1
+
+# Config B — filter off. Regression guard: prove AAAA answers come back
+# when the flag isn't set, so a network failure in Config A can't silently
+# pass as "filter working".
+cat > "$CONFIG" << 'CONF'
+[server]
+bind_addr = "127.0.0.1:5354"
+api_port = 5381
+
+[upstream]
+mode = "forward"
+address = "9.9.9.9"
+port = 53
+
+[cache]
+max_entries = 10000
+
+[blocking]
+enabled = false
+
+[proxy]
+enabled = false
+CONF
+
+RUST_LOG=info "$BINARY" "$CONFIG" > "$LOG" 2>&1 &
+NUMA_PID=$!
+sleep 3
+
+echo ""
+echo "=== filter_aaaa unset (regression guard) ==="
+
+check "AAAA returns real answers with filter off" \
+    ":" \
+    "$($DIG google.com AAAA +short | head -1)"
+
+kill "$NUMA_PID" 2>/dev/null || true
+wait "$NUMA_PID" 2>/dev/null || true
+sleep 1
+
+fi  # end Suite 7
+
+# ---- Suite 8: ODoH (Oblivious DoH via public relay + target) ----
+# Exercises the full client pipeline: /.well-known/odohconfigs fetch,
+# HPKE seal/unseal, URL-query target routing (RFC 9230 §5), dashboard
+# QueryPath::Odoh counter. Depends on the public ecosystem being up —
+# the probe-odoh-ecosystem.sh script guards against flaky runs.
+if should_run_suite 8; then
+echo ""
+echo "╔══════════════════════════════════════════╗"
+echo "║  Suite 8: ODoH (Anonymous DNS)           ║"
+echo "╚══════════════════════════════════════════╝"
+
+run_test_suite "ODoH via edgecompute.app relay → Cloudflare target" "
+[server]
+bind_addr = \"127.0.0.1:5354\"
+api_port = 5381
+
+[upstream]
+mode = \"odoh\"
+relay = \"https://odoh-relay.edgecompute.app/proxy\"
+target = \"https://odoh.cloudflare-dns.com/dns-query\"
+
+[cache]
+max_entries = 10000
+min_ttl = 60
+max_ttl = 86400
+
+[blocking]
+enabled = false
+
+[proxy]
+enabled = false
+"
+
+# Re-start briefly to assert ODoH-specific observability: the odoh counter
+# has to tick above zero after a query, and the stats label has to reflect
+# the oblivious path. These guard against silent regressions in the
+# QueryPath::Odoh tagging and the /stats serialisation.
+RUST_LOG=info "$BINARY" "$CONFIG" > "$LOG" 2>&1 &
+NUMA_PID=$!
+for _ in $(seq 1 30); do
+    curl -sf "http://127.0.0.1:$API_PORT/health" >/dev/null 2>&1 && break
+    sleep 0.1
+done
+
+$DIG example.com A +short > /dev/null 2>&1 || true
+sleep 1
+
+STATS=$(curl -sf http://127.0.0.1:$API_PORT/stats 2>/dev/null)
+# upstream_transport.odoh lives inside the upstream_transport object.
+ODOH_COUNT=$(echo "$STATS" | grep -o '"upstream_transport":{[^}]*}' \
+    | grep -o '"odoh":[0-9]*' | cut -d: -f2)
+check "upstream_transport.odoh > 0 after a query" "[1-9]" "${ODOH_COUNT:-0}"
+
+check "Upstream label advertises odoh://" \
+    "odoh://" \
+    "$(echo "$STATS" | grep -o '"upstream":"[^"]*"')"
+
+check "Stats mode field is 'odoh'" \
+    '"mode":"odoh"' \
+    "$(echo "$STATS" | grep -o '"mode":"odoh"')"
+
+# Strict-mode failure path: a clearly-unreachable relay must produce
+# SERVFAIL without silent downgrade. We hijack the config to point at
+# an .invalid host so we don't rely on external uptime.
+kill "$NUMA_PID" 2>/dev/null || true
+wait "$NUMA_PID" 2>/dev/null || true
+sleep 1
+
+cat > "$CONFIG" << 'CONF'
+[server]
+bind_addr = "127.0.0.1:5354"
+api_port = 5381
+
+[upstream]
+mode = "odoh"
+relay = "https://relay.invalid/proxy"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+strict = true
+
+[cache]
+max_entries = 10000
+
+[blocking]
+enabled = false
+
+[proxy]
+enabled = false
+CONF
+
+RUST_LOG=info "$BINARY" "$CONFIG" > "$LOG" 2>&1 &
+NUMA_PID=$!
+for _ in $(seq 1 30); do
+    curl -sf "http://127.0.0.1:$API_PORT/health" >/dev/null 2>&1 && break
+    sleep 0.1
+done
+
+check "Strict-mode relay outage returns SERVFAIL" \
+    "SERVFAIL" \
+    "$($DIG example.com A 2>&1 | grep 'status:')"
+
+kill "$NUMA_PID" 2>/dev/null || true
+wait "$NUMA_PID" 2>/dev/null || true
+sleep 1
+
+# Negative: relay and target on the same host must be rejected at startup.
+cat > "$CONFIG" << 'CONF'
+[server]
+bind_addr = "127.0.0.1:5354"
+api_port = 5381
+
+[upstream]
+mode = "odoh"
+relay = "https://odoh.cloudflare-dns.com/proxy"
+target = "https://odoh.cloudflare-dns.com/dns-query"
+CONF
+
+STARTUP_OUT=$("$BINARY" "$CONFIG" 2>&1 || true)
+check "Same-host relay+target rejected at startup" \
+    "same host" \
+    "$STARTUP_OUT"
+
+fi  # end Suite 8
+
+# ---- Suite 9: Numa's own ODoH relay (--relay-mode) ----
+# Exercises `numa relay PORT` as a forwarding proxy to a real ODoH target.
+# Validates the RFC 9230 §5 relay behaviour: URL-query routing, content-type
+# gating, body-size cap, and /health observability.
+if should_run_suite 9; then
+echo ""
+echo "╔══════════════════════════════════════════╗"
+echo "║  Suite 9: Numa ODoH Relay (own)          ║"
+echo "╚══════════════════════════════════════════╝"
+
+RELAY_PORT=18443
+"$BINARY" relay $RELAY_PORT > "$LOG" 2>&1 &
+NUMA_PID=$!
+for _ in $(seq 1 30); do
+    curl -sf "http://127.0.0.1:$RELAY_PORT/health" >/dev/null 2>&1 && break
+    sleep 0.1
+done
+
+echo ""
+echo "=== Relay Endpoints ==="
+
+check "Health endpoint returns ok" \
+    "ok" \
+    "$(curl -sf http://127.0.0.1:$RELAY_PORT/health | head -1)"
+
+# Happy path: forwards arbitrary body to Cloudflare's ODoH target. The
+# target will reject the garbage envelope with HTTP 400 — which is exactly
+# what proves our relay faithfully forwarded (otherwise we'd see our own
+# 4xx from the relay itself).
+HAPPY_STATUS=$(curl -sS -o /dev/null -w "%{http_code}" -X POST \
+    -H "Content-Type: application/oblivious-dns-message" \
+    --data-binary "garbage-forwarded-end-to-end" \
+    "http://127.0.0.1:$RELAY_PORT/relay?targethost=odoh.cloudflare-dns.com&targetpath=/dns-query")
+check "Relay forwards to target (target rejects garbage → 400)" \
+    "400" \
+    "$HAPPY_STATUS"
+
+echo ""
+echo "=== Guards ==="
+
+check "Missing content-type → 415" \
+    "415" \
+    "$(curl -sS -o /dev/null -w '%{http_code}' -X POST --data-binary 'x' \
+        'http://127.0.0.1:'$RELAY_PORT'/relay?targethost=odoh.cloudflare-dns.com&targetpath=/dns-query')"
+
+check "Oversized body (>4 KiB) → 413" \
+    "413" \
+    "$(head -c 5000 /dev/urandom | curl -sS -o /dev/null -w '%{http_code}' -X POST \
+        -H 'Content-Type: application/oblivious-dns-message' --data-binary @- \
+        'http://127.0.0.1:'$RELAY_PORT'/relay?targethost=odoh.cloudflare-dns.com&targetpath=/dns-query')"
+
+check "Invalid targethost (no dot) → 400" \
+    "400" \
+    "$(curl -sS -o /dev/null -w '%{http_code}' -X POST \
+        -H 'Content-Type: application/oblivious-dns-message' --data-binary 'x' \
+        'http://127.0.0.1:'$RELAY_PORT'/relay?targethost=invalid&targetpath=/dns-query')"
+
+echo ""
+echo "=== Counters ==="
+
+HEALTH=$(curl -sf "http://127.0.0.1:$RELAY_PORT/health")
+check "Relay counted at least one forwarded_ok" \
+    "[1-9]" \
+    "$(echo "$HEALTH" | grep 'forwarded_ok' | awk '{print $2}')"
+check "Relay counted at least one rejected_bad_request" \
+    "[1-9]" \
+    "$(echo "$HEALTH" | grep 'rejected_bad_request' | awk '{print $2}')"
+
+kill "$NUMA_PID" 2>/dev/null || true
+wait "$NUMA_PID" 2>/dev/null || true
+sleep 1
+
+fi  # end Suite 9
+
 # Summary
 echo ""
 TOTAL=$((PASSED + FAILED))
--- a/tests/probe-odoh-ecosystem.sh
+++ b/tests/probe-odoh-ecosystem.sh
@@ -0,0 +1,101 @@
+#!/usr/bin/env bash
+# Probe the public ODoH ecosystem.
+#
+# Source of truth: DNSCrypt's curated list at
+#   https://github.com/DNSCrypt/dnscrypt-resolvers/tree/master/v3
+#   - v3/odoh-servers.md  (ODoH targets)
+#   - v3/odoh-relays.md   (ODoH relays)
+#
+# As of commit 2025-09-16 ("odohrelay-crypto-sx seems to be the only ODoH
+# relay left"), the full public ecosystem is 4 targets + 1 relay. Re-run this
+# script against the upstream list before making any "only N public relays"
+# claim publicly.
+#
+# Usage: ./tests/probe-odoh-ecosystem.sh
+
+set -uo pipefail
+
+GREEN="\033[32m"
+RED="\033[31m"
+YELLOW="\033[33m"
+DIM="\033[90m"
+RESET="\033[0m"
+
+UP=0
+DOWN=0
+
+probe_target() {
+    local name="$1"
+    local host="$2"
+    local url="https://${host}/.well-known/odohconfigs"
+    local start=$(date +%s%N)
+    local headers
+    headers=$(curl -sS -o /tmp/odoh-probe-body -D - --max-time 5 -A "numa-odoh-probe/0.1" "$url" 2>&1) || {
+        DOWN=$((DOWN + 1))
+        printf "  ${RED}✗${RESET} %-25s ${DIM}unreachable${RESET}\n" "$name"
+        return
+    }
+    local elapsed_ms=$((($(date +%s%N) - start) / 1000000))
+    local status
+    status=$(echo "$headers" | head -1 | awk '{print $2}')
+    local ctype
+    ctype=$(echo "$headers" | grep -i '^content-type:' | head -1 | tr -d '\r')
+    local size
+    size=$(stat -f%z /tmp/odoh-probe-body 2>/dev/null || stat -c%s /tmp/odoh-probe-body 2>/dev/null || echo 0)
+
+    if [[ "$status" == "200" ]] && [[ "$size" -gt 0 ]]; then
+        UP=$((UP + 1))
+        printf "  ${GREEN}✓${RESET} %-25s ${DIM}%4dms  %s bytes  %s${RESET}\n" "$name" "$elapsed_ms" "$size" "$ctype"
+    else
+        DOWN=$((DOWN + 1))
+        printf "  ${RED}✗${RESET} %-25s ${DIM}status=%s size=%s${RESET}\n" "$name" "$status" "$size"
+    fi
+    rm -f /tmp/odoh-probe-body
+}
+
+probe_relay() {
+    # Relays don't expose /.well-known/odohconfigs — we just verify TLS reachability
+    # and that the endpoint responds to a malformed POST with an HTTP error
+    # (indicating the relay path exists). A real ODoH validation requires HPKE.
+    local name="$1"
+    local url="$2"
+    local start=$(date +%s%N)
+    local status
+    status=$(curl -sS -o /dev/null -w "%{http_code}" --max-time 5 -A "numa-odoh-probe/0.1" \
+        -X POST -H "Content-Type: application/oblivious-dns-message" \
+        --data-binary "" "$url" 2>&1) || {
+        DOWN=$((DOWN + 1))
+        printf "  ${RED}✗${RESET} %-25s ${DIM}unreachable${RESET}\n" "$name"
+        return
+    }
+    local elapsed_ms=$((($(date +%s%N) - start) / 1000000))
+    # Any 2xx or 4xx means the endpoint is live (TLS works, HTTP responded).
+    # 5xx or 000 (curl failure) means broken.
+    if [[ "$status" =~ ^[24] ]]; then
+        UP=$((UP + 1))
+        printf "  ${GREEN}✓${RESET} %-25s ${DIM}%4dms  status=%s (endpoint live)${RESET}\n" "$name" "$elapsed_ms" "$status"
+    else
+        DOWN=$((DOWN + 1))
+        printf "  ${RED}✗${RESET} %-25s ${DIM}status=%s${RESET}\n" "$name" "$status"
+    fi
+}
+
+echo "ODoH targets:"
+probe_target "Cloudflare"   "odoh.cloudflare-dns.com"
+probe_target "crypto.sx"    "odoh.crypto.sx"
+probe_target "Snowstorm"    "dope.snowstorm.love"
+probe_target "Tiarap"       "doh.tiarap.org"
+
+echo
+echo "ODoH relays:"
+probe_relay  "Frank Denis (Fastly)" "https://odoh-relay.edgecompute.app/proxy"
+
+echo
+TOTAL=$((UP + DOWN))
+if [[ "$DOWN" -eq 0 ]]; then
+    printf "${GREEN}All %d endpoints up${RESET}\n" "$TOTAL"
+    exit 0
+else
+    printf "${YELLOW}%d/%d up, %d down${RESET}\n" "$UP" "$TOTAL" "$DOWN"
+    exit 1
+fi
Author	SHA1	Message	Date
Razvan Dimescu	07c321f749	chore(release): bump to v0.14.0 Headline: ODoH (RFC 9230) — client + self-hosted relay. Set mode = "odoh" in [upstream] to seal queries before they leave the machine; run `numa relay` to add to the public ODoH ecosystem.	2026-04-20 17:07:31 +03:00
Razvan Dimescu	12a06a1410	Merge pull request #121 from razvandimescu/feat/odoh feat(odoh): ship ODoH client + self-hosted relay (RFC 9230)	2026-04-20 16:26:54 +03:00
Razvan Dimescu	eb5ea3b645	refactor(odoh): deduplicate post-audit findings - Hoist ODOH_CONTENT_TYPE to a single pub(crate) constant in odoh.rs; relay.rs imports it instead of declaring its own. - Generalize dashboard encryptionPct(data, encryptedKeys, allKeys) so both Inbound Wire and Outbound Wire panels share the same math instead of drifting independently. - Extract RelayState::new() and build_app() helpers in relay.rs so the test spawn_relay() and production run() wire the same router + body-limit layer. Prevents future middleware from landing in one path but not the other. All 344 lib tests pass; no behavior change.	2026-04-20 16:03:34 +03:00
Razvan Dimescu	be60f6ccbc	chore(packaging): docker-compose + Caddyfile for ODoH relay deploy Two-container deploy: Caddy terminates TLS (auto-provisions Let's Encrypt via ACME) and reverse-proxies to a Numa relay on an internal Docker network. The relay never reads sealed payloads; Caddy's access log is discarded so per-request observability doesn't defeat the oblivious property. Validated against Hetzner CX22 + DNS at odoh-relay.numa.rs: - TLS-ALPN-01 challenge succeeded on first attempt - /health returned the relay's counter block - End-to-end ODoH client → relay → Cloudflare works Operators only need to: set a DNS A record, edit Caddyfile's hostname, docker compose up -d. README walks through the steps and the DNSCrypt v3/odoh-relays.md submission to claim a public listing.	2026-04-20 15:44:29 +03:00
Razvan Dimescu	a3cc64c94f	feat(odoh): relay bind-address CLI arg + dashboard Outbound Wire panel - `numa relay [PORT] [BIND]` accepts an optional bind address (defaults to 127.0.0.1, matching the Caddy reverse-proxy deployment shape). Required for Docker, where the relay needs 0.0.0.0 inside the container so Caddy can reach it across the bridge network. - Dashboard now surfaces the upstream_transport dimension as an "Outbound Wire" panel alongside the existing "Inbound Wire" (renamed from "Transport" for directional clarity). Sub-headers — "apps → numa" / "numa → internet" — make the threat-model split obvious without jargon. Bars: UDP/DoH/DoT/ODoH, headline "X% encrypted outbound". The PR description's promise that "the dashboard answers how much of my DNS traffic left in cleartext honestly" is now true.	2026-04-20 15:44:20 +03:00
Razvan Dimescu	cf128c19af	feat(odoh): bootstrap-IP overrides + zero hedge for ODoH (post-deploy fixes) Two issues surfaced from running mode = "odoh" against the live Hetzner relay as system DNS: 1. Bootstrap deadlock. The reqwest HTTPS client resolves the relay and target hostnames via system DNS. When numa is itself the system resolver, the ODoH client loops trying to resolve through itself. Adds optional `relay_ip` and `target_ip` to `[upstream]`, plumbed into reqwest's `resolve()` so the HTTPS client bypasses system DNS for those two hostnames. TLS still validates against the URL hostname, so a stale IP fails loudly rather than silently MITM'ing. 2. 2x relay load. Default `hedge_ms = 10` triggers a duplicate in-flight query for every request. Useful for UDP/DoH/DoT (rescues tail latency cheaply); wasteful for ODoH (doubles HPKE seal/unseal, doubles sealed-byte footprint a passive observer can correlate, no latency win — relay hop dominates either way). Force-zero in oblivious mode regardless of configured hedge_ms. Validated end-to-end against odoh-relay.numa.rs → Cloudflare: 3 digs produced 3 forwarded_ok on the relay (was 6 before the hedge fix), upstream_transport.odoh ticks correctly.	2026-04-20 15:44:09 +03:00
Razvan Dimescu	241c40553b	feat(odoh): ship ODoH client + self-hosted relay (RFC 9230) Client (mode = "odoh"): URL-query target routing per RFC 9230 §5, /.well-known/odohconfigs TTL cache with 60s backoff on failure, HPKE seal/open via odoh-rs, strict-mode default that SERVFAILs on relay failure instead of silently downgrading. Host-equality config validation rejects same-operator relay/target pairs. Relay (`numa relay [PORT]`): axum server with /relay + /health. SSRF-hardened hostname validator (RFC 1035 ASCII + dot + dash), 4 KiB body cap at the axum layer, 5s full-transaction timeout, and static 502 on target failure (reqwest internals logged, not leaked). Aggregate counters only — no per-request logs. Observability: new `UpstreamTransport { Udp, Doh, Dot, Odoh }` orthogonal to `QueryPath`, so /stats can tally wire protocols symmetrically. Recursive mode records `Some(Udp)` for honest "bytes egressing in cleartext" accounting. Tests: Suite 8 exercises the client end-to-end via Frank Denis's public relay + Cloudflare target; Suite 9 exercises `numa relay` forwarding + guards against Cloudflare as the real far end. Full probe script at tests/probe-odoh-ecosystem.sh verifies the entire public ODoH ecosystem (4 targets + 1 relay per DNSCrypt's curated list — confirms deploying Numa's relay doubles global supply).	2026-04-20 12:34:04 +03:00
Razvan Dimescu	f6cfb3ce1b	Merge pull request #120 from razvandimescu/feat/named-record-types feat(question): name SVCB/LOC/NAPTR record types in logs	2026-04-19 08:08:54 +03:00
Razvan Dimescu	5725f94ff3	refactor(question): collapse QueryType impls behind define_qtypes! macro Adding a record type used to require 5 edits across the file (enum variant, to_num, from_num, as_str, parse_str). The macro takes a single (variant, num, str) tuple per type and generates the enum plus all four methods. UNKNOWN(u16) stays hand-coded since it carries data and can't sit in the table. src/question.rs: 156 lines -> 92 lines, no behavior change.	2026-04-19 08:01:18 +03:00
Razvan Dimescu	24610ae3fe	feat(question): add SVCB, LOC, NAPTR variants to QueryType Logs were printing UNKNOWN(64), UNKNOWN(29), UNKNOWN(35) for SVCB, LOC, and NAPTR — three RR types that have been registered for years and show up in the wild (notably SVCB via RFC 9462 DDR clients querying _dns.resolver.arpa). Adds the variants and replaces the SVCB_QTYPE u16 const introduced in #119 with QueryType::SVCB.to_num(), matching the HTTPS path. Closes #114.	2026-04-19 07:49:35 +03:00
Razvan Dimescu	6bc02982f0	Merge pull request #119 from razvandimescu/feat/filter-aaaa feat(resolver): filter_aaaa for IPv4-only networks	2026-04-19 07:31:27 +03:00
Razvan Dimescu	f9e996ae78	fmt: drop redundant comments per house style	2026-04-19 06:53:47 +03:00
Razvan Dimescu	5e85b147b9	feat(resolver): apply ipv6hint strip to SVCB (type 64) too HTTPS (65) and SVCB (64) share the RDATA wire format, so the existing parser already handles both — only the call site was HTTPS-only. Widen the qtype check and extend the existing pipeline test with a second query for SVCB.	2026-04-19 06:52:30 +03:00
Razvan Dimescu	d6bb9a0f01	fmt: rustfmt vec literal wrapping + signature collapse	2026-04-19 06:24:54 +03:00
Razvan Dimescu	61ea2e510d	refactor: dedupe HTTPS_TYPE, record-walk, and test rdata builder - Drop `const HTTPS_TYPE: u16 = 65;` in favor of `QueryType::HTTPS.to_num()` at the single call site — avoids a fresh magic number alongside the existing enum mapping in question.rs. - Add `DnsPacket::for_each_record_mut` so `strip_https_ipv6_hints` stops hand-rolling the answers/authorities/resources walk; future section rewrites go through the same helper. - Promote the SVCB test-rdata builder from `svcb::tests` to module scope as `pub(crate) #[cfg(test)] fn build_rdata`, and reuse it in the two pipeline tests in ctx.rs — kills ~20 lines of byte-fiddling and keeps one RDATA-construction code path.	2026-04-19 05:58:47 +03:00
Razvan Dimescu	22dd3cd222	fix(resolver): skip ipv6hint strip for DO-bit clients Modifying HTTPS rdata invalidates any accompanying RRSIG, so a DNSSEC- validating downstream would reject the response as Bogus. Gate the strip on !client_do, matching the existing DNSSEC-records strip. Adds a regression test that catches the gate being removed: builds a query with EDNS DO=1, asserts the HTTPS rdata round-trips untouched.	2026-04-19 05:52:37 +03:00
Razvan Dimescu	8014ebac9e	test(integration): add Suite 7 for filter_aaaa + SUITES env filter Suite 7 exercises the full pipeline end-to-end: A resolves, AAAA returns NODATA, local [[zones]] AAAA bypasses the filter, and HTTPS ipv6hint is stripped from a real cloudflare.com response. A second config run with the flag unset guards against network-failure false-positives. SUITES=N (comma list) runs a subset, e.g. `SUITES=7 bash tests/integration.sh` skips suites 1-6 for fast iteration.	2026-04-19 05:52:29 +03:00
Razvan Dimescu	70400187d0	Merge pull request #118 from razvandimescu/feat/linux-drop-privileges feat(linux): run systemd service as unprivileged numa user	2026-04-18 22:04:53 +03:00
Razvan Dimescu	fb41a6f8b5	test(linux): systemd service install verification Three scenarios CI cannot run: every advertised port is functional (DNS resolves, TLS chain validates against numa's CA, HTTP/API respond), CA fingerprint survives upgrade from pre-drop layout, binary staging fallback from a 0700 source dir. Self-bootstraps a privileged systemd-as-PID1 container — no dependency on long-lived test containers. MainPID user assertion retries until comm=numa to avoid a race where systemctl reports active while MainPID still points at a transitional process.	2026-04-18 22:00:54 +03:00
Razvan Dimescu	b02b607fb9	ci(linux): assert numa daemon does not run as root Locks in the invariant this branch establishes: a regression that reverts to User=root would otherwise ship green.	2026-04-18 20:07:24 +03:00
Razvan Dimescu	be98a02e49	feat(resolver): filter_aaaa for IPv4-only networks (#112 ) When enabled, AAAA queries short-circuit to NODATA (NOERROR + empty answer) so Happy Eyeballs clients don't stall waiting on a v6 address they can't use. Also strips `ipv6hint` SvcParam from HTTPS/SVCB answers (RFC 9460) so Chrome ≥103, Firefox, and Safari don't bypass the AAAA filter via the HTTPS record path. Local data is preserved: overrides, zones, the .numa proxy, and the blocklist sinkhole keep whatever v6 addresses they configure — the filter only kicks in on the cache/forward/recursive path. NODATA is correct per RFC 2308 here; NXDOMAIN would incorrectly imply the name doesn't exist for A queries either. Off by default. Opt in via `filter_aaaa = true` under `[server]`.	2026-04-18 19:52:06 +03:00
Razvan Dimescu	763131478f	fmt: rustfmt format! macro split	2026-04-18 12:15:44 +03:00
Razvan Dimescu	067195f2ab	fix(linux): atomic binary copy + restart instead of start on re-install Re-install failed with ETXTBSY (Text file busy) because std::fs::copy can't overwrite a binary that's currently being executed by the running service. Switch to copy-then-rename: write the new binary to /usr/local/bin/numa.new, then rename over /usr/local/bin/numa. Rename swaps the path while the running process keeps the old inode alive, so DNS keeps serving from the previous binary until restart. Bump systemctl start to restart so the new binary actually loads on re-install (start is a no-op when the unit is already active, which would silently leave the old binary running). Locally verified the full CI sequence: install → curl → reinstall → curl → uninstall → curl-fails. All three assertions pass.	2026-04-18 12:12:11 +03:00
Razvan Dimescu	e19505aa95	fix(linux): narrow replace_exe_path cfg to macos after Linux inlined the substitution Linux install_service_linux now does the {{exe_path}} substitution inline because it uses the (potentially copied) binary path returned by install_service_binary_linux, not current_exe(). The shared replace_exe_path helper is dead on Linux — clippy -D warnings caught it. Narrow the function to macos and split the placeholder test: keep the "both templates contain {{exe_path}}" assertion as a cross-platform test (catches placeholder removal on either file), keep the substitution test gated to macos where the function lives.	2026-04-18 11:57:54 +03:00
Razvan Dimescu	3970a9f45c	fix(linux): copy binary to /usr/local/bin when source path isn't world-traversable DynamicUser=yes' transient account can only traverse world-x directories. The CI binary at /home/runner/work/numa/numa/target/release/numa fails exec with EACCES because /home/runner is mode 0700; same applies to a build under /home/<user>/, ~/.cargo/bin, or any private $HOME tree. install_service_binary_linux now walks the binary's path. If every ancestor grants world-execute (Linuxbrew /home/linuxbrew is 0755, /usr/local/bin is fine, install.sh layout works), keep the source path so brew/distro upgrades propagate in place. Otherwise copy to /usr/local/bin/numa and reference that in the unit. Locally verified both branches in an Ubuntu 24.04 systemd container: - CI-like /home/runner (0700) → copies + service binds 5380 - Brew-like /home/linuxbrew (0755) → keeps source path + service binds 5380	2026-04-18 11:51:32 +03:00
Razvan Dimescu	7b9db9e889	fix(linux): drop ProtectHome=true — blocks exec when binary lives under /home Integration-linux journalctl showed status=203/EXEC: systemd couldn't exec /home/runner/work/numa/numa/target/release/numa because ProtectHome=yes makes /home invisible to the sandboxed process. My local Docker test passed because the binary was at /workspace, not /home. DynamicUser=yes already implies ProtectHome=read-only, which preserves exec access to binaries living under /home (cargo install, source builds, CI) while blocking writes to user $HOMEs. Keep that default rather than over-restricting. Follow-up worth tracking: install_service_linux could copy the binary to /usr/local/bin/numa the way Windows does at windows_service_exe_path, making the unit's ExecStart independent of where `numa install` was invoked from — then we could set ProtectHome=yes again.	2026-04-18 08:54:34 +03:00
Razvan Dimescu	dfeca53e21	ci: dump journalctl + systemctl status on integration-linux failure	2026-04-18 08:48:53 +03:00
Razvan Dimescu	4f6159d961	refactor(linux): switch to DynamicUser=yes, drop install-time user creation AUR installs never call `numa install` — PKGBUILD drops the unit straight into /usr/lib/systemd/system and the user runs `systemctl enable numa`. With User=numa the Rust installer's useradd code never fires there, breaking Arch out of the box. DynamicUser=yes sidesteps packaging entirely — systemd allocates a transient UID per start and remaps StateDirectory ownership (including legacy root-owned trees) automatically. Works on any modern systemd. Drops the ensure_numa_user_linux/chown helpers plus NUMA_USER; the unit file alone now captures the privilege-drop story.	2026-04-18 08:20:07 +03:00
Razvan Dimescu	41aea1dd12	fix(linux): drop risky sandbox directives that break Rust network daemons Integration test failed with exit 7 on curl to /health after a successful install — service started but never listened. The likely culprits are MemoryDenyWriteExecute (breaks jemalloc/some crypto), SystemCallFilter ~@privileged @resources (blocks setrlimit and friends tokio may use), and RestrictNamespaces/LockPersonality (occasional foot-guns). Pull them and keep a conservative hardening set that's well-tested with Rust network services: no-new-privs, protect-system/home, private tmp and devices, protect-kernel-*, restrict-realtime/suid/address-families. Layer the aggressive bits back in follow-up PRs once tested individually.	2026-04-18 08:10:04 +03:00
Razvan Dimescu	695a8b963c	feat(linux): run systemd service as unprivileged numa user - numa.service: User=numa + CAP_NET_BIND_SERVICE + sandboxing block (ProtectSystem=strict, PrivateTmp, seccomp @system-service, etc) - install_service_linux: create numa system user + chown data_dir before first start so TLS-cert generation and state writes land on a numa-owned tree Runtime verified root-free on Linux — network_watch_loop only reads /etc/resolv.conf; all system-DNS mutation stays in the installer, which continues to run as root via sudo.	2026-04-18 07:56:59 +03:00
Razvan Dimescu	34e2182ae4	Merge pull request #104 from razvandimescu/feat/forwarding-array-upstream feat: accept array of upstreams in [[forwarding]]	2026-04-17 23:25:04 +03:00
Razvan Dimescu	5f77af55e9	fix(forward): track SRTT for DoT upstreams, not just UDP The SRTT ordering + failure penalty path was UDP-only, so a DoT primary in a forwarding-rule pool was never deprioritized on failure and all DoT entries tied at INITIAL_SRTT_MS in the sort key. With [[forwarding]] now accepting arrays of upstreams, DoT pools are a first-class case and need the same healthiest-first behavior the default pool gets for UDP. - Add Upstream::tracked_ip() → Some(ip) for Udp/Dot, None for Doh (DoH has no stable IP — reqwest pools connections by hostname). - Rewire the three SRTT call sites in forward_with_failover_raw. - Hoist srtt.read() out of the candidate-scoring loop — one lock per query instead of N (matters now that pools commonly have N>1). - Drop unused #[derive(Debug)] on UpstreamPool and ForwardingRule. - Regression tests: udp_failure_records_in_srtt + dot_failure_records_in_srtt.	2026-04-17 03:39:21 +03:00
Razvan Dimescu	ab6cda0c91	Merge branch 'main' into feat/forwarding-array-upstream Resolves src/main.rs conflict: serve loop was extracted into src/serve.rs on main (PR #107). Ported the forwarding-rule log change to serve.rs — fwd.upstream is now Vec<String>, logged with join(", ").	2026-04-17 03:14:09 +03:00
Razvan Dimescu	4afc56a052	Merge main into feat/forwarding-array-upstream Resolves conflict in src/ctx.rs — both sides added independent tokio tests (forwarding fail-over on this branch, default-pool upstream path on main from #103). Keep both.	2026-04-15 21:28:04 +03:00
Razvan Dimescu	fef43635d6	fix(ci): rustfmt import order and gate Upstream import for Windows	2026-04-15 04:11:27 +03:00
Razvan Dimescu	9a0d586b13	feat: accept array of upstreams in [[forwarding]] Mirrors `[upstream] address` — `upstream` accepts string or array of strings, builds an `UpstreamPool` and routes queries through `forward_with_failover_raw` so SRTT ordering and failover apply to matched `[[forwarding]]` rules the same way they do for the default pool. Single-string rules keep their current behavior (one-element pool, equivalent single-upstream path). Empty array errors at config load. Addresses item 1 of issue #102. Plan: docs/102_item1.md.	2026-04-15 04:03:38 +03:00