privacy: fix a real leak - global IPv6 addresses were uploaded verbatim
Setting out to build the machine-readable schema, the first step was checking whether the anonymizer covers the fields the schema declares sensitive. It did not, and five identifying values were going out at the `balanced` level: networks[].link.addresses[].addr the device's own global IPv6 address networks[].link.routes[].gateway the ISP allocation networks[].link.dns.servers[] the configured resolver private_dns_hostname an internal hostname search_domains[] the internal domain The settings screen describes that level as pseudonymizing addresses. Root cause: classification keyed on field names, and the schema's actual names were never added to the table. Every existing test passed, because each checked a field somebody had remembered to write a case for - an unfalsifiable design for a privacy control. So beyond adding the names, classification now falls back to the *value* when the name is unknown: anything shaped like an IPv4/IPv6 address or a MAC is treated as one. Hostnames deliberately are not inferred by shape, since train.udp_updown is indistinguishable from a domain and mangling a test type would corrupt the document to protect nothing. LeakTest is the guard, and is written to fail for fields nobody thought of: it plants identifying values wherever one can occur and asserts none survive. It also pins that RFC1918 addresses stay readable, so it cannot pass by over-redacting. Route prefixes and :: needed care - 0.0.0.0/0 must stay itself or a routing table becomes unreadable for no privacy gain. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Fable 5
parent
e7afc2210f
commit
172afb421d
@@ -848,3 +848,48 @@ itself to table rows, so the prose can keep explaining which codes were retired
|
||||
|
||||
Six tests: uniqueness, declared-vs-listed, prefix↔category agreement, naming convention, a
|
||||
word-order-anagram check (the shape the duplication actually took), and the document agreement.
|
||||
|
||||
### A real privacy leak, found by starting on the machine-readable schema (2026-08-01)
|
||||
The intent was `measurement.schema.json` (§8's promised companion). The first step — checking
|
||||
whether the anonymizer actually covers the fields the schema declares as sensitive — found that it
|
||||
did not, so that became the work.
|
||||
|
||||
**At the `balanced` level, five identifying values were being uploaded verbatim:**
|
||||
|
||||
| value | field | why it matters |
|
||||
|---|---|---|
|
||||
| `2001:…::150` | `networks[].link.addresses[].addr` | the device's own global IPv6 address — a strong, geolocatable device identifier |
|
||||
| `2a02:…::1` | `networks[].link.routes[].gateway` | identifies the ISP allocation |
|
||||
| `203.0.113.77` | `networks[].link.dns.servers[]` | the configured resolver |
|
||||
| `nas.example.lan` | `private_dns_hostname` | an internal hostname |
|
||||
| `example.lan` | `search_domains[]` | the internal domain |
|
||||
|
||||
The settings screen describes that level as pseudonymizing addresses. It was not.
|
||||
|
||||
**Root cause:** classification keyed on field *names*, and the schema's actual names (`addr`,
|
||||
`gateway`, `dst`, `servers`, `search_domains`, `private_dns_hostname`) had never been added to the
|
||||
table. Not a subtle bug — just an unfalsifiable design. The existing tests all passed, because each
|
||||
one checked a field somebody had remembered to write a case for.
|
||||
|
||||
**Two fixes, one of them structural:**
|
||||
1. The missing names were added.
|
||||
2. More importantly, a **shape-based backstop**: when a field name is unrecognised, the *value* is
|
||||
inspected, and anything shaped like an IPv4/IPv6 address or a MAC is treated as one. A name
|
||||
table can only protect fields someone thought of, which is precisely the wrong property for a
|
||||
privacy control. Hostnames are deliberately *not* inferred by shape — `train.udp_updown` is
|
||||
indistinguishable from a domain, and mangling a test type would corrupt the document to protect
|
||||
nothing.
|
||||
|
||||
`LeakTest` is the new guard and is written to fail for fields nobody has considered: it plants
|
||||
identifying values wherever one can actually occur and asserts none survive, rather than checking
|
||||
a list of known cases. It also pins that RFC1918 addresses still come through readable, so the
|
||||
test cannot pass by over-redacting everything.
|
||||
|
||||
Route prefixes and the unspecified address needed care in the transform: `0.0.0.0/0` and `::/0`
|
||||
must stay themselves, or a routing table becomes unreadable for no privacy gain.
|
||||
|
||||
**Still outstanding:** `measurement.schema.json` itself. Worth noting what this episode implies for
|
||||
it — much of a document's payload lives in `evidence`/`metrics`/`params`, which are per-test-type
|
||||
`JsonObject` by design and therefore *outside* any schema. A schema-driven anonymizer would have
|
||||
less coverage there than the name-plus-shape one now does, so the schema should be built for
|
||||
validation and external tooling, not as a replacement for the classifier.
|
||||
|
||||
@@ -1,237 +1,256 @@
|
||||
// SPDX-FileCopyrightText: 2026 Echolot contributors
|
||||
// SPDX-License-Identifier: GPL-3.0-or-later
|
||||
|
||||
// Package privacy implements the anonymization contract of measurement-schema.md §8.
|
||||
//
|
||||
// The threat model is specific. An engineer running their own server wants the full document —
|
||||
// SSIDs and MACs are what make a run useful a week later. Someone measuring against a stranger's
|
||||
// server wants the numbers to survive and the identifiers not to. So this is a *transform*, not a
|
||||
// filter: the output is still a valid measurement document with the same tests, metrics and
|
||||
// findings; only the identifying scalars change, and consistently, so "same SSID as last run" is
|
||||
// still answerable from pseudonyms alone.
|
||||
//
|
||||
// Two properties are load-bearing and are what the tests pin:
|
||||
// - Consistency within a document: one input value always maps to one pseudonym, so
|
||||
// correlations inside a run survive.
|
||||
// - No consistency *across* documents unless the user asks for it: the salt is per-run by
|
||||
// default, so pseudonyms cannot be used to track a device between uploads. A stable salt is
|
||||
// opt-in (`Salt.stable`) for people diffing their own history on their own server.
|
||||
package app.echo_lot.privacy
|
||||
|
||||
import kotlinx.serialization.json.*
|
||||
import java.security.MessageDigest
|
||||
import java.util.Locale
|
||||
|
||||
/** How much to strip. Ordered: FULL < BALANCED < STRICT. Wire values match the server's. */
|
||||
enum class PrivacyLevel(val wire: String) {
|
||||
/** Nothing removed. The right choice for your own server. */
|
||||
FULL("full"),
|
||||
|
||||
/**
|
||||
* Identifiers pseudonymized, neighbour inventory dropped. Topology and timing survive:
|
||||
* you can still see that the gateway is a MikroTik at a /24 boundary with 3 % loss, but not
|
||||
* which MikroTik, on which SSID, next to whose Chromecast.
|
||||
*/
|
||||
BALANCED("balanced"),
|
||||
|
||||
/**
|
||||
* Numbers only: tests keep their metrics and status, evidence is dropped, findings keep their
|
||||
* codes and severities but lose descriptions (which quote real names). What is left cannot
|
||||
* identify a network, and is still enough for aggregate "how common is this fault" work.
|
||||
*/
|
||||
STRICT("strict");
|
||||
|
||||
companion object {
|
||||
fun fromWire(s: String?): PrivacyLevel =
|
||||
entries.firstOrNull { it.wire == s?.lowercase(Locale.ROOT) } ?: FULL
|
||||
|
||||
/** The stricter of two levels — used to honour a server's minimum. */
|
||||
fun max(a: PrivacyLevel, b: PrivacyLevel): PrivacyLevel = if (a.ordinal >= b.ordinal) a else b
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* The pseudonymization salt. Per-run by default: a fresh random salt means the same SSID uploaded
|
||||
* twice yields two different pseudonyms, so an upload endpoint cannot link runs to a device.
|
||||
* A stable salt trades that away for cross-run diffing and is only appropriate on a server you
|
||||
* own — the app makes that an explicit choice, not a default.
|
||||
*/
|
||||
class Salt private constructor(internal val bytes: ByteArray, val stable: Boolean) {
|
||||
companion object {
|
||||
fun perRun(random: ByteArray): Salt = Salt(random.copyOf(), stable = false)
|
||||
fun stable(secret: ByteArray): Salt = Salt(secret.copyOf(), stable = true)
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Transforms a measurement document to [level].
|
||||
*
|
||||
* Field classification is by JSON key name, because the schema names things consistently
|
||||
* (`ssid`, `bssid`, `mac`, `ip4`, `ip6`, `fqdn`, …) and a name-driven pass is auditable by
|
||||
* reading one table. Anything unrecognized is treated as identifying when it is a string inside
|
||||
* a known-sensitive container, and left alone otherwise — see [Classification].
|
||||
*/
|
||||
class Anonymizer(private val level: PrivacyLevel, private val salt: Salt) {
|
||||
|
||||
private val cache = HashMap<String, String>()
|
||||
|
||||
fun anonymize(doc: JsonObject): JsonObject {
|
||||
if (level == PrivacyLevel.FULL) return stamp(doc)
|
||||
val walked = walkObject(doc, path = emptyList())
|
||||
val out = if (level == PrivacyLevel.STRICT) strip(walked) else walked
|
||||
return stamp(out)
|
||||
}
|
||||
|
||||
/** Records what was done, so a reader of the archived/uploaded document is never guessing. */
|
||||
private fun stamp(doc: JsonObject): JsonObject {
|
||||
val run = doc["run"]?.jsonObject ?: return doc
|
||||
val privacy = buildJsonObject {
|
||||
put("anonymization", level.wire)
|
||||
put("salt", if (salt.stable) "stable" else "per_run")
|
||||
}
|
||||
return JsonObject(doc + ("run" to JsonObject(run + ("privacy" to privacy))))
|
||||
}
|
||||
|
||||
// ---- the tree walk -------------------------------------------------------------------
|
||||
|
||||
private fun walkObject(obj: JsonObject, path: List<String>): JsonObject = buildJsonObject {
|
||||
for ((k, v) in obj) {
|
||||
val childPath = path + k
|
||||
when {
|
||||
Classification.dropAtBalanced(childPath) -> Unit // omit entirely
|
||||
else -> put(k, walk(k, v, childPath))
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
private fun walk(key: String, v: JsonElement, path: List<String>): JsonElement = when (v) {
|
||||
is JsonObject -> walkObject(v, path)
|
||||
is JsonArray -> JsonArray(v.map { walk(key, it, path) })
|
||||
is JsonPrimitive ->
|
||||
if (v.isString) transform(Classification.typeOf(key, path), v.content).let(::JsonPrimitive)
|
||||
else v
|
||||
}
|
||||
|
||||
private fun transform(type: LogicalType?, value: String): String = when (type) {
|
||||
null -> value
|
||||
LogicalType.SSID -> pseudo("ssid", value) { "net-" + it.take(6) }
|
||||
LogicalType.MAC, LogicalType.BSSID -> macPreservingOui(value)
|
||||
LogicalType.IP4 -> ip4(value)
|
||||
LogicalType.IP6 -> ip6(value)
|
||||
LogicalType.FQDN -> fqdn(value)
|
||||
LogicalType.OPAQUE_ID -> "redacted"
|
||||
LogicalType.FREETEXT -> "[removed: may contain identifying text]"
|
||||
}
|
||||
|
||||
// ---- per-type transforms -------------------------------------------------------------
|
||||
|
||||
/**
|
||||
* Keeps the OUI (which vendor) and pseudonymizes the NIC part (which unit). Vendor is the
|
||||
* diagnostically valuable half — "the RA comes from a MikroTik" survives, "…from THAT
|
||||
* MikroTik" does not.
|
||||
*/
|
||||
private fun macPreservingOui(value: String): String {
|
||||
val sep = if (value.contains('-')) '-' else ':'
|
||||
val parts = value.split(sep)
|
||||
if (parts.size != 6 || parts.any { it.length != 2 }) return pseudo("mac", value) { "mac-" + it.take(8) }
|
||||
val nic = pseudo("mac", value) { it }
|
||||
return (parts.take(3) + listOf(nic.substring(0, 2), nic.substring(2, 4), nic.substring(4, 6)))
|
||||
.joinToString(sep.toString())
|
||||
.lowercase(Locale.ROOT)
|
||||
}
|
||||
|
||||
/**
|
||||
* Prefix-preserving within the same class, with reserved ranges kept verbatim: RFC1918 and
|
||||
* CGNAT addresses say something about the topology and nothing about the person, and a run
|
||||
* where 192.168.1.1 became a random public address would be actively misleading to read.
|
||||
* Public addresses keep only their /16 so the network is still locatable at ISP granularity.
|
||||
*/
|
||||
private fun ip4(value: String): String {
|
||||
val o = value.split(".")
|
||||
if (o.size != 4 || o.any { it.toIntOrNull() == null }) return value
|
||||
val n = o.map { it.toInt() }
|
||||
val reserved = n[0] == 10 ||
|
||||
(n[0] == 172 && n[1] in 16..31) ||
|
||||
(n[0] == 192 && n[1] == 168) ||
|
||||
(n[0] == 169 && n[1] == 254) ||
|
||||
(n[0] == 100 && n[1] in 64..127) ||
|
||||
n[0] == 127 || n[0] == 0 || n[0] >= 224
|
||||
if (reserved) return value
|
||||
val h = pseudo("ip4", value) { it }
|
||||
return "${n[0]}.${n[1]}.${h.substring(0, 2).toInt(16)}.${h.substring(2, 4).toInt(16)}"
|
||||
}
|
||||
|
||||
/**
|
||||
* IPv6 keeps the scope and the first 32 bits (so 2001:db8:… still reads as global unicast in
|
||||
* the same allocation) and pseudonymizes the rest — the interface identifier is the part that
|
||||
* is a device fingerprint, especially with EUI-64.
|
||||
*/
|
||||
private fun ip6(value: String): String {
|
||||
val v = value.lowercase(Locale.ROOT)
|
||||
if (v == "::1" || v == "::" || v.startsWith("fe80:") || v.startsWith("ff")) return v
|
||||
val groups = v.substringBefore('%').split(":")
|
||||
if (groups.size < 3) return v
|
||||
val h = pseudo("ip6", value) { it }
|
||||
return "${groups[0]}:${groups[1]}:${h.substring(0, 4)}:${h.substring(4, 8)}::${h.substring(8, 12)}"
|
||||
}
|
||||
|
||||
/**
|
||||
* Per-label pseudonyms with the public suffix kept, so "it resolved somewhere under .local"
|
||||
* or "…under example.com" survives without naming the host. The suffix list is deliberately
|
||||
* short: guessing wrong keeps *more* pseudonymized, never less.
|
||||
*/
|
||||
private fun fqdn(value: String): String {
|
||||
if (value.isEmpty()) return value
|
||||
val trailing = value.endsWith(".")
|
||||
val labels = value.trimEnd('.').split(".")
|
||||
if (labels.size == 1) return pseudo("fqdn", value) { "host-" + it.take(6) }
|
||||
val keep = if (labels.last() in publicSuffixes) 1 else 0
|
||||
val head = labels.dropLast(keep).map { l -> pseudo("label", l) { "l-" + it.take(6) } }
|
||||
return (head + labels.takeLast(keep)).joinToString(".") + if (trailing) "." else ""
|
||||
}
|
||||
|
||||
// ---- STRICT ---------------------------------------------------------------------------
|
||||
|
||||
/**
|
||||
* STRICT keeps the shape of the document and the numbers, and nothing that quotes the
|
||||
* network back. Evidence goes (trains carry addresses and hostnames), finding prose goes
|
||||
* (it interpolates real names), networks go entirely.
|
||||
*/
|
||||
private fun strip(doc: JsonObject): JsonObject = buildJsonObject {
|
||||
for ((k, v) in doc) {
|
||||
when (k) {
|
||||
"networks", "server_sessions" -> Unit
|
||||
"tests" -> put(k, JsonArray((v as? JsonArray ?: JsonArray(emptyList())).map { t ->
|
||||
val o = t.jsonObject
|
||||
JsonObject(o.filterKeys { it != "evidence" && it != "params" })
|
||||
}))
|
||||
"findings" -> put(k, JsonArray((v as? JsonArray ?: JsonArray(emptyList())).map { f ->
|
||||
val o = f.jsonObject
|
||||
JsonObject(o.filterKeys { it != "description" && it != "title" && it != "evidence_refs" })
|
||||
}))
|
||||
"run" -> put(k, JsonObject(v.jsonObject.filterKeys { it != "notes" }))
|
||||
else -> put(k, v)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// ---- pseudonym machinery ---------------------------------------------------------------
|
||||
|
||||
/** Deterministic per (domain, value, salt); memoized so one value maps to one pseudonym. */
|
||||
private fun pseudo(domain: String, value: String, shape: (String) -> String): String =
|
||||
cache.getOrPut("$domain | ||||