RPC API Security (JSON-RPC / XML-RPC)
Purpose
Exploit RPC-style APIs: enumerate callable methods, invoke privileged or internal methods with lower-privilege sessions, abuse batch semantics, and exploit ID/param handling — applying the same differential evidence contracts as REST skills.
Preconditions
- An identified JSON-RPC (POST with
{"jsonrpc":"2.0","method":...}) or XML-RPC (<methodCall>) endpoint - At least one valid session for differential testing
- The target's error behavior catalogued (invalid method, invalid params, authz failure)
Attack Surface Signals
- JSON bodies with
jsonrpc,method,params,idkeys Accept: application/jsonendpoints rejecting normal REST calls- Known frameworks: Ethereum nodes (
eth_*), Solr, supervisor, MantisBT, WordPress XML-RPC, CMS RPC layers /rpc,/jsonrpc,/xmlrpc.php,/api/jsonrpcpaths
Methodology
Step 1: Error-Triplet Baseline
curl -sk -X POST "https://TARGET/rpc" -H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","method":"no_such_method","params":[],"id":1}'
curl -sk -X POST "https://TARGET/rpc" -H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","method":"target_method","params":[],"id":1}'
Compare: unknown-method error vs real-method response (authz error, param error, or execution). Different error strings = method EXISTS and is callable to some depth. Catalog these before concluding anything.
Step 2: Method Enumeration
for m in user.list user.get admin.list system.listMethods rpc.discover \
getRoles updateUser deleteUser debug.status internal.exec importData; do
resp=$(curl -sk -X POST "https://TARGET/rpc" -H "Content-Type: application/json" \
-d "{\"jsonrpc\":\"2.0\",\"method\":\"$m\",\"params\":{},\"id\":1}")
echo "$m => $resp" | head -c 200; echo
done
# Framework-specific discovery
# system.listMethods (XML-RPC), rpc.discover (JSON-RPC OpenRPC), eth_* on chain nodes
Treat readable responses as CANDIDATES — a finding requires a concrete unauthorized outcome.
Step 3: Privileged Method Invocation
1. Invoke the privileged method legitimately as the admin/high session → record the effect (baseline)
2. Same exact call with the lower-privilege session → if it executes identically: BFLA finding
3. Same call anonymous → additional finding surface
Verify the privileged effect with a second read (role changed, user created, data returned).
Step 4: Param and Type Abuse
- Positional vs named params:
params:[victim_id]vsparams:{"id":victim_id}— some stacks authorize one shape - Extra params:
params:[own_id, {"role":"admin"}] - Type confusion: string IDs vs numeric; null params; nested objects where scalars expected
- Parameter injection:
params:{"id": {"$ne": null}}(NoSQL operator reaching the RPC layer)
Step 5: Batch Authorization Inconsistencies
curl -sk -X POST "https://TARGET/rpc" -H "Content-Type: application/json" -d '[
{"jsonrpc":"2.0","method":"user.get","params":[123],"id":1},
{"jsonrpc":"2.0","method":"user.get","params":[456],"id":2},
{"jsonrpc":"2.0","method":"admin.purgeUsers","params":[],"id":3}
]'
Some implementations authorize the batch container, not each method — mixed-role batches can smuggle privileged calls. Per-item responses must be checked individually: one executed admin.purgeUsers result = finding.
Step 6: ID and Notification Handling
idreflection: does the response echo attacker-controlledidvalues into logs/views (injection into ID fields)?- Notifications (
idomitted): fire-and-forget calls that may skip authz/logging idcollisions in batches: response misattribution between calls- Error responses leaking stack traces, file paths, versions (misconfig chain)
Step 7: XML-RPC-Specific
<methodCall><methodName>system.listMethods</methodName><params></params></methodCall>
<!-- WordPress/MantisBT: system.multicall as auth-bypass batch primitive
(password brute force via multicall, historically); pingback.ping as
SSRF primitive (chains into api-ssrf) -->
Evidence Contract
An RPC finding requires:
- Baseline: the method's legitimate behavior (privileged execution observed or documented)
- Manipulation: the lower-privilege/anonymous/param-abused invocation
- Differential: the privileged effect actually happened for the unauthorized caller (verified by a second read), or privileged data returned in the result
NOT evidence: a different error string, method existence, a 200 with {"result":null}, batch accepted but no per-item effects
Common Misses
- Error-string differences never catalogued, so method existence is missed
- Batch per-item results never inspected individually
- Named/positional param shapes not both tried
- Notifications (no
id) never tested for authz bypass - Framework-specific hidden methods (system., internal., debug.*) not probed
- XML-RPC system.multicall/pingback primitives forgotten
False Positives / Non-Findings
- Unknown-method errors (routing working)
result: nullon authorized methods without state change- Batch rejected wholesale (per-method authz working)
- Enumerated methods that exist but correctly enforce authorization
Xalgorix Tool Strategy
http_request/curl for precise RPC envelopes; Python for batch construction and per-item parsingoob_callbackwhere RPC methods accept URLs (pingback SSRF chains)authz_matrixfor multi-role method differentials
Stopping Rule
Every enumerated method x every session tier x positional/named/extra/type-abused params x batch and notification variants.
Handoff
Report the method, session tier, param manipulation, and the verified unauthorized effect. URL-accepting methods chain into api-ssrf; chain-node RPC (eth_*) into blockchain-specific skills.
