V8 Memory Corruption and Stack Overflow (fixed in Node v0.8.28 and v0.10.30)
The Node.js Project
A memory corruption vulnerability, which results in a denial-of-service, was
identified in the versions of V8 that ship with Node.js 0.8 and 0.10. In
certain circumstances, a particularly deep recursive workload that may trigger
a GC and receive an interrupt may overflow the stack and result in a
segmentation fault. For instance, if your work load involves successive
JSON.parse
calls and the parsed objects are significantly deep, you may
experience the process aborting while parsing.
This issue was identified by Tom Steele of ^Lift Security and Fedor Indunty, Node.js Core Team member worked closely with the V8 team to find our resolution.
The V8 issue is described here https://codereview.chromium.org/339883002
It has landed in the Node repository here: https://github.com/joyent/node/commit/530af9cb8e700e7596b3ec812bad123c9fa06356
And has been released in the following versions:
The Fix
The backport of the fix for Node.js is
diff --git a/deps/v8/src/isolate.h b/deps/v8/src/isolate.h
index b90191d..2769ca7 100644
--- a/deps/v8/src/isolate.h
+++ b/deps/v8/src/isolate.h
@@ -1392,14 +1392,9 @@ class StackLimitCheck BASE_EMBEDDED {
public:
explicit StackLimitCheck(Isolate* isolate) : isolate_(isolate) { }
- bool HasOverflowed() const {
+ inline bool HasOverflowed() const {
StackGuard* stack_guard = isolate_->stack_guard();
- // Stack has overflowed in C++ code only if stack pointer exceeds the C++
- // stack guard and the limits are not set to interrupt values.
- // TODO(214): Stack overflows are ignored if a interrupt is pending. This
- // code should probably always use the initial C++ limit.
- return (reinterpret_cast<uintptr_t>(this) < stack_guard->climit()) &&
- stack_guard->IsStackOverflow();
+ return reinterpret_cast<uintptr_t>(this) < stack_guard->real_climit();
}
private:
Isolate* isolate_;
Remediation
The best course of action is to patch or upgrade Node.js.
Mitigation
To mitigate against deep JSON parsing you can limit the size of the string you
parse against, or ban clients who trigger a RangeError
for parsing JSON.
There is no specific maximum size of a JSON string, though keeping the max to the size of your known message bodies is suggested. If your message bodies cannot be over 20K, there's no reason to accept 1MB bodies.
For web frameworks that do automatic JSON parsing, you may need to configure the routes that accept JSON payloads to have a maximum body size.
- expressjs and krakenjs used with the body-parser plugin accepts a
limit
parameter in your JSON config - Hapi.js has
payload.maxBytes
https://github.com/spumko/hapi/blob/master/docs/Reference.md - restify bundled
bodyParser
accepts amaxBodySize