Segfault node v0.11.14 related to TLS

[matejkramny]

Running https://github.com/CastawayLabs/dproxy.js.. Occasionally, it quits with segmentation fault..

It is caught using segfault-handler.

Stack trace:

PID 20 received SIGSEGV for address: 0x11
/srv/app/node_modules/segfault-handler/build/Release/segfault-handler.node(+0x1185)[0x7fc0d1619185]
/lib/x86_64-linux-gnu/libpthread.so.0(+0x10340)[0x7fc0d1dfe340]
/usr/local/bin/node(uv_write+0x1c)[0xb8a84c]
/usr/local/bin/node(_ZN4node12TLSCallbacks6EncOutEv+0x23a)[0xb621ea]
/usr/local/bin/node(_ZN4node12TLSCallbacks21OnClientHelloParseEndEPv+0x30)[0xb64340]
/usr/local/bin/node(_ZN4node6crypto7SSLWrapINS_12TLSCallbacksEE9EndParserERKN2v820FunctionCallbackInfoINS4_5ValueEEE+0x5d)[0xb49b1d]
/usr/local/bin/node(_ZN2v88internal25FunctionCallbackArguments4CallEPFvRKNS_20FunctionCallbackInfoINS_5ValueEEEE+0x92)[0x781002]
/usr/local/bin/node[0x79e5f8]
[0x1fb117c0634e]

If i'm reading this correctly, a client isn't sending proper TLS request and causes an internal node failure.

I'm determined to find the root cause of this. If this is genuinely not my fault (dproxy project), then other projects using TLS and SNI may be endangered.

Happened for v0.11.13 and v.0.11.14

/cc @tjfontaine as it seems related to TLS

[migounette]

@matejkramny please provide more details, for instance distro (Debian, ...), and how to produce the issue. A simple test case help for faster debugging and for get the time to look at it.
eg. npm start
My 2 cents

[misterdjules]

@matejkramny as @migounette suggested, could you please provide us with more information on how to reproduce this issue somewhat reliably? Ideally, we would come up with a test case that doesn't have any external dependency (a Node.js program that doesn't use any external module). However, anything that documents how to reproduce the issue with clear steps would already help.

[matejkramny]

I'll find time for this by end of the week.

It is really difficult to reproduce however, and seems to happen * randomly *.

The server seemed to crash every time i used the iOS simulator. Then quitting the simulator and relaunching it fixed the server. Next time i'll try to capture the network :p

[misterdjules]

One way to gather more information regarding what's going on would be to:

• Try to reproduce the problem either with a debug build or with a 32 bits binary. The problem with optimized x64 builds (like the one you're using) is that it can be difficult to examine a core dump due to the calling conventions used.
• Let the operating system generate a core dump when the program receives SIGSEGV. You'll have to make sure that the core file limits are set to unlimited with ulimit -c unlimited.
• Examine this core dump to understand what input causes this segfault. You can use gdb to do that.

If you need help doing this, please ping me (jgi) on #libuv.

Thanks!

[matejkramny]

@misterdjules Thanks. is that on freenode?

[misterdjules]

Yes, on Freenode.

[matejkramny]

Right. I compiled node.js on the system as 32 bit (i think). I couldn't get the server to crash (no matter what.). It crashed this morning for some reason.

Here's the log.. doesn't look very different from the 64 bit one. ulimit is set to unlimited

/home/proxy/node_modules/segfault-handler/build/Debug/segfault-handler.node(+0x2ca9)[0x7fa00e2e5ca9]
/lib/x86_64-linux-gnu/libpthread.so.0(+0xf030)[0x7fa00ea99030]
node(uv_write2+0x27)[0xbb83b7]
node(_ZN4node12TLSCallbacks6EncOutEv+0x239)[0xb92c49]
node(_ZN4node12TLSCallbacks21OnClientHelloParseEndEPv+0x48)[0xb93a28]
node(_ZN4node6crypto7SSLWrapINS_12TLSCallbacksEE9EndParserERKN2v820FunctionCallbackInfoINS4_5ValueEEE+0x5d)[0xb7c12d]
node(_ZN2v88internal25FunctionCallbackArguments4CallEPFvRKNS_20FunctionCallbackInfoINS_5ValueEEEE+0x9b)[0x7a675b]
node[0x7c577d]
[0x2047f850634e]

Is it of any use? I thought of having gdb set up, but i've never done that with node.js before (apart from using node-inspector which i suspect uses that).

[migounette]

@misterdjules Well, I found a problem a week ago but I didn't succeed to reproduce, it was the source of a crash in intensive TLS traffic.

The enc_out_ may be null... my 2 cents

void TLSCallbacks::EncOutCb(uv_write_t* req, int status) {
TLSCallbacks* callbacks = static_cast<TLSCallbacks*>(req->data);

// Handle error
if (status) {
// Ignore errors after shutdown
if (callbacks->shutdown_)
return;

// Notify about error
callbacks->InvokeQueued(status);
return;

}

// Commit

// FIX <<------- START
if (callbacks->enc_out_) {
NodeBIO::FromBIO(callbacks->enc_out_)->Read(NULL, callbacks->write_size_);
} // FIX <<--- END
// Try writing more data
callbacks->write_size_ = 0;
callbacks->EncOut();
}

[matejkramny]

@migounette can you make a PR?

[indutny]

May I ask you to give a try to nodejs/node#1702 ?

[matejkramny]

Ok will do.

Thanks @indutny!