src: fix freeing unintialized pointer bug in ParseSoaReply

[AasthaGupta]

ares_expand_name doesn't guarantee that pointer variable is initialized if
return code is ARES_EBADNAME or ARES_ENOMEM. But current usage of the function
in the codebase thinks otherwise.

There seems to be an assumption that pointer is always initialized even though
it is a local variable and we create a unique pointer soon after calling
ares_expand_name. This could potentially crash
the program with an invalid free pointer.

I was able to crash it by poisoning the memory and some manual hooks.

node(9118,0x111471dc0) malloc: *** error for object 0x7b: pointer being freed was not allocated
node(9118,0x111471dc0) malloc: *** set a breakpoint in malloc_error_break to debug
[1] 9118 abort node ../temp.js

By moving the unique_ptr after checking the return code we can fix the problem.
As the underlying function guarantees that pointer is initialized when the
status is ARES_SUCCESS.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[nodejs-github-bot]

Review requested:

• @nodejs/net

[addaleax]

Would it make sense to initialize the pointer variables with nullptr instead? That way the code is a bit more “obviously correct”, because then it’s clear (without checking the c-ares API documentation ) that 1. we never leak memory, because unique_ptr always takes ownership, and 2. we never access uninitialized memory, because if the call fails, the unique_ptr will just be empty.

[addaleax]

LGTM either way, but I think initializing the pointers with nullptr instead might be a bit clearer :)

[AasthaGupta]

I totally agree with your comment @addaleax

Let me also initialize the variable with nullptr :)

[AasthaGupta]

@addaleax Is there anything else that needs to be done?

[addaleax]

@AasthaGupta We currently have a 48-hour waiting period before PRs are merged, so that’s why this isn’t merged yet. :)

(Btw, my original suggestion here was to initialize with nullptrinstead of moving the unique_ptr construction sites, not in addition to – that way it would be clear that any memory allocated by c-ares would always be released without knowing how c-ares works, because even if c-ares did allocate and store data, that would be okay then. But since that’s not what’s currently happening in practice, this isn’t really important to me. :))

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/33396/ (:green_heart:)

[Trott]

Landed in 0f41bca.

Thanks for the contribution! 🎉