1447996 - Assertion failure: !inUnsafeRegion ([AutoAssertNoGC] possible GC in GC-unsafe region), at js/src/vm/JSContext.h:591

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect, P1)

Description

[Christian Holler (:decoder)]

The following testcase crashes on mozilla-central revision 7771df14ea18+ (build with --enable-posix-nspr-emulation --enable-valgrind --enable-gczeal --disable-tests --disable-profiling --enable-debug --enable-optimize, run with --fuzzing-safe --cpu-count=2):

var global = this;
var ReflectApply = global.Reflect.apply;
var RegExpPrototypeExec = global.RegExp.prototype.exec;
var StringPrototypeIndexOf = global.String.prototype.indexOf;
function StringSplit(str, delimiter) {
    var last = 0;
    var i = ReflectApply(StringPrototypeIndexOf, str, [delimiter, last]);
}
function printStatus(msg) {
    var lines = StringSplit(msg, "\n");
}
function reportMatch(expectedRegExp, actual, description) {
    var matches = ReflectApply(RegExpPrototypeExec, expectedRegExp, [actual]) !== null;
}
var actual = 'No Crash';
test();
function test() {
    printStatus(actual);
    expect = /TypeError: .+ is not a function/;
    try {
        test("{ 1; ");
    } catch (ex) {}
    reportMatch(expect, actual)
}


Backtrace:

received signal SIGSEGV, Segmentation fault.
0x0000000000ed526a in JSContext::verifyIsSafeToGC (this=0x7ffff5f16000) at js/src/vm/JSContext.h:591
#0  0x0000000000ed526a in JSContext::verifyIsSafeToGC (this=0x7ffff5f16000) at js/src/vm/JSContext.h:591
#1  js::gc::GCRuntime::checkAllocatorState<(js::AllowGC)1> (this=<optimized out>, cx=0x7ffff5f16000, kind=<optimized out>) at js/src/gc/Allocator.cpp:291
#2  0x0000000000ed5555 in js::AllocateString<JSFatInlineString, (js::AllowGC)1> (cx=0x7ffff5f16000, heap=heap@entry=js::gc::DefaultHeap) at js/src/gc/Allocator.cpp:179
#3  0x0000000000c4bd81 in js::Allocate<JSFatInlineString, (js::AllowGC)1> (heap=js::gc::DefaultHeap, cx=<optimized out>) at js/src/gc/Allocator.h:57
#4  JSFatInlineString::new_<(js::AllowGC)1> (cx=<optimized out>) at js/src/vm/StringType-inl.h:283
#5  js::AllocateInlineString<(js::AllowGC)1, unsigned char> (cx=<optimized out>, len=18, chars=0x7fffffdfd840) at js/src/vm/StringType-inl.h:38
#6  0x0000000000c79ae1 in js::NewInlineString<(js::AllowGC)1, unsigned char> (cx=0x7ffff5f16000, chars=...) at js/src/vm/StringType-inl.h:57
#7  js::NewStringCopyNDontDeflate<(js::AllowGC)1, unsigned char> (cx=0x7ffff5f16000, s=<optimized out>, n=<optimized out>) at js/src/vm/StringType.cpp:1534
#8  0x0000000000c79e8b in js::NewStringCopyUTF8N<(js::AllowGC)1> (cx=cx@entry=0x7ffff5f16000, utf8=...) at js/src/vm/StringType.cpp:1605
#9  0x00000000009c788b in js::NewStringCopyUTF8Z<(js::AllowGC)1> (utf8=..., cx=0x7ffff5f16000) at js/src/vm/StringType.h:1499
#10 JS_NewStringCopyUTF8Z (cx=0x7ffff5f16000, s=...) at js/src/jsapi.cpp:5778
#11 0x00000000009d9103 in js::ErrorToException (cx=0x7ffff5f16000, reportp=0x7fffffdfdb30, callback=<optimized out>, userRef=<optimized out>) at js/src/jsexn.cpp:678
#12 0x0000000000b3ccd3 in js::ReportErrorNumberVA (cx=cx@entry=0x7ffff5f16000, flags=flags@entry=0, callback=callback@entry=0xb26df0 <js::GetErrorMessage(void*, unsigned int)>, userRef=userRef@entry=0x0, errorNumber=errorNumber@entry=116, argumentsType=argumentsType@entry=js::ArgumentsAreASCII, ap=0x7fffffdfdc10) at js/src/vm/JSContext.cpp:903
#13 0x00000000009c10a8 in JS_ReportErrorNumberASCIIVA (cx=0x7ffff5f16000, errorCallback=0xb26df0 <js::GetErrorMessage(void*, unsigned int)>, userRef=0x0, errorNumber=116, ap=ap@entry=0x7fffffdfdc10) at js/src/jsapi.cpp:6440
#14 0x00000000009c1158 in JS_ReportErrorNumberASCII (cx=cx@entry=0x7ffff5f16000, errorCallback=errorCallback@entry=0xb26df0 <js::GetErrorMessage(void*, unsigned int)>, userRef=userRef@entry=0x0, errorNumber=errorNumber@entry=116) at js/src/jsapi.cpp:6429
#15 0x0000000000b3a876 in js::ReportOverRecursed (maybecx=0x7ffff5f16000, errorNumber=116) at js/src/vm/JSContext.cpp:402
#16 0x00000000004d56b8 in js::CheckRecursionLimit (cx=0x7ffff5f16000) at js/src/jsfriendapi.h:1098
#17 0x0000000000f26a80 in js::irregexp::RegExpCompiler::CheckOverRecursed (this=<optimized out>) at js/src/irregexp/RegExpEngine.cpp:2540
#18 js::irregexp::BoyerMooreLookahead::CheckOverRecursed (this=0x7ffff559a850) at js/src/irregexp/RegExpEngine.cpp:2551
#19 js::irregexp::ActionNode::FillInBMInfo (this=0x7ffff559a668, offset=0, budget=200, bm=0x7ffff559a850, not_at_start=<optimized out>) at js/src/irregexp/RegExpEngine.cpp:633
#20 0x0000000000f4237a in js::irregexp::ChoiceNode::Emit (this=0x7ffff559a6e0, compiler=0x7fffffdfe470, trace=0x7fffffdfe240) at js/src/irregexp/RegExpEngine.cpp:4362
#21 0x0000000000f38f6c in js::irregexp::RegExpCompiler::Assemble (this=this@entry=0x7fffffdfe470, cx=cx@entry=0x7ffff5f16000, assembler=<optimized out>, start=start@entry=0x7ffff559a6e0, capture_count=<optimized out>) at js/src/irregexp/RegExpEngine.cpp:1684
#22 0x0000000000f401bc in js::irregexp::CompilePattern (cx=cx@entry=0x7ffff5f16000, shared=..., shared@entry=..., data=data@entry=0x7fffffdff690, sample=..., sample@entry=..., is_global=is_global@entry=false, ignore_case=<optimized out>, is_latin1=true, match_only=false, force_bytecode=false, sticky=false, unicode=false, tables=...) at js/src/irregexp/RegExpEngine.cpp:1852
#23 0x0000000000be9548 in js::RegExpShared::compile (cx=0x7ffff5f16000, re=re@entry=..., pattern=..., pattern@entry=..., input=..., input@entry=..., mode=mode@entry=js::RegExpShared::Normal, force=force@entry=js::RegExpShared::DontForceByteCode) at js/src/vm/RegExpObject.cpp:1029
#24 0x0000000000bea022 in js::RegExpShared::compile (cx=0x7ffff5f16000, re=..., input=..., mode=js::RegExpShared::Normal, force=js::RegExpShared::DontForceByteCode) at js/src/vm/RegExpObject.cpp:994
#25 0x0000000000bea1b2 in js::RegExpShared::compileIfNecessary (cx=cx@entry=0x7ffff5f16000, re=..., re@entry=..., input=..., input@entry=..., mode=mode@entry=js::RegExpShared::Normal, force=force@entry=js::RegExpShared::DontForceByteCode) at js/src/vm/RegExpObject.cpp:1062
#26 0x0000000000bf0c0a in js::RegExpShared::execute (cx=cx@entry=0x7ffff5f16000, re=..., re@entry=..., input=..., input@entry=..., start=start@entry=0, matches=matches@entry=0x7fffffe00020, endIndex=endIndex@entry=0x0) at js/src/vm/RegExpObject.cpp:1076
#27 0x00000000004e8539 in ExecuteRegExpImpl (cx=cx@entry=0x7ffff5f16000, res=0x7ffff542db80, re=re@entry=..., input=input@entry=..., searchIndex=0, matches=matches@entry=0x7fffffe00020, endIndex=0x0) at js/src/builtin/RegExp.cpp:125
#28 0x00000000004e89d9 in ExecuteRegExp (cx=cx@entry=0x7ffff5f16000, regexp=regexp@entry=..., string=..., string@entry=..., lastIndex=<optimized out>, matches=matches@entry=0x7fffffe00020, endIndex=endIndex@entry=0x0) at js/src/builtin/RegExp.cpp:963
#29 0x00000000004ea165 in RegExpMatcherImpl (rval=..., lastIndex=<optimized out>, string=..., regexp=..., cx=0x7ffff5f16000) at js/src/builtin/RegExp.cpp:984
#30 js::RegExpMatcher (cx=0x7ffff5f16000, argc=<optimized out>, vp=<optimized out>) at js/src/builtin/RegExp.cpp:1018
#31 0x0000302fd2893791 in ?? ()
#32 0x00007fffffe00150 in ?? ()
#33 0x00007fffffe00100 in ?? ()
#34 0x0000000000000000 in ?? ()
rax	0x0	0
rbx	0x7ffff5f16000	140737319624704
rcx	0x7ffff6c282ad	140737333330605
rdx	0x0	0
rsi	0x7ffff6ef7770	140737336276848
rdi	0x7ffff6ef6540	140737336272192
rbp	0x7fffffdfd7d0	140737486247888
rsp	0x7fffffdfd7b0	140737486247856
r8	0x7ffff6ef7770	140737336276848
r9	0x7ffff7fe4780	140737354024832
r10	0x58	88
r11	0x7ffff6b9e7a0	140737332766624
r12	0x6	6
r13	0x1	1
r14	0x12	18
r15	0x7fffffdfda50	140737486248528
rip	0xed526a <js::gc::GCRuntime::checkAllocatorState<(js::AllowGC)1>(JSContext*, js::gc::AllocKind)+250>
=> 0xed526a <js::gc::GCRuntime::checkAllocatorState<(js::AllowGC)1>(JSContext*, js::gc::AllocKind)+250>:	movl   $0x0,0x0
   0xed5275 <js::gc::GCRuntime::checkAllocatorState<(js::AllowGC)1>(JSContext*, js::gc::AllocKind)+261>:	ud2


Marking s-s because this assert is about an unsafe GC.

Comment 1

[Jan de Mooij [:jandem]]

This might be from bug 1447578.

Steve, can you help me figure out why the static analysis didn't catch this? :)

Comment 2

[Jan de Mooij [:jandem]]

Oh we have a

    Maybe<NativeRegExpMacroAssembler> native_assembler;

And that contains our MacroAssembler, so that probably explains it. I can put an explicit JS::AutoCheckCannotGC on top of this, see if that reports the hazard.

I'm not sure if it's a real hazard - when we hit overrecursion we call SetRegExpTooBig(). I'll look into it more; I'm not sure if we check that flag also after we Assemble()...

Comment 3

[Jan de Mooij [:jandem]]

(In reply to Jan de Mooij [:jandem] from comment #2)
> I'm not sure if we check that flag also after we Assemble()...

Oh we do, we just don't call the getter for it:

https://searchfox.org/mozilla-central/rev/c217fbde244344fedfd07b57a740c694a456dbca/js/src/irregexp/RegExpEngine.cpp#1695

So I think we're good here security-wise, we just need to check for overrecursion without reporting an exception. I'll check if that's the only thing here that can GC.

Comment 4

[Fuzzing Team]

JSBugMon: Bisection requested, failed due to error (try manually).

Comment 5

[Jan de Mooij [:jandem]]

Attachment: Patch

This adds an explicit AutoCheckCannotGC because the static analysis doesn't understand that Maybe<NativeRegExpMacroAssembler> contains one (via StackMacroAssembler).

The rest of the patch is dealing with some issues the analysis reported.

Comment 6

[Jan de Mooij [:jandem]]

. The only places that can GC are calls that will abort regex compilation anyway.

Comment 7

[Steve Fink [:sfink] [:s:]]

(In reply to Jan de Mooij [:jandem] from comment #2)
> Oh we have a
> 
>     Maybe<NativeRegExpMacroAssembler> native_assembler;
> 
> And that contains our MacroAssembler, so that probably explains it. I can
> put an explicit JS::AutoCheckCannotGC on top of this, see if that reports
> the hazard.

Yeah, the Maybe<> blocks it. And I don't want the analysis to consider Maybe<T> to be a T, because it is often used precisely to limit the region of influence of an RAII guard, eg Maybe<AutoSuppressGC> or whatever. I suppose I could try to get fancy and say that any part of the graph that is reachable by an emplace() should see through Maybe<>, but I don't know if it's worth the effort. (And what if the GC in question always has an intervening reset()?)

Filed bug 1448131.

Comment 8

[Steve Fink [:sfink] [:s:]]

Comment on attachment 8961498 [details] [diff] [review]
Patch

Review of attachment 8961498 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/irregexp/RegExpEngine.cpp
@@ +1693,5 @@
>          return RegExpCode();
>  
>      if (reg_exp_too_big_) {
>          code.destroy();
> +        js::gc::AutoSuppressGC suppress(cx);

Ugh. This really shouldn't be necessary, since there's nothing held live across it that matters. AutoSuppressGCAnalysis would be better, but that's not going to work either (because it asserts if we GC.)

So yeah, this seems like the best of our available options. I mean, you could use

    struct IgnoreForAnalysis {} JS_HAZ_GC_SUPPRESSED ignore;

instead, but I prefer what you have.

Comment 9

[Jan de Mooij [:jandem]]

(In reply to Steve Fink [:sfink] [:s:] from comment #8)
> So yeah, this seems like the best of our available options. I mean, you
> could use
> 
>     struct IgnoreForAnalysis {} JS_HAZ_GC_SUPPRESSED ignore;
> 
> instead, but I prefer what you have.

I don't think this would work, because we're still in an AutoCheckCannotGC scope (two of them in fact) which also does runtime checks, so it's not just the static analysis we're trying to appease.

Comment 10

[Jan de Mooij [:jandem]]

Not a security bug.

Comment 11

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/7a682f7ab2f5
Don't GC when hitting overrecursion in RegExpCompiler; make the static analysis detect this. r=sfink

Comment 12

[Alex Gaynor [:Alex_Gaynor]]

FYI: OSS-Fuzz also found this crash (about 3 hours before Christian filed this bug, but for whatever reason it took a day for their automation to report the finding :-)).

Comment 13

[Raul Gurzau (:RaulG)]

https://hg.mozilla.org/mozilla-central/rev/7a682f7ab2f5