Closed Bug 610040 Opened 14 years ago Closed 14 years ago

OOM abort under nsBufferedStream::Init

Categories

(Core :: Networking: JAR, defect)

defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla2.0b8
Tracking Status
blocking2.0 --- betaN+

People

(Reporter: jorendorff, Assigned: taras.mozilla)

References

Details

Attachments

(1 file)

Apparently caused by bug 594172: Do anything that causes the firefox process to get huge and you'll eventually abort. In short, nsZipWriter::Open calls NS_BufferedOutputStream with the hard-coded argument 3 * 1024 * 1024. That ends up calling moz_xmalloc(3 * 1024 * 1024), which aborts if there's not enough memory. mozalloc.dll!mozalloc_abort(const char * const msg) Line 77 C++ mozalloc.dll!mozalloc_handle_oom() Line 54 + 0xa bytes C++ mozalloc.dll!moz_xmalloc(unsigned int size) Line 101 C++ xul.dll!operator new[](unsigned int size) Line 238 + 0xa bytes C++ xul.dll!nsBufferedStream::Init(nsISupports * stream, unsigned int bufferSize) Line 105 + 0x9 bytes C++ xul.dll!nsBufferedOutputStream::Init(nsIOutputStream * stream, unsigned int bufferSize) Line 562 C++ xul.dll!NS_NewBufferedOutputStream(nsIOutputStream * * result, nsIOutputStream * str, unsigned int bufferSize) Line 1096 + 0x21 bytes C++ xul.dll!nsZipWriter::Open(nsIFile * aFile, int aIoFlags) Line 304 + 0x2e bytes C++ xul.dll!mozilla::scache::StartupCache::PutBuffer(const char * id, const char * inbuf, unsigned int len) Line 289 + 0x39 bytes C++ xul.dll!mozJSComponentLoader::WriteScript(mozilla::scache::StartupCache * cache, JSScript * script, nsIFile * component, nsIURI * uri, JSContext * cx) Line 934 + 0x1e bytes C++ xul.dll!mozJSComponentLoader::GlobalForLocation(nsILocalFile * aComponentFile, nsIURI * aURI, JSObject * * aGlobal, char * * aLocation, jsval_layout * exception) Line 1221 + 0x2d bytes C++ xul.dll!mozJSComponentLoader::LoadModuleImpl(nsILocalFile * aSourceFile, nsAString_internal & aKey, nsIURI * aComponentURI) Line 729 + 0x2d bytes C++ xul.dll!mozJSComponentLoader::LoadModule(nsILocalFile * aComponentFile) Line 664 + 0x1f bytes C++ xul.dll!nsComponentManagerImpl::KnownModule::Load() Line 952 + 0x27 bytes C++ xul.dll!nsFactoryEntry::GetFactory() Line 1936 + 0xb bytes C++ xul.dll!nsComponentManagerImpl::CreateInstance(const nsID & aClass, nsISupports * aDelegate, const nsID & aIID, void * * aResult) Line 1213 + 0xc bytes C++ xul.dll!nsJSCID::CreateInstance(nsISupports * * _retval) Line 771 + 0x43 bytes C++ xul.dll!NS_InvokeByIndex_P(nsISupports * that, unsigned int methodIndex, unsigned int paramCount, nsXPTCVariant * params) Line 103 C++ xul.dll!CallMethodHelper::Invoke() Line 3054 + 0x1c bytes C++ xul.dll!CallMethodHelper::Call() Line 2321 + 0x8 bytes C++ xul.dll!XPCWrappedNative::CallMethod(XPCCallContext & ccx, XPCWrappedNative::CallMode mode) Line 2285 + 0x16 bytes C++ xul.dll!XPC_WN_CallMethod(JSContext * cx, unsigned int argc, jsval_layout * vp) Line 1631 + 0xe bytes C++ mozjs.dll!js::CallJSNative(JSContext * cx, int (JSContext *, unsigned int, js::Value *)* native, unsigned int argc, js::Value * vp) Line 684 + 0xf bytes C++ mozjs.dll!js::Interpret(JSContext * cx, JSStackFrame * entryFrame, unsigned int inlineCallCount, JSInterpMode interpMode) Line 4765 + 0x21 bytes C++ mozjs.dll!js::RunScript(JSContext * cx, JSScript * script, JSStackFrame * fp) Line 665 + 0x11 bytes C++ mozjs.dll!js::Invoke(JSContext * cx, const js::CallArgs & argsRef, unsigned int flags) Line 768 + 0x11 bytes C++ mozjs.dll!js::ExternalInvoke(JSContext * cx, const js::Value & thisv, const js::Value & fval, unsigned int argc, js::Value * argv, js::Value * rval) Line 881 + 0xf bytes C++ mozjs.dll!js::ExternalInvoke(JSContext * cx, JSObject * obj, const js::Value & fval, unsigned int argc, js::Value * argv, js::Value * rval) Line 954 + 0x2a bytes C++ mozjs.dll!JS_CallFunctionValue(JSContext * cx, JSObject * obj, jsval_layout fval, unsigned int argc, jsval_layout * argv, jsval_layout * rval) Line 4961 + 0x38 bytes C++ xul.dll!nsXPCWrappedJSClass::CallMethod(nsXPCWrappedJS * wrapper, unsigned short methodIndex, const XPTMethodDescriptor * info, nsXPTCMiniVariant * nativeParams) Line 1694 + 0x35 bytes C++ xul.dll!nsXPCWrappedJS::CallMethod(unsigned short methodIndex, const XPTMethodDescriptor * info, nsXPTCMiniVariant * params) Line 578 C++ xul.dll!PrepareAndDispatch(nsXPTCStubBase * self, unsigned int methodIndex, unsigned int * args, unsigned int * stackBytesToPop) Line 114 + 0x21 bytes C++ xul.dll!SharedStub() Line 142 C++ xul.dll!nsHttpChannel::CallOnStartRequest() Line 770 + 0x44 bytes C++ xul.dll!nsHttpChannel::ContinueProcessNormal(unsigned int rv) Line 1228 + 0x8 bytes C++ xul.dll!nsHttpChannel::ProcessNormal() Line 1166 C++ xul.dll!nsHttpChannel::ProcessResponse() Line 1052 + 0x8 bytes C++ xul.dll!nsHttpChannel::OnStartRequest(nsIRequest * request, nsISupports * ctxt) Line 3805 + 0xe bytes C++ xul.dll!nsInputStreamPump::OnStateStart() Line 441 + 0x2c bytes C++ xul.dll!nsInputStreamPump::OnInputStreamReady(nsIAsyncInputStream * stream) Line 397 + 0xb bytes C++ xul.dll!nsInputStreamReadyEvent::Run() Line 113 C++ xul.dll!nsThread::ProcessNextEvent(int mayWait, int * result) Line 609 + 0x19 bytes C++ xul.dll!NS_ProcessNextEvent_P(nsIThread * thread, int mayWait) Line 250 + 0x16 bytes C++ xul.dll!mozilla::ipc::MessagePump::Run(base::MessagePump::Delegate * aDelegate) Line 134 + 0xe bytes C++ xul.dll!MessageLoop::RunInternal() Line 220 C++ xul.dll!MessageLoop::RunHandler() Line 203 C++ xul.dll!MessageLoop::Run() Line 177 C++ xul.dll!nsBaseAppShell::Run() Line 187 C++ xul.dll!nsAppShell::Run() Line 243 + 0x9 bytes C++ xul.dll!nsAppStartup::Run() Line 191 + 0x1c bytes C++ xul.dll!XRE_main(int argc, char * * argv, const nsXREAppData * aAppData) Line 3682 + 0x25 bytes C++ firefox.exe!NS_internal_main(int argc, char * * argv) Line 158 + 0x12 bytes C++ firefox.exe!wmain(int argc, wchar_t * * argv) Line 129 + 0xd bytes C++ firefox.exe!__tmainCRTStartup() Line 552 + 0x19 bytes C firefox.exe!wmainCRTStartup() Line 371 C
Did this actually happen in the wild? Julian pointed this out from his valgrind exploration today.
Yeah, I had the ES5 spec loaded (about 1.5MB of HTML) and was using jQuery on it. firefox bloated to about 1.5GB and then aborted.
a) What sort of large buffer size is safer? I don't really want to cap it at 64K b)Sounds like we should mod bufferedoutputstream to use fallible malloc.
Assignee: nobody → tglek
(In reply to comment #1) Yeah, I also saw that; stats at https://bugzilla.mozilla.org/show_bug.cgi?id=610064#c0 Not only is the buffer 3MB, which seems a bit extravagant, at least in the case I was looking at 98% of the bytes in it were neither read nor written. (In reply to comment #3) > I don't really want to cap it at 64K Why not? Originally it was 2048 bytes. Changing it to 64k will reduce the per-buffer-fill costs by a factor of 32 without spiking heap usage in this way.
(In reply to comment #5) > (In reply to comment #1) > Yeah, I also saw that; stats at > https://bugzilla.mozilla.org/show_bug.cgi?id=610064#c0 > > Not only is the buffer 3MB, which seems a bit extravagant, > at least in the case I was looking at 98% of the bytes in > it were neither read nor written. > > (In reply to comment #3) > > I don't really want to cap it at 64K > > Why not? Originally it was 2048 bytes. Changing it to 64k will > reduce the per-buffer-fill costs by a factor of 32 without > spiking heap usage in this way. I'm not concerned about fill-costs. This was a workaround for crappy operating systems fragmenting the filesystem more if lots of little writes are used(big buffer is a way to reduce the number of write calls). I was curious to see if 3mb allocation would cause any issues, now I know :) I'll change it to 64K and use another way of dealing with the filesystem.
Attached patch oom fixSplinter Review
I'll let bufferedwriter be for now
Attachment #488610 - Flags: review?(dtownsend)
Can I nominate this fix to make it into Fx 4.0 ? Considering that the mobile folks just this week have been battling some space problems, iirc.
Attachment #488610 - Flags: review?(dtownsend) → review+
blocking2.0: --- → betaN+
Keywords: checkin-needed
Status: NEW → RESOLVED
Closed: 14 years ago
Keywords: checkin-needed
Resolution: --- → FIXED
Target Milestone: --- → mozilla2.0b8
Version: Other Branch → Trunk
Blocks: 598466
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: