[External] : Re: use-after-free in C2
Justin King
jcking at google.com
Mon Feb 13 22:06:21 UTC 2023
I modified Arena to make right-sized allocations, that's why the sizes are
odd. It's close to pointing to the root cause, but to get exact I need to
rework Arena more. I'll see if I get time.
On Mon, Feb 13, 2023 at 2:00 PM <dean.long at oracle.com> wrote:
> That's strange. Doesn't Arena::grow() allocate a chunk of size
> Chunk::size (~32K) or larger?
>
> dl
> On 2/10/23 12:42 PM, Justin King wrote:
>
> Hm. That size is still too big, I don't think the node is 224 bytes. Let
> me double check again.
>
> On Fri, Feb 10, 2023 at 12:06 PM Justin King <jcking at google.com> wrote:
>
>> Looks to be the temporary node created by clone_map is leaking. I think
>> there is a missing call to undo some of the work done by clone_map.
>>
>> ==3591618==ERROR: AddressSanitizer: use-after-poison on address
>> 0x6110000e20b8 at pc 0x7f7d8cac1875 bp 0x7f7d1028be20 sp 0x7f7d1028be18
>> READ of size 4 at 0x6110000e20b8 thread T13
>> #0 0x7f7d8cac1874 in
>> Unique_Node_List::remove_useless_nodes(VectorSet&)
>> src/hotspot/share/opto/node.cpp:2967
>> #1 0x7f7d8cc90b03 in
>> PhaseRemoveUseless::PhaseRemoveUseless(PhaseGVN*, Unique_Node_List*,
>> Phase::PhaseNumber) src/hotspot/share/opto/phaseX.cpp:423
>> #2 0x7f7d8b03853b in Compile::Compile(ciEnv*, ciMethod*, int,
>> Options, DirectiveSet*) src/hotspot/share/opto/compile.cpp:797
>> #3 0x7f7d8ace8ece in C2Compiler::compile_method(ciEnv*, ciMethod*,
>> int, bool, DirectiveSet*) src/hotspot/share/opto/c2compiler.cpp:113
>> #4 0x7f7d8b0507f8 in
>> CompileBroker::invoke_compiler_on_method(CompileTask*)
>> src/hotspot/share/compiler/compileBroker.cpp:2237
>> #5 0x7f7d8b053e57 in CompileBroker::compiler_thread_loop()
>> src/hotspot/share/compiler/compileBroker.cpp:1916
>> #6 0x7f7d8bc1f3a8 in JavaThread::thread_main_inner()
>> src/hotspot/share/runtime/javaThread.cpp:710
>> #7 0x7f7d8bc1f99f in JavaThread::thread_main_inner()
>> src/hotspot/share/runtime/javaThread.cpp:689
>> #8 0x7f7d8bc1f99f in JavaThread::run()
>> src/hotspot/share/runtime/javaThread.cpp:695
>> #9 0x7f7d8d535b35 in Thread::call_run()
>> src/hotspot/share/runtime/thread.cpp:224
>> #10 0x7f7d8cb7de8f in thread_native_entry
>> src/hotspot/os/linux/os_linux.cpp:737
>> #11 0x7f7d8faa7fd3 in start_thread nptl/pthread_create.c:442
>> #12 0x7f7d8fb2866b in clone3
>> ../sysdeps/unix/sysv/linux/x86_64/clone3.S:81
>>
>> 0x6110000e20b8 is located 56 bytes inside of 224-byte region
>> [0x6110000e2080,0x6110000e2160)
>> allocated by thread T13 here:
>> #0 0x7f7d8fcb89cf in __interceptor_malloc
>> ../../../../src/libsanitizer/asan/asan_malloc_linux.cpp:69
>> #1 0x7f7d8cb65b9a in os::malloc(unsigned long, MEMFLAGS,
>> NativeCallStack const&) src/hotspot/share/runtime/os.cpp:672
>> #2 0x7f7d8a68c47e in Chunk::operator new(unsigned long,
>> AllocFailStrategy::AllocFailEnum, unsigned long)
>> src/hotspot/share/memory/arena.cpp:190
>> #3 0x7f7d8a68c47e in Arena::grow(unsigned long,
>> AllocFailStrategy::AllocFailEnum) src/hotspot/share/memory/arena.cpp:345
>> #4 0x7f7d8cac1c44 in Arena::internal_amalloc(unsigned long,
>> AllocFailStrategy::AllocFailEnum) src/hotspot/share/memory/arena.hpp:113
>> #5 0x7f7d8cac1c44 in Arena::AmallocWords(unsigned long,
>> AllocFailStrategy::AllocFailEnum) src/hotspot/share/memory/arena.hpp:140
>> #6 0x7f7d8cac1c44 in Node::clone() const
>> src/hotspot/share/opto/node.cpp:495
>> #7 0x7f7d8b980c38 in GraphKit::clone_map()
>> src/hotspot/share/opto/graphKit.cpp:727
>> #8 0x7f7d8c538b44 in
>> LibraryCallKit::inline_unsafe_load_store(BasicType,
>> LibraryCallKit::LoadStoreKind, LibraryCallKit::AccessKind)
>> src/hotspot/share/opto/library_call.cpp:2599
>> #9 0x7f7d8c5aad3e in LibraryCallKit::try_to_inline(int)
>> src/hotspot/share/opto/library_call.cpp:416
>> #10 0x7f7d8c5ae34b in LibraryIntrinsic::generate(JVMState*)
>> src/hotspot/share/opto/library_call.cpp:116
>> #11 0x7f7d8b4430e9 in Parse::do_call()
>> src/hotspot/share/opto/doCall.cpp:662
>> #12 0x7f7d8cc4c1ef in Parse::do_one_bytecode()
>> src/hotspot/share/opto/parse2.cpp:2704
>> #13 0x7f7d8cc1f3f6 in Parse::do_one_block()
>> src/hotspot/share/opto/parse1.cpp:1554
>> #14 0x7f7d8cc207ce in Parse::do_all_blocks()
>> src/hotspot/share/opto/parse1.cpp:706
>> #15 0x7f7d8cc2a214 in Parse::Parse(JVMState*, ciMethod*, float)
>> src/hotspot/share/opto/parse1.cpp:613
>> #16 0x7f7d8acec235 in ParseGenerator::generate(JVMState*)
>> src/hotspot/share/opto/callGenerator.cpp:99
>> #17 0x7f7d8b03738c in Compile::Compile(ciEnv*, ciMethod*, int,
>> Options, DirectiveSet*) src/hotspot/share/opto/compile.cpp:763
>> #18 0x7f7d8ace8ece in C2Compiler::compile_method(ciEnv*, ciMethod*,
>> int, bool, DirectiveSet*) src/hotspot/share/opto/c2compiler.cpp:113
>> #19 0x7f7d8b0507f8 in
>> CompileBroker::invoke_compiler_on_method(CompileTask*)
>> src/hotspot/share/compiler/compileBroker.cpp:2237
>> #20 0x7f7d8b053e57 in CompileBroker::compiler_thread_loop()
>> src/hotspot/share/compiler/compileBroker.cpp:1916
>> #21 0x7f7d8bc1f3a8 in JavaThread::thread_main_inner()
>> src/hotspot/share/runtime/javaThread.cpp:710
>> #22 0x7f7d8bc1f99f in JavaThread::thread_main_inner()
>> src/hotspot/share/runtime/javaThread.cpp:689
>> #23 0x7f7d8bc1f99f in JavaThread::run()
>> src/hotspot/share/runtime/javaThread.cpp:695
>> #24 0x7f7d8d535b35 in Thread::call_run()
>> src/hotspot/share/runtime/thread.cpp:224
>> #25 0x7f7d8cb7de8f in thread_native_entry
>> src/hotspot/os/linux/os_linux.cpp:737
>> #26 0x7f7d8faa7fd3 in start_thread nptl/pthread_create.c:442
>>
>>
>>
>> On Thu, Feb 9, 2023 at 2:31 PM Justin King <jcking at google.com> wrote:
>>
>>> Hm. It unfortunately does not show where it was poisoned, as the Arena
>>> uses large chunks with multiple separate allocations per chunk. ASan only
>>> keeps track of malloc/free. But if I change the Arena implementation when
>>> building under ASan to just use a right-sized chunk for each request, it
>>> should be able to show us. I'll try that and bump this once I get something
>>> more definitive.
>>>
>>> On Thu, Feb 9, 2023 at 1:41 PM <dean.long at oracle.com> wrote:
>>>
>>>> Can ASan show where the memory was freed? We've had crashes in the
>>>> past (like JDK-8270028) that could be related to memory corruption or how
>>>> ResourceArea recycles memory. The allocation below seems to be using an
>>>> Arena without a ResourceArea, but if some other code used the same arena
>>>> wrapped in a ResourceArea, then it seems like that could lead to potential
>>>> problems.
>>>>
>>>> dl
>>>> On 2/9/23 8:59 AM, Justin King wrote:
>>>>
>>>> Hi,
>>>>
>>>> I was looking at instrumenting Arena again for ASan. The WIP patch is
>>>> 047d4aa9a091cf5a84b9308454862e39666ca253
>>>> <https://github.com/jcking/jdk/commit/047d4aa9a091cf5a84b9308454862e39666ca253>.
>>>> I ran back into the suspicious logic in C2
>>>> <https://bugs.openjdk.org/browse/JDK-8298984> where nodes are used
>>>> after calling Arena::Afree. The first issue is present in Node::destruct,
>>>> which I fixed by moving the call to Arena::Afree to the bottom of the
>>>> function (addressed in patch). The second issue came up after in
>>>> Compile::Compile, the stack trace is below. It looks like there are
>>>> residual freed nodes being operated on? Maybe we are failing to unregister
>>>> a temporary node from the node list? Maybe related to clone_map which
>>>> returns SafePointNode?
>>>>
>>>> ==3146540==ERROR: AddressSanitizer: use-after-poison on address
>>>> 0x62d00996b370 at pc 0x7f9f93048335 bp 0x7f9ed29fae20 sp 0x7f9ed29fae18
>>>> READ of size 4 at 0x62d00996b370 thread T13
>>>> #0 0x7f9f93048334 in
>>>> Unique_Node_List::remove_useless_nodes(VectorSet&)
>>>> src/hotspot/share/opto/node.cpp:2967
>>>> #1 0x7f9f932124b3 in
>>>> PhaseRemoveUseless::PhaseRemoveUseless(PhaseGVN*, Unique_Node_List*,
>>>> Phase::PhaseNumber) src/hotspot/share/opto/phaseX.cpp:423
>>>> #2 0x7f9f91621beb in Compile::Compile(ciEnv*, ciMethod*, int,
>>>> Options, DirectiveSet*) src/hotspot/share/opto/compile.cpp:797
>>>> #3 0x7f9f912e37fe in C2Compiler::compile_method(ciEnv*, ciMethod*,
>>>> int, bool, DirectiveSet*) src/hotspot/share/opto/c2compiler.cpp:113
>>>> #4 0x7f9f91638e07 in
>>>> CompileBroker::invoke_compiler_on_method(CompileTask*)
>>>> src/hotspot/share/compiler/compileBroker.cpp:2237
>>>> #5 0x7f9f9163bfd7 in CompileBroker::compiler_thread_loop()
>>>> src/hotspot/share/compiler/compileBroker.cpp:1916
>>>> #6 0x7f9f921e3eec in JavaThread::thread_main_inner()
>>>> src/hotspot/share/runtime/javaThread.cpp:710
>>>> #7 0x7f9f921e434f in JavaThread::thread_main_inner()
>>>> src/hotspot/share/runtime/javaThread.cpp:689
>>>> #8 0x7f9f921e434f in JavaThread::run()
>>>> src/hotspot/share/runtime/javaThread.cpp:695
>>>> #9 0x7f9f93aa3f55 in Thread::call_run()
>>>> src/hotspot/share/runtime/thread.cpp:224
>>>> #10 0x7f9f9310144f in thread_native_entry
>>>> src/hotspot/os/linux/os_linux.cpp:737
>>>> #11 0x7f9f960a7fd3 in start_thread nptl/pthread_create.c:442
>>>> #12 0x7f9f9612866b in clone3
>>>> ../sysdeps/unix/sysv/linux/x86_64/clone3.S:81
>>>>
>>>> 0x62d00996b370 is located 20336 bytes inside of 32744-byte region
>>>> [0x62d009966400,0x62d00996e3e8)
>>>> allocated by thread T13 here:
>>>> #0 0x7f9f962b89cf in __interceptor_malloc
>>>> ../../../../src/libsanitizer/asan/asan_malloc_linux.cpp:69
>>>> #1 0x7f9f930e915a in os::malloc(unsigned long, MEMFLAGS,
>>>> NativeCallStack const&) src/hotspot/share/runtime/os.cpp:672
>>>> #2 0x7f9f90c8a08a in Chunk::operator new(unsigned long,
>>>> AllocFailStrategy::AllocFailEnum, unsigned long)
>>>> src/hotspot/share/memory/arena.cpp:190
>>>> #3 0x7f9f90c8a08a in Arena::grow(unsigned long,
>>>> AllocFailStrategy::AllocFailEnum) src/hotspot/share/memory/arena.cpp:325
>>>> #4 0x7f9f932109f5 in Arena::internal_amalloc(unsigned long,
>>>> AllocFailStrategy::AllocFailEnum) src/hotspot/share/memory/arena.hpp:113
>>>> #5 0x7f9f932109f5 in Arena::Amalloc(unsigned long,
>>>> AllocFailStrategy::AllocFailEnum) src/hotspot/share/memory/arena.hpp:133
>>>> #6 0x7f9f932109f5 in NodeHash::NodeHash(Arena*, unsigned int)
>>>> src/hotspot/share/opto/phaseX.cpp:68
>>>> #7 0x7f9f932293c7 in PhaseValues::PhaseValues(Arena*, unsigned int)
>>>> src/hotspot/share/opto/phaseX.cpp:697
>>>> #8 0x7f9f9161f678 in PhaseGVN::PhaseGVN(Arena*, unsigned int)
>>>> src/hotspot/share/opto/phaseX.hpp:415
>>>> #9 0x7f9f9161f678 in Compile::Compile(ciEnv*, ciMethod*, int,
>>>> Options, DirectiveSet*) src/hotspot/share/opto/compile.cpp:714
>>>> #10 0x7f9f912e37fe in C2Compiler::compile_method(ciEnv*, ciMethod*,
>>>> int, bool, DirectiveSet*) src/hotspot/share/opto/c2compiler.cpp:113
>>>> #11 0x7f9f91638e07 in
>>>> CompileBroker::invoke_compiler_on_method(CompileTask*)
>>>> src/hotspot/share/compiler/compileBroker.cpp:2237
>>>> #12 0x7f9f9163bfd7 in CompileBroker::compiler_thread_loop()
>>>> src/hotspot/share/compiler/compileBroker.cpp:1916
>>>> #13 0x7f9f921e3eec in JavaThread::thread_main_inner()
>>>> src/hotspot/share/runtime/javaThread.cpp:710
>>>> #14 0x7f9f921e434f in JavaThread::thread_main_inner()
>>>> src/hotspot/share/runtime/javaThread.cpp:689
>>>> #15 0x7f9f921e434f in JavaThread::run()
>>>> src/hotspot/share/runtime/javaThread.cpp:695
>>>> #16 0x7f9f93aa3f55 in Thread::call_run()
>>>> src/hotspot/share/runtime/thread.cpp:224
>>>> #17 0x7f9f9310144f in thread_native_entry
>>>> src/hotspot/os/linux/os_linux.cpp:737
>>>> #18 0x7f9f960a7fd3 in start_thread nptl/pthread_create.c:442
>>>>
>>>> Thread T13 created by T1 here:
>>>> #0 0x7f9f96249726 in __interceptor_pthread_create
>>>> ../../../../src/libsanitizer/asan/asan_interceptors.cpp:207
>>>> #1 0x7f9f93102d88 in os::create_thread(Thread*, os::ThreadType,
>>>> unsigned long) src/hotspot/os/linux/os_linux.cpp:888
>>>> #2 0x7f9f91693d93 in CompilerThread::CompilerThread(CompileQueue*,
>>>> CompilerCounters*) src/hotspot/share/compiler/compilerThread.cpp:34
>>>> #3 0x7f9f91625c7c in
>>>> CompileBroker::make_thread(CompileBroker::ThreadType, _jobject*,
>>>> CompileQueue*, AbstractCompiler*, JavaThread*)
>>>> src/hotspot/share/compiler/compileBroker.cpp:842
>>>> #4 0x7f9f91628f71 in CompileBroker::init_compiler_threads()
>>>> src/hotspot/share/compiler/compileBroker.cpp:943
>>>> #5 0x7f9f9162a464 in
>>>> CompileBroker::compilation_init_phase1(JavaThread*)
>>>> src/hotspot/share/compiler/compileBroker.cpp:654
>>>> #6 0x7f9f93adc3a4 in Threads::create_vm(JavaVMInitArgs*, bool*)
>>>> src/hotspot/share/runtime/threads.cpp:701
>>>> #7 0x7f9f92465b51 in JNI_CreateJavaVM_inner
>>>> src/hotspot/share/prims/jni.cpp:3588
>>>> #8 0x7f9f92465b51 in JNI_CreateJavaVM
>>>> src/hotspot/share/prims/jni.cpp:3674
>>>> #9 0x7f9f968d2e25 in InitializeJVM
>>>> src/java.base/share/native/libjli/java.c:1459
>>>> #10 0x7f9f968d2e25 in JavaMain
>>>> src/java.base/share/native/libjli/java.c:413
>>>> #11 0x7f9f968db708 in ThreadJavaMain
>>>> src/java.base/unix/native/libjli/java_md.c:650
>>>> #12 0x7f9f960a7fd3 in start_thread nptl/pthread_create.c:442
>>>>
>>>> Thread T1 created by T0 here:
>>>> #0 0x7f9f96249726 in __interceptor_pthread_create
>>>> ../../../../src/libsanitizer/asan/asan_interceptors.cpp:207
>>>> #1 0x7f9f968dd3a1 in CallJavaMainInNewThread
>>>> src/java.base/unix/native/libjli/java_md.c:691
>>>> #2 0x7f9f968d822d in ContinueInNewThread
>>>> src/java.base/share/native/libjli/java.c:2280
>>>> #3 0x7f9f968d96ae in JLI_Launch
>>>> src/java.base/share/native/libjli/java.c:340
>>>> #4 0x5594a81c337c in main
>>>> src/java.base/share/native/launcher/main.c:166
>>>> #5 0x7f9f96046189 in __libc_start_call_main
>>>> ../sysdeps/nptl/libc_start_call_main.h:58
>>>>
>>>> --
>>>>
>>>> [image: Google Logo]
>>>> Justin King
>>>> Software Engineer
>>>> jcking at google.com
>>>>
>>>>
>>>>
>>>
>>> --
>>>
>>> [image: Google Logo]
>>> Justin King
>>> Software Engineer
>>> jcking at google.com
>>>
>>>
>>>
>>
>> --
>>
>> [image: Google Logo]
>> Justin King
>> Software Engineer
>> jcking at google.com
>>
>>
>>
>
> --
>
> [image: Google Logo]
> Justin King
> Software Engineer
> jcking at google.com
>
>
>
--
[image: Google Logo]
Justin King
Software Engineer
jcking at google.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-compiler-dev/attachments/20230213/e420976a/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 3999 bytes
Desc: S/MIME Cryptographic Signature
URL: <https://mail.openjdk.org/pipermail/hotspot-compiler-dev/attachments/20230213/e420976a/smime-0001.p7s>
More information about the hotspot-compiler-dev
mailing list