diff options
| author | hexcoder <hexcoder-@users.noreply.github.com> | 2021-06-01 12:14:53 +0200 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2021-06-01 12:14:53 +0200 |
| commit | d20d03114179e7c1dbd142972cafbf4499978cfc (patch) | |
| tree | f3f568e5ec7b53ff5f296c9fdbd2d41b1dd873aa /instrumentation | |
| parent | 67293b298d2f6146022ac7adebdf17aebf27bea7 (diff) | |
| download | afl++-d20d03114179e7c1dbd142972cafbf4499978cfc.tar.gz | |
Dev (#949)
* use atomic read-modify-write increment for LLVM CLASSIC * Change other LLVM modes to atomic increments * sync (#886) * Create FUNDING.yml * Update FUNDING.yml * moved custom_mutator examples * unicorn speedtest makefile cleanup * fixed example location * fix qdbi * update util readme * work in progress: not working correctly yet * Frida persistent (#880) * Added x64 support for persistent mode (function call only), in-memory teest cases and complog * Review changes, fix NeverZero and code to parse the .text section of the main executable. Excluded ranges TBC * Various minor fixes and finished support for AFL_INST_LIBS * Review changes Co-authored-by: Your Name <you@example.com> * nits * fix frida mode * Integer overflow/underflow fixes in libdislocator (#889) * libdislocator: fixing integer overflow in 'max_mem' variable and setting 'max_mem' type to 'size_t' * libdislocator: fixing potential integer underflow in 'total_mem' variable due to its different values in different threads * Bumped warnings up to the max and fixed remaining issues (#890) Co-authored-by: Your Name <you@example.com> * nits * frida mode - support non-pie * nits * nit * update grammar mutator * Fixes for aarch64, OSX and other minor issues (#891) Co-authored-by: Your Name <you@example.com> * nits * nits * fix PCGUARD, build aflpp_driver with fPIC * Added representative fuzzbench test and test for libxml (#893) * Added representative fuzzbench test and test for libxml * Added support for building FRIDA from source with FRIDA_SOURCE=1 Co-authored-by: Your Name <you@example.com> * nits * update changelog * typos * still not working * fixed potential double free in custom trim (#881) * error handling, freeing mem * frida: complog -> cmplog * fix statsd writing * let aflpp_qemu_driver_hook.so build fail gracefully * fix stdin trimming * Support for AFL_ENTRYPOINT (#898) Co-authored-by: Your Name <you@example.com> * remove the input file .cur_input at the end of the fuzzing, if AFL_TMPDIR is used * reverse push (#901) * Create FUNDING.yml * Update FUNDING.yml * disable QEMU static pie Co-authored-by: Andrea Fioraldi <andreafioraldi@gmail.com> * clarify that no modifications are required. * add new test for frida_mode (please review) * typos * fix persistent mode (64-bit) * set ARCH for linux intel 32-bit for frida-gum-devkit * prepare for 32-bit support (later) * not on qemu 3 anymore * unicorn mips fixes * instrumentation further move to C++11 (#900) * unicorn fixes * first working NeverZero implementation * more unicorn fixes * Fix memory errors when trim causes testcase growth (#881) (#903) * Revert "fixed potential double free in custom trim (#881)" This reverts commit e9d2f72382cab75832721d859c3e731da071435d. * Revert "fix custom trim for increasing data" This reverts commit 86a8ef168dda766d2f25f15c15c4d3ecf21d0667. * Fix memory errors when trim causes testcase growth Modify trim_case_custom to avoid writing into in_buf because some custom mutators can cause the testcase to grow rather than shrink. Instead of modifying in_buf directly, we write the update out to the disk when trimming is complete, and then the caller is responsible for refreshing the in-memory buffer from the file. This is still a bit sketchy because it does need to modify q->len in order to notify the upper layers that something changed, and it could end up telling upper layer code that the q->len is *bigger* than the buffer (q->testcase_buf) that contains it, which is asking for trouble down the line somewhere... * Fix an unlikely situation Put back some `unlikely()` calls that were in the e9d2f72382cab75832721d859c3e731da071435d commit that was reverted. * add some comments * typo * Exit on time (#904) * Variable AFL_EXIT_ON_TIME description has been added. Variables AFL_EXIT_ON_TIME and afl_exit_on_time has been added. afl->exit_on_time variable initialization has been added. The asignment of a value to the afl->afl_env.afl_exit_on_time variable from environment variables has been added. Code to exit on timeout if new path not found has been added. * Type of afl_exit_on_time variable has been changed. Variable exit_on_time has been added to the afl_state_t structure. * Command `export AFL_EXIT_WHEN_DONE=1` has been added. * Millisecond to second conversion has been added. Call get_cur_time() has been added. * Revert to using the saved current time value. * Useless check has been removed. * fix new path to custom-mutators * ensure crashes/README.txt exists * fix * Changes to bump FRIDA version and to clone FRIDA repo in to build directory rather than use a submodule as the FRIDA build scripts don't like it (#906) Co-authored-by: Your Name <you@example.com> * Fix numeric overflow in cmplog implementation (#907) Co-authored-by: Your Name <you@example.com> * testcase fixes for unicorn * remove merge conflict artifacts * fix afl-plot * Changes to remove binaries from frida_mode (#913) Co-authored-by: Your Name <you@example.com> * Frida cmplog fail fast (#914) * Changes to remove binaries from frida_mode * Changes to make cmplog fail fast Co-authored-by: Your Name <you@example.com> * afl-plot: relative time * arch linux and mac os support for afl-system-config * typo * code-format * update documentation * github workflow for qemu * OSX-specific improvements (#912) * Fix afl-cc to work correctly by default on OSX using xcode - CLANG_ENV_VAR must be set for afl-as to work - Use clang mode by default if no specific compiler selected * Add OSX-specific documentation for configuring shared memory * Fixes to memory operands for complog (#916) Co-authored-by: Your Name <you@example.com> * fix a few cur_time uses * added bounds check to pivot_inputs (fixes #921) * additional safety checks for restarts * restrict afl-showmap in_file size * fix seed crash disable * add warning for afl-showmap partial read * no core dumps * AFL_PRINT_FILENAMES added * more documentation for AFL_EXIT_ON_TIME * Flushing for AFL_PRINT_FILENAMES * FASAN Support (#918) * FASAN Support * Fix handling of Address Sanitizer DSO * Changes to identification of Address Sanitizer DSO Co-authored-by: Your Name <you@example.com> * Support for x86 (#920) Co-authored-by: Your Name <you@example.com> * Update frida_mode readme (#925) * libqasan: use syscalls for read and write * update readme * Minor integration tweaks (#926) Co-authored-by: Your Name <you@example.com> * merge * fix afl-fuzz.c frida preload * cleaned up AFL_PRINT_FILENAMES env * Changes to have persistent mode exit at the end of the loop (#928) Co-authored-by: Your Name <you@example.com> * fix llvm-dict2file * push to stable (#931) (#932) * sync (#886) * Create FUNDING.yml * Update FUNDING.yml * moved custom_mutator examples * unicorn speedtest makefile cleanup * fixed example location * fix qdbi * update util readme * Frida persistent (#880) * Added x64 support for persistent mode (function call only), in-memory teest cases and complog * Review changes, fix NeverZero and code to parse the .text section of the main executable. Excluded ranges TBC * Various minor fixes and finished support for AFL_INST_LIBS * Review changes Co-authored-by: Your Name <you@example.com> * nits * fix frida mode * Integer overflow/underflow fixes in libdislocator (#889) * libdislocator: fixing integer overflow in 'max_mem' variable and setting 'max_mem' type to 'size_t' * libdislocator: fixing potential integer underflow in 'total_mem' variable due to its different values in different threads * Bumped warnings up to the max and fixed remaining issues (#890) Co-authored-by: Your Name <you@example.com> * nits * frida mode - support non-pie * nits * nit * update grammar mutator * Fixes for aarch64, OSX and other minor issues (#891) Co-authored-by: Your Name <you@example.com> * nits * nits * fix PCGUARD, build aflpp_driver with fPIC * Added representative fuzzbench test and test for libxml (#893) * Added representative fuzzbench test and test for libxml * Added support for building FRIDA from source with FRIDA_SOURCE=1 Co-authored-by: Your Name <you@example.com> * nits * update changelog * typos * fixed potential double free in custom trim (#881) * error handling, freeing mem * frida: complog -> cmplog * fix statsd writing * let aflpp_qemu_driver_hook.so build fail gracefully * fix stdin trimming * Support for AFL_ENTRYPOINT (#898) Co-authored-by: Your Name <you@example.com> * remove the input file .cur_input at the end of the fuzzing, if AFL_TMPDIR is used * reverse push (#901) * Create FUNDING.yml * Update FUNDING.yml * disable QEMU static pie Co-authored-by: Andrea Fioraldi <andreafioraldi@gmail.com> * clarify that no modifications are required. * add new test for frida_mode (please review) * typos * fix persistent mode (64-bit) * set ARCH for linux intel 32-bit for frida-gum-devkit * prepare for 32-bit support (later) * not on qemu 3 anymore * unicorn mips fixes * instrumentation further move to C++11 (#900) * unicorn fixes * more unicorn fixes * Fix memory errors when trim causes testcase growth (#881) (#903) * Revert "fixed potential double free in custom trim (#881)" This reverts commit e9d2f72382cab75832721d859c3e731da071435d. * Revert "fix custom trim for increasing data" This reverts commit 86a8ef168dda766d2f25f15c15c4d3ecf21d0667. * Fix memory errors when trim causes testcase growth Modify trim_case_custom to avoid writing into in_buf because some custom mutators can cause the testcase to grow rather than shrink. Instead of modifying in_buf directly, we write the update out to the disk when trimming is complete, and then the caller is responsible for refreshing the in-memory buffer from the file. This is still a bit sketchy because it does need to modify q->len in order to notify the upper layers that something changed, and it could end up telling upper layer code that the q->len is *bigger* than the buffer (q->testcase_buf) that contains it, which is asking for trouble down the line somewhere... * Fix an unlikely situation Put back some `unlikely()` calls that were in the e9d2f72382cab75832721d859c3e731da071435d commit that was reverted. * typo * Exit on time (#904) * Variable AFL_EXIT_ON_TIME description has been added. Variables AFL_EXIT_ON_TIME and afl_exit_on_time has been added. afl->exit_on_time variable initialization has been added. The asignment of a value to the afl->afl_env.afl_exit_on_time variable from environment variables has been added. Code to exit on timeout if new path not found has been added. * Type of afl_exit_on_time variable has been changed. Variable exit_on_time has been added to the afl_state_t structure. * Command `export AFL_EXIT_WHEN_DONE=1` has been added. * Millisecond to second conversion has been added. Call get_cur_time() has been added. * Revert to using the saved current time value. * Useless check has been removed. * fix new path to custom-mutators * ensure crashes/README.txt exists * fix * Changes to bump FRIDA version and to clone FRIDA repo in to build directory rather than use a submodule as the FRIDA build scripts don't like it (#906) Co-authored-by: Your Name <you@example.com> * Fix numeric overflow in cmplog implementation (#907) Co-authored-by: Your Name <you@example.com> * testcase fixes for unicorn * remove merge conflict artifacts * fix afl-plot * Changes to remove binaries from frida_mode (#913) Co-authored-by: Your Name <you@example.com> * Frida cmplog fail fast (#914) * Changes to remove binaries from frida_mode * Changes to make cmplog fail fast Co-authored-by: Your Name <you@example.com> * afl-plot: relative time * arch linux and mac os support for afl-system-config * typo * code-format * update documentation * github workflow for qemu * OSX-specific improvements (#912) * Fix afl-cc to work correctly by default on OSX using xcode - CLANG_ENV_VAR must be set for afl-as to work - Use clang mode by default if no specific compiler selected * Add OSX-specific documentation for configuring shared memory * Fixes to memory operands for complog (#916) Co-authored-by: Your Name <you@example.com> * fix a few cur_time uses * added bounds check to pivot_inputs (fixes #921) * additional safety checks for restarts * restrict afl-showmap in_file size * fix seed crash disable * add warning for afl-showmap partial read * no core dumps * AFL_PRINT_FILENAMES added * more documentation for AFL_EXIT_ON_TIME * Flushing for AFL_PRINT_FILENAMES * FASAN Support (#918) * FASAN Support * Fix handling of Address Sanitizer DSO * Changes to identification of Address Sanitizer DSO Co-authored-by: Your Name <you@example.com> * Support for x86 (#920) Co-authored-by: Your Name <you@example.com> * Update frida_mode readme (#925) * libqasan: use syscalls for read and write * update readme * Minor integration tweaks (#926) Co-authored-by: Your Name <you@example.com> * merge * fix afl-fuzz.c frida preload * cleaned up AFL_PRINT_FILENAMES env * Changes to have persistent mode exit at the end of the loop (#928) Co-authored-by: Your Name <you@example.com> * fix llvm-dict2file Co-authored-by: Dominik Maier <domenukk@gmail.com> Co-authored-by: WorksButNotTested <62701594+WorksButNotTested@users.noreply.github.com> Co-authored-by: Your Name <you@example.com> Co-authored-by: Dmitry Zheregelya <zheregelya.d@gmail.com> Co-authored-by: hexcoder <hexcoder-@users.noreply.github.com> Co-authored-by: hexcoder- <heiko@hexco.de> Co-authored-by: Andrea Fioraldi <andreafioraldi@gmail.com> Co-authored-by: David CARLIER <devnexen@gmail.com> Co-authored-by: realmadsci <71108352+realmadsci@users.noreply.github.com> Co-authored-by: Roman M. Iudichev <SecNotice@ya.ru> Co-authored-by: Dustin Spicuzza <dustin@virtualroadside.com> Co-authored-by: Dominik Maier <domenukk@gmail.com> Co-authored-by: WorksButNotTested <62701594+WorksButNotTested@users.noreply.github.com> Co-authored-by: Your Name <you@example.com> Co-authored-by: Dmitry Zheregelya <zheregelya.d@gmail.com> Co-authored-by: hexcoder <hexcoder-@users.noreply.github.com> Co-authored-by: hexcoder- <heiko@hexco.de> Co-authored-by: Andrea Fioraldi <andreafioraldi@gmail.com> Co-authored-by: David CARLIER <devnexen@gmail.com> Co-authored-by: realmadsci <71108352+realmadsci@users.noreply.github.com> Co-authored-by: Roman M. Iudichev <SecNotice@ya.ru> Co-authored-by: Dustin Spicuzza <dustin@virtualroadside.com> * improve error msg * Added documentation for wine LoadLibrary workaround (#933) * Fix cmake target compilation command example (#934) - Fix typo DCMAKE_C_COMPILERC -> DCMAKE_C_COMPILER. - Add `cd build` after `mkdir build`. * showmap passes queue items in alphabetical order * added tmp files to gitignore * lenient dict parsing, no map size enum for binary fuzzing * added info about showmap queue directions * update binary-only doc * turn off map size detection if skip_bin_check is set * Typo * update docs * update afl-system-config * Set kill signal before using it in afl-showmap (#935) * fix afl-cc help output * add libafl to binary-only doc * update docs * less executions on variable paths * AFL_SKIP_CRASHES is obsolete since 3.0 * add AFL_TRY_AFFINITY * Typo * Typo * Typo/wording * tweaks * typos * fix afl-whatsup help output * fix afl-plot output * fix for MacOS * fix cmpcov doc for qemu * fix tmpfile removal * update dockerfile * Frida (#940) * Added re2 test * Added libpcap test * Fix validation of setting of ADDR_NO_RANDOMIZE * Added support for printing original and instrumented code Co-authored-by: Your Name <you@example.com> * Support for AFL_FRIDA_PERSISTENT_RET (#941) Co-authored-by: Your Name <you@example.com> * Changes to add missing exclusion of ranges (#943) Co-authored-by: Your Name <you@example.com> * add --afl-noopt to afl-cc * docs: fix link to README in QuickStartGuide (#946) * Support writing Stalker stats (#945) * Support writing Stalker stats * Fixed string handling in print functions Co-authored-by: Your Name <you@example.com> * afl-cmin help fix, aflpp_driver - + @@ support * fix for afl-showmap * support new env var AFL_LLVM_THREADSAFE_INST to enable atomic counters. add new test case for that. * add documentation for AFL_LLVM_THREADSAFE_INST * add support for AFL_LLVM_THREADSAFE_INST to other LLVM passes * add missing include for _exit() * threadsafe doc fixes, code format * Wording: "never zero" -> NeverZero * fix afl_custom_post_process with multiple custom mutators * fix docs * debug ck_write * fixed potential diff by 0 * fixes * fix classic threadsafe counters Co-authored-by: van Hauser <vh@thc.org> Co-authored-by: Dominik Maier <domenukk@gmail.com> Co-authored-by: WorksButNotTested <62701594+WorksButNotTested@users.noreply.github.com> Co-authored-by: Your Name <you@example.com> Co-authored-by: Dmitry Zheregelya <zheregelya.d@gmail.com> Co-authored-by: Andrea Fioraldi <andreafioraldi@gmail.com> Co-authored-by: David CARLIER <devnexen@gmail.com> Co-authored-by: realmadsci <71108352+realmadsci@users.noreply.github.com> Co-authored-by: Roman M. Iudichev <SecNotice@ya.ru> Co-authored-by: Dustin Spicuzza <dustin@virtualroadside.com> Co-authored-by: 0x4d5a-ctf <51098072+0x4d5a-ctf@users.noreply.github.com> Co-authored-by: Tommy Chiang <oToToT@users.noreply.github.com> Co-authored-by: buherator <buherator@silentsignal.hu> Co-authored-by: Dag Heyman Kajevic <dag.heyman@gmail.com>
Diffstat (limited to 'instrumentation')
| -rw-r--r-- | instrumentation/README.llvm.md | 5 | ||||
| -rw-r--r-- | instrumentation/README.neverzero.md | 14 | ||||
| -rw-r--r-- | instrumentation/SanitizerCoverageLTO.so.cc | 35 | ||||
| -rw-r--r-- | instrumentation/SanitizerCoveragePCGUARD.so.cc | 33 | ||||
| -rw-r--r-- | instrumentation/afl-llvm-lto-instrumentation.so.cc | 37 | ||||
| -rw-r--r-- | instrumentation/afl-llvm-pass.so.cc | 234 |
6 files changed, 291 insertions, 67 deletions
diff --git a/instrumentation/README.llvm.md b/instrumentation/README.llvm.md index cfe537d5..8ce5afb9 100644 --- a/instrumentation/README.llvm.md +++ b/instrumentation/README.llvm.md @@ -144,6 +144,11 @@ is not optimal and was only fixed in llvm 9. You can set this with AFL_LLVM_NOT_ZERO=1 See [README.neverzero.md](README.neverzero.md) +Support for thread safe counters has been added for all modes. +Activate it with `AFL_LLVM_THREADSAFE_INST=1`. The tradeoff is better precision +in multi threaded apps for a slightly higher instrumentation overhead. +This also disables the nozero counter default for performance reasons. + ## 4) Snapshot feature To speed up fuzzing you can use a linux loadable kernel module which enables diff --git a/instrumentation/README.neverzero.md b/instrumentation/README.neverzero.md index 49104e00..9bcae324 100644 --- a/instrumentation/README.neverzero.md +++ b/instrumentation/README.neverzero.md @@ -16,11 +16,12 @@ at a very little cost (one instruction per edge). (The alternative of saturated counters has been tested also and proved to be inferior in terms of path discovery.) -This is implemented in afl-gcc and afl-gcc-fast, however for llvm_mode this is optional if -the llvm version is below 9 - as there is a perfomance bug that is only fixed -in version 9 and onwards. +This is implemented in afl-gcc and afl-gcc-fast, however for llvm_mode this is +optional if multithread safe counters are selected or the llvm version is below +9 - as there are severe performance costs in these cases. -If you want to enable this for llvm versions below 9 then set +If you want to enable this for llvm versions below 9 or thread safe counters +then set ``` export AFL_LLVM_NOT_ZERO=1 @@ -33,3 +34,8 @@ AFL_LLVM_SKIP_NEVERZERO=1 ``` If the target does not have extensive loops or functions that are called a lot then this can give a small performance boost. + +Please note that the default counter implementations are not thread safe! + +Support for thread safe counters in mode LLVM CLASSIC can be activated with setting +`AFL_LLVM_THREADSAFE_INST=1`. \ No newline at end of file diff --git a/instrumentation/SanitizerCoverageLTO.so.cc b/instrumentation/SanitizerCoverageLTO.so.cc index 2f4337eb..20f1856e 100644 --- a/instrumentation/SanitizerCoverageLTO.so.cc +++ b/instrumentation/SanitizerCoverageLTO.so.cc @@ -236,7 +236,8 @@ class ModuleSanitizerCoverage { uint32_t inst = 0; uint32_t afl_global_id = 0; uint64_t map_addr = 0; - char * skip_nozero = NULL; + const char * skip_nozero = NULL; + const char * use_threadsafe_counters = nullptr; std::vector<BasicBlock *> BlockList; DenseMap<Value *, std::string *> valueMap; std::vector<std::string> dictionary; @@ -437,6 +438,7 @@ bool ModuleSanitizerCoverage::instrumentModule( be_quiet = 1; skip_nozero = getenv("AFL_LLVM_SKIP_NEVERZERO"); + use_threadsafe_counters = getenv("AFL_LLVM_THREADSAFE_INST"); if ((ptr = getenv("AFL_LLVM_LTO_STARTID")) != NULL) if ((afl_global_id = atoi(ptr)) < 0) @@ -1208,7 +1210,7 @@ void ModuleSanitizerCoverage::instrumentFunction( return; // Should not instrument sanitizer init functions. if (F.getName().startswith("__sanitizer_")) return; // Don't instrument __sanitizer_* callbacks. - // Don't touch available_externally functions, their actual body is elewhere. + // Don't touch available_externally functions, their actual body is elsewhere. if (F.getLinkage() == GlobalValue::AvailableExternallyLinkage) return; // Don't instrument MSVC CRT configuration helpers. They may run before normal // initialization. @@ -1495,22 +1497,31 @@ void ModuleSanitizerCoverage::InjectCoverageAtBlock(Function &F, BasicBlock &BB, } /* Update bitmap */ + if (use_threadsafe_counters) { /* Atomic */ - LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); - Counter->setMetadata(Mo->getMDKindID("nosanitize"), MDNode::get(*Ct, None)); + IRB.CreateAtomicRMW(llvm::AtomicRMWInst::BinOp::Add, MapPtrIdx, One, + llvm::AtomicOrdering::Monotonic); - Value *Incr = IRB.CreateAdd(Counter, One); + } else { - if (skip_nozero == NULL) { + LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); + Counter->setMetadata(Mo->getMDKindID("nosanitize"), + MDNode::get(*Ct, None)); - auto cf = IRB.CreateICmpEQ(Incr, Zero); - auto carry = IRB.CreateZExt(cf, Int8Tyi); - Incr = IRB.CreateAdd(Incr, carry); + Value *Incr = IRB.CreateAdd(Counter, One); - } + if (skip_nozero == NULL) { - IRB.CreateStore(Incr, MapPtrIdx) - ->setMetadata(Mo->getMDKindID("nosanitize"), MDNode::get(*Ct, None)); + auto cf = IRB.CreateICmpEQ(Incr, Zero); + auto carry = IRB.CreateZExt(cf, Int8Tyi); + Incr = IRB.CreateAdd(Incr, carry); + + } + + IRB.CreateStore(Incr, MapPtrIdx) + ->setMetadata(Mo->getMDKindID("nosanitize"), MDNode::get(*Ct, None)); + + } // done :) diff --git a/instrumentation/SanitizerCoveragePCGUARD.so.cc b/instrumentation/SanitizerCoveragePCGUARD.so.cc index 8878d3b1..4a8c9e28 100644 --- a/instrumentation/SanitizerCoveragePCGUARD.so.cc +++ b/instrumentation/SanitizerCoveragePCGUARD.so.cc @@ -86,7 +86,8 @@ const char SanCovPCsSectionName[] = "sancov_pcs"; const char SanCovLowestStackName[] = "__sancov_lowest_stack"; -static char *skip_nozero; +static const char *skip_nozero; +static const char *use_threadsafe_counters; namespace { @@ -386,6 +387,7 @@ bool ModuleSanitizerCoverage::instrumentModule( be_quiet = 1; skip_nozero = getenv("AFL_LLVM_SKIP_NEVERZERO"); + use_threadsafe_counters = getenv("AFL_LLVM_THREADSAFE_INST"); initInstrumentList(); scanForDangerousFunctions(&M); @@ -1067,22 +1069,31 @@ void ModuleSanitizerCoverage::InjectCoverageAtBlock(Function &F, BasicBlock &BB, /* Load counter for CurLoc */ - Value * MapPtrIdx = IRB.CreateGEP(MapPtr, CurLoc); - LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); + Value *MapPtrIdx = IRB.CreateGEP(MapPtr, CurLoc); - /* Update bitmap */ + if (use_threadsafe_counters) { - Value *Incr = IRB.CreateAdd(Counter, One); + IRB.CreateAtomicRMW(llvm::AtomicRMWInst::BinOp::Add, MapPtrIdx, One, + llvm::AtomicOrdering::Monotonic); - if (skip_nozero == NULL) { + } else { - auto cf = IRB.CreateICmpEQ(Incr, Zero); - auto carry = IRB.CreateZExt(cf, Int8Ty); - Incr = IRB.CreateAdd(Incr, carry); + LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); + /* Update bitmap */ - } + Value *Incr = IRB.CreateAdd(Counter, One); + + if (skip_nozero == NULL) { + + auto cf = IRB.CreateICmpEQ(Incr, Zero); + auto carry = IRB.CreateZExt(cf, Int8Ty); + Incr = IRB.CreateAdd(Incr, carry); - IRB.CreateStore(Incr, MapPtrIdx); + } + + IRB.CreateStore(Incr, MapPtrIdx); + + } // done :) diff --git a/instrumentation/afl-llvm-lto-instrumentation.so.cc b/instrumentation/afl-llvm-lto-instrumentation.so.cc index 68bd2fa5..fe43fbe5 100644 --- a/instrumentation/afl-llvm-lto-instrumentation.so.cc +++ b/instrumentation/afl-llvm-lto-instrumentation.so.cc @@ -93,7 +93,8 @@ class AFLLTOPass : public ModulePass { uint32_t function_minimum_size = 1; uint32_t inst_blocks = 0, inst_funcs = 0, total_instr = 0; unsigned long long int map_addr = 0x10000; - char * skip_nozero = NULL; + const char * skip_nozero = NULL; + const char * use_threadsafe_counters = nullptr; }; @@ -131,6 +132,8 @@ bool AFLLTOPass::runOnModule(Module &M) { be_quiet = 1; + use_threadsafe_counters = getenv("AFL_LLVM_THREADSAFE_INST"); + if ((ptr = getenv("AFL_LLVM_DOCUMENT_IDS")) != NULL) { if ((documentFile = fopen(ptr, "a")) == NULL) @@ -839,22 +842,32 @@ bool AFLLTOPass::runOnModule(Module &M) { /* Update bitmap */ - LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); - Counter->setMetadata(M.getMDKindID("nosanitize"), - MDNode::get(C, None)); + if (use_threadsafe_counters) { + + IRB.CreateAtomicRMW(llvm::AtomicRMWInst::BinOp::Add, MapPtrIdx, One, + llvm::AtomicOrdering::Monotonic); - Value *Incr = IRB.CreateAdd(Counter, One); + } else { - if (skip_nozero == NULL) { + LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); + Counter->setMetadata(M.getMDKindID("nosanitize"), + MDNode::get(C, None)); - auto cf = IRB.CreateICmpEQ(Incr, Zero); - auto carry = IRB.CreateZExt(cf, Int8Ty); - Incr = IRB.CreateAdd(Incr, carry); + Value *Incr = IRB.CreateAdd(Counter, One); - } + if (skip_nozero == NULL) { - IRB.CreateStore(Incr, MapPtrIdx) - ->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); + auto cf = IRB.CreateICmpEQ(Incr, Zero); + auto carry = IRB.CreateZExt(cf, Int8Ty); + Incr = IRB.CreateAdd(Incr, carry); + + } + + IRB.CreateStore(Incr, MapPtrIdx) + ->setMetadata(M.getMDKindID("nosanitize"), + MDNode::get(C, None)); + + } // done :) diff --git a/instrumentation/afl-llvm-pass.so.cc b/instrumentation/afl-llvm-pass.so.cc index 0f773aba..a8f1baff 100644 --- a/instrumentation/afl-llvm-pass.so.cc +++ b/instrumentation/afl-llvm-pass.so.cc @@ -81,11 +81,12 @@ class AFLCoverage : public ModulePass { bool runOnModule(Module &M) override; protected: - uint32_t ngram_size = 0; - uint32_t ctx_k = 0; - uint32_t map_size = MAP_SIZE; - uint32_t function_minimum_size = 1; - char * ctx_str = NULL, *caller_str = NULL, *skip_nozero = NULL; + uint32_t ngram_size = 0; + uint32_t ctx_k = 0; + uint32_t map_size = MAP_SIZE; + uint32_t function_minimum_size = 1; + const char *ctx_str = NULL, *caller_str = NULL, *skip_nozero = NULL; + const char *use_threadsafe_counters = nullptr; }; @@ -182,6 +183,38 @@ bool AFLCoverage::runOnModule(Module &M) { char *neverZero_counters_str = getenv("AFL_LLVM_NOT_ZERO"); #endif skip_nozero = getenv("AFL_LLVM_SKIP_NEVERZERO"); + use_threadsafe_counters = getenv("AFL_LLVM_THREADSAFE_INST"); + + if ((isatty(2) && !getenv("AFL_QUIET")) || !!getenv("AFL_DEBUG")) { + + if (use_threadsafe_counters) { + + // disabled unless there is support for other modules as well + // (increases documentation complexity) + /* if (!getenv("AFL_LLVM_NOT_ZERO")) { */ + + skip_nozero = "1"; + SAYF(cCYA "afl-llvm-pass" VERSION cRST " using thread safe counters\n"); + + /* + + } else { + + SAYF(cCYA "afl-llvm-pass" VERSION cRST + " using thread safe not-zero-counters\n"); + + } + + */ + + } else { + + SAYF(cCYA "afl-llvm-pass" VERSION cRST + " using non-thread safe instrumentation\n"); + + } + + } unsigned PrevLocSize = 0; unsigned PrevCallerSize = 0; @@ -388,7 +421,6 @@ bool AFLCoverage::runOnModule(Module &M) { #endif // other constants we need - ConstantInt *Zero = ConstantInt::get(Int8Ty, 0); ConstantInt *One = ConstantInt::get(Int8Ty, 1); Value * PrevCtx = NULL; // CTX sensitive coverage @@ -410,6 +442,7 @@ bool AFLCoverage::runOnModule(Module &M) { if (F.size() < function_minimum_size) continue; + std::list<Value *> todo; for (auto &BB : F) { BasicBlock::iterator IP = BB.getFirstInsertionPt(); @@ -628,37 +661,68 @@ bool AFLCoverage::runOnModule(Module &M) { /* Update bitmap */ - LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); - Counter->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); + if (use_threadsafe_counters) { /* Atomic */ + /* + #if LLVM_VERSION_MAJOR < 9 + if (neverZero_counters_str != + NULL) { // with llvm 9 we make this the default as the bug + in llvm + // is then fixed + #else + if (!skip_nozero) { + + #endif + // register MapPtrIdx in a todo list + todo.push_back(MapPtrIdx); + + } else { + + */ + IRB.CreateAtomicRMW(llvm::AtomicRMWInst::BinOp::Add, MapPtrIdx, One, + llvm::AtomicOrdering::Monotonic); + /* + + } + + */ - Value *Incr = IRB.CreateAdd(Counter, One); + } else { + + LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); + Counter->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); + + Value *Incr = IRB.CreateAdd(Counter, One); #if LLVM_VERSION_MAJOR < 9 - if (neverZero_counters_str != - NULL) { // with llvm 9 we make this the default as the bug in llvm is - // then fixed + if (neverZero_counters_str != + NULL) { // with llvm 9 we make this the default as the bug in llvm + // is then fixed #else - if (!skip_nozero) { + if (!skip_nozero) { #endif - /* hexcoder: Realize a counter that skips zero during overflow. - * Once this counter reaches its maximum value, it next increments to 1 - * - * Instead of - * Counter + 1 -> Counter - * we inject now this - * Counter + 1 -> {Counter, OverflowFlag} - * Counter + OverflowFlag -> Counter - */ + /* hexcoder: Realize a counter that skips zero during overflow. + * Once this counter reaches its maximum value, it next increments to + * 1 + * + * Instead of + * Counter + 1 -> Counter + * we inject now this + * Counter + 1 -> {Counter, OverflowFlag} + * Counter + OverflowFlag -> Counter + */ + + ConstantInt *Zero = ConstantInt::get(Int8Ty, 0); + auto cf = IRB.CreateICmpEQ(Incr, Zero); + auto carry = IRB.CreateZExt(cf, Int8Ty); + Incr = IRB.CreateAdd(Incr, carry); - auto cf = IRB.CreateICmpEQ(Incr, Zero); - auto carry = IRB.CreateZExt(cf, Int8Ty); - Incr = IRB.CreateAdd(Incr, carry); + } - } + IRB.CreateStore(Incr, MapPtrIdx) + ->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); - IRB.CreateStore(Incr, MapPtrIdx) - ->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); + } /* non atomic case */ /* Update prev_loc history vector (by placing cur_loc at the head of the vector and shuffle the other elements back by one) */ @@ -715,6 +779,120 @@ bool AFLCoverage::runOnModule(Module &M) { } +#if 0 + if (use_threadsafe_counters) { /*Atomic NeverZero */ + // handle the list of registered blocks to instrument + for (auto val : todo) { + + /* hexcoder: Realize a thread-safe counter that skips zero during + * overflow. Once this counter reaches its maximum value, it next + * increments to 1 + * + * Instead of + * Counter + 1 -> Counter + * we inject now this + * Counter + 1 -> {Counter, OverflowFlag} + * Counter + OverflowFlag -> Counter + */ + + /* equivalent c code looks like this + * Thanks to + https://preshing.com/20150402/you-can-do-any-kind-of-atomic-read-modify-write-operation/ + + int old = atomic_load_explicit(&Counter, memory_order_relaxed); + int new; + do { + + if (old == 255) { + + new = 1; + + } else { + + new = old + 1; + + } + + } while (!atomic_compare_exchange_weak_explicit(&Counter, &old, new, + + memory_order_relaxed, memory_order_relaxed)); + + */ + + Value * MapPtrIdx = val; + Instruction * MapPtrIdxInst = cast<Instruction>(val); + BasicBlock::iterator it0(&(*MapPtrIdxInst)); + ++it0; + IRBuilder<> IRB(&(*it0)); + + // load the old counter value atomically + LoadInst *Counter = IRB.CreateLoad(MapPtrIdx); + Counter->setAlignment(llvm::Align()); + Counter->setAtomic(llvm::AtomicOrdering::Monotonic); + Counter->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); + + BasicBlock *BB = IRB.GetInsertBlock(); + // insert a basic block with the corpus of a do while loop + // the calculation may need to repeat, if atomic compare_exchange is not + // successful + + BasicBlock::iterator it(*Counter); + it++; // split after load counter + BasicBlock *end_bb = BB->splitBasicBlock(it); + end_bb->setName("injected"); + + // insert the block before the second half of the split + BasicBlock *do_while_bb = + BasicBlock::Create(C, "injected", end_bb->getParent(), end_bb); + + // set terminator of BB from target end_bb to target do_while_bb + auto term = BB->getTerminator(); + BranchInst::Create(do_while_bb, BB); + term->eraseFromParent(); + + // continue to fill instructions into the do_while loop + IRB.SetInsertPoint(do_while_bb, do_while_bb->getFirstInsertionPt()); + + PHINode *PN = IRB.CreatePHI(Int8Ty, 2); + + // compare with maximum value 0xff + auto *Cmp = IRB.CreateICmpEQ(Counter, ConstantInt::get(Int8Ty, -1)); + + // increment the counter + Value *Incr = IRB.CreateAdd(Counter, One); + + // select the counter value or 1 + auto *Select = IRB.CreateSelect(Cmp, One, Incr); + + // try to save back the new counter value + auto *CmpXchg = IRB.CreateAtomicCmpXchg( + MapPtrIdx, PN, Select, llvm::AtomicOrdering::Monotonic, + llvm::AtomicOrdering::Monotonic); + CmpXchg->setAlignment(llvm::Align()); + CmpXchg->setWeak(true); + CmpXchg->setMetadata(M.getMDKindID("nosanitize"), MDNode::get(C, None)); + + // get the result of trying to update the Counter + Value *Success = + IRB.CreateExtractValue(CmpXchg, ArrayRef<unsigned>({1})); + // get the (possibly updated) value of Counter + Value *OldVal = + IRB.CreateExtractValue(CmpXchg, ArrayRef<unsigned>({0})); + + // initially we use Counter + PN->addIncoming(Counter, BB); + // on retry, we use the updated value + PN->addIncoming(OldVal, do_while_bb); + + // if the cmpXchg was not successful, retry + IRB.CreateCondBr(Success, end_bb, do_while_bb); + + } + + } + +#endif + } /* |
