Files
AFLplusplus/instrumentation/split-switches-pass.so.cc
van Hauser fff7f1c558 Dev (#1962)
* Pure Python (3.6) port of benchmark.sh as benchmark.py, no other changes

* Test standard and persistent modes separately

* Add support for multi-core benchmarking

* Save the results to a json file

* Allow config of all experiment params, average across runs

* Add start_time_of_run and total_execs_per_sec, cleanup for PR

* benchmark: cleanup, add results, add a data exploration notebook

* benchmark: add a README, lower default runs from 5 to 3

* benchmark: notebook wording tweaks

* copy 'detect_leaks=0' from ASAN to LSAN

fix for issue #1733, set "detect_leaks=0" when ASAN_OPTIONS contains it and LSAN_OPTIONS are not set.

* fix of fix: make sure ASAN_OPTIONS and LSAN_OPTIONS agree on leak detection

* fix lsan fix

* clang-format 16->17

* Add missing initialisation for havoc_queued during the custom mutator's stage.

* fix dictionary and cmin

* Use direct call to write to OpenBSD

The linker on OpenBSD emits a warning when linking this file:
warning: syscall() may go away, please rewrite code to use direct calls

* Fix possible doc inconsistency for custom mutator's queue_get function.

* update todos

* benchmark: Add support for COMPARISON file

* benchmark: show the number of cores used in COMPARISON

* benchmark: lower minimum Python version to 3.8

* benchmark: use afl's execs/s; increase CPU model width

* benchmark: disallow duplicate entries for the same CPU in COMPARISON

* Update benchmark.py

* fix inf in stats

* Fix benchmark.py

* missing closing parenthesis

* Update benchmark.py

* benchmark: remove self-calculation of execs/sec

* benchmark: update COMPARISON

* benchmark: Update Jupyter notebook and results file.

* benchmark: rename afl_execs_per_sec to execs_per_sec

* benchmark: update README

* update

* add benchmark

* nits

* add benchmarks

* Update unicornafl ref

* Pass correct Nyx ID when creating a Nyx runner

* Fix typo in docker pull command, add exampe to mount current dir as volume (#1914)

* mini fix

* add custom_post_run.c

* update afl-fuzz-run

* update python module

* format code

* update

* merge function

* changes

* code format

* improve cmplog

* nit

* nit

* fix

* fix

* Stop hardcoding the path /usr/local/lib/afl in afl-ld-lto.c and respect the configured PREFIX.

* Add benchmark for Raspberry Pi 5

* ryzen 5950 benchmark

* add missing raspery5

* comparison -> comparison.md

* removing options "-Wl,-rpath" "LLVM_LIBDIR" when using gcc

* fixing -Wl,-rpath=<LLVM_LIBDIR>

* nits

* fix

* afl-cc fixes

* nit

* add n_fuzz to ignore_timeouts

* fix

* Fix #1927

* in-depth blog post

* add AFL_FUZZER_LOOPCOUNT

* AFL_FUZZER_LOOPCOUNT

* fix 2 mutation bugs

* v4.09c release

* v4.10a init

* switch to explore powerschedule as default

* fix MUT_INSERTASCIINUM

* fix MUT_STRATEGY_ARRAY_SIZE

* fix bad fix for MUT_STRATEGY_ARRAY_SIZE

* remove afl-network-client on uninstall

* update nyx

* Improve binary-only related docs

* llvm 18 build fixes.

* code format

* Fix custom_send link

Add a leading '/' to walk in the repo root instead of current dir.

* Use ../ instead

* initial simple injection detection support

* inject docs

* fix for issue #1916, iLLVM crash in split-floatingpoint-compares

* LLVM 17 bug workaround

* finish injection implementation

* remove tmp todo

* update changelog

* forgot to add the injection pass

* Output afl-clang-fast stuffs only if necessary (#1912)

* afl-cc header

* afl-cc common declarations

 - Add afl-cc-state.c
 - Strip includes, find_object, debug/be_quiet/have_*/callname setting from afl-cc.c
 - Use debugf_args in main
 - Modify execvp stuffs to fit new aflcc struct

* afl-cc show usage

* afl-cc mode selecting

1. compiler_mode by callname in argv[0]
2. compiler_mode by env "AFL_CC_COMPILER"
3. compiler_mode/instrument_mode by command line options "--afl-..."
4. instrument_mode/compiler_mode by various env vars including "AFL_LLVM_INSTRUMENT"
5. final checking steps
6. print "... - mode: %s-%s\n"
7. determine real argv[0] according to compiler_mode

* afl-cc macro defs

* afl-cc linking behaviors

* afl-cc fsanitize behaviors

* afl-cc misc

* afl-cc body update

* afl-cc all-in-one

formated with custom-format.py

* nits

---------

Co-authored-by: vanhauser-thc <vh@thc.org>

* changelog

* update grammar mutator

* lto llvm 12+

* docs(custom_mutators): fix missing ':' (#1953)

* Fix broken LTO mode and response file support (#1948)

* Strip `-Wl,-no-undefined` during compilation (#1952)

Make the compiler wrapper stripping `-Wl,-no-undefined` in addition to `-Wl,--no-undefined`.
Both versions of the flag are accepted by clang and, therefore, used by building systems in the wild (e.g., samba will not build without this fix).

* Remove dead code in write_to_testcase (#1955)

The custom_mutators_count check in if case is duplicate with if condition.
The else case is custom_mutators_count == 0, neither custom_mutator_list iteration nor sent check needed.

Signed-off-by: Xeonacid <h.dwwwwww@gmail.com>

* update qemuafl

* WIP: Add ability to generate drcov trace using QEMU backend (#1956)

* Document new drcov QEMU plugin

* Add link to lightkeeper for QEMU drcov file loading

---------

Co-authored-by: Jean-Romain Garnier <jean-romain.garnier@airbus.com>

* code format

* changelog

* sleep on uid != 0 afl-system-config

* fix segv about skip_next, warn on unsupported cases of linking options (#1958)

* todos

* ensure afl-cc only allows available compiler modes

* update grammar mutator

* disable aslr on apple

* fix for arm64

* help selective instrumentation

* typos

* macos

* add compiler test script

* apple fixes

---------

Signed-off-by: Xeonacid <h.dwwwwww@gmail.com>
Co-authored-by: Chris Ball <chris@printf.net>
Co-authored-by: hexcoder <hexcoder-@users.noreply.github.com>
Co-authored-by: hexcoder- <heiko@hexco.de>
Co-authored-by: Manuel Carrasco <m.carrasco@imperial.ac.uk>
Co-authored-by: Jasper Lievisse Adriaanse <j@jasper.la>
Co-authored-by: ifyGecko <26214995+ifyGecko@users.noreply.github.com>
Co-authored-by: Dominik Maier <domenukk@gmail.com>
Co-authored-by: Christian Holler (:decoder) <choller@mozilla.com>
Co-authored-by: Carlo Maragno <ste.maragno@gmail.com>
Co-authored-by: yangzao <yangzaocn@outlook.com>
Co-authored-by: Romain Geissler <romain.geissler@amadeus.com>
Co-authored-by: Jakob Lell <jakob@jakoblell.com>
Co-authored-by: vincenzo MEZZELA <vincenzo.mezzela@amadeus.com>
Co-authored-by: Andrea Fioraldi <andreafioraldi@gmail.com>
Co-authored-by: Bet4 <0xbet4@gmail.com>
Co-authored-by: David Carlier <devnexen@gmail.com>
Co-authored-by: Xeonacid <h.dwwwwww@gmail.com>
Co-authored-by: Sonic <50692172+SonicStark@users.noreply.github.com>
Co-authored-by: Nils Bars <nils.bars@rub.de>
Co-authored-by: Jean-Romain Garnier <7504819+JRomainG@users.noreply.github.com>
Co-authored-by: Jean-Romain Garnier <jean-romain.garnier@airbus.com>
2024-01-20 10:19:46 +00:00

562 lines
15 KiB
C++

/*
* Copyright 2016 laf-intel
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* https://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <list>
#include <string>
#include <fstream>
#include <sys/time.h>
#include "llvm/Config/llvm-config.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/IR/IRBuilder.h"
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
#include "llvm/Passes/PassPlugin.h"
#include "llvm/Passes/PassBuilder.h"
#include "llvm/IR/PassManager.h"
#else
#include "llvm/IR/LegacyPassManager.h"
#include "llvm/Transforms/IPO/PassManagerBuilder.h"
#endif
#include "llvm/IR/Module.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/ValueTracking.h"
#if LLVM_VERSION_MAJOR >= 14 /* how about stable interfaces? */
#include "llvm/Passes/OptimizationLevel.h"
#endif
#include "llvm/IR/IRBuilder.h"
#if LLVM_VERSION_MAJOR >= 4 || \
(LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR > 4)
#include "llvm/IR/Verifier.h"
#include "llvm/IR/DebugInfo.h"
#else
#include "llvm/Analysis/Verifier.h"
#include "llvm/DebugInfo.h"
#define nullptr 0
#endif
#include <set>
#include "afl-llvm-common.h"
using namespace llvm;
namespace {
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
class SplitSwitchesTransform : public PassInfoMixin<SplitSwitchesTransform> {
public:
SplitSwitchesTransform() {
#else
class SplitSwitchesTransform : public ModulePass {
public:
static char ID;
SplitSwitchesTransform() : ModulePass(ID) {
#endif
initInstrumentList();
}
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
PreservedAnalyses run(Module &M, ModuleAnalysisManager &MAM);
#else
bool runOnModule(Module &M) override;
#if LLVM_VERSION_MAJOR >= 4
StringRef getPassName() const override {
#else
const char *getPassName() const override {
#endif
return "splits switch constructs";
}
#endif
struct CaseExpr {
ConstantInt *Val;
BasicBlock *BB;
CaseExpr(ConstantInt *val = nullptr, BasicBlock *bb = nullptr)
: Val(val), BB(bb) {
}
};
using CaseVector = std::vector<CaseExpr>;
private:
bool splitSwitches(Module &M);
bool transformCmps(Module &M, const bool processStrcmp,
const bool processMemcmp);
BasicBlock *switchConvert(CaseVector Cases, std::vector<bool> bytesChecked,
BasicBlock *OrigBlock, BasicBlock *NewDefault,
Value *Val, unsigned level);
};
} // namespace
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
extern "C" ::llvm::PassPluginLibraryInfo LLVM_ATTRIBUTE_WEAK
llvmGetPassPluginInfo() {
return {LLVM_PLUGIN_API_VERSION, "splitswitches", "v0.1",
/* lambda to insert our pass into the pass pipeline. */
[](PassBuilder &PB) {
#if 1
#if LLVM_VERSION_MAJOR <= 13
using OptimizationLevel = typename PassBuilder::OptimizationLevel;
#endif
PB.registerOptimizerLastEPCallback(
[](ModulePassManager &MPM, OptimizationLevel OL) {
MPM.addPass(SplitSwitchesTransform());
});
/* TODO LTO registration */
#else
using PipelineElement = typename PassBuilder::PipelineElement;
PB.registerPipelineParsingCallback([](StringRef Name,
ModulePassManager &MPM,
ArrayRef<PipelineElement>) {
if (Name == "splitswitches") {
MPM.addPass(SplitSwitchesTransform());
return true;
} else {
return false;
}
});
#endif
}};
}
#else
char SplitSwitchesTransform::ID = 0;
#endif
/* switchConvert - Transform simple list of Cases into list of CaseRange's */
BasicBlock *SplitSwitchesTransform::switchConvert(
CaseVector Cases, std::vector<bool> bytesChecked, BasicBlock *OrigBlock,
BasicBlock *NewDefault, Value *Val, unsigned level) {
unsigned ValTypeBitWidth = Cases[0].Val->getBitWidth();
IntegerType *ValType =
IntegerType::get(OrigBlock->getContext(), ValTypeBitWidth);
IntegerType *ByteType = IntegerType::get(OrigBlock->getContext(), 8);
unsigned BytesInValue = bytesChecked.size();
std::vector<uint8_t> setSizes;
std::vector<std::set<uint8_t> > byteSets(BytesInValue, std::set<uint8_t>());
/* for each of the possible cases we iterate over all bytes of the values
* build a set of possible values at each byte position in byteSets */
for (CaseExpr &Case : Cases) {
for (unsigned i = 0; i < BytesInValue; i++) {
uint8_t byte = (Case.Val->getZExtValue() >> (i * 8)) & 0xFF;
byteSets[i].insert(byte);
}
}
/* find the index of the first byte position that was not yet checked. then
* save the number of possible values at that byte position */
unsigned smallestIndex = 0;
unsigned smallestSize = 257;
for (unsigned i = 0; i < byteSets.size(); i++) {
if (bytesChecked[i]) continue;
if (byteSets[i].size() < smallestSize) {
smallestIndex = i;
smallestSize = byteSets[i].size();
}
}
assert(bytesChecked[smallestIndex] == false);
/* there are only smallestSize different bytes at index smallestIndex */
Instruction *Shift, *Trunc;
Function *F = OrigBlock->getParent();
BasicBlock *NewNode = BasicBlock::Create(Val->getContext(), "NodeBlock", F);
Shift = BinaryOperator::Create(Instruction::LShr, Val,
ConstantInt::get(ValType, smallestIndex * 8));
#if LLVM_VERSION_MAJOR >= 16
Shift->insertInto(NewNode, NewNode->end());
#else
NewNode->getInstList().push_back(Shift);
#endif
if (ValTypeBitWidth > 8) {
Trunc = new TruncInst(Shift, ByteType);
#if LLVM_VERSION_MAJOR >= 16
Trunc->insertInto(NewNode, NewNode->end());
#else
NewNode->getInstList().push_back(Trunc);
#endif
} else {
/* not necessary to trunc */
Trunc = Shift;
}
/* this is a trivial case, we can directly check for the byte,
* if the byte is not found go to default. if the byte was found
* mark the byte as checked. if this was the last byte to check
* we can finally execute the block belonging to this case */
if (smallestSize == 1) {
uint8_t byte = *(byteSets[smallestIndex].begin());
/* insert instructions to check whether the value we are switching on is
* equal to byte */
ICmpInst *Comp =
new ICmpInst(ICmpInst::ICMP_EQ, Trunc, ConstantInt::get(ByteType, byte),
"byteMatch");
#if LLVM_VERSION_MAJOR >= 16
Comp->insertInto(NewNode, NewNode->end());
#else
NewNode->getInstList().push_back(Comp);
#endif
bytesChecked[smallestIndex] = true;
bool allBytesAreChecked = true;
for (std::vector<bool>::iterator BCI = bytesChecked.begin(),
E = bytesChecked.end();
BCI != E; ++BCI) {
if (!*BCI) {
allBytesAreChecked = false;
break;
}
}
// if (std::all_of(bytesChecked.begin(), bytesChecked.end(),
// [](bool b) { return b; })) {
if (allBytesAreChecked) {
assert(Cases.size() == 1);
BranchInst::Create(Cases[0].BB, NewDefault, Comp, NewNode);
/* we have to update the phi nodes! */
for (BasicBlock::iterator I = Cases[0].BB->begin();
I != Cases[0].BB->end(); ++I) {
if (!isa<PHINode>(&*I)) { continue; }
PHINode *PN = cast<PHINode>(I);
/* Only update the first occurrence. */
unsigned Idx = 0, E = PN->getNumIncomingValues();
for (; Idx != E; ++Idx) {
if (PN->getIncomingBlock(Idx) == OrigBlock) {
PN->setIncomingBlock(Idx, NewNode);
break;
}
}
}
} else {
BasicBlock *BB = switchConvert(Cases, bytesChecked, OrigBlock, NewDefault,
Val, level + 1);
BranchInst::Create(BB, NewDefault, Comp, NewNode);
}
}
/* there is no byte which we can directly check on, split the tree */
else {
std::vector<uint8_t> byteVector;
std::copy(byteSets[smallestIndex].begin(), byteSets[smallestIndex].end(),
std::back_inserter(byteVector));
std::sort(byteVector.begin(), byteVector.end());
uint8_t pivot = byteVector[byteVector.size() / 2];
/* we already chose to divide the cases based on the value of byte at index
* smallestIndex the pivot value determines the threshold for the decicion;
* if a case value
* is smaller at this byte index move it to the LHS vector, otherwise to the
* RHS vector */
CaseVector LHSCases, RHSCases;
for (CaseExpr &Case : Cases) {
uint8_t byte = (Case.Val->getZExtValue() >> (smallestIndex * 8)) & 0xFF;
if (byte < pivot) {
LHSCases.push_back(Case);
} else {
RHSCases.push_back(Case);
}
}
BasicBlock *LBB, *RBB;
LBB = switchConvert(LHSCases, bytesChecked, OrigBlock, NewDefault, Val,
level + 1);
RBB = switchConvert(RHSCases, bytesChecked, OrigBlock, NewDefault, Val,
level + 1);
/* insert instructions to check whether the value we are switching on is
* equal to byte */
ICmpInst *Comp =
new ICmpInst(ICmpInst::ICMP_ULT, Trunc,
ConstantInt::get(ByteType, pivot), "byteMatch");
#if LLVM_VERSION_MAJOR >= 16
Comp->insertInto(NewNode, NewNode->end());
#else
NewNode->getInstList().push_back(Comp);
#endif
BranchInst::Create(LBB, RBB, Comp, NewNode);
}
return NewNode;
}
bool SplitSwitchesTransform::splitSwitches(Module &M) {
#if (LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR < 7)
LLVMContext &C = M.getContext();
#endif
std::vector<SwitchInst *> switches;
/* iterate over all functions, bbs and instruction and add
* all switches to switches vector for later processing */
for (auto &F : M) {
if (!isInInstrumentList(&F, MNAME)) continue;
for (auto &BB : F) {
SwitchInst *switchInst = nullptr;
if ((switchInst = dyn_cast<SwitchInst>(BB.getTerminator()))) {
if (switchInst->getNumCases() < 1) continue;
switches.push_back(switchInst);
}
}
}
if (!switches.size()) return false;
/*
if (!be_quiet)
errs() << "Rewriting " << switches.size() << " switch statements "
<< "\n";
*/
for (auto &SI : switches) {
BasicBlock *CurBlock = SI->getParent();
BasicBlock *OrigBlock = CurBlock;
Function *F = CurBlock->getParent();
/* this is the value we are switching on */
Value *Val = SI->getCondition();
BasicBlock *Default = SI->getDefaultDest();
unsigned bitw = Val->getType()->getIntegerBitWidth();
/*
if (!be_quiet)
errs() << "switch: " << SI->getNumCases() << " cases " << bitw
<< " bit\n";
*/
/* If there is only the default destination or the condition checks 8 bit or
* less, don't bother with the code below. */
if (SI->getNumCases() < 2 || bitw % 8 || bitw > 64) {
// if (!be_quiet) errs() << "skip switch..\n";
continue;
}
/* Create a new, empty default block so that the new hierarchy of
* if-then statements go to this and the PHI nodes are happy.
* if the default block is set as an unreachable we avoid creating one
* because will never be a valid target.*/
BasicBlock *NewDefault = nullptr;
NewDefault = BasicBlock::Create(SI->getContext(), "NewDefault", F, Default);
BranchInst::Create(Default, NewDefault);
/* Prepare cases vector. */
CaseVector Cases;
for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end(); i != e;
++i)
#if LLVM_VERSION_MAJOR >= 5
Cases.push_back(CaseExpr(i->getCaseValue(), i->getCaseSuccessor()));
#else
Cases.push_back(CaseExpr(i.getCaseValue(), i.getCaseSuccessor()));
#endif
/* bugfix thanks to pbst
* round up bytesChecked (in case getBitWidth() % 8 != 0) */
std::vector<bool> bytesChecked((7 + Cases[0].Val->getBitWidth()) / 8,
false);
BasicBlock *SwitchBlock =
switchConvert(Cases, bytesChecked, OrigBlock, NewDefault, Val, 0);
/* Branch to our shiny new if-then stuff... */
BranchInst::Create(SwitchBlock, OrigBlock);
/* We are now done with the switch instruction, delete it. */
#if LLVM_VERSION_MAJOR >= 16
SI->eraseFromParent();
#else
CurBlock->getInstList().erase(SI);
#endif
/* we have to update the phi nodes! */
for (BasicBlock::iterator I = Default->begin(); I != Default->end(); ++I) {
if (!isa<PHINode>(&*I)) { continue; }
PHINode *PN = cast<PHINode>(I);
/* Only update the first occurrence. */
unsigned Idx = 0, E = PN->getNumIncomingValues();
for (; Idx != E; ++Idx) {
if (PN->getIncomingBlock(Idx) == OrigBlock) {
PN->setIncomingBlock(Idx, NewDefault);
break;
}
}
}
}
verifyModule(M);
return true;
}
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
PreservedAnalyses SplitSwitchesTransform::run(Module &M,
ModuleAnalysisManager &MAM) {
#else
bool SplitSwitchesTransform::runOnModule(Module &M) {
#endif
if ((isatty(2) && getenv("AFL_QUIET") == NULL) || getenv("AFL_DEBUG") != NULL)
printf("Running split-switches-pass by laf.intel@gmail.com\n");
else
be_quiet = 1;
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
auto PA = PreservedAnalyses::all();
#endif
splitSwitches(M);
verifyModule(M);
#if LLVM_VERSION_MAJOR >= 11 /* use new pass manager */
/* if (modified) {
PA.abandon<XX_Manager>();
}*/
return PA;
#else
return true;
#endif
}
#if LLVM_VERSION_MAJOR < 11 /* use old pass manager */
static void registerSplitSwitchesTransPass(const PassManagerBuilder &,
legacy::PassManagerBase &PM) {
auto p = new SplitSwitchesTransform();
PM.add(p);
}
static RegisterStandardPasses RegisterSplitSwitchesTransPass(
PassManagerBuilder::EP_OptimizerLast, registerSplitSwitchesTransPass);
static RegisterStandardPasses RegisterSplitSwitchesTransPass0(
PassManagerBuilder::EP_EnabledOnOptLevel0, registerSplitSwitchesTransPass);
#if LLVM_VERSION_MAJOR >= 11
static RegisterStandardPasses RegisterSplitSwitchesTransPassLTO(
PassManagerBuilder::EP_FullLinkTimeOptimizationLast,
registerSplitSwitchesTransPass);
#endif
#endif