about summary refs log tree commit diff
path: root/llvm_mode
diff options
context:
space:
mode:
Diffstat (limited to 'llvm_mode')
-rw-r--r--llvm_mode/GNUmakefile3
-rw-r--r--llvm_mode/README.lto.md41
-rw-r--r--llvm_mode/afl-clang-fast.c3
-rw-r--r--llvm_mode/afl-llvm-lto-instrumentation.so.cc330
-rw-r--r--llvm_mode/afl-llvm-rt.o.c170
5 files changed, 485 insertions, 62 deletions
diff --git a/llvm_mode/GNUmakefile b/llvm_mode/GNUmakefile
index 7432b061..b176a24f 100644
--- a/llvm_mode/GNUmakefile
+++ b/llvm_mode/GNUmakefile
@@ -273,6 +273,7 @@ endif
 ../afl-llvm-lto-instrumentation.so: afl-llvm-lto-instrumentation.so.cc
 ifeq "$(LLVM_LTO)" "1"
 	$(CXX) $(CLANG_CFL) -Wno-writable-strings -fno-rtti -fPIC -std=$(LLVM_STDCXX) -shared $< -o $@ $(CLANG_LFL)
+	$(CC) $(CFLAGS) -O0 $(AFL_CLANG_FLTO) -fPIC -c afl-llvm-rt-lto.o.c -o ../afl-llvm-rt-lto.o
 endif
 
 # laf
@@ -318,7 +319,7 @@ all_done: test_build
 install: all
 	install -d -m 755 $${DESTDIR}$(BIN_PATH) $${DESTDIR}$(HELPER_PATH) $${DESTDIR}$(DOC_PATH) $${DESTDIR}$(MISC_PATH)
 	if [ -f ../afl-clang-fast -a -f ../libLLVMInsTrim.so -a -f ../afl-llvm-rt.o ]; then set -e; install -m 755 ../afl-clang-fast $${DESTDIR}$(BIN_PATH); ln -sf afl-clang-fast $${DESTDIR}$(BIN_PATH)/afl-clang-fast++; install -m 755 ../libLLVMInsTrim.so ../afl-llvm-pass.so ../afl-llvm-rt.o $${DESTDIR}$(HELPER_PATH); fi
-	if [ -f ../afl-clang-lto ]; then set -e; ln -sf afl-clang-fast $${DESTDIR}$(BIN_PATH)/afl-clang-lto; ln -sf afl-clang-fast $${DESTDIR}$(BIN_PATH)/afl-clang-lto++; install -m 755 ../afl-llvm-lto-instrumentation.so $${DESTDIR}$(HELPER_PATH); install -m 755 ../afl-llvm-lto-whitelist.so $${DESTDIR}$(HELPER_PATH); fi
+	if [ -f ../afl-clang-lto ]; then set -e; ln -sf afl-clang-fast $${DESTDIR}$(BIN_PATH)/afl-clang-lto; ln -sf afl-clang-fast $${DESTDIR}$(BIN_PATH)/afl-clang-lto++; install -m 755 ../afl-llvm-lto-instrumentation.so ../afl-llvm-rt-lto.o ../afl-llvm-lto-whitelist.so $${DESTDIR}$(HELPER_PATH); fi
 	if [ -f ../afl-llvm-rt-32.o ]; then set -e; install -m 755 ../afl-llvm-rt-32.o $${DESTDIR}$(HELPER_PATH); fi
 	if [ -f ../afl-llvm-rt-64.o ]; then set -e; install -m 755 ../afl-llvm-rt-64.o $${DESTDIR}$(HELPER_PATH); fi
 	if [ -f ../compare-transform-pass.so ]; then set -e; install -m 755 ../compare-transform-pass.so $${DESTDIR}$(HELPER_PATH); fi
diff --git a/llvm_mode/README.lto.md b/llvm_mode/README.lto.md
index a3c7ddc3..48d0e36c 100644
--- a/llvm_mode/README.lto.md
+++ b/llvm_mode/README.lto.md
@@ -12,6 +12,8 @@ This version requires a current llvm 11 compiled from the github master.
 
 3. It only works with llvm 11 (current github master state)
 
+4. AUTODICTIONARY feature! see below
+
 ## Introduction and problem description
 
 A big issue with how afl/afl++ works is that the basic block IDs that are
@@ -33,33 +35,22 @@ and many dead ends until we got to this:
  * Our compiler (afl-clang-lto/afl-clang-lto++) takes care of setting the
    correct LTO options and runs our own afl-ld linker instead of the system
    linker
- * Our linker collects all LTO files to link and instruments them so that
+ * The LLVM linker collects all LTO files to link and instruments them so that
    we have non-colliding edge overage
  * We use a new (for afl) edge coverage - which is the same as in llvm
    -fsanitize=coverage edge coverage mode :)
- * after inserting our instrumentation in all interesting edges we link
-   all parts of the program together to our executable
 
 The result:
- * 10-15% speed gain compared to llvm_mode
+ * 10-20% speed gain compared to llvm_mode
  * guaranteed non-colliding edge coverage :-)
  * The compile time especially for libraries can be longer
 
 Example build output from a libtiff build:
 ```
-/bin/bash ../libtool  --tag=CC   --mode=link afl-clang-lto  -g -O2 -Wall -W   -o thumbnail thumbnail.o ../libtiff/libtiff.la ../port/libport.la -llzma -ljbig -ljpeg -lz -lm 
 libtool: link: afl-clang-lto -g -O2 -Wall -W -o thumbnail thumbnail.o  ../libtiff/.libs/libtiff.a ../port/.libs/libport.a -llzma -ljbig -ljpeg -lz -lm
-afl-clang-lto++2.62d by Marc "vanHauser" Heuse <mh@mh-sec.de>
-afl-ld++2.62d by Marc "vanHauser" Heuse <mh@mh-sec.de> (level 0)
-[+] Running ar unpacker on /prg/tests/lto/tiff-4.0.4/tools/../libtiff/.libs/libtiff.a into /tmp/.afl-3914343-1583339800.dir
-[+] Running ar unpacker on /prg/tests/lto/tiff-4.0.4/tools/../port/.libs/libport.a into /tmp/.afl-3914343-1583339800.dir
-[+] Running bitcode linker, creating /tmp/.afl-3914343-1583339800-1.ll
-[+] Performing optimization via opt, creating /tmp/.afl-3914343-1583339800-2.bc
-[+] Performing instrumentation via opt, creating /tmp/.afl-3914343-1583339800-3.bc
-afl-llvm-lto++2.62d by Marc "vanHauser" Heuse <mh@mh-sec.de>
-[+] Instrumented 15833 locations with no collisions (on average 1767 collisions would be in afl-gcc/afl-clang-fast) (non-hardened mode).
-[+] Running real linker /bin/x86_64-linux-gnu-ld
-[+] Linker was successful
+afl-clang-lto++2.63d by Marc "vanHauser" Heuse <mh@mh-sec.de> in mode LTO
+afl-llvm-lto++2.63d by Marc "vanHauser" Heuse <mh@mh-sec.de>
+[+] Instrumented 11836 locations with no collisions (on average 1007 collisions would be in afl-gcc/afl-clang-fast) (non-hardened mode).
 ```
 
 ## Building llvm 11
@@ -70,8 +61,8 @@ $ git clone https://github.com/llvm/llvm-project
 $ cd llvm-project
 $ mkdir build
 $ cd build
-$ cmake -DLLVM_ENABLE_PROJECTS='clang;clang-tools-extra;compiler-rt;libclc;libcxx;libcxxabi;libunwind;lld' -DLLVM_BINUTILS_INCDIR=/usr/include/ ../llvm/
-$ make
+$ cmake -DLLVM_ENABLE_PROJECTS='clang;clang-tools-extra;compiler-rt;libclc;libcxx;libcxxabi;libunwind;lld' -DCMAKE_BUILD_TYPE=Release -DLLVM_BINUTILS_INCDIR=/usr/include/ ../llvm/
+$ make -j $(nproc)
 $ export PATH=`pwd`/bin:$PATH
 $ export LLVM_CONFIG=`pwd`/bin/llcm-config
 $ cd /path/to/AFLplusplus/
@@ -96,6 +87,13 @@ CC=afl-clang-lto CXX=afl-clang-lto++ ./configure
 make
 ```
 
+## AUTODICTIONARY feature
+
+Setting `AFL_LLVM_LTO_AUTODICTIONARY` will generate a dictionary in the
+target binary based on string compare and memory compare functions.
+afl-fuzz will automatically get these transmitted when starting to fuzz.
+This improves coverage on a lot of targets.
+
 ## Potential issues
 
 ### compiling libraries fails
@@ -121,11 +119,8 @@ Please report issues at:
 
 ## Upcoming Work
 
-1. Currently the LTO whitelist feature does not allow to not instrument main, start and init functions
-2. Modify the forkserver + afl-fuzz so that only the necessary map size is
-   loaded and used - and communicated to afl-fuzz too.
-   Result: faster fork in the target and faster map analysis in afl-fuzz
-   => more speed :-)
+1. Currently the LTO whitelist feature does not allow to not instrument main,
+   start and init functions
 
 ## History
 
diff --git a/llvm_mode/afl-clang-fast.c b/llvm_mode/afl-clang-fast.c
index 26ee0bab..cdb22cb9 100644
--- a/llvm_mode/afl-clang-fast.c
+++ b/llvm_mode/afl-clang-fast.c
@@ -477,6 +477,9 @@ static void edit_params(u32 argc, char **argv, char **envp) {
 
   }
 
+  if (instrument_mode == INSTRUMENT_LTO)
+    cc_params[cc_par_cnt++] = alloc_printf("%s/afl-llvm-rt-lto.o", obj_path);
+
 #ifndef __ANDROID__
   switch (bit_mode) {
 
diff --git a/llvm_mode/afl-llvm-lto-instrumentation.so.cc b/llvm_mode/afl-llvm-lto-instrumentation.so.cc
index 93968984..f387e79c 100644
--- a/llvm_mode/afl-llvm-lto-instrumentation.so.cc
+++ b/llvm_mode/afl-llvm-lto-instrumentation.so.cc
@@ -38,17 +38,24 @@
 #include <sys/time.h>
 
 #include "llvm/Config/llvm-config.h"
+#include "llvm/ADT/Statistic.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/LegacyPassManager.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/DebugInfo.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Verifier.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/IPO/PassManagerBuilder.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/MemorySSAUpdater.h"
-#include "llvm/IR/DebugInfo.h"
-#include "llvm/IR/CFG.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Pass.h"
+
+#include <set>
 
 using namespace llvm;
 
@@ -145,7 +152,7 @@ class AFLLTOPass : public ModulePass {
   bool runOnModule(Module &M) override;
 
  protected:
-  int      afl_global_id = 1, debug = 0;
+  int      afl_global_id = 1, debug = 0, autodictionary = 0;
   uint32_t be_quiet = 0, inst_blocks = 0, inst_funcs = 0, total_instr = 0;
 
 };
@@ -154,7 +161,9 @@ class AFLLTOPass : public ModulePass {
 
 bool AFLLTOPass::runOnModule(Module &M) {
 
-  LLVMContext &C = M.getContext();
+  LLVMContext &            C = M.getContext();
+  std::vector<std::string> dictionary;
+  std::vector<CallInst *>  calls;
 
   IntegerType *Int8Ty = IntegerType::getInt8Ty(C);
   IntegerType *Int32Ty = IntegerType::getInt32Ty(C);
@@ -172,6 +181,10 @@ bool AFLLTOPass::runOnModule(Module &M) {
 
     be_quiet = 1;
 
+  if (getenv("AFL_LLVM_AUTODICTIONARY") ||
+      getenv("AFL_LLVM_LTO_AUTODICTIONARY"))
+    autodictionary = 1;
+
   /* Get globals for the SHM region and the previous location. Note that
      __afl_prev_loc is thread-local. */
 
@@ -193,6 +206,110 @@ bool AFLLTOPass::runOnModule(Module &M) {
 
     std::vector<BasicBlock *> InsBlocks;
 
+    if (autodictionary) {
+
+      for (auto &BB : F) {
+
+        for (auto &IN : BB) {
+
+          CallInst *callInst = nullptr;
+
+          if ((callInst = dyn_cast<CallInst>(&IN))) {
+
+            bool isStrcmp = true;
+            bool isMemcmp = true;
+            bool isStrncmp = true;
+            bool isStrcasecmp = true;
+            bool isStrncasecmp = true;
+
+            Function *Callee = callInst->getCalledFunction();
+            if (!Callee) continue;
+            if (callInst->getCallingConv() != llvm::CallingConv::C) continue;
+            StringRef FuncName = Callee->getName();
+            isStrcmp &= !FuncName.compare(StringRef("strcmp"));
+            isMemcmp &= !FuncName.compare(StringRef("memcmp"));
+            isStrncmp &= !FuncName.compare(StringRef("strncmp"));
+            isStrcasecmp &= !FuncName.compare(StringRef("strcasecmp"));
+            isStrncasecmp &= !FuncName.compare(StringRef("strncasecmp"));
+
+            if (!isStrcmp && !isMemcmp && !isStrncmp && !isStrcasecmp &&
+                !isStrncasecmp)
+              continue;
+
+            /* Verify the strcmp/memcmp/strncmp/strcasecmp/strncasecmp function
+             * prototype */
+            FunctionType *FT = Callee->getFunctionType();
+
+            isStrcmp &= FT->getNumParams() == 2 &&
+                        FT->getReturnType()->isIntegerTy(32) &&
+                        FT->getParamType(0) == FT->getParamType(1) &&
+                        FT->getParamType(0) ==
+                            IntegerType::getInt8PtrTy(M.getContext());
+            isStrcasecmp &= FT->getNumParams() == 2 &&
+                            FT->getReturnType()->isIntegerTy(32) &&
+                            FT->getParamType(0) == FT->getParamType(1) &&
+                            FT->getParamType(0) ==
+                                IntegerType::getInt8PtrTy(M.getContext());
+            isMemcmp &= FT->getNumParams() == 3 &&
+                        FT->getReturnType()->isIntegerTy(32) &&
+                        FT->getParamType(0)->isPointerTy() &&
+                        FT->getParamType(1)->isPointerTy() &&
+                        FT->getParamType(2)->isIntegerTy();
+            isStrncmp &= FT->getNumParams() == 3 &&
+                         FT->getReturnType()->isIntegerTy(32) &&
+                         FT->getParamType(0) == FT->getParamType(1) &&
+                         FT->getParamType(0) ==
+                             IntegerType::getInt8PtrTy(M.getContext()) &&
+                         FT->getParamType(2)->isIntegerTy();
+            isStrncasecmp &= FT->getNumParams() == 3 &&
+                             FT->getReturnType()->isIntegerTy(32) &&
+                             FT->getParamType(0) == FT->getParamType(1) &&
+                             FT->getParamType(0) ==
+                                 IntegerType::getInt8PtrTy(M.getContext()) &&
+                             FT->getParamType(2)->isIntegerTy();
+
+            if (!isStrcmp && !isMemcmp && !isStrncmp && !isStrcasecmp &&
+                !isStrncasecmp)
+              continue;
+
+            /* is a str{n,}{case,}cmp/memcmp, check if we have
+             * str{case,}cmp(x, "const") or str{case,}cmp("const", x)
+             * strn{case,}cmp(x, "const", ..) or strn{case,}cmp("const", x, ..)
+             * memcmp(x, "const", ..) or memcmp("const", x, ..) */
+            Value *Str1P = callInst->getArgOperand(0),
+                  *Str2P = callInst->getArgOperand(1);
+            StringRef Str1, Str2;
+            bool      HasStr1 = getConstantStringInfo(Str1P, Str1);
+            bool      HasStr2 = getConstantStringInfo(Str2P, Str2);
+
+            /* handle cases of one string is const, one string is variable */
+            if (!(HasStr1 ^ HasStr2)) continue;
+
+            if (isMemcmp || isStrncmp || isStrncasecmp) {
+
+              /* check if third operand is a constant integer
+               * strlen("constStr") and sizeof() are treated as constant */
+              Value *      op2 = callInst->getArgOperand(2);
+              ConstantInt *ilen = dyn_cast<ConstantInt>(op2);
+              if (!ilen) continue;
+              /* final precaution: if size of compare is larger than constant
+               * string skip it*/
+              uint64_t literalLength =
+                  HasStr1 ? GetStringLength(Str1P) : GetStringLength(Str2P);
+              if (literalLength < ilen->getZExtValue()) continue;
+
+            }
+
+            calls.push_back(callInst);
+
+          }
+
+        }
+
+      }
+
+    }
+
     for (auto &BB : F) {
 
       uint32_t succ = 0;
@@ -282,32 +399,201 @@ bool AFLLTOPass::runOnModule(Module &M) {
 
     }
 
+    // save highest location ID to global variable
+    // do this after each function to fail faster
+    if (afl_global_id > MAP_SIZE) {
+
+      uint32_t pow2map = 1, map = afl_global_id;
+      while ((map = map >> 1))
+        pow2map++;
+      FATAL(
+          "We have %u blocks to instrument but the map size is only %u! Edit "
+          "config.h and set MAP_SIZE_POW2 from %u to %u, then recompile "
+          "afl-fuzz and llvm_mode.",
+          afl_global_id, MAP_SIZE, MAP_SIZE_POW2, pow2map);
+
+    }
+
   }
 
-  // save highest location ID to global variable
+  if (calls.size()) {
 
-  if (afl_global_id > MAP_SIZE) {
+    for (auto &callInst : calls) {
 
-    uint32_t pow2map = 1, map = afl_global_id;
-    while ((map = map >> 1))
-      pow2map++;
-    FATAL(
-        "We have %u blocks to instrument but the map size is only %u! Edit "
-        "config.h and set MAP_SIZE_POW2 from %u to %u, then recompile "
-        "afl-fuzz and llvm_mode.",
-        afl_global_id, MAP_SIZE, MAP_SIZE_POW2, pow2map);
+      Value *Str1P = callInst->getArgOperand(0),
+            *Str2P = callInst->getArgOperand(1);
+      StringRef   Str1, Str2, ConstStr;
+      std::string TmpConstStr;
+      Value *     VarStr;
+      bool        HasStr1 = getConstantStringInfo(Str1P, Str1);
+      getConstantStringInfo(Str2P, Str2);
+      uint64_t constLen, sizedLen;
+      bool     isMemcmp = !callInst->getCalledFunction()->getName().compare(
+          StringRef("memcmp"));
+      bool isSizedcmp = isMemcmp ||
+                        !callInst->getCalledFunction()->getName().compare(
+                            StringRef("strncmp")) ||
+                        !callInst->getCalledFunction()->getName().compare(
+                            StringRef("strncasecmp"));
+
+      if (isSizedcmp) {
+
+        Value *      op2 = callInst->getArgOperand(2);
+        ConstantInt *ilen = dyn_cast<ConstantInt>(op2);
+        sizedLen = ilen->getZExtValue();
+
+      } else {
+
+        sizedLen = 0;
+
+      }
+
+      if (HasStr1) {
+
+        TmpConstStr = Str1.str();
+        VarStr = Str2P;
+        constLen = isMemcmp ? sizedLen : GetStringLength(Str1P);
+
+      } else {
+
+        TmpConstStr = Str2.str();
+        VarStr = Str1P;
+        constLen = isMemcmp ? sizedLen : GetStringLength(Str2P);
+
+      }
+
+      /* properly handle zero terminated C strings by adding the terminating 0
+       * to the StringRef (in comparison to std::string a StringRef has built-in
+       * runtime bounds checking, which makes debugging easier) */
+      TmpConstStr.append("\0", 1);
+      ConstStr = StringRef(TmpConstStr);
+
+      if (isSizedcmp && constLen > sizedLen) { constLen = sizedLen; }
+
+      /*
+            if (!be_quiet)
+              errs() << callInst->getCalledFunction()->getName() << ": len "
+                     << constLen << ": " << ConstStr << "\n";
+      */
+
+      if (constLen && constLen < MAX_DICT_FILE)
+        dictionary.push_back(ConstStr.str().substr(0, constLen));
+
+    }
 
   }
 
-  if (getenv("AFL_LLVM_LTO_DONTWRITEID") == NULL) {
+  if (getenv("AFL_LLVM_LTO_DONTWRITEID") == NULL || dictionary.size()) {
+
+    // yes we could create our own function, insert it into ctors ...
+    // but this would be a pain in the butt ... so we use afl-llvm-rt-lto.o
 
-    GlobalVariable *AFLFinalLoc = new GlobalVariable(
-        M, Int32Ty, true, GlobalValue::ExternalLinkage, 0, "__afl_final_loc", 0,
-        GlobalVariable::GeneralDynamicTLSModel, 0, false);
-    ConstantInt *const_loc = ConstantInt::get(Int32Ty, afl_global_id);
-    MaybeAlign   Align = MaybeAlign(4);
-    AFLFinalLoc->setAlignment(Align);
-    AFLFinalLoc->setInitializer(const_loc);
+    Function *f = M.getFunction("__afl_auto_init_globals");
+
+    if (!f) {
+
+      fprintf(stderr,
+              "Error: init function could not be found (this hould not "
+              "happen)\n");
+      exit(-1);
+
+    }
+
+    BasicBlock *bb = &f->getEntryBlock();
+    if (!bb) {
+
+      fprintf(stderr,
+              "Error: init function does not have an EntryBlock (this should "
+              "not happen)\n");
+      exit(-1);
+
+    }
+
+    BasicBlock::iterator IP = bb->getFirstInsertionPt();
+    IRBuilder<>          IRB(&(*IP));
+
+    if (getenv("AFL_LLVM_LTO_DONTWRITEID") == NULL) {
+
+      GlobalVariable *AFLFinalLoc = new GlobalVariable(
+          M, Int32Ty, true, GlobalValue::ExternalLinkage, 0, "__afl_final_loc",
+          0, GlobalVariable::GeneralDynamicTLSModel, 0, false);
+      ConstantInt *const_loc = ConstantInt::get(Int32Ty, (((afl_global_id + 8) >> 3) << 3));
+      StoreInst *  StoreFinalLoc = IRB.CreateStore(const_loc, AFLFinalLoc);
+      StoreFinalLoc->setMetadata(M.getMDKindID("nosanitize"),
+                                 MDNode::get(C, None));
+
+    }
+
+    if (dictionary.size()) {
+
+      size_t memlen = 0, count = 0, offset = 0;
+      char * ptr;
+
+      for (auto token : dictionary) {
+
+        memlen += token.length();
+        count++;
+
+      }
+
+      if (!be_quiet) printf("AUTODICTIONARY: %lu strings found\n", count);
+
+      if (count) {
+
+        if ((ptr = (char *)malloc(memlen + count)) == NULL) {
+
+          fprintf(stderr, "Error: malloc for %lu bytes failed!\n",
+                  memlen + count);
+          exit(-1);
+
+        }
+
+        for (auto token : dictionary) {
+
+          if (offset + token.length() < 0xfffff0) {
+
+            ptr[offset++] = (uint8_t)token.length();
+            memcpy(ptr + offset, token.c_str(), token.length());
+            offset += token.length();
+
+          }
+
+        }
+
+        GlobalVariable *AFLDictionaryLen = new GlobalVariable(
+            M, Int32Ty, false, GlobalValue::ExternalLinkage, 0,
+            "__afl_dictionary_len", 0, GlobalVariable::GeneralDynamicTLSModel,
+            0, false);
+        ConstantInt *const_len = ConstantInt::get(Int32Ty, offset);
+        StoreInst *StoreDictLen = IRB.CreateStore(const_len, AFLDictionaryLen);
+        StoreDictLen->setMetadata(M.getMDKindID("nosanitize"),
+                                  MDNode::get(C, None));
+
+        ArrayType *ArrayTy = ArrayType::get(IntegerType::get(C, 8), offset);
+        GlobalVariable *AFLInternalDictionary = new GlobalVariable(
+            M, ArrayTy, true, GlobalValue::ExternalLinkage,
+            ConstantDataArray::get(C,
+                                   *(new ArrayRef<char>((char *)ptr, offset))),
+            "__afl_internal_dictionary", 0,
+            GlobalVariable::GeneralDynamicTLSModel, 0, false);
+        AFLInternalDictionary->setInitializer(ConstantDataArray::get(
+            C, *(new ArrayRef<char>((char *)ptr, offset))));
+        AFLInternalDictionary->setConstant(true);
+
+        GlobalVariable *AFLDictionary = new GlobalVariable(
+            M, PointerType::get(Int8Ty, 0), false, GlobalValue::ExternalLinkage,
+            0, "__afl_dictionary");
+
+        Value *AFLDictOff = IRB.CreateGEP(AFLInternalDictionary, Zero);
+        Value *AFLDictPtr =
+            IRB.CreatePointerCast(AFLDictOff, PointerType::get(Int8Ty, 0));
+        StoreInst *StoreDict = IRB.CreateStore(AFLDictPtr, AFLDictionary);
+        StoreDict->setMetadata(M.getMDKindID("nosanitize"),
+                               MDNode::get(C, None));
+
+      }
+
+    }
 
   }
 
diff --git a/llvm_mode/afl-llvm-rt.o.c b/llvm_mode/afl-llvm-rt.o.c
index 3651fd97..cbc4648d 100644
--- a/llvm_mode/afl-llvm-rt.o.c
+++ b/llvm_mode/afl-llvm-rt.o.c
@@ -50,11 +50,7 @@
    Basically, we need to make sure that the forkserver is initialized after
    the LLVM-generated runtime initialization pass, not before. */
 
-#ifdef USE_TRACE_PC
 #define CONST_PRIO 5
-#else
-#define CONST_PRIO 0
-#endif                                                     /* ^USE_TRACE_PC */
 
 #include <sys/mman.h>
 #include <fcntl.h>
@@ -65,17 +61,20 @@
 
 u8  __afl_area_initial[MAP_SIZE];
 u8 *__afl_area_ptr = __afl_area_initial;
+u8 *__afl_dictionary;
 
 #ifdef __ANDROID__
 PREV_LOC_T __afl_prev_loc[NGRAM_SIZE_MAX];
 u32        __afl_final_loc;
 u32        __afl_prev_ctx;
 u32        __afl_cmp_counter;
+u32        __afl_dictionary_len;
 #else
 __thread PREV_LOC_T __afl_prev_loc[NGRAM_SIZE_MAX];
 __thread u32        __afl_final_loc;
 __thread u32        __afl_prev_ctx;
 __thread u32        __afl_cmp_counter;
+__thread u32        __afl_dictionary_len;
 #endif
 
 struct cmp_map *__afl_cmp_map;
@@ -100,6 +99,10 @@ static void __afl_map_shm(void) {
     const char *   shm_file_path = id_str;
     int            shm_fd = -1;
     unsigned char *shm_base = NULL;
+    unsigned int   map_size = MAP_SIZE
+
+        if (__afl_final_loc > 1 && __afl_final_loc < MAP_SIZE) map_size =
+            __afl_final_loc;
 
     /* create the shared memory segment as if it was a file */
     shm_fd = shm_open(shm_file_path, O_RDWR, 0600);
@@ -111,7 +114,7 @@ static void __afl_map_shm(void) {
     }
 
     /* map the shared memory segment to the address space of the process */
-    shm_base = mmap(0, MAP_SIZE, PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0);
+    shm_base = mmap(0, map_size, PROT_READ | PROT_WRITE, MAP_SHARED, shm_fd, 0);
     if (shm_base == MAP_FAILED) {
 
       close(shm_fd);
@@ -187,8 +190,15 @@ static void __afl_map_shm(void) {
 #ifdef __linux__
 static void __afl_start_snapshots(void) {
 
-  static u8 tmp[4];
+  static u8 tmp[4] = {0, 0, 0, 0};
   s32       child_pid;
+  u32       status = 0;
+  u32       map_size = MAP_SIZE;
+  u32       already_read_first = 0;
+  u32       was_killed;
+
+  if (__afl_final_loc > 1 && __afl_final_loc < MAP_SIZE)
+    map_size = __afl_final_loc;
 
   u8 child_stopped = 0;
 
@@ -197,16 +207,74 @@ static void __afl_start_snapshots(void) {
   /* Phone home and tell the parent that we're OK. If parent isn't there,
      assume we're not running in forkserver mode and just execute program. */
 
+  status |= (FS_OPT_ENABLED | FS_OPT_SNAPSHOT);
+  if (map_size <= 0x1000000)
+    status |= (FS_OPT_SET_MAPSIZE(map_size) | FS_OPT_MAPSIZE);
+  if (__afl_dictionary_len > 0 && __afl_dictionary) status |= FS_OPT_AUTODICT;
+  memcpy(tmp, &status, 4);
+
   if (write(FORKSRV_FD + 1, tmp, 4) != 4) return;
 
+  if (__afl_dictionary_len > 0 && __afl_dictionary) {
+
+    if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+
+    if ((was_killed & (FS_OPT_ENABLED | FS_OPT_AUTODICT)) ==
+        (FS_OPT_ENABLED | FS_OPT_AUTODICT)) {
+
+      // great lets pass the dictionary through the forkserver FD
+      u32 len = __afl_dictionary_len, offset = 0;
+      s32 ret;
+
+      if (write(FORKSRV_FD + 1, &len, 4) != 4) {
+
+        write(2, "Error: could not send dictionary len\n",
+              strlen("Error: could not send dictionary len\n"));
+        _exit(1);
+
+      }
+
+      while (len != 0) {
+
+        ret = write(FORKSRV_FD + 1, __afl_dictionary + offset, len);
+
+        if (ret < 1) {
+
+          write(2, "Error: could not send dictionary\n",
+                strlen("Error: could not send dictionary\n"));
+          _exit(1);
+
+        }
+
+        len -= ret;
+        offset += ret;
+
+      }
+
+    } else {
+
+      // uh this forkserver master does not understand extended option passing
+      // or does not want the dictionary
+      already_read_first = 1;
+
+    }
+
+  }
+
   while (1) {
 
-    u32 was_killed;
     int status;
 
-    /* Wait for parent by reading from the pipe. Abort if read fails. */
+    if (already_read_first) {
 
-    if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+      already_read_first = 0;
+
+    } else {
+
+      /* Wait for parent by reading from the pipe. Abort if read fails. */
+      if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+
+    }
 
     /* If we stopped the child in persistent mode, but there was a race
        condition and afl-fuzz already issued SIGKILL, write off the old
@@ -291,26 +359,92 @@ static void __afl_start_forkserver(void) {
 
 #endif
 
-  static u8 tmp[4];
-  s32       child_pid;
+  u8  tmp[4] = {0, 0, 0, 0};
+  s32 child_pid;
+  u32 status = 0;
+  u32 map_size = MAP_SIZE;
+  u32 already_read_first = 0;
+  u32 was_killed;
+
+  if (__afl_final_loc > 1 && __afl_final_loc < MAP_SIZE)
+    map_size = __afl_final_loc;
 
   u8 child_stopped = 0;
 
   void (*old_sigchld_handler)(int) = 0;  // = signal(SIGCHLD, SIG_DFL);
 
+  if (map_size <= 0x1000000)
+    status |= (FS_OPT_SET_MAPSIZE(map_size) | FS_OPT_MAPSIZE);
+  if (__afl_dictionary_len > 0 && __afl_dictionary) status |= FS_OPT_AUTODICT;
+  if (status) status |= (FS_OPT_ENABLED);
+  memcpy(tmp, &status, 4);
+
   /* Phone home and tell the parent that we're OK. If parent isn't there,
      assume we're not running in forkserver mode and just execute program. */
 
   if (write(FORKSRV_FD + 1, tmp, 4) != 4) return;
 
+  if (__afl_dictionary_len > 0 && __afl_dictionary) {
+
+    if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+
+    if ((was_killed & (FS_OPT_ENABLED | FS_OPT_AUTODICT)) ==
+        (FS_OPT_ENABLED | FS_OPT_AUTODICT)) {
+
+      // great lets pass the dictionary through the forkserver FD
+      u32 len = __afl_dictionary_len, offset = 0;
+      s32 ret;
+
+      if (write(FORKSRV_FD + 1, &len, 4) != 4) {
+
+        write(2, "Error: could not send dictionary len\n",
+              strlen("Error: could not send dictionary len\n"));
+        _exit(1);
+
+      }
+
+      while (len != 0) {
+
+        ret = write(FORKSRV_FD + 1, __afl_dictionary + offset, len);
+
+        if (ret < 1) {
+
+          write(2, "Error: could not send dictionary\n",
+                strlen("Error: could not send dictionary\n"));
+          _exit(1);
+
+        }
+
+        len -= ret;
+        offset += ret;
+
+      }
+
+    } else {
+
+      // uh this forkserver master does not understand extended option passing
+      // or does not want the dictionary
+      already_read_first = 1;
+
+    }
+
+  }
+
   while (1) {
 
-    u32 was_killed;
     int status;
 
     /* Wait for parent by reading from the pipe. Abort if read fails. */
 
-    if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+    if (already_read_first) {
+
+      already_read_first = 0;
+
+    } else {
+
+      if (read(FORKSRV_FD, &was_killed, 4) != 4) _exit(1);
+
+    }
 
     /* If we stopped the child in persistent mode, but there was a race
        condition and afl-fuzz already issued SIGKILL, write off the old
@@ -378,8 +512,12 @@ static void __afl_start_forkserver(void) {
 
 int __afl_persistent_loop(unsigned int max_cnt) {
 
-  static u8  first_pass = 1;
-  static u32 cycle_cnt;
+  static u8    first_pass = 1;
+  static u32   cycle_cnt;
+  unsigned int map_size = MAP_SIZE;
+
+  if (__afl_final_loc > 1 && __afl_final_loc < MAP_SIZE)
+    map_size = __afl_final_loc;
 
   if (first_pass) {
 
@@ -390,7 +528,7 @@ int __afl_persistent_loop(unsigned int max_cnt) {
 
     if (is_persistent) {
 
-      memset(__afl_area_ptr, 0, MAP_SIZE);
+      memset(__afl_area_ptr, 0, map_size);
       __afl_area_ptr[0] = 1;
       memset(__afl_prev_loc, 0, NGRAM_SIZE_MAX * sizeof(PREV_LOC_T));