Merge pull request #2421 from wtdcode/rename-afl-san-no-inst

Rename `AFL_SAN_NO_INST` to `AFL_FSRV_ONLY`

afl-cmin.py

@@ -28,6 +28,38 @@ import shutil
 import subprocess
 import sys
 
+# https://more-itertools.readthedocs.io/en/stable/_modules/more_itertools/recipes.html#batched
+from sys import hexversion
+
+def _batched(iterable, n, *, strict=False):
+    """Batch data into tuples of length *n*. If the number of items in
+    *iterable* is not divisible by *n*:
+    * The last batch will be shorter if *strict* is ``False``.
+    * :exc:`ValueError` will be raised if *strict* is ``True``.
+
+    >>> list(batched('ABCDEFG', 3))
+    [('A', 'B', 'C'), ('D', 'E', 'F'), ('G',)]
+
+    On Python 3.13 and above, this is an alias for :func:`itertools.batched`.
+    """
+    if n < 1:
+        raise ValueError('n must be at least one')
+    iterator = iter(iterable)
+    while batch := tuple(itertools.islice(iterator, n)):
+        if strict and len(batch) != n:
+            raise ValueError('batched(): incomplete batch')
+        yield batch
+
+
+if hexversion >= 0x30D00A2:  # pragma: no cover
+    from itertools import batched as itertools_batched
+    def batched(iterable, n, *, strict=False):
+        return itertools_batched(iterable, n, strict=strict)
+else:
+    batched = _batched
+
+    batched.__doc__ = _batched.__doc__
+
 try:
     from tqdm import tqdm
 except ImportError:
@@ -574,7 +606,7 @@ def main():
         workers.append(p)
 
     chunk = max(1, min(128, len(files) // args.workers))
-    jobs = list(itertools.batched(enumerate(files), chunk))
+    jobs = list(batched(enumerate(files), chunk))
     jobs += [None] * args.workers  # sentinel
 
     dispatcher = JobDispatcher(job_queue, jobs)

docs/SAND.md

@@ -21,7 +21,7 @@ For a normal fuzzing workflow, we have:
 For SAND fuzzing workflow, this is slightly different:
 
 1. Build target project _without_ any sanitizers to get `target_native`, which we will define as a "native binary". It is usually done by using `afl-clang-fast/lto(++)` to compile your project _without_ `AFL_USE_ASAN/UBSAN/MSAN`.
-2. Build target project with AFL_USE_ASAN=1 AFL_SAN_NO_INST=1 to get `target_asan`. Do note this step can be repeated for multiple sanitizers, like MSAN, UBSAN etc. It is also possible to have ASAN and UBSAN to build together.
+2. Build target project with AFL_USE_ASAN=1 AFL_LLVM_ONLY_FSRV=1 to get `target_asan`. Do note this step can be repeated for multiple sanitizers, like MSAN, UBSAN etc. It is also possible to have ASAN and UBSAN to build together.
 3. Fuzz the target with `afl-fuzz -i seeds -o out -w ./target_asan -- ./target_native`. Note `-w` can be specified multiple times.
 
 Then you get:
@@ -44,11 +44,11 @@ Just like the normal building process, except using `afl-clang-fast`
 2. Build the sanitizers-enabled binaries.
 
 ```bash
-AFL_SAN_NO_INST=1 AFL_USE_UBSAN=1 AFL_USE_ASAN=1 afl-clang-fast test-instr.c -o ./asanubsan
+AFL_LLVM_ONLY_FSRV=1 AFL_USE_UBSAN=1 AFL_USE_ASAN=1 afl-clang-fast test-instr.c -o ./asanubsan
-AFL_SAN_NO_INST=1 AFL_USE_MSAN=1 afl-clang-fast test-instr.c -o ./msan
+AFL_LLVM_ONLY_FSRV=1 AFL_USE_MSAN=1 afl-clang-fast test-instr.c -o ./msan
 ```
 
-Do note `AFL_SAN_NO_INST=1` is crucial, this enables forkservers but disables pc instrumentation. Do not reuse sanitizers-enabled binaries built _without_ `AFL_SAN_NO_INST=1`. This will mess up SAND execution pattern.
+Do note `AFL_LLVM_ONLY_FSRV=1` is crucial, this enables forkservers but disables pc instrumentation. You are allowed to reuse sanitizers-enabled binaries, i.e. binaries built _without_ `AFL_LLVM_ONLY_FSRV=1`, at a cost of reduced speed. 
 
 3. Start fuzzing
 

docs/env_variables.md

@@ -111,6 +111,10 @@ fairly broad use of environment variables instead:
 
     - Note: both `AFL_CFISAN_VERBOSE=1` and `AFL_UBSAN_VERBOSE=1` are disabled by default as verbose output can significantly slow down fuzzing performance. Use these options only during debugging or when additional crash diagnostics are required
 
+  - `AFL_LLVM_ONLY_FSRV`/`AFL_GCC_ONLY_FSRV` will inject forkserver but not pc instrumentation. Please note this is different compared to `AFL_LLVM_DISABLE_INSTRUMENTATION`, which will totally disable forkserver implementation. This env is pretty useful in two cases:
+    - [SAND](./SAND.md). In this case, the binaries built in this way will serve as extra oracles. Check the corresponding documents for details.
+    - Compatible with LibAFL ForkserverExecutor implementation and thus faster to repeatedly run, compared to simple CommandExecutor.
+
   - `TMPDIR` is used by afl-as for temporary files; if this variable is not set,
     the tool defaults to /tmp.
 

include/envs.h

@@ -10,6 +10,7 @@ static char *afl_environment_deprecated[] = {
     "AFL_DEFER_FORKSRV",
     "AFL_POST_LIBRARY",
     "AFL_PERSISTENT",
+    "AFL_SAN_NO_INST",
     NULL
 
 };
@@ -118,7 +119,7 @@ static char *afl_environment_variables[] = {
     "AFL_CFISAN_VERBOSE", "AFL_USE_LSAN", "AFL_WINE_PATH", "AFL_NO_SNAPSHOT",
     "AFL_EXPAND_HAVOC_NOW", "AFL_USE_FASAN", "AFL_USE_QASAN",
     "AFL_PRINT_FILENAMES", "AFL_PIZZA_MODE", "AFL_NO_FASTRESUME",
-    "AFL_SAN_ABSTRACTION", "AFL_SAN_NO_INST", "AFL_SAN_RECOVER",
+    "AFL_SAN_ABSTRACTION", "AFL_LLVM_ONLY_FSRV", "AFL_GCC_ONLY_FRSV", "AFL_SAN_RECOVER",
     "AFL_PRELOAD_DISCRIMINATE_FORKSERVER_PARENT", NULL};
 
 extern char *afl_environment_variables[];

instrumentation/SanitizerCoverageLTO.so.cc

@@ -327,7 +327,7 @@ class ModuleSanitizerCoverageLTOLegacyPass : public ModulePass {
 
     };
 
-    if (!getenv("AFL_SAN_NO_INST")) {
+    if (!getenv("AFL_LLVM_ONLY_FSRV")) {
 
       return ModuleSancov.instrumentModule(M, DTCallback, PDTCallback);
 
@@ -389,7 +389,7 @@ PreservedAnalyses ModuleSanitizerCoverageLTO::run(Module                &M,
 
   };
 
-  if (!getenv("AFL_SAN_NO_INST")) {
+  if (!getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (ModuleSancov.instrumentModule(M, DTCallback, PDTCallback))
       return PreservedAnalyses::none();

instrumentation/SanitizerCoveragePCGUARD.so.cc

@@ -272,7 +272,7 @@ PreservedAnalyses ModuleSanitizerCoverageAFL::run(Module                &M,
   // TODO: Support LTO or llvm classic?
   // Note we still need afl-compiler-rt so we just disable the instrumentation
   // here.
-  if (!getenv("AFL_SAN_NO_INST")) {
+  if (!getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (ModuleSancov.instrumentModule(M, DTCallback, PDTCallback))
       return PreservedAnalyses::none();

instrumentation/afl-compiler-rt.o.c

@@ -1241,6 +1241,11 @@ void __afl_manual_init(void) {
 
   }
 
+  if (getenv("AFL_LLVM_ONLY_FSRV") || getenv("AFL_GCC_ONLY_FRSV")) {
+    fprintf(stderr, "DEBUG: Overwrite area_ptr to dummy due to AFL_LLVM_ONLY_FSRV/AFL_GCC_ONLY_FRSV\n");
+    __afl_area_ptr = __afl_area_ptr_dummy;
+  }
+
   if (!init_done) {
 
     __afl_start_forkserver();

instrumentation/afl-gcc-pass.so.cc

@@ -462,6 +462,7 @@ static struct plugin_info afl_plugin = {
     .help = G_("AFL gcc plugin\n\
 \n\
 Set AFL_QUIET in the environment to silence it.\n\
+Set AFL_GCC_ONLY_FRSV in the environment to disable instrumentation.\n\
 \n\
 Set AFL_INST_RATIO in the environment to a number from 0 to 100\n\
 to control how likely a block will be chosen for instrumentation.\n\
@@ -502,9 +503,12 @@ int plugin_init(struct plugin_name_args   *info,
      case it was specified in the command line's -frandom-seed for
      reproducible instrumentation.  */
   srandom(get_random_seed(false));
+  bool fsrv_only = !!getenv("AFL_GCC_ONLY_FRSV");
 
   const char *name = info->base_name;
-  register_callback(name, PLUGIN_INFO, NULL, &afl_plugin);
+  if (!fsrv_only) {
+    register_callback(name, PLUGIN_INFO, NULL, &afl_plugin);
+  }
 
   afl_pass                 *aflp = new afl_pass(quiet, inst_ratio);
   struct register_pass_info pass_info = {
@@ -516,14 +520,18 @@ int plugin_init(struct plugin_name_args   *info,
 
   };
 
-  register_callback(name, PLUGIN_PASS_MANAGER_SETUP, NULL, &pass_info);
+  if (!fsrv_only) {
-  register_callback(name, PLUGIN_FINISH, afl_pass::plugin_finalize,
+    register_callback(name, PLUGIN_PASS_MANAGER_SETUP, NULL, &pass_info);
-                    pass_info.pass);
+    register_callback(name, PLUGIN_FINISH, afl_pass::plugin_finalize,
+                      pass_info.pass);
+  }
 
   if (!quiet)
     ACTF(G_("%s instrumentation at ratio of %u%% in %s mode."),
          aflp->out_of_line ? G_("Call-based") : G_("Inline"), inst_ratio,
          getenv("AFL_HARDEN") ? G_("hardened") : G_("non-hardened"));
+  else if (fsrv_only)
+    ACTF("Instrumentation disabled due to AFL_GCC_ONLY_FRSV");
 
   return 0;
 

instrumentation/afl-llvm-pass.so.cc

@@ -225,7 +225,7 @@ bool AFLCoverage::runOnModule(Module &M) {
   if (getenv("AFL_DEBUG")) debug = 1;
 
 #if LLVM_VERSION_MAJOR >= 11                        /* use new pass manager */
-  if (getenv("AFL_SAN_NO_INST")) {
+  if (getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (debug) { fprintf(stderr, "Instrument disabled\n"); }
     return PreservedAnalyses::all();
@@ -233,7 +233,7 @@ bool AFLCoverage::runOnModule(Module &M) {
   }
 
 #else
-  if (getenv("AFL_SAN_NO_INST")) {
+  if (getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (debug) { fprintf(stderr, "Instrument disabled\n"); }
     return true;

src/afl-cc.c

@@ -244,7 +244,7 @@ static inline void insert_object(aflcc_state_t *aflcc, u8 *obj, u8 *fmt,
 /* Insert params into the new argv, make clang load the pass. */
 static inline void load_llvm_pass(aflcc_state_t *aflcc, u8 *pass) {
 
-  if (getenv("AFL_SAN_NO_INST")) {
+  if (getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (!be_quiet) { DEBUGF("SAND: Coverage instrumentation disabled\n"); }
     return;
@@ -2097,7 +2097,7 @@ void add_native_pcguard(aflcc_state_t *aflcc) {
    * anyway.
    */
   if (aflcc->have_rust_asanrt) { return; }
-  if (getenv("AFL_SAN_NO_INST")) {
+  if (getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (!be_quiet) { DEBUGF("SAND: Coverage instrumentation disabled\n"); }
     return;
@@ -2138,7 +2138,7 @@ void add_native_pcguard(aflcc_state_t *aflcc) {
 */
 void add_optimized_pcguard(aflcc_state_t *aflcc) {
 
-  if (getenv("AFL_SAN_NO_INST")) {
+  if (getenv("AFL_LLVM_ONLY_FSRV")) {
 
     if (!be_quiet) { DEBUGF("SAND: Coverage instrumentation disabled\n"); }
     return;
@@ -2600,7 +2600,7 @@ void add_assembler(aflcc_state_t *aflcc) {
 /* Add params to launch the gcc plugins for instrumentation. */
 void add_gcc_plugin(aflcc_state_t *aflcc) {
 
-  if (getenv("AFL_SAN_NO_INST")) {
+  if (getenv("AFL_GCC_ONLY_FSRV")) {
 
     if (!be_quiet) { DEBUGF("SAND: Coverage instrumentation disabled\n"); }
     return;

src/afl-common.c

@@ -819,8 +819,15 @@ void check_environment_vars(char **envp) {
 
           WARNF("AFL environment variable %s is deprecated!",
                 afl_environment_deprecated[i]);
-          issue_detected = 1;
+          
-
+          if (strncmp(afl_environment_deprecated[i], 
+            "AFL_SAN_NO_INST", strlen(afl_environment_deprecated[i])) == 0) {
+              WARNF("AFL_LLVM_ONLY_FSRV/AFL_GCC_ONLY_FSRV is induced and set instead.");
+              setenv("AFL_GCC_ONLY_FSRV", "1", 0);
+              setenv("AFL_LLVM_ONLY_FSRV", "1", 0);
+          } else {
+            issue_detected = 1;
+          }
         } else {
 
           i++;

src/afl-fuzz-sanfuzz.c

@@ -36,9 +36,11 @@ void sanfuzz_exec_child(afl_forkserver_t *fsrv, char **argv) {
       argv[0] != fsrv->asanfuzz_binary) {
 
     argv[0] = fsrv->asanfuzz_binary;
-
   }
 
+  // In case users provide the normally instrumented binaries, this servers as the last
+  // resort to avoid collecting incorrect coverage.
+  setenv("AFL_LLVM_ONLY_FSRV", "1", 0);
   execv(fsrv->target_path, argv);
 
 }