Begin work on a "structured" mutator mode (#795)

* Begin work on a "structured" mutator mode

* mutator: fill in structured mode

* worker: re-enable udis86

* README: document the MODE env var

README.md

@@ -46,6 +46,8 @@ Run the fuzzer for a bit:
 * `V=1` enables verbose output on `stderr`
 * `D=1` enables the "dummy" mutation mode for debugging purposes
 * `M=1` enables the "manual" mutation mode (i.e., read from `stdin`)
+* `MODE=mode` can be used to configure the mutation mode in the absence of `D` and `M`
+    * Valid mutation modes are `sliding` (default), `havoc`, and `structured`
 
 Convert mishegos's raw output into JSONL suitable for analysis:
 

src/include/mish_common.h

@@ -102,6 +102,7 @@ typedef enum {
 typedef enum {
   M_HAVOC = 0,
   M_SLIDING,
+  M_STRUCTURED,
   M_DUMMY,
   M_MANUAL,
 } mutator_mode;

src/mishegos/mishegos.c

@@ -181,6 +181,25 @@ static void mishegos_sem_init() {
   }
 }
 
+static mutator_mode get_mut_mode() {
+  const char *mode = getenv("MODE");
+
+  /* default to the "sliding" strategy in the mutator. */
+  if (mode == NULL) {
+    return M_SLIDING;
+  }
+
+  if (strcmp(mode, "sliding") == 0) {
+    return M_SLIDING;
+  } else if (strcmp(mode, "havoc") == 0) {
+    return M_HAVOC;
+  } else if (strcmp(mode, "structured") == 0) {
+    return M_STRUCTURED;
+  }
+
+  errx(1, "unknown mutator mode requested: %s", mode);
+}
+
 static void config_init() {
   /* TODO(ww): Configurable RNG seed.
    */
@@ -195,7 +214,10 @@ static void config_init() {
   } else if (debugging) {
     GET_CONFIG()->mut_mode = M_DUMMY;
   } else {
-    GET_CONFIG()->mut_mode = M_SLIDING;
+    /* If we're not in a manual or debugging mode, try to figure out what
+     * actual fuzzing mode the user wants from us.
+     */
+    GET_CONFIG()->mut_mode = get_mut_mode();
   }
 }
 

src/mishegos/mutator.c

@@ -237,6 +237,72 @@ static bool sliding_candidate(input_slot *slot) {
   return true;
 }
 
+/* Structured: generate an instruction candidate with the
+ * "structured" approach.
+ */
+static bool structured_candidate(input_slot *slot) {
+  /* We mirror build_sliding_candidate here, but with the constraint that
+   * we never overapproximate: we constrain ourselves to trying
+   * to build something that looks like an instruction of no more
+   * than 15 bytes.
+   */
+
+  uint8_t len = 0;
+
+  /* Up to 4 legacy prefixes. Like sliding, we don't try to enforce group rules.
+   * Unlike sliding, we allow for the possibility of no legacy prefixes.
+   * Running max: 4.
+   */
+  uint8_t prefix_count = (rand_byte() % 5);
+  for (int i = 0; i < prefix_count; ++i) {
+    slot->raw_insn[i] = legacy_prefixes[rand_byte() % sizeof(legacy_prefixes)];
+  }
+  len = prefix_count;
+
+  /* One or none REX prefixes.
+   * Always choose a valid REX prefix if we're inserting one.
+   * Running max: 5.
+   */
+  if (rand_byte() % 2) {
+    slot->raw_insn[len] = rex_prefixes[rand_byte() % sizeof(rex_prefixes)];
+    len++;
+  }
+
+  /* Random (but structured) opcode. Same as sliding.
+   * Running max: 8
+   */
+  opcode opc;
+  rand_opcode(&opc);
+  memcpy(slot->raw_insn + len, opc.op, opc.len);
+  len += opc.len;
+
+  /* One or none ModR/M bytes, and one or none SIB bytes.
+   * Both of these are just 8-bit LUTs, so they can be fully random.
+   * Running max: 10.
+   */
+  if (rand_byte() % 2) {
+    slot->raw_insn[len] = rand_byte();
+    len++;
+  }
+
+  if (rand_byte() % 2) {
+    slot->raw_insn[len] = rand_byte();
+    len++;
+  }
+
+  /* Finally, we have up to 5 bytes to play with for the immediate and
+   * displacement. Fill some amount of that (maybe not all) with randomness.
+   */
+  uint64_t tail = rand_long();
+  uint8_t tail_size = rand_byte() % 6;
+  memcpy(slot->raw_insn + len, &tail, tail_size);
+  len += tail_size;
+
+  slot->len = len;
+
+  return true;
+}
+
 /* Dummy: Generates a single NOP for debugging purposes.
  */
 static bool dummy_candidate(input_slot *slot) {
@@ -285,25 +351,23 @@ void mutator_init() {
  * Returns false if the configured mutation mode has been exhausted.
  */
 bool candidate(input_slot *slot) {
-  bool exhausted;
-
   switch (mut_mode) {
   case M_HAVOC: {
-    exhausted = havoc_candidate(slot);
-    break;
+    return havoc_candidate(slot);
   }
   case M_SLIDING: {
-    exhausted = sliding_candidate(slot);
-    break;
+    return sliding_candidate(slot);
+  }
+  case M_STRUCTURED: {
+    return structured_candidate(slot);
   }
   case M_DUMMY: {
-    exhausted = dummy_candidate(slot);
-    break;
+    return dummy_candidate(slot);
   }
   case M_MANUAL: {
-    exhausted = manual_candidate(slot);
-    break;
+    return manual_candidate(slot);
   }
   }
-  return exhausted;
+
+  __builtin_unreachable();
 }