diff --git a/devel/fuzz_llvm.sh b/devel/fuzz_llvm.sh
new file mode 100755
index 0000000..36a797d
--- /dev/null
+++ b/devel/fuzz_llvm.sh
@@ -0,0 +1,5 @@
+#!/bin/bash -e
+set -x
+clang -g -O -fsanitize=fuzzer,address \
+  fuzz_llvm_make_safe_uri.c -o fuzz_llvm_make_safe_uri
+./fuzz_llvm_make_safe_uri -only_ascii=1 $* fuzz_testcases/
diff --git a/devel/fuzz_llvm_make_safe_uri.c b/devel/fuzz_llvm_make_safe_uri.c
new file mode 100644
index 0000000..a82de59
--- /dev/null
+++ b/devel/fuzz_llvm_make_safe_uri.c
@@ -0,0 +1,136 @@
+// Wrapper around make_safe_url() for fuzzing with LLVM.
+// Aborts if the output is deemed safe but contains /../ or /./
+#include <assert.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+/* Consolidate slashes in-place by shifting parts of the string over repeated
+ * slashes.
+ */
+static void consolidate_slashes(char *s) {
+  size_t left = 0, right = 0;
+  int saw_slash = 0;
+
+  assert(s != NULL);
+  while (s[right] != '\0') {
+    if (saw_slash) {
+      if (s[right] == '/')
+        right++;
+      else {
+        saw_slash = 0;
+        s[left++] = s[right++];
+      }
+    } else {
+      if (s[right] == '/') saw_slash++;
+      s[left++] = s[right++];
+    }
+  }
+  s[left] = '\0';
+}
+
+/* Resolve /./ and /../ in a URL, in-place.  Also strip out query params.
+ * Returns NULL if the URL is invalid/unsafe, or the original buffer if
+ * successful.
+ */
+static char *make_safe_url(char *url) {
+  struct {
+    char *start;
+    size_t len;
+  } * chunks;
+  unsigned int num_slashes, num_chunks;
+  size_t urllen, i, j, pos;
+  int ends_in_slash;
+
+  /* strip query params */
+  for (pos = 0; url[pos] != '\0'; pos++) {
+    if (url[pos] == '?') {
+      url[pos] = '\0';
+      break;
+    }
+  }
+
+  if (url[0] != '/') return NULL;
+
+  consolidate_slashes(url);
+  urllen = strlen(url);
+  if (urllen > 0)
+    ends_in_slash = (url[urllen - 1] == '/');
+  else
+    ends_in_slash = 1;
+
+  /* count the slashes */
+  for (i = 0, num_slashes = 0; i < urllen; i++)
+    if (url[i] == '/') num_slashes++;
+
+  /* make an array for the URL elements */
+  assert(num_slashes > 0);
+  chunks = malloc(sizeof(*chunks) * num_slashes);
+
+  /* split by slashes and build chunks array */
+  num_chunks = 0;
+  for (i = 1; i < urllen;) {
+    /* look for the next slash */
+    for (j = i; j < urllen && url[j] != '/'; j++)
+      ;
+
+    /* process url[i,j) */
+    if ((j == i + 1) && (url[i] == '.'))
+      /* "." */;
+    else if ((j == i + 2) && (url[i] == '.') && (url[i + 1] == '.')) {
+      /* ".." */
+      if (num_chunks == 0) {
+        /* unsafe string so free chunks */
+        free(chunks);
+        return (NULL);
+      } else
+        num_chunks--;
+    } else {
+      chunks[num_chunks].start = url + i;
+      chunks[num_chunks].len = j - i;
+      num_chunks++;
+    }
+
+    i = j + 1; /* url[j] is a slash - move along one */
+  }
+
+  /* reassemble in-place */
+  pos = 0;
+  for (i = 0; i < num_chunks; i++) {
+    assert(pos <= urllen);
+    url[pos++] = '/';
+
+    assert(pos + chunks[i].len <= urllen);
+    assert(url + pos <= chunks[i].start);
+
+    if (url + pos < chunks[i].start)
+      memmove(url + pos, chunks[i].start, chunks[i].len);
+    pos += chunks[i].len;
+  }
+  free(chunks);
+
+  if ((num_chunks == 0) || ends_in_slash) url[pos++] = '/';
+  assert(pos <= urllen);
+  url[pos] = '\0';
+  return url;
+}
+
+int LLVMFuzzerTestOneInput(const uint8_t *data, size_t size) {
+  char *buf = malloc(size + 1);
+  memcpy(buf, data, size);
+  buf[size] = 0;
+  // Enable this to make sure the fuzzer is working.
+#if 0
+  if (size > 2 && buf[2] == 'x') __builtin_trap();
+#endif
+  char *safe = make_safe_url(buf);
+  if (safe) {
+    if (strstr(safe, "/../") != NULL) __builtin_trap();
+    if (strstr(safe, "/./") != NULL) __builtin_trap();
+  }
+  free(buf);  // Don't leak memory.
+  return 0;
+}
+
+/* vim:set ts=2 sw=2 sts=2 expandtab tw=78: */