json_format: Add html_safe option

FU/Util.pm

@@ -212,13 +212,6 @@ roughly similar to:
 
   JSON::PP->new->allow_nonref->core_bools->convert_blessed->encode($scalar);
 
-Some modules escape the slash character in encoded strings to prevent a
-potential XSS vulnerability when embedding JSON inside C<< <script> ..
-</script> >> tags.  This function does I<not> do that because it might not even
-be sufficient. The following is probably an improvement:
-
-  json_format($data) =~ s{</}{<\\/}rg =~ s/<!--/<\\u0021--/rg;
-
 This function generates invalid JSON if you pass it a string with invalid
 Unicode characters; I don't see how you'd ever accidentally end up with such a
 string, anyway.
@@ -244,6 +237,26 @@ versions.
 
 Boolean, returns a UTF-8 encoded byte string instead of a Perl Unicode string.
 
+=item html_safe
+
+Boolean. When set, the encoded JSON is safe for (unescaped) inclusion into HTML
+or XML content. This encodes C<< < >>, C<< > >> and C<< & >> as Unicode escapes.
+Commonly used to embed data inside a HTML page:
+
+  $html = '<script id="site_data" type="application/json">'
+        . json_format($data, html_safe => 1)
+        . '</script>';
+
+This option does NOT make it safe to include the encoded JSON as an attribute
+value. There is no way to do that without violating JSON specs, so you should
+use entity escaping instead.
+
+Some JSON modules escape the forward slash (C</>) character instead, but that
+is, at best, B<only> sufficient for embedding inside a C<< <script> >> tag (I'm
+not sure how C<< <!-- >> and C<< <![CDATA[ >> are treated in that context). In
+any other context, you'll need the more thourough escaping provided by this
+C<html_safe> option.
+
 =item max_size
 
 Maximum permitted size, in bytes, of the generated JSON string. Defaults to 1 GiB.

c/jsonfmt.c

@@ -3,6 +3,7 @@ typedef struct {
     UV depth;
     int canon;
     int pretty; /* <0 when disabled, current nesting level otherwise */
+    int htmlsafe;
 } fujson_fmt_ctx;
 
 static void fujson_fmt(pTHX_ fujson_fmt_ctx *, SV *);
@@ -27,21 +28,22 @@ static void fujson_fmt_str(pTHX_ fujson_fmt_ctx *ctx, const char *stri, size_t l
     while (off < len) {
         /* Fast path: no escaping needed */
         loff = off;
+
+#define SKIPUNTIL(cond) \
+                while (off < len) { \
+                    x = str[off]; \
+                    if (x <= 0x1f || x == '"' || x == '\\' || cond) break; \
+                    off++;\
+                }
         if (utf8) {
-            /* assume >=0x80 is valid utf8 */
+            if (!ctx->htmlsafe) { SKIPUNTIL(x == 0x7f) }
-            while (off < len) {
+            else                { SKIPUNTIL(x == 0x7f || x == '<' || x == '>' || x == '&') }
-                x = str[off];
-                if (x <= 0x1f || x == '"' || x == '\\' || x == 0x7f) break;
-                off++;
-            }
         } else {
-            /* binary strings need special handling for >=0x80 */
+            if (!ctx->htmlsafe) { SKIPUNTIL(x >= 0x7f) }
-            while (off < len) {
+            else                { SKIPUNTIL(x >= 0x7f || x == '<' || x == '>' || x == '&') }
-                x = str[off];
-                if (x <= 0x1f || x == '"' || x == '\\' || x >= 0x7f) break;
-                off++;
-            }
         }
+#undef SKIPUNTIL
+
         fustr_write(ctx->out, (char *)str+loff, off-loff);
 
         if (off < len) { /* early break, which means current byte needs special processing */
@@ -279,7 +281,7 @@ static SV *fujson_fmt_xs(pTHX_ I32 ax, I32 argc, SV *val) {
     ctx.out = &out;
     ctx.depth = 0;
     ctx.pretty = INT_MIN;
-    ctx.canon = 0;
+    ctx.canon = ctx.htmlsafe = 0;
     while (i < argc) {
         arg = SvPV_nolen(ST(i));
         i++;
@@ -289,6 +291,7 @@ static SV *fujson_fmt_xs(pTHX_ I32 ax, I32 argc, SV *val) {
 
         if (strcmp(arg, "canonical") == 0) ctx.canon = SvPVXtrue(r);
         else if (strcmp(arg, "pretty") == 0) ctx.pretty = SvPVXtrue(r) ? 0 : INT_MIN;
+        else if (strcmp(arg, "html_safe") == 0) ctx.htmlsafe = !!SvPVXtrue(r);
         else if (strcmp(arg, "utf8") == 0) encutf8 = SvPVXtrue(r);
         else if (strcmp(arg, "max_size") == 0) out.maxlen = SvUV(r);
         else if (strcmp(arg, "max_depth") == 0) ctx.depth = SvUV(r);

t/json_format.t

@@ -110,6 +110,7 @@ is json_format(
 }
 _
 
+is json_format('<hello & world>', html_safe => 1), '"\u003chello \u0026 world\u003e"';
 
 eval { json_format [[]], max_depth => 2 };
 like $@, qr/max_depth exceeded while formatting JSON/;