Don't crash when memory is exhausted

src/attribute.c

@@ -38,6 +38,8 @@ GumboAttribute* gumbo_get_attribute(
 
 void gumbo_destroy_attribute(
     struct GumboInternalParser* parser, GumboAttribute* attribute) {
+  if (NULL == attribute)
+    return;
   gumbo_parser_deallocate(parser, (void*) attribute->name);
   gumbo_parser_deallocate(parser, (void*) attribute->value);
   gumbo_parser_deallocate(parser, (void*) attribute);

src/error.c

@@ -258,6 +258,8 @@ void gumbo_print_caret_diagnostic(
 }
 
 void gumbo_error_destroy(GumboParser* parser, GumboError* error) {
+  if (NULL == error)
+    return;
   if (error->type == GUMBO_ERR_PARSER ||
       error->type == GUMBO_ERR_UNACKNOWLEDGED_SELF_CLOSING_TAG) {
     gumbo_vector_destroy(parser, &error->v.parser.tag_stack);

src/parser.c

@@ -20,6 +20,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <strings.h>
+#include <setjmp.h>
 
 #include "attribute.h"
 #include "error.h"
@@ -39,6 +40,12 @@
 #define TERMINATOR \
   { "", 0 }
 
+#ifndef _WIN32
+#define SETJMP(x) (sigsetjmp(x, 0))
+#else
+#define SETJMP(x) (setjmp(x))
+#endif
+
 typedef char gumbo_tagset[GUMBO_TAG_LAST];
 #define TAG(tag) [GUMBO_TAG_##tag] = (1 << GUMBO_NAMESPACE_HTML)
 #define TAG_SVG(tag) [GUMBO_TAG_##tag] = (1 << GUMBO_NAMESPACE_SVG)
@@ -398,6 +405,13 @@ typedef struct GumboInternalParserState {
   // flag appropriately.
   bool _closed_body_tag;
   bool _closed_html_tag;
+
+  // Jump buf to jump to on memory exhaustion.
+#ifdef _WIN32
+  jmp_buf oom_buf;
+#else
+  sigjmp_buf oom_buf;
+#endif
 } GumboParserState;
 
 static bool token_has_attribute(const GumboToken* token, const char* name) {
@@ -476,6 +490,7 @@ static void output_init(GumboParser* parser) {
 static void parser_state_init(GumboParser* parser) {
   GumboParserState* parser_state =
       gumbo_parser_allocate(parser, sizeof(GumboParserState));
+  if (NULL == parser_state) return;
   parser_state->_insertion_mode = GUMBO_INSERTION_MODE_INITIAL;
   parser_state->_reprocess_current_token = false;
   parser_state->_frameset_ok = true;
@@ -2350,6 +2365,8 @@ static bool handle_after_head(GumboParser* parser, GumboToken* token) {
 }
 
 static void destroy_node(GumboParser* parser, GumboNode* node) {
+  if (NULL == node)
+    return;
   switch (node->type) {
     case GUMBO_NODE_DOCUMENT: {
       GumboDocument* doc = &node->v.document;
@@ -4075,11 +4092,24 @@ GumboOutput* gumbo_parse(const char* buffer) {
 GumboOutput* gumbo_parse_with_options(
     const GumboOptions* options, const char* buffer, size_t length) {
   GumboParser parser;
+  memset(&parser, 0, sizeof parser);
   parser._options = options;
+
+  if (SETJMP(parser._oom_buf)) {
+    if (parser._parser_state)
+      parser_state_destroy(&parser);
+    if (parser._tokenizer_state)
+      gumbo_tokenizer_state_destroy(&parser);
+    return NULL;
+  }
+
   output_init(&parser);
   gumbo_tokenizer_state_init(&parser, buffer, length);
   parser_state_init(&parser);
 
+  if (NULL == parser._parser_state)
+    return NULL;
+
   if (options->fragment_context != GUMBO_TAG_LAST) {
     fragment_parser_init(
         &parser, options->fragment_context, options->fragment_namespace);
@@ -4179,6 +4209,8 @@ void gumbo_destroy_node(GumboOptions* options, GumboNode* node) {
 }
 
 void gumbo_destroy_output(const GumboOptions* options, GumboOutput* output) {
+  if (NULL == output)
+    return;
   // Need a dummy GumboParser because the allocator comes along with the
   // options object.
   GumboParser parser;

src/parser.h

@@ -20,6 +20,8 @@
 #ifndef GUMBO_PARSER_H_
 #define GUMBO_PARSER_H_
 
+#include <setjmp.h>
+
 #ifdef __cplusplus
 extern "C" {
 #endif
@@ -48,6 +50,13 @@ typedef struct GumboInternalParser {
   // The internal parser state.  Initialized on parse start and destroyed on
   // parse end; end-users will never see a non-garbage value in this pointer.
   struct GumboInternalParserState* _parser_state;
+
+  // jmp_buf to jump to on out of memory
+#ifdef _WIN32
+  jmp_buf _oom_buf;
+#else
+  sigjmp_buf _oom_buf;
+#endif
 } GumboParser;
 
 #ifdef __cplusplus

src/util.c

@@ -17,11 +17,12 @@
 #include "util.h"
 
 #include <assert.h>
+#include <setjmp.h>
+#include <stdarg.h>
+#include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <strings.h>
-#include <stdarg.h>
-#include <stdio.h>
 
 #include "gumbo.h"
 #include "parser.h"
@@ -32,7 +33,16 @@
 const GumboSourcePosition kGumboEmptySourcePosition = {0, 0, 0};
 
 void* gumbo_parser_allocate(GumboParser* parser, size_t num_bytes) {
-  return parser->_options->allocator(parser->_options->userdata, num_bytes);
+  void *v = parser->_options->allocator(parser->_options->userdata, num_bytes);
+  if (NULL == v) {
+#ifndef _WIN32
+    siglongjmp(parser->_oom_buf, 1);
+#else
+    longjmp(parser->_oom_buf, 1);
+#endif
+  } else {
+    return memset(v, '\0', num_bytes);
+  }
 }
 
 void gumbo_parser_deallocate(GumboParser* parser, void* ptr) {

tests/parser.cc

@@ -1954,6 +1954,32 @@ TEST_F(GumboParserTest, FragmentWithNamespace) {
   EXPECT_EQ(0, GetChildCount(div));
 }
 
+TEST_F(GumboParserTest, OutOfMemory) {
+  GumboOptions options;
+  memcpy((void*)&options, (void*)&kGumboDefaultOptions, sizeof options);
+  struct Count {
+    size_t count = 0;
+    size_t count_max = 0;
+  } count;
+  options.userdata = (void*)&count;
+  options.allocator = [](void *userdata, size_t size) {
+    auto count = (Count*)userdata;
+    if (count->count) {
+      count->count--;
+      return malloc(size);
+    } else {
+      count->count = ++count->count_max;
+      return (void*)nullptr;
+    }
+  };
+  const char buf[] = "<html><head><title>dummy</title></head><body>some text</body></html>";
+  GumboOutput *output;
+  do {
+     output = gumbo_parse_with_options(&options, buf, sizeof buf - 1);
+  } while (output == nullptr);
+  gumbo_destroy_output(&options, output);
+}
+
 TEST_F(GumboParserTest, FragmentWithTwoNodes) {
   ParseFragment("<h1>Hi</h1><br>", GUMBO_TAG_BODY, GUMBO_NAMESPACE_HTML);