gdb显示奇怪的堆栈跟踪

Question

I have a C++ daemon which segfaults after few days of work. I compiled it with debug options (I'm sure I did it well, because I tested it with premeditated crashes and gdb showed correct stack trace), but in "real" crash on production I see only following trace:

(gdb) where
#0  0x00007ffff674d5a7 in ?? () from /lib/x86_64-linux-gnu/libc.so.6
#1  0xffffffffffffffff in ?? ()
#2  0x0000000000000000 in ?? ()

What does it mean?

The following source code has potential issue, because it is the only new code since daemon become unstable:

namespace Foo {
    Bar* Bar::instance = NULL;

    Bar* Bar::getInstance() {
        if (!instance)
            instance = new Bar();

        return instance;
    }

    Bar::Bar() {
        curl = curl_easy_init();

        if(CURLE_OK != curl_easy_setopt(curl, CURLOPT_WRITEFUNCTION, &data_write)
        || CURLE_OK != curl_easy_setopt(curl, CURLOPT_NOPROGRESS, 1L)
        || CURLE_OK != curl_easy_setopt(curl, CURLOPT_FOLLOWLOCATION, 1L)
        || CURLE_OK != curl_easy_setopt(curl, CURLOPT_TIMEOUT_MS, Bar::timeout)) {
            throw std::runtime_error(std::string("Can't initialize curl."));
        }
    }

    Bar::~Bar() {
        curl_easy_cleanup(curl);
    }

    std::string Bar::getByIp(const std::string &id) {
        Bar *self = getInstance();
        std::string url = "example.com";
        url.append(id); 

        std::ostringstream oss;

        if (CURLE_OK == self->curl_read(url, oss)) {
            std::string output(oss.str());

            if (output.empty())
                return NULL_OBJECT;

            TiXmlDocument xml;
            xml.Parse(output.c_str());

           if (
                xml.Error()
                || !xml.FirstChild("a")
                || !xml.FirstChild("a")->FirstChild("b")
                || !xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lat")
                || !xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lng")
            )
                return NULL_OBJECT;

            std::string lat = xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lat")->GetText();
            std::string lng = xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lng")->GetText();

            return Region::getByCoordinates(lng, lat);
        }

        return NULL_OBJECT;
    }

    size_t Bar::data_write(void* buf, size_t size, size_t nmemb, void* userp)
    {
        if(userp)
        {
            std::ostream& os = *static_cast<std::ostream*>(userp);
            std::streamsize len = size * nmemb;
            if(os.write(static_cast<char*>(buf), len))
                return len;
        }

        return 0;
    }

    CURLcode Bar::curl_read(const std::string& url, std::ostringstream& os)
    {
        CURLcode code(CURLE_FAILED_INIT);

        if(curl)
        {
            if(
                CURLE_OK == (code = curl_easy_setopt(curl, CURLOPT_FILE, &os))
                && CURLE_OK == (code = curl_easy_setopt(curl, CURLOPT_URL, url.c_str()))
            ) {
                code = curl_easy_perform(curl);
            }
        }

        return code;
    }
}

Answer 1

Shouldn't the second if in your getByIp(...) method look like this?

if (xml.Error()
    || !xml.FirstChild("a")
    || !xml.FirstChild("a")->FirstChild("b")
    || !xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lat")
    || !xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lng")) // <- added missing parenthesis
{ // <- added
    std::string lat = xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lat")->GetText();
    std::string lng = xml.FirstChild("a")->FirstChild("b")->FirstChildElement("lng")->GetText();
    return Region::getByCoordinates(lng, lat);
} // <- added

If you are indeed missing the curly braces, you might be dereferencing invalid pointers when you retrieve the lng string, because its retrieval is then not a part of the conditional statement.

Answer 2

It looks like a memory corruption affecting the stack: writing outside of the allocated memory.

You can write a little program to exercise your Bar class the same way your daemon does, possibly in a loop. You can also run this program with MALLOC_CHECK_ , electric fence , Valgrind or any other memory checking tool.

It could be curl, TiXmlDocument, or the code calling your class.

Answer 3

I expect you have a few issues with pointers. As StackTrace #0 shows a function within libc.so.6 that does not have a name. And #1 has a NULL -1 pointer. We need a lot more to help you fix your bug.