Through Reddit I've found a nice blog post, "Using Static Analysis And Clang To Find Heartbleed":
http://blog.trailofbits.com/2014/04/27/using-static-analysis-and-clang-to-find-heartbleed/


The part about the range analysis is nice:

Ranges of symbol values:
 conj_$2{int} : { [-2147483648, -2], [0, 2147483647] }
 conj_$9{uint32_t} : { [0, 6] }

Unlike D, here the Clang compiler is keeping and managing those ranges across different lines of code. This means that here C/C++ is more modern&advanced than D.


Then the blog post shows a small C program that contains a Heartbleed-like bug:


int data_array[] = { 0, 18, 21, 95, 43, 32, 51};

int main(int argc, char *argv[]) {
  int   fd;
  char  buf[512] = {0};

  fd = open("dtin", O_RDONLY);

  if(fd != -1) {
    int size;
    int res;

    res = read(fd, &size, sizeof(int));

    if(res == sizeof(int)) {
      size = ntohl(size);

      if(size < sizeof(data_array)) {
        memcpy(buf, data_array, size);
      }

      memcpy(buf, data_array, size);
    }

    close(fd);
  }

  return 0;
}



Using a language a little less primitive than C helps avoid those bugs and to avoid several other mistakes that can happen in that C example code.

This is hypothetical equivalent D code that I think it's less bug-prone (perhaps I've missed some little part of the original C code):


immutable int[$] data = [0, 18, 21, 95, 43, 32, 51];

void main() {
    ubyte[512] buf;

    auto fIn = "dtin".File;
    scope(exit) fIn.close;

    immutable sizeBytes = fIn.tryRead!uint.Maybe!bswap;

    if (!sizeBytes.isNull) {
        if (sizeBytes.get < data.sizeof) {
            copyBytes(buf, data, sizeBytes.get);
        }

        copyBytes(buf, data, sizeBytes.get);
    }
}


Notes:

The "int[$] =" syntax means data is a fixed size array where the length is inferred. This is a possible syntax to avoid bug-prone code like (that currently compiles):
immutable int[8] data = [0, 18, 21, 95, 43, 32, 51];
An alternative syntax:
immutable data = [0, 18, 21, 95, 43, 32, 51]s;

scope(exit) helps avoid to forget to close the file when the scope ends. It's not essential in D, that closes the File using RAII and reference counting.

Instead of the ugly and bug-prone mess of "read(fd, &size, sizeof(int));" I have used something nicer. "tryRead!uint" tries to read an uint, and returns a Nullable!uint. The "Maybe" is a function currently missing in Phobos (https://issues.dlang.org/show_bug.cgi?id=12679 ), it's used to chain other functions on nullables, so bswap gets called only if tryRead doesn't return a null value. If tryRead returns a null, then Maybe returns a nulled Nullable (currently I think "tryRead" is not present in Phobos, but it's not hard to add something similar, using parse).

copyBytes is a template function that accepts both dynamic arrays and fixed-static arrays (if the input is a fixed-static array it doesn't throw away this precious compile-time information, as usually done by Phobos), that is much safer than memcpy. In non-release mode the code inside copyBytes will give a out of range error.

There are simple ways to reduce the template bloat caused by handling the length of fixed size arrays in library code: introducing in D a simple syntax to handle ghost typing (or here ghost values), that allows to manage types (and compile-time values) to enforce compile-time invariants, but that do not generate distinct instantiations of code and data structures, like templates on regular types (and regular compile-time values) do.

The D version also uses better names that make the semantics more clear, uses immutability (that is partially available in C too, with "const" that is quite underused in C code), and is less cluttered and noisy, this helps spot the bugs better.

In general I suggest to avoid using the raw C functions in D. Better to wrap them inside something a little safer. If such wrappers are standard Phobos functions it's even better:
https://d.puremagic.com/issues/show_bug.cgi?id=12548

Bye,
bearophile

Reply via email to