Memory leak in `Linux::getdents64`

[tom91136]

@FinnWilkinson The function's malloc is missing the corresponding free:

SimEng/src/lib/kernel/Linux.cc

Line 542 in 62c2890

result.d_name = (char*)malloc(result.d_namlen + 1);

Reading through the getdents64 implementation, the whole logic around populating this result.d_name field is a bit weird.

• Some calls to memcpy is not qualified.
• Computation of result.d_namlen creates a copy of next_direct->d_name as std::string only to call size() and discard the string.
• Memory for that field is allocated but not freed, and this happens in a while loop where linux_dirent64 result does not escape.

Because dirent->d_name is guaranteed to be \0 terminated, we can just do this:

...
    linux_dirent64 result;
    result.d_ino = next_direct->d_ino;
#ifdef __MACH__
    result.d_off = next_direct->d_seekoff;
#else
    result.d_off = next_direct->d_off;
#endif
    std::string d_name = next_direct->d_name;
    result.d_type = next_direct->d_type;
    result.d_namlen = d_name.size();
    result.d_name = d_name.data();  // C++17's non-const overload, mutation of the returned char* is permitted
    // Get size of struct before alignment
    // 20 = combined size of d_ino, d_off, d_reclen, d_type, and d_name's
    // null-terminator
...

d_name's memory is now managed by RAII so all is sound.
Let me know if we want this version or just call free and I'll get a PR going.

PS: When we eventually switch to C++20, we probably want to rewrite the initialisation of linux_dirent64 to use designations.

[FinnWilkinson]

The proposed solution looks cleaner and avoids malloc and memcpy so I would say this one as long as the tests still pass