STL's std::regex_search(buffers, match, pattern) leads to stack overflow and segfault in Gramine

[dimakuv]

The std::regex_search(buffers, match, pattern) function from STL C++ standard library may pose problems in Gramine.

STL regex-search implementation is recursive, so it may create a bunch of stack frames during execution. In fact, you can see that even the native execution overflows the stack when the stack size is limited on trivial programs:

$ g++ -o stl_regex_search stl_regex_search.cxx -fsanitize=address

$ ulimit -s 256
# this command sets the stack size of the current session (and all programs executed in this session) to 256KB

$ ./stl_regex_search
AddressSanitizer:DEADLYSIGNAL
=================================================================
==2411980==ERROR: AddressSanitizer: stack-overflow on address 0x7ffea7880ff8
... 248 stack frames ...
SUMMARY: AddressSanitizer: stack-overflow (stl_regex_search+0x96a1b) in std::__detail::_BracketMatcher<std::__cxx11::regex_traits<char>, false, false>::_M_apply(char, std::integral_constant<bool, true>) const
==2411980==ABORTING

When running such programs under Gramine, Gramine will segfault or go crazy because it has a small default stack size of 256KB.

In fact, this is a known issue:

• https://groups.google.com/g/comp.lang.c++/c/iaqXIVHn9d4
• https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86164

[dimakuv]

Solutions

1. Set sys.stack.size = "1M" or a greater value.
2. Use Boost because its implementation of regex-match is non-recursive on Linux: https://www.boost.org/doc/libs/1_82_0/libs/regex/doc/html/boost_regex/configuration/algorithm.html