Fix leak in early data

[duke8253]

The leak is happening due to unused 16k block when early data read failed.

[ywkaras]

It seems like a common pattern in our code that we call a free() member function to deallocate class instances. Maybe we should add a utility to itscore that uses std::unique_ptr to call free(). https://godbolt.org/z/G-g7CH It would help avoid accidental leaks like this in future code.

[duke8253]

I like this idea.

[zwoop]

Not sure how that's useful here, but maybe I'm missing something. You can't free things here when it goes out of scope, and you don't want to let it sit there if it's not going to be used either. Once used, it'll have to be freed up later, when it's no longer needed (you are collecting this early data as part of the handshake, to be used later by the HttpSM).

[ywkaras]

netvc->early_data_buf->append_block(block); would have to change to netvc->early_data_buf->append_block(block.release());

[zwoop]

Looking some more, I wonder if we could optimize for the common case (which is that the client fails to negotiate early data, at least for now). E.g. move this into inside the while() loop, and after the check if the early data read succeeded or not:

    if (netvc->early_data_buf == nullptr) {
      netvc->early_data_buf    = new_MIOBuffer(BUFFER_SIZE_INDEX_16K);
      netvc->early_data_reader = netvc->early_data_buf->alloc_reader();
    }

[duke8253]

Looking some more, I wonder if we could optimize for the common case (which is that the client fails to negotiate early data, at least for now). E.g. move this into inside the while() loop, and after the check if the early data read succeeded or not:

    if (netvc->early_data_buf == nullptr) {
      netvc->early_data_buf    = new_MIOBuffer(BUFFER_SIZE_INDEX_16K);
      netvc->early_data_reader = netvc->early_data_buf->alloc_reader();
    }

The reason I left it out was to not do the if check everytime it loops around, but I guess changing it will make it cleaner, and the check doesn't really take much time.

[duke8253]

[approve ci autest]

[zwoop]

[approve ci autest]

[shinrich]

Looks good to me. Seems everything in the discussion has been addressed.

[zwoop]

Tested on docs.

[zwoop]

Cherry-picked to v9.0.x branch.