ktls: Fix an off-by-one bug in tls13_find_record_type()

[markj]

Summary

If the entire plaintext is zero-filled, the backwards walk in
tls13_find_record_type() would return the offset of the last byte of the
TLS header. This causes an underflow when decrypting, resulting in a
null pointer dereference.

Fix the condition and add a regression test.

[jhb]

sys/kern/uipc_ktls.c

2393

I wonder if it would have been clearer if I had written this as last_offset = offset + m_len - 1; originally as then then you wouldn't need the + 1 when computing trailer_len at the end and the expression you are fixing would have been correct (it is clearly wrong now). From the comment above the while loop it does seem like it should be offset of the last non-zero byte. Mostly because while reviewing this fix it took me a bit to understand what last_offset meant, and in particular the disconnect between m_len - 1 being the array index into cp but m_len being used to compute last_offset was a bit jarring when re-reading this.

tests/sys/kern/ktls_test.c

2837–2842

Maybe just make this a compound sentence so it reads easier

[markj]

Apply John's suggestions