Check length before reading in stringmatchlen

[thaliaarchi]

Fixes four cases where stringmatchlen could overrun the pattern if it is not terminated with NUL.

These commits are cherry-picked from my fork which extracts stringmatch as a library and compares it to other projects by antirez which use the same matcher.

[zuiderkwast]

Thanks!

This is not a security issue, right? We always call this with null-terminated strings, I believe.

It's still good to remove the assumption about null-terminated strings and never look beyond the last char of the strings.

[thaliaarchi]

I count 20 calls in Valkey to stringmatch/stringmatchlen. I presume calls to stringmatch are safe, since callers know it expects a NUL-terminated string. Every call to stringmatchlen passes a pattern from an sds type using length sdslen(pattern). The sds API seems to guarantee that its strings are NUL-terminated. Thus, unless there's a bug in sds or callers break the stringmatch expectation, this bug is a non-issue. It's just defensive, not a security issue. However, I haven't traced the pattern initializations to be sure, because many are stored in more complex data structures.

[zuiderkwast]

Thanks! I edited the top comment because we use it as the commit message when squash-merging.

[thaliaarchi]

Thanks for the edit. I would have put the extra context in another comment if I knew it would be in the message. Oh well.