Stop wasting time on malloc in snprintf_zstd_header

[rincebrain]

Profiling zdb -vvvvv on datasets with a lot of zstd blocks, we find ourselves spending quite a lot of time on malloc/free, because we allocate a 16M abd each call, and never free it, so we're leaking 16M per call as well.

This seems sub-optimal. So let's just keep the buffer around and reuse it.

Motivation and Context

Leaking 16M per zstd block seems bad, actually.

Description

It's a static allocation. That's really it.

How Has This Been Tested?

It ran and I was no longer spending all my time in malloc/free/clear_page_erms.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Library ABI change (libzfs, libzfs_core, libnvpair, libuutil and libzfsbootenv)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the OpenZFS code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• I have run the ZFS Test Suite with this change applied.
• All commit messages are properly formatted and contain Signed-off-by.

[amotin]

Sure leaks per block are bad, but why do we need a leek at all? Couldn't we allocate just as much as we need instead of 16MB and then free?

[rincebrain]

The problem is that if you allocate and free each loop, then you spend most of your time in malloc/free/the bzero equivalent.

So yes, I could punt the allocation up a level and make all the callers hand in an abd to use, and refactor them all to not allocate naively, but it's not really clear to me why that's a better fit here.

[robn]

I don't love it, but it does the job and is obviously better than before.

I did enough code read to satisfy myself that there's only gonna be one thread calling this.

[rincebrain]

If nothing else, I'm pretty sure the print output would get horrendously mangled if it were called in parallel without explicit locking.