in_getmulti: recursing on if_addr_lock on malloc failure

[rlibby]

Summary

We recurse on the if_addr_lock on malloc M_NOWAIT failure in in_getmtuli. I believe the fix is just to invert the unlock and if_delmulti_ifma here. It looks like the obvious race is already possible between if_addmulti and the IF_ADDR_WLOCK above, therefore that race must be okay.

Test Plan

Isilon internal. Reproduce assertion with M_NOWAIT failure injection.

[jhb]

Yes, this is fine. This is just dropping a reference count that it acquired earlier in the function on failure.

[cem]

Rlibby doesn't have a src bit — do you want to commit this, JHB, or should I? Or should I wait for other reviews?

[jhb]

Go for it. You can mark me as having reviewed it.