tree 4212e284fa989d57ba8b694ed2fcd77abbdcb6c0
parent 498babbf1367b6558002cb23b1ff5993a65c76b3
author Kostya Kortchinsky <kostyak@google.com> 1580925532 -0800
committer Copybara-Service <copybara-worker@google.com> 1598597808 -0700

[scudo][standalone] Fix a race in the secondary release

Summary:
I tried to move the `madvise` calls outside of one of the secondary
mutexes, but this backfired. There is situation when a low release
interval is set combined with secondary pressure that leads to a race:
a thread can get a block from the cache, while another thread is
`madvise`'ing that block, resulting in a null header.

I changed the secondary race test so that this situation would be
triggered, and moved the release into the cache mutex scope.

Reviewers: cferris, pcc, eugenis, hctim, morehouse

Subscribers: jfb, #sanitizers, llvm-commits

Tags: #sanitizers, #llvm

Differential Revision: https://reviews.llvm.org/D74072

GitOrigin-RevId: a9d5f8989d83dee1ed01b7f8eaaab89ad190116a