The Wiz post says the root cause is fatal errors (mostly out of memory) leaving corrupted state behind, which "a later request or worker can reuse". So the memory disclosure looks like it depends on the same FPM worker living on into the next request.
Would a low pm.max_requests narrow that part until people can patch? It obviously wouldn't help with the OPcache type confusion, since OPcache is shared memory across workers. airtnp also points out the old-branch backport just disables session upload progress, so session.upload_progress.enabled=0 looks like the cheaper interim knob.
Would a low pm.max_requests narrow that part until people can patch? It obviously wouldn't help with the OPcache type confusion, since OPcache is shared memory across workers. airtnp also points out the old-branch backport just disables session upload progress, so session.upload_progress.enabled=0 looks like the cheaper interim knob.
https://www.wiz.io/blog/php2shell-php-rce