bind9.16/bind-9.16-isc_hp-CVE-2023-50387.patch
Petr Menšík fdd3436a6d Downstream specific changes related to KeyTrap
Fix for CVE-2023-50387 introduced new additional thread. But because
isc_hp functions were removed from later bind 9.16 release, their
changes did not contain increase of hazard pointers max thread limit.
To prevent obscure memory corruption increase thread max size.

In addition place at least few INSISTs to check this is catched before
random memory overwrites begins. It would be quite difficult to track
without any check.

Resolves: RHEL-25392
; Resolves: CVE-2023-50387
2024-02-21 20:57:49 +01:00

67 lines
2.2 KiB
Diff

From 103b09187466b2afbff7e204d166d21e2fbb057c Mon Sep 17 00:00:00 2001
From: Petr Mensik <pemensik@redhat.com>
Date: Wed, 21 Feb 2024 11:54:27 +0100
Subject: [PATCH] Downstream specific changes related to KeyTrap
Fix for CVE-2023-50387 introduced new additional thread. But because
isc_hp functions were removed from later bind 9.16 release, their
changes did not contain increase of hazard pointers max thread limit.
To prevent obscure memory corruption increase thread max size.
In addition place at least few INSISTs to check this is catched before
random memory overwrites begins. It would be quite difficult to track
without any check.
---
lib/isc/hp.c | 3 +++
lib/isc/managers.c | 5 +++--
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/lib/isc/hp.c b/lib/isc/hp.c
index 92d160b..5f9bbf7 100644
--- a/lib/isc/hp.c
+++ b/lib/isc/hp.c
@@ -138,6 +138,7 @@ isc_hp_destroy(isc_hp_t *hp) {
void
isc_hp_clear(isc_hp_t *hp) {
+ INSIST(tid() < isc__hp_max_threads);
for (int i = 0; i < hp->max_hps; i++) {
atomic_store_release(&hp->hp[tid()][i], 0);
}
@@ -152,6 +153,7 @@ uintptr_t
isc_hp_protect(isc_hp_t *hp, int ihp, atomic_uintptr_t *atom) {
uintptr_t n = 0;
uintptr_t ret;
+ INSIST(tid() < isc__hp_max_threads);
while ((ret = atomic_load(atom)) != n) {
atomic_store(&hp->hp[tid()][ihp], ret);
n = ret;
@@ -173,6 +175,7 @@ isc_hp_protect_release(isc_hp_t *hp, int ihp, atomic_uintptr_t ptr) {
void
isc_hp_retire(isc_hp_t *hp, uintptr_t ptr) {
+ INSIST(tid() < isc__hp_max_threads);
hp->rl[tid()]->list[hp->rl[tid()]->size++] = ptr;
INSIST(hp->rl[tid()]->size < isc__hp_max_retired);
diff --git a/lib/isc/managers.c b/lib/isc/managers.c
index c39a650..3bdca99 100644
--- a/lib/isc/managers.c
+++ b/lib/isc/managers.c
@@ -25,9 +25,10 @@ isc_managers_create(isc_mem_t *mctx, size_t workers, size_t quantum,
/*
* We have ncpus network threads, ncpus old network threads - make
- * it 4x just to be on the safe side.
+ * it 4x just to be on the safe side. One additional for slow netmgr
+ * thread.
*/
- isc_hp_init(4 * workers);
+ isc_hp_init(5 * workers);
REQUIRE(netmgrp != NULL && *netmgrp == NULL);
isc__netmgr_create(mctx, workers, &netmgr);
--
2.43.2