171 lines
6.9 KiB
Diff
171 lines
6.9 KiB
Diff
From 321a498e73e9fa00d69ee9cfa4756505e5e5c8e8 Mon Sep 17 00:00:00 2001
|
|
From: Evan Hunt <each@isc.org>
|
|
Date: Thu, 25 May 2023 23:53:50 -0700
|
|
Subject: [PATCH] CVE-2023-2828
|
|
|
|
Every `named` instance configured to run as a recursive resolver maintains a cache database holding the responses to the queries it has recently sent to authoritative servers. The size limit for that cache database can be configured using the `max-cache-size` statement in the configuration file; it defaults to 90% of the total amount of memory available on the host. When the size of the cache reaches 7/8 of the configured limit, a cache-cleaning algorithm starts to remove expired and/or least-recently used RRsets from the cache, to keep memory use below the configured limit. It has been discovered that the effectiveness of the cache-cleaning algorithm used in `named` can be severely diminished by querying the resolver for specific RRsets in a certain order, effectively allowing the configured `max-cache-size` limit to be significantly exceeded.
|
|
|
|
---
|
|
lib/dns/rbtdb.c | 105 ++++++++++++++++++++++++++++++------------------
|
|
1 file changed, 65 insertions(+), 40 deletions(-)
|
|
|
|
diff --git a/lib/dns/rbtdb.c b/lib/dns/rbtdb.c
|
|
index 1795f9d..c25023c 100644
|
|
--- a/lib/dns/rbtdb.c
|
|
+++ b/lib/dns/rbtdb.c
|
|
@@ -599,7 +599,7 @@ static void
|
|
expire_header(dns_rbtdb_t *rbtdb, rdatasetheader_t *header, bool tree_locked,
|
|
expire_t reason);
|
|
static void
|
|
-overmem_purge(dns_rbtdb_t *rbtdb, unsigned int locknum_start, isc_stdtime_t now,
|
|
+overmem_purge(dns_rbtdb_t *rbtdb, unsigned int locknum_start, size_t purgesize,
|
|
bool tree_locked);
|
|
static isc_result_t
|
|
resign_insert(dns_rbtdb_t *rbtdb, int idx, rdatasetheader_t *newheader);
|
|
@@ -6794,6 +6794,16 @@ cleanup:
|
|
|
|
static dns_dbmethods_t zone_methods;
|
|
|
|
+static size_t
|
|
+rdataset_size(rdatasetheader_t *header) {
|
|
+ if (!NONEXISTENT(header)) {
|
|
+ return (dns_rdataslab_size((unsigned char *)header,
|
|
+ sizeof(*header)));
|
|
+ }
|
|
+
|
|
+ return (sizeof(*header));
|
|
+}
|
|
+
|
|
static isc_result_t
|
|
addrdataset(dns_db_t *db, dns_dbnode_t *node, dns_dbversion_t *version,
|
|
isc_stdtime_t now, dns_rdataset_t *rdataset, unsigned int options,
|
|
@@ -6957,7 +6967,8 @@ addrdataset(dns_db_t *db, dns_dbnode_t *node, dns_dbversion_t *version,
|
|
}
|
|
|
|
if (cache_is_overmem) {
|
|
- overmem_purge(rbtdb, rbtnode->locknum, now, tree_locked);
|
|
+ overmem_purge(rbtdb, rbtnode->locknum, rdataset_size(newheader),
|
|
+ tree_locked);
|
|
}
|
|
|
|
NODE_LOCK(&rbtdb->node_locks[rbtnode->locknum].lock,
|
|
@@ -6976,10 +6987,18 @@ addrdataset(dns_db_t *db, dns_dbnode_t *node, dns_dbversion_t *version,
|
|
}
|
|
|
|
header = isc_heap_element(rbtdb->heaps[rbtnode->locknum], 1);
|
|
- if (header != NULL && header->rdh_ttl + rbtdb->serve_stale_ttl <
|
|
- now - RBTDB_VIRTUAL)
|
|
- {
|
|
- expire_header(rbtdb, header, tree_locked, expire_ttl);
|
|
+ if (header != NULL) {
|
|
+ dns_ttl_t rdh_ttl = header->rdh_ttl;
|
|
+
|
|
+ /* Only account for stale TTL if cache is not overmem */
|
|
+ if (!cache_is_overmem) {
|
|
+ rdh_ttl += rbtdb->serve_stale_ttl;
|
|
+ }
|
|
+
|
|
+ if (rdh_ttl < now - RBTDB_VIRTUAL) {
|
|
+ expire_header(rbtdb, header, tree_locked,
|
|
+ expire_ttl);
|
|
+ }
|
|
}
|
|
|
|
/*
|
|
@@ -10495,52 +10514,58 @@ update_header(dns_rbtdb_t *rbtdb, rdatasetheader_t *header, isc_stdtime_t now) {
|
|
ISC_LIST_PREPEND(rbtdb->rdatasets[header->node->locknum], header, link);
|
|
}
|
|
|
|
+static size_t
|
|
+expire_lru_headers(dns_rbtdb_t *rbtdb, unsigned int locknum, size_t purgesize,
|
|
+ bool tree_locked) {
|
|
+ rdatasetheader_t *header, *header_prev;
|
|
+ size_t purged = 0;
|
|
+
|
|
+ for (header = ISC_LIST_TAIL(rbtdb->rdatasets[locknum]);
|
|
+ header != NULL && purged <= purgesize; header = header_prev)
|
|
+ {
|
|
+ header_prev = ISC_LIST_PREV(header, link);
|
|
+ /*
|
|
+ * Unlink the entry at this point to avoid checking it
|
|
+ * again even if it's currently used someone else and
|
|
+ * cannot be purged at this moment. This entry won't be
|
|
+ * referenced any more (so unlinking is safe) since the
|
|
+ * TTL was reset to 0.
|
|
+ */
|
|
+ ISC_LIST_UNLINK(rbtdb->rdatasets[locknum], header, link);
|
|
+ size_t header_size = rdataset_size(header);
|
|
+ expire_header(rbtdb, header, tree_locked, expire_lru);
|
|
+ purged += header_size;
|
|
+ }
|
|
+
|
|
+ return (purged);
|
|
+}
|
|
+
|
|
/*%
|
|
- * Purge some expired and/or stale (i.e. unused for some period) cache entries
|
|
- * under an overmem condition. To recover from this condition quickly, up to
|
|
- * 2 entries will be purged. This process is triggered while adding a new
|
|
- * entry, and we specifically avoid purging entries in the same LRU bucket as
|
|
- * the one to which the new entry will belong. Otherwise, we might purge
|
|
- * entries of the same name of different RR types while adding RRsets from a
|
|
- * single response (consider the case where we're adding A and AAAA glue records
|
|
- * of the same NS name).
|
|
+ * Purge some stale (i.e. unused for some period - LRU based cleaning) cache
|
|
+ * entries under the overmem condition. To recover from this condition quickly,
|
|
+ * we cleanup entries up to the size of newly added rdata (passed as purgesize).
|
|
+ *
|
|
+ * This process is triggered while adding a new entry, and we specifically avoid
|
|
+ * purging entries in the same LRU bucket as the one to which the new entry will
|
|
+ * belong. Otherwise, we might purge entries of the same name of different RR
|
|
+ * types while adding RRsets from a single response (consider the case where
|
|
+ * we're adding A and AAAA glue records of the same NS name).
|
|
*/
|
|
static void
|
|
-overmem_purge(dns_rbtdb_t *rbtdb, unsigned int locknum_start, isc_stdtime_t now,
|
|
+overmem_purge(dns_rbtdb_t *rbtdb, unsigned int locknum_start, size_t purgesize,
|
|
bool tree_locked) {
|
|
- rdatasetheader_t *header, *header_prev;
|
|
unsigned int locknum;
|
|
- int purgecount = 2;
|
|
+ size_t purged = 0;
|
|
|
|
for (locknum = (locknum_start + 1) % rbtdb->node_lock_count;
|
|
- locknum != locknum_start && purgecount > 0;
|
|
+ locknum != locknum_start && purged <= purgesize;
|
|
locknum = (locknum + 1) % rbtdb->node_lock_count)
|
|
{
|
|
NODE_LOCK(&rbtdb->node_locks[locknum].lock,
|
|
isc_rwlocktype_write);
|
|
|
|
- header = isc_heap_element(rbtdb->heaps[locknum], 1);
|
|
- if (header && header->rdh_ttl < now - RBTDB_VIRTUAL) {
|
|
- expire_header(rbtdb, header, tree_locked, expire_ttl);
|
|
- purgecount--;
|
|
- }
|
|
-
|
|
- for (header = ISC_LIST_TAIL(rbtdb->rdatasets[locknum]);
|
|
- header != NULL && purgecount > 0; header = header_prev)
|
|
- {
|
|
- header_prev = ISC_LIST_PREV(header, link);
|
|
- /*
|
|
- * Unlink the entry at this point to avoid checking it
|
|
- * again even if it's currently used someone else and
|
|
- * cannot be purged at this moment. This entry won't be
|
|
- * referenced any more (so unlinking is safe) since the
|
|
- * TTL was reset to 0.
|
|
- */
|
|
- ISC_LIST_UNLINK(rbtdb->rdatasets[locknum], header,
|
|
- link);
|
|
- expire_header(rbtdb, header, tree_locked, expire_lru);
|
|
- purgecount--;
|
|
- }
|
|
+ purged += expire_lru_headers(rbtdb, locknum, purgesize - purged,
|
|
+ tree_locked);
|
|
|
|
NODE_UNLOCK(&rbtdb->node_locks[locknum].lock,
|
|
isc_rwlocktype_write);
|
|
--
|
|
2.33.0
|
|
|