diff --git a/CHANGES b/CHANGES index 1d225c08ee..e04878f596 100644 --- a/CHANGES +++ b/CHANGES @@ -1,3 +1,16 @@ + --- 9.16.42 released --- + +6192. [security] A query that prioritizes stale data over lookup + triggers a fetch to refresh the stale data in cache. + If the fetch is aborted for exceeding the recursion + quota, it was possible for 'named' to enter an infinite + callback loop and crash due to stack overflow. This has + been fixed. (CVE-2023-2911) [GL #4089] + +6190. [security] Improve the overmem cleaning process to prevent the + cache going over the configured limit. (CVE-2023-2828) + [GL #4055] + 6183. [bug] Fix a serve-stale bug where a delegation from cache could be returned to the client. [GL #3950] diff --git a/doc/arm/notes.rst b/doc/arm/notes.rst index 10dd7caedf..e3b60315ef 100644 --- a/doc/arm/notes.rst +++ b/doc/arm/notes.rst @@ -47,6 +47,7 @@ for Microsoft Windows operating systems. .. include:: ../notes/notes-known-issues.rst .. include:: ../notes/notes-current.rst +.. include:: ../notes/notes-9.16.42.rst .. include:: ../notes/notes-9.16.41.rst .. include:: ../notes/notes-9.16.40.rst .. include:: ../notes/notes-9.16.39.rst diff --git a/doc/arm/reference.rst b/doc/arm/reference.rst index 9a2f4af1ba..55af7b8b42 100644 --- a/doc/arm/reference.rst +++ b/doc/arm/reference.rst @@ -3095,6 +3095,11 @@ system. default value of that option (90% of physical memory for each individual cache) may lead to memory exhaustion over time. + .. note:: + + ``max-cache-size`` does not work reliably for the maximum + amount of memory of 100 MB or lower. + Upon startup and reconfiguration, caches with a limited size preallocate a small amount of memory (less than 1% of ``max-cache-size`` for a given view). This preallocation serves as an diff --git a/doc/notes/notes-9.16.42.rst b/doc/notes/notes-9.16.42.rst new file mode 100644 index 0000000000..85b0edeabd --- /dev/null +++ b/doc/notes/notes-9.16.42.rst @@ -0,0 +1,45 @@ +.. Copyright (C) Internet Systems Consortium, Inc. ("ISC") +.. +.. SPDX-License-Identifier: MPL-2.0 +.. +.. This Source Code Form is subject to the terms of the Mozilla Public +.. License, v. 2.0. If a copy of the MPL was not distributed with this +.. file, you can obtain one at https://mozilla.org/MPL/2.0/. +.. +.. See the COPYRIGHT file distributed with this work for additional +.. information regarding copyright ownership. + +Notes for BIND 9.16.42 +---------------------- + +Security Fixes +~~~~~~~~~~~~~~ + +- The overmem cleaning process has been improved, to prevent the cache + from significantly exceeding the configured ``max-cache-size`` limit. + (CVE-2023-2828) + + ISC would like to thank Shoham Danino from Reichman University, Anat + Bremler-Barr from Tel-Aviv University, Yehuda Afek from Tel-Aviv + University, and Yuval Shavitt from Tel-Aviv University for bringing + this vulnerability to our attention. :gl:`#4055` + +- A query that prioritizes stale data over lookup triggers a fetch to + refresh the stale data in cache. If the fetch is aborted for exceeding + the recursion quota, it was possible for :iscman:`named` to enter an + infinite callback loop and crash due to stack overflow. This has been + fixed. (CVE-2023-2911) :gl:`#4089` + +Bug Fixes +~~~~~~~~~ + +- Previously, it was possible for a delegation from cache to be returned + to the client after the ``stale-answer-client-timeout`` duration. + This has been fixed. :gl:`#3950` + +Known Issues +~~~~~~~~~~~~ + +- There are no new known issues with this release. See :ref:`above + ` for a list of all known issues affecting this + BIND 9 branch. diff --git a/lib/dns/rbtdb.c b/lib/dns/rbtdb.c index c32db44cad..ee06c51e92 100644 --- a/lib/dns/rbtdb.c +++ b/lib/dns/rbtdb.c @@ -605,7 +605,7 @@ 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 void resign_insert(dns_rbtdb_t *rbtdb, int idx, rdatasetheader_t *newheader); @@ -6823,6 +6823,16 @@ 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, @@ -6987,7 +6997,8 @@ 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, @@ -7006,10 +7017,18 @@ 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); + } } /* @@ -10547,52 +10566,58 @@ 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); diff --git a/lib/ns/query.c b/lib/ns/query.c index 1d7f70ae1c..5cf502eef0 100644 --- a/lib/ns/query.c +++ b/lib/ns/query.c @@ -5719,6 +5719,7 @@ query_refresh_rrset(query_ctx_t *orig_qctx) { qctx.client->query.dboptions &= ~(DNS_DBFIND_STALETIMEOUT | DNS_DBFIND_STALEOK | DNS_DBFIND_STALEENABLED); + qctx.client->nodetach = false; /* * We'll need some resources... @@ -5976,7 +5977,14 @@ query_lookup(query_ctx_t *qctx) { "%s stale answer used, an attempt to " "refresh the RRset will still be made", namebuf); + qctx->refresh_rrset = STALE(qctx->rdataset); + + /* + * If we are refreshing the RRSet, we must not + * detach from the client in query_send(). + */ + qctx->client->nodetach = qctx->refresh_rrset; } } else { /* @@ -6324,7 +6332,7 @@ ns_query_recurse(ns_client_t *client, dns_rdatatype_t qtype, dns_name_t *qname, if (recparam_match(&client->query.recparam, qtype, qname, qdomain)) { ns_client_log(client, NS_LOGCATEGORY_CLIENT, NS_LOGMODULE_QUERY, ISC_LOG_INFO, "recursion loop detected"); - return (ISC_R_FAILURE); + return (ISC_R_ALREADYRUNNING); } recparam_update(&client->query.recparam, qtype, qname, qdomain); @@ -7288,10 +7296,21 @@ query_usestale(query_ctx_t *qctx, isc_result_t result) { return (false); } - if (result == DNS_R_DUPLICATE || result == DNS_R_DROP) { + if (qctx->refresh_rrset) { + /* + * This is a refreshing query, we have already prioritized + * stale data, so don't enable serve-stale again. + */ + return (false); + } + + if (result == DNS_R_DUPLICATE || result == DNS_R_DROP || + result == ISC_R_ALREADYRUNNING) + { /* * Don't enable serve-stale if the result signals a duplicate - * query or query that is being dropped. + * query or a query that is being dropped or can't proceed + * because of a recursion loop. */ return (false); } @@ -11554,12 +11573,7 @@ ns_query_done(query_ctx_t *qctx) { /* * Client may have been detached after query_send(), so * we test and store the flag state here, for safety. - * If we are refreshing the RRSet, we must not detach from the client - * in the query_send(), so we need to override the flag. */ - if (qctx->refresh_rrset) { - qctx->client->nodetach = true; - } nodetach = qctx->client->nodetach; query_send(qctx->client);