When a catalog zone is updated using AXFR, the zone database is changed,
so it is required to unregister the update notification callback from
the old database, and register it for the new one.
Currently, here is the order of the steps happening in such scenario:
1. The zone.c:zone_startload() function registers the notify callback
on the new database using dns_zone_catz_enable_db()
2. The callback, when called, notices that the new 'db' is different
than 'catz->db', and unregisters the old callback for 'catz->db',
marks that it's unregistered by setting 'catz->db_registered' to
false, then it schedules an update if it isn't already scheduled.
3. The offloaded update process, after completing its job, notices that
'catz->db_registered' is false, and (re)registers the update callback
for the current database it is working on. There is no harm here even
if it was registered also on step 1, and we can't skip it, because
this function can also be called "artificially" during a
reconfiguration, and in that case the registration step is required
here.
A problem arises when before step 1 an update process was already
in a running state, operating on the old database, and finishing its
work only after step 2. As described in step 3, dns__catz_update_cb()
notices that 'catz->db_registered' is false and registers the callback
on the current database it is working on, which, at that state, is
already obsolete and unused by the zone. When it detaches the database,
the function which is responsible for its cleanup (e.g. free_rbtdb())
asserts because there is a registered update notify callback there.
To fix the problem, instead of delaying the (re)registration to step 3,
make sure that the new callback is registered and 'catz->db_registered'
is accordingly marked on step 2.
The purpose of the check is to verify the server has survived the
previous barrage of queries. This is done by sending a query and
checking we get a NOERROR response back.
Previously, that query could've been affected by a servfail cache - the
server would return a SERVFAIL answer, thus failing the check, despite
being up and running. Use version.bind txt ch query to avoid the
interference of servfail cache.
When create_fetch() in the dns_validator unit detects deadlock, it
returns DNS_R_NOVALIDSIG, but it didn't attach to the validator. The
other condition to returning result != ISC_R_SUCCESS would be error from
dns_resolver_createfetch(). The caller (in two places out of three)
would detect the error condition and always detach from the validator.
Move the dns_validator_detach() on dns_resolver_createfetch() error
condition to create_fetch() function and cleanup the extra detaches in
seek_dnskey() and get_dsset().
This commit changes send buffers allocation strategy for stream based
transports. Before that change we would allocate a dynamic buffers
sized at 64Kb even when we do not need that much. That could lead to
high memory usage on server. Now we resize the send buffer to match
the size of the actual data, freeing the memory at the end of the
buffer for being reused later.
dns_view_find* may be called after the final call to dns_view_detach
is made which detaches view->zonetable to permit the server to
shutdown. We need to detect if view->zonetable is NULL during this
stage and appropriately recover.
In some cases, the inlined version rcu_dereference() would not compile
when working on pointer to opaque struct (namely Ubuntu Jammy). Detect
such condition in the autoconf and disable the inlining of the small
functions if it breaks the build.
The number of clients per query is calculated using the pending
fetch responses in the list. The dns_resolver_createfetch() function
includes every item in the list when deciding whether the limit is
reached (i.e. fctx->spilled is true). Then, when the limit is reached,
there is another calculation in fctx_sendevents(), when deciding
whether it is needed to increase the limit, but this time the TRYSTALE
responses are not included in the calculation (because of early break
from the loop), and because of that the limit is never increased.
A single client can have more than one associated response/event in the
list (currently max. two), and calculating them as separate "clients"
is unexpected. E.g. if 'stale-answer-enable' is enabled and
'stale-answer-client-timeout' is enabled and is larger than 0, then
each client will have two events, which will effectively halve the
clients-per-query limit.
Fix the dns_resolver_createfetch() function to calculate only the
regular FETCHDONE responses/events.
Change the fctx_sendevents() function to also calculate only FETCHDONE
responses/events. Currently, this second change doesn't have any impact,
because the TRYSTALE events were already skipped, but having the same
condition in both places will help prevent similar bugs in the future
if a new type of response/event is ever added.
Prepare the fetchlimit system test for adding a clients-per-query
check. Change some functions and commands to accept a destination
NS IP address instead of using the hardcoded 10.53.0.3.
Tom Krizek
Arаm Sаrgsyаn
Michal Nowak
Ondřej Surý
Artem BoldarievandOndřej Surý
Mark Andrews