While working on the serve-stale backports, I noticed the following
oddities:
1. In the serve-stale system test, in one case we keep track of the
time how long it took for dig to complete. In commit
aaed7f9d8c, the code removed the
exception to check for result == ISC_R_SUCCESS on stale found
answers, and adjusted the test accordingly. This failed to update
the time tracking accordingly. Move the t1/t2 time track variables
back around the two dig commands to ensure the lookups resolved
faster than the resolver-query-timeout.
2. We can remove the setting of NS_QUERYATTR_STALEOK and
DNS_RDATASETATTR_STALE_ADDED on the "else if (stale_timeout)"
code path, because they are added later when we know we have
actually found a stale answer on a stale timeout lookup.
3. We should clear the NS_QUERYATTR_STALEOK flag from the client
query attributes instead of DNS_RDATASETATTR_STALE_ADDED (that
flag is set on the rdataset attributes).
4. In 'bin/named/config.c' we should set the configuration options
in alpabetical order.
5. In the ARM, in the backports we have added "(stale)" between
"cached" and "RRset" to make more clear a stale RRset may be
returned in this scenario.
Exerting excessive I/O load on the host running system tests should be
avoided in order to limit the number of false positives reported by the
system test suite. In some cases, running named with "-d 99" (which is
the default for system tests) results in a massive amount of logs being
generated, most of which are useless. Implement a log file size check
to draw developers' attention to overly verbose named instances used in
system tests. The warning threshold of 200,000 lines was chosen
arbitrarily.
The regression test for CVE-2020-8620 causes a lot of useless messages
to be logged. However, globally decreasing the log level for the
affected named instance would be a step too far as debugging information
may be useful for troubleshooting other checks in the "tcp" system test.
Starting a separate named instance for a single check should be avoided
when possible and thus is also not a good solution. As a compromise,
run "rndc trace 1" for the affected named instance before starting the
regression test for CVE-2020-8620.
The system test framework starts all named instances with the "-d 99"
command line option (unless it is overridden by a named.args file in a
given instance's working directory). This causes a lot of log messages
to be written to named.run files - currently over 5 million lines for a
single test suite run. While debugging information preserved in the log
files is essential for troubleshooting intermittent test failures, some
system tests involve sending hundreds or even thousands of queries,
which causes the relevant log files to explode in size. When multiple
tests (or even multiple test suites) are run in parallel, excessive
logging contributes considerably to the I/O load on the test host,
increasing the odds of intermittent test failures getting triggered.
Decrease the debug level for the seven most verbose named instances:
- use "-d 3" for ns2 in the "cacheclean" system test (it is the lowest
logging level at which the test still passes without the need to
apply any changes to tests.sh),
- use "-d 1" for the other six named instances.
This roughly halves the number of lines logged by each test suite run
while still leaving enough information in the logs to allow at least
basic troubleshooting in case of test failures.
This approach was chosen as it results in a greater decrease in the
number of lines logged than running all named instances with "-d 3",
without causing any test failures.
The malloc attribute allows compiler to do some optmizations on
functions that behave like malloc/calloc, like assuming that the
returned pointer do not alias other pointers.
There is no possibility for mpctx->items to be NULL at the point where
the code was removed, since we enforce that fillcount > 0, if
mpctx->items == NULL when isc_mempool_get is called, then we will
allocate fillcount more items and add to the mpctx->items list.
_query_detach function was incorrectly unliking the query object from
the lookup->q query list, this made it impossible to follow a query
lookup failure with the next one in the list (possibly using a separate
resolver), as the link to the next query in the list was dissolved.
Fix by unliking the node only when the query object is about to be
destroyed, i.e. there is no more references to the object.
When the keymgr needs to create new keys, it is possible it needs to
create multiple keys. The keymgr checks for keyid conflicts with
already existing keys, but it should also check against that it just
created.
GitLab CI pipelines do not currently include a Linux job that would have
GSSAPI support disabled. Add the "--without-gssapi" option to the
./configure invocation on Debian 9 to address that deficiency and also
to continuously test that build-time switch.
If "tkey-gssapi-credential" is set in the configuration and GSSAPI
support is not available, named will refuse to start. As the test
system framework does not support starting named instances
conditionally, ensure that "tkey-gssapi-credential" is only present in
named.conf if GSSAPI support is available.
as with TLS, the destruction of a client stream on failed read
needs to be conditional: if we reached failed_read_cb() as a
result of a timeout on a timer which has subsequently been
reset, the stream must not be closed.
Add a check to the "dnssec" system test which ensures that RRSIG(SOA)
RRsets present anywhere else than at the zone apex are automatically
removed after a zone containing such RRsets is loaded.
If there happens to be a RRSIG(SOA) that is not at the zone apex
for any reason it should not be considered as a stopping condition
for incremental zone signing.
the destruction of the socket in tls_failed_read_cb() needs to be
conditional; if reached due to a timeout on a timer that has
subsequently been reset, the socket must not be destroyed.
this is similar in structure to the UDP timeout recovery test.
this commit adds a new mechanism to the netmgr test allowing the
listen socket to accept incoming TCP connections but never send
a response. this forces the client to time out on read.
when running read callbacks, if the event result is not ISC_R_SUCCESS,
the callback is always run asynchronously. this is a problem on timeout,
because there's no chance to reset the timer before the socket has
already been destroyed. this commit allows read callbacks to run
synchronously for both ISC_R_SUCCESS and ISC_R_TIMEDOUT result codes.
this test sets up a server socket that listens for UDP connections
but never responds. the client will always time out; it should retry
five times before giving up.
The test spawns 4 parallel workers that keep adding, modifying and
deleting zones, the main thread repeatedly checks wheter rndc
status responds within a reasonable period.
While environment and timing issues may affect the test, in most
test cases the deadlock that was taking place before the fix used to
trigger in less than 7 seconds in a machine with at least 2 cores.
It follows a description of the steps that were leading to the deadlock:
1. `do_addzone` calls `isc_task_beginexclusive`.
2. `isc_task_beginexclusive` waits for (N_WORKERS - 1) halted tasks,
this blocks waiting for those (no. workers -1) workers to halt.
...
isc_task_beginexclusive(isc_task_t *task0) {
...
while (manager->halted + 1 < manager->workers) {
wake_all_queues(manager);
WAIT(&manager->halt_cond, &manager->halt_lock);
}
```
3. It is possible that in `task.c / dispatch()` a worker is running a
task event, if that event blocks it will not allow this worker to
halt.
4. `do_addzone` acquires `LOCK(&view->new_zone_lock);`,
5. `rmzone` event is called from some worker's `dispatch()`, `rmzone`
blocks waiting for the same lock.
6. `do_addzone` calls `isc_task_beginexclusive`.
7. Deadlock triggered, since:
- `rmzone` is wating for the lock.
- `isc_task_beginexclusive` is waiting for (no. workers - 1) to
be halted
- since `rmzone` event is blocked it won't allow the worker to halt.
To fix this, we updated do_addzone code to call isc_task_beginexclusive
before the lock is acquired, we postpone locking to the nearest required
place, same for isc_task_beginexclusive.
The same could happen with rndc modzone, so that was addressed as well.
