There are existing tests for simulating timeouts, read errors, and
refused connecion errors. Implement also "network unreachable"
simulation.
Use "fixed" string search mode `-F` for `grep` in more places where
it is appropriate to do so.
DiG implements different logic in the `recv_done()` callback function
when processing a failure:
1. For a timed-out query it applies the "retries" logic first, then,
when it fails, fail-overs to the next server.
2. For an EOF (end-of-file, or unexpected disconnect) error it tries to
make a single retry attempt (even if the user has requested more
retries), then, when it fails, fail-overs to the next server.
3. For other types of failures, DiG does not apply the "retries" logic,
and tries to fail-over to the next servers (again, even if the user
has requested to make retries).
Simplify the logic and apply the same logic (1) of first retries, and
then fail-over, for different types of failures in `recv_done()`.
When the `send_done()` callback function gets called with a failure
result code, DiG erroneously cancels the lookup.
Stop canceling the lookup and give DiG a chance to retry the failed
query, or fail-over to another server, using the logic implemented in
the `recv_done()` callback function.
When the `udp_ready()` callback function gets called with a failure
result code, DiG erroneously cancels the lookup.
Copy the logic behind `tcp_connected()` callback function into
`udp_ready()` so that DiG will now retry the failed query (if retries
are enabled) and then, if it fails again, it will fail-over to the next
server in the list, which synchronizes the behavior between TCP and UDP
modes.
Also, `udp_ready()` was calling `lookup_detach()` without calling
`lookup_attach()` first, but the issue was masked behind the fact
that `clear_current_lookup()` wasn't being called when needed, and
`lookup_detach()` was compensating for that. This also has been fixed.
Luckily we don't rely on SphinxDirective functionality which does not
exist in 1.6.7. Replace it with docutils Directive.
transform_content() callback was added only in Sphinx 3.0.0.
Detect if it was not called and call it manually.
The transform_content() function requires access to inner "contentnode"
which is created inside run(). This workaround relies on the order of
node as it was in the pre-3.0.0 versions, but it should not matter as
new versions will not trigger the workaround.
Since !6413 we discourage opt-out, so we should not be advertising it in
the examples. Even worse, it was just thrown into the command line
without even mentioning its meaning in the surrounding text.
Related: !6413
Reject loading of zones with TTL higher than the max-zone-ttl
from the dnssec-policy.
With this change, any zone with a dnssec-policy in use will ignore
the max-zone-ttl option in zone/view/options.
The "max-zone-ttl" option should now be configured as part of
"dnssec-policy". The option with the same name in "zone" and
"options" is hereby flagged as deprecated, and its functionality
will be removed in a future release.
Now that the respdiff tests can detect memory leaks, it is worth running
them for every merge request. However, the existing respdiff-based
tests take a while to complete (about half an hour with our current CI
infrastructure), which does not make them a good fit for this purpose.
Add a new GitLab CI job, "respdiff-short", which uses a smaller query
set that gets processed within a couple of minutes on our current CI
infrastructure. Rename the existing respdiff-based jobs to make
distinguishing them easier.
Ensure the common parts of all jobs using respdiff are available in the
form of a reusable YAML anchor, to reduce code duplication and to
simplify adding more respdiff-based jobs to GitLab CI.
The "respdiff" GitLab CI job compares DNS responses produced by the
current version of named with those produced by a reference version.
The latter is built from source in each "respdiff" job, despite the fact
that the reference version changes very rarely. Use a pre-built named
executable as the reference version instead, assuming it is available in
the OS image used for "respdiff" tests.
These calls have not been needed since OpenSSL 0.9.7h.
This dates to commit 704d6eeab1, "Work
around non-reentrancy in openssl by disabling precomputation in keys".
This was in the bundled OpenSSL 0.9.3a era and made two changes. First,
it registered a locking callback because, in those days, OpenSSL needed
a callback to support locks. Second, it set flags to disable various
bits of cached state on DH, DSA, and RSA objects.
Looking back in OpenSSL 0.9.3a, that cached state was not protected by a
lock:
https://github.com/openssl/openssl/blob/OpenSSL_0_9_3a/crypto/rsa/rsa_eay.c#L137-L142
However, this was fixed in OpenSSL 0.9.7h:
6ec8e63af6
The other flags (DSA and RSA) have since fallen away, DSA with the
removal of DSA altogether (3994b1f9c2) and
RSA with 3a8d4a316e, "openssl 0.9.6a and
higher don't have the RSA locking bug [...] other algorithms still don't
do locking when performing precomputation [...]".
That seems to be referring to this OpenSSL change, which indeed fixed it
for RSA but not others:
bb617a9646
The 0.9.7h change above fixed it across the board, but there was never a
similar update to the workaround for DSA and DH. With such OpenSSL
versions long since out of support, the last remains of this workaround
can finally be removed.
When callback was NULL, bind9 would use BN_GENCB_set_old to set a NULL
callback because OpenSSL happened to allow a NULL "old" callback, but
not a NULL "new" callback. Instead, the way to turn off the callback is
to pass a NULL BN_GENCB itself.
Switch to doing that.
The BUFSIZ value varies between platforms, it could be 8K on Linux and
512 bytes on mingw. Make sure the buffers are always big enough for the
output data to prevent truncation of the output by appropriately
enlarging or sizing the buffers.
Remove "external" from the list of legal values for the -M command-line
option as it has not been allowed since the internal memory allocator
was removed by commit 55ace5d3aa.
Make the style of the relevant paragraph more in line with the next one
and split its contents up into an unordered list of options for improved
readability.
When a thread calls dns_dispatch_connect() on an unconnected TCP socket
it sets `tcpstate` from `DNS_DISPATCHSTATE_NONE` to `_CONNECTING`.
Previously, it then INSISTed that there were no pending connections
before calling isc_nm_tcpdnsconnect().
If a second thread called dns_dispatch_connect() during that window
of time, it could add a pending connection to the list, and trigger
an assertion failure.
This commit removes the INSIST since the condition is actually
harmless.
The statistics system test makes a query to foo.info to check for the
pending connections because the ans4 doesn't respond to the query.
This might or might not (depending on exact timing) increment the failed
TCP connection counter when the query is retried over TCP because ans4
doesn't listen on the TCP.
Wait for the 'connection refused' in the ns3 log file to be able to
count the exactly 1 failed TCP connection.
The STATID_CONNECT and STATID_CONNECTFAIL statistics were used
incorrectly. The STATID_CONNECT was incremented twice (once in
the *_connect_direct() and once in the callback) and STATID_CONNECTFAIL
would not be incremented at all if the failure happened in the callback.
Closes: #3452
On FreeBSD (and perhaps other *BSD) systems, the TCP connect() call (via
uv_tcp_connect()) can fail with transient UV_EADDRINUSE error. The UDP
code already handles this by trying three times (is a charm) before
giving up. Add a code for the TCP, TCPDNS and TLSDNS layers to also try
three times before giving up by calling uv_tcp_connect() from the
callback two more time on UV_EADDRINUSE error.
Additionally, stop the timer only if we succeed or on hard error via
isc__nm_failed_connect_cb().
