Aram Sargsyan [Thu, 17 Mar 2022 17:11:04 +0000 (17:11 +0000)]
Document catalog zones member zone reset by change of unique label
The DNS catalog zones draft version 5 document requires that catalog
zones consumers must reset the member zone's internal zone state when
its unique label changes (either within the same catalog zone or
during change of ownership performed using the "coo" property).
BIND already behaves like that, and, in fact, doesn't support keeping
the zone state during change of ownership even if the unique label
has been kept the same, because BIND always removes the member zone
and adds it back during unique label renaming or change of ownership.
Document the described behavior and add a log message to inform when
unique label renaming occurs.
Add a system test case with unique label renaming.
We check the `rdclass` to be of type IN in `dns_catz_update_process()`
function, and all the other static functions where similar checks exist
are called after (and in the result of) that function being called,
so they are effectively redundant.
Add new catz system test to check a version property with invalid type
There is already a check for the missing version property case
(catalog-bad1.example), and this new test should result in the same
outcome, but differs in a way that there exists a version record in the
zone, but it is of a wrong type (A instead of the expected TXT).
Add a system test to check catalog zones with non-IN rdata class fail
According to DNS catalog zones draft version 5 document, the CLASS field
of every RR in a catalog zone MUST be IN.
Add a new check in the catz system test to verify that a non-IN class
catalog zone (in this case CH) fails to load.
BIND does not support having a non-IN class RR in an IN class zone, or
non-IN class zone in an IN class view, so to verify that BIND respects
the mentioned restriction we must try to add a non-IN class catalog
zone and check that it didn't succeed.
The `named` configuration files had to be restructured to put all the
zones inside views, which also resulted in some corresponding changes
in the tests.sh script.
Add a new warning message when processing view configuration
When parsing the configuration file, log a warning message in
configure_view() function when encountering a `catalog-zones`
option in a view with non-IN rdata class.
Add DNS extended errors 3 (Stale Answer) and 19 (Stale NXDOMAIN Answer)
to responses. Add extra text with the reason why the stale answer was
returned.
To test, we need to change the configuration such that for the first
set of tests the stale-refresh-time window does not interfer with the
expected extended errors.
The shutdown test sends 'rdnc status' commands in parallel with
'rndc stop' A new rndc connection arriving will reference the ACL
environment to see whether the client is allowed to connect.
Commit c0995bc380 added a mutex lock to ns_interfacemgr_getaclenv(),
but if the new connection arrives while the interfaces are being
purged during shutdown, that lock is already being held. If the
the connection event slips in ahead of one of the netmgr's "stop
listening" events on a worker thread, a deadlock can occur.
The fix is not to hold the interfacemgr lock while shutting down
interfaces; only while actually traversing the interface list to
identify interfaces needing shutdown.
previously fctx_done() detached the fctx but did not clear the pointer
passed into it from the caller. in some conditions, when rctx_done()
was reached while waiting for a validator to complete, fctx_done()
could be called twice on the same fetch, causing a double detach.
fctx_done() now clears the fctx pointer, to reduce the chances of
such mistakes.
Handle ISC_R_SUCCESS on a deactivated response in udp_recv()
There is a possibility for `udp_recv()` to be called with `eresult`
being `ISC_R_SUCCESS`, but nevertheless with already deactivated `resp`,
which can happen when the request has been canceled in the meantime.
TLSDNS: call send callbacks after only the data was sent
This commit ensures that write callbacks are getting called only after
the data has been sent via the network.
Without this fix, a situation could appear when a write callback could
get called before the actual encrypted data would have been sent to
the network. Instead, it would get called right after it would have
been passed to the OpenSSL (i.e. encrypted).
Most likely, the issue does not reveal itself often because the
callback call was asynchronous, so in most cases it should have been
called after the data has been sent, but that was not guaranteed by
the code logic.
Also, this commit removes one memory allocation (netievent) from a hot
path, as there is no need to call this callback asynchronously
anymore.
There was a query_detach() call missing in dig, which could lead to
dig hanging on TLS context creation errors. This commit fixes.
The error was introduced because the Strict TLS implementation was
initially made over an older version of the code, where this extra
query_detach() call was not needed.
Petr Špaček [Wed, 27 Apr 2022 06:51:41 +0000 (08:51 +0200)]
Pin Sphinx related package versions to match ReadTheDocs and our CI
This seems to be most appropriate way to ensure consistency between
release tarballs and public presentation on ReadTheDocs.
Previous attempt with removing docutils constraint, which relied on pip
depedency solver to pick the same packages as in CI was flawed. RTD
installs a bit different set of packages so it was inherently
unreliable.
As a result RTD pulled in sphinx-rtd-theme==0.4.3 while CI
had 1.0.0, and this inconsistency caused Table of Contents in Release
Notes to render incorrectly. Previous solution was to downgrade
docutils to < 0.17, but I think we should rather pin exact versions.
There was an error in AX_PROG_CC_FOR_BUILD macro that cached literal
name of the cache variable `saved_ac_cv_c_compiler_gnu` instead of the
value of said variable breaking the consecutive runs of ./configure
script with caching enabled.
Petr Špaček [Tue, 26 Apr 2022 12:28:11 +0000 (14:28 +0200)]
Fix mismatch between docutils version in CI and ReadTheDocs
Currently our CI images we use to build docs (which subsequently get
into release tarballs) are using docutils 0.17.1, which is latest version
which fulfills Sphinx 4.5.0 requirement for docutils < 0.18.
The old requirement for docutils < 0.17 was causing discrepancy between
the way we build release artifacts and the docs on ReadTheDocs.org which
uses doc/arm/requirements.txt from our repo.
Remove the limit for RDT with hope that it will pull latest permissible
version of docutils.
lock find when unlinking adbname->finds in dns_adb_cancelfind()
In dns_adb_cancelfind(), we need to release the find lock and
then acquire the bucket and find locks in that order, for
consistency with locking hierarchy elsehwere. Previously we
were only acquiring the bucket lock.
Petr Špaček [Mon, 25 Apr 2022 14:28:02 +0000 (16:28 +0200)]
Split negative and positive dig/mdig/delv options to support Sphinx 1.4.9
Man pages for dig/mdig/delv used `.. option:: +[no]bla` to describe two
options at once, and very old Sphinx does not support that [] in option
names.
Solution is to split negative and positive options into `+bla, +nobla`
form. In the end it improves readability because it transforms hard to
read strings with double brackets from
`+[no]subnet=addr[/prefix-length]` to
`+subnet=addr[/prefix-length], +nosubnet`.
As a side-effect it also allows easier linking to dig/mdig/delv options
using their name directly instead of always overriding the link target
to `+[no]bla` form.
Transformation was done using regex:
s/:: +\[no\]\(.*\)/:: +\1, +no\1
... and manual review around occurences matching regex
+no.*=
Set IP(V6)_RECVERR on connect UDP sockets (via libuv)
The connect()ed UDP socket provides feedback on a variety of ICMP
errors (eg port unreachable) which bind can then use to decide what to
do with errors (report them to the client, try again with a different
nameserver etc). However, Linux's implementation does not report what
it considers "transient" conditions, which is defined as Destination
host Unreachable, Destination network unreachable, Source Route Failed
and Message Too Big.
Explicitly enable IP_RECVERR / IPV6_RECVERR (via libuv uv_udp_bind()
flag) to learn about ICMP destination network/host unreachable.
Abort when libuv at runtime mismatches libuv at compile time
When we compile with libuv that has some capabilities via flags passed
to f.e. uv_udp_listen() or uv_udp_bind(), the call with such flags would
fail with invalid arguments when older libuv version is linked at the
runtime that doesn't understand the flag that was available at the
compile time.
Enforce minimal libuv version when flags have been available at the
compile time, but are not available at the runtime. This check is less
strict than enforcing the runtime libuv version to be same or higher
than compile time libuv version.
The route socket and its storage was detached while still reading
The interfacemgr and the .route was being detached while the network
manager had pending read from the socket. Instead of detaching from the
socket, we need to cancel the read which in turn will detach the route
socket and the associated interfacemgr.
Petr Špaček [Mon, 25 Apr 2022 10:07:24 +0000 (12:07 +0200)]
Use unique program + option names for link anchors to support Sphinx 1.8.5
Sphinx "standard domain" provides directive types ".. program::" and
".. option::" to create link anchor for a program name + option combination.
These can be referenced using :ref:`program option` syntax.
The problem is that Sphinx 1.8.5 (e.g. in Ubuntu 18.04) generates
conflicting link targets if a page contains two option directives
starting with the same word, e.g.:
.. program:: dnssec-settime
.. option:: -P date
.. option:: -P ds date
The reason is that option directive consumes only first word as "option
name" (-P) and all the rest is considered "option argument" (date, ds
date). Newer versions of Sphinx (e.g. 4.5.0) handle this by creating
numbered link anchors, but older versions warn and BIND build system
turns the warning into a hard error.
To handle that we use method recommended by Sphinx maintainer:
https://github.com/sphinx-doc/sphinx/issues/10218#issuecomment-1059925508
As a bonus it provides more accurate link anchors for sub-options.
Alternatives considered:
- Replacing standard domain definition of .. option - causes more
problems, see BIND issue #3294.
- Removing hyperlinks for options - that would be a step back.
Tony Finch [Fri, 22 Apr 2022 13:35:36 +0000 (14:35 +0100)]
Move random number re-seeding out of the hot path
Instead of checking if we need to re-seed for every isc_random call,
seed the random number generator in the libisc global initializer
and the per-thread initializer.
Tony Finch [Thu, 14 Apr 2022 17:18:12 +0000 (18:18 +0100)]
Make isc_random_uniform() nearly divisionless
It used to require two 32-bit integer divisions to get a random number
less than some limit. Now we use Daniel Lemire's "nearly-divisionless"
algorithm for unbiased bounded random numbers, which requires one
64-bit integer multiply in the usual case, and one 32-bit integer
division in rare slow cases. Even the slow cases are faster than
before; there are also fewer branches.
I think this algorithm is exceptionally beautiful. It also has more
clever tricks than lines of code, so I have done my best to explain
how it works.
The dns__adb_attach() had an assertion failure that prevented to attach
to dns_adb if the dns_adb was shutting down. There was a race between
checking for .exiting in dns_adb_createfind and creating new_adbfind() -
other thread could have set the .exiting to true between the check.
Remove the assertion failure and allow attaching to dns_adb even while
shutting down. The process of dns_adb shutting down would be noticed
only a moments later when any other callback is called.
The rctx_chaseds() function calls dns_resolver_createfetch(), passing
fctx->task as the target task to run resume_dslookup() from. This
breaks task-based serialization of events as fctx->task is the task that
the dns_resolver_createfetch() caller wants to receive its fetch
completion event in; meanwhile, intermediate fetches started by the
resolver itself (e.g. related to QNAME minimization) must use
res->buckets[bucketnum].task instead. This discrepancy may cause
trouble if the resume_dslookup() callback happens to be run concurrently
with e.g. fctx_doshutdown().
Fix by passing the correct task to dns_resolver_createfetch() in
rctx_chaseds().
Michał Kępień [Fri, 22 Apr 2022 11:27:12 +0000 (13:27 +0200)]
Fix loading plugins using just their filenames
BIND 9 plugins are installed using Automake's pkglib_LTLIBRARIES stanza,
which causes the relevant shared objects to be placed in the
$(libdir)/@PACKAGE@/ directory, where @PACKAGE@ is expanded to the
lowercase form of the first argument passed to AC_INIT(), i.e. "bind".
Meanwhile, NAMED_PLUGINDIR - the preprocessor macro that the
ns_plugin_expandpath() function uses for determining the absolute path
to a plugin for which only a filename has been provided (rather than a
path) - is set to $(libdir)/named. This discrepancy breaks loading
plugins using just their filenames. Fix the issue (and also prevent it
from reoccurring) by setting NAMED_PLUGINDIR to $(pkglibdir).
Regenerate the man pages to match contents produced in a Sphinx
environment using the above package versions. This is necessary to
prevent the "docs" GitLab CI job from failing.
Michał Kępień [Fri, 22 Apr 2022 09:25:27 +0000 (11:25 +0200)]
Fix a PyLint 2.13.7 error
PyLint 2.13.7 reports the following error:
bin/tests/system/doth/conftest.py:34:28: E0601: Using variable 'stderr' before assignment (used-before-assignment)
The reason the current code has not caused problems before is that
invoking gnutls-cli with just the --logfile=/dev/null argument causes it
to always return with a non-zero exit code, either due to the option not
being supported or due to the hostname argument not being provided. In
other words, the 'except' branch has always been taken. PyLint is
obviously right on a syntactical level, though.
Instead of relying on a less than obvious code flow (where the 'except'
branch is always taken), rework the flagged code by employing
subprocess.run(..., check=False) instead of subprocess.check_output(),
making exception handling redundant.
While this issue was investigated, it was also noticed that
subprocess.check_output() was incorrectly used as a context manager:
Popen objects are context managers, but subprocess.check_output() and
subprocess.run() are not. Fix by dropping the relevant 'with'
statement.
Michał Kępień [Fri, 22 Apr 2022 09:25:27 +0000 (11:25 +0200)]
Fix "digdelv" system test requirements
Commit 3ec5d2d6ede18e502ef0ef4521fe917f6531f258 added a Python-based
name server (bin/tests/system/digdelv/ans8/ans.py) to the "digdelv"
system test, but did not update bin/tests/system/Makefile.am to ensure
Python is present in the test environment before the "digdelv" system
test is run. Update bin/tests/system/Makefile.am to enforce that
requirement.
Michał Kępień [Fri, 22 Apr 2022 09:25:27 +0000 (11:25 +0200)]
Require Python 3.6+ for running Python-based tests
configure.ac currently requires Python 3.4 for running Python-based
system tests. Meanwhile, there are some features in Python 3.6+ that we
would like to use for making our Python code cleaner (e.g. f-strings).
Update the minimum Python version required for running Python-based
system tests to 3.6, noting that:
- Python 3.4 has reached end-of-life on March 18th, 2019.
- Python 3.5 has reached end-of-life on September 5th, 2020.
Michał Kępień [Thu, 21 Apr 2022 12:19:39 +0000 (14:19 +0200)]
Prevent memory bloat caused by a jemalloc quirk
Since version 5.0.0, decay-based purging is the only available dirty
page cleanup mechanism in jemalloc. It relies on so-called tickers,
which are simple data structures used for ensuring that certain actions
are taken "once every N times". Ticker data (state) is stored in a
thread-specific data structure called tsd in jemalloc parlance. Ticks
are triggered when extents are allocated and deallocated. Once every
1000 ticks, jemalloc attempts to release some of the dirty pages hanging
around (if any). This allows memory use to be kept in check over time.
This dirty page cleanup mechanism has a quirk. If the first
allocator-related action for a given thread is a free(), a
minimally-initialized tsd is set up which does not include ticker data.
When that thread subsequently calls *alloc(), the tsd transitions to its
nominal state, but due to a certain flag being set during minimal tsd
initialization, ticker data remains unallocated. This prevents
decay-based dirty page purging from working, effectively enabling memory
exhaustion over time. [1]
The quirk described above has been addressed (by moving ticker state to
a different structure) in jemalloc's development branch [2], but not in
any numbered jemalloc version released to date (the latest one being
5.2.1 as of this writing).
Work around the problem by ensuring that every thread spawned by
isc_thread_create() starts with a malloc() call. Avoid immediately
calling free() for the dummy allocation to prevent an optimizing
compiler from stripping away the malloc() + free() pair altogether.
An alternative implementation of this workaround was considered that
used a pair of isc_mem_create() + isc_mem_destroy() calls instead of
malloc() + free(), enabling the change to be fully contained within
isc__trampoline_run() (i.e. to not touch struct isc__trampoline), as the
compiler is not allowed to strip away arbitrary function calls.
However, that solution was eventually dismissed as it triggered
ThreadSanitizer reports when tools like dig, nsupdate, or rndc exited
abruptly without waiting for all worker threads to finish their work.
Mark Andrews [Mon, 14 Feb 2022 05:13:46 +0000 (16:13 +1100)]
Tighten DBC restrictions on message sections
dns_message_findname and dns_message_sectiontotext incorrectly accepted
DNS_SECTION_ANY. If DNS_SECTION_ANY was passed the section array could
be incorrectly accessed at (-1).
dns_message_pseudosectiontotext and dns_message_pseudosectiontoyaml
incorrectly accepted DNS_PSEUDOSECTION_ANY. These functions are
designed to process a single section.
Tony Finch [Tue, 12 Apr 2022 10:49:21 +0000 (11:49 +0100)]
Avoid timeouts in the notify system test
There were two problems in the notify system test when it waited for
log messages to appear: the shellcheck refactoring introduced a call
to `wait_for_log` with a regex, but `wait_for_log` only supports fixed
strings, so it always ran for the full 45 second timeout; and the new
test to ensure that notify messages time out failed to reset the
nextpart pointer, so if the notify messages timed out before the test
ran, it would fail to see them.
This change adds a `wait_for_log_re` helper that matches a regex, and
uses it where appropriate in the notify system test, which stops the
test from waiting longer than necessary; and it resets the nextpart
pointer so that the notify timeout test works reliably.
Ondřej Surý [Sat, 26 Mar 2022 13:37:38 +0000 (14:37 +0100)]
Add detailed tracing when TASKMGR_TRACE is defined
When TASKMGR_TRACE=1 is defined, the task and event objects have
detailed tracing information about function, file, line, and
backtrace (to the extent tracked by gcc) where it was created.
At exit, when there are unfinished tasks, they will be printed along
with the detailed information.
Ondřej Surý [Sat, 26 Mar 2022 13:37:38 +0000 (14:37 +0100)]
Remove isc_task_sendto(anddetach) functions
The only place where isc_task_sendto() was used was in dns_resolver
unit, where the "sendto" part was actually no-op, because dns_resolver
uses bound tasks. Remove the isc_task_sendto() and
isc_task_sendtoanddetach() functions in favor of using bound tasks
create with isc_task_create_bound().
Additionally, cache the number of running netmgr threads (nworkers)
locally to reduce the number of function calls.
projects / thirdparty / bind9.git / log
