The find invocation used by the bin/tests/system/get_ports.sh script
("find . -maxdepth 1 -mindepth 1 -type d") assumes the list of
directories in bin/tests/system/ remains unchanged throughout the run
time of a single system test suite. With pytest in use and the
conftest.py file now present in bin/tests/system/, that assumption is no
longer true as a __pycache__ directory may be created when the first
pytest-based test is started. Since the list of names returned by the
above find invocation serves as a fixed-size array of "port range
slots", any changes to that list during a system test suite run may lead
to port assignment collisions [1].
Fix by making the find invocation more nuanced, so that it only returns
names of directories containing test code. Squash a grep / cut pipeline
into a single awk invocation.
[1] see commit 31e5ca4bd9
Most Python-based system tests need to know which ports were assigned to
a given test by bin/tests/system/get_ports.sh. This is currently
handled by inspecting the values of various environment variables (set
by bin/tests/system/run.sh) and passing the port numbers to Python
scripts via pytest fixtures. However, this glue code has so far been
copy-pasted into each system test using it, rather than reused.
Since pytest also looks for conftest.py files in parent directories,
move commonly used fixtures to bin/tests/system/conftest.py. Set the
scope of all the moved fixtures to "session" as their return values are
only based on environment variables, so there is no point in recreating
them for every test requesting them. Adjust test code accordingly.
Building BIND 9 with older version of BIND 9 installed would result in
build failure. Fix the last two remaining cases where <prog>_CFLAGS was
being used leading to wrong order of the build flags on the command line.
Both utilities were included as one man page, but this caused a problem:
Sphinx directive .. include was used twice on the same file, which
prevented us from using labels (or anything with unique identifier) in
the man pages. This effectivelly prevented linking to them.
Splitting man pages allows us to solve the linking problems and also
clearly make text easier to follow because it does not mention two tools
at the same time.
This change causes duplication of text, but given the frequecy of changes
to these tools I think it is acceptable. I've considered deduplication
using smaller .rst snippets which get included into both man pages,
but it would require more sed scripting to handle defaults etc. and
I think it would be way too complex solution for this problem.
Related: #2799
Both utilities were included as one man page, but this caused a problem:
Sphinx directive .. include was used twice on the same file, which
prevented us from using labels (or anything with unique identifier) in
the man pages. This effectivelly prevented linking to them.
Splitting man pages allows us to solve the linking problems and also
clearly make text easier to follow because it does not mention two tools
at the same time.
This change causes duplication of text, but given the frequecy of changes
to these tools I think it is acceptable.
Related: #2799
When dig was built without IDN support, it reported an error if the
+noidnin and/or +noidnout options were used. This means the options
were not useful for a script that wants consistent lack of IDN
translation regardless of how BIND is built.
Make dig complain about lack of built-in IDN support only when the
user asks for IDN translation.
Closes#3188
The C17 standard deprecated ATOMIC_VAR_INIT() macro (see [1]). Follow
the suite and remove the ATOMIC_VAR_INIT() usage in favor of simple
assignment of the value as this is what all supported stdatomic.h
implementations do anyway:
* MacOSX.plaform: #define ATOMIC_VAR_INIT(__v) {__v}
* Gcc stdatomic.h: #define ATOMIC_VAR_INIT(VALUE) (VALUE)
1. http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p1138r0.pdf
Previously, the function(s) in the commit subject could fail for various
reasons - mostly allocation failures, or other functions returning
different return code than ISC_R_SUCCESS. Now, the aforementioned
function(s) cannot ever fail and they would always return ISC_R_SUCCESS.
Change the function(s) to return void and remove the extra checks in
the code that uses them.
Previously, the function(s) in the commit subject could fail for various
reasons - mostly allocation failures, or other functions returning
different return code than ISC_R_SUCCESS. Now, the aforementioned
function(s) cannot ever fail and they would always return ISC_R_SUCCESS.
Change the function(s) to return void and remove the extra checks in
the code that uses them.
bad-ksk-without-zsk.conf only has a ksk defined without a
matching zsk for the same algorithm.
bad-zsk-without-ksk.conf only has a zsk defined without a
matching ksk for the same algorithm.
bad-unpaired-keys.conf has two keys of different algorithms
one ksk only and the other zsk only
The current implementation of isc_queue uses Michael-Scott lock-free
queue that in turn uses hazard pointers. It was discovered that the way
we use the isc_queue, such complicated mechanism isn't really needed,
because most of the time, we either execute the work directly when on
nmthread (in case of UDP) or schedule the work from the matching
nmthreads.
Replace the current implementation of the isc_queue with a simple locked
ISC_LIST. There's a slight improvement - since copying the whole list
is very lightweight - we move the queue into a new list before we start
the processing and locking just for moving the queue and not for every
single item on the list.
NOTE: There's a room for future improvements - since we don't guarantee
the order in which the netievents are processed, we could have two lists
- one unlocked that would be used when scheduling the work from the
matching thread and one locked that would be used from non-matching
thread.
If the dns_request send callback is delayed, the dst API would get
deinitialized and then the detach from the tsig key would cause an
assertion failure.
Shutdown the isc_managers early, and only then dereference the dst
objects when cleaning up the resources used by nsupdate.
For the reference, the _cancel_lookup() function iterates through
the lookup's queries list and detaches them. In the ideal scenario,
that should be the last reference and the query will be destroyed
after that, but it is also possible that we are still expecting a
callback, which also holds a reference (for example, _cancel_lookup()
could have been called from recv_done(), when send_done() was still
not executed).
The start_udp() and start_tcp() functions are currently designed in
slightly different ways: start_udp() creates a new query attachment
`connectquery`, to be called in the callback function, while
start_tcp() does not, which is a bug, but is hidden by the fact
that when the query is being erroneously destroyed prematurely (before
_cancel_lookup() is called) in the result of that, it also gets
de-listed from the lookup's queries' list, so _cancel_lookup() doesn't
even try to detach it.
For better understanding, here's an illustration of the query's
references count changes, and from where it was changed:
UDP
---
1. _new_query() -> refcount = 1 (initial)
2. start_udp() -> refcount = 2 (lookup->current_query)
3. start_udp() -> refcount = 3 (connectquery)
4. udp_ready() -> refcount = 4 (readquery)
5. udp_ready() -> refcount = 5 (sendquery)
6. udp_ready() -> refcount = 4 (lookup->current_query)
7. udp_ready() -> refcount = 3 (connectquery)
8. send_done() -> refcount = 2 (sendquery)
9. recv_done() -> refcount = 1 (readquery)
10. _cancel_lookup() -> refcount = 0 (initial)
11. the query gets destroyed and removed from `lookup->q`
TCP, fortunate scenario
-----------------------
1. _new_query() -> refcount = 1 (initial)
2. start_tcp() -> refcount = 2 (lookup->current_query)
3. launch_next_query() -> refcount = 3 (readquery)
4. launch_next_query() -> refcount = 4 (sendquery)
5. tcp_connected() -> refcount = 3 (lookup->current_query)
6. tcp_connected() -> refcount = 2 (bug, there was no connectquery)
7. send_done() -> refcount = 1 (sendquery)
8. recv_done() -> refcount = 0 (readquery)
9. the query gets prematurely destroyed and removed from `lookup->q`
10. _cancel_lookup() -> the query is not in `lookup->q`
TCP, unfortunate scenario, revealing the bug
--------------------------------------------
1. _new_query() -> refcount = 1 (initial)
2. start_tcp() -> refcount = 2 (lookup->current_query)
3. launch_next_query() -> refcount = 3 (readquery)
4. launch_next_query() -> refcount = 4 (sendquery)
5. tcp_connected() -> refcount = 3 (lookup->current_query)
6. tcp_connected() -> refcount = 2 (bug, there was no connectquery)
7. recv_done() -> refcount = 1 (readquery)
8. _cancel_lookup() -> refcount = 0 (the query was in `lookup->q`)
9. we hit an assertion here when trying to destroy the query, because
sendhandle is not detached (which is done by send_done()).
10. send_done() -> this never happens
This commit does the following:
1. Add a `connectquery` attachment in start_tcp(), like done in
start_udp().
2. Add missing _cancel_lookup() calls for error scenarios, which
were possibly missing because before fixing the bug, calling
_cancel_lookup() and then calling query_detach() would cause
an assertion.
3. Log a debug message and call isc_nm_cancelread(query->readhandle)
for every query in the lookup from inside the _cancel_lookup()
function, like it is done in _cancel_all().
4. Add a `canceled` property for the query which becomes `true` when
the lookup (and subsequently, its queries) are canceled.
5. Use the `canceled` property in the network manager callbacks to
know that the query was canceled, and act like `eresult` was equal
to `ISC_R_CANCELED`.
There was a missing UNLOCK_LOOKUP in the recv_done() callback when
the operation had been canceled. That omission could result in a
deadlock situation.
the "zone" clause can be documented using, for instance,
`cfg_test --zonegrammar primary", which prints only
options that are valid in primary zones. this was not
the method being used when generating the named.conf
man page; instead, "zone" was documented with all possible
options, and no zone types at all.
this commit removes "zone" from the generic documentation
and adds include statements in named.conf.rst so that
correct zone grammars will be included in the man page.
"masters" and "default-masters" are now flagged so they will
not be included in the named.conf man page, despite being
accepted as valid options by the parser for backward
compatibiility.
The keep-response-order option has been obsoleted, and in this commit,
remove the keep-response-order ACL map rendering the option no-op, the
call the isc_nm_sequential() and the now unused isc_nm_sequential()
function itself.
The keep-response-order option has been introduced when TCP pipelining
has been introduced to BIND 9 as a failsafe for possibly non-compliant
clients.
Declare the keep-response-order obsolete as all DNS clients should
either support out-of-order processing or don't send more DNS queries
until the DNS response for the previous one has been received.
Extend the timeouts system test to ensure that the maximum outgoing
transfer time (max-transfer-time-out) and maximum outgoing transfer idle
time (max-transfer-idle-out) works as expected. This is done by
lowering the limits to 5/1 minutes and testing that the connection has
been dropped while sleeping between the individual XFR messages.
bin/tests/system/makejournal needs to ignore DNS_R_SEENINCLUDE
when calling dns_db_load(), otherwise it cannot generate a journal
for a zone file with a $INCLUDE statement.
Extend the timeouts system test that bursts the queries for large TXT
record and never read any responses back filling up the server TCP write
buffer. The test should work with the default wmem_max value on
Linux (208k).
There was a bug in the checking of the "blackhole" ACL in
dns_request_create*(), causing an address to be treated as included
in the ACL if it was explicitly *excluded*. Thus, leaving "blackhole"
unset had no effect, but setting it to "none" would cause any
destination addresses to be rejected for dns_request purposes. This
would cause zone transfer requests and SOA queries to fail, among
other things.
The bug has been fixed, and "blackhole { none; };" was added to the
xfer system test as a regression test.
Replace the hard-coded paths for various BIND 9 files (configuration,
pid, etc.) in the man pages and ARM with compile-time values using the
sphinx-build replace system.
This is more complicated, because the restructured text specification
doesn't allow |substitions| inside ``code-blocks``, so for each specific
file we had to create own substition which is sub-optimal, but it is
only way how to do this without adding Sphinx extension.
The isc_thread_setaffinity call was removed in !5265 and we are not
going to restore it because it was proven that the performance is better
without it. Additionally, remove the already disabled cpu system test.
The isc_thread_setconcurrency function is unused and also calling
pthread_setconcurrency() on Linux has no meaning, formerly it was
added because of Solaris in 2001 and it was removed when taskmgr was
refactored to run on top of netmgr in !4918.
When there are more than one tokens initialized in SoftHSMv2,
care must be taken to correctly identify them.
Use a SoftHSMv2 token label which will uniquely identify the
token used for this test.
Use the "--token-label" parameter for the `pkcs11-tool` program
to make sure that it finds and uses the correct token.
The 'id' variable is either keyfromlabel-ksk or keyfromlabel-zsk and is
set in the 'keygen' and 'keyfromlabel' functions. It should not be used
outside these functions.
This test was originally in the pkcs11 system test. While this crash
happened in the native pkcs11 of BIND 9, and that code has been
removed in 9.17, there is no need for this test. Nevertheless, it
doesn't hurt having the test case persist.
Add a system test for engine_pkcs11 interactions that replaces the
tests that are done in the native PKCS#11 system test.
The native PKCS#11 code was removed in 9.17 but without copying the
pkcs11 system test.
Apparently we forgot about DLZ when updating DNS_CLIENTINFO_VERSION
constant for ECS, which is at value "3" since ECS was introduced.
The code in example drivers and tests now hardcodes version numbers
2 (without ECS) and 3 (with ECS) depending on what a given code path
requires.
System test used to have sequential system tests, which can't run in
parallel with the rest of system tests. As there are no such tests
anymore the underlying infrastructure can be dropped.
"parallel.sh" script was used on Windows to run system tests in
parallel. Since Windows support was removed from BIND 9, the script is
not needed anymore.
Michał Kępień
Ondřej Surý
Tony FinchandPetr Špaček
Tony Finch
Mark Andrews
Aram SargsyanandEvan Hunt
Matthijs Mekking
Michal Nowak