Due to the recent improvements to the TCP processing, much higher loads
can be handled by BIND9 without causing client timeouts. The updated
parameters give us useful data for both cold and hot cache testing.
Backport of MR !9494
Merge branch 'backport-nicki/increase-tcp-dot-shotgun-load-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9548
Due to the recent improvements to the TCP processing, much higher loads
can be handled by BIND9 without causing client timeouts. The updated
parameters give us useful data for both cold and hot cache testing.
(cherry picked from commit e89afd3349)
The ans4 server wasn't shutting down cleanly sometimes. Check that read returns the
expected value for the message length or exit read loop.
Closes#4301
Backport of MR !9537
Merge branch 'backport-4301-check-read-value-in-ans4-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9545
The ans4 server wasn't shutting down cleanly sometimes. Check that
read returns the expected value for the message length or exit read
loop.
(cherry picked from commit 21baad0a8f)
The shutdown system test sends queries when named is shutting down, not
in an attempt to get answers but to destabilize the server into a crash.
With isctest.query.udp() defaulting to try up to ten times with a
ten-second timeout to get a response we don't care about from a likely
terminated server, we make the test run much longer than needed because
of retries and long timeouts.
Also, see isc-projects/bind9#4943.
Backport of MR !9507
Merge branch 'backport-mnowak/shutdown-downgrade-timeout-and-attempts-arguments-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9541
The shutdown system test sends queries when named is shutting down, not
in an attempt to get answers but to destabilize the server into a crash.
With isctest.query.udp() defaulting to try up to ten times with a
ten-second timeout to get a response we don't care about from a likely
terminated server, we make the test run much longer than needed because
of retries and long timeouts.
(cherry picked from commit 463ab2f3f5)
Incoming transfers that took longer than 30 seconds would stop reading from the TCP stream and the incoming transfer would be indefinitely stuck causing BIND 9 to hang during shutdown.
This has been fixed and the `max-transfer-time-in` and `max-transfer-idle-in` timeouts are now honoured.
Closes#4949
Backport of MR !9531
Merge branch 'backport-4949-fix-ignored-and-invalid-dispatch-timeout-in-dns_xfrin-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9536
The dns_dispatch_add() call in the dns_xfrin unit had hardcoded 30
second limit. This meant that any incoming transfer would be stopped in
it didn't finish within 30 seconds limit. Additionally, dns_xfrin
callback was ignoring the return value from dns_dispatch_getnext() when
restarting the reading from the TCP stream; this could cause transfers
to get stuck waiting for a callback that would never come due to the
dns_dispatch having already been shut down.
Call the dns_dispatch_add() without a timeout and properly handle the
result code from the dns_dispatch_getnext().
(cherry picked from commit 96ef98558c)
The dns_dispatch_add() has timeout parameter that could not be 0 (for
not timeout). Modify the dns_dispatch implementation to accept a zero
timeout for cases where the timeouts are undesirable because they are
managed externally.
(cherry picked from commit 0f810b3144)
The issue of long-running operations potentially blocking query resolution has been fixed. Revert this temporary workaround and restore the number of threadpool threads.
Related #4898
Backport of MR !9530
Merge branch 'backport-4898-remove-workaround-and-note-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9532
when the QP cache was adapted from the RBTDB, some variable names weren't changed, and still incorrectly reference the RBT.
Backport of MR !9515
Merge branch 'backport-each-qpcache-nits-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9525
when the QP cache was adapted from the RBT database, some names
weren't changed. this could be confusing, so let's change them now.
also, we no longer need to include rbt.h.
(cherry picked from commit 5a444838db)
Some 'notify' and 'xfer-in' related log messages were logged at the
'general' category instead of their own category. This has been fixed.
Closes#2730
Backport of MR !9451
Merge branch 'backport-2730-logging-category-for-notify-and-xfer-related-messages-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9514
Some notify/xfer related log messages are logged at the general
category. Set a more suitable caterogry for those messages.
(cherry picked from commit 7c45caa8a5)
If the operating system UDP queue gets full and the outgoing UDP sending
starts to be delayed, BIND 9 could exhibit memory spikes as it tries to
enqueue all the outgoing UDP messages. Try a bit harder to deliver the
outgoing UDP messages synchronously and if that fails, drop the outgoing
DNS message that would get queued up and then timeout on the client side.
Closes#4930
Backport of MR !9506
Merge branch 'backport-4930-limit-the-UDP-send-queue-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9511
If the operating system UDP queue gets full and the outgoing UDP sending
starts to be delayed, BIND 9 could exhibit memory spikes as it tries to
enqueue all the outgoing UDP messages. As those are not going to be
delivered anyway (as we argued when we stopped enlarging the operating
system send and receive buffers), try to send the UDP messages directly
using `uv_udp_try_send()` and if that fails, drop the outgoing UDP
message.
(cherry picked from commit b576c4c977)
We currently set SO_INCOMING_CPU incorrectly, and testing by Ondrej
shows that fixing the issue by setting affinities is worse than letting
the kernel schedule threads without constraints. So we should not set
SO_INCOMING_CPU anymore.
Closes#4936
Backport of MR !9497
Merge branch 'backport-4936-remove-so-incoming-cpu-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9504
We currently set SO_INCOMING_CPU incorrectly, and testing by Ondrej
shows that fixing the issue and setting affinities is worse than letting
the kernel schedule threads without constraints. So we should not set
SO_INCOMING_CPU anymore.
(cherry picked from commit 8b8149cdd2)
When resolving a zone with a 'forward only' policy, and
finding out that all the forwarders are marked as "bad",
the 'ServerQuota' counter of the statistics channel was
incorrectly increased. This has been fixed.
Closes#1793
Backport of MR !9493
Merge branch 'backport-1793-serverquota-counter-bug-with-forward-only-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9502
Check that the fix in the previous commit works and that the
'ServerQuota' counter in the statistics channel is still unset
after a SERVFAIL result in a 'forward only' zone.
(cherry picked from commit 81b3c5d908)
The 'all_spilled' local variable in resolver.c:fctx_getaddresses()
is 'true' by default, and only becomes false when there is at least
one successfully found NS address. However, when a 'forward only;'
configuration is used, the code jumps over the part where it looks
for NS addresses and doesn't reset the 'all_spilled' to false, which
results in incorretly increased 'serverquota' statistics variable,
and also in invalid return error code from the function. The result
code error didn't make any differences, because all codes other than
'ISC_R_SUCCESS' or 'DNS_R_WAIT' were treated in the same way, and
the result code was never logged anywhere.
Set the default value of 'all_spilled' to 'false', and only make it
'true' before actually starting to look up NS addresses.
(cherry picked from commit e430ce7039)
Silence Coverity CID 468757 and 468767 (DATA RACE read not locked) by converting dnssec-signzone to use atomics for statistics counters rather than using a lock.
Closes#4939
Backport of MR !9496
Merge branch 'backport-4939-remove-stats-lock-from-dnssec-signzone-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9500
Silence Coverity CID 468757 and 468767 (DATA RACE read not locked)
by converting dnssec-signzone to use atomics for statistics counters
rather than using a lock. This should be marginally faster than
using the lock as well when statistics are requested.
(cherry picked from commit 473cbd4e87)
As part of the KeyTrap \[CVE-2023-50387\] mitigation, the DNSSEC CPU-intensive operations were offloaded to a separate threadpool that we use to run other tasks that could affect the networking latency.
If that threadpool is running some long-running tasks like RPZ, catalog zone processing, or zone file operations, it would delay DNSSEC validations to a point where the resolving signed DNS records would fail.
Split the CPU-intensive and long-running tasks into separate threadpools in a way that the long-running tasks don't block the CPU-intensive operations.
Closes#4898
Backport of MR !9473
Merge branch 'backport-4898-move-offloaded-DNSSEC-to-own-threads-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9495
Currently, the isc_work API is overloaded. It runs both the
CPU-intensive operations like DNSSEC validations and long-term tasks
like RPZ processing, CATZ processing, zone file loading/dumping and few
others.
Under specific circumstances, when many large zones are being loaded, or
RPZ zones processed, this stops the CPU-intensive tasks and the DNSSEC
validation is practically stopped until the long-running tasks are
finished.
As this is undesireable, this commit moves the CPU-intensive operations
from the isc_work API to the isc_helper API that only runs fast memory
cleanups now.
(cherry picked from commit 8a96a3af6a)
Add an extra thread that can be used to offload operations that would
affect latency, but are not long-running tasks; those are handled by
isc_work API.
Each isc_loop now has matching isc_helper thread that also built on top
of uv_loop. In fact, it matches most of the isc_loop functionality, but
only the `isc_helper_run()` asynchronous call is exposed.
(cherry picked from commit 6370e9b311)
Closes#4634
Backport of MR !9150
Merge branch 'backport-4634-drop-dns.resolver-module-from-system-tests-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9491
When verifying a message in an offloaded thread there is a race with
the worker thread which writes to the same buffer. Clone the message
buffer before offloading.
Closes#4929
Backport of MR !9481
Merge branch 'backport-4929-data-race-in-dns_dnssec_verifymessage-memmove-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9490
When verifying a message in an offloaded thread there is a race with
the worker thread which writes to the same buffer. Clone the message
buffer before offloading.
(cherry picked from commit 35ef25e5ea)
When the libxml2 and libjson-c libraries are not supported, the statistics channel can't return anything useful, so it is now disabled. Use of `statistics-channel` in `named.conf` is a fatal error.
Closes#4895
Backport of MR !9423
Merge branch 'backport-4895-link-style-sheet-to-libxml2-support-9.20' into 'bind-9.20'
See merge request isc-projects/bind9!9486
If neither libxml2 nor libjson_c are available have named-checkconf
fail if a statistics-channels block is specified.
(cherry picked from commit b9246418e8)
Nicki Křížek
Mark Andrews
Michal Nowak
Ondřej Surý
Evan Hunt
Arаm Sаrgsyаn
Alessio Podda