Replace the isc_mutex in the dns_adb unit with isc_rwlock for better
performance. Both ADB names and ADB entries hashtables and LRU are now
using isc_rwlock.
This changes the internal isc_rwlock implementation to:
Irina Calciu, Dave Dice, Yossi Lev, Victor Luchangco, Virendra
J. Marathe, and Nir Shavit. 2013. NUMA-aware reader-writer locks.
SIGPLAN Not. 48, 8 (August 2013), 157–166.
DOI:https://doi.org/10.1145/2517327.24425
(The full article available from:
http://mcg.cs.tau.ac.il/papers/ppopp2013-rwlocks.pdf)
The implementation is based on the The Writer-Preference Lock (C-RW-WP)
variant (see the 3.4 section of the paper for the rationale).
The implemented algorithm has been modified for simplicity and for usage
patterns in rbtdb.c.
The changes compared to the original algorithm:
* We haven't implemented the cohort locks because that would require a
knowledge of NUMA nodes, instead a simple atomic_bool is used as
synchronization point for writer lock.
* The per-thread reader counters are not being used - this would
require the internal thread id (isc_tid_v) to be always initialized,
even in the utilities; the change has a slight performance penalty,
so we might revisit this change in the future. However, this change
also saves a lot of memory, because cache-line aligned counters were
used, so on 32-core machine, the rwlock would be 4096+ bytes big.
* The readers use a writer_barrier that will raise after a while when
readers lock can't be acquired to prevent readers starvation.
* Separate ingress and egress readers counters queues to reduce both
inter and intra-thread contention.
The dns_badcache was pulling the <isc/atomic.h> header only indirectly
via <isc/rwlock.h>, add the direct include as the <isc/rwlock.h> no
longer pulls the header when pthread_rwlock is used.
I misunderstood the purpose of the `heap_index` rdataset header
member; I thought it identified which heap to use, and could therefore
be smaller, the same size as `locknum` indexes. But in fact it is a
position within a heap, so it needs to be able to count up to the
total number of rdatasets in the rbtdb.
So this changes `heap_index` from `uint16_t` back to `unsigned int`.
To avoid re-embiggening the rdatasetheader, shrink the `count` member
from `uint32` to `uint16`. The `count` is used to rotate RRsets in
`dns_rdataset_towiresorted()`, so 16 bits is more than large enough.
This change also means we no longer need to avoid colliding with
`DNS_RDATASET_COUNT_UNDEFINED` i.e. UINT32_MAX.
Closes#3862
A dns_rpz_unref_rpzs() call is missing when taking the 'goto unlock;'
path on shutdown, in order to compensate for the earlier
dns_rpz_ref_rpzs() call.
Move the dns_rpz_ref_rpzs() call after the shutdown check.
When there are multiple managed trust anchors we need to know the
name of the trust anchor that is failing. Extend the error message
to include the trust anchor name.
removed some functions that are no longer used and unlikely to
be resurrected, and also some that were only used to support Windows
and can now be replaced with generic versions.
Add magic value to the fctxcount, to check for completely invalid
counters, or counters that have been already destroyed.
Improve the locking around the counters, and because of that we can drop
the atomics and use simple integers - the counters were already locked
and the tiny bits that used the atomics were not worth the extra effort.
- removed documentation of -S option from named man page
- removed documentation of reserved-sockets from ARM
- simplified documentation of dnssec-secure-to-insecure - it
now just says it's obsolete rather than describing what it
doesn't do anymore
- marked three formerly obsolete options as ancient:
parent-registration-delay, reserved-sockets, and
suppress-initial-notify
isc_bind9 was a global bool used to indicate whether the library
was being used internally by BIND or by an external caller. external
use is no longer supported, but the variable was retained for use
by dyndb, which needed it only when being built without libtool.
building without libtool is *also* no longer supported, so the variable
can go away.
Instead of using an extra rarely-used paramater to dns_clientinfo_init()
to set ECS information for a client, this commit adds a function
dns_clientinfo_setecs() which can be called only when ECS is needed.
A recent refactoring in 7e4e125e5e
had introduced a logical error which could result in calling the
dns_resolver_createfetch() function with 'nameservers' pointer set
to NULL, but with 'domain' not set to NULL, which is not allowed
by the function.
Make sure 'domain' is set only when 'nsrdataset' is valid.
the optional 'port' option, when used with notify-source,
transfer-source, etc, is used to set up UDP dispatches with a
particular source port, but when the actual UDP connection was
established the port would be overridden with a random one. this
has been fixed.
(configuring source ports is deprecated in 9.20 and slated for
removal in 9.22, but should still work correctly until then.)
it was possible for a managed trust anchor needing to send a key
refresh query to be unable to do so because an authoritative zone
was not yet loaded. this has been corrected by delaying the
synchronization of managed-keys zones until after all zones are
loaded.
* rbt node chains were sized to allow for bitstring labels, so they
had 256 levels; but in the absence of bistrings, 128 is enough.
* dns_byaddr_createptrname() had a redundant options argument,
and a very outdated doc comment.
* A number of comments referred to bitstring labels in a way that is
no longer helpful. (A few informative comments remain.)
Commit 7695c36a5d added a new parameter,
'options', to the prototype of the 'allrdatasets' function pointer in
struct dns_dbmethods. Handle this new parameter accordingly in
rpsdb_allrdatasets().
the rate limter now uses loop callbacks rather than task events.
the API for isc_ratelimiter_enqueue() has been changed; we now pass
in a loop, a callback function and a callback argument, and
receive back a rate limiter event object (isc_rlevent_t). it
is no longer necessary for the caller to allocate the event.
the callback argument needs to include a pointer to the rlevent
object so that it can be freed using isc_rlevent_free(), or by
dequeueing.
The ADB hashmaps are stored in extra memory contexts, so the hash
tables are excluded from the overmem accounting. The new memory
context was unnamed, give it a proper name.
Same thing has happened with extra memory context used for named
global log context - give the extra memory context a proper name.
Set the DS state after issuing 'rndc dnssec -checkds'. If the DS
was published, it should go in RUMOURED state, regardless whether it
is already safe to do so according to the state machine.
Leaving it in HIDDEN (or if it was magically already in OMNIPRESENT or
UNRETENTIVE) would allow for easy shoot in the foot situations.
Similar, if the DS was withdrawn, the state should be set to
UNRETENTIVE. Leaving it in OMNIPRESENT (or RUMOURED/HIDDEN)
would also allow for easy shoot in the foot situations.
Add check for extracting the public 'n' component on OpenSSL 3.0
path. This is mandatory component, and it's presence is checked
already on the other code path.
Also document the reason why private key component getting errors
are ignored.
The total memory counter had again little or no meaning when we removed
the internal memory allocator. It was just a monotonic counter that
would count add the allocation sizes but never subtracted anything, so
it would be just a "big number".
The maxinuse memory counter indicated the highest amount of
memory allocated in the past. Checking and updating this high-
water mark value every time memory was allocated had an impact
on server performance, so it has been removed. Memory size can
be monitored more efficiently via an external tool logging RSS.
Both receive_secure_serial() and setnsec3param() run on the same zone
loop, therefore they are serialized. Remove the mechanism to enqueue
the nsec3param and secure serial updates in case one of them is
running (as they can not) and replace it with sanity check.
Previously, the zone loading and dumping was effectively serialized by
the dns_io_t mechanism. In theory, more IO operations could be run in
parallel, but the zone manager .iolimit was set to 1 and never increased
as dns_zonemgr_setiolimit() was never ever called.
As the dns_master asynchronous load and dump was already offloaded to
non-worker threads with isc_work mechanism, drop the whole dns_io_t
and just rely on the isc_work to do the load and dump scheduling.
Instead of dumping the signed zone contents node by node during the
signing, dump the entire zone at the end. This was already done for the
raw zone format, but it shows that the IO is better utilized when the
zone dump is done in one single write rather than in small chunks.
A side effect of dumping node by node was that all names were printed
relative to the zone origin rather than being grouped under different
$ORIGINs as would normally be the case when dumping a zone. Also, state
was not maintained from one node to the next regarding whether the CLASS
has already been printed, so it was always included with the first
record of each node.
Since dnssec-signzone uses the dns_master_style_explicittl text format
style, and is the only application that does so, we can revise that
style and add a new DNS_STYLEFLAG_CLASS_PERNAME flag to get the output
back to what it was before this change.
Add support for loading and validating the 'tls' parameter from
the forwarders' configuration.
This prepares ground for adding support to forward queries to
DoT-enabled upstream servers.
The write node lock needs to be held when setting node->wild in
add_wildcard_magic except when being called from loading_addrdataset
which is used to load the zone without locking during its initial
load.
Return 'isc_result_t' type value instead of 'bool' to indicate
the actual failure. Rename the function to something not suggesting
a boolean type result. Make changes in the places where the API
function is being used to check for the result code instead of
a boolean value.
It is allowed to point parental-agents to a resolver. Therefore, the
RD bit should be set on requests.
Upon receiving a DS response, ensure that the message has either the
AA or the RA bit set.
Detaching the views in the zone_shutdown() could lead to
lock-order-inversion between adb->namelocks[bucket], adb->lock,
view->lock and zone->lock. Detach the views outside of the section that
zone-locked.
The .view (and possibly .prev_view) would be kept attached to the
removed zone until the zone is fully removed from the memory in
zone_free(). If this process is delayed because server is busy
something else like doing constant `rndc reconfig`, it could take
seconds to detach the view, possibly keeping multiple dead views in the
memory. This could quickly lead to a massive memory bloat.
Release the views early in the zone_shutdown() call, and don't wait
until the zone is freed.
During XoT it is important to check for "dot" ALPN tag to be
negotiated (according to the RFC 9103). We were doing that, however, the
situation was not handled properly, leading to non-cancelled zone
transfers that would crash (abort()) BIND on shutdown.
In this particular case 'result' might equal 'ISC_R_SUCCESS'. When
this is the case, the part of the code supposed to handle failures
will not cancel the zone transfer.
This situation cannot happen when BIND is a secondary of other BIND
instance. Only primaries following the RFC not closely enough could
trigger such a behaviour.
Although 'dns_fetch_t' fetch can have two associated events, one for
each of 'DNS_EVENT_FETCHDONE' and 'DNS_EVENT_TRYSTALE' types, the
dns_resolver_cancelfetch() function is designed in a way that it
expects only one existing event, which it must cancel, and when it
happens so that 'stale-answer-client-timeout' is enabled and there
are two events, only one of them is canceled, and it results in an
assertion in dns_resolver_destroyfetch(), when it finds a dangling
event.
Change the logic of dns_resolver_cancelfetch() function so that it
cancels both the events (if they exist), and in the right order.
As shown in the previous commit, using sizeof(type_t) is a little
bit more error-prone when copy-pasting code, so extracting the
size information from the pointer which is being dealt with seems
like a better alternative.
Free 'sizeof(dns_forwarder_t)' bytes of memory instead of
'sizeof(dns_sockaddr_t)' bytes, because `fwd` is a pointer
to a 'dns_forwarder_t' type structure.
Ondřej Surý
Ondřej SurýandOndřej Surý
Tony FinchandOndřej Surý
Aram Sargsyan
Mark Andrews
Evan Hunt
Michał Kępień
Matthijs Mekking
Timo Teräs
Artem Boldariev
Petr Menšík