If we did not attempt a fetch due to fetch-limits, we should not start
the stale-refresh-time window.
Introduce a new flag DNS_DBFIND_STALESTART to differentiate between
a resolver failure and unexpected error. If we are resuming, this
indicates a resolver failure, then start the stale-refresh-time window,
otherwise don't start the stale-refresh-time window, but still fall
back to using stale data.
(This commit also wraps some docstrings to 80 characters width)
First of all, there was a flaw in the code related to the
'stale-refresh-time' option. If stale answers are enabled, and we
returned stale data, then it was assumed that it was because we were
in the 'stale-refresh-time' window. But now we could also have returned
stale data because of a 'stale-answer-client-timeout'. To fix this,
introduce a rdataset attribute DNS_RDATASETATTR_STALE_WINDOW to
indicate whether the stale cache entry was returned because the
'stale-refresh-time' window is active.
Second, remove the special case handling when the result is
DNS_R_NCACHENXRRSET. This can be done more generic in the code block
when dealing with stale data.
Putting all stale case handling in the code block when dealing with
stale data makes the code more easy to follow.
Update documentation to be more verbose and to match then new code
flow.
The general logic behind the addition of this new feature works as
folows:
When a client query arrives, the basic path (query.c / ns_query_recurse)
was to create a fetch, waiting for completion in fetch_callback.
With the introduction of stale-answer-client-timeout, a new event of
type DNS_EVENT_TRYSTALE may invoke fetch_callback, whenever stale
answers are enabled and the fetch took longer than
stale-answer-client-timeout to complete.
When an event of type DNS_EVENT_TRYSTALE triggers fetch_callback, we
must ensure that the folowing happens:
1. Setup a new query context with the sole purpose of looking up for
stale RRset only data, for that matters a new flag was added
'DNS_DBFIND_STALEONLY' used in database lookups.
. If a stale RRset is found, mark the original client query as
answered (with a new query attribute named NS_QUERYATTR_ANSWERED),
so when the fetch completion event is received later, we avoid
answering the client twice.
. If a stale RRset is not found, cleanup and wait for the normal
fetch completion event.
2. In ns_query_done, we must change this part:
/*
* If we're recursing then just return; the query will
* resume when recursion ends.
*/
if (RECURSING(qctx->client)) {
return (qctx->result);
}
To this:
if (RECURSING(qctx->client) && !QUERY_STALEONLY(qctx->client)) {
return (qctx->result);
}
Otherwise we would not proceed to answer the client if it happened
that a stale answer was found when looking up for stale only data.
When an event of type DNS_EVENT_FETCHDONE triggers fetch_callback, we
proceed as before, resuming query, updating stats, etc, but a few
exceptions had to be added, most important of which are two:
1. Before answering the client (ns_client_send), check if the query
wasn't already answered before.
2. Before detaching a client, e.g.
isc_nmhandle_detach(&client->reqhandle), ensure that this is the
fetch completion event, and not the one triggered due to
stale-answer-client-timeout, so a correct call would be:
if (!QUERY_STALEONLY(client)) {
isc_nmhandle_detach(&client->reqhandle);
}
Other than these notes, comments were added in code in attempt to make
these updates easier to follow.
It is possible to have two threads destroying an rbtdb at the same
time when detachnode() executes and removes the last reference to
a node between exiting being set to true for the node and testing
if the references are zero in maybe_free_rbtdb(). Move NODE_UNLOCK()
to after checking if references is zero to prevent detachnode()
changing the reference count too early.
Before this update, BIND would attempt to do a full recursive resolution
process for each query received if the requested rrset had its ttl
expired. If the resolution fails for any reason, only then BIND would
check for stale rrset in cache (if 'stale-cache-enable' and
'stale-answer-enable' is on).
The problem with this approach is that if an authoritative server is
unreachable or is failing to respond, it is very unlikely that the
problem will be fixed in the next seconds.
A better approach to improve performance in those cases, is to mark the
moment in which a resolution failed, and if new queries arrive for that
same rrset, try to respond directly from the stale cache, and do that
for a window of time configured via 'stale-refresh-time'.
Only when this interval expires we then try to do a normal refresh of
the rrset.
The logic behind this commit is as following:
- In query.c / query_gotanswer(), if the test of 'result' variable falls
to the default case, an error is assumed to have happened, and a call
to 'query_usestale()' is made to check if serving of stale rrset is
enabled in configuration.
- If serving of stale answers is enabled, a flag will be turned on in
the query context to look for stale records:
query.c:6839
qctx->client->query.dboptions |= DNS_DBFIND_STALEOK;
- A call to query_lookup() will be made again, inside it a call to
'dns_db_findext()' is made, which in turn will invoke rbdb.c /
cache_find().
- In rbtdb.c / cache_find() the important bits of this change is the
call to 'check_stale_header()', which is a function that yields true
if we should skip the stale entry, or false if we should consider it.
- In check_stale_header() we now check if the DNS_DBFIND_STALEOK option
is set, if that is the case we know that this new search for stale
records was made due to a failure in a normal resolution, so we keep
track of the time in which the failured occured in rbtdb.c:4559:
header->last_refresh_fail_ts = search->now;
- In check_stale_header(), if DNS_DBFIND_STALEOK is not set, then we
know this is a normal lookup, if the record is stale and the query
time is between last failure time + stale-refresh-time window, then
we return false so cache_find() knows it can consider this stale
rrset entry to return as a response.
The last additions are two new methods to the database interface:
- setservestale_refresh
- getservestale_refresh
Those were added so rbtdb can be aware of the value set in configuration
option, since in that level we have no access to the view object.
After backporting #1870 to 9.11-S I saw that the condition check there
is different than in the main branch. In 9.11-S "stale" can mean
stale and serve-stale, or not active (awaiting cleanup). In 9.16 and
later versions, "stale" is stale and serve-stale, and "ancient" means
not active (awaiting cleanup). An "ancient" RRset is one that is not
active (TTL expired) and is not eligble for serve-stale.
Update the condition for rndc dumpdb -expired to closer match what is
in 9.11-S.
By changing the check in 'rdatasetiter_first' and 'rdatasetiter_next'
from "now > header->rdh_ttl" to "now - RBDTB_VIRTUAL > header->rdh_ttl"
we include expired rdataset entries so that they can be used for
"rndc dumpdb -expired".
The rbtdb version glue_table has been refactored similarly to rbt.c hash
table, so it does use 32-bit hash function return values and apply
Fibonacci Hashing to lookup the index to the hash table instead of
modulo. For more details, see the lib/dns/rbt.c commit log.
When a received RRSet has TTL 0, they would be preserved for
serve-stale (default `max-stale-cache` is 12 hours) rather than expiring
them quickly from the cache database.
This commit makes sure the RRSet didn't have TTL 0 before marking the
entry in the database as "stale".
Created isc_refcount_decrement_expect macro to test conditionally
the return value to ensure it is in expected range. Converted
unchecked isc_refcount_decrement to use isc_refcount_decrement_expect.
Converted INSIST(isc_refcount_decrement()...) to isc_refcount_decrement_expect.
There were several problems with rbt hashtable implementation:
1. Our internal hashing function returns uint64_t value, but it was
silently truncated to unsigned int in dns_name_hash() and
dns_name_fullhash() functions. As the SipHash 2-4 higher bits are
more random, we need to use the upper half of the return value.
2. The hashtable implementation in rbt.c was using modulo to pick the
slot number for the hash table. This has several problems because
modulo is: a) slow, b) oblivious to patterns in the input data. This
could lead to very uneven distribution of the hashed data in the
hashtable. Combined with the single-linked lists we use, it could
really hog-down the lookup and removal of the nodes from the rbt
tree[a]. The Fibonacci Hashing is much better fit for the hashtable
function here. For longer description, read "Fibonacci Hashing: The
Optimization that the World Forgot"[b] or just look at the Linux
kernel. Also this will make Diego very happy :).
3. The hashtable would rehash every time the number of nodes in the rbt
tree would exceed 3 * (hashtable size). The overcommit will make the
uneven distribution in the hashtable even worse, but the main problem
lies in the rehashing - every time the database grows beyond the
limit, each subsequent rehashing will be much slower. The mitigation
here is letting the rbt know how big the cache can grown and
pre-allocate the hashtable to be big enough to actually never need to
rehash. This will consume more memory at the start, but since the
size of the hashtable is capped to `1 << 32` (e.g. 4 mio entries), it
will only consume maximum of 32GB of memory for hashtable in the
worst case (and max-cache-size would need to be set to more than
4TB). Calling the dns_db_adjusthashsize() will also cap the maximum
size of the hashtable to the pre-computed number of bits, so it won't
try to consume more gigabytes of memory than available for the
database.
FIXME: What is the average size of the rbt node that gets hashed? I
chose the pagesize (4k) as initial value to precompute the size of
the hashtable, but the value is based on feeling and not any real
data.
For future work, there are more places where we use result of the hash
value modulo some small number and that would benefit from Fibonacci
Hashing to get better distribution.
Notes:
a. A doubly linked list should be used here to speedup the removal of
the entries from the hashtable.
b. https://probablydance.com/2018/06/16/fibonacci-hashing-the-optimization-that-the-world-forgot-or-a-better-alternative-to-integer-modulo/
The ThreadSanitizer found a data race when updating the stale header.
Instead of trying to acquire the write lock and failing occasionally
which would skew the statistics, the dns_rdatasetheader_t.attributes
field has been promoted to use stdatomics. Updating the attributes in
the mark_header_ancient() and mark_header_stale() now uses the cmpxchg
to update the attributes forfeiting the need to hold the write lock on
the tree. Please note that mark_header_ancient() still needs to hold
the lock because .dirty is being updated in the same go.
RBTDB node can now appear on the deadnodes lists following the changes
to decrement_reference in 176b23b6cd98e5b58f832902fdbe964ee5f762d0 to
defer checking of node->down when the tree write lock is not held. The
node should be unlinked instead.
in addition to being more efficient, this prevents a possible crash by
looking up the node name before the tree sructure can be changed when
cleaning up dead nodes in addrdataset().
The rewrite of BIND 9 build system is a large work and cannot be reasonable
split into separate merge requests. Addition of the automake has a positive
effect on the readability and maintainability of the build system as it is more
declarative, it allows conditional and we are able to drop all of the custom
make code that BIND 9 developed over the years to overcome the deficiencies of
autoconf + custom Makefile.in files.
This squashed commit contains following changes:
- conversion (or rather fresh rewrite) of all Makefile.in files to Makefile.am
by using automake
- the libtool is now properly integrated with automake (the way we used it
was rather hackish as the only official way how to use libtool is via
automake
- the dynamic module loading was rewritten from a custom patchwork to libtool's
libltdl (which includes the patchwork to support module loading on different
systems internally)
- conversion of the unit test executor from kyua to automake parallel driver
- conversion of the system test executor from custom make/shell to automake
parallel driver
- The GSSAPI has been refactored, the custom SPNEGO on the basis that
all major KRB5/GSSAPI (mit-krb5, heimdal and Windows) implementations
support SPNEGO mechanism.
- The various defunct tests from bin/tests have been removed:
bin/tests/optional and bin/tests/pkcs11
- The text files generated from the MD files have been removed, the
MarkDown has been designed to be readable by both humans and computers
- The xsl header is now generated by a simple sed command instead of
perl helper
- The <irs/platform.h> header has been removed
- cleanups of configure.ac script to make it more simpler, addition of multiple
macros (there's still work to be done though)
- the tarball can now be prepared with `make dist`
- the system tests are partially able to run in oot build
Here's a list of unfinished work that needs to be completed in subsequent merge
requests:
- `make distcheck` doesn't yet work (because of system tests oot run is not yet
finished)
- documentation is not yet built, there's a different merge request with docbook
to sphinx-build rst conversion that needs to be rebased and adapted on top of
the automake
- msvc build is non functional yet and we need to decide whether we will just
cross-compile bind9 using mingw-w64 or fix the msvc build
- contributed dlz modules are not included neither in the autoconf nor automake
Save 'i' to 'locknum' and use that rather than using
'header->node->locknum' when performing the deferred
unlock as 'header->node->locknum' can theoretically be
different to 'i'.
- change name of 'bytes' to 'xfrsize' in dns_db_getsize() parameter list
and related variables; this is a more accurate representation of what
the function is doing
- change the size calculations in dns_db_getsize() to more accurately
represent the space needed for a *XFR message or journal file to contain
the data in the database. previously we returned the sizes of all
rdataslabs, including header overhead and offset tables, which
resulted in the database size being reported as much larger than the
equivalent *XFR or journal.
- map files caused a particular problem here: the fullname can't be
determined from the node while a file is being deserialized, because
the uppernode pointers aren't set yet. so we store "full name length"
in the dns_rbtnode structure while serializing, and clear it after
deserialization is complete.
Both clang-tidy and uncrustify chokes on statement like this:
for (...)
if (...)
break;
This commit uses a very simple semantic patch (below) to add braces around such
statements.
Semantic patch used:
@@
statement S;
expression E;
@@
while (...)
- if (E) S
+ { if (E) { S } }
@@
statement S;
expression E;
@@
for (...;...;...)
- if (E) S
+ { if (E) { S } }
@@
statement S;
expression E;
@@
if (...)
- if (E) S
+ { if (E) { S } }
6412 cleanup:
6413 dns_rdataset_disassociate(&neg);
6414 dns_rdataset_disassociate(&negsig);
CID 1452700 (#1 of 1): Dereference before null check (REVERSE_INULL)
check_after_deref: Null-checking closest suggests that it
may be null, but it has already been dereferenced on all
paths leading to the check.
6415 if (closest != NULL)
6416 free_noqname(mctx, &closest);
6367cleanup:
6368 dns_rdataset_disassociate(&neg);
6369 dns_rdataset_disassociate(&negsig);
CID 1452704 (#1 of 1): Dereference before null check
(REVERSE_INULL) check_after_deref: Null-checking noqname
suggests that it may be null, but it has already been
dereferenced on all paths leading to the check.
6370 if (noqname != NULL)
6371 free_noqname(mctx, &noqname);
This commit simplifies the cachedb rrset statistics in two ways:
- Introduce new rdtypecounter arithmetics, allowing bitwise
operations.
- Remove the special DLV statistic counter.
New rdtypecounter arithmetics
-----------------------------
"The rdtypecounter arithmetics is a brain twister". Replace the
enum counters with some defines. A rdtypecounter is now 8 bits for
RRtypes and 3 bits for flags:
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
| | | | | | S |NX| RRType |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
If the 8 bits for RRtype are all zero, this is an Other RRtype.
Bit 7 is the NXRRSET (NX) flag and indicates whether this is a
positive (0) or a negative (1) RRset.
Then bit 5 and 6 mostly tell you if this counter is for an active,
stale, or ancient RRtype:
S = 0x00 means Active
S = 0x01 means Stale
S = 0x10 means Ancient
Since a counter cannot be stale and ancient at the same time, we
treat S = 0x11 as a special case to deal with NXDOMAIN counters.
S = 0x11 indicates an NXDOMAIN counter and in this case the RRtype
field signals the expiry of this cached item:
RRType = 0 means Active
RRType = 1 means Stale
RRType = 2 means Ancient
The isc_buffer_allocate() function now cannot fail with ISC_R_MEMORY.
This commit removes all the checks on the return code using the semantic
patch from previous commit, as isc_buffer_allocate() now returns void.
This commits removes superfluous checks when using the isc_refcount API.
Examples of superfluous checks:
1. The isc_refcount_decrement function ensures there was not underflow,
so this check is superfluous:
INSIST(isc_refcount_decrement(&r) > 0);
2 .The isc_refcount_destroy() includes check whether the counter
is zero, therefore this is superfluous:
INSIST(isc_refcount_decrement(&r) == 1 && isc_refcount_destroy(&r));
In decrement_reference only test node->down if the tree lock
is held. As node->down is not always tested in
decrement_reference we need to test that it is non NULL in
cleanup_dead_nodes prior to removing the node from the rbt
tree. Additionally it is not always possible to aquire the
node lock and reactivate a node when adding parent nodes.
Reactivate such nodes in cleanup_dead_nodes if required.
