Instead of using on hash of the name modulo number of the buckets,
assign the locknum randomly with isc_random_uniform(). This makes
the locknum assignment aligned with qpcache and allows the bucket
number to be non-prime in the future.
Reduce the number of qpzone_ref() and qpzone_unref() calls in
qpzone_detachnode() by relying on the call_rcu to delay
the destruction of the lock buckets.
Instead of having many node_lock_count * sizeof(<member>) arrays, pack
all the members into a qpzone_bucket_t that is cacheline aligned and have
a single array of those.
Instead of having many node_lock_count * sizeof(<member>) arrays, pack
all the members into a qpcache_bucket_t struct that is cacheline aligned
and have a single array of those.
Additionaly, make both the head and the tail of isc_queue_t padded, not
just the head, to prevent false sharing of the lock-free structure with
the lock that follows it.
When query logging is enabled, named will now include the destination
address port in the logged message.
Example messages for before and after this change:
before: client @0x7608b2026000 10.53.0.1#52136 (example.test): query: example.test IN A +E(0)K (10.53.0.1)
after: client @0x729bf5c26000 10.53.0.1#35976 (example.test): query: example.test IN A +E(0)K (10.53.0.1#53)
In #1870, the expiration time of ANCIENT records were printed, but
actually the ancient records are very short lived, and the information
carries a little value.
Instead of printing the expiration of ANCIENT records, print the
expiration time of STALE records.
The original .ttl field was actually used as TTL in the dns_qpzone unit.
Restore the field by adding it to union with the .expire struct member
and cleanup all the code that added or subtracted 'now' from the ttl
field as that was misleading as 'now' would be always 0 for qpzone
database.
The find_coveringnsec() was getting the 'now' from two sources -
search->now and separate now argument. Things like this are ticking
bombs, remove the extra 'now' argument and use single source of 'now'.
When the mark_ancient() helper function was introduced, couple of places
with duplicate (or almost duplicate) code was missed. Move the
mark_ancient() function closer to the top of the file, and correctly use
it in places that mark the header as ANCIENT.
If we know that the header has ZEROTTL set, the server should never send
stale records for it and the TTL should never be anything else than 0.
The comment was already there, but the code was not matching the
comment.
The old name was misleading as it never meant time-to-live, e.g. number
of seconds from now when the header should expire. The true meaning was
an expiration time e.g. now + ttl. This was the original design bug
that caused the slip when we assigned header->ttl to rdataset->ttl.
Because the name was matching, nobody has questioned the correctness of
the code both during the MR review and during the numerous re-reviews
when we were searching for the cause of the 54 year TTL.
When the header has been marked as ANCIENT, but the ttl hasn't been
reset (this happens in couple of places), the rdataset TTL would be
set to the header timestamp instead to a reasonable TTL value.
Since this header has been already expired (ANCIENT is set), set the
rdataset TTL to 0 and don't reuse this field to print the expiration
time when dumping the cache. Instead of printing the time, we now
just print 'expired (awaiting cleanup'.
When there are parent and child zones on the same server, the DNSKEY
lookup was failing as the pending record we are validating is needed
to fetch the DNSKEY records. This change allows that to happen.
The caller is already setting STARTATZONE when the name being looked
up is a subdomain of the current domain.
The address lookups from ADB were not being validated, allowing
spoofed responses to be accepted and used for other lookups.
Validate the answers except when CD=1 is set in the triggering
request. Separate ADB names looked up with CD=1 from those without
CD=1, to prevent the use of unvalidated answers in the normal lookup
case (CD=0). Set the TTL on unvalidated (pending) responses to
ADB_CACHE_MINIMUM when adding them to the ADB.
the search for the deepest known zone cut in the cache could
improperly reject a node containing stale data, even if the
NS rdataset wasn't the data that was stale.
this change also improves the efficiency of the search by
stopping it when both NS and RRSIG(NS) have been found.
Change the names of the node reference counting functions
and add comments to make the mechanism easier to understand:
- newref() and decref() are now called qpcnode_acquire()/
qpznode_acquire() and qpcnode_release()/qpznode_release()
respectively; this reflects the fact that they modify both
the internal and external reference counters for a node.
- qpcnode_newref() and qpznode_newref() are now called
qpcnode_erefs_increment() and qpznode_erefs_increment(), and
qpcnode_decref() and qpznode_decref() are now called
qpcnode_erefs_decrement() and qpznode_erefs_decrement(),
to reflect that they only increase and decrease the node's
external reference counters, not internal.
This removes the db_nodelock_t structure and changes the node_locks
array to be composed only of isc_rwlock_t pointers. The .reference
member has been moved to qpdb->references in addition to
common.references that's external to dns_db API users. The .exiting
members has been completely removed as it has no use when the reference
counting is used correctly.
The origin_node in qpcache was always NULL, so we can remove the
getoriginode() function and origin_node pointer as the
dns_db_getoriginnode() correctly returns ISC_R_NOTFOUND when the
function is not implemented.
Cleanup the pattern in the decref() functions in both qpcache.c and
qpzone.c, so it follows the similar patter as we already have in
newref() function.
In order to avoid to loop to find the next position to store an EDE in
a dns_edectx_t, add a "nextede" state which holds the next available
position.
Also, in order ot avoid to loop to find if an EDE is already existing in
a dns_edectx_t, and avoid a duplicate, use a bitmap to immediately know
if the EDE is there or not.
Those both changes applies for adding or copying EDE.
Also make the direction of dns_ede_copy more explicit/avoid errors by
making "edectx_from" a const pointer.
Migrate tests cases in client_test code which were exclusively testing
code which is now all wrapped inside ede compilation unit. Those are
testing maximum number of EDE, duplicate EDE as well as truncation of
text of an EDE.
Also add coverage for the copy of EDE from an edectx to another one, as
well as checking the assertion of the maximum EDE info code which can be
used.
Instead of mixing the dns_resolver and dns_validator units directly with
the EDE code, split-out the dns_ede functionality into own separate
compilation unit and hide the implementation details behind abstraction.
Additionally, the EDE codes are directly copied into the ns_client
buffers by passing the EDE context to dns_resolver_createfetch().
This makes the dns_ede implementation simpler to use, although sligtly
more complicated on the inside.
Co-authored-by: Colin Vidal <colin@isc.org>
Co-authored-by: Ondřej Surý <ondrej@isc.org>
When an EDE code is added to a message, the code is converted early in a
big-endian order so it can be memcpy-ed directly in the EDE buffer that
will go on the wire.
This previous change forget to update debug logs which still assume the
EDE code was in host byte order. Add a separate variable to
differentiate both and avoid ambiguities
Extended DNS error 22 (No reachable authority) was previously detected
when `fctx_expired` fired. It turns out this function is used as a
"safety net" and the timeout detection should be caught earlier.
It was working though, because of another issue fixed by !9927. Since
this change, the recursive request timed out detection occurs before
`fctx_expired` so EDE 22 is not added to the response message anymore.
The fix of the problem is to add the EDE 22 code in two situations:
- When the dispatch code timed out (rctx_timedout) the resolver code
checks various properties to figure out if it needs to make another
fetch attempt. One of the paramters if the fetch expiration time. If
it expires, the whole recursion is canceled, so it now adds the EDE 22
code.
- If the fetch expiration time doesn't expires in the case above (and
other parameters allows it) a new fetch attempt is made (fctx_query).
But before the new request is actually made, the fetch expiration time
is re-checked. It might then has elapsed, and the whole recursion is
canceled. So it now also adds the EDE 22 code here as well.
Add support for EDE codes 1 (Unsupported DNSKEY Algorithm) and 2
(Unsupported DS Digest Type) which might occurs during DNSSEC
validation in case of unsupported DNSKEY algorithm or DS digest type.
Because DNSSEC internally kicks off various fetches, we need to copy
all encountered extended errors from fetch responses to the fetch
context. Upon an event, the errors from the fetch context are copied
to the client response.
ISCCC_R_SYNTAX, ISCCC_R_EXPIRED, and ISCCC_R_CLOCKSKEW have the
same usage and text formats as DNS_R_SYNTAX, DNS_R_EXPIRED and
DNS_R_CLOCKSCREW respectively. this was originally done because
result codes were defined in separate libraries, and some tool
might be linked with libisccc but not libdns. as the result codes
are now defined in only one place, there's no need to retain the
duplicates.
the isc_mem allocation functions can no longer fail; as a result,
ISC_R_NOMEMORY is now rarely used: only when an external library
such as libjson-c or libfstrm could return NULL. (even in
these cases, arguably we should assert rather than returning
ISC_R_NOMEMORY.)
code and comments that mentioned ISC_R_NOMEMORY have been
cleaned up, and the following functions have been changed to
type void, since (in most cases) the only value they could
return was ISC_R_SUCCESS:
- dns_dns64_create()
- dns_dyndb_create()
- dns_ipkeylist_resize()
- dns_kasp_create()
- dns_kasp_key_create()
- dns_keystore_create()
- dns_order_create()
- dns_order_add()
- dns_peerlist_new()
- dns_tkeyctx_create()
- dns_view_create()
- dns_zone_setorigin()
- dns_zone_setfile()
- dns_zone_setstream()
- dns_zone_getdbtype()
- dns_zone_setjournal()
- dns_zone_setkeydirectory()
- isc_lex_openstream()
- isc_portset_create()
- isc_symtab_create()
(the exception is dns_view_create(), which could have returned
other error codes in the event of a crypto library failure when
calling isc_file_sanitize(), but that should be a RUNTIME_CHECK
anyway.)
Track inside the dns_dnsseckey structure whether we have seen the
private key, or if this key only has a public key file.
If the key only has a public key file, or a DNSKEY reference in the
zone, mark the key 'pubkey'. In dnssec-signzone, if the key only
has a public key available, consider the key to be offline. Any
signatures that should be refreshed for which the key is not available,
retain the signature.
So in the code, 'expired' becomes 'refresh', and the new 'expired'
is only used to determine whether we need to keep the signature if
the corresponding key is not available (retaining the signature if
it is not expired).
In the 'keysthatsigned' function, we can remove:
- key->force_publish = false;
- key->force_sign = false;
because they are redundant ('dns_dnsseckey_create' already sets these
values to false).
Extended DNS error mechanism (EDE) enables to have several EDE raised
during a DNS resolution (typically, a DNSSEC query will do multiple
fetches which each of them can have an error). Add support to up to 3
EDE errors in an DNS response. If duplicates occur (two EDEs with the
same code, the extra text is not compared), only the first one will be
part of the DNS answer.
Because the maximum number of EDE is statically fixed, `ns_client_t`
object own a static vector of `DNS_DE_MAX_ERRORS` (instead of a linked
list, for instance). The array can be fully filled (all slots point to
an allocated `dns_ednsopt_t` object) or partially filled (or
empty). In such case, the first NULL slot means there is no more EDE
objects.
When TCP is used, 'fctx_query()' adds one second to the rtt
(round-trip time) value, but there's a bug when the decision
about using TCP is made already after the calculation. Move the
block of the code which looks up the peers list to decide
whether to use TCP into a place that's before the rtt calculation
is performed. This commit doesn't add or remove any code, it just
moves the code and adds a comment block.
When 'ISC_R_TIMEDOUT' is received in 'tcp_recv()', it times out the
oldest response in the active responses queue, and only after that it
checks whether other active responses have also timed out. So when
setting a timeout value for a read operation after a successful
connection, it makes sense to take the timeout value from the oldest
response in the active queue too, because, theoretically, the responses
can have different timeout values, e.g. when the TCP dispatch is shared.
Currently 'resp' is always NULL. Previously when connect and read
timeouts were not separated in dispatch this affected only logging, but
now since we are setting a new timeout after a successful connection,
we need to choose a suitable response from the active queue.
Previously, this function always acquires a node write lock if it
might need node cleanup in case the reference decrements to 0. In
fact, the lock is unnecessary if the reference is larger than 1 and it
can be optimized as an "easy" case. This optimization could even be
"necessary". In some extreme cases, many worker threads could repeat
acquring and releasing the reference on the same node, resulting in
severe lock contention for nothing (as the ref wouldn't decrement to 0
in most cases). This change would prevent noticeable performance
drop like query timeout for such cases.
Co-authored-by: JINMEI Tatuya <jtatuya@infoblox.com>
Co-authored-by: Ondřej Surý <ondrej@isc.org>
Currently, the fetch context will continue running even when the last
fetch (response) has been removed from the context, so named can process
and cache the answer. This can lead to a situation where the number of
outgoing recursing clients exceeds the the configured number for
recursive-clients.
Be more stringent about the recursive-clients limit and shutdown the
fetch context immediately after the last fetch has been canceled from
that particular fetch context.
Address Database (ADB) shares the memory for the short lived ADB
objects (finds, fetches, addrinfo) and the long lived ADB
objects (names, entries, namehooks). This could lead to a situation
where the resolver-heavy load would force evict ADB objects from the
database to point where ADB is completely empty, leading to even more
resolver-heavy load.
Make the short lived ADB objects use the other memory context that we
already created for the hashmaps. This makes the ADB overmem condition
to not be triggered by the ongoing resolver fetches.
In some places there was a limitation of the maximum timeout
value of INT16_MAX, which is only about 32 seconds. Refactor
the code to remove the limitation.
The network manager layer has two different timers with their
own timeout values for TCP connections: connect timeout and read
timeout. Separate the connect and the read TCP timeouts in the
dispatch module too.
struct fetchctx does have a list of pending validators as well as a
pointer to the HEAD validator. Remove the validator pointer to avoid
confusion, as there is no perticular reasons to have it directly
accessible outside of the list.
We started using isc_nm_bad_request() more actively throughout
codebase. In the case of HTTP/2 it can lead to a large count of
useless "Bad Request" messages in the BIND log, as often we attempt to
send such request over effectively finished HTTP/2 sessions.
This commit fixes that.
A call to isc_nm_read_stop() would always stop reading timer even in
manual timer control mode which was added with StreamDNS in mind. That
looks like an omission that happened due to how timers are controlled
in StreamDNS where we always stop the timer before pausing reading
anyway (see streamdns_on_complete_dnsmessage()). That would not work
well for HTTP, though, where we might want pause reading without
stopping the timer in the case we want to split incoming data into
multiple chunks to be processed independently.
I suppose that it happened due to NM refactoring in the middle of
StreamDNS development (at the time isc_nm_cancelread() and
isc_nm_pauseread() were removed), as the StreamDNS code seems to be
written as if timers are not stoping during a call to
isc_nm_read_stop().
This commit introduces manual read timer control as used by StreamDNS
and its underlying transports. Before that, DoH code would rely on the
timer control provided by TCP, which would reset the timer any time
some data arrived. Now, the timer is restarted only when a full DNS
message is processed in line with other DNS transports.
That change is required because we should not stop the timer when
reading from the network is paused due to throttling. We need a way to
drop timed-out clients, particularly those who refuse to read the data
we send.
