The TCP module has been updated to use the generic functions from
netmgr.c instead of its own local copies. This brings the module
mostly up to par with the TCPDNS and TLSDNS modules.
Serveral problems were discovered and fixed after the change in
the connection timeout in the previous commits:
* In TLSDNS, the connection callback was not called at all under some
circumstances when the TCP connection had been established, but the
TLS handshake hadn't been completed yet. Additional checks have
been put in place so that tls_cycle() will end early when the
nmsocket is invalidated by the isc__nm_tlsdns_shutdown() call.
* In TCP, TCPDNS and TLSDNS, new connections would be established
even when the network manager was shutting down. The new
call isc__nm_closing() has been added and is used to bail out
early even before uv_tcp_connect() is attempted.
Similarly to the read timeout, it's now possible to recover from
ISC_R_TIMEDOUT event by restarting the timer from the connect callback.
The change here also fixes platforms that missing the socket() options
to set the TCP connection timeout, by moving the timeout code into user
space. On platforms that support setting the connect timeout via a
socket option, the timeout has been hardcoded to 2 minutes (the maximum
value of tcp-initial-timeout).
Previously, when the client timed out on read, the client socket would
be automatically closed and destroyed when the nmhandle was detached.
This commit changes the logic so that it's possible for the callback to
recover from the ISC_R_TIMEDOUT event by restarting the timer. This is
done by calling isc_nmhandle_settimeout(), which prevents the timeout
handling code from destroying the socket; instead, it continues to wait
for data.
One specific use case for multiple timeouts is serve-stale - the client
socket could be created with shorter timeout (as specified with
stale-answer-client-timeout), so we can serve the requestor with stale
answer, but keep the original query running for a longer time.
The full netmgr test suite is unstable when run in CI due to various
timing issues. Previously, we enabled the full test suite only when
CI_ENABLE_ALL_TESTS environment variable was set, but that went against
original intent of running the full suite when an individual developer
would run it locally.
This change disables the full test suite only when running in the CI and
the CI_ENABLE_ALL_TESTS is not set.
Fix race between zone_maintenance and dns_zone_notifyreceive functions,
zone_maintenance was attempting to read a zone flag calling
DNS_ZONE_FLAG(zone, flag) while dns_zone_notifyreceive was updating
a flag in the same zone calling DNS_ZONE_SETFLAG(zone, ...).
The code reading the flag in zone_maintenance was not protected by the
zone's lock, to avoid a race the zone's lock is now being acquired
before an attempt to read the zone flag is made.
When a unit test binary hangs, the GitLab CI job in which it is run is
stuck until its run time limit is exceeded. Furthermore, it is not
trivial to determine which test(s) hung in a given GitLab CI job based
on its log. To prevent these issues, enforce a run time limit on every
binary executed by the lib/unit-test-driver.sh script. Use a timeout of
5 minutes for consistency with older BIND 9 branches, which employed
Kyua for running unit tests. Report an exit code of 124 when the run
time limit is exceeded for a unit test binary, for consistency with the
"timeout" tool included in GNU coreutils.
See "BUGS" section at:
https://www.openssl.org/docs/man1.1.1/man3/SSL_get_error.html
It is mentioned there that when TLS status equals SSL_ERROR_SYSCALL
AND errno == 0 it means that underlying transport layer returned EOF
prematurely. However, we are managing the transport ourselves, so we
should just resume reading from the TCP socket.
It seems that this case has been handled properly on modern versions
of OpenSSL. That being said, the situation goes in line with the
manual: it is briefly mentioned there that SSL_ERROR_SYSCALL might be
returned not only in a case of low-level errors (like system call
failures).
When we are recursing, RPZ processing is not allowed. But when we are
performing a lookup due to "stale-answer-client-timeout", we are still
recursing. This effectively means that RPZ processing is disabled on
such a lookup.
In this case, bail the "stale-answer-client-timeout" lookup and wait
for recursion to complete, as we we can't perform the RPZ rewrite
rules reliably.
The dboption DNS_DBFIND_STALEONLY caused confusion because it implies
we are looking for stale data **only** and ignore any active RRsets in
the cache. Rename it to DNS_DBFIND_STALETIMEOUT as it is more clear
the option is related to a lookup due to "stale-answer-client-timeout".
Rename other usages of "staleonly", instead use "lookup due to...".
Also rename related function and variable names.
When doing a staleonly lookup we don't want to fallback to recursion.
After all, there are obviously problems with recursion, otherwise we
wouldn't do a staleonly lookup.
When resuming from recursion however, we should restore the
RECURSIONOK flag, allowing future required lookups for this client
to recurse.
When implementing "stale-answer-client-timeout", we decided that
we should only return positive answers prematurely to clients. A
negative response is not useful, and in that case it is better to
wait for the recursion to complete.
To do so, we check the result and if it is not ISC_R_SUCCESS, we
decide that it is not good enough. However, there are more return
codes that could lead to a positive answer (e.g. CNAME chains).
This commit removes the exception and now uses the same logic that
other stale lookups use to determine if we found a useful stale
answer (stale_found == true).
This means we can simplify two test cases in the serve-stale system
test: nodata.example is no longer treated differently than data.example.
The NS_QUERYATTR_ANSWERED attribute is to prevent sending a response
twice. Without the attribute, this may happen if a staleonly lookup
found a useful answer and sends a response to the client, and later
recursion ends and also tries to send a response.
The attribute was also used to mask adding a duplicate RRset. This is
considered harmful. When we created a response to the client with a
stale only lookup (regardless if we actually have send the response),
we should clear the rdatasets that were added during that lookup.
Mark such rdatasets with the a new attribute,
DNS_RDATASETATTR_STALE_ADDED. Set a query attribute
NS_QUERYATTR_STALEOK if we may have added rdatasets during a stale
only lookup. Before creating a response on a normal lookup, check if
we can expect rdatasets to have been added during a staleonly lookup.
If so, clear the rdatasets from the message with the attribute
DNS_RDATASETATTR_STALE_ADDED set.
With stale-answer-client-timeout, we may send a response to the client,
but we may want to hold on to the network manager handle, because
recursion is going on in the background, or we need to refresh a
stale RRset.
Simplify the setting of 'nodetach':
* During a staleonly lookup we should not detach the nmhandle, so just
set it prior to 'query_lookup()'.
* During a staleonly "stalefirst" lookup set the 'nodetach' to true
if we are going to refresh the RRset.
Now there is no longer the need to clear the 'nodetach' if we go
through the "dbfind_stale", "stale_refresh_window", or "stale_only"
paths.
When doing a staleonly lookup, ignore active RRsets from cache. If we
don't, we may add a duplicate RRset to the message, and hit an
assertion failure in query.c because adding the duplicate RRset to the
ANSWER section failed.
This can happen on a race condition. When a client query is received,
the recursion is started. When 'stale-answer-client-timeout' triggers
around the same time the recursion completes, the following sequence
of events may happen:
1. Queue the "try stale" fetch_callback() event to the client task.
2. Add the RRsets from the authoritative response to the cache.
3. Queue the "fetch complete" fetch_callback() event to the client task.
4. Execute the "try stale" fetch_callback(), which retrieves the
just-inserted RRset from the database.
5. In "ns_query_done()" we are still recursing, but the "staleonly"
query attribute has already been cleared. In other words, the
query will resume when recursion ends (it already has ended but is
still on the task queue).
6. Execute the "fetch complete" fetch_callback(). It finds the answer
from recursion in the cache again and tries to add the duplicate to
the answer section.
This commit changes the logic for finding stale answers in the cache,
such that on "stale_only" lookups actually only stale RRsets are
considered. It refactors the code so that code paths for "dbfind_stale",
"stale_refresh_window", and "stale_only" are more clear.
First we call some generic code that applies in all three cases,
formatting the domain name for logging purposes, increment the
trystale stats, and check if we actually found stale data that we can
use.
The "dbfind_stale" lookup will return SERVFAIL if we didn't found a
usable answer, otherwise we will continue with the lookup
(query_gotanswer()). This is no different as before the introduction of
"stale-answer-client-timeout" and "stale-refresh-time".
The "stale_refresh_window" lookup is similar to the "dbfind_stale"
lookup: return SERVFAIL if we didn't found a usable answer, otherwise
continue with the lookup (query_gotanswer()).
Finally the "stale_only" lookup.
If the "stale_only" lookup was triggered because of an actual client
timeout (stale-answer-client-timeout > 0), and if database lookup
returned a stale usable RRset, trigger a response to the client.
Otherwise return and wait until the recursion completes (or the
resolver query times out).
If the "stale_only" lookup is a "stale-anwer-client-timeout 0" lookup,
preferring stale data over a lookup. In this case if there was no stale
data, or the data was not a positive answer, retry the lookup with the
stale options cleared, a.k.a. a normal lookup. Otherwise, continue
with the lookup (query_gotanswer()) and refresh the stale RRset. This
will trigger a response to the client, but will not detach the handle
because a fetch will be created to refresh the RRset.
The stale-answer-client-timeout feature introduced a dependancy on
when a client may be detached from the handle. The dboption
DNS_DBFIND_STALEONLY was reused to track this attribute. This overloads
the meaning of this database option, and actually introduced a bug
because the option was checked in other places. In particular, in
'ns_query_done()' there is a check for 'RECURSING(qctx->client) &&
(!QUERY_STALEONLY(&qctx->client->query) || ...' and the condition is
satisfied because recursion has not completed yet and
DNS_DBFIND_STALEONLY is already cleared by that time (in
query_lookup()), because we found a useful answer and we should detach
the client from the handle after sending the response.
Add a new boolean to the client structure to keep track of client
detach from handle is allowed or not. It is only disallowed if we are
in a staleonly lookup and we didn't found a useful answer.
This commit fixes crash in dig when it encounters non-expected header
value. The bug was introduced at some point late in the last DoH
development cycle. Also, refactors the relevant code a little bit to
ensure better incoming data validation for client-side DoH
connections.
add matching macros to pass arguments from called methods
to generic methods. This will reduce the amount of work
required when extending methods.
Also cleanup unnecessary UNUSED declarations.
util.h requires ISC_CONSTRUCTOR definition, which depends on config.h
inclusion. It does not include it from isc/util.h (or any other header).
Using isc/util.h fails hard when isc/util.h is used without including
bind's config.h.
Move the check to c file, where ISC_CONSTRUCTOR is used. Ensure config.h
is included there.
CDS/CDNSKEY DELETE records are only useful if they are signed,
otherwise the parent cannot verify these RRsets anyway. So once the DS
has been removed (and signaled to BIND), we can remove the DNSKEY and
RRSIG records, and at this point we can also remove the CDS/CDNSKEY
records.
The 'keymgr_key_init()' function initializes key states if they have
not been set previously. It looks at the key timing metadata and
determines using the given times whether a state should be set to
RUMOURED or OMNIPRESENT.
However, the DNSKEY and ZRRSIG states were mixed up: When looking
at the Activate timing metadata we should set the ZRRSIG state, and
when looking at the Published timing metadata we should set the
DNSKEY state.
The current isc_time_now uses CLOCK_REALTIME_COARSE which only updates
on a timer tick. This clock is generally fine for millisecond accuracy,
but on servers with 100hz clocks, this clock is nowhere near accurate
enough for microsecond accuracy.
This commit adds a new isc_time_now_hires function that uses
CLOCK_REALTIME, which gives the current time, though it is somewhat
expensive to call. When microsecond accuracy is required, it may be
required to use extra resources for higher accuracy.
When NETMGR_TRACE(_VERBOSE) is enabled, the build would fail on some
non-Linux non-glibc platforms because:
* Use <stdint.h> print macros because uint_fast32_t is not always
unsigned long
* The header <execinfo.h> is not available on non-glibc, thus commit
adds dummy backtrace() and backtrace_symbols_fd() functions for
platforms without HAVE_BACKTRACE
The netmgr unit tests were designed to push the system limits to maximum
by sending as many queries as possible in the busy loop from multiple
threads. This mostly works with UDP, but in the stateful protocol where
establishing the connection takes more time, it failed quite often in
the CI. On FreeBSD, this happened more often, because the socket() call
would fail spuriosly making the problem even worse.
This commit does several things to improve reliability:
* return value of isc_nm_<proto>connect() is always checked and retried
when scheduling the connection fails
* The busy while loop has been slowed down with usleep(1000); so the
netmgr threads could schedule the work and get executed.
* The isc_thread_yield() was replaced with usleep(1000); also to allow
the other threads to do any work.
* Instead of waiting on just one variable, we wait for multiple
variables to reach the final value
* We are wrapping the netmgr operations (connects, reads, writes,
accepts) with reference counting and waiting for all the callbacks to
be accounted for.
This has two effects:
a) the isc_nm_t is always clean of active sockets and handles when
destroyed, so it will prevent the spurious INSIST(references == 1)
from isc_nm_destroy()
b) the unit test now ensures that all the callbacks are always called
when they should be called, so any stuck test means that there was
a missing callback call and it is always a real bug
These changes allows us to remove the workaround that would not run
certain tests on systems without port load-balancing.
In tls_error(), we now call isc__nm_tlsdns_failed_read() instead of just
stopping timer and reading from the socket. This allows us to properly
cleanup any pending operation on the socket.
When shutting down, calling the isc__nm_failed_connect_cb() was delayed
until the connect callback would be called. It turned out that the
connect callback might not get called at all when the socket is being
shut down. Call the failed_connect_cb() directly in the
tlsdns_shutdown() instead of waiting for the connect callback to call it.
After a partial write the tls.senddata buffer would be rearranged to
contain only the data tha wasn't sent and the len part would be made
shorter, which would lead to attempt to free only part of a socket's
tls.senddata buffer.
The tlsdns_cycle() might call uv_write() to write data to the socket,
when this happens and the socket is shutdown before the callback
completes, the uvreq structure was not freed because the callback would
be called with non-zero status code.
The RFC7828 specifies the keepalive interval to be 16-bit, specified in
units of 100 milliseconds and the configuration options tcp-*-timeouts
are following the suit. The units of 100 milliseconds are very
unintuitive and while we can't change the configuration and presentation
format, we should not follow this weird unit in the API.
This commit changes the isc_nm_(get|set)timeouts() functions to work
with milliseconds and convert the values to milliseconds before passing
them to the function, not just internally.
The udp, tcpdns and tlsdns contained lot of cut&paste code or code that
was very similar making the stack harder to maintain as any change to
one would have to be copied to the the other protocols.
In this commit, we merge the common parts into the common functions
under isc__nm_<foo> namespace and just keep the little differences based
on the socket type.
After the TCPDNS refactoring the initial and idle timers were broken and
only the tcp-initial-timeout was always applied on the whole TCP
connection.
This broke any TCP connection that took longer than tcp-initial-timeout,
most often this would affect large zone AXFRs.
This commit changes the timeout logic in this way:
* On TCP connection accept the tcp-initial-timeout is applied
and the timer is started
* When we are processing and/or sending any DNS message the timer is
stopped
* When we stop processing all DNS messages, the tcp-idle-timeout
is applied and the timer is started again
Dynamic zones with dnssec-policy could not be thawed because KASP
zones were considered always dynamic. But a dynamic KASP zone should
also check whether updates are disabled.
This commit fixes loading the certificate chain files so that the full
chain could be sent to the clients which require that for
verification. Before that fix only the top most certificate would be
loaded from the chain and sent to clients preventing some of them to
perform certificate validation (e.g. Windows 10 DoH client).
The transport should also be detached when we skip a master, otherwise
named will crash when sending a SOA query to the next master over TLS,
because the transport must be NULL when we enter
'dns_view_gettransport'.
When we query the resolver for a domain name that is in the same zone
for which is already one or more fetches outstanding, we could
potentially hit the fetch limits. If so, recursion fails immediately
for the incoming query and if serve-stale is enabled, we may try to
return a stale answer.
If the resolver is also is authoritative for the parent zone (for
example the root zone), first a delegation is found, but we first
check the cache for a better response.
Nothing is found in the cache, so we try to recurse to find the
answer to the query.
Because of fetch-limits 'dns_resolver_createfetch()' returns an error,
which 'ns_query_recurse()' propagates to the caller,
'query_delegation_recurse()'.
Because serve-stale is enabled, 'query_usestale()' is called,
setting 'qctx->db' to the cache db, but leaving 'qctx->version'
untouched. Now 'query_lookup()' is called to search for stale data
in the cache database with a non-NULL 'qctx->version'
(which is set to a zone db version), and thus we hit an assertion
in rbtdb.
This crash was introduced in 'main' by commit
8bcd7fe69e.
- use a value less than 2^32 for DNS_ZONEFLG_FIXJOURNAL; a larger value
could cause problems in some build environments. the zone flag
DNS_ZONEFLG_DIFFONRELOAD, which was no longer in use, has now been
deleted and its value reused for _FIXJOURNAL.
*** CID 329157: Null pointer dereferences (REVERSE_INULL)
/lib/dns/journal.c: 754 in journal_open()
748 j->header.index_size * sizeof(journal_rawpos_t));
749 }
750 if (j->index != NULL) {
751 isc_mem_put(j->mctx, j->index,
752 j->header.index_size * sizeof(journal_pos_t));
753 }
CID 329157: Null pointer dereferences (REVERSE_INULL)
Null-checking "j->filename" suggests that it may be null, but it has already been dereferenced on all paths leading to the check.
754 if (j->filename != NULL) {
755 isc_mem_free(j->mctx, j->filename);
756 }
757 if (j->fp != NULL) {
758 (void)isc_stdio_close(j->fp);
759 }
It is advisable to disable Nagle's algorithm for HTTP/2 connections
because multiple HTTP/2 streams could be multiplexed over one
transport connection. Thus, delays when delivering small packets could
bring down performance for the whole session. HTTP/2 is meant to be
used this way.
when called from within the context of a network thread,
isc_nm_tlsconnect() hangs. it is waiting for the socket's
result code to be updated, but that update is supposed to happen
asynchronously in the network thread, and if we're already blocking
in the network thread, it can never occur.
we can kluge around this by setting the socket result code
early; this works for most clients (including "dig"), but it causes
inconsistent behaviors that manifest as test failures in the DoH unit
test.
so we kluged around it even more by setting the socket result code
early *only when running in the network thread*. we need a better
solution for this problem, but this will do for now.
