When isc_nmhandle_t gets deactivated, it could be just put onto array
stack to be reused later to safe some initialization time.
Unfortunately, this might hide some use-after-free errors.
Disable the inactive handles caching when compiled with Address or
Thread Sanitizer.
The isc__nmsocket_t has locked array of isc_nmhandle_t that's not used
for anything. The isc__nmhandle_get() adds the isc_nmhandle_t to the
locked array (and resized if necessary) and removed when
isc_nmhandle_put() finally destroys the handle. That's all it does, so
it serves no useful purpose.
Remove the .ah_handles, .ah_size, and .ah_frees members of the
isc__nmsocket_t and .ah_pos member of the isc_nmhandle_t struct.
When the TCP, TCPDNS or TLSDNS connection times out, the isc__nm_uvreq_t
would be pushed into sock->inactivereqs before the uv_tcp_connect()
callback finishes. Because the isc__nmsocket_t keeps the list of
inactive isc__nm_uvreq_t, this would cause use-after-free only when the
sock->inactivereqs is full (which could never happen because the failure
happens in connection timeout callback) or when the sock->inactivereqs
mechanism is completely removed (f.e. when running under Address or
Thread Sanitizer).
Delay isc__nm_uvreq_t deallocation to the connection callback and only
signal the connection callback should be called by shutting down the
libuv socket from the connection timeout callback.
Commit aab691d512 did not fix all possible
scenarios in which the ns_statscounter_recursclients counter underflows.
The solution implemented therein can be ineffective e.g. when CNAME
chaining happens with prefetching enabled.
Here is an example recursive resolution scenario in which the
ns_statscounter_recursclients counter can underflow with the current
logic in effect:
1. Query processing starts, the answer is not found in the cache, so
recursion is started. The NS_CLIENTATTR_RECURSING attribute is set.
ns_statscounter_recursclients is incremented (Δ = +1).
2. Recursion completes, returning a CNAME. client->recursionquota is
non-NULL, so the NS_CLIENTATTR_RECURSING attribute remains set.
ns_statscounter_recursclients is decremented (Δ = 0).
3. Query processing restarts.
4. The current QNAME (the target of the CNAME from step 2) is found in
the cache, with a TTL low enough to trigger a prefetch.
5. query_prefetch() attaches to client->recursionquota.
ns_statscounter_recursclients is not incremented because
query_prefetch() does not do that (Δ = 0).
6. Query processing restarts.
7. The current QNAME (the target of the CNAME from step 4) is not found
in the cache, so recursion is started. client->recursionquota is
already attached to (since step 5) and the NS_CLIENTATTR_RECURSING
attribute is set (since step 1), so ns_statscounter_recursclients is
not incremented (Δ = 0).
8. The prefetch from step 5 completes. client->recursionquota is
detached from in prefetch_done(). ns_statscounter_recursclients is
not decremented because prefetch_done() does not do that (Δ = 0).
9. Recursion for the current QNAME completes. client->recursionquota
is already detached from, i.e. set to NULL (since step 8), and the
NS_CLIENTATTR_RECURSING attribute is set (since step 1), so
ns_statscounter_recursclients is decremented (Δ = -1).
Another possible scenario is that after step 7, recursion for the target
of the CNAME from step 4 completes before the prefetch for the CNAME
itself. fetch_callback() then notices that client->recursionquota is
non-NULL and decrements ns_statscounter_recursclients, even though
client->recursionquota was attached to by query_prefetch() and therefore
not accompanied by an incrementation of ns_statscounter_recursclients.
The net result is also an underflow.
Instead of trying to properly handle all possible orderings of events
set into motion by normal recursion and prefetch-triggered recursion,
adjust ns_statscounter_recursclients whenever the recursive clients
quota is successfully attached to or detached from. Remove the
NS_CLIENTATTR_RECURSING attribute altogether as its only purpose is made
obsolete by this change.
Commit b6d40b3c4e removed most uses of the
'fctx' variable from the rctx_dispfail() function: it is now only needed
by the FCTXTRACE3() macro. However, when --enable-querytrace is not in
effect, that macro evaluates to a list of UNUSED() macros that does not
include "UNUSED(fctx);". This triggers the following compilation
warning when building without --enable-querytrace:
resolver.c: In function 'rctx_dispfail':
resolver.c:7888:21: warning: unused variable 'fctx' [-Wunused-variable]
7888 | fetchctx_t *fctx = rctx->fctx;
| ^~~~
Fix by adding "UNUSED(fctx);" lines to all FCTXTRACE*() macros. This is
safe to do because all of those macros use the 'fctx' local variable, so
there is no danger of introducing new errors caused by use of undeclared
identifiers.
Due to a bug in gcc-11, the build fails when AddressSanitizer is
enabled. Downgrading the -Wstringop-overread to just a warning in the
gcc:asan build allows the code to compile.
The keep-response-order option has been obsoleted, and in this commit,
remove the keep-response-order ACL map rendering the option no-op, the
call the isc_nm_sequential() and the now unused isc_nm_sequential()
function itself.
The keep-response-order option has been introduced when TCP pipelining
has been introduced to BIND 9 as a failsafe for possibly non-compliant
clients.
Declare the keep-response-order obsolete as all DNS clients should
either support out-of-order processing or don't send more DNS queries
until the DNS response for the previous one has been received.
There was an artificial limit of 23 on the number of simultaneous
pipelined queries in the single TCP connection. The new network
managers is capable of handling "unlimited" (limited only by the TCP
read buffer size ) queries similar to "unlimited" handling of the DNS
queries receive over UDP.
Don't limit the number of TCP queries that we can process within a
single TCP read callback.
Extend the timeouts system test to ensure that the maximum outgoing
transfer time (max-transfer-time-out) and maximum outgoing transfer idle
time (max-transfer-idle-out) works as expected. This is done by
lowering the limits to 5/1 minutes and testing that the connection has
been dropped while sleeping between the individual XFR messages.
While refactoring the libns to use the new network manager, the
max-transfer-*-out options were not implemented and they were turned
non-operational.
Reimplement the max-transfer-idle-out functionality using the write
timer and max-transfer-time-out using the new isc_nm_timer API.
While refactoring the lib/ns/xfrout.c, it was discovered that .shutdown
and .shutdown_arg members of ns_client_t structure are unused.
Remove the unused members and associated code that was using in it in
the ns_xfrout.
bin/tests/system/makejournal needs to ignore DNS_R_SEENINCLUDE
when calling dns_db_load(), otherwise it cannot generate a journal
for a zone file with a $INCLUDE statement.
When invalid DNS message is received, there was a handling mechanism for
DoH that would be called to return proper HTTP response.
Reuse this mechanism and reset the TCP connection when the client is
blackholed, DNS message is completely bogus or the ns_client receives
response instead of query.
- certain TCP result codes, including ISC_R_EOF and
ISC_R_CONNECTIONRESET, were being mapped to ISC_R_SHUTTINGDOWN
before calling the response handler in tcp_recv_cancelall().
the result codes should be passed through to the response handler
without being changed.
- the response handlers, resquery_response() and req_response(), had
code to return immediately if encountering ISC_R_EOF, but this is
not the correct behavior; that should only happen in the case of
ISC_R_CANCELED when it was the caller that canceled the operation
- ISC_R_CONNECTIONRESET was not being caught in rctx_dispfail().
- removed code in rctx_dispfail() to retry queries without EDNS
when receiving ISC_R_EOF; this is now treated the same as any
other connection failure.
Use the isc_nmhandle_setwritetimeout() function in the netmgr unit test
to allow more time for writing and reading the responses because some of
the intervals that are used in the unit tests are really small leaving a
little room for any delays.
In some situations (unit test and forthcoming XFR timeouts MR), we need
to modify the write timeout independently of the read timeout. Add a
isc_nmhandle_setwritetimeout() function that could be called before
isc_nm_send() to specify a custom write timeout interval.
Extend the timeouts system test that bursts the queries for large TXT
record and never read any responses back filling up the server TCP write
buffer. The test should work with the default wmem_max value on
Linux (208k).
When the outgoing TCP write buffers are full because the other party is
not reading the data, the uv_write() could wait indefinitely on the
uv_loop and never calling the callback. Add a new write timer that uses
the `tcp-idle-timeout` value to interrupt the TCP connection when we are
not able to send data for defined period of time.
Ondřej Surý
Michał Kępień
Petr Špaček
Matthijs Mekking
