DNS_RDATA_UPDATE flag.
531. [func] Rdata really should be initalized before being
assigned to (dns_rdata_fromwire(), dns_rdata_fromtext(),
dns_rdata_clone(), dns_rdata_fromregion()),
check that it is.
the result code check after the loop deal with ISC_R_SUCCESS
instead. The previous fix had the side effect of keeping multiple
answers to non-ANY queries, causing assertion failures in cache_name()
which is not prepared to deal with this situation
the multiple answer RRsets by validating each one separately. Also,
eliminated the "done" variable in answer_response() because in the
rare situations where it got set to ISC_TRUE, it caused the function
to return prematurely by exiting a loop with a result of ISC_R_SUCCESS
and hitting a "if (result != ISC_R_NOMORE) return (result);" test
immediately following following the loop.
This should fix [RT #109], "ANY query in secure zone crashes server".
isc_socket_cancel() has been called. resquery_senddone() has been fixed
to account for this (there was a race condition because of the faulty
assumption that only ISC_R_CANCELED would be returned after
isc_socket_cancel()).
The interaction of resquery_response() with the dispatcher now has to be
closely reviewed to make sure it does not have a similar problem with
a canceled query. If the faulty assumption hadn't been faulty, it was
fine, but now I'm not so sure.
the DNS_DB_NOEXACT flag. This means that the resolver will start look for the
key at or above the parent zone, and will return the key set from the parent
zone if key sets are present in both the parent and child.
isc_socket_sendto() calls that did not yet have their
completion events posted, so it could (rarely) end up
destroying the query context and then want to use
it again when the send event posted, triggering an
assertion as it tried to cancel an already-canceled
query. (RT #77)
... so now the resolver counts the number of outstanding isc_socket_sendto()s
and only destroys the query context when there are no more.
state. Set DNS_DISPATCHATTR_CONNECTED when calling
dns_dispatch_createtcp() for connected TCP sockets
or call dns_dispatch_starttcp() when the socket is
connected.
error during shutdown of a busy caching server (specifically, one
making many outgoing TCP connections).
Bug 1: resquery_destroy() accessed query->fctx->res->mctx, even though
query->fctx could already be destroyed at this point.
Bug 2: resquery_connected() called fctx_done(query->fctx, result)
immediately after fctx_cancelquery(), which could destroy *query.
Also added some explanatory comments.
cast to dns_rdatatype_t via macros of their same name
so that they are of the proper integral type wherever
a dns_rdatatype_t is needed.
(And a few prototypes that I recently bogusly changed to take "int" parameters
in place of "dns_rdatatype_t" to accomodate the type of an enum identifier
have been reverted to again properly take a dns_rdatatype_t.)
Mostly, several functions that take pointers as arguments, almost
always char * pointers, had those pointers qualified with "const".
Those that returned pointers to previously const-qualified arguments
had their return values qualified as const. Some structure members
were qualified as const to retain that attribute from the variables
from which they were assigned.
The macro DE_CONST is used to deal with a handful of very special
places where something is qualified as const but really needs to have
its const qualifier removed.
rdata.c now defines macros for the prototypes of the basic rdata functions,
and all of the lib/dns/rdata/**/*.c files now use them.
Some minor integer-compatibility issues. (IE, ~0x03 is a signed int,
so assigning it to an unsigned int should use a cast. The type of an
enum member is int, so there are some conversion issues there, too.)
A pointers-to-function should not be cast to a pointer-to-object.
Variables should not be named for C reserved identifiers.
One or two set-but-not-used variables removed.
Minor other ISC style cleanups.
