upstream/bind9
1
0
Fork 0
mirror of https://github.com/isc-projects/bind9.git synced 2026-06-09 05:12:05 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Remove all locking from XFR

Browse source 
Instead of locking the struct dns_xfrin members that get accessed from
the statistics, convert those into atomic types and use atomic accesses
to prevent ThreadSanitizer from blowing up.

In fact, even the atomic operations are not really needed here, because
all writes are done from a single thread and we don't really require
consistency from the statistics.  It's easier to use atomics here, but
it is slightly confusing as it suggests there might be multithreaded
accesses to those variables while in fact, the only off-thread access
happens when collecting the statistics.
This commit is contained in:
Ondřej Surý 2023-10-13 12:16:37 +02:00
parent 109dc883e7
commit e5c79261c0
No known key found for this signature in database
GPG key ID: 2820F37E873DEA41
1 changed files with 56 additions and 82 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

138
lib/dns/xfrin.c
View file

@ -134,14 +134,20 @@ struct dns_xfrin {
bool edns, expireoptset;
atomic_bool is_ixfr;
isc_mutex_t statslock;
/* Locked by statslock; can be accessed from the statistics channel. */
unsigned int nmsg; /*%< Number of messages recvd */
unsigned int nrecs; /*%< Number of records recvd */
uint64_t nbytes; /*%< Number of bytes received */
isc_time_t start; /*%< Start time of the transfer */
dns_transport_type_t soa_transport_type;
uint32_t end_serial;
/*
* Following variable were made atomic only for loading the values for
* the statistics channelr, thus all accesses can be **relaxed** because
* all store and load operations that affect XFR are done on the same
* thread and only the statistics channel thread could perform a load
* operation from a different thread and it's ok to not be precise in
* the statistics.
*/
atomic_uint nmsg; /*%< Number of messages recvd */
atomic_uint nrecs; /*%< Number of records recvd */
atomic_uint_fast64_t nbytes; /*%< Number of bytes received */
_Atomic(isc_time_t) start; /*%< Start time of the transfer */
_Atomic(dns_transport_type_t) soa_transport_type;
atomic_uint_fast32_t end_serial;
unsigned int maxrecords; /*%< The maximum number of
* records set for the zone */
@ -470,10 +476,9 @@ failure:
static isc_result_t
xfr_rr(dns_xfrin_t *xfr, dns_name_t *name, uint32_t ttl, dns_rdata_t *rdata) {
isc_result_t result;
uint_fast32_t end_serial;
LOCK(&xfr->statslock);
xfr->nrecs++;
UNLOCK(&xfr->statslock);
atomic_fetch_add_relaxed(&xfr->nrecs, 1);
if (rdata->type == dns_rdatatype_none ||
dns_rdatatype_ismeta(rdata->type))
@ -511,16 +516,15 @@ redo:
result = DNS_R_FORMERR;
goto failure;
}
LOCK(&xfr->statslock);
xfr->end_serial = dns_soa_getserial(rdata);
UNLOCK(&xfr->statslock);
if (!DNS_SERIAL_GT(xfr->end_serial, xfr->ixfr.request_serial) &&
end_serial = dns_soa_getserial(rdata);
atomic_store_relaxed(&xfr->end_serial, end_serial);
if (!DNS_SERIAL_GT(end_serial, xfr->ixfr.request_serial) &&
!dns_zone_isforced(xfr->zone))
{
xfrin_log(xfr, ISC_LOG_DEBUG(3),
"requested serial %u, "
"primary has %u, not updating",
xfr->ixfr.request_serial, xfr->end_serial);
"primary has %" PRIuFAST32 ", not updating",
xfr->ixfr.request_serial, end_serial);
result = DNS_R_UPTODATE;
goto failure;
}
@ -544,11 +548,10 @@ redo:
* Remember the serial number in the initial SOA.
* We need it to recognize the end of an IXFR.
*/
LOCK(&xfr->statslock);
xfr->end_serial = dns_soa_getserial(rdata);
UNLOCK(&xfr->statslock);
end_serial = dns_soa_getserial(rdata);
atomic_store_relaxed(&xfr->end_serial, end_serial);
if (xfr->reqtype == dns_rdatatype_ixfr &&
!DNS_SERIAL_GT(xfr->end_serial, xfr->ixfr.request_serial) &&
!DNS_SERIAL_GT(end_serial, xfr->ixfr.request_serial) &&
!dns_zone_isforced(xfr->zone))
{
/*
@ -558,8 +561,8 @@ redo:
*/
xfrin_log(xfr, ISC_LOG_DEBUG(3),
"requested serial %u, "
"primary has %u, not updating",
xfr->ixfr.request_serial, xfr->end_serial);
"primary has %" PRIuFAST32 ", not updating",
xfr->ixfr.request_serial, end_serial);
result = DNS_R_UPTODATE;
goto failure;
}
@ -619,7 +622,8 @@ redo:
case XFRST_IXFR_ADD:
if (rdata->type == dns_rdatatype_soa) {
uint32_t soa_serial = dns_soa_getserial(rdata);
if (soa_serial == xfr->end_serial) {
if (soa_serial == atomic_load_relaxed(&xfr->end_serial))
{
CHECK(ixfr_commit(xfr));
atomic_store(&xfr->state, XFRST_IXFR_END);
break;
@ -764,9 +768,7 @@ dns_xfrin_getstarttime(dns_xfrin_t *xfr) {
REQUIRE(VALID_XFRIN(xfr));
LOCK(&xfr->statslock);
start = xfr->start;
UNLOCK(&xfr->statslock);
return (atomic_load_relaxed(&xfr->start));
return (start);
}
@ -818,15 +820,9 @@ dns_xfrin_getstate(const dns_xfrin_t *xfr, const char **statestr,
uint32_t
dns_xfrin_getendserial(dns_xfrin_t *xfr) {
uint32_t end_serial;
REQUIRE(VALID_XFRIN(xfr));
LOCK(&xfr->statslock);
end_serial = xfr->end_serial;
UNLOCK(&xfr->statslock);
return (end_serial);
return (atomic_load_relaxed(&xfr->end_serial));
}
void
@ -835,11 +831,9 @@ dns_xfrin_getstats(dns_xfrin_t *xfr, unsigned int *nmsgp, unsigned int *nrecsp,
REQUIRE(VALID_XFRIN(xfr));
REQUIRE(nmsgp != NULL && nrecsp != NULL && nbytesp != NULL);
LOCK(&xfr->statslock);
*nmsgp = xfr->nmsg;
*nrecsp = xfr->nrecs;
*nbytesp = xfr->nbytes;
UNLOCK(&xfr->statslock);
SET_IF_NOT_NULL(nmsgp, atomic_load_relaxed(&xfr->nmsg));
SET_IF_NOT_NULL(nrecsp, atomic_load_relaxed(&xfr->nrecs));
SET_IF_NOT_NULL(nbytesp, atomic_load_relaxed(&xfr->nbytes));
}
const isc_sockaddr_t *
@ -869,15 +863,9 @@ dns_xfrin_gettransporttype(const dns_xfrin_t *xfr) {
dns_transport_type_t
dns_xfrin_getsoatransporttype(dns_xfrin_t *xfr) {
dns_transport_type_t soa_transport_type;
REQUIRE(VALID_XFRIN(xfr));
LOCK(&xfr->statslock);
soa_transport_type = xfr->soa_transport_type;
UNLOCK(&xfr->statslock);
return (soa_transport_type);
return (atomic_load_relaxed(&xfr->soa_transport_type));
}
const dns_name_t *
@ -1007,8 +995,6 @@ xfrin_create(isc_mem_t *mctx, dns_zone_t *zone, dns_db_t *db,
dns_view_weakattach(dns_zone_getview(zone), &xfr->view);
dns_name_init(&xfr->name, NULL);
isc_mutex_init(&xfr->statslock);
atomic_init(&xfr->shuttingdown, false);
atomic_init(&xfr->is_ixfr, false);
@ -1024,7 +1010,7 @@ xfrin_create(isc_mem_t *mctx, dns_zone_t *zone, dns_db_t *db,
atomic_init(&xfr->state, XFRST_ZONEXFRREQUEST);
}
xfr->start = isc_time_now();
atomic_init(&xfr->start, isc_time_now());
if (tsigkey != NULL) {
dns_tsigkey_attach(tsigkey, &xfr->tsigkey);
@ -1116,9 +1102,8 @@ xfrin_start(dns_xfrin_t *xfr) {
* The "SOA before" mode is used, where the SOA request is
* using the same transport as the XFR.
*/
LOCK(&xfr->statslock);
xfr->soa_transport_type = dns_xfrin_gettransporttype(xfr);
UNLOCK(&xfr->statslock);
atomic_init(&xfr->soa_transport_type,
dns_xfrin_gettransporttype(xfr));
}
/* Set the maximum timer */
@ -1414,12 +1399,11 @@ xfrin_send_request(dns_xfrin_t *xfr) {
CHECK(add_opt(msg, udpsize, reqnsid, reqexpire));
}
LOCK(&xfr->statslock);
xfr->nmsg = 0;
xfr->nrecs = 0;
xfr->nbytes = 0;
xfr->start = isc_time_now();
UNLOCK(&xfr->statslock);
atomic_init(&xfr->nmsg, 0);
atomic_init(&xfr->nrecs, 0);
atomic_init(&xfr->nbytes, 0);
atomic_init(&xfr->start, isc_time_now());
msg->id = xfr->id;
if (xfr->tsigctx != NULL) {
dst_context_destroy(&xfr->tsigctx);
@ -1551,11 +1535,7 @@ xfrin_recv_done(isc_result_t result, isc_region_t *region, void *arg) {
dns_message_setclass(msg, xfr->rdclass);
LOCK(&xfr->statslock);
if (xfr->nmsg > 0) {
msg->tcp_continuation = 1;
}
UNLOCK(&xfr->statslock);
msg->tcp_continuation = (atomic_load_relaxed(&xfr->nmsg) > 0) ? 1 : 0;
isc_buffer_init(&buffer, region->base, region->length);
isc_buffer_add(&buffer, region->length);
@ -1758,29 +1738,21 @@ xfrin_recv_done(isc_result_t result, isc_region_t *region, void *arg) {
CHECK(dns_message_getquerytsig(msg, xfr->mctx, &xfr->lasttsig));
} else if (dns_message_gettsigkey(msg) != NULL) {
xfr->sincetsig++;
LOCK(&xfr->statslock);
if (xfr->sincetsig > 100 || xfr->nmsg == 0 ||
if (xfr->sincetsig > 100 ||
atomic_load_relaxed(&xfr->nmsg) == 0 ||
atomic_load(&xfr->state) == XFRST_AXFR_END ||
atomic_load(&xfr->state) == XFRST_IXFR_END)
{
UNLOCK(&xfr->statslock);
result = DNS_R_EXPECTEDTSIG;
goto failure;
}
UNLOCK(&xfr->statslock);
}
/*
* Update the number of messages received.
* Update the number of messages and bytes received.
*/
LOCK(&xfr->statslock);
xfr->nmsg++;
/*
* Update the number of bytes received.
*/
xfr->nbytes += buffer.used;
UNLOCK(&xfr->statslock);
atomic_fetch_add_relaxed(&xfr->nmsg, 1);
atomic_fetch_add_relaxed(&xfr->nbytes, buffer.used);
/*
* Take the context back.
@ -1884,18 +1856,21 @@ xfrin_destroy(dns_xfrin_t *xfr) {
* Calculate the length of time the transfer took,
* and print a log message with the bytes and rate.
*/
msecs = isc_time_microdiff(&now, &xfr->start) / 1000;
isc_time_t start = atomic_load_relaxed(&xfr->start);
msecs = isc_time_microdiff(&now, &start) / 1000;
if (msecs == 0) {
msecs = 1;
}
persec = (xfr->nbytes * 1000) / msecs;
persec = (atomic_load_relaxed(&xfr->nbytes) * 1000) / msecs;
xfrin_log(xfr, ISC_LOG_INFO,
"Transfer completed: %d messages, %d records, "
"%" PRIu64 " bytes, "
"%u.%03u secs (%u bytes/sec) (serial %u)",
xfr->nmsg, xfr->nrecs, xfr->nbytes,
"%u.%03u secs (%u bytes/sec) (serial %" PRIuFAST32 ")",
atomic_load_relaxed(&xfr->nmsg),
atomic_load_relaxed(&xfr->nrecs),
atomic_load_relaxed(&xfr->nbytes),
(unsigned int)(msecs / 1000), (unsigned int)(msecs % 1000),
(unsigned int)persec, xfr->end_serial);
(unsigned int)persec, atomic_load_relaxed(&xfr->end_serial));
if (xfr->dispentry != NULL) {
dns_dispatch_done(&xfr->dispentry);
@ -1971,7 +1946,6 @@ xfrin_destroy(dns_xfrin_t *xfr) {
isc_timer_destroy(&xfr->max_idle_timer);
isc_timer_destroy(&xfr->max_time_timer);
isc_mutex_destroy(&xfr->statslock);
isc_mem_putanddetach(&xfr->mctx, xfr, sizeof(*xfr));
}