bind/bind-9.11-CVE-2023-50387.patch
Petr Menšík 5573a679ad Fail the DNSSEC validation on the first failure
Be more strict when encountering DNSSEC validation failures - fail on
the first failure.  This will break domains that have DNSSEC signing
keys with duplicate key ids, but this is something that's much easier
to fix on the authoritative side, so we are just going to be strict
on the resolver side where it is causing performance problems.

(cherry picked from commit 8b7ecba9885e163c07c2dd3e1ceab79b2ba89e34)

Add normal and slow task queues

Split the task manager queues into normal and slow task queues, so we
can move the tasks that blocks processing for a long time (like DNSSEC
validation) into the slow queue which doesn't block fast
operations (like responding from the cache).  This mitigates the whole
class of KeyTrap-like issues.

(cherry picked from commit db083a21726300916fa0b9fd8a433a796fedf636)

Don't iterate from start every time we select new signing key

Improve the selecting of the new signing key by remembering where
we stopped the iteration and just continue from that place instead
of iterating from the start over and over again each time.

(cherry picked from commit 75faeefcab47e4f1e12b358525190b4be90f97de)

Optimize selecting the signing key

Don't parse the crypto data before parsing and matching the id and the
algorithm.

(cherry picked from commit b38552cca7200a72658e482f8407f57516efc5db)

6322.	[security]	Specific DNS answers could cause a denial-of-service
			condition due to DNS validation taking a long time.
			(CVE-2023-50387) [GL #4424]

			The same code change also addresses another problem:
			preparing NSEC3 closest encloser proofs could exhaust
			available CPU resources. (CVE-2023-50868) [GL #4459]

; Resolves: CVE-2023-50387 CVE-2023-50868
Resolves: RHEL-25681 RHEL-25649
2024-03-26 18:07:39 +01:00

738 lines
24 KiB
Diff

From 4c20ab54ec503f65d8ee0b863cbf41103d95130a Mon Sep 17 00:00:00 2001
From: Mark Andrews <marka@isc.org>
Date: Wed, 22 Nov 2023 16:59:03 +1100
Subject: [PATCH] Fail the DNSSEC validation on the first failure
Be more strict when encountering DNSSEC validation failures - fail on
the first failure. This will break domains that have DNSSEC signing
keys with duplicate key ids, but this is something that's much easier
to fix on the authoritative side, so we are just going to be strict
on the resolver side where it is causing performance problems.
(cherry picked from commit 8b7ecba9885e163c07c2dd3e1ceab79b2ba89e34)
Add normal and slow task queues
Split the task manager queues into normal and slow task queues, so we
can move the tasks that blocks processing for a long time (like DNSSEC
validation) into the slow queue which doesn't block fast
operations (like responding from the cache). This mitigates the whole
class of KeyTrap-like issues.
(cherry picked from commit db083a21726300916fa0b9fd8a433a796fedf636)
Don't iterate from start every time we select new signing key
Improve the selecting of the new signing key by remembering where
we stopped the iteration and just continue from that place instead
of iterating from the start over and over again each time.
(cherry picked from commit 75faeefcab47e4f1e12b358525190b4be90f97de)
Optimize selecting the signing key
Don't parse the crypto data before parsing and matching the id and the
algorithm.
(cherry picked from commit b38552cca7200a72658e482f8407f57516efc5db)
6322. [security] Specific DNS answers could cause a denial-of-service
condition due to DNS validation taking a long time.
(CVE-2023-50387) [GL #4424]
The same code change also addresses another problem:
preparing NSEC3 closest encloser proofs could exhaust
available CPU resources. (CVE-2023-50868) [GL #4459]
---
lib/dns/dst_api.c | 25 ++++--
lib/dns/include/dns/validator.h | 1 +
lib/dns/include/dst/dst.h | 4 +
lib/dns/resolver.c | 2 +-
lib/dns/validator.c | 97 +++++++++-----------
lib/dns/win32/libdns.def.in | 1 +
lib/isc/include/isc/task.h | 11 ++-
lib/isc/task.c | 153 ++++++++++++++++++++++----------
8 files changed, 186 insertions(+), 108 deletions(-)
diff --git a/lib/dns/dst_api.c b/lib/dns/dst_api.c
index 2156384ec1..6bcd99796c 100644
--- a/lib/dns/dst_api.c
+++ b/lib/dns/dst_api.c
@@ -105,6 +105,7 @@ static isc_result_t frombuffer(dns_name_t *name,
dns_rdataclass_t rdclass,
isc_buffer_t *source,
isc_mem_t *mctx,
+ bool no_rdata,
dst_key_t **keyp);
static isc_result_t algorithm_status(unsigned int alg);
@@ -764,6 +765,13 @@ isc_result_t
dst_key_fromdns(dns_name_t *name, dns_rdataclass_t rdclass,
isc_buffer_t *source, isc_mem_t *mctx, dst_key_t **keyp)
{
+ return (dst_key_fromdns_ex(name, rdclass, source, mctx, false, keyp));
+}
+
+isc_result_t
+dst_key_fromdns_ex(dns_name_t *name, dns_rdataclass_t rdclass,
+ isc_buffer_t *source, isc_mem_t *mctx, bool no_rdata,
+ dst_key_t **keyp) {
uint8_t alg, proto;
uint32_t flags, extflags;
dst_key_t *key = NULL;
@@ -792,7 +800,7 @@ dst_key_fromdns(dns_name_t *name, dns_rdataclass_t rdclass,
}
result = frombuffer(name, alg, flags, proto, rdclass, source,
- mctx, &key);
+ mctx, no_rdata, &key);
if (result != ISC_R_SUCCESS)
return (result);
key->key_id = id;
@@ -814,7 +822,7 @@ dst_key_frombuffer(dns_name_t *name, unsigned int alg,
REQUIRE(dst_initialized);
result = frombuffer(name, alg, flags, protocol, rdclass, source,
- mctx, &key);
+ mctx, false, &key);
if (result != ISC_R_SUCCESS)
return (result);
@@ -1915,7 +1923,8 @@ computeid(dst_key_t *key) {
static isc_result_t
frombuffer(dns_name_t *name, unsigned int alg, unsigned int flags,
unsigned int protocol, dns_rdataclass_t rdclass,
- isc_buffer_t *source, isc_mem_t *mctx, dst_key_t **keyp)
+ isc_buffer_t *source, isc_mem_t *mctx, bool no_rdata,
+ dst_key_t **keyp)
{
dst_key_t *key;
isc_result_t ret;
@@ -1940,10 +1949,12 @@ frombuffer(dns_name_t *name, unsigned int alg, unsigned int flags,
return (DST_R_UNSUPPORTEDALG);
}
- ret = key->func->fromdns(key, source);
- if (ret != ISC_R_SUCCESS) {
- dst_key_free(&key);
- return (ret);
+ if (!no_rdata) {
+ ret = key->func->fromdns(key, source);
+ if (ret != ISC_R_SUCCESS) {
+ dst_key_free(&key);
+ return (ret);
+ }
}
}
diff --git a/lib/dns/include/dns/validator.h b/lib/dns/include/dns/validator.h
index cc4478d6d4..b4bf8f29db 100644
--- a/lib/dns/include/dns/validator.h
+++ b/lib/dns/include/dns/validator.h
@@ -160,6 +160,7 @@ struct dns_validator {
unsigned int depth;
unsigned int authcount;
unsigned int authfail;
+ bool failed;
isc_stdtime_t start;
};
diff --git a/lib/dns/include/dst/dst.h b/lib/dns/include/dst/dst.h
index 180c841307..a8be2daf67 100644
--- a/lib/dns/include/dst/dst.h
+++ b/lib/dns/include/dst/dst.h
@@ -435,6 +435,10 @@ dst_key_tofile(const dst_key_t *key, int type, const char *directory);
*/
isc_result_t
+dst_key_fromdns_ex(dns_name_t *name, dns_rdataclass_t rdclass,
+ isc_buffer_t *source, isc_mem_t *mctx, bool no_rdata,
+ dst_key_t **keyp);
+isc_result_t
dst_key_fromdns(dns_name_t *name, dns_rdataclass_t rdclass,
isc_buffer_t *source, isc_mem_t *mctx, dst_key_t **keyp);
/*%<
diff --git a/lib/dns/resolver.c b/lib/dns/resolver.c
index 4f71f48039..487107614c 100644
--- a/lib/dns/resolver.c
+++ b/lib/dns/resolver.c
@@ -9267,7 +9267,7 @@ dns_resolver_create(dns_view_t *view,
if (result != ISC_R_SUCCESS)
goto cleanup_buckets;
res->buckets[i].task = NULL;
- result = isc_task_create(taskmgr, 0, &res->buckets[i].task);
+ result = isc_task_create(taskmgr, ISC_TASK_QUANTUM_SLOW, &res->buckets[i].task);
if (result != ISC_R_SUCCESS) {
DESTROYLOCK(&res->buckets[i].lock);
goto cleanup_buckets;
diff --git a/lib/dns/validator.c b/lib/dns/validator.c
index 2a5c3caa6a..0b257fe874 100644
--- a/lib/dns/validator.c
+++ b/lib/dns/validator.c
@@ -1207,6 +1207,12 @@ create_validator(dns_validator_t *val, dns_name_t *name, dns_rdatatype_t type,
* val->key at it.
*
* If val->key is non-NULL, this returns the next matching key.
+ * If val->key is already non-NULL, start searching from the next position in
+ * 'rdataset' to find the *next* key that could have signed 'siginfo', then
+ * set val->key to that.
+ *
+ * Returns ISC_R_SUCCESS if a possible matching key has been found,
+ * ISC_R_NOTFOUND if not. Any other value indicates error.
*/
static isc_result_t
get_dst_key(dns_validator_t *val, dns_rdata_rrsig_t *siginfo,
@@ -1216,54 +1222,59 @@ get_dst_key(dns_validator_t *val, dns_rdata_rrsig_t *siginfo,
isc_buffer_t b;
dns_rdata_t rdata = DNS_RDATA_INIT;
dst_key_t *oldkey = val->key;
- bool foundold;
+ bool no_rdata = false;
- if (oldkey == NULL)
- foundold = true;
- else {
- foundold = false;
+ if (oldkey == NULL) {
+ result = dns_rdataset_first(rdataset);
+ } else {
+ dst_key_free(&oldkey);
val->key = NULL;
+ result = dns_rdataset_next(rdataset);
+ }
+
+ if (result != ISC_R_SUCCESS) {
+ goto done;
}
- result = dns_rdataset_first(rdataset);
- if (result != ISC_R_SUCCESS)
- goto failure;
do {
dns_rdataset_current(rdataset, &rdata);
isc_buffer_init(&b, rdata.data, rdata.length);
isc_buffer_add(&b, rdata.length);
INSIST(val->key == NULL);
- result = dst_key_fromdns(&siginfo->signer, rdata.rdclass, &b,
- val->view->mctx, &val->key);
+ result = dst_key_fromdns_ex(&siginfo->signer, rdata.rdclass, &b,
+ val->view->mctx, no_rdata,
+ &val->key);
if (result == ISC_R_SUCCESS) {
if (siginfo->algorithm ==
(dns_secalg_t)dst_key_alg(val->key) &&
siginfo->keyid ==
(dns_keytag_t)dst_key_id(val->key) &&
+ (dst_key_flags(val->key) & DNS_KEYFLAG_REVOKE) ==
+ 0 &&
dst_key_iszonekey(val->key))
{
- if (foundold) {
- /*
- * This is the key we're looking for.
- */
- return (ISC_R_SUCCESS);
- } else if (dst_key_compare(oldkey, val->key)) {
- foundold = true;
- dst_key_free(&oldkey);
+ if (no_rdata) {
+ /* Retry with full key */
+ dns_rdata_reset(&rdata);
+ dst_key_free(&val->key);
+ no_rdata = false;
+ continue;
}
+ /* This is the key we're looking for. */
+ goto done;
}
dst_key_free(&val->key);
}
dns_rdata_reset(&rdata);
result = dns_rdataset_next(rdataset);
+ no_rdata = true;
} while (result == ISC_R_SUCCESS);
- if (result == ISC_R_NOMORE)
- result = ISC_R_NOTFOUND;
- failure:
- if (oldkey != NULL)
- dst_key_free(&oldkey);
+done:
+ if (result == ISC_R_NOMORE) {
+ result = ISC_R_NOTFOUND;
+ }
return (result);
}
@@ -1633,37 +1644,13 @@ validate(dns_validator_t *val, bool resume) {
continue;
}
- do {
- vresult = verify(val, val->key, &rdata,
- val->siginfo->keyid);
- if (vresult == ISC_R_SUCCESS)
- break;
- if (val->keynode != NULL) {
- dns_keynode_t *nextnode = NULL;
- result = dns_keytable_findnextkeynode(
- val->keytable,
- val->keynode,
- &nextnode);
- dns_keytable_detachkeynode(val->keytable,
- &val->keynode);
- val->keynode = nextnode;
- if (result != ISC_R_SUCCESS) {
- val->key = NULL;
- break;
- }
- val->key = dns_keynode_key(val->keynode);
- if (val->key == NULL)
- break;
- } else {
- if (get_dst_key(val, val->siginfo, val->keyset)
- != ISC_R_SUCCESS)
- break;
- }
- } while (1);
- if (vresult != ISC_R_SUCCESS)
+ vresult = verify(val, val->key, &rdata,
+ val->siginfo->keyid);
+ if (vresult != ISC_R_SUCCESS) {
+ val->failed = true;
validator_log(val, ISC_LOG_DEBUG(3),
"failed to verify rdataset");
- else {
+ } else {
dns_rdataset_trimttl(event->rdataset,
event->sigrdataset,
val->siginfo, val->start,
@@ -1700,9 +1687,13 @@ validate(dns_validator_t *val, bool resume) {
} else {
validator_log(val, ISC_LOG_DEBUG(3),
"verify failure: %s",
- isc_result_totext(result));
+ isc_result_totext(vresult));
resume = false;
}
+ if (val->failed) {
+ result = ISC_R_NOMORE;
+ break;
+ }
}
if (result != ISC_R_NOMORE) {
validator_log(val, ISC_LOG_DEBUG(3),
diff --git a/lib/dns/win32/libdns.def.in b/lib/dns/win32/libdns.def.in
index f597049493..7320653439 100644
--- a/lib/dns/win32/libdns.def.in
+++ b/lib/dns/win32/libdns.def.in
@@ -1439,6 +1439,7 @@ dst_key_format
dst_key_free
dst_key_frombuffer
dst_key_fromdns
+dst_key_fromdns_ex
dst_key_fromfile
dst_key_fromgssapi
dst_key_fromlabel
diff --git a/lib/isc/include/isc/task.h b/lib/isc/include/isc/task.h
index 28e5e25fc6..42f7763869 100644
--- a/lib/isc/include/isc/task.h
+++ b/lib/isc/include/isc/task.h
@@ -98,8 +98,15 @@ ISC_LANG_BEGINDECLS
***/
typedef enum {
- isc_taskmgrmode_normal = 0,
- isc_taskmgrmode_privileged
+ isc_taskqueue_normal = 0,
+ isc_taskqueue_slow = 1,
+} isc_taskqueue_t;
+
+#define ISC_TASK_QUANTUM_SLOW 1024
+
+typedef enum {
+ isc_taskmgrmode_normal = 0,
+ isc_taskmgrmode_privileged
} isc_taskmgrmode_t;
/*% Task and task manager methods */
diff --git a/lib/isc/task.c b/lib/isc/task.c
index 048639350b..cc83269df2 100644
--- a/lib/isc/task.c
+++ b/lib/isc/task.c
@@ -107,6 +107,7 @@ struct isc__task {
isc_eventlist_t on_shutdown;
unsigned int nevents;
unsigned int quantum;
+ unsigned int qid;
unsigned int flags;
isc_stdtime_t now;
isc_time_t tnow;
@@ -141,11 +142,11 @@ struct isc__taskmgr {
/* Locked by task manager lock. */
unsigned int default_quantum;
LIST(isc__task_t) tasks;
- isc__tasklist_t ready_tasks;
- isc__tasklist_t ready_priority_tasks;
+ isc__tasklist_t ready_tasks[2];
+ isc__tasklist_t ready_priority_tasks[2];
isc_taskmgrmode_t mode;
#ifdef ISC_PLATFORM_USETHREADS
- isc_condition_t work_available;
+ isc_condition_t work_available[2];
isc_condition_t exclusive_granted;
isc_condition_t paused;
#endif /* ISC_PLATFORM_USETHREADS */
@@ -247,13 +248,13 @@ isc_taskmgrmode_t
isc__taskmgr_mode(isc_taskmgr_t *manager0);
static inline bool
-empty_readyq(isc__taskmgr_t *manager);
+empty_readyq(isc__taskmgr_t *manager, isc_taskqueue_t qid);
static inline isc__task_t *
-pop_readyq(isc__taskmgr_t *manager);
+pop_readyq(isc__taskmgr_t *manager, isc_taskqueue_t qid);
static inline void
-push_readyq(isc__taskmgr_t *manager, isc__task_t *task);
+push_readyq(isc__taskmgr_t *manager, isc__task_t *task, isc_taskqueue_t qid);
static struct isc__taskmethods {
isc_taskmethods_t methods;
@@ -324,7 +325,8 @@ task_finished(isc__task_t *task) {
* any idle worker threads so they
* can exit.
*/
- BROADCAST(&manager->work_available);
+ BROADCAST(&manager->work_available[isc_taskqueue_normal]);
+ BROADCAST(&manager->work_available[isc_taskqueue_slow]);
}
#endif /* USE_WORKER_THREADS */
UNLOCK(&manager->lock);
@@ -364,7 +366,13 @@ isc__task_create(isc_taskmgr_t *manager0, unsigned int quantum,
INIT_LIST(task->events);
INIT_LIST(task->on_shutdown);
task->nevents = 0;
- task->quantum = quantum;
+ if (quantum >= ISC_TASK_QUANTUM_SLOW) {
+ task->qid = isc_taskqueue_slow;
+ task->quantum = quantum - ISC_TASK_QUANTUM_SLOW;
+ } else {
+ task->qid = isc_taskqueue_normal;
+ task->quantum = quantum;
+ }
task->flags = 0;
task->now = 0;
isc_time_settoepoch(&task->tnow);
@@ -476,11 +484,11 @@ task_ready(isc__task_t *task) {
LOCK(&manager->lock);
LOCK(&task->lock);
- push_readyq(manager, task);
+ push_readyq(manager, task, task->qid);
UNLOCK(&task->lock);
#ifdef USE_WORKER_THREADS
if (manager->mode == isc_taskmgrmode_normal || has_privilege)
- SIGNAL(&manager->work_available);
+ SIGNAL(&manager->work_available[task->qid]);
#endif /* USE_WORKER_THREADS */
UNLOCK(&manager->lock);
}
@@ -961,13 +969,13 @@ isc__task_getcurrenttimex(isc_task_t *task0, isc_time_t *t) {
* Caller must hold the task manager lock.
*/
static inline bool
-empty_readyq(isc__taskmgr_t *manager) {
+empty_readyq(isc__taskmgr_t *manager, isc_taskqueue_t qid) {
isc__tasklist_t queue;
if (manager->mode == isc_taskmgrmode_normal)
- queue = manager->ready_tasks;
+ queue = manager->ready_tasks[qid];
else
- queue = manager->ready_priority_tasks;
+ queue = manager->ready_priority_tasks[qid];
return (EMPTY(queue));
}
@@ -981,18 +989,18 @@ empty_readyq(isc__taskmgr_t *manager) {
* Caller must hold the task manager lock.
*/
static inline isc__task_t *
-pop_readyq(isc__taskmgr_t *manager) {
+pop_readyq(isc__taskmgr_t *manager, isc_taskqueue_t qid) {
isc__task_t *task;
if (manager->mode == isc_taskmgrmode_normal)
- task = HEAD(manager->ready_tasks);
+ task = HEAD(manager->ready_tasks[qid]);
else
- task = HEAD(manager->ready_priority_tasks);
+ task = HEAD(manager->ready_priority_tasks[qid]);
if (task != NULL) {
- DEQUEUE(manager->ready_tasks, task, ready_link);
+ DEQUEUE(manager->ready_tasks[qid], task, ready_link);
if (ISC_LINK_LINKED(task, ready_priority_link))
- DEQUEUE(manager->ready_priority_tasks, task,
+ DEQUEUE(manager->ready_priority_tasks[qid], task,
ready_priority_link);
}
@@ -1006,16 +1014,16 @@ pop_readyq(isc__taskmgr_t *manager) {
* Caller must hold the task manager lock.
*/
static inline void
-push_readyq(isc__taskmgr_t *manager, isc__task_t *task) {
- ENQUEUE(manager->ready_tasks, task, ready_link);
+push_readyq(isc__taskmgr_t *manager, isc__task_t *task, isc_taskqueue_t qid) {
+ ENQUEUE(manager->ready_tasks[qid], task, ready_link);
if ((task->flags & TASK_F_PRIVILEGED) != 0)
- ENQUEUE(manager->ready_priority_tasks, task,
+ ENQUEUE(manager->ready_priority_tasks[qid], task,
ready_priority_link);
manager->tasks_ready++;
}
static void
-dispatch(isc__taskmgr_t *manager) {
+dispatch(isc__taskmgr_t *manager, isc_taskqueue_t qid) {
isc__task_t *task;
#ifndef USE_WORKER_THREADS
unsigned int total_dispatch_count = 0;
@@ -1094,13 +1102,13 @@ dispatch(isc__taskmgr_t *manager) {
* If a pause has been requested, don't do any work
* until it's been released.
*/
- while ((empty_readyq(manager) || manager->pause_requested ||
+ while ((empty_readyq(manager, qid) || manager->pause_requested ||
manager->exclusive_requested) && !FINISHED(manager))
{
XTHREADTRACE(isc_msgcat_get(isc_msgcat,
ISC_MSGSET_GENERAL,
ISC_MSG_WAIT, "wait"));
- WAIT(&manager->work_available, &manager->lock);
+ WAIT(&manager->work_available[qid], &manager->lock);
XTHREADTRACE(isc_msgcat_get(isc_msgcat,
ISC_MSGSET_TASK,
ISC_MSG_AWAKE, "awake"));
@@ -1113,7 +1121,7 @@ dispatch(isc__taskmgr_t *manager) {
XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_TASK,
ISC_MSG_WORKING, "working"));
- task = pop_readyq(manager);
+ task = pop_readyq(manager, qid);
if (task != NULL) {
unsigned int dispatch_count = 0;
bool done = false;
@@ -1278,7 +1286,7 @@ dispatch(isc__taskmgr_t *manager) {
*/
#ifdef USE_WORKER_THREADS
LOCK(&task->lock);
- push_readyq(manager, task);
+ push_readyq(manager, task, qid);
UNLOCK(&task->lock);
#else
ENQUEUE(new_ready_tasks, task, ready_link);
@@ -1297,10 +1305,14 @@ dispatch(isc__taskmgr_t *manager) {
* we're stuck. Automatically drop privileges at that
* point and continue with the regular ready queue.
*/
- if (manager->tasks_running == 0 && empty_readyq(manager)) {
+ if (manager->tasks_running == 0 && empty_readyq(manager, isc_taskqueue_normal) && empty_readyq(manager, isc_taskqueue_slow)) {
manager->mode = isc_taskmgrmode_normal;
- if (!empty_readyq(manager))
- BROADCAST(&manager->work_available);
+ if (!empty_readyq(manager, isc_taskqueue_normal)) {
+ BROADCAST(&manager->work_available[isc_taskqueue_normal]);
+ }
+ if (!empty_readyq(manager, isc_taskqueue_slow)) {
+ BROADCAST(&manager->work_available[isc_taskqueue_slow]);
+ }
}
#endif
}
@@ -1322,13 +1334,37 @@ static isc_threadresult_t
#ifdef _WIN32
WINAPI
#endif
-run(void *uap) {
+run_normal(void *uap) {
isc__taskmgr_t *manager = uap;
XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
ISC_MSG_STARTING, "starting"));
- dispatch(manager);
+ dispatch(manager, isc_taskqueue_normal);
+
+ XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
+ ISC_MSG_EXITING, "exiting"));
+
+#ifdef OPENSSL_LEAKS
+ ERR_remove_state(0);
+#endif
+
+ return ((isc_threadresult_t)0);
+}
+#endif /* USE_WORKER_THREADS */
+
+#ifdef USE_WORKER_THREADS
+static isc_threadresult_t
+#ifdef _WIN32
+WINAPI
+#endif
+run_slow(void *uap) {
+ isc__taskmgr_t *manager = uap;
+
+ XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
+ ISC_MSG_STARTING, "starting"));
+
+ dispatch(manager, isc_taskqueue_slow);
XTHREADTRACE(isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
ISC_MSG_EXITING, "exiting"));
@@ -1347,7 +1383,8 @@ manager_free(isc__taskmgr_t *manager) {
#ifdef USE_WORKER_THREADS
(void)isc_condition_destroy(&manager->exclusive_granted);
- (void)isc_condition_destroy(&manager->work_available);
+ (void)isc_condition_destroy(&manager->work_available[isc_taskqueue_normal]);
+ (void)isc_condition_destroy(&manager->work_available[isc_taskqueue_slow]);
(void)isc_condition_destroy(&manager->paused);
isc_mem_free(manager->mctx, manager->threads);
#endif /* USE_WORKER_THREADS */
@@ -1414,12 +1451,20 @@ isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
#ifdef USE_WORKER_THREADS
manager->workers = 0;
manager->threads = isc_mem_allocate(mctx,
- workers * sizeof(isc_thread_t));
+ 2 * workers * sizeof(isc_thread_t));
if (manager->threads == NULL) {
result = ISC_R_NOMEMORY;
goto cleanup_lock;
}
- if (isc_condition_init(&manager->work_available) != ISC_R_SUCCESS) {
+ if (isc_condition_init(&manager->work_available[isc_taskqueue_normal]) != ISC_R_SUCCESS) {
+ UNEXPECTED_ERROR(__FILE__, __LINE__,
+ "isc_condition_init() %s",
+ isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
+ ISC_MSG_FAILED, "failed"));
+ result = ISC_R_UNEXPECTED;
+ goto cleanup_threads;
+ }
+ if (isc_condition_init(&manager->work_available[isc_taskqueue_slow]) != ISC_R_SUCCESS) {
UNEXPECTED_ERROR(__FILE__, __LINE__,
"isc_condition_init() %s",
isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
@@ -1448,8 +1493,10 @@ isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
default_quantum = DEFAULT_DEFAULT_QUANTUM;
manager->default_quantum = default_quantum;
INIT_LIST(manager->tasks);
- INIT_LIST(manager->ready_tasks);
- INIT_LIST(manager->ready_priority_tasks);
+ INIT_LIST(manager->ready_tasks[isc_taskqueue_normal]);
+ INIT_LIST(manager->ready_tasks[isc_taskqueue_slow]);
+ INIT_LIST(manager->ready_priority_tasks[isc_taskqueue_normal]);
+ INIT_LIST(manager->ready_priority_tasks[isc_taskqueue_slow]);
manager->tasks_running = 0;
manager->tasks_ready = 0;
manager->exclusive_requested = false;
@@ -1465,7 +1512,19 @@ isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
* Start workers.
*/
for (i = 0; i < workers; i++) {
- if (isc_thread_create(run, manager,
+ if (isc_thread_create(run_normal, manager,
+ &manager->threads[manager->workers]) ==
+ ISC_R_SUCCESS) {
+ char name[21]; /* thread name limit on Linux */
+ snprintf(name, sizeof(name), "isc-worker%04u", i);
+ isc_thread_setname(manager->threads[manager->workers],
+ name);
+ manager->workers++;
+ started++;
+ }
+ }
+ for (; i < workers * 2; i++) {
+ if (isc_thread_create(run_slow, manager,
&manager->threads[manager->workers]) ==
ISC_R_SUCCESS) {
char name[21]; /* thread name limit on Linux */
@@ -1482,7 +1541,7 @@ isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
manager_free(manager);
return (ISC_R_NOTHREADS);
}
- isc_thread_setconcurrency(workers);
+ isc_thread_setconcurrency(workers * 2);
#endif /* USE_WORKER_THREADS */
#ifdef USE_SHARED_MANAGER
manager->refs = 1;
@@ -1497,7 +1556,8 @@ isc__taskmgr_create(isc_mem_t *mctx, unsigned int workers,
cleanup_exclusivegranted:
(void)isc_condition_destroy(&manager->exclusive_granted);
cleanup_workavailable:
- (void)isc_condition_destroy(&manager->work_available);
+ (void)isc_condition_destroy(&manager->work_available[isc_taskqueue_slow]);
+ (void)isc_condition_destroy(&manager->work_available[isc_taskqueue_normal]);
cleanup_threads:
isc_mem_free(mctx, manager->threads);
cleanup_lock:
@@ -1582,7 +1642,7 @@ isc__taskmgr_destroy(isc_taskmgr_t **managerp) {
task = NEXT(task, link)) {
LOCK(&task->lock);
if (task_shutdown(task))
- push_readyq(manager, task);
+ push_readyq(manager, task, task->qid);
UNLOCK(&task->lock);
}
#ifdef USE_WORKER_THREADS
@@ -1591,7 +1651,8 @@ isc__taskmgr_destroy(isc_taskmgr_t **managerp) {
* there's work left to do, and if there are already no tasks left
* it will cause the workers to see manager->exiting.
*/
- BROADCAST(&manager->work_available);
+ BROADCAST(&manager->work_available[isc_taskqueue_normal]);
+ BROADCAST(&manager->work_available[isc_taskqueue_slow]);
UNLOCK(&manager->lock);
/*
@@ -1693,7 +1754,8 @@ isc__taskmgr_resume(isc_taskmgr_t *manager0) {
LOCK(&manager->lock);
if (manager->pause_requested) {
manager->pause_requested = false;
- BROADCAST(&manager->work_available);
+ BROADCAST(&manager->work_available[isc_taskqueue_normal]);
+ BROADCAST(&manager->work_available[isc_taskqueue_slow]);
}
UNLOCK(&manager->lock);
}
@@ -1778,7 +1840,8 @@ isc__task_endexclusive(isc_task_t *task0) {
LOCK(&manager->lock);
REQUIRE(manager->exclusive_requested);
manager->exclusive_requested = false;
- BROADCAST(&manager->work_available);
+ BROADCAST(&manager->work_available[isc_taskqueue_normal]);
+ BROADCAST(&manager->work_available[isc_taskqueue_slow]);
UNLOCK(&manager->lock);
#else
UNUSED(task0);
@@ -1804,10 +1867,10 @@ isc__task_setprivilege(isc_task_t *task0, bool priv) {
LOCK(&manager->lock);
if (priv && ISC_LINK_LINKED(task, ready_link))
- ENQUEUE(manager->ready_priority_tasks, task,
+ ENQUEUE(manager->ready_priority_tasks[task->qid], task,
ready_priority_link);
else if (!priv && ISC_LINK_LINKED(task, ready_priority_link))
- DEQUEUE(manager->ready_priority_tasks, task,
+ DEQUEUE(manager->ready_priority_tasks[task->qid], task,
ready_priority_link);
UNLOCK(&manager->lock);
}
--
2.43.2