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.
This commit is contained in:
Ondřej Surý 2024-01-09 19:59:54 +01:00 committed by Michał Kępień
parent 8b7ecba988
commit db083a2172
No known key found for this signature in database
3 changed files with 118 additions and 48 deletions

View file

@ -9261,7 +9261,7 @@ 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;

View file

@ -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 */

View file

@ -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);
}