1/* Kernel thread helper functions.
2 * Copyright (C) 2004 IBM Corporation, Rusty Russell.
3 *
4 * Creation is done via kthreadd, so that we get a clean environment
5 * even if we're invoked from userspace (think modprobe, hotplug cpu,
6 * etc.).
7 */
8#include <uapi/linux/sched/types.h>
9#include <linux/sched.h>
10#include <linux/sched/task.h>
11#include <linux/kthread.h>
12#include <linux/completion.h>
13#include <linux/err.h>
14#include <linux/cpuset.h>
15#include <linux/unistd.h>
16#include <linux/file.h>
17#include <linux/export.h>
18#include <linux/mutex.h>
19#include <linux/slab.h>
20#include <linux/freezer.h>
21#include <linux/ptrace.h>
22#include <linux/uaccess.h>
23#include <linux/numa.h>
24#include <trace/events/sched.h>
25
26static DEFINE_SPINLOCK(kthread_create_lock);
27static LIST_HEAD(kthread_create_list);
28struct task_struct *kthreadd_task;
29
30struct kthread_create_info
31{
32 /* Information passed to kthread() from kthreadd. */
33 int (*threadfn)(void *data);
34 void *data;
35 int node;
36
37 /* Result passed back to kthread_create() from kthreadd. */
38 struct task_struct *result;
39 struct completion *done;
40
41 struct list_head list;
42};
43
44struct kthread {
45 unsigned long flags;
46 unsigned int cpu;
47 void *data;
48 struct completion parked;
49 struct completion exited;
50#ifdef CONFIG_BLK_CGROUP
51 struct cgroup_subsys_state *blkcg_css;
52#endif
53};
54
55enum KTHREAD_BITS {
56 KTHREAD_IS_PER_CPU = 0,
57 KTHREAD_SHOULD_STOP,
58 KTHREAD_SHOULD_PARK,
59};
60
61static inline void set_kthread_struct(void *kthread)
62{
63 /*
64 * We abuse ->set_child_tid to avoid the new member and because it
65 * can't be wrongly copied by copy_process(). We also rely on fact
66 * that the caller can't exec, so PF_KTHREAD can't be cleared.
67 */
68 current->set_child_tid = (__force void __user *)kthread;
69}
70
71static inline struct kthread *to_kthread(struct task_struct *k)
72{
73 WARN_ON(!(k->flags & PF_KTHREAD));
74 return (__force void *)k->set_child_tid;
75}
76
77void free_kthread_struct(struct task_struct *k)
78{
79 struct kthread *kthread;
80
81 /*
82 * Can be NULL if this kthread was created by kernel_thread()
83 * or if kmalloc() in kthread() failed.
84 */
85 kthread = to_kthread(k);
86#ifdef CONFIG_BLK_CGROUP
87 WARN_ON_ONCE(kthread && kthread->blkcg_css);
88#endif
89 kfree(kthread);
90}
91
92/**
93 * kthread_should_stop - should this kthread return now?
94 *
95 * When someone calls kthread_stop() on your kthread, it will be woken
96 * and this will return true. You should then return, and your return
97 * value will be passed through to kthread_stop().
98 */
99bool kthread_should_stop(void)
100{
101 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
102}
103EXPORT_SYMBOL(kthread_should_stop);
104
105bool __kthread_should_park(struct task_struct *k)
106{
107 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
108}
109EXPORT_SYMBOL_GPL(__kthread_should_park);
110
111/**
112 * kthread_should_park - should this kthread park now?
113 *
114 * When someone calls kthread_park() on your kthread, it will be woken
115 * and this will return true. You should then do the necessary
116 * cleanup and call kthread_parkme()
117 *
118 * Similar to kthread_should_stop(), but this keeps the thread alive
119 * and in a park position. kthread_unpark() "restarts" the thread and
120 * calls the thread function again.
121 */
122bool kthread_should_park(void)
123{
124 return __kthread_should_park(current);
125}
126EXPORT_SYMBOL_GPL(kthread_should_park);
127
128/**
129 * kthread_freezable_should_stop - should this freezable kthread return now?
130 * @was_frozen: optional out parameter, indicates whether %current was frozen
131 *
132 * kthread_should_stop() for freezable kthreads, which will enter
133 * refrigerator if necessary. This function is safe from kthread_stop() /
134 * freezer deadlock and freezable kthreads should use this function instead
135 * of calling try_to_freeze() directly.
136 */
137bool kthread_freezable_should_stop(bool *was_frozen)
138{
139 bool frozen = false;
140
141 might_sleep();
142
143 if (unlikely(freezing(current)))
144 frozen = __refrigerator(true);
145
146 if (was_frozen)
147 *was_frozen = frozen;
148
149 return kthread_should_stop();
150}
151EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
152
153/**
154 * kthread_data - return data value specified on kthread creation
155 * @task: kthread task in question
156 *
157 * Return the data value specified when kthread @task was created.
158 * The caller is responsible for ensuring the validity of @task when
159 * calling this function.
160 */
161void *kthread_data(struct task_struct *task)
162{
163 return to_kthread(task)->data;
164}
165
166/**
167 * kthread_probe_data - speculative version of kthread_data()
168 * @task: possible kthread task in question
169 *
170 * @task could be a kthread task. Return the data value specified when it
171 * was created if accessible. If @task isn't a kthread task or its data is
172 * inaccessible for any reason, %NULL is returned. This function requires
173 * that @task itself is safe to dereference.
174 */
175void *kthread_probe_data(struct task_struct *task)
176{
177 struct kthread *kthread = to_kthread(task);
178 void *data = NULL;
179
180 probe_kernel_read(&data, &kthread->data, sizeof(data));
181 return data;
182}
183
184static void __kthread_parkme(struct kthread *self)
185{
186 for (;;) {
187 /*
188 * TASK_PARKED is a special state; we must serialize against
189 * possible pending wakeups to avoid store-store collisions on
190 * task->state.
191 *
192 * Such a collision might possibly result in the task state
193 * changin from TASK_PARKED and us failing the
194 * wait_task_inactive() in kthread_park().
195 */
196 set_special_state(TASK_PARKED);
197 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
198 break;
199
200 complete(&self->parked);
201 schedule();
202 }
203 __set_current_state(TASK_RUNNING);
204}
205
206void kthread_parkme(void)
207{
208 __kthread_parkme(to_kthread(current));
209}
210EXPORT_SYMBOL_GPL(kthread_parkme);
211
212static int kthread(void *_create)
213{
214 /* Copy data: it's on kthread's stack */
215 struct kthread_create_info *create = _create;
216 int (*threadfn)(void *data) = create->threadfn;
217 void *data = create->data;
218 struct completion *done;
219 struct kthread *self;
220 int ret;
221
222 self = kzalloc(sizeof(*self), GFP_KERNEL);
223 set_kthread_struct(self);
224
225 /* If user was SIGKILLed, I release the structure. */
226 done = xchg(&create->done, NULL);
227 if (!done) {
228 kfree(create);
229 do_exit(-EINTR);
230 }
231
232 if (!self) {
233 create->result = ERR_PTR(-ENOMEM);
234 complete(done);
235 do_exit(-ENOMEM);
236 }
237
238 self->data = data;
239 init_completion(&self->exited);
240 init_completion(&self->parked);
241 current->vfork_done = &self->exited;
242
243 /* OK, tell user we're spawned, wait for stop or wakeup */
244 __set_current_state(TASK_UNINTERRUPTIBLE);
245 create->result = current;
246 complete(done);
247 schedule();
248
249 ret = -EINTR;
250 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
251 cgroup_kthread_ready();
252 __kthread_parkme(self);
253 ret = threadfn(data);
254 }
255 do_exit(ret);
256}
257
258/* called from do_fork() to get node information for about to be created task */
259int tsk_fork_get_node(struct task_struct *tsk)
260{
261#ifdef CONFIG_NUMA
262 if (tsk == kthreadd_task)
263 return tsk->pref_node_fork;
264#endif
265 return NUMA_NO_NODE;
266}
267
268static void create_kthread(struct kthread_create_info *create)
269{
270 int pid;
271
272#ifdef CONFIG_NUMA
273 current->pref_node_fork = create->node;
274#endif
275 /* We want our own signal handler (we take no signals by default). */
276 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
277 if (pid < 0) {
278 /* If user was SIGKILLed, I release the structure. */
279 struct completion *done = xchg(&create->done, NULL);
280
281 if (!done) {
282 kfree(create);
283 return;
284 }
285 create->result = ERR_PTR(pid);
286 complete(done);
287 }
288}
289
290static __printf(4, 0)
291struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
292 void *data, int node,
293 const char namefmt[],
294 va_list args)
295{
296 DECLARE_COMPLETION_ONSTACK(done);
297 struct task_struct *task;
298 struct kthread_create_info *create = kmalloc(sizeof(*create),
299 GFP_KERNEL);
300
301 if (!create)
302 return ERR_PTR(-ENOMEM);
303 create->threadfn = threadfn;
304 create->data = data;
305 create->node = node;
306 create->done = &done;
307
308 spin_lock(&kthread_create_lock);
309 list_add_tail(&create->list, &kthread_create_list);
310 spin_unlock(&kthread_create_lock);
311
312 wake_up_process(kthreadd_task);
313 /*
314 * Wait for completion in killable state, for I might be chosen by
315 * the OOM killer while kthreadd is trying to allocate memory for
316 * new kernel thread.
317 */
318 if (unlikely(wait_for_completion_killable(&done))) {
319 /*
320 * If I was SIGKILLed before kthreadd (or new kernel thread)
321 * calls complete(), leave the cleanup of this structure to
322 * that thread.
323 */
324 if (xchg(&create->done, NULL))
325 return ERR_PTR(-EINTR);
326 /*
327 * kthreadd (or new kernel thread) will call complete()
328 * shortly.
329 */
330 wait_for_completion(&done);
331 }
332 task = create->result;
333 if (!IS_ERR(task)) {
334 static const struct sched_param param = { .sched_priority = 0 };
335 char name[TASK_COMM_LEN];
336
337 /*
338 * task is already visible to other tasks, so updating
339 * COMM must be protected.
340 */
341 vsnprintf(name, sizeof(name), namefmt, args);
342 set_task_comm(task, name);
343 /*
344 * root may have changed our (kthreadd's) priority or CPU mask.
345 * The kernel thread should not inherit these properties.
346 */
347 sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
348 set_cpus_allowed_ptr(task, cpu_all_mask);
349 }
350 kfree(create);
351 return task;
352}
353
354/**
355 * kthread_create_on_node - create a kthread.
356 * @threadfn: the function to run until signal_pending(current).
357 * @data: data ptr for @threadfn.
358 * @node: task and thread structures for the thread are allocated on this node
359 * @namefmt: printf-style name for the thread.
360 *
361 * Description: This helper function creates and names a kernel
362 * thread. The thread will be stopped: use wake_up_process() to start
363 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
364 * is affine to all CPUs.
365 *
366 * If thread is going to be bound on a particular cpu, give its node
367 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
368 * When woken, the thread will run @threadfn() with @data as its
369 * argument. @threadfn() can either call do_exit() directly if it is a
370 * standalone thread for which no one will call kthread_stop(), or
371 * return when 'kthread_should_stop()' is true (which means
372 * kthread_stop() has been called). The return value should be zero
373 * or a negative error number; it will be passed to kthread_stop().
374 *
375 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
376 */
377struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
378 void *data, int node,
379 const char namefmt[],
380 ...)
381{
382 struct task_struct *task;
383 va_list args;
384
385 va_start(args, namefmt);
386 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
387 va_end(args);
388
389 return task;
390}
391EXPORT_SYMBOL(kthread_create_on_node);
392
393static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
394{
395 unsigned long flags;
396
397 if (!wait_task_inactive(p, state)) {
398 WARN_ON(1);
399 return;
400 }
401
402 /* It's safe because the task is inactive. */
403 raw_spin_lock_irqsave(&p->pi_lock, flags);
404 do_set_cpus_allowed(p, mask);
405 p->flags |= PF_NO_SETAFFINITY;
406 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
407}
408
409static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
410{
411 __kthread_bind_mask(p, cpumask_of(cpu), state);
412}
413
414void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
415{
416 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
417}
418
419/**
420 * kthread_bind - bind a just-created kthread to a cpu.
421 * @p: thread created by kthread_create().
422 * @cpu: cpu (might not be online, must be possible) for @k to run on.
423 *
424 * Description: This function is equivalent to set_cpus_allowed(),
425 * except that @cpu doesn't need to be online, and the thread must be
426 * stopped (i.e., just returned from kthread_create()).
427 */
428void kthread_bind(struct task_struct *p, unsigned int cpu)
429{
430 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
431}
432EXPORT_SYMBOL(kthread_bind);
433
434/**
435 * kthread_create_on_cpu - Create a cpu bound kthread
436 * @threadfn: the function to run until signal_pending(current).
437 * @data: data ptr for @threadfn.
438 * @cpu: The cpu on which the thread should be bound,
439 * @namefmt: printf-style name for the thread. Format is restricted
440 * to "name.*%u". Code fills in cpu number.
441 *
442 * Description: This helper function creates and names a kernel thread
443 * The thread will be woken and put into park mode.
444 */
445struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
446 void *data, unsigned int cpu,
447 const char *namefmt)
448{
449 struct task_struct *p;
450
451 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
452 cpu);
453 if (IS_ERR(p))
454 return p;
455 kthread_bind(p, cpu);
456 /* CPU hotplug need to bind once again when unparking the thread. */
457 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
458 to_kthread(p)->cpu = cpu;
459 return p;
460}
461
462/**
463 * kthread_unpark - unpark a thread created by kthread_create().
464 * @k: thread created by kthread_create().
465 *
466 * Sets kthread_should_park() for @k to return false, wakes it, and
467 * waits for it to return. If the thread is marked percpu then its
468 * bound to the cpu again.
469 */
470void kthread_unpark(struct task_struct *k)
471{
472 struct kthread *kthread = to_kthread(k);
473
474 /*
475 * Newly created kthread was parked when the CPU was offline.
476 * The binding was lost and we need to set it again.
477 */
478 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
479 __kthread_bind(k, kthread->cpu, TASK_PARKED);
480
481 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
482 /*
483 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
484 */
485 wake_up_state(k, TASK_PARKED);
486}
487EXPORT_SYMBOL_GPL(kthread_unpark);
488
489/**
490 * kthread_park - park a thread created by kthread_create().
491 * @k: thread created by kthread_create().
492 *
493 * Sets kthread_should_park() for @k to return true, wakes it, and
494 * waits for it to return. This can also be called after kthread_create()
495 * instead of calling wake_up_process(): the thread will park without
496 * calling threadfn().
497 *
498 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
499 * If called by the kthread itself just the park bit is set.
500 */
501int kthread_park(struct task_struct *k)
502{
503 struct kthread *kthread = to_kthread(k);
504
505 if (WARN_ON(k->flags & PF_EXITING))
506 return -ENOSYS;
507
508 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
509 return -EBUSY;
510
511 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
512 if (k != current) {
513 wake_up_process(k);
514 /*
515 * Wait for __kthread_parkme() to complete(), this means we
516 * _will_ have TASK_PARKED and are about to call schedule().
517 */
518 wait_for_completion(&kthread->parked);
519 /*
520 * Now wait for that schedule() to complete and the task to
521 * get scheduled out.
522 */
523 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
524 }
525
526 return 0;
527}
528EXPORT_SYMBOL_GPL(kthread_park);
529
530/**
531 * kthread_stop - stop a thread created by kthread_create().
532 * @k: thread created by kthread_create().
533 *
534 * Sets kthread_should_stop() for @k to return true, wakes it, and
535 * waits for it to exit. This can also be called after kthread_create()
536 * instead of calling wake_up_process(): the thread will exit without
537 * calling threadfn().
538 *
539 * If threadfn() may call do_exit() itself, the caller must ensure
540 * task_struct can't go away.
541 *
542 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
543 * was never called.
544 */
545int kthread_stop(struct task_struct *k)
546{
547 struct kthread *kthread;
548 int ret;
549
550 trace_sched_kthread_stop(k);
551
552 get_task_struct(k);
553 kthread = to_kthread(k);
554 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
555 kthread_unpark(k);
556 wake_up_process(k);
557 wait_for_completion(&kthread->exited);
558 ret = k->exit_code;
559 put_task_struct(k);
560
561 trace_sched_kthread_stop_ret(ret);
562 return ret;
563}
564EXPORT_SYMBOL(kthread_stop);
565
566int kthreadd(void *unused)
567{
568 struct task_struct *tsk = current;
569
570 /* Setup a clean context for our children to inherit. */
571 set_task_comm(tsk, "kthreadd");
572 ignore_signals(tsk);
573 set_cpus_allowed_ptr(tsk, cpu_all_mask);
574 set_mems_allowed(node_states[N_MEMORY]);
575
576 current->flags |= PF_NOFREEZE;
577 cgroup_init_kthreadd();
578
579 for (;;) {
580 set_current_state(TASK_INTERRUPTIBLE);
581 if (list_empty(&kthread_create_list))
582 schedule();
583 __set_current_state(TASK_RUNNING);
584
585 spin_lock(&kthread_create_lock);
586 while (!list_empty(&kthread_create_list)) {
587 struct kthread_create_info *create;
588
589 create = list_entry(kthread_create_list.next,
590 struct kthread_create_info, list);
591 list_del_init(&create->list);
592 spin_unlock(&kthread_create_lock);
593
594 create_kthread(create);
595
596 spin_lock(&kthread_create_lock);
597 }
598 spin_unlock(&kthread_create_lock);
599 }
600
601 return 0;
602}
603
604void __kthread_init_worker(struct kthread_worker *worker,
605 const char *name,
606 struct lock_class_key *key)
607{
608 memset(worker, 0, sizeof(struct kthread_worker));
609 raw_spin_lock_init(&worker->lock);
610 lockdep_set_class_and_name(&worker->lock, key, name);
611 INIT_LIST_HEAD(&worker->work_list);
612 INIT_LIST_HEAD(&worker->delayed_work_list);
613}
614EXPORT_SYMBOL_GPL(__kthread_init_worker);
615
616/**
617 * kthread_worker_fn - kthread function to process kthread_worker
618 * @worker_ptr: pointer to initialized kthread_worker
619 *
620 * This function implements the main cycle of kthread worker. It processes
621 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
622 * is empty.
623 *
624 * The works are not allowed to keep any locks, disable preemption or interrupts
625 * when they finish. There is defined a safe point for freezing when one work
626 * finishes and before a new one is started.
627 *
628 * Also the works must not be handled by more than one worker at the same time,
629 * see also kthread_queue_work().
630 */
631int kthread_worker_fn(void *worker_ptr)
632{
633 struct kthread_worker *worker = worker_ptr;
634 struct kthread_work *work;
635
636 /*
637 * FIXME: Update the check and remove the assignment when all kthread
638 * worker users are created using kthread_create_worker*() functions.
639 */
640 WARN_ON(worker->task && worker->task != current);
641 worker->task = current;
642
643 if (worker->flags & KTW_FREEZABLE)
644 set_freezable();
645
646repeat:
647 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
648
649 if (kthread_should_stop()) {
650 __set_current_state(TASK_RUNNING);
651 raw_spin_lock_irq(&worker->lock);
652 worker->task = NULL;
653 raw_spin_unlock_irq(&worker->lock);
654 return 0;
655 }
656
657 work = NULL;
658 raw_spin_lock_irq(&worker->lock);
659 if (!list_empty(&worker->work_list)) {
660 work = list_first_entry(&worker->work_list,
661 struct kthread_work, node);
662 list_del_init(&work->node);
663 }
664 worker->current_work = work;
665 raw_spin_unlock_irq(&worker->lock);
666
667 if (work) {
668 __set_current_state(TASK_RUNNING);
669 work->func(work);
670 } else if (!freezing(current))
671 schedule();
672
673 try_to_freeze();
674 cond_resched();
675 goto repeat;
676}
677EXPORT_SYMBOL_GPL(kthread_worker_fn);
678
679static __printf(3, 0) struct kthread_worker *
680__kthread_create_worker(int cpu, unsigned int flags,
681 const char namefmt[], va_list args)
682{
683 struct kthread_worker *worker;
684 struct task_struct *task;
685 int node = NUMA_NO_NODE;
686
687 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
688 if (!worker)
689 return ERR_PTR(-ENOMEM);
690
691 kthread_init_worker(worker);
692
693 if (cpu >= 0)
694 node = cpu_to_node(cpu);
695
696 task = __kthread_create_on_node(kthread_worker_fn, worker,
697 node, namefmt, args);
698 if (IS_ERR(task))
699 goto fail_task;
700
701 if (cpu >= 0)
702 kthread_bind(task, cpu);
703
704 worker->flags = flags;
705 worker->task = task;
706 wake_up_process(task);
707 return worker;
708
709fail_task:
710 kfree(worker);
711 return ERR_CAST(task);
712}
713
714/**
715 * kthread_create_worker - create a kthread worker
716 * @flags: flags modifying the default behavior of the worker
717 * @namefmt: printf-style name for the kthread worker (task).
718 *
719 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
720 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
721 * when the worker was SIGKILLed.
722 */
723struct kthread_worker *
724kthread_create_worker(unsigned int flags, const char namefmt[], ...)
725{
726 struct kthread_worker *worker;
727 va_list args;
728
729 va_start(args, namefmt);
730 worker = __kthread_create_worker(-1, flags, namefmt, args);
731 va_end(args);
732
733 return worker;
734}
735EXPORT_SYMBOL(kthread_create_worker);
736
737/**
738 * kthread_create_worker_on_cpu - create a kthread worker and bind it
739 * it to a given CPU and the associated NUMA node.
740 * @cpu: CPU number
741 * @flags: flags modifying the default behavior of the worker
742 * @namefmt: printf-style name for the kthread worker (task).
743 *
744 * Use a valid CPU number if you want to bind the kthread worker
745 * to the given CPU and the associated NUMA node.
746 *
747 * A good practice is to add the cpu number also into the worker name.
748 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
749 *
750 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
751 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
752 * when the worker was SIGKILLed.
753 */
754struct kthread_worker *
755kthread_create_worker_on_cpu(int cpu, unsigned int flags,
756 const char namefmt[], ...)
757{
758 struct kthread_worker *worker;
759 va_list args;
760
761 va_start(args, namefmt);
762 worker = __kthread_create_worker(cpu, flags, namefmt, args);
763 va_end(args);
764
765 return worker;
766}
767EXPORT_SYMBOL(kthread_create_worker_on_cpu);
768
769/*
770 * Returns true when the work could not be queued at the moment.
771 * It happens when it is already pending in a worker list
772 * or when it is being cancelled.
773 */
774static inline bool queuing_blocked(struct kthread_worker *worker,
775 struct kthread_work *work)
776{
777 lockdep_assert_held(&worker->lock);
778
779 return !list_empty(&work->node) || work->canceling;
780}
781
782static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
783 struct kthread_work *work)
784{
785 lockdep_assert_held(&worker->lock);
786 WARN_ON_ONCE(!list_empty(&work->node));
787 /* Do not use a work with >1 worker, see kthread_queue_work() */
788 WARN_ON_ONCE(work->worker && work->worker != worker);
789}
790
791/* insert @work before @pos in @worker */
792static void kthread_insert_work(struct kthread_worker *worker,
793 struct kthread_work *work,
794 struct list_head *pos)
795{
796 kthread_insert_work_sanity_check(worker, work);
797
798 list_add_tail(&work->node, pos);
799 work->worker = worker;
800 if (!worker->current_work && likely(worker->task))
801 wake_up_process(worker->task);
802}
803
804/**
805 * kthread_queue_work - queue a kthread_work
806 * @worker: target kthread_worker
807 * @work: kthread_work to queue
808 *
809 * Queue @work to work processor @task for async execution. @task
810 * must have been created with kthread_worker_create(). Returns %true
811 * if @work was successfully queued, %false if it was already pending.
812 *
813 * Reinitialize the work if it needs to be used by another worker.
814 * For example, when the worker was stopped and started again.
815 */
816bool kthread_queue_work(struct kthread_worker *worker,
817 struct kthread_work *work)
818{
819 bool ret = false;
820 unsigned long flags;
821
822 raw_spin_lock_irqsave(&worker->lock, flags);
823 if (!queuing_blocked(worker, work)) {
824 kthread_insert_work(worker, work, &worker->work_list);
825 ret = true;
826 }
827 raw_spin_unlock_irqrestore(&worker->lock, flags);
828 return ret;
829}
830EXPORT_SYMBOL_GPL(kthread_queue_work);
831
832/**
833 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
834 * delayed work when the timer expires.
835 * @t: pointer to the expired timer
836 *
837 * The format of the function is defined by struct timer_list.
838 * It should have been called from irqsafe timer with irq already off.
839 */
840void kthread_delayed_work_timer_fn(struct timer_list *t)
841{
842 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
843 struct kthread_work *work = &dwork->work;
844 struct kthread_worker *worker = work->worker;
845 unsigned long flags;
846
847 /*
848 * This might happen when a pending work is reinitialized.
849 * It means that it is used a wrong way.
850 */
851 if (WARN_ON_ONCE(!worker))
852 return;
853
854 raw_spin_lock_irqsave(&worker->lock, flags);
855 /* Work must not be used with >1 worker, see kthread_queue_work(). */
856 WARN_ON_ONCE(work->worker != worker);
857
858 /* Move the work from worker->delayed_work_list. */
859 WARN_ON_ONCE(list_empty(&work->node));
860 list_del_init(&work->node);
861 kthread_insert_work(worker, work, &worker->work_list);
862
863 raw_spin_unlock_irqrestore(&worker->lock, flags);
864}
865EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
866
867void __kthread_queue_delayed_work(struct kthread_worker *worker,
868 struct kthread_delayed_work *dwork,
869 unsigned long delay)
870{
871 struct timer_list *timer = &dwork->timer;
872 struct kthread_work *work = &dwork->work;
873
874 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
875
876 /*
877 * If @delay is 0, queue @dwork->work immediately. This is for
878 * both optimization and correctness. The earliest @timer can
879 * expire is on the closest next tick and delayed_work users depend
880 * on that there's no such delay when @delay is 0.
881 */
882 if (!delay) {
883 kthread_insert_work(worker, work, &worker->work_list);
884 return;
885 }
886
887 /* Be paranoid and try to detect possible races already now. */
888 kthread_insert_work_sanity_check(worker, work);
889
890 list_add(&work->node, &worker->delayed_work_list);
891 work->worker = worker;
892 timer->expires = jiffies + delay;
893 add_timer(timer);
894}
895
896/**
897 * kthread_queue_delayed_work - queue the associated kthread work
898 * after a delay.
899 * @worker: target kthread_worker
900 * @dwork: kthread_delayed_work to queue
901 * @delay: number of jiffies to wait before queuing
902 *
903 * If the work has not been pending it starts a timer that will queue
904 * the work after the given @delay. If @delay is zero, it queues the
905 * work immediately.
906 *
907 * Return: %false if the @work has already been pending. It means that
908 * either the timer was running or the work was queued. It returns %true
909 * otherwise.
910 */
911bool kthread_queue_delayed_work(struct kthread_worker *worker,
912 struct kthread_delayed_work *dwork,
913 unsigned long delay)
914{
915 struct kthread_work *work = &dwork->work;
916 unsigned long flags;
917 bool ret = false;
918
919 raw_spin_lock_irqsave(&worker->lock, flags);
920
921 if (!queuing_blocked(worker, work)) {
922 __kthread_queue_delayed_work(worker, dwork, delay);
923 ret = true;
924 }
925
926 raw_spin_unlock_irqrestore(&worker->lock, flags);
927 return ret;
928}
929EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
930
931struct kthread_flush_work {
932 struct kthread_work work;
933 struct completion done;
934};
935
936static void kthread_flush_work_fn(struct kthread_work *work)
937{
938 struct kthread_flush_work *fwork =
939 container_of(work, struct kthread_flush_work, work);
940 complete(&fwork->done);
941}
942
943/**
944 * kthread_flush_work - flush a kthread_work
945 * @work: work to flush
946 *
947 * If @work is queued or executing, wait for it to finish execution.
948 */
949void kthread_flush_work(struct kthread_work *work)
950{
951 struct kthread_flush_work fwork = {
952 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
953 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
954 };
955 struct kthread_worker *worker;
956 bool noop = false;
957
958 worker = work->worker;
959 if (!worker)
960 return;
961
962 raw_spin_lock_irq(&worker->lock);
963 /* Work must not be used with >1 worker, see kthread_queue_work(). */
964 WARN_ON_ONCE(work->worker != worker);
965
966 if (!list_empty(&work->node))
967 kthread_insert_work(worker, &fwork.work, work->node.next);
968 else if (worker->current_work == work)
969 kthread_insert_work(worker, &fwork.work,
970 worker->work_list.next);
971 else
972 noop = true;
973
974 raw_spin_unlock_irq(&worker->lock);
975
976 if (!noop)
977 wait_for_completion(&fwork.done);
978}
979EXPORT_SYMBOL_GPL(kthread_flush_work);
980
981/*
982 * This function removes the work from the worker queue. Also it makes sure
983 * that it won't get queued later via the delayed work's timer.
984 *
985 * The work might still be in use when this function finishes. See the
986 * current_work proceed by the worker.
987 *
988 * Return: %true if @work was pending and successfully canceled,
989 * %false if @work was not pending
990 */
991static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
992 unsigned long *flags)
993{
994 /* Try to cancel the timer if exists. */
995 if (is_dwork) {
996 struct kthread_delayed_work *dwork =
997 container_of(work, struct kthread_delayed_work, work);
998 struct kthread_worker *worker = work->worker;
999
1000 /*
1001 * del_timer_sync() must be called to make sure that the timer
1002 * callback is not running. The lock must be temporary released
1003 * to avoid a deadlock with the callback. In the meantime,
1004 * any queuing is blocked by setting the canceling counter.
1005 */
1006 work->canceling++;
1007 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1008 del_timer_sync(&dwork->timer);
1009 raw_spin_lock_irqsave(&worker->lock, *flags);
1010 work->canceling--;
1011 }
1012
1013 /*
1014 * Try to remove the work from a worker list. It might either
1015 * be from worker->work_list or from worker->delayed_work_list.
1016 */
1017 if (!list_empty(&work->node)) {
1018 list_del_init(&work->node);
1019 return true;
1020 }
1021
1022 return false;
1023}
1024
1025/**
1026 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1027 * @worker: kthread worker to use
1028 * @dwork: kthread delayed work to queue
1029 * @delay: number of jiffies to wait before queuing
1030 *
1031 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1032 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1033 * @work is guaranteed to be queued immediately.
1034 *
1035 * Return: %true if @dwork was pending and its timer was modified,
1036 * %false otherwise.
1037 *
1038 * A special case is when the work is being canceled in parallel.
1039 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1040 * or yet another kthread_mod_delayed_work() call. We let the other command
1041 * win and return %false here. The caller is supposed to synchronize these
1042 * operations a reasonable way.
1043 *
1044 * This function is safe to call from any context including IRQ handler.
1045 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1046 * for details.
1047 */
1048bool kthread_mod_delayed_work(struct kthread_worker *worker,
1049 struct kthread_delayed_work *dwork,
1050 unsigned long delay)
1051{
1052 struct kthread_work *work = &dwork->work;
1053 unsigned long flags;
1054 int ret = false;
1055
1056 raw_spin_lock_irqsave(&worker->lock, flags);
1057
1058 /* Do not bother with canceling when never queued. */
1059 if (!work->worker)
1060 goto fast_queue;
1061
1062 /* Work must not be used with >1 worker, see kthread_queue_work() */
1063 WARN_ON_ONCE(work->worker != worker);
1064
1065 /* Do not fight with another command that is canceling this work. */
1066 if (work->canceling)
1067 goto out;
1068
1069 ret = __kthread_cancel_work(work, true, &flags);
1070fast_queue:
1071 __kthread_queue_delayed_work(worker, dwork, delay);
1072out:
1073 raw_spin_unlock_irqrestore(&worker->lock, flags);
1074 return ret;
1075}
1076EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1077
1078static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1079{
1080 struct kthread_worker *worker = work->worker;
1081 unsigned long flags;
1082 int ret = false;
1083
1084 if (!worker)
1085 goto out;
1086
1087 raw_spin_lock_irqsave(&worker->lock, flags);
1088 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1089 WARN_ON_ONCE(work->worker != worker);
1090
1091 ret = __kthread_cancel_work(work, is_dwork, &flags);
1092
1093 if (worker->current_work != work)
1094 goto out_fast;
1095
1096 /*
1097 * The work is in progress and we need to wait with the lock released.
1098 * In the meantime, block any queuing by setting the canceling counter.
1099 */
1100 work->canceling++;
1101 raw_spin_unlock_irqrestore(&worker->lock, flags);
1102 kthread_flush_work(work);
1103 raw_spin_lock_irqsave(&worker->lock, flags);
1104 work->canceling--;
1105
1106out_fast:
1107 raw_spin_unlock_irqrestore(&worker->lock, flags);
1108out:
1109 return ret;
1110}
1111
1112/**
1113 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1114 * @work: the kthread work to cancel
1115 *
1116 * Cancel @work and wait for its execution to finish. This function
1117 * can be used even if the work re-queues itself. On return from this
1118 * function, @work is guaranteed to be not pending or executing on any CPU.
1119 *
1120 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1121 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1122 *
1123 * The caller must ensure that the worker on which @work was last
1124 * queued can't be destroyed before this function returns.
1125 *
1126 * Return: %true if @work was pending, %false otherwise.
1127 */
1128bool kthread_cancel_work_sync(struct kthread_work *work)
1129{
1130 return __kthread_cancel_work_sync(work, false);
1131}
1132EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1133
1134/**
1135 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1136 * wait for it to finish.
1137 * @dwork: the kthread delayed work to cancel
1138 *
1139 * This is kthread_cancel_work_sync() for delayed works.
1140 *
1141 * Return: %true if @dwork was pending, %false otherwise.
1142 */
1143bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1144{
1145 return __kthread_cancel_work_sync(&dwork->work, true);
1146}
1147EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1148
1149/**
1150 * kthread_flush_worker - flush all current works on a kthread_worker
1151 * @worker: worker to flush
1152 *
1153 * Wait until all currently executing or pending works on @worker are
1154 * finished.
1155 */
1156void kthread_flush_worker(struct kthread_worker *worker)
1157{
1158 struct kthread_flush_work fwork = {
1159 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1160 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1161 };
1162
1163 kthread_queue_work(worker, &fwork.work);
1164 wait_for_completion(&fwork.done);
1165}
1166EXPORT_SYMBOL_GPL(kthread_flush_worker);
1167
1168/**
1169 * kthread_destroy_worker - destroy a kthread worker
1170 * @worker: worker to be destroyed
1171 *
1172 * Flush and destroy @worker. The simple flush is enough because the kthread
1173 * worker API is used only in trivial scenarios. There are no multi-step state
1174 * machines needed.
1175 */
1176void kthread_destroy_worker(struct kthread_worker *worker)
1177{
1178 struct task_struct *task;
1179
1180 task = worker->task;
1181 if (WARN_ON(!task))
1182 return;
1183
1184 kthread_flush_worker(worker);
1185 kthread_stop(task);
1186 WARN_ON(!list_empty(&worker->work_list));
1187 kfree(worker);
1188}
1189EXPORT_SYMBOL(kthread_destroy_worker);
1190
1191#ifdef CONFIG_BLK_CGROUP
1192/**
1193 * kthread_associate_blkcg - associate blkcg to current kthread
1194 * @css: the cgroup info
1195 *
1196 * Current thread must be a kthread. The thread is running jobs on behalf of
1197 * other threads. In some cases, we expect the jobs attach cgroup info of
1198 * original threads instead of that of current thread. This function stores
1199 * original thread's cgroup info in current kthread context for later
1200 * retrieval.
1201 */
1202void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1203{
1204 struct kthread *kthread;
1205
1206 if (!(current->flags & PF_KTHREAD))
1207 return;
1208 kthread = to_kthread(current);
1209 if (!kthread)
1210 return;
1211
1212 if (kthread->blkcg_css) {
1213 css_put(kthread->blkcg_css);
1214 kthread->blkcg_css = NULL;
1215 }
1216 if (css) {
1217 css_get(css);
1218 kthread->blkcg_css = css;
1219 }
1220}
1221EXPORT_SYMBOL(kthread_associate_blkcg);
1222
1223/**
1224 * kthread_blkcg - get associated blkcg css of current kthread
1225 *
1226 * Current thread must be a kthread.
1227 */
1228struct cgroup_subsys_state *kthread_blkcg(void)
1229{
1230 struct kthread *kthread;
1231
1232 if (current->flags & PF_KTHREAD) {
1233 kthread = to_kthread(current);
1234 if (kthread)
1235 return kthread->blkcg_css;
1236 }
1237 return NULL;
1238}
1239EXPORT_SYMBOL(kthread_blkcg);
1240#endif
1241