Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/* Kernel thread helper functions.
3 * Copyright (C) 2004 IBM Corporation, Rusty Russell.
4 *
5 * Creation is done via kthreadd, so that we get a clean environment
6 * even if we're invoked from userspace (think modprobe, hotplug cpu,
7 * etc.).
8 */
9#include <uapi/linux/sched/types.h>
10#include <linux/sched.h>
11#include <linux/sched/task.h>
12#include <linux/kthread.h>
13#include <linux/completion.h>
14#include <linux/err.h>
15#include <linux/cgroup.h>
16#include <linux/cpuset.h>
17#include <linux/unistd.h>
18#include <linux/file.h>
19#include <linux/export.h>
20#include <linux/mutex.h>
21#include <linux/slab.h>
22#include <linux/freezer.h>
23#include <linux/ptrace.h>
24#include <linux/uaccess.h>
25#include <linux/numa.h>
26#include <trace/events/sched.h>
27
28static DEFINE_SPINLOCK(kthread_create_lock);
29static LIST_HEAD(kthread_create_list);
30struct task_struct *kthreadd_task;
31
32struct kthread_create_info
33{
34 /* Information passed to kthread() from kthreadd. */
35 int (*threadfn)(void *data);
36 void *data;
37 int node;
38
39 /* Result passed back to kthread_create() from kthreadd. */
40 struct task_struct *result;
41 struct completion *done;
42
43 struct list_head list;
44};
45
46struct kthread {
47 unsigned long flags;
48 unsigned int cpu;
49 void *data;
50 struct completion parked;
51 struct completion exited;
52#ifdef CONFIG_BLK_CGROUP
53 struct cgroup_subsys_state *blkcg_css;
54#endif
55};
56
57enum KTHREAD_BITS {
58 KTHREAD_IS_PER_CPU = 0,
59 KTHREAD_SHOULD_STOP,
60 KTHREAD_SHOULD_PARK,
61};
62
63static inline void set_kthread_struct(void *kthread)
64{
65 /*
66 * We abuse ->set_child_tid to avoid the new member and because it
67 * can't be wrongly copied by copy_process(). We also rely on fact
68 * that the caller can't exec, so PF_KTHREAD can't be cleared.
69 */
70 current->set_child_tid = (__force void __user *)kthread;
71}
72
73static inline struct kthread *to_kthread(struct task_struct *k)
74{
75 WARN_ON(!(k->flags & PF_KTHREAD));
76 return (__force void *)k->set_child_tid;
77}
78
79void free_kthread_struct(struct task_struct *k)
80{
81 struct kthread *kthread;
82
83 /*
84 * Can be NULL if this kthread was created by kernel_thread()
85 * or if kmalloc() in kthread() failed.
86 */
87 kthread = to_kthread(k);
88#ifdef CONFIG_BLK_CGROUP
89 WARN_ON_ONCE(kthread && kthread->blkcg_css);
90#endif
91 kfree(kthread);
92}
93
94/**
95 * kthread_should_stop - should this kthread return now?
96 *
97 * When someone calls kthread_stop() on your kthread, it will be woken
98 * and this will return true. You should then return, and your return
99 * value will be passed through to kthread_stop().
100 */
101bool kthread_should_stop(void)
102{
103 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
104}
105EXPORT_SYMBOL(kthread_should_stop);
106
107bool __kthread_should_park(struct task_struct *k)
108{
109 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
110}
111EXPORT_SYMBOL_GPL(__kthread_should_park);
112
113/**
114 * kthread_should_park - should this kthread park now?
115 *
116 * When someone calls kthread_park() on your kthread, it will be woken
117 * and this will return true. You should then do the necessary
118 * cleanup and call kthread_parkme()
119 *
120 * Similar to kthread_should_stop(), but this keeps the thread alive
121 * and in a park position. kthread_unpark() "restarts" the thread and
122 * calls the thread function again.
123 */
124bool kthread_should_park(void)
125{
126 return __kthread_should_park(current);
127}
128EXPORT_SYMBOL_GPL(kthread_should_park);
129
130/**
131 * kthread_freezable_should_stop - should this freezable kthread return now?
132 * @was_frozen: optional out parameter, indicates whether %current was frozen
133 *
134 * kthread_should_stop() for freezable kthreads, which will enter
135 * refrigerator if necessary. This function is safe from kthread_stop() /
136 * freezer deadlock and freezable kthreads should use this function instead
137 * of calling try_to_freeze() directly.
138 */
139bool kthread_freezable_should_stop(bool *was_frozen)
140{
141 bool frozen = false;
142
143 might_sleep();
144
145 if (unlikely(freezing(current)))
146 frozen = __refrigerator(true);
147
148 if (was_frozen)
149 *was_frozen = frozen;
150
151 return kthread_should_stop();
152}
153EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
154
155/**
156 * kthread_data - return data value specified on kthread creation
157 * @task: kthread task in question
158 *
159 * Return the data value specified when kthread @task was created.
160 * The caller is responsible for ensuring the validity of @task when
161 * calling this function.
162 */
163void *kthread_data(struct task_struct *task)
164{
165 return to_kthread(task)->data;
166}
167
168/**
169 * kthread_probe_data - speculative version of kthread_data()
170 * @task: possible kthread task in question
171 *
172 * @task could be a kthread task. Return the data value specified when it
173 * was created if accessible. If @task isn't a kthread task or its data is
174 * inaccessible for any reason, %NULL is returned. This function requires
175 * that @task itself is safe to dereference.
176 */
177void *kthread_probe_data(struct task_struct *task)
178{
179 struct kthread *kthread = to_kthread(task);
180 void *data = NULL;
181
182 probe_kernel_read(&data, &kthread->data, sizeof(data));
183 return data;
184}
185
186static void __kthread_parkme(struct kthread *self)
187{
188 for (;;) {
189 /*
190 * TASK_PARKED is a special state; we must serialize against
191 * possible pending wakeups to avoid store-store collisions on
192 * task->state.
193 *
194 * Such a collision might possibly result in the task state
195 * changin from TASK_PARKED and us failing the
196 * wait_task_inactive() in kthread_park().
197 */
198 set_special_state(TASK_PARKED);
199 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
200 break;
201
202 complete(&self->parked);
203 schedule();
204 }
205 __set_current_state(TASK_RUNNING);
206}
207
208void kthread_parkme(void)
209{
210 __kthread_parkme(to_kthread(current));
211}
212EXPORT_SYMBOL_GPL(kthread_parkme);
213
214static int kthread(void *_create)
215{
216 /* Copy data: it's on kthread's stack */
217 struct kthread_create_info *create = _create;
218 int (*threadfn)(void *data) = create->threadfn;
219 void *data = create->data;
220 struct completion *done;
221 struct kthread *self;
222 int ret;
223
224 self = kzalloc(sizeof(*self), GFP_KERNEL);
225 set_kthread_struct(self);
226
227 /* If user was SIGKILLed, I release the structure. */
228 done = xchg(&create->done, NULL);
229 if (!done) {
230 kfree(create);
231 do_exit(-EINTR);
232 }
233
234 if (!self) {
235 create->result = ERR_PTR(-ENOMEM);
236 complete(done);
237 do_exit(-ENOMEM);
238 }
239
240 self->data = data;
241 init_completion(&self->exited);
242 init_completion(&self->parked);
243 current->vfork_done = &self->exited;
244
245 /* OK, tell user we're spawned, wait for stop or wakeup */
246 __set_current_state(TASK_UNINTERRUPTIBLE);
247 create->result = current;
248 complete(done);
249 schedule();
250
251 ret = -EINTR;
252 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
253 cgroup_kthread_ready();
254 __kthread_parkme(self);
255 ret = threadfn(data);
256 }
257 do_exit(ret);
258}
259
260/* called from do_fork() to get node information for about to be created task */
261int tsk_fork_get_node(struct task_struct *tsk)
262{
263#ifdef CONFIG_NUMA
264 if (tsk == kthreadd_task)
265 return tsk->pref_node_fork;
266#endif
267 return NUMA_NO_NODE;
268}
269
270static void create_kthread(struct kthread_create_info *create)
271{
272 int pid;
273
274#ifdef CONFIG_NUMA
275 current->pref_node_fork = create->node;
276#endif
277 /* We want our own signal handler (we take no signals by default). */
278 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
279 if (pid < 0) {
280 /* If user was SIGKILLed, I release the structure. */
281 struct completion *done = xchg(&create->done, NULL);
282
283 if (!done) {
284 kfree(create);
285 return;
286 }
287 create->result = ERR_PTR(pid);
288 complete(done);
289 }
290}
291
292static __printf(4, 0)
293struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
294 void *data, int node,
295 const char namefmt[],
296 va_list args)
297{
298 DECLARE_COMPLETION_ONSTACK(done);
299 struct task_struct *task;
300 struct kthread_create_info *create = kmalloc(sizeof(*create),
301 GFP_KERNEL);
302
303 if (!create)
304 return ERR_PTR(-ENOMEM);
305 create->threadfn = threadfn;
306 create->data = data;
307 create->node = node;
308 create->done = &done;
309
310 spin_lock(&kthread_create_lock);
311 list_add_tail(&create->list, &kthread_create_list);
312 spin_unlock(&kthread_create_lock);
313
314 wake_up_process(kthreadd_task);
315 /*
316 * Wait for completion in killable state, for I might be chosen by
317 * the OOM killer while kthreadd is trying to allocate memory for
318 * new kernel thread.
319 */
320 if (unlikely(wait_for_completion_killable(&done))) {
321 /*
322 * If I was SIGKILLed before kthreadd (or new kernel thread)
323 * calls complete(), leave the cleanup of this structure to
324 * that thread.
325 */
326 if (xchg(&create->done, NULL))
327 return ERR_PTR(-EINTR);
328 /*
329 * kthreadd (or new kernel thread) will call complete()
330 * shortly.
331 */
332 wait_for_completion(&done);
333 }
334 task = create->result;
335 if (!IS_ERR(task)) {
336 static const struct sched_param param = { .sched_priority = 0 };
337 char name[TASK_COMM_LEN];
338
339 /*
340 * task is already visible to other tasks, so updating
341 * COMM must be protected.
342 */
343 vsnprintf(name, sizeof(name), namefmt, args);
344 set_task_comm(task, name);
345 /*
346 * root may have changed our (kthreadd's) priority or CPU mask.
347 * The kernel thread should not inherit these properties.
348 */
349 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
350 set_cpus_allowed_ptr(task, cpu_all_mask);
351 }
352 kfree(create);
353 return task;
354}
355
356/**
357 * kthread_create_on_node - create a kthread.
358 * @threadfn: the function to run until signal_pending(current).
359 * @data: data ptr for @threadfn.
360 * @node: task and thread structures for the thread are allocated on this node
361 * @namefmt: printf-style name for the thread.
362 *
363 * Description: This helper function creates and names a kernel
364 * thread. The thread will be stopped: use wake_up_process() to start
365 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
366 * is affine to all CPUs.
367 *
368 * If thread is going to be bound on a particular cpu, give its node
369 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
370 * When woken, the thread will run @threadfn() with @data as its
371 * argument. @threadfn() can either call do_exit() directly if it is a
372 * standalone thread for which no one will call kthread_stop(), or
373 * return when 'kthread_should_stop()' is true (which means
374 * kthread_stop() has been called). The return value should be zero
375 * or a negative error number; it will be passed to kthread_stop().
376 *
377 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
378 */
379struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
380 void *data, int node,
381 const char namefmt[],
382 ...)
383{
384 struct task_struct *task;
385 va_list args;
386
387 va_start(args, namefmt);
388 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
389 va_end(args);
390
391 return task;
392}
393EXPORT_SYMBOL(kthread_create_on_node);
394
395static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
396{
397 unsigned long flags;
398
399 if (!wait_task_inactive(p, state)) {
400 WARN_ON(1);
401 return;
402 }
403
404 /* It's safe because the task is inactive. */
405 raw_spin_lock_irqsave(&p->pi_lock, flags);
406 do_set_cpus_allowed(p, mask);
407 p->flags |= PF_NO_SETAFFINITY;
408 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
409}
410
411static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
412{
413 __kthread_bind_mask(p, cpumask_of(cpu), state);
414}
415
416void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
417{
418 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
419}
420
421/**
422 * kthread_bind - bind a just-created kthread to a cpu.
423 * @p: thread created by kthread_create().
424 * @cpu: cpu (might not be online, must be possible) for @k to run on.
425 *
426 * Description: This function is equivalent to set_cpus_allowed(),
427 * except that @cpu doesn't need to be online, and the thread must be
428 * stopped (i.e., just returned from kthread_create()).
429 */
430void kthread_bind(struct task_struct *p, unsigned int cpu)
431{
432 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
433}
434EXPORT_SYMBOL(kthread_bind);
435
436/**
437 * kthread_create_on_cpu - Create a cpu bound kthread
438 * @threadfn: the function to run until signal_pending(current).
439 * @data: data ptr for @threadfn.
440 * @cpu: The cpu on which the thread should be bound,
441 * @namefmt: printf-style name for the thread. Format is restricted
442 * to "name.*%u". Code fills in cpu number.
443 *
444 * Description: This helper function creates and names a kernel thread
445 * The thread will be woken and put into park mode.
446 */
447struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
448 void *data, unsigned int cpu,
449 const char *namefmt)
450{
451 struct task_struct *p;
452
453 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
454 cpu);
455 if (IS_ERR(p))
456 return p;
457 kthread_bind(p, cpu);
458 /* CPU hotplug need to bind once again when unparking the thread. */
459 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
460 to_kthread(p)->cpu = cpu;
461 return p;
462}
463
464/**
465 * kthread_unpark - unpark a thread created by kthread_create().
466 * @k: thread created by kthread_create().
467 *
468 * Sets kthread_should_park() for @k to return false, wakes it, and
469 * waits for it to return. If the thread is marked percpu then its
470 * bound to the cpu again.
471 */
472void kthread_unpark(struct task_struct *k)
473{
474 struct kthread *kthread = to_kthread(k);
475
476 /*
477 * Newly created kthread was parked when the CPU was offline.
478 * The binding was lost and we need to set it again.
479 */
480 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
481 __kthread_bind(k, kthread->cpu, TASK_PARKED);
482
483 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
484 /*
485 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
486 */
487 wake_up_state(k, TASK_PARKED);
488}
489EXPORT_SYMBOL_GPL(kthread_unpark);
490
491/**
492 * kthread_park - park a thread created by kthread_create().
493 * @k: thread created by kthread_create().
494 *
495 * Sets kthread_should_park() for @k to return true, wakes it, and
496 * waits for it to return. This can also be called after kthread_create()
497 * instead of calling wake_up_process(): the thread will park without
498 * calling threadfn().
499 *
500 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
501 * If called by the kthread itself just the park bit is set.
502 */
503int kthread_park(struct task_struct *k)
504{
505 struct kthread *kthread = to_kthread(k);
506
507 if (WARN_ON(k->flags & PF_EXITING))
508 return -ENOSYS;
509
510 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
511 return -EBUSY;
512
513 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
514 if (k != current) {
515 wake_up_process(k);
516 /*
517 * Wait for __kthread_parkme() to complete(), this means we
518 * _will_ have TASK_PARKED and are about to call schedule().
519 */
520 wait_for_completion(&kthread->parked);
521 /*
522 * Now wait for that schedule() to complete and the task to
523 * get scheduled out.
524 */
525 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
526 }
527
528 return 0;
529}
530EXPORT_SYMBOL_GPL(kthread_park);
531
532/**
533 * kthread_stop - stop a thread created by kthread_create().
534 * @k: thread created by kthread_create().
535 *
536 * Sets kthread_should_stop() for @k to return true, wakes it, and
537 * waits for it to exit. This can also be called after kthread_create()
538 * instead of calling wake_up_process(): the thread will exit without
539 * calling threadfn().
540 *
541 * If threadfn() may call do_exit() itself, the caller must ensure
542 * task_struct can't go away.
543 *
544 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
545 * was never called.
546 */
547int kthread_stop(struct task_struct *k)
548{
549 struct kthread *kthread;
550 int ret;
551
552 trace_sched_kthread_stop(k);
553
554 get_task_struct(k);
555 kthread = to_kthread(k);
556 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
557 kthread_unpark(k);
558 wake_up_process(k);
559 wait_for_completion(&kthread->exited);
560 ret = k->exit_code;
561 put_task_struct(k);
562
563 trace_sched_kthread_stop_ret(ret);
564 return ret;
565}
566EXPORT_SYMBOL(kthread_stop);
567
568int kthreadd(void *unused)
569{
570 struct task_struct *tsk = current;
571
572 /* Setup a clean context for our children to inherit. */
573 set_task_comm(tsk, "kthreadd");
574 ignore_signals(tsk);
575 set_cpus_allowed_ptr(tsk, cpu_all_mask);
576 set_mems_allowed(node_states[N_MEMORY]);
577
578 current->flags |= PF_NOFREEZE;
579 cgroup_init_kthreadd();
580
581 for (;;) {
582 set_current_state(TASK_INTERRUPTIBLE);
583 if (list_empty(&kthread_create_list))
584 schedule();
585 __set_current_state(TASK_RUNNING);
586
587 spin_lock(&kthread_create_lock);
588 while (!list_empty(&kthread_create_list)) {
589 struct kthread_create_info *create;
590
591 create = list_entry(kthread_create_list.next,
592 struct kthread_create_info, list);
593 list_del_init(&create->list);
594 spin_unlock(&kthread_create_lock);
595
596 create_kthread(create);
597
598 spin_lock(&kthread_create_lock);
599 }
600 spin_unlock(&kthread_create_lock);
601 }
602
603 return 0;
604}
605
606void __kthread_init_worker(struct kthread_worker *worker,
607 const char *name,
608 struct lock_class_key *key)
609{
610 memset(worker, 0, sizeof(struct kthread_worker));
611 raw_spin_lock_init(&worker->lock);
612 lockdep_set_class_and_name(&worker->lock, key, name);
613 INIT_LIST_HEAD(&worker->work_list);
614 INIT_LIST_HEAD(&worker->delayed_work_list);
615}
616EXPORT_SYMBOL_GPL(__kthread_init_worker);
617
618/**
619 * kthread_worker_fn - kthread function to process kthread_worker
620 * @worker_ptr: pointer to initialized kthread_worker
621 *
622 * This function implements the main cycle of kthread worker. It processes
623 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
624 * is empty.
625 *
626 * The works are not allowed to keep any locks, disable preemption or interrupts
627 * when they finish. There is defined a safe point for freezing when one work
628 * finishes and before a new one is started.
629 *
630 * Also the works must not be handled by more than one worker at the same time,
631 * see also kthread_queue_work().
632 */
633int kthread_worker_fn(void *worker_ptr)
634{
635 struct kthread_worker *worker = worker_ptr;
636 struct kthread_work *work;
637
638 /*
639 * FIXME: Update the check and remove the assignment when all kthread
640 * worker users are created using kthread_create_worker*() functions.
641 */
642 WARN_ON(worker->task && worker->task != current);
643 worker->task = current;
644
645 if (worker->flags & KTW_FREEZABLE)
646 set_freezable();
647
648repeat:
649 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
650
651 if (kthread_should_stop()) {
652 __set_current_state(TASK_RUNNING);
653 raw_spin_lock_irq(&worker->lock);
654 worker->task = NULL;
655 raw_spin_unlock_irq(&worker->lock);
656 return 0;
657 }
658
659 work = NULL;
660 raw_spin_lock_irq(&worker->lock);
661 if (!list_empty(&worker->work_list)) {
662 work = list_first_entry(&worker->work_list,
663 struct kthread_work, node);
664 list_del_init(&work->node);
665 }
666 worker->current_work = work;
667 raw_spin_unlock_irq(&worker->lock);
668
669 if (work) {
670 __set_current_state(TASK_RUNNING);
671 work->func(work);
672 } else if (!freezing(current))
673 schedule();
674
675 try_to_freeze();
676 cond_resched();
677 goto repeat;
678}
679EXPORT_SYMBOL_GPL(kthread_worker_fn);
680
681static __printf(3, 0) struct kthread_worker *
682__kthread_create_worker(int cpu, unsigned int flags,
683 const char namefmt[], va_list args)
684{
685 struct kthread_worker *worker;
686 struct task_struct *task;
687 int node = NUMA_NO_NODE;
688
689 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
690 if (!worker)
691 return ERR_PTR(-ENOMEM);
692
693 kthread_init_worker(worker);
694
695 if (cpu >= 0)
696 node = cpu_to_node(cpu);
697
698 task = __kthread_create_on_node(kthread_worker_fn, worker,
699 node, namefmt, args);
700 if (IS_ERR(task))
701 goto fail_task;
702
703 if (cpu >= 0)
704 kthread_bind(task, cpu);
705
706 worker->flags = flags;
707 worker->task = task;
708 wake_up_process(task);
709 return worker;
710
711fail_task:
712 kfree(worker);
713 return ERR_CAST(task);
714}
715
716/**
717 * kthread_create_worker - create a kthread worker
718 * @flags: flags modifying the default behavior of the worker
719 * @namefmt: printf-style name for the kthread worker (task).
720 *
721 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
722 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
723 * when the worker was SIGKILLed.
724 */
725struct kthread_worker *
726kthread_create_worker(unsigned int flags, const char namefmt[], ...)
727{
728 struct kthread_worker *worker;
729 va_list args;
730
731 va_start(args, namefmt);
732 worker = __kthread_create_worker(-1, flags, namefmt, args);
733 va_end(args);
734
735 return worker;
736}
737EXPORT_SYMBOL(kthread_create_worker);
738
739/**
740 * kthread_create_worker_on_cpu - create a kthread worker and bind it
741 * it to a given CPU and the associated NUMA node.
742 * @cpu: CPU number
743 * @flags: flags modifying the default behavior of the worker
744 * @namefmt: printf-style name for the kthread worker (task).
745 *
746 * Use a valid CPU number if you want to bind the kthread worker
747 * to the given CPU and the associated NUMA node.
748 *
749 * A good practice is to add the cpu number also into the worker name.
750 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
751 *
752 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
753 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
754 * when the worker was SIGKILLed.
755 */
756struct kthread_worker *
757kthread_create_worker_on_cpu(int cpu, unsigned int flags,
758 const char namefmt[], ...)
759{
760 struct kthread_worker *worker;
761 va_list args;
762
763 va_start(args, namefmt);
764 worker = __kthread_create_worker(cpu, flags, namefmt, args);
765 va_end(args);
766
767 return worker;
768}
769EXPORT_SYMBOL(kthread_create_worker_on_cpu);
770
771/*
772 * Returns true when the work could not be queued at the moment.
773 * It happens when it is already pending in a worker list
774 * or when it is being cancelled.
775 */
776static inline bool queuing_blocked(struct kthread_worker *worker,
777 struct kthread_work *work)
778{
779 lockdep_assert_held(&worker->lock);
780
781 return !list_empty(&work->node) || work->canceling;
782}
783
784static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
785 struct kthread_work *work)
786{
787 lockdep_assert_held(&worker->lock);
788 WARN_ON_ONCE(!list_empty(&work->node));
789 /* Do not use a work with >1 worker, see kthread_queue_work() */
790 WARN_ON_ONCE(work->worker && work->worker != worker);
791}
792
793/* insert @work before @pos in @worker */
794static void kthread_insert_work(struct kthread_worker *worker,
795 struct kthread_work *work,
796 struct list_head *pos)
797{
798 kthread_insert_work_sanity_check(worker, work);
799
800 list_add_tail(&work->node, pos);
801 work->worker = worker;
802 if (!worker->current_work && likely(worker->task))
803 wake_up_process(worker->task);
804}
805
806/**
807 * kthread_queue_work - queue a kthread_work
808 * @worker: target kthread_worker
809 * @work: kthread_work to queue
810 *
811 * Queue @work to work processor @task for async execution. @task
812 * must have been created with kthread_worker_create(). Returns %true
813 * if @work was successfully queued, %false if it was already pending.
814 *
815 * Reinitialize the work if it needs to be used by another worker.
816 * For example, when the worker was stopped and started again.
817 */
818bool kthread_queue_work(struct kthread_worker *worker,
819 struct kthread_work *work)
820{
821 bool ret = false;
822 unsigned long flags;
823
824 raw_spin_lock_irqsave(&worker->lock, flags);
825 if (!queuing_blocked(worker, work)) {
826 kthread_insert_work(worker, work, &worker->work_list);
827 ret = true;
828 }
829 raw_spin_unlock_irqrestore(&worker->lock, flags);
830 return ret;
831}
832EXPORT_SYMBOL_GPL(kthread_queue_work);
833
834/**
835 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
836 * delayed work when the timer expires.
837 * @t: pointer to the expired timer
838 *
839 * The format of the function is defined by struct timer_list.
840 * It should have been called from irqsafe timer with irq already off.
841 */
842void kthread_delayed_work_timer_fn(struct timer_list *t)
843{
844 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
845 struct kthread_work *work = &dwork->work;
846 struct kthread_worker *worker = work->worker;
847 unsigned long flags;
848
849 /*
850 * This might happen when a pending work is reinitialized.
851 * It means that it is used a wrong way.
852 */
853 if (WARN_ON_ONCE(!worker))
854 return;
855
856 raw_spin_lock_irqsave(&worker->lock, flags);
857 /* Work must not be used with >1 worker, see kthread_queue_work(). */
858 WARN_ON_ONCE(work->worker != worker);
859
860 /* Move the work from worker->delayed_work_list. */
861 WARN_ON_ONCE(list_empty(&work->node));
862 list_del_init(&work->node);
863 kthread_insert_work(worker, work, &worker->work_list);
864
865 raw_spin_unlock_irqrestore(&worker->lock, flags);
866}
867EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
868
869static void __kthread_queue_delayed_work(struct kthread_worker *worker,
870 struct kthread_delayed_work *dwork,
871 unsigned long delay)
872{
873 struct timer_list *timer = &dwork->timer;
874 struct kthread_work *work = &dwork->work;
875
876 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
877
878 /*
879 * If @delay is 0, queue @dwork->work immediately. This is for
880 * both optimization and correctness. The earliest @timer can
881 * expire is on the closest next tick and delayed_work users depend
882 * on that there's no such delay when @delay is 0.
883 */
884 if (!delay) {
885 kthread_insert_work(worker, work, &worker->work_list);
886 return;
887 }
888
889 /* Be paranoid and try to detect possible races already now. */
890 kthread_insert_work_sanity_check(worker, work);
891
892 list_add(&work->node, &worker->delayed_work_list);
893 work->worker = worker;
894 timer->expires = jiffies + delay;
895 add_timer(timer);
896}
897
898/**
899 * kthread_queue_delayed_work - queue the associated kthread work
900 * after a delay.
901 * @worker: target kthread_worker
902 * @dwork: kthread_delayed_work to queue
903 * @delay: number of jiffies to wait before queuing
904 *
905 * If the work has not been pending it starts a timer that will queue
906 * the work after the given @delay. If @delay is zero, it queues the
907 * work immediately.
908 *
909 * Return: %false if the @work has already been pending. It means that
910 * either the timer was running or the work was queued. It returns %true
911 * otherwise.
912 */
913bool kthread_queue_delayed_work(struct kthread_worker *worker,
914 struct kthread_delayed_work *dwork,
915 unsigned long delay)
916{
917 struct kthread_work *work = &dwork->work;
918 unsigned long flags;
919 bool ret = false;
920
921 raw_spin_lock_irqsave(&worker->lock, flags);
922
923 if (!queuing_blocked(worker, work)) {
924 __kthread_queue_delayed_work(worker, dwork, delay);
925 ret = true;
926 }
927
928 raw_spin_unlock_irqrestore(&worker->lock, flags);
929 return ret;
930}
931EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
932
933struct kthread_flush_work {
934 struct kthread_work work;
935 struct completion done;
936};
937
938static void kthread_flush_work_fn(struct kthread_work *work)
939{
940 struct kthread_flush_work *fwork =
941 container_of(work, struct kthread_flush_work, work);
942 complete(&fwork->done);
943}
944
945/**
946 * kthread_flush_work - flush a kthread_work
947 * @work: work to flush
948 *
949 * If @work is queued or executing, wait for it to finish execution.
950 */
951void kthread_flush_work(struct kthread_work *work)
952{
953 struct kthread_flush_work fwork = {
954 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
955 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
956 };
957 struct kthread_worker *worker;
958 bool noop = false;
959
960 worker = work->worker;
961 if (!worker)
962 return;
963
964 raw_spin_lock_irq(&worker->lock);
965 /* Work must not be used with >1 worker, see kthread_queue_work(). */
966 WARN_ON_ONCE(work->worker != worker);
967
968 if (!list_empty(&work->node))
969 kthread_insert_work(worker, &fwork.work, work->node.next);
970 else if (worker->current_work == work)
971 kthread_insert_work(worker, &fwork.work,
972 worker->work_list.next);
973 else
974 noop = true;
975
976 raw_spin_unlock_irq(&worker->lock);
977
978 if (!noop)
979 wait_for_completion(&fwork.done);
980}
981EXPORT_SYMBOL_GPL(kthread_flush_work);
982
983/*
984 * This function removes the work from the worker queue. Also it makes sure
985 * that it won't get queued later via the delayed work's timer.
986 *
987 * The work might still be in use when this function finishes. See the
988 * current_work proceed by the worker.
989 *
990 * Return: %true if @work was pending and successfully canceled,
991 * %false if @work was not pending
992 */
993static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
994 unsigned long *flags)
995{
996 /* Try to cancel the timer if exists. */
997 if (is_dwork) {
998 struct kthread_delayed_work *dwork =
999 container_of(work, struct kthread_delayed_work, work);
1000 struct kthread_worker *worker = work->worker;
1001
1002 /*
1003 * del_timer_sync() must be called to make sure that the timer
1004 * callback is not running. The lock must be temporary released
1005 * to avoid a deadlock with the callback. In the meantime,
1006 * any queuing is blocked by setting the canceling counter.
1007 */
1008 work->canceling++;
1009 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1010 del_timer_sync(&dwork->timer);
1011 raw_spin_lock_irqsave(&worker->lock, *flags);
1012 work->canceling--;
1013 }
1014
1015 /*
1016 * Try to remove the work from a worker list. It might either
1017 * be from worker->work_list or from worker->delayed_work_list.
1018 */
1019 if (!list_empty(&work->node)) {
1020 list_del_init(&work->node);
1021 return true;
1022 }
1023
1024 return false;
1025}
1026
1027/**
1028 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1029 * @worker: kthread worker to use
1030 * @dwork: kthread delayed work to queue
1031 * @delay: number of jiffies to wait before queuing
1032 *
1033 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1034 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1035 * @work is guaranteed to be queued immediately.
1036 *
1037 * Return: %true if @dwork was pending and its timer was modified,
1038 * %false otherwise.
1039 *
1040 * A special case is when the work is being canceled in parallel.
1041 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1042 * or yet another kthread_mod_delayed_work() call. We let the other command
1043 * win and return %false here. The caller is supposed to synchronize these
1044 * operations a reasonable way.
1045 *
1046 * This function is safe to call from any context including IRQ handler.
1047 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1048 * for details.
1049 */
1050bool kthread_mod_delayed_work(struct kthread_worker *worker,
1051 struct kthread_delayed_work *dwork,
1052 unsigned long delay)
1053{
1054 struct kthread_work *work = &dwork->work;
1055 unsigned long flags;
1056 int ret = false;
1057
1058 raw_spin_lock_irqsave(&worker->lock, flags);
1059
1060 /* Do not bother with canceling when never queued. */
1061 if (!work->worker)
1062 goto fast_queue;
1063
1064 /* Work must not be used with >1 worker, see kthread_queue_work() */
1065 WARN_ON_ONCE(work->worker != worker);
1066
1067 /* Do not fight with another command that is canceling this work. */
1068 if (work->canceling)
1069 goto out;
1070
1071 ret = __kthread_cancel_work(work, true, &flags);
1072fast_queue:
1073 __kthread_queue_delayed_work(worker, dwork, delay);
1074out:
1075 raw_spin_unlock_irqrestore(&worker->lock, flags);
1076 return ret;
1077}
1078EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1079
1080static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1081{
1082 struct kthread_worker *worker = work->worker;
1083 unsigned long flags;
1084 int ret = false;
1085
1086 if (!worker)
1087 goto out;
1088
1089 raw_spin_lock_irqsave(&worker->lock, flags);
1090 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1091 WARN_ON_ONCE(work->worker != worker);
1092
1093 ret = __kthread_cancel_work(work, is_dwork, &flags);
1094
1095 if (worker->current_work != work)
1096 goto out_fast;
1097
1098 /*
1099 * The work is in progress and we need to wait with the lock released.
1100 * In the meantime, block any queuing by setting the canceling counter.
1101 */
1102 work->canceling++;
1103 raw_spin_unlock_irqrestore(&worker->lock, flags);
1104 kthread_flush_work(work);
1105 raw_spin_lock_irqsave(&worker->lock, flags);
1106 work->canceling--;
1107
1108out_fast:
1109 raw_spin_unlock_irqrestore(&worker->lock, flags);
1110out:
1111 return ret;
1112}
1113
1114/**
1115 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1116 * @work: the kthread work to cancel
1117 *
1118 * Cancel @work and wait for its execution to finish. This function
1119 * can be used even if the work re-queues itself. On return from this
1120 * function, @work is guaranteed to be not pending or executing on any CPU.
1121 *
1122 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1123 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1124 *
1125 * The caller must ensure that the worker on which @work was last
1126 * queued can't be destroyed before this function returns.
1127 *
1128 * Return: %true if @work was pending, %false otherwise.
1129 */
1130bool kthread_cancel_work_sync(struct kthread_work *work)
1131{
1132 return __kthread_cancel_work_sync(work, false);
1133}
1134EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1135
1136/**
1137 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1138 * wait for it to finish.
1139 * @dwork: the kthread delayed work to cancel
1140 *
1141 * This is kthread_cancel_work_sync() for delayed works.
1142 *
1143 * Return: %true if @dwork was pending, %false otherwise.
1144 */
1145bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1146{
1147 return __kthread_cancel_work_sync(&dwork->work, true);
1148}
1149EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1150
1151/**
1152 * kthread_flush_worker - flush all current works on a kthread_worker
1153 * @worker: worker to flush
1154 *
1155 * Wait until all currently executing or pending works on @worker are
1156 * finished.
1157 */
1158void kthread_flush_worker(struct kthread_worker *worker)
1159{
1160 struct kthread_flush_work fwork = {
1161 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1162 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1163 };
1164
1165 kthread_queue_work(worker, &fwork.work);
1166 wait_for_completion(&fwork.done);
1167}
1168EXPORT_SYMBOL_GPL(kthread_flush_worker);
1169
1170/**
1171 * kthread_destroy_worker - destroy a kthread worker
1172 * @worker: worker to be destroyed
1173 *
1174 * Flush and destroy @worker. The simple flush is enough because the kthread
1175 * worker API is used only in trivial scenarios. There are no multi-step state
1176 * machines needed.
1177 */
1178void kthread_destroy_worker(struct kthread_worker *worker)
1179{
1180 struct task_struct *task;
1181
1182 task = worker->task;
1183 if (WARN_ON(!task))
1184 return;
1185
1186 kthread_flush_worker(worker);
1187 kthread_stop(task);
1188 WARN_ON(!list_empty(&worker->work_list));
1189 kfree(worker);
1190}
1191EXPORT_SYMBOL(kthread_destroy_worker);
1192
1193#ifdef CONFIG_BLK_CGROUP
1194/**
1195 * kthread_associate_blkcg - associate blkcg to current kthread
1196 * @css: the cgroup info
1197 *
1198 * Current thread must be a kthread. The thread is running jobs on behalf of
1199 * other threads. In some cases, we expect the jobs attach cgroup info of
1200 * original threads instead of that of current thread. This function stores
1201 * original thread's cgroup info in current kthread context for later
1202 * retrieval.
1203 */
1204void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1205{
1206 struct kthread *kthread;
1207
1208 if (!(current->flags & PF_KTHREAD))
1209 return;
1210 kthread = to_kthread(current);
1211 if (!kthread)
1212 return;
1213
1214 if (kthread->blkcg_css) {
1215 css_put(kthread->blkcg_css);
1216 kthread->blkcg_css = NULL;
1217 }
1218 if (css) {
1219 css_get(css);
1220 kthread->blkcg_css = css;
1221 }
1222}
1223EXPORT_SYMBOL(kthread_associate_blkcg);
1224
1225/**
1226 * kthread_blkcg - get associated blkcg css of current kthread
1227 *
1228 * Current thread must be a kthread.
1229 */
1230struct cgroup_subsys_state *kthread_blkcg(void)
1231{
1232 struct kthread *kthread;
1233
1234 if (current->flags & PF_KTHREAD) {
1235 kthread = to_kthread(current);
1236 if (kthread)
1237 return kthread->blkcg_css;
1238 }
1239 return NULL;
1240}
1241EXPORT_SYMBOL(kthread_blkcg);
1242#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/* Kernel thread helper functions.
3 * Copyright (C) 2004 IBM Corporation, Rusty Russell.
4 * Copyright (C) 2009 Red Hat, Inc.
5 *
6 * Creation is done via kthreadd, so that we get a clean environment
7 * even if we're invoked from userspace (think modprobe, hotplug cpu,
8 * etc.).
9 */
10#include <uapi/linux/sched/types.h>
11#include <linux/mm.h>
12#include <linux/mmu_context.h>
13#include <linux/sched.h>
14#include <linux/sched/mm.h>
15#include <linux/sched/task.h>
16#include <linux/kthread.h>
17#include <linux/completion.h>
18#include <linux/err.h>
19#include <linux/cgroup.h>
20#include <linux/cpuset.h>
21#include <linux/unistd.h>
22#include <linux/file.h>
23#include <linux/export.h>
24#include <linux/mutex.h>
25#include <linux/slab.h>
26#include <linux/freezer.h>
27#include <linux/ptrace.h>
28#include <linux/uaccess.h>
29#include <linux/numa.h>
30#include <linux/sched/isolation.h>
31#include <trace/events/sched.h>
32
33
34static DEFINE_SPINLOCK(kthread_create_lock);
35static LIST_HEAD(kthread_create_list);
36struct task_struct *kthreadd_task;
37
38struct kthread_create_info
39{
40 /* Information passed to kthread() from kthreadd. */
41 int (*threadfn)(void *data);
42 void *data;
43 int node;
44
45 /* Result passed back to kthread_create() from kthreadd. */
46 struct task_struct *result;
47 struct completion *done;
48
49 struct list_head list;
50};
51
52struct kthread {
53 unsigned long flags;
54 unsigned int cpu;
55 int (*threadfn)(void *);
56 void *data;
57 mm_segment_t oldfs;
58 struct completion parked;
59 struct completion exited;
60#ifdef CONFIG_BLK_CGROUP
61 struct cgroup_subsys_state *blkcg_css;
62#endif
63};
64
65enum KTHREAD_BITS {
66 KTHREAD_IS_PER_CPU = 0,
67 KTHREAD_SHOULD_STOP,
68 KTHREAD_SHOULD_PARK,
69};
70
71static inline void set_kthread_struct(void *kthread)
72{
73 /*
74 * We abuse ->set_child_tid to avoid the new member and because it
75 * can't be wrongly copied by copy_process(). We also rely on fact
76 * that the caller can't exec, so PF_KTHREAD can't be cleared.
77 */
78 current->set_child_tid = (__force void __user *)kthread;
79}
80
81static inline struct kthread *to_kthread(struct task_struct *k)
82{
83 WARN_ON(!(k->flags & PF_KTHREAD));
84 return (__force void *)k->set_child_tid;
85}
86
87void free_kthread_struct(struct task_struct *k)
88{
89 struct kthread *kthread;
90
91 /*
92 * Can be NULL if this kthread was created by kernel_thread()
93 * or if kmalloc() in kthread() failed.
94 */
95 kthread = to_kthread(k);
96#ifdef CONFIG_BLK_CGROUP
97 WARN_ON_ONCE(kthread && kthread->blkcg_css);
98#endif
99 kfree(kthread);
100}
101
102/**
103 * kthread_should_stop - should this kthread return now?
104 *
105 * When someone calls kthread_stop() on your kthread, it will be woken
106 * and this will return true. You should then return, and your return
107 * value will be passed through to kthread_stop().
108 */
109bool kthread_should_stop(void)
110{
111 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
112}
113EXPORT_SYMBOL(kthread_should_stop);
114
115bool __kthread_should_park(struct task_struct *k)
116{
117 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
118}
119EXPORT_SYMBOL_GPL(__kthread_should_park);
120
121/**
122 * kthread_should_park - should this kthread park now?
123 *
124 * When someone calls kthread_park() on your kthread, it will be woken
125 * and this will return true. You should then do the necessary
126 * cleanup and call kthread_parkme()
127 *
128 * Similar to kthread_should_stop(), but this keeps the thread alive
129 * and in a park position. kthread_unpark() "restarts" the thread and
130 * calls the thread function again.
131 */
132bool kthread_should_park(void)
133{
134 return __kthread_should_park(current);
135}
136EXPORT_SYMBOL_GPL(kthread_should_park);
137
138/**
139 * kthread_freezable_should_stop - should this freezable kthread return now?
140 * @was_frozen: optional out parameter, indicates whether %current was frozen
141 *
142 * kthread_should_stop() for freezable kthreads, which will enter
143 * refrigerator if necessary. This function is safe from kthread_stop() /
144 * freezer deadlock and freezable kthreads should use this function instead
145 * of calling try_to_freeze() directly.
146 */
147bool kthread_freezable_should_stop(bool *was_frozen)
148{
149 bool frozen = false;
150
151 might_sleep();
152
153 if (unlikely(freezing(current)))
154 frozen = __refrigerator(true);
155
156 if (was_frozen)
157 *was_frozen = frozen;
158
159 return kthread_should_stop();
160}
161EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
162
163/**
164 * kthread_func - return the function specified on kthread creation
165 * @task: kthread task in question
166 *
167 * Returns NULL if the task is not a kthread.
168 */
169void *kthread_func(struct task_struct *task)
170{
171 if (task->flags & PF_KTHREAD)
172 return to_kthread(task)->threadfn;
173 return NULL;
174}
175EXPORT_SYMBOL_GPL(kthread_func);
176
177/**
178 * kthread_data - return data value specified on kthread creation
179 * @task: kthread task in question
180 *
181 * Return the data value specified when kthread @task was created.
182 * The caller is responsible for ensuring the validity of @task when
183 * calling this function.
184 */
185void *kthread_data(struct task_struct *task)
186{
187 return to_kthread(task)->data;
188}
189EXPORT_SYMBOL_GPL(kthread_data);
190
191/**
192 * kthread_probe_data - speculative version of kthread_data()
193 * @task: possible kthread task in question
194 *
195 * @task could be a kthread task. Return the data value specified when it
196 * was created if accessible. If @task isn't a kthread task or its data is
197 * inaccessible for any reason, %NULL is returned. This function requires
198 * that @task itself is safe to dereference.
199 */
200void *kthread_probe_data(struct task_struct *task)
201{
202 struct kthread *kthread = to_kthread(task);
203 void *data = NULL;
204
205 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
206 return data;
207}
208
209static void __kthread_parkme(struct kthread *self)
210{
211 for (;;) {
212 /*
213 * TASK_PARKED is a special state; we must serialize against
214 * possible pending wakeups to avoid store-store collisions on
215 * task->state.
216 *
217 * Such a collision might possibly result in the task state
218 * changin from TASK_PARKED and us failing the
219 * wait_task_inactive() in kthread_park().
220 */
221 set_special_state(TASK_PARKED);
222 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
223 break;
224
225 /*
226 * Thread is going to call schedule(), do not preempt it,
227 * or the caller of kthread_park() may spend more time in
228 * wait_task_inactive().
229 */
230 preempt_disable();
231 complete(&self->parked);
232 schedule_preempt_disabled();
233 preempt_enable();
234 }
235 __set_current_state(TASK_RUNNING);
236}
237
238void kthread_parkme(void)
239{
240 __kthread_parkme(to_kthread(current));
241}
242EXPORT_SYMBOL_GPL(kthread_parkme);
243
244static int kthread(void *_create)
245{
246 /* Copy data: it's on kthread's stack */
247 struct kthread_create_info *create = _create;
248 int (*threadfn)(void *data) = create->threadfn;
249 void *data = create->data;
250 struct completion *done;
251 struct kthread *self;
252 int ret;
253
254 self = kzalloc(sizeof(*self), GFP_KERNEL);
255 set_kthread_struct(self);
256
257 /* If user was SIGKILLed, I release the structure. */
258 done = xchg(&create->done, NULL);
259 if (!done) {
260 kfree(create);
261 do_exit(-EINTR);
262 }
263
264 if (!self) {
265 create->result = ERR_PTR(-ENOMEM);
266 complete(done);
267 do_exit(-ENOMEM);
268 }
269
270 self->threadfn = threadfn;
271 self->data = data;
272 init_completion(&self->exited);
273 init_completion(&self->parked);
274 current->vfork_done = &self->exited;
275
276 /* OK, tell user we're spawned, wait for stop or wakeup */
277 __set_current_state(TASK_UNINTERRUPTIBLE);
278 create->result = current;
279 /*
280 * Thread is going to call schedule(), do not preempt it,
281 * or the creator may spend more time in wait_task_inactive().
282 */
283 preempt_disable();
284 complete(done);
285 schedule_preempt_disabled();
286 preempt_enable();
287
288 ret = -EINTR;
289 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
290 cgroup_kthread_ready();
291 __kthread_parkme(self);
292 ret = threadfn(data);
293 }
294 do_exit(ret);
295}
296
297/* called from do_fork() to get node information for about to be created task */
298int tsk_fork_get_node(struct task_struct *tsk)
299{
300#ifdef CONFIG_NUMA
301 if (tsk == kthreadd_task)
302 return tsk->pref_node_fork;
303#endif
304 return NUMA_NO_NODE;
305}
306
307static void create_kthread(struct kthread_create_info *create)
308{
309 int pid;
310
311#ifdef CONFIG_NUMA
312 current->pref_node_fork = create->node;
313#endif
314 /* We want our own signal handler (we take no signals by default). */
315 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
316 if (pid < 0) {
317 /* If user was SIGKILLed, I release the structure. */
318 struct completion *done = xchg(&create->done, NULL);
319
320 if (!done) {
321 kfree(create);
322 return;
323 }
324 create->result = ERR_PTR(pid);
325 complete(done);
326 }
327}
328
329static __printf(4, 0)
330struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
331 void *data, int node,
332 const char namefmt[],
333 va_list args)
334{
335 DECLARE_COMPLETION_ONSTACK(done);
336 struct task_struct *task;
337 struct kthread_create_info *create = kmalloc(sizeof(*create),
338 GFP_KERNEL);
339
340 if (!create)
341 return ERR_PTR(-ENOMEM);
342 create->threadfn = threadfn;
343 create->data = data;
344 create->node = node;
345 create->done = &done;
346
347 spin_lock(&kthread_create_lock);
348 list_add_tail(&create->list, &kthread_create_list);
349 spin_unlock(&kthread_create_lock);
350
351 wake_up_process(kthreadd_task);
352 /*
353 * Wait for completion in killable state, for I might be chosen by
354 * the OOM killer while kthreadd is trying to allocate memory for
355 * new kernel thread.
356 */
357 if (unlikely(wait_for_completion_killable(&done))) {
358 /*
359 * If I was SIGKILLed before kthreadd (or new kernel thread)
360 * calls complete(), leave the cleanup of this structure to
361 * that thread.
362 */
363 if (xchg(&create->done, NULL))
364 return ERR_PTR(-EINTR);
365 /*
366 * kthreadd (or new kernel thread) will call complete()
367 * shortly.
368 */
369 wait_for_completion(&done);
370 }
371 task = create->result;
372 if (!IS_ERR(task)) {
373 static const struct sched_param param = { .sched_priority = 0 };
374 char name[TASK_COMM_LEN];
375
376 /*
377 * task is already visible to other tasks, so updating
378 * COMM must be protected.
379 */
380 vsnprintf(name, sizeof(name), namefmt, args);
381 set_task_comm(task, name);
382 /*
383 * root may have changed our (kthreadd's) priority or CPU mask.
384 * The kernel thread should not inherit these properties.
385 */
386 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
387 set_cpus_allowed_ptr(task,
388 housekeeping_cpumask(HK_FLAG_KTHREAD));
389 }
390 kfree(create);
391 return task;
392}
393
394/**
395 * kthread_create_on_node - create a kthread.
396 * @threadfn: the function to run until signal_pending(current).
397 * @data: data ptr for @threadfn.
398 * @node: task and thread structures for the thread are allocated on this node
399 * @namefmt: printf-style name for the thread.
400 *
401 * Description: This helper function creates and names a kernel
402 * thread. The thread will be stopped: use wake_up_process() to start
403 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
404 * is affine to all CPUs.
405 *
406 * If thread is going to be bound on a particular cpu, give its node
407 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
408 * When woken, the thread will run @threadfn() with @data as its
409 * argument. @threadfn() can either call do_exit() directly if it is a
410 * standalone thread for which no one will call kthread_stop(), or
411 * return when 'kthread_should_stop()' is true (which means
412 * kthread_stop() has been called). The return value should be zero
413 * or a negative error number; it will be passed to kthread_stop().
414 *
415 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
416 */
417struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
418 void *data, int node,
419 const char namefmt[],
420 ...)
421{
422 struct task_struct *task;
423 va_list args;
424
425 va_start(args, namefmt);
426 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
427 va_end(args);
428
429 return task;
430}
431EXPORT_SYMBOL(kthread_create_on_node);
432
433static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
434{
435 unsigned long flags;
436
437 if (!wait_task_inactive(p, state)) {
438 WARN_ON(1);
439 return;
440 }
441
442 /* It's safe because the task is inactive. */
443 raw_spin_lock_irqsave(&p->pi_lock, flags);
444 do_set_cpus_allowed(p, mask);
445 p->flags |= PF_NO_SETAFFINITY;
446 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
447}
448
449static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
450{
451 __kthread_bind_mask(p, cpumask_of(cpu), state);
452}
453
454void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
455{
456 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
457}
458
459/**
460 * kthread_bind - bind a just-created kthread to a cpu.
461 * @p: thread created by kthread_create().
462 * @cpu: cpu (might not be online, must be possible) for @k to run on.
463 *
464 * Description: This function is equivalent to set_cpus_allowed(),
465 * except that @cpu doesn't need to be online, and the thread must be
466 * stopped (i.e., just returned from kthread_create()).
467 */
468void kthread_bind(struct task_struct *p, unsigned int cpu)
469{
470 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
471}
472EXPORT_SYMBOL(kthread_bind);
473
474/**
475 * kthread_create_on_cpu - Create a cpu bound kthread
476 * @threadfn: the function to run until signal_pending(current).
477 * @data: data ptr for @threadfn.
478 * @cpu: The cpu on which the thread should be bound,
479 * @namefmt: printf-style name for the thread. Format is restricted
480 * to "name.*%u". Code fills in cpu number.
481 *
482 * Description: This helper function creates and names a kernel thread
483 */
484struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
485 void *data, unsigned int cpu,
486 const char *namefmt)
487{
488 struct task_struct *p;
489
490 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
491 cpu);
492 if (IS_ERR(p))
493 return p;
494 kthread_bind(p, cpu);
495 /* CPU hotplug need to bind once again when unparking the thread. */
496 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
497 to_kthread(p)->cpu = cpu;
498 return p;
499}
500
501/**
502 * kthread_unpark - unpark a thread created by kthread_create().
503 * @k: thread created by kthread_create().
504 *
505 * Sets kthread_should_park() for @k to return false, wakes it, and
506 * waits for it to return. If the thread is marked percpu then its
507 * bound to the cpu again.
508 */
509void kthread_unpark(struct task_struct *k)
510{
511 struct kthread *kthread = to_kthread(k);
512
513 /*
514 * Newly created kthread was parked when the CPU was offline.
515 * The binding was lost and we need to set it again.
516 */
517 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
518 __kthread_bind(k, kthread->cpu, TASK_PARKED);
519
520 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
521 /*
522 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
523 */
524 wake_up_state(k, TASK_PARKED);
525}
526EXPORT_SYMBOL_GPL(kthread_unpark);
527
528/**
529 * kthread_park - park a thread created by kthread_create().
530 * @k: thread created by kthread_create().
531 *
532 * Sets kthread_should_park() for @k to return true, wakes it, and
533 * waits for it to return. This can also be called after kthread_create()
534 * instead of calling wake_up_process(): the thread will park without
535 * calling threadfn().
536 *
537 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
538 * If called by the kthread itself just the park bit is set.
539 */
540int kthread_park(struct task_struct *k)
541{
542 struct kthread *kthread = to_kthread(k);
543
544 if (WARN_ON(k->flags & PF_EXITING))
545 return -ENOSYS;
546
547 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
548 return -EBUSY;
549
550 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
551 if (k != current) {
552 wake_up_process(k);
553 /*
554 * Wait for __kthread_parkme() to complete(), this means we
555 * _will_ have TASK_PARKED and are about to call schedule().
556 */
557 wait_for_completion(&kthread->parked);
558 /*
559 * Now wait for that schedule() to complete and the task to
560 * get scheduled out.
561 */
562 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
563 }
564
565 return 0;
566}
567EXPORT_SYMBOL_GPL(kthread_park);
568
569/**
570 * kthread_stop - stop a thread created by kthread_create().
571 * @k: thread created by kthread_create().
572 *
573 * Sets kthread_should_stop() for @k to return true, wakes it, and
574 * waits for it to exit. This can also be called after kthread_create()
575 * instead of calling wake_up_process(): the thread will exit without
576 * calling threadfn().
577 *
578 * If threadfn() may call do_exit() itself, the caller must ensure
579 * task_struct can't go away.
580 *
581 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
582 * was never called.
583 */
584int kthread_stop(struct task_struct *k)
585{
586 struct kthread *kthread;
587 int ret;
588
589 trace_sched_kthread_stop(k);
590
591 get_task_struct(k);
592 kthread = to_kthread(k);
593 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
594 kthread_unpark(k);
595 wake_up_process(k);
596 wait_for_completion(&kthread->exited);
597 ret = k->exit_code;
598 put_task_struct(k);
599
600 trace_sched_kthread_stop_ret(ret);
601 return ret;
602}
603EXPORT_SYMBOL(kthread_stop);
604
605int kthreadd(void *unused)
606{
607 struct task_struct *tsk = current;
608
609 /* Setup a clean context for our children to inherit. */
610 set_task_comm(tsk, "kthreadd");
611 ignore_signals(tsk);
612 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
613 set_mems_allowed(node_states[N_MEMORY]);
614
615 current->flags |= PF_NOFREEZE;
616 cgroup_init_kthreadd();
617
618 for (;;) {
619 set_current_state(TASK_INTERRUPTIBLE);
620 if (list_empty(&kthread_create_list))
621 schedule();
622 __set_current_state(TASK_RUNNING);
623
624 spin_lock(&kthread_create_lock);
625 while (!list_empty(&kthread_create_list)) {
626 struct kthread_create_info *create;
627
628 create = list_entry(kthread_create_list.next,
629 struct kthread_create_info, list);
630 list_del_init(&create->list);
631 spin_unlock(&kthread_create_lock);
632
633 create_kthread(create);
634
635 spin_lock(&kthread_create_lock);
636 }
637 spin_unlock(&kthread_create_lock);
638 }
639
640 return 0;
641}
642
643void __kthread_init_worker(struct kthread_worker *worker,
644 const char *name,
645 struct lock_class_key *key)
646{
647 memset(worker, 0, sizeof(struct kthread_worker));
648 raw_spin_lock_init(&worker->lock);
649 lockdep_set_class_and_name(&worker->lock, key, name);
650 INIT_LIST_HEAD(&worker->work_list);
651 INIT_LIST_HEAD(&worker->delayed_work_list);
652}
653EXPORT_SYMBOL_GPL(__kthread_init_worker);
654
655/**
656 * kthread_worker_fn - kthread function to process kthread_worker
657 * @worker_ptr: pointer to initialized kthread_worker
658 *
659 * This function implements the main cycle of kthread worker. It processes
660 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
661 * is empty.
662 *
663 * The works are not allowed to keep any locks, disable preemption or interrupts
664 * when they finish. There is defined a safe point for freezing when one work
665 * finishes and before a new one is started.
666 *
667 * Also the works must not be handled by more than one worker at the same time,
668 * see also kthread_queue_work().
669 */
670int kthread_worker_fn(void *worker_ptr)
671{
672 struct kthread_worker *worker = worker_ptr;
673 struct kthread_work *work;
674
675 /*
676 * FIXME: Update the check and remove the assignment when all kthread
677 * worker users are created using kthread_create_worker*() functions.
678 */
679 WARN_ON(worker->task && worker->task != current);
680 worker->task = current;
681
682 if (worker->flags & KTW_FREEZABLE)
683 set_freezable();
684
685repeat:
686 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
687
688 if (kthread_should_stop()) {
689 __set_current_state(TASK_RUNNING);
690 raw_spin_lock_irq(&worker->lock);
691 worker->task = NULL;
692 raw_spin_unlock_irq(&worker->lock);
693 return 0;
694 }
695
696 work = NULL;
697 raw_spin_lock_irq(&worker->lock);
698 if (!list_empty(&worker->work_list)) {
699 work = list_first_entry(&worker->work_list,
700 struct kthread_work, node);
701 list_del_init(&work->node);
702 }
703 worker->current_work = work;
704 raw_spin_unlock_irq(&worker->lock);
705
706 if (work) {
707 __set_current_state(TASK_RUNNING);
708 work->func(work);
709 } else if (!freezing(current))
710 schedule();
711
712 try_to_freeze();
713 cond_resched();
714 goto repeat;
715}
716EXPORT_SYMBOL_GPL(kthread_worker_fn);
717
718static __printf(3, 0) struct kthread_worker *
719__kthread_create_worker(int cpu, unsigned int flags,
720 const char namefmt[], va_list args)
721{
722 struct kthread_worker *worker;
723 struct task_struct *task;
724 int node = NUMA_NO_NODE;
725
726 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
727 if (!worker)
728 return ERR_PTR(-ENOMEM);
729
730 kthread_init_worker(worker);
731
732 if (cpu >= 0)
733 node = cpu_to_node(cpu);
734
735 task = __kthread_create_on_node(kthread_worker_fn, worker,
736 node, namefmt, args);
737 if (IS_ERR(task))
738 goto fail_task;
739
740 if (cpu >= 0)
741 kthread_bind(task, cpu);
742
743 worker->flags = flags;
744 worker->task = task;
745 wake_up_process(task);
746 return worker;
747
748fail_task:
749 kfree(worker);
750 return ERR_CAST(task);
751}
752
753/**
754 * kthread_create_worker - create a kthread worker
755 * @flags: flags modifying the default behavior of the worker
756 * @namefmt: printf-style name for the kthread worker (task).
757 *
758 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
759 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
760 * when the worker was SIGKILLed.
761 */
762struct kthread_worker *
763kthread_create_worker(unsigned int flags, const char namefmt[], ...)
764{
765 struct kthread_worker *worker;
766 va_list args;
767
768 va_start(args, namefmt);
769 worker = __kthread_create_worker(-1, flags, namefmt, args);
770 va_end(args);
771
772 return worker;
773}
774EXPORT_SYMBOL(kthread_create_worker);
775
776/**
777 * kthread_create_worker_on_cpu - create a kthread worker and bind it
778 * it to a given CPU and the associated NUMA node.
779 * @cpu: CPU number
780 * @flags: flags modifying the default behavior of the worker
781 * @namefmt: printf-style name for the kthread worker (task).
782 *
783 * Use a valid CPU number if you want to bind the kthread worker
784 * to the given CPU and the associated NUMA node.
785 *
786 * A good practice is to add the cpu number also into the worker name.
787 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
788 *
789 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
790 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
791 * when the worker was SIGKILLed.
792 */
793struct kthread_worker *
794kthread_create_worker_on_cpu(int cpu, unsigned int flags,
795 const char namefmt[], ...)
796{
797 struct kthread_worker *worker;
798 va_list args;
799
800 va_start(args, namefmt);
801 worker = __kthread_create_worker(cpu, flags, namefmt, args);
802 va_end(args);
803
804 return worker;
805}
806EXPORT_SYMBOL(kthread_create_worker_on_cpu);
807
808/*
809 * Returns true when the work could not be queued at the moment.
810 * It happens when it is already pending in a worker list
811 * or when it is being cancelled.
812 */
813static inline bool queuing_blocked(struct kthread_worker *worker,
814 struct kthread_work *work)
815{
816 lockdep_assert_held(&worker->lock);
817
818 return !list_empty(&work->node) || work->canceling;
819}
820
821static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
822 struct kthread_work *work)
823{
824 lockdep_assert_held(&worker->lock);
825 WARN_ON_ONCE(!list_empty(&work->node));
826 /* Do not use a work with >1 worker, see kthread_queue_work() */
827 WARN_ON_ONCE(work->worker && work->worker != worker);
828}
829
830/* insert @work before @pos in @worker */
831static void kthread_insert_work(struct kthread_worker *worker,
832 struct kthread_work *work,
833 struct list_head *pos)
834{
835 kthread_insert_work_sanity_check(worker, work);
836
837 list_add_tail(&work->node, pos);
838 work->worker = worker;
839 if (!worker->current_work && likely(worker->task))
840 wake_up_process(worker->task);
841}
842
843/**
844 * kthread_queue_work - queue a kthread_work
845 * @worker: target kthread_worker
846 * @work: kthread_work to queue
847 *
848 * Queue @work to work processor @task for async execution. @task
849 * must have been created with kthread_worker_create(). Returns %true
850 * if @work was successfully queued, %false if it was already pending.
851 *
852 * Reinitialize the work if it needs to be used by another worker.
853 * For example, when the worker was stopped and started again.
854 */
855bool kthread_queue_work(struct kthread_worker *worker,
856 struct kthread_work *work)
857{
858 bool ret = false;
859 unsigned long flags;
860
861 raw_spin_lock_irqsave(&worker->lock, flags);
862 if (!queuing_blocked(worker, work)) {
863 kthread_insert_work(worker, work, &worker->work_list);
864 ret = true;
865 }
866 raw_spin_unlock_irqrestore(&worker->lock, flags);
867 return ret;
868}
869EXPORT_SYMBOL_GPL(kthread_queue_work);
870
871/**
872 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
873 * delayed work when the timer expires.
874 * @t: pointer to the expired timer
875 *
876 * The format of the function is defined by struct timer_list.
877 * It should have been called from irqsafe timer with irq already off.
878 */
879void kthread_delayed_work_timer_fn(struct timer_list *t)
880{
881 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
882 struct kthread_work *work = &dwork->work;
883 struct kthread_worker *worker = work->worker;
884 unsigned long flags;
885
886 /*
887 * This might happen when a pending work is reinitialized.
888 * It means that it is used a wrong way.
889 */
890 if (WARN_ON_ONCE(!worker))
891 return;
892
893 raw_spin_lock_irqsave(&worker->lock, flags);
894 /* Work must not be used with >1 worker, see kthread_queue_work(). */
895 WARN_ON_ONCE(work->worker != worker);
896
897 /* Move the work from worker->delayed_work_list. */
898 WARN_ON_ONCE(list_empty(&work->node));
899 list_del_init(&work->node);
900 kthread_insert_work(worker, work, &worker->work_list);
901
902 raw_spin_unlock_irqrestore(&worker->lock, flags);
903}
904EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
905
906static void __kthread_queue_delayed_work(struct kthread_worker *worker,
907 struct kthread_delayed_work *dwork,
908 unsigned long delay)
909{
910 struct timer_list *timer = &dwork->timer;
911 struct kthread_work *work = &dwork->work;
912
913 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
914
915 /*
916 * If @delay is 0, queue @dwork->work immediately. This is for
917 * both optimization and correctness. The earliest @timer can
918 * expire is on the closest next tick and delayed_work users depend
919 * on that there's no such delay when @delay is 0.
920 */
921 if (!delay) {
922 kthread_insert_work(worker, work, &worker->work_list);
923 return;
924 }
925
926 /* Be paranoid and try to detect possible races already now. */
927 kthread_insert_work_sanity_check(worker, work);
928
929 list_add(&work->node, &worker->delayed_work_list);
930 work->worker = worker;
931 timer->expires = jiffies + delay;
932 add_timer(timer);
933}
934
935/**
936 * kthread_queue_delayed_work - queue the associated kthread work
937 * after a delay.
938 * @worker: target kthread_worker
939 * @dwork: kthread_delayed_work to queue
940 * @delay: number of jiffies to wait before queuing
941 *
942 * If the work has not been pending it starts a timer that will queue
943 * the work after the given @delay. If @delay is zero, it queues the
944 * work immediately.
945 *
946 * Return: %false if the @work has already been pending. It means that
947 * either the timer was running or the work was queued. It returns %true
948 * otherwise.
949 */
950bool kthread_queue_delayed_work(struct kthread_worker *worker,
951 struct kthread_delayed_work *dwork,
952 unsigned long delay)
953{
954 struct kthread_work *work = &dwork->work;
955 unsigned long flags;
956 bool ret = false;
957
958 raw_spin_lock_irqsave(&worker->lock, flags);
959
960 if (!queuing_blocked(worker, work)) {
961 __kthread_queue_delayed_work(worker, dwork, delay);
962 ret = true;
963 }
964
965 raw_spin_unlock_irqrestore(&worker->lock, flags);
966 return ret;
967}
968EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
969
970struct kthread_flush_work {
971 struct kthread_work work;
972 struct completion done;
973};
974
975static void kthread_flush_work_fn(struct kthread_work *work)
976{
977 struct kthread_flush_work *fwork =
978 container_of(work, struct kthread_flush_work, work);
979 complete(&fwork->done);
980}
981
982/**
983 * kthread_flush_work - flush a kthread_work
984 * @work: work to flush
985 *
986 * If @work is queued or executing, wait for it to finish execution.
987 */
988void kthread_flush_work(struct kthread_work *work)
989{
990 struct kthread_flush_work fwork = {
991 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
992 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
993 };
994 struct kthread_worker *worker;
995 bool noop = false;
996
997 worker = work->worker;
998 if (!worker)
999 return;
1000
1001 raw_spin_lock_irq(&worker->lock);
1002 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1003 WARN_ON_ONCE(work->worker != worker);
1004
1005 if (!list_empty(&work->node))
1006 kthread_insert_work(worker, &fwork.work, work->node.next);
1007 else if (worker->current_work == work)
1008 kthread_insert_work(worker, &fwork.work,
1009 worker->work_list.next);
1010 else
1011 noop = true;
1012
1013 raw_spin_unlock_irq(&worker->lock);
1014
1015 if (!noop)
1016 wait_for_completion(&fwork.done);
1017}
1018EXPORT_SYMBOL_GPL(kthread_flush_work);
1019
1020/*
1021 * This function removes the work from the worker queue. Also it makes sure
1022 * that it won't get queued later via the delayed work's timer.
1023 *
1024 * The work might still be in use when this function finishes. See the
1025 * current_work proceed by the worker.
1026 *
1027 * Return: %true if @work was pending and successfully canceled,
1028 * %false if @work was not pending
1029 */
1030static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
1031 unsigned long *flags)
1032{
1033 /* Try to cancel the timer if exists. */
1034 if (is_dwork) {
1035 struct kthread_delayed_work *dwork =
1036 container_of(work, struct kthread_delayed_work, work);
1037 struct kthread_worker *worker = work->worker;
1038
1039 /*
1040 * del_timer_sync() must be called to make sure that the timer
1041 * callback is not running. The lock must be temporary released
1042 * to avoid a deadlock with the callback. In the meantime,
1043 * any queuing is blocked by setting the canceling counter.
1044 */
1045 work->canceling++;
1046 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1047 del_timer_sync(&dwork->timer);
1048 raw_spin_lock_irqsave(&worker->lock, *flags);
1049 work->canceling--;
1050 }
1051
1052 /*
1053 * Try to remove the work from a worker list. It might either
1054 * be from worker->work_list or from worker->delayed_work_list.
1055 */
1056 if (!list_empty(&work->node)) {
1057 list_del_init(&work->node);
1058 return true;
1059 }
1060
1061 return false;
1062}
1063
1064/**
1065 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1066 * @worker: kthread worker to use
1067 * @dwork: kthread delayed work to queue
1068 * @delay: number of jiffies to wait before queuing
1069 *
1070 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1071 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1072 * @work is guaranteed to be queued immediately.
1073 *
1074 * Return: %true if @dwork was pending and its timer was modified,
1075 * %false otherwise.
1076 *
1077 * A special case is when the work is being canceled in parallel.
1078 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1079 * or yet another kthread_mod_delayed_work() call. We let the other command
1080 * win and return %false here. The caller is supposed to synchronize these
1081 * operations a reasonable way.
1082 *
1083 * This function is safe to call from any context including IRQ handler.
1084 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1085 * for details.
1086 */
1087bool kthread_mod_delayed_work(struct kthread_worker *worker,
1088 struct kthread_delayed_work *dwork,
1089 unsigned long delay)
1090{
1091 struct kthread_work *work = &dwork->work;
1092 unsigned long flags;
1093 int ret = false;
1094
1095 raw_spin_lock_irqsave(&worker->lock, flags);
1096
1097 /* Do not bother with canceling when never queued. */
1098 if (!work->worker)
1099 goto fast_queue;
1100
1101 /* Work must not be used with >1 worker, see kthread_queue_work() */
1102 WARN_ON_ONCE(work->worker != worker);
1103
1104 /* Do not fight with another command that is canceling this work. */
1105 if (work->canceling)
1106 goto out;
1107
1108 ret = __kthread_cancel_work(work, true, &flags);
1109fast_queue:
1110 __kthread_queue_delayed_work(worker, dwork, delay);
1111out:
1112 raw_spin_unlock_irqrestore(&worker->lock, flags);
1113 return ret;
1114}
1115EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1116
1117static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1118{
1119 struct kthread_worker *worker = work->worker;
1120 unsigned long flags;
1121 int ret = false;
1122
1123 if (!worker)
1124 goto out;
1125
1126 raw_spin_lock_irqsave(&worker->lock, flags);
1127 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1128 WARN_ON_ONCE(work->worker != worker);
1129
1130 ret = __kthread_cancel_work(work, is_dwork, &flags);
1131
1132 if (worker->current_work != work)
1133 goto out_fast;
1134
1135 /*
1136 * The work is in progress and we need to wait with the lock released.
1137 * In the meantime, block any queuing by setting the canceling counter.
1138 */
1139 work->canceling++;
1140 raw_spin_unlock_irqrestore(&worker->lock, flags);
1141 kthread_flush_work(work);
1142 raw_spin_lock_irqsave(&worker->lock, flags);
1143 work->canceling--;
1144
1145out_fast:
1146 raw_spin_unlock_irqrestore(&worker->lock, flags);
1147out:
1148 return ret;
1149}
1150
1151/**
1152 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1153 * @work: the kthread work to cancel
1154 *
1155 * Cancel @work and wait for its execution to finish. This function
1156 * can be used even if the work re-queues itself. On return from this
1157 * function, @work is guaranteed to be not pending or executing on any CPU.
1158 *
1159 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1160 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1161 *
1162 * The caller must ensure that the worker on which @work was last
1163 * queued can't be destroyed before this function returns.
1164 *
1165 * Return: %true if @work was pending, %false otherwise.
1166 */
1167bool kthread_cancel_work_sync(struct kthread_work *work)
1168{
1169 return __kthread_cancel_work_sync(work, false);
1170}
1171EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1172
1173/**
1174 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1175 * wait for it to finish.
1176 * @dwork: the kthread delayed work to cancel
1177 *
1178 * This is kthread_cancel_work_sync() for delayed works.
1179 *
1180 * Return: %true if @dwork was pending, %false otherwise.
1181 */
1182bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1183{
1184 return __kthread_cancel_work_sync(&dwork->work, true);
1185}
1186EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1187
1188/**
1189 * kthread_flush_worker - flush all current works on a kthread_worker
1190 * @worker: worker to flush
1191 *
1192 * Wait until all currently executing or pending works on @worker are
1193 * finished.
1194 */
1195void kthread_flush_worker(struct kthread_worker *worker)
1196{
1197 struct kthread_flush_work fwork = {
1198 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1199 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1200 };
1201
1202 kthread_queue_work(worker, &fwork.work);
1203 wait_for_completion(&fwork.done);
1204}
1205EXPORT_SYMBOL_GPL(kthread_flush_worker);
1206
1207/**
1208 * kthread_destroy_worker - destroy a kthread worker
1209 * @worker: worker to be destroyed
1210 *
1211 * Flush and destroy @worker. The simple flush is enough because the kthread
1212 * worker API is used only in trivial scenarios. There are no multi-step state
1213 * machines needed.
1214 */
1215void kthread_destroy_worker(struct kthread_worker *worker)
1216{
1217 struct task_struct *task;
1218
1219 task = worker->task;
1220 if (WARN_ON(!task))
1221 return;
1222
1223 kthread_flush_worker(worker);
1224 kthread_stop(task);
1225 WARN_ON(!list_empty(&worker->work_list));
1226 kfree(worker);
1227}
1228EXPORT_SYMBOL(kthread_destroy_worker);
1229
1230/**
1231 * kthread_use_mm - make the calling kthread operate on an address space
1232 * @mm: address space to operate on
1233 */
1234void kthread_use_mm(struct mm_struct *mm)
1235{
1236 struct mm_struct *active_mm;
1237 struct task_struct *tsk = current;
1238
1239 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1240 WARN_ON_ONCE(tsk->mm);
1241
1242 task_lock(tsk);
1243 /* Hold off tlb flush IPIs while switching mm's */
1244 local_irq_disable();
1245 active_mm = tsk->active_mm;
1246 if (active_mm != mm) {
1247 mmgrab(mm);
1248 tsk->active_mm = mm;
1249 }
1250 tsk->mm = mm;
1251 switch_mm_irqs_off(active_mm, mm, tsk);
1252 local_irq_enable();
1253 task_unlock(tsk);
1254#ifdef finish_arch_post_lock_switch
1255 finish_arch_post_lock_switch();
1256#endif
1257
1258 if (active_mm != mm)
1259 mmdrop(active_mm);
1260
1261 to_kthread(tsk)->oldfs = force_uaccess_begin();
1262}
1263EXPORT_SYMBOL_GPL(kthread_use_mm);
1264
1265/**
1266 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1267 * @mm: address space to operate on
1268 */
1269void kthread_unuse_mm(struct mm_struct *mm)
1270{
1271 struct task_struct *tsk = current;
1272
1273 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1274 WARN_ON_ONCE(!tsk->mm);
1275
1276 force_uaccess_end(to_kthread(tsk)->oldfs);
1277
1278 task_lock(tsk);
1279 sync_mm_rss(mm);
1280 local_irq_disable();
1281 tsk->mm = NULL;
1282 /* active_mm is still 'mm' */
1283 enter_lazy_tlb(mm, tsk);
1284 local_irq_enable();
1285 task_unlock(tsk);
1286}
1287EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1288
1289#ifdef CONFIG_BLK_CGROUP
1290/**
1291 * kthread_associate_blkcg - associate blkcg to current kthread
1292 * @css: the cgroup info
1293 *
1294 * Current thread must be a kthread. The thread is running jobs on behalf of
1295 * other threads. In some cases, we expect the jobs attach cgroup info of
1296 * original threads instead of that of current thread. This function stores
1297 * original thread's cgroup info in current kthread context for later
1298 * retrieval.
1299 */
1300void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1301{
1302 struct kthread *kthread;
1303
1304 if (!(current->flags & PF_KTHREAD))
1305 return;
1306 kthread = to_kthread(current);
1307 if (!kthread)
1308 return;
1309
1310 if (kthread->blkcg_css) {
1311 css_put(kthread->blkcg_css);
1312 kthread->blkcg_css = NULL;
1313 }
1314 if (css) {
1315 css_get(css);
1316 kthread->blkcg_css = css;
1317 }
1318}
1319EXPORT_SYMBOL(kthread_associate_blkcg);
1320
1321/**
1322 * kthread_blkcg - get associated blkcg css of current kthread
1323 *
1324 * Current thread must be a kthread.
1325 */
1326struct cgroup_subsys_state *kthread_blkcg(void)
1327{
1328 struct kthread *kthread;
1329
1330 if (current->flags & PF_KTHREAD) {
1331 kthread = to_kthread(current);
1332 if (kthread)
1333 return kthread->blkcg_css;
1334 }
1335 return NULL;
1336}
1337EXPORT_SYMBOL(kthread_blkcg);
1338#endif