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 char *full_name;
42 int (*threadfn)(void *data);
43 void *data;
44 int node;
45
46 /* Result passed back to kthread_create() from kthreadd. */
47 struct task_struct *result;
48 struct completion *done;
49
50 struct list_head list;
51};
52
53struct kthread {
54 unsigned long flags;
55 unsigned int cpu;
56 int result;
57 int (*threadfn)(void *);
58 void *data;
59 struct completion parked;
60 struct completion exited;
61#ifdef CONFIG_BLK_CGROUP
62 struct cgroup_subsys_state *blkcg_css;
63#endif
64 /* To store the full name if task comm is truncated. */
65 char *full_name;
66};
67
68enum KTHREAD_BITS {
69 KTHREAD_IS_PER_CPU = 0,
70 KTHREAD_SHOULD_STOP,
71 KTHREAD_SHOULD_PARK,
72};
73
74static inline struct kthread *to_kthread(struct task_struct *k)
75{
76 WARN_ON(!(k->flags & PF_KTHREAD));
77 return k->worker_private;
78}
79
80/*
81 * Variant of to_kthread() that doesn't assume @p is a kthread.
82 *
83 * Per construction; when:
84 *
85 * (p->flags & PF_KTHREAD) && p->worker_private
86 *
87 * the task is both a kthread and struct kthread is persistent. However
88 * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and
89 * begin_new_exec()).
90 */
91static inline struct kthread *__to_kthread(struct task_struct *p)
92{
93 void *kthread = p->worker_private;
94 if (kthread && !(p->flags & PF_KTHREAD))
95 kthread = NULL;
96 return kthread;
97}
98
99void get_kthread_comm(char *buf, size_t buf_size, struct task_struct *tsk)
100{
101 struct kthread *kthread = to_kthread(tsk);
102
103 if (!kthread || !kthread->full_name) {
104 strscpy(buf, tsk->comm, buf_size);
105 return;
106 }
107
108 strscpy_pad(buf, kthread->full_name, buf_size);
109}
110
111bool set_kthread_struct(struct task_struct *p)
112{
113 struct kthread *kthread;
114
115 if (WARN_ON_ONCE(to_kthread(p)))
116 return false;
117
118 kthread = kzalloc(sizeof(*kthread), GFP_KERNEL);
119 if (!kthread)
120 return false;
121
122 init_completion(&kthread->exited);
123 init_completion(&kthread->parked);
124 p->vfork_done = &kthread->exited;
125
126 p->worker_private = kthread;
127 return true;
128}
129
130void free_kthread_struct(struct task_struct *k)
131{
132 struct kthread *kthread;
133
134 /*
135 * Can be NULL if kmalloc() in set_kthread_struct() failed.
136 */
137 kthread = to_kthread(k);
138 if (!kthread)
139 return;
140
141#ifdef CONFIG_BLK_CGROUP
142 WARN_ON_ONCE(kthread->blkcg_css);
143#endif
144 k->worker_private = NULL;
145 kfree(kthread->full_name);
146 kfree(kthread);
147}
148
149/**
150 * kthread_should_stop - should this kthread return now?
151 *
152 * When someone calls kthread_stop() on your kthread, it will be woken
153 * and this will return true. You should then return, and your return
154 * value will be passed through to kthread_stop().
155 */
156bool kthread_should_stop(void)
157{
158 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
159}
160EXPORT_SYMBOL(kthread_should_stop);
161
162static bool __kthread_should_park(struct task_struct *k)
163{
164 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
165}
166
167/**
168 * kthread_should_park - should this kthread park now?
169 *
170 * When someone calls kthread_park() on your kthread, it will be woken
171 * and this will return true. You should then do the necessary
172 * cleanup and call kthread_parkme()
173 *
174 * Similar to kthread_should_stop(), but this keeps the thread alive
175 * and in a park position. kthread_unpark() "restarts" the thread and
176 * calls the thread function again.
177 */
178bool kthread_should_park(void)
179{
180 return __kthread_should_park(current);
181}
182EXPORT_SYMBOL_GPL(kthread_should_park);
183
184bool kthread_should_stop_or_park(void)
185{
186 struct kthread *kthread = __to_kthread(current);
187
188 if (!kthread)
189 return false;
190
191 return kthread->flags & (BIT(KTHREAD_SHOULD_STOP) | BIT(KTHREAD_SHOULD_PARK));
192}
193
194/**
195 * kthread_freezable_should_stop - should this freezable kthread return now?
196 * @was_frozen: optional out parameter, indicates whether %current was frozen
197 *
198 * kthread_should_stop() for freezable kthreads, which will enter
199 * refrigerator if necessary. This function is safe from kthread_stop() /
200 * freezer deadlock and freezable kthreads should use this function instead
201 * of calling try_to_freeze() directly.
202 */
203bool kthread_freezable_should_stop(bool *was_frozen)
204{
205 bool frozen = false;
206
207 might_sleep();
208
209 if (unlikely(freezing(current)))
210 frozen = __refrigerator(true);
211
212 if (was_frozen)
213 *was_frozen = frozen;
214
215 return kthread_should_stop();
216}
217EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
218
219/**
220 * kthread_func - return the function specified on kthread creation
221 * @task: kthread task in question
222 *
223 * Returns NULL if the task is not a kthread.
224 */
225void *kthread_func(struct task_struct *task)
226{
227 struct kthread *kthread = __to_kthread(task);
228 if (kthread)
229 return kthread->threadfn;
230 return NULL;
231}
232EXPORT_SYMBOL_GPL(kthread_func);
233
234/**
235 * kthread_data - return data value specified on kthread creation
236 * @task: kthread task in question
237 *
238 * Return the data value specified when kthread @task was created.
239 * The caller is responsible for ensuring the validity of @task when
240 * calling this function.
241 */
242void *kthread_data(struct task_struct *task)
243{
244 return to_kthread(task)->data;
245}
246EXPORT_SYMBOL_GPL(kthread_data);
247
248/**
249 * kthread_probe_data - speculative version of kthread_data()
250 * @task: possible kthread task in question
251 *
252 * @task could be a kthread task. Return the data value specified when it
253 * was created if accessible. If @task isn't a kthread task or its data is
254 * inaccessible for any reason, %NULL is returned. This function requires
255 * that @task itself is safe to dereference.
256 */
257void *kthread_probe_data(struct task_struct *task)
258{
259 struct kthread *kthread = __to_kthread(task);
260 void *data = NULL;
261
262 if (kthread)
263 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
264 return data;
265}
266
267static void __kthread_parkme(struct kthread *self)
268{
269 for (;;) {
270 /*
271 * TASK_PARKED is a special state; we must serialize against
272 * possible pending wakeups to avoid store-store collisions on
273 * task->state.
274 *
275 * Such a collision might possibly result in the task state
276 * changin from TASK_PARKED and us failing the
277 * wait_task_inactive() in kthread_park().
278 */
279 set_special_state(TASK_PARKED);
280 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
281 break;
282
283 /*
284 * Thread is going to call schedule(), do not preempt it,
285 * or the caller of kthread_park() may spend more time in
286 * wait_task_inactive().
287 */
288 preempt_disable();
289 complete(&self->parked);
290 schedule_preempt_disabled();
291 preempt_enable();
292 }
293 __set_current_state(TASK_RUNNING);
294}
295
296void kthread_parkme(void)
297{
298 __kthread_parkme(to_kthread(current));
299}
300EXPORT_SYMBOL_GPL(kthread_parkme);
301
302/**
303 * kthread_exit - Cause the current kthread return @result to kthread_stop().
304 * @result: The integer value to return to kthread_stop().
305 *
306 * While kthread_exit can be called directly, it exists so that
307 * functions which do some additional work in non-modular code such as
308 * module_put_and_kthread_exit can be implemented.
309 *
310 * Does not return.
311 */
312void __noreturn kthread_exit(long result)
313{
314 struct kthread *kthread = to_kthread(current);
315 kthread->result = result;
316 do_exit(0);
317}
318EXPORT_SYMBOL(kthread_exit);
319
320/**
321 * kthread_complete_and_exit - Exit the current kthread.
322 * @comp: Completion to complete
323 * @code: The integer value to return to kthread_stop().
324 *
325 * If present, complete @comp and then return code to kthread_stop().
326 *
327 * A kernel thread whose module may be removed after the completion of
328 * @comp can use this function to exit safely.
329 *
330 * Does not return.
331 */
332void __noreturn kthread_complete_and_exit(struct completion *comp, long code)
333{
334 if (comp)
335 complete(comp);
336
337 kthread_exit(code);
338}
339EXPORT_SYMBOL(kthread_complete_and_exit);
340
341static int kthread(void *_create)
342{
343 static const struct sched_param param = { .sched_priority = 0 };
344 /* Copy data: it's on kthread's stack */
345 struct kthread_create_info *create = _create;
346 int (*threadfn)(void *data) = create->threadfn;
347 void *data = create->data;
348 struct completion *done;
349 struct kthread *self;
350 int ret;
351
352 self = to_kthread(current);
353
354 /* Release the structure when caller killed by a fatal signal. */
355 done = xchg(&create->done, NULL);
356 if (!done) {
357 kfree(create->full_name);
358 kfree(create);
359 kthread_exit(-EINTR);
360 }
361
362 self->full_name = create->full_name;
363 self->threadfn = threadfn;
364 self->data = data;
365
366 /*
367 * The new thread inherited kthreadd's priority and CPU mask. Reset
368 * back to default in case they have been changed.
369 */
370 sched_setscheduler_nocheck(current, SCHED_NORMAL, ¶m);
371 set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_TYPE_KTHREAD));
372
373 /* OK, tell user we're spawned, wait for stop or wakeup */
374 __set_current_state(TASK_UNINTERRUPTIBLE);
375 create->result = current;
376 /*
377 * Thread is going to call schedule(), do not preempt it,
378 * or the creator may spend more time in wait_task_inactive().
379 */
380 preempt_disable();
381 complete(done);
382 schedule_preempt_disabled();
383 preempt_enable();
384
385 ret = -EINTR;
386 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
387 cgroup_kthread_ready();
388 __kthread_parkme(self);
389 ret = threadfn(data);
390 }
391 kthread_exit(ret);
392}
393
394/* called from kernel_clone() to get node information for about to be created task */
395int tsk_fork_get_node(struct task_struct *tsk)
396{
397#ifdef CONFIG_NUMA
398 if (tsk == kthreadd_task)
399 return tsk->pref_node_fork;
400#endif
401 return NUMA_NO_NODE;
402}
403
404static void create_kthread(struct kthread_create_info *create)
405{
406 int pid;
407
408#ifdef CONFIG_NUMA
409 current->pref_node_fork = create->node;
410#endif
411 /* We want our own signal handler (we take no signals by default). */
412 pid = kernel_thread(kthread, create, create->full_name,
413 CLONE_FS | CLONE_FILES | SIGCHLD);
414 if (pid < 0) {
415 /* Release the structure when caller killed by a fatal signal. */
416 struct completion *done = xchg(&create->done, NULL);
417
418 kfree(create->full_name);
419 if (!done) {
420 kfree(create);
421 return;
422 }
423 create->result = ERR_PTR(pid);
424 complete(done);
425 }
426}
427
428static __printf(4, 0)
429struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
430 void *data, int node,
431 const char namefmt[],
432 va_list args)
433{
434 DECLARE_COMPLETION_ONSTACK(done);
435 struct task_struct *task;
436 struct kthread_create_info *create = kmalloc(sizeof(*create),
437 GFP_KERNEL);
438
439 if (!create)
440 return ERR_PTR(-ENOMEM);
441 create->threadfn = threadfn;
442 create->data = data;
443 create->node = node;
444 create->done = &done;
445 create->full_name = kvasprintf(GFP_KERNEL, namefmt, args);
446 if (!create->full_name) {
447 task = ERR_PTR(-ENOMEM);
448 goto free_create;
449 }
450
451 spin_lock(&kthread_create_lock);
452 list_add_tail(&create->list, &kthread_create_list);
453 spin_unlock(&kthread_create_lock);
454
455 wake_up_process(kthreadd_task);
456 /*
457 * Wait for completion in killable state, for I might be chosen by
458 * the OOM killer while kthreadd is trying to allocate memory for
459 * new kernel thread.
460 */
461 if (unlikely(wait_for_completion_killable(&done))) {
462 /*
463 * If I was killed by a fatal signal before kthreadd (or new
464 * kernel thread) calls complete(), leave the cleanup of this
465 * structure to that thread.
466 */
467 if (xchg(&create->done, NULL))
468 return ERR_PTR(-EINTR);
469 /*
470 * kthreadd (or new kernel thread) will call complete()
471 * shortly.
472 */
473 wait_for_completion(&done);
474 }
475 task = create->result;
476free_create:
477 kfree(create);
478 return task;
479}
480
481/**
482 * kthread_create_on_node - create a kthread.
483 * @threadfn: the function to run until signal_pending(current).
484 * @data: data ptr for @threadfn.
485 * @node: task and thread structures for the thread are allocated on this node
486 * @namefmt: printf-style name for the thread.
487 *
488 * Description: This helper function creates and names a kernel
489 * thread. The thread will be stopped: use wake_up_process() to start
490 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
491 * is affine to all CPUs.
492 *
493 * If thread is going to be bound on a particular cpu, give its node
494 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
495 * When woken, the thread will run @threadfn() with @data as its
496 * argument. @threadfn() can either return directly if it is a
497 * standalone thread for which no one will call kthread_stop(), or
498 * return when 'kthread_should_stop()' is true (which means
499 * kthread_stop() has been called). The return value should be zero
500 * or a negative error number; it will be passed to kthread_stop().
501 *
502 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
503 */
504struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
505 void *data, int node,
506 const char namefmt[],
507 ...)
508{
509 struct task_struct *task;
510 va_list args;
511
512 va_start(args, namefmt);
513 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
514 va_end(args);
515
516 return task;
517}
518EXPORT_SYMBOL(kthread_create_on_node);
519
520static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state)
521{
522 unsigned long flags;
523
524 if (!wait_task_inactive(p, state)) {
525 WARN_ON(1);
526 return;
527 }
528
529 /* It's safe because the task is inactive. */
530 raw_spin_lock_irqsave(&p->pi_lock, flags);
531 do_set_cpus_allowed(p, mask);
532 p->flags |= PF_NO_SETAFFINITY;
533 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
534}
535
536static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state)
537{
538 __kthread_bind_mask(p, cpumask_of(cpu), state);
539}
540
541void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
542{
543 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
544}
545
546/**
547 * kthread_bind - bind a just-created kthread to a cpu.
548 * @p: thread created by kthread_create().
549 * @cpu: cpu (might not be online, must be possible) for @k to run on.
550 *
551 * Description: This function is equivalent to set_cpus_allowed(),
552 * except that @cpu doesn't need to be online, and the thread must be
553 * stopped (i.e., just returned from kthread_create()).
554 */
555void kthread_bind(struct task_struct *p, unsigned int cpu)
556{
557 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
558}
559EXPORT_SYMBOL(kthread_bind);
560
561/**
562 * kthread_create_on_cpu - Create a cpu bound kthread
563 * @threadfn: the function to run until signal_pending(current).
564 * @data: data ptr for @threadfn.
565 * @cpu: The cpu on which the thread should be bound,
566 * @namefmt: printf-style name for the thread. Format is restricted
567 * to "name.*%u". Code fills in cpu number.
568 *
569 * Description: This helper function creates and names a kernel thread
570 */
571struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
572 void *data, unsigned int cpu,
573 const char *namefmt)
574{
575 struct task_struct *p;
576
577 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
578 cpu);
579 if (IS_ERR(p))
580 return p;
581 kthread_bind(p, cpu);
582 /* CPU hotplug need to bind once again when unparking the thread. */
583 to_kthread(p)->cpu = cpu;
584 return p;
585}
586EXPORT_SYMBOL(kthread_create_on_cpu);
587
588void kthread_set_per_cpu(struct task_struct *k, int cpu)
589{
590 struct kthread *kthread = to_kthread(k);
591 if (!kthread)
592 return;
593
594 WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
595
596 if (cpu < 0) {
597 clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
598 return;
599 }
600
601 kthread->cpu = cpu;
602 set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
603}
604
605bool kthread_is_per_cpu(struct task_struct *p)
606{
607 struct kthread *kthread = __to_kthread(p);
608 if (!kthread)
609 return false;
610
611 return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
612}
613
614/**
615 * kthread_unpark - unpark a thread created by kthread_create().
616 * @k: thread created by kthread_create().
617 *
618 * Sets kthread_should_park() for @k to return false, wakes it, and
619 * waits for it to return. If the thread is marked percpu then its
620 * bound to the cpu again.
621 */
622void kthread_unpark(struct task_struct *k)
623{
624 struct kthread *kthread = to_kthread(k);
625
626 if (!test_bit(KTHREAD_SHOULD_PARK, &kthread->flags))
627 return;
628 /*
629 * Newly created kthread was parked when the CPU was offline.
630 * The binding was lost and we need to set it again.
631 */
632 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
633 __kthread_bind(k, kthread->cpu, TASK_PARKED);
634
635 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
636 /*
637 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
638 */
639 wake_up_state(k, TASK_PARKED);
640}
641EXPORT_SYMBOL_GPL(kthread_unpark);
642
643/**
644 * kthread_park - park a thread created by kthread_create().
645 * @k: thread created by kthread_create().
646 *
647 * Sets kthread_should_park() for @k to return true, wakes it, and
648 * waits for it to return. This can also be called after kthread_create()
649 * instead of calling wake_up_process(): the thread will park without
650 * calling threadfn().
651 *
652 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
653 * If called by the kthread itself just the park bit is set.
654 */
655int kthread_park(struct task_struct *k)
656{
657 struct kthread *kthread = to_kthread(k);
658
659 if (WARN_ON(k->flags & PF_EXITING))
660 return -ENOSYS;
661
662 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
663 return -EBUSY;
664
665 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
666 if (k != current) {
667 wake_up_process(k);
668 /*
669 * Wait for __kthread_parkme() to complete(), this means we
670 * _will_ have TASK_PARKED and are about to call schedule().
671 */
672 wait_for_completion(&kthread->parked);
673 /*
674 * Now wait for that schedule() to complete and the task to
675 * get scheduled out.
676 */
677 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
678 }
679
680 return 0;
681}
682EXPORT_SYMBOL_GPL(kthread_park);
683
684/**
685 * kthread_stop - stop a thread created by kthread_create().
686 * @k: thread created by kthread_create().
687 *
688 * Sets kthread_should_stop() for @k to return true, wakes it, and
689 * waits for it to exit. This can also be called after kthread_create()
690 * instead of calling wake_up_process(): the thread will exit without
691 * calling threadfn().
692 *
693 * If threadfn() may call kthread_exit() itself, the caller must ensure
694 * task_struct can't go away.
695 *
696 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
697 * was never called.
698 */
699int kthread_stop(struct task_struct *k)
700{
701 struct kthread *kthread;
702 int ret;
703
704 trace_sched_kthread_stop(k);
705
706 get_task_struct(k);
707 kthread = to_kthread(k);
708 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
709 kthread_unpark(k);
710 set_tsk_thread_flag(k, TIF_NOTIFY_SIGNAL);
711 wake_up_process(k);
712 wait_for_completion(&kthread->exited);
713 ret = kthread->result;
714 put_task_struct(k);
715
716 trace_sched_kthread_stop_ret(ret);
717 return ret;
718}
719EXPORT_SYMBOL(kthread_stop);
720
721/**
722 * kthread_stop_put - stop a thread and put its task struct
723 * @k: thread created by kthread_create().
724 *
725 * Stops a thread created by kthread_create() and put its task_struct.
726 * Only use when holding an extra task struct reference obtained by
727 * calling get_task_struct().
728 */
729int kthread_stop_put(struct task_struct *k)
730{
731 int ret;
732
733 ret = kthread_stop(k);
734 put_task_struct(k);
735 return ret;
736}
737EXPORT_SYMBOL(kthread_stop_put);
738
739int kthreadd(void *unused)
740{
741 struct task_struct *tsk = current;
742
743 /* Setup a clean context for our children to inherit. */
744 set_task_comm(tsk, "kthreadd");
745 ignore_signals(tsk);
746 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_TYPE_KTHREAD));
747 set_mems_allowed(node_states[N_MEMORY]);
748
749 current->flags |= PF_NOFREEZE;
750 cgroup_init_kthreadd();
751
752 for (;;) {
753 set_current_state(TASK_INTERRUPTIBLE);
754 if (list_empty(&kthread_create_list))
755 schedule();
756 __set_current_state(TASK_RUNNING);
757
758 spin_lock(&kthread_create_lock);
759 while (!list_empty(&kthread_create_list)) {
760 struct kthread_create_info *create;
761
762 create = list_entry(kthread_create_list.next,
763 struct kthread_create_info, list);
764 list_del_init(&create->list);
765 spin_unlock(&kthread_create_lock);
766
767 create_kthread(create);
768
769 spin_lock(&kthread_create_lock);
770 }
771 spin_unlock(&kthread_create_lock);
772 }
773
774 return 0;
775}
776
777void __kthread_init_worker(struct kthread_worker *worker,
778 const char *name,
779 struct lock_class_key *key)
780{
781 memset(worker, 0, sizeof(struct kthread_worker));
782 raw_spin_lock_init(&worker->lock);
783 lockdep_set_class_and_name(&worker->lock, key, name);
784 INIT_LIST_HEAD(&worker->work_list);
785 INIT_LIST_HEAD(&worker->delayed_work_list);
786}
787EXPORT_SYMBOL_GPL(__kthread_init_worker);
788
789/**
790 * kthread_worker_fn - kthread function to process kthread_worker
791 * @worker_ptr: pointer to initialized kthread_worker
792 *
793 * This function implements the main cycle of kthread worker. It processes
794 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
795 * is empty.
796 *
797 * The works are not allowed to keep any locks, disable preemption or interrupts
798 * when they finish. There is defined a safe point for freezing when one work
799 * finishes and before a new one is started.
800 *
801 * Also the works must not be handled by more than one worker at the same time,
802 * see also kthread_queue_work().
803 */
804int kthread_worker_fn(void *worker_ptr)
805{
806 struct kthread_worker *worker = worker_ptr;
807 struct kthread_work *work;
808
809 /*
810 * FIXME: Update the check and remove the assignment when all kthread
811 * worker users are created using kthread_create_worker*() functions.
812 */
813 WARN_ON(worker->task && worker->task != current);
814 worker->task = current;
815
816 if (worker->flags & KTW_FREEZABLE)
817 set_freezable();
818
819repeat:
820 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
821
822 if (kthread_should_stop()) {
823 __set_current_state(TASK_RUNNING);
824 raw_spin_lock_irq(&worker->lock);
825 worker->task = NULL;
826 raw_spin_unlock_irq(&worker->lock);
827 return 0;
828 }
829
830 work = NULL;
831 raw_spin_lock_irq(&worker->lock);
832 if (!list_empty(&worker->work_list)) {
833 work = list_first_entry(&worker->work_list,
834 struct kthread_work, node);
835 list_del_init(&work->node);
836 }
837 worker->current_work = work;
838 raw_spin_unlock_irq(&worker->lock);
839
840 if (work) {
841 kthread_work_func_t func = work->func;
842 __set_current_state(TASK_RUNNING);
843 trace_sched_kthread_work_execute_start(work);
844 work->func(work);
845 /*
846 * Avoid dereferencing work after this point. The trace
847 * event only cares about the address.
848 */
849 trace_sched_kthread_work_execute_end(work, func);
850 } else if (!freezing(current)) {
851 schedule();
852 } else {
853 /*
854 * Handle the case where the current remains
855 * TASK_INTERRUPTIBLE. try_to_freeze() expects
856 * the current to be TASK_RUNNING.
857 */
858 __set_current_state(TASK_RUNNING);
859 }
860
861 try_to_freeze();
862 cond_resched();
863 goto repeat;
864}
865EXPORT_SYMBOL_GPL(kthread_worker_fn);
866
867static __printf(3, 0) struct kthread_worker *
868__kthread_create_worker(int cpu, unsigned int flags,
869 const char namefmt[], va_list args)
870{
871 struct kthread_worker *worker;
872 struct task_struct *task;
873 int node = NUMA_NO_NODE;
874
875 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
876 if (!worker)
877 return ERR_PTR(-ENOMEM);
878
879 kthread_init_worker(worker);
880
881 if (cpu >= 0)
882 node = cpu_to_node(cpu);
883
884 task = __kthread_create_on_node(kthread_worker_fn, worker,
885 node, namefmt, args);
886 if (IS_ERR(task))
887 goto fail_task;
888
889 if (cpu >= 0)
890 kthread_bind(task, cpu);
891
892 worker->flags = flags;
893 worker->task = task;
894 wake_up_process(task);
895 return worker;
896
897fail_task:
898 kfree(worker);
899 return ERR_CAST(task);
900}
901
902/**
903 * kthread_create_worker - create a kthread worker
904 * @flags: flags modifying the default behavior of the worker
905 * @namefmt: printf-style name for the kthread worker (task).
906 *
907 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
908 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
909 * when the caller was killed by a fatal signal.
910 */
911struct kthread_worker *
912kthread_create_worker(unsigned int flags, const char namefmt[], ...)
913{
914 struct kthread_worker *worker;
915 va_list args;
916
917 va_start(args, namefmt);
918 worker = __kthread_create_worker(-1, flags, namefmt, args);
919 va_end(args);
920
921 return worker;
922}
923EXPORT_SYMBOL(kthread_create_worker);
924
925/**
926 * kthread_create_worker_on_cpu - create a kthread worker and bind it
927 * to a given CPU and the associated NUMA node.
928 * @cpu: CPU number
929 * @flags: flags modifying the default behavior of the worker
930 * @namefmt: printf-style name for the kthread worker (task).
931 *
932 * Use a valid CPU number if you want to bind the kthread worker
933 * to the given CPU and the associated NUMA node.
934 *
935 * A good practice is to add the cpu number also into the worker name.
936 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
937 *
938 * CPU hotplug:
939 * The kthread worker API is simple and generic. It just provides a way
940 * to create, use, and destroy workers.
941 *
942 * It is up to the API user how to handle CPU hotplug. They have to decide
943 * how to handle pending work items, prevent queuing new ones, and
944 * restore the functionality when the CPU goes off and on. There are a
945 * few catches:
946 *
947 * - CPU affinity gets lost when it is scheduled on an offline CPU.
948 *
949 * - The worker might not exist when the CPU was off when the user
950 * created the workers.
951 *
952 * Good practice is to implement two CPU hotplug callbacks and to
953 * destroy/create the worker when the CPU goes down/up.
954 *
955 * Return:
956 * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
957 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
958 * when the caller was killed by a fatal signal.
959 */
960struct kthread_worker *
961kthread_create_worker_on_cpu(int cpu, unsigned int flags,
962 const char namefmt[], ...)
963{
964 struct kthread_worker *worker;
965 va_list args;
966
967 va_start(args, namefmt);
968 worker = __kthread_create_worker(cpu, flags, namefmt, args);
969 va_end(args);
970
971 return worker;
972}
973EXPORT_SYMBOL(kthread_create_worker_on_cpu);
974
975/*
976 * Returns true when the work could not be queued at the moment.
977 * It happens when it is already pending in a worker list
978 * or when it is being cancelled.
979 */
980static inline bool queuing_blocked(struct kthread_worker *worker,
981 struct kthread_work *work)
982{
983 lockdep_assert_held(&worker->lock);
984
985 return !list_empty(&work->node) || work->canceling;
986}
987
988static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
989 struct kthread_work *work)
990{
991 lockdep_assert_held(&worker->lock);
992 WARN_ON_ONCE(!list_empty(&work->node));
993 /* Do not use a work with >1 worker, see kthread_queue_work() */
994 WARN_ON_ONCE(work->worker && work->worker != worker);
995}
996
997/* insert @work before @pos in @worker */
998static void kthread_insert_work(struct kthread_worker *worker,
999 struct kthread_work *work,
1000 struct list_head *pos)
1001{
1002 kthread_insert_work_sanity_check(worker, work);
1003
1004 trace_sched_kthread_work_queue_work(worker, work);
1005
1006 list_add_tail(&work->node, pos);
1007 work->worker = worker;
1008 if (!worker->current_work && likely(worker->task))
1009 wake_up_process(worker->task);
1010}
1011
1012/**
1013 * kthread_queue_work - queue a kthread_work
1014 * @worker: target kthread_worker
1015 * @work: kthread_work to queue
1016 *
1017 * Queue @work to work processor @task for async execution. @task
1018 * must have been created with kthread_worker_create(). Returns %true
1019 * if @work was successfully queued, %false if it was already pending.
1020 *
1021 * Reinitialize the work if it needs to be used by another worker.
1022 * For example, when the worker was stopped and started again.
1023 */
1024bool kthread_queue_work(struct kthread_worker *worker,
1025 struct kthread_work *work)
1026{
1027 bool ret = false;
1028 unsigned long flags;
1029
1030 raw_spin_lock_irqsave(&worker->lock, flags);
1031 if (!queuing_blocked(worker, work)) {
1032 kthread_insert_work(worker, work, &worker->work_list);
1033 ret = true;
1034 }
1035 raw_spin_unlock_irqrestore(&worker->lock, flags);
1036 return ret;
1037}
1038EXPORT_SYMBOL_GPL(kthread_queue_work);
1039
1040/**
1041 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
1042 * delayed work when the timer expires.
1043 * @t: pointer to the expired timer
1044 *
1045 * The format of the function is defined by struct timer_list.
1046 * It should have been called from irqsafe timer with irq already off.
1047 */
1048void kthread_delayed_work_timer_fn(struct timer_list *t)
1049{
1050 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
1051 struct kthread_work *work = &dwork->work;
1052 struct kthread_worker *worker = work->worker;
1053 unsigned long flags;
1054
1055 /*
1056 * This might happen when a pending work is reinitialized.
1057 * It means that it is used a wrong way.
1058 */
1059 if (WARN_ON_ONCE(!worker))
1060 return;
1061
1062 raw_spin_lock_irqsave(&worker->lock, flags);
1063 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1064 WARN_ON_ONCE(work->worker != worker);
1065
1066 /* Move the work from worker->delayed_work_list. */
1067 WARN_ON_ONCE(list_empty(&work->node));
1068 list_del_init(&work->node);
1069 if (!work->canceling)
1070 kthread_insert_work(worker, work, &worker->work_list);
1071
1072 raw_spin_unlock_irqrestore(&worker->lock, flags);
1073}
1074EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
1075
1076static void __kthread_queue_delayed_work(struct kthread_worker *worker,
1077 struct kthread_delayed_work *dwork,
1078 unsigned long delay)
1079{
1080 struct timer_list *timer = &dwork->timer;
1081 struct kthread_work *work = &dwork->work;
1082
1083 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
1084
1085 /*
1086 * If @delay is 0, queue @dwork->work immediately. This is for
1087 * both optimization and correctness. The earliest @timer can
1088 * expire is on the closest next tick and delayed_work users depend
1089 * on that there's no such delay when @delay is 0.
1090 */
1091 if (!delay) {
1092 kthread_insert_work(worker, work, &worker->work_list);
1093 return;
1094 }
1095
1096 /* Be paranoid and try to detect possible races already now. */
1097 kthread_insert_work_sanity_check(worker, work);
1098
1099 list_add(&work->node, &worker->delayed_work_list);
1100 work->worker = worker;
1101 timer->expires = jiffies + delay;
1102 add_timer(timer);
1103}
1104
1105/**
1106 * kthread_queue_delayed_work - queue the associated kthread work
1107 * after a delay.
1108 * @worker: target kthread_worker
1109 * @dwork: kthread_delayed_work to queue
1110 * @delay: number of jiffies to wait before queuing
1111 *
1112 * If the work has not been pending it starts a timer that will queue
1113 * the work after the given @delay. If @delay is zero, it queues the
1114 * work immediately.
1115 *
1116 * Return: %false if the @work has already been pending. It means that
1117 * either the timer was running or the work was queued. It returns %true
1118 * otherwise.
1119 */
1120bool kthread_queue_delayed_work(struct kthread_worker *worker,
1121 struct kthread_delayed_work *dwork,
1122 unsigned long delay)
1123{
1124 struct kthread_work *work = &dwork->work;
1125 unsigned long flags;
1126 bool ret = false;
1127
1128 raw_spin_lock_irqsave(&worker->lock, flags);
1129
1130 if (!queuing_blocked(worker, work)) {
1131 __kthread_queue_delayed_work(worker, dwork, delay);
1132 ret = true;
1133 }
1134
1135 raw_spin_unlock_irqrestore(&worker->lock, flags);
1136 return ret;
1137}
1138EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1139
1140struct kthread_flush_work {
1141 struct kthread_work work;
1142 struct completion done;
1143};
1144
1145static void kthread_flush_work_fn(struct kthread_work *work)
1146{
1147 struct kthread_flush_work *fwork =
1148 container_of(work, struct kthread_flush_work, work);
1149 complete(&fwork->done);
1150}
1151
1152/**
1153 * kthread_flush_work - flush a kthread_work
1154 * @work: work to flush
1155 *
1156 * If @work is queued or executing, wait for it to finish execution.
1157 */
1158void kthread_flush_work(struct kthread_work *work)
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 struct kthread_worker *worker;
1165 bool noop = false;
1166
1167 worker = work->worker;
1168 if (!worker)
1169 return;
1170
1171 raw_spin_lock_irq(&worker->lock);
1172 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1173 WARN_ON_ONCE(work->worker != worker);
1174
1175 if (!list_empty(&work->node))
1176 kthread_insert_work(worker, &fwork.work, work->node.next);
1177 else if (worker->current_work == work)
1178 kthread_insert_work(worker, &fwork.work,
1179 worker->work_list.next);
1180 else
1181 noop = true;
1182
1183 raw_spin_unlock_irq(&worker->lock);
1184
1185 if (!noop)
1186 wait_for_completion(&fwork.done);
1187}
1188EXPORT_SYMBOL_GPL(kthread_flush_work);
1189
1190/*
1191 * Make sure that the timer is neither set nor running and could
1192 * not manipulate the work list_head any longer.
1193 *
1194 * The function is called under worker->lock. The lock is temporary
1195 * released but the timer can't be set again in the meantime.
1196 */
1197static void kthread_cancel_delayed_work_timer(struct kthread_work *work,
1198 unsigned long *flags)
1199{
1200 struct kthread_delayed_work *dwork =
1201 container_of(work, struct kthread_delayed_work, work);
1202 struct kthread_worker *worker = work->worker;
1203
1204 /*
1205 * del_timer_sync() must be called to make sure that the timer
1206 * callback is not running. The lock must be temporary released
1207 * to avoid a deadlock with the callback. In the meantime,
1208 * any queuing is blocked by setting the canceling counter.
1209 */
1210 work->canceling++;
1211 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1212 del_timer_sync(&dwork->timer);
1213 raw_spin_lock_irqsave(&worker->lock, *flags);
1214 work->canceling--;
1215}
1216
1217/*
1218 * This function removes the work from the worker queue.
1219 *
1220 * It is called under worker->lock. The caller must make sure that
1221 * the timer used by delayed work is not running, e.g. by calling
1222 * kthread_cancel_delayed_work_timer().
1223 *
1224 * The work might still be in use when this function finishes. See the
1225 * current_work proceed by the worker.
1226 *
1227 * Return: %true if @work was pending and successfully canceled,
1228 * %false if @work was not pending
1229 */
1230static bool __kthread_cancel_work(struct kthread_work *work)
1231{
1232 /*
1233 * Try to remove the work from a worker list. It might either
1234 * be from worker->work_list or from worker->delayed_work_list.
1235 */
1236 if (!list_empty(&work->node)) {
1237 list_del_init(&work->node);
1238 return true;
1239 }
1240
1241 return false;
1242}
1243
1244/**
1245 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1246 * @worker: kthread worker to use
1247 * @dwork: kthread delayed work to queue
1248 * @delay: number of jiffies to wait before queuing
1249 *
1250 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1251 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1252 * @work is guaranteed to be queued immediately.
1253 *
1254 * Return: %false if @dwork was idle and queued, %true otherwise.
1255 *
1256 * A special case is when the work is being canceled in parallel.
1257 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1258 * or yet another kthread_mod_delayed_work() call. We let the other command
1259 * win and return %true here. The return value can be used for reference
1260 * counting and the number of queued works stays the same. Anyway, the caller
1261 * is supposed to synchronize these operations a reasonable way.
1262 *
1263 * This function is safe to call from any context including IRQ handler.
1264 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1265 * for details.
1266 */
1267bool kthread_mod_delayed_work(struct kthread_worker *worker,
1268 struct kthread_delayed_work *dwork,
1269 unsigned long delay)
1270{
1271 struct kthread_work *work = &dwork->work;
1272 unsigned long flags;
1273 int ret;
1274
1275 raw_spin_lock_irqsave(&worker->lock, flags);
1276
1277 /* Do not bother with canceling when never queued. */
1278 if (!work->worker) {
1279 ret = false;
1280 goto fast_queue;
1281 }
1282
1283 /* Work must not be used with >1 worker, see kthread_queue_work() */
1284 WARN_ON_ONCE(work->worker != worker);
1285
1286 /*
1287 * Temporary cancel the work but do not fight with another command
1288 * that is canceling the work as well.
1289 *
1290 * It is a bit tricky because of possible races with another
1291 * mod_delayed_work() and cancel_delayed_work() callers.
1292 *
1293 * The timer must be canceled first because worker->lock is released
1294 * when doing so. But the work can be removed from the queue (list)
1295 * only when it can be queued again so that the return value can
1296 * be used for reference counting.
1297 */
1298 kthread_cancel_delayed_work_timer(work, &flags);
1299 if (work->canceling) {
1300 /* The number of works in the queue does not change. */
1301 ret = true;
1302 goto out;
1303 }
1304 ret = __kthread_cancel_work(work);
1305
1306fast_queue:
1307 __kthread_queue_delayed_work(worker, dwork, delay);
1308out:
1309 raw_spin_unlock_irqrestore(&worker->lock, flags);
1310 return ret;
1311}
1312EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1313
1314static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1315{
1316 struct kthread_worker *worker = work->worker;
1317 unsigned long flags;
1318 int ret = false;
1319
1320 if (!worker)
1321 goto out;
1322
1323 raw_spin_lock_irqsave(&worker->lock, flags);
1324 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1325 WARN_ON_ONCE(work->worker != worker);
1326
1327 if (is_dwork)
1328 kthread_cancel_delayed_work_timer(work, &flags);
1329
1330 ret = __kthread_cancel_work(work);
1331
1332 if (worker->current_work != work)
1333 goto out_fast;
1334
1335 /*
1336 * The work is in progress and we need to wait with the lock released.
1337 * In the meantime, block any queuing by setting the canceling counter.
1338 */
1339 work->canceling++;
1340 raw_spin_unlock_irqrestore(&worker->lock, flags);
1341 kthread_flush_work(work);
1342 raw_spin_lock_irqsave(&worker->lock, flags);
1343 work->canceling--;
1344
1345out_fast:
1346 raw_spin_unlock_irqrestore(&worker->lock, flags);
1347out:
1348 return ret;
1349}
1350
1351/**
1352 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1353 * @work: the kthread work to cancel
1354 *
1355 * Cancel @work and wait for its execution to finish. This function
1356 * can be used even if the work re-queues itself. On return from this
1357 * function, @work is guaranteed to be not pending or executing on any CPU.
1358 *
1359 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1360 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1361 *
1362 * The caller must ensure that the worker on which @work was last
1363 * queued can't be destroyed before this function returns.
1364 *
1365 * Return: %true if @work was pending, %false otherwise.
1366 */
1367bool kthread_cancel_work_sync(struct kthread_work *work)
1368{
1369 return __kthread_cancel_work_sync(work, false);
1370}
1371EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1372
1373/**
1374 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1375 * wait for it to finish.
1376 * @dwork: the kthread delayed work to cancel
1377 *
1378 * This is kthread_cancel_work_sync() for delayed works.
1379 *
1380 * Return: %true if @dwork was pending, %false otherwise.
1381 */
1382bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1383{
1384 return __kthread_cancel_work_sync(&dwork->work, true);
1385}
1386EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1387
1388/**
1389 * kthread_flush_worker - flush all current works on a kthread_worker
1390 * @worker: worker to flush
1391 *
1392 * Wait until all currently executing or pending works on @worker are
1393 * finished.
1394 */
1395void kthread_flush_worker(struct kthread_worker *worker)
1396{
1397 struct kthread_flush_work fwork = {
1398 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1399 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1400 };
1401
1402 kthread_queue_work(worker, &fwork.work);
1403 wait_for_completion(&fwork.done);
1404}
1405EXPORT_SYMBOL_GPL(kthread_flush_worker);
1406
1407/**
1408 * kthread_destroy_worker - destroy a kthread worker
1409 * @worker: worker to be destroyed
1410 *
1411 * Flush and destroy @worker. The simple flush is enough because the kthread
1412 * worker API is used only in trivial scenarios. There are no multi-step state
1413 * machines needed.
1414 *
1415 * Note that this function is not responsible for handling delayed work, so
1416 * caller should be responsible for queuing or canceling all delayed work items
1417 * before invoke this function.
1418 */
1419void kthread_destroy_worker(struct kthread_worker *worker)
1420{
1421 struct task_struct *task;
1422
1423 task = worker->task;
1424 if (WARN_ON(!task))
1425 return;
1426
1427 kthread_flush_worker(worker);
1428 kthread_stop(task);
1429 WARN_ON(!list_empty(&worker->delayed_work_list));
1430 WARN_ON(!list_empty(&worker->work_list));
1431 kfree(worker);
1432}
1433EXPORT_SYMBOL(kthread_destroy_worker);
1434
1435/**
1436 * kthread_use_mm - make the calling kthread operate on an address space
1437 * @mm: address space to operate on
1438 */
1439void kthread_use_mm(struct mm_struct *mm)
1440{
1441 struct mm_struct *active_mm;
1442 struct task_struct *tsk = current;
1443
1444 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1445 WARN_ON_ONCE(tsk->mm);
1446
1447 /*
1448 * It is possible for mm to be the same as tsk->active_mm, but
1449 * we must still mmgrab(mm) and mmdrop_lazy_tlb(active_mm),
1450 * because these references are not equivalent.
1451 */
1452 mmgrab(mm);
1453
1454 task_lock(tsk);
1455 /* Hold off tlb flush IPIs while switching mm's */
1456 local_irq_disable();
1457 active_mm = tsk->active_mm;
1458 tsk->active_mm = mm;
1459 tsk->mm = mm;
1460 membarrier_update_current_mm(mm);
1461 switch_mm_irqs_off(active_mm, mm, tsk);
1462 local_irq_enable();
1463 task_unlock(tsk);
1464#ifdef finish_arch_post_lock_switch
1465 finish_arch_post_lock_switch();
1466#endif
1467
1468 /*
1469 * When a kthread starts operating on an address space, the loop
1470 * in membarrier_{private,global}_expedited() may not observe
1471 * that tsk->mm, and not issue an IPI. Membarrier requires a
1472 * memory barrier after storing to tsk->mm, before accessing
1473 * user-space memory. A full memory barrier for membarrier
1474 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1475 * mmdrop_lazy_tlb().
1476 */
1477 mmdrop_lazy_tlb(active_mm);
1478}
1479EXPORT_SYMBOL_GPL(kthread_use_mm);
1480
1481/**
1482 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1483 * @mm: address space to operate on
1484 */
1485void kthread_unuse_mm(struct mm_struct *mm)
1486{
1487 struct task_struct *tsk = current;
1488
1489 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1490 WARN_ON_ONCE(!tsk->mm);
1491
1492 task_lock(tsk);
1493 /*
1494 * When a kthread stops operating on an address space, the loop
1495 * in membarrier_{private,global}_expedited() may not observe
1496 * that tsk->mm, and not issue an IPI. Membarrier requires a
1497 * memory barrier after accessing user-space memory, before
1498 * clearing tsk->mm.
1499 */
1500 smp_mb__after_spinlock();
1501 local_irq_disable();
1502 tsk->mm = NULL;
1503 membarrier_update_current_mm(NULL);
1504 mmgrab_lazy_tlb(mm);
1505 /* active_mm is still 'mm' */
1506 enter_lazy_tlb(mm, tsk);
1507 local_irq_enable();
1508 task_unlock(tsk);
1509
1510 mmdrop(mm);
1511}
1512EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1513
1514#ifdef CONFIG_BLK_CGROUP
1515/**
1516 * kthread_associate_blkcg - associate blkcg to current kthread
1517 * @css: the cgroup info
1518 *
1519 * Current thread must be a kthread. The thread is running jobs on behalf of
1520 * other threads. In some cases, we expect the jobs attach cgroup info of
1521 * original threads instead of that of current thread. This function stores
1522 * original thread's cgroup info in current kthread context for later
1523 * retrieval.
1524 */
1525void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1526{
1527 struct kthread *kthread;
1528
1529 if (!(current->flags & PF_KTHREAD))
1530 return;
1531 kthread = to_kthread(current);
1532 if (!kthread)
1533 return;
1534
1535 if (kthread->blkcg_css) {
1536 css_put(kthread->blkcg_css);
1537 kthread->blkcg_css = NULL;
1538 }
1539 if (css) {
1540 css_get(css);
1541 kthread->blkcg_css = css;
1542 }
1543}
1544EXPORT_SYMBOL(kthread_associate_blkcg);
1545
1546/**
1547 * kthread_blkcg - get associated blkcg css of current kthread
1548 *
1549 * Current thread must be a kthread.
1550 */
1551struct cgroup_subsys_state *kthread_blkcg(void)
1552{
1553 struct kthread *kthread;
1554
1555 if (current->flags & PF_KTHREAD) {
1556 kthread = to_kthread(current);
1557 if (kthread)
1558 return kthread->blkcg_css;
1559 }
1560 return NULL;
1561}
1562#endif