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 
34 static DEFINE_SPINLOCK(kthread_create_lock);
35 static LIST_HEAD(kthread_create_list);
36 struct task_struct *kthreadd_task;
37 
38 struct 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 
53 struct 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 
68 enum KTHREAD_BITS {
69 	KTHREAD_IS_PER_CPU = 0,
70 	KTHREAD_SHOULD_STOP,
71 	KTHREAD_SHOULD_PARK,
72 };
73 
to_kthread(struct task_struct * k)74 static 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  */
__to_kthread(struct task_struct * p)91 static 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 
get_kthread_comm(char * buf,size_t buf_size,struct task_struct * tsk)99 void 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 		__get_task_comm(buf, buf_size, tsk);
105 		return;
106 	}
107 
108 	strscpy_pad(buf, kthread->full_name, buf_size);
109 }
110 
set_kthread_struct(struct task_struct * p)111 bool 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 
free_kthread_struct(struct task_struct * k)130 void 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  */
kthread_should_stop(void)156 bool kthread_should_stop(void)
157 {
158 	return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
159 }
160 EXPORT_SYMBOL(kthread_should_stop);
161 
__kthread_should_park(struct task_struct * k)162 static 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  */
kthread_should_park(void)178 bool kthread_should_park(void)
179 {
180 	return __kthread_should_park(current);
181 }
182 EXPORT_SYMBOL_GPL(kthread_should_park);
183 
kthread_should_stop_or_park(void)184 bool 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  */
kthread_freezable_should_stop(bool * was_frozen)203 bool 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 }
217 EXPORT_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  */
kthread_func(struct task_struct * task)225 void *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 }
232 EXPORT_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  */
kthread_data(struct task_struct * task)242 void *kthread_data(struct task_struct *task)
243 {
244 	return to_kthread(task)->data;
245 }
246 EXPORT_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  */
kthread_probe_data(struct task_struct * task)257 void *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 
__kthread_parkme(struct kthread * self)267 static 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 
kthread_parkme(void)296 void kthread_parkme(void)
297 {
298 	__kthread_parkme(to_kthread(current));
299 }
300 EXPORT_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  */
kthread_exit(long result)312 void __noreturn kthread_exit(long result)
313 {
314 	struct kthread *kthread = to_kthread(current);
315 	kthread->result = result;
316 	do_exit(0);
317 }
318 
319 /**
320  * kthread_complete_and_exit - Exit the current kthread.
321  * @comp: Completion to complete
322  * @code: The integer value to return to kthread_stop().
323  *
324  * If present, complete @comp and then return code to kthread_stop().
325  *
326  * A kernel thread whose module may be removed after the completion of
327  * @comp can use this function to exit safely.
328  *
329  * Does not return.
330  */
kthread_complete_and_exit(struct completion * comp,long code)331 void __noreturn kthread_complete_and_exit(struct completion *comp, long code)
332 {
333 	if (comp)
334 		complete(comp);
335 
336 	kthread_exit(code);
337 }
338 EXPORT_SYMBOL(kthread_complete_and_exit);
339 
kthread(void * _create)340 static int kthread(void *_create)
341 {
342 	static const struct sched_param param = { .sched_priority = 0 };
343 	/* Copy data: it's on kthread's stack */
344 	struct kthread_create_info *create = _create;
345 	int (*threadfn)(void *data) = create->threadfn;
346 	void *data = create->data;
347 	struct completion *done;
348 	struct kthread *self;
349 	int ret;
350 
351 	self = to_kthread(current);
352 
353 	/* Release the structure when caller killed by a fatal signal. */
354 	done = xchg(&create->done, NULL);
355 	if (!done) {
356 		kfree(create->full_name);
357 		kfree(create);
358 		kthread_exit(-EINTR);
359 	}
360 
361 	self->full_name = create->full_name;
362 	self->threadfn = threadfn;
363 	self->data = data;
364 
365 	/*
366 	 * The new thread inherited kthreadd's priority and CPU mask. Reset
367 	 * back to default in case they have been changed.
368 	 */
369 	sched_setscheduler_nocheck(current, SCHED_NORMAL, &param);
370 	set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_TYPE_KTHREAD));
371 
372 	/* OK, tell user we're spawned, wait for stop or wakeup */
373 	__set_current_state(TASK_UNINTERRUPTIBLE);
374 	create->result = current;
375 	/*
376 	 * Thread is going to call schedule(), do not preempt it,
377 	 * or the creator may spend more time in wait_task_inactive().
378 	 */
379 	preempt_disable();
380 	complete(done);
381 	schedule_preempt_disabled();
382 	preempt_enable();
383 
384 	ret = -EINTR;
385 	if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
386 		cgroup_kthread_ready();
387 		__kthread_parkme(self);
388 		ret = threadfn(data);
389 	}
390 	kthread_exit(ret);
391 }
392 
393 /* called from kernel_clone() to get node information for about to be created task */
tsk_fork_get_node(struct task_struct * tsk)394 int tsk_fork_get_node(struct task_struct *tsk)
395 {
396 #ifdef CONFIG_NUMA
397 	if (tsk == kthreadd_task)
398 		return tsk->pref_node_fork;
399 #endif
400 	return NUMA_NO_NODE;
401 }
402 
create_kthread(struct kthread_create_info * create)403 static void create_kthread(struct kthread_create_info *create)
404 {
405 	int pid;
406 
407 #ifdef CONFIG_NUMA
408 	current->pref_node_fork = create->node;
409 #endif
410 	/* We want our own signal handler (we take no signals by default). */
411 	pid = kernel_thread(kthread, create, create->full_name,
412 			    CLONE_FS | CLONE_FILES | SIGCHLD);
413 	if (pid < 0) {
414 		/* Release the structure when caller killed by a fatal signal. */
415 		struct completion *done = xchg(&create->done, NULL);
416 
417 		kfree(create->full_name);
418 		if (!done) {
419 			kfree(create);
420 			return;
421 		}
422 		create->result = ERR_PTR(pid);
423 		complete(done);
424 	}
425 }
426 
427 static __printf(4, 0)
__kthread_create_on_node(int (* threadfn)(void * data),void * data,int node,const char namefmt[],va_list args)428 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
429 						    void *data, int node,
430 						    const char namefmt[],
431 						    va_list args)
432 {
433 	DECLARE_COMPLETION_ONSTACK(done);
434 	struct task_struct *task;
435 	struct kthread_create_info *create = kmalloc(sizeof(*create),
436 						     GFP_KERNEL);
437 
438 	if (!create)
439 		return ERR_PTR(-ENOMEM);
440 	create->threadfn = threadfn;
441 	create->data = data;
442 	create->node = node;
443 	create->done = &done;
444 	create->full_name = kvasprintf(GFP_KERNEL, namefmt, args);
445 	if (!create->full_name) {
446 		task = ERR_PTR(-ENOMEM);
447 		goto free_create;
448 	}
449 
450 	spin_lock(&kthread_create_lock);
451 	list_add_tail(&create->list, &kthread_create_list);
452 	spin_unlock(&kthread_create_lock);
453 
454 	wake_up_process(kthreadd_task);
455 	/*
456 	 * Wait for completion in killable state, for I might be chosen by
457 	 * the OOM killer while kthreadd is trying to allocate memory for
458 	 * new kernel thread.
459 	 */
460 	if (unlikely(wait_for_completion_killable(&done))) {
461 		/*
462 		 * If I was killed by a fatal signal before kthreadd (or new
463 		 * kernel thread) calls complete(), leave the cleanup of this
464 		 * structure to that thread.
465 		 */
466 		if (xchg(&create->done, NULL))
467 			return ERR_PTR(-EINTR);
468 		/*
469 		 * kthreadd (or new kernel thread) will call complete()
470 		 * shortly.
471 		 */
472 		wait_for_completion(&done);
473 	}
474 	task = create->result;
475 free_create:
476 	kfree(create);
477 	return task;
478 }
479 
480 /**
481  * kthread_create_on_node - create a kthread.
482  * @threadfn: the function to run until signal_pending(current).
483  * @data: data ptr for @threadfn.
484  * @node: task and thread structures for the thread are allocated on this node
485  * @namefmt: printf-style name for the thread.
486  *
487  * Description: This helper function creates and names a kernel
488  * thread.  The thread will be stopped: use wake_up_process() to start
489  * it.  See also kthread_run().  The new thread has SCHED_NORMAL policy and
490  * is affine to all CPUs.
491  *
492  * If thread is going to be bound on a particular cpu, give its node
493  * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
494  * When woken, the thread will run @threadfn() with @data as its
495  * argument. @threadfn() can either return directly if it is a
496  * standalone thread for which no one will call kthread_stop(), or
497  * return when 'kthread_should_stop()' is true (which means
498  * kthread_stop() has been called).  The return value should be zero
499  * or a negative error number; it will be passed to kthread_stop().
500  *
501  * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
502  */
kthread_create_on_node(int (* threadfn)(void * data),void * data,int node,const char namefmt[],...)503 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
504 					   void *data, int node,
505 					   const char namefmt[],
506 					   ...)
507 {
508 	struct task_struct *task;
509 	va_list args;
510 
511 	va_start(args, namefmt);
512 	task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
513 	va_end(args);
514 
515 	return task;
516 }
517 EXPORT_SYMBOL(kthread_create_on_node);
518 
__kthread_bind_mask(struct task_struct * p,const struct cpumask * mask,unsigned int state)519 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state)
520 {
521 	unsigned long flags;
522 
523 	if (!wait_task_inactive(p, state)) {
524 		WARN_ON(1);
525 		return;
526 	}
527 
528 	/* It's safe because the task is inactive. */
529 	raw_spin_lock_irqsave(&p->pi_lock, flags);
530 	do_set_cpus_allowed(p, mask);
531 	p->flags |= PF_NO_SETAFFINITY;
532 	raw_spin_unlock_irqrestore(&p->pi_lock, flags);
533 }
534 
__kthread_bind(struct task_struct * p,unsigned int cpu,unsigned int state)535 static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state)
536 {
537 	__kthread_bind_mask(p, cpumask_of(cpu), state);
538 }
539 
kthread_bind_mask(struct task_struct * p,const struct cpumask * mask)540 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
541 {
542 	__kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
543 }
544 
545 /**
546  * kthread_bind - bind a just-created kthread to a cpu.
547  * @p: thread created by kthread_create().
548  * @cpu: cpu (might not be online, must be possible) for @k to run on.
549  *
550  * Description: This function is equivalent to set_cpus_allowed(),
551  * except that @cpu doesn't need to be online, and the thread must be
552  * stopped (i.e., just returned from kthread_create()).
553  */
kthread_bind(struct task_struct * p,unsigned int cpu)554 void kthread_bind(struct task_struct *p, unsigned int cpu)
555 {
556 	__kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
557 }
558 EXPORT_SYMBOL(kthread_bind);
559 
560 /**
561  * kthread_create_on_cpu - Create a cpu bound kthread
562  * @threadfn: the function to run until signal_pending(current).
563  * @data: data ptr for @threadfn.
564  * @cpu: The cpu on which the thread should be bound,
565  * @namefmt: printf-style name for the thread. Format is restricted
566  *	     to "name.*%u". Code fills in cpu number.
567  *
568  * Description: This helper function creates and names a kernel thread
569  */
kthread_create_on_cpu(int (* threadfn)(void * data),void * data,unsigned int cpu,const char * namefmt)570 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
571 					  void *data, unsigned int cpu,
572 					  const char *namefmt)
573 {
574 	struct task_struct *p;
575 
576 	p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
577 				   cpu);
578 	if (IS_ERR(p))
579 		return p;
580 	kthread_bind(p, cpu);
581 	/* CPU hotplug need to bind once again when unparking the thread. */
582 	to_kthread(p)->cpu = cpu;
583 	return p;
584 }
585 EXPORT_SYMBOL(kthread_create_on_cpu);
586 
kthread_set_per_cpu(struct task_struct * k,int cpu)587 void kthread_set_per_cpu(struct task_struct *k, int cpu)
588 {
589 	struct kthread *kthread = to_kthread(k);
590 	if (!kthread)
591 		return;
592 
593 	WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
594 
595 	if (cpu < 0) {
596 		clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
597 		return;
598 	}
599 
600 	kthread->cpu = cpu;
601 	set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
602 }
603 
kthread_is_per_cpu(struct task_struct * p)604 bool kthread_is_per_cpu(struct task_struct *p)
605 {
606 	struct kthread *kthread = __to_kthread(p);
607 	if (!kthread)
608 		return false;
609 
610 	return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
611 }
612 
613 /**
614  * kthread_unpark - unpark a thread created by kthread_create().
615  * @k:		thread created by kthread_create().
616  *
617  * Sets kthread_should_park() for @k to return false, wakes it, and
618  * waits for it to return. If the thread is marked percpu then its
619  * bound to the cpu again.
620  */
kthread_unpark(struct task_struct * k)621 void kthread_unpark(struct task_struct *k)
622 {
623 	struct kthread *kthread = to_kthread(k);
624 
625 	/*
626 	 * Newly created kthread was parked when the CPU was offline.
627 	 * The binding was lost and we need to set it again.
628 	 */
629 	if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
630 		__kthread_bind(k, kthread->cpu, TASK_PARKED);
631 
632 	clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
633 	/*
634 	 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
635 	 */
636 	wake_up_state(k, TASK_PARKED);
637 }
638 EXPORT_SYMBOL_GPL(kthread_unpark);
639 
640 /**
641  * kthread_park - park a thread created by kthread_create().
642  * @k: thread created by kthread_create().
643  *
644  * Sets kthread_should_park() for @k to return true, wakes it, and
645  * waits for it to return. This can also be called after kthread_create()
646  * instead of calling wake_up_process(): the thread will park without
647  * calling threadfn().
648  *
649  * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
650  * If called by the kthread itself just the park bit is set.
651  */
kthread_park(struct task_struct * k)652 int kthread_park(struct task_struct *k)
653 {
654 	struct kthread *kthread = to_kthread(k);
655 
656 	if (WARN_ON(k->flags & PF_EXITING))
657 		return -ENOSYS;
658 
659 	if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
660 		return -EBUSY;
661 
662 	set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
663 	if (k != current) {
664 		wake_up_process(k);
665 		/*
666 		 * Wait for __kthread_parkme() to complete(), this means we
667 		 * _will_ have TASK_PARKED and are about to call schedule().
668 		 */
669 		wait_for_completion(&kthread->parked);
670 		/*
671 		 * Now wait for that schedule() to complete and the task to
672 		 * get scheduled out.
673 		 */
674 		WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
675 	}
676 
677 	return 0;
678 }
679 EXPORT_SYMBOL_GPL(kthread_park);
680 
681 /**
682  * kthread_stop - stop a thread created by kthread_create().
683  * @k: thread created by kthread_create().
684  *
685  * Sets kthread_should_stop() for @k to return true, wakes it, and
686  * waits for it to exit. This can also be called after kthread_create()
687  * instead of calling wake_up_process(): the thread will exit without
688  * calling threadfn().
689  *
690  * If threadfn() may call kthread_exit() itself, the caller must ensure
691  * task_struct can't go away.
692  *
693  * Returns the result of threadfn(), or %-EINTR if wake_up_process()
694  * was never called.
695  */
kthread_stop(struct task_struct * k)696 int kthread_stop(struct task_struct *k)
697 {
698 	struct kthread *kthread;
699 	int ret;
700 
701 	trace_sched_kthread_stop(k);
702 
703 	get_task_struct(k);
704 	kthread = to_kthread(k);
705 	set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
706 	kthread_unpark(k);
707 	set_tsk_thread_flag(k, TIF_NOTIFY_SIGNAL);
708 	wake_up_process(k);
709 	wait_for_completion(&kthread->exited);
710 	ret = kthread->result;
711 	put_task_struct(k);
712 
713 	trace_sched_kthread_stop_ret(ret);
714 	return ret;
715 }
716 EXPORT_SYMBOL(kthread_stop);
717 
kthreadd(void * unused)718 int kthreadd(void *unused)
719 {
720 	struct task_struct *tsk = current;
721 
722 	/* Setup a clean context for our children to inherit. */
723 	set_task_comm(tsk, "kthreadd");
724 	ignore_signals(tsk);
725 	set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_TYPE_KTHREAD));
726 	set_mems_allowed(node_states[N_MEMORY]);
727 
728 	current->flags |= PF_NOFREEZE;
729 	cgroup_init_kthreadd();
730 
731 	for (;;) {
732 		set_current_state(TASK_INTERRUPTIBLE);
733 		if (list_empty(&kthread_create_list))
734 			schedule();
735 		__set_current_state(TASK_RUNNING);
736 
737 		spin_lock(&kthread_create_lock);
738 		while (!list_empty(&kthread_create_list)) {
739 			struct kthread_create_info *create;
740 
741 			create = list_entry(kthread_create_list.next,
742 					    struct kthread_create_info, list);
743 			list_del_init(&create->list);
744 			spin_unlock(&kthread_create_lock);
745 
746 			create_kthread(create);
747 
748 			spin_lock(&kthread_create_lock);
749 		}
750 		spin_unlock(&kthread_create_lock);
751 	}
752 
753 	return 0;
754 }
755 
__kthread_init_worker(struct kthread_worker * worker,const char * name,struct lock_class_key * key)756 void __kthread_init_worker(struct kthread_worker *worker,
757 				const char *name,
758 				struct lock_class_key *key)
759 {
760 	memset(worker, 0, sizeof(struct kthread_worker));
761 	raw_spin_lock_init(&worker->lock);
762 	lockdep_set_class_and_name(&worker->lock, key, name);
763 	INIT_LIST_HEAD(&worker->work_list);
764 	INIT_LIST_HEAD(&worker->delayed_work_list);
765 }
766 EXPORT_SYMBOL_GPL(__kthread_init_worker);
767 
768 /**
769  * kthread_worker_fn - kthread function to process kthread_worker
770  * @worker_ptr: pointer to initialized kthread_worker
771  *
772  * This function implements the main cycle of kthread worker. It processes
773  * work_list until it is stopped with kthread_stop(). It sleeps when the queue
774  * is empty.
775  *
776  * The works are not allowed to keep any locks, disable preemption or interrupts
777  * when they finish. There is defined a safe point for freezing when one work
778  * finishes and before a new one is started.
779  *
780  * Also the works must not be handled by more than one worker at the same time,
781  * see also kthread_queue_work().
782  */
kthread_worker_fn(void * worker_ptr)783 int kthread_worker_fn(void *worker_ptr)
784 {
785 	struct kthread_worker *worker = worker_ptr;
786 	struct kthread_work *work;
787 
788 	/*
789 	 * FIXME: Update the check and remove the assignment when all kthread
790 	 * worker users are created using kthread_create_worker*() functions.
791 	 */
792 	WARN_ON(worker->task && worker->task != current);
793 	worker->task = current;
794 
795 	if (worker->flags & KTW_FREEZABLE)
796 		set_freezable();
797 
798 repeat:
799 	set_current_state(TASK_INTERRUPTIBLE);	/* mb paired w/ kthread_stop */
800 
801 	if (kthread_should_stop()) {
802 		__set_current_state(TASK_RUNNING);
803 		raw_spin_lock_irq(&worker->lock);
804 		worker->task = NULL;
805 		raw_spin_unlock_irq(&worker->lock);
806 		return 0;
807 	}
808 
809 	work = NULL;
810 	raw_spin_lock_irq(&worker->lock);
811 	if (!list_empty(&worker->work_list)) {
812 		work = list_first_entry(&worker->work_list,
813 					struct kthread_work, node);
814 		list_del_init(&work->node);
815 	}
816 	worker->current_work = work;
817 	raw_spin_unlock_irq(&worker->lock);
818 
819 	if (work) {
820 		kthread_work_func_t func = work->func;
821 		__set_current_state(TASK_RUNNING);
822 		trace_sched_kthread_work_execute_start(work);
823 		work->func(work);
824 		/*
825 		 * Avoid dereferencing work after this point.  The trace
826 		 * event only cares about the address.
827 		 */
828 		trace_sched_kthread_work_execute_end(work, func);
829 	} else if (!freezing(current))
830 		schedule();
831 
832 	try_to_freeze();
833 	cond_resched();
834 	goto repeat;
835 }
836 EXPORT_SYMBOL_GPL(kthread_worker_fn);
837 
838 static __printf(3, 0) struct kthread_worker *
__kthread_create_worker(int cpu,unsigned int flags,const char namefmt[],va_list args)839 __kthread_create_worker(int cpu, unsigned int flags,
840 			const char namefmt[], va_list args)
841 {
842 	struct kthread_worker *worker;
843 	struct task_struct *task;
844 	int node = NUMA_NO_NODE;
845 
846 	worker = kzalloc(sizeof(*worker), GFP_KERNEL);
847 	if (!worker)
848 		return ERR_PTR(-ENOMEM);
849 
850 	kthread_init_worker(worker);
851 
852 	if (cpu >= 0)
853 		node = cpu_to_node(cpu);
854 
855 	task = __kthread_create_on_node(kthread_worker_fn, worker,
856 						node, namefmt, args);
857 	if (IS_ERR(task))
858 		goto fail_task;
859 
860 	if (cpu >= 0)
861 		kthread_bind(task, cpu);
862 
863 	worker->flags = flags;
864 	worker->task = task;
865 	wake_up_process(task);
866 	return worker;
867 
868 fail_task:
869 	kfree(worker);
870 	return ERR_CAST(task);
871 }
872 
873 /**
874  * kthread_create_worker - create a kthread worker
875  * @flags: flags modifying the default behavior of the worker
876  * @namefmt: printf-style name for the kthread worker (task).
877  *
878  * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
879  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
880  * when the caller was killed by a fatal signal.
881  */
882 struct kthread_worker *
kthread_create_worker(unsigned int flags,const char namefmt[],...)883 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
884 {
885 	struct kthread_worker *worker;
886 	va_list args;
887 
888 	va_start(args, namefmt);
889 	worker = __kthread_create_worker(-1, flags, namefmt, args);
890 	va_end(args);
891 
892 	return worker;
893 }
894 EXPORT_SYMBOL(kthread_create_worker);
895 
896 /**
897  * kthread_create_worker_on_cpu - create a kthread worker and bind it
898  *	to a given CPU and the associated NUMA node.
899  * @cpu: CPU number
900  * @flags: flags modifying the default behavior of the worker
901  * @namefmt: printf-style name for the kthread worker (task).
902  *
903  * Use a valid CPU number if you want to bind the kthread worker
904  * to the given CPU and the associated NUMA node.
905  *
906  * A good practice is to add the cpu number also into the worker name.
907  * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
908  *
909  * CPU hotplug:
910  * The kthread worker API is simple and generic. It just provides a way
911  * to create, use, and destroy workers.
912  *
913  * It is up to the API user how to handle CPU hotplug. They have to decide
914  * how to handle pending work items, prevent queuing new ones, and
915  * restore the functionality when the CPU goes off and on. There are a
916  * few catches:
917  *
918  *    - CPU affinity gets lost when it is scheduled on an offline CPU.
919  *
920  *    - The worker might not exist when the CPU was off when the user
921  *      created the workers.
922  *
923  * Good practice is to implement two CPU hotplug callbacks and to
924  * destroy/create the worker when the CPU goes down/up.
925  *
926  * Return:
927  * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
928  * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
929  * when the caller was killed by a fatal signal.
930  */
931 struct kthread_worker *
kthread_create_worker_on_cpu(int cpu,unsigned int flags,const char namefmt[],...)932 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
933 			     const char namefmt[], ...)
934 {
935 	struct kthread_worker *worker;
936 	va_list args;
937 
938 	va_start(args, namefmt);
939 	worker = __kthread_create_worker(cpu, flags, namefmt, args);
940 	va_end(args);
941 
942 	return worker;
943 }
944 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
945 
946 /*
947  * Returns true when the work could not be queued at the moment.
948  * It happens when it is already pending in a worker list
949  * or when it is being cancelled.
950  */
queuing_blocked(struct kthread_worker * worker,struct kthread_work * work)951 static inline bool queuing_blocked(struct kthread_worker *worker,
952 				   struct kthread_work *work)
953 {
954 	lockdep_assert_held(&worker->lock);
955 
956 	return !list_empty(&work->node) || work->canceling;
957 }
958 
kthread_insert_work_sanity_check(struct kthread_worker * worker,struct kthread_work * work)959 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
960 					     struct kthread_work *work)
961 {
962 	lockdep_assert_held(&worker->lock);
963 	WARN_ON_ONCE(!list_empty(&work->node));
964 	/* Do not use a work with >1 worker, see kthread_queue_work() */
965 	WARN_ON_ONCE(work->worker && work->worker != worker);
966 }
967 
968 /* insert @work before @pos in @worker */
kthread_insert_work(struct kthread_worker * worker,struct kthread_work * work,struct list_head * pos)969 static void kthread_insert_work(struct kthread_worker *worker,
970 				struct kthread_work *work,
971 				struct list_head *pos)
972 {
973 	kthread_insert_work_sanity_check(worker, work);
974 
975 	trace_sched_kthread_work_queue_work(worker, work);
976 
977 	list_add_tail(&work->node, pos);
978 	work->worker = worker;
979 	if (!worker->current_work && likely(worker->task))
980 		wake_up_process(worker->task);
981 }
982 
983 /**
984  * kthread_queue_work - queue a kthread_work
985  * @worker: target kthread_worker
986  * @work: kthread_work to queue
987  *
988  * Queue @work to work processor @task for async execution.  @task
989  * must have been created with kthread_worker_create().  Returns %true
990  * if @work was successfully queued, %false if it was already pending.
991  *
992  * Reinitialize the work if it needs to be used by another worker.
993  * For example, when the worker was stopped and started again.
994  */
kthread_queue_work(struct kthread_worker * worker,struct kthread_work * work)995 bool kthread_queue_work(struct kthread_worker *worker,
996 			struct kthread_work *work)
997 {
998 	bool ret = false;
999 	unsigned long flags;
1000 
1001 	raw_spin_lock_irqsave(&worker->lock, flags);
1002 	if (!queuing_blocked(worker, work)) {
1003 		kthread_insert_work(worker, work, &worker->work_list);
1004 		ret = true;
1005 	}
1006 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1007 	return ret;
1008 }
1009 EXPORT_SYMBOL_GPL(kthread_queue_work);
1010 
1011 /**
1012  * kthread_delayed_work_timer_fn - callback that queues the associated kthread
1013  *	delayed work when the timer expires.
1014  * @t: pointer to the expired timer
1015  *
1016  * The format of the function is defined by struct timer_list.
1017  * It should have been called from irqsafe timer with irq already off.
1018  */
kthread_delayed_work_timer_fn(struct timer_list * t)1019 void kthread_delayed_work_timer_fn(struct timer_list *t)
1020 {
1021 	struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
1022 	struct kthread_work *work = &dwork->work;
1023 	struct kthread_worker *worker = work->worker;
1024 	unsigned long flags;
1025 
1026 	/*
1027 	 * This might happen when a pending work is reinitialized.
1028 	 * It means that it is used a wrong way.
1029 	 */
1030 	if (WARN_ON_ONCE(!worker))
1031 		return;
1032 
1033 	raw_spin_lock_irqsave(&worker->lock, flags);
1034 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1035 	WARN_ON_ONCE(work->worker != worker);
1036 
1037 	/* Move the work from worker->delayed_work_list. */
1038 	WARN_ON_ONCE(list_empty(&work->node));
1039 	list_del_init(&work->node);
1040 	if (!work->canceling)
1041 		kthread_insert_work(worker, work, &worker->work_list);
1042 
1043 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1044 }
1045 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
1046 
__kthread_queue_delayed_work(struct kthread_worker * worker,struct kthread_delayed_work * dwork,unsigned long delay)1047 static void __kthread_queue_delayed_work(struct kthread_worker *worker,
1048 					 struct kthread_delayed_work *dwork,
1049 					 unsigned long delay)
1050 {
1051 	struct timer_list *timer = &dwork->timer;
1052 	struct kthread_work *work = &dwork->work;
1053 
1054 	WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
1055 
1056 	/*
1057 	 * If @delay is 0, queue @dwork->work immediately.  This is for
1058 	 * both optimization and correctness.  The earliest @timer can
1059 	 * expire is on the closest next tick and delayed_work users depend
1060 	 * on that there's no such delay when @delay is 0.
1061 	 */
1062 	if (!delay) {
1063 		kthread_insert_work(worker, work, &worker->work_list);
1064 		return;
1065 	}
1066 
1067 	/* Be paranoid and try to detect possible races already now. */
1068 	kthread_insert_work_sanity_check(worker, work);
1069 
1070 	list_add(&work->node, &worker->delayed_work_list);
1071 	work->worker = worker;
1072 	timer->expires = jiffies + delay;
1073 	add_timer(timer);
1074 }
1075 
1076 /**
1077  * kthread_queue_delayed_work - queue the associated kthread work
1078  *	after a delay.
1079  * @worker: target kthread_worker
1080  * @dwork: kthread_delayed_work to queue
1081  * @delay: number of jiffies to wait before queuing
1082  *
1083  * If the work has not been pending it starts a timer that will queue
1084  * the work after the given @delay. If @delay is zero, it queues the
1085  * work immediately.
1086  *
1087  * Return: %false if the @work has already been pending. It means that
1088  * either the timer was running or the work was queued. It returns %true
1089  * otherwise.
1090  */
kthread_queue_delayed_work(struct kthread_worker * worker,struct kthread_delayed_work * dwork,unsigned long delay)1091 bool kthread_queue_delayed_work(struct kthread_worker *worker,
1092 				struct kthread_delayed_work *dwork,
1093 				unsigned long delay)
1094 {
1095 	struct kthread_work *work = &dwork->work;
1096 	unsigned long flags;
1097 	bool ret = false;
1098 
1099 	raw_spin_lock_irqsave(&worker->lock, flags);
1100 
1101 	if (!queuing_blocked(worker, work)) {
1102 		__kthread_queue_delayed_work(worker, dwork, delay);
1103 		ret = true;
1104 	}
1105 
1106 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1107 	return ret;
1108 }
1109 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1110 
1111 struct kthread_flush_work {
1112 	struct kthread_work	work;
1113 	struct completion	done;
1114 };
1115 
kthread_flush_work_fn(struct kthread_work * work)1116 static void kthread_flush_work_fn(struct kthread_work *work)
1117 {
1118 	struct kthread_flush_work *fwork =
1119 		container_of(work, struct kthread_flush_work, work);
1120 	complete(&fwork->done);
1121 }
1122 
1123 /**
1124  * kthread_flush_work - flush a kthread_work
1125  * @work: work to flush
1126  *
1127  * If @work is queued or executing, wait for it to finish execution.
1128  */
kthread_flush_work(struct kthread_work * work)1129 void kthread_flush_work(struct kthread_work *work)
1130 {
1131 	struct kthread_flush_work fwork = {
1132 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1133 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1134 	};
1135 	struct kthread_worker *worker;
1136 	bool noop = false;
1137 
1138 	worker = work->worker;
1139 	if (!worker)
1140 		return;
1141 
1142 	raw_spin_lock_irq(&worker->lock);
1143 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1144 	WARN_ON_ONCE(work->worker != worker);
1145 
1146 	if (!list_empty(&work->node))
1147 		kthread_insert_work(worker, &fwork.work, work->node.next);
1148 	else if (worker->current_work == work)
1149 		kthread_insert_work(worker, &fwork.work,
1150 				    worker->work_list.next);
1151 	else
1152 		noop = true;
1153 
1154 	raw_spin_unlock_irq(&worker->lock);
1155 
1156 	if (!noop)
1157 		wait_for_completion(&fwork.done);
1158 }
1159 EXPORT_SYMBOL_GPL(kthread_flush_work);
1160 
1161 /*
1162  * Make sure that the timer is neither set nor running and could
1163  * not manipulate the work list_head any longer.
1164  *
1165  * The function is called under worker->lock. The lock is temporary
1166  * released but the timer can't be set again in the meantime.
1167  */
kthread_cancel_delayed_work_timer(struct kthread_work * work,unsigned long * flags)1168 static void kthread_cancel_delayed_work_timer(struct kthread_work *work,
1169 					      unsigned long *flags)
1170 {
1171 	struct kthread_delayed_work *dwork =
1172 		container_of(work, struct kthread_delayed_work, work);
1173 	struct kthread_worker *worker = work->worker;
1174 
1175 	/*
1176 	 * del_timer_sync() must be called to make sure that the timer
1177 	 * callback is not running. The lock must be temporary released
1178 	 * to avoid a deadlock with the callback. In the meantime,
1179 	 * any queuing is blocked by setting the canceling counter.
1180 	 */
1181 	work->canceling++;
1182 	raw_spin_unlock_irqrestore(&worker->lock, *flags);
1183 	del_timer_sync(&dwork->timer);
1184 	raw_spin_lock_irqsave(&worker->lock, *flags);
1185 	work->canceling--;
1186 }
1187 
1188 /*
1189  * This function removes the work from the worker queue.
1190  *
1191  * It is called under worker->lock. The caller must make sure that
1192  * the timer used by delayed work is not running, e.g. by calling
1193  * kthread_cancel_delayed_work_timer().
1194  *
1195  * The work might still be in use when this function finishes. See the
1196  * current_work proceed by the worker.
1197  *
1198  * Return: %true if @work was pending and successfully canceled,
1199  *	%false if @work was not pending
1200  */
__kthread_cancel_work(struct kthread_work * work)1201 static bool __kthread_cancel_work(struct kthread_work *work)
1202 {
1203 	/*
1204 	 * Try to remove the work from a worker list. It might either
1205 	 * be from worker->work_list or from worker->delayed_work_list.
1206 	 */
1207 	if (!list_empty(&work->node)) {
1208 		list_del_init(&work->node);
1209 		return true;
1210 	}
1211 
1212 	return false;
1213 }
1214 
1215 /**
1216  * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1217  * @worker: kthread worker to use
1218  * @dwork: kthread delayed work to queue
1219  * @delay: number of jiffies to wait before queuing
1220  *
1221  * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1222  * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1223  * @work is guaranteed to be queued immediately.
1224  *
1225  * Return: %false if @dwork was idle and queued, %true otherwise.
1226  *
1227  * A special case is when the work is being canceled in parallel.
1228  * It might be caused either by the real kthread_cancel_delayed_work_sync()
1229  * or yet another kthread_mod_delayed_work() call. We let the other command
1230  * win and return %true here. The return value can be used for reference
1231  * counting and the number of queued works stays the same. Anyway, the caller
1232  * is supposed to synchronize these operations a reasonable way.
1233  *
1234  * This function is safe to call from any context including IRQ handler.
1235  * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1236  * for details.
1237  */
kthread_mod_delayed_work(struct kthread_worker * worker,struct kthread_delayed_work * dwork,unsigned long delay)1238 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1239 			      struct kthread_delayed_work *dwork,
1240 			      unsigned long delay)
1241 {
1242 	struct kthread_work *work = &dwork->work;
1243 	unsigned long flags;
1244 	int ret;
1245 
1246 	raw_spin_lock_irqsave(&worker->lock, flags);
1247 
1248 	/* Do not bother with canceling when never queued. */
1249 	if (!work->worker) {
1250 		ret = false;
1251 		goto fast_queue;
1252 	}
1253 
1254 	/* Work must not be used with >1 worker, see kthread_queue_work() */
1255 	WARN_ON_ONCE(work->worker != worker);
1256 
1257 	/*
1258 	 * Temporary cancel the work but do not fight with another command
1259 	 * that is canceling the work as well.
1260 	 *
1261 	 * It is a bit tricky because of possible races with another
1262 	 * mod_delayed_work() and cancel_delayed_work() callers.
1263 	 *
1264 	 * The timer must be canceled first because worker->lock is released
1265 	 * when doing so. But the work can be removed from the queue (list)
1266 	 * only when it can be queued again so that the return value can
1267 	 * be used for reference counting.
1268 	 */
1269 	kthread_cancel_delayed_work_timer(work, &flags);
1270 	if (work->canceling) {
1271 		/* The number of works in the queue does not change. */
1272 		ret = true;
1273 		goto out;
1274 	}
1275 	ret = __kthread_cancel_work(work);
1276 
1277 fast_queue:
1278 	__kthread_queue_delayed_work(worker, dwork, delay);
1279 out:
1280 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1281 	return ret;
1282 }
1283 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1284 
__kthread_cancel_work_sync(struct kthread_work * work,bool is_dwork)1285 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1286 {
1287 	struct kthread_worker *worker = work->worker;
1288 	unsigned long flags;
1289 	int ret = false;
1290 
1291 	if (!worker)
1292 		goto out;
1293 
1294 	raw_spin_lock_irqsave(&worker->lock, flags);
1295 	/* Work must not be used with >1 worker, see kthread_queue_work(). */
1296 	WARN_ON_ONCE(work->worker != worker);
1297 
1298 	if (is_dwork)
1299 		kthread_cancel_delayed_work_timer(work, &flags);
1300 
1301 	ret = __kthread_cancel_work(work);
1302 
1303 	if (worker->current_work != work)
1304 		goto out_fast;
1305 
1306 	/*
1307 	 * The work is in progress and we need to wait with the lock released.
1308 	 * In the meantime, block any queuing by setting the canceling counter.
1309 	 */
1310 	work->canceling++;
1311 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1312 	kthread_flush_work(work);
1313 	raw_spin_lock_irqsave(&worker->lock, flags);
1314 	work->canceling--;
1315 
1316 out_fast:
1317 	raw_spin_unlock_irqrestore(&worker->lock, flags);
1318 out:
1319 	return ret;
1320 }
1321 
1322 /**
1323  * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1324  * @work: the kthread work to cancel
1325  *
1326  * Cancel @work and wait for its execution to finish.  This function
1327  * can be used even if the work re-queues itself. On return from this
1328  * function, @work is guaranteed to be not pending or executing on any CPU.
1329  *
1330  * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1331  * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1332  *
1333  * The caller must ensure that the worker on which @work was last
1334  * queued can't be destroyed before this function returns.
1335  *
1336  * Return: %true if @work was pending, %false otherwise.
1337  */
kthread_cancel_work_sync(struct kthread_work * work)1338 bool kthread_cancel_work_sync(struct kthread_work *work)
1339 {
1340 	return __kthread_cancel_work_sync(work, false);
1341 }
1342 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1343 
1344 /**
1345  * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1346  *	wait for it to finish.
1347  * @dwork: the kthread delayed work to cancel
1348  *
1349  * This is kthread_cancel_work_sync() for delayed works.
1350  *
1351  * Return: %true if @dwork was pending, %false otherwise.
1352  */
kthread_cancel_delayed_work_sync(struct kthread_delayed_work * dwork)1353 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1354 {
1355 	return __kthread_cancel_work_sync(&dwork->work, true);
1356 }
1357 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1358 
1359 /**
1360  * kthread_flush_worker - flush all current works on a kthread_worker
1361  * @worker: worker to flush
1362  *
1363  * Wait until all currently executing or pending works on @worker are
1364  * finished.
1365  */
kthread_flush_worker(struct kthread_worker * worker)1366 void kthread_flush_worker(struct kthread_worker *worker)
1367 {
1368 	struct kthread_flush_work fwork = {
1369 		KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1370 		COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1371 	};
1372 
1373 	kthread_queue_work(worker, &fwork.work);
1374 	wait_for_completion(&fwork.done);
1375 }
1376 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1377 
1378 /**
1379  * kthread_destroy_worker - destroy a kthread worker
1380  * @worker: worker to be destroyed
1381  *
1382  * Flush and destroy @worker.  The simple flush is enough because the kthread
1383  * worker API is used only in trivial scenarios.  There are no multi-step state
1384  * machines needed.
1385  *
1386  * Note that this function is not responsible for handling delayed work, so
1387  * caller should be responsible for queuing or canceling all delayed work items
1388  * before invoke this function.
1389  */
kthread_destroy_worker(struct kthread_worker * worker)1390 void kthread_destroy_worker(struct kthread_worker *worker)
1391 {
1392 	struct task_struct *task;
1393 
1394 	task = worker->task;
1395 	if (WARN_ON(!task))
1396 		return;
1397 
1398 	kthread_flush_worker(worker);
1399 	kthread_stop(task);
1400 	WARN_ON(!list_empty(&worker->delayed_work_list));
1401 	WARN_ON(!list_empty(&worker->work_list));
1402 	kfree(worker);
1403 }
1404 EXPORT_SYMBOL(kthread_destroy_worker);
1405 
1406 /**
1407  * kthread_use_mm - make the calling kthread operate on an address space
1408  * @mm: address space to operate on
1409  */
kthread_use_mm(struct mm_struct * mm)1410 void kthread_use_mm(struct mm_struct *mm)
1411 {
1412 	struct mm_struct *active_mm;
1413 	struct task_struct *tsk = current;
1414 
1415 	WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1416 	WARN_ON_ONCE(tsk->mm);
1417 
1418 	/*
1419 	 * It is possible for mm to be the same as tsk->active_mm, but
1420 	 * we must still mmgrab(mm) and mmdrop_lazy_tlb(active_mm),
1421 	 * because these references are not equivalent.
1422 	 */
1423 	mmgrab(mm);
1424 
1425 	task_lock(tsk);
1426 	/* Hold off tlb flush IPIs while switching mm's */
1427 	local_irq_disable();
1428 	active_mm = tsk->active_mm;
1429 	tsk->active_mm = mm;
1430 	tsk->mm = mm;
1431 	membarrier_update_current_mm(mm);
1432 	switch_mm_irqs_off(active_mm, mm, tsk);
1433 	local_irq_enable();
1434 	task_unlock(tsk);
1435 #ifdef finish_arch_post_lock_switch
1436 	finish_arch_post_lock_switch();
1437 #endif
1438 
1439 	/*
1440 	 * When a kthread starts operating on an address space, the loop
1441 	 * in membarrier_{private,global}_expedited() may not observe
1442 	 * that tsk->mm, and not issue an IPI. Membarrier requires a
1443 	 * memory barrier after storing to tsk->mm, before accessing
1444 	 * user-space memory. A full memory barrier for membarrier
1445 	 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1446 	 * mmdrop_lazy_tlb().
1447 	 */
1448 	mmdrop_lazy_tlb(active_mm);
1449 }
1450 EXPORT_SYMBOL_GPL(kthread_use_mm);
1451 
1452 /**
1453  * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1454  * @mm: address space to operate on
1455  */
kthread_unuse_mm(struct mm_struct * mm)1456 void kthread_unuse_mm(struct mm_struct *mm)
1457 {
1458 	struct task_struct *tsk = current;
1459 
1460 	WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1461 	WARN_ON_ONCE(!tsk->mm);
1462 
1463 	task_lock(tsk);
1464 	/*
1465 	 * When a kthread stops operating on an address space, the loop
1466 	 * in membarrier_{private,global}_expedited() may not observe
1467 	 * that tsk->mm, and not issue an IPI. Membarrier requires a
1468 	 * memory barrier after accessing user-space memory, before
1469 	 * clearing tsk->mm.
1470 	 */
1471 	smp_mb__after_spinlock();
1472 	sync_mm_rss(mm);
1473 	local_irq_disable();
1474 	tsk->mm = NULL;
1475 	membarrier_update_current_mm(NULL);
1476 	mmgrab_lazy_tlb(mm);
1477 	/* active_mm is still 'mm' */
1478 	enter_lazy_tlb(mm, tsk);
1479 	local_irq_enable();
1480 	task_unlock(tsk);
1481 
1482 	mmdrop(mm);
1483 }
1484 EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1485 
1486 #ifdef CONFIG_BLK_CGROUP
1487 /**
1488  * kthread_associate_blkcg - associate blkcg to current kthread
1489  * @css: the cgroup info
1490  *
1491  * Current thread must be a kthread. The thread is running jobs on behalf of
1492  * other threads. In some cases, we expect the jobs attach cgroup info of
1493  * original threads instead of that of current thread. This function stores
1494  * original thread's cgroup info in current kthread context for later
1495  * retrieval.
1496  */
kthread_associate_blkcg(struct cgroup_subsys_state * css)1497 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1498 {
1499 	struct kthread *kthread;
1500 
1501 	if (!(current->flags & PF_KTHREAD))
1502 		return;
1503 	kthread = to_kthread(current);
1504 	if (!kthread)
1505 		return;
1506 
1507 	if (kthread->blkcg_css) {
1508 		css_put(kthread->blkcg_css);
1509 		kthread->blkcg_css = NULL;
1510 	}
1511 	if (css) {
1512 		css_get(css);
1513 		kthread->blkcg_css = css;
1514 	}
1515 }
1516 EXPORT_SYMBOL(kthread_associate_blkcg);
1517 
1518 /**
1519  * kthread_blkcg - get associated blkcg css of current kthread
1520  *
1521  * Current thread must be a kthread.
1522  */
kthread_blkcg(void)1523 struct cgroup_subsys_state *kthread_blkcg(void)
1524 {
1525 	struct kthread *kthread;
1526 
1527 	if (current->flags & PF_KTHREAD) {
1528 		kthread = to_kthread(current);
1529 		if (kthread)
1530 			return kthread->blkcg_css;
1531 	}
1532 	return NULL;
1533 }
1534 #endif
1535