projects/linux·blob v5.10-rc5 · refs

1339 lines · 36.4 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/* Kernel thread helper functions.
3 * Copyright (C) 2004 IBM Corporation, Rusty Russell.
4 * Copyright (C) 2009 Red Hat, Inc.
5 *
6 * Creation is done via kthreadd, so that we get a clean environment
7 * even if we're invoked from userspace (think modprobe, hotplug cpu,
8 * etc.).
9 */
10#include <uapi/linux/sched/types.h>
11#include <linux/mm.h>
12#include <linux/mmu_context.h>
13#include <linux/sched.h>
14#include <linux/sched/mm.h>
15#include <linux/sched/task.h>
16#include <linux/kthread.h>
17#include <linux/completion.h>
18#include <linux/err.h>
19#include <linux/cgroup.h>
20#include <linux/cpuset.h>
21#include <linux/unistd.h>
22#include <linux/file.h>
23#include <linux/export.h>
24#include <linux/mutex.h>
25#include <linux/slab.h>
26#include <linux/freezer.h>
27#include <linux/ptrace.h>
28#include <linux/uaccess.h>
29#include <linux/numa.h>
30#include <linux/sched/isolation.h>
31#include <trace/events/sched.h>
32
33
34static DEFINE_SPINLOCK(kthread_create_lock);
35static LIST_HEAD(kthread_create_list);
36struct task_struct *kthreadd_task;
37
38struct kthread_create_info
39{
40 /* Information passed to kthread() from kthreadd. */
41 int (*threadfn)(void *data);
42 void *data;
43 int node;
44
45 /* Result passed back to kthread_create() from kthreadd. */
46 struct task_struct *result;
47 struct completion *done;
48
49 struct list_head list;
50};
51
52struct kthread {
53 unsigned long flags;
54 unsigned int cpu;
55 int (*threadfn)(void *);
56 void *data;
57 mm_segment_t oldfs;
58 struct completion parked;
59 struct completion exited;
60#ifdef CONFIG_BLK_CGROUP
61 struct cgroup_subsys_state *blkcg_css;
62#endif
63};
64
65enum KTHREAD_BITS {
66 KTHREAD_IS_PER_CPU = 0,
67 KTHREAD_SHOULD_STOP,
68 KTHREAD_SHOULD_PARK,
69};
70
71static inline void set_kthread_struct(void *kthread)
72{
73 /*
74 * We abuse ->set_child_tid to avoid the new member and because it
75 * can't be wrongly copied by copy_process(). We also rely on fact
76 * that the caller can't exec, so PF_KTHREAD can't be cleared.
77 */
78 current->set_child_tid = (__force void __user *)kthread;
79}
80
81static inline struct kthread *to_kthread(struct task_struct *k)
82{
83 WARN_ON(!(k->flags & PF_KTHREAD));
84 return (__force void *)k->set_child_tid;
85}
86
87void free_kthread_struct(struct task_struct *k)
88{
89 struct kthread *kthread;
90
91 /*
92 * Can be NULL if this kthread was created by kernel_thread()
93 * or if kmalloc() in kthread() failed.
94 */
95 kthread = to_kthread(k);
96#ifdef CONFIG_BLK_CGROUP
97 WARN_ON_ONCE(kthread && kthread->blkcg_css);
98#endif
99 kfree(kthread);
100}
101
102/**
103 * kthread_should_stop - should this kthread return now?
104 *
105 * When someone calls kthread_stop() on your kthread, it will be woken
106 * and this will return true. You should then return, and your return
107 * value will be passed through to kthread_stop().
108 */
109bool kthread_should_stop(void)
110{
111 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
112}
113EXPORT_SYMBOL(kthread_should_stop);
114
115bool __kthread_should_park(struct task_struct *k)
116{
117 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
118}
119EXPORT_SYMBOL_GPL(__kthread_should_park);
120
121/**
122 * kthread_should_park - should this kthread park now?
123 *
124 * When someone calls kthread_park() on your kthread, it will be woken
125 * and this will return true. You should then do the necessary
126 * cleanup and call kthread_parkme()
127 *
128 * Similar to kthread_should_stop(), but this keeps the thread alive
129 * and in a park position. kthread_unpark() "restarts" the thread and
130 * calls the thread function again.
131 */
132bool kthread_should_park(void)
133{
134 return __kthread_should_park(current);
135}
136EXPORT_SYMBOL_GPL(kthread_should_park);
137
138/**
139 * kthread_freezable_should_stop - should this freezable kthread return now?
140 * @was_frozen: optional out parameter, indicates whether %current was frozen
141 *
142 * kthread_should_stop() for freezable kthreads, which will enter
143 * refrigerator if necessary. This function is safe from kthread_stop() /
144 * freezer deadlock and freezable kthreads should use this function instead
145 * of calling try_to_freeze() directly.
146 */
147bool kthread_freezable_should_stop(bool *was_frozen)
148{
149 bool frozen = false;
150
151 might_sleep();
152
153 if (unlikely(freezing(current)))
154 frozen = __refrigerator(true);
155
156 if (was_frozen)
157 *was_frozen = frozen;
158
159 return kthread_should_stop();
160}
161EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
162
163/**
164 * kthread_func - return the function specified on kthread creation
165 * @task: kthread task in question
166 *
167 * Returns NULL if the task is not a kthread.
168 */
169void *kthread_func(struct task_struct *task)
170{
171 if (task->flags & PF_KTHREAD)
172 return to_kthread(task)->threadfn;
173 return NULL;
174}
175EXPORT_SYMBOL_GPL(kthread_func);
176
177/**
178 * kthread_data - return data value specified on kthread creation
179 * @task: kthread task in question
180 *
181 * Return the data value specified when kthread @task was created.
182 * The caller is responsible for ensuring the validity of @task when
183 * calling this function.
184 */
185void *kthread_data(struct task_struct *task)
186{
187 return to_kthread(task)->data;
188}
189EXPORT_SYMBOL_GPL(kthread_data);
190
191/**
192 * kthread_probe_data - speculative version of kthread_data()
193 * @task: possible kthread task in question
194 *
195 * @task could be a kthread task. Return the data value specified when it
196 * was created if accessible. If @task isn't a kthread task or its data is
197 * inaccessible for any reason, %NULL is returned. This function requires
198 * that @task itself is safe to dereference.
199 */
200void *kthread_probe_data(struct task_struct *task)
201{
202 struct kthread *kthread = to_kthread(task);
203 void *data = NULL;
204
205 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
206 return data;
207}
208
209static void __kthread_parkme(struct kthread *self)
210{
211 for (;;) {
212 /*
213 * TASK_PARKED is a special state; we must serialize against
214 * possible pending wakeups to avoid store-store collisions on
215 * task->state.
216 *
217 * Such a collision might possibly result in the task state
218 * changin from TASK_PARKED and us failing the
219 * wait_task_inactive() in kthread_park().
220 */
221 set_special_state(TASK_PARKED);
222 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
223 break;
224
225 /*
226 * Thread is going to call schedule(), do not preempt it,
227 * or the caller of kthread_park() may spend more time in
228 * wait_task_inactive().
229 */
230 preempt_disable();
231 complete(&self->parked);
232 schedule_preempt_disabled();
233 preempt_enable();
234 }
235 __set_current_state(TASK_RUNNING);
236}
237
238void kthread_parkme(void)
239{
240 __kthread_parkme(to_kthread(current));
241}
242EXPORT_SYMBOL_GPL(kthread_parkme);
243
244static int kthread(void *_create)
245{
246 /* Copy data: it's on kthread's stack */
247 struct kthread_create_info *create = _create;
248 int (*threadfn)(void *data) = create->threadfn;
249 void *data = create->data;
250 struct completion *done;
251 struct kthread *self;
252 int ret;
253
254 self = kzalloc(sizeof(*self), GFP_KERNEL);
255 set_kthread_struct(self);
256
257 /* If user was SIGKILLed, I release the structure. */
258 done = xchg(&create->done, NULL);
259 if (!done) {
260 kfree(create);
261 do_exit(-EINTR);
262 }
263
264 if (!self) {
265 create->result = ERR_PTR(-ENOMEM);
266 complete(done);
267 do_exit(-ENOMEM);
268 }
269
270 self->threadfn = threadfn;
271 self->data = data;
272 init_completion(&self->exited);
273 init_completion(&self->parked);
274 current->vfork_done = &self->exited;
275
276 /* OK, tell user we're spawned, wait for stop or wakeup */
277 __set_current_state(TASK_UNINTERRUPTIBLE);
278 create->result = current;
279 /*
280 * Thread is going to call schedule(), do not preempt it,
281 * or the creator may spend more time in wait_task_inactive().
282 */
283 preempt_disable();
284 complete(done);
285 schedule_preempt_disabled();
286 preempt_enable();
287
288 ret = -EINTR;
289 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
290 cgroup_kthread_ready();
291 __kthread_parkme(self);
292 ret = threadfn(data);
293 }
294 do_exit(ret);
295}
296
297/* called from do_fork() to get node information for about to be created task */
298int tsk_fork_get_node(struct task_struct *tsk)
299{
300#ifdef CONFIG_NUMA
301 if (tsk == kthreadd_task)
302 return tsk->pref_node_fork;
303#endif
304 return NUMA_NO_NODE;
305}
306
307static void create_kthread(struct kthread_create_info *create)
308{
309 int pid;
310
311#ifdef CONFIG_NUMA
312 current->pref_node_fork = create->node;
313#endif
314 /* We want our own signal handler (we take no signals by default). */
315 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
316 if (pid < 0) {
317 /* If user was SIGKILLed, I release the structure. */
318 struct completion *done = xchg(&create->done, NULL);
319
320 if (!done) {
321 kfree(create);
322 return;
323 }
324 create->result = ERR_PTR(pid);
325 complete(done);
326 }
327}
328
329static __printf(4, 0)
330struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
331 void *data, int node,
332 const char namefmt[],
333 va_list args)
334{
335 DECLARE_COMPLETION_ONSTACK(done);
336 struct task_struct *task;
337 struct kthread_create_info *create = kmalloc(sizeof(*create),
338 GFP_KERNEL);
339
340 if (!create)
341 return ERR_PTR(-ENOMEM);
342 create->threadfn = threadfn;
343 create->data = data;
344 create->node = node;
345 create->done = &done;
346
347 spin_lock(&kthread_create_lock);
348 list_add_tail(&create->list, &kthread_create_list);
349 spin_unlock(&kthread_create_lock);
350
351 wake_up_process(kthreadd_task);
352 /*
353 * Wait for completion in killable state, for I might be chosen by
354 * the OOM killer while kthreadd is trying to allocate memory for
355 * new kernel thread.
356 */
357 if (unlikely(wait_for_completion_killable(&done))) {
358 /*
359 * If I was SIGKILLed before kthreadd (or new kernel thread)
360 * calls complete(), leave the cleanup of this structure to
361 * that thread.
362 */
363 if (xchg(&create->done, NULL))
364 return ERR_PTR(-EINTR);
365 /*
366 * kthreadd (or new kernel thread) will call complete()
367 * shortly.
368 */
369 wait_for_completion(&done);
370 }
371 task = create->result;
372 if (!IS_ERR(task)) {
373 static const struct sched_param param = { .sched_priority = 0 };
374 char name[TASK_COMM_LEN];
375
376 /*
377 * task is already visible to other tasks, so updating
378 * COMM must be protected.
379 */
380 vsnprintf(name, sizeof(name), namefmt, args);
381 set_task_comm(task, name);
382 /*
383 * root may have changed our (kthreadd's) priority or CPU mask.
384 * The kernel thread should not inherit these properties.
385 */
386 sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
387 set_cpus_allowed_ptr(task,
388 housekeeping_cpumask(HK_FLAG_KTHREAD));
389 }
390 kfree(create);
391 return task;
392}
393
394/**
395 * kthread_create_on_node - create a kthread.
396 * @threadfn: the function to run until signal_pending(current).
397 * @data: data ptr for @threadfn.
398 * @node: task and thread structures for the thread are allocated on this node
399 * @namefmt: printf-style name for the thread.
400 *
401 * Description: This helper function creates and names a kernel
402 * thread. The thread will be stopped: use wake_up_process() to start
403 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
404 * is affine to all CPUs.
405 *
406 * If thread is going to be bound on a particular cpu, give its node
407 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
408 * When woken, the thread will run @threadfn() with @data as its
409 * argument. @threadfn() can either call do_exit() directly if it is a
410 * standalone thread for which no one will call kthread_stop(), or
411 * return when 'kthread_should_stop()' is true (which means
412 * kthread_stop() has been called). The return value should be zero
413 * or a negative error number; it will be passed to kthread_stop().
414 *
415 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
416 */
417struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
418 void *data, int node,
419 const char namefmt[],
420 ...)
421{
422 struct task_struct *task;
423 va_list args;
424
425 va_start(args, namefmt);
426 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
427 va_end(args);
428
429 return task;
430}
431EXPORT_SYMBOL(kthread_create_on_node);
432
433static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
434{
435 unsigned long flags;
436
437 if (!wait_task_inactive(p, state)) {
438 WARN_ON(1);
439 return;
440 }
441
442 /* It's safe because the task is inactive. */
443 raw_spin_lock_irqsave(&p->pi_lock, flags);
444 do_set_cpus_allowed(p, mask);
445 p->flags |= PF_NO_SETAFFINITY;
446 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
447}
448
449static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
450{
451 __kthread_bind_mask(p, cpumask_of(cpu), state);
452}
453
454void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
455{
456 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
457}
458
459/**
460 * kthread_bind - bind a just-created kthread to a cpu.
461 * @p: thread created by kthread_create().
462 * @cpu: cpu (might not be online, must be possible) for @k to run on.
463 *
464 * Description: This function is equivalent to set_cpus_allowed(),
465 * except that @cpu doesn't need to be online, and the thread must be
466 * stopped (i.e., just returned from kthread_create()).
467 */
468void kthread_bind(struct task_struct *p, unsigned int cpu)
469{
470 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
471}
472EXPORT_SYMBOL(kthread_bind);
473
474/**
475 * kthread_create_on_cpu - Create a cpu bound kthread
476 * @threadfn: the function to run until signal_pending(current).
477 * @data: data ptr for @threadfn.
478 * @cpu: The cpu on which the thread should be bound,
479 * @namefmt: printf-style name for the thread. Format is restricted
480 * to "name.*%u". Code fills in cpu number.
481 *
482 * Description: This helper function creates and names a kernel thread
483 */
484struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
485 void *data, unsigned int cpu,
486 const char *namefmt)
487{
488 struct task_struct *p;
489
490 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
491 cpu);
492 if (IS_ERR(p))
493 return p;
494 kthread_bind(p, cpu);
495 /* CPU hotplug need to bind once again when unparking the thread. */
496 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
497 to_kthread(p)->cpu = cpu;
498 return p;
499}
500
501/**
502 * kthread_unpark - unpark a thread created by kthread_create().
503 * @k: thread created by kthread_create().
504 *
505 * Sets kthread_should_park() for @k to return false, wakes it, and
506 * waits for it to return. If the thread is marked percpu then its
507 * bound to the cpu again.
508 */
509void kthread_unpark(struct task_struct *k)
510{
511 struct kthread *kthread = to_kthread(k);
512
513 /*
514 * Newly created kthread was parked when the CPU was offline.
515 * The binding was lost and we need to set it again.
516 */
517 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
518 __kthread_bind(k, kthread->cpu, TASK_PARKED);
519
520 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
521 /*
522 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
523 */
524 wake_up_state(k, TASK_PARKED);
525}
526EXPORT_SYMBOL_GPL(kthread_unpark);
527
528/**
529 * kthread_park - park a thread created by kthread_create().
530 * @k: thread created by kthread_create().
531 *
532 * Sets kthread_should_park() for @k to return true, wakes it, and
533 * waits for it to return. This can also be called after kthread_create()
534 * instead of calling wake_up_process(): the thread will park without
535 * calling threadfn().
536 *
537 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
538 * If called by the kthread itself just the park bit is set.
539 */
540int kthread_park(struct task_struct *k)
541{
542 struct kthread *kthread = to_kthread(k);
543
544 if (WARN_ON(k->flags & PF_EXITING))
545 return -ENOSYS;
546
547 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
548 return -EBUSY;
549
550 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
551 if (k != current) {
552 wake_up_process(k);
553 /*
554 * Wait for __kthread_parkme() to complete(), this means we
555 * _will_ have TASK_PARKED and are about to call schedule().
556 */
557 wait_for_completion(&kthread->parked);
558 /*
559 * Now wait for that schedule() to complete and the task to
560 * get scheduled out.
561 */
562 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
563 }
564
565 return 0;
566}
567EXPORT_SYMBOL_GPL(kthread_park);
568
569/**
570 * kthread_stop - stop a thread created by kthread_create().
571 * @k: thread created by kthread_create().
572 *
573 * Sets kthread_should_stop() for @k to return true, wakes it, and
574 * waits for it to exit. This can also be called after kthread_create()
575 * instead of calling wake_up_process(): the thread will exit without
576 * calling threadfn().
577 *
578 * If threadfn() may call do_exit() itself, the caller must ensure
579 * task_struct can't go away.
580 *
581 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
582 * was never called.
583 */
584int kthread_stop(struct task_struct *k)
585{
586 struct kthread *kthread;
587 int ret;
588
589 trace_sched_kthread_stop(k);
590
591 get_task_struct(k);
592 kthread = to_kthread(k);
593 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
594 kthread_unpark(k);
595 wake_up_process(k);
596 wait_for_completion(&kthread->exited);
597 ret = k->exit_code;
598 put_task_struct(k);
599
600 trace_sched_kthread_stop_ret(ret);
601 return ret;
602}
603EXPORT_SYMBOL(kthread_stop);
604
605int kthreadd(void *unused)
606{
607 struct task_struct *tsk = current;
608
609 /* Setup a clean context for our children to inherit. */
610 set_task_comm(tsk, "kthreadd");
611 ignore_signals(tsk);
612 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
613 set_mems_allowed(node_states[N_MEMORY]);
614
615 current->flags |= PF_NOFREEZE;
616 cgroup_init_kthreadd();
617
618 for (;;) {
619 set_current_state(TASK_INTERRUPTIBLE);
620 if (list_empty(&kthread_create_list))
621 schedule();
622 __set_current_state(TASK_RUNNING);
623
624 spin_lock(&kthread_create_lock);
625 while (!list_empty(&kthread_create_list)) {
626 struct kthread_create_info *create;
627
628 create = list_entry(kthread_create_list.next,
629 struct kthread_create_info, list);
630 list_del_init(&create->list);
631 spin_unlock(&kthread_create_lock);
632
633 create_kthread(create);
634
635 spin_lock(&kthread_create_lock);
636 }
637 spin_unlock(&kthread_create_lock);
638 }
639
640 return 0;
641}
642
643void __kthread_init_worker(struct kthread_worker *worker,
644 const char *name,
645 struct lock_class_key *key)
646{
647 memset(worker, 0, sizeof(struct kthread_worker));
648 raw_spin_lock_init(&worker->lock);
649 lockdep_set_class_and_name(&worker->lock, key, name);
650 INIT_LIST_HEAD(&worker->work_list);
651 INIT_LIST_HEAD(&worker->delayed_work_list);
652}
653EXPORT_SYMBOL_GPL(__kthread_init_worker);
654
655/**
656 * kthread_worker_fn - kthread function to process kthread_worker
657 * @worker_ptr: pointer to initialized kthread_worker
658 *
659 * This function implements the main cycle of kthread worker. It processes
660 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
661 * is empty.
662 *
663 * The works are not allowed to keep any locks, disable preemption or interrupts
664 * when they finish. There is defined a safe point for freezing when one work
665 * finishes and before a new one is started.
666 *
667 * Also the works must not be handled by more than one worker at the same time,
668 * see also kthread_queue_work().
669 */
670int kthread_worker_fn(void *worker_ptr)
671{
672 struct kthread_worker *worker = worker_ptr;
673 struct kthread_work *work;
674
675 /*
676 * FIXME: Update the check and remove the assignment when all kthread
677 * worker users are created using kthread_create_worker*() functions.
678 */
679 WARN_ON(worker->task && worker->task != current);
680 worker->task = current;
681
682 if (worker->flags & KTW_FREEZABLE)
683 set_freezable();
684
685repeat:
686 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
687
688 if (kthread_should_stop()) {
689 __set_current_state(TASK_RUNNING);
690 raw_spin_lock_irq(&worker->lock);
691 worker->task = NULL;
692 raw_spin_unlock_irq(&worker->lock);
693 return 0;
694 }
695
696 work = NULL;
697 raw_spin_lock_irq(&worker->lock);
698 if (!list_empty(&worker->work_list)) {
699 work = list_first_entry(&worker->work_list,
700 struct kthread_work, node);
701 list_del_init(&work->node);
702 }
703 worker->current_work = work;
704 raw_spin_unlock_irq(&worker->lock);
705
706 if (work) {
707 __set_current_state(TASK_RUNNING);
708 work->func(work);
709 } else if (!freezing(current))
710 schedule();
711
712 try_to_freeze();
713 cond_resched();
714 goto repeat;
715}
716EXPORT_SYMBOL_GPL(kthread_worker_fn);
717
718static __printf(3, 0) struct kthread_worker *
719__kthread_create_worker(int cpu, unsigned int flags,
720 const char namefmt[], va_list args)
721{
722 struct kthread_worker *worker;
723 struct task_struct *task;
724 int node = NUMA_NO_NODE;
725
726 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
727 if (!worker)
728 return ERR_PTR(-ENOMEM);
729
730 kthread_init_worker(worker);
731
732 if (cpu >= 0)
733 node = cpu_to_node(cpu);
734
735 task = __kthread_create_on_node(kthread_worker_fn, worker,
736 node, namefmt, args);
737 if (IS_ERR(task))
738 goto fail_task;
739
740 if (cpu >= 0)
741 kthread_bind(task, cpu);
742
743 worker->flags = flags;
744 worker->task = task;
745 wake_up_process(task);
746 return worker;
747
748fail_task:
749 kfree(worker);
750 return ERR_CAST(task);
751}
752
753/**
754 * kthread_create_worker - create a kthread worker
755 * @flags: flags modifying the default behavior of the worker
756 * @namefmt: printf-style name for the kthread worker (task).
757 *
758 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
759 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
760 * when the worker was SIGKILLed.
761 */
762struct kthread_worker *
763kthread_create_worker(unsigned int flags, const char namefmt[], ...)
764{
765 struct kthread_worker *worker;
766 va_list args;
767
768 va_start(args, namefmt);
769 worker = __kthread_create_worker(-1, flags, namefmt, args);
770 va_end(args);
771
772 return worker;
773}
774EXPORT_SYMBOL(kthread_create_worker);
775
776/**
777 * kthread_create_worker_on_cpu - create a kthread worker and bind it
778 * to a given CPU and the associated NUMA node.
779 * @cpu: CPU number
780 * @flags: flags modifying the default behavior of the worker
781 * @namefmt: printf-style name for the kthread worker (task).
782 *
783 * Use a valid CPU number if you want to bind the kthread worker
784 * to the given CPU and the associated NUMA node.
785 *
786 * A good practice is to add the cpu number also into the worker name.
787 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
788 *
789 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
790 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
791 * when the worker was SIGKILLed.
792 */
793struct kthread_worker *
794kthread_create_worker_on_cpu(int cpu, unsigned int flags,
795 const char namefmt[], ...)
796{
797 struct kthread_worker *worker;
798 va_list args;
799
800 va_start(args, namefmt);
801 worker = __kthread_create_worker(cpu, flags, namefmt, args);
802 va_end(args);
803
804 return worker;
805}
806EXPORT_SYMBOL(kthread_create_worker_on_cpu);
807
808/*
809 * Returns true when the work could not be queued at the moment.
810 * It happens when it is already pending in a worker list
811 * or when it is being cancelled.
812 */
813static inline bool queuing_blocked(struct kthread_worker *worker,
814 struct kthread_work *work)
815{
816 lockdep_assert_held(&worker->lock);
817
818 return !list_empty(&work->node) || work->canceling;
819}
820
821static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
822 struct kthread_work *work)
823{
824 lockdep_assert_held(&worker->lock);
825 WARN_ON_ONCE(!list_empty(&work->node));
826 /* Do not use a work with >1 worker, see kthread_queue_work() */
827 WARN_ON_ONCE(work->worker && work->worker != worker);
828}
829
830/* insert @work before @pos in @worker */
831static void kthread_insert_work(struct kthread_worker *worker,
832 struct kthread_work *work,
833 struct list_head *pos)
834{
835 kthread_insert_work_sanity_check(worker, work);
836
837 list_add_tail(&work->node, pos);
838 work->worker = worker;
839 if (!worker->current_work && likely(worker->task))
840 wake_up_process(worker->task);
841}
842
843/**
844 * kthread_queue_work - queue a kthread_work
845 * @worker: target kthread_worker
846 * @work: kthread_work to queue
847 *
848 * Queue @work to work processor @task for async execution. @task
849 * must have been created with kthread_worker_create(). Returns %true
850 * if @work was successfully queued, %false if it was already pending.
851 *
852 * Reinitialize the work if it needs to be used by another worker.
853 * For example, when the worker was stopped and started again.
854 */
855bool kthread_queue_work(struct kthread_worker *worker,
856 struct kthread_work *work)
857{
858 bool ret = false;
859 unsigned long flags;
860
861 raw_spin_lock_irqsave(&worker->lock, flags);
862 if (!queuing_blocked(worker, work)) {
863 kthread_insert_work(worker, work, &worker->work_list);
864 ret = true;
865 }
866 raw_spin_unlock_irqrestore(&worker->lock, flags);
867 return ret;
868}
869EXPORT_SYMBOL_GPL(kthread_queue_work);
870
871/**
872 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
873 * delayed work when the timer expires.
874 * @t: pointer to the expired timer
875 *
876 * The format of the function is defined by struct timer_list.
877 * It should have been called from irqsafe timer with irq already off.
878 */
879void kthread_delayed_work_timer_fn(struct timer_list *t)
880{
881 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
882 struct kthread_work *work = &dwork->work;
883 struct kthread_worker *worker = work->worker;
884 unsigned long flags;
885
886 /*
887 * This might happen when a pending work is reinitialized.
888 * It means that it is used a wrong way.
889 */
890 if (WARN_ON_ONCE(!worker))
891 return;
892
893 raw_spin_lock_irqsave(&worker->lock, flags);
894 /* Work must not be used with >1 worker, see kthread_queue_work(). */
895 WARN_ON_ONCE(work->worker != worker);
896
897 /* Move the work from worker->delayed_work_list. */
898 WARN_ON_ONCE(list_empty(&work->node));
899 list_del_init(&work->node);
900 if (!work->canceling)
901 kthread_insert_work(worker, work, &worker->work_list);
902
903 raw_spin_unlock_irqrestore(&worker->lock, flags);
904}
905EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
906
907static void __kthread_queue_delayed_work(struct kthread_worker *worker,
908 struct kthread_delayed_work *dwork,
909 unsigned long delay)
910{
911 struct timer_list *timer = &dwork->timer;
912 struct kthread_work *work = &dwork->work;
913
914 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
915
916 /*
917 * If @delay is 0, queue @dwork->work immediately. This is for
918 * both optimization and correctness. The earliest @timer can
919 * expire is on the closest next tick and delayed_work users depend
920 * on that there's no such delay when @delay is 0.
921 */
922 if (!delay) {
923 kthread_insert_work(worker, work, &worker->work_list);
924 return;
925 }
926
927 /* Be paranoid and try to detect possible races already now. */
928 kthread_insert_work_sanity_check(worker, work);
929
930 list_add(&work->node, &worker->delayed_work_list);
931 work->worker = worker;
932 timer->expires = jiffies + delay;
933 add_timer(timer);
934}
935
936/**
937 * kthread_queue_delayed_work - queue the associated kthread work
938 * after a delay.
939 * @worker: target kthread_worker
940 * @dwork: kthread_delayed_work to queue
941 * @delay: number of jiffies to wait before queuing
942 *
943 * If the work has not been pending it starts a timer that will queue
944 * the work after the given @delay. If @delay is zero, it queues the
945 * work immediately.
946 *
947 * Return: %false if the @work has already been pending. It means that
948 * either the timer was running or the work was queued. It returns %true
949 * otherwise.
950 */
951bool kthread_queue_delayed_work(struct kthread_worker *worker,
952 struct kthread_delayed_work *dwork,
953 unsigned long delay)
954{
955 struct kthread_work *work = &dwork->work;
956 unsigned long flags;
957 bool ret = false;
958
959 raw_spin_lock_irqsave(&worker->lock, flags);
960
961 if (!queuing_blocked(worker, work)) {
962 __kthread_queue_delayed_work(worker, dwork, delay);
963 ret = true;
964 }
965
966 raw_spin_unlock_irqrestore(&worker->lock, flags);
967 return ret;
968}
969EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
970
971struct kthread_flush_work {
972 struct kthread_work work;
973 struct completion done;
974};
975
976static void kthread_flush_work_fn(struct kthread_work *work)
977{
978 struct kthread_flush_work *fwork =
979 container_of(work, struct kthread_flush_work, work);
980 complete(&fwork->done);
981}
982
983/**
984 * kthread_flush_work - flush a kthread_work
985 * @work: work to flush
986 *
987 * If @work is queued or executing, wait for it to finish execution.
988 */
989void kthread_flush_work(struct kthread_work *work)
990{
991 struct kthread_flush_work fwork = {
992 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
993 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
994 };
995 struct kthread_worker *worker;
996 bool noop = false;
997
998 worker = work->worker;
999 if (!worker)
1000 return;
1001
1002 raw_spin_lock_irq(&worker->lock);
1003 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1004 WARN_ON_ONCE(work->worker != worker);
1005
1006 if (!list_empty(&work->node))
1007 kthread_insert_work(worker, &fwork.work, work->node.next);
1008 else if (worker->current_work == work)
1009 kthread_insert_work(worker, &fwork.work,
1010 worker->work_list.next);
1011 else
1012 noop = true;
1013
1014 raw_spin_unlock_irq(&worker->lock);
1015
1016 if (!noop)
1017 wait_for_completion(&fwork.done);
1018}
1019EXPORT_SYMBOL_GPL(kthread_flush_work);
1020
1021/*
1022 * This function removes the work from the worker queue. Also it makes sure
1023 * that it won't get queued later via the delayed work's timer.
1024 *
1025 * The work might still be in use when this function finishes. See the
1026 * current_work proceed by the worker.
1027 *
1028 * Return: %true if @work was pending and successfully canceled,
1029 * %false if @work was not pending
1030 */
1031static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
1032 unsigned long *flags)
1033{
1034 /* Try to cancel the timer if exists. */
1035 if (is_dwork) {
1036 struct kthread_delayed_work *dwork =
1037 container_of(work, struct kthread_delayed_work, work);
1038 struct kthread_worker *worker = work->worker;
1039
1040 /*
1041 * del_timer_sync() must be called to make sure that the timer
1042 * callback is not running. The lock must be temporary released
1043 * to avoid a deadlock with the callback. In the meantime,
1044 * any queuing is blocked by setting the canceling counter.
1045 */
1046 work->canceling++;
1047 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1048 del_timer_sync(&dwork->timer);
1049 raw_spin_lock_irqsave(&worker->lock, *flags);
1050 work->canceling--;
1051 }
1052
1053 /*
1054 * Try to remove the work from a worker list. It might either
1055 * be from worker->work_list or from worker->delayed_work_list.
1056 */
1057 if (!list_empty(&work->node)) {
1058 list_del_init(&work->node);
1059 return true;
1060 }
1061
1062 return false;
1063}
1064
1065/**
1066 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1067 * @worker: kthread worker to use
1068 * @dwork: kthread delayed work to queue
1069 * @delay: number of jiffies to wait before queuing
1070 *
1071 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1072 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1073 * @work is guaranteed to be queued immediately.
1074 *
1075 * Return: %true if @dwork was pending and its timer was modified,
1076 * %false otherwise.
1077 *
1078 * A special case is when the work is being canceled in parallel.
1079 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1080 * or yet another kthread_mod_delayed_work() call. We let the other command
1081 * win and return %false here. The caller is supposed to synchronize these
1082 * operations a reasonable way.
1083 *
1084 * This function is safe to call from any context including IRQ handler.
1085 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1086 * for details.
1087 */
1088bool kthread_mod_delayed_work(struct kthread_worker *worker,
1089 struct kthread_delayed_work *dwork,
1090 unsigned long delay)
1091{
1092 struct kthread_work *work = &dwork->work;
1093 unsigned long flags;
1094 int ret = false;
1095
1096 raw_spin_lock_irqsave(&worker->lock, flags);
1097
1098 /* Do not bother with canceling when never queued. */
1099 if (!work->worker)
1100 goto fast_queue;
1101
1102 /* Work must not be used with >1 worker, see kthread_queue_work() */
1103 WARN_ON_ONCE(work->worker != worker);
1104
1105 /* Do not fight with another command that is canceling this work. */
1106 if (work->canceling)
1107 goto out;
1108
1109 ret = __kthread_cancel_work(work, true, &flags);
1110fast_queue:
1111 __kthread_queue_delayed_work(worker, dwork, delay);
1112out:
1113 raw_spin_unlock_irqrestore(&worker->lock, flags);
1114 return ret;
1115}
1116EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1117
1118static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1119{
1120 struct kthread_worker *worker = work->worker;
1121 unsigned long flags;
1122 int ret = false;
1123
1124 if (!worker)
1125 goto out;
1126
1127 raw_spin_lock_irqsave(&worker->lock, flags);
1128 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1129 WARN_ON_ONCE(work->worker != worker);
1130
1131 ret = __kthread_cancel_work(work, is_dwork, &flags);
1132
1133 if (worker->current_work != work)
1134 goto out_fast;
1135
1136 /*
1137 * The work is in progress and we need to wait with the lock released.
1138 * In the meantime, block any queuing by setting the canceling counter.
1139 */
1140 work->canceling++;
1141 raw_spin_unlock_irqrestore(&worker->lock, flags);
1142 kthread_flush_work(work);
1143 raw_spin_lock_irqsave(&worker->lock, flags);
1144 work->canceling--;
1145
1146out_fast:
1147 raw_spin_unlock_irqrestore(&worker->lock, flags);
1148out:
1149 return ret;
1150}
1151
1152/**
1153 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1154 * @work: the kthread work to cancel
1155 *
1156 * Cancel @work and wait for its execution to finish. This function
1157 * can be used even if the work re-queues itself. On return from this
1158 * function, @work is guaranteed to be not pending or executing on any CPU.
1159 *
1160 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1161 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1162 *
1163 * The caller must ensure that the worker on which @work was last
1164 * queued can't be destroyed before this function returns.
1165 *
1166 * Return: %true if @work was pending, %false otherwise.
1167 */
1168bool kthread_cancel_work_sync(struct kthread_work *work)
1169{
1170 return __kthread_cancel_work_sync(work, false);
1171}
1172EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1173
1174/**
1175 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1176 * wait for it to finish.
1177 * @dwork: the kthread delayed work to cancel
1178 *
1179 * This is kthread_cancel_work_sync() for delayed works.
1180 *
1181 * Return: %true if @dwork was pending, %false otherwise.
1182 */
1183bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1184{
1185 return __kthread_cancel_work_sync(&dwork->work, true);
1186}
1187EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1188
1189/**
1190 * kthread_flush_worker - flush all current works on a kthread_worker
1191 * @worker: worker to flush
1192 *
1193 * Wait until all currently executing or pending works on @worker are
1194 * finished.
1195 */
1196void kthread_flush_worker(struct kthread_worker *worker)
1197{
1198 struct kthread_flush_work fwork = {
1199 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1200 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1201 };
1202
1203 kthread_queue_work(worker, &fwork.work);
1204 wait_for_completion(&fwork.done);
1205}
1206EXPORT_SYMBOL_GPL(kthread_flush_worker);
1207
1208/**
1209 * kthread_destroy_worker - destroy a kthread worker
1210 * @worker: worker to be destroyed
1211 *
1212 * Flush and destroy @worker. The simple flush is enough because the kthread
1213 * worker API is used only in trivial scenarios. There are no multi-step state
1214 * machines needed.
1215 */
1216void kthread_destroy_worker(struct kthread_worker *worker)
1217{
1218 struct task_struct *task;
1219
1220 task = worker->task;
1221 if (WARN_ON(!task))
1222 return;
1223
1224 kthread_flush_worker(worker);
1225 kthread_stop(task);
1226 WARN_ON(!list_empty(&worker->work_list));
1227 kfree(worker);
1228}
1229EXPORT_SYMBOL(kthread_destroy_worker);
1230
1231/**
1232 * kthread_use_mm - make the calling kthread operate on an address space
1233 * @mm: address space to operate on
1234 */
1235void kthread_use_mm(struct mm_struct *mm)
1236{
1237 struct mm_struct *active_mm;
1238 struct task_struct *tsk = current;
1239
1240 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1241 WARN_ON_ONCE(tsk->mm);
1242
1243 task_lock(tsk);
1244 /* Hold off tlb flush IPIs while switching mm's */
1245 local_irq_disable();
1246 active_mm = tsk->active_mm;
1247 if (active_mm != mm) {
1248 mmgrab(mm);
1249 tsk->active_mm = mm;
1250 }
1251 tsk->mm = mm;
1252 switch_mm_irqs_off(active_mm, mm, tsk);
1253 local_irq_enable();
1254 task_unlock(tsk);
1255#ifdef finish_arch_post_lock_switch
1256 finish_arch_post_lock_switch();
1257#endif
1258
1259 if (active_mm != mm)
1260 mmdrop(active_mm);
1261
1262 to_kthread(tsk)->oldfs = force_uaccess_begin();
1263}
1264EXPORT_SYMBOL_GPL(kthread_use_mm);
1265
1266/**
1267 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1268 * @mm: address space to operate on
1269 */
1270void kthread_unuse_mm(struct mm_struct *mm)
1271{
1272 struct task_struct *tsk = current;
1273
1274 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1275 WARN_ON_ONCE(!tsk->mm);
1276
1277 force_uaccess_end(to_kthread(tsk)->oldfs);
1278
1279 task_lock(tsk);
1280 sync_mm_rss(mm);
1281 local_irq_disable();
1282 tsk->mm = NULL;
1283 /* active_mm is still 'mm' */
1284 enter_lazy_tlb(mm, tsk);
1285 local_irq_enable();
1286 task_unlock(tsk);
1287}
1288EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1289
1290#ifdef CONFIG_BLK_CGROUP
1291/**
1292 * kthread_associate_blkcg - associate blkcg to current kthread
1293 * @css: the cgroup info
1294 *
1295 * Current thread must be a kthread. The thread is running jobs on behalf of
1296 * other threads. In some cases, we expect the jobs attach cgroup info of
1297 * original threads instead of that of current thread. This function stores
1298 * original thread's cgroup info in current kthread context for later
1299 * retrieval.
1300 */
1301void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1302{
1303 struct kthread *kthread;
1304
1305 if (!(current->flags & PF_KTHREAD))
1306 return;
1307 kthread = to_kthread(current);
1308 if (!kthread)
1309 return;
1310
1311 if (kthread->blkcg_css) {
1312 css_put(kthread->blkcg_css);
1313 kthread->blkcg_css = NULL;
1314 }
1315 if (css) {
1316 css_get(css);
1317 kthread->blkcg_css = css;
1318 }
1319}
1320EXPORT_SYMBOL(kthread_associate_blkcg);
1321
1322/**
1323 * kthread_blkcg - get associated blkcg css of current kthread
1324 *
1325 * Current thread must be a kthread.
1326 */
1327struct cgroup_subsys_state *kthread_blkcg(void)
1328{
1329 struct kthread *kthread;
1330
1331 if (current->flags & PF_KTHREAD) {
1332 kthread = to_kthread(current);
1333 if (kthread)
1334 return kthread->blkcg_css;
1335 }
1336 return NULL;
1337}
1338EXPORT_SYMBOL(kthread_blkcg);
1339#endif