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