[linux-block.git] / fs / btrfs / async-thread.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Copyright (C) 2007 Oracle.  All rights reserved.
 * Copyright (C) 2014 Fujitsu.  All rights reserved.
 */

#include <linux/kthread.h>
#include <linux/slab.h>
#include <linux/list.h>
#include <linux/spinlock.h>
#include <linux/freezer.h>
#include "async-thread.h"
#include "ctree.h"

enum {
	WORK_DONE_BIT,
	WORK_ORDER_DONE_BIT,
	WORK_HIGH_PRIO_BIT,
};

#define NO_THRESHOLD (-1)
#define DFT_THRESHOLD (32)

struct __btrfs_workqueue {
	struct workqueue_struct *normal_wq;

	/* File system this workqueue services */
	struct btrfs_fs_info *fs_info;

	/* List head pointing to ordered work list */
	struct list_head ordered_list;

	/* Spinlock for ordered_list */
	spinlock_t list_lock;

	/* Thresholding related variants */
	atomic_t pending;

	/* Up limit of concurrency workers */
	int limit_active;

	/* Current number of concurrency workers */
	int current_active;

	/* Threshold to change current_active */
	int thresh;
	unsigned int count;
	spinlock_t thres_lock;
};

struct btrfs_workqueue {
	struct __btrfs_workqueue *normal;
	struct __btrfs_workqueue *high;
};

struct btrfs_fs_info * __pure btrfs_workqueue_owner(const struct __btrfs_workqueue *wq)
{
	return wq->fs_info;
}

struct btrfs_fs_info * __pure btrfs_work_owner(const struct btrfs_work *work)
{
	return work->wq->fs_info;
}

bool btrfs_workqueue_normal_congested(const struct btrfs_workqueue *wq)
{
	/*
	 * We could compare wq->normal->pending with num_online_cpus()
	 * to support "thresh == NO_THRESHOLD" case, but it requires
	 * moving up atomic_inc/dec in thresh_queue/exec_hook. Let's
	 * postpone it until someone needs the support of that case.
	 */
	if (wq->normal->thresh == NO_THRESHOLD)
		return false;

	return atomic_read(&wq->normal->pending) > wq->normal->thresh * 2;
}

static struct __btrfs_workqueue *
__btrfs_alloc_workqueue(struct btrfs_fs_info *fs_info, const char *name,
			unsigned int flags, int limit_active, int thresh)
{
	struct __btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_KERNEL);

	if (!ret)
		return NULL;

	ret->fs_info = fs_info;
	ret->limit_active = limit_active;
	atomic_set(&ret->pending, 0);
	if (thresh == 0)
		thresh = DFT_THRESHOLD;
	/* For low threshold, disabling threshold is a better choice */
	if (thresh < DFT_THRESHOLD) {
		ret->current_active = limit_active;
		ret->thresh = NO_THRESHOLD;
	} else {
		/*
		 * For threshold-able wq, let its concurrency grow on demand.
		 * Use minimal max_active at alloc time to reduce resource
		 * usage.
		 */
		ret->current_active = 1;
		ret->thresh = thresh;
	}

	if (flags & WQ_HIGHPRI)
		ret->normal_wq = alloc_workqueue("btrfs-%s-high", flags,
						 ret->current_active, name);
	else
		ret->normal_wq = alloc_workqueue("btrfs-%s", flags,
						 ret->current_active, name);
	if (!ret->normal_wq) {
		kfree(ret);
		return NULL;
	}

	INIT_LIST_HEAD(&ret->ordered_list);
	spin_lock_init(&ret->list_lock);
	spin_lock_init(&ret->thres_lock);
	trace_btrfs_workqueue_alloc(ret, name, flags & WQ_HIGHPRI);
	return ret;
}

static inline void
__btrfs_destroy_workqueue(struct __btrfs_workqueue *wq);

struct btrfs_workqueue *btrfs_alloc_workqueue(struct btrfs_fs_info *fs_info,
					      const char *name,
					      unsigned int flags,
					      int limit_active,
					      int thresh)
{
	struct btrfs_workqueue *ret = kzalloc(sizeof(*ret), GFP_KERNEL);

	if (!ret)
		return NULL;

	ret->normal = __btrfs_alloc_workqueue(fs_info, name,
					      flags & ~WQ_HIGHPRI,
					      limit_active, thresh);
	if (!ret->normal) {
		kfree(ret);
		return NULL;
	}

	if (flags & WQ_HIGHPRI) {
		ret->high = __btrfs_alloc_workqueue(fs_info, name, flags,
						    limit_active, thresh);
		if (!ret->high) {
			__btrfs_destroy_workqueue(ret->normal);
			kfree(ret);
			return NULL;
		}
	}
	return ret;
}

/*
 * Hook for threshold which will be called in btrfs_queue_work.
 * This hook WILL be called in IRQ handler context,
 * so workqueue_set_max_active MUST NOT be called in this hook
 */
static inline void thresh_queue_hook(struct __btrfs_workqueue *wq)
{
	if (wq->thresh == NO_THRESHOLD)
		return;
	atomic_inc(&wq->pending);
}

/*
 * Hook for threshold which will be called before executing the work,
 * This hook is called in kthread content.
 * So workqueue_set_max_active is called here.
 */
static inline void thresh_exec_hook(struct __btrfs_workqueue *wq)
{
	int new_current_active;
	long pending;
	int need_change = 0;

	if (wq->thresh == NO_THRESHOLD)
		return;

	atomic_dec(&wq->pending);
	spin_lock(&wq->thres_lock);
	/*
	 * Use wq->count to limit the calling frequency of
	 * workqueue_set_max_active.
	 */
	wq->count++;
	wq->count %= (wq->thresh / 4);
	if (!wq->count)
		goto  out;
	new_current_active = wq->current_active;

	/*
	 * pending may be changed later, but it's OK since we really
	 * don't need it so accurate to calculate new_max_active.
	 */
	pending = atomic_read(&wq->pending);
	if (pending > wq->thresh)
		new_current_active++;
	if (pending < wq->thresh / 2)
		new_current_active--;
	new_current_active = clamp_val(new_current_active, 1, wq->limit_active);
	if (new_current_active != wq->current_active)  {
		need_change = 1;
		wq->current_active = new_current_active;
	}
out:
	spin_unlock(&wq->thres_lock);

	if (need_change) {
		workqueue_set_max_active(wq->normal_wq, wq->current_active);
	}
}

static void run_ordered_work(struct __btrfs_workqueue *wq,
			     struct btrfs_work *self)
{
	struct list_head *list = &wq->ordered_list;
	struct btrfs_work *work;
	spinlock_t *lock = &wq->list_lock;
	unsigned long flags;
	bool free_self = false;

	while (1) {
		spin_lock_irqsave(lock, flags);
		if (list_empty(list))
			break;
		work = list_entry(list->next, struct btrfs_work,
				  ordered_list);
		if (!test_bit(WORK_DONE_BIT, &work->flags))
			break;
		/*
		 * Orders all subsequent loads after reading WORK_DONE_BIT,
		 * paired with the smp_mb__before_atomic in btrfs_work_helper
		 * this guarantees that the ordered function will see all
		 * updates from ordinary work function.
		 */
		smp_rmb();

		/*
		 * we are going to call the ordered done function, but
		 * we leave the work item on the list as a barrier so
		 * that later work items that are done don't have their
		 * functions called before this one returns
		 */
		if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
			break;
		trace_btrfs_ordered_sched(work);
		spin_unlock_irqrestore(lock, flags);
		work->ordered_func(work);

		/* now take the lock again and drop our item from the list */
		spin_lock_irqsave(lock, flags);
		list_del(&work->ordered_list);
		spin_unlock_irqrestore(lock, flags);

		if (work == self) {
			/*
			 * This is the work item that the worker is currently
			 * executing.
			 *
			 * The kernel workqueue code guarantees non-reentrancy
			 * of work items. I.e., if a work item with the same
			 * address and work function is queued twice, the second
			 * execution is blocked until the first one finishes. A
			 * work item may be freed and recycled with the same
			 * work function; the workqueue code assumes that the
			 * original work item cannot depend on the recycled work
			 * item in that case (see find_worker_executing_work()).
			 *
			 * Note that different types of Btrfs work can depend on
			 * each other, and one type of work on one Btrfs
			 * filesystem may even depend on the same type of work
			 * on another Btrfs filesystem via, e.g., a loop device.
			 * Therefore, we must not allow the current work item to
			 * be recycled until we are really done, otherwise we
			 * break the above assumption and can deadlock.
			 */
			free_self = true;
		} else {
			/*
			 * We don't want to call the ordered free functions with
			 * the lock held.
			 */
			work->ordered_free(work);
			/* NB: work must not be dereferenced past this point. */
			trace_btrfs_all_work_done(wq->fs_info, work);
		}
	}
	spin_unlock_irqrestore(lock, flags);

	if (free_self) {
		self->ordered_free(self);
		/* NB: self must not be dereferenced past this point. */
		trace_btrfs_all_work_done(wq->fs_info, self);
	}
}

static void btrfs_work_helper(struct work_struct *normal_work)
{
	struct btrfs_work *work = container_of(normal_work, struct btrfs_work,
					       normal_work);
	struct __btrfs_workqueue *wq;
	int need_order = 0;

	/*
	 * We should not touch things inside work in the following cases:
	 * 1) after work->func() if it has no ordered_free
	 *    Since the struct is freed in work->func().
	 * 2) after setting WORK_DONE_BIT
	 *    The work may be freed in other threads almost instantly.
	 * So we save the needed things here.
	 */
	if (work->ordered_func)
		need_order = 1;
	wq = work->wq;

	trace_btrfs_work_sched(work);
	thresh_exec_hook(wq);
	work->func(work);
	if (need_order) {
		/*
		 * Ensures all memory accesses done in the work function are
		 * ordered before setting the WORK_DONE_BIT. Ensuring the thread
		 * which is going to executed the ordered work sees them.
		 * Pairs with the smp_rmb in run_ordered_work.
		 */
		smp_mb__before_atomic();
		set_bit(WORK_DONE_BIT, &work->flags);
		run_ordered_work(wq, work);
	} else {
		/* NB: work must not be dereferenced past this point. */
		trace_btrfs_all_work_done(wq->fs_info, work);
	}
}

void btrfs_init_work(struct btrfs_work *work, btrfs_func_t func,
		     btrfs_func_t ordered_func, btrfs_func_t ordered_free)
{
	work->func = func;
	work->ordered_func = ordered_func;
	work->ordered_free = ordered_free;
	INIT_WORK(&work->normal_work, btrfs_work_helper);
	INIT_LIST_HEAD(&work->ordered_list);
	work->flags = 0;
}

static inline void __btrfs_queue_work(struct __btrfs_workqueue *wq,
				      struct btrfs_work *work)
{
	unsigned long flags;

	work->wq = wq;
	thresh_queue_hook(wq);
	if (work->ordered_func) {
		spin_lock_irqsave(&wq->list_lock, flags);
		list_add_tail(&work->ordered_list, &wq->ordered_list);
		spin_unlock_irqrestore(&wq->list_lock, flags);
	}
	trace_btrfs_work_queued(work);
	queue_work(wq->normal_wq, &work->normal_work);
}

void btrfs_queue_work(struct btrfs_workqueue *wq,
		      struct btrfs_work *work)
{
	struct __btrfs_workqueue *dest_wq;

	if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags) && wq->high)
		dest_wq = wq->high;
	else
		dest_wq = wq->normal;
	__btrfs_queue_work(dest_wq, work);
}

static inline void
__btrfs_destroy_workqueue(struct __btrfs_workqueue *wq)
{
	destroy_workqueue(wq->normal_wq);
	trace_btrfs_workqueue_destroy(wq);
	kfree(wq);
}

void btrfs_destroy_workqueue(struct btrfs_workqueue *wq)
{
	if (!wq)
		return;
	if (wq->high)
		__btrfs_destroy_workqueue(wq->high);
	__btrfs_destroy_workqueue(wq->normal);
	kfree(wq);
}

void btrfs_workqueue_set_max(struct btrfs_workqueue *wq, int limit_active)
{
	if (!wq)
		return;
	wq->normal->limit_active = limit_active;
	if (wq->high)
		wq->high->limit_active = limit_active;
}

void btrfs_set_work_high_priority(struct btrfs_work *work)
{
	set_bit(WORK_HIGH_PRIO_BIT, &work->flags);
}

void btrfs_flush_workqueue(struct btrfs_workqueue *wq)
{
	if (wq->high)
		flush_workqueue(wq->high->normal_wq);

	flush_workqueue(wq->normal->normal_wq);
}
Commit	Line	Data
c1d7c514	1	// SPDX-License-Identifier: GPL-2.0
8b712842 CM	2	/*
8b712842 CM	3	* Copyright (C) 2007 Oracle. All rights reserved.
08a9ff32	4	* Copyright (C) 2014 Fujitsu. All rights reserved.
8b712842 CM	5	*/
	6
	7	#include <linux/kthread.h>
5a0e3ad6	8	#include <linux/slab.h>
8b712842 CM	9	#include <linux/list.h>
8b712842 CM	10	#include <linux/spinlock.h>
b51912c9	11	#include <linux/freezer.h>
8b712842	12	#include "async-thread.h"
52483bc2	13	#include "ctree.h"
8b712842	14
f64ce7b8 DS	15	enum {
	16	WORK_DONE_BIT,
	17	WORK_ORDER_DONE_BIT,
	18	WORK_HIGH_PRIO_BIT,
	19	};
4a69a410	20
0bd9289c QW	21	#define NO_THRESHOLD (-1)
	22	#define DFT_THRESHOLD (32)
	23
d458b054	24	struct __btrfs_workqueue {
08a9ff32	25	struct workqueue_struct *normal_wq;
cb001095 JM	26
	27	/* File system this workqueue services */
	28	struct btrfs_fs_info *fs_info;
	29
08a9ff32 QW	30	/* List head pointing to ordered work list */
	31	struct list_head ordered_list;
	32
	33	/* Spinlock for ordered_list */
	34	spinlock_t list_lock;
0bd9289c QW	35
	36	/* Thresholding related variants */
	37	atomic_t pending;
c6dd6ea5 QW	38
	39	/* Up limit of concurrency workers */
	40	int limit_active;
	41
	42	/* Current number of concurrency workers */
	43	int current_active;
	44
	45	/* Threshold to change current_active */
0bd9289c QW	46	int thresh;
	47	unsigned int count;
	48	spinlock_t thres_lock;
08a9ff32 QW	49	};
08a9ff32 QW	50
d458b054 QW	51	struct btrfs_workqueue {
	52	struct __btrfs_workqueue *normal;
	53	struct __btrfs_workqueue *high;
1ca08976 QW	54	};
1ca08976 QW	55
e1f60a65	56	struct btrfs_fs_info * __pure btrfs_workqueue_owner(const struct __btrfs_workqueue *wq)
cb001095 JM	57	{
	58	return wq->fs_info;
	59	}
	60
e1f60a65	61	struct btrfs_fs_info * __pure btrfs_work_owner(const struct btrfs_work *work)
cb001095 JM	62	{
	63	return work->wq->fs_info;
	64	}
	65
9a35b637	66	bool btrfs_workqueue_normal_congested(const struct btrfs_workqueue *wq)
2939e1a8 MP	67	{
	68	/*
	69	* We could compare wq->normal->pending with num_online_cpus()
	70	* to support "thresh == NO_THRESHOLD" case, but it requires
	71	* moving up atomic_inc/dec in thresh_queue/exec_hook. Let's
	72	* postpone it until someone needs the support of that case.
	73	*/
	74	if (wq->normal->thresh == NO_THRESHOLD)
	75	return false;
	76
	77	return atomic_read(&wq->normal->pending) > wq->normal->thresh * 2;
	78	}
	79
9e0af237	80	static struct __btrfs_workqueue *
cb001095 JM	81	__btrfs_alloc_workqueue(struct btrfs_fs_info fs_info, const char name,
cb001095 JM	82	unsigned int flags, int limit_active, int thresh)
1ca08976	83	{
61dd5ae6	84	struct __btrfs_workqueue ret = kzalloc(sizeof(ret), GFP_KERNEL);
1ca08976	85
5d99a998	86	if (!ret)
1ca08976 QW	87	return NULL;
1ca08976 QW	88
cb001095	89	ret->fs_info = fs_info;
c6dd6ea5	90	ret->limit_active = limit_active;
0bd9289c QW	91	atomic_set(&ret->pending, 0);
	92	if (thresh == 0)
	93	thresh = DFT_THRESHOLD;
	94	/* For low threshold, disabling threshold is a better choice */
	95	if (thresh < DFT_THRESHOLD) {
c6dd6ea5	96	ret->current_active = limit_active;
0bd9289c QW	97	ret->thresh = NO_THRESHOLD;
0bd9289c QW	98	} else {
c6dd6ea5 QW	99	/*
	100	* For threshold-able wq, let its concurrency grow on demand.
	101	* Use minimal max_active at alloc time to reduce resource
	102	* usage.
	103	*/
	104	ret->current_active = 1;
0bd9289c QW	105	ret->thresh = thresh;
	106	}
	107
1ca08976	108	if (flags & WQ_HIGHPRI)
ce3ded10 DS	109	ret->normal_wq = alloc_workqueue("btrfs-%s-high", flags,
ce3ded10 DS	110	ret->current_active, name);
1ca08976	111	else
ce3ded10 DS	112	ret->normal_wq = alloc_workqueue("btrfs-%s", flags,
ce3ded10 DS	113	ret->current_active, name);
5d99a998	114	if (!ret->normal_wq) {
1ca08976 QW	115	kfree(ret);
	116	return NULL;
	117	}
	118
	119	INIT_LIST_HEAD(&ret->ordered_list);
	120	spin_lock_init(&ret->list_lock);
0bd9289c	121	spin_lock_init(&ret->thres_lock);
c3a46891	122	trace_btrfs_workqueue_alloc(ret, name, flags & WQ_HIGHPRI);
1ca08976 QW	123	return ret;
	124	}
	125
	126	static inline void
d458b054	127	__btrfs_destroy_workqueue(struct __btrfs_workqueue *wq);
1ca08976	128
cb001095 JM	129	struct btrfs_workqueue btrfs_alloc_workqueue(struct btrfs_fs_info fs_info,
cb001095 JM	130	const char *name,
6f011058	131	unsigned int flags,
c6dd6ea5	132	int limit_active,
d458b054	133	int thresh)
08a9ff32	134	{
61dd5ae6	135	struct btrfs_workqueue ret = kzalloc(sizeof(ret), GFP_KERNEL);
08a9ff32	136
5d99a998	137	if (!ret)
08a9ff32 QW	138	return NULL;
08a9ff32 QW	139
cb001095 JM	140	ret->normal = __btrfs_alloc_workqueue(fs_info, name,
cb001095 JM	141	flags & ~WQ_HIGHPRI,
c6dd6ea5	142	limit_active, thresh);
5d99a998	143	if (!ret->normal) {
08a9ff32 QW	144	kfree(ret);
	145	return NULL;
	146	}
	147
1ca08976	148	if (flags & WQ_HIGHPRI) {
cb001095 JM	149	ret->high = __btrfs_alloc_workqueue(fs_info, name, flags,
cb001095 JM	150	limit_active, thresh);
5d99a998	151	if (!ret->high) {
1ca08976 QW	152	__btrfs_destroy_workqueue(ret->normal);
	153	kfree(ret);
	154	return NULL;
	155	}
	156	}
08a9ff32 QW	157	return ret;
	158	}
	159
0bd9289c QW	160	/*
	161	* Hook for threshold which will be called in btrfs_queue_work.
	162	* This hook WILL be called in IRQ handler context,
	163	* so workqueue_set_max_active MUST NOT be called in this hook
	164	*/
d458b054	165	static inline void thresh_queue_hook(struct __btrfs_workqueue *wq)
0bd9289c QW	166	{
	167	if (wq->thresh == NO_THRESHOLD)
	168	return;
	169	atomic_inc(&wq->pending);
	170	}
	171
	172	/*
	173	* Hook for threshold which will be called before executing the work,
	174	* This hook is called in kthread content.
	175	* So workqueue_set_max_active is called here.
	176	*/
d458b054	177	static inline void thresh_exec_hook(struct __btrfs_workqueue *wq)
0bd9289c	178	{
c6dd6ea5	179	int new_current_active;
0bd9289c QW	180	long pending;
	181	int need_change = 0;
	182
	183	if (wq->thresh == NO_THRESHOLD)
	184	return;
	185
	186	atomic_dec(&wq->pending);
	187	spin_lock(&wq->thres_lock);
	188	/*
	189	* Use wq->count to limit the calling frequency of
	190	* workqueue_set_max_active.
	191	*/
	192	wq->count++;
	193	wq->count %= (wq->thresh / 4);
	194	if (!wq->count)
	195	goto out;
c6dd6ea5	196	new_current_active = wq->current_active;
0bd9289c QW	197
	198	/*
	199	* pending may be changed later, but it's OK since we really
	200	* don't need it so accurate to calculate new_max_active.
	201	*/
	202	pending = atomic_read(&wq->pending);
	203	if (pending > wq->thresh)
c6dd6ea5	204	new_current_active++;
0bd9289c	205	if (pending < wq->thresh / 2)
c6dd6ea5 QW	206	new_current_active--;
	207	new_current_active = clamp_val(new_current_active, 1, wq->limit_active);
	208	if (new_current_active != wq->current_active) {
0bd9289c	209	need_change = 1;
c6dd6ea5	210	wq->current_active = new_current_active;
0bd9289c QW	211	}
	212	out:
	213	spin_unlock(&wq->thres_lock);
	214
	215	if (need_change) {
c6dd6ea5	216	workqueue_set_max_active(wq->normal_wq, wq->current_active);
0bd9289c QW	217	}
	218	}
	219
c495dcd6 OS	220	static void run_ordered_work(struct __btrfs_workqueue *wq,
c495dcd6 OS	221	struct btrfs_work *self)
08a9ff32 QW	222	{
08a9ff32 QW	223	struct list_head *list = &wq->ordered_list;
d458b054	224	struct btrfs_work *work;
08a9ff32 QW	225	spinlock_t *lock = &wq->list_lock;
08a9ff32 QW	226	unsigned long flags;
c495dcd6	227	bool free_self = false;
08a9ff32 QW	228
	229	while (1) {
	230	spin_lock_irqsave(lock, flags);
	231	if (list_empty(list))
	232	break;
d458b054	233	work = list_entry(list->next, struct btrfs_work,
08a9ff32 QW	234	ordered_list);
	235	if (!test_bit(WORK_DONE_BIT, &work->flags))
	236	break;
45da9c17 NB	237	/*
	238	* Orders all subsequent loads after reading WORK_DONE_BIT,
	239	* paired with the smp_mb__before_atomic in btrfs_work_helper
	240	* this guarantees that the ordered function will see all
	241	* updates from ordinary work function.
	242	*/
	243	smp_rmb();
08a9ff32 QW	244
	245	/*
	246	* we are going to call the ordered done function, but
	247	* we leave the work item on the list as a barrier so
	248	* that later work items that are done don't have their
	249	* functions called before this one returns
	250	*/
	251	if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags))
	252	break;
52483bc2	253	trace_btrfs_ordered_sched(work);
08a9ff32 QW	254	spin_unlock_irqrestore(lock, flags);
	255	work->ordered_func(work);
	256
	257	/* now take the lock again and drop our item from the list */
	258	spin_lock_irqsave(lock, flags);
	259	list_del(&work->ordered_list);
	260	spin_unlock_irqrestore(lock, flags);
	261
c495dcd6 OS	262	if (work == self) {
	263	/*
	264	* This is the work item that the worker is currently
	265	* executing.
	266	*
	267	* The kernel workqueue code guarantees non-reentrancy
	268	* of work items. I.e., if a work item with the same
	269	* address and work function is queued twice, the second
	270	* execution is blocked until the first one finishes. A
	271	* work item may be freed and recycled with the same
	272	* work function; the workqueue code assumes that the
	273	* original work item cannot depend on the recycled work
	274	* item in that case (see find_worker_executing_work()).
	275	*
a0cac0ec OS	276	* Note that different types of Btrfs work can depend on
	277	* each other, and one type of work on one Btrfs
	278	* filesystem may even depend on the same type of work
	279	* on another Btrfs filesystem via, e.g., a loop device.
	280	* Therefore, we must not allow the current work item to
	281	* be recycled until we are really done, otherwise we
	282	* break the above assumption and can deadlock.
c495dcd6 OS	283	*/
	284	free_self = true;
	285	} else {
	286	/*
	287	* We don't want to call the ordered free functions with
c9eb55db	288	* the lock held.
c495dcd6	289	*/
c495dcd6	290	work->ordered_free(work);
c9eb55db OS	291	/* NB: work must not be dereferenced past this point. */
c9eb55db OS	292	trace_btrfs_all_work_done(wq->fs_info, work);
c495dcd6	293	}
08a9ff32 QW	294	}
08a9ff32 QW	295	spin_unlock_irqrestore(lock, flags);
c495dcd6 OS	296
c495dcd6 OS	297	if (free_self) {
c495dcd6	298	self->ordered_free(self);
c9eb55db OS	299	/* NB: self must not be dereferenced past this point. */
c9eb55db OS	300	trace_btrfs_all_work_done(wq->fs_info, self);
c495dcd6	301	}
08a9ff32 QW	302	}
08a9ff32 QW	303
a0cac0ec	304	static void btrfs_work_helper(struct work_struct *normal_work)
08a9ff32	305	{
a0cac0ec OS	306	struct btrfs_work *work = container_of(normal_work, struct btrfs_work,
a0cac0ec OS	307	normal_work);
d458b054	308	struct __btrfs_workqueue *wq;
08a9ff32 QW	309	int need_order = 0;
08a9ff32 QW	310
08a9ff32 QW	311	/*
	312	* We should not touch things inside work in the following cases:
	313	* 1) after work->func() if it has no ordered_free
	314	* Since the struct is freed in work->func().
	315	* 2) after setting WORK_DONE_BIT
	316	* The work may be freed in other threads almost instantly.
	317	* So we save the needed things here.
	318	*/
	319	if (work->ordered_func)
	320	need_order = 1;
	321	wq = work->wq;
	322
52483bc2	323	trace_btrfs_work_sched(work);
0bd9289c	324	thresh_exec_hook(wq);
08a9ff32 QW	325	work->func(work);
08a9ff32 QW	326	if (need_order) {
45da9c17 NB	327	/*
	328	* Ensures all memory accesses done in the work function are
	329	* ordered before setting the WORK_DONE_BIT. Ensuring the thread
	330	* which is going to executed the ordered work sees them.
	331	* Pairs with the smp_rmb in run_ordered_work.
	332	*/
	333	smp_mb__before_atomic();
08a9ff32	334	set_bit(WORK_DONE_BIT, &work->flags);
c495dcd6	335	run_ordered_work(wq, work);
c9eb55db OS	336	} else {
	337	/* NB: work must not be dereferenced past this point. */
	338	trace_btrfs_all_work_done(wq->fs_info, work);
08a9ff32 QW	339	}
	340	}
	341
a0cac0ec OS	342	void btrfs_init_work(struct btrfs_work *work, btrfs_func_t func,
a0cac0ec OS	343	btrfs_func_t ordered_func, btrfs_func_t ordered_free)
08a9ff32 QW	344	{
	345	work->func = func;
	346	work->ordered_func = ordered_func;
	347	work->ordered_free = ordered_free;
a0cac0ec	348	INIT_WORK(&work->normal_work, btrfs_work_helper);
08a9ff32 QW	349	INIT_LIST_HEAD(&work->ordered_list);
	350	work->flags = 0;
	351	}
	352
d458b054 QW	353	static inline void __btrfs_queue_work(struct __btrfs_workqueue *wq,
d458b054 QW	354	struct btrfs_work *work)
08a9ff32 QW	355	{
	356	unsigned long flags;
	357
	358	work->wq = wq;
0bd9289c	359	thresh_queue_hook(wq);
08a9ff32 QW	360	if (work->ordered_func) {
	361	spin_lock_irqsave(&wq->list_lock, flags);
	362	list_add_tail(&work->ordered_list, &wq->ordered_list);
	363	spin_unlock_irqrestore(&wq->list_lock, flags);
	364	}
52483bc2	365	trace_btrfs_work_queued(work);
0a95b851	366	queue_work(wq->normal_wq, &work->normal_work);
08a9ff32 QW	367	}
08a9ff32 QW	368
d458b054 QW	369	void btrfs_queue_work(struct btrfs_workqueue *wq,
d458b054 QW	370	struct btrfs_work *work)
1ca08976	371	{
d458b054	372	struct __btrfs_workqueue *dest_wq;
1ca08976 QW	373
	374	if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags) && wq->high)
	375	dest_wq = wq->high;
	376	else
	377	dest_wq = wq->normal;
	378	__btrfs_queue_work(dest_wq, work);
	379	}
	380
	381	static inline void
d458b054	382	__btrfs_destroy_workqueue(struct __btrfs_workqueue *wq)
08a9ff32 QW	383	{
08a9ff32 QW	384	destroy_workqueue(wq->normal_wq);
c3a46891	385	trace_btrfs_workqueue_destroy(wq);
08a9ff32 QW	386	kfree(wq);
	387	}
	388
d458b054	389	void btrfs_destroy_workqueue(struct btrfs_workqueue *wq)
1ca08976 QW	390	{
	391	if (!wq)
	392	return;
	393	if (wq->high)
	394	__btrfs_destroy_workqueue(wq->high);
	395	__btrfs_destroy_workqueue(wq->normal);
ef66af10	396	kfree(wq);
1ca08976 QW	397	}
1ca08976 QW	398
c6dd6ea5	399	void btrfs_workqueue_set_max(struct btrfs_workqueue *wq, int limit_active)
08a9ff32	400	{
800ee224 ST	401	if (!wq)
800ee224 ST	402	return;
c6dd6ea5	403	wq->normal->limit_active = limit_active;
1ca08976	404	if (wq->high)
c6dd6ea5	405	wq->high->limit_active = limit_active;
1ca08976 QW	406	}
1ca08976 QW	407
d458b054	408	void btrfs_set_work_high_priority(struct btrfs_work *work)
1ca08976 QW	409	{
1ca08976 QW	410	set_bit(WORK_HIGH_PRIO_BIT, &work->flags);
08a9ff32	411	}
f0cc2cd7 FM	412
	413	void btrfs_flush_workqueue(struct btrfs_workqueue *wq)
	414	{
	415	if (wq->high)
	416	flush_workqueue(wq->high->normal_wq);
	417
	418	flush_workqueue(wq->normal->normal_wq);
	419	}