cgroup.c

 * css_next_descendant_post - find the next descendant for post-order walk
 * @pos: the current position (%NULL to initiate traversal)
 * @root: css whose descendants to walk
 *
 * To be used by css_for_each_descendant_post().  Find the next descendant
 * to visit for post-order traversal of @root's descendants.  @root is
 * included in the iteration and the last node to be visited.
 *
 * While this function requires RCU read locking, it doesn't require the
 * whole traversal to be contained in a single RCU critical section.  This
 * function will return the correct next descendant as long as both @pos
 * and @cgroup are accessible and @pos is a descendant of @cgroup.
 */
struct cgroup_subsys_state *
css_next_descendant_post(struct cgroup_subsys_state *pos,
			 struct cgroup_subsys_state *root)
{
	struct cgroup_subsys_state *next;

	WARN_ON_ONCE(!rcu_read_lock_held());

	/* if first iteration, visit leftmost descendant which may be @root */
	if (!pos)
		return css_leftmost_descendant(root);

	/* if we visited @root, we're done */
	if (pos == root)
		return NULL;

	/* if there's an unvisited sibling, visit its leftmost descendant */
	next = css_next_child(pos, css_parent(pos));
	if (next)
		return css_leftmost_descendant(next);

	/* no sibling left, visit parent */
	return css_parent(pos);
}
EXPORT_SYMBOL_GPL(css_next_descendant_post);

/**
 * css_advance_task_iter - advance a task itererator to the next css_set
 * @it: the iterator to advance
 *
 * Advance @it to the next css_set to walk.
 */
static void css_advance_task_iter(struct css_task_iter *it)
{
	struct list_head *l = it->cset_link;
	struct cgrp_cset_link *link;
	struct css_set *cset;

	/* Advance to the next non-empty css_set */
	do {
		l = l->next;
		if (l == &it->origin_css->cgroup->cset_links) {
			it->cset_link = NULL;
			return;
		}
		link = list_entry(l, struct cgrp_cset_link, cset_link);
		cset = link->cset;
	} while (list_empty(&cset->tasks));
	it->cset_link = l;
	it->task = cset->tasks.next;
}

/**
 * css_task_iter_start - initiate task iteration
 * @css: the css to walk tasks of
 * @it: the task iterator to use
 *
 * Initiate iteration through the tasks of @css.  The caller can call
 * css_task_iter_next() to walk through the tasks until the function
 * returns NULL.  On completion of iteration, css_task_iter_end() must be
 * called.
 *
 * Note that this function acquires a lock which is released when the
 * iteration finishes.  The caller can't sleep while iteration is in
 * progress.
 */
void css_task_iter_start(struct cgroup_subsys_state *css,
			 struct css_task_iter *it)
	__acquires(css_set_lock)
{
	/*
	 * The first time anyone tries to iterate across a css, we need to
	 * enable the list linking each css_set to its tasks, and fix up
	 * all existing tasks.
	 */
	if (!use_task_css_set_links)
		cgroup_enable_task_cg_lists();

	read_lock(&css_set_lock);

	it->origin_css = css;
	it->cset_link = &css->cgroup->cset_links;

	css_advance_task_iter(it);
}

/**
 * css_task_iter_next - return the next task for the iterator
 * @it: the task iterator being iterated
 *
 * The "next" function for task iteration.  @it should have been
 * initialized via css_task_iter_start().  Returns NULL when the iteration
 * reaches the end.
 */
struct task_struct *css_task_iter_next(struct css_task_iter *it)
{
	struct task_struct *res;
	struct list_head *l = it->task;
	struct cgrp_cset_link *link;

	/* If the iterator cg is NULL, we have no tasks */
	if (!it->cset_link)
		return NULL;
	res = list_entry(l, struct task_struct, cg_list);
	/* Advance iterator to find next entry */
	l = l->next;
	link = list_entry(it->cset_link, struct cgrp_cset_link, cset_link);
	if (l == &link->cset->tasks) {
		/*
		 * We reached the end of this task list - move on to the
		 * next cgrp_cset_link.
		 */
		css_advance_task_iter(it);
	} else {
		it->task = l;
	}
	return res;
}

/**
 * css_task_iter_end - finish task iteration
 * @it: the task iterator to finish
 *
 * Finish task iteration started by css_task_iter_start().
 */
void css_task_iter_end(struct css_task_iter *it)
	__releases(css_set_lock)
{
	read_unlock(&css_set_lock);
}

static inline int started_after_time(struct task_struct *t1,
				     struct timespec *time,
				     struct task_struct *t2)
{
	int start_diff = timespec_compare(&t1->start_time, time);
	if (start_diff > 0) {
		return 1;
	} else if (start_diff < 0) {
		return 0;
	} else {
		/*
		 * Arbitrarily, if two processes started at the same
		 * time, we'll say that the lower pointer value
		 * started first. Note that t2 may have exited by now
		 * so this may not be a valid pointer any longer, but
		 * that's fine - it still serves to distinguish
		 * between two tasks started (effectively) simultaneously.
		 */
		return t1 > t2;
	}
}

/*
 * This function is a callback from heap_insert() and is used to order
 * the heap.
 * In this case we order the heap in descending task start time.
 */
static inline int started_after(void *p1, void *p2)
{
	struct task_struct *t1 = p1;
	struct task_struct *t2 = p2;
	return started_after_time(t1, &t2->start_time, t2);
}

/**
 * css_scan_tasks - iterate though all the tasks in a css
 * @css: the css to iterate tasks of
 * @test: optional test callback
 * @process: process callback
 * @data: data passed to @test and @process
 * @heap: optional pre-allocated heap used for task iteration
 *
 * Iterate through all the tasks in @css, calling @test for each, and if it
 * returns %true, call @process for it also.
 *
 * @test may be NULL, meaning always true (select all tasks), which
 * effectively duplicates css_task_iter_{start,next,end}() but does not
 * lock css_set_lock for the call to @process.
 *
 * It is guaranteed that @process will act on every task that is a member
 * of @css for the duration of this call.  This function may or may not
 * call @process for tasks that exit or move to a different css during the
 * call, or are forked or move into the css during the call.
 *
 * Note that @test may be called with locks held, and may in some
 * situations be called multiple times for the same task, so it should be
 * cheap.
 *
 * If @heap is non-NULL, a heap has been pre-allocated and will be used for
 * heap operations (and its "gt" member will be overwritten), else a
 * temporary heap will be used (allocation of which may cause this function
 * to fail).
 */
int css_scan_tasks(struct cgroup_subsys_state *css,
		   bool (*test)(struct task_struct *, void *),
		   void (*process)(struct task_struct *, void *),
		   void *data, struct ptr_heap *heap)
{
	int retval, i;
	struct css_task_iter it;
	struct task_struct *p, *dropped;
	/* Never dereference latest_task, since it's not refcounted */
	struct task_struct *latest_task = NULL;
	struct ptr_heap tmp_heap;
	struct timespec latest_time = { 0, 0 };

	if (heap) {
		/* The caller supplied our heap and pre-allocated its memory */
		heap->gt = &started_after;
	} else {
		/* We need to allocate our own heap memory */
		heap = &tmp_heap;
		retval = heap_init(heap, PAGE_SIZE, GFP_KERNEL, &started_after);
		if (retval)
			/* cannot allocate the heap */
			return retval;
	}

 again:
	/*
	 * Scan tasks in the css, using the @test callback to determine
	 * which are of interest, and invoking @process callback on the
	 * ones which need an update.  Since we don't want to hold any
	 * locks during the task updates, gather tasks to be processed in a
	 * heap structure.  The heap is sorted by descending task start
	 * time.  If the statically-sized heap fills up, we overflow tasks
	 * that started later, and in future iterations only consider tasks
	 * that started after the latest task in the previous pass. This
	 * guarantees forward progress and that we don't miss any tasks.
	 */
	heap->size = 0;
	css_task_iter_start(css, &it);
	while ((p = css_task_iter_next(&it))) {
		/*
		 * Only affect tasks that qualify per the caller's callback,
		 * if he provided one
		 */
		if (test && !test(p, data))
			continue;
		/*
		 * Only process tasks that started after the last task
		 * we processed
		 */
		if (!started_after_time(p, &latest_time, latest_task))
			continue;
		dropped = heap_insert(heap, p);
		if (dropped == NULL) {
			/*
			 * The new task was inserted; the heap wasn't
			 * previously full
			 */
			get_task_struct(p);
		} else if (dropped != p) {
			/*
			 * The new task was inserted, and pushed out a
			 * different task
			 */
			get_task_struct(p);
			put_task_struct(dropped);
		}
		/*
		 * Else the new task was newer than anything already in
		 * the heap and wasn't inserted
		 */
	}
	css_task_iter_end(&it);

	if (heap->size) {
		for (i = 0; i < heap->size; i++) {
			struct task_struct *q = heap->ptrs[i];
			if (i == 0) {
				latest_time = q->start_time;
				latest_task = q;
			}
			/* Process the task per the caller's callback */
			process(q, data);
			put_task_struct(q);
		}
		/*
		 * If we had to process any tasks at all, scan again
		 * in case some of them were in the middle of forking
		 * children that didn't get processed.
		 * Not the most efficient way to do it, but it avoids
		 * having to take callback_mutex in the fork path
		 */
		goto again;
	}
	if (heap == &tmp_heap)
		heap_free(&tmp_heap);
	return 0;
}

static void cgroup_transfer_one_task(struct task_struct *task, void *data)
{
	struct cgroup *new_cgroup = data;

	mutex_lock(&cgroup_mutex);
	cgroup_attach_task(new_cgroup, task, false);
	mutex_unlock(&cgroup_mutex);
}

/**
 * cgroup_trasnsfer_tasks - move tasks from one cgroup to another
 * @to: cgroup to which the tasks will be moved
 * @from: cgroup in which the tasks currently reside
 */
int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
{
	return css_scan_tasks(&from->dummy_css, NULL, cgroup_transfer_one_task,
			      to, NULL);
}

/*
 * Stuff for reading the 'tasks'/'procs' files.
 *
 * Reading this file can return large amounts of data if a cgroup has
 * *lots* of attached tasks. So it may need several calls to read(),
 * but we cannot guarantee that the information we produce is correct
 * unless we produce it entirely atomically.
 *
 */

/* which pidlist file are we talking about? */
enum cgroup_filetype {
	CGROUP_FILE_PROCS,
	CGROUP_FILE_TASKS,
};

/*
 * A pidlist is a list of pids that virtually represents the contents of one
 * of the cgroup files ("procs" or "tasks"). We keep a list of such pidlists,
 * a pair (one each for procs, tasks) for each pid namespace that's relevant
 * to the cgroup.
 */
struct cgroup_pidlist {
	/*
	 * used to find which pidlist is wanted. doesn't change as long as
	 * this particular list stays in the list.
	*/
	struct { enum cgroup_filetype type; struct pid_namespace *ns; } key;
	/* array of xids */
	pid_t *list;
	/* how many elements the above list has */
	int length;
	/* each of these stored in a list by its cgroup */
	struct list_head links;
	/* pointer to the cgroup we belong to, for list removal purposes */
	struct cgroup *owner;
	/* for delayed destruction */
	struct delayed_work destroy_dwork;
};

/*
 * The following two functions "fix" the issue where there are more pids
 * than kmalloc will give memory for; in such cases, we use vmalloc/vfree.
 * TODO: replace with a kernel-wide solution to this problem
 */
#define PIDLIST_TOO_LARGE(c) ((c) * sizeof(pid_t) > (PAGE_SIZE * 2))
static void *pidlist_allocate(int count)
{
	if (PIDLIST_TOO_LARGE(count))
		return vmalloc(count * sizeof(pid_t));
	else
		return kmalloc(count * sizeof(pid_t), GFP_KERNEL);
}

static void pidlist_free(void *p)
{
	if (is_vmalloc_addr(p))
		vfree(p);
	else
		kfree(p);
}

/*
 * Used to destroy all pidlists lingering waiting for destroy timer.  None
 * should be left afterwards.
 */
static void cgroup_pidlist_destroy_all(struct cgroup *cgrp)
{
	struct cgroup_pidlist *l, *tmp_l;

	mutex_lock(&cgrp->pidlist_mutex);
	list_for_each_entry_safe(l, tmp_l, &cgrp->pidlists, links)
		mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork, 0);
	mutex_unlock(&cgrp->pidlist_mutex);

	flush_workqueue(cgroup_pidlist_destroy_wq);
	BUG_ON(!list_empty(&cgrp->pidlists));
}

static void cgroup_pidlist_destroy_work_fn(struct work_struct *work)
{
	struct delayed_work *dwork = to_delayed_work(work);
	struct cgroup_pidlist *l = container_of(dwork, struct cgroup_pidlist,
						destroy_dwork);
	struct cgroup_pidlist *tofree = NULL;

	mutex_lock(&l->owner->pidlist_mutex);

	/*
	 * Destroy iff we didn't get queued again.  The state won't change
	 * as destroy_dwork can only be queued while locked.
	 */
	if (!delayed_work_pending(dwork)) {
		list_del(&l->links);
		pidlist_free(l->list);
		put_pid_ns(l->key.ns);
		tofree = l;
	}

	mutex_unlock(&l->owner->pidlist_mutex);
	kfree(tofree);
}

/*
 * pidlist_uniq - given a kmalloc()ed list, strip out all duplicate entries
 * Returns the number of unique elements.
 */
static int pidlist_uniq(pid_t *list, int length)
{
	int src, dest = 1;

	/*
	 * we presume the 0th element is unique, so i starts at 1. trivial
	 * edge cases first; no work needs to be done for either
	 */
	if (length == 0 || length == 1)
		return length;
	/* src and dest walk down the list; dest counts unique elements */
	for (src = 1; src < length; src++) {
		/* find next unique element */
		while (list[src] == list[src-1]) {
			src++;
			if (src == length)
				goto after;
		}
		/* dest always points to where the next unique element goes */
		list[dest] = list[src];
		dest++;
	}
after:
	return dest;
}

/*
 * The two pid files - task and cgroup.procs - guaranteed that the result
 * is sorted, which forced this whole pidlist fiasco.  As pid order is
 * different per namespace, each namespace needs differently sorted list,
 * making it impossible to use, for example, single rbtree of member tasks
 * sorted by task pointer.  As pidlists can be fairly large, allocating one
 * per open file is dangerous, so cgroup had to implement shared pool of
 * pidlists keyed by cgroup and namespace.
 *
 * All this extra complexity was caused by the original implementation
 * committing to an entirely unnecessary property.  In the long term, we
 * want to do away with it.  Explicitly scramble sort order if
 * sane_behavior so that no such expectation exists in the new interface.
 *
 * Scrambling is done by swapping every two consecutive bits, which is
 * non-identity one-to-one mapping which disturbs sort order sufficiently.
 */
static pid_t pid_fry(pid_t pid)
{
	unsigned a = pid & 0x55555555;
	unsigned b = pid & 0xAAAAAAAA;

	return (a << 1) | (b >> 1);
}

static pid_t cgroup_pid_fry(struct cgroup *cgrp, pid_t pid)
{
	if (cgroup_sane_behavior(cgrp))
		return pid_fry(pid);
	else
		return pid;
}

static int cmppid(const void *a, const void *b)
{
	return *(pid_t *)a - *(pid_t *)b;
}

static int fried_cmppid(const void *a, const void *b)
{
	return pid_fry(*(pid_t *)a) - pid_fry(*(pid_t *)b);
}

static struct cgroup_pidlist *cgroup_pidlist_find(struct cgroup *cgrp,
						  enum cgroup_filetype type)
{
	struct cgroup_pidlist *l;
	/* don't need task_nsproxy() if we're looking at ourself */
	struct pid_namespace *ns = task_active_pid_ns(current);

	lockdep_assert_held(&cgrp->pidlist_mutex);

	list_for_each_entry(l, &cgrp->pidlists, links)
		if (l->key.type == type && l->key.ns == ns)
			return l;
	return NULL;
}

/*
 * find the appropriate pidlist for our purpose (given procs vs tasks)
 * returns with the lock on that pidlist already held, and takes care
 * of the use count, or returns NULL with no locks held if we're out of
 * memory.
 */
static struct cgroup_pidlist *cgroup_pidlist_find_create(struct cgroup *cgrp,
						enum cgroup_filetype type)
{
	struct cgroup_pidlist *l;

	lockdep_assert_held(&cgrp->pidlist_mutex);

	l = cgroup_pidlist_find(cgrp, type);
	if (l)
		return l;

	/* entry not found; create a new one */
	l = kzalloc(sizeof(struct cgroup_pidlist), GFP_KERNEL);
	if (!l)
		return l;

	INIT_DELAYED_WORK(&l->destroy_dwork, cgroup_pidlist_destroy_work_fn);
	l->key.type = type;
	/* don't need task_nsproxy() if we're looking at ourself */
	l->key.ns = get_pid_ns(task_active_pid_ns(current));
	l->owner = cgrp;
	list_add(&l->links, &cgrp->pidlists);
	return l;
}

/*
 * Load a cgroup's pidarray with either procs' tgids or tasks' pids
 */
static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
			      struct cgroup_pidlist **lp)
{
	pid_t *array;
	int length;
	int pid, n = 0; /* used for populating the array */
	struct css_task_iter it;
	struct task_struct *tsk;
	struct cgroup_pidlist *l;

	lockdep_assert_held(&cgrp->pidlist_mutex);

	/*
	 * If cgroup gets more users after we read count, we won't have
	 * enough space - tough.  This race is indistinguishable to the
	 * caller from the case that the additional cgroup users didn't
	 * show up until sometime later on.
	 */
	length = cgroup_task_count(cgrp);
	array = pidlist_allocate(length);
	if (!array)
		return -ENOMEM;
	/* now, populate the array */
	css_task_iter_start(&cgrp->dummy_css, &it);
	while ((tsk = css_task_iter_next(&it))) {
		if (unlikely(n == length))
			break;
		/* get tgid or pid for procs or tasks file respectively */
		if (type == CGROUP_FILE_PROCS)
			pid = task_tgid_vnr(tsk);
		else
			pid = task_pid_vnr(tsk);
		if (pid > 0) /* make sure to only use valid results */
			array[n++] = pid;
	}
	css_task_iter_end(&it);
	length = n;
	/* now sort & (if procs) strip out duplicates */
	if (cgroup_sane_behavior(cgrp))
		sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
	else
		sort(array, length, sizeof(pid_t), cmppid, NULL);
	if (type == CGROUP_FILE_PROCS)
		length = pidlist_uniq(array, length);

	l = cgroup_pidlist_find_create(cgrp, type);
	if (!l) {
		mutex_unlock(&cgrp->pidlist_mutex);
		pidlist_free(array);
		return -ENOMEM;
	}

	/* store array, freeing old if necessary */
	pidlist_free(l->list);
	l->list = array;
	l->length = length;
	*lp = l;
	return 0;
}

/**
 * cgroupstats_build - build and fill cgroupstats
 * @stats: cgroupstats to fill information into
 * @dentry: A dentry entry belonging to the cgroup for which stats have
 * been requested.
 *
 * Build and fill cgroupstats so that taskstats can export it to user
 * space.
 */
int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry)
{
	int ret = -EINVAL;
	struct cgroup *cgrp;
	struct css_task_iter it;
	struct task_struct *tsk;

	/*
	 * Validate dentry by checking the superblock operations,
	 * and make sure it's a directory.
	 */
	if (dentry->d_sb->s_op != &cgroup_ops ||
	    !S_ISDIR(dentry->d_inode->i_mode))
		 goto err;

	ret = 0;
	cgrp = dentry->d_fsdata;

	css_task_iter_start(&cgrp->dummy_css, &it);
	while ((tsk = css_task_iter_next(&it))) {
		switch (tsk->state) {
		case TASK_RUNNING:
			stats->nr_running++;
			break;
		case TASK_INTERRUPTIBLE:
			stats->nr_sleeping++;
			break;
		case TASK_UNINTERRUPTIBLE:
			stats->nr_uninterruptible++;
			break;
		case TASK_STOPPED:
			stats->nr_stopped++;
			break;
		default:
			if (delayacct_is_task_waiting_on_io(tsk))
				stats->nr_io_wait++;
			break;
		}
	}
	css_task_iter_end(&it);

err:
	return ret;
}


/*
 * seq_file methods for the tasks/procs files. The seq_file position is the
 * next pid to display; the seq_file iterator is a pointer to the pid
 * in the cgroup->l->list array.
 */

static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
{
	/*
	 * Initially we receive a position value that corresponds to
	 * one more than the last pid shown (or 0 on the first call or
	 * after a seek to the start). Use a binary-search to find the
	 * next pid to display, if any
	 */
	struct cgroup_open_file *of = s->private;
	struct cgroup *cgrp = seq_css(s)->cgroup;
	struct cgroup_pidlist *l;
	enum cgroup_filetype type = seq_cft(s)->private;
	int index = 0, pid = *pos;
	int *iter, ret;

	mutex_lock(&cgrp->pidlist_mutex);

	/*
	 * !NULL @of->priv indicates that this isn't the first start()
	 * after open.  If the matching pidlist is around, we can use that.
	 * Look for it.  Note that @of->priv can't be used directly.  It
	 * could already have been destroyed.
	 */
	if (of->priv)
		of->priv = cgroup_pidlist_find(cgrp, type);

	/*
	 * Either this is the first start() after open or the matching
	 * pidlist has been destroyed inbetween.  Create a new one.
	 */
	if (!of->priv) {
		ret = pidlist_array_load(cgrp, type,
					 (struct cgroup_pidlist **)&of->priv);
		if (ret)
			return ERR_PTR(ret);
	}
	l = of->priv;

	if (pid) {
		int end = l->length;

		while (index < end) {
			int mid = (index + end) / 2;
			if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
				index = mid;
				break;
			} else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
				index = mid + 1;
			else
				end = mid;
		}
	}
	/* If we're off the end of the array, we're done */
	if (index >= l->length)
		return NULL;
	/* Update the abstract position to be the actual pid that we found */
	iter = l->list + index;
	*pos = cgroup_pid_fry(cgrp, *iter);
	return iter;
}

static void cgroup_pidlist_stop(struct seq_file *s, void *v)
{
	struct cgroup_open_file *of = s->private;
	struct cgroup_pidlist *l = of->priv;

	if (l)
		mod_delayed_work(cgroup_pidlist_destroy_wq, &l->destroy_dwork,
				 CGROUP_PIDLIST_DESTROY_DELAY);
	mutex_unlock(&seq_css(s)->cgroup->pidlist_mutex);
}

static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
{
	struct cgroup_open_file *of = s->private;
	struct cgroup_pidlist *l = of->priv;
	pid_t *p = v;
	pid_t *end = l->list + l->length;
	/*
	 * Advance to the next pid in the array. If this goes off the
	 * end, we're done
	 */
	p++;
	if (p >= end) {
		return NULL;
	} else {
		*pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
		return p;
	}
}

static int cgroup_pidlist_show(struct seq_file *s, void *v)
{
	return seq_printf(s, "%d\n", *(int *)v);
}

/*
 * seq_operations functions for iterating on pidlists through seq_file -
 * independent of whether it's tasks or procs
 */
static const struct seq_operations cgroup_pidlist_seq_operations = {
	.start = cgroup_pidlist_start,
	.stop = cgroup_pidlist_stop,
	.next = cgroup_pidlist_next,
	.show = cgroup_pidlist_show,
};

static const struct file_operations cgroup_pidlist_operations = {
	.read = seq_read,
	.llseek = seq_lseek,
	.write = cgroup_file_write,
	.release = seq_release_private,
};

/*
 * The following functions handle opens on a file that displays a pidlist
 * (tasks or procs). Prepare an array of the process/thread IDs of whoever's
 * in the cgroup.
 */
/* helper function for the two below it */
static int cgroup_pidlist_open(struct inode *unused, struct file *file)
{
	struct cfent *cfe = __d_cfe(file->f_dentry);
	struct cgroup_open_file *of;

	/* configure file information */
	file->f_op = &cgroup_pidlist_operations;

	of = __seq_open_private(file, &cgroup_pidlist_seq_operations,
				sizeof(*of));
	if (!of)
		return -ENOMEM;

	of->cfe = cfe;
	return 0;
}

static u64 cgroup_read_notify_on_release(struct cgroup_subsys_state *css,
					 struct cftype *cft)
{
	return notify_on_release(css->cgroup);
}

static int cgroup_write_notify_on_release(struct cgroup_subsys_state *css,
					  struct cftype *cft, u64 val)
{
	clear_bit(CGRP_RELEASABLE, &css->cgroup->flags);
	if (val)
		set_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
	else
		clear_bit(CGRP_NOTIFY_ON_RELEASE, &css->cgroup->flags);
	return 0;
}

/*
 * When dput() is called asynchronously, if umount has been done and
 * then deactivate_super() in cgroup_free_fn() kills the superblock,
 * there's a small window that vfs will see the root dentry with non-zero
 * refcnt and trigger BUG().
 *
 * That's why we hold a reference before dput() and drop it right after.
 */
static void cgroup_dput(struct cgroup *cgrp)
{
	struct super_block *sb = cgrp->root->sb;

	atomic_inc(&sb->s_active);
	dput(cgrp->dentry);
	deactivate_super(sb);
}

static u64 cgroup_clone_children_read(struct cgroup_subsys_state *css,
				      struct cftype *cft)
{
	return test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
}

static int cgroup_clone_children_write(struct cgroup_subsys_state *css,
				       struct cftype *cft, u64 val)
{
	if (val)
		set_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
	else
		clear_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags);
	return 0;
}

static struct cftype cgroup_base_files[] = {
	{
		.name = "cgroup.procs",
		.open = cgroup_pidlist_open,
		.private = CGROUP_FILE_PROCS,
		.write_u64 = cgroup_procs_write,
		.mode = S_IRUGO | S_IWUSR,
	},
	{
		.name = "cgroup.clone_children",
		.flags = CFTYPE_INSANE,
		.read_u64 = cgroup_clone_children_read,
		.write_u64 = cgroup_clone_children_write,
	},
	{
		.name = "cgroup.sane_behavior",
		.flags = CFTYPE_ONLY_ON_ROOT,
		.read_seq_string = cgroup_sane_behavior_show,
	},

	/*
	 * Historical crazy stuff.  These don't have "cgroup."  prefix and
	 * don't exist if sane_behavior.  If you're depending on these, be
	 * prepared to be burned.
	 */
	{
		.name = "tasks",
		.flags = CFTYPE_INSANE,		/* use "procs" instead */
		.open = cgroup_pidlist_open,
		.private = CGROUP_FILE_TASKS,
		.write_u64 = cgroup_tasks_write,
		.mode = S_IRUGO | S_IWUSR,
	},
	{
		.name = "notify_on_release",
		.flags = CFTYPE_INSANE,
		.read_u64 = cgroup_read_notify_on_release,
		.write_u64 = cgroup_write_notify_on_release,
	},
	{
		.name = "release_agent",
		.flags = CFTYPE_INSANE | CFTYPE_ONLY_ON_ROOT,
		.read_seq_string = cgroup_release_agent_show,
		.write_string = cgroup_release_agent_write,
		.max_write_len = PATH_MAX,
	},
	{ }	/* terminate */
};

/**
 * cgroup_populate_dir - create subsys files in a cgroup directory
 * @cgrp: target cgroup
 * @subsys_mask: mask of the subsystem ids whose files should be added
 *
 * On failure, no file is added.
 */
static int cgroup_populate_dir(struct cgroup *cgrp, unsigned long subsys_mask)
{
	struct cgroup_subsys *ss;
	int i, ret = 0;

	/* process cftsets of each subsystem */
	for_each_subsys(ss, i) {
		struct cftype_set *set;

		if (!test_bit(i, &subsys_mask))
			continue;

		list_for_each_entry(set, &ss->cftsets, node) {
			ret = cgroup_addrm_files(cgrp, set->cfts, true);
			if (ret < 0)
				goto err;
		}
	}
	return 0;
err:
	cgroup_clear_dir(cgrp, subsys_mask);
	return ret;
}

/*
 * css destruction is four-stage process.
 *
 * 1. Destruction starts.  Killing of the percpu_ref is initiated.
 *    Implemented in kill_css().
 *
 * 2. When the percpu_ref is confirmed to be visible as killed on all CPUs
 *    and thus css_tryget() is guaranteed to fail, the css can be offlined
 *    by invoking offline_css().  After offlining, the base ref is put.
 *    Implemented in css_killed_work_fn().
 *
 * 3. When the percpu_ref reaches zero, the only possible remaining
 *    accessors are inside RCU read sections.  css_release() schedules the
 *    RCU callback.
 *
 * 4. After the grace period, the css can be freed.  Implemented in
 *    css_free_work_fn().
 *
 * It is actually hairier because both step 2 and 4 require process context
 * and thus involve punting to css->destroy_work adding two additional
 * steps to the already complex sequence.
 */
static void css_free_work_fn(struct work_struct *work)
{
	struct cgroup_subsys_state *css =
		container_of(work, struct cgroup_subsys_state, destroy_work);
	struct cgroup *cgrp = css->cgroup;

	if (css->parent)
		css_put(css->parent);

	css->ss->css_free(css);
	cgroup_dput(cgrp);
}

static void css_free_rcu_fn(struct rcu_head *rcu_head)
{
	struct cgroup_subsys_state *css =
		container_of(rcu_head, struct cgroup_subsys_state, rcu_head);

	/*
	 * css holds an extra ref to @cgrp->dentry which is put on the last
	 * css_put().  dput() requires process context which we don't have.
	 */
	INIT_WORK(&css->destroy_work, css_free_work_fn);
	queue_work(cgroup_destroy_wq, &css->destroy_work);
}

static void css_release(struct percpu_ref *ref)
{
	struct cgroup_subsys_state *css =
		container_of(ref, struct cgroup_subsys_state, refcnt);

	call_rcu(&css->rcu_head, css_free_rcu_fn);
}

static void init_css(struct cgroup_subsys_state *css, struct cgroup_subsys *ss,
		     struct cgroup *cgrp)
{
	css->cgroup = cgrp;
	css->ss = ss;