kernfs, cgroup: reimplement "cgroup.procs" reading for v2

index 78219d5..d0b6fb1 100644
@@ -516,7 +516,7 @@ static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
 		goto out_put;
 
 	rc = 0;
-	of->mmapped = 1;
+	of->mmapped = true;
 	of->vm_ops = vma->vm_ops;
 	vma->vm_ops = &kernfs_vm_ops;
 out_put:

index 7056238..afd4e5a 100644
@@ -185,7 +185,7 @@ struct kernfs_open_file {
 	char			*prealloc_buf;
 
 	size_t			atomic_write_len;
-	bool			mmapped;
+	bool			mmapped:1;
 	const struct vm_operations_struct *vm_ops;
 };
 

index cf4c636..0b9d23b 100644
@@ -468,7 +468,7 @@ static void kernfs_drain(struct kernfs_node *kn)
 		rwsem_release(&kn->dep_map, 1, _RET_IP_);
 	}
 
-	kernfs_unmap_bin_file(kn);
+	kernfs_drain_open_files(kn);
 
 	mutex_lock(&kernfs_mutex);
 }

104 lines not on this page · raw message

index bfd551b..3100987 100644
@@ -104,7 +104,7 @@ struct kernfs_node *kernfs_new_node(struct kernfs_node *parent,
  */
 extern const struct file_operations kernfs_file_fops;
 
-void kernfs_unmap_bin_file(struct kernfs_node *kn);
+void kernfs_drain_open_files(struct kernfs_node *kn);
 
 /*
  * symlink.c

index afd4e5a..a9b11b8 100644
@@ -46,6 +46,7 @@ enum kernfs_node_flag {
 	KERNFS_SUICIDAL		= 0x0400,
 	KERNFS_SUICIDED		= 0x0800,
 	KERNFS_EMPTY_DIR	= 0x1000,
+	KERNFS_HAS_RELEASE	= 0x2000,
 };
 
 /* @flags for kernfs_create_root() */

@@ -175,6 +176,7 @@ struct kernfs_open_file {
 	/* published fields */
 	struct kernfs_node	*kn;
 	struct file		*file;
+	struct seq_file		*seq_file;
 	void			*priv;
 
 	/* private fields, do not use outside kernfs proper */

@@ -186,11 +188,19 @@ struct kernfs_open_file {
 
 	size_t			atomic_write_len;
 	bool			mmapped:1;
+	bool			released:1;
 	const struct vm_operations_struct *vm_ops;
 };
 
 struct kernfs_ops {
 	/*
+	 * Optional open/release methods.  Both are called with
+	 * @of->seq_file populated.
+	 */
+	int (*open)(struct kernfs_open_file *of);
+	void (*release)(struct kernfs_open_file *of);
+
+	/*
 	 * Read is handled by either seq_file or raw_read().
 	 *
 	 * If seq_show() is present, seq_file path is active.  Other seq

index 861b467..8a916dc 100644
@@ -388,6 +388,9 @@ struct cftype {
 	struct list_head node;		/* anchored at ss->cfts */
 	struct kernfs_ops *kf_ops;
 
+	int (*open)(struct kernfs_open_file *of);
+	void (*release)(struct kernfs_open_file *of);
+
 	/*
 	 * read_u64() is a shortcut for the common case of returning a
 	 * single integer. Use it in place of read()

index 2ee9ec3..87167e4 100644
@@ -3503,6 +3503,23 @@ static int cgroup_events_show(struct seq_file *seq, void *v)
 	return 0;
 }
 
+static int cgroup_file_open(struct kernfs_open_file *of)
+{
+	struct cftype *cft = of->kn->priv;
+
+	if (cft->open)
+		return cft->open(of);
+	return 0;
+}
+
+static void cgroup_file_release(struct kernfs_open_file *of)
+{
+	struct cftype *cft = of->kn->priv;
+
+	if (cft->release)
+		cft->release(of);
+}
+
 static ssize_t cgroup_file_write(struct kernfs_open_file *of, char *buf,
 				 size_t nbytes, loff_t off)
 {

@@ -3553,7 +3570,8 @@ static void *cgroup_seqfile_next(struct seq_file *seq, void *v, loff_t *ppos)
 
 static void cgroup_seqfile_stop(struct seq_file *seq, void *v)
 {
-	seq_cft(seq)->seq_stop(seq, v);
+	if (seq_cft(seq)->seq_stop)
+		seq_cft(seq)->seq_stop(seq, v);
 }
 
 static int cgroup_seqfile_show(struct seq_file *m, void *arg)

@@ -3575,12 +3593,16 @@ static int cgroup_seqfile_show(struct seq_file *m, void *arg)
 
 static struct kernfs_ops cgroup_kf_single_ops = {
 	.atomic_write_len	= PAGE_SIZE,
+	.open			= cgroup_file_open,
+	.release		= cgroup_file_release,
 	.write			= cgroup_file_write,
 	.seq_show		= cgroup_seqfile_show,
 };
 
 static struct kernfs_ops cgroup_kf_ops = {
 	.atomic_write_len	= PAGE_SIZE,
+	.open			= cgroup_file_open,
+	.release		= cgroup_file_release,
 	.write			= cgroup_file_write,
 	.seq_start		= cgroup_seqfile_start,
 	.seq_next		= cgroup_seqfile_next,

index 87167e4..fd684bfd 100644
@@ -4426,6 +4426,60 @@ int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from)
 	return ret;
 }
 
+static void cgroup_procs_release(struct kernfs_open_file *of)
+{
+	if (of->priv) {
+		css_task_iter_end(of->priv);
+		kfree(of->priv);
+	}
+}
+
+static void *cgroup_procs_next(struct seq_file *s, void *v, loff_t *pos)
+{
+	struct kernfs_open_file *of = s->private;
+	struct css_task_iter *it = of->priv;
+	struct task_struct *task;
+
+	do {
+		task = css_task_iter_next(it);
+	} while (task && !thread_group_leader(task));
+
+	return task;
+}
+
+static void *cgroup_procs_start(struct seq_file *s, loff_t *pos)
+{
+	struct kernfs_open_file *of = s->private;
+	struct cgroup *cgrp = seq_css(s)->cgroup;
+	struct css_task_iter *it = of->priv;
+
+	/*
+	 * When a seq_file is seeked, it's always traversed sequentially
+	 * from position 0, so we can simply keep iterating on !0 *pos.
+	 */
+	if (!it) {
+		if (WARN_ON_ONCE((*pos)++))
+			return ERR_PTR(-EINVAL);
+
+		it = kzalloc(sizeof(*it), GFP_KERNEL);
+		if (!it)
+			return ERR_PTR(-ENOMEM);
+		of->priv = it;
+		css_task_iter_start(&cgrp->self, it);
+	} else if (!(*pos)++) {
+		css_task_iter_end(it);
+		css_task_iter_start(&cgrp->self, it);
+	}
+
+	return cgroup_procs_next(s, NULL, NULL);
+}
+
+static int cgroup_procs_show(struct seq_file *s, void *v)
+{
+	seq_printf(s, "%d\n", task_tgid_vnr(v));
+	return 0;
+}
+
 /*
  * Stuff for reading the 'tasks'/'procs' files.
  *

@@ -4914,11 +4968,10 @@ static struct cftype cgroup_dfl_base_files[] = {
 	{
 		.name = "cgroup.procs",
 		.file_offset = offsetof(struct cgroup, procs_file),
-		.seq_start = cgroup_pidlist_start,
-		.seq_next = cgroup_pidlist_next,
-		.seq_stop = cgroup_pidlist_stop,
-		.seq_show = cgroup_pidlist_show,
-		.private = CGROUP_FILE_PROCS,
+		.release = cgroup_procs_release,
+		.seq_start = cgroup_procs_start,
+		.seq_next = cgroup_procs_next,
+		.seq_show = cgroup_procs_show,
 		.write = cgroup_procs_write,
 	},
 	{

index fd684bfd..b9e2d85 100644
@@ -4618,42 +4618,12 @@ static int pidlist_uniq(pid_t *list, int length)
  * 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 on the
- * default hierarchy 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_on_dfl(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)
 {

@@ -4741,10 +4711,7 @@ static int pidlist_array_load(struct cgroup *cgrp, enum cgroup_filetype type,
 	css_task_iter_end(&it);
 	length = n;
 	/* now sort & (if procs) strip out duplicates */
-	if (cgroup_on_dfl(cgrp))
-		sort(array, length, sizeof(pid_t), fried_cmppid, NULL);
-	else
-		sort(array, length, sizeof(pid_t), cmppid, NULL);
+	sort(array, length, sizeof(pid_t), cmppid, NULL);
 	if (type == CGROUP_FILE_PROCS)
 		length = pidlist_uniq(array, length);
 

@@ -4876,10 +4843,10 @@ static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
 
 		while (index < end) {
 			int mid = (index + end) / 2;
-			if (cgroup_pid_fry(cgrp, l->list[mid]) == pid) {
+			if (l->list[mid] == pid) {
 				index = mid;
 				break;
-			} else if (cgroup_pid_fry(cgrp, l->list[mid]) <= pid)
+			} else if (l->list[mid] <= pid)
 				index = mid + 1;
 			else
 				end = mid;

@@ -4890,7 +4857,7 @@ static void *cgroup_pidlist_start(struct seq_file *s, loff_t *pos)
 		return NULL;
 	/* Update the abstract position to be the actual pid that we found */
 	iter = l->list + index;
-	*pos = cgroup_pid_fry(cgrp, *iter);
+	*pos = *iter;
 	return iter;
 }
 

@@ -4919,7 +4886,7 @@ static void *cgroup_pidlist_next(struct seq_file *s, void *v, loff_t *pos)
 	if (p >= end) {
 		return NULL;
 	} else {
-		*pos = cgroup_pid_fry(seq_css(s)->cgroup, *p);
+		*pos = *p;
 		return p;
 	}
 }