[Devel] [PATCH rh7 v3 2/2] ve/fs/sync: fix CT's mountpoints traversal

[Andrey Ryabinin]

Currently sync reads 've->root_path.mnt' mount and iterate over it childs.
This doesn't work, because
 a) not all in-container mounts are in childs list of root mount.
 b) ve->root_path.mnt points to incorrect 'struct mount *'.

This patch slightly rework's mounts traversal. Now sync iterates over all
mounts of mount namespaces in ve. List of ve's mount namespaces maintained
via mntns_list.

https://jira.sw.ru/browse/PSBM-44125

Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
---
 fs/mount.h         |  1 +
 fs/namespace.c     |  7 +++++++
 fs/sync.c          | 40 +++++++++++++++++++++-------------------
 include/linux/ve.h |  1 +
 kernel/ve/ve.c     |  4 ++++
 5 files changed, 34 insertions(+), 19 deletions(-)

diff --git a/fs/mount.h b/fs/mount.h
index b496064..285484c 100644
--- a/fs/mount.h
+++ b/fs/mount.h
@@ -7,6 +7,7 @@ struct mnt_namespace {
 	unsigned int		proc_inum;
 	struct mount *	root;
 	struct list_head	list;
+	struct list_head	mntns_list;
 	struct user_namespace	*user_ns;
 	u64			seq;	/* Sequence number to prevent loops */
 	wait_queue_head_t poll;
diff --git a/fs/namespace.c b/fs/namespace.c
index fa9ee9e..8c02d0c 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2569,6 +2569,7 @@ static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
 	atomic_set(&new_ns->count, 1);
 	new_ns->root = NULL;
 	INIT_LIST_HEAD(&new_ns->list);
+	INIT_LIST_HEAD(&new_ns->mntns_list);
 	init_waitqueue_head(&new_ns->poll);
 	new_ns->event = 0;
 	new_ns->user_ns = get_user_ns(user_ns);
@@ -2609,6 +2610,8 @@ static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
 	list_add_tail(&new_ns->list, &new->mnt_list);
 	br_write_unlock(&vfsmount_lock);
 
+	list_add(&new_ns->mntns_list, &get_exec_env()->mntns_list);
+
 	/*
 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
 	 * as belonging to new namespace.  We have already acquired a private
@@ -2673,7 +2676,10 @@ static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
 		struct mount *mnt = real_mount(m);
 		mnt->mnt_ns = new_ns;
 		new_ns->root = mnt;
+		down_write(&namespace_sem);
 		list_add(&mnt->mnt_list, &new_ns->list);
+		list_add(&new_ns->mntns_list, &get_exec_env()->mntns_list);
+		up_write(&namespace_sem);
 	} else {
 		mntput(m);
 	}
@@ -2954,6 +2960,7 @@ void put_mnt_ns(struct mnt_namespace *ns)
 	namespace_lock();
 	br_write_lock(&vfsmount_lock);
 	umount_tree(ns->root, 0);
+	list_del(&ns->mntns_list);
 	br_write_unlock(&vfsmount_lock);
 	namespace_unlock();
 	free_mnt_ns(ns);
diff --git a/fs/sync.c b/fs/sync.c
index 0eb621f..abb52f5 100644
--- a/fs/sync.c
+++ b/fs/sync.c
@@ -129,33 +129,35 @@ static int sync_filesystem_collected(struct list_head *sync_list, struct super_b
 
 static int sync_collect_filesystems(struct ve_struct *ve, struct list_head *sync_list)
 {
-	struct mount *root = real_mount(ve->root_path.mnt);
 	struct mount *mnt;
+	struct mnt_namespace *mnt_ns;
 	struct sync_sb *ss;
 	int ret = 0;
 
 	BUG_ON(!list_empty(sync_list));
 
 	down_read(&namespace_sem);
-	for (mnt = root; mnt; mnt = next_mnt(mnt, root)) {
-		if (sync_filesystem_collected(sync_list, mnt->mnt.mnt_sb))
-			continue;
-
-		ss = kmalloc(sizeof(*ss), GFP_KERNEL);
-		if (ss == NULL) {
-			ret = -ENOMEM;
-			break;
+	list_for_each_entry(mnt_ns, &ve->mntns_list, mntns_list) {
+		list_for_each_entry(mnt, &mnt_ns->list, mnt_list) {
+			if (sync_filesystem_collected(sync_list, mnt->mnt.mnt_sb))
+				continue;
+
+			ss = kmalloc(sizeof(*ss), GFP_KERNEL);
+			if (ss == NULL) {
+				ret = -ENOMEM;
+				break;
+			}
+			ss->sb = mnt->mnt.mnt_sb;
+			/*
+			 * We hold mount point and thus can be sure, that superblock is
+			 * alive. And it means, that we can safely increase it's usage
+			 * counter.
+			 */
+			spin_lock(&sb_lock);
+			ss->sb->s_count++;
+			spin_unlock(&sb_lock);
+			list_add_tail(&ss->list, sync_list);
 		}
-		ss->sb = mnt->mnt.mnt_sb;
-		/*
-		 * We hold mount point and thus can be sure, that superblock is
-		 * alive. And it means, that we can safely increase it's usage
-		 * counter.
-		 */
-		spin_lock(&sb_lock);
-		ss->sb->s_count++;
-		spin_unlock(&sb_lock);
-		list_add_tail(&ss->list, sync_list);
 	}
 	up_read(&namespace_sem);
 	return ret;
diff --git a/include/linux/ve.h b/include/linux/ve.h
index e603d9e..243bea1 100644
--- a/include/linux/ve.h
+++ b/include/linux/ve.h
@@ -115,6 +115,7 @@ struct ve_struct {
 	struct list_head	devmnt_list;
 	struct mutex		devmnt_mutex;
 
+	struct list_head	mntns_list;
 	struct kmapset_key	ve_sysfs_perms;
 
 #ifdef CONFIG_AIO
diff --git a/kernel/ve/ve.c b/kernel/ve/ve.c
index 231f398..ffd55e4 100644
--- a/kernel/ve/ve.c
+++ b/kernel/ve/ve.c
@@ -80,6 +80,7 @@ struct ve_struct ve0 = {
 	.sched_lat_ve.cur	= &ve0_lat_stats,
 	.init_cred		= &init_cred,
 	.mnt_nr			= 0,
+	.mntns_list		= LIST_HEAD_INIT(ve0.mntns_list),
 };
 EXPORT_SYMBOL(ve0);
 
@@ -652,6 +653,7 @@ do_init:
 	INIT_LIST_HEAD(&ve->devices);
 	INIT_LIST_HEAD(&ve->ve_list);
 	INIT_LIST_HEAD(&ve->devmnt_list);
+	INIT_LIST_HEAD(&ve->mntns_list);
 	mutex_init(&ve->devmnt_mutex);
 	kmapset_init_key(&ve->ve_sysfs_perms);
 
@@ -706,6 +708,8 @@ static void ve_destroy(struct cgroup *cg)
 	kmapset_unlink(&ve->ve_sysfs_perms, &ve_sysfs_perms);
 	free_ve_devmnts(ve);
 
+	WARN_ON(!list_empty(&ve->mntns_list));
+
 	ve_log_destroy(ve);
 #if IS_ENABLED(CONFIG_BINFMT_MISC)
 	kfree(ve->binfmt_misc);
-- 
2.4.10