[RFC v3 01/24] fs: unify locking semantics for fs freeze / thaw

[Luis Chamberlain]

On Mon, Jan 16, 2023 at 04:14:55PM +0100, Jan Kara wrote:
> So I  think we may need to also block attempts to unmount frozen filesystem -
> actually GFS2 needs this as well [1].
> 
> [1] lore.kernel.org/r/20221129230736.3462830-1-agruenba at redhat.com

Yes, I reviewed Andreas's patch and I think we end up just complicating
things by allowing us to continue to "support" unmounting frozen
filesystems. Instead it is easier for us to just block that insanity.

Current attempt / non-boot tested or anything.

From: Luis Chamberlain <mcgrof at kernel.org>
Date: Sat, 6 May 2023 20:13:49 -0700
Subject: [RFC] fs: prevent mount / umount of frozen filesystems

Today you can unmount a frozen filesystem. Doing that turns it into
a zombie filesystem, you cannot shut it down until first you remounting
it and then unthawing it.

Enabling this sort of behaviour is madness.

Simplify this by instead just preventing us to unmount frozen
filesystems, and likewise prevent mounting frozen filesystems.

Suggested-by: Jan Kara <jack at suse.cz>
Reported-by: Andreas Gruenbacher <agruenba at redhat.com>
Signed-off-by: Luis Chamberlain <mcgrof at kernel.org>
---
 fs/namespace.c |  3 +++
 fs/super.c     | 14 ++++++++++++++
 2 files changed, 17 insertions(+)

diff --git a/fs/namespace.c b/fs/namespace.c
index 54847db5b819..9c21d8662fc8 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -1636,6 +1636,9 @@ static int do_umount(struct mount *mnt, int flags)
 	if (retval)
 		return retval;
 
+	if (!(sb_is_unfrozen(sb)))
+		return -EBUSY;
+
 	/*
 	 * Allow userspace to request a mountpoint be expired rather than
 	 * unmounting unconditionally. Unmount only happens if:
diff --git a/fs/super.c b/fs/super.c
index 34afe411cf2b..55f5728f5090 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -441,6 +441,7 @@ void retire_super(struct super_block *sb)
 {
 	WARN_ON(!sb->s_bdev);
 	down_write(&sb->s_umount);
+	WARN_ON_ONCE(!(sb_is_unfrozen(sb)));
 	if (sb->s_iflags & SB_I_PERSB_BDI) {
 		bdi_unregister(sb->s_bdi);
 		sb->s_iflags &= ~SB_I_PERSB_BDI;
@@ -468,6 +469,7 @@ void generic_shutdown_super(struct super_block *sb)
 {
 	const struct super_operations *sop = sb->s_op;
 
+	WARN_ON_ONCE(!(sb_is_unfrozen(sb)));
 	if (sb->s_root) {
 		shrink_dcache_for_umount(sb);
 		sync_filesystem(sb);
@@ -1354,6 +1356,12 @@ struct dentry *mount_bdev(struct file_system_type *fs_type,
 	if (IS_ERR(s))
 		goto error_s;
 
+	if (!(sb_is_unfrozen(sb))) {
+			deactivate_locked_super(s);
+			error = -EBUSY;
+			goto error_bdev;
+	}
+
 	if (s->s_root) {
 		if ((flags ^ s->s_flags) & SB_RDONLY) {
 			deactivate_locked_super(s);
@@ -1473,6 +1481,10 @@ struct dentry *mount_single(struct file_system_type *fs_type,
 	s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
 	if (IS_ERR(s))
 		return ERR_CAST(s);
+	if (!(sb_is_unfrozen(sb))) {
+		deactivate_locked_super(s);
+		return ERR_PTR(-EBUSY);
+	}
 	if (!s->s_root) {
 		error = fill_super(s, data, flags & SB_SILENT ? 1 : 0);
 		if (!error)
@@ -1522,6 +1534,8 @@ int vfs_get_tree(struct fs_context *fc)
 
 	sb = fc->root->d_sb;
 	WARN_ON(!sb->s_bdi);
+	if (!(sb_is_unfrozen(sb)))
+		return -EBUSY;
 
 	/*
 	 * Write barrier is for super_cache_count(). We place it before setting
-- 
2.39.2