Re: Upcoming: Notifications, FS notifications and fsinfo()

[Miklos Szeredi <miklos@szeredi.hu>]

[-- Attachment #1: Type: text/plain, Size: 1346 bytes --]

On Wed, Apr 1, 2020 at 11:05 AM Karel Zak <kzak@redhat.com> wrote:
>
> On Tue, Mar 31, 2020 at 10:52:52PM +0100, David Howells wrote:
> > Christian Brauner <christian.brauner@ubuntu.com> wrote:
> >
> > > querying all properties of a mount atomically all-at-once,
> >
> > I don't actually offer that, per se.
> >
> > Having an atomic all-at-once query for a single mount is actually quite a
> > burden on the system.  There's potentially a lot of state involved, much of
> > which you don't necessarily need.
>
> If all means "all possible attributes" than it is unnecessary, for
> example ext4 timestamps or volume uuid/label are rarely necessary.
> We usually need together (as consistent set):
>
>     source
>     mountpoint
>     FS type
>     FS root (FSINFO_ATTR_MOUNT_PATH)
>     FS options (FSINFO_ATTR_CONFIGURATION)
>     VFS attributes
>     VFS propagation flags
>     mount ID
>     parent ID
>     devno (or maj:min)

This is trivial with mountfs (reuse format of /proc/PID/mountinfo):

# cat /mnt/30/info
30 20 0:14 / /mnt rw,relatime - mountfs none rw

Attached patch applies against readfile patch.

We might want something more generic, and it won't get any simpler:

 mount.h          |    1 +
 mountfs/super.c  |   12 +++++++++++-
 proc_namespace.c |    2 +-
 3 files changed, 13 insertions(+), 2 deletions(-)

Thanks,
Miklos

[-- Attachment #2: mountfs-info.patch --]
[-- Type: text/x-patch, Size: 1769 bytes --]

---
 fs/mount.h          |    1 +
 fs/mountfs/super.c  |   12 +++++++++++-
 fs/proc_namespace.c |    2 +-
 3 files changed, 13 insertions(+), 2 deletions(-)

--- a/fs/mount.h
+++ b/fs/mount.h
@@ -186,3 +186,4 @@ void mountfs_create(struct mount *mnt);
 extern void mountfs_remove(struct mount *mnt);
 int mountfs_lookup_internal(struct vfsmount *m, struct path *path);
 
+int show_mountinfo(struct seq_file *m, struct vfsmount *mnt);
--- a/fs/mountfs/super.c
+++ b/fs/mountfs/super.c
@@ -22,7 +22,7 @@ struct mountfs_entry {
 
 static const char *mountfs_attrs[] = {
 	"root", "mountpoint", "id", "parent", "options", "children",
-	"group", "master", "propagate_from", "counter"
+	"group", "master", "propagate_from", "counter", "info"
 };
 
 #define MOUNTFS_INO(id) (((unsigned long) id + 1) * \
@@ -126,11 +126,21 @@ static int mountfs_attr_show(struct seq_
 		if (IS_MNT_SLAVE(mnt)) {
 			get_fs_root(current->fs, &root);
 			tmp = get_dominating_id(mnt, &root);
+			path_put(&root);
 			if (tmp)
 				seq_printf(sf, "%i\n", tmp);
 		}
 	} else if (strcmp(name, "counter") == 0) {
 		seq_printf(sf, "%u\n", atomic_read(&mnt->mnt_topology_changes));
+	} else if (strcmp(name, "info") == 0) {
+		struct proc_mounts p = {};
+
+		WARN_ON(sf->private);
+		sf->private = &p;
+		get_fs_root(current->fs, &p.root);
+		err = show_mountinfo(sf, m);
+		path_put(&p.root);
+		sf->private = NULL;
 	} else {
 		WARN_ON(1);
 		err = -EIO;
--- a/fs/proc_namespace.c
+++ b/fs/proc_namespace.c
@@ -110,7 +110,7 @@ static int show_vfsmnt(struct seq_file *
 	return err;
 }
 
-static int show_mountinfo(struct seq_file *m, struct vfsmount *mnt)
+int show_mountinfo(struct seq_file *m, struct vfsmount *mnt)
 {
 	struct proc_mounts *p = m->private;
 	struct mount *r = real_mount(mnt);