Re: [PATCH 08/25] vfs: Allow fsinfo() to look up a mount object by ID [ver #14]

[Christian Brauner]

On Mon, Jun 24, 2019 at 03:09:58PM +0100, David Howells wrote:
> Allow the fsinfo() syscall to look up a mount object by ID rather than by
> pathname.  This is necessary as there can be multiple mounts stacked up at
> the same pathname and there's no way to look through them otherwise.
> 
> This is done by passing AT_FSINFO_MOUNTID_PATH to fsinfo() in the
> parameters and then passing the mount ID as a string to fsinfo() in place
> of the filename:
> 
> 	struct fsinfo_params params = {
> 		.at_flags = AT_FSINFO_MOUNTID_PATH,
> 		.request = FSINFO_ATTR_IDS,
> 	};
> 
> 	ret = fsinfo(AT_FDCWD, "21", &params, buffer, sizeof(buffer));
> 
> The caller is only permitted to query a mount object if the root directory
> of that mount connects directly to the current chroot if dfd == AT_FDCWD[*]
> or the directory specified by dfd otherwise.  Note that this is not
> available to the pathwalk of any other syscall.
> 
> [*] This needs to be something other than AT_FDCWD, perhaps AT_FDROOT.
> 
> [!] This probably needs an LSM hook.
> 
> [!] This might want to check the permissions on all the intervening dirs -
>     but it would have to do that under RCU conditions.
> 
> [!] This might want to check a CAP_* flag.

I would recommend you drop that patch for now.

First, because that patchset itself is already so huge that it's hard to
get enough reviewers to look at it in detail.
Second, from your "[!]" comments above it's obvious that the details of
this feature are not clear enough and need more discussion.
Third, (relatest to the first point) this patchset could use a little
shrinking. :)

> 
> Signed-off-by: David Howells <dhowells@xxxxxxxxxx>
> ---
> 
>  fs/fsinfo.c                |   56 +++++++++++++++++++++
>  fs/internal.h              |    2 +
>  fs/namespace.c             |  117 +++++++++++++++++++++++++++++++++++++++++++-
>  include/uapi/linux/fcntl.h |    1 
>  4 files changed, 173 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/fsinfo.c b/fs/fsinfo.c
> index 127538193b77..3b218f9fedb7 100644
> --- a/fs/fsinfo.c
> +++ b/fs/fsinfo.c
> @@ -516,6 +516,57 @@ static int vfs_fsinfo_fscontext(int fd, struct fsinfo_kparams *params)
>  	return ret;
>  }
>  
> +/*
> + * Look up the root of a mount object.  This allows access to mount objects
> + * (and their attached superblocks) that can't be retrieved by path because
> + * they're entirely covered.
> + *
> + * We only permit access to a mount that has a direct path between either the
> + * dentry pointed to by dfd or to our chroot (if dfd is AT_FDCWD).
> + */
> +static int vfs_fsinfo_mount(int dfd, const char __user *filename,
> +			    struct fsinfo_kparams *params)
> +{
> +	struct path path;
> +	struct fd f = {};
> +	char *name;
> +	long mnt_id;
> +	int ret;
> +
> +	if ((params->at_flags & ~AT_FSINFO_MOUNTID_PATH) ||
> +	    !filename)
> +		return -EINVAL;
> +
> +	name = strndup_user(filename, 32);
> +	if (IS_ERR(name))
> +		return PTR_ERR(name);
> +	ret = kstrtoul(name, 0, &mnt_id);
> +	if (ret < 0)
> +		goto out_name;
> +	if (mnt_id > INT_MAX)
> +		goto out_name;
> +
> +	if (dfd != AT_FDCWD) {
> +		ret = -EBADF;
> +		f = fdget_raw(dfd);
> +		if (!f.file)
> +			goto out_name;
> +	}
> +
> +	ret = lookup_mount_object(f.file ? &f.file->f_path : NULL,
> +				  mnt_id, &path);
> +	if (ret < 0)
> +		goto out_fd;
> +
> +	ret = vfs_fsinfo(&path, params);
> +	path_put(&path);
> +out_fd:
> +	fdput(f);
> +out_name:
> +	kfree(name);
> +	return ret;
> +}
> +
>  /*
>   * Return buffer information by requestable attribute.
>   *
> @@ -629,6 +680,9 @@ SYSCALL_DEFINE5(fsinfo,
>  
>  		if ((params.at_flags & AT_FSINFO_FROM_FSOPEN) && filename)
>  			return -EINVAL;
> +		if ((params.at_flags & (AT_FSINFO_FROM_FSOPEN | AT_FSINFO_MOUNTID_PATH)) ==
> +		    (AT_FSINFO_FROM_FSOPEN | AT_FSINFO_MOUNTID_PATH))
> +			return -EINVAL;
>  	} else {
>  		params.request = FSINFO_ATTR_STATFS;
>  	}
> @@ -688,6 +742,8 @@ SYSCALL_DEFINE5(fsinfo,
>  
>  	if (params.at_flags & AT_FSINFO_FROM_FSOPEN)
>  		ret = vfs_fsinfo_fscontext(dfd, &params);
> +	else if (params.at_flags & AT_FSINFO_MOUNTID_PATH)
> +		ret = vfs_fsinfo_mount(dfd, filename, &params);
>  	else if (filename)
>  		ret = vfs_fsinfo_path(dfd, filename, &params);
>  	else
> diff --git a/fs/internal.h b/fs/internal.h
> index b089a489da1f..074b1c65e3bd 100644
> --- a/fs/internal.h
> +++ b/fs/internal.h
> @@ -97,6 +97,8 @@ extern int __mnt_want_write_file(struct file *);
>  extern void __mnt_drop_write_file(struct file *);
>  
>  extern void dissolve_on_fput(struct vfsmount *);
> +extern int lookup_mount_object(struct path *, int, struct path *);
> +
>  /*
>   * fs_struct.c
>   */
> diff --git a/fs/namespace.c b/fs/namespace.c
> index 1141641dff96..a49a7d9ed482 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -62,7 +62,7 @@ static int __init set_mphash_entries(char *str)
>  __setup("mphash_entries=", set_mphash_entries);
>  
>  static u64 event;
> -static DEFINE_IDA(mnt_id_ida);
> +static DEFINE_IDR(mnt_id_ida);
>  static DEFINE_IDA(mnt_group_ida);
>  
>  static struct hlist_head *mount_hashtable __read_mostly;
> @@ -101,17 +101,27 @@ static inline struct hlist_head *mp_hash(struct dentry *dentry)
>  
>  static int mnt_alloc_id(struct mount *mnt)
>  {
> -	int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
> +	int res;
>  
> +	/* Allocate an ID, but don't set the pointer back to the mount until
> +	 * later, as once we do that, we have to follow RCU protocols to get
> +	 * rid of the mount struct.
> +	 */
> +	res = idr_alloc(&mnt_id_ida, NULL, 0, INT_MAX, GFP_KERNEL);
>  	if (res < 0)
>  		return res;
>  	mnt->mnt_id = res;
>  	return 0;
>  }
>  
> +static void mnt_publish_id(struct mount *mnt)
> +{
> +	idr_replace(&mnt_id_ida, mnt, mnt->mnt_id);
> +}
> +
>  static void mnt_free_id(struct mount *mnt)
>  {
> -	ida_free(&mnt_id_ida, mnt->mnt_id);
> +	idr_remove(&mnt_id_ida, mnt->mnt_id);
>  }
>  
>  /*
> @@ -974,6 +984,7 @@ struct vfsmount *vfs_create_mount(struct fs_context *fc)
>  	lock_mount_hash();
>  	list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
>  	unlock_mount_hash();
> +	mnt_publish_id(mnt);
>  	return &mnt->mnt;
>  }
>  EXPORT_SYMBOL(vfs_create_mount);
> @@ -1067,6 +1078,7 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
>  	lock_mount_hash();
>  	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
>  	unlock_mount_hash();
> +	mnt_publish_id(mnt);
>  
>  	if ((flag & CL_SLAVE) ||
>  	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
> @@ -3988,3 +4000,102 @@ const struct proc_ns_operations mntns_operations = {
>  	.install	= mntns_install,
>  	.owner		= mntns_owner,
>  };
> +
> +/*
> + * See if one path point connects directly to another by ancestral relationship
> + * across mountpoints.  Must call with the RCU read lock held.
> + */
> +static bool are_paths_connected(struct path *ancestor, struct path *to_check)
> +{
> +	struct mount *mnt, *parent;
> +	struct path cursor;
> +	unsigned seq;
> +	bool connected;
> +
> +	seq = 0;
> +restart:
> +	cursor = *to_check;
> +
> +	read_seqbegin_or_lock(&rename_lock, &seq);
> +	while (cursor.mnt != ancestor->mnt) {
> +		mnt = real_mount(cursor.mnt);
> +		parent = READ_ONCE(mnt->mnt_parent);
> +		if (mnt == parent)
> +			goto failed;
> +		cursor.dentry = READ_ONCE(mnt->mnt_mountpoint);
> +		cursor.mnt = &parent->mnt;
> +	}
> +
> +	while (cursor.dentry != ancestor->dentry) {
> +		if (cursor.dentry == cursor.mnt->mnt_root ||
> +		    IS_ROOT(cursor.dentry))
> +			goto failed;
> +		cursor.dentry = READ_ONCE(cursor.dentry->d_parent);
> +	}
> +
> +	connected = true;
> +out:
> +	done_seqretry(&rename_lock, seq);
> +	return connected;
> +
> +failed:
> +	if (need_seqretry(&rename_lock, seq)) {
> +		seq = 1;
> +		goto restart;
> +	}
> +	connected = false;
> +	goto out;
> +}
> +
> +/**
> + * lookup_mount_object - Look up a vfsmount object by ID
> + * @root: The mount root must connect backwards to this point (or chroot if NULL).
> + * @id: The ID of the mountpoint.
> + * @_mntpt: Where to return the resulting mountpoint path.
> + *
> + * Look up the root of the mount with the corresponding ID.  This is only
> + * permitted if that mount connects directly to the specified root/chroot.
> + */
> +int lookup_mount_object(struct path *root, int mnt_id, struct path *_mntpt)
> +{
> +	struct mount *mnt;
> +	struct path stop, mntpt = {};
> +	int ret = -EPERM;
> +
> +	if (!root)
> +		get_fs_root(current->fs, &stop);
> +	else
> +		stop = *root;
> +
> +	rcu_read_lock();
> +	lock_mount_hash();
> +	mnt = idr_find(&mnt_id_ida, mnt_id);
> +	if (!mnt)
> +		goto out_unlock_mh;
> +	if (mnt->mnt.mnt_flags & (MNT_SYNC_UMOUNT | MNT_UMOUNT | MNT_DOOMED))
> +		goto out_unlock_mh;
> +	if (mnt_get_count(mnt) == 0)
> +		goto out_unlock_mh;
> +	mnt_add_count(mnt, 1);
> +	mntpt.mnt = &mnt->mnt;
> +	mntpt.dentry = dget(mnt->mnt.mnt_root);
> +	unlock_mount_hash();
> +
> +	if (are_paths_connected(&stop, &mntpt)) {
> +		*_mntpt = mntpt;
> +		mntpt.mnt = NULL;
> +		mntpt.dentry = NULL;
> +		ret = 0;
> +	}
> +
> +out_unlock:
> +	rcu_read_unlock();
> +	if (!root)
> +		path_put(&stop);
> +	path_put(&mntpt);
> +	return ret;
> +
> +out_unlock_mh:
> +	unlock_mount_hash();
> +	goto out_unlock;
> +}
> diff --git a/include/uapi/linux/fcntl.h b/include/uapi/linux/fcntl.h
> index 6a2402a8fa30..5fda91cfca8a 100644
> --- a/include/uapi/linux/fcntl.h
> +++ b/include/uapi/linux/fcntl.h
> @@ -92,6 +92,7 @@
>  #define AT_STATX_DONT_SYNC	0x4000	/* - Don't sync attributes with the server */
>  
>  #define AT_FSINFO_FROM_FSOPEN	0x2000	/* Examine the fs_context attached to dfd by fsopen() */
> +#define AT_FSINFO_MOUNTID_PATH	0x4000	/* The path is a mount object ID, not an actual path */
>  
>  #define AT_RECURSIVE		0x8000	/* Apply to the entire subtree */
>  
>