From 95964f1b13921d245ca7741d2d209e4754cb6ea8 Mon Sep 17 00:00:00 2001
From: Pavel Emelyanov <xemul@parallels.com>
Date: Wed, 27 Jun 2012 20:57:36 +0400
Subject: [PATCH] mount: Core for creating mount namespace

Namespace restore goes in two steps -- first, clean old mountpoints
(except root) that are left after clone(CLONE_NEWNS) call. Next,
repopulate the namespace with new mountpoints (except root) read
from the image file.

This patch adds the skeleton for the above -- reading the mountpoints
from current proc and image and calling for mount/umount. The real
action will come in further patches.

Signed-off-by: Pavel Emelyanov <xemul@parallels.com>
---
 include/mount.h |   1 +
 mount.c         | 136 +++++++++++++++++++++++++++++++++++++++++++++++-
 namespaces.c    |   2 +
 3 files changed, 137 insertions(+), 2 deletions(-)

diff --git a/include/mount.h b/include/mount.h
index 22f89c2a7..300807c68 100644
--- a/include/mount.h
+++ b/include/mount.h
@@ -10,5 +10,6 @@ struct cr_fdset;
 extern int dump_mnt_ns(int pid, struct cr_fdset *);
 struct cr_options;
 extern void show_mountpoints(int fd, struct cr_options *);
+int prepare_mnt_ns(int pid);
 
 #endif /* MOUNT_H__ */
diff --git a/mount.c b/mount.c
index ce7a00b2b..649cd647e 100644
--- a/mount.c
+++ b/mount.c
@@ -266,8 +266,6 @@ int dump_mnt_ns(int ns_pid, struct cr_fdset *fdset)
 		return -1;
 	}
 
-	mnt_build_tree(pm);
-
 	pr_info("Dumping mountpoints\n");
 
 	img_fd = fdset_fd(fdset, CR_FD_MOUNTPOINTS);
@@ -284,6 +282,140 @@ int dump_mnt_ns(int ns_pid, struct cr_fdset *fdset)
 	return 0;
 }
 
+static int mnt_tree_for_each(struct mount_info *m,
+		int (*fn)(struct mount_info *))
+{
+	pr_err("NOT IMPLEMENTED\n");
+	return -1;
+}
+
+static int mnt_tree_for_each_reverse(struct mount_info *m,
+		int (*fn)(struct mount_info *))
+{
+	pr_err("NOT IMPLEMENTED\n");
+	return -1;
+}
+
+static int do_mount_one(struct mount_info *mi)
+{
+	if (!mi->parent)
+		return 0;
+
+	pr_debug("\tMounting %s @%s\n", mi->fstype, mi->mountpoint);
+	return 0;
+}
+
+static int do_umount_one(struct mount_info *mi)
+{
+	if (!mi->parent)
+		return 0;
+
+	pr_debug("\tUmounting %s\n", mi->mountpoint);
+	return 0;
+}
+
+static int clean_mnt_ns(void)
+{
+	struct mount_info *pm;
+
+	pr_info("Cleaning mount namespace\n");
+
+	pm = parse_mountinfo(getpid());
+	if (!pm) {
+		pr_err("Can't parse my new mount namespace\n");
+		return -1;
+	}
+
+	pm = mnt_build_tree(pm);
+	if (!pm)
+		return -1;
+
+	return mnt_tree_for_each_reverse(pm, do_umount_one);
+}
+
+static int populate_mnt_ns(int ns_pid)
+{
+	int img, ret;
+	struct mount_info *pms = NULL;
+
+	pr_info("Populating mount namespace\n");
+
+	img = open_image_ro(CR_FD_MOUNTPOINTS, ns_pid);
+	if (img < 0)
+		return -1;
+
+	pr_debug("Reading mountpoint images\n");
+
+	while (1) {
+		struct mnt_entry me;
+		struct mount_info *pm;
+
+		ret = read_img_eof(img, &me);
+		if (ret <= 0)
+			break;
+
+		ret = -1;
+		pm = xmalloc(sizeof(*pm));
+		if (!pm)
+			break;
+
+		mnt_entry_init(pm);
+
+		pm->mnt_id = me.mnt_id;
+		pm->parent_mnt_id = me.parent_mnt_id;
+		pm->s_dev = me.root_dev;
+		pm->flags = me.flags;
+		pm->fstype = decode_fstype(me.fstype); /* FIXME: abort unsupported early */
+
+		pr_debug("\t\tGetting root for %d\n", pm->mnt_id);
+		if (read_img_str(img, &pm->root, me.root_dentry_len) < 0)
+			break;
+
+		pr_debug("\t\tGetting mpt for %d\n", pm->mnt_id);
+		if (read_img_str(img, &pm->mountpoint, me.mountpoint_path_len) < 0)
+			break;
+
+		pr_debug("\t\tGetting source for %d\n", pm->mnt_id);
+		if (read_img_str(img, &pm->source, me.source_len) < 0)
+			break;
+
+		pr_debug("\t\tGetting opts for %d\n", pm->mnt_id);
+		if (read_img_str(img, &pm->options, me.options_len) < 0)
+			break;
+
+		pr_debug("\tRead %d mp @ %s\n", pm->mnt_id, pm->mountpoint);
+		pm->next = pms;
+		pms = pm;
+	}
+
+	close(img);
+
+	pms = mnt_build_tree(pms);
+	if (!pms)
+		return -1;
+
+	return mnt_tree_for_each(pms, do_mount_one);
+}
+
+int prepare_mnt_ns(int ns_pid)
+{
+	int ret;
+
+	pr_info("Restoring mount namespace\n");
+
+	/*
+	 * The new mount namespace is filled with the mountpoint
+	 * clones from the original one. We have to umount them
+	 * prior to recreating new ones.
+	 */
+
+	ret = clean_mnt_ns();
+	if (!ret)
+		ret = populate_mnt_ns(ns_pid);
+
+	return ret;
+}
+
 void show_mountpoints(int fd, struct cr_options *o)
 {
 	struct mnt_entry me;
diff --git a/namespaces.c b/namespaces.c
index 67d3d8eeb..9e3644193 100644
--- a/namespaces.c
+++ b/namespaces.c
@@ -124,6 +124,8 @@ int prepare_namespace(int pid, unsigned long clone_flags)
 		ret = prepare_utsns(pid);
 	if (clone_flags & CLONE_NEWIPC)
 		ret = prepare_ipc_ns(pid);
+	if (clone_flags & CLONE_NEWNS)
+		ret = prepare_mnt_ns(pid);
 
 	return ret;
 }