[CRIU] [PATCH 2/3] mount: allow to dump content even if a part of fs is overmounted (v2)

[Andrey Vagin]

for that the mount point is bind-mounted in a temporary place.

v2: * check, that the fs you get access to at the end is _really_ the
      one you wanted to
    * use switch_ns/restore_ns helpers

Signed-off-by: Andrey Vagin <avagin at openvz.org>
---
 mount.c | 95 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 89 insertions(+), 6 deletions(-)

diff --git a/mount.c b/mount.c
index 68965bb..1ba0ddc 100644
--- a/mount.c
+++ b/mount.c
@@ -197,18 +197,13 @@ static struct mount_info *mnt_build_tree(struct mount_info *list)
 	return tree;
 }
 
-static DIR *open_mountpoint(struct mount_info *pm)
+static DIR *__open_mountpoint(struct mount_info *pm)
 {
 	int fd, ret;
 	char path[PATH_MAX + 1];
 	struct stat st;
 	DIR *fdir;
 
-	if (!list_empty(&pm->children)) {
-		pr_err("Something is mounted on top of %s\n", pm->fstype->name);
-		return NULL;
-	}
-
 	snprintf(path, sizeof(path), ".%s", pm->mountpoint);
 	fd = openat(mntns_root, path, O_RDONLY);
 	if (fd < 0) {
@@ -249,6 +244,94 @@ static int close_mountpoint(DIR *dfd)
 	return 0;
 }
 
+static DIR *open_mountpoint(struct mount_info *pm)
+{
+	int fd = -1, ns_old = -1;
+	DIR *fdir = NULL;
+	char buf[PATH_MAX];
+	struct stat st;
+	char mnt_path[] = "/tmp/cr-tmpfs.XXXXXX";
+	struct ns_desc ns_desc = NS_DESC_ENTRY(CLONE_NEWNS, "mnt");
+
+	/*
+	 * If a mount doesn't have children, we can open a mount point,
+	 * otherwise we need to create a "private" copy.
+	 */
+	if (list_empty(&pm->children))
+		return __open_mountpoint(pm);
+
+	pr_info("Something is mounted on top of %s\n", pm->mountpoint);
+
+	/*
+	 * To create a "private" copy, the target mount is bind-mounted
+	 * in a temporary place w/o MS_REC (non-recursively).
+	 * A mount point can't be bind-mounted in criu's namespace, it will be
+	 * mounted in a target namespace. The sequence of actions is
+	 * mkdtemp, setns(tgt), mount, open, detach, setns(old).
+	 */
+
+	if (switch_ns(root_item->pid.real, &ns_desc, &ns_old) < 0)
+		return NULL;
+
+	if (mkdtemp(mnt_path) == NULL) {
+		pr_perror("Can't create a temporary directory");
+		goto out;
+	}
+
+	snprintf(buf, sizeof(buf), "/proc/self/root/%s", pm->mountpoint);
+	if (mount(buf, mnt_path, NULL, MS_BIND, NULL)) {
+		pr_perror("Can't bind-mount %d:%s to %s",
+				pm->mnt_id, pm->mountpoint, mnt_path);
+		rmdir(mnt_path);
+		goto out;
+	}
+
+	fd = open(mnt_path, O_RDONLY | O_DIRECTORY);
+	if (fd < 0)
+		pr_perror("Can't open %s\n", mnt_path);
+
+	if (umount2(mnt_path, MNT_DETACH)) {
+		pr_perror("Can't umount %s", mnt_path);
+		goto out;
+	}
+
+	if (rmdir(mnt_path)) {
+		pr_perror("Can't remove the directory %s", mnt_path);
+		goto out;
+	}
+
+	if (fd < 0)
+		goto out;
+
+	if (fstat(fd, &st)) {
+		pr_perror("Can't get file status");
+		goto out;
+	}
+
+	if (pm->s_dev != st.st_dev) {
+		pr_err("Can't create a temporary mount "
+			"(device mismatch %lx != %x\n", st.st_dev, pm->s_dev);
+		goto out;
+	}
+
+	if (restore_ns(ns_old, &ns_desc))
+		goto out;
+
+	fdir = fdopendir(fd);
+	if (fdir == NULL) {
+		close(fd);
+		pr_perror("Can't open %s", pm->mountpoint);
+		return NULL;
+	}
+
+	return fdir;
+out:
+	if (ns_old >= 0 && restore_ns(ns_old, &ns_desc))
+		pr_perror("Can't return into the origin mount namespace");
+	close_safe(&fd);
+	return NULL;
+}
+
 static int tmpfs_dump(struct mount_info *pm)
 {
 	int ret = -1;
-- 
1.8.3.1