Dear Linus,
Here is the patch that we discussed at LinuxCon Europe. My plan was to send it
the week after LinuxCon Europe (last week) as you had asked. However, I caught
somethng at the conference and was sick last week, so that was delayed.
The history of the patch is that the v1 was done in July. I sent v2 shortly
afterward upon realizing that v1 did not take the opportunity atomically close
the race with respect to bind mounting and marking readonly. v3 and v4 followed
with feedback from Mateusz Guzik, but aside from him, no one else replied.
Then at LinuxCon Europe, I spoke briefly to Mauro Carvalho Chehab, Ted T'so and
yourself about it. Mauro thought it looked okay, but he told me not to include
his ACKed by because he is not involved in this subsystem. Ted was surprised
that this was even a bug. The two of us spoke about so many filesystem things
that I neglected to ask if including him as ACKed by was okay. I have CCed
Mateusz, Ted, Mauro, the usual people from get_maintainers.pl and other
interested parties.
Yours truly,
Richard Yao
Richard Yao (1):
vfs: Respect MS_RDONLY at bind mount creation
fs/namespace.c | 20 ++++++++++++++++----
fs/pnode.h | 17 +++++++++--------
2 files changed, 25 insertions(+), 12 deletions(-)
--
2.0.4
`mount -o bind,ro ...` suffers from a silent failure where the readonly
flag is ignored. The bind mount will be created rw whenever the target
is rw. Users typically workaround this by remounting readonly, but that
does not work when you want to define readonly bind mounts in fstab.
This is a major annoyance when dealing with recursive bind mounts
because the userland mount command does not expose the option to
recursively remount a subtree as readonly.
Signed-off-by: Richard Yao <[email protected]>
---
fs/namespace.c | 20 ++++++++++++++++----
fs/pnode.h | 17 +++++++++--------
2 files changed, 25 insertions(+), 12 deletions(-)
diff --git a/fs/namespace.c b/fs/namespace.c
index 5b66b2b..6f07336 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -990,6 +990,14 @@ static struct mount *clone_mnt(struct mount *old, struct dentry *root,
if (flag & CL_MAKE_SHARED)
set_mnt_shared(mnt);
+ /*
+ * We set the flag directly because the mount point is not yet visible.
+ * This means there are no writers that require the checks in
+ * mnt_make_readonly().
+ */
+ if (flag & CL_MAKE_RDONLY)
+ mnt->mnt.mnt_flags |= MNT_READONLY;
+
/* stick the duplicate mount on the same expiry list
* as the original if that was on one */
if (flag & CL_EXPIRE) {
@@ -1992,11 +2000,13 @@ static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
* do loopback mount.
*/
static int do_loopback(struct path *path, const char *old_name,
- int recurse)
+ unsigned long flags)
{
struct path old_path;
struct mount *mnt = NULL, *old, *parent;
struct mountpoint *mp;
+ int recurse = flags & MS_REC;
+ int clflags = (flags & MS_RDONLY) ? CL_MAKE_RDONLY : 0;
int err;
if (!old_name || !*old_name)
return -EINVAL;
@@ -2027,9 +2037,10 @@ static int do_loopback(struct path *path, const char *old_name,
goto out2;
if (recurse)
- mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
+ mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE |
+ clflags);
else
- mnt = clone_mnt(old, old_path.dentry, 0);
+ mnt = clone_mnt(old, old_path.dentry, clflags);
if (IS_ERR(mnt)) {
err = PTR_ERR(mnt);
@@ -2625,7 +2636,8 @@ long do_mount(const char *dev_name, const char __user *dir_name,
retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
data_page);
else if (flags & MS_BIND)
- retval = do_loopback(&path, dev_name, flags & MS_REC);
+ retval = do_loopback(&path, dev_name, flags & (MS_REC |
+ MS_RDONLY));
else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
retval = do_change_type(&path, flags);
else if (flags & MS_MOVE)
diff --git a/fs/pnode.h b/fs/pnode.h
index 4a24635..326f5be 100644
--- a/fs/pnode.h
+++ b/fs/pnode.h
@@ -20,14 +20,15 @@
#define SET_MNT_MARK(m) ((m)->mnt.mnt_flags |= MNT_MARKED)
#define CLEAR_MNT_MARK(m) ((m)->mnt.mnt_flags &= ~MNT_MARKED)
-#define CL_EXPIRE 0x01
-#define CL_SLAVE 0x02
-#define CL_COPY_UNBINDABLE 0x04
-#define CL_MAKE_SHARED 0x08
-#define CL_PRIVATE 0x10
-#define CL_SHARED_TO_SLAVE 0x20
-#define CL_UNPRIVILEGED 0x40
-#define CL_COPY_MNT_NS_FILE 0x80
+#define CL_EXPIRE 0x001
+#define CL_SLAVE 0x002
+#define CL_COPY_UNBINDABLE 0x004
+#define CL_MAKE_SHARED 0x008
+#define CL_PRIVATE 0x010
+#define CL_SHARED_TO_SLAVE 0x020
+#define CL_UNPRIVILEGED 0x040
+#define CL_COPY_MNT_NS_FILE 0x080
+#define CL_MAKE_RDONLY 0x100
#define CL_COPY_ALL (CL_COPY_UNBINDABLE | CL_COPY_MNT_NS_FILE)
--
2.0.4