[PATCH] VFS: lookup path of "-o remount" the same way as "umount"

From: NeilBrown
Date: Thu Aug 10 2017 - 02:33:09 EST



"mount -o remount" does not need to validate the target directory
for the same reasons that "umount" doesn't. It just needs to
find the mountpoint and verify it is a mountpount.

So change do_mount() to use user_path_mountpoint_at() in the
MS_REMOUNT case.

This means that mount(.., MS_REMOUNT|MS_RO, ..) on a network
attached filesystem will only access the network if there
is data to be written out. This reduces the chance of a hang
when the network is down.

Systemd-shutdown currently calls
mount(NULL, path, NULL, MS_REMOUNT|MS_RDONLY, ...);
umount2(path, 0);

which is not unreasonable for local filesystems which may still
be in use, but causes a hang for NFS filesystems which have not
been unmounted.

Note that this change does not affect
/usr/bin/mount -o remount,ro ...
as that currently calls "lstat()" on the mount point.

Signed-off-by: NeilBrown <neilb@xxxxxxxx>
---
fs/namespace.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/fs/namespace.c b/fs/namespace.c
index f8893dc6a989..31ded3a1cdff 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2718,7 +2718,10 @@ long do_mount(const char *dev_name, const char __user *dir_name,
((char *)data_page)[PAGE_SIZE - 1] = 0;

/* ... and get the mountpoint */
- retval = user_path(dir_name, &path);
+ if (flags & MS_REMOUNT)
+ retval = user_path_mountpoint_at(AT_FDCWD, dir_name, 0, &path);
+ else
+ retval = user_path(dir_name, &path);
if (retval)
return retval;

--
2.12.2

Attachment: signature.asc
Description: PGP signature