Re: [PATCH] fs/namespace.c: fix use-after-free of mount in mnt_warn_timestamp_expiry()

[Eric Biggers]

On Sun, Oct 13, 2019 at 07:04:10PM -0700, Deepa Dinamani wrote:
> Thanks for the fix.
> 
> Would it be better to move the check and warning to a place where the
> access is still safe?
> 
> -Deepa

True, we could just do

diff --git a/fs/namespace.c b/fs/namespace.c
index fe0e9e1410fe..9f2ceb542f05 100644
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -2764,14 +2764,14 @@ static int do_new_mount_fc(struct fs_context *fc, 
struct path *mountpoint,
if (IS_ERR(mnt))
return PTR_ERR(mnt);
 
+   mnt_warn_timestamp_expiry(mountpoint, mnt);
+
error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags);
if (error < 0) {
mntput(mnt);
return error;
}
 
-   mnt_warn_timestamp_expiry(mountpoint, mnt);
-
return error;
 }

But then the warning ("Mounted %s file system ...") is printed even if
do_add_mount() fails so nothing was actually mounted.

Though, it's just a warning message and I think failures here are rare, so maybe
we don't care.  Al, what do you think?

- Eric

Re: [PATCH] fs/namespace.c: fix use-after-free of mount in mnt_warn_timestamp_expiry()

[Deepa Dinamani]

Thanks for the fix.

Would it be better to move the check and warning to a place where the
access is still safe?

-Deepa

> On Oct 9, 2019, at 12:19 AM, Eric Biggers  wrote:
>
> From: Eric Biggers 


On Wed, Oct 9, 2019 at 12:19 AM Eric Biggers  wrote:
>
> From: Eric Biggers 
>
> After do_add_mount() returns success, the caller doesn't hold a
> reference to the 'struct mount' anymore.  So it's invalid to access it
> in mnt_warn_timestamp_expiry().
>
> Fix it by instead passing the super_block and the mnt_flags.  It's safe
> to access the super_block because it's pinned by fs_context::root.
>
> Reported-by: syzbot+da4f525235510683d...@syzkaller.appspotmail.com
> Fixes: f8b92ba67c5d ("mount: Add mount warning for impending timestamp 
> expiry")
> Signed-off-by: Eric Biggers 
> ---
>  fs/namespace.c | 15 +++
>  1 file changed, 7 insertions(+), 8 deletions(-)
>
> diff --git a/fs/namespace.c b/fs/namespace.c
> index fe0e9e1410fe..7ef8edaaed69 100644
> --- a/fs/namespace.c
> +++ b/fs/namespace.c
> @@ -2466,12 +2466,11 @@ static void set_mount_attributes(struct mount *mnt, 
> unsigned int mnt_flags)
> unlock_mount_hash();
>  }
>
> -static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct 
> vfsmount *mnt)
> +static void mnt_warn_timestamp_expiry(struct path *mountpoint,
> + struct super_block *sb, int mnt_flags)
>  {
> -   struct super_block *sb = mnt->mnt_sb;
> -
> -   if (!__mnt_is_readonly(mnt) &&
> -  (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) 
> {
> +   if (!(mnt_flags & MNT_READONLY) && !sb_rdonly(sb) &&
> +   (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > 
> sb->s_time_max)) {
> char *buf = (char *)__get_free_page(GFP_KERNEL);
> char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : 
> ERR_PTR(-ENOMEM);
> struct tm tm;
> @@ -2512,7 +2511,7 @@ static int do_reconfigure_mnt(struct path *path, 
> unsigned int mnt_flags)
> set_mount_attributes(mnt, mnt_flags);
> up_write(>s_umount);
>
> -   mnt_warn_timestamp_expiry(path, >mnt);
> +   mnt_warn_timestamp_expiry(path, sb, mnt_flags);
>
> return ret;
>  }
> @@ -2555,7 +2554,7 @@ static int do_remount(struct path *path, int ms_flags, 
> int sb_flags,
> up_write(>s_umount);
> }
>
> -   mnt_warn_timestamp_expiry(path, >mnt);
> +   mnt_warn_timestamp_expiry(path, sb, mnt_flags);
>
> put_fs_context(fc);
> return err;
> @@ -2770,7 +2769,7 @@ static int do_new_mount_fc(struct fs_context *fc, 
> struct path *mountpoint,
> return error;
> }
>
> -   mnt_warn_timestamp_expiry(mountpoint, mnt);
> +   mnt_warn_timestamp_expiry(mountpoint, sb, mnt_flags);
>
> return error;
>  }
> --
> 2.23.0
>