(RESEND) [PATCH] ovl: stacked file operation for mmap

(RESEND) [PATCH] ovl: stacked file operation for mmap

[Chengguang Xu]

Currently only mmap does not behave as stacked file operation,
although in practice there is less change to open a file in
RDONLY mode and take long time to do mmap but the fix looks
reasonable.

Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
---
Hi Miklos,

I'm sorry that I did a mistake about signed-off-by tag in previous
email, so I resend this patch.

 fs/overlayfs/file.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c
index 3582c3ae819c..f98b1c0c975b 100644
--- a/fs/overlayfs/file.c
+++ b/fs/overlayfs/file.c
@@ -461,6 +461,7 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	struct file *realfile = file->private_data;
 	const struct cred *old_cred;
+	struct fd real;
 	int ret;
 
 	if (!realfile->f_op->mmap)
@@ -469,7 +470,11 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 	if (WARN_ON(file != vma->vm_file))
 		return -EIO;
 
-	vma->vm_file = get_file(realfile);
+	ret = ovl_real_fdget(file, &real);
+	if (ret)
+		return ret;
+
+	vma->vm_file = get_file(real.file);
 
 	old_cred = ovl_override_creds(file_inode(file)->i_sb);
 	ret = call_mmap(vma->vm_file, vma);
@@ -477,13 +482,14 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 
 	if (ret) {
 		/* Drop reference count from new vm_file value */
-		fput(realfile);
+		fput(real.file);
 	} else {
 		/* Drop reference count from previous vm_file value */
 		fput(file);
 	}
 
 	ovl_file_accessed(file);
+	fdput(real);
 
 	return ret;
 }
-- 
2.26.2

Re: (RESEND) [PATCH] ovl: stacked file operation for mmap

[Amir Goldstein]

On Fri, Oct 16, 2020 at 7:38 PM Chengguang Xu <cgxu519@mykernel.net> wrote:
>
> Currently only mmap does not behave as stacked file operation,
> although in practice there is less change to open a file in
> RDONLY mode and take long time to do mmap but the fix looks
> reasonable.

I suppose you do not have a real life use case where this fix
is relevant?

The thing is that this change is not without consequence.
It could result in 2 overlapping mmaps on the same RDONLY fd
mapping a different file and it can be even more confusing
than different fd mapping different files.

It is not clear which non-standard behavior is preferred, so without
any evidence that one strange behavior is preferred over the other
I don't think we should change anything.

Thanks,
Amir.

>
> Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
> ---
> Hi Miklos,
>
> I'm sorry that I did a mistake about signed-off-by tag in previous
> email, so I resend this patch.
>
>  fs/overlayfs/file.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c
> index 3582c3ae819c..f98b1c0c975b 100644
> --- a/fs/overlayfs/file.c
> +++ b/fs/overlayfs/file.c
> @@ -461,6 +461,7 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
>  {
>         struct file *realfile = file->private_data;
>         const struct cred *old_cred;
> +       struct fd real;
>         int ret;
>
>         if (!realfile->f_op->mmap)
> @@ -469,7 +470,11 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
>         if (WARN_ON(file != vma->vm_file))
>                 return -EIO;
>
> -       vma->vm_file = get_file(realfile);
> +       ret = ovl_real_fdget(file, &real);
> +       if (ret)
> +               return ret;
> +
> +       vma->vm_file = get_file(real.file);
>
>         old_cred = ovl_override_creds(file_inode(file)->i_sb);
>         ret = call_mmap(vma->vm_file, vma);
> @@ -477,13 +482,14 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
>
>         if (ret) {
>                 /* Drop reference count from new vm_file value */
> -               fput(realfile);
> +               fput(real.file);
>         } else {
>                 /* Drop reference count from previous vm_file value */
>                 fput(file);
>         }
>
>         ovl_file_accessed(file);
> +       fdput(real);
>
>         return ret;
>  }
> --
> 2.26.2
>
>

Re: (RESEND) [PATCH] ovl: stacked file operation for mmap

[Chengguang Xu]

 ---- 在 星期日, 2020-10-18 15:53:23 Amir Goldstein <amir73il@gmail.com> 撰写 ----
 > On Fri, Oct 16, 2020 at 7:38 PM Chengguang Xu <cgxu519@mykernel.net> wrote:
 > >
 > > Currently only mmap does not behave as stacked file operation,
 > > although in practice there is less change to open a file in
 > > RDONLY mode and take long time to do mmap but the fix looks
 > > reasonable.
 > 
 > I suppose you do not have a real life use case where this fix
 > is relevant?

Detected by some unexpectedly running test scripts.

 > 
 > The thing is that this change is not without consequence.
 > It could result in 2 overlapping mmaps on the same RDONLY fd
 > mapping a different file and it can be even more confusing
 > than different fd mapping different files.
 > 
 > It is not clear which non-standard behavior is preferred, so without
 > any evidence that one strange behavior is preferred over the other
 > I don't think we should change anything.
 > 
Makes sense.