diff mbox series

[RFC,v3,06/10] ovl: mark overlayfs' inode dirty on shared mmap

Message ID 20201108140307.1385745-7-cgxu519@mykernel.net (mailing list archive)
State New
Headers show
Series implement containerized syncfs for overlayfs | expand

Commit Message

Chengguang Xu Nov. 8, 2020, 2:03 p.m. UTC
Overlayfs cannot be notified when mmapped area gets dirty,
so we need to proactively mark inode dirty in ->mmap operation.

Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
---
 fs/overlayfs/file.c | 2 ++
 1 file changed, 2 insertions(+)

Comments

Chengguang Xu Nov. 11, 2020, 1:05 p.m. UTC | #1
---- 在 星期日, 2020-11-08 22:03:03 Chengguang Xu <cgxu519@mykernel.net> 撰写 ----
 > Overlayfs cannot be notified when mmapped area gets dirty,
 > so we need to proactively mark inode dirty in ->mmap operation.
 > 
 > Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
 > ---
 >  fs/overlayfs/file.c | 2 ++
 >  1 file changed, 2 insertions(+)
 > 
 > diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c
 > index efccb7c1f9bc..662252047fff 100644
 > --- a/fs/overlayfs/file.c
 > +++ b/fs/overlayfs/file.c
 > @@ -486,6 +486,8 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 >          /* Drop reference count from new vm_file value */
 >          fput(realfile);
 >      } else {
 > +        if (vma->vm_flags & (VM_SHARED|VM_MAYSHARE))

Maybe it's better to mark dirty only having upper inode.


 > +            ovl_mark_inode_dirty(file_inode(file));
 >          /* Drop reference count from previous vm_file value */
 >          fput(file);
 >      }
 > -- 
 > 2.26.2
 > 
 > 
 >
Amir Goldstein Nov. 11, 2020, 3:20 p.m. UTC | #2
On Wed, Nov 11, 2020 at 3:05 PM Chengguang Xu <cgxu519@mykernel.net> wrote:
>
>  ---- 在 星期日, 2020-11-08 22:03:03 Chengguang Xu <cgxu519@mykernel.net> 撰写 ----
>  > Overlayfs cannot be notified when mmapped area gets dirty,
>  > so we need to proactively mark inode dirty in ->mmap operation.
>  >
>  > Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
>  > ---
>  >  fs/overlayfs/file.c | 2 ++
>  >  1 file changed, 2 insertions(+)
>  >
>  > diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c
>  > index efccb7c1f9bc..662252047fff 100644
>  > --- a/fs/overlayfs/file.c
>  > +++ b/fs/overlayfs/file.c
>  > @@ -486,6 +486,8 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
>  >          /* Drop reference count from new vm_file value */
>  >          fput(realfile);
>  >      } else {
>  > +        if (vma->vm_flags & (VM_SHARED|VM_MAYSHARE))
>
> Maybe it's better to mark dirty only having upper inode.
>

Yeh.

And since mapping_map_writable() is only called if VM_SHARED flag
is set (and not VM_MAYSHARE), we are not going to re-dirty an inode on
account of VM_MAYSHARE alone, so I wonder why we need to mark it
dirty here on account of VM_MAYSHARE?

Thanks,
Amir.
Chengguang Xu Nov. 11, 2020, 4:09 p.m. UTC | #3
---- 在 星期三, 2020-11-11 23:20:56 Amir Goldstein <amir73il@gmail.com> 撰写 ----
 > On Wed, Nov 11, 2020 at 3:05 PM Chengguang Xu <cgxu519@mykernel.net> wrote:
 > >
 > >  ---- 在 星期日, 2020-11-08 22:03:03 Chengguang Xu <cgxu519@mykernel.net> 撰写 ----
 > >  > Overlayfs cannot be notified when mmapped area gets dirty,
 > >  > so we need to proactively mark inode dirty in ->mmap operation.
 > >  >
 > >  > Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
 > >  > ---
 > >  >  fs/overlayfs/file.c | 2 ++
 > >  >  1 file changed, 2 insertions(+)
 > >  >
 > >  > diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c
 > >  > index efccb7c1f9bc..662252047fff 100644
 > >  > --- a/fs/overlayfs/file.c
 > >  > +++ b/fs/overlayfs/file.c
 > >  > @@ -486,6 +486,8 @@ static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 > >  >          /* Drop reference count from new vm_file value */
 > >  >          fput(realfile);
 > >  >      } else {
 > >  > +        if (vma->vm_flags & (VM_SHARED|VM_MAYSHARE))
 > >
 > > Maybe it's better to mark dirty only having upper inode.
 > >
 > 
 > Yeh.
 > 
 > And since mapping_map_writable() is only called if VM_SHARED flag
 > is set (and not VM_MAYSHARE), we are not going to re-dirty an inode on
 > account of VM_MAYSHARE alone, so I wonder why we need to mark it
 > dirty here on account of VM_MAYSHARE?
 > 

Yeah, you are right. It just means the pages in this memory region can be mapped
by other process with shared+write mode and this is actually meaningless in our case.
So let's just ignore it. :-)


Thanks,
Chengguang
diff mbox series

Patch

diff --git a/fs/overlayfs/file.c b/fs/overlayfs/file.c
index efccb7c1f9bc..662252047fff 100644
--- a/fs/overlayfs/file.c
+++ b/fs/overlayfs/file.c
@@ -486,6 +486,8 @@  static int ovl_mmap(struct file *file, struct vm_area_struct *vma)
 		/* Drop reference count from new vm_file value */
 		fput(realfile);
 	} else {
+		if (vma->vm_flags & (VM_SHARED|VM_MAYSHARE))
+			ovl_mark_inode_dirty(file_inode(file));
 		/* Drop reference count from previous vm_file value */
 		fput(file);
 	}