[RFC PATCH] 9p: create writeback fid on shared mmap

[RFC PATCH] 9p: create writeback fid on shared mmap

[Chengguang Xu]

If vma is shared and the file was opened for writing,
we should also create writeback fid because vma may be
mprotected writable even if now readonly.

Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
---
Caveat: Only compile tested.

 fs/9p/vfs_file.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/fs/9p/vfs_file.c b/fs/9p/vfs_file.c
index b177fd3b1eb3..791839c2dd5c 100644
--- a/fs/9p/vfs_file.c
+++ b/fs/9p/vfs_file.c
@@ -516,8 +516,7 @@ v9fs_mmap_file_mmap(struct file *filp, struct vm_area_struct *vma)
 	v9inode = V9FS_I(inode);
 	mutex_lock(&v9inode->v_mutex);
 	if (!v9inode->writeback_fid &&
-	    (vma->vm_flags & VM_SHARED) &&
-	    (vma->vm_flags & VM_WRITE)) {
+	    mapping_writably_mapped(filp->f_mapping)) {
 		/*
 		 * clone a fid and add it to writeback_fid
 		 * we do it during mmap instead of
-- 
2.26.2

Re: [RFC PATCH] 9p: create writeback fid on shared mmap

[Dominique Martinet]

Chengguang Xu wrote on Sat, Dec 05, 2020:
> If vma is shared and the file was opened for writing,
> we should also create writeback fid because vma may be
> mprotected writable even if now readonly.

Hm, I guess it makes sense.

> Signed-off-by: Chengguang Xu <cgxu519@mykernel.net>
> ---
> Caveat: Only compile tested.

Will test later and add it to next then, might be a bit.


>  fs/9p/vfs_file.c | 3 +--
>  1 file changed, 1 insertion(+), 2 deletions(-)
> 
> diff --git a/fs/9p/vfs_file.c b/fs/9p/vfs_file.c
> index b177fd3b1eb3..791839c2dd5c 100644
> --- a/fs/9p/vfs_file.c
> +++ b/fs/9p/vfs_file.c
> @@ -516,8 +516,7 @@ v9fs_mmap_file_mmap(struct file *filp, struct vm_area_struct *vma)
>  	v9inode = V9FS_I(inode);
>  	mutex_lock(&v9inode->v_mutex);
>  	if (!v9inode->writeback_fid &&
> -	    (vma->vm_flags & VM_SHARED) &&
> -	    (vma->vm_flags & VM_WRITE)) {
> +	    mapping_writably_mapped(filp->f_mapping)) {
>  		/*
>  		 * clone a fid and add it to writeback_fid
>  		 * we do it during mmap instead of
-- 
Dominique

Re: [V9fs-developer] [RFC PATCH] 9p: create writeback fid on shared mmap

[Dominique Martinet]

Dominique Martinet wrote on Sun, Dec 06, 2020:
> Chengguang Xu wrote on Sat, Dec 05, 2020:
> > If vma is shared and the file was opened for writing,
> > we should also create writeback fid because vma may be
> > mprotected writable even if now readonly.
> 
> Hm, I guess it makes sense.

I had a second look, and generic_file_readonly_mmap uses vma's
`vma->vm_flags & VM_MAYWRITE` instead (together with VM_SHARED),
while mapping_writably_mapped ultimately basically only seems to
validate that the mapping is shared from a look at mapping_map_writable
callers? It's not very clear to me.

OTOH, VM_MAYWRITE is set anytime we have a shared map where file has
been opened read-write, which seems to be what you want with regards to
protecting from mprotect calls.

How about simply changing check from WRITE to MAYWRITE?

 	v9inode = V9FS_I(inode);
 	mutex_lock(&v9inode->v_mutex);
 	if (!v9inode->writeback_fid &&
 	    (vma->vm_flags & VM_SHARED) &&
-	    (vma->vm_flags & VM_WRITE)) {
+	    (vma->vm_flags & VM_MAYWRITE)) {
 		/*
 		 * clone a fid and add it to writeback_fid
 		 * we do it during mmap instead of
-- 
Dominique

Re: [V9fs-developer] [RFC PATCH] 9p: create writeback fid on shared mmap

[Chengguang Xu]

 ---- 在 星期一, 2020-12-07 04:53:18 Dominique Martinet <asmadeus@codewreck.org> 撰写 ----
 > Dominique Martinet wrote on Sun, Dec 06, 2020:
 > > Chengguang Xu wrote on Sat, Dec 05, 2020:
 > > > If vma is shared and the file was opened for writing,
 > > > we should also create writeback fid because vma may be
 > > > mprotected writable even if now readonly.
 > > 
 > > Hm, I guess it makes sense.
 > 
 > I had a second look, and generic_file_readonly_mmap uses vma's
 > `vma->vm_flags & VM_MAYWRITE` instead (together with VM_SHARED),
 > while mapping_writably_mapped ultimately basically only seems to
 > validate that the mapping is shared from a look at mapping_map_writable
 > callers? It's not very clear to me.
 > 
 > , VM_MAYWRITE is set anytime we have a shared map where file has
 > been opened read-write, which seems to be what you want with regards to
 > protecting from mprotect calls.
 > 
 > How about simply changing check from WRITE to MAYWRITE?

It would be fine and based on the code in do_mmap(), it  seems we even don't
need extra check here.  The condition (vma->vm_flags & VM_SHARED) will be enough.
Am I missing something?

Thanks,
Chengguang

 > 
 >      v9inode = V9FS_I(inode);
 >      mutex_lock(&v9inode->v_mutex);
 >      if (!v9inode->writeback_fid &&
 >          (vma->vm_flags & VM_SHARED) &&
 > -        (vma->vm_flags & VM_WRITE)) {
 > +        (vma->vm_flags & VM_MAYWRITE)) {
 >          /*
 >           * clone a fid and add it to writeback_fid
 >           * we do it during mmap instead of

Re: [V9fs-developer] [RFC PATCH] 9p: create writeback fid on shared mmap

[Dominique Martinet]

Chengguang Xu wrote on Mon, Dec 07, 2020:
>  > , VM_MAYWRITE is set anytime we have a shared map where file has
>  > been opened read-write, which seems to be what you want with regards to
>  > protecting from mprotect calls.
>  > 
>  > How about simply changing check from WRITE to MAYWRITE?
> 
> It would be fine and based on the code in do_mmap(), it  seems we even don't
> need extra check here.  The condition (vma->vm_flags & VM_SHARED) will be enough.
> Am I missing something?

VM_MAYWRITE is unset if the file hasn't been open for writing (in which
case the mapping can't be mprotect()ed to writable map), so checking it
is a bit more efficient.

Anyway I'd like to obsolete the writeback fid uses now that fids have a
refcount (this usecase can be a simple refcount increase), in which case
efficiency is less of a problem, but we're not there yet...

Please resend with MAYWRITE if you want authorship and I'll try to take
some time to test incl. the mprotect usecase.

-- 
Dominique

Re: [V9fs-developer] [RFC PATCH] 9p: create writeback fid on shared mmap

[Chengguang Xu]

 ---- 在 星期一, 2020-12-07 19:24:10 Dominique Martinet <asmadeus@codewreck.org> 撰写 ----
 > Chengguang Xu wrote on Mon, Dec 07, 2020:
 > >  > , VM_MAYWRITE is set anytime we have a shared map where file has
 > >  > been opened read-write, which seems to be what you want with regards to
 > >  > protecting from mprotect calls.
 > >  > 
 > >  > How about simply changing check from WRITE to MAYWRITE?
 > > 
 > > It would be fine and based on the code in do_mmap(), it  seems we even don't
 > > need extra check here.  The condition (vma->vm_flags & VM_SHARED) will be enough.
 > > Am I missing something?
 > 
 > VM_MAYWRITE is unset if the file hasn't been open for writing (in which
 > case the mapping can't be mprotect()ed to writable map), so checking it
 > is a bit more efficient.
 > 
 > Anyway I'd like to obsolete the writeback fid uses now that fids have a
 > refcount (this usecase can be a simple refcount increase), in which case
 > efficiency is less of a problem, but we're not there yet...
 > 
 > Please resend with MAYWRITE if you want authorship and I'll try to take
 > some time to test incl. the mprotect usecase.
 > 

Thanks for the review, I'll send revised version later.

Thanks,
Chengguang