Re: [PATCH 1/2] msync: support syncing a small part of the file

[Andrew Morton <akpm@linux-foundation.org>]

On Thu, 31 May 2012 22:43:54 +0200
Paolo Bonzini <pbonzini@redhat.com> wrote:

> msync does not need to flush changes to the entire file, even with MS_SYNC.
> Instead, it can use vfs_fsync_range to only synchronize a part of the file.
> 
> In addition, not all metadata has to be synced; msync is closer to
> fdatasync than it is to msync.  So, pass 1 to vfs_fsync_range.

Would be nice, but if applications were previously assuming that an
msync() was syncing the whole file, this patch will secretly and subtly
break them.  Convince me that this change won't weaken anyone's data
integrity ;)

> @@ -77,18 +79,23 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags)
>  			goto out_unlock;
>  		}
>  		file = vma->vm_file;
> -		start = vma->vm_end;
> +		next = min(end, vma->vm_end);
>  		if ((flags & MS_SYNC) && file &&
>  				(vma->vm_flags & VM_SHARED)) {
> +			file_offset = vma->vm_pgoff * PAGE_SIZE;
>  			get_file(file);
>  			up_read(&mm->mmap_sem);
> -			error = vfs_fsync(file, 0);
> +			error = vfs_fsync_range(file,
> +					start - vma->vm_start + file_offset,
> +					next - vma->vm_start + file_offset, 1);

Is that an off-by-one?  The "end" argument to vfs_fsync_range() is
"inclusive".  ie, it indexes the last byte, not the last byte+1.  It's
done this way so we can refer to end-of-file with -1UL.