[CRIU] [PATCH 13/15] cr-restore: remove unshared pages from inherited private mappings

[Pavel Emelyanov]

On 11/02/2012 05:32 PM, Andrey Vagin wrote:
> A parent process can change a few pages after forking a child and
> all this pages should not be avaliable from the child.
> 
> Each vma has a bitmap of existent pages. Parent's and child's bitmaps
> can be compared and all pages which are not present in a child bitmap
> are dropped.
> 
> Signed-off-by: Andrey Vagin <avagin at openvz.org>
> ---
>  cr-restore.c      | 42 +++++++++++++++++++++++++++++++++++++++++-
>  include/crtools.h |  2 ++
>  2 files changed, 43 insertions(+), 1 deletion(-)
> 
> diff --git a/cr-restore.c b/cr-restore.c
> index 9749634..eb759cb 100644
> --- a/cr-restore.c
> +++ b/cr-restore.c
> @@ -187,6 +187,7 @@ static int map_private_vma(pid_t pid, struct vma_area *vma, void *tgt_addr,
>  {
>  	int ret;
>  	void *addr;
> +	unsigned long nr_pages;
>  	struct vma_area *p = *pvma;
>  
>  	if (vma_entry_is(&vma->vma, VMA_FILE_PRIVATE)) {
> @@ -200,6 +201,11 @@ static int map_private_vma(pid_t pid, struct vma_area *vma, void *tgt_addr,
>  		vma->vma.shmid = 0;
>  	}
>  
> +	nr_pages = vma_entry_len(&vma->vma) / PAGE_SIZE;
> +	vma->page_bitmap = xzalloc(BITS_TO_LONGS(nr_pages) * sizeof(long));
> +	if (vma->page_bitmap == NULL)
> +		return -1;
> +
>  	list_for_each_entry_continue(p, pvma_list, list) {
>  		if (p->vma.start > vma->vma.start)
>  			 break;
> @@ -231,6 +237,8 @@ static int map_private_vma(pid_t pid, struct vma_area *vma, void *tgt_addr,
>  		}
>  		vma_premmaped_start(&(vma->vma)) = (unsigned long) addr;
>  	} else {
> +		vma->ppage_bitmap = p->page_bitmap;
> +
>  		addr = mremap((void *)vma_premmaped_start(&vma->vma),
>  				vma_area_len(vma), vma_area_len(vma),
>  				MREMAP_FIXED | MREMAP_MAYMOVE, tgt_addr);
> @@ -263,7 +271,7 @@ static int restore_priv_vma_content(pid_t pid)
>  	 * Read page contents.
>  	 */
>  	while (1) {
> -		u64 va;
> +		u64 va, page_offset;
>  		char buf[PAGE_SIZE];
>  		void *p;
>  
> @@ -283,6 +291,12 @@ static int restore_priv_vma_content(pid_t pid)
>  			vma = list_entry(vma->list.next, struct vma_area, list);
>  		}
>  
> +		page_offset = (va - vma->vma.start) / PAGE_SIZE;
> +		if (vma->page_bitmap)
> +			set_bit(page_offset, vma->page_bitmap);

Can page_bitmap be NULL?

> +		if (vma->ppage_bitmap)
> +			clear_bit(page_offset, vma->ppage_bitmap);



> +
>  		ret = read(fd, buf, PAGE_SIZE);
>  		if (ret != PAGE_SIZE) {
>  			pr_err("Can'r read mapping page %d\n", ret);
> @@ -298,6 +312,32 @@ static int restore_priv_vma_content(pid_t pid)
>  	}
>  	close(fd);
>  
> +	/* Remove pages, which were not shared with a child */
> +	list_for_each_entry(vma, &rst_vma_list, list) {
> +		unsigned long size, i = 0;
> +		void *addr = (void *) vma_premmaped_start(&vma->vma);
> +
> +		if (vma->ppage_bitmap == NULL)
> +			continue;
> +
> +		size = vma_entry_len(&vma->vma) / PAGE_SIZE;
> +		while (1) {
> +			/* Find all pages, which avaliable only for a parent */
> +			i = find_next_bit(vma->ppage_bitmap, size, i);

This code means "find all pages available for a parent", w/o only.

> +
> +			if ( i >= size)
> +				break;
> +
> +			ret = madvise(addr + PAGE_SIZE * i,
> +						PAGE_SIZE, MADV_DONTNEED);
> +			if (ret < 0) {
> +				pr_perror("madvise failed\n");
> +				return -1;
> +			}
> +			i++;
> +		}
> +	}
> +
>  	return 0;
>  }
>  
> diff --git a/include/crtools.h b/include/crtools.h
> index 37e50ca..ec76ec1 100644
> --- a/include/crtools.h
> +++ b/include/crtools.h
> @@ -215,6 +215,8 @@ struct vma_area {
>  	struct list_head	list;
>  	VmaEntry		vma;
>  	int			vm_file_fd;
> +	unsigned long		*page_bitmap;  /* existent pages */

Not used anywhere for read, this is set-only field.

> +	unsigned long		*ppage_bitmap; /* parent's existent pages */
>  };
>  
>  #define vma_area_is(vma_area, s)	vma_entry_is(&((vma_area)->vma), s)
>