Re: [PATCH] mm: Improve dump_page() for compound pages

["Kirill A. Shutemov" <kirill@shutemov.name>]

On Fri, Feb 07, 2020 at 08:44:15PM -0800, Matthew Wilcox wrote:
> From: "Matthew Wilcox (Oracle)" <willy@infradead.org>
> 
> There was no protection against a corrupted struct page having an
> implausible compound_head().  Sanity check that a compound page has
> a head within reach of the maximum allocatable page (this will need
> to be adjusted if one of the plans to allocate 1GB pages comes to
> fruition).  In addition,
> 
>  - Print the mapping pointer using %p insted of %px.  The actual value of
>    the pointer can be read out of the raw page dump and using %p gives a
>    chance to correlate it with an earlier printk of the mapping pointer
>  - Print the mapping pointer from the head page, not the tail page
>    (the tail ->mapping pointer may be in use for other purposes, eg part
>    of a list_head)
>  - Print the order of the page for compound pages
>  - Dump the raw head page as well as the raw page
>  - Print the refcount from the head page, not the tail page
> 
> Suggested-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
> Co-developed-by: John Hubbard <jhubbard@nvidia.com>
> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
> ---
>  mm/debug.c | 33 +++++++++++++++++++++++----------
>  1 file changed, 23 insertions(+), 10 deletions(-)
> 
> diff --git a/mm/debug.c b/mm/debug.c
> index ecccd9f17801..3594951cc408 100644
> --- a/mm/debug.c
> +++ b/mm/debug.c
> @@ -44,8 +44,10 @@ const struct trace_print_flags vmaflag_names[] = {
>  
>  void __dump_page(struct page *page, const char *reason)
>  {
> +	struct page *head = compound_head(page);
>  	struct address_space *mapping;
>  	bool page_poisoned = PagePoisoned(page);
> +	bool compound = PageCompound(page);
>  	/*
>  	 * Accessing the pageblock without the zone lock. It could change to
>  	 * "isolate" again in the meantime, but since we are just dumping the
> @@ -66,25 +68,32 @@ void __dump_page(struct page *page, const char *reason)
>  		goto hex_only;
>  	}
>  
> -	mapping = page_mapping(page);
> +	if (page < head || (page >= head + MAX_ORDER_NR_PAGES)) {
> +		/* Corrupt page, cannot call page_mapping */
> +		mapping = page->mapping;
> +		head = page;
> +		compound = false;
> +	} else {
> +		mapping = page_mapping(page);
> +	}
>  
>  	/*
>  	 * Avoid VM_BUG_ON() in page_mapcount().
>  	 * page->_mapcount space in struct page is used by sl[aou]b pages to
>  	 * encode own info.
>  	 */
> -	mapcount = PageSlab(page) ? 0 : page_mapcount(page);
> +	mapcount = PageSlab(head) ? 0 : page_mapcount(head);

This is wrong. We want to see mapcount for the tail page, not head.

-- 
 Kirill A. Shutemov