On Fri, 19 Mar 2021, Johannes Weiner wrote:
> When the freeing of a higher-order page block (non-compound) races
> with a speculative page cache lookup, __free_pages() needs to leave
> the first order-0 page in the chunk to the lookup but free the buddy
> pages that the lookup doesn't know about separately.
>
> However, if such a higher-order page is charged to a memcg (e.g. !vmap
> kernel stack)), only the first page of the block has page->memcg
> set. That means we'll uncharge only one order-0 page from the entire
> block, and leak the remainder.
>
> Add a split_page_memcg() to __free_pages() right before it starts
> taking the higher-order page apart and freeing its individual
> constituent pages. This ensures all of them will have the memcg
> linkage set up for correct uncharging. Also update the comments a bit
> to clarify what exactly is happening to the page during that race.
>
> This bug is old and has its roots in the speculative page cache patch
> and adding cgroup accounting of kernel pages. There are no known user
> reports. A backport to stable is therefor not warranted.
>
> Reported-by: Matthew Wilcox <willy@xxxxxxxxxxxxx>
> Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx>
Acked-by: Hugh Dickins <hughd@xxxxxxxxxx>
to the split_page_memcg() addition etc, but a doubt just hit me on the
original e320d3012d25 ("mm/page_alloc.c: fix freeing non-compound pages"):
see comment below.
> ---
> mm/page_alloc.c | 33 +++++++++++++++++++++++++++------
> 1 file changed, 27 insertions(+), 6 deletions(-)
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index c53fe4fa10bf..f4bd56656402 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5112,10 +5112,9 @@ static inline void free_the_page(struct page *page, unsigned int order)
> * the allocation, so it is easy to leak memory. Freeing more memory
> * than was allocated will probably emit a warning.
> *
> - * If the last reference to this page is speculative, it will be released
> - * by put_page() which only frees the first page of a non-compound
> - * allocation. To prevent the remaining pages from being leaked, we free
> - * the subsequent pages here. If you want to use the page's reference
> + * This function isn't a put_page(). Don't let the put_page_testzero()
> + * fool you, it's only to deal with speculative cache references. It
> + * WILL free pages directly. If you want to use the page's reference
> * count to decide when to free the allocation, you should allocate a
> * compound page, and use put_page() instead of __free_pages().
> *
> @@ -5124,11 +5123,33 @@ static inline void free_the_page(struct page *page, unsigned int order)
> */
> void __free_pages(struct page *page, unsigned int order)
> {
> - if (put_page_testzero(page))
> + /*
> + * Drop the base reference from __alloc_pages and free. In
> + * case there is an outstanding speculative reference, from
> + * e.g. the page cache, it will put and free the page later.
> + */
> + if (likely(put_page_testzero(page))) {
> free_the_page(page, order);
> - else if (!PageHead(page))
> + return;
> + }
> +
> + /*
> + * The speculative reference will put and free the page.
> + *
> + * However, if the speculation was into a higher-order page
> + * chunk that isn't marked compound, the other side will know
> + * nothing about our buddy pages and only free the order-0
> + * page at the start of our chunk! We must split off and free
> + * the buddy pages here.
> + *
> + * The buddy pages aren't individually refcounted, so they
> + * can't have any pending speculative references themselves.
> + */
> + if (!PageHead(page) && order > 0) {
The put_page_testzero() has released our reference to the first
subpage of page: it's now under the control of the racing speculative
lookup. So it seems to me unsafe to be checking PageHead(page) here:
if it was actually a compound page, PageHead might already be cleared
by now, and we doubly free its tail pages below? I think we need to
use a "bool compound = PageHead(page)" on entry to __free_pages().
Or alternatively, it's wrong to call __free_pages() on a compound
page anyway, so we should not check PageHead at all, except in a
WARN_ON_ONCE(PageCompound(page)) at the start?
And would it be wrong to fix that too in this patch?
Though it ought then to be backported to 5.10 stable.
> + split_page_memcg(page, 1 << order);
> while (order-- > 0)
> free_the_page(page + (1 << order), order);
> + }
> }
> EXPORT_SYMBOL(__free_pages);
>
> --
> 2.30.1
