As mentioned in commit c07aea3ef4d4 ("mm: add a signature in struct page"): "The page->signature field is aliased to page->lru.next and page->compound_head."
And as the comment in page_is_pfmemalloc(): "lru.next has bit 1 set if the page is allocated from the pfmemalloc reserves. Callers may simply overwrite it if they do not need to preserve that information."
The page->signature is or’ed with PP_SIGNATURE when a page is allocated in page pool, see __page_pool_alloc_pages_slow(), and page->signature is checked directly with PP_SIGNATURE in page_pool_return_skb_page(), which might cause resoure leaking problem for a page from page pool if bit 1 of lru.next is set for a pfmemalloc page.
As bit 0 is page->compound_head, So mask both bit 0 and 1 before the checking in page_pool_return_skb_page().
Fixes: 6a5bcd84e886 ("page_pool: Allow drivers to hint on SKB recycling") Signed-off-by: Yunsheng Lin linyunsheng@huawei.com --- net/core/page_pool.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/net/core/page_pool.c b/net/core/page_pool.c index 5e4eb45..33b7dd7 100644 --- a/net/core/page_pool.c +++ b/net/core/page_pool.c @@ -634,7 +634,7 @@ bool page_pool_return_skb_page(struct page *page) struct page_pool *pp;
page = compound_head(page); - if (unlikely(page->pp_magic != PP_SIGNATURE)) + if (unlikely((page->pp_magic & ~0x3UL) != PP_SIGNATURE)) return false;
pp = page->pp;
On Thu, Aug 05, 2021 at 09:06:57AM +0800, Yunsheng Lin wrote:
As mentioned in commit c07aea3ef4d4 ("mm: add a signature in struct page"): "The page->signature field is aliased to page->lru.next and page->compound_head."
And as the comment in page_is_pfmemalloc(): "lru.next has bit 1 set if the page is allocated from the pfmemalloc reserves. Callers may simply overwrite it if they do not need to preserve that information."
The page->signature is or’ed with PP_SIGNATURE when a page is allocated in page pool, see __page_pool_alloc_pages_slow(), and page->signature is checked directly with PP_SIGNATURE in page_pool_return_skb_page(), which might cause resoure leaking problem for a page from page pool if bit 1 of lru.next is set for a pfmemalloc page.
As bit 0 is page->compound_head, So mask both bit 0 and 1 before the checking in page_pool_return_skb_page().
No, you don't understand. We *want* the check to fail if we were low on memory so we return the emergency allocation.
On 2021/8/5 9:50, Matthew Wilcox wrote:
On Thu, Aug 05, 2021 at 09:06:57AM +0800, Yunsheng Lin wrote:
As mentioned in commit c07aea3ef4d4 ("mm: add a signature in struct page"): "The page->signature field is aliased to page->lru.next and page->compound_head."
And as the comment in page_is_pfmemalloc(): "lru.next has bit 1 set if the page is allocated from the pfmemalloc reserves. Callers may simply overwrite it if they do not need to preserve that information."
The page->signature is or’ed with PP_SIGNATURE when a page is allocated in page pool, see __page_pool_alloc_pages_slow(), and page->signature is checked directly with PP_SIGNATURE in page_pool_return_skb_page(), which might cause resoure leaking problem for a page from page pool if bit 1 of lru.next is set for a pfmemalloc page.
As bit 0 is page->compound_head, So mask both bit 0 and 1 before the checking in page_pool_return_skb_page().
No, you don't understand. We *want* the check to fail if we were low on memory so we return the emergency allocation.
If the check failed, but the page pool assume the page is not from page pool and will not do the resource cleaning(like dma unmapping), as the page pool still use the page with pfmemalloc set and dma map the page if pp_flags & PP_FLAG_DMA_MAP is true in __page_pool_alloc_pages_slow().
The returning the emergency allocation you mentioned seems to be handled in __page_pool_put_page(), see:
https://elixir.bootlin.com/linux/latest/source/net/core/page_pool.c#L411
We just use the page with pfmemalloc one time and do the resource cleaning before returning the page back to page allocator. Or did I miss something here?
.
On Thu, Aug 05, 2021 at 10:14:39AM +0800, Yunsheng Lin wrote:
On 2021/8/5 9:50, Matthew Wilcox wrote:
On Thu, Aug 05, 2021 at 09:06:57AM +0800, Yunsheng Lin wrote:
As mentioned in commit c07aea3ef4d4 ("mm: add a signature in struct page"): "The page->signature field is aliased to page->lru.next and page->compound_head."
And as the comment in page_is_pfmemalloc(): "lru.next has bit 1 set if the page is allocated from the pfmemalloc reserves. Callers may simply overwrite it if they do not need to preserve that information."
The page->signature is or???ed with PP_SIGNATURE when a page is allocated in page pool, see __page_pool_alloc_pages_slow(), and page->signature is checked directly with PP_SIGNATURE in page_pool_return_skb_page(), which might cause resoure leaking problem for a page from page pool if bit 1 of lru.next is set for a pfmemalloc page.
As bit 0 is page->compound_head, So mask both bit 0 and 1 before the checking in page_pool_return_skb_page().
No, you don't understand. We *want* the check to fail if we were low on memory so we return the emergency allocation.
If the check failed, but the page pool assume the page is not from page pool and will not do the resource cleaning(like dma unmapping), as the page pool still use the page with pfmemalloc set and dma map the page if pp_flags & PP_FLAG_DMA_MAP is true in __page_pool_alloc_pages_slow().
The returning the emergency allocation you mentioned seems to be handled in __page_pool_put_page(), see:
https://elixir.bootlin.com/linux/latest/source/net/core/page_pool.c#L411
We just use the page with pfmemalloc one time and do the resource cleaning before returning the page back to page allocator. Or did I miss something here?
.
I think you are right here. What happens is that the original pp->signature is OR'ed after the allocation in order to preserve any existing bits. When those are present though the if which will trigger the recycling will fail and those DMA mapping will be left stale.
If we mask the bits during the check (as your patch does), we'll end up not recycling the page anyway since it has the pfmemalloc bit set. The page pool recycle function will end up releasing the page and the DMA mappings right?
Regards /Ilias
On 2021/8/5 16:54, Ilias Apalodimas wrote:
On Thu, Aug 05, 2021 at 10:14:39AM +0800, Yunsheng Lin wrote:
On 2021/8/5 9:50, Matthew Wilcox wrote:
On Thu, Aug 05, 2021 at 09:06:57AM +0800, Yunsheng Lin wrote:
As mentioned in commit c07aea3ef4d4 ("mm: add a signature in struct page"): "The page->signature field is aliased to page->lru.next and page->compound_head."
And as the comment in page_is_pfmemalloc(): "lru.next has bit 1 set if the page is allocated from the pfmemalloc reserves. Callers may simply overwrite it if they do not need to preserve that information."
The page->signature is or???ed with PP_SIGNATURE when a page is allocated in page pool, see __page_pool_alloc_pages_slow(), and page->signature is checked directly with PP_SIGNATURE in page_pool_return_skb_page(), which might cause resoure leaking problem for a page from page pool if bit 1 of lru.next is set for a pfmemalloc page.
As bit 0 is page->compound_head, So mask both bit 0 and 1 before the checking in page_pool_return_skb_page().
No, you don't understand. We *want* the check to fail if we were low on memory so we return the emergency allocation.
If the check failed, but the page pool assume the page is not from page pool and will not do the resource cleaning(like dma unmapping), as the page pool still use the page with pfmemalloc set and dma map the page if pp_flags & PP_FLAG_DMA_MAP is true in __page_pool_alloc_pages_slow().
The returning the emergency allocation you mentioned seems to be handled in __page_pool_put_page(), see:
https://elixir.bootlin.com/linux/latest/source/net/core/page_pool.c#L411
We just use the page with pfmemalloc one time and do the resource cleaning before returning the page back to page allocator. Or did I miss something here?
.
I think you are right here. What happens is that the original pp->signature is OR'ed after the allocation in order to preserve any existing bits. When those are present though the if which will trigger the recycling will fail and those DMA mapping will be left stale.
If we mask the bits during the check (as your patch does), we'll end up not recycling the page anyway since it has the pfmemalloc bit set. The page pool recycle function will end up releasing the page and the DMA mappings right?
Yes. The problem might be magnified when frag page in page pool is added, because page pool only hold one ref of the page, and page_pool_return_skb_page() might dec the page ref twice if the frag page has two users, supposing the above checking fail with the pfmemalloc page, leaving to the below log:
[ 49.584990] BUG: Bad page state in process iperf pfn:20af242 [ 49.584992] page:(____ptrval____) refcount:-1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x20af242
Regards /Ilias .
Right, mind sending a v2 with a comment explaining why we need to mask?
Other than that
Reviewed-by: Ilias Apalodimas ilias.apalodimas@linaro.org
On Thu, 5 Aug 2021 at 12:31, Yunsheng Lin linyunsheng@huawei.com wrote:
On 2021/8/5 16:54, Ilias Apalodimas wrote:
On Thu, Aug 05, 2021 at 10:14:39AM +0800, Yunsheng Lin wrote:
On 2021/8/5 9:50, Matthew Wilcox wrote:
On Thu, Aug 05, 2021 at 09:06:57AM +0800, Yunsheng Lin wrote:
As mentioned in commit c07aea3ef4d4 ("mm: add a signature in struct page"): "The page->signature field is aliased to page->lru.next and page->compound_head."
And as the comment in page_is_pfmemalloc(): "lru.next has bit 1 set if the page is allocated from the pfmemalloc reserves. Callers may simply overwrite it if they do not need to preserve that information."
The page->signature is or???ed with PP_SIGNATURE when a page is allocated in page pool, see __page_pool_alloc_pages_slow(), and page->signature is checked directly with PP_SIGNATURE in page_pool_return_skb_page(), which might cause resoure leaking problem for a page from page pool if bit 1 of lru.next is set for a pfmemalloc page.
As bit 0 is page->compound_head, So mask both bit 0 and 1 before the checking in page_pool_return_skb_page().
No, you don't understand. We *want* the check to fail if we were low on memory so we return the emergency allocation.
If the check failed, but the page pool assume the page is not from page pool and will not do the resource cleaning(like dma unmapping), as the page pool still use the page with pfmemalloc set and dma map the page if pp_flags & PP_FLAG_DMA_MAP is true in __page_pool_alloc_pages_slow().
The returning the emergency allocation you mentioned seems to be handled in __page_pool_put_page(), see:
https://elixir.bootlin.com/linux/latest/source/net/core/page_pool.c#L411
We just use the page with pfmemalloc one time and do the resource cleaning before returning the page back to page allocator. Or did I miss something here?
.
I think you are right here. What happens is that the original pp->signature is OR'ed after the allocation in order to preserve any existing bits. When those are present though the if which will trigger the recycling will fail and those DMA mapping will be left stale.
If we mask the bits during the check (as your patch does), we'll end up not recycling the page anyway since it has the pfmemalloc bit set. The page pool recycle function will end up releasing the page and the DMA mappings right?
Yes. The problem might be magnified when frag page in page pool is added, because page pool only hold one ref of the page, and page_pool_return_skb_page() might dec the page ref twice if the frag page has two users, supposing the above checking fail with the pfmemalloc page, leaving to the below log:
[ 49.584990] BUG: Bad page state in process iperf pfn:20af242 [ 49.584992] page:(____ptrval____) refcount:-1 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x20af242
Regards /Ilias .
-
Ilias Apalodimas -
Matthew Wilcox -
Yunsheng Lin