mm: make page ref count overflow check tighter and more explicit
authorLinus Torvalds <torvalds@linux-foundation.org>
Thu, 11 Apr 2019 17:06:20 +0000 (10:06 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 11 Jun 2019 10:22:45 +0000 (12:22 +0200)
commit f958d7b528b1b40c44cfda5eabe2d82760d868c3 upstream.

We have a VM_BUG_ON() to check that the page reference count doesn't
underflow (or get close to overflow) by checking the sign of the count.

That's all fine, but we actually want to allow people to use a "get page
ref unless it's already very high" helper function, and we want that one
to use the sign of the page ref (without triggering this VM_BUG_ON).

Change the VM_BUG_ON to only check for small underflows (or _very_ close
to overflowing), and ignore overflows which have strayed into negative
territory.

Acked-by: Matthew Wilcox <willy@infradead.org>
Cc: Jann Horn <jannh@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Ben Hutchings <ben.hutchings@codethink.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/linux/mm.h

index e3c8d40a18b5b517fc10cd01cc28e1571f9587d6..478466081265d181d4b25a93d4e8faec7cc74605 100644 (file)
@@ -763,6 +763,10 @@ static inline bool is_zone_device_page(const struct page *page)
 }
 #endif
 
+/* 127: arbitrary random number, small enough to assemble well */
+#define page_ref_zero_or_close_to_overflow(page) \
+       ((unsigned int) page_ref_count(page) + 127u <= 127u)
+
 static inline void get_page(struct page *page)
 {
        page = compound_head(page);
@@ -770,7 +774,7 @@ static inline void get_page(struct page *page)
         * Getting a normal page or the head of a compound page
         * requires to already have an elevated page->_refcount.
         */
-       VM_BUG_ON_PAGE(page_ref_count(page) <= 0, page);
+       VM_BUG_ON_PAGE(page_ref_zero_or_close_to_overflow(page), page);
        page_ref_inc(page);
 
        if (unlikely(is_zone_device_page(page)))