diff mbox series

[2/2] mm: introduce __GFP_TRACKLEAK to track in-kernel allocation

Message ID 1663126621-26926-2-git-send-email-zhaoyang.huang@unisoc.com (mailing list archive)
State New
Headers show
Series [1/2] mm: fix logic error of page_expected_state | expand

Commit Message

zhaoyang.huang Sept. 14, 2022, 3:37 a.m. UTC
From: Zhaoyang Huang <zhaoyang.huang@unisoc.com>

Kthread and drivers could fetch memory via alloc_pages directly which make them
hard to debug when leaking. Solve this by introducing __GFP_TRACELEAK and reuse
kmemleak mechanism which unified most of kernel cosuming pages into kmemleak.

This patch has been tested with alloc_pages(__GFP_TRACKLEAK) & (__GFP_TRACKLEAK|__GFP_COMP)
and got proved as effective.

unreferenced object 0xffffff807c620000 (size 65536):
comm "allocator@4.0-s", pid 745, jiffies 4294906308 (age 5136.616s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
backtrace:
[<00000000ffefbfdf>] __alloc_pages_nodemask+0x108/0x3a4
[<0000000083595277>] ion_page_pool_alloc+0x178/0x234
[<000000008267995a>] ion_system_heap_allocate+0x13c/0x708
[<00000000d4df5a5e>] ion_buffer_create+0x98/0x67c
[<0000000043fa6683>] ion_dmabuf_alloc+0xcc/0x1c0
[<000000000d1db17e>] ion_ioctl+0x150/0x350
[<00000000a2b89048>] do_vfs_ioctl+0x5d4/0xa94
[<000000008e9b61d3>] __arm64_sys_ioctl+0x14c/0x164
[<00000000114425a9>] el0_svc_common+0xd0/0x23c
[<00000000ec9cb1b1>] el0_svc_handler+0x2c/0x3c
[<00000000e44a2c21>] el0_svc+0x8/0x100
unreferenced object 0xffffff807c189000 (size 4096):
comm "allocator@4.0-s", pid 745, jiffies 4294906309 (age 5136.612s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
backtrace:
[<00000000ffefbfdf>] __alloc_pages_nodemask+0x108/0x3a4
[<0000000083595277>] ion_page_pool_alloc+0x178/0x234
[<00000000b30c4562>] ion_system_heap_allocate+0x160/0x708
[<00000000d4df5a5e>] ion_buffer_create+0x98/0x67c
[<0000000043fa6683>] ion_dmabuf_alloc+0xcc/0x1c0
[<000000000d1db17e>] ion_ioctl+0x150/0x350
[<00000000a2b89048>] do_vfs_ioctl+0x5d4/0xa94
[<000000008e9b61d3>] __arm64_sys_ioctl+0x14c/0x164
[<00000000114425a9>] el0_svc_common+0xd0/0x23c
[<00000000ec9cb1b1>] el0_svc_handler+0x2c/0x3c
[<00000000e44a2c21>] el0_svc+0x8/0x100

Signed-off-by: Zhaoyang Huang <zhaoyang.huang@unisoc.com>
---
v2: code update
v3: update code and Documentation
---
---
 Documentation/dev-tools/kmemleak.rst |  5 ++++-
 include/linux/gfp.h                  |  8 +++++++-
 include/linux/page-flags.h           |  3 +++
 mm/page_alloc.c                      | 14 ++++++++++++++
 4 files changed, 28 insertions(+), 2 deletions(-)

Comments

Matthew Wilcox Sept. 14, 2022, 8:26 a.m. UTC | #1
On Wed, Sep 14, 2022 at 11:37:01AM +0800, zhaoyang.huang wrote:
> ---
> v2: code update
> v3: update code and Documentation

This is really not good enough.  What changed?

The documentation is also not good enough.  It needs to mention:

 - This cannot be used for GFP_HIGHMEM allocations.
 - This cannot be used for pages which are mapped into userspace.

I also still want to see selftests.  order-0, order-N (with and without
__GFP_COMP).  What happens if you allocate an order-N page without
GFP_COMP, take an extra ref on the first page, call free_pages() and
then one of the recently-freed pages is allocated again while you still
have the reference on the first page?

I believe Andrew also suggested that 

if (PageTrackleak(page))

become always-false if the CONFIG option is disabled.

> +#ifdef CONFIG_HAVE_DEBUG_KMEMLEAK

This is the wrong CONFIG option, it should be CONFIG_DEBUG_KMEMLEAK.


Add to this the very real question of how useful is this, and I'm not
getting warm fuzzy feelings about where this patchset is heading.
diff mbox series

Patch

diff --git a/Documentation/dev-tools/kmemleak.rst b/Documentation/dev-tools/kmemleak.rst
index 1c935f4..b1128fe 100644
--- a/Documentation/dev-tools/kmemleak.rst
+++ b/Documentation/dev-tools/kmemleak.rst
@@ -81,7 +81,7 @@  Basic Algorithm
 ---------------
 
 The memory allocations via :c:func:`kmalloc`, :c:func:`vmalloc`,
-:c:func:`kmem_cache_alloc` and
+:c:func:`kmem_cache_alloc`, :c:func:`alloc_pages(__GFP_TRACKLEAK)` (1)and
 friends are traced and the pointers, together with additional
 information like size and stack trace, are stored in a rbtree.
 The corresponding freeing function calls are tracked and the pointers
@@ -257,3 +257,6 @@  memory leaks``. Then read the file to see then::
 
 Removing the module with ``rmmod kmemleak_test`` should also trigger some
 kmemleak results.
+
+(1)Don't use __GFP_TRACKLEAK when getting pages for vm_iomap_memory which map
+physical address from kernel to userspace.
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index 2d2ccae..53464c6 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -68,6 +68,11 @@ 
 #else
 #define ___GFP_NOLOCKDEP	0
 #endif
+#ifdef CONFIG_HAVE_DEBUG_KMEMLEAK
+#define ___GFP_TRACKLEAK	0x10000000u
+#else
+#define ___GFP_TRACKLEAK	0
+#endif
 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
 
 /*
@@ -259,12 +264,13 @@ 
 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
 #define __GFP_SKIP_KASAN_UNPOISON ((__force gfp_t)___GFP_SKIP_KASAN_UNPOISON)
 #define __GFP_SKIP_KASAN_POISON   ((__force gfp_t)___GFP_SKIP_KASAN_POISON)
+#define __GFP_TRACKLEAK   ((__force gfp_t)___GFP_TRACKLEAK)
 
 /* Disable lockdep for GFP context tracking */
 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
 
 /* Room for N __GFP_FOO bits */
-#define __GFP_BITS_SHIFT (27 + IS_ENABLED(CONFIG_LOCKDEP))
+#define __GFP_BITS_SHIFT (27 + IS_ENABLED(CONFIG_LOCKDEP) + IS_ENABLED(CONFIG_HAVE_DEBUG_KMEMLEAK))
 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
 
 /**
diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index 5d3274b..1374e29 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -942,6 +942,7 @@  static inline bool is_page_hwpoison(struct page *page)
 #define PG_offline	0x00000100
 #define PG_table	0x00000200
 #define PG_guard	0x00000400
+#define PG_trackleak	0x00000800
 
 #define PageType(page, flag)						\
 	((page->page_type & (PAGE_TYPE_BASE | flag)) == PAGE_TYPE_BASE)
@@ -1012,6 +1013,8 @@  static inline bool page_has_type(struct page *page)
  */
 PAGE_TYPE_OPS(Guard, guard)
 
+PAGE_TYPE_OPS(Trackleak, trackleak)
+
 extern bool is_free_buddy_page(struct page *page);
 
 PAGEFLAG(Isolated, isolated, PF_ANY);
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 3714680..9e036f1 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1357,6 +1357,10 @@  static __always_inline bool free_pages_prepare(struct page *page,
 			(page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
 		}
 	}
+	if (PageTrackleak(page)) {
+		__ClearPageTrackleak(page);
+		kmemleak_free(page_address(page));
+	}
 	if (PageMappingFlags(page))
 		page->mapping = NULL;
 	if (memcg_kmem_enabled() && PageMemcgKmem(page))
@@ -1521,6 +1525,11 @@  static void free_pcppages_bulk(struct zone *zone, int count,
 			if (unlikely(isolated_pageblocks))
 				mt = get_pageblock_migratetype(page);
 
+			if (PageTrackleak(page)) {
+				__ClearPageTrackleak(page);
+				kmemleak_free(page_address(page));
+			}
+
 			__free_one_page(page, page_to_pfn(page), zone, order, mt, FPI_NONE);
 			trace_mm_page_pcpu_drain(page, order, mt);
 		} while (count > 0 && !list_empty(list));
@@ -2468,6 +2477,11 @@  static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags
 		set_page_pfmemalloc(page);
 	else
 		clear_page_pfmemalloc(page);
+
+	if (gfp_flags & __GFP_TRACKLEAK) {
+		kmemleak_alloc(page_address(page), PAGE_SIZE << order, 1, gfp_flags & ~__GFP_TRACKLEAK);
+		__SetPageTrackleak(page);
+	}
 }
 
 /*