Skip to contentSkip to conversation

mm: kfence: remove kfence_pool from leak trace

PATCH In review [email protected] 2 participants 7 messages Jun 23, 2022 – Jun 28, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover mm: kfence: remove kfence_pool from leak trace

From: Yee Lee <[email protected]>

Since the latest patches of kmemleak separated its address management to two, phys and virt, rb-trees, the kmemleak_free failed to free kfence_pool in the virt tree. It caused unexpected scan the blocks and triggered kfence faults.

Reference:
https://lore.kernel.org/linux-mm/[email protected]/

==================================================================

 BUG: KFENCE: invalid read in scan_block+0x13c/0x838

Invalid read at 0x000000003dc26873:

  scan_block+0x13c/0x838
  scan_gray_list+0x1f0/0x478
  kmemleak_scan+0x63c/0xd64
  kmemleak_write+0x618/0x8cc
  full_proxy_write+0x70/0x138
  vfs_write+0x108/0x314
  ksys_write+0x7c/0x14c
  __arm64_sys_write+0x20/0x30
  el0_svc_common+0xa8/0x144
  do_el0_svc+0x30/0xd4
  el0_svc+0x38/0x15c
  el0t_64_sync_handler+0x88/0xf8
  el0t_64_sync+0x1a0/0x1a4

 CPU: 0 PID: 128 Comm: sh Not tainted 5.18.0-mainline-40996-g7d83a175ff4a-dirty #1
 Hardware name: linux,dummy-virt (DT)
 ==================================================================

This patch change to call the memblock alloc with the flag to skip leak tracking. And the latter kmemleak_free can be removed too.

Yee Lee (1):
mm: kfence: skip kmemleak alloc in kfence_pool

 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
Signature · 1 line
2.18.0
#1 lore raw

PATCH 1/1 mm: kfence: skip kmemleak alloc in kfence_pool

From: Yee Lee <[email protected]>

Use MEMBLOCK_ALLOC_NOLEAKTRACE to skip kmemleak registration when the kfence pool is allocated from memblock. And the kmemleak_free later can be removed too.

---
 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
mm/kfence/core.c +8 -10
index 4e7cd4c8e687..0d33d83f5244 100644
@@ -600,14 +600,6 @@ static unsigned long kfence_init_pool(void)
 		addr += 2 * PAGE_SIZE;
 	}
 
-	/*
-	 * The pool is live and will never be deallocated from this point on.
-	 * Remove the pool object from the kmemleak object tree, as it would
-	 * otherwise overlap with allocations returned by kfence_alloc(), which
-	 * are registered with kmemleak through the slab post-alloc hook.
-	 */
-	kmemleak_free(__kfence_pool);
-
 	return 0;
 }
 
@@ -831,8 +823,14 @@ void __init kfence_alloc_pool(void)
 {
 	if (!kfence_sample_interval)
 		return;
-
-	__kfence_pool = memblock_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
+	/*
+	 * The pool is live and will never be deallocated from this point on.
+	 * Skip the pool object from the kmemleak object allocation, as it would
+	 * otherwise overlap with allocations returned by kfence_alloc(), which
+	 * are registered with kmemleak through the slab post-alloc hook.
+	 */
+	__kfence_pool = memblock_alloc_try_nid(KFENCE_POOL_SIZE, PAGE_SIZE,
+		 MEMBLOCK_LOW_LIMIT, MEMBLOCK_ALLOC_NOLEAKTRACE, NUMA_NO_NODE);
 
 	if (!__kfence_pool)
 		pr_err("failed to allocate pool\n");
Signature · 1 line
2.18.0
maintainer
#2 lore raw

Re: [PATCH 1/1] mm: kfence: skip kmemleak alloc in kfence_pool

Quoted 6 lines · yee.lee via kasan-dev

From: Yee Lee <[email protected]>

Use MEMBLOCK_ALLOC_NOLEAKTRACE to skip kmemleak registration when the kfence pool is allocated from memblock. And the kmemleak_free later can be removed too.

Is this purely meant to be a cleanup and non-functional change?

Quoted 11 lines
---
 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
mm/kfence/core.c +0 -0
index 4e7cd4c8e687..0d33d83f5244 100644

mm/kfence/core.c static unsigned long kfence_init_pool(void) · line 609

Quoted 10 lines
mm/kfence/core.c +0 -7
@@ -600,14 +600,6 @@ static unsigned long kfence_init_pool(void)
                addr += 2 * PAGE_SIZE;
        }
 
-       /*
-        * The pool is live and will never be deallocated from this point on.
-        * Remove the pool object from the kmemleak object tree, as it would
-        * otherwise overlap with allocations returned by kfence_alloc(), which
-        * are registered with kmemleak through the slab post-alloc hook.
-        */
-       kmemleak_free(__kfence_pool);

This appears to only be a non-functional change if the pool is allocated early. If the pool is allocated late using page-alloc, then there'll not be a kmemleak_free() on that memory and we'll have the same problem.

author
#3 lore raw

Re: [PATCH 1/1] mm: kfence: skip kmemleak alloc in kfence_pool

Quoted 25 lines · Marco Elver
Quoted 6 lines · yee.lee via kasan-dev

From: Yee Lee <[email protected]>

Use MEMBLOCK_ALLOC_NOLEAKTRACE to skip kmemleak registration when the kfence pool is allocated from memblock. And the kmemleak_free later can be removed too.

Is this purely meant to be a cleanup and non-functional change?

Quoted 14 lines
---
 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
mm/kfence/core.c +0 -0
index 4e7cd4c8e687..0d33d83f5244 100644
@@ -600,14 +600,6 @@ static unsigned long kfence_init_pool(void)
                addr += 2 * PAGE_SIZE;
        }
 
Quoted 16 lines

mm/kfence/core.c static unsigned long kfence_init_pool(void) · line 609

Quoted 11 lines
mm/kfence/core.c +0 -2
-       /*
-        * The pool is live and will never be deallocated from this

point on.
- * Remove the pool object from the kmemleak object tree, as
it would
- * otherwise overlap with allocations returned by
kfence_alloc(), which
- * are registered with kmemleak through the slab post-alloc
hook.

mm/kfence/core.c +0 -2
-        */
-       kmemleak_free(__kfence_pool);

This appears to only be a non-functional change if the pool is allocated early. If the pool is allocated late using page-alloc, then there'll not be a kmemleak_free() on that memory and we'll have the same problem.

Do you mean the kzalloc(slab_is_available) in memblock_allc()? That implies that MEMBLOCK_ALLOC_NOLEAKTRACE has no guarantee skipping kmemleak_alloc from this. (Maybe add it?)

If so, we cannot identify later the block is stored in the phys tree(memblock) or the virt tree(page_alloc).

maintainer
#4 lore raw

Re: [PATCH 1/1] mm: kfence: skip kmemleak alloc in kfence_pool

Quoted 44 lines · Yee Lee' via kasan-dev
Quoted 25 lines · Marco Elver
Quoted 6 lines · yee.lee via kasan-dev

From: Yee Lee <[email protected]>

Use MEMBLOCK_ALLOC_NOLEAKTRACE to skip kmemleak registration when the kfence pool is allocated from memblock. And the kmemleak_free later can be removed too.

Is this purely meant to be a cleanup and non-functional change?

Quoted 14 lines
---
 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
mm/kfence/core.c +0 -0
index 4e7cd4c8e687..0d33d83f5244 100644
@@ -600,14 +600,6 @@ static unsigned long kfence_init_pool(void)
                addr += 2 * PAGE_SIZE;
        }
 
Quoted 16 lines

mm/kfence/core.c static unsigned long kfence_init_pool(void) · line 609

Quoted 11 lines
mm/kfence/core.c +0 -2
-       /*
-        * The pool is live and will never be deallocated from this

point on.
- * Remove the pool object from the kmemleak object tree, as
it would
- * otherwise overlap with allocations returned by
kfence_alloc(), which
- * are registered with kmemleak through the slab post-alloc
hook.

mm/kfence/core.c +0 -2
-        */
-       kmemleak_free(__kfence_pool);

This appears to only be a non-functional change if the pool is allocated early. If the pool is allocated late using page-alloc, then there'll not be a kmemleak_free() on that memory and we'll have the same problem.

Quoted 3 lines

Do you mean the kzalloc(slab_is_available) in memblock_allc()? That implies that MEMBLOCK_ALLOC_NOLEAKTRACE has no guarantee skipping kmemleak_alloc from this. (Maybe add it?)

No, if KFENCE is initialized through kfence_init_late() -> kfence_init_pool_late() -> kfence_init_pool().

author
#5 lore raw

Re: [PATCH 1/1] mm: kfence: skip kmemleak alloc in kfence_pool

Quoted 54 lines · Marco Elver
Quoted 52 lines · Yee Lee' via kasan-dev
Quoted 49 lines · Marco Elver
Quoted 8 lines · yee.lee via kasan-dev

From: Yee Lee <[email protected]>

Use MEMBLOCK_ALLOC_NOLEAKTRACE to skip kmemleak registration when
the kfence pool is allocated from memblock. And the kmemleak_free
later can be removed too.

Is this purely meant to be a cleanup and non-functional change?

Quoted 27 lines
---
 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
mm/kfence/core.c +0 -2
index 4e7cd4c8e687..0d33d83f5244 100644
@@ -600,14 +600,6 @@ static unsigned long
 kfence_init_pool(void)
                addr += 2 * PAGE_SIZE;
        }
 
-       /*
-        * The pool is live and will never be deallocated from

this
point on.
- * Remove the pool object from the kmemleak object
tree, as
it would
- * otherwise overlap with allocations returned by
kfence_alloc(), which
- * are registered with kmemleak through the slab post-
alloc
hook.

mm/kfence/core.c

Quoted 2 lines
mm/kfence/core.c +0 -2
-        */
-       kmemleak_free(__kfence_pool);

This appears to only be a non-functional change if the pool is allocated early. If the pool is allocated late using page-alloc, then
there'll not be a kmemleak_free() on that memory and we'll have the
same problem.

Quoted 6 lines
Quoted 3 lines

Do you mean the kzalloc(slab_is_available) in memblock_allc()? That implies that MEMBLOCK_ALLOC_NOLEAKTRACE has no guarantee skipping kmemleak_alloc from this. (Maybe add it?)

No, if KFENCE is initialized through kfence_init_late() -> kfence_init_pool_late() -> kfence_init_pool().

Thanks for the information.

But as I known, page-alloc does not request kmemleak areas. So the current kfence_pool_init_late() would cause another kmemleak warning on unknown freeing.

Reproducing test: (kfence late enable + kmemleak debug on)

/ # echo 500 > /sys/module/kfence/parameters/sample_interval

[  153.433518] kmemleak: Freeing unknown object at 0xffff0000c0600000
[  153.433804] CPU: 0 PID: 100 Comm: sh Not tainted 5.19.0-rc3-74069-
gde5c208d533a-dirty #1
[  153.434027] Hardware name: linux,dummy-virt (DT)
[  153.434265] Call trace:
[  153.434331]  dump_backtrace+0xdc/0xfc
[  153.434962]  show_stack+0x18/0x24
[  153.435106]  dump_stack_lvl+0x64/0x7c
[  153.435232]  dump_stack+0x18/0x38
[  153.435347]  kmemleak_free+0x184/0x1c8
[  153.435462]  kfence_init_pool+0x16c/0x194
[  153.435587]  param_set_sample_interval+0xe0/0x1c4
[  153.435694]  param_attr_store+0x98/0xf4
[  153.435804]  module_attr_store+0x24/0x3c
[  153.435910]  sysfs_kf_write+0x3c/0x50
...(skip)
[  153.444496] kfence: initialized - using 524288 bytes for 63 objects

at 0x00000000a3236b01-0x00000000901655d3
/ #

Hence, now there are two issues to solve.
(1) (The original)To prevent the undesired kmemleak scanning on the kfence pool. As Cataline's suggestion, we can just apply kmemleak_ignore_phys instead of free it at all.
ref: https://lore.kernel.org/linux-mm/[email protected]/

(2) The late-allocated kfence pool doesn't need to go through kmemleak_free. We can relocate the opeartion to kfence_init_pool_early() to seperate them.
That is, kfence_init_pool_early(memblock) has it and kfence_init_pool_late(page alloc) does not.

The draft is like the following.

mm/kfence/core.c +0 -2
index 11a954763be9..a52db7f06c04 100644
@@ -591,14 +591,6 @@ static unsigned long kfence_init_pool(void)
                addr += 2 * PAGE_SIZE;
        }
 
-       /*
-        * The pool is live and will never be deallocated from this

point on.
- * Remove the pool object from the kmemleak object tree, as it
would
- * otherwise overlap with allocations returned by
kfence_alloc(), which
- * are registered with kmemleak through the slab post-alloc
hook.
- */
- kmemleak_free(__kfence_pool);
-

        return 0;

}

@@ -611,8 +603,16 @@ static bool __init kfence_init_pool_early(void)

        addr = kfence_init_pool();

- if (!addr)
+ if (!addr) {
+ /*
+ * The pool is live and will never be deallocated from
this point on.
+ * Ignore the pool object from the kmemleak phys object
tree, as it would
+ * otherwise overlap with allocations returned by
kfence_alloc(), which
+ * are registered with kmemleak through the slab post-
alloc hook.
+ */

+               kmemleak_ignore_phys(__pa(__kfence_pool));
                return true;

+ }

/*
* Only release unprotected pages, and do not try to go back
and change

maintainer
#6 lore raw

Re: [PATCH 1/1] mm: kfence: skip kmemleak alloc in kfence_pool

Quoted 89 lines · Yee Lee

On Fri, 2022-06-24 at 10:28 +0200, Marco Elver wrote:

On Fri, 24 Jun 2022 at 10:20, 'Yee Lee' via kasan-dev <[email protected]> wrote:

On Thu, 2022-06-23 at 13:59 +0200, Marco Elver wrote:

On Thu, 23 Jun 2022 at 13:20, yee.lee via kasan-dev <[email protected]> wrote:

From: Yee Lee <[email protected]>

Use MEMBLOCK_ALLOC_NOLEAKTRACE to skip kmemleak registration when the kfence pool is allocated from memblock. And the kmemleak_free later can be removed too.

Is this purely meant to be a cleanup and non-functional change?

---
 mm/kfence/core.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
mm/kfence/core.c +0 -2
index 4e7cd4c8e687..0d33d83f5244 100644
@@ -600,14 +600,6 @@ static unsigned long kfence_init_pool(void)
                addr += 2 * PAGE_SIZE;
        }
 
-       /*
-        * The pool is live and will never be deallocated from this

point on.
- * Remove the pool object from the kmemleak object tree, as
it would
- * otherwise overlap with allocations returned by
kfence_alloc(), which
- * are registered with kmemleak through the slab post-alloc
hook.

mm/kfence/core.c +0 -2
-        */
-       kmemleak_free(__kfence_pool);

This appears to only be a non-functional change if the pool is allocated early. If the pool is allocated late using page-alloc, then there'll not be a kmemleak_free() on that memory and we'll have the same problem.

Do you mean the kzalloc(slab_is_available) in memblock_allc()? That implies that MEMBLOCK_ALLOC_NOLEAKTRACE has no guarantee skipping kmemleak_alloc from this. (Maybe add it?)

No, if KFENCE is initialized through kfence_init_late() -> kfence_init_pool_late() -> kfence_init_pool().

Thanks for the information.

But as I known, page-alloc does not request kmemleak areas.
So the current kfence_pool_init_late() would cause another kmemleak warning on unknown freeing.

Reproducing test: (kfence late enable + kmemleak debug on)

/ # echo 500 > /sys/module/kfence/parameters/sample_interval

[  153.433518] kmemleak: Freeing unknown object at 0xffff0000c0600000
[  153.433804] CPU: 0 PID: 100 Comm: sh Not tainted 5.19.0-rc3-74069-gde5c208d533a-dirty #1
[  153.434027] Hardware name: linux,dummy-virt (DT)
[  153.434265] Call trace:
[  153.434331]  dump_backtrace+0xdc/0xfc
[  153.434962]  show_stack+0x18/0x24
[  153.435106]  dump_stack_lvl+0x64/0x7c
[  153.435232]  dump_stack+0x18/0x38
[  153.435347]  kmemleak_free+0x184/0x1c8
[  153.435462]  kfence_init_pool+0x16c/0x194
[  153.435587]  param_set_sample_interval+0xe0/0x1c4
[  153.435694]  param_attr_store+0x98/0xf4
[  153.435804]  module_attr_store+0x24/0x3c
[  153.435910]  sysfs_kf_write+0x3c/0x50
...(skip)
[  153.444496] kfence: initialized - using 524288 bytes for 63 objects at 0x00000000a3236b01-0x00000000901655d3

/ #

Quoted 8 lines

Hence, now there are two issues to solve.
(1) (The original)To prevent the undesired kmemleak scanning on the kfence pool. As Cataline's suggestion, we can just apply kmemleak_ignore_phys instead of free it at all.
ref: https://lore.kernel.org/linux-mm/[email protected]/

(2) The late-allocated kfence pool doesn't need to go through kmemleak_free. We can relocate the opeartion to kfence_init_pool_early() to seperate them. That is, kfence_init_pool_early(memblock) has it and kfence_init_pool_late(page alloc) does not.

The draft is like the following.

Looks reasonable - feel free to send v2.

Thanks,
-- Marco