Skip to contentSkip to conversation

stackdepot, kasan, workqueue: Avoid expanding stackdepot slabs when holding raw_spin_lock

PATCH Superseded Marco Elver 6 participants 17 messages Sep 7, 2021 – Sep 14, 2021

Subsystems (5)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover stackdepot, kasan, workqueue: Avoid expanding stackdepot slabs when holding raw_spin_lock

Shuah Khan reported [1]:

 | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
 | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
 | it tries to allocate memory attempting to acquire spinlock in page
 | allocation code while holding workqueue pool raw_spinlock.
 |
 | There are several instances of this problem when block layer tries
 | to __queue_work(). Call trace from one of these instances is below:
 |
 |     kblockd_mod_delayed_work_on()
 |       mod_delayed_work_on()
 |         __queue_delayed_work()
 |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
 |             insert_work()
 |               kasan_record_aux_stack()
 |                 kasan_save_stack()
 |                   stack_depot_save()
 |                     alloc_pages()
 |                       __alloc_pages()
 |                         get_page_from_freelist()
 |                           rm_queue()
 |                             rm_queue_pcplist()
 |                               local_lock_irqsave(&pagesets.lock, flags);
 |                               [ BUG: Invalid wait context triggered ]

[1] https://lkml.kernel.org/r/[email protected]

PROVE_RAW_LOCK_NESTING is pointing out that (on RT kernels) the locking rules are being violated. More generally, memory is being allocated from a non-preemptive context (raw_spin_lock'd c-s) where it is not allowed.

To properly fix this, we must prevent stackdepot from replenishing its "stack slab" pool if memory allocations cannot be done in the current
context: it's a bug to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

The only downside is that saving a stack trace may fail if: stackdepot runs out of space AND the same stack trace has not been recorded before. I expect this to be unlikely, and a simple experiment (boot the kernel) didn't result in any failure to record stack trace from insert_work().

The series includes a few minor fixes to stackdepot that I noticed in preparing the series. It then introduces __stack_depot_save(), which exposes the option to force stackdepot to not allocate any memory. Finally, KASAN is changed to use the new stackdepot interface and provide kasan_record_aux_stack_noalloc(), which is then used by workqueue code.

Marco Elver (6):
lib/stackdepot: include gfp.h
lib/stackdepot: remove unused function argument lib/stackdepot: introduce __stack_depot_save()
kasan: common: provide can_alloc in kasan_save_stack()
kasan: generic: introduce kasan_record_aux_stack_noalloc() workqueue, kasan: avoid alloc_pages() when recording stack

 include/linux/kasan.h      |  2 ++
 include/linux/stackdepot.h |  6 +++++
 kernel/workqueue.c         |  2 +-
 lib/stackdepot.c           | 51 ++++++++++++++++++++++++++++++--------
 mm/kasan/common.c          |  6 ++---
 mm/kasan/generic.c         | 14 +++++++++--
 mm/kasan/kasan.h           |  2 +-
 7 files changed, 65 insertions(+), 18 deletions(-)
Signature · 1 line
2.33.0.153.gba50c8fa24-goog
#1 lore raw

PATCH 1/6 lib/stackdepot: include gfp.h

<linux/stackdepot.h> refers to gfp_t, but doesn't include gfp.h.

Fix it by including <linux/gfp.h>.

---
 include/linux/stackdepot.h | 2 ++
 1 file changed, 2 insertions(+)
include/linux/stackdepot.h +2 -0
index 6bb4bc1a5f54..97b36dc53301 100644
@@ -11,6 +11,8 @@
 #ifndef _LINUX_STACKDEPOT_H
 #define _LINUX_STACKDEPOT_H
 
+#include <linux/gfp.h>
+
 typedef u32 depot_stack_handle_t;
 
 depot_stack_handle_t stack_depot_save(unsigned long *entries,
Signature · 1 line
2.33.0.153.gba50c8fa24-goog
reviewer Reviewed-by on v1 1/6
#2 lore raw

Re: [PATCH 1/6] lib/stackdepot: include gfp.h

Quoted 6 lines · Marco Elver

<linux/stackdepot.h> refers to gfp_t, but doesn't include gfp.h.

Fix it by including <linux/gfp.h>.

Quoted 20 lines
---
 include/linux/stackdepot.h | 2 ++
 1 file changed, 2 insertions(+)
include/linux/stackdepot.h +2 -1
index 6bb4bc1a5f54..97b36dc53301 100644
@@ -11,6 +11,8 @@
 #ifndef _LINUX_STACKDEPOT_H
 #define _LINUX_STACKDEPOT_H
 
+#include <linux/gfp.h>
+
 typedef u32 depot_stack_handle_t;
 
 depot_stack_handle_t stack_depot_save(unsigned long *entries,
--

2.33.0.153.gba50c8fa24-goog

Signature · 10 lines
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg
#3 lore raw

PATCH 2/6 lib/stackdepot: remove unused function argument

In reply to Marco Elver

alloc_flags in depot_alloc_stack() is no longer used; remove it.

---
 lib/stackdepot.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
lib/stackdepot.c +4 -5
index 0a2e417f83cb..c80a9f734253 100644
@@ -102,8 +102,8 @@ static bool init_stack_slab(void **prealloc)
 }
 
 /* Allocation of a new stack in raw storage */
-static struct stack_record *depot_alloc_stack(unsigned long *entries, int size,
-		u32 hash, void **prealloc, gfp_t alloc_flags)
+static struct stack_record *
+depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc)
 {
 	struct stack_record *stack;
 	size_t required_size = struct_size(stack, entries, size);
@@ -309,9 +309,8 @@ depot_stack_handle_t stack_depot_save(unsigned long *entries,
 
 	found = find_stack(*bucket, entries, nr_entries, hash);
 	if (!found) {
-		struct stack_record *new =
-			depot_alloc_stack(entries, nr_entries,
-					  hash, &prealloc, alloc_flags);
+		struct stack_record *new = depot_alloc_stack(entries, nr_entries, hash, &prealloc);
+
 		if (new) {
 			new->next = *bucket;
 			/*
Signature · 1 line
2.33.0.153.gba50c8fa24-goog
reviewer Reviewed-by on v1 2/6
#4 lore raw

Re: [PATCH 2/6] lib/stackdepot: remove unused function argument

Quoted 4 lines · Marco Elver

alloc_flags in depot_alloc_stack() is no longer used; remove it.

Quoted 34 lines
---
 lib/stackdepot.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
lib/stackdepot.c +4 -6
index 0a2e417f83cb..c80a9f734253 100644
@@ -102,8 +102,8 @@ static bool init_stack_slab(void **prealloc)
 }
 
 /* Allocation of a new stack in raw storage */
-static struct stack_record *depot_alloc_stack(unsigned long *entries, int size,
-               u32 hash, void **prealloc, gfp_t alloc_flags)
+static struct stack_record *
+depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc)
 {
        struct stack_record *stack;
        size_t required_size = struct_size(stack, entries, size);
@@ -309,9 +309,8 @@ depot_stack_handle_t stack_depot_save(unsigned long *entries,
 
        found = find_stack(*bucket, entries, nr_entries, hash);
        if (!found) {
-               struct stack_record *new =
-                       depot_alloc_stack(entries, nr_entries,
-                                         hash, &prealloc, alloc_flags);
+               struct stack_record *new = depot_alloc_stack(entries, nr_entries, hash, &prealloc);
+
                if (new) {
                        new->next = *bucket;
                        /*
--

2.33.0.153.gba50c8fa24-goog

Signature · 10 lines
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg
#5 lore raw

PATCH 3/6 lib/stackdepot: introduce __stack_depot_save()

In reply to Marco Elver

Add __stack_depot_save(), which provides more fine-grained control over stackdepot's memory allocation behaviour, in case stackdepot runs out of "stack slabs".

Normally stackdepot uses alloc_pages() in case it runs out of space; passing can_alloc==false to __stack_depot_save() prohibits this, at the cost of more likely failure to record a stack trace.

---
 include/linux/stackdepot.h |  4 ++++
 lib/stackdepot.c           | 42 ++++++++++++++++++++++++++++++++------
 2 files changed, 40 insertions(+), 6 deletions(-)
include/linux/stackdepot.h +4 -0
index 97b36dc53301..b2f7e7c6ba54 100644
@@ -15,6 +15,10 @@
 
 typedef u32 depot_stack_handle_t;
 
+depot_stack_handle_t __stack_depot_save(unsigned long *entries,
+					unsigned int nr_entries,
+					gfp_t gfp_flags, bool can_alloc);
+
 depot_stack_handle_t stack_depot_save(unsigned long *entries,
 				      unsigned int nr_entries, gfp_t gfp_flags);
 
lib/stackdepot.c +36 -6

66 lines not on this page · raw message

Signature · 1 line
2.33.0.153.gba50c8fa24-goog
#6 lore raw

Re: [PATCH 3/6] lib/stackdepot: introduce __stack_depot_save()

Quoted 8 lines · Marco Elver
lib/stackdepot.c +0 -0
index c80a9f734253..cab6cf117290 100644
@@ -339,6 +350,25 @@ depot_stack_handle_t stack_depot_save(unsigned long *entries,
 fast_exit:
 	return retval;
 }

lib/stackdepot.c depot_stack_handle_t stack_depot_save(unsigned long *entries, · line 362

Quoted 10 lines
lib/stackdepot.c +10 -0
+EXPORT_SYMBOL_GPL(__stack_depot_save);
+
+/**
+ * stack_depot_save - Save a stack trace from an array
+ *
+ * @entries:		Pointer to storage array
+ * @nr_entries:		Size of the storage array
+ * @alloc_flags:	Allocation gfp flags
+ *
+ * Context: Contexts where allocations via alloc_pages() are allowed.

Could we add here something like (see __stack_depot_save() for details) since it has more verbose.

Sebastian

#7 lore raw

PATCH 4/6 kasan: common: provide can_alloc in kasan_save_stack()

In reply to Marco Elver

Add another argument, can_alloc, to kasan_save_stack() which is passed as-is to __stack_depot_save().

No functional change intended.

---
 mm/kasan/common.c  | 6 +++---
 mm/kasan/generic.c | 2 +-
 mm/kasan/kasan.h   | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)
mm/kasan/common.c +3 -3
index 2baf121fb8c5..3e0999892c36 100644
@@ -30,20 +30,20 @@
 #include "kasan.h"
 #include "../slab.h"
 
-depot_stack_handle_t kasan_save_stack(gfp_t flags)
+depot_stack_handle_t kasan_save_stack(gfp_t flags, bool can_alloc)
 {
 	unsigned long entries[KASAN_STACK_DEPTH];
 	unsigned int nr_entries;
 
 	nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 0);
 	nr_entries = filter_irq_stacks(entries, nr_entries);
-	return stack_depot_save(entries, nr_entries, flags);
+	return __stack_depot_save(entries, nr_entries, flags, can_alloc);
 }
 
 void kasan_set_track(struct kasan_track *track, gfp_t flags)
 {
 	track->pid = current->pid;
-	track->stack = kasan_save_stack(flags);
+	track->stack = kasan_save_stack(flags, true);
 }
 
 #if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
mm/kasan/generic.c +1 -1
index c3f5ba7a294a..2a8e59e6326d 100644
@@ -345,7 +345,7 @@ void kasan_record_aux_stack(void *addr)
 		return;
 
 	alloc_meta->aux_stack[1] = alloc_meta->aux_stack[0];
-	alloc_meta->aux_stack[0] = kasan_save_stack(GFP_NOWAIT);
+	alloc_meta->aux_stack[0] = kasan_save_stack(GFP_NOWAIT, true);
 }
 
 void kasan_set_free_info(struct kmem_cache *cache,
mm/kasan/kasan.h +1 -1
index fa02c88b6948..e442d94a8f6e 100644
@@ -260,7 +260,7 @@ void kasan_report_invalid_free(void *object, unsigned long ip);
 
 struct page *kasan_addr_to_page(const void *addr);
 
-depot_stack_handle_t kasan_save_stack(gfp_t flags);
+depot_stack_handle_t kasan_save_stack(gfp_t flags, bool can_alloc);
 void kasan_set_track(struct kasan_track *track, gfp_t flags);
 void kasan_set_free_info(struct kmem_cache *cache, void *object, u8 tag);
 struct kasan_track *kasan_get_free_track(struct kmem_cache *cache,
Signature · 1 line
2.33.0.153.gba50c8fa24-goog
#8 lore raw

PATCH 5/6 kasan: generic: introduce kasan_record_aux_stack_noalloc()

In reply to Marco Elver

Introduce a variant of kasan_record_aux_stack() that does not do any memory allocation through stackdepot. This will permit using it in contexts that cannot allocate any memory.

---
 include/linux/kasan.h |  2 ++
 mm/kasan/generic.c    | 14 ++++++++++++--
 2 files changed, 14 insertions(+), 2 deletions(-)
include/linux/kasan.h +2 -0
index dd874a1ee862..736d7b458996 100644
@@ -370,12 +370,14 @@ static inline void kasan_unpoison_task_stack(struct task_struct *task) {}
 void kasan_cache_shrink(struct kmem_cache *cache);
 void kasan_cache_shutdown(struct kmem_cache *cache);
 void kasan_record_aux_stack(void *ptr);
+void kasan_record_aux_stack_noalloc(void *ptr);
 
 #else /* CONFIG_KASAN_GENERIC */
 
 static inline void kasan_cache_shrink(struct kmem_cache *cache) {}
 static inline void kasan_cache_shutdown(struct kmem_cache *cache) {}
 static inline void kasan_record_aux_stack(void *ptr) {}
+static inline void kasan_record_aux_stack_noalloc(void *ptr) {}
 
 #endif /* CONFIG_KASAN_GENERIC */
 
mm/kasan/generic.c +12 -2
index 2a8e59e6326d..84a038b07c6f 100644
@@ -328,7 +328,7 @@ DEFINE_ASAN_SET_SHADOW(f3);
 DEFINE_ASAN_SET_SHADOW(f5);
 DEFINE_ASAN_SET_SHADOW(f8);
 
-void kasan_record_aux_stack(void *addr)
+static void __kasan_record_aux_stack(void *addr, bool can_alloc)
 {
 	struct page *page = kasan_addr_to_page(addr);
 	struct kmem_cache *cache;
@@ -345,7 +345,17 @@ void kasan_record_aux_stack(void *addr)
 		return;
 
 	alloc_meta->aux_stack[1] = alloc_meta->aux_stack[0];
-	alloc_meta->aux_stack[0] = kasan_save_stack(GFP_NOWAIT, true);
+	alloc_meta->aux_stack[0] = kasan_save_stack(GFP_NOWAIT, can_alloc);
+}
+
+void kasan_record_aux_stack(void *addr)
+{
+	return __kasan_record_aux_stack(addr, true);
+}
+
+void kasan_record_aux_stack_noalloc(void *addr)
+{
+	return __kasan_record_aux_stack(addr, false);
 }
 
 void kasan_set_free_info(struct kmem_cache *cache,
Signature · 1 line
2.33.0.153.gba50c8fa24-goog
#9 lore raw

PATCH 6/6 workqueue, kasan: avoid alloc_pages() when recording stack

In reply to Marco Elver

Shuah Khan reported:

 | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
 | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
 | it tries to allocate memory attempting to acquire spinlock in page
 | allocation code while holding workqueue pool raw_spinlock.
 |
 | There are several instances of this problem when block layer tries
 | to __queue_work(). Call trace from one of these instances is below:
 |
 |     kblockd_mod_delayed_work_on()
 |       mod_delayed_work_on()
 |         __queue_delayed_work()
 |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
 |             insert_work()
 |               kasan_record_aux_stack()
 |                 kasan_save_stack()
 |                   stack_depot_save()
 |                     alloc_pages()
 |                       __alloc_pages()
 |                         get_page_from_freelist()
 |                           rm_queue()
 |                             rm_queue_pcplist()
 |                               local_lock_irqsave(&pagesets.lock, flags);
 |                               [ BUG: Invalid wait context triggered ]

The default kasan_record_aux_stack() calls stack_depot_save() with GFP_NOWAIT, which in turn can then call alloc_pages(GFP_NOWAIT, ...). In general, however, it is not even possible to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

Fix it by instructing stackdepot to not expand stack storage via alloc_pages() in case it runs out by using kasan_record_aux_stack_noalloc().

While there is an increased risk of failing to insert the stack trace, this is typically unlikely, especially if the same insertion had already succeeded previously (stack depot hit). For frequent calls from the same location, it therefore becomes extremely unlikely that kasan_record_aux_stack_noalloc() fails.

---
 kernel/workqueue.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/workqueue.c +1 -1
index 50142fc08902..0681774e6908 100644
@@ -1329,7 +1329,7 @@ static void insert_work(struct pool_workqueue *pwq, struct work_struct *work,
 	struct worker_pool *pool = pwq->pool;
 
 	/* record the work call stack in order to print it in KASAN reports */
-	kasan_record_aux_stack(work);
+	kasan_record_aux_stack_noalloc(work);
 
 	/* we own @work, set data and link */
 	set_work_pwq(work, pwq, extra_flags);
Signature · 1 line
2.33.0.153.gba50c8fa24-goog
maintainer Acked-by on v1 6/6
#10 lore raw

Re: [PATCH 6/6] workqueue, kasan: avoid alloc_pages() when recording stack

Quoted 41 lines · Marco Elver

Shuah Khan reported:

 | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
 | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
 | it tries to allocate memory attempting to acquire spinlock in page
 | allocation code while holding workqueue pool raw_spinlock.
 |
 | There are several instances of this problem when block layer tries
 | to __queue_work(). Call trace from one of these instances is below:
 |
 |     kblockd_mod_delayed_work_on()
 |       mod_delayed_work_on()
 |         __queue_delayed_work()
 |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
 |             insert_work()
 |               kasan_record_aux_stack()
 |                 kasan_save_stack()
 |                   stack_depot_save()
 |                     alloc_pages()
 |                       __alloc_pages()
 |                         get_page_from_freelist()
 |                           rm_queue()
 |                             rm_queue_pcplist()
 |                               local_lock_irqsave(&pagesets.lock, flags);
 |                               [ BUG: Invalid wait context triggered ]

The default kasan_record_aux_stack() calls stack_depot_save() with GFP_NOWAIT, which in turn can then call alloc_pages(GFP_NOWAIT, ...). In general, however, it is not even possible to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

Fix it by instructing stackdepot to not expand stack storage via alloc_pages() in case it runs out by using kasan_record_aux_stack_noalloc().

While there is an increased risk of failing to insert the stack trace, this is typically unlikely, especially if the same insertion had already succeeded previously (stack depot hit). For frequent calls from the same location, it therefore becomes extremely unlikely that kasan_record_aux_stack_noalloc() fails.

Quoted 3 lines

Thanks.

Signature · 1 line
tejun
#11 lore raw
In reply to Marco Elver

[+Cc: Thomas, Sebastian]

Sorry, forgot to Cc you... :-/

Quoted 71 lines · Marco Elver

Shuah Khan reported [1]:

 | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
 | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
 | it tries to allocate memory attempting to acquire spinlock in page
 | allocation code while holding workqueue pool raw_spinlock.
 |
 | There are several instances of this problem when block layer tries
 | to __queue_work(). Call trace from one of these instances is below:
 |
 |     kblockd_mod_delayed_work_on()
 |       mod_delayed_work_on()
 |         __queue_delayed_work()
 |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
 |             insert_work()
 |               kasan_record_aux_stack()
 |                 kasan_save_stack()
 |                   stack_depot_save()
 |                     alloc_pages()
 |                       __alloc_pages()
 |                         get_page_from_freelist()
 |                           rm_queue()
 |                             rm_queue_pcplist()
 |                               local_lock_irqsave(&pagesets.lock, flags);
 |                               [ BUG: Invalid wait context triggered ]

[1] https://lkml.kernel.org/r/[email protected]

PROVE_RAW_LOCK_NESTING is pointing out that (on RT kernels) the locking rules are being violated. More generally, memory is being allocated from a non-preemptive context (raw_spin_lock'd c-s) where it is not allowed.

To properly fix this, we must prevent stackdepot from replenishing its "stack slab" pool if memory allocations cannot be done in the current
context: it's a bug to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

The only downside is that saving a stack trace may fail if: stackdepot runs out of space AND the same stack trace has not been recorded before. I expect this to be unlikely, and a simple experiment (boot the kernel) didn't result in any failure to record stack trace from insert_work().

The series includes a few minor fixes to stackdepot that I noticed in preparing the series. It then introduces __stack_depot_save(), which exposes the option to force stackdepot to not allocate any memory. Finally, KASAN is changed to use the new stackdepot interface and provide kasan_record_aux_stack_noalloc(), which is then used by workqueue code.

Marco Elver (6):
lib/stackdepot: include gfp.h
lib/stackdepot: remove unused function argument lib/stackdepot: introduce __stack_depot_save()
kasan: common: provide can_alloc in kasan_save_stack()
kasan: generic: introduce kasan_record_aux_stack_noalloc() workqueue, kasan: avoid alloc_pages() when recording stack

 include/linux/kasan.h      |  2 ++
 include/linux/stackdepot.h |  6 +++++
 kernel/workqueue.c         |  2 +-
 lib/stackdepot.c           | 51 ++++++++++++++++++++++++++++++--------
 mm/kasan/common.c          |  6 ++---
 mm/kasan/generic.c         | 14 +++++++++--
 mm/kasan/kasan.h           |  2 +-
 7 files changed, 65 insertions(+), 18 deletions(-)

--
2.33.0.153.gba50c8fa24-goog

Tested-by on the v1 series
#12 lore raw
In reply to Marco Elver
Quoted 58 lines · Marco Elver

Shuah Khan reported [1]:

  | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
  | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
  | it tries to allocate memory attempting to acquire spinlock in page
  | allocation code while holding workqueue pool raw_spinlock.
  |
  | There are several instances of this problem when block layer tries
  | to __queue_work(). Call trace from one of these instances is below:
  |
  |     kblockd_mod_delayed_work_on()
  |       mod_delayed_work_on()
  |         __queue_delayed_work()
  |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
  |             insert_work()
  |               kasan_record_aux_stack()
  |                 kasan_save_stack()
  |                   stack_depot_save()
  |                     alloc_pages()
  |                       __alloc_pages()
  |                         get_page_from_freelist()
  |                           rm_queue()
  |                             rm_queue_pcplist()
  |                               local_lock_irqsave(&pagesets.lock, flags);
  |                               [ BUG: Invalid wait context triggered ]

[1] https://lkml.kernel.org/r/[email protected]

PROVE_RAW_LOCK_NESTING is pointing out that (on RT kernels) the locking rules are being violated. More generally, memory is being allocated from a non-preemptive context (raw_spin_lock'd c-s) where it is not allowed.

To properly fix this, we must prevent stackdepot from replenishing its "stack slab" pool if memory allocations cannot be done in the current
context: it's a bug to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

The only downside is that saving a stack trace may fail if: stackdepot runs out of space AND the same stack trace has not been recorded before. I expect this to be unlikely, and a simple experiment (boot the kernel) didn't result in any failure to record stack trace from insert_work().

The series includes a few minor fixes to stackdepot that I noticed in preparing the series. It then introduces __stack_depot_save(), which exposes the option to force stackdepot to not allocate any memory. Finally, KASAN is changed to use the new stackdepot interface and provide kasan_record_aux_stack_noalloc(), which is then used by workqueue code.

Marco Elver (6):
lib/stackdepot: include gfp.h
lib/stackdepot: remove unused function argument lib/stackdepot: introduce __stack_depot_save()
kasan: common: provide can_alloc in kasan_save_stack()
kasan: generic: introduce kasan_record_aux_stack_noalloc() workqueue, kasan: avoid alloc_pages() when recording stack

Quoted 9 lines
  include/linux/kasan.h      |  2 ++
  include/linux/stackdepot.h |  6 +++++
  kernel/workqueue.c         |  2 +-
  lib/stackdepot.c           | 51 ++++++++++++++++++++++++++++++--------
  mm/kasan/common.c          |  6 ++---
  mm/kasan/generic.c         | 14 +++++++++--
  mm/kasan/kasan.h           |  2 +-
  7 files changed, 65 insertions(+), 18 deletions(-)

Thank you. Tested all the 6 patches in this series on Linux 5.14. This problem exists in 5.13 and needs to be marked for both 5.14 and 5.13 stable releases.

Here is my

thanks,
-- Shuah

#13 lore raw
Quoted 70 lines · Shuah Khan
Quoted 59 lines · Marco Elver

Shuah Khan reported [1]:

  | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
  | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
  | it tries to allocate memory attempting to acquire spinlock in page
  | allocation code while holding workqueue pool raw_spinlock.
  |
  | There are several instances of this problem when block layer tries
  | to __queue_work(). Call trace from one of these instances is below:
  |
  |     kblockd_mod_delayed_work_on()
  |       mod_delayed_work_on()
  |         __queue_delayed_work()
  |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
  |             insert_work()
  |               kasan_record_aux_stack()
  |                 kasan_save_stack()
  |                   stack_depot_save()
  |                     alloc_pages()
  |                       __alloc_pages()
  |                         get_page_from_freelist()
  |                           rm_queue()
  |                             rm_queue_pcplist()
  |                               local_lock_irqsave(&pagesets.lock, flags);
  |                               [ BUG: Invalid wait context triggered ]

[1]
https://lkml.kernel.org/r/[email protected]

PROVE_RAW_LOCK_NESTING is pointing out that (on RT kernels) the locking rules are being violated. More generally, memory is being allocated from a non-preemptive context (raw_spin_lock'd c-s) where it is not allowed.

To properly fix this, we must prevent stackdepot from replenishing its "stack slab" pool if memory allocations cannot be done in the current
context: it's a bug to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

The only downside is that saving a stack trace may fail if: stackdepot runs out of space AND the same stack trace has not been recorded before. I expect this to be unlikely, and a simple experiment (boot the kernel) didn't result in any failure to record stack trace from insert_work().

The series includes a few minor fixes to stackdepot that I noticed in preparing the series. It then introduces __stack_depot_save(), which exposes the option to force stackdepot to not allocate any memory. Finally, KASAN is changed to use the new stackdepot interface and provide kasan_record_aux_stack_noalloc(), which is then used by workqueue code.

Marco Elver (6):
   lib/stackdepot: include gfp.h
   lib/stackdepot: remove unused function argument lib/stackdepot: introduce __stack_depot_save()
   kasan: common: provide can_alloc in kasan_save_stack()
   kasan: generic: introduce kasan_record_aux_stack_noalloc() workqueue, kasan: avoid alloc_pages() when recording stack

Quoted 9 lines

  include/linux/kasan.h      |  2 ++ include/linux/stackdepot.h |  6 +++++
  kernel/workqueue.c         |  2 +-
  lib/stackdepot.c           | 51 ++++++++++++++++++++++++++++++-------- mm/kasan/common.c          |  6 ++---
  mm/kasan/generic.c         | 14 +++++++++-- mm/kasan/kasan.h           |  2 +-

  7 files changed, 65 insertions(+), 18 deletions(-)
Quoted 2 lines

Thank you. Tested all the 6 patches in this series on Linux 5.14. This problem exists in 5.13 and needs to be marked for both 5.14 and 5.13 stable releases.

I think if this problem manifests only with CONFIG_PROVE_RAW_LOCK_NESTING then it shouldn't be backported to stable. CONFIG_PROVE_RAW_LOCK_NESTING is an experimental/development option to earlier discover what will collide with RT lock semantics, without needing the full RT tree.
Thus, good to fix going forward, but not necessary to stable backport.

Quoted 7 lines

Here is my

thanks,
-- Shuah

#14 lore raw
Quoted 8 lines · Vlastimil Babka
Quoted 2 lines

Thank you. Tested all the 6 patches in this series on Linux 5.14. This problem exists in 5.13 and needs to be marked for both 5.14 and 5.13 stable releases.

I think if this problem manifests only with CONFIG_PROVE_RAW_LOCK_NESTING then it shouldn't be backported to stable. CONFIG_PROVE_RAW_LOCK_NESTING is an experimental/development option to earlier discover what will collide with RT lock semantics, without needing the full RT tree.
Thus, good to fix going forward, but not necessary to stable backport.

for the series. Thank you.

As for the backport I agree here with Vlastimil.

I pulled it into my RT tree for some testing and it looked good. I had to

kernel/rcu/tree.c +1 -1
@@ -3030,7 +3030,7 @@ __call_rcu(struct rcu_head *head, rcu_callback_t func)
        head->func = func;
        head->next = NULL;
        local_irq_save(flags);
-       kasan_record_aux_stack(head);
+       kasan_record_aux_stack_noalloc(head);
        rdp = this_cpu_ptr(&rcu_data);
 
        /* Add the callback to our list. */

We could move kasan_record_aux_stack() before that local_irq_save() but then call_rcu() can be called preempt-disabled section so we would have the same problem.

The second warning came from kasan_quarantine_remove_cache(). At the end per_cpu_remove_cache() -> qlist_free_all() will free memory with disabled interrupts (due to that smp-function call).
Moving it to kworker would solve the problem. I don't mind keeping that smp_function call assuming that it is all debug-code and it increases overall latency anyway. But then could we maybe move all those objects to a single list which freed after on_each_cpu()?

Otherwise I haven't seen any new warnings showing up with KASAN enabled.

Sebastian

#15 lore raw
Quoted 4 lines · Sebastian Andrzej Siewior
Quoted 3 lines · Vlastimil Babka
Quoted 2 lines

Thank you. Tested all the 6 patches in this series on Linux 5.14. This problem exists in 5.13 and needs to be marked for both 5.14 and 5.13 stable releases.

Quoted 8 lines
Quoted 5 lines

I think if this problem manifests only with CONFIG_PROVE_RAW_LOCK_NESTING then it shouldn't be backported to stable. CONFIG_PROVE_RAW_LOCK_NESTING is an experimental/development option to earlier discover what will collide with RT lock semantics, without needing the full RT tree.
Thus, good to fix going forward, but not necessary to stable backport.

for the series. Thank you.

Thank you. I'll send v2 with Acks/Tested-by added and the comment addition you suggested.

Quoted 20 lines

As for the backport I agree here with Vlastimil.

I pulled it into my RT tree for some testing and it looked good. I had to

kernel/rcu/tree.c +1 -1
@@ -3030,7 +3030,7 @@ __call_rcu(struct rcu_head *head, rcu_callback_t func)
        head->func = func;
        head->next = NULL;
        local_irq_save(flags);
-       kasan_record_aux_stack(head);
+       kasan_record_aux_stack_noalloc(head);
        rdp = this_cpu_ptr(&rcu_data);
 
        /* Add the callback to our list. */

We could move kasan_record_aux_stack() before that local_irq_save() but then call_rcu() can be called preempt-disabled section so we would have the same problem.

Quoted 7 lines

The second warning came from kasan_quarantine_remove_cache(). At the end per_cpu_remove_cache() -> qlist_free_all() will free memory with disabled interrupts (due to that smp-function call).
Moving it to kworker would solve the problem. I don't mind keeping that smp_function call assuming that it is all debug-code and it increases overall latency anyway. But then could we maybe move all those objects to a single list which freed after on_each_cpu()?

The quarantine is per-CPU, and I think what you suggest would fundamentally change its design. If you have something that works on RT without a fundamental change would be ideal (it is all debug code and not used on non-KASAN kernels).

Quoted 3 lines

Otherwise I haven't seen any new warnings showing up with KASAN enabled.

Sebastian

reviewer Acked-by on the v1 series
#16 lore raw
In reply to Marco Elver
Quoted 51 lines · Marco Elver

Shuah Khan reported [1]:

 | When CONFIG_PROVE_RAW_LOCK_NESTING=y and CONFIG_KASAN are enabled,
 | kasan_record_aux_stack() runs into "BUG: Invalid wait context" when
 | it tries to allocate memory attempting to acquire spinlock in page
 | allocation code while holding workqueue pool raw_spinlock.
 |
 | There are several instances of this problem when block layer tries
 | to __queue_work(). Call trace from one of these instances is below:
 |
 |     kblockd_mod_delayed_work_on()
 |       mod_delayed_work_on()
 |         __queue_delayed_work()
 |           __queue_work() (rcu_read_lock, raw_spin_lock pool->lock held)
 |             insert_work()
 |               kasan_record_aux_stack()
 |                 kasan_save_stack()
 |                   stack_depot_save()
 |                     alloc_pages()
 |                       __alloc_pages()
 |                         get_page_from_freelist()
 |                           rm_queue()
 |                             rm_queue_pcplist()
 |                               local_lock_irqsave(&pagesets.lock, flags);
 |                               [ BUG: Invalid wait context triggered ]

[1] https://lkml.kernel.org/r/[email protected]

PROVE_RAW_LOCK_NESTING is pointing out that (on RT kernels) the locking rules are being violated. More generally, memory is being allocated from a non-preemptive context (raw_spin_lock'd c-s) where it is not allowed.

To properly fix this, we must prevent stackdepot from replenishing its "stack slab" pool if memory allocations cannot be done in the current
context: it's a bug to use either GFP_ATOMIC nor GFP_NOWAIT in certain non-preemptive contexts, including raw_spin_locks (see gfp.h and ab00db216c9c7).

The only downside is that saving a stack trace may fail if: stackdepot runs out of space AND the same stack trace has not been recorded before. I expect this to be unlikely, and a simple experiment (boot the kernel) didn't result in any failure to record stack trace from insert_work().

The series includes a few minor fixes to stackdepot that I noticed in preparing the series. It then introduces __stack_depot_save(), which exposes the option to force stackdepot to not allocate any memory. Finally, KASAN is changed to use the new stackdepot interface and provide kasan_record_aux_stack_noalloc(), which is then used by workqueue code.

Quoted 7 lines

Marco Elver (6):
lib/stackdepot: include gfp.h
lib/stackdepot: remove unused function argument lib/stackdepot: introduce __stack_depot_save()
kasan: common: provide can_alloc in kasan_save_stack()
kasan: generic: introduce kasan_record_aux_stack_noalloc() workqueue, kasan: avoid alloc_pages() when recording stack

for the whole series.

Quoted 13 lines
 include/linux/kasan.h      |  2 ++
 include/linux/stackdepot.h |  6 +++++
 kernel/workqueue.c         |  2 +-
 lib/stackdepot.c           | 51 ++++++++++++++++++++++++++++++--------
 mm/kasan/common.c          |  6 ++---
 mm/kasan/generic.c         | 14 +++++++++--
 mm/kasan/kasan.h           |  2 +-
 7 files changed, 65 insertions(+), 18 deletions(-)

--
2.33.0.153.gba50c8fa24-goog

Signature · 10 lines
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg