kfence: save freeing stack trace at calling time instead of freeing time

For kmem_cache with SLAB_TYPESAFE_BY_RCU, the freeing trace stack at
calling kmem_cache_free() is more useful. While the following stack is
meaningless and provides no help:
  freed by task 46 on cpu 0 at 656.840729s:
   rcu_do_batch+0x1ab/0x540
   nocb_cb_wait+0x8f/0x260
   rcu_nocb_cb_kthread+0x25/0x80
   kthread+0xd2/0x100
   ret_from_fork+0x34/0x50
   ret_from_fork_asm+0x1a/0x30

Link: https://lkml.kernel.org/r/20240812095517.2357-1-dtcccc@linux.alibaba.com
Signed-off-by: Tianchen Ding <dtcccc@linux.alibaba.com>
Reviewed-by: Marco Elver <elver@google.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

3 files changed, 34 insertions, 13 deletions

diff --git a/mm/kfence/core.c b/mm/kfence/core.c
index c3ef7eb8d4dc..67fc321db79b 100644
--- a/mm/kfence/core.c
+++ b/mm/kfence/core.c
@@ -273,6 +273,13 @@ static inline unsigned long metadata_to_pageaddr(const struct kfence_metadata *m
 	return pageaddr;
 }
 
+static inline bool kfence_obj_allocated(const struct kfence_metadata *meta)
+{
+	enum kfence_object_state state = READ_ONCE(meta->state);
+
+	return state == KFENCE_OBJECT_ALLOCATED || state == KFENCE_OBJECT_RCU_FREEING;
+}
+
 /*
  * Update the object's metadata state, including updating the alloc/free stacks
  * depending on the state transition.
@@ -282,10 +289,14 @@ metadata_update_state(struct kfence_metadata *meta, enum kfence_object_state nex
 		      unsigned long *stack_entries, size_t num_stack_entries)
 {
 	struct kfence_track *track =
-		next == KFENCE_OBJECT_FREED ? &meta->free_track : &meta->alloc_track;
+		next == KFENCE_OBJECT_ALLOCATED ? &meta->alloc_track : &meta->free_track;
 
 	lockdep_assert_held(&meta->lock);
 
+	/* Stack has been saved when calling rcu, skip. */
+	if (READ_ONCE(meta->state) == KFENCE_OBJECT_RCU_FREEING)
+		goto out;
+
 	if (stack_entries) {
 		memcpy(track->stack_entries, stack_entries,
 		       num_stack_entries * sizeof(stack_entries[0]));
@@ -301,6 +312,7 @@ metadata_update_state(struct kfence_metadata *meta, enum kfence_object_state nex
 	track->cpu = raw_smp_processor_id();
 	track->ts_nsec = local_clock(); /* Same source as printk timestamps. */
 
+out:
 	/*
 	 * Pairs with READ_ONCE() in
 	 *	kfence_shutdown_cache(),
@@ -506,7 +518,7 @@ static void kfence_guarded_free(void *addr, struct kfence_metadata *meta, bool z
 
 	raw_spin_lock_irqsave(&meta->lock, flags);
 
-	if (meta->state != KFENCE_OBJECT_ALLOCATED || meta->addr != (unsigned long)addr) {
+	if (!kfence_obj_allocated(meta) || meta->addr != (unsigned long)addr) {
 		/* Invalid or double-free, bail out. */
 		atomic_long_inc(&counters[KFENCE_COUNTER_BUGS]);
 		kfence_report_error((unsigned long)addr, false, NULL, meta,
@@ -784,7 +796,7 @@ static void kfence_check_all_canary(void)
 	for (i = 0; i < CONFIG_KFENCE_NUM_OBJECTS; i++) {
 		struct kfence_metadata *meta = &kfence_metadata[i];
 
-		if (meta->state == KFENCE_OBJECT_ALLOCATED)
+		if (kfence_obj_allocated(meta))
 			check_canary(meta);
 	}
 }
@@ -1010,12 +1022,11 @@ void kfence_shutdown_cache(struct kmem_cache *s)
 		 * the lock will not help, as different critical section
 		 * serialization will have the same outcome.
 		 */
-		if (READ_ONCE(meta->cache) != s ||
-		    READ_ONCE(meta->state) != KFENCE_OBJECT_ALLOCATED)
+		if (READ_ONCE(meta->cache) != s || !kfence_obj_allocated(meta))
 			continue;
 
 		raw_spin_lock_irqsave(&meta->lock, flags);
-		in_use = meta->cache == s && meta->state == KFENCE_OBJECT_ALLOCATED;
+		in_use = meta->cache == s && kfence_obj_allocated(meta);
 		raw_spin_unlock_irqrestore(&meta->lock, flags);
 
 		if (in_use) {
@@ -1160,11 +1171,19 @@ void __kfence_free(void *addr)
 	 * the object, as the object page may be recycled for other-typed
 	 * objects once it has been freed. meta->cache may be NULL if the cache
 	 * was destroyed.
+	 * Save the stack trace here so that reports show where the user freed
+	 * the object.
 	 */
-	if (unlikely(meta->cache && (meta->cache->flags & SLAB_TYPESAFE_BY_RCU)))
+	if (unlikely(meta->cache && (meta->cache->flags & SLAB_TYPESAFE_BY_RCU))) {
+		unsigned long flags;
+
+		raw_spin_lock_irqsave(&meta->lock, flags);
+		metadata_update_state(meta, KFENCE_OBJECT_RCU_FREEING, NULL, 0);
+		raw_spin_unlock_irqrestore(&meta->lock, flags);
 		call_rcu(&meta->rcu_head, rcu_guarded_free);
-	else
+	} else {
 		kfence_guarded_free(addr, meta, false);
+	}
 }
 
 bool kfence_handle_page_fault(unsigned long addr, bool is_write, struct pt_regs *regs)
@@ -1188,14 +1207,14 @@ bool kfence_handle_page_fault(unsigned long addr, bool is_write, struct pt_regs
 		int distance = 0;
 
 		meta = addr_to_metadata(addr - PAGE_SIZE);
-		if (meta && READ_ONCE(meta->state) == KFENCE_OBJECT_ALLOCATED) {
+		if (meta && kfence_obj_allocated(meta)) {
 			to_report = meta;
 			/* Data race ok; distance calculation approximate. */
 			distance = addr - data_race(meta->addr + meta->size);
 		}
 
 		meta = addr_to_metadata(addr + PAGE_SIZE);
-		if (meta && READ_ONCE(meta->state) == KFENCE_OBJECT_ALLOCATED) {
+		if (meta && kfence_obj_allocated(meta)) {
 			/* Data race ok; distance calculation approximate. */
 			if (!to_report || distance > data_race(meta->addr) - addr)
 				to_report = meta;
diff --git a/mm/kfence/kfence.h b/mm/kfence/kfence.h
index db87a05047bd..dfba5ea06b01 100644
--- a/mm/kfence/kfence.h
+++ b/mm/kfence/kfence.h
@@ -38,6 +38,7 @@
 enum kfence_object_state {
 	KFENCE_OBJECT_UNUSED,		/* Object is unused. */
 	KFENCE_OBJECT_ALLOCATED,	/* Object is currently allocated. */
+	KFENCE_OBJECT_RCU_FREEING,	/* Object was allocated, and then being freed by rcu. */
 	KFENCE_OBJECT_FREED,		/* Object was allocated, and then freed. */
 };
 
diff --git a/mm/kfence/report.c b/mm/kfence/report.c
index 73a6fe42845a..451991a3a8f2 100644
--- a/mm/kfence/report.c
+++ b/mm/kfence/report.c
@@ -114,7 +114,8 @@ static void kfence_print_stack(struct seq_file *seq, const struct kfence_metadat
 
 	/* Timestamp matches printk timestamp format. */
 	seq_con_printf(seq, "%s by task %d on cpu %d at %lu.%06lus (%lu.%06lus ago):\n",
-		       show_alloc ? "allocated" : "freed", track->pid,
+		       show_alloc ? "allocated" : meta->state == KFENCE_OBJECT_RCU_FREEING ?
+		       "rcu freeing" : "freed", track->pid,
 		       track->cpu, (unsigned long)ts_sec, rem_nsec / 1000,
 		       (unsigned long)interval_nsec, rem_interval_nsec / 1000);
 
@@ -149,7 +150,7 @@ void kfence_print_object(struct seq_file *seq, const struct kfence_metadata *met
 
 	kfence_print_stack(seq, meta, true);
 
-	if (meta->state == KFENCE_OBJECT_FREED) {
+	if (meta->state == KFENCE_OBJECT_FREED || meta->state == KFENCE_OBJECT_RCU_FREEING) {
 		seq_con_printf(seq, "\n");
 		kfence_print_stack(seq, meta, false);
 	}
@@ -318,7 +319,7 @@ bool __kfence_obj_info(struct kmem_obj_info *kpp, void *object, struct slab *sla
 	kpp->kp_slab_cache = meta->cache;
 	kpp->kp_objp = (void *)meta->addr;
 	kfence_to_kp_stack(&meta->alloc_track, kpp->kp_stack);
-	if (meta->state == KFENCE_OBJECT_FREED)
+	if (meta->state == KFENCE_OBJECT_FREED || meta->state == KFENCE_OBJECT_RCU_FREEING)
 		kfence_to_kp_stack(&meta->free_track, kpp->kp_free_stack);
 	/* get_stack_skipnr() ensures the first entry is outside allocator. */
 	kpp->kp_ret = kpp->kp_stack[0];