[PATCH] kasan: unpoison task stack below watermark only in generic mode

[Andrey Ryabinin <ryabinin.a.a@gmail.com>]

CPU resume and BPF exception handling can discard stack frames without
running their compiler-generated epilogues. Generic KASAN needs
kasan_unpoison_task_stack_below() to clear the redzones left behind by
those frames before the stack is reused.

CONFIG_KASAN_STACK also enables this helper for SW_TAGS. The helper
derives the stack base from an untagged stack pointer, so
kasan_unpoison() writes KASAN_TAG_KERNEL (0xff) into the shadow for
[base, watermark). For a vmapped task stack, vm_area->addr still
carries the original random allocation tag. This creates a tag mismatch
at the stack base even if that memory has never held an instrumented
stack object.

When the task exits and its stack is not cached, thread_stack_free_rcu()
passes vm_area->addr to vfree(). In RCU callback context this reaches
vfree_atomic(), whose llist_add() writes to the allocation base through
the tagged pointer and triggers a false KASAN invalid-access report:

  BUG: KASAN: invalid-access in vfree_atomic+0x90/0x150
  Write of size 8 at addr c2ffffc0a8f70000 by task rcuop/7/75
  Pointer tag: [c2], memory tag: [ff]
  Call trace:
   __hwasan_store8_noabort+0xe8/0xf8
   vfree_atomic+0x90/0x150
   vfree+0x220/0x298
   thread_stack_free_rcu+0x3c/0x4c
   rcu_do_batch+0x308/0xaf0
   rcu_nocb_cb_kthread+0x33c/0x708

The deferred-free path started using the tagged vm_area->addr in
commit 449e0b4ed5a1 ("fork: clean-up naming of vm_stack/vm_struct
variables in vmap stacks code"). Previously it freed the untagged
pointer derived from tsk->stack, so the write was never tag-checked.

SW_TAGS does not need the blanket shadow reset to make subsequent stack
accesses valid: untagged kernel pointers have the match-all 0xff tag,
and the compiler initializes the shadow tags of instrumented stack
objects before use.

Return early unless CONFIG_KASAN_GENERIC is enabled. Keep the mode check
in the common helper so it covers both resume and bpf_throw().

Fixes: 449e0b4ed5a1 ("fork: clean-up naming of vm_stack/vm_struct variables in vmap stacks code")
Cc: stable@vger.kernel.org
Reported-by: Shaobo Huang <huangshaobo3@xiaomi.com>
Closes: https://lore.kernel.org/all/20260806123020.90869-1-huangshaobo3@xiaomi.com/
Assisted-by: LLM
Signed-off-by: Andrey Ryabinin <ryabinin.a.a@gmail.com>
---
 mm/kasan/common.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/mm/kasan/common.c b/mm/kasan/common.c
index b7d05c2a6d93..7d21c4db68d2 100644
--- a/mm/kasan/common.c
+++ b/mm/kasan/common.c
@@ -114,12 +114,23 @@ void kasan_unpoison_task_stack(struct task_struct *task)
 /* Unpoison the stack for the current task beyond a watermark sp value. */
 asmlinkage void kasan_unpoison_task_stack_below(const void *watermark)
 {
+	void *base;
+
+	/*
+	 * Only the generic mode needs this. In the tag-based mode the
+	 * compiler fully initializes the shadow of stack variables on
+	 * function entry, so stale tags left below the watermark are
+	 * harmless.
+	 */
+	if (!IS_ENABLED(CONFIG_KASAN_GENERIC))
+		return;
+
 	/*
 	 * Calculate the task stack base address.  Avoid using 'current'
 	 * because this function is called by early resume code which hasn't
 	 * yet set up the percpu register (%gs).
 	 */
-	void *base = (void *)((unsigned long)watermark & ~(THREAD_SIZE - 1));
+	base = (void *)((unsigned long)watermark & ~(THREAD_SIZE - 1));
 
 	kasan_unpoison(base, watermark - base, false);
 }
-- 
2.55.0