[PATCH] mm/vmalloc: Bail out if in softirq context

["Uladzislau Rezki (Sony)" <urezki@gmail.com>]

vmalloc() is not allowed to be invoked from the NMI,
hard-irq and softirq contexts because doing so may
deadlock on internal locks.

Commit 04aa71da5f35 ("mm/vmalloc: do not trigger BUG()
on BH disabled context") replaced the deprecated in_interrupt()
macro with explicit in_nmi() and in_hardirq() checks, but
unintentionally dropped the softirq context check.

Example:

<snip>
vzalloc() in a task
 spin_lock(&vn->busy.lock)
  <------ Hard irq
   irq_exit()
    process softirq on the same CPU
     vmalloc()
      spin_lock(&vn->busy.lock)
       spins forever
<snip>

Restore the check by checking the caller's context with
the in_serving_softirq() helper. Note that, it is still
allowed to invoke vmalloc() with disabled bottom halves.

Replace BUG_ON() with WARN_ON_ONCE() for in_nmi() and
in_hardirq() case.

Cc: David Hildenbrand (Arm) <david@kernel.org>
Fixes: 04aa71da5f35 ("mm/vmalloc: do not trigger BUG() on BH disabled context")
Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
---
 mm/vmalloc.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 7521b9b3bfd2..36e6e3be6ba4 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -3252,7 +3252,13 @@ struct vm_struct *__get_vm_area_node(unsigned long size,
 	struct vm_struct *area;
 	unsigned long requested_size = size;
 
-	BUG_ON(in_nmi() || in_hardirq());
+	/* IRQ re-entrancy -> irq_exit() -> process softirq. */
+	if (in_serving_softirq())
+		return NULL;
+
+	if (WARN_ON_ONCE(in_nmi() || in_hardirq()))
+		return NULL;
+
 	size = ALIGN(size, 1ul << shift);
 	if (unlikely(!size))
 		return NULL;
-- 
2.47.3