net: simplify sk_page_frag
authorBenjamin Coddington <bcodding@redhat.com>
Fri, 16 Dec 2022 12:45:28 +0000 (07:45 -0500)
committerJakub Kicinski <kuba@kernel.org>
Tue, 20 Dec 2022 01:28:50 +0000 (17:28 -0800)
Now that in-kernel socket users that may recurse during reclaim have benn
converted to sk_use_task_frag = false, we can have sk_page_frag() simply
check that value.

Signed-off-by: Benjamin Coddington <bcodding@redhat.com>
Reviewed-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
include/net/sock.h

index fefe1f4abf191de3e0b8719e5e50d83a0e28ef20..dcd72e6285b23006051d651630bdd966741cbb01 100644 (file)
@@ -2564,19 +2564,14 @@ static inline void sk_stream_moderate_sndbuf(struct sock *sk)
  * Both direct reclaim and page faults can nest inside other
  * socket operations and end up recursing into sk_page_frag()
  * while it's already in use: explicitly avoid task page_frag
- * usage if the caller is potentially doing any of them.
- * This assumes that page fault handlers use the GFP_NOFS flags or
- * explicitly disable sk_use_task_frag.
+ * when users disable sk_use_task_frag.
  *
  * Return: a per task page_frag if context allows that,
  * otherwise a per socket one.
  */
 static inline struct page_frag *sk_page_frag(struct sock *sk)
 {
-       if (sk->sk_use_task_frag &&
-           (sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC |
-                                 __GFP_FS)) ==
-           (__GFP_DIRECT_RECLAIM | __GFP_FS))
+       if (sk->sk_use_task_frag)
                return &current->task_frag;
 
        return &sk->sk_frag;