Received-SPF: pass (google.com: best guess record for domain of linux-wireless-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Denis Kenzior <denkenz@gmail.com>
To:     johannes@sipsolutions.net, linux-wireless@vger.kernel.org
Cc:     Denis Kenzior <denkenz@gmail.com>
Subject: [RFCv1 2/2] nl80211: Don't split-dump for clients with large buffers
Date:   Thu,  1 Aug 2019 02:14:55 -0500
Message-Id: <20190801071455.4974-2-denkenz@gmail.com>
In-Reply-To: <20190801071455.4974-1-denkenz@gmail.com>
References: <20190801071455.4974-1-denkenz@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-wireless-owner@vger.kernel.org
Precedence: bulk

---
 net/wireless/nl80211.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
index 6774072e836f..e1707cfd9076 100644
--- a/net/wireless/nl80211.c
+++ b/net/wireless/nl80211.c
@@ -2496,6 +2496,22 @@ static int nl80211_dump_wiphy(struct sk_buff *skb, struct netlink_callback *cb)
 			rtnl_unlock();
 			return ret;
 		}
+
+		/*
+		 * auto-detect support for large buffer sizes: af_netlink
+		 * will allocate skbufs larger than 4096 in cases where
+		 * it detects that the client receive buffer (given to
+		 * recvmsg) is bigger.  In such cases we can assume that
+		 * performing split dumps is wasteful since the client
+		 * can likely safely consume the entire un-split wiphy
+		 * message in one go without the extra message header
+		 * overhead.
+		 */
+		if (skb_tailroom(skb) > 4096) {
+			state->large_message = true;
+			state->split = false;
+		}
+
 		cb->args[0] = (long)state;
 	}
 
@@ -2529,6 +2545,7 @@ static int nl80211_dump_wiphy(struct sk_buff *skb, struct netlink_callback *cb)
 				 * We can then retry with the larger buffer.
 				 */
 				if ((ret == -ENOBUFS || ret == -EMSGSIZE) &&
+				    !state->large_message &&
 				    !skb->len && !state->split &&
 				    cb->min_dump_alloc < 4096) {
 					cb->min_dump_alloc = 4096;
-- 
2.21.0