[PATCHv5 1/2] mac80211: avoid extra memcpy in A-MSDU head creation

[PATCHv5 1/2] mac80211: avoid extra memcpy in A-MSDU head creation

[Michael Braun]

Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
---
 net/mac80211/tx.c | 23 +++++++++++++----------
 1 file changed, 13 insertions(+), 10 deletions(-)

diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
index 5023966..56a883b 100644
--- a/net/mac80211/tx.c
+++ b/net/mac80211/tx.c
@@ -3046,11 +3046,11 @@ static bool ieee80211_amsdu_prepare_head(struct ieee80211_sub_if_data *sdata,
 	struct ieee80211_local *local = sdata->local;
 	struct ieee80211_tx_info *info = IEEE80211_SKB_CB(skb);
 	struct ieee80211_hdr *hdr;
-	struct ethhdr amsdu_hdr;
+	struct ethhdr *amsdu_hdr;
 	int hdr_len = fast_tx->hdr_len - sizeof(rfc1042_header);
 	int subframe_len = skb->len - hdr_len;
 	void *data;
-	u8 *qc;
+	u8 *qc, *h_80211_src, *h_80211_dst;
 
 	if (info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE)
 		return false;
@@ -3058,19 +3058,22 @@ static bool ieee80211_amsdu_prepare_head(struct ieee80211_sub_if_data *sdata,
 	if (info->control.flags & IEEE80211_TX_CTRL_AMSDU)
 		return true;
 
-	if (!ieee80211_amsdu_realloc_pad(local, skb, sizeof(amsdu_hdr),
+	if (!ieee80211_amsdu_realloc_pad(local, skb, sizeof(*amsdu_hdr),
 					 &subframe_len))
 		return false;
 
-	amsdu_hdr.h_proto = cpu_to_be16(subframe_len);
-	memcpy(amsdu_hdr.h_source, skb->data + fast_tx->sa_offs, ETH_ALEN);
-	memcpy(amsdu_hdr.h_dest, skb->data + fast_tx->da_offs, ETH_ALEN);
+	data = skb_push(skb, sizeof(*amsdu_hdr));
+	memmove(data, data + sizeof(*amsdu_hdr), hdr_len);
+	hdr = data;
+	amsdu_hdr = data + hdr_len;
+	/* h_80211_src/dst is addr* field within hdr */
+	h_80211_src = data + fast_tx->sa_offs;
+	h_80211_dst = data + fast_tx->da_offs;
 
-	data = skb_push(skb, sizeof(amsdu_hdr));
-	memmove(data, data + sizeof(amsdu_hdr), hdr_len);
-	memcpy(data + hdr_len, &amsdu_hdr, sizeof(amsdu_hdr));
+	amsdu_hdr->h_proto = cpu_to_be16(subframe_len);
+	ether_addr_copy(amsdu_hdr->h_source, h_80211_src);
+	ether_addr_copy(amsdu_hdr->h_dest, h_80211_dst);
 
-	hdr = data;
 	qc = ieee80211_get_qos_ctl(hdr);
 	*qc |= IEEE80211_QOS_CTL_A_MSDU_PRESENT;
 
-- 
2.1.4

[PATCHv5 2/2] mac80211: fix A-MSDU outer SA/DA

[Michael Braun]

According to IEEE 802.11-2012 section 8.3.2 table 8-19, the outer SA/DA
of A-MSDU frames need to be changed depending on FromDS/ToDS values.

Signed-off-by: Michael Braun <michael-dev@fami-braun.de>

--
v5:
 - single out amsdu_hdr to ptr conversion before
v4:
 - h_80211_src/dst has been memmove'd and thus needs to be fixed
v3:
 - write to outer 802.11 header instead of inner amsdu subframe header
v2:
 - avoid the extra write to amsdu_hdr
 - avoid copy of asmdu_hdr into skb, use ptr instead
---
 net/mac80211/tx.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/net/mac80211/tx.c b/net/mac80211/tx.c
index 56a883b..7fcd4b6 100644
--- a/net/mac80211/tx.c
+++ b/net/mac80211/tx.c
@@ -3051,6 +3051,7 @@ static bool ieee80211_amsdu_prepare_head(struct ieee80211_sub_if_data *sdata,
 	int subframe_len = skb->len - hdr_len;
 	void *data;
 	u8 *qc, *h_80211_src, *h_80211_dst;
+	const u8 *bssid;
 
 	if (info->flags & IEEE80211_TX_CTL_RATE_CTRL_PROBE)
 		return false;
@@ -3074,6 +3075,28 @@ static bool ieee80211_amsdu_prepare_head(struct ieee80211_sub_if_data *sdata,
 	ether_addr_copy(amsdu_hdr->h_source, h_80211_src);
 	ether_addr_copy(amsdu_hdr->h_dest, h_80211_dst);
 
+	/* according to IEEE 802.11-2012 8.3.2 table 8-19, the outer SA/DA
+	 * fields needs to be changed to BSSID for A-MSDU frames depending
+	 * on FromDS/ToDS values.
+	 */
+	switch (sdata->vif.type) {
+	case NL80211_IFTYPE_STATION:
+		bssid = sdata->u.mgd.bssid;
+		break;
+	case NL80211_IFTYPE_AP:
+	case NL80211_IFTYPE_AP_VLAN:
+		bssid = sdata->vif.addr;
+		break;
+	default:
+		bssid = NULL;
+	}
+
+	if (bssid && ieee80211_has_fromds(hdr->frame_control))
+		memcpy(h_80211_src, bssid, ETH_ALEN);
+
+	if (bssid && ieee80211_has_tods(hdr->frame_control))
+		memcpy(h_80211_dst, bssid, ETH_ALEN);
+
 	qc = ieee80211_get_qos_ctl(hdr);
 	*qc |= IEEE80211_QOS_CTL_A_MSDU_PRESENT;
 
-- 
2.1.4

Re: [PATCHv5 1/2] mac80211: avoid extra memcpy in A-MSDU head creation

[Johannes Berg]

On Sat, 2016-10-15 at 13:28 +0200, Michael Braun wrote:
> Signed-off-by: Michael Braun <michael-dev@fami-braun.de>
> 
Applied.

johannes

Re: [PATCHv5 2/2] mac80211: fix A-MSDU outer SA/DA

[Johannes Berg]

On Sat, 2016-10-15 at 13:28 +0200, Michael Braun wrote:
> According to IEEE 802.11-2012 section 8.3.2 table 8-19, the outer
> SA/DA
> of A-MSDU frames need to be changed depending on FromDS/ToDS values.

Also applied,

> +	if (bssid && ieee80211_has_fromds(hdr->frame_control))
> +		memcpy(h_80211_src, bssid, ETH_ALEN);
> +
> +	if (bssid && ieee80211_has_tods(hdr->frame_control))
> +		memcpy(h_80211_dst, bssid, ETH_ALEN);

but I also changed these to ether_addr_copy()

johannes

Re: [Projekt-wlan] [PATCHv5 2/2] mac80211: fix A-MSDU outer SA/DA

[michael-dev]

Am 17.10.2016 11:44, schrieb Johannes Berg:
>> +	if (bssid && ieee80211_has_fromds(hdr->frame_control))
>> +		memcpy(h_80211_src, bssid, ETH_ALEN);
>> +
>> +	if (bssid && ieee80211_has_tods(hdr->frame_control))
>> +		memcpy(h_80211_dst, bssid, ETH_ALEN);
> 
> but I also changed these to ether_addr_copy()

ether_addr_copy requires both arguments to be __aligned(2) according to 
include/linux/etherdevice.h.
bssid might be sdata->u.mgd.bssid, which is bssid in struct 
ieee80211_if_managed, but neither sdata->u, sdata->u.mgd nor 
sdata->u.mgd.bssid seem to be declared as __aligned(2).
So ether_addr_copy should not be used.

What am I missing?

Regards,
M. Braun

Re: [Projekt-wlan] [PATCHv5 2/2] mac80211: fix A-MSDU outer SA/DA

[Johannes Berg]

On Mon, 2016-10-17 at 12:29 +0200, michael-dev wrote:
> 
> ether_addr_copy requires both arguments to be __aligned(2) according
> to 
> include/linux/etherdevice.h.
> bssid might be sdata->u.mgd.bssid, which is bssid in struct 
> ieee80211_if_managed, but neither sdata->u, sdata->u.mgd nor 
> sdata->u.mgd.bssid seem to be declared as __aligned(2).So
> ether_addr_copy should not be used.

They are in fact aligned today, given that they occur after a pointer.
However, I did add the __aligned(2) annotation to make that plenty
clear for the future.

johannes