[PATCH net-next 0/7] net: move skb->dropcount to skb->cb[]

[PATCH net-next 0/7] net: move skb->dropcount to skb->cb[]

[Eyal Birger]

Commit 977750076d98 ("af_packet: add interframe drop cmsg (v6)")
unionized skb->mark and skb->dropcount in order to allow recording
of the socket drop count while maintaining struct sk_buff size.

skb->dropcount was introduced since there was no available room
in skb->cb[] in packet sockets. However, its introduction led to
the inability to export skb->mark to userspace.

It was considered to alias skb->priority instead of skb->mark.
However, that would lead to the inabilty to export skb->priority
to userspace if desired. Such change may also lead to hard-to-find
issues as skb->priority is assumed to be alias free, and, as noted
by Shmulik Ladkani, is not 'naturally orthogonal' with other skb
fields.

This patch series follows the suggestions made by Eric Dumazet moving
the dropcount metric to skb->cb[], eliminating this problem
at the expense of 4 bytes less in skb->cb[] for protocol families
using it.

The patch series include compactization of bluetooth and packet
use of skb->cb[] as well as the infrastructure for placing dropcount
in skb->cb[].

Eyal Birger (7):
  net: bluetooth: compact struct bt_skb_cb by inlining struct
    hci_req_ctrl
  net: bluetooth: compact struct bt_skb_cb by converting boolean fields
    to bit fields
  net: rxrpc: change call to sock_recv_ts_and_drops() on rxrpc recvmsg
    to sock_recv_timestamp()
  net: packet: use skb->dev as storage for skb orig len instead of
    skb->cb[]
  net: use common macro for assering skb->cb[] available size in
    protocol families
  net: add common accessor for setting dropcount on packets
  net: move skb->dropcount to skb->cb[]

 include/linux/skbuff.h            |  2 --
 include/net/bluetooth/bluetooth.h | 14 +++++---------
 include/net/sock.h                | 22 ++++++++++++++++++++++
 net/bluetooth/af_bluetooth.c      |  3 +--
 net/bluetooth/hci_core.c          | 12 ++++++------
 net/bluetooth/hci_event.c         |  4 ++--
 net/bluetooth/hci_request.c       |  6 +++---
 net/bluetooth/hci_sock.c          |  2 +-
 net/can/bcm.c                     |  2 +-
 net/can/raw.c                     |  6 +++---
 net/core/sock.c                   |  2 +-
 net/ipv4/af_inet.c                |  2 +-
 net/ipv4/tcp.c                    |  3 +--
 net/ipv6/af_inet6.c               |  2 +-
 net/packet/af_packet.c            | 12 +++++-------
 net/rxrpc/ar-recvmsg.c            |  2 +-
 net/sctp/protocol.c               |  3 +--
 net/socket.c                      |  4 ++--
 18 files changed, 57 insertions(+), 46 deletions(-)

-- 
2.1.4

[PATCH net-next 1/7] net: bluetooth: compact struct bt_skb_cb by inlining struct hci_req_ctrl

[Eyal Birger]

struct hci_req_ctrl is never used outside of struct bt_skb_cb;
Inlining it frees 8 bytes on a 64 bit system in skb->cb[] allowing
the addition of more ancillary data.

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 include/net/bluetooth/bluetooth.h | 10 +++-------
 net/bluetooth/hci_core.c          | 12 ++++++------
 net/bluetooth/hci_event.c         |  4 ++--
 net/bluetooth/hci_request.c       |  6 +++---
 net/bluetooth/hci_sock.c          |  2 +-
 5 files changed, 15 insertions(+), 19 deletions(-)

diff --git a/include/net/bluetooth/bluetooth.h b/include/net/bluetooth/bluetooth.h
index e00455a..0989366 100644
--- a/include/net/bluetooth/bluetooth.h
+++ b/include/net/bluetooth/bluetooth.h
@@ -275,21 +275,17 @@ struct hci_dev;
 
 typedef void (*hci_req_complete_t)(struct hci_dev *hdev, u8 status, u16 opcode);
 
-struct hci_req_ctrl {
-	bool			start;
-	u8			event;
-	hci_req_complete_t	complete;
-};
-
 struct bt_skb_cb {
 	__u8 pkt_type;
 	__u8 incoming;
 	__u16 opcode;
 	__u16 expect;
 	__u8 force_active;
+	bool req_start;
+	u8 req_event;
+	hci_req_complete_t req_complete;
 	struct l2cap_chan *chan;
 	struct l2cap_ctrl control;
-	struct hci_req_ctrl req;
 	bdaddr_t bdaddr;
 	__le16 psm;
 };
diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 3322d3f..85a0655 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -3517,7 +3517,7 @@ int hci_send_cmd(struct hci_dev *hdev, __u16 opcode, __u32 plen,
 	/* Stand-alone HCI commands must be flagged as
 	 * single-command requests.
 	 */
-	bt_cb(skb)->req.start = true;
+	bt_cb(skb)->req_start = true;
 
 	skb_queue_tail(&hdev->cmd_q, skb);
 	queue_work(hdev->workqueue, &hdev->cmd_work);
@@ -4195,7 +4195,7 @@ static bool hci_req_is_complete(struct hci_dev *hdev)
 	if (!skb)
 		return true;
 
-	return bt_cb(skb)->req.start;
+	return bt_cb(skb)->req_start;
 }
 
 static void hci_resend_last(struct hci_dev *hdev)
@@ -4255,14 +4255,14 @@ void hci_req_cmd_complete(struct hci_dev *hdev, u16 opcode, u8 status)
 	 * command queue (hdev->cmd_q).
 	 */
 	if (hdev->sent_cmd) {
-		req_complete = bt_cb(hdev->sent_cmd)->req.complete;
+		req_complete = bt_cb(hdev->sent_cmd)->req_complete;
 
 		if (req_complete) {
 			/* We must set the complete callback to NULL to
 			 * avoid calling the callback more than once if
 			 * this function gets called again.
 			 */
-			bt_cb(hdev->sent_cmd)->req.complete = NULL;
+			bt_cb(hdev->sent_cmd)->req_complete = NULL;
 
 			goto call_complete;
 		}
@@ -4271,12 +4271,12 @@ void hci_req_cmd_complete(struct hci_dev *hdev, u16 opcode, u8 status)
 	/* Remove all pending commands belonging to this request */
 	spin_lock_irqsave(&hdev->cmd_q.lock, flags);
 	while ((skb = __skb_dequeue(&hdev->cmd_q))) {
-		if (bt_cb(skb)->req.start) {
+		if (bt_cb(skb)->req_start) {
 			__skb_queue_head(&hdev->cmd_q, skb);
 			break;
 		}
 
-		req_complete = bt_cb(skb)->req.complete;
+		req_complete = bt_cb(skb)->req_complete;
 		kfree_skb(skb);
 	}
 	spin_unlock_irqrestore(&hdev->cmd_q.lock, flags);
diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
index a3fb094..8e8c433 100644
--- a/net/bluetooth/hci_event.c
+++ b/net/bluetooth/hci_event.c
@@ -3106,7 +3106,7 @@ static void hci_cmd_status_evt(struct hci_dev *hdev, struct sk_buff *skb)
 		cancel_delayed_work(&hdev->cmd_timer);
 
 	if (ev->status ||
-	    (hdev->sent_cmd && !bt_cb(hdev->sent_cmd)->req.event))
+	    (hdev->sent_cmd && !bt_cb(hdev->sent_cmd)->req_event))
 		hci_req_cmd_complete(hdev, opcode, ev->status);
 
 	if (ev->ncmd && !test_bit(HCI_RESET, &hdev->flags)) {
@@ -5039,7 +5039,7 @@ void hci_event_packet(struct hci_dev *hdev, struct sk_buff *skb)
 
 	skb_pull(skb, HCI_EVENT_HDR_SIZE);
 
-	if (hdev->sent_cmd && bt_cb(hdev->sent_cmd)->req.event == event) {
+	if (hdev->sent_cmd && bt_cb(hdev->sent_cmd)->req_event == event) {
 		struct hci_command_hdr *cmd_hdr = (void *) hdev->sent_cmd->data;
 		u16 opcode = __le16_to_cpu(cmd_hdr->opcode);
 
diff --git a/net/bluetooth/hci_request.c b/net/bluetooth/hci_request.c
index b59f92c..db2f45a 100644
--- a/net/bluetooth/hci_request.c
+++ b/net/bluetooth/hci_request.c
@@ -55,7 +55,7 @@ int hci_req_run(struct hci_request *req, hci_req_complete_t complete)
 		return -ENODATA;
 
 	skb = skb_peek_tail(&req->cmd_q);
-	bt_cb(skb)->req.complete = complete;
+	bt_cb(skb)->req_complete = complete;
 
 	spin_lock_irqsave(&hdev->cmd_q.lock, flags);
 	skb_queue_splice_tail(&req->cmd_q, &hdev->cmd_q);
@@ -116,9 +116,9 @@ void hci_req_add_ev(struct hci_request *req, u16 opcode, u32 plen,
 	}
 
 	if (skb_queue_empty(&req->cmd_q))
-		bt_cb(skb)->req.start = true;
+		bt_cb(skb)->req_start = true;
 
-	bt_cb(skb)->req.event = event;
+	bt_cb(skb)->req_event = event;
 
 	skb_queue_tail(&req->cmd_q, skb);
 }
diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
index 1d65c5b..f003818 100644
--- a/net/bluetooth/hci_sock.c
+++ b/net/bluetooth/hci_sock.c
@@ -965,7 +965,7 @@ static int hci_sock_sendmsg(struct kiocb *iocb, struct socket *sock,
 			/* Stand-alone HCI commands must be flagged as
 			 * single-command requests.
 			 */
-			bt_cb(skb)->req.start = true;
+			bt_cb(skb)->req_start = true;
 
 			skb_queue_tail(&hdev->cmd_q, skb);
 			queue_work(hdev->workqueue, &hdev->cmd_work);
-- 
2.1.4

Re: [PATCH net-next 1/7] net: bluetooth: compact struct bt_skb_cb by inlining struct hci_req_ctrl

[Shmulik Ladkani]

On Thu, 26 Feb 2015 04:10:06 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
> struct hci_req_ctrl is never used outside of struct bt_skb_cb;
> Inlining it frees 8 bytes on a 64 bit system in skb->cb[] allowing
> the addition of more ancillary data.
> 
> Signed-off-by: Eyal Birger <eyal.birger@gmail.com>

Reviewed-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>

Re: [PATCH net-next 1/7] net: bluetooth: compact struct bt_skb_cb by inlining struct hci_req_ctrl

[Marcel Holtmann]

Hi Eyal,

> struct hci_req_ctrl is never used outside of struct bt_skb_cb;
> Inlining it frees 8 bytes on a 64 bit system in skb->cb[] allowing
> the addition of more ancillary data.
> 
> Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
> ---
> include/net/bluetooth/bluetooth.h | 10 +++-------
> net/bluetooth/hci_core.c          | 12 ++++++------
> net/bluetooth/hci_event.c         |  4 ++--
> net/bluetooth/hci_request.c       |  6 +++---
> net/bluetooth/hci_sock.c          |  2 +-
> 5 files changed, 15 insertions(+), 19 deletions(-)

it would be a really good idea to actually send this to linux-bluetooth@vger.kernel.org for review.

Regards

Marcel

Re: [PATCH net-next 1/7] net: bluetooth: compact struct bt_skb_cb by inlining struct hci_req_ctrl

[Eyal Birger]

Hi Marcel,

On Thu, Feb 26, 2015 at 6:15 PM, Marcel Holtmann <marcel@holtmann.org> wrote:
> Hi Eyal,
>
>> struct hci_req_ctrl is never used outside of struct bt_skb_cb;
>> Inlining it frees 8 bytes on a 64 bit system in skb->cb[] allowing
>> the addition of more ancillary data.
>>
>> Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
>> ---
>> include/net/bluetooth/bluetooth.h | 10 +++-------
>> net/bluetooth/hci_core.c          | 12 ++++++------
>> net/bluetooth/hci_event.c         |  4 ++--
>> net/bluetooth/hci_request.c       |  6 +++---
>> net/bluetooth/hci_sock.c          |  2 +-
>> 5 files changed, 15 insertions(+), 19 deletions(-)
>
> it would be a really good idea to actually send this to linux-bluetooth@vger.kernel.org for review.
>

Thanks! I have done that in v2.

Eyal.

[PATCH net-next 2/7] net: bluetooth: compact struct bt_skb_cb by converting boolean fields to bit fields

[Eyal Birger]

On 64 bit systems, struct bt_skb_cb size is padded by 6 bytes in order
to reach 8 byte alignment.
Convert boolean fields force_active, incoming and req_start to bit fields
in order to eliminate the padding.

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 include/net/bluetooth/bluetooth.h | 6 +++---
 net/bluetooth/hci_core.c          | 2 +-
 net/bluetooth/hci_request.c       | 2 +-
 net/bluetooth/hci_sock.c          | 2 +-
 4 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/include/net/bluetooth/bluetooth.h b/include/net/bluetooth/bluetooth.h
index 0989366..bb1851e 100644
--- a/include/net/bluetooth/bluetooth.h
+++ b/include/net/bluetooth/bluetooth.h
@@ -277,11 +277,11 @@ typedef void (*hci_req_complete_t)(struct hci_dev *hdev, u8 status, u16 opcode);
 
 struct bt_skb_cb {
 	__u8 pkt_type;
-	__u8 incoming;
 	__u16 opcode;
 	__u16 expect;
-	__u8 force_active;
-	bool req_start;
+	__u8 force_active:1;
+	__u8 incoming:1;
+	__u8 req_start:1;
 	u8 req_event;
 	hci_req_complete_t req_complete;
 	struct l2cap_chan *chan;
diff --git a/net/bluetooth/hci_core.c b/net/bluetooth/hci_core.c
index 85a0655..80f40e8 100644
--- a/net/bluetooth/hci_core.c
+++ b/net/bluetooth/hci_core.c
@@ -3517,7 +3517,7 @@ int hci_send_cmd(struct hci_dev *hdev, __u16 opcode, __u32 plen,
 	/* Stand-alone HCI commands must be flagged as
 	 * single-command requests.
 	 */
-	bt_cb(skb)->req_start = true;
+	bt_cb(skb)->req_start = 1;
 
 	skb_queue_tail(&hdev->cmd_q, skb);
 	queue_work(hdev->workqueue, &hdev->cmd_work);
diff --git a/net/bluetooth/hci_request.c b/net/bluetooth/hci_request.c
index db2f45a..f857e76 100644
--- a/net/bluetooth/hci_request.c
+++ b/net/bluetooth/hci_request.c
@@ -116,7 +116,7 @@ void hci_req_add_ev(struct hci_request *req, u16 opcode, u32 plen,
 	}
 
 	if (skb_queue_empty(&req->cmd_q))
-		bt_cb(skb)->req_start = true;
+		bt_cb(skb)->req_start = 1;
 
 	bt_cb(skb)->req_event = event;
 
diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
index f003818..37198fb 100644
--- a/net/bluetooth/hci_sock.c
+++ b/net/bluetooth/hci_sock.c
@@ -965,7 +965,7 @@ static int hci_sock_sendmsg(struct kiocb *iocb, struct socket *sock,
 			/* Stand-alone HCI commands must be flagged as
 			 * single-command requests.
 			 */
-			bt_cb(skb)->req_start = true;
+			bt_cb(skb)->req_start = 1;
 
 			skb_queue_tail(&hdev->cmd_q, skb);
 			queue_work(hdev->workqueue, &hdev->cmd_work);
-- 
2.1.4

Re: [PATCH net-next 2/7] net: bluetooth: compact struct bt_skb_cb by converting boolean fields to bit fields

[Shmulik Ladkani]

On Thu, 26 Feb 2015 04:10:07 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
> @@ -277,11 +277,11 @@ typedef void (*hci_req_complete_t)(struct hci_dev *hdev, u8 status, u16 opcode);
>  
>  struct bt_skb_cb {
>  	__u8 pkt_type;
> -	__u8 incoming;
>  	__u16 opcode;
>  	__u16 expect;
> -	__u8 force_active;
> -	bool req_start;
> +	__u8 force_active:1;
> +	__u8 incoming:1;
> +	__u8 req_start:1;
>  	u8 req_event;
>  	hci_req_complete_t req_complete;
>  	struct l2cap_chan *chan;

There's a:
    bt_cb(skb)->force_active = BT_POWER_FORCE_ACTIVE_ON;
assignment in l2cap_core.c.

No bluetooth expert, no idea if BT_POWER_FORCE_ACTIVE_ON is 1 by definition.
If not, maybe prefer:
    bt_cb(skb)->force_active = !!BT_POWER_FORCE_ACTIVE_ON

Besides that,
Reviewed-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>

Re: [PATCH net-next 2/7] net: bluetooth: compact struct bt_skb_cb by converting boolean fields to bit fields

[Eyal Birger]

On Thu, Feb 26, 2015 at 9:30 AM, Shmulik Ladkani
<shmulik.ladkani@gmail.com> wrote:
> On Thu, 26 Feb 2015 04:10:07 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
>> @@ -277,11 +277,11 @@ typedef void (*hci_req_complete_t)(struct hci_dev *hdev, u8 status, u16 opcode);
>>
>>  struct bt_skb_cb {
>>       __u8 pkt_type;
>> -     __u8 incoming;
>>       __u16 opcode;
>>       __u16 expect;
>> -     __u8 force_active;
>> -     bool req_start;
>> +     __u8 force_active:1;
>> +     __u8 incoming:1;
>> +     __u8 req_start:1;
>>       u8 req_event;
>>       hci_req_complete_t req_complete;
>>       struct l2cap_chan *chan;
>
> There's a:
>     bt_cb(skb)->force_active = BT_POWER_FORCE_ACTIVE_ON;
> assignment in l2cap_core.c.
>
> No bluetooth expert, no idea if BT_POWER_FORCE_ACTIVE_ON is 1 by definition.
> If not, maybe prefer:
>     bt_cb(skb)->force_active = !!BT_POWER_FORCE_ACTIVE_ON
>

I think I can keep force_active as __u8.

Will update on v2.

Thanks!

> Besides that,
> Reviewed-by: Shmulik Ladkani <shmulik.ladkani@gmail.com>

RE: [PATCH net-next 2/7] net: bluetooth: compact struct bt_skb_cb by converting boolean fields to bit fields

[David Laight]

From: Eyal Birger
> On 64 bit systems, struct bt_skb_cb size is padded by 6 bytes in order
> to reach 8 byte alignment.
> Convert boolean fields force_active, incoming and req_start to bit fields
> in order to eliminate the padding.
...
> +++ b/include/net/bluetooth/bluetooth.h
...
>  struct bt_skb_cb {
>  	__u8 pkt_type;
> -	__u8 incoming;
>  	__u16 opcode;
>  	__u16 expect;
> -	__u8 force_active;
> -	bool req_start;
> +	__u8 force_active:1;
> +	__u8 incoming:1;
> +	__u8 req_start:1;
>  	u8 req_event;
>  	hci_req_complete_t req_complete;
>  	struct l2cap_chan *chan;

You've generated some pad bytes, best to put everything on its natural alignment.
The old version had req_complete at offset 9, the new one at offset 8.
Not sure how this removes 6 bytes of pad.

	David

Re: [PATCH net-next 2/7] net: bluetooth: compact struct bt_skb_cb by converting boolean fields to bit fields

[Eyal Birger]

Hi,

On Thu, Feb 26, 2015 at 1:31 PM, David Laight <David.Laight@aculab.com> wrote:
> From: Eyal Birger
>> On 64 bit systems, struct bt_skb_cb size is padded by 6 bytes in order
>> to reach 8 byte alignment.
>> Convert boolean fields force_active, incoming and req_start to bit fields
>> in order to eliminate the padding.
> ...
>> +++ b/include/net/bluetooth/bluetooth.h
> ...
>>  struct bt_skb_cb {
>>       __u8 pkt_type;
>> -     __u8 incoming;
>>       __u16 opcode;
>>       __u16 expect;
>> -     __u8 force_active;
>> -     bool req_start;
>> +     __u8 force_active:1;
>> +     __u8 incoming:1;
>> +     __u8 req_start:1;
>>       u8 req_event;
>>       hci_req_complete_t req_complete;
>>       struct l2cap_chan *chan;
>
> You've generated some pad bytes, best to put everything on its natural alignment.
> The old version had req_complete at offset 9, the new one at offset 8.
> Not sure how this removes 6 bytes of pad.
>

Hmm, I tried different variations here. At a certain point, I had 6
bytes of padding.
Seems pahole is playing peekaboo on my holes :)

Anyway, since req_complete has to be pointer aligned, it does not
reside at offset 9,
but at offset 16, which creates a hole.

Thanks. I will update appropriately in v2.

[PATCH net-next 3/7] net: rxrpc: change call to sock_recv_ts_and_drops() on rxrpc recvmsg to sock_recv_timestamp()

[Eyal Birger]

Commit 3b885787ea4112 ("net: Generalize socket rx gap / receive queue overflow cmsg")
allowed receiving packet dropcount information as a socket level option.
RXRPC sockets recvmsg function was changed to support this by calling
sock_recv_ts_and_drops() instead of sock_recv_timestamp().

However, protocol families wishing to receive dropcount should call
sock_queue_rcv_skb() or set the dropcount specifically (as done
in packet_rcv()). This was not done for rxrpc and thus this feature
never worked on these sockets.

Formalizing this by not calling sock_recv_ts_and_drops() in rxrpc as
part of an effort to move skb->dropcount into skb->cb[]

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 net/rxrpc/ar-recvmsg.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/rxrpc/ar-recvmsg.c b/net/rxrpc/ar-recvmsg.c
index 4575485..d58ba70 100644
--- a/net/rxrpc/ar-recvmsg.c
+++ b/net/rxrpc/ar-recvmsg.c
@@ -150,7 +150,7 @@ int rxrpc_recvmsg(struct kiocb *iocb, struct socket *sock,
 				       &call->conn->trans->peer->srx, len);
 				msg->msg_namelen = len;
 			}
-			sock_recv_ts_and_drops(msg, &rx->sk, skb);
+			sock_recv_timestamp(msg, &rx->sk, skb);
 		}
 
 		/* receive the message */
-- 
2.1.4

Re: [PATCH net-next 3/7] net: rxrpc: change call to sock_recv_ts_and_drops() on rxrpc recvmsg to sock_recv_timestamp()

[Eric Dumazet]

On Thu, 2015-02-26 at 04:10 +0200, Eyal Birger wrote:
> Commit 3b885787ea4112 ("net: Generalize socket rx gap / receive queue overflow cmsg")
> allowed receiving packet dropcount information as a socket level option.
> RXRPC sockets recvmsg function was changed to support this by calling
> sock_recv_ts_and_drops() instead of sock_recv_timestamp().
> 
> However, protocol families wishing to receive dropcount should call
> sock_queue_rcv_skb() or set the dropcount specifically (as done
> in packet_rcv()). This was not done for rxrpc and thus this feature
> never worked on these sockets.
> 
> Formalizing this by not calling sock_recv_ts_and_drops() in rxrpc as
> part of an effort to move skb->dropcount into skb->cb[]
> 
> Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
> ---
>  net/rxrpc/ar-recvmsg.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/rxrpc/ar-recvmsg.c b/net/rxrpc/ar-recvmsg.c
> index 4575485..d58ba70 100644
> --- a/net/rxrpc/ar-recvmsg.c
> +++ b/net/rxrpc/ar-recvmsg.c
> @@ -150,7 +150,7 @@ int rxrpc_recvmsg(struct kiocb *iocb, struct socket *sock,
>  				       &call->conn->trans->peer->srx, len);
>  				msg->msg_namelen = len;
>  			}
> -			sock_recv_ts_and_drops(msg, &rx->sk, skb);
> +			sock_recv_timestamp(msg, &rx->sk, skb);
>  		}

I guess this should be the last patch of the serie, as this will
overwrite skb->mark, and rxrpc seems to care about skb->mark 

Re: [PATCH net-next 3/7] net: rxrpc: change call to sock_recv_ts_and_drops() on rxrpc recvmsg to sock_recv_timestamp()

[Eyal Birger]

>> However, protocol families wishing to receive dropcount should call
>> sock_queue_rcv_skb() or set the dropcount specifically (as done
>> in packet_rcv()). This was not done for rxrpc and thus this feature
>> never worked on these sockets.
>>
>> Formalizing this by not calling sock_recv_ts_and_drops() in rxrpc as
>> part of an effort to move skb->dropcount into skb->cb[]
>>
>> Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
>> ---
>>  net/rxrpc/ar-recvmsg.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/net/rxrpc/ar-recvmsg.c b/net/rxrpc/ar-recvmsg.c
>> index 4575485..d58ba70 100644
>> --- a/net/rxrpc/ar-recvmsg.c
>> +++ b/net/rxrpc/ar-recvmsg.c
>> @@ -150,7 +150,7 @@ int rxrpc_recvmsg(struct kiocb *iocb, struct socket *sock,
>>                                      &call->conn->trans->peer->srx, len);
>>                               msg->msg_namelen = len;
>>                       }
>> -                     sock_recv_ts_and_drops(msg, &rx->sk, skb);
>> +                     sock_recv_timestamp(msg, &rx->sk, skb);
>>               }
>
> I guess this should be the last patch of the serie, as this will
> overwrite skb->mark, and rxrpc seems to care about skb->mark
>

Maybe I didn't understand your comment. Note this patch does not fix
SO_RXQ_OVFL on rxrpc sockets.

I essentially reverted 3b885787ea4112 for these sockets.

It may be possible to compact rxrpc skb->cb[] usage by tricking the size
of the resend_at variable as you suggested.

However, since SO_RXQ_OVFL never really worked on these sockets,
I limited the patch series scope to moving skb->dropcount; This patch only
signals that rxrpc can't support this feature unless some space is freed in its
skb->cb[] use.

Re: [PATCH net-next 3/7] net: rxrpc: change call to sock_recv_ts_and_drops() on rxrpc recvmsg to sock_recv_timestamp()

[Eric Dumazet]

On Thu, 2015-02-26 at 09:30 +0200, Eyal Birger wrote:

> Maybe I didn't understand your comment. Note this patch does not fix
> SO_RXQ_OVFL on rxrpc sockets.
> 
> I essentially reverted 3b885787ea4112 for these sockets.
> 
> It may be possible to compact rxrpc skb->cb[] usage by tricking the size
> of the resend_at variable as you suggested.
> 
> However, since SO_RXQ_OVFL never really worked on these sockets,
> I limited the patch series scope to moving skb->dropcount; This patch only
> signals that rxrpc can't support this feature unless some space is freed in its
> skb->cb[] use.
> --

My point is : If you stick this patch early in the serie, then skb->mark
will be overwritten by sock_recv_timestamp() (to store skb->dropcount)
and rxrpc breaks as it actively relies on skb->mark (apparently)

We dont care if SO_RXQ_OVFL is broken for rxprpc : Nobody noticed yet.

But breaking rxrpc skb->mark early in the serie does not help bisection
if needed later.

Re: [PATCH net-next 3/7] net: rxrpc: change call to sock_recv_ts_and_drops() on rxrpc recvmsg to sock_recv_timestamp()

[Eyal Birger]

On Thu, Feb 26, 2015 at 4:31 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Thu, 2015-02-26 at 09:30 +0200, Eyal Birger wrote:
>
>> Maybe I didn't understand your comment. Note this patch does not fix
>> SO_RXQ_OVFL on rxrpc sockets.
>>
>> I essentially reverted 3b885787ea4112 for these sockets.
>>
>> It may be possible to compact rxrpc skb->cb[] usage by tricking the size
>> of the resend_at variable as you suggested.
>>
>> However, since SO_RXQ_OVFL never really worked on these sockets,
>> I limited the patch series scope to moving skb->dropcount; This patch only
>> signals that rxrpc can't support this feature unless some space is freed in its
>> skb->cb[] use.
>> --
>
> My point is : If you stick this patch early in the serie, then skb->mark
> will be overwritten by sock_recv_timestamp() (to store skb->dropcount)
> and rxrpc breaks as it actively relies on skb->mark (apparently)
>

skb->dropcount is stored prior to enqueue in sock_queue_rcv_skb().
This function is not called in rxrpc.

sock_recv_timestamp() is called from the recvmsg code, and does not alter
skb->dropcount.

What am I missing?

> We dont care if SO_RXQ_OVFL is broken for rxprpc : Nobody noticed yet.
>

This was my point in this patch - to make it explicit. When rxrpc called
sock_recv_ts_and_drops() it allegedly supported receiving drops, when
dropcount was never set.

> But breaking rxrpc skb->mark early in the serie does not help bisection
> if needed later.
>

[PATCH net-next 4/7] net: packet: use skb->dev as storage for skb orig len instead of skb->cb[]

[Eyal Birger]

As part of an effort to move skb->dropcount to skb->cb[], 4 bytes
of additional room are needed in skb->cb[] in packet sockets.

Store the skb original length in skb->dev instead of skb->cb[] for
this purpose.

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 net/packet/af_packet.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 9c28cec..26889fd 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -216,7 +216,6 @@ static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
 static void packet_flush_mclist(struct sock *sk);
 
 struct packet_skb_cb {
-	unsigned int origlen;
 	union {
 		struct sockaddr_pkt pkt;
 		struct sockaddr_ll ll;
@@ -224,6 +223,7 @@ struct packet_skb_cb {
 };
 
 #define PACKET_SKB_CB(__skb)	((struct packet_skb_cb *)((__skb)->cb))
+#define PACKET_SKB_ORIGLEN(__skb) ((unsigned long *)&(__skb)->dev)
 
 #define GET_PBDQC_FROM_RB(x)	((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
 #define GET_PBLOCK_DESC(x, bid)	\
@@ -1825,13 +1825,12 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
 
 	sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
 
-	PACKET_SKB_CB(skb)->origlen = skb->len;
+	*PACKET_SKB_ORIGLEN(skb) = skb->len;
 
 	if (pskb_trim(skb, snaplen))
 		goto drop_n_acct;
 
 	skb_set_owner_r(skb, sk);
-	skb->dev = NULL;
 	skb_dst_drop(skb);
 
 	/* drop conntrack reference */
@@ -3006,7 +3005,7 @@ static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
 		aux.tp_status = TP_STATUS_USER;
 		if (skb->ip_summed == CHECKSUM_PARTIAL)
 			aux.tp_status |= TP_STATUS_CSUMNOTREADY;
-		aux.tp_len = PACKET_SKB_CB(skb)->origlen;
+		aux.tp_len = *PACKET_SKB_ORIGLEN(skb);
 		aux.tp_snaplen = skb->len;
 		aux.tp_mac = 0;
 		aux.tp_net = skb_network_offset(skb);
-- 
2.1.4

Re: [PATCH net-next 4/7] net: packet: use skb->dev as storage for skb orig len instead of skb->cb[]

[Eyal Birger]

>
>         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
>
> -       PACKET_SKB_CB(skb)->origlen = skb->len;
> +       *PACKET_SKB_ORIGLEN(skb) = skb->len;
>
>         if (pskb_trim(skb, snaplen))
>                 goto drop_n_acct;
>
>         skb_set_owner_r(skb, sk);
> -       skb->dev = NULL;
>         skb_dst_drop(skb);

There is a problem here. skb->dev must still be valid on the call to
skb_set_owner_r().

Sorry about that. I'll spin a second version along with any other
comments raised on this series.

Eyal.

Re: [PATCH net-next 4/7] net: packet: use skb->dev as storage for skb orig len instead of skb->cb[]

[Willem de Bruijn]

On Wed, Feb 25, 2015 at 9:10 PM, Eyal Birger <eyal.birger@gmail.com> wrote:
> As part of an effort to move skb->dropcount to skb->cb[], 4 bytes
> of additional room are needed in skb->cb[] in packet sockets.
>
> Store the skb original length in skb->dev instead of skb->cb[] for
> this purpose.

Another option is to delay preparation of the full sockaddr_ll struct until
when it's needed. It often is not used at all (if msg_name == NULL).

sll_family, sll_protocol and sll_pkttype can be derived when needed
in packet_recvmsg. The first two are the head of sockaddr_ll, so
a shorter struct filled in in packet_rcv only from sll_ifindex onwards
would save the 4B needed for origlen. Actually, a union is simpler:

struct __packet_cb_sockaddr_ll {
        union {
                struct {
                        unsigned short  sll_family;
                        __be16          sll_protocol;
                 }
                 unsigned int sll_origlen;
        }
    /* etc.. */
};

The family and protocol can just overwrite sll_origlen in packet_recvmsg
before the memcpy into msg->msg_name.

Also interesting would be to avoid copying the address in the fast path
as it may never be used, especially for long addresses (INFINIBAND_ALEN
and FWNET_ALEN). From what I can tell, all but one header_ops.parse
implementations return bytes from inside the packet, so could take an
unsigned char ** as argument, store that pointer and postpone the
memcpy to packet_recvmsg (handling skb_trim correctly). The exception is
fwnet_header_parse, which would require some workaround.

Re: [PATCH net-next 4/7] net: packet: use skb->dev as storage for skb orig len instead of skb->cb[]

[Eyal Birger]

Hi,

On Thu, Feb 26, 2015 at 7:26 PM, Willem de Bruijn <willemb@google.com> wrote:
> On Wed, Feb 25, 2015 at 9:10 PM, Eyal Birger <eyal.birger@gmail.com> wrote:
>> As part of an effort to move skb->dropcount to skb->cb[], 4 bytes
>> of additional room are needed in skb->cb[] in packet sockets.
>>
>> Store the skb original length in skb->dev instead of skb->cb[] for
>> this purpose.
>
> Another option is to delay preparation of the full sockaddr_ll struct until
> when it's needed. It often is not used at all (if msg_name == NULL).
>
> sll_family, sll_protocol and sll_pkttype can be derived when needed
> in packet_recvmsg. The first two are the head of sockaddr_ll, so
> a shorter struct filled in in packet_rcv only from sll_ifindex onwards
> would save the 4B needed for origlen. Actually, a union is simpler:
>
> struct __packet_cb_sockaddr_ll {
>         union {
>                 struct {
>                         unsigned short  sll_family;
>                         __be16          sll_protocol;
>                  }
>                  unsigned int sll_origlen;
>         }
>     /* etc.. */
> };
>
> The family and protocol can just overwrite sll_origlen in packet_recvmsg
> before the memcpy into msg->msg_name.
>

That's an interesting option.
My personal inclination is not to inline sockaddr_ll in the packet cb structure.
IMHO it would probably make sense to do it as part of your below suggestion.

> Also interesting would be to avoid copying the address in the fast path
> as it may never be used, especially for long addresses (INFINIBAND_ALEN
> and FWNET_ALEN). From what I can tell, all but one header_ops.parse
> implementations return bytes from inside the packet, so could take an
> unsigned char ** as argument, store that pointer and postpone the
> memcpy to packet_recvmsg (handling skb_trim correctly). The exception is
> fwnet_header_parse, which would require some workaround.

I think it may be interesting to pursue this option, though I don't
think it should
be part of this specific effort.

Thanks!
Eyal.

Re: [PATCH net-next 4/7] net: packet: use skb->dev as storage for skb orig len instead of skb->cb[]

[Willem de Bruijn]

On Thu, Feb 26, 2015 at 1:38 PM, Eyal Birger <eyal.birger@gmail.com> wrote:
> Hi,
>
> On Thu, Feb 26, 2015 at 7:26 PM, Willem de Bruijn <willemb@google.com> wrote:
>> On Wed, Feb 25, 2015 at 9:10 PM, Eyal Birger <eyal.birger@gmail.com> wrote:
>>> As part of an effort to move skb->dropcount to skb->cb[], 4 bytes
>>> of additional room are needed in skb->cb[] in packet sockets.
>>>
>>> Store the skb original length in skb->dev instead of skb->cb[] for
>>> this purpose.
>>
>> Another option is to delay preparation of the full sockaddr_ll struct until
>> when it's needed. It often is not used at all (if msg_name == NULL).
>>
>> sll_family, sll_protocol and sll_pkttype can be derived when needed
>> in packet_recvmsg. The first two are the head of sockaddr_ll, so
>> a shorter struct filled in in packet_rcv only from sll_ifindex onwards
>> would save the 4B needed for origlen. Actually, a union is simpler:
>>
>> struct __packet_cb_sockaddr_ll {
>>         union {
>>                 struct {
>>                         unsigned short  sll_family;
>>                         __be16          sll_protocol;
>>                  }
>>                  unsigned int sll_origlen;
>>         }
>>     /* etc.. */
>> };
>>
>> The family and protocol can just overwrite sll_origlen in packet_recvmsg
>> before the memcpy into msg->msg_name.
>>
>
> That's an interesting option.
> My personal inclination is not to inline sockaddr_ll in the packet cb structure.

It already is, isn't it?

> IMHO it would probably make sense to do it as part of your below suggestion.
>
>> Also interesting would be to avoid copying the address in the fast path
>> as it may never be used, especially for long addresses (INFINIBAND_ALEN
>> and FWNET_ALEN). From what I can tell, all but one header_ops.parse
>> implementations return bytes from inside the packet, so could take an
>> unsigned char ** as argument, store that pointer and postpone the
>> memcpy to packet_recvmsg (handling skb_trim correctly). The exception is
>> fwnet_header_parse, which would require some workaround.
>
> I think it may be interesting to pursue this option, though I don't
> think it should
> be part of this specific effort.

Absolutely. This is a lot more work. Too much to fold into this patchset.
I was more speculating in general. I may look into this at some point.

>
> Thanks!
> Eyal.

Re: [PATCH net-next 4/7] net: packet: use skb->dev as storage for skb orig len instead of skb->cb[]

[Eyal Birger]

On Thu, Feb 26, 2015 at 8:51 PM, Willem de Bruijn <willemb@google.com> wrote:
> On Thu, Feb 26, 2015 at 1:38 PM, Eyal Birger <eyal.birger@gmail.com> wrote:
>> Hi,
>>
>> On Thu, Feb 26, 2015 at 7:26 PM, Willem de Bruijn <willemb@google.com> wrote:
>>>
>>> Another option is to delay preparation of the full sockaddr_ll struct until
>>> when it's needed. It often is not used at all (if msg_name == NULL).
>>>
>>> sll_family, sll_protocol and sll_pkttype can be derived when needed
>>> in packet_recvmsg. The first two are the head of sockaddr_ll, so
>>> a shorter struct filled in in packet_rcv only from sll_ifindex onwards
>>> would save the 4B needed for origlen. Actually, a union is simpler:
>>>
>>> struct __packet_cb_sockaddr_ll {
>>>         union {
>>>                 struct {
>>>                         unsigned short  sll_family;
>>>                         __be16          sll_protocol;
>>>                  }
>>>                  unsigned int sll_origlen;
>>>         }
>>>     /* etc.. */
>>> };
>>>
>>> The family and protocol can just overwrite sll_origlen in packet_recvmsg
>>> before the memcpy into msg->msg_name.
>>>
>>
>> That's an interesting option.
>> My personal inclination is not to inline sockaddr_ll in the packet cb structure.
>
> It already is, isn't it?
>

I meant I prefer not to break out sockaddr_ll's individual fields in
packet_skb_cb.
Sorry for being ambiguous.

>> IMHO it would probably make sense to do it as part of your below suggestion.
>>
>>> Also interesting would be to avoid copying the address in the fast path
>>> as it may never be used, especially for long addresses (INFINIBAND_ALEN
>>> and FWNET_ALEN). From what I can tell, all but one header_ops.parse
>>> implementations return bytes from inside the packet, so could take an
>>> unsigned char ** as argument, store that pointer and postpone the
>>> memcpy to packet_recvmsg (handling skb_trim correctly). The exception is
>>> fwnet_header_parse, which would require some workaround.
>>
>> I think it may be interesting to pursue this option, though I don't
>> think it should
>> be part of this specific effort.
>
> Absolutely. This is a lot more work. Too much to fold into this patchset.
> I was more speculating in general. I may look into this at some point.
>
>>
>> Thanks!
>> Eyal.

[PATCH net-next 5/7] net: use common macro for assering skb->cb[] available size in protocol families

[Eyal Birger]

As part of an effort to move skb->dropcount to skb->cb[] use a common
macro in protocol families using skb->cb[] for ancillary data to
validate available room in skb->cb[].

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 include/net/sock.h           | 3 +++
 net/bluetooth/af_bluetooth.c | 3 +--
 net/can/bcm.c                | 2 +-
 net/can/raw.c                | 6 +++---
 net/ipv4/af_inet.c           | 2 +-
 net/ipv4/tcp.c               | 3 +--
 net/ipv6/af_inet6.c          | 2 +-
 net/packet/af_packet.c       | 3 +--
 net/sctp/protocol.c          | 3 +--
 9 files changed, 13 insertions(+), 14 deletions(-)

diff --git a/include/net/sock.h b/include/net/sock.h
index ab186b1..a2502d2 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2078,6 +2078,9 @@ static inline int sock_intr_errno(long timeo)
 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
 }
 
+#define sock_skb_cb_check_size(size) \
+	BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
+
 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 			   struct sk_buff *skb);
 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
diff --git a/net/bluetooth/af_bluetooth.c b/net/bluetooth/af_bluetooth.c
index ce22e0c..4b904c9 100644
--- a/net/bluetooth/af_bluetooth.c
+++ b/net/bluetooth/af_bluetooth.c
@@ -711,10 +711,9 @@ EXPORT_SYMBOL_GPL(bt_debugfs);
 
 static int __init bt_init(void)
 {
-	struct sk_buff *skb;
 	int err;
 
-	BUILD_BUG_ON(sizeof(struct bt_skb_cb) > sizeof(skb->cb));
+	sock_skb_cb_check_size(sizeof(struct bt_skb_cb));
 
 	BT_INFO("Core ver %s", VERSION);
 
diff --git a/net/can/bcm.c b/net/can/bcm.c
index ee9ffd9..d559f92 100644
--- a/net/can/bcm.c
+++ b/net/can/bcm.c
@@ -328,7 +328,7 @@ static void bcm_send_to_user(struct bcm_op *op, struct bcm_msg_head *head,
 	 *  containing the interface index.
 	 */
 
-	BUILD_BUG_ON(sizeof(skb->cb) < sizeof(struct sockaddr_can));
+	sock_skb_cb_check_size(sizeof(struct sockaddr_can));
 	addr = (struct sockaddr_can *)skb->cb;
 	memset(addr, 0, sizeof(*addr));
 	addr->can_family  = AF_CAN;
diff --git a/net/can/raw.c b/net/can/raw.c
index 00c13ef..94601b7 100644
--- a/net/can/raw.c
+++ b/net/can/raw.c
@@ -95,8 +95,8 @@ struct raw_sock {
  */
 static inline unsigned int *raw_flags(struct sk_buff *skb)
 {
-	BUILD_BUG_ON(sizeof(skb->cb) <= (sizeof(struct sockaddr_can) +
-					 sizeof(unsigned int)));
+	sock_skb_cb_check_size(sizeof(struct sockaddr_can) +
+			       sizeof(unsigned int));
 
 	/* return pointer after struct sockaddr_can */
 	return (unsigned int *)(&((struct sockaddr_can *)skb->cb)[1]);
@@ -135,7 +135,7 @@ static void raw_rcv(struct sk_buff *oskb, void *data)
 	 *  containing the interface index.
 	 */
 
-	BUILD_BUG_ON(sizeof(skb->cb) < sizeof(struct sockaddr_can));
+	sock_skb_cb_check_size(sizeof(struct sockaddr_can));
 	addr = (struct sockaddr_can *)skb->cb;
 	memset(addr, 0, sizeof(*addr));
 	addr->can_family  = AF_CAN;
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index d2e49ba..4ce954c 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -1675,7 +1675,7 @@ static int __init inet_init(void)
 	struct list_head *r;
 	int rc = -EINVAL;
 
-	BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
+	sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
 
 	rc = proto_register(&tcp_prot, 1);
 	if (rc)
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 9d72a0f..4b57ea8 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -3005,12 +3005,11 @@ static void __init tcp_init_mem(void)
 
 void __init tcp_init(void)
 {
-	struct sk_buff *skb = NULL;
 	unsigned long limit;
 	int max_rshare, max_wshare, cnt;
 	unsigned int i;
 
-	BUILD_BUG_ON(sizeof(struct tcp_skb_cb) > sizeof(skb->cb));
+	sock_skb_cb_check_size(sizeof(struct tcp_skb_cb));
 
 	percpu_counter_init(&tcp_sockets_allocated, 0, GFP_KERNEL);
 	percpu_counter_init(&tcp_orphan_count, 0, GFP_KERNEL);
diff --git a/net/ipv6/af_inet6.c b/net/ipv6/af_inet6.c
index e8c4400..6bafcc2 100644
--- a/net/ipv6/af_inet6.c
+++ b/net/ipv6/af_inet6.c
@@ -824,7 +824,7 @@ static int __init inet6_init(void)
 	struct list_head *r;
 	int err = 0;
 
-	BUILD_BUG_ON(sizeof(struct inet6_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
+	sock_skb_cb_check_size(sizeof(struct inet6_skb_parm));
 
 	/* Register the socket-side information for inet6_create.  */
 	for (r = &inetsw6[0]; r < &inetsw6[SOCK_MAX]; ++r)
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 26889fd..6ce6e00 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -1810,8 +1810,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
 		skb = nskb;
 	}
 
-	BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
-		     sizeof(skb->cb));
+	sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
 
 	sll = &PACKET_SKB_CB(skb)->sa.ll;
 	sll->sll_family = AF_PACKET;
diff --git a/net/sctp/protocol.c b/net/sctp/protocol.c
index 8f34b27..53b7acd 100644
--- a/net/sctp/protocol.c
+++ b/net/sctp/protocol.c
@@ -1322,8 +1322,7 @@ static __init int sctp_init(void)
 	int max_share;
 	int order;
 
-	BUILD_BUG_ON(sizeof(struct sctp_ulpevent) >
-		     sizeof(((struct sk_buff *) 0)->cb));
+	sock_skb_cb_check_size(sizeof(struct sctp_ulpevent));
 
 	/* Allocate bind_bucket and chunk caches. */
 	status = -ENOBUFS;
-- 
2.1.4

[PATCH net-next 6/7] net: add common accessor for setting dropcount on packets

[Eyal Birger]

As part of an effort to move skb->dropcount to skb->cb[], use
a common function in order to set dropcount in struct sk_buff.

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 include/net/sock.h     | 5 +++++
 net/core/sock.c        | 2 +-
 net/packet/af_packet.c | 2 +-
 3 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/include/net/sock.h b/include/net/sock.h
index a2502d2..d462f5e 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2081,6 +2081,11 @@ static inline int sock_intr_errno(long timeo)
 #define sock_skb_cb_check_size(size) \
 	BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
 
+static inline void sock_skb_set_dropcount(struct sock *sk, struct sk_buff *skb)
+{
+	skb->dropcount = atomic_read(&sk->sk_drops);
+}
+
 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
 			   struct sk_buff *skb);
 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
diff --git a/net/core/sock.c b/net/core/sock.c
index 93c8b20..9c74fc8 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -466,7 +466,7 @@ int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 	skb_dst_force(skb);
 
 	spin_lock_irqsave(&list->lock, flags);
-	skb->dropcount = atomic_read(&sk->sk_drops);
+	sock_skb_set_dropcount(sk, skb);
 	__skb_queue_tail(list, skb);
 	spin_unlock_irqrestore(&list->lock, flags);
 
diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index 6ce6e00..cc51b87 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -1837,7 +1837,7 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
 
 	spin_lock(&sk->sk_receive_queue.lock);
 	po->stats.stats1.tp_packets++;
-	skb->dropcount = atomic_read(&sk->sk_drops);
+	sock_skb_set_dropcount(sk, skb);
 	__skb_queue_tail(&sk->sk_receive_queue, skb);
 	spin_unlock(&sk->sk_receive_queue.lock);
 	sk->sk_data_ready(sk);
-- 
2.1.4

Re: [PATCH net-next 6/7] net: add common accessor for setting dropcount on packets

[Eric Dumazet]

On Thu, 2015-02-26 at 04:10 +0200, Eyal Birger wrote:
> As part of an effort to move skb->dropcount to skb->cb[], use
> a common function in order to set dropcount in struct sk_buff.
> 
> Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
> ---
>  include/net/sock.h     | 5 +++++
>  net/core/sock.c        | 2 +-
>  net/packet/af_packet.c | 2 +-
>  3 files changed, 7 insertions(+), 2 deletions(-)
> 
> diff --git a/include/net/sock.h b/include/net/sock.h
> index a2502d2..d462f5e 100644
> --- a/include/net/sock.h
> +++ b/include/net/sock.h
> @@ -2081,6 +2081,11 @@ static inline int sock_intr_errno(long timeo)
>  #define sock_skb_cb_check_size(size) \
>  	BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
>  
> +static inline void sock_skb_set_dropcount(struct sock *sk, struct sk_buff *skb)

const struct sock *sk


> +{
> +	skb->dropcount = atomic_read(&sk->sk_drops);
> +}

[PATCH net-next 7/7] net: move skb->dropcount to skb->cb[]

[Eyal Birger]

Commit 977750076d98 ("af_packet: add interframe drop cmsg (v6)")
unionized skb->mark and skb->dropcount in order to allow recording
of the socket drop count while maintaining struct sk_buff size.

skb->dropcount was introduced since there was no available room
in skb->cb[] in packet sockets. However, its introduction led to
the inability to export skb->mark, or any other aliased field to
userspace if so desired.

Moving the dropcount metric to skb->cb[] eliminates this problem
at the expense of 4 bytes less in skb->cb[] for protocol families
using it.

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>
---
 include/linux/skbuff.h |  2 --
 include/net/sock.h     | 18 ++++++++++++++++--
 net/socket.c           |  4 ++--
 3 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index d898b32..bba1330 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -492,7 +492,6 @@ static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
   *	@napi_id: id of the NAPI struct this skb came from
  *	@secmark: security marking
  *	@mark: Generic packet mark
- *	@dropcount: total number of sk_receive_queue overflows
  *	@vlan_proto: vlan encapsulation protocol
  *	@vlan_tci: vlan tag control information
  *	@inner_protocol: Protocol (encapsulation)
@@ -641,7 +640,6 @@ struct sk_buff {
 #endif
 	union {
 		__u32		mark;
-		__u32		dropcount;
 		__u32		reserved_tailroom;
 	};
 
diff --git a/include/net/sock.h b/include/net/sock.h
index d462f5e..8807586 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2078,12 +2078,26 @@ static inline int sock_intr_errno(long timeo)
 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
 }
 
+struct sock_skb_cb {
+	u32 dropcount;
+};
+
+/* Store sock_skb_cb at the end of skb->cb[] so protocol families
+ * using skb->cb[] would keep using it directly and utilize its
+ * alignement guarantee.
+ */
+#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
+			    sizeof(struct sock_skb_cb)))
+
+#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
+			    SOCK_SKB_CB_OFFSET))
+
 #define sock_skb_cb_check_size(size) \
-	BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
+	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
 
 static inline void sock_skb_set_dropcount(struct sock *sk, struct sk_buff *skb)
 {
-	skb->dropcount = atomic_read(&sk->sk_drops);
+	SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
 }
 
 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
diff --git a/net/socket.c b/net/socket.c
index bbedbfc..b78cf60 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -731,9 +731,9 @@ EXPORT_SYMBOL_GPL(__sock_recv_wifi_status);
 static inline void sock_recv_drops(struct msghdr *msg, struct sock *sk,
 				   struct sk_buff *skb)
 {
-	if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && skb->dropcount)
+	if (sock_flag(sk, SOCK_RXQ_OVFL) && skb && SOCK_SKB_CB(skb)->dropcount)
 		put_cmsg(msg, SOL_SOCKET, SO_RXQ_OVFL,
-			sizeof(__u32), &skb->dropcount);
+			sizeof(__u32), &SOCK_SKB_CB(skb)->dropcount);
 }
 
 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
-- 
2.1.4

Re: [PATCH net-next 7/7] net: move skb->dropcount to skb->cb[]

[Shmulik Ladkani]

On Thu, 26 Feb 2015 04:10:12 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
> diff --git a/include/net/sock.h b/include/net/sock.h
> index d462f5e..8807586 100644
> --- a/include/net/sock.h
> +++ b/include/net/sock.h
> @@ -2078,12 +2078,26 @@ static inline int sock_intr_errno(long timeo)
>  	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
>  }
>  
> +struct sock_skb_cb {
> +	u32 dropcount;
> +};
> +
> +/* Store sock_skb_cb at the end of skb->cb[] so protocol families
> + * using skb->cb[] would keep using it directly and utilize its
> + * alignement guarantee.
> + */
> +#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
> +			    sizeof(struct sock_skb_cb)))
> +
> +#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
> +			    SOCK_SKB_CB_OFFSET))
> +
>  #define sock_skb_cb_check_size(size) \
> -	BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
> +	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
>  

You didn't take care of aligning the 'sock_skb_cb'.
(Althogh in practive, dropcount is 4 and cb[] is 48 so you're golden).

How about:

struct sock_skb_cb {
	/* protocol families specifc CBs */
	u8 reserved[__SOCK_SKB_CB_OFFSET];
	struct __sock_skb_cb x;			// name me better!
};

Where '__SOCK_SKB_CB_OFFSET' and '__sock_skb_cb' are as follows:

struct __sock_skb_cb {	// name me better!
	u32 dropcount;
};

#define __SOCK_SKB_CB_OFFSET_UNALIGNED \
	(FIELD_SIZEOF(struct sk_buff, cb) - sizeof(struct __sock_skb_cb))

#define __SOCK_SKB_CB_OFFSET \
	(__SOCK_SKB_CB_OFFSET_UNALIGNED - \
	 (__SOCK_SKB_CB_OFFSET_UNALIGNED % __alignof__(struct __sock_skb_cb)))

This also takes care for alignement of '__sock_skb_cb x' within the CB.

Then,

#define sock_skb_cb_check_size(size) \
	BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sock_skb_cb, reserved))

Regards,
Shmulik

Re: [PATCH net-next 7/7] net: move skb->dropcount to skb->cb[]

[Eyal Birger]

On Thu, Feb 26, 2015 at 10:49 AM, Shmulik Ladkani
<shmulik.ladkani@gmail.com> wrote:
> On Thu, 26 Feb 2015 04:10:12 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
>> diff --git a/include/net/sock.h b/include/net/sock.h
>> index d462f5e..8807586 100644
>> --- a/include/net/sock.h
>> +++ b/include/net/sock.h
>> @@ -2078,12 +2078,26 @@ static inline int sock_intr_errno(long timeo)
>>       return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
>>  }
>>
>> +struct sock_skb_cb {
>> +     u32 dropcount;
>> +};
>> +
>> +/* Store sock_skb_cb at the end of skb->cb[] so protocol families
>> + * using skb->cb[] would keep using it directly and utilize its
>> + * alignement guarantee.
>> + */
>> +#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
>> +                         sizeof(struct sock_skb_cb)))
>> +
>> +#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
>> +                         SOCK_SKB_CB_OFFSET))
>> +
>>  #define sock_skb_cb_check_size(size) \
>> -     BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
>> +     BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
>>
>
> You didn't take care of aligning the 'sock_skb_cb'.
> (Althogh in practive, dropcount is 4 and cb[] is 48 so you're golden).
>
> How about:
>
> struct sock_skb_cb {
>         /* protocol families specifc CBs */
>         u8 reserved[__SOCK_SKB_CB_OFFSET];
>         struct __sock_skb_cb x;                 // name me better!
> };
>
> Where '__SOCK_SKB_CB_OFFSET' and '__sock_skb_cb' are as follows:
>
> struct __sock_skb_cb {  // name me better!
>         u32 dropcount;
> };
>
> #define __SOCK_SKB_CB_OFFSET_UNALIGNED \
>         (FIELD_SIZEOF(struct sk_buff, cb) - sizeof(struct __sock_skb_cb))
>
> #define __SOCK_SKB_CB_OFFSET \
>         (__SOCK_SKB_CB_OFFSET_UNALIGNED - \
>          (__SOCK_SKB_CB_OFFSET_UNALIGNED % __alignof__(struct __sock_skb_cb)))
>
> This also takes care for alignement of '__sock_skb_cb x' within the CB.
>
> Then,
>
> #define sock_skb_cb_check_size(size) \
>         BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sock_skb_cb, reserved))
>

Agreed. Will add to v2.

Re: [PATCH net-next 7/7] net: move skb->dropcount to skb->cb[]

[Eyal Birger]

On Thu, Feb 26, 2015 at 12:17 PM, Eyal Birger <eyal.birger@gmail.com> wrote:
> On Thu, Feb 26, 2015 at 10:49 AM, Shmulik Ladkani
> <shmulik.ladkani@gmail.com> wrote:
>> On Thu, 26 Feb 2015 04:10:12 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
>>> diff --git a/include/net/sock.h b/include/net/sock.h
>>> index d462f5e..8807586 100644
>>> --- a/include/net/sock.h
>>> +++ b/include/net/sock.h
>>> @@ -2078,12 +2078,26 @@ static inline int sock_intr_errno(long timeo)
>>>       return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
>>>  }
>>>
>>> +struct sock_skb_cb {
>>> +     u32 dropcount;
>>> +};
>>> +
>>> +/* Store sock_skb_cb at the end of skb->cb[] so protocol families
>>> + * using skb->cb[] would keep using it directly and utilize its
>>> + * alignement guarantee.
>>> + */
>>> +#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
>>> +                         sizeof(struct sock_skb_cb)))
>>> +
>>> +#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
>>> +                         SOCK_SKB_CB_OFFSET))
>>> +
>>>  #define sock_skb_cb_check_size(size) \
>>> -     BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sk_buff, cb))
>>> +     BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
>>>
>>
>> You didn't take care of aligning the 'sock_skb_cb'.
>> (Althogh in practive, dropcount is 4 and cb[] is 48 so you're golden).
>>
>> How about:
>>
>> struct sock_skb_cb {
>>         /* protocol families specifc CBs */
>>         u8 reserved[__SOCK_SKB_CB_OFFSET];
>>         struct __sock_skb_cb x;                 // name me better!
>> };
>>
>> Where '__SOCK_SKB_CB_OFFSET' and '__sock_skb_cb' are as follows:
>>
>> struct __sock_skb_cb {  // name me better!
>>         u32 dropcount;
>> };
>>
>> #define __SOCK_SKB_CB_OFFSET_UNALIGNED \
>>         (FIELD_SIZEOF(struct sk_buff, cb) - sizeof(struct __sock_skb_cb))
>>
>> #define __SOCK_SKB_CB_OFFSET \
>>         (__SOCK_SKB_CB_OFFSET_UNALIGNED - \
>>          (__SOCK_SKB_CB_OFFSET_UNALIGNED % __alignof__(struct __sock_skb_cb)))
>>
>> This also takes care for alignement of '__sock_skb_cb x' within the CB.
>>
>> Then,
>>
>> #define sock_skb_cb_check_size(size) \
>>         BUILD_BUG_ON((size) > FIELD_SIZEOF(struct sock_skb_cb, reserved))
>>
>
> Agreed. Will add to v2.

After giving it some additional thought and research, I don't think
this is necessary.

The sizeof operator on struct sock_skb_cb would give a padded result
which would take care of the structure alignment.

This is under the assumption that skb->cb[] size is of a proper multiple (which
I think is rather safe).

Objections?
Eyal.

Re: [PATCH net-next 7/7] net: move skb->dropcount to skb->cb[]

[Shmulik Ladkani]

On Thu, 26 Feb 2015 12:48:14 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
> After giving it some additional thought and research, I don't think
> this is necessary.
> 
> The sizeof operator on struct sock_skb_cb would give a padded result
> which would take care of the structure alignment.
> 
> This is under the assumption that skb->cb[] size is of a proper multiple (which
> I think is rather safe).

Yes, you are correct.

With the assumption that skb->cb and 'cb' size are properly aligned and
sized (compile time asserted, btw?), you can simply:

struct your_thing {     // name it
	u32 dropcount;
};

struct sock_skb_cb {
	/* protocol families specifc CBs */
	u8 pf_reserved[PAD];
	struct your_thing sock_cb;
};

where:

#define PAD (FIELD_SIZEOF(struct sk_buff, cb) - sizeof(struct your_thing))

Re: [PATCH net-next 7/7] net: move skb->dropcount to skb->cb[]

[Eyal Birger]

On Thu, Feb 26, 2015 at 5:32 PM, Shmulik Ladkani
<shmulik.ladkani@gmail.com> wrote:
> On Thu, 26 Feb 2015 12:48:14 +0200 Eyal Birger <eyal.birger@gmail.com> wrote:
>> After giving it some additional thought and research, I don't think
>> this is necessary.
>>
>> The sizeof operator on struct sock_skb_cb would give a padded result
>> which would take care of the structure alignment.
>>
>> This is under the assumption that skb->cb[] size is of a proper multiple (which
>> I think is rather safe).
>
> Yes, you are correct.
>
> With the assumption that skb->cb and 'cb' size are properly aligned and
> sized (compile time asserted, btw?), you can simply:
>
> struct your_thing {     // name it
>         u32 dropcount;
> };
>
> struct sock_skb_cb {
>         /* protocol families specifc CBs */
>         u8 pf_reserved[PAD];
>         struct your_thing sock_cb;
> };
>
> where:
>
> #define PAD (FIELD_SIZEOF(struct sk_buff, cb) - sizeof(struct your_thing))
>

struct sock_skb_cb as I see it represents the sock generic metadata
required in the skb->cb[] and its placement at the end of skb->cb[]
is an implementation detail - so it, in my view, is what you referred to as
"struct your_thing".

Coding-wise I would rather avoid the additional enclosing struct and stick
with the proposed implementation.

Eyal.