[PATCH 01/11] netlink: add symbolic value for congested state

[PATCH 01/11] netlink: add symbolic value for congested state

[kaber]

From: Patrick McHardy <kaber@trash.net>

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 net/netlink/af_netlink.c |   18 +++++++++++-------
 1 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 0a4db02..fc63ca5 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -88,6 +88,10 @@ struct listeners {
 	unsigned long		masks[0];
 };
 
+/* state bits */
+#define NETLINK_CONGESTED	0x0
+
+/* flags */
 #define NETLINK_KERNEL_SOCKET	0x1
 #define NETLINK_RECV_PKTINFO	0x2
 #define NETLINK_BROADCAST_SEND_ERROR	0x4
@@ -737,7 +741,7 @@ static void netlink_overrun(struct sock *sk)
 	struct netlink_sock *nlk = nlk_sk(sk);
 
 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
-		if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
+		if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
 			sk->sk_err = ENOBUFS;
 			sk->sk_error_report(sk);
 		}
@@ -798,7 +802,7 @@ int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
 	nlk = nlk_sk(sk);
 
 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
-	    test_bit(0, &nlk->state)) {
+	    test_bit(NETLINK_CONGESTED, &nlk->state)) {
 		DECLARE_WAITQUEUE(wait, current);
 		if (!*timeo) {
 			if (!ssk || netlink_is_kernel(ssk))
@@ -812,7 +816,7 @@ int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
 		add_wait_queue(&nlk->wait, &wait);
 
 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
-		     test_bit(0, &nlk->state)) &&
+		     test_bit(NETLINK_CONGESTED, &nlk->state)) &&
 		    !sock_flag(sk, SOCK_DEAD))
 			*timeo = schedule_timeout(*timeo);
 
@@ -876,8 +880,8 @@ static inline void netlink_rcv_wake(struct sock *sk)
 	struct netlink_sock *nlk = nlk_sk(sk);
 
 	if (skb_queue_empty(&sk->sk_receive_queue))
-		clear_bit(0, &nlk->state);
-	if (!test_bit(0, &nlk->state))
+		clear_bit(NETLINK_CONGESTED, &nlk->state);
+	if (!test_bit(NETLINK_CONGESTED, &nlk->state))
 		wake_up_interruptible(&nlk->wait);
 }
 
@@ -958,7 +962,7 @@ static inline int netlink_broadcast_deliver(struct sock *sk,
 	struct netlink_sock *nlk = nlk_sk(sk);
 
 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
-	    !test_bit(0, &nlk->state)) {
+	    !test_bit(NETLINK_CONGESTED, &nlk->state)) {
 		skb_set_owner_r(skb, sk);
 		skb_queue_tail(&sk->sk_receive_queue, skb);
 		sk->sk_data_ready(sk, skb->len);
@@ -1236,7 +1240,7 @@ static int netlink_setsockopt(struct socket *sock, int level, int optname,
 	case NETLINK_NO_ENOBUFS:
 		if (val) {
 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
-			clear_bit(0, &nlk->state);
+			clear_bit(NETLINK_CONGESTED, &nlk->state);
 			wake_up_interruptible(&nlk->wait);
 		} else
 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
-- 
1.7.4.4

[PATCH 00/11] netlink: memory mapped I/O

[Patrick McHardy]

The following patches contain an implementation of memory mapped I/O for
netlink, rebased onto the current net-next tree. The implementation is
modelled after AF_PACKET memory mapped I/O with a few differences:

- In order to perform memory mapped I/O to userspace, the kernel allocates
  skbs with the data area pointing to the data area of the mapped frames.
  All netlink subsystems assume a linear data area, so for the sake of
  simplicity, the mapped data area is not attached to the paged area but
  to skb->data. This requires introduction of a special skb alloction
  function that just allocates an skb head without the data area. Since this
  is a quite rare use case, I introduced a new function based on __alloc_skb
  instead of splitting it up into head and data alloction. The alternative
  would be to   introduce an __alloc_skb_head and __alloc_skb_data function,
  which would actually be useful for a specific error case in memory mapped
  netlink, but would require a couple of extra instructions for the common
  skb allocation case, so it doesn't really seem worth it.

  In order to get the destination memory area for skb->data before message
  construction, memory mapped netlink I/O needs to look up the destination
  socket during allocation instead of during transmission because the
  ring is owned by the receiveing socket/process. A special skb allocation
  function (netlink_alloc_skb) taking the destination pid as an argument is
  used for this, all subsystems that want to support memory mapped I/O need
  to use this function, automatic fallback to the receive queue happens
  for unconverted subsystems. Dumps automatically use memory mapped I/O if
  the receiving socket has enabled it.

  The visible effect of looking up the destination socket during allocation
  instead of transmission is that message ordering in userspace might
  change in case allocation and transmission aren't performed atomically.
  This usually doesn't matter since most subsystems have a BKL-like lock
  like the rtnl mutex, to my knowledge the currently only existing case
  where it might matter is nfnetlink_queue combined with the recently
  introduced batched verdicts, but a) that subsystem already includes
  sequence numbers which allow userspace to reorder messages in case it
  cares to, also the reodering window is quite small and b) with memory
  mapped transmission batching can be performed in a subsystem indepandant
  manner.

- AF_NETLINK contains flow control for database dumps, with regular I/O
  dump continuation are triggered based on the sockets receive queue space
  and by recvmsg() calls. Since with memory mapped I/O there are no
  recvmsg() calls under normal operation, this is done in netlink_poll(),
  under the assumption that userspace has processed all pending frames
  before invoking poll(), thus the ring is expected to have room for new
  messages. Dumps currently don't benefit as much as they could from
  memory mapped I/O because each single continuation requires a poll()
  call. A more agressive approach seems like a good idea to me, especially
  in case the socket is not subscribed to any multicast groups (IOW only
  receiving explicitly requested data).

Besides that, the memory mapped netlink implementation extends the states
defined by AF_PACKET between userspace and the kernel by a SKIP status, this
is intended for the case that userspace wants to queue frames (specifically
when using nfnetlink_queue, an IDS and stream reassembly, requested by
Eric Leblond) for a longer period of time. The kernel skips over all frames
marked with SKIP when looking or unused frames and only fails when not finding
a free frame or when having skipped the entire ring.

Also noteworthy is memory mapped sendmsg: the kernel performs validation
of messages before accepting and processing them, in order to prevent
userspace from changing the messages contents after validation, the
kernel checks that the ring is only mapped once and the file descriptor
is not shared (in order to avoid having userspace set up another mapping
after the first mentioned check). If either of both is not true, the
message copied to an allocated skb and processed as with regular I/O.

As an example, nfnetlink_queue is convererted to support memory mapped
I/O. Other subsystems that would probably benefit are nfnetlink_log,
audit and maybe ISCSI.

Since the last posting only a minor bug in ring teardown has been fixed.
I'm still working with Florian on solving the nfnetlink_queue ordering
issue, besides that there are no known issues.

Some older performance numbers with nfnetlink_queue from Florian:

nfq recv: regular netlink I/O
mnl recv: mmap'ed netlink I/O
batch: number of batched verdicts

1400 byte UDP packets, 8 cores, NFQUEUE balancing using 4 queues, only
mmap'ed RX, regular TX, userspace running inline Snort with all rules
and preprocessors enabled:

nfq recv, batch 0  1250 MBit total rx
mnl recv, batch 0  1230 MBit total rx
nfq recv, batch 10 1590 MBit total rx
mnl recv, batch 10 1770 MBit total rx

I'll try to get some new numbers soon, including TX and dumps.


The patches are available in a git tree at:

git://github.com/kaber/netlink-mmap master

once git push has finished.


Patrick McHardy (11):
      netlink: add symbolic value for congested state
      net: add function to allocate skbuff head without data area
      netlink: don't orphan skb in netlink_trim()
      netlink: add netlink_skb_set_owner_r()
      netlink: mmaped netlink: ring setup
      netlink: add mmap'ed netlink helper functions
      netlink: implement memory mapped sendmsg()
      netlink: implement memory mapped recvmsg()
      nfnetlink: add support for memory mapped netlink
      netlink: add flow control for memory mapped I/O
      netlink: add documentation for memory mapped I/O

 Documentation/networking/netlink_mmap.txt |  337 ++++++++++++
 include/linux/netfilter/nfnetlink.h       |    2 +
 include/linux/netlink.h                   |   42 ++
 include/linux/skbuff.h                    |    6 +
 net/Kconfig                               |    9 +
 net/core/skbuff.c                         |   31 +-
 net/netfilter/nfnetlink.c                 |    7 +
 net/netfilter/nfnetlink_log.c             |    9 +-
 net/netfilter/nfnetlink_queue_core.c      |    2 +-
 net/netlink/af_netlink.c                  |  849 +++++++++++++++++++++++++++--
 10 files changed, 1250 insertions(+), 44 deletions(-)

[PATCH 01/11] netlink: add symbolic value for congested state

[Patrick McHardy]

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 net/netlink/af_netlink.c |   18 +++++++++++-------
 1 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 5463969..1bdfa52 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -89,6 +89,10 @@ struct listeners {
 	unsigned long		masks[0];
 };
 
+/* state bits */
+#define NETLINK_CONGESTED	0x0
+
+/* flags */
 #define NETLINK_KERNEL_SOCKET	0x1
 #define NETLINK_RECV_PKTINFO	0x2
 #define NETLINK_BROADCAST_SEND_ERROR	0x4
@@ -762,7 +766,7 @@ static void netlink_overrun(struct sock *sk)
 	struct netlink_sock *nlk = nlk_sk(sk);
 
 	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
-		if (!test_and_set_bit(0, &nlk_sk(sk)->state)) {
+		if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
 			sk->sk_err = ENOBUFS;
 			sk->sk_error_report(sk);
 		}
@@ -823,7 +827,7 @@ int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
 	nlk = nlk_sk(sk);
 
 	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
-	    test_bit(0, &nlk->state)) {
+	    test_bit(NETLINK_CONGESTED, &nlk->state)) {
 		DECLARE_WAITQUEUE(wait, current);
 		if (!*timeo) {
 			if (!ssk || netlink_is_kernel(ssk))
@@ -837,7 +841,7 @@ int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
 		add_wait_queue(&nlk->wait, &wait);
 
 		if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
-		     test_bit(0, &nlk->state)) &&
+		     test_bit(NETLINK_CONGESTED, &nlk->state)) &&
 		    !sock_flag(sk, SOCK_DEAD))
 			*timeo = schedule_timeout(*timeo);
 
@@ -907,8 +911,8 @@ static void netlink_rcv_wake(struct sock *sk)
 	struct netlink_sock *nlk = nlk_sk(sk);
 
 	if (skb_queue_empty(&sk->sk_receive_queue))
-		clear_bit(0, &nlk->state);
-	if (!test_bit(0, &nlk->state))
+		clear_bit(NETLINK_CONGESTED, &nlk->state);
+	if (!test_bit(NETLINK_CONGESTED, &nlk->state))
 		wake_up_interruptible(&nlk->wait);
 }
 
@@ -990,7 +994,7 @@ static int netlink_broadcast_deliver(struct sock *sk, struct sk_buff *skb)
 	struct netlink_sock *nlk = nlk_sk(sk);
 
 	if (atomic_read(&sk->sk_rmem_alloc) <= sk->sk_rcvbuf &&
-	    !test_bit(0, &nlk->state)) {
+	    !test_bit(NETLINK_CONGESTED, &nlk->state)) {
 		skb_set_owner_r(skb, sk);
 		__netlink_sendskb(sk, skb);
 		return atomic_read(&sk->sk_rmem_alloc) > (sk->sk_rcvbuf >> 1);
@@ -1270,7 +1274,7 @@ static int netlink_setsockopt(struct socket *sock, int level, int optname,
 	case NETLINK_NO_ENOBUFS:
 		if (val) {
 			nlk->flags |= NETLINK_RECV_NO_ENOBUFS;
-			clear_bit(0, &nlk->state);
+			clear_bit(NETLINK_CONGESTED, &nlk->state);
 			wake_up_interruptible(&nlk->wait);
 		} else {
 			nlk->flags &= ~NETLINK_RECV_NO_ENOBUFS;
-- 
1.7.7.6

[PATCH 02/11] net: add function to allocate skbuff head without data area

[Patrick McHardy]

Add a function to allocate a skbuff head without any data. This will be
used by memory mapped netlink to attached data from the mmaped area to
the skb.

Additionally change skb_release_all() to check whether the skb has a
data area to allow the skb destructor to clear the data pointer in
case only a head has been allocated.

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 include/linux/skbuff.h |    6 ++++++
 net/core/skbuff.c      |   31 ++++++++++++++++++++++++++++++-
 2 files changed, 36 insertions(+), 1 deletions(-)

diff --git a/include/linux/skbuff.h b/include/linux/skbuff.h
index b33a3a1..93fe80f 100644
--- a/include/linux/skbuff.h
+++ b/include/linux/skbuff.h
@@ -589,6 +589,12 @@ static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
 	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
 }
 
+extern struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
+static inline struct sk_buff *alloc_skb_head(gfp_t priority)
+{
+	return __alloc_skb_head(priority, -1);
+}
+
 extern void skb_recycle(struct sk_buff *skb);
 extern bool skb_recycle_check(struct sk_buff *skb, int skb_size);
 
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index fe00d12..4046f6e 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -188,6 +188,34 @@ out:
  *
  */
 
+struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
+{
+	struct sk_buff *skb;
+
+	/* Get the HEAD */
+	skb = kmem_cache_alloc_node(skbuff_head_cache,
+				    gfp_mask & ~__GFP_DMA, node);
+	if (!skb)
+		goto out;
+	prefetchw(skb);
+
+	/*
+	 * Only clear those fields we need to clear, not those that we will
+	 * actually initialise below. Hence, don't put any more fields after
+	 * the tail pointer in struct sk_buff!
+	 */
+	memset(skb, 0, offsetof(struct sk_buff, tail));
+	skb->data = NULL;
+	skb->truesize = sizeof(struct sk_buff);
+	atomic_set(&skb->users, 1);
+
+#ifdef NET_SKBUFF_DATA_USES_OFFSET
+	skb->mac_header = ~0U;
+#endif
+out:
+	return skb;
+}
+
 /**
  *	__alloc_skb	-	allocate a network buffer
  *	@size: size to allocate
@@ -581,7 +609,8 @@ static void skb_release_head_state(struct sk_buff *skb)
 static void skb_release_all(struct sk_buff *skb)
 {
 	skb_release_head_state(skb);
-	skb_release_data(skb);
+	if (likely(skb->data))
+		skb_release_data(skb);
 }
 
 /**
-- 
1.7.7.6

[PATCH 03/11] netlink: don't orphan skb in netlink_trim()

[Patrick McHardy]

Netlink doesn't account skbs to the sending socket, so the there's no
need to orphan the skb before trimming it.

Removing the skb_orphan() call is required for mmap'ed netlink, which uses
a netlink specific skb destructor that must not be invoked before the
final freeing of the skb.

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 net/netlink/af_netlink.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 1bdfa52..3233601 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -886,7 +886,7 @@ static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
 {
 	int delta;
 
-	skb_orphan(skb);
+	WARN_ON(skb->sk != NULL);
 
 	delta = skb->end - skb->tail;
 	if (delta * 2 < skb->truesize)
-- 
1.7.7.6

[PATCH 04/11] netlink: add netlink_skb_set_owner_r()

[Patrick McHardy]

For mmap'ed I/O a netlink specific skb destructor needs to be invoked after the
final kfree_skb() to clean up state. This doesn't work currently since the skb's
ownership is transfered to the receiving socket using skb_set_owner_r(), which
orphans the skb, thereby invoking the destructor prematurely.

Since netlink doesn't account skbs to the originating socket, there's no need
to orphan the skb. Add a netlink specific skb_set_owner_r() variant that does
not orphan the skb and use a netlink specific destructor to call sock_rfree().

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 net/netlink/af_netlink.c |   18 ++++++++++++++++--
 1 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 3233601..e2d0177 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -166,6 +166,20 @@ static void netlink_consume_callback(struct netlink_callback *cb)
 	kfree(cb);
 }
 
+static void netlink_skb_destructor(struct sk_buff *skb)
+{
+	sock_rfree(skb);
+}
+
+static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
+{
+	WARN_ON(skb->sk != NULL);
+	skb->sk = sk;
+	skb->destructor = netlink_skb_destructor;
+	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
+	sk_mem_charge(sk, skb->truesize);
+}
+
 static void netlink_sock_destruct(struct sock *sk)
 {
 	struct netlink_sock *nlk = nlk_sk(sk);
@@ -855,7 +869,7 @@ int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
 		}
 		return 1;
 	}
-	skb_set_owner_r(skb, sk);
+	netlink_skb_set_owner_r(skb, sk);
 	return 0;
 }
 
@@ -924,7 +938,7 @@ static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
 	ret = -ECONNREFUSED;
 	if (nlk->netlink_rcv != NULL) {
 		ret = skb->len;
-		skb_set_owner_r(skb, sk);
+		netlink_skb_set_owner_r(skb, sk);
 		nlk->netlink_rcv(skb);
 		consume_skb(skb);
 	} else {
-- 
1.7.7.6

[PATCH 05/11] netlink: mmaped netlink: ring setup

[Patrick McHardy]

Add support for mmap'ed RX and TX ring setup and teardown based on the
af_packet.c code. The following patches will use this to add the real
mmap'ed receive and transmit functionality.

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 include/linux/netlink.h  |   32 +++++
 net/Kconfig              |    9 ++
 net/netlink/af_netlink.c |  288 +++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 327 insertions(+), 2 deletions(-)

diff --git a/include/linux/netlink.h b/include/linux/netlink.h
index f74dd13..3600461 100644
--- a/include/linux/netlink.h
+++ b/include/linux/netlink.h
@@ -1,6 +1,7 @@
 #ifndef __LINUX_NETLINK_H
 #define __LINUX_NETLINK_H
 
+#include <linux/kernel.h>
 #include <linux/socket.h> /* for __kernel_sa_family_t */
 #include <linux/types.h>
 
@@ -105,11 +106,42 @@ struct nlmsgerr {
 #define NETLINK_PKTINFO		3
 #define NETLINK_BROADCAST_ERROR	4
 #define NETLINK_NO_ENOBUFS	5
+#define NETLINK_RX_RING		6
+#define NETLINK_TX_RING		7
 
 struct nl_pktinfo {
 	__u32	group;
 };
 
+struct nl_mmap_req {
+	unsigned int	nm_block_size;
+	unsigned int	nm_block_nr;
+	unsigned int	nm_frame_size;
+	unsigned int	nm_frame_nr;
+};
+
+struct nl_mmap_hdr {
+	unsigned int	nm_status;
+	unsigned int	nm_len;
+	__u32		nm_group;
+	/* credentials */
+	__u32		nm_pid;
+	__u32		nm_uid;
+	__u32		nm_gid;
+};
+
+enum nl_mmap_status {
+	NL_MMAP_STATUS_UNUSED,
+	NL_MMAP_STATUS_RESERVED,
+	NL_MMAP_STATUS_VALID,
+	NL_MMAP_STATUS_COPY,
+	NL_MMAP_STATUS_SKIP,
+};
+
+#define NL_MMAP_MSG_ALIGNMENT		NLMSG_ALIGNTO
+#define NL_MMAP_MSG_ALIGN(sz)		__ALIGN_KERNEL(sz, NL_MMAP_MSG_ALIGNMENT)
+#define NL_MMAP_HDRLEN			NL_MMAP_MSG_ALIGN(sizeof(struct nl_mmap_hdr))
+
 #define NET_MAJOR 36		/* Major 36 is reserved for networking 						*/
 
 enum {
diff --git a/net/Kconfig b/net/Kconfig
index 245831b..7f48691 100644
--- a/net/Kconfig
+++ b/net/Kconfig
@@ -23,6 +23,15 @@ menuconfig NET
 
 if NET
 
+config NETLINK_MMAP
+	bool "Netlink: mmaped IO"
+	help
+	  This option enables support for memory mapped netlink IO. This
+	  reduces overhead by avoiding copying data between kernel- and
+	  userspace.
+
+	  If unsure, say N.
+
 config WANT_COMPAT_NETLINK_MESSAGES
 	bool
 	help
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index e2d0177..e20a918 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -55,6 +55,7 @@
 #include <linux/types.h>
 #include <linux/audit.h>
 #include <linux/mutex.h>
+#include <linux/vmalloc.h>
 
 #include <net/net_namespace.h>
 #include <net/sock.h>
@@ -64,6 +65,20 @@
 #define NLGRPSZ(x)	(ALIGN(x, sizeof(unsigned long) * 8) / 8)
 #define NLGRPLONGS(x)	(NLGRPSZ(x)/sizeof(unsigned long))
 
+struct netlink_ring {
+	void			**pg_vec;
+	unsigned int		head;
+	unsigned int		frames_per_block;
+	unsigned int		frame_size;
+	unsigned int		frame_max;
+
+	unsigned int		pg_vec_order;
+	unsigned int		pg_vec_pages;
+	unsigned int		pg_vec_len;
+
+	atomic_t		pending;
+};
+
 struct netlink_sock {
 	/* struct sock has to be the first member of netlink_sock */
 	struct sock		sk;
@@ -82,6 +97,12 @@ struct netlink_sock {
 	void			(*netlink_rcv)(struct sk_buff *skb);
 	void			(*netlink_bind)(int group);
 	struct module		*module;
+#ifdef CONFIG_NETLINK_MMAP
+	struct mutex		pg_vec_lock;
+	struct netlink_ring	rx_ring;
+	struct netlink_ring	tx_ring;
+	atomic_t		mapped;
+#endif /* CONFIG_NETLINK_MMAP */
 };
 
 struct listeners {
@@ -154,6 +175,234 @@ static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid
 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
 }
 
+#ifdef CONFIG_NETLINK_MMAP
+static __pure struct page *pgvec_to_page(const void *addr)
+{
+	if (is_vmalloc_addr(addr))
+		return vmalloc_to_page(addr);
+	else
+		return virt_to_page(addr);
+}
+
+static void free_pg_vec(void **pg_vec, unsigned int order, unsigned int len)
+{
+	unsigned int i;
+
+	for (i = 0; i < len; i++) {
+		if (pg_vec[i] != NULL) {
+			if (is_vmalloc_addr(pg_vec[i]))
+				vfree(pg_vec[i]);
+			else
+				free_pages((unsigned long)pg_vec[i], order);
+		}
+	}
+	kfree(pg_vec);
+}
+
+static void *alloc_one_pg_vec_page(unsigned long order)
+{
+	void *buffer;
+	gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO |
+			  __GFP_NOWARN | __GFP_NORETRY;
+
+	buffer = (void *)__get_free_pages(gfp_flags, order);
+	if (buffer != NULL)
+		return buffer;
+
+	buffer = vzalloc((1 << order) * PAGE_SIZE);
+	if (buffer != NULL)
+		return buffer;
+
+	gfp_flags &= ~__GFP_NORETRY;
+	return (void *)__get_free_pages(gfp_flags, order);
+}
+
+static void **alloc_pg_vec(struct netlink_sock *nlk,
+			   struct nl_mmap_req *req, unsigned int order)
+{
+	unsigned int block_nr = req->nm_block_nr;
+	unsigned int i;
+	void **pg_vec, *ptr;
+
+	pg_vec = kcalloc(block_nr, sizeof(void *), GFP_KERNEL);
+	if (pg_vec == NULL)
+		return NULL;
+
+	for (i = 0; i < block_nr; i++) {
+		pg_vec[i] = ptr = alloc_one_pg_vec_page(order);
+		if (pg_vec[i] == NULL)
+			goto err1;
+	}
+
+	return pg_vec;
+err1:
+	free_pg_vec(pg_vec, order, block_nr);
+	return NULL;
+}
+
+static int netlink_set_ring(struct sock *sk, struct nl_mmap_req *req,
+			    bool closing, bool tx_ring)
+{
+	struct netlink_sock *nlk = nlk_sk(sk);
+	struct netlink_ring *ring;
+	struct sk_buff_head *queue;
+	void **pg_vec = NULL;
+	unsigned int order = 0;
+	int err;
+
+	ring  = tx_ring ? &nlk->tx_ring : &nlk->rx_ring;
+	queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
+
+	if (!closing) {
+		if (atomic_read(&nlk->mapped))
+			return -EBUSY;
+		if (atomic_read(&ring->pending))
+			return -EBUSY;
+	}
+
+	if (req->nm_block_nr) {
+		if (ring->pg_vec != NULL)
+			return -EBUSY;
+
+		if ((int)req->nm_block_size <= 0)
+			return -EINVAL;
+		if (!IS_ALIGNED(req->nm_block_size, PAGE_SIZE))
+			return -EINVAL;
+		if (req->nm_frame_size < NL_MMAP_HDRLEN)
+			return -EINVAL;
+		if (!IS_ALIGNED(req->nm_frame_size, NL_MMAP_MSG_ALIGNMENT))
+			return -EINVAL;
+
+		ring->frames_per_block = req->nm_block_size /
+					 req->nm_frame_size;
+		if (ring->frames_per_block == 0)
+			return -EINVAL;
+		if (ring->frames_per_block * req->nm_block_nr !=
+		    req->nm_frame_nr)
+			return -EINVAL;
+
+		order = get_order(req->nm_block_size);
+		pg_vec = alloc_pg_vec(nlk, req, order);
+		if (pg_vec == NULL)
+			return -ENOMEM;
+	} else {
+		if (req->nm_frame_nr)
+			return -EINVAL;
+	}
+
+	err = -EBUSY;
+	mutex_lock(&nlk->pg_vec_lock);
+	if (closing || atomic_read(&nlk->mapped) == 0) {
+		err = 0;
+		spin_lock_bh(&queue->lock);
+
+		ring->frame_max		= req->nm_frame_nr - 1;
+		ring->head		= 0;
+		ring->frame_size	= req->nm_frame_size;
+		ring->pg_vec_pages	= req->nm_block_size / PAGE_SIZE;
+
+		swap(ring->pg_vec_len, req->nm_block_nr);
+		swap(ring->pg_vec_order, order);
+		swap(ring->pg_vec, pg_vec);
+
+		__skb_queue_purge(queue);
+		spin_unlock_bh(&queue->lock);
+
+		WARN_ON(atomic_read(&nlk->mapped));
+	}
+	mutex_unlock(&nlk->pg_vec_lock);
+
+	if (pg_vec)
+		free_pg_vec(pg_vec, order, req->nm_block_nr);
+	return err;
+}
+
+static void netlink_mm_open(struct vm_area_struct *vma)
+{
+	struct file *file = vma->vm_file;
+	struct socket *sock = file->private_data;
+	struct sock *sk = sock->sk;
+
+	if (sk)
+		atomic_inc(&nlk_sk(sk)->mapped);
+}
+
+static void netlink_mm_close(struct vm_area_struct *vma)
+{
+	struct file *file = vma->vm_file;
+	struct socket *sock = file->private_data;
+	struct sock *sk = sock->sk;
+
+	if (sk)
+		atomic_dec(&nlk_sk(sk)->mapped);
+}
+
+static const struct vm_operations_struct netlink_mmap_ops = {
+	.open	= netlink_mm_open,
+	.close	= netlink_mm_close,
+};
+
+static int netlink_mmap(struct file *file, struct socket *sock,
+			struct vm_area_struct *vma)
+{
+	struct sock *sk = sock->sk;
+	struct netlink_sock *nlk = nlk_sk(sk);
+	struct netlink_ring *ring;
+	unsigned long start, size, expected;
+	unsigned int i;
+	int err = -EINVAL;
+
+	if (vma->vm_pgoff)
+		return -EINVAL;
+
+	mutex_lock(&nlk->pg_vec_lock);
+
+	expected = 0;
+	for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
+		if (ring->pg_vec == NULL)
+			continue;
+		expected += ring->pg_vec_len * ring->pg_vec_pages * PAGE_SIZE;
+	}
+
+	if (expected == 0)
+		goto out;
+
+	size = vma->vm_end - vma->vm_start;
+	if (size != expected)
+		goto out;
+
+	start = vma->vm_start;
+	for (ring = &nlk->rx_ring; ring <= &nlk->tx_ring; ring++) {
+		if (ring->pg_vec == NULL)
+			continue;
+
+		for (i = 0; i < ring->pg_vec_len; i++) {
+			struct page *page;
+			void *kaddr = ring->pg_vec[i];
+			unsigned int pg_num;
+
+			for (pg_num = 0; pg_num < ring->pg_vec_pages; pg_num++) {
+				page = pgvec_to_page(kaddr);
+				err = vm_insert_page(vma, start, page);
+				if (err < 0)
+					goto out;
+				start += PAGE_SIZE;
+				kaddr += PAGE_SIZE;
+			}
+		}
+	}
+
+	atomic_inc(&nlk->mapped);
+	vma->vm_ops = &netlink_mmap_ops;
+	err = 0;
+out:
+	mutex_unlock(&nlk->pg_vec_lock);
+	return 0;
+}
+#else /* CONFIG_NETLINK_MMAP */
+#define netlink_mmap			sock_no_mmap
+#endif /* CONFIG_NETLINK_MMAP */
+
 static void netlink_destroy_callback(struct netlink_callback *cb)
 {
 	kfree_skb(cb->skb);
@@ -191,6 +440,18 @@ static void netlink_sock_destruct(struct sock *sk)
 	}
 
 	skb_queue_purge(&sk->sk_receive_queue);
+#ifdef CONFIG_NETLINK_MMAP
+	if (1) {
+		struct nl_mmap_req req;
+
+		memset(&req, 0, sizeof(req));
+		if (nlk->rx_ring.pg_vec)
+			netlink_set_ring(sk, &req, true, false);
+		memset(&req, 0, sizeof(req));
+		if (nlk->tx_ring.pg_vec)
+			netlink_set_ring(sk, &req, true, true);
+	}
+#endif /* CONFIG_NETLINK_MMAP */
 
 	if (!sock_flag(sk, SOCK_DEAD)) {
 		printk(KERN_ERR "Freeing alive netlink socket %p\n", sk);
@@ -452,6 +713,9 @@ static int __netlink_create(struct net *net, struct socket *sock,
 		mutex_init(nlk->cb_mutex);
 	}
 	init_waitqueue_head(&nlk->wait);
+#ifdef CONFIG_NETLINK_MMAP
+	mutex_init(&nlk->pg_vec_lock);
+#endif
 
 	sk->sk_destruct = netlink_sock_destruct;
 	sk->sk_protocol = protocol;
@@ -1246,7 +1510,8 @@ static int netlink_setsockopt(struct socket *sock, int level, int optname,
 	if (level != SOL_NETLINK)
 		return -ENOPROTOOPT;
 
-	if (optlen >= sizeof(int) &&
+	if (optname != NETLINK_RX_RING && optname != NETLINK_TX_RING &&
+	    optlen >= sizeof(int) &&
 	    get_user(val, (unsigned int __user *)optval))
 		return -EFAULT;
 
@@ -1295,6 +1560,25 @@ static int netlink_setsockopt(struct socket *sock, int level, int optname,
 		}
 		err = 0;
 		break;
+#ifdef CONFIG_NETLINK_MMAP
+	case NETLINK_RX_RING:
+	case NETLINK_TX_RING: {
+		struct nl_mmap_req req;
+
+		/* Rings might consume more memory than queue limits, require
+		 * CAP_NET_ADMIN.
+		 */
+		if (!capable(CAP_NET_ADMIN))
+			return -EPERM;
+		if (optlen < sizeof(req))
+			return -EINVAL;
+		if (copy_from_user(&req, optval, sizeof(req)))
+			return -EFAULT;
+		err = netlink_set_ring(sk, &req, false,
+				       optname == NETLINK_TX_RING);
+		break;
+	}
+#endif /* CONFIG_NETLINK_MMAP */
 	default:
 		err = -ENOPROTOOPT;
 	}
@@ -2124,7 +2408,7 @@ static const struct proto_ops netlink_ops = {
 	.getsockopt =	netlink_getsockopt,
 	.sendmsg =	netlink_sendmsg,
 	.recvmsg =	netlink_recvmsg,
-	.mmap =		sock_no_mmap,
+	.mmap =		netlink_mmap,
 	.sendpage =	sock_no_sendpage,
 };
 
-- 
1.7.7.6

[PATCH 06/11] netlink: add mmap'ed netlink helper functions

[Patrick McHardy]

Add helper functions for looking up mmap'ed frame headers, reading and writing
their status, allocating skbs with mmap'ed data areas and a poll function.

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 include/linux/netlink.h  |    8 ++
 net/netlink/af_netlink.c |  185 +++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 191 insertions(+), 2 deletions(-)

diff --git a/include/linux/netlink.h b/include/linux/netlink.h
index 3600461..98754e8 100644
--- a/include/linux/netlink.h
+++ b/include/linux/netlink.h
@@ -193,10 +193,18 @@ static inline struct nlmsghdr *nlmsg_hdr(const struct sk_buff *skb)
 	return (struct nlmsghdr *)skb->data;
 }
 
+enum netlink_skb_flags {
+	NETLINK_SKB_MMAPED	= 0x1,		/* Packet data is mmapped */
+	NETLINK_SKB_TX		= 0x2,		/* Packet was sent by userspace */
+	NETLINK_SKB_DELIVERED	= 0x4,		/* Packet was delivered */
+};
+
 struct netlink_skb_parms {
 	struct ucred		creds;		/* Skb credentials	*/
 	__u32			pid;
 	__u32			dst_group;
+	__u32			flags;
+	struct sock		*sk;		/* socket owning mmaped ring */
 };
 
 #define NETLINK_CB(skb)		(*(struct netlink_skb_parms*)&((skb)->cb))
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index e20a918..3810911 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -56,6 +56,7 @@
 #include <linux/audit.h>
 #include <linux/mutex.h>
 #include <linux/vmalloc.h>
+#include <asm/cacheflush.h>
 
 #include <net/net_namespace.h>
 #include <net/sock.h>
@@ -159,6 +160,7 @@ static struct netlink_table *nl_table;
 static DECLARE_WAIT_QUEUE_HEAD(nl_table_wait);
 
 static int netlink_dump(struct sock *sk);
+static void netlink_skb_destructor(struct sk_buff *skb);
 
 static DEFINE_RWLOCK(nl_table_lock);
 static atomic_t nl_table_users = ATOMIC_INIT(0);
@@ -176,6 +178,11 @@ static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid
 }
 
 #ifdef CONFIG_NETLINK_MMAP
+static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
+{
+	return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
+}
+
 static __pure struct page *pgvec_to_page(const void *addr)
 {
 	if (is_vmalloc_addr(addr))
@@ -399,8 +406,154 @@ out:
 	mutex_unlock(&nlk->pg_vec_lock);
 	return 0;
 }
+
+static void netlink_frame_flush_dcache(const struct nl_mmap_hdr *hdr)
+{
+#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
+	struct page *p_start, *p_end;
+
+	/* First page is flushed through netlink_{get,set}_status */
+	p_start = pgvec_to_page(hdr + PAGE_SIZE);
+	p_end   = pgvec_to_page((void *)hdr + NL_MMAP_MSG_HDRLEN + hdr->nm_len - 1);
+	while (p_start <= p_end) {
+		flush_dcache_page(p_start);
+		p_start++;
+	}
+#endif
+}
+
+static enum nl_mmap_status netlink_get_status(const struct nl_mmap_hdr *hdr)
+{
+	smp_rmb();
+	flush_dcache_page(pgvec_to_page(hdr));
+	return hdr->nm_status;
+}
+
+static void netlink_set_status(struct nl_mmap_hdr *hdr,
+			       enum nl_mmap_status status)
+{
+	hdr->nm_status = status;
+	flush_dcache_page(pgvec_to_page(hdr));
+	smp_wmb();
+}
+
+static struct nl_mmap_hdr *
+__netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos)
+{
+	unsigned int pg_vec_pos, frame_off;
+
+	pg_vec_pos = pos / ring->frames_per_block;
+	frame_off  = pos % ring->frames_per_block;
+
+	return ring->pg_vec[pg_vec_pos] + (frame_off * ring->frame_size);
+}
+
+static struct nl_mmap_hdr *
+netlink_lookup_frame(const struct netlink_ring *ring, unsigned int pos,
+		     enum nl_mmap_status status)
+{
+	struct nl_mmap_hdr *hdr;
+
+	hdr = __netlink_lookup_frame(ring, pos);
+	if (netlink_get_status(hdr) != status)
+		return NULL;
+
+	return hdr;
+}
+
+static struct nl_mmap_hdr *
+netlink_current_frame(const struct netlink_ring *ring,
+		      enum nl_mmap_status status)
+{
+	return netlink_lookup_frame(ring, ring->head, status);
+}
+
+static struct nl_mmap_hdr *
+netlink_previous_frame(const struct netlink_ring *ring,
+		       enum nl_mmap_status status)
+{
+	unsigned int prev;
+
+	prev = ring->head ? ring->head - 1 : ring->frame_max;
+	return netlink_lookup_frame(ring, prev, status);
+}
+
+static void netlink_increment_head(struct netlink_ring *ring)
+{
+	ring->head = ring->head != ring->frame_max ? ring->head + 1 : 0;
+}
+
+static void netlink_forward_ring(struct netlink_ring *ring)
+{
+	unsigned int head = ring->head, pos = head;
+	const struct nl_mmap_hdr *hdr;
+
+	do {
+		hdr = __netlink_lookup_frame(ring, pos);
+		if (hdr->nm_status == NL_MMAP_STATUS_UNUSED)
+			break;
+		if (hdr->nm_status != NL_MMAP_STATUS_SKIP)
+			break;
+		netlink_increment_head(ring);
+	} while (ring->head != head);
+}
+
+static unsigned int netlink_poll(struct file *file, struct socket *sock,
+				 poll_table *wait)
+{
+	struct sock *sk = sock->sk;
+	struct netlink_sock *nlk = nlk_sk(sk);
+	unsigned int mask;
+
+	mask = datagram_poll(file, sock, wait);
+
+	spin_lock_bh(&sk->sk_receive_queue.lock);
+	if (nlk->rx_ring.pg_vec) {
+		netlink_forward_ring(&nlk->rx_ring);
+		if (!netlink_previous_frame(&nlk->rx_ring, NL_MMAP_STATUS_UNUSED))
+			mask |= POLLIN | POLLRDNORM;
+	}
+	spin_unlock_bh(&sk->sk_receive_queue.lock);
+
+	spin_lock_bh(&sk->sk_write_queue.lock);
+	if (nlk->tx_ring.pg_vec) {
+		if (netlink_current_frame(&nlk->tx_ring, NL_MMAP_STATUS_UNUSED))
+			mask |= POLLOUT | POLLWRNORM;
+	}
+	spin_unlock_bh(&sk->sk_write_queue.lock);
+
+	return mask;
+}
+
+static struct nl_mmap_hdr *netlink_mmap_hdr(struct sk_buff *skb)
+{
+	return (struct nl_mmap_hdr *)(skb->head - NL_MMAP_HDRLEN);
+}
+
+static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
+				   struct netlink_ring *ring,
+				   struct nl_mmap_hdr *hdr)
+{
+	unsigned int size;
+	void *data;
+
+	size = ring->frame_size - NL_MMAP_HDRLEN;
+	data = (void *)hdr + NL_MMAP_HDRLEN;
+
+	skb->head	= data;
+	skb->data	= data;
+	skb_reset_tail_pointer(skb);
+	skb->end	= skb->tail + size;
+	skb->len	= 0;
+
+	skb->destructor	= netlink_skb_destructor;
+	NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
+	NETLINK_CB(skb).sk = sk;
+}
 #else /* CONFIG_NETLINK_MMAP */
+#define netlink_skb_is_mmaped(skb)	false
 #define netlink_mmap			sock_no_mmap
+#define netlink_poll			datagram_poll
 #endif /* CONFIG_NETLINK_MMAP */
 
 static void netlink_destroy_callback(struct netlink_callback *cb)
@@ -417,7 +570,35 @@ static void netlink_consume_callback(struct netlink_callback *cb)
 
 static void netlink_skb_destructor(struct sk_buff *skb)
 {
-	sock_rfree(skb);
+#ifdef CONFIG_NETLINK_MMAP
+	struct nl_mmap_hdr *hdr;
+	struct netlink_ring *ring;
+	struct sock *sk;
+
+	/* If a packet from the kernel to userspace was freed because of an
+	 * error without being delivered to userspace, the kernel must reset
+	 * the status. In the direction userspace to kernel, the status is
+	 * always reset here after the packet was processed and freed.
+	 */
+	if (netlink_skb_is_mmaped(skb)) {
+		hdr = netlink_mmap_hdr(skb);
+		sk = NETLINK_CB(skb).sk;
+
+		if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
+			hdr->nm_len = 0;
+			netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
+		}
+		ring = &nlk_sk(sk)->rx_ring;
+
+		WARN_ON(atomic_read(&ring->pending) == 0);
+		atomic_dec(&ring->pending);
+		sock_put(sk);
+
+		skb->data = NULL;
+	}
+#endif
+	if (skb->sk != NULL)
+		sock_rfree(skb);
 }
 
 static void netlink_skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
@@ -2400,7 +2581,7 @@ static const struct proto_ops netlink_ops = {
 	.socketpair =	sock_no_socketpair,
 	.accept =	sock_no_accept,
 	.getname =	netlink_getname,
-	.poll =		datagram_poll,
+	.poll =		netlink_poll,
 	.ioctl =	sock_no_ioctl,
 	.listen =	sock_no_listen,
 	.shutdown =	sock_no_shutdown,
-- 
1.7.7.6

[PATCH 07/11] netlink: implement memory mapped sendmsg()

[Patrick McHardy]

Add support for mmap'ed sendmsg() to netlink.

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 net/netlink/af_netlink.c |  132 +++++++++++++++++++++++++++++++++++++++++++--
 1 files changed, 126 insertions(+), 6 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 3810911..65867fd 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -183,6 +183,11 @@ static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
 	return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
 }
 
+static bool netlink_tx_is_mmaped(struct sock *sk)
+{
+	return nlk_sk(sk)->tx_ring.pg_vec != NULL;
+}
+
 static __pure struct page *pgvec_to_page(const void *addr)
 {
 	if (is_vmalloc_addr(addr))
@@ -505,6 +510,9 @@ static unsigned int netlink_poll(struct file *file, struct socket *sock,
 	struct netlink_sock *nlk = nlk_sk(sk);
 	unsigned int mask;
 
+	if (nlk->cb != NULL && nlk->rx_ring.pg_vec != NULL)
+		netlink_dump(sk);
+
 	mask = datagram_poll(file, sock, wait);
 
 	spin_lock_bh(&sk->sk_receive_queue.lock);
@@ -550,10 +558,108 @@ static void netlink_ring_setup_skb(struct sk_buff *skb, struct sock *sk,
 	NETLINK_CB(skb).flags |= NETLINK_SKB_MMAPED;
 	NETLINK_CB(skb).sk = sk;
 }
+
+static int netlink_mmap_sendmsg(struct sock *sk, struct msghdr *msg,
+				u32 dst_pid, u32 dst_group,
+				struct sock_iocb *siocb)
+{
+	struct netlink_sock *nlk = nlk_sk(sk);
+	struct netlink_ring *ring;
+	struct nl_mmap_hdr *hdr;
+	struct sk_buff *skb;
+	unsigned int maxlen;
+	bool excl = true;
+	int err = 0, len = 0;
+
+	/* Netlink messages are validated by the receiver before processing.
+	 * In order to avoid userspace changing the contents of the message
+	 * after validation, the socket and the ring may only be used by a
+	 * single process, otherwise we fall back to copying.
+	 */
+	if (atomic_long_read(&sk->sk_socket->file->f_count) > 2 ||
+	    atomic_read(&nlk->mapped) > 1)
+		excl = false;
+
+	mutex_lock(&nlk->pg_vec_lock);
+
+	ring   = &nlk->tx_ring;
+	maxlen = ring->frame_size - NL_MMAP_HDRLEN;
+
+	do {
+		hdr = netlink_current_frame(ring, NL_MMAP_STATUS_VALID);
+		if (hdr == NULL) {
+			if (!(msg->msg_flags & MSG_DONTWAIT) &&
+			    atomic_read(&nlk->tx_ring.pending))
+				schedule();
+			continue;
+		}
+		if (hdr->nm_len > maxlen) {
+			err = -EINVAL;
+			goto out;
+		}
+
+		netlink_frame_flush_dcache(hdr);
+
+		if (likely(dst_pid == 0 && dst_group == 0 && excl)) {
+			skb = alloc_skb_head(GFP_KERNEL);
+			if (skb == NULL) {
+				err = -ENOBUFS;
+				goto out;
+			}
+			sock_hold(sk);
+			netlink_ring_setup_skb(skb, sk, ring, hdr);
+			NETLINK_CB(skb).flags |= NETLINK_SKB_TX;
+			__skb_put(skb, hdr->nm_len);
+			netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
+			atomic_inc(&ring->pending);
+		} else {
+			skb = alloc_skb(hdr->nm_len, GFP_KERNEL);
+			if (skb == NULL) {
+				err = -ENOBUFS;
+				goto out;
+			}
+			__skb_put(skb, hdr->nm_len);
+			memcpy(skb->data, (void *)hdr + NL_MMAP_HDRLEN, hdr->nm_len);
+			netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
+		}
+
+		netlink_increment_head(ring);
+
+		NETLINK_CB(skb).pid	  = nlk->pid;
+		NETLINK_CB(skb).dst_group = dst_group;
+		NETLINK_CB(skb).creds	  = siocb->scm->creds;
+
+		err = security_netlink_send(sk, skb);
+		if (err) {
+			kfree_skb(skb);
+			goto out;
+		}
+
+		if (unlikely(dst_group)) {
+			atomic_inc(&skb->users);
+			netlink_broadcast(sk, skb, dst_pid, dst_group, GFP_KERNEL);
+		}
+		err = netlink_unicast(sk, skb, dst_pid, msg->msg_flags & MSG_DONTWAIT);
+		if (err < 0)
+			goto out;
+		len += err;
+
+	} while (hdr != NULL ||
+		 (!(msg->msg_flags & MSG_DONTWAIT) &&
+		  atomic_read(&nlk->tx_ring.pending)));
+
+	if (len > 0)
+		err = len;
+out:
+	mutex_unlock(&nlk->pg_vec_lock);
+	return err;
+}
 #else /* CONFIG_NETLINK_MMAP */
 #define netlink_skb_is_mmaped(skb)	false
+#define netlink_tx_is_mmaped(sk)	false
 #define netlink_mmap			sock_no_mmap
 #define netlink_poll			datagram_poll
+#define netlink_mmap_sendmsg(sk, msg, dst_pid, dst_group, siocb)	0
 #endif /* CONFIG_NETLINK_MMAP */
 
 static void netlink_destroy_callback(struct netlink_callback *cb)
@@ -584,11 +690,16 @@ static void netlink_skb_destructor(struct sk_buff *skb)
 		hdr = netlink_mmap_hdr(skb);
 		sk = NETLINK_CB(skb).sk;
 
-		if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
-			hdr->nm_len = 0;
-			netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
+		if (NETLINK_CB(skb).flags & NETLINK_SKB_TX) {
+			netlink_set_status(hdr, NL_MMAP_STATUS_UNUSED);
+			ring = &nlk_sk(sk)->tx_ring;
+		} else {
+			if (!(NETLINK_CB(skb).flags & NETLINK_SKB_DELIVERED)) {
+				hdr->nm_len = 0;
+				netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
+			}
+			ring = &nlk_sk(sk)->rx_ring;
 		}
-		ring = &nlk_sk(sk)->rx_ring;
 
 		WARN_ON(atomic_read(&ring->pending) == 0);
 		atomic_dec(&ring->pending);
@@ -1285,8 +1396,9 @@ int netlink_attachskb(struct sock *sk, struct sk_buff *skb,
 
 	nlk = nlk_sk(sk);
 
-	if (atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
-	    test_bit(NETLINK_CONGESTED, &nlk->state)) {
+	if ((atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf ||
+	     test_bit(NETLINK_CONGESTED, &nlk->state)) &&
+	    !netlink_skb_is_mmaped(skb)) {
 		DECLARE_WAITQUEUE(wait, current);
 		if (!*timeo) {
 			if (!ssk || netlink_is_kernel(ssk))
@@ -1346,6 +1458,8 @@ static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
 	int delta;
 
 	WARN_ON(skb->sk != NULL);
+	if (netlink_skb_is_mmaped(skb))
+		return skb;
 
 	delta = skb->end - skb->tail;
 	if (delta * 2 < skb->truesize)
@@ -1869,6 +1983,12 @@ static int netlink_sendmsg(struct kiocb *kiocb, struct socket *sock,
 			goto out;
 	}
 
+	if (netlink_tx_is_mmaped(sk) &&
+	    msg->msg_iov->iov_base == NULL) {
+		err = netlink_mmap_sendmsg(sk, msg, dst_pid, dst_group, siocb);
+		goto out;
+	}
+
 	err = -EMSGSIZE;
 	if (len > sk->sk_sndbuf - 32)
 		goto out;
-- 
1.7.7.6

[PATCH 08/11] netlink: implement memory mapped recvmsg()

[Patrick McHardy]

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 include/linux/netlink.h  |    2 +
 net/netlink/af_netlink.c |  145 ++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 143 insertions(+), 4 deletions(-)

diff --git a/include/linux/netlink.h b/include/linux/netlink.h
index 98754e8..144ef3a 100644
--- a/include/linux/netlink.h
+++ b/include/linux/netlink.h
@@ -232,6 +232,8 @@ extern void __netlink_clear_multicast_users(struct sock *sk, unsigned int group)
 extern void netlink_clear_multicast_users(struct sock *sk, unsigned int group);
 extern void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err);
 extern int netlink_has_listeners(struct sock *sk, unsigned int group);
+extern struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
+					 u32 dst_pid, gfp_t gfp_mask);
 extern int netlink_unicast(struct sock *ssk, struct sk_buff *skb, __u32 pid, int nonblock);
 extern int netlink_broadcast(struct sock *ssk, struct sk_buff *skb, __u32 pid,
 			     __u32 group, gfp_t allocation);
diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index 65867fd..c42a601 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -183,6 +183,11 @@ static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
 	return NETLINK_CB(skb).flags & NETLINK_SKB_MMAPED;
 }
 
+static bool netlink_rx_is_mmaped(struct sock *sk)
+{
+	return nlk_sk(sk)->rx_ring.pg_vec != NULL;
+}
+
 static bool netlink_tx_is_mmaped(struct sock *sk)
 {
 	return nlk_sk(sk)->tx_ring.pg_vec != NULL;
@@ -654,8 +659,54 @@ out:
 	mutex_unlock(&nlk->pg_vec_lock);
 	return err;
 }
+
+static void netlink_queue_mmaped_skb(struct sock *sk, struct sk_buff *skb)
+{
+	struct nl_mmap_hdr *hdr;
+
+	hdr = netlink_mmap_hdr(skb);
+	hdr->nm_len	= skb->len;
+	hdr->nm_group	= NETLINK_CB(skb).dst_group;
+	hdr->nm_pid	= NETLINK_CB(skb).creds.pid;
+	hdr->nm_uid	= NETLINK_CB(skb).creds.uid;
+	hdr->nm_gid	= NETLINK_CB(skb).creds.gid;
+	netlink_frame_flush_dcache(hdr);
+	netlink_set_status(hdr, NL_MMAP_STATUS_VALID);
+
+	NETLINK_CB(skb).flags |= NETLINK_SKB_DELIVERED;
+	kfree_skb(skb);
+}
+
+static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
+{
+	struct netlink_sock *nlk = nlk_sk(sk);
+	struct netlink_ring *ring = &nlk->rx_ring;
+	struct nl_mmap_hdr *hdr;
+
+	spin_lock_bh(&sk->sk_receive_queue.lock);
+	hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
+	if (hdr == NULL) {
+		spin_unlock_bh(&sk->sk_receive_queue.lock);
+		kfree_skb(skb);
+		sk->sk_err = ENOBUFS;
+		sk->sk_error_report(sk);
+		return;
+	}
+	netlink_increment_head(ring);
+	__skb_queue_tail(&sk->sk_receive_queue, skb);
+	spin_unlock_bh(&sk->sk_receive_queue.lock);
+
+	hdr->nm_len	= skb->len;
+	hdr->nm_group	= NETLINK_CB(skb).dst_group;
+	hdr->nm_pid	= NETLINK_CB(skb).creds.pid;
+	hdr->nm_uid	= NETLINK_CB(skb).creds.uid;
+	hdr->nm_gid	= NETLINK_CB(skb).creds.gid;
+	netlink_set_status(hdr, NL_MMAP_STATUS_COPY);
+}
+
 #else /* CONFIG_NETLINK_MMAP */
 #define netlink_skb_is_mmaped(skb)	false
+#define netlink_rx_is_mmaped(sk)	false
 #define netlink_tx_is_mmaped(sk)	false
 #define netlink_mmap			sock_no_mmap
 #define netlink_poll			datagram_poll
@@ -1434,7 +1485,14 @@ static int __netlink_sendskb(struct sock *sk, struct sk_buff *skb)
 {
 	int len = skb->len;
 
-	skb_queue_tail(&sk->sk_receive_queue, skb);
+#ifdef CONFIG_NETLINK_MMAP
+	if (netlink_skb_is_mmaped(skb))
+		netlink_queue_mmaped_skb(sk, skb);
+	else if (netlink_rx_is_mmaped(sk))
+		netlink_ring_set_copied(sk, skb);
+	else
+#endif /* CONFIG_NETLINK_MMAP */
+		skb_queue_tail(&sk->sk_receive_queue, skb);
 	sk->sk_data_ready(sk, len);
 	return len;
 }
@@ -1543,6 +1601,68 @@ retry:
 }
 EXPORT_SYMBOL(netlink_unicast);
 
+struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
+				  u32 dst_pid, gfp_t gfp_mask)
+{
+#ifdef CONFIG_NETLINK_MMAP
+	struct sock *sk = NULL;
+	struct sk_buff *skb;
+	struct netlink_ring *ring;
+	struct nl_mmap_hdr *hdr;
+	unsigned int maxlen;
+
+	sk = netlink_getsockbypid(ssk, dst_pid);
+	if (IS_ERR(sk))
+		goto out;
+
+	ring = &nlk_sk(sk)->rx_ring;
+	/* fast-path without atomic ops for common case: non-mmaped receiver */
+	if (ring->pg_vec == NULL)
+		goto out_put;
+
+	skb = alloc_skb_head(gfp_mask);
+	if (skb == NULL)
+		goto err1;
+
+	spin_lock_bh(&sk->sk_receive_queue.lock);
+	/* check again under lock */
+	if (ring->pg_vec == NULL)
+		goto out_free;
+
+	maxlen = ring->frame_size - NL_MMAP_HDRLEN;
+	if (maxlen < size)
+		goto out_free;
+
+	netlink_forward_ring(ring);
+	hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
+	if (hdr == NULL)
+		goto err2;
+	netlink_ring_setup_skb(skb, sk, ring, hdr);
+	netlink_set_status(hdr, NL_MMAP_STATUS_RESERVED);
+	atomic_inc(&ring->pending);
+	netlink_increment_head(ring);
+
+	spin_unlock_bh(&sk->sk_receive_queue.lock);
+	return skb;
+
+err2:
+	kfree_skb(skb);
+	spin_unlock_bh(&sk->sk_receive_queue.lock);
+err1:
+	sock_put(sk);
+	return NULL;
+
+out_free:
+	kfree_skb(skb);
+	spin_unlock_bh(&sk->sk_receive_queue.lock);
+out_put:
+	sock_put(sk);
+out:
+#endif
+	return alloc_skb(size, gfp_mask);
+}
+EXPORT_SYMBOL_GPL(netlink_alloc_skb);
+
 int netlink_has_listeners(struct sock *sk, unsigned int group)
 {
 	int res = 0;
@@ -2328,9 +2448,13 @@ static int netlink_dump(struct sock *sk)
 
 	alloc_size = max_t(int, cb->min_dump_alloc, NLMSG_GOODSIZE);
 
-	skb = sock_rmalloc(sk, alloc_size, 0, GFP_KERNEL);
+	if (!netlink_rx_is_mmaped(sk) &&
+	    atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
+		goto errout_skb;
+	skb = netlink_alloc_skb(sk, alloc_size, nlk->pid, GFP_KERNEL);
 	if (!skb)
 		goto errout_skb;
+	netlink_skb_set_owner_r(skb, sk);
 
 	len = cb->dump(skb, cb);
 
@@ -2384,12 +2508,24 @@ int netlink_dump_start(struct sock *ssk, struct sk_buff *skb,
 	if (cb == NULL)
 		return -ENOBUFS;
 
+	/* Memory mapped dump requests need to be copied to avoid looping
+	 * on the pending state in netlink_mmap_sendmsg() while the cb holds
+	 * a reference to the skb.
+	 */
+	if (netlink_skb_is_mmaped(skb)) {
+		skb = skb_copy(skb, GFP_KERNEL);
+		if (skb == NULL) {
+			kfree(cb);
+			return -ENOBUFS;
+		}
+	} else
+		atomic_inc(&skb->users);
+
 	cb->dump = control->dump;
 	cb->done = control->done;
 	cb->nlh = nlh;
 	cb->data = control->data;
 	cb->min_dump_alloc = control->min_dump_alloc;
-	atomic_inc(&skb->users);
 	cb->skb = skb;
 
 	sk = netlink_lookup(sock_net(ssk), ssk->sk_protocol, NETLINK_CB(skb).pid);
@@ -2434,7 +2570,8 @@ void netlink_ack(struct sk_buff *in_skb, struct nlmsghdr *nlh, int err)
 	if (err)
 		payload += nlmsg_len(nlh);
 
-	skb = nlmsg_new(payload, GFP_KERNEL);
+	skb = netlink_alloc_skb(in_skb->sk, nlmsg_total_size(payload),
+				NETLINK_CB(in_skb).pid, GFP_KERNEL);
 	if (!skb) {
 		struct sock *sk;
 
-- 
1.7.7.6

[PATCH 09/11] nfnetlink: add support for memory mapped netlink

[Patrick McHardy]

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 include/linux/netfilter/nfnetlink.h  |    2 ++
 net/netfilter/nfnetlink.c            |    7 +++++++
 net/netfilter/nfnetlink_log.c        |    9 +++++----
 net/netfilter/nfnetlink_queue_core.c |    2 +-
 4 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/include/linux/netfilter/nfnetlink.h b/include/linux/netfilter/nfnetlink.h
index 18341cd..447d1a7 100644
--- a/include/linux/netfilter/nfnetlink.h
+++ b/include/linux/netfilter/nfnetlink.h
@@ -81,6 +81,8 @@ extern int nfnetlink_subsys_register(const struct nfnetlink_subsystem *n);
 extern int nfnetlink_subsys_unregister(const struct nfnetlink_subsystem *n);
 
 extern int nfnetlink_has_listeners(struct net *net, unsigned int group);
+extern struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
+					   u32 dst_pid, gfp_t gfp_mask);
 extern int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 pid, unsigned int group,
 			  int echo, gfp_t flags);
 extern int nfnetlink_set_err(struct net *net, u32 pid, u32 group, int error);
diff --git a/net/netfilter/nfnetlink.c b/net/netfilter/nfnetlink.c
index a265033..d0aa9fc 100644
--- a/net/netfilter/nfnetlink.c
+++ b/net/netfilter/nfnetlink.c
@@ -111,6 +111,13 @@ int nfnetlink_has_listeners(struct net *net, unsigned int group)
 }
 EXPORT_SYMBOL_GPL(nfnetlink_has_listeners);
 
+struct sk_buff *nfnetlink_alloc_skb(struct net *net, unsigned int size,
+				    u32 dst_pid, gfp_t gfp_mask)
+{
+	return netlink_alloc_skb(net->nfnl, size, dst_pid, gfp_mask);
+}
+EXPORT_SYMBOL_GPL(nfnetlink_alloc_skb);
+
 int nfnetlink_send(struct sk_buff *skb, struct net *net, u32 pid,
 		   unsigned int group, int echo, gfp_t flags)
 {
diff --git a/net/netfilter/nfnetlink_log.c b/net/netfilter/nfnetlink_log.c
index 169ab59..6232ca8 100644
--- a/net/netfilter/nfnetlink_log.c
+++ b/net/netfilter/nfnetlink_log.c
@@ -296,7 +296,7 @@ nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
 }
 
 static struct sk_buff *
-nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
+nfulnl_alloc_skb(u32 peer_pid, unsigned int inst_size, unsigned int pkt_size)
 {
 	struct sk_buff *skb;
 	unsigned int n;
@@ -305,13 +305,14 @@ nfulnl_alloc_skb(unsigned int inst_size, unsigned int pkt_size)
 	 * message.  WARNING: has to be <= 128k due to slab restrictions */
 
 	n = max(inst_size, pkt_size);
-	skb = alloc_skb(n, GFP_ATOMIC);
+	skb = nfnetlink_alloc_skb(&init_net, n, peer_pid, GFP_ATOMIC);
 	if (!skb) {
 		if (n > pkt_size) {
 			/* try to allocate only as much as we need for current
 			 * packet */
 
-			skb = alloc_skb(pkt_size, GFP_ATOMIC);
+			skb = nfnetlink_alloc_skb(&init_net, pkt_size, peer_pid,
+						  GFP_ATOMIC);
 			if (!skb)
 				pr_err("nfnetlink_log: can't even alloc %u bytes\n",
 				       pkt_size);
@@ -658,7 +659,7 @@ nfulnl_log_packet(u_int8_t pf,
 	}
 
 	if (!inst->skb) {
-		inst->skb = nfulnl_alloc_skb(inst->nlbufsiz, size);
+		inst->skb = nfulnl_alloc_skb(inst->peer_pid, inst->nlbufsiz, size);
 		if (!inst->skb)
 			goto alloc_failure;
 	}
diff --git a/net/netfilter/nfnetlink_queue_core.c b/net/netfilter/nfnetlink_queue_core.c
index c0496a5..6983b2e 100644
--- a/net/netfilter/nfnetlink_queue_core.c
+++ b/net/netfilter/nfnetlink_queue_core.c
@@ -272,7 +272,7 @@ nfqnl_build_packet_message(struct nfqnl_instance *queue,
 	if (queue->flags & NFQA_CFG_F_CONNTRACK)
 		ct = nfqnl_ct_get(entskb, &size, &ctinfo);
 
-	skb = alloc_skb(size, GFP_ATOMIC);
+	skb = nfnetlink_alloc_skb(&init_net, size, queue->peer_pid, GFP_ATOMIC);
 	if (!skb)
 		return NULL;
 
-- 
1.7.7.6

[PATCH 10/11] netlink: add flow control for memory mapped I/O

[Patrick McHardy]

Add flow control for memory mapped RX. Since user-space usually doesn't
invoke recvmsg() when using memory mapped I/O, flow control is performed
in netlink_poll(). Dumps are allowed to continue if at least half of the
ring frames are unused.

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 net/netlink/af_netlink.c |   87 +++++++++++++++++++++++++++++++--------------
 1 files changed, 60 insertions(+), 27 deletions(-)

diff --git a/net/netlink/af_netlink.c b/net/netlink/af_netlink.c
index c42a601..cc90aa8 100644
--- a/net/netlink/af_netlink.c
+++ b/net/netlink/af_netlink.c
@@ -177,6 +177,29 @@ static inline struct hlist_head *nl_pid_hashfn(struct nl_pid_hash *hash, u32 pid
 	return &hash->table[jhash_1word(pid, hash->rnd) & hash->mask];
 }
 
+static void netlink_overrun(struct sock *sk)
+{
+	struct netlink_sock *nlk = nlk_sk(sk);
+
+	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
+		if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
+			sk->sk_err = ENOBUFS;
+			sk->sk_error_report(sk);
+		}
+	}
+	atomic_inc(&sk->sk_drops);
+}
+
+static void netlink_rcv_wake(struct sock *sk)
+{
+	struct netlink_sock *nlk = nlk_sk(sk);
+
+	if (skb_queue_empty(&sk->sk_receive_queue))
+		clear_bit(NETLINK_CONGESTED, &nlk->state);
+	if (!test_bit(NETLINK_CONGESTED, &nlk->state))
+		wake_up_interruptible(&nlk->wait);
+}
+
 #ifdef CONFIG_NETLINK_MMAP
 static bool netlink_skb_is_mmaped(const struct sk_buff *skb)
 {
@@ -508,15 +531,48 @@ static void netlink_forward_ring(struct netlink_ring *ring)
 	} while (ring->head != head);
 }
 
+static bool netlink_dump_space(struct netlink_sock *nlk)
+{
+	struct netlink_ring *ring = &nlk->rx_ring;
+	struct nl_mmap_hdr *hdr;
+	unsigned int n;
+
+	hdr = netlink_current_frame(ring, NL_MMAP_STATUS_UNUSED);
+	if (hdr == NULL)
+		return false;
+
+	n = ring->head + ring->frame_max / 2;
+	if (n > ring->frame_max)
+		n -= ring->frame_max;
+
+	hdr = __netlink_lookup_frame(ring, n);
+
+	return hdr->nm_status == NL_MMAP_STATUS_UNUSED;
+}
+
 static unsigned int netlink_poll(struct file *file, struct socket *sock,
 				 poll_table *wait)
 {
 	struct sock *sk = sock->sk;
 	struct netlink_sock *nlk = nlk_sk(sk);
 	unsigned int mask;
+	int err;
 
-	if (nlk->cb != NULL && nlk->rx_ring.pg_vec != NULL)
-		netlink_dump(sk);
+	if (nlk->rx_ring.pg_vec != NULL) {
+		/* Memory mapped sockets don't call recvmsg(), so flow control
+		 * for dumps is performed here. A dump is allowed to continue
+		 * if at least half the ring is unused.
+		 */
+		while (nlk->cb != NULL && netlink_dump_space(nlk)) {
+			err = netlink_dump(sk);
+			if (err < 0) {
+				sk->sk_err = err;
+				sk->sk_error_report(sk);
+				break;
+			}
+		}
+		netlink_rcv_wake(sk);
+	}
 
 	mask = datagram_poll(file, sock, wait);
 
@@ -688,8 +744,7 @@ static void netlink_ring_set_copied(struct sock *sk, struct sk_buff *skb)
 	if (hdr == NULL) {
 		spin_unlock_bh(&sk->sk_receive_queue.lock);
 		kfree_skb(skb);
-		sk->sk_err = ENOBUFS;
-		sk->sk_error_report(sk);
+		netlink_overrun(sk);
 		return;
 	}
 	netlink_increment_head(ring);
@@ -1382,19 +1437,6 @@ static int netlink_getname(struct socket *sock, struct sockaddr *addr,
 	return 0;
 }
 
-static void netlink_overrun(struct sock *sk)
-{
-	struct netlink_sock *nlk = nlk_sk(sk);
-
-	if (!(nlk->flags & NETLINK_RECV_NO_ENOBUFS)) {
-		if (!test_and_set_bit(NETLINK_CONGESTED, &nlk_sk(sk)->state)) {
-			sk->sk_err = ENOBUFS;
-			sk->sk_error_report(sk);
-		}
-	}
-	atomic_inc(&sk->sk_drops);
-}
-
 static struct sock *netlink_getsockbypid(struct sock *ssk, u32 pid)
 {
 	struct sock *sock;
@@ -1537,16 +1579,6 @@ static struct sk_buff *netlink_trim(struct sk_buff *skb, gfp_t allocation)
 	return skb;
 }
 
-static void netlink_rcv_wake(struct sock *sk)
-{
-	struct netlink_sock *nlk = nlk_sk(sk);
-
-	if (skb_queue_empty(&sk->sk_receive_queue))
-		clear_bit(NETLINK_CONGESTED, &nlk->state);
-	if (!test_bit(NETLINK_CONGESTED, &nlk->state))
-		wake_up_interruptible(&nlk->wait);
-}
-
 static int netlink_unicast_kernel(struct sock *sk, struct sk_buff *skb)
 {
 	int ret;
@@ -1648,6 +1680,7 @@ struct sk_buff *netlink_alloc_skb(struct sock *ssk, unsigned int size,
 err2:
 	kfree_skb(skb);
 	spin_unlock_bh(&sk->sk_receive_queue.lock);
+	netlink_overrun(sk);
 err1:
 	sock_put(sk);
 	return NULL;
-- 
1.7.7.6

[PATCH 11/11] netlink: add documentation for memory mapped I/O

[Patrick McHardy]

Signed-off-by: Patrick McHardy <kaber@trash.net>
---
 Documentation/networking/netlink_mmap.txt |  337 +++++++++++++++++++++++++++++
 1 files changed, 337 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/networking/netlink_mmap.txt

diff --git a/Documentation/networking/netlink_mmap.txt b/Documentation/networking/netlink_mmap.txt
new file mode 100644
index 0000000..413c521
--- /dev/null
+++ b/Documentation/networking/netlink_mmap.txt
@@ -0,0 +1,337 @@
+This file documents how to use memory mapped I/O with netlink.
+
+Overview
+--------
+
+Memory mapped netlink I/O can be used to increase throughput and decrease
+overhead of unicast receive and transmit operations. Some netlink subsystems
+require high throughput, these are mainly the netfilter subsystems
+nfnetlink_queue and nfnetlink_log, but it can also help speed up large
+dump operations of f.i. the routing database.
+
+Memory mapped netlink I/O used two circular ring buffers for RX and TX which
+are mapped into the processes address space.
+
+The RX ring is used by the kernel to directly construct netlink messages into
+user-space memory without copying them as done with regular socket I/O,
+additionally as long as the ring contains messages no recvmsg() or poll()
+syscalls have to be issued by user-space to get more message.
+
+The TX ring is used to process messages directly from user-space memory, the
+kernel processes all messages contained in the ring using a single sendmsg()
+call.
+
+Usage overview
+--------------
+
+In order to use memory mapped netlink I/O, user-space needs three main changes:
+
+- ring setup
+- conversion of the RX path to get messages from the ring instead of recvmsg()
+- conversion of the TX path to construct messages into the ring
+
+Ring setup is done using setsockopt() to provide the ring parameters to the
+kernel, then a call to mmap() to map the ring into the processes address space:
+
+- setsockopt(fd, SOL_NETLINK, NETLINK_RX_RING, &params, sizeof(params));
+- setsockopt(fd, SOL_NETLINK, NETLINK_TX_RING, &params, sizeof(params));
+- ring = mmap(NULL, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0)
+
+Usage of either ring is optional, but even if only the RX ring is used the
+mapping still needs to be writable in order to update the frame status after
+processing.
+
+Conversion of the reception path involves calling poll() on the file
+descriptor, once the socket is readable the frames from the ring are
+processsed in order until no more messages are available, as indicated by
+a status word in the frame header.
+
+On kernel side, in order to make use of memory mapped I/O on receive, the
+originating netlink subsystem needs to support memory mapped I/O, otherwise
+it will use an allocated socket buffer as usual and the contents will be
+ copied to the ring on transmission, nullifying most of the performance gains.
+Dumps of kernel databases automatically support memory mapped I/O.
+
+Conversion of the transmit path involves changing message contruction to
+use memory from the TX ring instead of (usually) a buffer declared on the
+stack and setting up the frame header approriately. Optionally poll() can
+be used to wait for free frames in the TX ring.
+
+Structured and definitions for using memory mapped I/O are contained in
+<linux/netlink.h>.
+
+RX and TX rings
+----------------
+
+Each ring contains a number of continous memory blocks, containing frames of
+fixed size dependant on the parameters used for ring setup.
+
+Ring:	[ block 0 ]
+		[ frame 0 ]
+		[ frame 1 ]
+	[ block 1 ]
+		[ frame 2 ]
+		[ frame 3 ]
+	...
+	[ block n ]
+		[ frame 2 * n ]
+		[ frame 2 * n + 1 ]
+
+The blocks are only visible to the kernel, from the point of view of user-space
+the ring just contains the frames in a continous memory zone.
+
+The ring parameters used for setting up the ring are defined as follows:
+
+struct nl_mmap_req {
+	unsigned int	nm_block_size;
+	unsigned int	nm_block_nr;
+	unsigned int	nm_frame_size;
+	unsigned int	nm_frame_nr;
+};
+
+Frames are grouped into blocks, where each block is a continous region of memory
+and holds nm_block_size / nm_frame_size frames. The total number of frames in
+the ring is nm_frame_nr. The following invariants hold:
+
+- frames_per_block = nm_block_size / nm_frame_size
+
+- nm_frame_nr = frames_per_block * nm_block_nr
+
+Some parameters are constrained, specifically:
+
+- nm_block_size must be a multiple of the architectures memory page size.
+  The getpagesize() function can be used to get the page size.
+
+- nm_frame_size must be equal or larger to NL_MMAP_HDRLEN, IOW a frame must be
+  able to hold at least the frame header
+
+- nm_frame_size must be smaller or equal to nm_block_size
+
+- nm_frame_size must be a multiple of NL_MMAP_MSG_ALIGNMENT
+
+- nm_frame_nr must equal the actual number of frames as specified above.
+
+When the kernel can't allocate phsyically continous memory for a ring block,
+it will fall back to use physically discontinous memory. This might affect
+performance negatively, in order to avoid this the nm_frame_size parameter
+should be chosen to be as small as possible for the required frame size and
+the number of blocks should be increased instead.
+
+Ring frames
+------------
+
+Each frames contain a frame header, consisting of a synchronization word and some
+meta-data, and the message itself.
+
+Frame:	[ header message ]
+
+The frame header is defined as follows:
+
+struct nl_mmap_hdr {
+	unsigned int	nm_status;
+	unsigned int	nm_len;
+	__u32		nm_group;
+	/* credentials */
+	__u32		nm_pid;
+	__u32		nm_uid;
+	__u32		nm_gid;
+};
+
+- nm_status is used for synchronizing processing between the kernel and user-
+  space and specifies ownership of the frame as well as the operation to perform
+
+- nm_len contains the length of the message contained in the data area
+
+- nm_group specified the destination multicast group of message
+
+- nm_pid, nm_uid and nm_gid contain the netlink pid, UID and GID of the sending
+  process. These values correspond to the data available using SOCK_PASSCRED in
+  the SCM_CREDENTIALS cmsg.
+
+The possible values in the status word are:
+
+- NL_MMAP_STATUS_UNUSED:
+	RX ring:	frame belongs to the kernel and contains no message
+			for user-space. Approriate action is to invoke poll()
+			to wait for new messages.
+
+	TX ring:	frame belongs to user-space and can be used for
+			message construction.
+
+- NL_MMAP_STATUS_RESERVED:
+	RX ring only:	frame is currently used by the kernel for message
+			construction and contains no valid message yet.
+			Appropriate action is to invoke poll() to wait for
+			new messages.
+
+- NL_MMAP_STATUS_VALID:
+	RX ring:	frame contains a valid message. Approriate action is
+			to process the message and release the frame back to
+			the kernel by setting the status to
+			NL_MMAP_STATUS_UNUSED or queue the frame by setting the
+			status to NL_MMAP_STATUS_SKIP.
+
+	TX ring:	the frame contains a valid message from user-space to
+			be processed by the kernel. After completing processing
+			the kernel will release the frame back to user-space by
+			setting the status to NL_MMAP_STATUS_UNUSED.
+
+- NL_MMAP_STATUS_COPY:
+	RX ring only:	a message is ready to be processed but could not be
+			stored in the ring, either because it exceeded the
+			frame size or because the originating subsystem does
+			not support memory mapped I/O. Appropriate action is
+			to invoke recvmsg() to receive the message and release
+			the frame back to the kernel by setting the status to
+			NL_MMAP_STATUS_UNUSED.
+
+- NL_MMAP_STATUS_SKIP:
+	RX ring only:	user-space queued the message for later processing, but
+			processed some messages following it in the ring. The
+			kernel should skip this frame when looking for unused
+			frames.
+
+The data area of a frame begins at a offset of NL_MMAP_HDRLEN relative to the
+frame header.
+
+TX limitations
+--------------
+
+Kernel processing usually involves validation of the message received by
+user-space, then processing its contents. The kernel must assure that
+userspace is not able to modify the message contents after they have been
+validated. In order to do so, the message is copied from the ring frame
+to an allocated buffer if either of these conditions is false:
+
+- only a single mapping of the ring exists
+- the file descriptor is not shared between processes
+
+This means that for threaded programs, the kernel will fall back to copying.
+
+Example
+-------
+
+Ring setup:
+
+	unsigned int block_size = 16 * getpagesize();
+	struct nl_mmap_req req = {
+		.nm_block_size		= block_size,
+		.nm_block_nr		= 64,
+		.nm_frame_size		= 16384,
+		.nm_frame_nr		= 64 * block_size / 16384,
+	};
+	unsigned int ring_size;
+	void *rx_ring, *tx_ring;
+
+	/* Configure ring parameters */
+	if (setsockopt(fd, NETLINK_RX_RING, &req, sizeof(req)) < 0)
+		exit(1);
+	if (setsockopt(fd, NETLINK_TX_RING, &req, sizeof(req)) < 0)
+		exit(1)
+
+	/* Calculate size of each invididual ring */
+	ring_size = req.nm_block_nr * req.nm_block_size;
+
+	/* Map RX/TX rings. The TX ring is located after the RX ring */
+	rx_ring = mmap(NULL, 2 * ring_size, PROT_READ | PROT_WRITE,
+		       MAP_SHARED, fd, 0);
+	if ((long)rx_ring == -1L)
+		exit(1);
+	tx_ring = rx_ring + ring_size:
+
+Message reception:
+
+This example assumes some ring parameters of the ring setup are available.
+
+	unsigned int frame_offset = 0;
+	struct nl_mmap_hdr *hdr;
+	struct nlmsghdr *nlh;
+	unsigned char buf[16384];
+	ssize_t len;
+
+	while (1) {
+		struct pollfd pfds[1];
+
+		pfds[0].fd	= fd;
+		pfds[0].events	= POLLIN | POLLERR;
+		pfds[0].revents	= 0;
+
+		if (poll(pfds, 1, -1) < 0 && errno != -EINTR)
+			exit(1);
+
+		/* Check for errors. Error handling omitted */
+		if (pfds[0].revents & POLLERR)
+			<handle error>
+
+		/* If no new messages, poll again */
+		if (!(pfds[0].revents & POLLIN))
+			continue;
+
+		/* Process all frames */
+		while (1) {
+			/* Get next frame header */
+			hdr = rx_ring + frame_offset;
+
+			if (hdr->nm_status == NL_MMAP_STATUS_VALID)
+				/* Regular memory mapped frame */
+				nlh = (void *hdr) + NL_MMAP_HDRLEN;
+				len = hdr->nm_len;
+
+				/* Release empty message immediately. May happen
+				 * on error during message construction.
+				 */
+				if (len == 0)
+					goto release;
+			} else if (hdr->nm_status == NL_MMAP_STATUS_COPY) {
+				/* Frame queued to socket receive queue */
+				len = recv(fd, buf, sizeof(buf), MSG_DONTWAIT);
+				if (len <= 0)
+					break;
+				nlh = buf;
+			} else
+				/* No more messages to process, continue polling */
+				break;
+
+			process_msg(nlh);
+release:
+			/* Release frame back to the kernel */
+			hdr->nm_status = NL_MMAP_STATUS_UNUSED;
+
+			/* Advance frame offset to next frame */
+			frame_offset = (frame_offset + frame_size) % ring_size;
+		}
+	}
+
+Message transmission:
+
+This example assumes some ring parameters of the ring setup are available.
+A single message is constructed and transmitted, to send multiple messages
+at once they would be constructed in consecutive frames before a final call
+to sendto().
+
+	unsigned int frame_offset = 0;
+	struct nl_mmap_hdr *hdr;
+	struct nlmsghdr *nlh;
+	struct sockaddr_nl addr = {
+		.nl_family	= AF_NETLINK,
+	};
+
+	hdr = tx_ring + frame_offset;
+	if (hdr->nm_status != NL_MMAP_STATUS_UNUSED)
+		/* No frame available. Use poll() to avoid. */
+		exit(1);
+
+	nlh = (void *)hdr + NL_MMAP_HDRLEN;
+
+	/* Build message */
+	build_message(nlh);
+
+	/* Fill frame header: length and status need to be set */
+	hdr->nm_len	= nlh->nlmsg_len;
+	hdr->nm_status	= NL_MMAP_STATUS_VALID;
+
+	if (sendto(fd, NULL, 0, 0, &addr, sizeof(addr)) < 0)
+		exit(1);
+
+	/* Advance frame offset to next frame */
+	frame_offset = (frame_offset + frame_size) % ring_size;
-- 
1.7.7.6

Re: [PATCH 02/11] net: add function to allocate skbuff head without data area

[Eric Dumazet]

On Mon, 2012-08-20 at 08:18 +0200, Patrick McHardy wrote:
> Add a function to allocate a skbuff head without any data. This will be
> used by memory mapped netlink to attached data from the mmaped area to
> the skb.
> 
> Additionally change skb_release_all() to check whether the skb has a
> data area to allow the skb destructor to clear the data pointer in
> case only a head has been allocated.

      * ...

>  
> +struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
> +{
> +	struct sk_buff *skb;
> +
> +	/* Get the HEAD */
> +	skb = kmem_cache_alloc_node(skbuff_head_cache,
> +				    gfp_mask & ~__GFP_DMA, node);
> +	if (!skb)
> +		goto out;
> +	prefetchw(skb);
> +

This prefetch is probably not worth it, since you write on skb on the
following memset(). Cpu wont have enough 'time' to take benefit from
this hint.

> +	/*
> +	 * Only clear those fields we need to clear, not those that we will
> +	 * actually initialise below. Hence, don't put any more fields after
> +	 * the tail pointer in struct sk_buff!
> +	 */
> +	memset(skb, 0, offsetof(struct sk_buff, tail));

Re: [PATCH 02/11] net: add function to allocate skbuff head without data area

[Patrick McHardy]

On Mon, 20 Aug 2012, Eric Dumazet wrote:

> On Mon, 2012-08-20 at 08:18 +0200, Patrick McHardy wrote:
>> Add a function to allocate a skbuff head without any data. This will be
>> used by memory mapped netlink to attached data from the mmaped area to
>> the skb.
>>
>> Additionally change skb_release_all() to check whether the skb has a
>> data area to allow the skb destructor to clear the data pointer in
>> case only a head has been allocated.
>
>      * ...
>
>>
>> +struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
>> +{
>> +	struct sk_buff *skb;
>> +
>> +	/* Get the HEAD */
>> +	skb = kmem_cache_alloc_node(skbuff_head_cache,
>> +				    gfp_mask & ~__GFP_DMA, node);
>> +	if (!skb)
>> +		goto out;
>> +	prefetchw(skb);
>> +
>
> This prefetch is probably not worth it, since you write on skb on the
> following memset(). Cpu wont have enough 'time' to take benefit from
> this hint.

Right, I remember we talked about that before and I already removed it.
Must have forgotten to actually commit that change, thanks!

>
>> +	/*
>> +	 * Only clear those fields we need to clear, not those that we will
>> +	 * actually initialise below. Hence, don't put any more fields after
>> +	 * the tail pointer in struct sk_buff!
>> +	 */
>> +	memset(skb, 0, offsetof(struct sk_buff, tail));

Re: [PATCH 11/11] netlink: add documentation for memory mapped I/O

[Jan Engelhardt]

On Monday 2012-08-20 08:18, Patrick McHardy wrote:
>+
>+RX and TX rings
>+----------------
>+
>+Each ring contains a number of continous memory blocks, containing frames of
>+fixed size dependant on the parameters used for ring setup.

dependent

>+The blocks are only visible to the kernel, from the point of view of user-space
>+the ring just contains the frames in a continous memory zone.

...to the kernel; from the user-space point of view, the ring...

continuous (multiple occurrences)

>+Ring frames
>+------------
>+
>+Each frames contain a frame header, consisting of a synchronization word and some

Each frame contains a frame header,

>+The possible values in the status word are:
>+
>+- NL_MMAP_STATUS_UNUSED:
>+	RX ring:	frame belongs to the kernel and contains no message

The frame (mult)

>+			for user-space. Approriate action is to invoke poll()
>+			to wait for new messages.

The appropriate (mult)

>+			if (hdr->nm_status == NL_MMAP_STATUS_VALID)
>+				/* Regular memory mapped frame */
>+				nlh = (void *hdr) + NL_MMAP_HDRLEN;

(void *)hdr + NL_MMAP_HDRLEN

Re: [PATCH 00/11] netlink: memory mapped I/O

[David Miller]

From: Patrick McHardy <kaber@trash.net>
Date: Mon, 20 Aug 2012 08:18:41 +0200

> The following patches contain an implementation of memory mapped I/O for
> netlink, rebased onto the current net-next tree. The implementation is
> modelled after AF_PACKET memory mapped I/O with a few differences:

Thanks for taking this work so far.  Let me know when you have a version
with all the feedback integrated.

Re: [PATCH 11/11] netlink: add documentation for memory mapped I/O

[Patrick McHardy]

On Mon, 20 Aug 2012, Jan Engelhardt wrote:

> On Monday 2012-08-20 08:18, Patrick McHardy wrote:
>> +
>> +RX and TX rings
>> +----------------
>> +
>> +Each ring contains a number of continous memory blocks, containing frames of
>> +fixed size dependant on the parameters used for ring setup.
>
> dependent
> [...]

All fixed, thanks.