Phonet: Pipe End Point for Phonet Pipes protocol
Rémi Denis-Courmont [Sun, 5 Oct 2008 18:15:13 +0000 (11:15 -0700)]
This protocol provides some connection handling and negotiated
congestion control. Nokia cellular modems use it for bulk transfers.
It provides packet boundaries (hence SOCK_SEQPACKET). Congestion
control is per packet rather per byte, so we do not re-use the
generic socket memory accounting.

Signed-off-by: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/phonet.h
include/net/phonet/pep.h
net/phonet/Makefile
net/phonet/af_phonet.c
net/phonet/pep.c [new file with mode: 0644]

index 3a027f5..f921852 100644 (file)
@@ -27,7 +27,9 @@
 #define PN_PROTO_TRANSPORT     0
 /* Phonet datagram socket */
 #define PN_PROTO_PHONET                1
-#define PHONET_NPROTO          2
+/* Phonet pipe */
+#define PN_PROTO_PIPE          2
+#define PHONET_NPROTO          3
 
 #define PNADDR_ANY             0
 #define PNPORT_RESOURCE_ROUTING        0
index b2f8c54..fb024e1 100644 (file)
 struct pep_sock {
        struct pn_sock          pn_sk;
 
+       /* XXX: union-ify listening vs connected stuff ? */
        /* Listening socket stuff: */
        struct hlist_head       ackq;
+       struct hlist_head       hlist;
 
        /* Connected socket stuff: */
+       struct sock             *listener;
+       u16                     peer_type;      /* peer type/subtype */
+       u8                      pipe_handle;
+
+       u8                      rx_credits;
        u8                      tx_credits;
+       u8                      rx_fc;  /* RX flow control */
+       u8                      tx_fc;  /* TX flow control */
+       u8                      init_enable;    /* auto-enable at creation */
 };
 
 static inline struct pep_sock *pep_sk(struct sock *sk)
@@ -40,4 +50,108 @@ static inline struct pep_sock *pep_sk(struct sock *sk)
 
 extern const struct proto_ops phonet_stream_ops;
 
+/* Pipe protocol definitions */
+struct pnpipehdr {
+       u8                      utid; /* transaction ID */
+       u8                      message_id;
+       u8                      pipe_handle;
+       union {
+               u8              state_after_connect;    /* connect request */
+               u8              state_after_reset;      /* reset request */
+               u8              error_code;             /* any response */
+               u8              pep_type;               /* status indication */
+               u8              data[1];
+       };
+};
+#define other_pep_type         data[1]
+
+static inline struct pnpipehdr *pnp_hdr(struct sk_buff *skb)
+{
+       return (struct pnpipehdr *)skb_transport_header(skb);
+}
+
+#define MAX_PNPIPE_HEADER (MAX_PHONET_HEADER + 4)
+
+enum {
+       PNS_PIPE_DATA = 0x20,
+
+       PNS_PEP_CONNECT_REQ = 0x40,
+       PNS_PEP_CONNECT_RESP,
+       PNS_PEP_DISCONNECT_REQ,
+       PNS_PEP_DISCONNECT_RESP,
+       PNS_PEP_RESET_REQ,
+       PNS_PEP_RESET_RESP,
+       PNS_PEP_ENABLE_REQ,
+       PNS_PEP_ENABLE_RESP,
+       PNS_PEP_CTRL_REQ,
+       PNS_PEP_CTRL_RESP,
+       PNS_PEP_DISABLE_REQ = 0x4C,
+       PNS_PEP_DISABLE_RESP,
+
+       PNS_PEP_STATUS_IND = 0x60,
+       PNS_PIPE_CREATED_IND,
+       PNS_PIPE_RESET_IND = 0x63,
+       PNS_PIPE_ENABLED_IND,
+       PNS_PIPE_REDIRECTED_IND,
+       PNS_PIPE_DISABLED_IND = 0x66,
+};
+
+#define PN_PIPE_INVALID_HANDLE 0xff
+#define PN_PEP_TYPE_COMMON     0x00
+
+/* Phonet pipe status indication */
+enum {
+       PN_PEP_IND_FLOW_CONTROL,
+       PN_PEP_IND_ID_MCFC_GRANT_CREDITS,
+};
+
+/* Phonet pipe error codes */
+enum {
+       PN_PIPE_NO_ERROR,
+       PN_PIPE_ERR_INVALID_PARAM,
+       PN_PIPE_ERR_INVALID_HANDLE,
+       PN_PIPE_ERR_INVALID_CTRL_ID,
+       PN_PIPE_ERR_NOT_ALLOWED,
+       PN_PIPE_ERR_PEP_IN_USE,
+       PN_PIPE_ERR_OVERLOAD,
+       PN_PIPE_ERR_DEV_DISCONNECTED,
+       PN_PIPE_ERR_TIMEOUT,
+       PN_PIPE_ERR_ALL_PIPES_IN_USE,
+       PN_PIPE_ERR_GENERAL,
+       PN_PIPE_ERR_NOT_SUPPORTED,
+};
+
+/* Phonet pipe states */
+enum {
+       PN_PIPE_DISABLE,
+       PN_PIPE_ENABLE,
+};
+
+/* Phonet pipe sub-block types */
+enum {
+       PN_PIPE_SB_CREATE_REQ_PEP_SUB_TYPE,
+       PN_PIPE_SB_CONNECT_REQ_PEP_SUB_TYPE,
+       PN_PIPE_SB_REDIRECT_REQ_PEP_SUB_TYPE,
+       PN_PIPE_SB_NEGOTIATED_FC,
+       PN_PIPE_SB_REQUIRED_FC_TX,
+       PN_PIPE_SB_PREFERRED_FC_RX,
+};
+
+/* Phonet pipe flow control models */
+enum {
+       PN_NO_FLOW_CONTROL,
+       PN_LEGACY_FLOW_CONTROL,
+       PN_ONE_CREDIT_FLOW_CONTROL,
+       PN_MULTI_CREDIT_FLOW_CONTROL,
+};
+
+#define pn_flow_safe(fc) ((fc) >> 1)
+
+/* Phonet pipe flow control states */
+enum {
+       PEP_IND_EMPTY,
+       PEP_IND_BUSY,
+       PEP_IND_READY,
+};
+
 #endif
index ae9c3ed..505df2a 100644 (file)
@@ -1,4 +1,4 @@
-obj-$(CONFIG_PHONET) += phonet.o
+obj-$(CONFIG_PHONET) += phonet.o pn_pep.o
 
 phonet-objs := \
        pn_dev.o \
@@ -7,3 +7,5 @@ phonet-objs := \
        datagram.o \
        sysctl.o \
        af_phonet.o
+
+pn_pep-objs := pep.o
index 0a74aea..9e9c6fc 100644 (file)
@@ -58,6 +58,9 @@ static int pn_socket_create(struct net *net, struct socket *sock, int protocol)
                case SOCK_DGRAM:
                        protocol = PN_PROTO_PHONET;
                        break;
+               case SOCK_SEQPACKET:
+                       protocol = PN_PROTO_PIPE;
+                       break;
                default:
                        return -EPROTONOSUPPORT;
                }
diff --git a/net/phonet/pep.c b/net/phonet/pep.c
new file mode 100644 (file)
index 0000000..c5dfecb
--- /dev/null
@@ -0,0 +1,908 @@
+/*
+ * File: pep.c
+ *
+ * Phonet pipe protocol end point socket
+ *
+ * Copyright (C) 2008 Nokia Corporation.
+ *
+ * Author: Rémi Denis-Courmont <remi.denis-courmont@nokia.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA
+ * 02110-1301 USA
+ */
+
+#include <linux/kernel.h>
+#include <linux/socket.h>
+#include <net/sock.h>
+#include <net/tcp_states.h>
+#include <asm/ioctls.h>
+
+#include <linux/phonet.h>
+#include <net/phonet/phonet.h>
+#include <net/phonet/pep.h>
+
+/* sk_state values:
+ * TCP_CLOSE           sock not in use yet
+ * TCP_CLOSE_WAIT      disconnected pipe
+ * TCP_LISTEN          listening pipe endpoint
+ * TCP_SYN_RECV                connected pipe in disabled state
+ * TCP_ESTABLISHED     connected pipe in enabled state
+ *
+ * pep_sock locking:
+ *  - sk_state, ackq, hlist: sock lock needed
+ *  - listener: read only
+ *  - pipe_handle: read only
+ */
+
+#define CREDITS_MAX    10
+#define CREDITS_THR    7
+
+static const struct sockaddr_pn pipe_srv = {
+       .spn_family = AF_PHONET,
+       .spn_resource = 0xD9, /* pipe service */
+};
+
+#define pep_sb_size(s) (((s) + 5) & ~3) /* 2-bytes head, 32-bits aligned */
+
+/* Get the next TLV sub-block. */
+static unsigned char *pep_get_sb(struct sk_buff *skb, u8 *ptype, u8 *plen,
+                                       void *buf)
+{
+       void *data = NULL;
+       struct {
+               u8 sb_type;
+               u8 sb_len;
+       } *ph, h;
+       int buflen = *plen;
+
+       ph = skb_header_pointer(skb, 0, 2, &h);
+       if (ph == NULL || ph->sb_len < 2 || !pskb_may_pull(skb, ph->sb_len))
+               return NULL;
+       ph->sb_len -= 2;
+       *ptype = ph->sb_type;
+       *plen = ph->sb_len;
+
+       if (buflen > ph->sb_len)
+               buflen = ph->sb_len;
+       data = skb_header_pointer(skb, 2, buflen, buf);
+       __skb_pull(skb, 2 + ph->sb_len);
+       return data;
+}
+
+static int pep_reply(struct sock *sk, struct sk_buff *oskb,
+                       u8 code, const void *data, int len, gfp_t priority)
+{
+       const struct pnpipehdr *oph = pnp_hdr(oskb);
+       struct pnpipehdr *ph;
+       struct sk_buff *skb;
+
+       skb = alloc_skb(MAX_PNPIPE_HEADER + len, priority);
+       if (!skb)
+               return -ENOMEM;
+       skb_set_owner_w(skb, sk);
+
+       skb_reserve(skb, MAX_PNPIPE_HEADER);
+       __skb_put(skb, len);
+       skb_copy_to_linear_data(skb, data, len);
+       __skb_push(skb, sizeof(*ph));
+       skb_reset_transport_header(skb);
+       ph = pnp_hdr(skb);
+       ph->utid = oph->utid;
+       ph->message_id = oph->message_id + 1; /* REQ -> RESP */
+       ph->pipe_handle = oph->pipe_handle;
+       ph->error_code = code;
+
+       return pn_skb_send(sk, skb, &pipe_srv);
+}
+
+#define PAD 0x00
+static int pep_accept_conn(struct sock *sk, struct sk_buff *skb)
+{
+       static const u8 data[20] = {
+               PAD, PAD, PAD, 2 /* sub-blocks */,
+               PN_PIPE_SB_REQUIRED_FC_TX, pep_sb_size(5), 3, PAD,
+                       PN_MULTI_CREDIT_FLOW_CONTROL,
+                       PN_ONE_CREDIT_FLOW_CONTROL,
+                       PN_LEGACY_FLOW_CONTROL,
+                       PAD,
+               PN_PIPE_SB_PREFERRED_FC_RX, pep_sb_size(5), 3, PAD,
+                       PN_MULTI_CREDIT_FLOW_CONTROL,
+                       PN_ONE_CREDIT_FLOW_CONTROL,
+                       PN_LEGACY_FLOW_CONTROL,
+                       PAD,
+       };
+
+       might_sleep();
+       return pep_reply(sk, skb, PN_PIPE_NO_ERROR, data, sizeof(data),
+                               GFP_KERNEL);
+}
+
+static int pep_reject_conn(struct sock *sk, struct sk_buff *skb, u8 code)
+{
+       static const u8 data[4] = { PAD, PAD, PAD, 0 /* sub-blocks */ };
+       WARN_ON(code == PN_PIPE_NO_ERROR);
+       return pep_reply(sk, skb, code, data, sizeof(data), GFP_ATOMIC);
+}
+
+/* Control requests are not sent by the pipe service and have a specific
+ * message format. */
+static int pep_ctrlreq_error(struct sock *sk, struct sk_buff *oskb, u8 code)
+{
+       const struct pnpipehdr *oph = pnp_hdr(oskb);
+       struct sk_buff *skb;
+       struct pnpipehdr *ph;
+       struct sockaddr_pn dst;
+
+       skb = alloc_skb(MAX_PNPIPE_HEADER + 4, GFP_ATOMIC);
+       if (!skb)
+               return -ENOMEM;
+       skb_set_owner_w(skb, sk);
+
+       skb_reserve(skb, MAX_PHONET_HEADER);
+       ph = (struct pnpipehdr *)skb_put(skb, sizeof(*ph) + 4);
+
+       ph->utid = oph->utid;
+       ph->message_id = PNS_PEP_CTRL_RESP;
+       ph->pipe_handle = oph->pipe_handle;
+       ph->data[0] = oph->data[1]; /* CTRL id */
+       ph->data[1] = oph->data[0]; /* PEP type */
+       ph->data[2] = code; /* error code, at an usual offset */
+       ph->data[3] = PAD;
+       ph->data[4] = PAD;
+
+       pn_skb_get_src_sockaddr(oskb, &dst);
+       return pn_skb_send(sk, skb, &dst);
+}
+
+static int pipe_snd_status(struct sock *sk, u8 type, u8 status, gfp_t priority)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct pnpipehdr *ph;
+       struct sk_buff *skb;
+
+       skb = alloc_skb(MAX_PNPIPE_HEADER + 4, priority);
+       if (!skb)
+               return -ENOMEM;
+       skb_set_owner_w(skb, sk);
+
+       skb_reserve(skb, MAX_PNPIPE_HEADER + 4);
+       __skb_push(skb, sizeof(*ph) + 4);
+       skb_reset_transport_header(skb);
+       ph = pnp_hdr(skb);
+       ph->utid = 0;
+       ph->message_id = PNS_PEP_STATUS_IND;
+       ph->pipe_handle = pn->pipe_handle;
+       ph->pep_type = PN_PEP_TYPE_COMMON;
+       ph->data[1] = type;
+       ph->data[2] = PAD;
+       ph->data[3] = PAD;
+       ph->data[4] = status;
+
+       return pn_skb_send(sk, skb, &pipe_srv);
+}
+
+/* Send our RX flow control information to the sender.
+ * Socket must be locked. */
+static void pipe_grant_credits(struct sock *sk)
+{
+       struct pep_sock *pn = pep_sk(sk);
+
+       BUG_ON(sk->sk_state != TCP_ESTABLISHED);
+
+       switch (pn->rx_fc) {
+       case PN_LEGACY_FLOW_CONTROL: /* TODO */
+               break;
+       case PN_ONE_CREDIT_FLOW_CONTROL:
+               pipe_snd_status(sk, PN_PEP_IND_FLOW_CONTROL,
+                               PEP_IND_READY, GFP_ATOMIC);
+               pn->rx_credits = 1;
+               break;
+       case PN_MULTI_CREDIT_FLOW_CONTROL:
+               if ((pn->rx_credits + CREDITS_THR) > CREDITS_MAX)
+                       break;
+               if (pipe_snd_status(sk, PN_PEP_IND_ID_MCFC_GRANT_CREDITS,
+                                       CREDITS_MAX - pn->rx_credits,
+                                       GFP_ATOMIC) == 0)
+                       pn->rx_credits = CREDITS_MAX;
+               break;
+       }
+}
+
+static int pipe_rcv_status(struct sock *sk, struct sk_buff *skb)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct pnpipehdr *hdr = pnp_hdr(skb);
+
+       if (!pskb_may_pull(skb, sizeof(*hdr) + 4))
+               return -EINVAL;
+
+       if (hdr->data[0] != PN_PEP_TYPE_COMMON) {
+               LIMIT_NETDEBUG(KERN_DEBUG"Phonet unknown PEP type: %u\n",
+                               (unsigned)hdr->data[0]);
+               return -EOPNOTSUPP;
+       }
+
+       switch (hdr->data[1]) {
+       case PN_PEP_IND_FLOW_CONTROL:
+               switch (pn->tx_fc) {
+               case PN_LEGACY_FLOW_CONTROL:
+                       switch (hdr->data[4]) {
+                       case PEP_IND_BUSY:
+                               pn->tx_credits = 0;
+                               break;
+                       case PEP_IND_READY:
+                               pn->tx_credits = 1;
+                               break;
+                       }
+                       break;
+               case PN_ONE_CREDIT_FLOW_CONTROL:
+                       if (hdr->data[4] == PEP_IND_READY)
+                               pn->tx_credits = 1;
+                       break;
+               }
+               break;
+
+       case PN_PEP_IND_ID_MCFC_GRANT_CREDITS:
+               if (pn->tx_fc != PN_MULTI_CREDIT_FLOW_CONTROL)
+                       break;
+               if (pn->tx_credits + hdr->data[4] > 0xff)
+                       pn->tx_credits = 0xff;
+               else
+                       pn->tx_credits += hdr->data[4];
+               break;
+
+       default:
+               LIMIT_NETDEBUG(KERN_DEBUG"Phonet unknown PEP indication: %u\n",
+                               (unsigned)hdr->data[1]);
+               return -EOPNOTSUPP;
+       }
+       if (pn->tx_credits)
+               sk->sk_write_space(sk);
+       return 0;
+}
+
+static int pipe_rcv_created(struct sock *sk, struct sk_buff *skb)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct pnpipehdr *hdr = pnp_hdr(skb);
+       u8 n_sb = hdr->data[0];
+
+       pn->rx_fc = pn->tx_fc = PN_LEGACY_FLOW_CONTROL;
+       __skb_pull(skb, sizeof(*hdr));
+       while (n_sb > 0) {
+               u8 type, buf[2], len = sizeof(buf);
+               u8 *data = pep_get_sb(skb, &type, &len, buf);
+
+               if (data == NULL)
+                       return -EINVAL;
+               switch (type) {
+               case PN_PIPE_SB_NEGOTIATED_FC:
+                       if (len < 2 || (data[0] | data[1]) > 3)
+                               break;
+                       pn->tx_fc = data[0] & 3;
+                       pn->rx_fc = data[1] & 3;
+                       break;
+               }
+               n_sb--;
+       }
+       return 0;
+}
+
+/* Queue an skb to a connected sock.
+ * Socket lock must be held. */
+static int pipe_do_rcv(struct sock *sk, struct sk_buff *skb)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct pnpipehdr *hdr = pnp_hdr(skb);
+       int err = 0;
+
+       BUG_ON(sk->sk_state == TCP_CLOSE_WAIT);
+
+       switch (hdr->message_id) {
+       case PNS_PEP_CONNECT_REQ:
+               pep_reject_conn(sk, skb, PN_PIPE_ERR_PEP_IN_USE);
+               break;
+
+       case PNS_PEP_DISCONNECT_REQ:
+               pep_reply(sk, skb, PN_PIPE_NO_ERROR, NULL, 0, GFP_ATOMIC);
+               sk->sk_state = TCP_CLOSE_WAIT;
+               if (!sock_flag(sk, SOCK_DEAD))
+                       sk->sk_state_change(sk);
+               break;
+
+       case PNS_PEP_ENABLE_REQ:
+               /* Wait for PNS_PIPE_(ENABLED|REDIRECTED)_IND */
+               pep_reply(sk, skb, PN_PIPE_NO_ERROR, NULL, 0, GFP_ATOMIC);
+               break;
+
+       case PNS_PEP_RESET_REQ:
+               switch (hdr->state_after_reset) {
+               case PN_PIPE_DISABLE:
+                       pn->init_enable = 0;
+                       break;
+               case PN_PIPE_ENABLE:
+                       pn->init_enable = 1;
+                       break;
+               default: /* not allowed to send an error here!? */
+                       err = -EINVAL;
+                       goto out;
+               }
+               /* fall through */
+       case PNS_PEP_DISABLE_REQ:
+               pn->tx_credits = 0;
+               pep_reply(sk, skb, PN_PIPE_NO_ERROR, NULL, 0, GFP_ATOMIC);
+               break;
+
+       case PNS_PEP_CTRL_REQ:
+               /* TODO */
+               pep_ctrlreq_error(sk, skb, PN_PIPE_NO_ERROR);
+               break;
+
+       case PNS_PIPE_DATA:
+               __skb_pull(skb, 3); /* Pipe data header */
+               if (!pn_flow_safe(pn->rx_fc)) {
+                       err = sock_queue_rcv_skb(sk, skb);
+                       if (!err)
+                               return 0;
+                       break;
+               }
+
+               if (pn->rx_credits == 0) {
+                       err = -ENOBUFS;
+                       break;
+               }
+               pn->rx_credits--;
+               skb->dev = NULL;
+               skb_set_owner_r(skb, sk);
+               err = skb->len;
+               skb_queue_tail(&sk->sk_receive_queue, skb);
+               if (!sock_flag(sk, SOCK_DEAD))
+                       sk->sk_data_ready(sk, err);
+               return 0;
+
+       case PNS_PEP_STATUS_IND:
+               pipe_rcv_status(sk, skb);
+               break;
+
+       case PNS_PIPE_REDIRECTED_IND:
+               err = pipe_rcv_created(sk, skb);
+               break;
+
+       case PNS_PIPE_CREATED_IND:
+               err = pipe_rcv_created(sk, skb);
+               if (err)
+                       break;
+               /* fall through */
+       case PNS_PIPE_RESET_IND:
+               if (!pn->init_enable)
+                       break;
+               /* fall through */
+       case PNS_PIPE_ENABLED_IND:
+               if (!pn_flow_safe(pn->tx_fc)) {
+                       pn->tx_credits = 1;
+                       sk->sk_write_space(sk);
+               }
+               if (sk->sk_state == TCP_ESTABLISHED)
+                       break; /* Nothing to do */
+               sk->sk_state = TCP_ESTABLISHED;
+               pipe_grant_credits(sk);
+               break;
+
+       case PNS_PIPE_DISABLED_IND:
+               sk->sk_state = TCP_SYN_RECV;
+               pn->rx_credits = 0;
+               break;
+
+       default:
+               LIMIT_NETDEBUG(KERN_DEBUG"Phonet unknown PEP message: %u\n",
+                               hdr->message_id);
+               err = -EINVAL;
+       }
+out:
+       kfree_skb(skb);
+       return err;
+}
+
+/* Destroy connected sock. */
+static void pipe_destruct(struct sock *sk)
+{
+       skb_queue_purge(&sk->sk_receive_queue);
+}
+
+static int pep_connreq_rcv(struct sock *sk, struct sk_buff *skb)
+{
+       struct sock *newsk;
+       struct pep_sock *newpn, *pn = pep_sk(sk);
+       struct pnpipehdr *hdr;
+       struct sockaddr_pn dst;
+       u16 peer_type;
+       u8 pipe_handle, enabled, n_sb;
+
+       if (!pskb_pull(skb, sizeof(*hdr) + 4))
+               return -EINVAL;
+
+       hdr = pnp_hdr(skb);
+       pipe_handle = hdr->pipe_handle;
+       switch (hdr->state_after_connect) {
+       case PN_PIPE_DISABLE:
+               enabled = 0;
+               break;
+       case PN_PIPE_ENABLE:
+               enabled = 1;
+               break;
+       default:
+               pep_reject_conn(sk, skb, PN_PIPE_ERR_INVALID_PARAM);
+               return -EINVAL;
+       }
+       peer_type = hdr->other_pep_type << 8;
+
+       if (unlikely(sk->sk_state != TCP_LISTEN) || sk_acceptq_is_full(sk)) {
+               pep_reject_conn(sk, skb, PN_PIPE_ERR_PEP_IN_USE);
+               return -ENOBUFS;
+       }
+
+       /* Parse sub-blocks (options) */
+       n_sb = hdr->data[4];
+       while (n_sb > 0) {
+               u8 type, buf[1], len = sizeof(buf);
+               const u8 *data = pep_get_sb(skb, &type, &len, buf);
+
+               if (data == NULL)
+                       return -EINVAL;
+               switch (type) {
+               case PN_PIPE_SB_CONNECT_REQ_PEP_SUB_TYPE:
+                       if (len < 1)
+                               return -EINVAL;
+                       peer_type = (peer_type & 0xff00) | data[0];
+                       break;
+               }
+               n_sb--;
+       }
+
+       skb = skb_clone(skb, GFP_ATOMIC);
+       if (!skb)
+               return -ENOMEM;
+
+       /* Create a new to-be-accepted sock */
+       newsk = sk_alloc(sock_net(sk), PF_PHONET, GFP_ATOMIC, sk->sk_prot);
+       if (!newsk) {
+               kfree_skb(skb);
+               return -ENOMEM;
+       }
+       sock_init_data(NULL, newsk);
+       newsk->sk_state = TCP_SYN_RECV;
+       newsk->sk_backlog_rcv = pipe_do_rcv;
+       newsk->sk_protocol = sk->sk_protocol;
+       newsk->sk_destruct = pipe_destruct;
+
+       newpn = pep_sk(newsk);
+       pn_skb_get_dst_sockaddr(skb, &dst);
+       newpn->pn_sk.sobject = pn_sockaddr_get_object(&dst);
+       newpn->pn_sk.resource = pn->pn_sk.resource;
+       newpn->pipe_handle = pipe_handle;
+       newpn->peer_type = peer_type;
+       newpn->rx_credits = newpn->tx_credits = 0;
+       newpn->rx_fc = newpn->tx_fc = PN_LEGACY_FLOW_CONTROL;
+       newpn->init_enable = enabled;
+
+       BUG_ON(!skb_queue_empty(&newsk->sk_receive_queue));
+       skb_queue_head(&newsk->sk_receive_queue, skb);
+       if (!sock_flag(sk, SOCK_DEAD))
+               sk->sk_data_ready(sk, 0);
+
+       sk_acceptq_added(sk);
+       sk_add_node(newsk, &pn->ackq);
+       return 0;
+}
+
+/* Listening sock must be locked */
+static struct sock *pep_find_pipe(const struct hlist_head *hlist,
+                                       const struct sockaddr_pn *dst,
+                                       u8 pipe_handle)
+{
+       struct hlist_node *node;
+       struct sock *sknode;
+       u16 dobj = pn_sockaddr_get_object(dst);
+
+       sk_for_each(sknode, node, hlist) {
+               struct pep_sock *pnnode = pep_sk(sknode);
+
+               /* Ports match, but addresses might not: */
+               if (pnnode->pn_sk.sobject != dobj)
+                       continue;
+               if (pnnode->pipe_handle != pipe_handle)
+                       continue;
+               if (sknode->sk_state == TCP_CLOSE_WAIT)
+                       continue;
+
+               sock_hold(sknode);
+               return sknode;
+       }
+       return NULL;
+}
+
+/*
+ * Deliver an skb to a listening sock.
+ * Socket lock must be held.
+ * We then queue the skb to the right connected sock (if any).
+ */
+static int pep_do_rcv(struct sock *sk, struct sk_buff *skb)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct sock *sknode;
+       struct pnpipehdr *hdr = pnp_hdr(skb);
+       struct sockaddr_pn dst;
+       int err = NET_RX_SUCCESS;
+       u8 pipe_handle;
+
+       if (!pskb_may_pull(skb, sizeof(*hdr)))
+               goto drop;
+
+       hdr = pnp_hdr(skb);
+       pipe_handle = hdr->pipe_handle;
+       if (pipe_handle == PN_PIPE_INVALID_HANDLE)
+               goto drop;
+
+       pn_skb_get_dst_sockaddr(skb, &dst);
+
+       /* Look for an existing pipe handle */
+       sknode = pep_find_pipe(&pn->hlist, &dst, pipe_handle);
+       if (sknode)
+               return sk_receive_skb(sknode, skb, 1);
+
+       /* Look for a pipe handle pending accept */
+       sknode = pep_find_pipe(&pn->ackq, &dst, pipe_handle);
+       if (sknode) {
+               sock_put(sknode);
+               if (net_ratelimit())
+                       printk(KERN_WARNING"Phonet unconnected PEP ignored");
+               err = NET_RX_DROP;
+               goto drop;
+       }
+
+       switch (hdr->message_id) {
+       case PNS_PEP_CONNECT_REQ:
+               err = pep_connreq_rcv(sk, skb);
+               break;
+
+       case PNS_PEP_DISCONNECT_REQ:
+               pep_reply(sk, skb, PN_PIPE_NO_ERROR, NULL, 0, GFP_ATOMIC);
+               break;
+
+       case PNS_PEP_CTRL_REQ:
+               pep_ctrlreq_error(sk, skb, PN_PIPE_INVALID_HANDLE);
+               break;
+
+       case PNS_PEP_RESET_REQ:
+       case PNS_PEP_ENABLE_REQ:
+       case PNS_PEP_DISABLE_REQ:
+               /* invalid handle is not even allowed here! */
+       default:
+               err = NET_RX_DROP;
+       }
+drop:
+       kfree_skb(skb);
+       return err;
+}
+
+/* associated socket ceases to exist */
+static void pep_sock_close(struct sock *sk, long timeout)
+{
+       struct pep_sock *pn = pep_sk(sk);
+
+       sk_common_release(sk);
+
+       lock_sock(sk);
+       if (sk->sk_state == TCP_LISTEN) {
+               /* Destroy the listen queue */
+               struct sock *sknode;
+               struct hlist_node *p, *n;
+
+               sk_for_each_safe(sknode, p, n, &pn->ackq)
+                       sk_del_node_init(sknode);
+               sk->sk_state = TCP_CLOSE;
+       }
+       release_sock(sk);
+}
+
+static int pep_wait_connreq(struct sock *sk, int noblock)
+{
+       struct task_struct *tsk = current;
+       struct pep_sock *pn = pep_sk(sk);
+       long timeo = sock_rcvtimeo(sk, noblock);
+
+       for (;;) {
+               DEFINE_WAIT(wait);
+
+               if (sk->sk_state != TCP_LISTEN)
+                       return -EINVAL;
+               if (!hlist_empty(&pn->ackq))
+                       break;
+               if (!timeo)
+                       return -EWOULDBLOCK;
+               if (signal_pending(tsk))
+                       return sock_intr_errno(timeo);
+
+               prepare_to_wait_exclusive(&sk->sk_socket->wait, &wait,
+                                               TASK_INTERRUPTIBLE);
+               release_sock(sk);
+               timeo = schedule_timeout(timeo);
+               lock_sock(sk);
+               finish_wait(&sk->sk_socket->wait, &wait);
+       }
+
+       return 0;
+}
+
+static struct sock *pep_sock_accept(struct sock *sk, int flags, int *errp)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct sock *newsk = NULL;
+       struct sk_buff *oskb;
+       int err;
+
+       lock_sock(sk);
+       err = pep_wait_connreq(sk, flags & O_NONBLOCK);
+       if (err)
+               goto out;
+
+       newsk = __sk_head(&pn->ackq);
+
+       oskb = skb_dequeue(&newsk->sk_receive_queue);
+       err = pep_accept_conn(newsk, oskb);
+       if (err) {
+               skb_queue_head(&newsk->sk_receive_queue, oskb);
+               newsk = NULL;
+               goto out;
+       }
+
+       sock_hold(sk);
+       pep_sk(newsk)->listener = sk;
+
+       sock_hold(newsk);
+       sk_del_node_init(newsk);
+       sk_acceptq_removed(sk);
+       sk_add_node(newsk, &pn->hlist);
+       __sock_put(newsk);
+
+out:
+       release_sock(sk);
+       *errp = err;
+       return newsk;
+}
+
+static int pep_ioctl(struct sock *sk, int cmd, unsigned long arg)
+{
+       int answ;
+
+       switch (cmd) {
+       case SIOCINQ:
+               if (sk->sk_state == TCP_LISTEN)
+                       return -EINVAL;
+
+               lock_sock(sk);
+               if (!skb_queue_empty(&sk->sk_receive_queue))
+                       answ = skb_peek(&sk->sk_receive_queue)->len;
+               else
+                       answ = 0;
+               release_sock(sk);
+               return put_user(answ, (int __user *)arg);
+       }
+
+       return -ENOIOCTLCMD;
+}
+
+static int pep_init(struct sock *sk)
+{
+       struct pep_sock *pn = pep_sk(sk);
+
+       INIT_HLIST_HEAD(&pn->ackq);
+       INIT_HLIST_HEAD(&pn->hlist);
+       pn->pipe_handle = PN_PIPE_INVALID_HANDLE;
+       return 0;
+}
+
+static int pep_sendmsg(struct kiocb *iocb, struct sock *sk,
+                       struct msghdr *msg, size_t len)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct sk_buff *skb = NULL;
+       struct pnpipehdr *ph;
+       long timeo;
+       int flags = msg->msg_flags;
+       int err, done;
+
+       if (msg->msg_flags & MSG_OOB || !(msg->msg_flags & MSG_EOR))
+               return -EOPNOTSUPP;
+
+       lock_sock(sk);
+       timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
+       if ((1 << sk->sk_state) & (TCPF_LISTEN|TCPF_CLOSE)) {
+               err = -ENOTCONN;
+               goto out;
+       }
+       if (sk->sk_state != TCP_ESTABLISHED) {
+               /* Wait until the pipe gets to enabled state */
+disabled:
+               err = sk_stream_wait_connect(sk, &timeo);
+               if (err)
+                       goto out;
+
+               if (sk->sk_state == TCP_CLOSE_WAIT) {
+                       err = -ECONNRESET;
+                       goto out;
+               }
+       }
+       BUG_ON(sk->sk_state != TCP_ESTABLISHED);
+
+       /* Wait until flow control allows TX */
+       done = pn->tx_credits > 0;
+       while (!done) {
+               DEFINE_WAIT(wait);
+
+               if (!timeo) {
+                       err = -EAGAIN;
+                       goto out;
+               }
+               if (signal_pending(current)) {
+                       err = sock_intr_errno(timeo);
+                       goto out;
+               }
+
+               prepare_to_wait(&sk->sk_socket->wait, &wait,
+                               TASK_INTERRUPTIBLE);
+               done = sk_wait_event(sk, &timeo, pn->tx_credits > 0);
+               finish_wait(&sk->sk_socket->wait, &wait);
+
+               if (sk->sk_state != TCP_ESTABLISHED)
+                       goto disabled;
+       }
+
+       if (!skb) {
+               skb = sock_alloc_send_skb(sk, MAX_PNPIPE_HEADER + len,
+                                               flags & MSG_DONTWAIT, &err);
+               if (skb == NULL)
+                       goto out;
+               skb_reserve(skb, MAX_PHONET_HEADER + 3);
+
+               if (sk->sk_state != TCP_ESTABLISHED || !pn->tx_credits)
+                       goto disabled; /* sock_alloc_send_skb might sleep */
+       }
+
+       err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
+       if (err < 0)
+               goto out;
+
+       __skb_push(skb, 3);
+       skb_reset_transport_header(skb);
+       ph = pnp_hdr(skb);
+       ph->utid = 0;
+       ph->message_id = PNS_PIPE_DATA;
+       ph->pipe_handle = pn->pipe_handle;
+       if (pn_flow_safe(pn->tx_fc)) /* credit-based flow control */
+               pn->tx_credits--;
+
+       err = pn_skb_send(sk, skb, &pipe_srv);
+       if (err >= 0)
+               err = len; /* success! */
+       skb = NULL;
+out:
+       release_sock(sk);
+       kfree_skb(skb);
+       return err;
+}
+
+static int pep_recvmsg(struct kiocb *iocb, struct sock *sk,
+                       struct msghdr *msg, size_t len, int noblock,
+                       int flags, int *addr_len)
+{
+       struct sk_buff *skb;
+       int err;
+
+       if (unlikely(flags & MSG_OOB))
+               return -EOPNOTSUPP;
+       if (unlikely(1 << sk->sk_state & (TCPF_LISTEN | TCPF_CLOSE)))
+               return -ENOTCONN;
+
+       skb = skb_recv_datagram(sk, flags, noblock, &err);
+       lock_sock(sk);
+       if (skb == NULL) {
+               if (err == -ENOTCONN && sk->sk_state == TCP_CLOSE_WAIT)
+                       err = -ECONNRESET;
+               release_sock(sk);
+               return err;
+       }
+
+       if (sk->sk_state == TCP_ESTABLISHED)
+               pipe_grant_credits(sk);
+       release_sock(sk);
+
+       msg->msg_flags |= MSG_EOR;
+
+       if (skb->len > len)
+               msg->msg_flags |= MSG_TRUNC;
+       else
+               len = skb->len;
+
+       err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, len);
+       if (!err)
+               err = (flags & MSG_TRUNC) ? skb->len : len;
+
+       skb_free_datagram(sk, skb);
+       return err;
+}
+
+static void pep_sock_unhash(struct sock *sk)
+{
+       struct pep_sock *pn = pep_sk(sk);
+       struct sock *skparent = NULL;
+
+       lock_sock(sk);
+       if ((1 << sk->sk_state) & ~(TCPF_CLOSE|TCPF_LISTEN)) {
+               skparent = pn->listener;
+               sk_del_node_init(sk);
+               release_sock(sk);
+
+               sk = skparent;
+               pn = pep_sk(skparent);
+               lock_sock(sk);
+       }
+       /* Unhash a listening sock only when it is closed
+        * and all of its active connected pipes are closed. */
+       if (hlist_empty(&pn->hlist))
+               pn_sock_unhash(&pn->pn_sk.sk);
+       release_sock(sk);
+
+       if (skparent)
+               sock_put(skparent);
+}
+
+static struct proto pep_proto = {
+       .close          = pep_sock_close,
+       .accept         = pep_sock_accept,
+       .ioctl          = pep_ioctl,
+       .init           = pep_init,
+       .sendmsg        = pep_sendmsg,
+       .recvmsg        = pep_recvmsg,
+       .backlog_rcv    = pep_do_rcv,
+       .hash           = pn_sock_hash,
+       .unhash         = pep_sock_unhash,
+       .get_port       = pn_sock_get_port,
+       .obj_size       = sizeof(struct pep_sock),
+       .owner          = THIS_MODULE,
+       .name           = "PNPIPE",
+};
+
+static struct phonet_protocol pep_pn_proto = {
+       .ops            = &phonet_stream_ops,
+       .prot           = &pep_proto,
+       .sock_type      = SOCK_SEQPACKET,
+};
+
+static int __init pep_register(void)
+{
+       return phonet_proto_register(PN_PROTO_PIPE, &pep_pn_proto);
+}
+
+static void __exit pep_unregister(void)
+{
+       phonet_proto_unregister(PN_PROTO_PIPE, &pep_pn_proto);
+}
+
+module_init(pep_register);
+module_exit(pep_unregister);
+MODULE_AUTHOR("Remi Denis-Courmont, Nokia");
+MODULE_DESCRIPTION("Phonet pipe protocol");
+MODULE_LICENSE("GPL");
+MODULE_ALIAS_NET_PF_PROTO(PF_PHONET, PN_PROTO_PIPE);