mctp: Add device handling and netlink interface
authorJeremy Kerr <jk@codeconstruct.com.au>
Thu, 29 Jul 2021 02:20:44 +0000 (10:20 +0800)
committerDavid S. Miller <davem@davemloft.net>
Thu, 29 Jul 2021 14:06:50 +0000 (15:06 +0100)
This change adds the infrastructure for managing MCTP netdevices; we add
a pointer to the AF_MCTP-specific data to struct netdevice, and hook up
the rtnetlink operations for adding and removing addresses.

Includes changes from Matt Johnston <matt@codeconstruct.com.au>.

Signed-off-by: Jeremy Kerr <jk@codeconstruct.com.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
MAINTAINERS
include/linux/netdevice.h
include/net/mctp.h
include/net/mctpdevice.h [new file with mode: 0644]
include/uapi/linux/if_ether.h
include/uapi/linux/if_link.h
include/uapi/linux/mctp.h
net/mctp/Makefile
net/mctp/af_mctp.c
net/mctp/device.c [new file with mode: 0644]

index e95eb3b00cd247ffe9457f69921725ce88964fa1..c2943c227ee6a7da3f28d1fe3d10dd6343079246 100644 (file)
@@ -11039,6 +11039,7 @@ L:      netdev@vger.kernel.org
 S:     Maintained
 F:     drivers/net/mctp/
 F:     include/net/mctp.h
+F:     include/net/mctpdevice.h
 F:     net/mctp/
 
 MAN-PAGES: MANUAL PAGES FOR LINUX -- Sections 2, 3, 4, 5, and 7
index 226bbee06730da15cebf68e75387573947813b62..d63a94ecbf3bbe54e195bf91858f3cb8e1fd9138 100644 (file)
@@ -1823,6 +1823,7 @@ enum netdev_ml_priv_type {
  *     @ieee802154_ptr: IEEE 802.15.4 low-rate Wireless Personal Area Network
  *                      device struct
  *     @mpls_ptr:      mpls_dev struct pointer
+ *     @mctp_ptr:      MCTP specific data
  *
  *     @dev_addr:      Hw address (before bcast,
  *                     because most packets are unicast)
@@ -2110,6 +2111,9 @@ struct net_device {
 #if IS_ENABLED(CONFIG_MPLS_ROUTING)
        struct mpls_dev __rcu   *mpls_ptr;
 #endif
+#if IS_ENABLED(CONFIG_MCTP)
+       struct mctp_dev __rcu   *mctp_ptr;
+#endif
 
 /*
  * Cache lines mostly used on receive path (including eth_type_trans())
index 4c01e083be45ec921a66dd6f832585916cd21579..61452e03aa85e34694b739512994ae3888dad6f2 100644 (file)
@@ -10,6 +10,7 @@
 #define __NET_MCTP_H
 
 #include <linux/bits.h>
+#include <linux/mctp.h>
 
 /* MCTP packet definitions */
 struct mctp_hdr {
@@ -32,4 +33,17 @@ struct mctp_hdr {
 #define MCTP_HDR_TAG_SHIFT     0
 #define MCTP_HDR_TAG_MASK      GENMASK(2, 0)
 
+static inline bool mctp_address_ok(mctp_eid_t eid)
+{
+       return eid >= 8 && eid < 255;
+}
+
+static inline struct mctp_hdr *mctp_hdr(struct sk_buff *skb)
+{
+       return (struct mctp_hdr *)skb_network_header(skb);
+}
+
+void mctp_device_init(void);
+void mctp_device_exit(void);
+
 #endif /* __NET_MCTP_H */
diff --git a/include/net/mctpdevice.h b/include/net/mctpdevice.h
new file mode 100644 (file)
index 0000000..71a1101
--- /dev/null
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Management Component Transport Protocol (MCTP) - device
+ * definitions.
+ *
+ * Copyright (c) 2021 Code Construct
+ * Copyright (c) 2021 Google
+ */
+
+#ifndef __NET_MCTPDEVICE_H
+#define __NET_MCTPDEVICE_H
+
+#include <linux/list.h>
+#include <linux/types.h>
+#include <linux/refcount.h>
+
+struct mctp_dev {
+       struct net_device       *dev;
+
+       unsigned int            net;
+
+       /* Only modified under RTNL. Reads have addrs_lock held */
+       u8                      *addrs;
+       size_t                  num_addrs;
+       spinlock_t              addrs_lock;
+
+       struct rcu_head         rcu;
+};
+
+#define MCTP_INITIAL_DEFAULT_NET       1
+
+struct mctp_dev *mctp_dev_get_rtnl(const struct net_device *dev);
+struct mctp_dev *__mctp_dev_get(const struct net_device *dev);
+
+#endif /* __NET_MCTPDEVICE_H */
index a0b637911d3cf97fa9c0785d20b31ab00bffc372..5f589c7a83824b54932242ea089d1e7c0d9d6e65 100644 (file)
 #define ETH_P_MAP      0x00F9          /* Qualcomm multiplexing and
                                         * aggregation protocol
                                         */
+#define ETH_P_MCTP     0x00FA          /* Management component transport
+                                        * protocol packets
+                                        */
 
 /*
  *     This is an Ethernet frame header.
index 4882e81514b664f5cb95f6a51d94a862efcd4a5c..49b22afab78f1fb1a4e069abb51cb5c59ded58f4 100644 (file)
@@ -1260,4 +1260,14 @@ struct ifla_rmnet_flags {
        __u32   mask;
 };
 
+/* MCTP section */
+
+enum {
+       IFLA_MCTP_UNSPEC,
+       IFLA_MCTP_NET,
+       __IFLA_MCTP_MAX,
+};
+
+#define IFLA_MCTP_MAX (__IFLA_MCTP_MAX - 1)
+
 #endif /* _UAPI_LINUX_IF_LINK_H */
index 52b54d13f3850fcd6e95dcc3119a2ad6bd613564..a9d8edb3402b6eef15e258da6daea337b0803412 100644 (file)
@@ -26,6 +26,7 @@ struct sockaddr_mctp {
 };
 
 #define MCTP_NET_ANY           0x0
+#define MCTP_NET_DEFAULT       0x0
 
 #define MCTP_ADDR_NULL         0x00
 #define MCTP_ADDR_ANY          0xff
index 7c056b1b7939d0c1d03946af82fc439ff2502569..2ea98c27b262ba0010559a41f0abccb1d1e07a1a 100644 (file)
@@ -1,3 +1,3 @@
 # SPDX-License-Identifier: GPL-2.0
 obj-$(CONFIG_MCTP) += mctp.o
-mctp-objs := af_mctp.o
+mctp-objs := af_mctp.o device.o
index b3aeca6486e37ce2d96e9b30a20507a2b4697b20..401b4fa141a5bbf0870ea1aa57ff4bf391bd8193 100644 (file)
@@ -6,13 +6,18 @@
  * Copyright (c) 2021 Google
  */
 
+#include <linux/if_arp.h>
 #include <linux/net.h>
 #include <linux/mctp.h>
 #include <linux/module.h>
 #include <linux/socket.h>
 
+#include <net/mctp.h>
+#include <net/mctpdevice.h>
 #include <net/sock.h>
 
+/* socket implementation */
+
 struct mctp_sock {
        struct sock     sk;
 };
@@ -152,6 +157,8 @@ static __init int mctp_init(void)
        if (rc)
                goto err_unreg_sock;
 
+       mctp_device_init();
+
        return 0;
 
 err_unreg_sock:
@@ -162,6 +169,7 @@ err_unreg_sock:
 
 static __exit void mctp_exit(void)
 {
+       mctp_device_exit();
        proto_unregister(&mctp_proto);
        sock_unregister(PF_MCTP);
 }
diff --git a/net/mctp/device.c b/net/mctp/device.c
new file mode 100644 (file)
index 0000000..877abe5
--- /dev/null
@@ -0,0 +1,414 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Management Component Transport Protocol (MCTP) - device implementation.
+ *
+ * Copyright (c) 2021 Code Construct
+ * Copyright (c) 2021 Google
+ */
+
+#include <linux/if_link.h>
+#include <linux/mctp.h>
+#include <linux/netdevice.h>
+#include <linux/rcupdate.h>
+#include <linux/rtnetlink.h>
+
+#include <net/addrconf.h>
+#include <net/netlink.h>
+#include <net/mctp.h>
+#include <net/mctpdevice.h>
+#include <net/sock.h>
+
+struct mctp_dump_cb {
+       int h;
+       int idx;
+       size_t a_idx;
+};
+
+/* unlocked: caller must hold rcu_read_lock */
+struct mctp_dev *__mctp_dev_get(const struct net_device *dev)
+{
+       return rcu_dereference(dev->mctp_ptr);
+}
+
+struct mctp_dev *mctp_dev_get_rtnl(const struct net_device *dev)
+{
+       return rtnl_dereference(dev->mctp_ptr);
+}
+
+static void mctp_dev_destroy(struct mctp_dev *mdev)
+{
+       struct net_device *dev = mdev->dev;
+
+       dev_put(dev);
+       kfree_rcu(mdev, rcu);
+}
+
+static int mctp_fill_addrinfo(struct sk_buff *skb, struct netlink_callback *cb,
+                             struct mctp_dev *mdev, mctp_eid_t eid)
+{
+       struct ifaddrmsg *hdr;
+       struct nlmsghdr *nlh;
+
+       nlh = nlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
+                       RTM_NEWADDR, sizeof(*hdr), NLM_F_MULTI);
+       if (!nlh)
+               return -EMSGSIZE;
+
+       hdr = nlmsg_data(nlh);
+       hdr->ifa_family = AF_MCTP;
+       hdr->ifa_prefixlen = 0;
+       hdr->ifa_flags = 0;
+       hdr->ifa_scope = 0;
+       hdr->ifa_index = mdev->dev->ifindex;
+
+       if (nla_put_u8(skb, IFA_LOCAL, eid))
+               goto cancel;
+
+       if (nla_put_u8(skb, IFA_ADDRESS, eid))
+               goto cancel;
+
+       nlmsg_end(skb, nlh);
+
+       return 0;
+
+cancel:
+       nlmsg_cancel(skb, nlh);
+       return -EMSGSIZE;
+}
+
+static int mctp_dump_dev_addrinfo(struct mctp_dev *mdev, struct sk_buff *skb,
+                                 struct netlink_callback *cb)
+{
+       struct mctp_dump_cb *mcb = (void *)cb->ctx;
+       int rc = 0;
+
+       for (; mcb->a_idx < mdev->num_addrs; mcb->a_idx++) {
+               rc = mctp_fill_addrinfo(skb, cb, mdev, mdev->addrs[mcb->a_idx]);
+               if (rc < 0)
+                       break;
+       }
+
+       return rc;
+}
+
+static int mctp_dump_addrinfo(struct sk_buff *skb, struct netlink_callback *cb)
+{
+       struct mctp_dump_cb *mcb = (void *)cb->ctx;
+       struct net *net = sock_net(skb->sk);
+       struct hlist_head *head;
+       struct net_device *dev;
+       struct ifaddrmsg *hdr;
+       struct mctp_dev *mdev;
+       int ifindex;
+       int idx, rc;
+
+       hdr = nlmsg_data(cb->nlh);
+       // filter by ifindex if requested
+       ifindex = hdr->ifa_index;
+
+       rcu_read_lock();
+       for (; mcb->h < NETDEV_HASHENTRIES; mcb->h++, mcb->idx = 0) {
+               idx = 0;
+               head = &net->dev_index_head[mcb->h];
+               hlist_for_each_entry_rcu(dev, head, index_hlist) {
+                       if (idx >= mcb->idx &&
+                           (ifindex == 0 || ifindex == dev->ifindex)) {
+                               mdev = __mctp_dev_get(dev);
+                               if (mdev) {
+                                       rc = mctp_dump_dev_addrinfo(mdev,
+                                                                   skb, cb);
+                                       // Error indicates full buffer, this
+                                       // callback will get retried.
+                                       if (rc < 0)
+                                               goto out;
+                               }
+                       }
+                       idx++;
+                       // reset for next iteration
+                       mcb->a_idx = 0;
+               }
+       }
+out:
+       rcu_read_unlock();
+       mcb->idx = idx;
+
+       return skb->len;
+}
+
+static const struct nla_policy ifa_mctp_policy[IFA_MAX + 1] = {
+       [IFA_ADDRESS]           = { .type = NLA_U8 },
+       [IFA_LOCAL]             = { .type = NLA_U8 },
+};
+
+static int mctp_rtm_newaddr(struct sk_buff *skb, struct nlmsghdr *nlh,
+                           struct netlink_ext_ack *extack)
+{
+       struct net *net = sock_net(skb->sk);
+       struct nlattr *tb[IFA_MAX + 1];
+       struct net_device *dev;
+       struct mctp_addr *addr;
+       struct mctp_dev *mdev;
+       struct ifaddrmsg *ifm;
+       unsigned long flags;
+       u8 *tmp_addrs;
+       int rc;
+
+       rc = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_mctp_policy,
+                        extack);
+       if (rc < 0)
+               return rc;
+
+       ifm = nlmsg_data(nlh);
+
+       if (tb[IFA_LOCAL])
+               addr = nla_data(tb[IFA_LOCAL]);
+       else if (tb[IFA_ADDRESS])
+               addr = nla_data(tb[IFA_ADDRESS]);
+       else
+               return -EINVAL;
+
+       /* find device */
+       dev = __dev_get_by_index(net, ifm->ifa_index);
+       if (!dev)
+               return -ENODEV;
+
+       mdev = mctp_dev_get_rtnl(dev);
+       if (!mdev)
+               return -ENODEV;
+
+       if (!mctp_address_ok(addr->s_addr))
+               return -EINVAL;
+
+       /* Prevent duplicates. Under RTNL so don't need to lock for reading */
+       if (memchr(mdev->addrs, addr->s_addr, mdev->num_addrs))
+               return -EEXIST;
+
+       tmp_addrs = kmalloc(mdev->num_addrs + 1, GFP_KERNEL);
+       if (!tmp_addrs)
+               return -ENOMEM;
+       memcpy(tmp_addrs, mdev->addrs, mdev->num_addrs);
+       tmp_addrs[mdev->num_addrs] = addr->s_addr;
+
+       /* Lock to write */
+       spin_lock_irqsave(&mdev->addrs_lock, flags);
+       mdev->num_addrs++;
+       swap(mdev->addrs, tmp_addrs);
+       spin_unlock_irqrestore(&mdev->addrs_lock, flags);
+
+       kfree(tmp_addrs);
+
+       return 0;
+}
+
+static int mctp_rtm_deladdr(struct sk_buff *skb, struct nlmsghdr *nlh,
+                           struct netlink_ext_ack *extack)
+{
+       struct net *net = sock_net(skb->sk);
+       struct nlattr *tb[IFA_MAX + 1];
+       struct net_device *dev;
+       struct mctp_addr *addr;
+       struct mctp_dev *mdev;
+       struct ifaddrmsg *ifm;
+       unsigned long flags;
+       u8 *pos;
+       int rc;
+
+       rc = nlmsg_parse(nlh, sizeof(*ifm), tb, IFA_MAX, ifa_mctp_policy,
+                        extack);
+       if (rc < 0)
+               return rc;
+
+       ifm = nlmsg_data(nlh);
+
+       if (tb[IFA_LOCAL])
+               addr = nla_data(tb[IFA_LOCAL]);
+       else if (tb[IFA_ADDRESS])
+               addr = nla_data(tb[IFA_ADDRESS]);
+       else
+               return -EINVAL;
+
+       /* find device */
+       dev = __dev_get_by_index(net, ifm->ifa_index);
+       if (!dev)
+               return -ENODEV;
+
+       mdev = mctp_dev_get_rtnl(dev);
+       if (!mdev)
+               return -ENODEV;
+
+       pos = memchr(mdev->addrs, addr->s_addr, mdev->num_addrs);
+       if (!pos)
+               return -ENOENT;
+
+       spin_lock_irqsave(&mdev->addrs_lock, flags);
+       memmove(pos, pos + 1, mdev->num_addrs - 1 - (pos - mdev->addrs));
+       mdev->num_addrs--;
+       spin_unlock_irqrestore(&mdev->addrs_lock, flags);
+
+       return 0;
+}
+
+static struct mctp_dev *mctp_add_dev(struct net_device *dev)
+{
+       struct mctp_dev *mdev;
+
+       ASSERT_RTNL();
+
+       mdev = kzalloc(sizeof(*mdev), GFP_KERNEL);
+       if (!mdev)
+               return ERR_PTR(-ENOMEM);
+
+       spin_lock_init(&mdev->addrs_lock);
+
+       mdev->net = MCTP_INITIAL_DEFAULT_NET;
+
+       /* associate to net_device */
+       rcu_assign_pointer(dev->mctp_ptr, mdev);
+       dev_hold(dev);
+       mdev->dev = dev;
+
+       return mdev;
+}
+
+static int mctp_fill_link_af(struct sk_buff *skb,
+                            const struct net_device *dev, u32 ext_filter_mask)
+{
+       struct mctp_dev *mdev;
+
+       mdev = mctp_dev_get_rtnl(dev);
+       if (!mdev)
+               return -ENODATA;
+       if (nla_put_u32(skb, IFLA_MCTP_NET, mdev->net))
+               return -EMSGSIZE;
+       return 0;
+}
+
+static size_t mctp_get_link_af_size(const struct net_device *dev,
+                                   u32 ext_filter_mask)
+{
+       struct mctp_dev *mdev;
+       unsigned int ret;
+
+       /* caller holds RCU */
+       mdev = __mctp_dev_get(dev);
+       if (!mdev)
+               return 0;
+       ret = nla_total_size(4); /* IFLA_MCTP_NET */
+       return ret;
+}
+
+static const struct nla_policy ifla_af_mctp_policy[IFLA_MCTP_MAX + 1] = {
+       [IFLA_MCTP_NET]         = { .type = NLA_U32 },
+};
+
+static int mctp_set_link_af(struct net_device *dev, const struct nlattr *attr,
+                           struct netlink_ext_ack *extack)
+{
+       struct nlattr *tb[IFLA_MCTP_MAX + 1];
+       struct mctp_dev *mdev;
+       int rc;
+
+       rc = nla_parse_nested(tb, IFLA_MCTP_MAX, attr, ifla_af_mctp_policy,
+                             NULL);
+       if (rc)
+               return rc;
+
+       mdev = mctp_dev_get_rtnl(dev);
+       if (!mdev)
+               return 0;
+
+       if (tb[IFLA_MCTP_NET])
+               WRITE_ONCE(mdev->net, nla_get_u32(tb[IFLA_MCTP_NET]));
+
+       return 0;
+}
+
+static void mctp_unregister(struct net_device *dev)
+{
+       struct mctp_dev *mdev;
+
+       mdev = mctp_dev_get_rtnl(dev);
+
+       if (!mdev)
+               return;
+
+       RCU_INIT_POINTER(mdev->dev->mctp_ptr, NULL);
+
+       kfree(mdev->addrs);
+
+       mctp_dev_destroy(mdev);
+}
+
+static int mctp_register(struct net_device *dev)
+{
+       struct mctp_dev *mdev;
+
+       /* Already registered? */
+       if (rtnl_dereference(dev->mctp_ptr))
+               return 0;
+
+       /* only register specific types; MCTP-specific and loopback for now */
+       if (dev->type != ARPHRD_MCTP && dev->type != ARPHRD_LOOPBACK)
+               return 0;
+
+       mdev = mctp_add_dev(dev);
+       if (IS_ERR(mdev))
+               return PTR_ERR(mdev);
+
+       return 0;
+}
+
+static int mctp_dev_notify(struct notifier_block *this, unsigned long event,
+                          void *ptr)
+{
+       struct net_device *dev = netdev_notifier_info_to_dev(ptr);
+       int rc;
+
+       switch (event) {
+       case NETDEV_REGISTER:
+               rc = mctp_register(dev);
+               if (rc)
+                       return notifier_from_errno(rc);
+               break;
+       case NETDEV_UNREGISTER:
+               mctp_unregister(dev);
+               break;
+       }
+
+       return NOTIFY_OK;
+}
+
+static struct rtnl_af_ops mctp_af_ops = {
+       .family = AF_MCTP,
+       .fill_link_af = mctp_fill_link_af,
+       .get_link_af_size = mctp_get_link_af_size,
+       .set_link_af = mctp_set_link_af,
+};
+
+static struct notifier_block mctp_dev_nb = {
+       .notifier_call = mctp_dev_notify,
+       .priority = ADDRCONF_NOTIFY_PRIORITY,
+};
+
+void __init mctp_device_init(void)
+{
+       register_netdevice_notifier(&mctp_dev_nb);
+
+       rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_GETADDR,
+                            NULL, mctp_dump_addrinfo, 0);
+       rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_NEWADDR,
+                            mctp_rtm_newaddr, NULL, 0);
+       rtnl_register_module(THIS_MODULE, PF_MCTP, RTM_DELADDR,
+                            mctp_rtm_deladdr, NULL, 0);
+       rtnl_af_register(&mctp_af_ops);
+}
+
+void __exit mctp_device_exit(void)
+{
+       rtnl_af_unregister(&mctp_af_ops);
+       rtnl_unregister(PF_MCTP, RTM_DELADDR);
+       rtnl_unregister(PF_MCTP, RTM_NEWADDR);
+       rtnl_unregister(PF_MCTP, RTM_GETADDR);
+
+       unregister_netdevice_notifier(&mctp_dev_nb);
+}