static LIST_HEAD(dps);
static struct vport *new_vport(const struct vport_parms *);
-static int queue_userspace_packets(struct datapath *, struct sk_buff *,
+static int queue_userspace_packets(struct datapath *, u32 pid, struct sk_buff *,
const struct dp_upcall_info *);
/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
}
+static int get_dpifindex(struct datapath *dp)
+{
+ struct vport *local;
+ int ifindex;
+
+ rcu_read_lock();
+
+ local = get_vport_protected(dp, OVSP_LOCAL);
+ if (local)
+ ifindex = vport_get_ifindex(local);
+ else
+ ifindex = 0;
+
+ rcu_read_unlock();
+
+ return ifindex;
+}
+
static inline size_t br_nlmsg_size(void)
{
return NLMSG_ALIGN(sizeof(struct ifinfomsg))
hdr->ifi_change = 0;
NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
- NLA_PUT_U32(skb, IFLA_MASTER,
- vport_get_ifindex(get_vport_protected(dp, OVSP_LOCAL)));
+ NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
#ifdef IFLA_OPERSTATE
NLA_PUT_U8(skb, IFLA_OPERSTATE,
upcall.cmd = OVS_PACKET_CMD_MISS;
upcall.key = &key;
- upcall.userdata = 0;
- upcall.sample_pool = 0;
- upcall.actions = NULL;
- upcall.actions_len = 0;
dp_upcall(dp, skb, &upcall);
stats_counter_off = offsetof(struct dp_stats_percpu, n_missed);
goto out;
.maxattr = OVS_PACKET_ATTR_MAX
};
-/* Generic Netlink multicast groups for upcalls.
- *
- * We really want three unique multicast groups per datapath, but we can't even
- * get one, because genl_register_mc_group() takes genl_lock, which is also
- * held during Generic Netlink message processing, so trying to acquire
- * multicast groups during OVS_DP_NEW processing deadlocks. Instead, we
- * preallocate a few groups and use them round-robin for datapaths. Collision
- * isn't fatal--multicast listeners should check that the family is the one
- * that they want and discard others--but it wastes time and memory to receive
- * unwanted messages.
- */
-#define PACKET_N_MC_GROUPS 16
-static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
-
-static u32 packet_mc_group(struct datapath *dp, u8 cmd)
-{
- u32 idx;
- BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
-
- idx = jhash_2words(dp->dp_ifindex, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
- return packet_mc_groups[idx].id;
-}
-
-static int packet_register_mc_groups(void)
-{
- int i;
-
- for (i = 0; i < PACKET_N_MC_GROUPS; i++) {
- struct genl_multicast_group *group = &packet_mc_groups[i];
- int error;
-
- sprintf(group->name, "packet%d", i);
- error = genl_register_mc_group(&dp_packet_genl_family, group);
- if (error)
- return error;
- }
- return 0;
-}
-
int dp_upcall(struct datapath *dp, struct sk_buff *skb, const struct dp_upcall_info *upcall_info)
{
struct dp_stats_percpu *stats;
+ u32 pid;
int err;
- WARN_ON_ONCE(skb_shared(skb));
+ if (OVS_CB(skb)->flow)
+ pid = OVS_CB(skb)->flow->upcall_pid;
+ else
+ pid = OVS_CB(skb)->vport->upcall_pid;
+
+ if (pid == 0) {
+ err = -ENOTCONN;
+ kfree_skb(skb);
+ goto err;
+ }
forward_ip_summed(skb, true);
skb = nskb;
}
- err = queue_userspace_packets(dp, skb, upcall_info);
+ err = queue_userspace_packets(dp, pid, skb, upcall_info);
if (err)
goto err;
* 'upcall_info'. There will be only one packet unless we broke up a GSO
* packet.
*/
-static int queue_userspace_packets(struct datapath *dp, struct sk_buff *skb,
- const struct dp_upcall_info *upcall_info)
+static int queue_userspace_packets(struct datapath *dp, u32 pid,
+ struct sk_buff *skb,
+ const struct dp_upcall_info *upcall_info)
{
- u32 group = packet_mc_group(dp, upcall_info->cmd);
+ int dp_ifindex;
struct sk_buff *nskb;
int err;
+ dp_ifindex = get_dpifindex(dp);
+ if (!dp_ifindex) {
+ err = -ENODEV;
+ nskb = skb->next;
+ goto err_kfree_skbs;
+ }
+
do {
struct ovs_header *upcall;
struct sk_buff *user_skb; /* to be queued to userspace */
if (unlikely(err))
goto err_kfree_skbs;
- if (nla_attr_size(skb->len) > USHRT_MAX)
+ if (nla_attr_size(skb->len) > USHRT_MAX) {
+ err = -EFBIG;
goto err_kfree_skbs;
+ }
len = sizeof(struct ovs_header);
len += nla_total_size(skb->len);
len += nla_total_size(FLOW_BUFSIZE);
- if (upcall_info->userdata)
+ if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
len += nla_total_size(8);
- if (upcall_info->sample_pool)
- len += nla_total_size(4);
- if (upcall_info->actions_len)
- len += nla_total_size(upcall_info->actions_len);
user_skb = genlmsg_new(len, GFP_ATOMIC);
if (!user_skb) {
- netlink_set_err(INIT_NET_GENL_SOCK, 0, group, -ENOBUFS);
+ err = -ENOMEM;
goto err_kfree_skbs;
}
upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
- upcall->dp_ifindex = dp->dp_ifindex;
+ upcall->dp_ifindex = dp_ifindex;
nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
flow_to_nlattrs(upcall_info->key, user_skb);
nla_nest_end(user_skb, nla);
- if (upcall_info->userdata)
- nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA, upcall_info->userdata);
- if (upcall_info->sample_pool)
- nla_put_u32(user_skb, OVS_PACKET_ATTR_SAMPLE_POOL, upcall_info->sample_pool);
- if (upcall_info->actions_len) {
- const struct nlattr *actions = upcall_info->actions;
- u32 actions_len = upcall_info->actions_len;
-
- nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
- memcpy(__skb_put(user_skb, actions_len), actions, actions_len);
- nla_nest_end(user_skb, nla);
- }
+ if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
+ nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
+ upcall_info->userdata);
nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
if (skb->ip_summed == CHECKSUM_PARTIAL)
else
skb_copy_bits(skb, 0, nla_data(nla), skb->len);
- err = genlmsg_multicast(user_skb, 0, group, GFP_ATOMIC);
+ err = genlmsg_unicast(&init_net, user_skb, pid);
if (err)
goto err_kfree_skbs;
return 0;
}
-static int validate_actions(const struct nlattr *attr)
+static int validate_actions(const struct nlattr *attr, int depth);
+
+static int validate_sample(const struct nlattr *attr, int depth)
+{
+ static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
+ {
+ [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
+ [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
+ };
+ struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
+ int error;
+
+ error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
+ if (error)
+ return error;
+
+ if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
+ return -EINVAL;
+ if (!a[OVS_SAMPLE_ATTR_ACTIONS])
+ return -EINVAL;
+
+ return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], (depth + 1));
+}
+
+static int validate_actions(const struct nlattr *attr, int depth)
{
const struct nlattr *a;
- int rem;
+ int rem, err;
+
+ if (depth >= SAMPLE_ACTION_DEPTH)
+ return -EOVERFLOW;
nla_for_each_nested(a, attr, rem) {
static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
};
int type = nla_type(a);
- if (type > OVS_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
+ /* Match expected attr len for given attr len except for
+ * OVS_ACTION_ATTR_SAMPLE, as it has nested actions list which
+ * is variable size. */
+ if (type > OVS_ACTION_ATTR_MAX ||
+ (nla_len(a) != action_lens[type] &&
+ type != OVS_ACTION_ATTR_SAMPLE))
return -EINVAL;
switch (type) {
return -EINVAL;
break;
+ case OVS_ACTION_ATTR_SAMPLE:
+ err = validate_sample(a, depth);
+ if (err)
+ return err;
+ break;
+
default:
return -EOPNOTSUPP;
}
nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
goto err;
- err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS]);
+ err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], 0);
if (err)
goto err;
flow->hash = flow_hash(&flow->key, key_len);
+ if (a[OVS_PACKET_ATTR_UPCALL_PID])
+ flow->upcall_pid = nla_get_u32(a[OVS_PACKET_ATTR_UPCALL_PID]);
+ else
+ flow->upcall_pid = NETLINK_CB(skb).pid;
+
acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
err = PTR_ERR(acts);
if (IS_ERR(acts))
err = -ENODEV;
if (!dp)
goto err_unlock;
+
+ if (flow->key.eth.in_port < DP_MAX_PORTS)
+ OVS_CB(packet)->vport = get_vport_protected(dp,
+ flow->key.eth.in_port);
+
err = execute_actions(dp, packet);
rcu_read_unlock();
[OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
+ [OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
};
static struct genl_ops dp_packet_genl_ops[] = {
}
}
-/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports.
- * Called with RTNL lock.
- */
-int dp_min_mtu(const struct datapath *dp)
-{
- struct vport *p;
- int mtu = 0;
-
- ASSERT_RTNL();
-
- list_for_each_entry (p, &dp->port_list, node) {
- int dev_mtu;
-
- /* Skip any internal ports, since that's what we're trying to
- * set. */
- if (is_internal_vport(p))
- continue;
-
- dev_mtu = vport_get_mtu(p);
- if (!dev_mtu)
- continue;
- if (!mtu || dev_mtu < mtu)
- mtu = dev_mtu;
- }
-
- return mtu ? mtu : ETH_DATA_LEN;
-}
-
-/* Sets the MTU of all datapath devices to the minimum of the ports
- * Called with RTNL lock.
- */
-void set_internal_devs_mtu(const struct datapath *dp)
-{
- struct vport *p;
- int mtu;
-
- ASSERT_RTNL();
-
- mtu = dp_min_mtu(dp);
-
- list_for_each_entry (p, &dp->port_list, node) {
- if (is_internal_vport(p))
- vport_set_mtu(p, mtu);
- }
-}
-
static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
+ [OVS_FLOW_ATTR_UPCALL_PID] = { .type = NLA_U32 },
[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
};
if (!ovs_header)
return -EMSGSIZE;
- ovs_header->dp_ifindex = dp->dp_ifindex;
+ ovs_header->dp_ifindex = get_dpifindex(dp);
nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
if (!nla)
goto error;
nla_nest_end(skb, nla);
+ NLA_PUT_U32(skb, OVS_FLOW_ATTR_UPCALL_PID, flow->upcall_pid);
+
spin_lock_bh(&flow->lock);
used = flow->used;
stats.n_packets = flow->packet_count;
/* Validate actions. */
if (a[OVS_FLOW_ATTR_ACTIONS]) {
- error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS]);
+ error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], 0);
if (error)
goto error;
} else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
flow->key = key;
clear_stats(flow);
+ if (a[OVS_FLOW_ATTR_UPCALL_PID])
+ flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
+ else
+ flow->upcall_pid = NETLINK_CB(skb).pid;
+
/* Obtain actions. */
acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
error = PTR_ERR(acts);
reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
info->snd_seq, OVS_FLOW_CMD_NEW);
+ if (a[OVS_FLOW_ATTR_UPCALL_PID])
+ flow->upcall_pid = nla_get_u32(a[OVS_FLOW_ATTR_UPCALL_PID]);
+
/* Clear stats. */
if (a[OVS_FLOW_ATTR_CLEAR]) {
spin_lock_bh(&flow->lock);
#ifdef HAVE_NLA_NUL_STRING
[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
#endif
+ [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
[OVS_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
- [OVS_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
};
static struct genl_family dp_datapath_genl_family = {
if (!ovs_header)
goto error;
- ovs_header->dp_ifindex = dp->dp_ifindex;
+ ovs_header->dp_ifindex = get_dpifindex(dp);
rcu_read_lock();
err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
NLA_PUT_U32(skb, OVS_DP_ATTR_IPV4_FRAGS,
dp->drop_frags ? OVS_DP_FRAG_DROP : OVS_DP_FRAG_ZERO);
- if (dp->sflow_probability)
- NLA_PUT_U32(skb, OVS_DP_ATTR_SAMPLING, dp->sflow_probability);
-
- nla = nla_nest_start(skb, OVS_DP_ATTR_MCGROUPS);
- if (!nla)
- goto nla_put_failure;
- NLA_PUT_U32(skb, OVS_PACKET_CMD_MISS, packet_mc_group(dp, OVS_PACKET_CMD_MISS));
- NLA_PUT_U32(skb, OVS_PACKET_CMD_ACTION, packet_mc_group(dp, OVS_PACKET_CMD_ACTION));
- NLA_PUT_U32(skb, OVS_PACKET_CMD_SAMPLE, packet_mc_group(dp, OVS_PACKET_CMD_SAMPLE));
- nla_nest_end(skb, nla);
-
return genlmsg_end(skb, ovs_header);
nla_put_failure:
{
if (a[OVS_DP_ATTR_IPV4_FRAGS])
dp->drop_frags = nla_get_u32(a[OVS_DP_ATTR_IPV4_FRAGS]) == OVS_DP_FRAG_DROP;
- if (a[OVS_DP_ATTR_SAMPLING])
- dp->sflow_probability = nla_get_u32(a[OVS_DP_ATTR_SAMPLING]);
}
static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
if (!dp->table)
goto err_free_dp;
+ dp->drop_frags = 0;
+ dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
+ if (!dp->stats_percpu) {
+ err = -ENOMEM;
+ goto err_destroy_table;
+ }
+
+ change_datapath(dp, a);
+
/* Set up our datapath device. */
parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
parms.type = OVS_VPORT_TYPE_INTERNAL;
parms.options = NULL;
parms.dp = dp;
parms.port_no = OVSP_LOCAL;
+ if (a[OVS_DP_ATTR_UPCALL_PID])
+ parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
+ else
+ parms.upcall_pid = NETLINK_CB(skb).pid;
+
vport = new_vport(&parms);
if (IS_ERR(vport)) {
err = PTR_ERR(vport);
if (err == -EBUSY)
err = -EEXIST;
- goto err_destroy_table;
- }
- dp->dp_ifindex = vport_get_ifindex(vport);
-
- dp->drop_frags = 0;
- dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
- if (!dp->stats_percpu) {
- err = -ENOMEM;
- goto err_destroy_local_port;
+ goto err_destroy_percpu;
}
- change_datapath(dp, a);
-
reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
err = PTR_ERR(reply);
if (IS_ERR(reply))
err_destroy_local_port:
dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
+err_destroy_percpu:
+ free_percpu(dp->stats_percpu);
err_destroy_table:
flow_tbl_destroy(get_table_protected(dp));
err_free_dp:
static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
#ifdef HAVE_NLA_NUL_STRING
[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
- [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
- [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
- [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct rtnl_link_stats64) },
+ [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
[OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
#else
- [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct rtnl_link_stats64) },
+ [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
[OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
#endif
+ [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
+ [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
};
if (!ovs_header)
return -EMSGSIZE;
- ovs_header->dp_ifindex = vport->dp->dp_ifindex;
+ ovs_header->dp_ifindex = get_dpifindex(vport->dp);
NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
+ NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
- nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct rtnl_link_stats64));
+ nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
if (!nla)
goto nla_put_failure;
- if (vport_get_stats(vport, nla_data(nla)))
- __skb_trim(skb, skb->len - nla->nla_len);
+
+ vport_get_stats(vport, nla_data(nla));
NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
{
int err = 0;
+
if (a[OVS_VPORT_ATTR_STATS])
- err = vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
- if (!err && a[OVS_VPORT_ATTR_ADDRESS])
+ vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
+
+ if (a[OVS_VPORT_ATTR_ADDRESS])
err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
+
return err;
}
parms.options = a[OVS_VPORT_ATTR_OPTIONS];
parms.dp = dp;
parms.port_no = port_no;
+ if (a[OVS_VPORT_ATTR_UPCALL_PID])
+ parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
+ else
+ parms.upcall_pid = NETLINK_CB(skb).pid;
vport = new_vport(&parms);
err = PTR_ERR(vport);
if (IS_ERR(vport))
goto exit_unlock;
- set_internal_devs_mtu(dp);
dp_sysfs_add_if(vport);
err = change_vport(vport, a);
err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
if (!err)
err = change_vport(vport, a);
+ if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
+ vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
OVS_VPORT_CMD_NEW);
}
}
- err = packet_register_mc_groups();
- if (err)
- goto error;
return 0;
error:
git://git.onelab.eu / sliver-openvswitch.git / blobdiff