#include <linux/init.h>
#include <linux/module.h>
-#include <linux/fs.h>
#include <linux/if_arp.h>
#include <linux/if_vlan.h>
#include <linux/in.h>
#include <linux/dmi.h>
#include <net/inet_ecn.h>
#include <net/genetlink.h>
-#include <linux/compat.h>
#include "openvswitch/datapath-protocol.h"
#include "checksum.h"
#include "flow.h"
#include "loop_counter.h"
#include "table.h"
+#include "vlan.h"
#include "vport-internal_dev.h"
int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
* each other.
*/
-/* Protected by genl_mutex. */
-static struct datapath __rcu *dps[256];
+/* Global list of datapaths to enable dumping them all out.
+ * Protected by genl_mutex.
+ */
+static LIST_HEAD(dps);
static struct vport *new_vport(const struct vport_parms *);
+static int queue_control_packets(struct datapath *, struct sk_buff *,
+ const struct dp_upcall_info *);
/* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
-struct datapath *get_dp(int dp_idx)
+struct datapath *get_dp(int dp_ifindex)
{
- if (dp_idx < 0 || dp_idx >= ARRAY_SIZE(dps))
- return NULL;
+ struct datapath *dp = NULL;
+ struct net_device *dev;
- return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
- lockdep_rtnl_is_held() ||
- lockdep_genl_is_held());
+ rcu_read_lock();
+ dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
+ if (dev) {
+ struct vport *vport = internal_dev_get_vport(dev);
+ if (vport)
+ dp = vport->dp;
+ }
+ rcu_read_unlock();
+
+ return dp;
}
EXPORT_SYMBOL_GPL(get_dp);
get_skb_csum_pointers(skb, &csum_start, &csum_offset);
csum_start -= skb_headroom(skb);
- BUG_ON(csum_start >= skb_headlen(skb));
skb_copy_bits(skb, 0, to, csum_start);
*(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
}
-static struct genl_family dp_packet_genl_family;
+static struct genl_family dp_packet_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct odp_header),
+ .name = ODP_PACKET_FAMILY,
+ .version = 1,
+ .maxattr = ODP_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 ODP_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 int packet_mc_group(struct datapath *dp, u8 cmd)
+static u32 packet_mc_group(struct datapath *dp, u8 cmd)
{
+ u32 idx;
BUILD_BUG_ON_NOT_POWER_OF_2(PACKET_N_MC_GROUPS);
- return jhash_2words(dp->dp_idx, cmd, 0) & (PACKET_N_MC_GROUPS - 1);
+
+ 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;
+ int err;
+
+ WARN_ON_ONCE(skb_shared(skb));
+
+ forward_ip_summed(skb);
+
+ err = vswitch_skb_checksum_setup(skb);
+ if (err)
+ goto err_kfree_skb;
+
+ /* Break apart GSO packets into their component pieces. Otherwise
+ * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
+ if (skb_is_gso(skb)) {
+ struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
+
+ kfree_skb(skb);
+ skb = nskb;
+ if (IS_ERR(skb)) {
+ err = PTR_ERR(skb);
+ goto err;
+ }
+ }
+
+ err = queue_control_packets(dp, skb, upcall_info);
+ if (err)
+ goto err;
+
+ return 0;
+
+err_kfree_skb:
+ kfree_skb(skb);
+err:
+ local_bh_disable();
+ stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
+
+ write_seqcount_begin(&stats->seqlock);
+ stats->n_lost++;
+ write_seqcount_end(&stats->seqlock);
+
+ local_bh_enable();
+
+ return err;
}
/* Send each packet in the 'skb' list to userspace for 'dp' as directed by
nskb = skb->next;
skb->next = NULL;
+ err = vlan_deaccel_tag(skb);
+ if (unlikely(err))
+ goto err_kfree_skbs;
+
+ if (nla_attr_size(skb->len) > USHRT_MAX)
+ goto err_kfree_skbs;
+
len = sizeof(struct odp_header);
- len += nla_total_size(4); /* ODP_PACKET_ATTR_TYPE. */
len += nla_total_size(skb->len);
len += nla_total_size(FLOW_BUFSIZE);
if (upcall_info->userdata)
}
upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family, 0, upcall_info->cmd);
- upcall->dp_idx = dp->dp_idx;
+ upcall->dp_ifindex = dp->dp_ifindex;
nla = nla_nest_start(user_skb, ODP_PACKET_ATTR_KEY);
flow_to_nlattrs(upcall_info->key, user_skb);
return err;
}
-/* 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 ODP_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.
- */
-static struct genl_multicast_group packet_mc_groups[PACKET_N_MC_GROUPS];
-
-static struct genl_family dp_packet_genl_family = {
- .id = GENL_ID_GENERATE,
- .hdrsize = sizeof(struct odp_header),
- .name = ODP_PACKET_FAMILY,
- .version = 1,
- .maxattr = ODP_PACKET_ATTR_MAX
-};
-
-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;
- int err;
-
- WARN_ON_ONCE(skb_shared(skb));
-
- forward_ip_summed(skb);
-
- err = vswitch_skb_checksum_setup(skb);
- if (err)
- goto err_kfree_skb;
-
- /* Break apart GSO packets into their component pieces. Otherwise
- * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */
- if (skb_is_gso(skb)) {
- struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
-
- kfree_skb(skb);
- skb = nskb;
- if (IS_ERR(skb)) {
- err = PTR_ERR(skb);
- goto err;
- }
- }
-
- return queue_control_packets(dp, skb, upcall_info);
-
-err_kfree_skb:
- kfree_skb(skb);
-err:
- local_bh_disable();
- stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
-
- write_seqcount_begin(&stats->seqlock);
- stats->n_lost++;
- write_seqcount_end(&stats->seqlock);
-
- local_bh_enable();
-
- return err;
-}
-
/* Called with genl_mutex. */
-static int flush_flows(int dp_idx)
+static int flush_flows(int dp_ifindex)
{
struct tbl *old_table;
struct tbl *new_table;
struct datapath *dp;
- dp = get_dp(dp_idx);
+ dp = get_dp(dp_ifindex);
if (!dp)
return -ENODEV;
return 0;
}
-static int validate_actions(const struct nlattr *actions, u32 actions_len)
+static int validate_actions(const struct nlattr *attr)
{
const struct nlattr *a;
int rem;
- nla_for_each_attr(a, actions, actions_len, rem) {
- static const u32 action_lens[ODPAT_MAX + 1] = {
- [ODPAT_OUTPUT] = 4,
- [ODPAT_CONTROLLER] = 8,
- [ODPAT_SET_DL_TCI] = 2,
- [ODPAT_STRIP_VLAN] = 0,
- [ODPAT_SET_DL_SRC] = ETH_ALEN,
- [ODPAT_SET_DL_DST] = ETH_ALEN,
- [ODPAT_SET_NW_SRC] = 4,
- [ODPAT_SET_NW_DST] = 4,
- [ODPAT_SET_NW_TOS] = 1,
- [ODPAT_SET_TP_SRC] = 2,
- [ODPAT_SET_TP_DST] = 2,
- [ODPAT_SET_TUNNEL] = 8,
- [ODPAT_SET_PRIORITY] = 4,
- [ODPAT_POP_PRIORITY] = 0,
- [ODPAT_DROP_SPOOFED_ARP] = 0,
+ nla_for_each_nested(a, attr, rem) {
+ static const u32 action_lens[ODP_ACTION_ATTR_MAX + 1] = {
+ [ODP_ACTION_ATTR_OUTPUT] = 4,
+ [ODP_ACTION_ATTR_CONTROLLER] = 8,
+ [ODP_ACTION_ATTR_SET_DL_TCI] = 2,
+ [ODP_ACTION_ATTR_STRIP_VLAN] = 0,
+ [ODP_ACTION_ATTR_SET_DL_SRC] = ETH_ALEN,
+ [ODP_ACTION_ATTR_SET_DL_DST] = ETH_ALEN,
+ [ODP_ACTION_ATTR_SET_NW_SRC] = 4,
+ [ODP_ACTION_ATTR_SET_NW_DST] = 4,
+ [ODP_ACTION_ATTR_SET_NW_TOS] = 1,
+ [ODP_ACTION_ATTR_SET_TP_SRC] = 2,
+ [ODP_ACTION_ATTR_SET_TP_DST] = 2,
+ [ODP_ACTION_ATTR_SET_TUNNEL] = 8,
+ [ODP_ACTION_ATTR_SET_PRIORITY] = 4,
+ [ODP_ACTION_ATTR_POP_PRIORITY] = 0,
+ [ODP_ACTION_ATTR_DROP_SPOOFED_ARP] = 0,
};
int type = nla_type(a);
- if (type > ODPAT_MAX || nla_len(a) != action_lens[type])
+ if (type > ODP_ACTION_ATTR_MAX || nla_len(a) != action_lens[type])
return -EINVAL;
switch (type) {
- case ODPAT_UNSPEC:
+ case ODP_ACTION_ATTR_UNSPEC:
return -EINVAL;
- case ODPAT_CONTROLLER:
- case ODPAT_STRIP_VLAN:
- case ODPAT_SET_DL_SRC:
- case ODPAT_SET_DL_DST:
- case ODPAT_SET_NW_SRC:
- case ODPAT_SET_NW_DST:
- case ODPAT_SET_TP_SRC:
- case ODPAT_SET_TP_DST:
- case ODPAT_SET_TUNNEL:
- case ODPAT_SET_PRIORITY:
- case ODPAT_POP_PRIORITY:
- case ODPAT_DROP_SPOOFED_ARP:
+ case ODP_ACTION_ATTR_CONTROLLER:
+ case ODP_ACTION_ATTR_STRIP_VLAN:
+ case ODP_ACTION_ATTR_SET_DL_SRC:
+ case ODP_ACTION_ATTR_SET_DL_DST:
+ case ODP_ACTION_ATTR_SET_NW_SRC:
+ case ODP_ACTION_ATTR_SET_NW_DST:
+ case ODP_ACTION_ATTR_SET_TP_SRC:
+ case ODP_ACTION_ATTR_SET_TP_DST:
+ case ODP_ACTION_ATTR_SET_TUNNEL:
+ case ODP_ACTION_ATTR_SET_PRIORITY:
+ case ODP_ACTION_ATTR_POP_PRIORITY:
+ case ODP_ACTION_ATTR_DROP_SPOOFED_ARP:
/* No validation needed. */
break;
- case ODPAT_OUTPUT:
+ case ODP_ACTION_ATTR_OUTPUT:
if (nla_get_u32(a) >= DP_MAX_PORTS)
return -EINVAL;
break;
- case ODPAT_SET_DL_TCI:
+ case ODP_ACTION_ATTR_SET_DL_TCI:
if (nla_get_be16(a) & htons(VLAN_CFI_MASK))
return -EINVAL;
break;
- case ODPAT_SET_NW_TOS:
+ case ODP_ACTION_ATTR_SET_NW_TOS:
if (nla_get_u8(a) & INET_ECN_MASK)
return -EINVAL;
break;
return 0;
}
-
-struct dp_flowcmd {
- u32 nlmsg_flags;
- u32 dp_idx;
- u32 total_len;
- struct sw_flow_key key;
- const struct nlattr *actions;
- u32 actions_len;
- bool clear;
- u64 state;
-};
-
-static struct sw_flow_actions *get_actions(const struct dp_flowcmd *flowcmd)
-{
- struct sw_flow_actions *actions;
-
- actions = flow_actions_alloc(flowcmd->actions_len);
- if (!IS_ERR(actions) && flowcmd->actions_len)
- memcpy(actions->actions, flowcmd->actions, flowcmd->actions_len);
- return actions;
-}
-
static void clear_stats(struct sw_flow *flow)
{
flow->used = 0;
struct odp_header *odp_header = info->userhdr;
struct nlattr **a = info->attrs;
struct sk_buff *packet;
- unsigned int actions_len;
- struct nlattr *actions;
struct sw_flow_key key;
struct datapath *dp;
struct ethhdr *eth;
err = -EINVAL;
if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
- goto exit;
+ goto err;
- actions = nla_data(a[ODP_PACKET_ATTR_ACTIONS]);
- actions_len = nla_len(a[ODP_PACKET_ATTR_ACTIONS]);
- err = validate_actions(actions, actions_len);
+ err = validate_actions(a[ODP_PACKET_ATTR_ACTIONS]);
if (err)
- goto exit;
+ goto err;
packet = skb_clone(skb, GFP_KERNEL);
err = -ENOMEM;
if (!packet)
- goto exit;
+ goto err;
packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
else
packet->protocol = htons(ETH_P_802_2);
+ /* Initialize OVS_CB (it came from Netlink so might not be zeroed). */
+ memset(OVS_CB(packet), 0, sizeof(struct ovs_skb_cb));
+
err = flow_extract(packet, -1, &key, &is_frag);
if (err)
- goto exit;
+ goto err_kfree_skb;
rcu_read_lock();
- dp = get_dp(odp_header->dp_idx);
+ dp = get_dp(odp_header->dp_ifindex);
err = -ENODEV;
- if (dp)
- err = execute_actions(dp, packet, &key, actions, actions_len);
+ if (!dp)
+ goto err_unlock;
+ err = execute_actions(dp, packet, &key,
+ nla_data(a[ODP_PACKET_ATTR_ACTIONS]),
+ nla_len(a[ODP_PACKET_ATTR_ACTIONS]));
rcu_read_unlock();
+ return err;
-exit:
+err_unlock:
+ rcu_read_unlock();
+err_kfree_skb:
+ kfree_skb(packet);
+err:
return err;
}
continue;
dev_mtu = vport_get_mtu(p);
+ if (!dev_mtu)
+ continue;
if (!mtu || dev_mtu < mtu)
mtu = dev_mtu;
}
[ODP_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
[ODP_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
[ODP_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
- [ODP_FLOW_ATTR_STATE] = { .type = NLA_U64 },
};
+static struct genl_family dp_flow_genl_family = {
+ .id = GENL_ID_GENERATE,
+ .hdrsize = sizeof(struct odp_header),
+ .name = ODP_FLOW_FAMILY,
+ .version = 1,
+ .maxattr = ODP_FLOW_ATTR_MAX
+};
+
+static struct genl_multicast_group dp_flow_multicast_group = {
+ .name = ODP_FLOW_MCGROUP
+};
-static int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp,
- struct sw_flow *flow, u32 total_len, u64 state)
+/* Called with genl_lock. */
+static int odp_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
+ struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
{
+ const int skb_orig_len = skb->len;
const struct sw_flow_actions *sf_acts;
struct odp_flow_stats stats;
- struct odp_flow *odp_flow;
- struct sk_buff *skb;
+ struct odp_header *odp_header;
struct nlattr *nla;
unsigned long used;
u8 tcp_flags;
sf_acts = rcu_dereference_protected(flow->sf_acts,
lockdep_genl_is_held());
- skb = alloc_skb(128 + FLOW_BUFSIZE + sf_acts->actions_len, GFP_KERNEL);
- err = -ENOMEM;
- if (!skb)
- goto exit;
+ odp_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
+ if (!odp_header)
+ return -EMSGSIZE;
- odp_flow = (struct odp_flow*)__skb_put(skb, sizeof(struct odp_flow));
- odp_flow->dp_idx = dp->dp_idx;
- odp_flow->total_len = total_len;
+ odp_header->dp_ifindex = dp->dp_ifindex;
nla = nla_nest_start(skb, ODP_FLOW_ATTR_KEY);
if (!nla)
goto nla_put_failure;
err = flow_to_nlattrs(&flow->key, skb);
if (err)
- goto exit_free;
- nla_nest_end(skb, nla);
-
- nla = nla_nest_start(skb, ODP_FLOW_ATTR_ACTIONS);
- if (!nla || skb_tailroom(skb) < sf_acts->actions_len)
- goto nla_put_failure;
- memcpy(__skb_put(skb, sf_acts->actions_len), sf_acts->actions, sf_acts->actions_len);
+ goto error;
nla_nest_end(skb, nla);
spin_lock_bh(&flow->lock);
spin_unlock_bh(&flow->lock);
if (used)
- NLA_PUT_MSECS(skb, ODP_FLOW_ATTR_USED, used);
+ NLA_PUT_U64(skb, ODP_FLOW_ATTR_USED, flow_used_time(used));
if (stats.n_packets)
NLA_PUT(skb, ODP_FLOW_ATTR_STATS, sizeof(struct odp_flow_stats), &stats);
if (tcp_flags)
NLA_PUT_U8(skb, ODP_FLOW_ATTR_TCP_FLAGS, tcp_flags);
- if (state)
- NLA_PUT_U64(skb, ODP_FLOW_ATTR_STATE, state);
-
- if (skb->len > total_len)
- goto nla_put_failure;
+ /* If ODP_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
+ * this is the first flow to be dumped into 'skb'. This is unusual for
+ * Netlink but individual action lists can be longer than
+ * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
+ * The userspace caller can always fetch the actions separately if it
+ * really wants them. (Most userspace callers in fact don't care.)
+ *
+ * This can only fail for dump operations because the skb is always
+ * properly sized for single flows.
+ */
+ err = nla_put(skb, ODP_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
+ sf_acts->actions);
+ if (err < 0 && skb_orig_len)
+ goto error;
- odp_flow->len = skb->len;
- err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0;
- goto exit_free;
+ return genlmsg_end(skb, odp_header);
nla_put_failure:
err = -EMSGSIZE;
-exit_free:
- kfree_skb(skb);
-exit:
+error:
+ genlmsg_cancel(skb, odp_header);
return err;
}
-/* Called with genl_mutex. */
-static struct sk_buff *copy_flow_from_user(struct odp_flow __user *uodp_flow,
- struct dp_flowcmd *flowcmd)
+static struct sk_buff *odp_flow_cmd_alloc_info(struct sw_flow *flow)
{
- struct nlattr *a[ODP_FLOW_ATTR_MAX + 1];
- struct odp_flow *odp_flow;
- struct sk_buff *skb;
- u32 len;
- int err;
-
- if (get_user(len, &uodp_flow->len))
- return ERR_PTR(-EFAULT);
- if (len < sizeof(struct odp_flow))
- return ERR_PTR(-EINVAL);
-
- skb = alloc_skb(len, GFP_KERNEL);
- if (!skb)
- return ERR_PTR(-ENOMEM);
-
- err = -EFAULT;
- if (copy_from_user(__skb_put(skb, len), uodp_flow, len))
- goto error_free_skb;
-
- odp_flow = (struct odp_flow *)skb->data;
- err = -EINVAL;
- if (odp_flow->len != len)
- goto error_free_skb;
-
- flowcmd->nlmsg_flags = odp_flow->nlmsg_flags;
- flowcmd->dp_idx = odp_flow->dp_idx;
- flowcmd->total_len = odp_flow->total_len;
-
- err = nla_parse(a, ODP_FLOW_ATTR_MAX,
- (struct nlattr *)(skb->data + sizeof(struct odp_flow)),
- skb->len - sizeof(struct odp_flow), flow_policy);
- if (err)
- goto error_free_skb;
+ const struct sw_flow_actions *sf_acts;
+ int len;
- /* ODP_FLOW_ATTR_KEY. */
- if (a[ODP_FLOW_ATTR_KEY]) {
- err = flow_from_nlattrs(&flowcmd->key, a[ODP_FLOW_ATTR_KEY]);
- if (err)
- goto error_free_skb;
- } else
- memset(&flowcmd->key, 0, sizeof(struct sw_flow_key));
+ sf_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_genl_is_held());
- /* ODP_FLOW_ATTR_ACTIONS. */
- if (a[ODP_FLOW_ATTR_ACTIONS]) {
- flowcmd->actions = nla_data(a[ODP_FLOW_ATTR_ACTIONS]);
- flowcmd->actions_len = nla_len(a[ODP_FLOW_ATTR_ACTIONS]);
- err = validate_actions(flowcmd->actions, flowcmd->actions_len);
- if (err)
- goto error_free_skb;
- } else {
- flowcmd->actions = NULL;
- flowcmd->actions_len = 0;
- }
+ len = nla_total_size(FLOW_BUFSIZE); /* ODP_FLOW_ATTR_KEY */
+ len += nla_total_size(sf_acts->actions_len); /* ODP_FLOW_ATTR_ACTIONS */
+ len += nla_total_size(sizeof(struct odp_flow_stats)); /* ODP_FLOW_ATTR_STATS */
+ len += nla_total_size(1); /* ODP_FLOW_ATTR_TCP_FLAGS */
+ len += nla_total_size(8); /* ODP_FLOW_ATTR_USED */
+ return genlmsg_new(NLMSG_ALIGN(sizeof(struct odp_header)) + len, GFP_KERNEL);
+}
- flowcmd->clear = a[ODP_FLOW_ATTR_CLEAR] != NULL;
+static struct sk_buff *odp_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
+ u32 pid, u32 seq, u8 cmd)
+{
+ struct sk_buff *skb;
+ int retval;
- flowcmd->state = a[ODP_FLOW_ATTR_STATE] ? nla_get_u64(a[ODP_FLOW_ATTR_STATE]) : 0;
+ skb = odp_flow_cmd_alloc_info(flow);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
+ retval = odp_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
+ BUG_ON(retval < 0);
return skb;
-
-error_free_skb:
- kfree_skb(skb);
- return ERR_PTR(err);
}
-static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow)
+static int odp_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
{
+ struct nlattr **a = info->attrs;
+ struct odp_header *odp_header = info->userhdr;
struct tbl_node *flow_node;
- struct dp_flowcmd flowcmd;
+ struct sw_flow_key key;
struct sw_flow *flow;
- struct sk_buff *skb;
+ struct sk_buff *reply;
struct datapath *dp;
struct tbl *table;
u32 hash;
int error;
- skb = copy_flow_from_user(uodp_flow, &flowcmd);
- error = PTR_ERR(skb);
- if (IS_ERR(skb))
- goto exit;
+ /* Extract key. */
+ error = -EINVAL;
+ if (!a[ODP_FLOW_ATTR_KEY])
+ goto error;
+ error = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
+ if (error)
+ goto error;
- dp = get_dp(flowcmd.dp_idx);
+ /* Validate actions. */
+ if (a[ODP_FLOW_ATTR_ACTIONS]) {
+ error = validate_actions(a[ODP_FLOW_ATTR_ACTIONS]);
+ if (error)
+ goto error;
+ } else if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW) {
+ error = -EINVAL;
+ goto error;
+ }
+
+ dp = get_dp(odp_header->dp_ifindex);
error = -ENODEV;
if (!dp)
- goto exit;
+ goto error;
- hash = flow_hash(&flowcmd.key);
+ hash = flow_hash(&key);
table = get_table_protected(dp);
- flow_node = tbl_lookup(table, &flowcmd.key, hash, flow_cmp);
+ flow_node = tbl_lookup(table, &key, hash, flow_cmp);
if (!flow_node) {
struct sw_flow_actions *acts;
/* Bail out if we're not allowed to create a new flow. */
error = -ENOENT;
- if (cmd == ODP_FLOW_SET)
- goto exit;
+ if (info->genlhdr->cmd == ODP_FLOW_CMD_SET)
+ goto error;
/* Expand table, if necessary, to make room. */
if (tbl_count(table) >= tbl_n_buckets(table)) {
error = expand_table(dp);
if (error)
- goto exit;
+ goto error;
table = get_table_protected(dp);
}
flow = flow_alloc();
if (IS_ERR(flow)) {
error = PTR_ERR(flow);
- goto exit;
+ goto error;
}
- flow->key = flowcmd.key;
+ flow->key = key;
clear_stats(flow);
/* Obtain actions. */
- acts = get_actions(&flowcmd);
+ acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
error = PTR_ERR(acts);
if (IS_ERR(acts))
goto error_free_flow;
rcu_assign_pointer(flow->sf_acts, acts);
- error = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0);
- if (error)
- goto error_free_flow;
-
/* Put flow in bucket. */
error = tbl_insert(table, &flow->tbl_node, hash);
if (error)
goto error_free_flow;
+
+ reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
+ info->snd_seq, ODP_FLOW_CMD_NEW);
} else {
/* We found a matching flow. */
struct sw_flow_actions *old_acts;
* gets fixed.
*/
error = -EEXIST;
- if (flowcmd.nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
- goto error_kfree_skb;
+ if (info->genlhdr->cmd == ODP_FLOW_CMD_NEW &&
+ info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
+ goto error;
/* Update actions. */
flow = flow_cast(flow_node);
old_acts = rcu_dereference_protected(flow->sf_acts,
lockdep_genl_is_held());
- if (flowcmd.actions &&
- (old_acts->actions_len != flowcmd.actions_len ||
- memcmp(old_acts->actions, flowcmd.actions,
- flowcmd.actions_len))) {
+ if (a[ODP_FLOW_ATTR_ACTIONS] &&
+ (old_acts->actions_len != nla_len(a[ODP_FLOW_ATTR_ACTIONS]) ||
+ memcmp(old_acts->actions, nla_data(a[ODP_FLOW_ATTR_ACTIONS]),
+ old_acts->actions_len))) {
struct sw_flow_actions *new_acts;
- new_acts = get_actions(&flowcmd);
+ new_acts = flow_actions_alloc(a[ODP_FLOW_ATTR_ACTIONS]);
error = PTR_ERR(new_acts);
if (IS_ERR(new_acts))
- goto error_kfree_skb;
+ goto error;
rcu_assign_pointer(flow->sf_acts, new_acts);
flow_deferred_free_acts(old_acts);
}
- error = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0);
- if (error)
- goto error_kfree_skb;
+ reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid,
+ info->snd_seq, ODP_FLOW_CMD_NEW);
/* Clear stats. */
- if (flowcmd.clear) {
+ if (a[ODP_FLOW_ATTR_CLEAR]) {
spin_lock_bh(&flow->lock);
clear_stats(flow);
spin_unlock_bh(&flow->lock);
}
}
- kfree_skb(skb);
+
+ if (!IS_ERR(reply))
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
+ else
+ netlink_set_err(INIT_NET_GENL_SOCK, 0,
+ dp_flow_multicast_group.id, PTR_ERR(reply));
return 0;
error_free_flow:
flow_put(flow);
-error_kfree_skb:
- kfree_skb(skb);
-exit:
+error:
return error;
}
-static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow)
+static int odp_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
{
+ struct nlattr **a = info->attrs;
+ struct odp_header *odp_header = info->userhdr;
+ struct sw_flow_key key;
struct tbl_node *flow_node;
- struct dp_flowcmd flowcmd;
+ struct sk_buff *reply;
struct sw_flow *flow;
- struct sk_buff *skb;
struct datapath *dp;
struct tbl *table;
int err;
- skb = copy_flow_from_user(uodp_flow, &flowcmd);
- if (IS_ERR(skb))
- return PTR_ERR(skb);
+ if (!a[ODP_FLOW_ATTR_KEY])
+ return -EINVAL;
+ err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
+ if (err)
+ return err;
- dp = get_dp(flowcmd.dp_idx);
+ dp = get_dp(odp_header->dp_ifindex);
if (!dp)
return -ENODEV;
table = get_table_protected(dp);
- flow_node = tbl_lookup(table, &flowcmd.key, flow_hash(&flowcmd.key), flow_cmp);
+ flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
if (!flow_node)
return -ENOENT;
- if (cmd == ODP_FLOW_DEL) {
- err = tbl_remove(table, flow_node);
- if (err)
- return err;
- }
-
flow = flow_cast(flow_node);
- err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len, 0);
- if (!err && cmd == ODP_FLOW_DEL)
- flow_deferred_free(flow);
+ reply = odp_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, ODP_FLOW_CMD_NEW);
+ if (IS_ERR(reply))
+ return PTR_ERR(reply);
- return err;
+ return genlmsg_reply(reply, info);
}
-static int dump_flow(struct odp_flow __user *uodp_flow)
+static int odp_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
{
+ struct nlattr **a = info->attrs;
+ struct odp_header *odp_header = info->userhdr;
+ struct sw_flow_key key;
struct tbl_node *flow_node;
- struct dp_flowcmd flowcmd;
+ struct sk_buff *reply;
struct sw_flow *flow;
- struct sk_buff *skb;
struct datapath *dp;
- u32 bucket, obj;
+ struct tbl *table;
int err;
- skb = copy_flow_from_user(uodp_flow, &flowcmd);
- err = PTR_ERR(skb);
- if (IS_ERR(skb))
- goto exit;
+ if (!a[ODP_FLOW_ATTR_KEY])
+ return flush_flows(odp_header->dp_ifindex);
+ err = flow_from_nlattrs(&key, a[ODP_FLOW_ATTR_KEY]);
+ if (err)
+ return err;
- dp = get_dp(flowcmd.dp_idx);
- err = -ENODEV;
+ dp = get_dp(odp_header->dp_ifindex);
if (!dp)
- goto exit_kfree_skb;
+ return -ENODEV;
- bucket = flowcmd.state >> 32;
- obj = flowcmd.state;
- flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
- err = -ENODEV;
+ table = get_table_protected(dp);
+ flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
if (!flow_node)
- goto exit_kfree_skb;
-
+ return -ENOENT;
flow = flow_cast(flow_node);
- err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len,
- ((u64)bucket << 32) | obj);
-exit_kfree_skb:
- kfree_skb(skb);
-exit:
- return err;
+ reply = odp_flow_cmd_alloc_info(flow);
+ if (!reply)
+ return -ENOMEM;
+
+ err = tbl_remove(table, flow_node);
+ if (err) {
+ kfree_skb(reply);
+ return err;
+ }
+
+ err = odp_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
+ info->snd_seq, 0, ODP_FLOW_CMD_DEL);
+ BUG_ON(err < 0);
+
+ flow_deferred_free(flow);
+
+ genl_notify(reply, genl_info_net(info), info->snd_pid,
+ dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
+ return 0;
}
+static int odp_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
+{
+ struct odp_header *odp_header = genlmsg_data(nlmsg_data(cb->nlh));
+ struct datapath *dp;
+
+ dp = get_dp(odp_header->dp_ifindex);
+ if (!dp)
+ return -ENODEV;
+
+ for (;;) {
+ struct tbl_node *flow_node;
+ struct sw_flow *flow;
+ u32 bucket, obj;
+
+ bucket = cb->args[0];
+ obj = cb->args[1];
+ flow_node = tbl_next(get_table_protected(dp), &bucket, &obj);
+ if (!flow_node)
+ break;
+
+ flow = flow_cast(flow_node);
+ if (odp_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
+ cb->nlh->nlmsg_seq, NLM_F_MULTI,
+ ODP_FLOW_CMD_NEW) < 0)
+ break;
+
+ cb->args[0] = bucket;
+ cb->args[1] = obj;
+ }
+ return skb->len;
+}
+
+static struct genl_ops dp_flow_genl_ops[] = {
+ { .cmd = ODP_FLOW_CMD_NEW,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = flow_policy,
+ .doit = odp_flow_cmd_new_or_set
+ },
+ { .cmd = ODP_FLOW_CMD_DEL,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = flow_policy,
+ .doit = odp_flow_cmd_del
+ },
+ { .cmd = ODP_FLOW_CMD_GET,
+ .flags = 0, /* OK for unprivileged users. */
+ .policy = flow_policy,
+ .doit = odp_flow_cmd_get,
+ .dumpit = odp_flow_cmd_dump
+ },
+ { .cmd = ODP_FLOW_CMD_SET,
+ .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
+ .policy = flow_policy,
+ .doit = odp_flow_cmd_new_or_set,
+ },
+};
+
static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
#ifdef HAVE_NLA_NUL_STRING
[ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
if (!odp_header)
goto error;
- odp_header->dp_idx = dp->dp_idx;
+ odp_header->dp_ifindex = dp->dp_ifindex;
rcu_read_lock();
err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
return -EINVAL;
}
- return VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
+ return CHECK_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
}
/* Called with genl_mutex and optionally with RTNL lock also. */
static struct datapath *lookup_datapath(struct odp_header *odp_header, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
{
- if (!a[ODP_DP_ATTR_NAME]) {
- struct datapath *dp = get_dp(odp_header->dp_idx);
- if (!dp)
- return ERR_PTR(-ENODEV);
- return dp;
- } else {
+ struct datapath *dp;
+
+ if (!a[ODP_DP_ATTR_NAME])
+ dp = get_dp(odp_header->dp_ifindex);
+ else {
struct vport *vport;
- int dp_idx;
rcu_read_lock();
vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
- dp_idx = vport && vport->port_no == ODPP_LOCAL ? vport->dp->dp_idx : -1;
+ dp = vport && vport->port_no == ODPP_LOCAL ? vport->dp : NULL;
rcu_read_unlock();
-
- if (dp_idx < 0)
- return ERR_PTR(-ENODEV);
- return vport->dp;
}
+ return dp ? dp : ERR_PTR(-ENODEV);
}
/* Called with genl_mutex. */
static int odp_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
{
struct nlattr **a = info->attrs;
- struct odp_header *odp_header = info->userhdr;
struct vport_parms parms;
struct sk_buff *reply;
struct datapath *dp;
struct vport *vport;
- int dp_idx;
int err;
err = -EINVAL;
if (!try_module_get(THIS_MODULE))
goto err_unlock_rtnl;
- dp_idx = odp_header->dp_idx;
- if (dp_idx < 0) {
- err = -EFBIG;
- for (dp_idx = 0; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
- if (get_dp(dp_idx))
- continue;
- err = 0;
- break;
- }
- } else if (dp_idx < ARRAY_SIZE(dps))
- err = get_dp(dp_idx) ? -EBUSY : 0;
- else
- err = -EINVAL;
- if (err)
- goto err_put_module;
-
err = -ENOMEM;
dp = kzalloc(sizeof(*dp), GFP_KERNEL);
if (dp == NULL)
goto err_put_module;
INIT_LIST_HEAD(&dp->port_list);
- dp->dp_idx = dp_idx;
/* Initialize kobject for bridge. This will be added as
* /sys/class/net/<devname>/brif later, if sysfs is enabled. */
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 (IS_ERR(reply))
goto err_destroy_local_port;
- rcu_assign_pointer(dps[dp_idx], dp);
+ list_add_tail(&dp->list_node, &dps);
dp_sysfs_add_dp(dp);
rtnl_unlock();
dp_detach_port(vport);
dp_sysfs_del_dp(dp);
- rcu_assign_pointer(dps[dp->dp_idx], NULL);
+ list_del(&dp->list_node);
dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
+ /* rtnl_unlock() will wait until all the references to devices that
+ * are pending unregistration have been dropped. We do it here to
+ * ensure that any internal devices (which contain DP pointers) are
+ * fully destroyed before freeing the datapath.
+ */
+ rtnl_unlock();
+
call_rcu(&dp->rcu, destroy_dp_rcu);
module_put(THIS_MODULE);
genl_notify(reply, genl_info_net(info), info->snd_pid,
dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
- err = 0;
+
+ return 0;
exit_unlock:
rtnl_unlock();
static int odp_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
{
- u32 dp_idx;
+ struct datapath *dp;
+ int skip = cb->args[0];
+ int i = 0;
- for (dp_idx = cb->args[0]; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
- struct datapath *dp = get_dp(dp_idx);
- if (!dp)
+ list_for_each_entry (dp, &dps, list_node) {
+ if (i < skip)
continue;
if (odp_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
cb->nlh->nlmsg_seq, NLM_F_MULTI,
ODP_DP_CMD_NEW) < 0)
break;
+ i++;
}
- cb->args[0] = dp_idx;
+ cb->args[0] = i;
+
return skb->len;
}
struct odp_header *odp_header;
struct nlattr *nla;
int ifindex, iflink;
+ int mtu;
int err;
odp_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
if (!odp_header)
return -EMSGSIZE;
- odp_header->dp_idx = vport->dp->dp_idx;
+ odp_header->dp_ifindex = vport->dp->dp_ifindex;
NLA_PUT_U32(skb, ODP_VPORT_ATTR_PORT_NO, vport->port_no);
NLA_PUT_U32(skb, ODP_VPORT_ATTR_TYPE, vport_get_type(vport));
NLA_PUT(skb, ODP_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
- NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, vport_get_mtu(vport));
+ mtu = vport_get_mtu(vport);
+ if (mtu)
+ NLA_PUT_U32(skb, ODP_VPORT_ATTR_MTU, mtu);
err = vport_get_options(vport, skb);
if (err == -EMSGSIZE)
static int odp_vport_cmd_validate(struct nlattr *a[ODP_VPORT_ATTR_MAX + 1])
{
- return VERIFY_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
+ return CHECK_NUL_STRING(a[ODP_VPORT_ATTR_NAME], IFNAMSIZ - 1);
}
/* Called with RTNL lock or RCU read lock. */
if (port_no >= DP_MAX_PORTS)
return ERR_PTR(-EFBIG);
- dp = get_dp(odp_header->dp_idx);
+ dp = get_dp(odp_header->dp_ifindex);
if (!dp)
return ERR_PTR(-ENODEV);
goto exit;
rtnl_lock();
- dp = get_dp(odp_header->dp_idx);
+ dp = get_dp(odp_header->dp_ifindex);
err = -ENODEV;
if (!dp)
goto exit_unlock;
u32 port_no;
int retval;
- dp = get_dp(odp_header->dp_idx);
+ dp = get_dp(odp_header->dp_ifindex);
if (!dp)
return -ENODEV;
},
};
-static long openvswitch_ioctl(struct file *f, unsigned int cmd,
- unsigned long argp)
-{
- int err;
-
- genl_lock();
- switch (cmd) {
- case ODP_FLOW_FLUSH:
- err = flush_flows(argp);
- goto exit;
-
- case ODP_FLOW_NEW:
- case ODP_FLOW_SET:
- err = new_flow(cmd, (struct odp_flow __user *)argp);
- goto exit;
-
- case ODP_FLOW_GET:
- case ODP_FLOW_DEL:
- err = get_or_del_flow(cmd, (struct odp_flow __user *)argp);
- goto exit;
-
- case ODP_FLOW_DUMP:
- err = dump_flow((struct odp_flow __user *)argp);
- goto exit;
-
- default:
- err = -ENOIOCTLCMD;
- break;
- }
-exit:
- genl_unlock();
- return err;
-}
-
-#ifdef CONFIG_COMPAT
-static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
-{
- switch (cmd) {
- case ODP_FLOW_FLUSH:
- /* Ioctls that don't need any translation at all. */
- return openvswitch_ioctl(f, cmd, argp);
-
- case ODP_FLOW_NEW:
- case ODP_FLOW_DEL:
- case ODP_FLOW_GET:
- case ODP_FLOW_SET:
- case ODP_FLOW_DUMP:
- /* Ioctls that just need their pointer argument extended. */
- return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
-
- default:
- return -ENOIOCTLCMD;
- }
-}
-#endif
-
-static struct file_operations openvswitch_fops = {
- .owner = THIS_MODULE,
- .unlocked_ioctl = openvswitch_ioctl,
-#ifdef CONFIG_COMPAT
- .compat_ioctl = openvswitch_compat_ioctl,
-#endif
-};
-
-static int major;
-
struct genl_family_and_ops {
struct genl_family *family;
struct genl_ops *ops;
{ &dp_vport_genl_family,
dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
&dp_vport_multicast_group },
+ { &dp_flow_genl_family,
+ dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
+ &dp_flow_multicast_group },
{ &dp_packet_genl_family,
dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
NULL },
if (err)
goto error_vport_exit;
- major = register_chrdev(0, "openvswitch", &openvswitch_fops);
- if (err < 0)
- goto error_unreg_notifier;
-
err = dp_register_genl();
if (err < 0)
- goto error_unreg_chrdev;
+ goto error_unreg_notifier;
return 0;
-error_unreg_chrdev:
- unregister_chrdev(major, "openvswitch");
error_unreg_notifier:
unregister_netdevice_notifier(&dp_device_notifier);
error_vport_exit:
{
rcu_barrier();
dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
- unregister_chrdev(major, "openvswitch");
unregister_netdevice_notifier(&dp_device_notifier);
vport_exit();
flow_exit();
git://git.onelab.eu / sliver-openvswitch.git / blobdiff