#include "actions.h"
#include "flow.h"
#include "loop_counter.h"
-#include "odp-compat.h"
#include "table.h"
#include "vport-internal_dev.h"
* It is safe to access the datapath and vport structures with just
* dp_mutex.
*/
-static struct datapath __rcu *dps[ODP_MAX];
+static struct datapath __rcu *dps[256];
static DEFINE_MUTEX(dp_mutex);
static struct vport *new_vport(const struct vport_parms *);
/* Must be called with rcu_read_lock or dp_mutex. */
struct datapath *get_dp(int dp_idx)
{
- if (dp_idx < 0 || dp_idx >= ODP_MAX)
+ if (dp_idx < 0 || dp_idx >= ARRAY_SIZE(dps))
return NULL;
return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() ||
lockdep_is_held(&dp_mutex));
.release = release_dp
};
-static int create_dp(int dp_idx, const char __user *devnamep)
-{
- struct vport_parms parms;
- char devname[IFNAMSIZ];
- struct vport *vport;
- struct datapath *dp;
- int err;
- int i;
-
- if (devnamep) {
- int retval = strncpy_from_user(devname, devnamep, IFNAMSIZ);
- if (retval < 0) {
- err = -EFAULT;
- goto err;
- } else if (retval >= IFNAMSIZ) {
- err = -ENAMETOOLONG;
- goto err;
- }
- } else {
- snprintf(devname, sizeof(devname), "of%d", dp_idx);
- }
-
- rtnl_lock();
- mutex_lock(&dp_mutex);
- err = -ENODEV;
- if (!try_module_get(THIS_MODULE))
- goto err_unlock;
-
- /* Exit early if a datapath with that number already exists.
- * (We don't use -EEXIST because that's ambiguous with 'devname'
- * conflicting with an existing network device name.) */
- err = -EBUSY;
- if (get_dp(dp_idx))
- 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);
- mutex_init(&dp->mutex);
- mutex_lock(&dp->mutex);
- dp->dp_idx = dp_idx;
- for (i = 0; i < DP_N_QUEUES; i++)
- skb_queue_head_init(&dp->queues[i]);
- init_waitqueue_head(&dp->waitqueue);
-
- /* Initialize kobject for bridge. This will be added as
- * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
- dp->ifobj.kset = NULL;
- kobject_init(&dp->ifobj, &dp_ktype);
-
- /* Allocate table. */
- err = -ENOMEM;
- rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
- if (!dp->table)
- goto err_free_dp;
-
- /* Set up our datapath device. */
- parms.name = devname;
- parms.type = ODP_VPORT_TYPE_INTERNAL;
- parms.options = NULL;
- parms.dp = dp;
- parms.port_no = ODPP_LOCAL;
- vport = new_vport(&parms);
- if (IS_ERR(vport)) {
- err = PTR_ERR(vport);
- if (err == -EBUSY)
- err = -EEXIST;
-
- goto err_destroy_table;
- }
-
- dp->drop_frags = 0;
- dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
- if (!dp->stats_percpu) {
- err = -ENOMEM;
- goto err_destroy_local_port;
- }
-
- rcu_assign_pointer(dps[dp_idx], dp);
- dp_sysfs_add_dp(dp);
-
- mutex_unlock(&dp->mutex);
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
-
- return 0;
-
-err_destroy_local_port:
- dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
-err_destroy_table:
- tbl_destroy(get_table_protected(dp), NULL);
-err_free_dp:
- mutex_unlock(&dp->mutex);
- kfree(dp);
-err_put_module:
- module_put(THIS_MODULE);
-err_unlock:
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
-err:
- return err;
-}
-
static void destroy_dp_rcu(struct rcu_head *rcu)
{
struct datapath *dp = container_of(rcu, struct datapath, rcu);
kobject_put(&dp->ifobj);
}
-static int destroy_dp(int dp_idx)
+/* Caller must hold RTNL, dp_mutex, and dp->mutex. */
+static void destroy_dp(struct datapath *dp)
{
- struct datapath *dp;
- int err = 0;
struct vport *p, *n;
- rtnl_lock();
- mutex_lock(&dp_mutex);
- dp = get_dp(dp_idx);
- if (!dp) {
- err = -ENODEV;
- goto out;
- }
-
- mutex_lock(&dp->mutex);
-
list_for_each_entry_safe (p, n, &dp->port_list, node)
if (p->port_no != ODPP_LOCAL)
dp_detach_port(p);
mutex_unlock(&dp->mutex);
call_rcu(&dp->rcu, destroy_dp_rcu);
module_put(THIS_MODULE);
-
-out:
- mutex_unlock(&dp_mutex);
- rtnl_unlock();
- return err;
}
/* Called with RTNL lock and dp->mutex. */
bool is_frag;
/* Extract flow from 'skb' into 'key'. */
- error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key, &is_frag);
+ error = flow_extract(skb, p->port_no, &key, &is_frag);
if (unlikely(error)) {
kfree_skb(skb);
return;
return err;
}
-static int flush_flows(struct datapath *dp)
+static int flush_flows(int dp_idx)
{
- struct tbl *old_table = get_table_protected(dp);
+ struct tbl *old_table;
struct tbl *new_table;
+ struct datapath *dp;
+ int err;
+
+ dp = get_dp_locked(dp_idx);
+ err = -ENODEV;
+ if (!dp)
+ goto exit;
+ old_table = get_table_protected(dp);
new_table = tbl_create(TBL_MIN_BUCKETS);
+ err = -ENOMEM;
if (!new_table)
- return -ENOMEM;
+ goto exit_unlock;
rcu_assign_pointer(dp->table, new_table);
tbl_deferred_destroy(old_table, flow_free_tbl);
- return 0;
+ err = 0;
+
+exit_unlock:
+ mutex_unlock(&dp->mutex);
+exit:
+ return err;
}
static int validate_actions(const struct nlattr *actions, u32 actions_len)
return 0;
}
-static struct sw_flow_actions *get_actions(const struct odp_flow *flow)
+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;
- int error;
-
- actions = flow_actions_alloc(flow->actions_len);
- error = PTR_ERR(actions);
- if (IS_ERR(actions))
- goto error;
-
- error = -EFAULT;
- if (copy_from_user(actions->actions,
- (struct nlattr __user __force *)flow->actions,
- flow->actions_len))
- goto error_free_actions;
- error = validate_actions(actions->actions, actions->actions_len);
- if (error)
- goto error_free_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;
-
-error_free_actions:
- kfree(actions);
-error:
- return ERR_PTR(error);
-}
-
-static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats)
-{
- if (flow->used) {
- struct timespec offset_ts, used, now_mono;
-
- ktime_get_ts(&now_mono);
- jiffies_to_timespec(jiffies - flow->used, &offset_ts);
- set_normalized_timespec(&used, now_mono.tv_sec - offset_ts.tv_sec,
- now_mono.tv_nsec - offset_ts.tv_nsec);
-
- stats->used_sec = used.tv_sec;
- stats->used_nsec = used.tv_nsec;
- } else {
- stats->used_sec = 0;
- stats->used_nsec = 0;
- }
-
- stats->n_packets = flow->packet_count;
- stats->n_bytes = flow->byte_count;
- stats->reserved = 0;
- stats->tcp_flags = flow->tcp_flags;
}
static void clear_stats(struct sw_flow *flow)
rcu_assign_pointer(dp->table, new_table);
tbl_deferred_destroy(old_table, NULL);
- return 0;
+ return 0;
}
-static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf,
- struct odp_flow_stats *stats)
+static const struct nla_policy execute_policy[ODP_PACKET_ATTR_MAX + 1] = {
+ [ODP_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
+ [ODP_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
+};
+
+static int execute_packet(const struct odp_packet __user *uodp_packet)
{
- struct tbl_node *flow_node;
+ struct nlattr *a[ODP_PACKET_ATTR_MAX + 1];
+ struct odp_packet *odp_packet;
+ struct sk_buff *skb, *packet;
+ unsigned int actions_len;
+ struct nlattr *actions;
struct sw_flow_key key;
- struct sw_flow *flow;
- struct tbl *table;
- struct sw_flow_actions *acts = NULL;
- int error;
- u32 hash;
-
- error = flow_copy_from_user(&key, (const struct nlattr __force __user *)uf->flow.key,
- uf->flow.key_len);
- if (error)
- return error;
+ struct datapath *dp;
+ struct ethhdr *eth;
+ bool is_frag;
+ u32 len;
+ int err;
- hash = flow_hash(&key);
- table = get_table_protected(dp);
- flow_node = tbl_lookup(table, &key, hash, flow_cmp);
- if (!flow_node) {
- /* No such flow. */
- error = -ENOENT;
- if (!(uf->flags & ODPPF_CREATE))
- goto error;
+ if (get_user(len, &uodp_packet->len))
+ return -EFAULT;
+ if (len < sizeof(struct odp_packet))
+ return -EINVAL;
- /* Expand table, if necessary, to make room. */
- if (tbl_count(table) >= tbl_n_buckets(table)) {
- error = expand_table(dp);
- if (error)
- goto error;
- table = get_table_protected(dp);
- }
+ skb = alloc_skb(len, GFP_KERNEL);
+ if (!skb)
+ return -ENOMEM;
- /* Allocate flow. */
- flow = flow_alloc();
- if (IS_ERR(flow)) {
- error = PTR_ERR(flow);
- goto error;
- }
- flow->key = key;
- clear_stats(flow);
+ err = -EFAULT;
+ if (copy_from_user(__skb_put(skb, len), uodp_packet, len))
+ goto exit_free_skb;
- /* Obtain actions. */
- acts = get_actions(&uf->flow);
- error = PTR_ERR(acts);
- if (IS_ERR(acts))
- goto error_free_flow;
- rcu_assign_pointer(flow->sf_acts, acts);
+ odp_packet = (struct odp_packet *)skb->data;
+ err = -EINVAL;
+ if (odp_packet->len != len)
+ goto exit_free_skb;
- /* Put flow in bucket. */
- error = tbl_insert(table, &flow->tbl_node, hash);
- if (error)
- goto error_free_flow_acts;
+ __skb_pull(skb, sizeof(struct odp_packet));
+ err = nla_parse(a, ODP_PACKET_ATTR_MAX, (struct nlattr *)skb->data,
+ skb->len, execute_policy);
+ if (err)
+ goto exit_free_skb;
- memset(stats, 0, sizeof(struct odp_flow_stats));
- } else {
- /* We found a matching flow. */
- struct sw_flow_actions *old_acts, *new_acts;
+ err = -EINVAL;
+ if (!a[ODP_PACKET_ATTR_PACKET] || !a[ODP_PACKET_ATTR_ACTIONS] ||
+ nla_len(a[ODP_PACKET_ATTR_PACKET]) < ETH_HLEN)
+ goto exit_free_skb;
- flow = flow_cast(flow_node);
+ actions = nla_data(a[ODP_PACKET_ATTR_ACTIONS]);
+ actions_len = nla_len(a[ODP_PACKET_ATTR_ACTIONS]);
+ err = validate_actions(actions, actions_len);
+ if (err)
+ goto exit_free_skb;
- /* Bail out if we're not allowed to modify an existing flow. */
- error = -EEXIST;
- if (!(uf->flags & ODPPF_MODIFY))
- goto error;
+ packet = skb_clone(skb, GFP_KERNEL);
+ err = -ENOMEM;
+ if (!packet)
+ goto exit_free_skb;
+ packet->data = nla_data(a[ODP_PACKET_ATTR_PACKET]);
+ packet->len = nla_len(a[ODP_PACKET_ATTR_PACKET]);
- /* Swap actions. */
- new_acts = get_actions(&uf->flow);
- error = PTR_ERR(new_acts);
- if (IS_ERR(new_acts))
- goto error;
+ skb_reset_mac_header(packet);
+ eth = eth_hdr(packet);
- old_acts = rcu_dereference_protected(flow->sf_acts,
- lockdep_is_held(&dp->mutex));
- if (old_acts->actions_len != new_acts->actions_len ||
- memcmp(old_acts->actions, new_acts->actions,
- old_acts->actions_len)) {
- rcu_assign_pointer(flow->sf_acts, new_acts);
- flow_deferred_free_acts(old_acts);
- } else {
- kfree(new_acts);
- }
+ /* Normally, setting the skb 'protocol' field would be handled by a
+ * call to eth_type_trans(), but it assumes there's a sending
+ * device, which we may not have. */
+ if (ntohs(eth->h_proto) >= 1536)
+ packet->protocol = eth->h_proto;
+ else
+ packet->protocol = htons(ETH_P_802_2);
- /* Fetch stats, then clear them if necessary. */
- spin_lock_bh(&flow->lock);
- get_stats(flow, stats);
- if (uf->flags & ODPPF_ZERO_STATS)
- clear_stats(flow);
- spin_unlock_bh(&flow->lock);
- }
+ err = flow_extract(packet, -1, &key, &is_frag);
+ if (err)
+ goto exit_free_skb;
- return 0;
+ rcu_read_lock();
+ dp = get_dp(odp_packet->dp_idx);
+ err = -ENODEV;
+ if (dp)
+ err = execute_actions(dp, packet, &key, actions, actions_len);
+ rcu_read_unlock();
-error_free_flow_acts:
- kfree(acts);
-error_free_flow:
- flow->sf_acts = NULL;
- flow_put(flow);
-error:
- return error;
+exit_free_skb:
+ kfree_skb(skb);
+ return err;
}
-static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp)
+static void get_dp_stats(struct datapath *dp, struct odp_stats *stats)
{
- struct odp_flow_stats stats;
- struct odp_flow_put uf;
- int error;
+ int i;
- if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put)))
- return -EFAULT;
+ stats->n_frags = stats->n_hit = stats->n_missed = stats->n_lost = 0;
+ for_each_possible_cpu(i) {
+ const struct dp_stats_percpu *percpu_stats;
+ struct dp_stats_percpu local_stats;
+ unsigned seqcount;
- error = do_put_flow(dp, &uf, &stats);
- if (error)
- return error;
+ percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
- if (copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
- return -EFAULT;
+ do {
+ seqcount = read_seqcount_begin(&percpu_stats->seqlock);
+ local_stats = *percpu_stats;
+ } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
- return 0;
+ stats->n_frags += local_stats.n_frags;
+ stats->n_hit += local_stats.n_hit;
+ stats->n_missed += local_stats.n_missed;
+ stats->n_lost += local_stats.n_lost;
+ }
}
-static int do_answer_query(struct datapath *dp, struct sw_flow *flow,
- u32 query_flags,
- struct odp_flow_stats __user *ustats,
- struct nlattr __user *actions,
- u32 __user *actions_lenp)
+/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
+int dp_min_mtu(const struct datapath *dp)
{
- struct sw_flow_actions *sf_acts;
- struct odp_flow_stats stats;
- u32 actions_len;
+ struct vport *p;
+ int mtu = 0;
- spin_lock_bh(&flow->lock);
- get_stats(flow, &stats);
- if (query_flags & ODPFF_ZERO_TCP_FLAGS)
- flow->tcp_flags = 0;
+ ASSERT_RTNL();
- spin_unlock_bh(&flow->lock);
+ list_for_each_entry_rcu (p, &dp->port_list, node) {
+ int dev_mtu;
- if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) ||
- get_user(actions_len, actions_lenp))
- return -EFAULT;
+ /* Skip any internal ports, since that's what we're trying to
+ * set. */
+ if (is_internal_vport(p))
+ continue;
- if (!actions_len)
- return 0;
+ dev_mtu = vport_get_mtu(p);
+ if (!mtu || dev_mtu < mtu)
+ mtu = dev_mtu;
+ }
- sf_acts = rcu_dereference_protected(flow->sf_acts,
- lockdep_is_held(&dp->mutex));
- if (put_user(sf_acts->actions_len, actions_lenp) ||
- (actions && copy_to_user(actions, sf_acts->actions,
- min(sf_acts->actions_len, actions_len))))
- return -EFAULT;
+ return mtu ? mtu : ETH_DATA_LEN;
+}
- return 0;
+/* Sets the MTU of all datapath devices to the minimum of the ports. Must
+ * be 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_rcu (p, &dp->port_list, node) {
+ if (is_internal_vport(p))
+ vport_set_mtu(p, mtu);
+ }
}
-static int answer_query(struct datapath *dp, struct sw_flow *flow,
- u32 query_flags, struct odp_flow __user *ufp)
+static int get_listen_mask(const struct file *f)
{
- struct nlattr __user *actions;
+ return (long)f->private_data;
+}
- if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions))
- return -EFAULT;
+static void set_listen_mask(struct file *f, int listen_mask)
+{
+ f->private_data = (void*)(long)listen_mask;
+}
+
+static const struct nla_policy flow_policy[ODP_FLOW_ATTR_MAX + 1] = {
+ [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 int copy_flow_to_user(struct odp_flow __user *dst, struct datapath *dp,
+ struct sw_flow *flow, u32 total_len, u64 state)
+{
+ const struct sw_flow_actions *sf_acts;
+ struct odp_flow_stats stats;
+ struct odp_flow *odp_flow;
+ struct sk_buff *skb;
+ struct nlattr *nla;
+ unsigned long used;
+ u8 tcp_flags;
+ int err;
+
+ sf_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_is_held(&dp->mutex));
+
+ skb = alloc_skb(128 + FLOW_BUFSIZE + sf_acts->actions_len, GFP_KERNEL);
+ err = -ENOMEM;
+ if (!skb)
+ goto exit;
- return do_answer_query(dp, flow, query_flags,
- &ufp->stats, actions, &ufp->actions_len);
+ rcu_read_lock();
+ 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;
+
+ 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_unlock;
+ 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);
+ nla_nest_end(skb, nla);
+
+ spin_lock_bh(&flow->lock);
+ used = flow->used;
+ stats.n_packets = flow->packet_count;
+ stats.n_bytes = flow->byte_count;
+ tcp_flags = flow->tcp_flags;
+ spin_unlock_bh(&flow->lock);
+
+ if (used)
+ NLA_PUT_MSECS(skb, ODP_FLOW_ATTR_USED, 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;
+
+ odp_flow->len = skb->len;
+ err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0;
+ goto exit_unlock;
+
+nla_put_failure:
+ err = -EMSGSIZE;
+exit_unlock:
+ rcu_read_unlock();
+ kfree_skb(skb);
+exit:
+ return err;
}
-static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len)
+static struct sk_buff *copy_flow_from_user(struct odp_flow __user *uodp_flow,
+ struct dp_flowcmd *flowcmd)
{
- struct tbl *table = get_table_protected(dp);
- struct tbl_node *flow_node;
- struct sw_flow_key swkey;
- int error;
+ 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);
- error = flow_copy_from_user(&swkey, key, key_len);
- if (error)
- return ERR_PTR(error);
+ skb = alloc_skb(len, GFP_KERNEL);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
- flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp);
- if (!flow_node)
- return ERR_PTR(-ENOENT);
+ 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;
+
+ /* 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));
- error = tbl_remove(table, flow_node);
- if (error)
- return ERR_PTR(error);
+ /* 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;
+ }
- /* XXX Returned flow_node's statistics might lose a few packets, since
- * other CPUs can be using this flow. We used to synchronize_rcu() to
- * make sure that we get completely accurate stats, but that blows our
- * performance, badly. */
- return flow_cast(flow_node);
+ flowcmd->clear = a[ODP_FLOW_ATTR_CLEAR] != NULL;
+
+ flowcmd->state = a[ODP_FLOW_ATTR_STATE] ? nla_get_u64(a[ODP_FLOW_ATTR_STATE]) : 0;
+
+ return skb;
+
+error_free_skb:
+ kfree_skb(skb);
+ return ERR_PTR(err);
}
-static int del_flow(struct datapath *dp, struct odp_flow __user *ufp)
+static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow)
{
+ struct tbl_node *flow_node;
+ struct dp_flowcmd flowcmd;
struct sw_flow *flow;
- struct odp_flow uf;
+ struct sk_buff *skb;
+ struct datapath *dp;
+ struct tbl *table;
+ u32 hash;
int error;
- if (copy_from_user(&uf, ufp, sizeof(uf)))
- return -EFAULT;
+ skb = copy_flow_from_user(uodp_flow, &flowcmd);
+ error = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
- flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len);
- if (IS_ERR(flow))
- return PTR_ERR(flow);
+ dp = get_dp_locked(flowcmd.dp_idx);
+ error = -ENODEV;
+ if (!dp)
+ goto error_kfree_skb;
- error = answer_query(dp, flow, 0, ufp);
- flow_deferred_free(flow);
+ hash = flow_hash(&flowcmd.key);
+ table = get_table_protected(dp);
+ flow_node = tbl_lookup(table, &flowcmd.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 error_unlock_dp;
+
+ /* Expand table, if necessary, to make room. */
+ if (tbl_count(table) >= tbl_n_buckets(table)) {
+ error = expand_table(dp);
+ if (error)
+ goto error_unlock_dp;
+ table = get_table_protected(dp);
+ }
+
+ /* Allocate flow. */
+ flow = flow_alloc();
+ if (IS_ERR(flow)) {
+ error = PTR_ERR(flow);
+ goto error_unlock_dp;
+ }
+ flow->key = flowcmd.key;
+ clear_stats(flow);
+
+ /* Obtain actions. */
+ acts = get_actions(&flowcmd);
+ 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;
+ } else {
+ /* We found a matching flow. */
+ struct sw_flow_actions *old_acts;
+
+ /* Bail out if we're not allowed to modify an existing flow.
+ * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
+ * because Generic Netlink treats the latter as a dump
+ * request. We also accept NLM_F_EXCL in case that bug ever
+ * gets fixed.
+ */
+ error = -EEXIST;
+ if (flowcmd.nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
+ goto error_kfree_skb;
+
+ /* Update actions. */
+ flow = flow_cast(flow_node);
+ old_acts = rcu_dereference_protected(flow->sf_acts,
+ lockdep_is_held(&dp->mutex));
+ if (flowcmd.actions &&
+ (old_acts->actions_len != flowcmd.actions_len ||
+ memcmp(old_acts->actions, flowcmd.actions,
+ flowcmd.actions_len))) {
+ struct sw_flow_actions *new_acts;
+
+ new_acts = get_actions(&flowcmd);
+ error = PTR_ERR(new_acts);
+ if (IS_ERR(new_acts))
+ goto error_kfree_skb;
+
+ 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;
+
+ /* Clear stats. */
+ if (flowcmd.clear) {
+ spin_lock_bh(&flow->lock);
+ clear_stats(flow);
+ spin_unlock_bh(&flow->lock);
+ }
+ }
+ kfree_skb(skb);
+ mutex_unlock(&dp->mutex);
+ return 0;
+
+error_free_flow:
+ flow_put(flow);
+error_unlock_dp:
+ mutex_unlock(&dp->mutex);
+error_kfree_skb:
+ kfree_skb(skb);
+exit:
return error;
}
-static int query_flow(struct datapath *dp, struct odp_flow __user *uflow)
+static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow)
{
- struct tbl *table = get_table_protected(dp);
struct tbl_node *flow_node;
- struct sw_flow_key key;
- struct odp_flow flow;
- int error;
+ struct dp_flowcmd flowcmd;
+ struct sw_flow *flow;
+ struct sk_buff *skb;
+ struct datapath *dp;
+ struct tbl *table;
+ int err;
- if (copy_from_user(&flow, uflow, sizeof(flow)))
- return -EFAULT;
+ skb = copy_flow_from_user(uodp_flow, &flowcmd);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
- error = flow_copy_from_user(&key, (const struct nlattr __force __user *)flow.key, flow.key_len);
- if (error)
- return error;
+ dp = get_dp_locked(flowcmd.dp_idx);
+ err = -ENODEV;
+ if (!dp)
+ goto exit_kfree_skb;
- flow_node = tbl_lookup(table, &flow.key, flow_hash(&key), flow_cmp);
+ table = get_table_protected(dp);
+ flow_node = tbl_lookup(table, &flowcmd.key, flow_hash(&flowcmd.key), flow_cmp);
+ err = -ENOENT;
if (!flow_node)
- return -ENOENT;
- return answer_query(dp, flow_cast(flow_node), flow.flags, uflow);
+ goto exit_unlock_dp;
+
+ if (cmd == ODP_FLOW_DEL) {
+ err = tbl_remove(table, flow_node);
+ if (err)
+ goto exit_unlock_dp;
+ }
+
+ 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);
+
+exit_unlock_dp:
+ mutex_unlock(&dp->mutex);
+exit_kfree_skb:
+ kfree_skb(skb);
+exit:
+ return err;
}
-static struct sw_flow *do_dump_flow(struct datapath *dp, u32 __user *state)
+static int dump_flow(struct odp_flow __user *uodp_flow)
{
- struct tbl *table = get_table_protected(dp);
- struct tbl_node *tbl_node;
+ struct tbl_node *flow_node;
+ struct dp_flowcmd flowcmd;
+ struct sw_flow *flow;
+ struct sk_buff *skb;
+ struct datapath *dp;
u32 bucket, obj;
+ int err;
- if (get_user(bucket, &state[0]) || get_user(obj, &state[1]))
- return ERR_PTR(-EFAULT);
+ skb = copy_flow_from_user(uodp_flow, &flowcmd);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
- tbl_node = tbl_next(table, &bucket, &obj);
+ dp = get_dp_locked(flowcmd.dp_idx);
+ err = -ENODEV;
+ if (!dp)
+ goto exit_free;
- if (put_user(bucket, &state[0]) || put_user(obj, &state[1]))
- return ERR_PTR(-EFAULT);
+ bucket = flowcmd.state >> 32;
+ obj = flowcmd.state;
+ flow_node = tbl_next(dp->table, &bucket, &obj);
+ err = -ENODEV;
+ if (!flow_node)
+ goto exit_unlock_dp;
+
+ flow = flow_cast(flow_node);
+ err = copy_flow_to_user(uodp_flow, dp, flow, flowcmd.total_len,
+ ((u64)bucket << 32) | obj);
- return tbl_node ? flow_cast(tbl_node) : NULL;
+exit_unlock_dp:
+ mutex_unlock(&dp->mutex);
+exit_free:
+ kfree_skb(skb);
+exit:
+ return err;
}
-static int dump_flow(struct datapath *dp, struct odp_flow_dump __user *udumpp)
+static const struct nla_policy datapath_policy[ODP_DP_ATTR_MAX + 1] = {
+ [ODP_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
+ [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NLA_U32 },
+ [ODP_DP_ATTR_SAMPLING] = { .type = NLA_U32 },
+};
+
+static int copy_datapath_to_user(void __user *dst, struct datapath *dp, uint32_t total_len)
{
- struct odp_flow __user *uflowp;
- struct nlattr __user *ukey;
- struct sw_flow *flow;
- u32 key_len;
+ struct odp_datapath *odp_datapath;
+ struct sk_buff *skb;
+ struct nlattr *nla;
+ int err;
- flow = do_dump_flow(dp, udumpp->state);
- if (IS_ERR(flow))
- return PTR_ERR(flow);
+ skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL);
+ err = -ENOMEM;
+ if (!skb)
+ goto exit;
- if (get_user(uflowp, (struct odp_flow __user *__user*)&udumpp->flow))
- return -EFAULT;
+ odp_datapath = (struct odp_datapath*)__skb_put(skb, sizeof(struct odp_datapath));
+ odp_datapath->dp_idx = dp->dp_idx;
+ odp_datapath->total_len = total_len;
- if (!flow)
- return put_user(ODPFF_EOF, &uflowp->flags);
+ rcu_read_lock();
+ err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp));
+ rcu_read_unlock();
+ if (err)
+ goto nla_put_failure;
- if (put_user(0, &uflowp->flags) ||
- get_user(ukey, (struct nlattr __user * __user*)&uflowp->key) ||
- get_user(key_len, &uflowp->key_len))
- return -EFAULT;
+ nla = nla_reserve(skb, ODP_DP_ATTR_STATS, sizeof(struct odp_stats));
+ if (!nla)
+ goto nla_put_failure;
+ get_dp_stats(dp, nla_data(nla));
- key_len = flow_copy_to_user(ukey, &flow->key, key_len);
- if (key_len < 0)
- return key_len;
- if (put_user(key_len, &uflowp->key_len))
- return -EFAULT;
+ NLA_PUT_U32(skb, ODP_DP_ATTR_IPV4_FRAGS,
+ dp->drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO);
+
+ if (dp->sflow_probability)
+ NLA_PUT_U32(skb, ODP_DP_ATTR_SAMPLING, dp->sflow_probability);
+
+ if (skb->len > total_len)
+ goto nla_put_failure;
+
+ odp_datapath->len = skb->len;
+ err = copy_to_user(dst, skb->data, skb->len) ? -EFAULT : 0;
+ goto exit_free_skb;
- return answer_query(dp, flow, 0, uflowp);
+nla_put_failure:
+ err = -EMSGSIZE;
+exit_free_skb:
+ kfree_skb(skb);
+exit:
+ return err;
}
-static int do_execute(struct datapath *dp, const struct odp_execute *execute)
+static struct sk_buff *copy_datapath_from_user(struct odp_datapath __user *uodp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
{
- struct sw_flow_key key;
+ struct odp_datapath *odp_datapath;
struct sk_buff *skb;
- struct sw_flow_actions *actions;
- struct ethhdr *eth;
- bool is_frag;
+ u32 len;
int err;
- err = -EINVAL;
- if (execute->length < ETH_HLEN || execute->length > 65535)
- goto error;
+ if (get_user(len, &uodp_datapath->len))
+ return ERR_PTR(-EFAULT);
+ if (len < sizeof(struct odp_datapath))
+ return ERR_PTR(-EINVAL);
- actions = flow_actions_alloc(execute->actions_len);
- if (IS_ERR(actions)) {
- err = PTR_ERR(actions);
- goto error;
- }
+ skb = alloc_skb(len, GFP_KERNEL);
+ if (!skb)
+ return ERR_PTR(-ENOMEM);
err = -EFAULT;
- if (copy_from_user(actions->actions,
- (struct nlattr __user __force *)execute->actions, execute->actions_len))
- goto error_free_actions;
+ if (copy_from_user(__skb_put(skb, len), uodp_datapath, len))
+ goto error_free_skb;
- err = validate_actions(actions->actions, execute->actions_len);
+ odp_datapath = (struct odp_datapath *)skb->data;
+ err = -EINVAL;
+ if (odp_datapath->len != len)
+ goto error_free_skb;
+
+ err = nla_parse(a, ODP_DP_ATTR_MAX,
+ (struct nlattr *)(skb->data + sizeof(struct odp_datapath)),
+ skb->len - sizeof(struct odp_datapath), datapath_policy);
if (err)
- goto error_free_actions;
+ goto error_free_skb;
- err = -ENOMEM;
- skb = alloc_skb(execute->length, GFP_KERNEL);
- if (!skb)
- goto error_free_actions;
+ if (a[ODP_DP_ATTR_IPV4_FRAGS]) {
+ u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]);
- err = -EFAULT;
- if (copy_from_user(skb_put(skb, execute->length),
- (const void __user __force *)execute->data,
- execute->length))
+ err = -EINVAL;
+ if (frags != ODP_DP_FRAG_ZERO && frags != ODP_DP_FRAG_DROP)
+ goto error_free_skb;
+ }
+
+ err = VERIFY_NUL_STRING(a[ODP_DP_ATTR_NAME], IFNAMSIZ - 1);
+ if (err)
goto error_free_skb;
- skb_reset_mac_header(skb);
- eth = eth_hdr(skb);
+ return skb;
- /* Normally, setting the skb 'protocol' field would be handled by a
- * call to eth_type_trans(), but it assumes there's a sending
- * device, which we may not have. */
- if (ntohs(eth->h_proto) >= 1536)
- skb->protocol = eth->h_proto;
- else
- skb->protocol = htons(ETH_P_802_2);
+error_free_skb:
+ kfree_skb(skb);
+ return ERR_PTR(err);
+}
+
+/* Called with dp_mutex and optionally with RTNL lock also.
+ * Holds the returned datapath's mutex on return.
+ */
+static struct datapath *lookup_datapath(struct odp_datapath *odp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
+{
+ WARN_ON_ONCE(!mutex_is_locked(&dp_mutex));
+
+ if (!a[ODP_DP_ATTR_NAME]) {
+ struct datapath *dp;
+
+ dp = get_dp(odp_datapath->dp_idx);
+ if (!dp)
+ return ERR_PTR(-ENODEV);
+ mutex_lock(&dp->mutex);
+ return dp;
+ } else {
+ struct datapath *dp;
+ struct vport *vport;
+ int dp_idx;
+
+ vport_lock();
+ vport = vport_locate(nla_data(a[ODP_DP_ATTR_NAME]));
+ dp_idx = vport && vport->port_no == ODPP_LOCAL ? vport->dp->dp_idx : -1;
+ vport_unlock();
+
+ if (dp_idx < 0)
+ return ERR_PTR(-ENODEV);
+
+ dp = get_dp(dp_idx);
+ mutex_lock(&dp->mutex);
+ return dp;
+ }
+}
- err = flow_extract(skb, -1, &key, &is_frag);
+static void change_datapath(struct datapath *dp, struct nlattr *a[ODP_DP_ATTR_MAX + 1])
+{
+ if (a[ODP_DP_ATTR_IPV4_FRAGS])
+ dp->drop_frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]) == ODP_DP_FRAG_DROP;
+ if (a[ODP_DP_ATTR_SAMPLING])
+ dp->sflow_probability = nla_get_u32(a[ODP_DP_ATTR_SAMPLING]);
+}
+
+static int new_datapath(struct odp_datapath __user *uodp_datapath)
+{
+ struct nlattr *a[ODP_DP_ATTR_MAX + 1];
+ struct odp_datapath *odp_datapath;
+ struct vport_parms parms;
+ struct sk_buff *skb;
+ struct datapath *dp;
+ struct vport *vport;
+ int dp_idx;
+ int err;
+ int i;
+
+ skb = copy_datapath_from_user(uodp_datapath, a);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto err;
+ odp_datapath = (struct odp_datapath *)skb->data;
+
+ err = -EINVAL;
+ if (!a[ODP_DP_ATTR_NAME])
+ goto err_free_skb;
+
+ rtnl_lock();
+ mutex_lock(&dp_mutex);
+ err = -ENODEV;
+ if (!try_module_get(THIS_MODULE))
+ goto err_unlock_dp_mutex;
+
+ dp_idx = odp_datapath->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 error_free_skb;
+ goto err_put_module;
- rcu_read_lock();
- err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len);
- rcu_read_unlock();
+ err = -ENOMEM;
+ dp = kzalloc(sizeof(*dp), GFP_KERNEL);
+ if (dp == NULL)
+ goto err_put_module;
+ INIT_LIST_HEAD(&dp->port_list);
+ mutex_init(&dp->mutex);
+ mutex_lock(&dp->mutex);
+ dp->dp_idx = dp_idx;
+ for (i = 0; i < DP_N_QUEUES; i++)
+ skb_queue_head_init(&dp->queues[i]);
+ init_waitqueue_head(&dp->waitqueue);
+
+ /* Initialize kobject for bridge. This will be added as
+ * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
+ dp->ifobj.kset = NULL;
+ kobject_init(&dp->ifobj, &dp_ktype);
+
+ /* Allocate table. */
+ err = -ENOMEM;
+ rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS));
+ if (!dp->table)
+ goto err_free_dp;
+
+ /* Set up our datapath device. */
+ parms.name = nla_data(a[ODP_DP_ATTR_NAME]);
+ parms.type = ODP_VPORT_TYPE_INTERNAL;
+ parms.options = NULL;
+ parms.dp = dp;
+ parms.port_no = ODPP_LOCAL;
+ vport = new_vport(&parms);
+ if (IS_ERR(vport)) {
+ err = PTR_ERR(vport);
+ if (err == -EBUSY)
+ err = -EEXIST;
+
+ goto err_destroy_table;
+ }
+
+ dp->drop_frags = 0;
+ dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
+ if (!dp->stats_percpu) {
+ err = -ENOMEM;
+ goto err_destroy_local_port;
+ }
+
+ change_datapath(dp, a);
+
+ rcu_assign_pointer(dps[dp_idx], dp);
+ dp_sysfs_add_dp(dp);
+
+ mutex_unlock(&dp->mutex);
+ mutex_unlock(&dp_mutex);
+ rtnl_unlock();
+
+ return 0;
+
+err_destroy_local_port:
+ dp_detach_port(get_vport_protected(dp, ODPP_LOCAL));
+err_destroy_table:
+ tbl_destroy(get_table_protected(dp), NULL);
+err_free_dp:
+ mutex_unlock(&dp->mutex);
+ kfree(dp);
+err_put_module:
+ module_put(THIS_MODULE);
+err_unlock_dp_mutex:
+ mutex_unlock(&dp_mutex);
+ rtnl_unlock();
+err_free_skb:
+ kfree_skb(skb);
+err:
+ return err;
+}
+
+static int del_datapath(struct odp_datapath __user *uodp_datapath)
+{
+ struct nlattr *a[ODP_DP_ATTR_MAX + 1];
+ struct datapath *dp;
+ struct sk_buff *skb;
+ int err;
+
+ skb = copy_datapath_from_user(uodp_datapath, a);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
+
+ rtnl_lock();
+ mutex_lock(&dp_mutex);
+ dp = lookup_datapath((struct odp_datapath *)skb->data, a);
+ err = PTR_ERR(dp);
+ if (IS_ERR(dp))
+ goto exit_free;
- kfree(actions);
- return err;
+ destroy_dp(dp);
+ err = 0;
-error_free_skb:
+exit_free:
kfree_skb(skb);
-error_free_actions:
- kfree(actions);
-error:
+ mutex_unlock(&dp_mutex);
+ rtnl_unlock();
+exit:
return err;
}
-static int execute_packet(struct datapath *dp, const struct odp_execute __user *executep)
-{
- struct odp_execute execute;
-
- if (copy_from_user(&execute, executep, sizeof(execute)))
- return -EFAULT;
-
- return do_execute(dp, &execute);
-}
-
-static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp)
+static int set_datapath(struct odp_datapath __user *uodp_datapath)
{
- struct odp_stats stats;
- int i;
+ struct nlattr *a[ODP_DP_ATTR_MAX + 1];
+ struct datapath *dp;
+ struct sk_buff *skb;
+ int err;
- stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0;
- for_each_possible_cpu(i) {
- const struct dp_stats_percpu *percpu_stats;
- struct dp_stats_percpu local_stats;
- unsigned seqcount;
+ skb = copy_datapath_from_user(uodp_datapath, a);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
- percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
+ mutex_lock(&dp_mutex);
+ dp = lookup_datapath((struct odp_datapath *)skb->data, a);
+ err = PTR_ERR(dp);
+ if (IS_ERR(dp))
+ goto exit_free;
- do {
- seqcount = read_seqcount_begin(&percpu_stats->seqlock);
- local_stats = *percpu_stats;
- } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
+ change_datapath(dp, a);
+ mutex_unlock(&dp->mutex);
+ err = 0;
- stats.n_frags += local_stats.n_frags;
- stats.n_hit += local_stats.n_hit;
- stats.n_missed += local_stats.n_missed;
- stats.n_lost += local_stats.n_lost;
- }
- return copy_to_user(statsp, &stats, sizeof(stats)) ? -EFAULT : 0;
+exit_free:
+ kfree_skb(skb);
+ mutex_unlock(&dp_mutex);
+exit:
+ return err;
}
-/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */
-int dp_min_mtu(const struct datapath *dp)
+static int get_datapath(struct odp_datapath __user *uodp_datapath)
{
- struct vport *p;
- int mtu = 0;
-
- ASSERT_RTNL();
+ struct nlattr *a[ODP_DP_ATTR_MAX + 1];
+ struct odp_datapath *odp_datapath;
+ struct datapath *dp;
+ struct sk_buff *skb;
+ int err;
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- int dev_mtu;
+ skb = copy_datapath_from_user(uodp_datapath, a);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
+ odp_datapath = (struct odp_datapath *)skb->data;
- /* Skip any internal ports, since that's what we're trying to
- * set. */
- if (is_internal_vport(p))
- continue;
+ mutex_lock(&dp_mutex);
+ dp = lookup_datapath(odp_datapath, a);
+ mutex_unlock(&dp_mutex);
- dev_mtu = vport_get_mtu(p);
- if (!mtu || dev_mtu < mtu)
- mtu = dev_mtu;
- }
+ err = PTR_ERR(dp);
+ if (IS_ERR(dp))
+ goto exit_free;
- return mtu ? mtu : ETH_DATA_LEN;
+ err = copy_datapath_to_user(uodp_datapath, dp, odp_datapath->total_len);
+ mutex_unlock(&dp->mutex);
+exit_free:
+ kfree_skb(skb);
+exit:
+ return err;
}
-/* Sets the MTU of all datapath devices to the minimum of the ports. Must
- * be called with RTNL lock. */
-void set_internal_devs_mtu(const struct datapath *dp)
+static int dump_datapath(struct odp_datapath __user *uodp_datapath)
{
- struct vport *p;
- int mtu;
+ struct nlattr *a[ODP_DP_ATTR_MAX + 1];
+ struct odp_datapath *odp_datapath;
+ struct sk_buff *skb;
+ u32 dp_idx;
+ int err;
- ASSERT_RTNL();
+ skb = copy_datapath_from_user(uodp_datapath, a);
+ err = PTR_ERR(skb);
+ if (IS_ERR(skb))
+ goto exit;
+ odp_datapath = (struct odp_datapath *)skb->data;
- mtu = dp_min_mtu(dp);
+ mutex_lock(&dp_mutex);
+ for (dp_idx = odp_datapath->dp_idx; dp_idx < ARRAY_SIZE(dps); dp_idx++) {
+ struct datapath *dp = get_dp(dp_idx);
+ if (!dp)
+ continue;
- list_for_each_entry_rcu (p, &dp->port_list, node) {
- if (is_internal_vport(p))
- vport_set_mtu(p, mtu);
+ mutex_lock(&dp->mutex);
+ mutex_unlock(&dp_mutex);
+ err = copy_datapath_to_user(uodp_datapath, dp, odp_datapath->total_len);
+ mutex_unlock(&dp->mutex);
+ goto exit_free;
}
-}
-
-static int get_listen_mask(const struct file *f)
-{
- return (long)f->private_data;
-}
+ mutex_unlock(&dp_mutex);
+ err = -ENODEV;
-static void set_listen_mask(struct file *f, int listen_mask)
-{
- f->private_data = (void*)(long)listen_mask;
+exit_free:
+ kfree_skb(skb);
+exit:
+ return err;
}
static const struct nla_policy vport_policy[ODP_VPORT_ATTR_MAX + 1] = {
{
int dp_idx = iminor(f->f_dentry->d_inode);
struct datapath *dp;
- int drop_frags, listeners;
- unsigned int sflow_probability;
+ int listeners;
int err;
/* Handle commands with special locking requirements up front. */
switch (cmd) {
- case ODP_DP_CREATE:
- err = create_dp(dp_idx, (char __user *)argp);
+ case ODP_DP_NEW:
+ err = new_datapath((struct odp_datapath __user *)argp);
+ goto exit;
+
+ case ODP_DP_GET:
+ err = get_datapath((struct odp_datapath __user *)argp);
+ goto exit;
+
+ case ODP_DP_DEL:
+ err = del_datapath((struct odp_datapath __user *)argp);
+ goto exit;
+
+ case ODP_DP_SET:
+ err = set_datapath((struct odp_datapath __user *)argp);
goto exit;
- case ODP_DP_DESTROY:
- err = destroy_dp(dp_idx);
+ case ODP_DP_DUMP:
+ err = dump_datapath((struct odp_datapath __user *)argp);
goto exit;
case ODP_VPORT_NEW:
case ODP_VPORT_DUMP:
err = dump_vport((struct odp_vport __user *)argp);
goto exit;
+
+ 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;
+
+ case ODP_EXECUTE:
+ err = execute_packet((struct odp_packet __user *)argp);
+ goto exit;
}
dp = get_dp_locked(dp_idx);
goto exit;
switch (cmd) {
- case ODP_DP_STATS:
- err = get_dp_stats(dp, (struct odp_stats __user *)argp);
- break;
-
- case ODP_GET_DROP_FRAGS:
- err = put_user(dp->drop_frags, (int __user *)argp);
- break;
-
- case ODP_SET_DROP_FRAGS:
- err = get_user(drop_frags, (int __user *)argp);
- if (err)
- break;
- err = -EINVAL;
- if (drop_frags != 0 && drop_frags != 1)
- break;
- dp->drop_frags = drop_frags;
- err = 0;
- break;
-
case ODP_GET_LISTEN_MASK:
err = put_user(get_listen_mask(f), (int __user *)argp);
break;
set_listen_mask(f, listeners);
break;
- case ODP_GET_SFLOW_PROBABILITY:
- err = put_user(dp->sflow_probability, (unsigned int __user *)argp);
- break;
-
- case ODP_SET_SFLOW_PROBABILITY:
- err = get_user(sflow_probability, (unsigned int __user *)argp);
- if (!err)
- dp->sflow_probability = sflow_probability;
- break;
-
- case ODP_FLOW_FLUSH:
- err = flush_flows(dp);
- break;
-
- case ODP_FLOW_PUT:
- err = put_flow(dp, (struct odp_flow_put __user *)argp);
- break;
-
- case ODP_FLOW_DEL:
- err = del_flow(dp, (struct odp_flow __user *)argp);
- break;
-
- case ODP_FLOW_GET:
- err = query_flow(dp, (struct odp_flow __user *)argp);
- break;
-
- case ODP_FLOW_DUMP:
- err = dump_flow(dp, (struct odp_flow_dump __user *)argp);
- break;
-
- case ODP_EXECUTE:
- err = execute_packet(dp, (struct odp_execute __user *)argp);
- break;
-
default:
err = -ENOIOCTLCMD;
break;
}
#ifdef CONFIG_COMPAT
-static int compat_get_flow(struct odp_flow *flow, const struct compat_odp_flow __user *compat)
-{
- compat_uptr_t key, actions;
-
- if (!access_ok(VERIFY_READ, compat, sizeof(struct compat_odp_flow)) ||
- __copy_from_user(&flow->stats, &compat->stats, sizeof(struct odp_flow_stats)) ||
- __get_user(key, &compat->key) ||
- __get_user(flow->key_len, &compat->key_len) ||
- __get_user(actions, &compat->actions) ||
- __get_user(flow->actions_len, &compat->actions_len) ||
- __get_user(flow->flags, &compat->flags))
- return -EFAULT;
-
- flow->key = (struct nlattr __force *)compat_ptr(key);
- flow->actions = (struct nlattr __force *)compat_ptr(actions);
- return 0;
-}
-
-static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __user *ufp)
-{
- struct odp_flow_stats stats;
- struct odp_flow_put fp;
- int error;
-
- if (compat_get_flow(&fp.flow, &ufp->flow) ||
- get_user(fp.flags, &ufp->flags))
- return -EFAULT;
-
- error = do_put_flow(dp, &fp, &stats);
- if (error)
- return error;
-
- if (copy_to_user(&ufp->flow.stats, &stats,
- sizeof(struct odp_flow_stats)))
- return -EFAULT;
-
- return 0;
-}
-
-static int compat_answer_query(struct datapath *dp, struct sw_flow *flow,
- u32 query_flags,
- struct compat_odp_flow __user *ufp)
-{
- compat_uptr_t actions;
-
- if (get_user(actions, &ufp->actions))
- return -EFAULT;
-
- return do_answer_query(dp, flow, query_flags, &ufp->stats,
- compat_ptr(actions), &ufp->actions_len);
-}
-
-static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *ufp)
-{
- struct sw_flow *flow;
- struct odp_flow uf;
- int error;
-
- if (compat_get_flow(&uf, ufp))
- return -EFAULT;
-
- flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len);
- if (IS_ERR(flow))
- return PTR_ERR(flow);
-
- error = compat_answer_query(dp, flow, 0, ufp);
- flow_deferred_free(flow);
- return error;
-}
-
-static int compat_query_flow(struct datapath *dp, struct compat_odp_flow __user *uflow)
-{
- struct tbl *table = get_table_protected(dp);
- struct tbl_node *flow_node;
- struct sw_flow_key key;
- struct odp_flow flow;
- int error;
-
- if (compat_get_flow(&flow, uflow))
- return -EFAULT;
-
- error = flow_copy_from_user(&key, (const struct nlattr __force __user *)flow.key, flow.key_len);
- if (error)
- return error;
-
- flow_node = tbl_lookup(table, &key, flow_hash(&key), flow_cmp);
- if (!flow_node)
- return -ENOENT;
- return compat_answer_query(dp, flow_cast(flow_node), flow.flags, uflow);
-}
-
-static int compat_dump_flow(struct datapath *dp, struct compat_odp_flow_dump __user *udumpp)
-{
- struct compat_odp_flow __user *uflowp;
- compat_uptr_t compat_ufp;
- struct sw_flow *flow;
- compat_uptr_t ukey;
- u32 key_len;
-
- flow = do_dump_flow(dp, udumpp->state);
- if (IS_ERR(flow))
- return PTR_ERR(flow);
-
- if (get_user(compat_ufp, &udumpp->flow))
- return -EFAULT;
- uflowp = compat_ptr(compat_ufp);
-
- if (!flow)
- return put_user(ODPFF_EOF, &uflowp->flags);
-
- if (put_user(0, &uflowp->flags) ||
- get_user(ukey, &uflowp->key) ||
- get_user(key_len, &uflowp->key_len))
- return -EFAULT;
-
- key_len = flow_copy_to_user(compat_ptr(ukey), &flow->key, key_len);
- if (key_len < 0)
- return key_len;
- if (put_user(key_len, &uflowp->key_len))
- return -EFAULT;
-
- return compat_answer_query(dp, flow, 0, uflowp);
-}
-
-static int compat_execute(struct datapath *dp, const struct compat_odp_execute __user *uexecute)
-{
- struct odp_execute execute;
- compat_uptr_t actions;
- compat_uptr_t data;
-
- if (!access_ok(VERIFY_READ, uexecute, sizeof(struct compat_odp_execute)) ||
- __get_user(actions, &uexecute->actions) ||
- __get_user(execute.actions_len, &uexecute->actions_len) ||
- __get_user(data, &uexecute->data) ||
- __get_user(execute.length, &uexecute->length))
- return -EFAULT;
-
- execute.actions = (struct nlattr __force *)compat_ptr(actions);
- execute.data = (const void __force *)compat_ptr(data);
-
- return do_execute(dp, &execute);
-}
-
static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp)
{
- int dp_idx = iminor(f->f_dentry->d_inode);
- struct datapath *dp;
- int err;
-
switch (cmd) {
- case ODP_DP_DESTROY:
case ODP_FLOW_FLUSH:
/* Ioctls that don't need any translation at all. */
return openvswitch_ioctl(f, cmd, argp);
- case ODP_DP_CREATE:
+ case ODP_DP_NEW:
+ case ODP_DP_GET:
+ case ODP_DP_DEL:
+ case ODP_DP_SET:
+ case ODP_DP_DUMP:
case ODP_VPORT_NEW:
case ODP_VPORT_DEL:
case ODP_VPORT_GET:
case ODP_VPORT_SET:
case ODP_VPORT_DUMP:
- case ODP_DP_STATS:
- case ODP_GET_DROP_FRAGS:
- case ODP_SET_DROP_FRAGS:
+ case ODP_FLOW_NEW:
+ case ODP_FLOW_DEL:
+ case ODP_FLOW_GET:
+ case ODP_FLOW_SET:
+ case ODP_FLOW_DUMP:
case ODP_SET_LISTEN_MASK:
case ODP_GET_LISTEN_MASK:
- case ODP_SET_SFLOW_PROBABILITY:
- case ODP_GET_SFLOW_PROBABILITY:
+ case ODP_EXECUTE:
/* Ioctls that just need their pointer argument extended. */
return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp));
- }
-
- dp = get_dp_locked(dp_idx);
- err = -ENODEV;
- if (!dp)
- goto exit;
-
- switch (cmd) {
- case ODP_FLOW_PUT32:
- err = compat_put_flow(dp, compat_ptr(argp));
- break;
-
- case ODP_FLOW_DEL32:
- err = compat_del_flow(dp, compat_ptr(argp));
- break;
-
- case ODP_FLOW_GET32:
- err = compat_query_flow(dp, compat_ptr(argp));
- break;
-
- case ODP_FLOW_DUMP32:
- err = compat_dump_flow(dp, compat_ptr(argp));
- break;
-
- case ODP_EXECUTE32:
- err = compat_execute(dp, compat_ptr(argp));
- break;
default:
- err = -ENOIOCTLCMD;
- break;
+ return -ENOIOCTLCMD;
}
- mutex_unlock(&dp->mutex);
-exit:
- return err;
}
#endif