From d656937779f5b987ec021c21189a0deab29fd64d Mon Sep 17 00:00:00 2001 From: Ben Pfaff Date: Wed, 26 Jan 2011 15:42:00 -0800 Subject: [PATCH] datapath: Convert datapath operations to use Netlink framing. Signed-off-by: Ben Pfaff Acked-by: Jesse Gross --- datapath/datapath.c | 1524 ++++++++++++----------- datapath/flow.c | 94 +- datapath/flow.h | 7 +- datapath/odp-compat.h | 25 - include/openvswitch/datapath-protocol.h | 118 +- lib/dpif-linux.c | 696 ++++++++--- 6 files changed, 1464 insertions(+), 1000 deletions(-) diff --git a/datapath/datapath.c b/datapath/datapath.c index f6a2824c..ea998edc 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -66,7 +66,7 @@ EXPORT_SYMBOL(dp_ioctl_hook); * 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 *); @@ -74,7 +74,7 @@ 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)); @@ -206,111 +206,6 @@ static struct kobj_type dp_ktype = { .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//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); @@ -324,22 +219,11 @@ static void destroy_dp_rcu(struct rcu_head *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); @@ -351,11 +235,6 @@ static int destroy_dp(int dp_idx) 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. */ @@ -745,54 +624,25 @@ 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) @@ -815,471 +665,912 @@ static int expand_table(struct datapath *dp) 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 int do_execute(struct datapath *dp, const struct odp_execute *execute) { - struct tbl_node *flow_node; 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; - - 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; + struct sk_buff *skb; + struct sw_flow_actions *actions; + struct ethhdr *eth; + bool is_frag; + int err; - /* 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); - } + err = -EINVAL; + if (execute->length < ETH_HLEN || execute->length > 65535) + goto error; - /* Allocate flow. */ - flow = flow_alloc(); - if (IS_ERR(flow)) { - error = PTR_ERR(flow); - goto error; - } - flow->key = key; - clear_stats(flow); + actions = flow_actions_alloc(execute->actions_len); + if (IS_ERR(actions)) { + err = PTR_ERR(actions); + goto error; + } - /* 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); + err = -EFAULT; + if (copy_from_user(actions->actions, + (struct nlattr __user __force *)execute->actions, execute->actions_len)) + goto error_free_actions; - /* Put flow in bucket. */ - error = tbl_insert(table, &flow->tbl_node, hash); - if (error) - goto error_free_flow_acts; + err = validate_actions(actions->actions, execute->actions_len); + if (err) + goto error_free_actions; - memset(stats, 0, sizeof(struct odp_flow_stats)); - } else { - /* We found a matching flow. */ - struct sw_flow_actions *old_acts, *new_acts; + err = -ENOMEM; + skb = alloc_skb(execute->length, GFP_KERNEL); + if (!skb) + goto error_free_actions; - flow = flow_cast(flow_node); + err = -EFAULT; + if (copy_from_user(skb_put(skb, execute->length), + (const void __user __force *)execute->data, + execute->length)) + goto error_free_skb; - /* Bail out if we're not allowed to modify an existing flow. */ - error = -EEXIST; - if (!(uf->flags & ODPPF_MODIFY)) - goto error; + skb_reset_mac_header(skb); + eth = eth_hdr(skb); - /* Swap actions. */ - new_acts = get_actions(&uf->flow); - error = PTR_ERR(new_acts); - if (IS_ERR(new_acts)) - goto error; + /* 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); - 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); - } + err = flow_extract(skb, -1, &key, &is_frag); + if (err) + goto error_free_skb; - /* 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); - } + rcu_read_lock(); + err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); + rcu_read_unlock(); - return 0; + kfree(actions); + return err; -error_free_flow_acts: - kfree(acts); -error_free_flow: - flow->sf_acts = NULL; - flow_put(flow); +error_free_skb: + kfree_skb(skb); +error_free_actions: + kfree(actions); error: - return error; + return err; } -static int put_flow(struct odp_flow_put __user *ufp) +static int execute_packet(const struct odp_execute __user *executep) { - struct odp_flow_stats stats; - struct odp_flow_put uf; + struct odp_execute execute; struct datapath *dp; int error; - if (copy_from_user(&uf, ufp, sizeof(struct odp_flow_put))) + if (copy_from_user(&execute, executep, sizeof(execute))) return -EFAULT; - dp = get_dp_locked(uf.flow.dp_idx); + dp = get_dp_locked(execute.dp_idx); if (!dp) return -ENODEV; - - error = do_put_flow(dp, &uf, &stats); - if (!error) { - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - error = -EFAULT; - } + error = do_execute(dp, &execute); mutex_unlock(&dp->mutex); return error; } -static int do_answer_query(struct datapath *dp, struct sw_flow *flow, - struct odp_flow_stats __user *ustats, - struct nlattr __user *actions, - u32 __user *actions_lenp) +static void get_dp_stats(struct datapath *dp, struct odp_stats *stats) { - struct sw_flow_actions *sf_acts; - struct odp_flow_stats stats; - u32 actions_len; + int i; - spin_lock_bh(&flow->lock); - get_stats(flow, &stats); - spin_unlock_bh(&flow->lock); + 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; - if (copy_to_user(ustats, &stats, sizeof(struct odp_flow_stats)) || - get_user(actions_len, actions_lenp)) - return -EFAULT; + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); - if (!actions_len) - return 0; + do { + seqcount = read_seqcount_begin(&percpu_stats->seqlock); + local_stats = *percpu_stats; + } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); - 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; + 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 0; +/* MTU of the dp pseudo-device: ETH_DATA_LEN or the minimum of the ports */ +int dp_min_mtu(const struct datapath *dp) +{ + struct vport *p; + int mtu = 0; + + ASSERT_RTNL(); + + list_for_each_entry_rcu (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 (!mtu || dev_mtu < mtu) + mtu = dev_mtu; + } + + return mtu ? mtu : ETH_DATA_LEN; } -static int answer_query(struct datapath *dp, struct sw_flow *flow, - struct odp_flow __user *ufp) +/* 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 nlattr __user *actions; + struct vport *p; + int mtu; - if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions)) - return -EFAULT; + ASSERT_RTNL(); + + mtu = dp_min_mtu(dp); - return do_answer_query(dp, flow, &ufp->stats, actions, &ufp->actions_len); + list_for_each_entry_rcu (p, &dp->port_list, node) { + if (is_internal_vport(p)) + vport_set_mtu(p, mtu); + } } -static struct sw_flow *do_del_flow(struct datapath *dp, const struct nlattr __user *key, u32 key_len) +static int get_listen_mask(const struct file *f) { - struct tbl *table = get_table_protected(dp); - struct tbl_node *flow_node; - struct sw_flow_key swkey; - int error; + return (long)f->private_data; +} + +static void set_listen_mask(struct file *f, int listen_mask) +{ + f->private_data = (void*)(long)listen_mask; +} - error = flow_copy_from_user(&swkey, key, key_len); - if (error) - return ERR_PTR(error); +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 }, +}; - flow_node = tbl_lookup(table, &swkey, flow_hash(&swkey), flow_cmp); - if (!flow_node) - return ERR_PTR(-ENOENT); +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; - error = tbl_remove(table, flow_node); - if (error) - return ERR_PTR(error); + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); - /* 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); + skb = alloc_skb(128 + FLOW_BUFSIZE + sf_acts->actions_len, GFP_KERNEL); + err = -ENOMEM; + if (!skb) + goto exit; + + 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 int del_flow(struct odp_flow __user *ufp) +static struct sk_buff *copy_flow_from_user(struct odp_flow __user *uodp_flow, + struct dp_flowcmd *flowcmd) { - struct sw_flow *flow; - struct datapath *dp; - struct odp_flow uf; - int error; + struct nlattr *a[ODP_FLOW_ATTR_MAX + 1]; + struct odp_flow *odp_flow; + struct sk_buff *skb; + u32 len; + int err; - if (copy_from_user(&uf, ufp, sizeof(uf))) - return -EFAULT; + if (get_user(len, &uodp_flow->len)) + return ERR_PTR(-EFAULT); + if (len < sizeof(struct odp_flow)) + return ERR_PTR(-EINVAL); - dp = get_dp_locked(uf.dp_idx); - if (!dp) - return -ENODEV; + skb = alloc_skb(len, GFP_KERNEL); + if (!skb) + return ERR_PTR(-ENOMEM); - flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len); - error = PTR_ERR(flow); - if (!IS_ERR(flow)) { - error = answer_query(dp, flow, ufp); - flow_deferred_free(flow); + 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)); + + /* 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; } - mutex_unlock(&dp->mutex); - return error; + 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 query_flow(struct odp_flow __user *uflow) +static int new_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) { struct tbl_node *flow_node; - struct sw_flow_key key; - struct odp_flow flow; + struct dp_flowcmd flowcmd; + struct sw_flow *flow; + struct sk_buff *skb; struct datapath *dp; + struct tbl *table; + u32 hash; int error; - if (copy_from_user(&flow, uflow, sizeof(flow))) - return -EFAULT; + skb = copy_flow_from_user(uodp_flow, &flowcmd); + error = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - dp = get_dp_locked(flow.dp_idx); + dp = get_dp_locked(flowcmd.dp_idx); + error = -ENODEV; if (!dp) - return -ENODEV; + goto error_kfree_skb; - error = flow_copy_from_user(&key, (const struct nlattr __force __user *)flow.key, flow.key_len); - if (!error) { - struct tbl *table = get_table_protected(dp); - flow_node = tbl_lookup(table, &flow.key, flow_hash(&key), flow_cmp); - if (flow_node) - error = answer_query(dp, flow_cast(flow_node), uflow); - else - error = -ENOENT; + 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 dump_flow(struct odp_flow_dump __user *udump) +static int get_or_del_flow(unsigned int cmd, struct odp_flow __user *uodp_flow) { - struct odp_flow __user *uflow; - struct nlattr __user *ukey; - struct tbl_node *tbl_node; - struct odp_flow_dump dump; + struct tbl_node *flow_node; + struct dp_flowcmd flowcmd; struct sw_flow *flow; + struct sk_buff *skb; struct datapath *dp; struct tbl *table; - u32 key_len; int err; - err = -EFAULT; - if (copy_from_user(&dump, udump, sizeof(struct odp_flow_dump))) + skb = copy_flow_from_user(uodp_flow, &flowcmd); + err = PTR_ERR(skb); + if (IS_ERR(skb)) goto exit; - uflow = (struct odp_flow __user __force *)dump.flow; - dp = get_dp_locked(dump.dp_idx); + dp = get_dp_locked(flowcmd.dp_idx); err = -ENODEV; if (!dp) - goto exit; + goto exit_kfree_skb; table = get_table_protected(dp); - tbl_node = tbl_next(table, &dump.state[0], &dump.state[1]); - if (!tbl_node) { - err = put_user(0, &uflow->key_len); - goto exit_unlock; + flow_node = tbl_lookup(table, &flowcmd.key, flow_hash(&flowcmd.key), flow_cmp); + err = -ENOENT; + if (!flow_node) + goto exit_unlock_dp; + + if (cmd == ODP_FLOW_DEL) { + err = tbl_remove(table, flow_node); + if (err) + goto exit_unlock_dp; } - flow = flow_cast(tbl_node); - err = -EFAULT; - if (copy_to_user(udump->state, dump.state, 2 * sizeof(uint32_t)) || - get_user(ukey, (struct nlattr __user * __user*)&uflow->key) || - get_user(key_len, &uflow->key_len)) - goto exit_unlock; + 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); - key_len = flow_copy_to_user(ukey, &flow->key, key_len); - err = key_len; - if (key_len < 0) - goto exit_unlock; - err = -EFAULT; - if (put_user(key_len, &uflow->key_len)) - goto exit_unlock; +exit_unlock_dp: + mutex_unlock(&dp->mutex); +exit_kfree_skb: + kfree_skb(skb); +exit: + return err; +} + +static int dump_flow(struct odp_flow __user *uodp_flow) +{ + 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; + + skb = copy_flow_from_user(uodp_flow, &flowcmd); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - err = answer_query(dp, flow, uflow); + dp = get_dp_locked(flowcmd.dp_idx); + err = -ENODEV; + if (!dp) + goto exit_free; -exit_unlock: + 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); + +exit_unlock_dp: mutex_unlock(&dp->mutex); +exit_free: + kfree_skb(skb); exit: return err; } -static int do_execute(struct datapath *dp, const struct odp_execute *execute) +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 sw_flow_key key; + struct odp_datapath *odp_datapath; struct sk_buff *skb; - struct sw_flow_actions *actions; - struct ethhdr *eth; - bool is_frag; + struct nlattr *nla; int err; - err = -EINVAL; - if (execute->length < ETH_HLEN || execute->length > 65535) - goto error; - - actions = flow_actions_alloc(execute->actions_len); - if (IS_ERR(actions)) { - err = PTR_ERR(actions); - goto error; - } + skb = alloc_skb(NLMSG_GOODSIZE, GFP_KERNEL); + err = -ENOMEM; + if (!skb) + goto exit; - err = -EFAULT; - if (copy_from_user(actions->actions, - (struct nlattr __user __force *)execute->actions, execute->actions_len)) - goto error_free_actions; + 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; - err = validate_actions(actions->actions, execute->actions_len); + rcu_read_lock(); + err = nla_put_string(skb, ODP_DP_ATTR_NAME, dp_name(dp)); + rcu_read_unlock(); if (err) - goto error_free_actions; + goto nla_put_failure; - err = -ENOMEM; - skb = alloc_skb(execute->length, GFP_KERNEL); + 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)); + + 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; + +nla_put_failure: + err = -EMSGSIZE; +exit_free_skb: + kfree_skb(skb); +exit: + return err; +} + +static struct sk_buff *copy_datapath_from_user(struct odp_datapath __user *uodp_datapath, struct nlattr *a[ODP_DP_ATTR_MAX + 1]) +{ + struct odp_datapath *odp_datapath; + struct sk_buff *skb; + u32 len; + int err; + + if (get_user(len, &uodp_datapath->len)) + return ERR_PTR(-EFAULT); + if (len < sizeof(struct odp_datapath)) + return ERR_PTR(-EINVAL); + + skb = alloc_skb(len, GFP_KERNEL); if (!skb) - goto error_free_actions; + return ERR_PTR(-ENOMEM); err = -EFAULT; - if (copy_from_user(skb_put(skb, execute->length), - (const void __user __force *)execute->data, - execute->length)) + if (copy_from_user(__skb_put(skb, len), uodp_datapath, len)) goto error_free_skb; - skb_reset_mac_header(skb); - eth = eth_hdr(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); + odp_datapath = (struct odp_datapath *)skb->data; + err = -EINVAL; + if (odp_datapath->len != len) + goto error_free_skb; - err = flow_extract(skb, -1, &key, &is_frag); + 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_skb; - rcu_read_lock(); - err = execute_actions(dp, skb, &key, actions->actions, actions->actions_len); - rcu_read_unlock(); + if (a[ODP_DP_ATTR_IPV4_FRAGS]) { + u32 frags = nla_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); - kfree(actions); - return err; + 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; + + return skb; error_free_skb: kfree_skb(skb); -error_free_actions: - kfree(actions); -error: + 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; + } +} + +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 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//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 execute_packet(const struct odp_execute __user *executep) +static int del_datapath(struct odp_datapath __user *uodp_datapath) { - struct odp_execute execute; + struct nlattr *a[ODP_DP_ATTR_MAX + 1]; struct datapath *dp; - int error; + struct sk_buff *skb; + int err; - if (copy_from_user(&execute, executep, sizeof(execute))) - return -EFAULT; + skb = copy_datapath_from_user(uodp_datapath, a); + err = PTR_ERR(skb); + if (IS_ERR(skb)) + goto exit; - dp = get_dp_locked(execute.dp_idx); - if (!dp) - return -ENODEV; - error = do_execute(dp, &execute); - mutex_unlock(&dp->mutex); + 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; - return error; + destroy_dp(dp); + err = 0; + +exit_free: + kfree_skb(skb); + mutex_unlock(&dp_mutex); + rtnl_unlock(); +exit: + return err; } -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] = { @@ -1681,18 +1972,29 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, { 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_DESTROY: - err = destroy_dp(dp_idx); + 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_DUMP: + err = dump_datapath((struct odp_datapath __user *)argp); goto exit; case ODP_VPORT_NEW: @@ -1719,20 +2021,18 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, err = flush_flows(argp); goto exit; - case ODP_FLOW_PUT: - err = put_flow((struct odp_flow_put __user *)argp); - goto exit; - - case ODP_FLOW_DEL: - err = del_flow((struct odp_flow __user *)argp); + case ODP_FLOW_NEW: + case ODP_FLOW_SET: + err = new_flow(cmd, (struct odp_flow __user *)argp); goto exit; case ODP_FLOW_GET: - err = query_flow((struct odp_flow __user *)argp); + 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_dump __user *)argp); + err = dump_flow((struct odp_flow __user *)argp); goto exit; case ODP_EXECUTE: @@ -1746,25 +2046,6 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, 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; @@ -1780,16 +2061,6 @@ static long openvswitch_ioctl(struct file *f, unsigned int cmd, 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; - default: err = -ENOIOCTLCMD; break; @@ -1810,170 +2081,6 @@ static int dp_has_packet_of_interest(struct datapath *dp, int listeners) } #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)) - 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 compat_odp_flow_put __user *ufp) -{ - struct odp_flow_stats stats; - struct odp_flow_put uf; - struct datapath *dp; - int error; - - if (compat_get_flow(&uf.flow, &ufp->flow) || - get_user(uf.flags, &ufp->flags)) - return -EFAULT; - - dp = get_dp_locked(uf.flow.dp_idx); - if (!dp) - return -ENODEV; - - error = do_put_flow(dp, &uf, &stats); - if (!error) { - if (copy_to_user(&ufp->flow.stats, &stats, - sizeof(struct odp_flow_stats))) - error = -EFAULT; - } - mutex_unlock(&dp->mutex); - - return error; -} - - -static int compat_answer_query(struct datapath *dp, struct sw_flow *flow, - struct compat_odp_flow __user *ufp) -{ - compat_uptr_t actions; - - if (get_user(actions, &ufp->actions)) - return -EFAULT; - - return do_answer_query(dp, flow, &ufp->stats, - compat_ptr(actions), &ufp->actions_len); -} - -static int compat_del_flow(struct compat_odp_flow __user *ufp) -{ - struct sw_flow *flow; - struct datapath *dp; - struct odp_flow uf; - int error; - - if (compat_get_flow(&uf, ufp)) - return -EFAULT; - - dp = get_dp_locked(uf.dp_idx); - if (!dp) - return -ENODEV; - - flow = do_del_flow(dp, (const struct nlattr __force __user *)uf.key, uf.key_len); - error = PTR_ERR(flow); - if (!IS_ERR(flow)) { - error = compat_answer_query(dp, flow, ufp); - flow_deferred_free(flow); - } - mutex_unlock(&dp->mutex); - - return error; -} - -static int compat_query_flow(struct compat_odp_flow __user *uflow) -{ - struct tbl_node *flow_node; - struct sw_flow_key key; - struct odp_flow flow; - struct datapath *dp; - int error; - - if (compat_get_flow(&flow, uflow)) - return -EFAULT; - - dp = get_dp_locked(flow.dp_idx); - if (!dp) - return -ENODEV; - - error = flow_copy_from_user(&key, (const struct nlattr __force __user *)flow.key, flow.key_len); - if (!error) { - struct tbl *table = get_table_protected(dp); - flow_node = tbl_lookup(table, &flow.key, flow_hash(&key), flow_cmp); - if (flow_node) - error = compat_answer_query(dp, flow_cast(flow_node), uflow); - else - error = -ENOENT; - } - mutex_unlock(&dp->mutex); - - return error; -} - -static int compat_dump_flow(struct compat_odp_flow_dump __user *udump) -{ - struct compat_odp_flow __user *uflow; - struct nlattr __user *ukey; - struct tbl_node *tbl_node; - struct compat_odp_flow_dump dump; - struct sw_flow *flow; - compat_uptr_t ukey32; - struct datapath *dp; - struct tbl *table; - u32 key_len; - int err; - - err = -EFAULT; - if (copy_from_user(&dump, udump, sizeof(struct compat_odp_flow_dump))) - goto exit; - uflow =compat_ptr(dump.flow); - - dp = get_dp_locked(dump.dp_idx); - err = -ENODEV; - if (!dp) - goto exit; - - table = get_table_protected(dp); - tbl_node = tbl_next(table, &dump.state[0], &dump.state[1]); - if (!tbl_node) { - err = put_user(0, &uflow->key_len); - goto exit_unlock; - } - flow = flow_cast(tbl_node); - - err = -EFAULT; - if (copy_to_user(udump->state, dump.state, 2 * sizeof(uint32_t)) || - get_user(ukey32, &uflow->key) || - get_user(key_len, &uflow->key_len)) - goto exit_unlock; - ukey = compat_ptr(ukey32); - - key_len = flow_copy_to_user(ukey, &flow->key, key_len); - err = key_len; - if (key_len < 0) - goto exit_unlock; - err = -EFAULT; - if (put_user(key_len, &uflow->key_len)) - goto exit_unlock; - - err = compat_answer_query(dp, flow, uflow); - -exit_unlock: - mutex_unlock(&dp->mutex); -exit: - return err; -} - static int compat_execute(const struct compat_odp_execute __user *uexecute) { struct odp_execute execute; @@ -2005,39 +2112,30 @@ static int compat_execute(const struct compat_odp_execute __user *uexecute) static long openvswitch_compat_ioctl(struct file *f, unsigned int cmd, unsigned long argp) { 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: /* Ioctls that just need their pointer argument extended. */ return openvswitch_ioctl(f, cmd, (unsigned long)compat_ptr(argp)); - case ODP_FLOW_PUT32: - return compat_put_flow(compat_ptr(argp)); - - case ODP_FLOW_DEL32: - return compat_del_flow(compat_ptr(argp)); - - case ODP_FLOW_GET32: - return compat_query_flow(compat_ptr(argp)); - - case ODP_FLOW_DUMP32: - return compat_dump_flow(compat_ptr(argp)); - case ODP_EXECUTE32: return compat_execute(compat_ptr(argp)); diff --git a/datapath/flow.c b/datapath/flow.c index 63f27520..0987fd2a 100644 --- a/datapath/flow.c +++ b/datapath/flow.c @@ -399,15 +399,15 @@ int flow_cmp(const struct tbl_node *node, void *key2_) /** * flow_from_nlattrs - parses Netlink attributes into a flow key. * @swkey: receives the extracted flow key. - * @key: start of %ODP_KEY_ATTR_* Netlink attribute sequence. - * @key_len: number of bytes in @key. + * @key: Netlink attribute holding nested %ODP_KEY_ATTR_* Netlink attribute + * sequence. * * This state machine accepts the following forms, with [] for optional * elements and | for alternatives: * * [tun_id] in_port ethernet [8021q] [ethertype [IP [TCP|UDP|ICMP] | ARP] */ -static int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *key, u32 key_len) +int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *attr) { const struct nlattr *nla; u16 prev_type; @@ -417,7 +417,7 @@ static int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *key swkey->dl_type = htons(ETH_P_802_2); prev_type = ODP_KEY_ATTR_UNSPEC; - nla_for_each_attr(nla, key, key_len, rem) { + nla_for_each_nested(nla, attr, rem) { static const u32 key_lens[ODP_KEY_ATTR_MAX + 1] = { [ODP_KEY_ATTR_TUN_ID] = 8, [ODP_KEY_ATTR_IN_PORT] = 4, @@ -572,39 +572,43 @@ static int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *key return -EINVAL; } -u32 flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb) +int flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb) { struct odp_key_ethernet *eth_key; - - if (skb_tailroom(skb) < FLOW_BUFSIZE) - return -EMSGSIZE; + struct nlattr *nla; if (swkey->tun_id != cpu_to_be64(0)) - nla_put_be64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id); + NLA_PUT_BE64(skb, ODP_KEY_ATTR_TUN_ID, swkey->tun_id); - nla_put_u32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port); + NLA_PUT_U32(skb, ODP_KEY_ATTR_IN_PORT, swkey->in_port); - eth_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key))); + nla = nla_reserve(skb, ODP_KEY_ATTR_ETHERNET, sizeof(*eth_key)); + if (!nla) + goto nla_put_failure; + eth_key = nla_data(nla); memcpy(eth_key->eth_src, swkey->dl_src, ETH_ALEN); memcpy(eth_key->eth_dst, swkey->dl_dst, ETH_ALEN); if (swkey->dl_tci != htons(0)) { - struct odp_key_8021q *q_key; + struct odp_key_8021q q_key; - q_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_8021Q, sizeof(*q_key))); - q_key->q_tpid = htons(ETH_P_8021Q); - q_key->q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT); + q_key.q_tpid = htons(ETH_P_8021Q); + q_key.q_tci = swkey->dl_tci & ~htons(VLAN_TAG_PRESENT); + NLA_PUT(skb, ODP_KEY_ATTR_8021Q, sizeof(q_key), &q_key); } if (swkey->dl_type == htons(ETH_P_802_2)) - goto exit; + return 0; - nla_put_be16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type); + NLA_PUT_BE16(skb, ODP_KEY_ATTR_ETHERTYPE, swkey->dl_type); if (swkey->dl_type == htons(ETH_P_IP)) { struct odp_key_ipv4 *ipv4_key; - ipv4_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key))); + nla = nla_reserve(skb, ODP_KEY_ATTR_IPV4, sizeof(*ipv4_key)); + if (!nla) + goto nla_put_failure; + ipv4_key = nla_data(nla); ipv4_key->ipv4_src = swkey->nw_src; ipv4_key->ipv4_dst = swkey->nw_dst; ipv4_key->ipv4_proto = swkey->nw_proto; @@ -613,63 +617,47 @@ u32 flow_to_nlattrs(const struct sw_flow_key *swkey, struct sk_buff *skb) if (swkey->nw_proto == IPPROTO_TCP) { struct odp_key_tcp *tcp_key; - tcp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key))); + nla = nla_reserve(skb, ODP_KEY_ATTR_TCP, sizeof(*tcp_key)); + if (!nla) + goto nla_put_failure; + tcp_key = nla_data(nla); tcp_key->tcp_src = swkey->tp_src; tcp_key->tcp_dst = swkey->tp_dst; } else if (swkey->nw_proto == IPPROTO_UDP) { struct odp_key_udp *udp_key; - udp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key))); + nla = nla_reserve(skb, ODP_KEY_ATTR_UDP, sizeof(*udp_key)); + if (!nla) + goto nla_put_failure; + udp_key = nla_data(nla); udp_key->udp_src = swkey->tp_src; udp_key->udp_dst = swkey->tp_dst; } else if (swkey->nw_proto == IPPROTO_ICMP) { struct odp_key_icmp *icmp_key; - icmp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key))); + nla = nla_reserve(skb, ODP_KEY_ATTR_ICMP, sizeof(*icmp_key)); + if (!nla) + goto nla_put_failure; + icmp_key = nla_data(nla); icmp_key->icmp_type = ntohs(swkey->tp_src); icmp_key->icmp_code = ntohs(swkey->tp_dst); } } else if (swkey->dl_type == htons(ETH_P_ARP)) { struct odp_key_arp *arp_key; - arp_key = nla_data(__nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key))); + nla = nla_reserve(skb, ODP_KEY_ATTR_ARP, sizeof(*arp_key)); + if (!nla) + goto nla_put_failure; + arp_key = nla_data(nla); arp_key->arp_sip = swkey->nw_src; arp_key->arp_tip = swkey->nw_dst; arp_key->arp_op = htons(swkey->nw_proto); } -exit: - return skb->len; -} - -int flow_copy_from_user(struct sw_flow_key *swkey, const struct nlattr __user *ukey, u32 ukey_len) -{ - char key[FLOW_BUFSIZE] __aligned(NLA_ALIGNTO); - - if (ukey_len > FLOW_BUFSIZE || ukey_len % NLA_ALIGNTO) - return -EINVAL; - - if (copy_from_user(key, ukey, ukey_len)) - return -EFAULT; - - return flow_from_nlattrs(swkey, (const struct nlattr *)key, ukey_len); -} - -int flow_copy_to_user(struct nlattr __user *ukey, const struct sw_flow_key *swkey, u32 ukey_len) -{ - struct sk_buff *skb; - int retval; - - skb = alloc_skb(FLOW_BUFSIZE, GFP_KERNEL); - if (!skb) - return -ENOMEM; - - retval = flow_to_nlattrs(swkey, skb); - if (copy_to_user(ukey, skb->data, min(skb->len, ukey_len))) - retval = -EFAULT; - kfree_skb(skb); + return 0; - return retval; +nla_put_failure: + return -EMSGSIZE; } /* Initializes the flow module. diff --git a/datapath/flow.h b/datapath/flow.h index 5e5ae5e1..f9aa44a4 100644 --- a/datapath/flow.h +++ b/datapath/flow.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2009, 2010 Nicira Networks. + * Copyright (c) 2009, 2010, 2011 Nicira Networks. * Distributed under the terms of the GNU GPL version 2. * * Significant portions of this file may be copied from parts of the Linux @@ -100,9 +100,8 @@ int flow_cmp(const struct tbl_node *, void *target); */ #define FLOW_BUFSIZE 96 -u32 flow_to_nlattrs(const struct sw_flow_key *, struct sk_buff *); -int flow_copy_from_user(struct sw_flow_key *, const struct nlattr __user *ukey, u32 key_len); -int flow_copy_to_user(struct nlattr __user *ukey, const struct sw_flow_key *, u32 key_len); +int flow_to_nlattrs(const struct sw_flow_key *, struct sk_buff *); +int flow_from_nlattrs(struct sw_flow_key *swkey, const struct nlattr *); static inline struct sw_flow *flow_cast(const struct tbl_node *node) { diff --git a/datapath/odp-compat.h b/datapath/odp-compat.h index ae997e67..b18c31a6 100644 --- a/datapath/odp-compat.h +++ b/datapath/odp-compat.h @@ -15,32 +15,7 @@ #include "openvswitch/datapath-protocol.h" #include -#define ODP_FLOW_GET32 _IOWR('O', 13, struct compat_odp_flow) -#define ODP_FLOW_PUT32 _IOWR('O', 14, struct compat_odp_flow) -#define ODP_FLOW_DUMP32 _IOWR('O', 15, struct compat_odp_flow_dump) -#define ODP_FLOW_DEL32 _IOWR('O', 17, struct compat_odp_flow) #define ODP_EXECUTE32 _IOR('O', 18, struct compat_odp_execute) -#define ODP_FLOW_DEL32 _IOWR('O', 17, struct compat_odp_flow) - -struct compat_odp_flow { - uint32_t dp_idx; - struct odp_flow_stats stats; - compat_uptr_t key; - u32 key_len; - compat_uptr_t actions; - u32 actions_len; -}; - -struct compat_odp_flow_put { - struct compat_odp_flow flow; - u32 flags; -}; - -struct compat_odp_flow_dump { - uint32_t dp_idx; - compat_uptr_t flow; - uint32_t state[2]; -}; struct compat_odp_execute { uint32_t dp_idx; diff --git a/include/openvswitch/datapath-protocol.h b/include/openvswitch/datapath-protocol.h index cf81cb31..23cf4ef8 100644 --- a/include/openvswitch/datapath-protocol.h +++ b/include/openvswitch/datapath-protocol.h @@ -70,14 +70,11 @@ #include #include -#define ODP_MAX 256 /* Maximum number of datapaths. */ - -#define ODP_DP_CREATE _IO('O', 0) -#define ODP_DP_DESTROY _IO('O', 1) -#define ODP_DP_STATS _IOW('O', 2, struct odp_stats) - -#define ODP_GET_DROP_FRAGS _IOW('O', 3, int) -#define ODP_SET_DROP_FRAGS _IOR('O', 4, int) +#define ODP_DP_NEW _IOWR('O', 0, struct odp_datapath) +#define ODP_DP_DEL _IOR('O', 1, struct odp_datapath) +#define ODP_DP_GET _IOWR('O', 2, struct odp_datapath) +#define ODP_DP_SET _IOWR('O', 3, struct odp_datapath) +#define ODP_DP_DUMP _IOWR('O', 4, struct odp_datapath) #define ODP_GET_LISTEN_MASK _IOW('O', 5, int) #define ODP_SET_LISTEN_MASK _IOR('O', 6, int) @@ -88,16 +85,48 @@ #define ODP_VPORT_SET _IOR('O', 22, struct odp_vport) #define ODP_VPORT_DUMP _IOWR('O', 10, struct odp_vport) -#define ODP_FLOW_GET _IOWR('O', 13, struct odp_flow) -#define ODP_FLOW_PUT _IOWR('O', 14, struct odp_flow) -#define ODP_FLOW_DUMP _IOWR('O', 15, struct odp_flow_dump) -#define ODP_FLOW_FLUSH _IO('O', 16) -#define ODP_FLOW_DEL _IOWR('O', 17, struct odp_flow) +#define ODP_FLOW_NEW _IOWR('O', 13, struct odp_flow) +#define ODP_FLOW_DEL _IOWR('O', 14, struct odp_flow) +#define ODP_FLOW_GET _IOWR('O', 15, struct odp_flow) +#define ODP_FLOW_SET _IOWR('O', 16, struct odp_flow) +#define ODP_FLOW_DUMP _IOWR('O', 17, struct odp_flow) +#define ODP_FLOW_FLUSH _IO('O', 19) #define ODP_EXECUTE _IOR('O', 18, struct odp_execute) -#define ODP_SET_SFLOW_PROBABILITY _IOR('O', 19, int) -#define ODP_GET_SFLOW_PROBABILITY _IOW('O', 20, int) +/** + * struct odp_datapath - header with basic information about a datapath. + * @dp_idx: Datapath index (-1 to make a request not specific to a datapath). + * @len: Length of this structure plus the Netlink attributes following it. + * @total_len: Total space available for kernel reply to request. + * + * Followed by &struct nlattr attributes, whose types are drawn from + * %ODP_DP_ATTR_*, up to a length of @len bytes including the &struct + * odp_datapath header. + */ +struct odp_datapath { + int32_t dp_idx; + uint32_t len; + uint32_t total_len; +}; + +enum odp_datapath_type { + ODP_DP_ATTR_UNSPEC, + ODP_DP_ATTR_NAME, /* name of dp_ifidx netdev */ + ODP_DP_ATTR_STATS, /* struct odp_stats */ + ODP_DP_ATTR_IPV4_FRAGS, /* 32-bit enum odp_frag_handling */ + ODP_DP_ATTR_SAMPLING, /* 32-bit fraction of packets to sample. */ + __ODP_DP_ATTR_MAX +}; + +#define ODP_DP_ATTR_MAX (__ODP_DP_ATTR_MAX - 1) + +/* Values for ODP_DP_ATTR_IPV4_FRAGS. */ +enum odp_frag_handling { + ODP_DP_FRAG_UNSPEC, + ODP_DP_FRAG_ZERO, /* Treat IP fragments as transport port 0. */ + ODP_DP_FRAG_DROP /* Drop IP fragments. */ +}; struct odp_stats { uint64_t n_frags; /* Number of dropped IP fragments. */ @@ -210,10 +239,6 @@ enum { struct odp_flow_stats { uint64_t n_packets; /* Number of matched packets. */ uint64_t n_bytes; /* Number of matched bytes. */ - uint64_t used_sec; /* Time last used, in system monotonic time. */ - uint32_t used_nsec; - uint8_t tcp_flags; - uint8_t reserved; }; enum odp_key_type { @@ -271,41 +296,36 @@ struct odp_key_arp { ovs_be16 arp_op; }; +/** + * struct odp_flow - header with basic information about a flow. + * @dp_idx: Datapath index. + * @len: Length of this structure plus the Netlink attributes following it. + * @total_len: Total space available for kernel reply to request. + * + * Followed by &struct nlattr attributes, whose types are drawn from + * %ODP_FLOW_ATTR_*, up to a length of @len bytes including the &struct + * odp_flow header. + */ struct odp_flow { - uint32_t dp_idx; - struct odp_flow_stats stats; - struct nlattr *key; - uint32_t key_len; - struct nlattr *actions; - uint32_t actions_len; + uint32_t nlmsg_flags; + uint32_t dp_idx; + uint32_t len; + uint32_t total_len; }; -/* Flags for ODP_FLOW_PUT. */ -#define ODPPF_CREATE (1 << 0) /* Allow creating a new flow. */ -#define ODPPF_MODIFY (1 << 1) /* Allow modifying an existing flow. */ -#define ODPPF_ZERO_STATS (1 << 2) /* Zero the stats of an existing flow. */ - -/* ODP_FLOW_PUT argument. */ -struct odp_flow_put { - struct odp_flow flow; - uint32_t flags; +enum odp_flow_type { + ODP_FLOW_ATTR_UNSPEC, + ODP_FLOW_ATTR_KEY, /* Sequence of ODP_KEY_ATTR_* attributes. */ + ODP_FLOW_ATTR_ACTIONS, /* Sequence of nested ODPAT_* attributes. */ + ODP_FLOW_ATTR_STATS, /* struct odp_flow_stats. */ + ODP_FLOW_ATTR_TCP_FLAGS, /* 8-bit OR'd TCP flags. */ + ODP_FLOW_ATTR_USED, /* u64 msecs last used in monotonic time. */ + ODP_FLOW_ATTR_CLEAR, /* Flag to clear stats, tcp_flags, used. */ + ODP_FLOW_ATTR_STATE, /* u64 state for ODP_FLOW_DUMP. */ + __ODP_FLOW_ATTR_MAX }; -/* ODP_FLOW_DUMP argument. - * - * This is used to iterate through the flow table flow-by-flow. Each - * ODP_FLOW_DUMP call either stores a new odp_flow into 'flow' or stores 0 into - * flow->key_len to indicate that the end of the table has been reached, and - * updates 'state' in-place. - * - * Before the first call, zero 'state'. The format of 'state' is otherwise - * unspecified. - */ -struct odp_flow_dump { - uint32_t dp_idx; - struct odp_flow *flow; - uint32_t state[2]; -}; +#define ODP_FLOW_ATTR_MAX (__ODP_FLOW_ATTR_MAX - 1) /* Action types. */ enum odp_action_type { diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index 3c11be2c..ea2f424e 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -47,11 +47,64 @@ #include "rtnetlink-link.h" #include "shash.h" #include "svec.h" +#include "unaligned.h" #include "util.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(dpif_linux); +struct dpif_linux_dp { + /* ioctl command argument. */ + int cmd; + + /* struct odp_datapath header. */ + uint32_t dp_idx; + + /* Attributes. */ + const char *name; /* ODP_DP_ATTR_NAME. */ + struct odp_stats stats; /* ODP_DP_ATTR_STATS. */ + enum odp_frag_handling ipv4_frags; /* ODP_DP_ATTR_IPV4_FRAGS. */ + const uint32_t *sampling; /* ODP_DP_ATTR_SAMPLING. */ +}; + +static void dpif_linux_dp_init(struct dpif_linux_dp *); +static int dpif_linux_dp_transact(const struct dpif_linux_dp *request, + struct dpif_linux_dp *reply, + struct ofpbuf **bufp); +static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply, + struct ofpbuf **bufp); + +struct dpif_linux_flow { + /* ioctl command argument. */ + int cmd; + + /* struct odp_flow header. */ + unsigned int nlmsg_flags; + uint32_t dp_idx; + + /* Attributes. + * + * The 'stats', 'used', and 'state' members point to 64-bit data that might + * only be aligned on 32-bit boundaries, so get_unaligned_u64() should be + * used to access their values. */ + const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */ + size_t key_len; + const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */ + size_t actions_len; + const struct odp_flow_stats *stats; /* ODP_FLOW_ATTR_STATS. */ + const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */ + const uint64_t *used; /* ODP_FLOW_ATTR_USED. */ + bool clear; /* ODP_FLOW_ATTR_CLEAR. */ + const uint64_t *state; /* ODP_FLOW_ATTR_STATE. */ +}; + +static void dpif_linux_flow_init(struct dpif_linux_flow *); +static int dpif_linux_flow_transact(const struct dpif_linux_flow *request, + struct dpif_linux_flow *reply, + struct ofpbuf **bufp); +static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *, + struct dpif_flow_stats *); + /* Datapath interface for the openvswitch Linux kernel module. */ struct dpif_linux { struct dpif dpif; @@ -74,7 +127,6 @@ static int do_ioctl(const struct dpif *, int cmd, const void *arg); static int open_dpif(const struct dpif_linux_vport *local_vport, struct dpif **); static int get_openvswitch_major(void); -static int create_minor(const char *name, int minor); static int open_minor(int minor, int *fdp); static int make_openvswitch_device(int minor, char **fnp); static void dpif_linux_port_changed(const struct rtnetlink_link_change *, @@ -90,9 +142,8 @@ dpif_linux_cast(const struct dpif *dpif) static int dpif_linux_enumerate(struct svec *all_dps) { + uint32_t dp_idx; int major; - int error; - int i; /* Check that the Open vSwitch module is loaded. */ major = get_openvswitch_major(); @@ -100,22 +151,28 @@ dpif_linux_enumerate(struct svec *all_dps) return -major; } - error = 0; - for (i = 0; i < ODP_MAX; i++) { - struct dpif *dpif; + dp_idx = 0; + for (;;) { + struct dpif_linux_dp request, reply; + struct ofpbuf *buf; char devname[16]; - int retval; - - sprintf(devname, "dp%d", i); - retval = dpif_open(devname, "system", &dpif); - if (!retval) { - svec_add(all_dps, devname); - dpif_uninit(dpif, true); - } else if (retval != ENODEV && !error) { - error = retval; + int error; + + dpif_linux_dp_init(&request); + request.dp_idx = dp_idx; + request.cmd = ODP_DP_DUMP; + + error = dpif_linux_dp_transact(&request, &reply, &buf); + if (error) { + return error == ENODEV ? 0 : error; } + ofpbuf_delete(buf); + + sprintf(devname, "dp%d", reply.dp_idx); + svec_add(all_dps, devname); + + dp_idx = reply.dp_idx + 1; } - return error; } static int @@ -130,27 +187,20 @@ dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, minor = !strncmp(name, "dp", 2) && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1; if (create) { - if (minor >= 0) { - error = create_minor(name, minor); - if (error) { - return error; - } - } else { - /* Scan for unused minor number. */ - for (minor = 0; ; minor++) { - if (minor >= ODP_MAX) { - /* All datapath numbers in use. */ - return ENOBUFS; - } + struct dpif_linux_dp request, reply; + struct ofpbuf *buf; + int error; - error = create_minor(name, minor); - if (!error) { - break; - } else if (error != EBUSY) { - return error; - } - } + dpif_linux_dp_init(&request); + request.cmd = ODP_DP_NEW; + request.dp_idx = minor; + request.name = name; + error = dpif_linux_dp_transact(&request, &reply, &buf); + if (error) { + return error; } + minor = reply.dp_idx; + ofpbuf_delete(buf); } dpif_linux_vport_init(&request); @@ -245,25 +295,41 @@ dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names) static int dpif_linux_destroy(struct dpif *dpif_) { - return do_ioctl(dpif_, ODP_DP_DESTROY, NULL); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp dp; + + dpif_linux_dp_init(&dp); + dp.cmd = ODP_DP_DEL; + dp.dp_idx = dpif->minor; + return dpif_linux_dp_transact(&dp, NULL, NULL); } static int dpif_linux_get_stats(const struct dpif *dpif_, struct odp_stats *stats) { - memset(stats, 0, sizeof *stats); - return do_ioctl(dpif_, ODP_DP_STATS, stats); + struct dpif_linux_dp dp; + struct ofpbuf *buf; + int error; + + error = dpif_linux_dp_get(dpif_, &dp, &buf); + if (!error) { + *stats = dp.stats; + ofpbuf_delete(buf); + } + return error; } static int dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) { - int drop_frags; + struct dpif_linux_dp dp; + struct ofpbuf *buf; int error; - error = do_ioctl(dpif_, ODP_GET_DROP_FRAGS, &drop_frags); + error = dpif_linux_dp_get(dpif_, &dp, &buf); if (!error) { - *drop_fragsp = drop_frags & 1; + *drop_fragsp = dp.ipv4_frags == ODP_DP_FRAG_DROP; + ofpbuf_delete(buf); } return error; } @@ -271,8 +337,14 @@ dpif_linux_get_drop_frags(const struct dpif *dpif_, bool *drop_fragsp) static int dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) { - int drop_frags_int = drop_frags; - return do_ioctl(dpif_, ODP_SET_DROP_FRAGS, &drop_frags_int); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp dp; + + dpif_linux_dp_init(&dp); + dp.cmd = ODP_DP_SET; + dp.dp_idx = dpif->minor; + dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO; + return dpif_linux_dp_transact(&dp, NULL, NULL); } static int @@ -460,48 +532,32 @@ dpif_linux_port_poll_wait(const struct dpif *dpif_) } } -static void -odp_flow_stats_to_dpif_flow_stats(const struct odp_flow_stats *ofs, - struct dpif_flow_stats *dfs) -{ - dfs->n_packets = ofs->n_packets; - dfs->n_bytes = ofs->n_bytes; - dfs->used = ofs->used_sec * 1000 + ofs->used_nsec / 1000000; - dfs->tcp_flags = ofs->tcp_flags; -} - static int dpif_linux_flow_get(const struct dpif *dpif_, const struct nlattr *key, size_t key_len, struct ofpbuf **actionsp, struct dpif_flow_stats *stats) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct ofpbuf *actions = NULL; - struct odp_flow odp_flow; + struct dpif_linux_flow request, reply; + struct ofpbuf *buf; int error; - memset(&odp_flow, 0, sizeof odp_flow); - odp_flow.dp_idx = dpif->minor; - odp_flow.key = (struct nlattr *) key; - odp_flow.key_len = key_len; - if (actionsp) { - actions = *actionsp = ofpbuf_new(65536); - odp_flow.actions = actions->base; - odp_flow.actions_len = actions->allocated; - } - - error = do_ioctl(dpif_, ODP_FLOW_GET, &odp_flow); + dpif_linux_flow_init(&request); + request.cmd = ODP_FLOW_GET; + request.dp_idx = dpif->minor; + request.key = key; + request.key_len = key_len; + error = dpif_linux_flow_transact(&request, &reply, &buf); if (!error) { if (stats) { - odp_flow_stats_to_dpif_flow_stats(&odp_flow.stats, stats); - } - if (actions) { - actions->size = odp_flow.actions_len; - ofpbuf_trim(actions); + dpif_linux_flow_get_stats(&reply, stats); } - } else { - if (actions) { - ofpbuf_delete(actions); + if (actionsp) { + buf->data = (void *) reply.actions; + buf->size = reply.actions_len; + *actionsp = buf; + } else { + ofpbuf_delete(buf); } } return error; @@ -514,28 +570,27 @@ dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, struct dpif_flow_stats *stats) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct odp_flow_put put; + struct dpif_linux_flow request, reply; + struct ofpbuf *buf; int error; - memset(&put, 0, sizeof put); - put.flow.dp_idx = dpif->minor; - put.flow.key = (struct nlattr *) key; - put.flow.key_len = key_len; - put.flow.actions = (struct nlattr *) actions; - put.flow.actions_len = actions_len; - put.flags = 0; - if (flags & DPIF_FP_CREATE) { - put.flags |= ODPPF_CREATE; - } - if (flags & DPIF_FP_MODIFY) { - put.flags |= ODPPF_MODIFY; - } + dpif_linux_flow_init(&request); + request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_NEW : ODP_FLOW_SET; + request.dp_idx = dpif->minor; + request.key = key; + request.key_len = key_len; + request.actions = actions; + request.actions_len = actions_len; if (flags & DPIF_FP_ZERO_STATS) { - put.flags |= ODPPF_ZERO_STATS; + request.clear = true; } - error = do_ioctl(dpif_, ODP_FLOW_PUT, &put); + request.nlmsg_flags = flags & DPIF_FP_MODIFY ? 0 : NLM_F_CREATE; + error = dpif_linux_flow_transact(&request, + stats ? &reply : NULL, + stats ? &buf : NULL); if (!error && stats) { - odp_flow_stats_to_dpif_flow_stats(&put.flow.stats, stats); + dpif_linux_flow_get_stats(&reply, stats); + ofpbuf_delete(buf); } return error; } @@ -546,84 +601,81 @@ dpif_linux_flow_del(struct dpif *dpif_, struct dpif_flow_stats *stats) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct odp_flow odp_flow; + struct dpif_linux_flow request, reply; + struct ofpbuf *buf; int error; - memset(&odp_flow, 0, sizeof odp_flow); - odp_flow.dp_idx = dpif->minor; - odp_flow.key = (struct nlattr *) key; - odp_flow.key_len = key_len; - error = do_ioctl(dpif_, ODP_FLOW_DEL, &odp_flow); + dpif_linux_flow_init(&request); + request.cmd = ODP_FLOW_DEL; + request.dp_idx = dpif->minor; + request.key = key; + request.key_len = key_len; + error = dpif_linux_flow_transact(&request, + stats ? &reply : NULL, + stats ? &buf : NULL); if (!error && stats) { - odp_flow_stats_to_dpif_flow_stats(&odp_flow.stats, stats); + dpif_linux_flow_get_stats(&reply, stats); + ofpbuf_delete(buf); } return error; } + struct dpif_linux_flow_state { - struct odp_flow_dump dump; - struct odp_flow flow; - uint32_t keybuf[ODPUTIL_FLOW_KEY_U32S]; - uint32_t actionsbuf[65536 / sizeof(uint32_t)]; + struct dpif_linux_flow flow; + struct ofpbuf *buf; struct dpif_flow_stats stats; }; static int -dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) +dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_flow_state *state; - - *statep = state = xmalloc(sizeof *state); - state->dump.dp_idx = dpif->minor; - state->dump.state[0] = 0; - state->dump.state[1] = 0; - state->dump.flow = &state->flow; + *statep = xzalloc(sizeof(struct dpif_linux_flow_state)); return 0; } static int -dpif_linux_flow_dump_next(const struct dpif *dpif, void *state_, +dpif_linux_flow_dump_next(const struct dpif *dpif_, void *state_, const struct nlattr **key, size_t *key_len, const struct nlattr **actions, size_t *actions_len, const struct dpif_flow_stats **stats) { + struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow_state *state = state_; + struct ofpbuf *old_buf = state->buf; + struct dpif_linux_flow request; int error; - memset(&state->flow, 0, sizeof state->flow); - state->flow.key = (struct nlattr *) state->keybuf; - state->flow.key_len = sizeof state->keybuf; - if (actions) { - state->flow.actions = (struct nlattr *) state->actionsbuf; - state->flow.actions_len = sizeof state->actionsbuf; - } + dpif_linux_flow_init(&request); + request.cmd = ODP_FLOW_DUMP; + request.dp_idx = dpif->minor; + request.state = state->flow.state; + error = dpif_linux_flow_transact(&request, &state->flow, &state->buf); + ofpbuf_delete(old_buf); - error = do_ioctl(dpif, ODP_FLOW_DUMP, &state->dump); if (!error) { - if (!state->flow.key_len) { - return EOF; - } if (key) { - *key = (const struct nlattr *) state->keybuf; + *key = state->flow.key; *key_len = state->flow.key_len; } if (actions) { - *actions = (const struct nlattr *) state->actionsbuf; + *actions = state->flow.actions; *actions_len = state->flow.actions_len; } if (stats) { - odp_flow_stats_to_dpif_flow_stats(&state->flow.stats, - &state->stats); + dpif_linux_flow_get_stats(&state->flow, &state->stats); *stats = &state->stats; } } - return error; + return error == ENODEV ? EOF : error; } static int -dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state) +dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) { + struct dpif_linux_flow_state *state = state_; + + ofpbuf_delete(state->buf); free(state); return 0; } @@ -661,13 +713,29 @@ static int dpif_linux_get_sflow_probability(const struct dpif *dpif_, uint32_t *probability) { - return do_ioctl(dpif_, ODP_GET_SFLOW_PROBABILITY, probability); + struct dpif_linux_dp dp; + struct ofpbuf *buf; + int error; + + error = dpif_linux_dp_get(dpif_, &dp, &buf); + if (!error) { + *probability = dp.sampling ? *dp.sampling : 0; + ofpbuf_delete(buf); + } + return error; } static int dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) { - return do_ioctl(dpif_, ODP_SET_SFLOW_PROBABILITY, &probability); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp dp; + + dpif_linux_dp_init(&dp); + dp.cmd = ODP_DP_SET; + dp.dp_idx = dpif->minor; + dp.sampling = &probability; + return dpif_linux_dp_transact(&dp, NULL, NULL); } static int @@ -985,22 +1053,6 @@ get_major(const char *target) return -ENODEV; } -static int -create_minor(const char *name, int minor) -{ - int error; - int fd; - - error = open_minor(minor, &fd); - if (error) { - return error; - } - - error = ioctl(fd, ODP_DP_CREATE, name) ? errno : 0; - close(fd); - return error; -} - static int open_minor(int minor, int *fdp) { @@ -1042,6 +1094,24 @@ dpif_linux_port_changed(const struct rtnetlink_link_change *change, dpif->change_error = true; } } + +static int +get_dp0_fd(int *dp0_fdp) +{ + static int dp0_fd = -1; + if (dp0_fd < 0) { + int error; + int fd; + + error = open_minor(0, &fd); + if (error) { + return error; + } + dp0_fd = fd; + } + *dp0_fdp = dp0_fd; + return 0; +} /* Parses the contents of 'buf', which contains a "struct odp_vport" followed * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a @@ -1183,25 +1253,21 @@ dpif_linux_vport_transact(const struct dpif_linux_vport *request, struct dpif_linux_vport *reply, struct ofpbuf **bufp) { - static int dp0_fd = -1; struct ofpbuf *buf = NULL; int error; + int fd; assert((reply != NULL) == (bufp != NULL)); - if (dp0_fd < 0) { - int fd; - error = open_minor(0, &fd); - if (error) { - goto error; - } - dp0_fd = fd; + error = get_dp0_fd(&fd); + if (error) { + goto error; } buf = ofpbuf_new(1024); dpif_linux_vport_to_ofpbuf(request, buf); - error = ioctl(dp0_fd, request->cmd, buf->data) ? errno : 0; + error = ioctl(fd, request->cmd, buf->data) ? errno : 0; if (error) { goto error; } @@ -1242,4 +1308,322 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, return dpif_linux_vport_transact(&request, reply, bufp); } + +/* Parses the contents of 'buf', which contains a "struct odp_datapath" + * followed by Netlink attributes, into 'dp'. Returns 0 if successful, + * otherwise a positive errno value. + * + * 'dp' will contain pointers into 'buf', so the caller should not free 'buf' + * while 'dp' is still in use. */ +static int +dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) +{ + static const struct nl_policy odp_datapath_policy[] = { + [ODP_DP_ATTR_NAME] = { .type = NL_A_STRING, .max_len = IFNAMSIZ }, + [ODP_DP_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct odp_stats), + .max_len = sizeof(struct odp_stats), + .optional = true }, + [ODP_DP_ATTR_IPV4_FRAGS] = { .type = NL_A_U32, .optional = true }, + [ODP_DP_ATTR_SAMPLING] = { .type = NL_A_U32, .optional = true }, + }; + + struct odp_datapath *odp_dp; + struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)]; + + dpif_linux_dp_init(dp); + + if (!nl_policy_parse(buf, sizeof *odp_dp, odp_datapath_policy, + a, ARRAY_SIZE(odp_datapath_policy))) { + return EINVAL; + } + odp_dp = buf->data; + + dp->dp_idx = odp_dp->dp_idx; + dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]); + if (a[ODP_DP_ATTR_STATS]) { + /* Can't use structure assignment because Netlink doesn't ensure + * sufficient alignment for 64-bit members. */ + memcpy(&dp->stats, nl_attr_get(a[ODP_DP_ATTR_STATS]), + sizeof dp->stats); + } + if (a[ODP_DP_ATTR_IPV4_FRAGS]) { + dp->ipv4_frags = nl_attr_get_u32(a[ODP_DP_ATTR_IPV4_FRAGS]); + } + if (a[ODP_DP_ATTR_SAMPLING]) { + dp->sampling = nl_attr_get(a[ODP_DP_ATTR_SAMPLING]); + } + return 0; +} + +/* Appends to 'buf' (which must initially be empty) a "struct odp_datapath" + * followed by Netlink attributes corresponding to 'dp'. */ +static void +dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) +{ + struct odp_datapath *odp_dp; + + ofpbuf_reserve(buf, sizeof odp_dp); + + if (dp->name) { + nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name); + } + + /* Skip ODP_DP_ATTR_STATS since we never have a reason to serialize it. */ + + if (dp->ipv4_frags) { + nl_msg_put_u32(buf, ODP_DP_ATTR_IPV4_FRAGS, dp->ipv4_frags); + } + + if (dp->sampling) { + nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling); + } + + odp_dp = ofpbuf_push_uninit(buf, sizeof *odp_dp); + odp_dp->dp_idx = dp->dp_idx; + odp_dp->len = buf->size; + odp_dp->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data; +} + +/* Clears 'dp' to "empty" values. */ +void +dpif_linux_dp_init(struct dpif_linux_dp *dp) +{ + memset(dp, 0, sizeof *dp); + dp->dp_idx = -1; +} + +/* Executes 'request' in the kernel datapath. If the command fails, returns a + * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 + * without doing anything else. If 'reply' and 'bufp' are nonnull, then the + * result of the command is expected to be an odp_datapath also, which is + * decoded and stored in '*reply' and '*bufp'. The caller must free '*bufp' + * when the reply is no longer needed ('reply' will contain pointers into + * '*bufp'). */ +int +dpif_linux_dp_transact(const struct dpif_linux_dp *request, + struct dpif_linux_dp *reply, struct ofpbuf **bufp) +{ + struct ofpbuf *buf = NULL; + int error; + int fd; + + assert((reply != NULL) == (bufp != NULL)); + + error = get_dp0_fd(&fd); + if (error) { + goto error; + } + + buf = ofpbuf_new(1024); + dpif_linux_dp_to_ofpbuf(request, buf); + + error = ioctl(fd, request->cmd, buf->data) ? errno : 0; + if (error) { + goto error; + } + + if (bufp) { + buf->size = ((struct odp_datapath *) buf->data)->len; + error = dpif_linux_dp_from_ofpbuf(reply, buf); + if (error) { + goto error; + } + *bufp = buf; + } else { + ofpbuf_delete(buf); + } + return 0; + +error: + ofpbuf_delete(buf); + if (bufp) { + memset(reply, 0, sizeof *reply); + *bufp = NULL; + } + return error; +} + +/* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'. + * The caller must free '*bufp' when the reply is no longer needed ('reply' + * will contain pointers into '*bufp'). */ +int +dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, + struct ofpbuf **bufp) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_dp request; + + dpif_linux_dp_init(&request); + request.cmd = ODP_DP_GET; + request.dp_idx = dpif->minor; + + return dpif_linux_dp_transact(&request, reply, bufp); +} + +/* Parses the contents of 'buf', which contains a "struct odp_flow" followed by + * Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a + * positive errno value. + * + * 'flow' will contain pointers into 'buf', so the caller should not free 'buf' + * while 'flow' is still in use. */ +static int +dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, + const struct ofpbuf *buf) +{ + static const struct nl_policy odp_flow_policy[] = { + [ODP_FLOW_ATTR_KEY] = { .type = NL_A_NESTED }, + [ODP_FLOW_ATTR_ACTIONS] = { .type = NL_A_NESTED, .optional = true }, + [ODP_FLOW_ATTR_STATS] = { .type = NL_A_UNSPEC, + .min_len = sizeof(struct odp_flow_stats), + .max_len = sizeof(struct odp_flow_stats), + .optional = true }, + [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, + [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, + /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */ + [ODP_FLOW_ATTR_STATE] = { .type = NL_A_U64, .optional = true }, + }; + + struct odp_flow *odp_flow; + struct nlattr *a[ARRAY_SIZE(odp_flow_policy)]; + + dpif_linux_flow_init(flow); + + if (!nl_policy_parse(buf, sizeof *odp_flow, odp_flow_policy, + a, ARRAY_SIZE(odp_flow_policy))) { + return EINVAL; + } + odp_flow = buf->data; + + flow->nlmsg_flags = odp_flow->nlmsg_flags; + flow->dp_idx = odp_flow->dp_idx; + flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]); + flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]); + if (a[ODP_FLOW_ATTR_ACTIONS]) { + flow->actions = nl_attr_get(a[ODP_FLOW_ATTR_ACTIONS]); + flow->actions_len = nl_attr_get_size(a[ODP_FLOW_ATTR_ACTIONS]); + } + if (a[ODP_FLOW_ATTR_STATS]) { + flow->stats = nl_attr_get(a[ODP_FLOW_ATTR_STATS]); + } + if (a[ODP_FLOW_ATTR_TCP_FLAGS]) { + flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]); + } + if (a[ODP_FLOW_ATTR_STATE]) { + flow->state = nl_attr_get(a[ODP_FLOW_ATTR_STATE]); + } + return 0; +} + +/* Appends to 'buf' (which must initially be empty) a "struct odp_flow" + * followed by Netlink attributes corresponding to 'flow'. */ +static void +dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, + struct ofpbuf *buf) +{ + struct odp_flow *odp_flow; + + ofpbuf_reserve(buf, sizeof odp_flow); + + if (flow->key_len) { + nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len); + } + + if (flow->actions_len) { + nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS, + flow->actions, flow->actions_len); + } + + /* We never need to send these to the kernel. */ + assert(!flow->stats); + assert(!flow->tcp_flags); + assert(!flow->used); + + if (flow->clear) { + nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR); + } + + if (flow->state) { + nl_msg_put_u64(buf, ODP_FLOW_ATTR_STATE, + get_unaligned_u64(flow->state)); + } + + odp_flow = ofpbuf_push_uninit(buf, sizeof *odp_flow); + odp_flow->nlmsg_flags = flow->nlmsg_flags; + odp_flow->dp_idx = flow->dp_idx; + odp_flow->len = buf->size; + odp_flow->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data; +} + +/* Clears 'flow' to "empty" values. */ +void +dpif_linux_flow_init(struct dpif_linux_flow *flow) +{ + memset(flow, 0, sizeof *flow); +} + +/* Executes 'request' in the kernel datapath. If the command fails, returns a + * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 + * without doing anything else. If 'reply' and 'bufp' are nonnull, then the + * result of the command is expected to be an odp_flow also, which is decoded + * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the + * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ +int +dpif_linux_flow_transact(const struct dpif_linux_flow *request, + struct dpif_linux_flow *reply, struct ofpbuf **bufp) +{ + struct ofpbuf *buf = NULL; + int error; + int fd; + + assert((reply != NULL) == (bufp != NULL)); + + error = get_dp0_fd(&fd); + if (error) { + goto error; + } + + buf = ofpbuf_new(1024); + dpif_linux_flow_to_ofpbuf(request, buf); + + error = ioctl(fd, request->cmd, buf->data) ? errno : 0; + if (error) { + goto error; + } + + if (bufp) { + buf->size = ((struct odp_flow *) buf->data)->len; + error = dpif_linux_flow_from_ofpbuf(reply, buf); + if (error) { + goto error; + } + *bufp = buf; + } else { + ofpbuf_delete(buf); + } + return 0; + +error: + ofpbuf_delete(buf); + if (bufp) { + memset(reply, 0, sizeof *reply); + *bufp = NULL; + } + return error; +} + +static void +dpif_linux_flow_get_stats(const struct dpif_linux_flow *flow, + struct dpif_flow_stats *stats) +{ + if (flow->stats) { + stats->n_packets = get_unaligned_u64(&flow->stats->n_packets); + stats->n_bytes = get_unaligned_u64(&flow->stats->n_bytes); + } else { + stats->n_packets = 0; + stats->n_bytes = 0; + } + stats->used = flow->used ? get_unaligned_u64(flow->used) : 0; + stats->tcp_flags = flow->tcp_flags ? *flow->tcp_flags : 0; +} -- 2.30.2