1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2014 Nicira, Inc.
6 #include <linux/etherdevice.h>
8 #include <linux/if_vlan.h>
9 #include <linux/jhash.h>
10 #include <linux/kernel.h>
11 #include <linux/list.h>
12 #include <linux/mutex.h>
13 #include <linux/percpu.h>
14 #include <linux/rcupdate.h>
15 #include <linux/rtnetlink.h>
16 #include <linux/compat.h>
17 #include <net/net_namespace.h>
18 #include <linux/module.h>
22 #include "vport-internal_dev.h"
24 static LIST_HEAD(vport_ops_list);
26 /* Protected by RCU read lock for reading, ovs_mutex for writing. */
27 static struct hlist_head *dev_table;
28 #define VPORT_HASH_BUCKETS 1024
31 * ovs_vport_init - initialize vport subsystem
33 * Called at module load time to initialize the vport subsystem.
35 int ovs_vport_init(void)
37 dev_table = kcalloc(VPORT_HASH_BUCKETS, sizeof(struct hlist_head),
46 * ovs_vport_exit - shutdown vport subsystem
48 * Called at module exit time to shutdown the vport subsystem.
50 void ovs_vport_exit(void)
55 static struct hlist_head *hash_bucket(const struct net *net, const char *name)
57 unsigned int hash = jhash(name, strlen(name), (unsigned long) net);
58 return &dev_table[hash & (VPORT_HASH_BUCKETS - 1)];
61 int __ovs_vport_ops_register(struct vport_ops *ops)
67 list_for_each_entry(o, &vport_ops_list, list)
68 if (ops->type == o->type)
71 list_add_tail(&ops->list, &vport_ops_list);
77 EXPORT_SYMBOL_GPL(__ovs_vport_ops_register);
79 void ovs_vport_ops_unregister(struct vport_ops *ops)
85 EXPORT_SYMBOL_GPL(ovs_vport_ops_unregister);
88 * ovs_vport_locate - find a port that has already been created
90 * @name: name of port to find
92 * Must be called with ovs or RCU read lock.
94 struct vport *ovs_vport_locate(const struct net *net, const char *name)
96 struct hlist_head *bucket = hash_bucket(net, name);
99 hlist_for_each_entry_rcu(vport, bucket, hash_node)
100 if (!strcmp(name, ovs_vport_name(vport)) &&
101 net_eq(ovs_dp_get_net(vport->dp), net))
108 * ovs_vport_alloc - allocate and initialize new vport
110 * @priv_size: Size of private data area to allocate.
111 * @ops: vport device ops
113 * Allocate and initialize a new vport defined by @ops. The vport will contain
114 * a private data area of size @priv_size that can be accessed using
115 * vport_priv(). vports that are no longer needed should be released with
118 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
119 const struct vport_parms *parms)
124 alloc_size = sizeof(struct vport);
126 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
127 alloc_size += priv_size;
130 vport = kzalloc(alloc_size, GFP_KERNEL);
132 return ERR_PTR(-ENOMEM);
134 vport->dp = parms->dp;
135 vport->port_no = parms->port_no;
137 INIT_HLIST_NODE(&vport->dp_hash_node);
139 if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
141 return ERR_PTR(-EINVAL);
146 EXPORT_SYMBOL_GPL(ovs_vport_alloc);
149 * ovs_vport_free - uninitialize and free vport
151 * @vport: vport to free
153 * Frees a vport allocated with vport_alloc() when it is no longer needed.
155 * The caller must ensure that an RCU grace period has passed since the last
156 * time @vport was in a datapath.
158 void ovs_vport_free(struct vport *vport)
160 /* vport is freed from RCU callback or error path, Therefore
161 * it is safe to use raw dereference.
163 kfree(rcu_dereference_raw(vport->upcall_portids));
166 EXPORT_SYMBOL_GPL(ovs_vport_free);
168 static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
170 struct vport_ops *ops;
172 list_for_each_entry(ops, &vport_ops_list, list)
173 if (ops->type == parms->type)
180 * ovs_vport_add - add vport device (for kernel callers)
182 * @parms: Information about new vport.
184 * Creates a new vport with the specified configuration (which is dependent on
185 * device type). ovs_mutex must be held.
187 struct vport *ovs_vport_add(const struct vport_parms *parms)
189 struct vport_ops *ops;
192 ops = ovs_vport_lookup(parms);
194 struct hlist_head *bucket;
196 if (!try_module_get(ops->owner))
197 return ERR_PTR(-EAFNOSUPPORT);
199 vport = ops->create(parms);
201 module_put(ops->owner);
205 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
206 ovs_vport_name(vport));
207 hlist_add_head_rcu(&vport->hash_node, bucket);
211 /* Unlock to attempt module load and return -EAGAIN if load
212 * was successful as we need to restart the port addition
216 request_module("vport-type-%d", parms->type);
219 if (!ovs_vport_lookup(parms))
220 return ERR_PTR(-EAFNOSUPPORT);
222 return ERR_PTR(-EAGAIN);
226 * ovs_vport_set_options - modify existing vport device (for kernel callers)
228 * @vport: vport to modify.
229 * @options: New configuration.
231 * Modifies an existing device with the specified configuration (which is
232 * dependent on device type). ovs_mutex must be held.
234 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
236 if (!vport->ops->set_options)
238 return vport->ops->set_options(vport, options);
242 * ovs_vport_del - delete existing vport device
244 * @vport: vport to delete.
246 * Detaches @vport from its datapath and destroys it. ovs_mutex must
249 void ovs_vport_del(struct vport *vport)
251 hlist_del_rcu(&vport->hash_node);
252 module_put(vport->ops->owner);
253 vport->ops->destroy(vport);
257 * ovs_vport_get_stats - retrieve device stats
259 * @vport: vport from which to retrieve the stats
260 * @stats: location to store stats
262 * Retrieves transmit, receive, and error stats for the given device.
264 * Must be called with ovs_mutex or rcu_read_lock.
266 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
268 const struct rtnl_link_stats64 *dev_stats;
269 struct rtnl_link_stats64 temp;
271 dev_stats = dev_get_stats(vport->dev, &temp);
272 stats->rx_errors = dev_stats->rx_errors;
273 stats->tx_errors = dev_stats->tx_errors;
274 stats->tx_dropped = dev_stats->tx_dropped;
275 stats->rx_dropped = dev_stats->rx_dropped;
277 stats->rx_bytes = dev_stats->rx_bytes;
278 stats->rx_packets = dev_stats->rx_packets;
279 stats->tx_bytes = dev_stats->tx_bytes;
280 stats->tx_packets = dev_stats->tx_packets;
284 * ovs_vport_get_options - retrieve device options
286 * @vport: vport from which to retrieve the options.
287 * @skb: sk_buff where options should be appended.
289 * Retrieves the configuration of the given device, appending an
290 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
291 * vport-specific attributes to @skb.
293 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
294 * negative error code if a real error occurred. If an error occurs, @skb is
297 * Must be called with ovs_mutex or rcu_read_lock.
299 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
304 if (!vport->ops->get_options)
307 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
311 err = vport->ops->get_options(vport, skb);
313 nla_nest_cancel(skb, nla);
317 nla_nest_end(skb, nla);
322 * ovs_vport_set_upcall_portids - set upcall portids of @vport.
324 * @vport: vport to modify.
325 * @ids: new configuration, an array of port ids.
327 * Sets the vport's upcall_portids to @ids.
329 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
330 * as an array of U32.
332 * Must be called with ovs_mutex.
334 int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
336 struct vport_portids *old, *vport_portids;
338 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
341 old = ovsl_dereference(vport->upcall_portids);
343 vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
348 vport_portids->n_ids = nla_len(ids) / sizeof(u32);
349 vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
350 nla_memcpy(vport_portids->ids, ids, nla_len(ids));
352 rcu_assign_pointer(vport->upcall_portids, vport_portids);
360 * ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
362 * @vport: vport from which to retrieve the portids.
363 * @skb: sk_buff where portids should be appended.
365 * Retrieves the configuration of the given vport, appending the
366 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
369 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
370 * If an error occurs, @skb is left unmodified. Must be called with
371 * ovs_mutex or rcu_read_lock.
373 int ovs_vport_get_upcall_portids(const struct vport *vport,
376 struct vport_portids *ids;
378 ids = rcu_dereference_ovsl(vport->upcall_portids);
380 if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
381 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
382 ids->n_ids * sizeof(u32), (void *)ids->ids);
384 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
388 * ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
390 * @vport: vport from which the missed packet is received.
391 * @skb: skb that the missed packet was received.
393 * Uses the skb_get_hash() to select the upcall portid to send the
396 * Returns the portid of the target socket. Must be called with rcu_read_lock.
398 u32 ovs_vport_find_upcall_portid(const struct vport *vport, struct sk_buff *skb)
400 struct vport_portids *ids;
404 ids = rcu_dereference(vport->upcall_portids);
406 if (ids->n_ids == 1 && ids->ids[0] == 0)
409 hash = skb_get_hash(skb);
410 ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
411 return ids->ids[ids_index];
415 * ovs_vport_receive - pass up received packet to the datapath for processing
417 * @vport: vport that received the packet
418 * @skb: skb that was received
419 * @tun_key: tunnel (if any) that carried packet
421 * Must be called with rcu_read_lock. The packet cannot be shared and
422 * skb->data should point to the Ethernet header.
424 int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
425 const struct ip_tunnel_info *tun_info)
427 struct sw_flow_key key;
430 OVS_CB(skb)->input_vport = vport;
431 OVS_CB(skb)->mru = 0;
432 OVS_CB(skb)->cutlen = 0;
433 if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
437 skb_scrub_packet(skb, true);
442 /* Extract flow from 'skb' into 'key'. */
443 error = ovs_flow_key_extract(tun_info, skb, &key);
444 if (unlikely(error)) {
448 ovs_dp_process_packet(skb, &key);
452 static int packet_length(const struct sk_buff *skb,
453 struct net_device *dev)
455 int length = skb->len - dev->hard_header_len;
457 if (!skb_vlan_tag_present(skb) &&
458 eth_type_vlan(skb->protocol))
461 /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
462 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
463 * account for 802.1ad. e.g. is_skb_forwardable().
466 return length > 0 ? length : 0;
469 void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
471 int mtu = vport->dev->mtu;
473 switch (vport->dev->type) {
475 if (mac_proto == MAC_PROTO_ETHERNET) {
476 skb_reset_network_header(skb);
477 skb_reset_mac_len(skb);
478 skb->protocol = htons(ETH_P_TEB);
479 } else if (mac_proto != MAC_PROTO_NONE) {
485 if (mac_proto != MAC_PROTO_ETHERNET)
492 if (unlikely(packet_length(skb, vport->dev) > mtu &&
494 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
496 packet_length(skb, vport->dev), mtu);
497 vport->dev->stats.tx_errors++;
501 skb->dev = vport->dev;
502 vport->ops->send(skb);