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 * @net: network namespace
91 * @name: name of port to find
93 * Must be called with ovs or RCU read lock.
95 struct vport *ovs_vport_locate(const struct net *net, const char *name)
97 struct hlist_head *bucket = hash_bucket(net, name);
100 hlist_for_each_entry_rcu(vport, bucket, hash_node,
101 lockdep_ovsl_is_held())
102 if (!strcmp(name, ovs_vport_name(vport)) &&
103 net_eq(ovs_dp_get_net(vport->dp), net))
110 * ovs_vport_alloc - allocate and initialize new vport
112 * @priv_size: Size of private data area to allocate.
113 * @ops: vport device ops
115 * Allocate and initialize a new vport defined by @ops. The vport will contain
116 * a private data area of size @priv_size that can be accessed using
117 * vport_priv(). vports that are no longer needed should be released with
120 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
121 const struct vport_parms *parms)
126 alloc_size = sizeof(struct vport);
128 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
129 alloc_size += priv_size;
132 vport = kzalloc(alloc_size, GFP_KERNEL);
134 return ERR_PTR(-ENOMEM);
136 vport->dp = parms->dp;
137 vport->port_no = parms->port_no;
139 INIT_HLIST_NODE(&vport->dp_hash_node);
141 if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
143 return ERR_PTR(-EINVAL);
148 EXPORT_SYMBOL_GPL(ovs_vport_alloc);
151 * ovs_vport_free - uninitialize and free vport
153 * @vport: vport to free
155 * Frees a vport allocated with vport_alloc() when it is no longer needed.
157 * The caller must ensure that an RCU grace period has passed since the last
158 * time @vport was in a datapath.
160 void ovs_vport_free(struct vport *vport)
162 /* vport is freed from RCU callback or error path, Therefore
163 * it is safe to use raw dereference.
165 kfree(rcu_dereference_raw(vport->upcall_portids));
168 EXPORT_SYMBOL_GPL(ovs_vport_free);
170 static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
172 struct vport_ops *ops;
174 list_for_each_entry(ops, &vport_ops_list, list)
175 if (ops->type == parms->type)
182 * ovs_vport_add - add vport device (for kernel callers)
184 * @parms: Information about new vport.
186 * Creates a new vport with the specified configuration (which is dependent on
187 * device type). ovs_mutex must be held.
189 struct vport *ovs_vport_add(const struct vport_parms *parms)
191 struct vport_ops *ops;
194 ops = ovs_vport_lookup(parms);
196 struct hlist_head *bucket;
198 if (!try_module_get(ops->owner))
199 return ERR_PTR(-EAFNOSUPPORT);
201 vport = ops->create(parms);
203 module_put(ops->owner);
207 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
208 ovs_vport_name(vport));
209 hlist_add_head_rcu(&vport->hash_node, bucket);
213 /* Unlock to attempt module load and return -EAGAIN if load
214 * was successful as we need to restart the port addition
218 request_module("vport-type-%d", parms->type);
221 if (!ovs_vport_lookup(parms))
222 return ERR_PTR(-EAFNOSUPPORT);
224 return ERR_PTR(-EAGAIN);
228 * ovs_vport_set_options - modify existing vport device (for kernel callers)
230 * @vport: vport to modify.
231 * @options: New configuration.
233 * Modifies an existing device with the specified configuration (which is
234 * dependent on device type). ovs_mutex must be held.
236 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
238 if (!vport->ops->set_options)
240 return vport->ops->set_options(vport, options);
244 * ovs_vport_del - delete existing vport device
246 * @vport: vport to delete.
248 * Detaches @vport from its datapath and destroys it. ovs_mutex must
251 void ovs_vport_del(struct vport *vport)
253 hlist_del_rcu(&vport->hash_node);
254 module_put(vport->ops->owner);
255 vport->ops->destroy(vport);
259 * ovs_vport_get_stats - retrieve device stats
261 * @vport: vport from which to retrieve the stats
262 * @stats: location to store stats
264 * Retrieves transmit, receive, and error stats for the given device.
266 * Must be called with ovs_mutex or rcu_read_lock.
268 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
270 const struct rtnl_link_stats64 *dev_stats;
271 struct rtnl_link_stats64 temp;
273 dev_stats = dev_get_stats(vport->dev, &temp);
274 stats->rx_errors = dev_stats->rx_errors;
275 stats->tx_errors = dev_stats->tx_errors;
276 stats->tx_dropped = dev_stats->tx_dropped;
277 stats->rx_dropped = dev_stats->rx_dropped;
279 stats->rx_bytes = dev_stats->rx_bytes;
280 stats->rx_packets = dev_stats->rx_packets;
281 stats->tx_bytes = dev_stats->tx_bytes;
282 stats->tx_packets = dev_stats->tx_packets;
286 * ovs_vport_get_options - retrieve device options
288 * @vport: vport from which to retrieve the options.
289 * @skb: sk_buff where options should be appended.
291 * Retrieves the configuration of the given device, appending an
292 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
293 * vport-specific attributes to @skb.
295 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
296 * negative error code if a real error occurred. If an error occurs, @skb is
299 * Must be called with ovs_mutex or rcu_read_lock.
301 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
306 if (!vport->ops->get_options)
309 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
313 err = vport->ops->get_options(vport, skb);
315 nla_nest_cancel(skb, nla);
319 nla_nest_end(skb, nla);
324 * ovs_vport_set_upcall_portids - set upcall portids of @vport.
326 * @vport: vport to modify.
327 * @ids: new configuration, an array of port ids.
329 * Sets the vport's upcall_portids to @ids.
331 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
332 * as an array of U32.
334 * Must be called with ovs_mutex.
336 int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
338 struct vport_portids *old, *vport_portids;
340 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
343 old = ovsl_dereference(vport->upcall_portids);
345 vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
350 vport_portids->n_ids = nla_len(ids) / sizeof(u32);
351 vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
352 nla_memcpy(vport_portids->ids, ids, nla_len(ids));
354 rcu_assign_pointer(vport->upcall_portids, vport_portids);
362 * ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
364 * @vport: vport from which to retrieve the portids.
365 * @skb: sk_buff where portids should be appended.
367 * Retrieves the configuration of the given vport, appending the
368 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
371 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
372 * If an error occurs, @skb is left unmodified. Must be called with
373 * ovs_mutex or rcu_read_lock.
375 int ovs_vport_get_upcall_portids(const struct vport *vport,
378 struct vport_portids *ids;
380 ids = rcu_dereference_ovsl(vport->upcall_portids);
382 if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
383 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
384 ids->n_ids * sizeof(u32), (void *)ids->ids);
386 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
390 * ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
392 * @vport: vport from which the missed packet is received.
393 * @skb: skb that the missed packet was received.
395 * Uses the skb_get_hash() to select the upcall portid to send the
398 * Returns the portid of the target socket. Must be called with rcu_read_lock.
400 u32 ovs_vport_find_upcall_portid(const struct vport *vport,
403 struct vport_portids *ids;
407 ids = rcu_dereference(vport->upcall_portids);
409 /* If there is only one portid, select it in the fast-path. */
413 hash = skb_get_hash(skb);
414 ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
415 return ids->ids[ids_index];
419 * ovs_vport_receive - pass up received packet to the datapath for processing
421 * @vport: vport that received the packet
422 * @skb: skb that was received
423 * @tun_info: tunnel (if any) that carried packet
425 * Must be called with rcu_read_lock. The packet cannot be shared and
426 * skb->data should point to the Ethernet header.
428 int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
429 const struct ip_tunnel_info *tun_info)
431 struct sw_flow_key key;
434 OVS_CB(skb)->input_vport = vport;
435 OVS_CB(skb)->mru = 0;
436 OVS_CB(skb)->cutlen = 0;
437 if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
441 skb_scrub_packet(skb, true);
446 /* Extract flow from 'skb' into 'key'. */
447 error = ovs_flow_key_extract(tun_info, skb, &key);
448 if (unlikely(error)) {
452 ovs_dp_process_packet(skb, &key);
456 static int packet_length(const struct sk_buff *skb,
457 struct net_device *dev)
459 int length = skb->len - dev->hard_header_len;
461 if (!skb_vlan_tag_present(skb) &&
462 eth_type_vlan(skb->protocol))
465 /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
466 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
467 * account for 802.1ad. e.g. is_skb_forwardable().
470 return length > 0 ? length : 0;
473 void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
475 int mtu = vport->dev->mtu;
477 switch (vport->dev->type) {
479 if (mac_proto == MAC_PROTO_ETHERNET) {
480 skb_reset_network_header(skb);
481 skb_reset_mac_len(skb);
482 skb->protocol = htons(ETH_P_TEB);
483 } else if (mac_proto != MAC_PROTO_NONE) {
489 if (mac_proto != MAC_PROTO_ETHERNET)
496 if (unlikely(packet_length(skb, vport->dev) > mtu &&
498 net_warn_ratelimited("%s: dropped over-mtu packet: %d > %d\n",
500 packet_length(skb, vport->dev), mtu);
501 vport->dev->stats.tx_errors++;
505 skb->dev = vport->dev;
507 vport->ops->send(skb);