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
114 * @parms: information about new vport.
116 * Allocate and initialize a new vport defined by @ops. The vport will contain
117 * a private data area of size @priv_size that can be accessed using
118 * vport_priv(). Some parameters of the vport will be initialized from @parms.
119 * @vports that are no longer needed should be released with
122 struct vport *ovs_vport_alloc(int priv_size, const struct vport_ops *ops,
123 const struct vport_parms *parms)
128 alloc_size = sizeof(struct vport);
130 alloc_size = ALIGN(alloc_size, VPORT_ALIGN);
131 alloc_size += priv_size;
134 vport = kzalloc(alloc_size, GFP_KERNEL);
136 return ERR_PTR(-ENOMEM);
138 vport->dp = parms->dp;
139 vport->port_no = parms->port_no;
141 INIT_HLIST_NODE(&vport->dp_hash_node);
143 if (ovs_vport_set_upcall_portids(vport, parms->upcall_portids)) {
145 return ERR_PTR(-EINVAL);
150 EXPORT_SYMBOL_GPL(ovs_vport_alloc);
153 * ovs_vport_free - uninitialize and free vport
155 * @vport: vport to free
157 * Frees a vport allocated with vport_alloc() when it is no longer needed.
159 * The caller must ensure that an RCU grace period has passed since the last
160 * time @vport was in a datapath.
162 void ovs_vport_free(struct vport *vport)
164 /* vport is freed from RCU callback or error path, Therefore
165 * it is safe to use raw dereference.
167 kfree(rcu_dereference_raw(vport->upcall_portids));
170 EXPORT_SYMBOL_GPL(ovs_vport_free);
172 static struct vport_ops *ovs_vport_lookup(const struct vport_parms *parms)
174 struct vport_ops *ops;
176 list_for_each_entry(ops, &vport_ops_list, list)
177 if (ops->type == parms->type)
184 * ovs_vport_add - add vport device (for kernel callers)
186 * @parms: Information about new vport.
188 * Creates a new vport with the specified configuration (which is dependent on
189 * device type). ovs_mutex must be held.
191 struct vport *ovs_vport_add(const struct vport_parms *parms)
193 struct vport_ops *ops;
196 ops = ovs_vport_lookup(parms);
198 struct hlist_head *bucket;
200 if (!try_module_get(ops->owner))
201 return ERR_PTR(-EAFNOSUPPORT);
203 vport = ops->create(parms);
205 module_put(ops->owner);
209 bucket = hash_bucket(ovs_dp_get_net(vport->dp),
210 ovs_vport_name(vport));
211 hlist_add_head_rcu(&vport->hash_node, bucket);
215 /* Unlock to attempt module load and return -EAGAIN if load
216 * was successful as we need to restart the port addition
220 request_module("vport-type-%d", parms->type);
223 if (!ovs_vport_lookup(parms))
224 return ERR_PTR(-EAFNOSUPPORT);
226 return ERR_PTR(-EAGAIN);
230 * ovs_vport_set_options - modify existing vport device (for kernel callers)
232 * @vport: vport to modify.
233 * @options: New configuration.
235 * Modifies an existing device with the specified configuration (which is
236 * dependent on device type). ovs_mutex must be held.
238 int ovs_vport_set_options(struct vport *vport, struct nlattr *options)
240 if (!vport->ops->set_options)
242 return vport->ops->set_options(vport, options);
246 * ovs_vport_del - delete existing vport device
248 * @vport: vport to delete.
250 * Detaches @vport from its datapath and destroys it. ovs_mutex must
253 void ovs_vport_del(struct vport *vport)
255 hlist_del_rcu(&vport->hash_node);
256 module_put(vport->ops->owner);
257 vport->ops->destroy(vport);
261 * ovs_vport_get_stats - retrieve device stats
263 * @vport: vport from which to retrieve the stats
264 * @stats: location to store stats
266 * Retrieves transmit, receive, and error stats for the given device.
268 * Must be called with ovs_mutex or rcu_read_lock.
270 void ovs_vport_get_stats(struct vport *vport, struct ovs_vport_stats *stats)
272 const struct rtnl_link_stats64 *dev_stats;
273 struct rtnl_link_stats64 temp;
275 dev_stats = dev_get_stats(vport->dev, &temp);
276 stats->rx_errors = dev_stats->rx_errors;
277 stats->tx_errors = dev_stats->tx_errors;
278 stats->tx_dropped = dev_stats->tx_dropped;
279 stats->rx_dropped = dev_stats->rx_dropped;
281 stats->rx_bytes = dev_stats->rx_bytes;
282 stats->rx_packets = dev_stats->rx_packets;
283 stats->tx_bytes = dev_stats->tx_bytes;
284 stats->tx_packets = dev_stats->tx_packets;
288 * ovs_vport_get_upcall_stats - retrieve upcall stats
290 * @vport: vport from which to retrieve the stats.
291 * @skb: sk_buff where upcall stats should be appended.
293 * Retrieves upcall stats for the given device.
295 * Must be called with ovs_mutex or rcu_read_lock.
297 int ovs_vport_get_upcall_stats(struct vport *vport, struct sk_buff *skb)
302 __u64 tx_success = 0;
305 for_each_possible_cpu(i) {
306 const struct vport_upcall_stats_percpu *stats;
309 stats = per_cpu_ptr(vport->upcall_stats, i);
311 start = u64_stats_fetch_begin(&stats->syncp);
312 tx_success += u64_stats_read(&stats->n_success);
313 tx_fail += u64_stats_read(&stats->n_fail);
314 } while (u64_stats_fetch_retry(&stats->syncp, start));
317 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_UPCALL_STATS);
321 if (nla_put_u64_64bit(skb, OVS_VPORT_UPCALL_ATTR_SUCCESS, tx_success,
322 OVS_VPORT_ATTR_PAD)) {
323 nla_nest_cancel(skb, nla);
327 if (nla_put_u64_64bit(skb, OVS_VPORT_UPCALL_ATTR_FAIL, tx_fail,
328 OVS_VPORT_ATTR_PAD)) {
329 nla_nest_cancel(skb, nla);
332 nla_nest_end(skb, nla);
338 * ovs_vport_get_options - retrieve device options
340 * @vport: vport from which to retrieve the options.
341 * @skb: sk_buff where options should be appended.
343 * Retrieves the configuration of the given device, appending an
344 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
345 * vport-specific attributes to @skb.
347 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
348 * negative error code if a real error occurred. If an error occurs, @skb is
351 * Must be called with ovs_mutex or rcu_read_lock.
353 int ovs_vport_get_options(const struct vport *vport, struct sk_buff *skb)
358 if (!vport->ops->get_options)
361 nla = nla_nest_start_noflag(skb, OVS_VPORT_ATTR_OPTIONS);
365 err = vport->ops->get_options(vport, skb);
367 nla_nest_cancel(skb, nla);
371 nla_nest_end(skb, nla);
376 * ovs_vport_set_upcall_portids - set upcall portids of @vport.
378 * @vport: vport to modify.
379 * @ids: new configuration, an array of port ids.
381 * Sets the vport's upcall_portids to @ids.
383 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
384 * as an array of U32.
386 * Must be called with ovs_mutex.
388 int ovs_vport_set_upcall_portids(struct vport *vport, const struct nlattr *ids)
390 struct vport_portids *old, *vport_portids;
392 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
395 old = ovsl_dereference(vport->upcall_portids);
397 vport_portids = kmalloc(sizeof(*vport_portids) + nla_len(ids),
402 vport_portids->n_ids = nla_len(ids) / sizeof(u32);
403 vport_portids->rn_ids = reciprocal_value(vport_portids->n_ids);
404 nla_memcpy(vport_portids->ids, ids, nla_len(ids));
406 rcu_assign_pointer(vport->upcall_portids, vport_portids);
414 * ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
416 * @vport: vport from which to retrieve the portids.
417 * @skb: sk_buff where portids should be appended.
419 * Retrieves the configuration of the given vport, appending the
420 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
423 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
424 * If an error occurs, @skb is left unmodified. Must be called with
425 * ovs_mutex or rcu_read_lock.
427 int ovs_vport_get_upcall_portids(const struct vport *vport,
430 struct vport_portids *ids;
432 ids = rcu_dereference_ovsl(vport->upcall_portids);
434 if (vport->dp->user_features & OVS_DP_F_VPORT_PIDS)
435 return nla_put(skb, OVS_VPORT_ATTR_UPCALL_PID,
436 ids->n_ids * sizeof(u32), (void *)ids->ids);
438 return nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, ids->ids[0]);
442 * ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
444 * @vport: vport from which the missed packet is received.
445 * @skb: skb that the missed packet was received.
447 * Uses the skb_get_hash() to select the upcall portid to send the
450 * Returns the portid of the target socket. Must be called with rcu_read_lock.
452 u32 ovs_vport_find_upcall_portid(const struct vport *vport,
455 struct vport_portids *ids;
459 ids = rcu_dereference(vport->upcall_portids);
461 /* If there is only one portid, select it in the fast-path. */
465 hash = skb_get_hash(skb);
466 ids_index = hash - ids->n_ids * reciprocal_divide(hash, ids->rn_ids);
467 return ids->ids[ids_index];
471 * ovs_vport_receive - pass up received packet to the datapath for processing
473 * @vport: vport that received the packet
474 * @skb: skb that was received
475 * @tun_info: tunnel (if any) that carried packet
477 * Must be called with rcu_read_lock. The packet cannot be shared and
478 * skb->data should point to the Ethernet header.
480 int ovs_vport_receive(struct vport *vport, struct sk_buff *skb,
481 const struct ip_tunnel_info *tun_info)
483 struct sw_flow_key key;
486 OVS_CB(skb)->input_vport = vport;
487 OVS_CB(skb)->mru = 0;
488 OVS_CB(skb)->cutlen = 0;
489 if (unlikely(dev_net(skb->dev) != ovs_dp_get_net(vport->dp))) {
493 skb_scrub_packet(skb, true);
498 /* Extract flow from 'skb' into 'key'. */
499 error = ovs_flow_key_extract(tun_info, skb, &key);
500 if (unlikely(error)) {
504 ovs_dp_process_packet(skb, &key);
508 static int packet_length(const struct sk_buff *skb,
509 struct net_device *dev)
511 int length = skb->len - dev->hard_header_len;
513 if (!skb_vlan_tag_present(skb) &&
514 eth_type_vlan(skb->protocol))
517 /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
518 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
519 * account for 802.1ad. e.g. is_skb_forwardable().
522 return length > 0 ? length : 0;
525 void ovs_vport_send(struct vport *vport, struct sk_buff *skb, u8 mac_proto)
527 int mtu = vport->dev->mtu;
529 switch (vport->dev->type) {
531 if (mac_proto == MAC_PROTO_ETHERNET) {
532 skb_reset_network_header(skb);
533 skb_reset_mac_len(skb);
534 skb->protocol = htons(ETH_P_TEB);
535 } else if (mac_proto != MAC_PROTO_NONE) {
541 if (mac_proto != MAC_PROTO_ETHERNET)
548 if (unlikely(packet_length(skb, vport->dev) > mtu &&
550 vport->dev->stats.tx_errors++;
551 if (vport->dev->flags & IFF_UP)
552 net_warn_ratelimited("%s: dropped over-mtu packet: "
553 "%d > %d\n", vport->dev->name,
554 packet_length(skb, vport->dev),
559 skb->dev = vport->dev;
560 skb_clear_tstamp(skb);
561 vport->ops->send(skb);