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_options - retrieve device options
290 * @vport: vport from which to retrieve the options.
291 * @skb: sk_buff where options should be appended.
293 * Retrieves the configuration of the given device, appending an
294 * %OVS_VPORT_ATTR_OPTIONS attribute that in turn contains nested
295 * vport-specific attributes to @skb.
297 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room, or another
298 * negative error code if a real error occurred. If an error occurs, @skb is
301 * Must be called with ovs_mutex or rcu_read_lock.
303 int ovs_vport_get_options(const struct vport
*vport
, struct sk_buff
*skb
)
308 if (!vport
->ops
->get_options
)
311 nla
= nla_nest_start_noflag(skb
, OVS_VPORT_ATTR_OPTIONS
);
315 err
= vport
->ops
->get_options(vport
, skb
);
317 nla_nest_cancel(skb
, nla
);
321 nla_nest_end(skb
, nla
);
326 * ovs_vport_set_upcall_portids - set upcall portids of @vport.
328 * @vport: vport to modify.
329 * @ids: new configuration, an array of port ids.
331 * Sets the vport's upcall_portids to @ids.
333 * Returns 0 if successful, -EINVAL if @ids is zero length or cannot be parsed
334 * as an array of U32.
336 * Must be called with ovs_mutex.
338 int ovs_vport_set_upcall_portids(struct vport
*vport
, const struct nlattr
*ids
)
340 struct vport_portids
*old
, *vport_portids
;
342 if (!nla_len(ids
) || nla_len(ids
) % sizeof(u32
))
345 old
= ovsl_dereference(vport
->upcall_portids
);
347 vport_portids
= kmalloc(sizeof(*vport_portids
) + nla_len(ids
),
352 vport_portids
->n_ids
= nla_len(ids
) / sizeof(u32
);
353 vport_portids
->rn_ids
= reciprocal_value(vport_portids
->n_ids
);
354 nla_memcpy(vport_portids
->ids
, ids
, nla_len(ids
));
356 rcu_assign_pointer(vport
->upcall_portids
, vport_portids
);
364 * ovs_vport_get_upcall_portids - get the upcall_portids of @vport.
366 * @vport: vport from which to retrieve the portids.
367 * @skb: sk_buff where portids should be appended.
369 * Retrieves the configuration of the given vport, appending the
370 * %OVS_VPORT_ATTR_UPCALL_PID attribute which is the array of upcall
373 * Returns 0 if successful, -EMSGSIZE if @skb has insufficient room.
374 * If an error occurs, @skb is left unmodified. Must be called with
375 * ovs_mutex or rcu_read_lock.
377 int ovs_vport_get_upcall_portids(const struct vport
*vport
,
380 struct vport_portids
*ids
;
382 ids
= rcu_dereference_ovsl(vport
->upcall_portids
);
384 if (vport
->dp
->user_features
& OVS_DP_F_VPORT_PIDS
)
385 return nla_put(skb
, OVS_VPORT_ATTR_UPCALL_PID
,
386 ids
->n_ids
* sizeof(u32
), (void *)ids
->ids
);
388 return nla_put_u32(skb
, OVS_VPORT_ATTR_UPCALL_PID
, ids
->ids
[0]);
392 * ovs_vport_find_upcall_portid - find the upcall portid to send upcall.
394 * @vport: vport from which the missed packet is received.
395 * @skb: skb that the missed packet was received.
397 * Uses the skb_get_hash() to select the upcall portid to send the
400 * Returns the portid of the target socket. Must be called with rcu_read_lock.
402 u32
ovs_vport_find_upcall_portid(const struct vport
*vport
,
405 struct vport_portids
*ids
;
409 ids
= rcu_dereference(vport
->upcall_portids
);
411 /* If there is only one portid, select it in the fast-path. */
415 hash
= skb_get_hash(skb
);
416 ids_index
= hash
- ids
->n_ids
* reciprocal_divide(hash
, ids
->rn_ids
);
417 return ids
->ids
[ids_index
];
421 * ovs_vport_receive - pass up received packet to the datapath for processing
423 * @vport: vport that received the packet
424 * @skb: skb that was received
425 * @tun_info: tunnel (if any) that carried packet
427 * Must be called with rcu_read_lock. The packet cannot be shared and
428 * skb->data should point to the Ethernet header.
430 int ovs_vport_receive(struct vport
*vport
, struct sk_buff
*skb
,
431 const struct ip_tunnel_info
*tun_info
)
433 struct sw_flow_key key
;
436 OVS_CB(skb
)->input_vport
= vport
;
437 OVS_CB(skb
)->mru
= 0;
438 OVS_CB(skb
)->cutlen
= 0;
439 if (unlikely(dev_net(skb
->dev
) != ovs_dp_get_net(vport
->dp
))) {
443 skb_scrub_packet(skb
, true);
448 /* Extract flow from 'skb' into 'key'. */
449 error
= ovs_flow_key_extract(tun_info
, skb
, &key
);
450 if (unlikely(error
)) {
454 ovs_dp_process_packet(skb
, &key
);
458 static int packet_length(const struct sk_buff
*skb
,
459 struct net_device
*dev
)
461 int length
= skb
->len
- dev
->hard_header_len
;
463 if (!skb_vlan_tag_present(skb
) &&
464 eth_type_vlan(skb
->protocol
))
467 /* Don't subtract for multiple VLAN tags. Most (all?) drivers allow
468 * (ETH_LEN + VLAN_HLEN) in addition to the mtu value, but almost none
469 * account for 802.1ad. e.g. is_skb_forwardable().
472 return length
> 0 ? length
: 0;
475 void ovs_vport_send(struct vport
*vport
, struct sk_buff
*skb
, u8 mac_proto
)
477 int mtu
= vport
->dev
->mtu
;
479 switch (vport
->dev
->type
) {
481 if (mac_proto
== MAC_PROTO_ETHERNET
) {
482 skb_reset_network_header(skb
);
483 skb_reset_mac_len(skb
);
484 skb
->protocol
= htons(ETH_P_TEB
);
485 } else if (mac_proto
!= MAC_PROTO_NONE
) {
491 if (mac_proto
!= MAC_PROTO_ETHERNET
)
498 if (unlikely(packet_length(skb
, vport
->dev
) > mtu
&&
500 vport
->dev
->stats
.tx_errors
++;
501 if (vport
->dev
->flags
& IFF_UP
)
502 net_warn_ratelimited("%s: dropped over-mtu packet: "
503 "%d > %d\n", vport
->dev
->name
,
504 packet_length(skb
, vport
->dev
),
509 skb
->dev
= vport
->dev
;
511 vport
->ops
->send(skb
);