aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/net/bonding/rte_eth_bond_alb.c
blob: d3e16d4bda63c9f4079f370ffb8e75dc35abb9be (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
/* SPDX-License-Identifier: BSD-3-Clause
 * Copyright(c) 2010-2015 Intel Corporation
 */

#include "rte_eth_bond_private.h"
#include "rte_eth_bond_alb.h"

static inline uint8_t
simple_hash(uint8_t *hash_start, int hash_size)
{
	int i;
	uint8_t hash;

	hash = 0;
	for (i = 0; i < hash_size; ++i)
		hash ^= hash_start[i];

	return hash;
}

static uint16_t
calculate_slave(struct bond_dev_private *internals)
{
	uint16_t idx;

	idx = (internals->mode6.last_slave + 1) % internals->active_slave_count;
	internals->mode6.last_slave = idx;
	return internals->active_slaves[idx];
}

int
bond_mode_alb_enable(struct rte_eth_dev *bond_dev)
{
	struct bond_dev_private *internals = bond_dev->data->dev_private;
	struct client_data *hash_table = internals->mode6.client_table;

	uint16_t data_size;
	char mem_name[RTE_ETH_NAME_MAX_LEN];
	int socket_id = bond_dev->data->numa_node;

	/* Fill hash table with initial values */
	memset(hash_table, 0, sizeof(struct client_data) * ALB_HASH_TABLE_SIZE);
	rte_spinlock_init(&internals->mode6.lock);
	internals->mode6.last_slave = ALB_NULL_INDEX;
	internals->mode6.ntt = 0;

	/* Initialize memory pool for ARP packets to send */
	if (internals->mode6.mempool == NULL) {
		/*
		 * 256 is size of ETH header, ARP header and nested VLAN headers.
		 * The value is chosen to be cache aligned.
		 */
		data_size = 256 + RTE_PKTMBUF_HEADROOM;
		snprintf(mem_name, sizeof(mem_name), "%s_ALB",
				bond_dev->device->name);
		internals->mode6.mempool = rte_pktmbuf_pool_create(mem_name,
			512 * RTE_MAX_ETHPORTS,
			RTE_MEMPOOL_CACHE_MAX_SIZE >= 32 ?
				32 : RTE_MEMPOOL_CACHE_MAX_SIZE,
			0, data_size, socket_id);

		if (internals->mode6.mempool == NULL) {
			RTE_BOND_LOG(ERR, "%s: Failed to initialize ALB mempool.\n",
				     bond_dev->device->name);
			goto mempool_alloc_error;
		}
	}

	return 0;

mempool_alloc_error:
	return -ENOMEM;
}

void bond_mode_alb_arp_recv(struct ether_hdr *eth_h, uint16_t offset,
		struct bond_dev_private *internals) {
	struct arp_hdr *arp;

	struct client_data *hash_table = internals->mode6.client_table;
	struct client_data *client_info;

	uint8_t hash_index;

	arp = (struct arp_hdr *) ((char *) (eth_h + 1) + offset);

	/* ARP Requests are forwarded to the application with no changes */
	if (arp->arp_op != rte_cpu_to_be_16(ARP_OP_REPLY))
		return;

	/* From now on, we analyze only ARP Reply packets */
	hash_index = simple_hash((uint8_t *) &arp->arp_data.arp_sip,
			sizeof(arp->arp_data.arp_sip));
	client_info = &hash_table[hash_index];

	/*
	 * We got reply for ARP Request send by the application. We need to
	 * update client table when received data differ from what is stored
	 * in ALB table and issue sending update packet to that slave.
	 */
	rte_spinlock_lock(&internals->mode6.lock);
	if (client_info->in_use == 0 ||
			client_info->app_ip != arp->arp_data.arp_tip ||
			client_info->cli_ip != arp->arp_data.arp_sip ||
			!is_same_ether_addr(&client_info->cli_mac, &arp->arp_data.arp_sha) ||
			client_info->vlan_count != offset / sizeof(struct vlan_hdr) ||
			memcmp(client_info->vlan, eth_h + 1, offset) != 0
	) {
		client_info->in_use = 1;
		client_info->app_ip = arp->arp_data.arp_tip;
		client_info->cli_ip = arp->arp_data.arp_sip;
		ether_addr_copy(&arp->arp_data.arp_sha, &client_info->cli_mac);
		client_info->slave_idx = calculate_slave(internals);
		rte_eth_macaddr_get(client_info->slave_idx, &client_info->app_mac);
		ether_addr_copy(&client_info->app_mac, &arp->arp_data.arp_tha);
		memcpy(client_info->vlan, eth_h + 1, offset);
		client_info->vlan_count = offset / sizeof(struct vlan_hdr);
	}
	internals->mode6.ntt = 1;
	rte_spinlock_unlock(&internals->mode6.lock);
}

uint16_t
bond_mode_alb_arp_xmit(struct ether_hdr *eth_h, uint16_t offset,
		struct bond_dev_private *internals)
{
	struct arp_hdr *arp;

	struct client_data *hash_table = internals->mode6.client_table;
	struct client_data *client_info;

	uint8_t hash_index;

	struct ether_addr bonding_mac;

	arp = (struct arp_hdr *)((char *)(eth_h + 1) + offset);

	/*
	 * Traffic with src MAC other than bonding should be sent on
	 * current primary port.
	 */
	rte_eth_macaddr_get(internals->port_id, &bonding_mac);
	if (!is_same_ether_addr(&bonding_mac, &arp->arp_data.arp_sha)) {
		rte_eth_macaddr_get(internals->current_primary_port,
				&arp->arp_data.arp_sha);
		return internals->current_primary_port;
	}

	hash_index = simple_hash((uint8_t *)&arp->arp_data.arp_tip,
			sizeof(uint32_t));
	client_info = &hash_table[hash_index];

	rte_spinlock_lock(&internals->mode6.lock);
	if (arp->arp_op == rte_cpu_to_be_16(ARP_OP_REPLY)) {
		if (client_info->in_use) {
			if (client_info->app_ip == arp->arp_data.arp_sip &&
				client_info->cli_ip == arp->arp_data.arp_tip) {
				/* Entry is already assigned to this client */
				if (!is_broadcast_ether_addr(&arp->arp_data.arp_tha)) {
					ether_addr_copy(&arp->arp_data.arp_tha,
							&client_info->cli_mac);
				}
				rte_eth_macaddr_get(client_info->slave_idx,
						&client_info->app_mac);
				ether_addr_copy(&client_info->app_mac, &arp->arp_data.arp_sha);
				memcpy(client_info->vlan, eth_h + 1, offset);
				client_info->vlan_count = offset / sizeof(struct vlan_hdr);
				rte_spinlock_unlock(&internals->mode6.lock);
				return client_info->slave_idx;
			}
		}

		/* Assign new slave to this client and update src mac in ARP */
		client_info->in_use = 1;
		client_info->ntt = 0;
		client_info->app_ip = arp->arp_data.arp_sip;
		ether_addr_copy(&arp->arp_data.arp_tha, &client_info->cli_mac);
		client_info->cli_ip = arp->arp_data.arp_tip;
		client_info->slave_idx = calculate_slave(internals);
		rte_eth_macaddr_get(client_info->slave_idx, &client_info->app_mac);
		ether_addr_copy(&client_info->app_mac, &arp->arp_data.arp_sha);
		memcpy(client_info->vlan, eth_h + 1, offset);
		client_info->vlan_count = offset / sizeof(struct vlan_hdr);
		rte_spinlock_unlock(&internals->mode6.lock);
		return client_info->slave_idx;
	}

	/* If packet is not ARP Reply, send it on current primary port. */
	rte_spinlock_unlock(&internals->mode6.lock);
	rte_eth_macaddr_get(internals->current_primary_port,
			&arp->arp_data.arp_sha);
	return internals->current_primary_port;
}

uint16_t
bond_mode_alb_arp_upd(struct client_data *client_info,
		struct rte_mbuf *pkt, struct bond_dev_private *internals)
{
	struct ether_hdr *eth_h;
	struct arp_hdr *arp_h;
	uint16_t slave_idx;

	rte_spinlock_lock(&internals->mode6.lock);
	eth_h = rte_pktmbuf_mtod(pkt, struct ether_hdr *);

	ether_addr_copy(&client_info->app_mac, &eth_h->s_addr);
	ether_addr_copy(&client_info->cli_mac, &eth_h->d_addr);
	if (client_info->vlan_count > 0)
		eth_h->ether_type = rte_cpu_to_be_16(ETHER_TYPE_VLAN);
	else
		eth_h->ether_type = rte_cpu_to_be_16(ETHER_TYPE_ARP);

	arp_h = (struct arp_hdr *)((char *)eth_h + sizeof(struct ether_hdr)
			+ client_info->vlan_count * sizeof(struct vlan_hdr));

	memcpy(eth_h + 1, client_info->vlan,
			client_info->vlan_count * sizeof(struct vlan_hdr));

	ether_addr_copy(&client_info->app_mac, &arp_h->arp_data.arp_sha);
	arp_h->arp_data.arp_sip = client_info->app_ip;
	ether_addr_copy(&client_info->cli_mac, &arp_h->arp_data.arp_tha);
	arp_h->arp_data.arp_tip = client_info->cli_ip;

	arp_h->arp_hrd = rte_cpu_to_be_16(ARP_HRD_ETHER);
	arp_h->arp_pro = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
	arp_h->arp_hln = ETHER_ADDR_LEN;
	arp_h->arp_pln = sizeof(uint32_t);
	arp_h->arp_op = rte_cpu_to_be_16(ARP_OP_REPLY);

	slave_idx = client_info->slave_idx;
	rte_spinlock_unlock(&internals->mode6.lock);

	return slave_idx;
}

void
bond_mode_alb_client_list_upd(struct rte_eth_dev *bond_dev)
{
	struct bond_dev_private *internals = bond_dev->data->dev_private;
	struct client_data *client_info;

	int i;

	/* If active slave count is 0, it's pointless to refresh alb table */
	if (internals->active_slave_count <= 0)
		return;

	rte_spinlock_lock(&internals->mode6.lock);
	internals->mode6.last_slave = ALB_NULL_INDEX;

	for (i = 0; i < ALB_HASH_TABLE_SIZE; i++) {
		client_info = &internals->mode6.client_table[i];
		if (client_info->in_use) {
			client_info->slave_idx = calculate_slave(internals);
			rte_eth_macaddr_get(client_info->slave_idx, &client_info->app_mac);
			internals->mode6.ntt = 1;
		}
	}
	rte_spinlock_unlock(&internals->mode6.lock);
}