1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 * Copyright 2012 OmniTI Computer Consulting, Inc All rights reserved.
25 * Copyright 2018 Joyent, Inc.
26 */
27
28 #ifndef _SYS_AGGR_IMPL_H
29 #define _SYS_AGGR_IMPL_H
30
31 #include <sys/types.h>
32 #include <sys/cred.h>
33 #include <sys/mac_ether.h>
34 #include <sys/mac_provider.h>
35 #include <sys/mac_client.h>
36 #include <sys/mac_client_priv.h>
37 #include <sys/aggr_lacp.h>
38
39 #ifdef __cplusplus
40 extern "C" {
41 #endif
42
43 #ifdef _KERNEL
44
45 #define AGGR_MINOR_CTL 1 /* control interface minor */
46
47 /* flags for aggr_grp_modify() */
48 #define AGGR_MODIFY_POLICY 0x01
49 #define AGGR_MODIFY_MAC 0x02
50 #define AGGR_MODIFY_LACP_MODE 0x04
51 #define AGGR_MODIFY_LACP_TIMER 0x08
52
53 /*
54 * Possible value of aggr_rseudo_rx_ring_t.arr_flags. Set when the ring entry
55 * in the pseudo RX group is used.
56 */
57 #define MAC_PSEUDO_RING_INUSE 0x01
58
59 /*
60 * VLAN filters placed on the Rx pseudo group.
61 */
62 typedef struct aggr_vlan {
63 list_node_t av_link;
64 uint16_t av_vid; /* VLAN ID */
65 uint_t av_refs; /* num aggr clients using this VID */
66 } aggr_vlan_t;
67
68 typedef struct aggr_unicst_addr_s {
69 uint8_t aua_addr[ETHERADDRL];
70 struct aggr_unicst_addr_s *aua_next;
71 } aggr_unicst_addr_t;
72
73 typedef struct aggr_pseudo_rx_ring_s {
74 mac_ring_handle_t arr_rh; /* filled in by aggr_fill_ring() */
75 struct aggr_port_s *arr_port;
76 mac_ring_handle_t arr_hw_rh;
77 uint_t arr_flags;
78 uint64_t arr_gen;
79 } aggr_pseudo_rx_ring_t;
80
81 typedef struct aggr_pseudo_rx_group_s {
82 struct aggr_grp_s *arg_grp; /* filled in by aggr_fill_group() */
83 mac_group_handle_t arg_gh; /* filled in by aggr_fill_group() */
84 aggr_unicst_addr_t *arg_macaddr;
85 aggr_pseudo_rx_ring_t arg_rings[MAX_RINGS_PER_GROUP];
86 uint_t arg_ring_cnt;
87 uint_t arg_untagged; /* num clients untagged */
88 list_t arg_vlans; /* VLANs on this group */
89 } aggr_pseudo_rx_group_t;
90
91 typedef struct aggr_pseudo_tx_ring_s {
92 mac_ring_handle_t atr_rh; /* filled in by aggr_fill_ring() */
93 struct aggr_port_s *atr_port;
94 mac_ring_handle_t atr_hw_rh;
95 uint_t atr_flags;
96 } aggr_pseudo_tx_ring_t;
97
98 typedef struct aggr_pseudo_tx_group_s {
99 mac_group_handle_t atg_gh; /* filled in by aggr_fill_group() */
100 uint_t atg_ring_cnt;
101 aggr_pseudo_tx_ring_t atg_rings[MAX_RINGS_PER_GROUP];
102 } aggr_pseudo_tx_group_t;
103
104 /*
105 * A link aggregation MAC port.
106 * Note that lp_next is protected by the lg_lock of the group the
107 * port is part of.
108 */
109 typedef struct aggr_port_s {
110 struct aggr_port_s *lp_next;
111 struct aggr_grp_s *lp_grp; /* back ptr to group */
112 datalink_id_t lp_linkid;
113 uint16_t lp_portid;
114 uint8_t lp_addr[ETHERADDRL]; /* port MAC address */
115 uint32_t lp_refs; /* refcount */
116 aggr_port_state_t lp_state;
117 uint32_t lp_started : 1,
118 lp_tx_enabled : 1,
119 lp_collector_enabled : 1,
120 lp_promisc_on : 1,
121 lp_no_link_update : 1,
122 lp_rx_grp_added : 1,
123 lp_tx_grp_added : 1,
124 lp_closing : 1,
125 lp_pad_bits : 24;
126 mac_handle_t lp_mh;
127 mac_client_handle_t lp_mch;
128 const mac_info_t *lp_mip;
129 mac_notify_handle_t lp_mnh;
130 uint_t lp_tx_idx; /* idx in group's tx array */
131 uint64_t lp_ifspeed;
132 link_state_t lp_link_state;
133 link_duplex_t lp_link_duplex;
134 uint64_t lp_stat[MAC_NSTAT];
135 uint64_t lp_ether_stat[ETHER_NSTAT];
136 aggr_lacp_port_t lp_lacp; /* LACP state */
137 lacp_stats_t lp_lacp_stats;
138 uint32_t lp_margin;
139 mac_promisc_handle_t lp_mphp;
140 mac_unicast_handle_t lp_mah;
141
142 /* List of non-primary addresses that requires promiscous mode set */
143 aggr_unicst_addr_t *lp_prom_addr;
144 /* handle of the underlying HW RX group */
145 mac_group_handle_t lp_hwgh;
146 int lp_tx_ring_cnt;
147 /* handles of the underlying HW TX rings */
148 mac_ring_handle_t *lp_tx_rings;
149 /*
150 * Handles of the pseudo TX rings. Each of them maps to
151 * corresponding hardware TX ring in lp_tx_rings[]. A
152 * pseudo TX ring is presented to aggr primary mac
153 * client even when underlying NIC has no TX ring.
154 */
155 mac_ring_handle_t *lp_pseudo_tx_rings;
156 void *lp_tx_notify_mh;
157 } aggr_port_t;
158
159 /*
160 * A link aggregation group.
161 *
162 * The following per-group flags are defined:
163 *
164 * - lg_addr_fixed: set when the MAC address has been explicitely set
165 * when the group was created, or by a m_unicst_set() request.
166 * If this flag is not set, the MAC address of the group will be
167 * set to the first port that is added to the group.
168 *
169 * - lg_add_set: used only when lg_addr_fixed is not set. Captures whether
170 * the MAC address was initialized according to the members of the group.
171 * When set, the lg_port field points to the port from which the
172 * MAC address was initialized.
173 *
174 */
175 typedef struct aggr_grp_s {
176 datalink_id_t lg_linkid;
177 uint16_t lg_key; /* key (group port number) */
178 uint32_t lg_refs; /* refcount */
179 uint16_t lg_nports; /* number of MAC ports */
180 uint8_t lg_addr[ETHERADDRL]; /* group MAC address */
181 uint16_t
182 lg_closing : 1,
183 lg_addr_fixed : 1, /* fixed MAC address? */
184 lg_started : 1, /* group started? */
185 lg_promisc : 1, /* in promiscuous mode? */
186 lg_zcopy : 1,
187 lg_vlan : 1,
188 lg_force : 1,
189 lg_lso : 1,
190 lg_pad_bits : 8;
191 aggr_port_t *lg_ports; /* list of configured ports */
192 aggr_port_t *lg_mac_addr_port;
193 mac_handle_t lg_mh;
194 zoneid_t lg_zoneid;
195 uint_t lg_nattached_ports;
196 krwlock_t lg_tx_lock;
197 uint_t lg_ntx_ports;
198 aggr_port_t **lg_tx_ports; /* array of tx ports */
199 uint_t lg_tx_ports_size; /* size of lg_tx_ports */
200 uint32_t lg_tx_policy; /* outbound policy */
201 uint8_t lg_mac_tx_policy;
202 link_state_t lg_link_state;
203
204
205 /*
206 * The lg_stat_lock must be held when accessing these fields.
207 */
208 kmutex_t lg_stat_lock;
209 uint64_t lg_ifspeed;
210 link_duplex_t lg_link_duplex;
211 uint64_t lg_stat[MAC_NSTAT];
212 uint64_t lg_ether_stat[ETHER_NSTAT];
213
214 aggr_lacp_mode_t lg_lacp_mode; /* off, active, or passive */
215 Agg_t aggr; /* 802.3ad data */
216 uint32_t lg_hcksum_txflags;
217 uint_t lg_max_sdu;
218 uint32_t lg_margin;
219 mac_capab_lso_t lg_cap_lso;
220
221 /*
222 * The following fields are used by the LACP packets processing.
223 * Specifically, as the LACP packets processing is not performance
224 * critical, all LACP packets will be handled by a dedicated thread
225 * instead of in the mac_rx() call. This is to avoid the dead lock
226 * with mac_unicast_remove(), which holding the mac perimeter of the
227 * aggr, and wait for the mr_refcnt of the RX ring to drop to zero.
228 */
229 kmutex_t lg_lacp_lock;
230 kcondvar_t lg_lacp_cv;
231 mblk_t *lg_lacp_head;
232 mblk_t *lg_lacp_tail;
233 kthread_t *lg_lacp_rx_thread;
234 boolean_t lg_lacp_done;
235
236 aggr_pseudo_rx_group_t lg_rx_group;
237 aggr_pseudo_tx_group_t lg_tx_group;
238
239 kmutex_t lg_tx_flowctl_lock;
240 kcondvar_t lg_tx_flowctl_cv;
241 uint_t lg_tx_blocked_cnt;
242 mac_ring_handle_t *lg_tx_blocked_rings;
243 kthread_t *lg_tx_notify_thread;
244 boolean_t lg_tx_notify_done;
245
246 /*
247 * The following fields are used by aggr to wait for all the
248 * aggr_port_notify_cb() and aggr_port_timer_thread() to finish
249 * before it calls mac_unregister() when the aggr is deleted.
250 */
251 kmutex_t lg_port_lock;
252 kcondvar_t lg_port_cv;
253 int lg_port_ref;
254 } aggr_grp_t;
255
256 #define AGGR_GRP_REFHOLD(grp) { \
257 atomic_inc_32(&(grp)->lg_refs); \
258 ASSERT((grp)->lg_refs != 0); \
259 }
260
261 #define AGGR_GRP_REFRELE(grp) { \
262 ASSERT((grp)->lg_refs != 0); \
263 membar_exit(); \
264 if (atomic_dec_32_nv(&(grp)->lg_refs) == 0) \
265 aggr_grp_free(grp); \
266 }
267
268 #define AGGR_PORT_REFHOLD(port) { \
269 atomic_inc_32(&(port)->lp_refs); \
270 ASSERT((port)->lp_refs != 0); \
271 }
272
273 #define AGGR_PORT_REFRELE(port) { \
274 ASSERT((port)->lp_refs != 0); \
275 membar_exit(); \
276 if (atomic_dec_32_nv(&(port)->lp_refs) == 0) \
277 aggr_port_free(port); \
278 }
279
280 extern dev_info_t *aggr_dip;
281 extern int aggr_ioc_init(void);
282 extern void aggr_ioc_fini(void);
283
284 typedef int (*aggr_grp_info_new_grp_fn_t)(void *, datalink_id_t, uint32_t,
285 uchar_t *, boolean_t, boolean_t, uint32_t, uint32_t, aggr_lacp_mode_t,
286 aggr_lacp_timer_t);
287 typedef int (*aggr_grp_info_new_port_fn_t)(void *, datalink_id_t, uchar_t *,
288 aggr_port_state_t, aggr_lacp_state_t *);
289
290 extern void aggr_grp_init(void);
291 extern void aggr_grp_fini(void);
292 extern int aggr_grp_create(datalink_id_t, uint32_t, uint_t, laioc_port_t *,
293 uint32_t, boolean_t, boolean_t, uchar_t *, aggr_lacp_mode_t,
294 aggr_lacp_timer_t, cred_t *);
295 extern int aggr_grp_delete(datalink_id_t, cred_t *);
296 extern void aggr_grp_free(aggr_grp_t *);
297
298 extern int aggr_grp_info(datalink_id_t, void *, aggr_grp_info_new_grp_fn_t,
299 aggr_grp_info_new_port_fn_t, cred_t *);
300 extern void aggr_grp_notify(aggr_grp_t *, uint32_t);
301 extern boolean_t aggr_grp_attach_port(aggr_grp_t *, aggr_port_t *);
302 extern boolean_t aggr_grp_detach_port(aggr_grp_t *, aggr_port_t *);
303 extern void aggr_grp_port_mac_changed(aggr_grp_t *, aggr_port_t *,
304 boolean_t *, boolean_t *);
305 extern int aggr_grp_add_ports(datalink_id_t, uint_t, boolean_t,
306 laioc_port_t *);
307 extern int aggr_grp_rem_ports(datalink_id_t, uint_t, laioc_port_t *);
308 extern boolean_t aggr_grp_update_ports_mac(aggr_grp_t *);
309 extern int aggr_grp_modify(datalink_id_t, uint8_t, uint32_t, boolean_t,
310 const uchar_t *, aggr_lacp_mode_t, aggr_lacp_timer_t);
311 extern void aggr_grp_multicst_port(aggr_port_t *, boolean_t);
312 extern uint_t aggr_grp_count(void);
313 extern void aggr_grp_update_default(aggr_grp_t *);
314
315 extern void aggr_port_init(void);
316 extern void aggr_port_fini(void);
317 extern int aggr_port_create(aggr_grp_t *, const datalink_id_t, boolean_t,
318 aggr_port_t **);
319 extern void aggr_port_delete(aggr_port_t *);
320 extern void aggr_port_free(aggr_port_t *);
321 extern int aggr_port_start(aggr_port_t *);
322 extern void aggr_port_stop(aggr_port_t *);
323 extern int aggr_port_promisc(aggr_port_t *, boolean_t);
324 extern int aggr_port_unicst(aggr_port_t *);
325 extern int aggr_port_multicst(void *, boolean_t, const uint8_t *);
326 extern uint64_t aggr_port_stat(aggr_port_t *, uint_t);
327 extern boolean_t aggr_port_notify_link(aggr_grp_t *, aggr_port_t *);
328 extern void aggr_port_init_callbacks(aggr_port_t *);
329
330 extern void aggr_recv_cb(void *, mac_resource_handle_t, mblk_t *, boolean_t);
331 extern void aggr_recv_promisc_cb(void *, mac_resource_handle_t, mblk_t *,
332 boolean_t);
333
334 extern void aggr_tx_ring_update(void *, uintptr_t);
335 extern void aggr_tx_notify_thread(void *);
336 extern void aggr_send_port_enable(aggr_port_t *);
337 extern void aggr_send_port_disable(aggr_port_t *);
338 extern void aggr_send_update_policy(aggr_grp_t *, uint32_t);
339
340 extern void aggr_lacp_init(void);
341 extern void aggr_lacp_fini(void);
342 extern void aggr_lacp_init_port(aggr_port_t *);
343 extern void aggr_lacp_init_grp(aggr_grp_t *);
344 extern void aggr_lacp_set_mode(aggr_grp_t *, aggr_lacp_mode_t,
345 aggr_lacp_timer_t);
346 extern void aggr_lacp_update_mode(aggr_grp_t *, aggr_lacp_mode_t);
347 extern void aggr_lacp_update_timer(aggr_grp_t *, aggr_lacp_timer_t);
348 extern void aggr_lacp_rx_enqueue(aggr_port_t *, mblk_t *);
349 extern void aggr_lacp_port_attached(aggr_port_t *);
350 extern void aggr_lacp_port_detached(aggr_port_t *);
351 extern void aggr_port_lacp_set_mode(aggr_grp_t *, aggr_port_t *);
352
353 extern void aggr_lacp_rx_thread(void *);
354 extern void aggr_recv_lacp(aggr_port_t *, mac_resource_handle_t, mblk_t *);
355
356 extern void aggr_grp_port_hold(aggr_port_t *);
357 extern void aggr_grp_port_rele(aggr_port_t *);
358 extern void aggr_grp_port_wait(aggr_grp_t *);
359
360 extern int aggr_port_addmac(aggr_port_t *, const uint8_t *);
361 extern void aggr_port_remmac(aggr_port_t *, const uint8_t *);
362
363 extern int aggr_port_addvlan(aggr_port_t *, uint16_t);
364 extern int aggr_port_remvlan(aggr_port_t *, uint16_t);
365
366 extern mblk_t *aggr_ring_tx(void *, mblk_t *);
367 extern mblk_t *aggr_find_tx_ring(void *, mblk_t *,
368 uintptr_t, mac_ring_handle_t *);
369
370 #endif /* _KERNEL */
371
372 #ifdef __cplusplus
373 }
374 #endif
375
376 #endif /* _SYS_AGGR_IMPL_H */