xref: /illumos-gate/usr/src/uts/common/sys/mac_client_impl.h (revision d48be21240dfd051b689384ce2b23479d757f2d8)
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 (c) 2012, Joyent, Inc.  All rights reserved.
25  */
26 /*
27  * Copyright 2018 Joyent, Inc.
28  */
29 
30 #ifndef	_SYS_MAC_CLIENT_IMPL_H
31 #define	_SYS_MAC_CLIENT_IMPL_H
32 
33 #include <sys/modhash.h>
34 #include <sys/mac_client.h>
35 #include <sys/mac_provider.h>
36 #include <sys/mac.h>
37 #include <sys/mac_impl.h>
38 #include <sys/mac_stat.h>
39 #include <net/if.h>
40 #include <sys/mac_flow_impl.h>
41 
42 #ifdef	__cplusplus
43 extern "C" {
44 #endif
45 
46 extern kmem_cache_t	*mac_client_impl_cache;
47 extern kmem_cache_t	*mac_unicast_impl_cache;
48 extern kmem_cache_t	*mac_promisc_impl_cache;
49 
50 /*
51  * Need a list to chain all VIDs assigned to a client. Normally, one
52  * MAC client only has one VID. But vsw might need multiple VIDs.
53  */
54 typedef struct mac_unicast_impl_s {			/* Protected by */
55 	struct mac_unicast_impl_s	*mui_next;	/* SL */
56 	mac_address_t			*mui_map;	/* SL */
57 	uint16_t			mui_vid;	/* SL */
58 } mac_unicast_impl_t;
59 
60 #define	MAC_CLIENT_FLAGS_PRIMARY		0x0001
61 #define	MAC_CLIENT_FLAGS_VNIC_PRIMARY		0x0002
62 #define	MAC_CLIENT_FLAGS_MULTI_PRIMARY		0x0004
63 #define	MAC_CLIENT_FLAGS_PASSIVE_PRIMARY	0x0008
64 
65 /*
66  * One of these is instantiated per MAC client promiscuous callback.
67  *
68  * Each element of this structure belongs to two linked list. One
69  * for the mac_client_impl_t (mci_promisc_list) which created allocated
70  * the callback, the other for the mac_impl_t (mi_promisc_list) corresponding
71  * to the MAC client.
72  * The former allows us to do bookkeeping, the latter allows us
73  * to more efficiently dispatch packets to the promiscuous callbacks.
74  */
75 typedef struct mac_promisc_impl_s {			/* Protected by */
76 	mac_cb_t			mpi_mci_link;	/* mi_promisc_lock */
77 	mac_cb_t			mpi_mi_link;	/* mi_promisc_lock */
78 	mac_client_promisc_type_t	mpi_type;	/* WO */
79 	mac_rx_t			mpi_fn;		/* WO */
80 	void				*mpi_arg;	/* WO */
81 	struct mac_client_impl_s	*mpi_mcip;	/* WO */
82 	boolean_t			mpi_no_tx_loop;	/* WO */
83 	boolean_t			mpi_no_phys;	/* WO */
84 	boolean_t			mpi_strip_vlan_tag;	/* WO */
85 	boolean_t			mpi_no_copy;	/* WO */
86 } mac_promisc_impl_t;
87 
88 typedef union mac_tx_percpu_s {
89 	struct {
90 		kmutex_t	_pcpu_tx_lock;
91 		uint_t		_pcpu_tx_refcnt;
92 	} pcpu_lr;
93 	uchar_t		pcpu_pad[64];
94 } mac_tx_percpu_t;
95 
96 #define	pcpu_tx_lock	pcpu_lr._pcpu_tx_lock
97 #define	pcpu_tx_refcnt	pcpu_lr._pcpu_tx_refcnt
98 
99 /*
100  * One of these is instantiated for each MAC client.
101  */
102 struct mac_client_impl_s {			/* Protected by */
103 	struct mac_client_impl_s *mci_client_next;	/* mi_rw_lock */
104 	char			mci_name[MAXNAMELEN];	/* mi_rw_lock */
105 	/*
106 	 * This flow entry will contain all the internal constructs
107 	 * such as SRS etc. for this MAC client. The MAC client may
108 	 * have more than one flow corresponding to each upper client
109 	 * sharing this mac_client_impl_t.
110 	 */
111 	flow_entry_t		*mci_flent;		/* mi_rw_lock */
112 	struct mac_impl_s	*mci_mip;		/* WO */
113 	/*
114 	 * If this is a client that has a pass thru MAC (e.g. a VNIC),
115 	 * then we also keep the handle for the client's upper MAC.
116 	 */
117 	struct mac_impl_s	*mci_upper_mip;		/* WO */
118 
119 	uint32_t		mci_state_flags;	/* WO */
120 	mac_rx_t		mci_rx_fn;		/* Rx Quiescence */
121 	void			*mci_rx_arg;		/* Rx Quiescence */
122 	mac_direct_rx_t		mci_direct_rx_fn;	/* SL */
123 	void			*mci_direct_rx_arg;	/* SL */
124 	mac_rx_t		mci_rx_p_fn;		/* Rx Quiescence */
125 	void			*mci_rx_p_arg;		/* Rx Quiescence */
126 	void			*mci_p_unicast_list;
127 
128 	mac_cb_t		*mci_promisc_list;	/* mi_promisc_lock */
129 
130 	mac_address_t		*mci_unicast;
131 	uint32_t		mci_flags;		/* SL */
132 	krwlock_t		mci_rw_lock;
133 	mac_unicast_impl_t	*mci_unicast_list;	/* mci_rw_lock */
134 
135 	/*
136 	 * The mac_client_impl_t may be shared by multiple clients, i.e
137 	 * multiple VLANs sharing the same MAC client. In this case the
138 	 * address/vid tuples differ and are each associated with their
139 	 * own flow entry, but the rest underlying components SRS, etc,
140 	 * are common.
141 	 *
142 	 * This is only needed to support sun4v vsw. There are several
143 	 * places in MAC we could simplify the code if we removed
144 	 * sun4v support.
145 	 */
146 	flow_entry_t		*mci_flent_list;	/* mci_rw_lock */
147 	uint_t			mci_nflents;		/* mci_rw_lock */
148 	uint_t			mci_nvids;		/* mci_rw_lock */
149 	volatile uint32_t	mci_vidcache;		/* VID cache */
150 
151 	/* Resource Management Functions */
152 	mac_resource_add_t	mci_resource_add;	/* SL */
153 	mac_resource_remove_t	mci_resource_remove;	/* SL */
154 	mac_resource_quiesce_t	mci_resource_quiesce;	/* SL */
155 	mac_resource_restart_t	mci_resource_restart;	/* SL */
156 	mac_resource_bind_t	mci_resource_bind;	/* SL */
157 	void			*mci_resource_arg;	/* SL */
158 
159 
160 	/* Tx notify callback */
161 	kmutex_t		mci_tx_cb_lock;
162 	mac_cb_info_t		mci_tx_notify_cb_info;	/* cb list info */
163 	mac_cb_t		*mci_tx_notify_cb_list;	/* The cb list */
164 	uintptr_t		mci_tx_notify_id;
165 
166 	/* per MAC client stats */			/* None */
167 	mac_misc_stats_t	mci_misc_stat;
168 
169 	flow_tab_t		*mci_subflow_tab;	/* Rx quiescence */
170 
171 	/*
172 	 * Priority range for this MAC client. This the range
173 	 * corresponding to the priority configured (nr_flow_priority).
174 	 */
175 	pri_t			mci_min_pri;
176 	pri_t			mci_max_pri;
177 
178 	/*
179 	 * Hybrid I/O related definitions.
180 	 */
181 	mac_share_handle_t	mci_share;
182 
183 	/* for multicast support */
184 	struct mac_mcast_addrs_s *mci_mcast_addrs;	/* mi_rw_lock */
185 
186 	/*
187 	 * Mac protection related fields
188 	 */
189 	kmutex_t		mci_protect_lock;
190 	uint32_t		mci_protect_flags;	/* SL */
191 	in6_addr_t		mci_v6_mac_token;	/* SL */
192 	in6_addr_t		mci_v6_local_addr;	/* SL */
193 	avl_tree_t		mci_v4_pending_txn;	/* mci_protect_lock */
194 	avl_tree_t		mci_v4_completed_txn;	/* mci_protect_lock */
195 	avl_tree_t		mci_v4_dyn_ip;		/* mci_protect_lock */
196 	avl_tree_t		mci_v6_pending_txn;	/* mci_protect_lock */
197 	avl_tree_t		mci_v6_cid;		/* mci_protect_lock */
198 	avl_tree_t		mci_v6_dyn_ip;		/* mci_protect_lock */
199 	avl_tree_t		mci_v6_slaac_ip;	/* mci_protect_lock */
200 	timeout_id_t		mci_txn_cleanup_tid;	/* mci_protect_lock */
201 
202 	/*
203 	 * Protected by mci_tx_pcpu[0].pcpu_tx_lock
204 	 */
205 	uint_t			mci_tx_flag;
206 	kcondvar_t		mci_tx_cv;
207 
208 	/* Must be last in the structure for dynamic sizing */
209 	mac_tx_percpu_t		mci_tx_pcpu[1];		/* SL */
210 };
211 
212 #define	MAC_CLIENT_IMPL_SIZE						\
213 	(sizeof (mac_client_impl_t) +					\
214 	    (mac_tx_percpu_cnt * sizeof (mac_tx_percpu_t)))
215 
216 extern	int	mac_tx_percpu_cnt;
217 
218 #define	MCIP_TX_SRS(mcip)	\
219 	((mcip)->mci_flent == NULL ? NULL : (mcip)->mci_flent->fe_tx_srs)
220 
221 /* Defensive coding, non-null mcip_flent could be an assert */
222 
223 #define	MCIP_DATAPATH_SETUP(mcip)		\
224 	((mcip)->mci_flent == NULL ? B_FALSE :	\
225 	!((mcip)->mci_flent->fe_flags & FE_MC_NO_DATAPATH))
226 
227 #define	MCIP_RESOURCE_PROPS(mcip)		\
228 	((mcip)->mci_flent == NULL ? NULL :	\
229 	&(mcip)->mci_flent->fe_resource_props)
230 
231 #define	MCIP_EFFECTIVE_PROPS(mcip)		\
232 	(mcip->mci_flent == NULL ? NULL :	\
233 	&(mcip)->mci_flent->fe_effective_props)
234 
235 #define	MCIP_RESOURCE_PROPS_MASK(mcip)		\
236 	((mcip)->mci_flent == NULL ? 0 :	\
237 	(mcip)->mci_flent->fe_resource_props.mrp_mask)
238 
239 #define	MCIP_RESOURCE_PROPS_MAXBW(mcip)		\
240 	((mcip)->mci_flent == NULL ? 0 :	\
241 	(mcip)->mci_flent->fe_resource_props.mrp_maxbw)
242 
243 #define	MCIP_RESOURCE_PROPS_PRIORITY(mcip)		\
244 	((mcip)->mci_flent == NULL ? 0 :	\
245 	(mcip)->mci_flent->fe_resource_props.mrp_priority)
246 
247 #define	MCIP_RESOURCE_PROPS_CPUS(mcip)		\
248 	((mcip)->mci_flent == NULL ? 0 :	\
249 	&(mcip)->mci_flent->fe_resource_props.mrp_cpus)
250 
251 #define	MCIP_RESOURCE_PROPS_NCPUS(mcip)		\
252 	((mcip)->mci_flent == NULL ? 0 :	\
253 	(mcip)->mci_flent->fe_resource_props.mrp_ncpus)
254 
255 #define	MCIP_RESOURCE_PROPS_CPU(mcip)		\
256 	((mcip)->mci_flent == NULL ? 0 :	\
257 	(mcip)->mci_flent->fe_resource_props.mrp_ncpu)
258 
259 /*
260  * We validate the VLAN id of the packet w.r.t the client's vid,
261  * if required (i.e. !MCIS_DISABLE_TX_VID_CHECK). DLS clients
262  * will have MCIS_DISABLE_TX_VID_CHECK set.
263  * (In the case of aggr when we get back packets, due to
264  * the underlying driver being flow controlled, we won't
265  * drop the packet even if it is VLAN tagged as we
266  * don't set MCIS_DISABLE_TX_VID_CHECK for an aggr.)
267  */
268 #define	MAC_VID_CHECK_NEEDED(mcip)					\
269 	(((mcip)->mci_state_flags & MCIS_DISABLE_TX_VID_CHECK) == 0 &&	\
270 	(mcip)->mci_mip->mi_info.mi_nativemedia == DL_ETHER)
271 
272 #define	MAC_VID_CHECK(mcip, mp, err) {					\
273 	if (ntohs(((struct ether_header *)(mp)->b_rptr)->ether_type) ==	\
274 	    ETHERTYPE_VLAN) {						\
275 		/*							\
276 		 * err is set to EINVAL (so the caller can take the	\
277 		 * appropriate action. e.g. freemsg()) for two cases:	\
278 		 * -client is not responsible for filling in the vid.	\
279 		 * -client is responsible for filling in the vid, but	\
280 		 *  the vid doesn't match the vid of the MAC client.	\
281 		 */							\
282 		(err) = EINVAL;						\
283 		if (((mcip)->mci_state_flags & MCIS_TAG_DISABLE) != 0) {\
284 			struct ether_vlan_header	*evhp;		\
285 			uint16_t			vlanid;		\
286 									\
287 			evhp = (struct ether_vlan_header *)(mp)->b_rptr;\
288 			vlanid = VLAN_ID(ntohs(evhp->ether_tci));	\
289 			if (mac_client_check_flow_vid((mcip), vlanid))	\
290 				(err) = 0;				\
291 		}							\
292 	}								\
293 }
294 
295 /*
296  * To allow the hot path to not grab any additional locks, we keep a single
297  * entry VLAN ID cache that caches whether or not a given VID belongs to a
298  * MAC client.
299  */
300 #define	MCIP_VIDCACHE_VALIDSHIFT	31
301 #define	MCIP_VIDCACHE_VIDSHIFT		1
302 #define	MCIP_VIDCACHE_VIDMASK		(UINT16_MAX << MCIP_VIDCACHE_VIDSHIFT)
303 #define	MCIP_VIDCACHE_BOOLSHIFT		0
304 
305 #define	MCIP_VIDCACHE_INVALID		0
306 
307 #define	MCIP_VIDCACHE_CACHE(vid, bool)	\
308 	((1U << MCIP_VIDCACHE_VALIDSHIFT) | \
309 	((vid) << MCIP_VIDCACHE_VIDSHIFT) | \
310 	((bool) ? (1U << MCIP_VIDCACHE_BOOLSHIFT) : 0))
311 
312 #define	MCIP_VIDCACHE_ISVALID(v)	((v) & (1U << MCIP_VIDCACHE_VALIDSHIFT))
313 #define	MCIP_VIDCACHE_VID(v)		\
314 	(((v) & MCIP_VIDCACHE_VIDMASK) >> MCIP_VIDCACHE_VIDSHIFT)
315 #define	MCIP_VIDCACHE_BOOL(v)		((v) & (1U << MCIP_VIDCACHE_BOOLSHIFT))
316 
317 #define	MAC_TAG_NEEDED(mcip)						\
318 	(((mcip)->mci_state_flags & MCIS_TAG_DISABLE) == 0 &&		\
319 	(mcip)->mci_nvids == 1)						\
320 
321 /*
322  * MAC Client Implementation State (mci_state_flags)
323  *
324  * MCIS_IS_VNIC
325  *
326  *	The client is a VNIC.
327  *
328  * MCIS_EXCLUSIVE
329  *
330  *	The client has exclusive control over the MAC, such that it is
331  *	the sole client of the MAC.
332  *
333  * MCIS_TAG_DISABLE
334  *
335  *	MAC will not add VLAN tags to outgoing traffic. If this flag
336  *	is set it is up to the client to add the correct VLAN tag.
337  *
338  * MCIS_STRIP_DISABLE
339  *
340  *	MAC will not strip the VLAN tags on incoming traffic before
341  *	passing it to mci_rx_fn. This only applies to non-bypass
342  *	traffic.
343  *
344  * MCIS_IS_AGGR_PORT
345  *
346  *	The client represents a port on an aggr.
347  *
348  * MCIS_CLIENT_POLL_CAPABLE
349  *
350  *	The client is capable of polling the Rx TCP/UDP softrings.
351  *
352  * MCIS_DESC_LOGGED
353  *
354  *	This flag is set when the client's link info has been logged
355  *	by the mac_log_linkinfo() timer. This ensures that the
356  *	client's link info is only logged once.
357  *
358  * MCIS_SHARE_BOUND
359  *
360  *	This client has an HIO share bound to it.
361  *
362  * MCIS_DISABLE_TX_VID_CHECK
363  *
364  *	MAC will not check the VID of the client's Tx traffic.
365  *
366  * MCIS_USE_DATALINK_NAME
367  *
368  *	The client is using the same name as its underlying MAC. This
369  *	happens when dlmgmtd is unreachable during client creation.
370  *
371  * MCIS_UNICAST_HW
372  *
373  *	The client requires MAC address hardware classification. This
374  *	is only used by sun4v vsw.
375  *
376  * MCIS_IS_AGGR_CLIENT
377  *
378  *	The client sits atop an aggr.
379  *
380  * MCIS_RX_BYPASS_DISABLE
381  *
382  *	Do not allow the client to enable DLS bypass.
383  *
384  * MCIS_NO_UNICAST_ADDR
385  *
386  *	This client has no MAC unicast addresss associated with it.
387  *
388  */
389 /* MCI state flags */
390 #define	MCIS_IS_VNIC			0x0001
391 #define	MCIS_EXCLUSIVE			0x0002
392 #define	MCIS_TAG_DISABLE		0x0004
393 #define	MCIS_STRIP_DISABLE		0x0008
394 #define	MCIS_IS_AGGR_PORT		0x0010
395 #define	MCIS_CLIENT_POLL_CAPABLE	0x0020
396 #define	MCIS_DESC_LOGGED		0x0040
397 #define	MCIS_SHARE_BOUND		0x0080
398 #define	MCIS_DISABLE_TX_VID_CHECK	0x0100
399 #define	MCIS_USE_DATALINK_NAME		0x0200
400 #define	MCIS_UNICAST_HW			0x0400
401 #define	MCIS_IS_AGGR_CLIENT		0x0800
402 #define	MCIS_RX_BYPASS_DISABLE		0x1000
403 #define	MCIS_NO_UNICAST_ADDR		0x2000
404 
405 /* Mac protection flags */
406 #define	MPT_FLAG_V6_LOCAL_ADDR_SET	0x0001
407 #define	MPT_FLAG_PROMISC_FILTERED	0x0002
408 
409 /* in mac_client.c */
410 extern void mac_promisc_client_dispatch(mac_client_impl_t *, mblk_t *);
411 extern void mac_client_init(void);
412 extern void mac_client_fini(void);
413 extern void mac_promisc_dispatch(mac_impl_t *, mblk_t *, mac_client_impl_t *,
414     boolean_t);
415 
416 extern int mac_validate_props(mac_impl_t *, mac_resource_props_t *);
417 
418 extern mac_client_impl_t *mac_vnic_lower(mac_impl_t *);
419 extern mac_client_impl_t *mac_primary_client_handle(mac_impl_t *);
420 extern uint16_t i_mac_flow_vid(flow_entry_t *);
421 extern boolean_t i_mac_capab_get(mac_handle_t, mac_capab_t, void *);
422 
423 extern void mac_unicast_update_clients(mac_impl_t *, mac_address_t *);
424 extern void mac_update_resources(mac_resource_props_t *,
425     mac_resource_props_t *, boolean_t);
426 
427 boolean_t mac_client_check_flow_vid(mac_client_impl_t *, uint16_t);
428 
429 extern boolean_t mac_is_primary_client(mac_client_impl_t *);
430 
431 extern int mac_client_set_rings_prop(mac_client_impl_t *,
432     mac_resource_props_t *, mac_resource_props_t *);
433 extern void mac_set_prim_vlan_rings(mac_impl_t *, mac_resource_props_t *);
434 
435 #ifdef	__cplusplus
436 }
437 #endif
438 
439 #endif	/* _SYS_MAC_CLIENT_IMPL_H */
440