[apple/xnu.git] / bsd / net / dlil.h

/*
 * Copyright (c) 1999-2019 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */
#ifndef DLIL_H
#define DLIL_H
#ifdef KERNEL

#include <sys/kernel_types.h>
#include <net/kpi_interface.h>

enum {
	BPF_TAP_DISABLE,
	BPF_TAP_INPUT,
	BPF_TAP_OUTPUT,
	BPF_TAP_INPUT_OUTPUT
};

/*
 * DLIL_DESC_ETYPE2 - native_type must point to 2 byte ethernet raw protocol,
 *                    variants.native_type_length must be set to 2
 * DLIL_DESC_SAP - native_type must point to 3 byte SAP protocol
 *                 variants.native_type_length must be set to 3
 * DLIL_DESC_SNAP - native_type must point to 5 byte SNAP protocol
 *                  variants.native_type_length must be set to 5
 *
 * All protocols must be in Network byte order.
 *
 * Future interface families may define more protocol types they know about.
 * The type implies the offset and context of the protocol data at native_type.
 * The length of the protocol data specified at native_type must be set in
 * variants.native_type_length.
 */
/* Ethernet specific types */
#define DLIL_DESC_ETYPE2        4
#define DLIL_DESC_SAP           5
#define DLIL_DESC_SNAP          6

#ifdef KERNEL_PRIVATE
#include <net/if.h>
#include <net/if_var.h>
#include <net/classq/classq.h>
#include <net/flowadv.h>
#include <sys/kern_event.h>
#include <kern/thread.h>
#include <kern/locks.h>

#ifdef BSD_KERNEL_PRIVATE
/* Operations on timespecs. */
#define net_timerclear(tvp)     (tvp)->tv_sec = (tvp)->tv_nsec = 0

#define net_timerisset(tvp)     ((tvp)->tv_sec || (tvp)->tv_nsec)

#define net_timercmp(tvp, uvp, cmp)                                     \
	(((tvp)->tv_sec == (uvp)->tv_sec) ?                             \
	((tvp)->tv_nsec cmp (uvp)->tv_nsec) :                           \
	((tvp)->tv_sec cmp (uvp)->tv_sec))

#define net_timeradd(tvp, uvp, vvp) do {                                \
	(vvp)->tv_sec = (tvp)->tv_sec + (uvp)->tv_sec;                  \
	(vvp)->tv_nsec = (tvp)->tv_nsec + (uvp)->tv_nsec;               \
	if ((vvp)->tv_nsec >= (long)NSEC_PER_SEC) {                     \
	        (vvp)->tv_sec++;                                        \
	        (vvp)->tv_nsec -= NSEC_PER_SEC;                         \
	}                                                               \
} while (0)

#define net_timersub(tvp, uvp, vvp) do {                                \
	(vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec;                  \
	(vvp)->tv_nsec = (tvp)->tv_nsec - (uvp)->tv_nsec;               \
	if ((vvp)->tv_nsec < 0) {                                       \
	        (vvp)->tv_sec--;                                        \
	        (vvp)->tv_nsec += NSEC_PER_SEC;                         \
	}                                                               \
} while (0)

#define net_timerusec(tvp, nsp) do {                                    \
	*(nsp) = (tvp)->tv_nsec / NSEC_PER_USEC;                        \
	if ((tvp)->tv_sec > 0)                                          \
	        *(nsp) += ((tvp)->tv_sec * USEC_PER_SEC);               \
} while (0)

#define net_timernsec(tvp, nsp) do {                                    \
	*(nsp) = (tvp)->tv_nsec;                                        \
	if ((tvp)->tv_sec > 0)                                          \
	        *(nsp) += ((tvp)->tv_sec * NSEC_PER_SEC);               \
} while (0)

#if defined(__x86_64__) || defined(__arm64__)
#define net_nsectimer(nsp, tvp) do {                                    \
	u_int64_t __nsp = *(nsp);                                       \
	net_timerclear(tvp);                                            \
	uint64_t __sec = __nsp / NSEC_PER_SEC;                          \
	(tvp)->tv_sec = (__darwin_time_t)__sec;                         \
	(tvp)->tv_nsec = (long)(__nsp - __sec * NSEC_PER_SEC);          \
} while (0)
#else /* 32 bit */
/*
 * NSEC needs to be < 2^31*10^9 to be representable in a struct timespec
 * because __darwin_time_t is 32 bit on 32-bit platforms. This bound
 * is < 2^61. We get a first approximation to convert into seconds using
 * the following values.
 * a = floor(NSEC / 2^29)
 * inv = floor(2^61 / 10^9)
 *
 * The approximation of seconds is correct or too low by 1 unit.
 * So we fix it by computing the remainder.
 */
#define net_nsectimer(nsp, tvp) do {                                    \
	u_int64_t __nsp = *(nsp);                                       \
	net_timerclear(tvp);                                            \
	uint32_t __a = (uint32_t)(__nsp >> 29);                         \
	const uint32_t __inv = 0x89705F41;                              \
	uint32_t __sec = (uint32_t)(((uint64_t)__a * __inv) >> 32);     \
	uint32_t __rem = (uint32_t)(__nsp - __sec * NSEC_PER_SEC);      \
	__sec += ((__rem >= NSEC_PER_SEC) ? 1 : 0);                     \
	(tvp)->tv_sec = (__darwin_time_t)__sec;                         \
	(tvp)->tv_nsec =                                                \
	    (long)((__rem >= NSEC_PER_SEC) ? (__rem - NSEC_PER_SEC) : __rem);   \
} while(0)
#endif /* 32 bit */

struct ifnet;
struct mbuf;
struct ether_header;
struct sockaddr_dl;
struct iff_filter;

#define DLIL_THREADNAME_LEN     32

/*
 * DLIL input thread info
 */
struct dlil_threading_info {
	decl_lck_mtx_data(, input_lck);
	lck_grp_t       *lck_grp;       /* lock group (for lock stats) */
	u_int32_t       input_waiting;  /* DLIL condition of thread */
	u_int32_t       wtot;           /* # of wakeup requests */
	char            input_name[DLIL_THREADNAME_LEN]; /* name storage */
	struct ifnet    *ifp;           /* pointer to interface */
	class_queue_t   rcvq_pkts;      /* queue of pkts */
	struct ifnet_stat_increment_param stats; /* incremental statistics */
	/*
	 * Thread affinity (workloop and DLIL threads).
	 */
	boolean_t       net_affinity;   /* affinity set is available */
	struct thread   *input_thr;     /* input thread */
	struct thread   *wloop_thr;     /* workloop thread */
	struct thread   *poll_thr;      /* poll thread */
	u_int32_t       tag;            /* affinity tag */
#if IFNET_INPUT_SANITY_CHK
	/*
	 * For debugging.
	 */
	u_int64_t       input_mbuf_cnt; /* total # of packets processed */
#endif
};

/*
 * DLIL input thread info (for main/loopback input thread)
 */
struct dlil_main_threading_info {
	struct dlil_threading_info      inp;
	class_queue_t                   lo_rcvq_pkts; /* queue of lo0 pkts */
};

/*
 * The following are shared with kpi_protocol.c so that it may wakeup
 * the input thread to run through packets queued for protocol input.
 */
#define DLIL_INPUT_RUNNING      0x80000000
#define DLIL_INPUT_WAITING      0x40000000
#define DLIL_PROTO_REGISTER     0x20000000
#define DLIL_PROTO_WAITING      0x10000000
#define DLIL_INPUT_TERMINATE    0x08000000
#define DLIL_INPUT_TERMINATE_COMPLETE   0x04000000

/*
 * Flags for dlil_attach_filter()
 */
#define DLIL_IFF_TSO            0x01    /* Interface filter supports TSO */
#define DLIL_IFF_INTERNAL       0x02    /* Apple internal -- do not count towards stats */

/* Input poll interval definitions */
#define IF_RXPOLL_INTERVALTIME_MIN      (1ULL * 1000)           /* 1 us */
#define IF_RXPOLL_INTERVALTIME          (1ULL * 1000 * 1000)    /* 1 ms */

extern int dlil_verbose;
extern uint32_t hwcksum_dbg;
extern uint32_t hwcksum_tx;
extern uint32_t hwcksum_rx;
extern struct dlil_threading_info *dlil_main_input_thread;
extern unsigned int net_rxpoll;
extern uint32_t if_rxpoll;
extern uint32_t if_rxpoll_decay;
extern uint32_t if_rxpoll_interval_pkts;
extern uint32_t if_rcvq_maxlen;

extern void dlil_init(void);

extern errno_t ifp_if_ioctl(struct ifnet *, unsigned long, void *);
extern errno_t ifp_if_output(struct ifnet *, struct mbuf *);
extern void ifp_if_start(struct ifnet *);

extern errno_t dlil_set_bpf_tap(ifnet_t, bpf_tap_mode, bpf_packet_func);

/*
 * Send arp internal bypasses the check for IPv4LL.
 */
extern errno_t dlil_send_arp_internal(ifnet_t, u_int16_t,
    const struct sockaddr_dl *, const struct sockaddr *,
    const struct sockaddr_dl *, const struct sockaddr *);

/*
 * The following constants are used with the net_thread_mark_apply and
 * net_thread_is_unmarked functions to control the bits in the uu_network_marks
 * field of the uthread structure.
 */
#define NET_THREAD_HELD_PF      0x1     /* thread is holding PF lock */
#define NET_THREAD_HELD_DOMAIN  0x2     /* thread is holding domain_proto_mtx */
#define NET_THREAD_CKREQ_LLADDR 0x4     /* thread reqs MACF check for LLADDR */

/*
 * net_thread_marks_t is a pointer to a phantom structure type used for
 * manipulating the uthread:uu_network_marks field.  As an example...
 *
 *   static const u_int32_t bits = NET_THREAD_CKREQ_LLADDR;
 *   struct uthread *uth = get_bsdthread_info(current_thread());
 *
 *   net_thread_marks_t marks = net_thread_marks_push(bits);
 *   VERIFY((uth->uu_network_marks & NET_THREAD_CKREQ_LLADDR) != 0);
 *   net_thread_marks_pop(marks);
 *
 * The net_thread_marks_push() function returns an encoding of the bits
 * that were changed from zero to one in the uu_network_marks field. When
 * the net_thread_marks_pop() function later processes that value, it
 * resets the bits to their previous value.
 *
 * The net_thread_unmarks_push() and net_thread_unmarks_pop() functions
 * are similar to net_thread_marks_push() and net_thread_marks_pop() except
 * they clear the marks bits in the guarded section rather than set them.
 *
 * The net_thread_is_marked() and net_thread_is_unmarked() functions return
 * the subset of the bits that are currently set or cleared (respectively)
 * in the uthread:uu_network_marks field.
 *
 * Finally, the value of the net_thread_marks_none constant is provided for
 * comparing for equality with the value returned when no bits in the marks
 * field are changed by the push.
 *
 * It is not significant that a value of type net_thread_marks_t may
 * compare as equal to the NULL pointer.
 */
struct net_thread_marks;
typedef const struct net_thread_marks *net_thread_marks_t;

extern const net_thread_marks_t net_thread_marks_none;

extern net_thread_marks_t net_thread_marks_push(u_int32_t);
extern net_thread_marks_t net_thread_unmarks_push(u_int32_t);
extern void net_thread_marks_pop(net_thread_marks_t);
extern void net_thread_unmarks_pop(net_thread_marks_t);
extern u_int32_t net_thread_is_marked(u_int32_t);
extern u_int32_t net_thread_is_unmarked(u_int32_t);

extern int dlil_output(ifnet_t, protocol_family_t, mbuf_t, void *,
    const struct sockaddr *, int, struct flowadv *);

extern void dlil_input_packet_list(struct ifnet *, struct mbuf *);
extern void dlil_input_packet_list_extended(struct ifnet *, struct mbuf *,
    u_int32_t, ifnet_model_t);

extern errno_t dlil_resolve_multi(struct ifnet *,
    const struct sockaddr *, struct sockaddr *, size_t);

extern errno_t dlil_send_arp(ifnet_t, u_int16_t, const struct sockaddr_dl *,
    const struct sockaddr *, const struct sockaddr_dl *,
    const struct sockaddr *, u_int32_t);

extern int dlil_attach_filter(ifnet_t, const struct iff_filter *,
    interface_filter_t *, u_int32_t);
extern void dlil_detach_filter(interface_filter_t);

extern void dlil_proto_unplumb_all(ifnet_t);

extern int dlil_post_msg(struct ifnet *, u_int32_t, u_int32_t,
    struct net_event_data *, u_int32_t);

extern void dlil_post_sifflags_msg(struct ifnet *);

extern int dlil_post_complete_msg(struct ifnet *, struct kev_msg *);

extern int dlil_alloc_local_stats(struct ifnet *);

extern void ifnet_filter_update_tso(boolean_t filter_enable);
extern errno_t dlil_rxpoll_validate_params(struct ifnet_poll_params *);
extern void dlil_rxpoll_update_params(struct ifnet *,
    struct ifnet_poll_params *);
extern void ifnet_poll(struct ifnet *);
extern errno_t ifnet_input_poll(struct ifnet *, struct mbuf *,
    struct mbuf *, const struct ifnet_stat_increment_param *);


/*
 * dlil_if_acquire is obsolete. Use ifnet_allocate.
 */
extern int dlil_if_acquire(u_int32_t, const void *, size_t, const char *, struct ifnet **);
/*
 * dlil_if_release is obsolete. The equivalent is called automatically when
 * an interface is detached.
 */
extern void dlil_if_release(struct ifnet *ifp);

extern errno_t dlil_if_ref(struct ifnet *);
extern errno_t dlil_if_free(struct ifnet *);

extern int dlil_node_present(struct ifnet *, struct sockaddr *, int32_t, int,
    int, u_int8_t[48]);
extern void dlil_node_absent(struct ifnet *, struct sockaddr *);
extern int dlil_node_present_v2(struct ifnet *, struct sockaddr *, struct sockaddr_dl *, int32_t, int,
    int, u_int8_t[48]);

extern const void *dlil_ifaddr_bytes(const struct sockaddr_dl *, size_t *,
    kauth_cred_t *);

extern void dlil_report_issues(struct ifnet *, u_int8_t[DLIL_MODIDLEN],
    u_int8_t[DLIL_MODARGLEN]);

#define PROTO_HASH_SLOTS        5

extern int proto_hash_value(u_int32_t);

extern const char *dlil_kev_dl_code_str(u_int32_t);

extern errno_t dlil_rxpoll_set_params(struct ifnet *,
    struct ifnet_poll_params *, boolean_t);
extern errno_t dlil_rxpoll_get_params(struct ifnet *,
    struct ifnet_poll_params *);

extern errno_t dlil_output_handler(struct ifnet *, struct mbuf *);
extern errno_t dlil_input_handler(struct ifnet *, struct mbuf *,
    struct mbuf *, const struct ifnet_stat_increment_param *,
    boolean_t, struct thread *);


/*
 * This is mostly called from the context of the DLIL input thread;
 * because of that there is no need for atomic operations.
 */
__attribute__((always_inline))
static inline void
ifp_inc_traffic_class_in(struct ifnet *ifp, struct mbuf *m)
{
	if (!(m->m_flags & M_PKTHDR)) {
		return;
	}

	switch (m_get_traffic_class(m)) {
	case MBUF_TC_BE:
		ifp->if_tc.ifi_ibepackets++;
		ifp->if_tc.ifi_ibebytes += m->m_pkthdr.len;
		break;
	case MBUF_TC_BK:
		ifp->if_tc.ifi_ibkpackets++;
		ifp->if_tc.ifi_ibkbytes += m->m_pkthdr.len;
		break;
	case MBUF_TC_VI:
		ifp->if_tc.ifi_ivipackets++;
		ifp->if_tc.ifi_ivibytes += m->m_pkthdr.len;
		break;
	case MBUF_TC_VO:
		ifp->if_tc.ifi_ivopackets++;
		ifp->if_tc.ifi_ivobytes += m->m_pkthdr.len;
		break;
	default:
		break;
	}

	if (mbuf_is_traffic_class_privileged(m)) {
		ifp->if_tc.ifi_ipvpackets++;
		ifp->if_tc.ifi_ipvbytes += m->m_pkthdr.len;
	}
}

/*
 * This is called from DLIL output, hence multiple threads could end
 * up modifying the statistics.  We trade off acccuracy for performance
 * by not using atomic operations here.
 */
__attribute__((always_inline))
static inline void
ifp_inc_traffic_class_out(struct ifnet *ifp, struct mbuf *m)
{
	if (!(m->m_flags & M_PKTHDR)) {
		return;
	}

	switch (m_get_traffic_class(m)) {
	case MBUF_TC_BE:
		ifp->if_tc.ifi_obepackets++;
		ifp->if_tc.ifi_obebytes += m->m_pkthdr.len;
		break;
	case MBUF_TC_BK:
		ifp->if_tc.ifi_obkpackets++;
		ifp->if_tc.ifi_obkbytes += m->m_pkthdr.len;
		break;
	case MBUF_TC_VI:
		ifp->if_tc.ifi_ovipackets++;
		ifp->if_tc.ifi_ovibytes += m->m_pkthdr.len;
		break;
	case MBUF_TC_VO:
		ifp->if_tc.ifi_ovopackets++;
		ifp->if_tc.ifi_ovobytes += m->m_pkthdr.len;
		break;
	default:
		break;
	}

	if (mbuf_is_traffic_class_privileged(m)) {
		ifp->if_tc.ifi_opvpackets++;
		ifp->if_tc.ifi_opvbytes += m->m_pkthdr.len;
	}
}
#endif /* BSD_KERNEL_PRIVATE */
#endif /* KERNEL_PRIVATE */
#endif /* KERNEL */
#endif /* DLIL_H */
Commit	Line	Data
1c79356b	1	/*
cb323159	2	* Copyright (c) 1999-2019 Apple Inc. All rights reserved.
5d5c5d0d	3	*
2d21ac55	4	* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
39236c6e	5	*
2d21ac55 A	6	* This file contains Original Code and/or Modifications of Original Code
	7	* as defined in and that are subject to the Apple Public Source License
	8	* Version 2.0 (the 'License'). You may not use this file except in
	9	* compliance with the License. The rights granted to you under the License
	10	* may not be used to create, or enable the creation or redistribution of,
	11	* unlawful or unlicensed copies of an Apple operating system, or to
	12	* circumvent, violate, or enable the circumvention or violation of, any
	13	* terms of an Apple operating system software license agreement.
39236c6e	14	*
2d21ac55 A	15	* Please obtain a copy of the License at
2d21ac55 A	16	* http://www.opensource.apple.com/apsl/ and read it before using this file.
39236c6e	17	*
2d21ac55 A	18	* The Original Code and all software distributed under the License are
2d21ac55 A	19	* distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
8f6c56a5 A	20	* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
8f6c56a5 A	21	* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
2d21ac55 A	22	* FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
	23	* Please see the License for the specific language governing rights and
	24	* limitations under the License.
39236c6e	25	*
2d21ac55	26	* @APPLE_OSREFERENCE_LICENSE_HEADER_END@
1c79356b	27	*/
1c79356b	28	#ifndef DLIL_H
0a7de745	29	#define DLIL_H
91447636	30	#ifdef KERNEL
2d21ac55	31
91447636 A	32	#include <sys/kernel_types.h>
91447636 A	33	#include <net/kpi_interface.h>
1c79356b	34
1c79356b A	35	enum {
	36	BPF_TAP_DISABLE,
	37	BPF_TAP_INPUT,
	38	BPF_TAP_OUTPUT,
	39	BPF_TAP_INPUT_OUTPUT
	40	};
	41
7b1edb79 A	42	/*
	43	* DLIL_DESC_ETYPE2 - native_type must point to 2 byte ethernet raw protocol,
	44	* variants.native_type_length must be set to 2
	45	* DLIL_DESC_SAP - native_type must point to 3 byte SAP protocol
	46	* variants.native_type_length must be set to 3
	47	* DLIL_DESC_SNAP - native_type must point to 5 byte SNAP protocol
	48	* variants.native_type_length must be set to 5
	49	*
	50	* All protocols must be in Network byte order.
	51	*
	52	* Future interface families may define more protocol types they know about.
	53	* The type implies the offset and context of the protocol data at native_type.
	54	* The length of the protocol data specified at native_type must be set in
	55	* variants.native_type_length.
	56	*/
316670eb	57	/* Ethernet specific types */
0a7de745 A	58	#define DLIL_DESC_ETYPE2 4
	59	#define DLIL_DESC_SAP 5
	60	#define DLIL_DESC_SNAP 6
1c79356b	61
91447636	62	#ifdef KERNEL_PRIVATE
2d21ac55 A	63	#include <net/if.h>
2d21ac55 A	64	#include <net/if_var.h>
316670eb A	65	#include <net/classq/classq.h>
316670eb A	66	#include <net/flowadv.h>
2d21ac55 A	67	#include <sys/kern_event.h>
	68	#include <kern/thread.h>
	69	#include <kern/locks.h>
1c79356b	70
316670eb A	71	#ifdef BSD_KERNEL_PRIVATE
316670eb A	72	/* Operations on timespecs. */
0a7de745	73	#define net_timerclear(tvp) (tvp)->tv_sec = (tvp)->tv_nsec = 0
316670eb	74
0a7de745	75	#define net_timerisset(tvp) ((tvp)->tv_sec \|\| (tvp)->tv_nsec)
316670eb	76
0a7de745 A	77	#define net_timercmp(tvp, uvp, cmp) \
	78	(((tvp)->tv_sec == (uvp)->tv_sec) ? \
	79	((tvp)->tv_nsec cmp (uvp)->tv_nsec) : \
39236c6e	80	((tvp)->tv_sec cmp (uvp)->tv_sec))
316670eb	81
0a7de745 A	82	#define net_timeradd(tvp, uvp, vvp) do { \
	83	(vvp)->tv_sec = (tvp)->tv_sec + (uvp)->tv_sec; \
	84	(vvp)->tv_nsec = (tvp)->tv_nsec + (uvp)->tv_nsec; \
	85	if ((vvp)->tv_nsec >= (long)NSEC_PER_SEC) { \
	86	(vvp)->tv_sec++; \
	87	(vvp)->tv_nsec -= NSEC_PER_SEC; \
	88	} \
316670eb A	89	} while (0)
316670eb A	90
0a7de745 A	91	#define net_timersub(tvp, uvp, vvp) do { \
	92	(vvp)->tv_sec = (tvp)->tv_sec - (uvp)->tv_sec; \
	93	(vvp)->tv_nsec = (tvp)->tv_nsec - (uvp)->tv_nsec; \
	94	if ((vvp)->tv_nsec < 0) { \
	95	(vvp)->tv_sec--; \
	96	(vvp)->tv_nsec += NSEC_PER_SEC; \
	97	} \
316670eb A	98	} while (0)
316670eb A	99
cb323159 A	100	#define net_timerusec(tvp, nsp) do { \
	101	*(nsp) = (tvp)->tv_nsec / NSEC_PER_USEC; \
	102	if ((tvp)->tv_sec > 0) \
	103	(nsp) += ((tvp)->tv_sec USEC_PER_SEC); \
	104	} while (0)
	105
0a7de745 A	106	#define net_timernsec(tvp, nsp) do { \
	107	*(nsp) = (tvp)->tv_nsec; \
	108	if ((tvp)->tv_sec > 0) \
	109	(nsp) += ((tvp)->tv_sec NSEC_PER_SEC); \
316670eb A	110	} while (0)
316670eb A	111
3e170ce0	112	#if defined(__x86_64__) \|\| defined(__arm64__)
0a7de745 A	113	#define net_nsectimer(nsp, tvp) do { \
	114	u_int64_t __nsp = *(nsp); \
	115	net_timerclear(tvp); \
	116	uint64_t __sec = __nsp / NSEC_PER_SEC; \
	117	(tvp)->tv_sec = (__darwin_time_t)__sec; \
	118	(tvp)->tv_nsec = (long)(__nsp - __sec * NSEC_PER_SEC); \
316670eb	119	} while (0)
3e170ce0 A	120	#else /* 32 bit */
	121	/*
	122	* NSEC needs to be < 2^31*10^9 to be representable in a struct timespec
	123	* because __darwin_time_t is 32 bit on 32-bit platforms. This bound
	124	* is < 2^61. We get a first approximation to convert into seconds using
	125	* the following values.
	126	* a = floor(NSEC / 2^29)
	127	* inv = floor(2^61 / 10^9)
	128	*
	129	* The approximation of seconds is correct or too low by 1 unit.
	130	* So we fix it by computing the remainder.
	131	*/
0a7de745 A	132	#define net_nsectimer(nsp, tvp) do { \
	133	u_int64_t __nsp = *(nsp); \
	134	net_timerclear(tvp); \
	135	uint32_t __a = (uint32_t)(__nsp >> 29); \
	136	const uint32_t __inv = 0x89705F41; \
	137	uint32_t __sec = (uint32_t)(((uint64_t)__a * __inv) >> 32); \
	138	uint32_t __rem = (uint32_t)(__nsp - __sec * NSEC_PER_SEC); \
	139	__sec += ((__rem >= NSEC_PER_SEC) ? 1 : 0); \
	140	(tvp)->tv_sec = (__darwin_time_t)__sec; \
	141	(tvp)->tv_nsec = \
	142	(long)((__rem >= NSEC_PER_SEC) ? (__rem - NSEC_PER_SEC) : __rem); \
3e170ce0 A	143	} while(0)
3e170ce0 A	144	#endif /* 32 bit */
1c79356b	145
2d21ac55 A	146	struct ifnet;
	147	struct mbuf;
	148	struct ether_header;
	149	struct sockaddr_dl;
2d21ac55	150	struct iff_filter;
91447636	151
0a7de745	152	#define DLIL_THREADNAME_LEN 32
6d2010ae	153
316670eb A	154	/*
	155	* DLIL input thread info
	156	*/
2d21ac55	157	struct dlil_threading_info {
6d2010ae	158	decl_lck_mtx_data(, input_lck);
0a7de745 A	159	lck_grp_t lck_grp; / lock group (for lock stats) */
	160	u_int32_t input_waiting; /* DLIL condition of thread */
	161	u_int32_t wtot; /* # of wakeup requests */
	162	char input_name[DLIL_THREADNAME_LEN]; /* name storage */
	163	struct ifnet ifp; / pointer to interface */
	164	class_queue_t rcvq_pkts; /* queue of pkts */
316670eb A	165	struct ifnet_stat_increment_param stats; /* incremental statistics */
	166	/*
	167	* Thread affinity (workloop and DLIL threads).
	168	*/
0a7de745 A	169	boolean_t net_affinity; /* affinity set is available */
	170	struct thread input_thr; / input thread */
	171	struct thread wloop_thr; / workloop thread */
	172	struct thread poll_thr; / poll thread */
	173	u_int32_t tag; /* affinity tag */
2d21ac55	174	#if IFNET_INPUT_SANITY_CHK
316670eb A	175	/*
	176	* For debugging.
	177	*/
0a7de745	178	u_int64_t input_mbuf_cnt; /* total # of packets processed */
2d21ac55 A	179	#endif
2d21ac55 A	180	};
1c79356b	181
316670eb A	182	/*
	183	* DLIL input thread info (for main/loopback input thread)
	184	*/
	185	struct dlil_main_threading_info {
0a7de745 A	186	struct dlil_threading_info inp;
0a7de745 A	187	class_queue_t lo_rcvq_pkts; /* queue of lo0 pkts */
316670eb A	188	};
316670eb A	189
2d21ac55	190	/*
6d2010ae A	191	* The following are shared with kpi_protocol.c so that it may wakeup
6d2010ae A	192	* the input thread to run through packets queued for protocol input.
0a7de745 A	193	*/
	194	#define DLIL_INPUT_RUNNING 0x80000000
	195	#define DLIL_INPUT_WAITING 0x40000000
	196	#define DLIL_PROTO_REGISTER 0x20000000
	197	#define DLIL_PROTO_WAITING 0x10000000
	198	#define DLIL_INPUT_TERMINATE 0x08000000
	199	#define DLIL_INPUT_TERMINATE_COMPLETE 0x04000000
1c79356b	200
39236c6e A	201	/*
	202	* Flags for dlil_attach_filter()
	203	*/
	204	#define DLIL_IFF_TSO 0x01 /* Interface filter supports TSO */
0a7de745	205	#define DLIL_IFF_INTERNAL 0x02 /* Apple internal -- do not count towards stats */
39236c6e	206
cb323159 A	207	/* Input poll interval definitions */
	208	#define IF_RXPOLL_INTERVALTIME_MIN (1ULL * 1000) /* 1 us */
	209	#define IF_RXPOLL_INTERVALTIME (1ULL * 1000 * 1000) /* 1 ms */
	210
39236c6e A	211	extern int dlil_verbose;
	212	extern uint32_t hwcksum_dbg;
	213	extern uint32_t hwcksum_tx;
	214	extern uint32_t hwcksum_rx;
	215	extern struct dlil_threading_info *dlil_main_input_thread;
cb323159 A	216	extern unsigned int net_rxpoll;
	217	extern uint32_t if_rxpoll;
	218	extern uint32_t if_rxpoll_decay;
	219	extern uint32_t if_rxpoll_interval_pkts;
	220	extern uint32_t if_rcvq_maxlen;
316670eb	221
6d2010ae	222	extern void dlil_init(void);
1c79356b	223
316670eb	224	extern errno_t ifp_if_ioctl(struct ifnet , unsigned long, void );
5ba3f43e A	225	extern errno_t ifp_if_output(struct ifnet , struct mbuf );
5ba3f43e A	226	extern void ifp_if_start(struct ifnet *);
316670eb	227
6d2010ae	228	extern errno_t dlil_set_bpf_tap(ifnet_t, bpf_tap_mode, bpf_packet_func);
91447636 A	229
91447636 A	230	/*
6d2010ae	231	* Send arp internal bypasses the check for IPv4LL.
91447636	232	*/
6d2010ae A	233	extern errno_t dlil_send_arp_internal(ifnet_t, u_int16_t,
	234	const struct sockaddr_dl , const struct sockaddr ,
	235	const struct sockaddr_dl , const struct sockaddr );
1c79356b	236
316670eb	237	/*
39236c6e A	238	* The following constants are used with the net_thread_mark_apply and
	239	* net_thread_is_unmarked functions to control the bits in the uu_network_marks
	240	* field of the uthread structure.
316670eb	241	*/
0a7de745 A	242	#define NET_THREAD_HELD_PF 0x1 /* thread is holding PF lock */
	243	#define NET_THREAD_HELD_DOMAIN 0x2 /* thread is holding domain_proto_mtx */
	244	#define NET_THREAD_CKREQ_LLADDR 0x4 /* thread reqs MACF check for LLADDR */
39236c6e A	245
	246	/*
	247	* net_thread_marks_t is a pointer to a phantom structure type used for
	248	* manipulating the uthread:uu_network_marks field. As an example...
	249	*
	250	* static const u_int32_t bits = NET_THREAD_CKREQ_LLADDR;
	251	* struct uthread *uth = get_bsdthread_info(current_thread());
	252	*
	253	* net_thread_marks_t marks = net_thread_marks_push(bits);
	254	* VERIFY((uth->uu_network_marks & NET_THREAD_CKREQ_LLADDR) != 0);
	255	* net_thread_marks_pop(marks);
	256	*
	257	* The net_thread_marks_push() function returns an encoding of the bits
	258	* that were changed from zero to one in the uu_network_marks field. When
	259	* the net_thread_marks_pop() function later processes that value, it
	260	* resets the bits to their previous value.
	261	*
	262	* The net_thread_unmarks_push() and net_thread_unmarks_pop() functions
	263	* are similar to net_thread_marks_push() and net_thread_marks_pop() except
	264	* they clear the marks bits in the guarded section rather than set them.
	265	*
	266	* The net_thread_is_marked() and net_thread_is_unmarked() functions return
	267	* the subset of the bits that are currently set or cleared (respectively)
	268	* in the uthread:uu_network_marks field.
	269	*
	270	* Finally, the value of the net_thread_marks_none constant is provided for
	271	* comparing for equality with the value returned when no bits in the marks
	272	* field are changed by the push.
	273	*
	274	* It is not significant that a value of type net_thread_marks_t may
	275	* compare as equal to the NULL pointer.
	276	*/
	277	struct net_thread_marks;
	278	typedef const struct net_thread_marks *net_thread_marks_t;
	279
	280	extern const net_thread_marks_t net_thread_marks_none;
316670eb	281
39236c6e A	282	extern net_thread_marks_t net_thread_marks_push(u_int32_t);
	283	extern net_thread_marks_t net_thread_unmarks_push(u_int32_t);
	284	extern void net_thread_marks_pop(net_thread_marks_t);
	285	extern void net_thread_unmarks_pop(net_thread_marks_t);
	286	extern u_int32_t net_thread_is_marked(u_int32_t);
	287	extern u_int32_t net_thread_is_unmarked(u_int32_t);
316670eb	288
6d2010ae	289	extern int dlil_output(ifnet_t, protocol_family_t, mbuf_t, void *,
316670eb	290	const struct sockaddr , int, struct flowadv );
1c79356b	291
6d2010ae	292	extern void dlil_input_packet_list(struct ifnet , struct mbuf );
316670eb A	293	extern void dlil_input_packet_list_extended(struct ifnet , struct mbuf ,
316670eb A	294	u_int32_t, ifnet_model_t);
9bccf70c	295
6d2010ae A	296	extern errno_t dlil_resolve_multi(struct ifnet *,
	297	const struct sockaddr , struct sockaddr , size_t);
	298
	299	extern errno_t dlil_send_arp(ifnet_t, u_int16_t, const struct sockaddr_dl *,
	300	const struct sockaddr , const struct sockaddr_dl ,
316670eb	301	const struct sockaddr *, u_int32_t);
9bccf70c	302
6d2010ae	303	extern int dlil_attach_filter(ifnet_t, const struct iff_filter *,
39236c6e	304	interface_filter_t *, u_int32_t);
6d2010ae	305	extern void dlil_detach_filter(interface_filter_t);
9bccf70c	306
6d2010ae A	307	extern void dlil_proto_unplumb_all(ifnet_t);
6d2010ae A	308
cb323159	309	extern int dlil_post_msg(struct ifnet *, u_int32_t, u_int32_t,
6d2010ae A	310	struct net_event_data *, u_int32_t);
6d2010ae A	311
a39ff7e2 A	312	extern void dlil_post_sifflags_msg(struct ifnet *);
a39ff7e2 A	313
39037602 A	314	extern int dlil_post_complete_msg(struct ifnet , struct kev_msg );
39037602 A	315
316670eb A	316	extern int dlil_alloc_local_stats(struct ifnet *);
316670eb A	317
cb323159 A	318	extern void ifnet_filter_update_tso(boolean_t filter_enable);
	319	extern errno_t dlil_rxpoll_validate_params(struct ifnet_poll_params *);
	320	extern void dlil_rxpoll_update_params(struct ifnet *,
	321	struct ifnet_poll_params *);
	322	extern void ifnet_poll(struct ifnet *);
	323	extern errno_t ifnet_input_poll(struct ifnet , struct mbuf ,
	324	struct mbuf , const struct ifnet_stat_increment_param );
	325
5ba3f43e	326
6d2010ae A	327	/*
	328	* dlil_if_acquire is obsolete. Use ifnet_allocate.
	329	*/
a39ff7e2	330	extern int dlil_if_acquire(u_int32_t, const void , size_t, const char , struct ifnet **);
6d2010ae	331	/*
2d21ac55 A	332	* dlil_if_release is obsolete. The equivalent is called automatically when
	333	* an interface is detached.
	334	*/
6d2010ae	335	extern void dlil_if_release(struct ifnet *ifp);
9bccf70c	336
6d2010ae A	337	extern errno_t dlil_if_ref(struct ifnet *);
6d2010ae A	338	extern errno_t dlil_if_free(struct ifnet *);
d1ecb069	339
cb323159	340	extern int dlil_node_present(struct ifnet , struct sockaddr , int32_t, int,
316670eb A	341	int, u_int8_t[48]);
316670eb A	342	extern void dlil_node_absent(struct ifnet , struct sockaddr );
cb323159 A	343	extern int dlil_node_present_v2(struct ifnet , struct sockaddr , struct sockaddr_dl *, int32_t, int,
cb323159 A	344	int, u_int8_t[48]);
316670eb	345
39236c6e A	346	extern const void dlil_ifaddr_bytes(const struct sockaddr_dl , size_t *,
	347	kauth_cred_t *);
	348
	349	extern void dlil_report_issues(struct ifnet *, u_int8_t[DLIL_MODIDLEN],
	350	u_int8_t[DLIL_MODARGLEN]);
	351
cb323159	352	#define PROTO_HASH_SLOTS 5
39236c6e A	353
	354	extern int proto_hash_value(u_int32_t);
	355
	356	extern const char *dlil_kev_dl_code_str(u_int32_t);
	357
	358	extern errno_t dlil_rxpoll_set_params(struct ifnet *,
	359	struct ifnet_poll_params *, boolean_t);
	360	extern errno_t dlil_rxpoll_get_params(struct ifnet *,
	361	struct ifnet_poll_params *);
	362
39037602 A	363	extern errno_t dlil_output_handler(struct ifnet , struct mbuf );
	364	extern errno_t dlil_input_handler(struct ifnet , struct mbuf ,
	365	struct mbuf , const struct ifnet_stat_increment_param ,
	366	boolean_t, struct thread *);
	367
5ba3f43e A	368
	369	/*
	370	* This is mostly called from the context of the DLIL input thread;
	371	* because of that there is no need for atomic operations.
	372	*/
	373	__attribute__((always_inline))
	374	static inline void
	375	ifp_inc_traffic_class_in(struct ifnet ifp, struct mbuf m)
	376	{
0a7de745	377	if (!(m->m_flags & M_PKTHDR)) {
5ba3f43e	378	return;
0a7de745	379	}
5ba3f43e A	380
	381	switch (m_get_traffic_class(m)) {
	382	case MBUF_TC_BE:
	383	ifp->if_tc.ifi_ibepackets++;
	384	ifp->if_tc.ifi_ibebytes += m->m_pkthdr.len;
	385	break;
	386	case MBUF_TC_BK:
	387	ifp->if_tc.ifi_ibkpackets++;
	388	ifp->if_tc.ifi_ibkbytes += m->m_pkthdr.len;
	389	break;
	390	case MBUF_TC_VI:
	391	ifp->if_tc.ifi_ivipackets++;
	392	ifp->if_tc.ifi_ivibytes += m->m_pkthdr.len;
	393	break;
	394	case MBUF_TC_VO:
	395	ifp->if_tc.ifi_ivopackets++;
	396	ifp->if_tc.ifi_ivobytes += m->m_pkthdr.len;
	397	break;
	398	default:
	399	break;
	400	}
	401
	402	if (mbuf_is_traffic_class_privileged(m)) {
	403	ifp->if_tc.ifi_ipvpackets++;
	404	ifp->if_tc.ifi_ipvbytes += m->m_pkthdr.len;
	405	}
	406	}
	407
	408	/*
	409	* This is called from DLIL output, hence multiple threads could end
	410	* up modifying the statistics. We trade off acccuracy for performance
	411	* by not using atomic operations here.
	412	*/
	413	__attribute__((always_inline))
	414	static inline void
	415	ifp_inc_traffic_class_out(struct ifnet ifp, struct mbuf m)
	416	{
0a7de745	417	if (!(m->m_flags & M_PKTHDR)) {
5ba3f43e	418	return;
0a7de745	419	}
5ba3f43e A	420
	421	switch (m_get_traffic_class(m)) {
	422	case MBUF_TC_BE:
	423	ifp->if_tc.ifi_obepackets++;
	424	ifp->if_tc.ifi_obebytes += m->m_pkthdr.len;
	425	break;
	426	case MBUF_TC_BK:
	427	ifp->if_tc.ifi_obkpackets++;
	428	ifp->if_tc.ifi_obkbytes += m->m_pkthdr.len;
	429	break;
	430	case MBUF_TC_VI:
	431	ifp->if_tc.ifi_ovipackets++;
	432	ifp->if_tc.ifi_ovibytes += m->m_pkthdr.len;
	433	break;
	434	case MBUF_TC_VO:
	435	ifp->if_tc.ifi_ovopackets++;
	436	ifp->if_tc.ifi_ovobytes += m->m_pkthdr.len;
	437	break;
	438	default:
	439	break;
	440	}
	441
	442	if (mbuf_is_traffic_class_privileged(m)) {
	443	ifp->if_tc.ifi_opvpackets++;
	444	ifp->if_tc.ifi_opvbytes += m->m_pkthdr.len;
	445	}
	446	}
316670eb	447	#endif /* BSD_KERNEL_PRIVATE */
91447636 A	448	#endif /* KERNEL_PRIVATE */
91447636 A	449	#endif /* KERNEL */
1c79356b	450	#endif /* DLIL_H */