xref: /xnu-10002.61.3/bsd/skywalk/packet/os_packet_private.h (revision 0f4c859e951fba394238ab619495c4e1d54d0f34)
1 /*
2  * Copyright (c) 2016-2022 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
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.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
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.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 
29 #ifndef _SKYWALK_OS_PACKET_PRIVATE_H_
30 #define _SKYWALK_OS_PACKET_PRIVATE_H_
31 
32 #if defined(PRIVATE) || defined(BSD_KERNEL_PRIVATE)
33 #include <skywalk/os_packet.h>
34 #include <skywalk/os_nexus_private.h>
35 #include <skywalk/os_channel_private.h>
36 #include <libkern/OSByteOrder.h>
37 #include <netinet/in.h>
38 #include <net/ethernet.h>
39 
40 #if defined(BSD_KERNEL_PRIVATE)
41 /*
42  * Flow (currently for kernel, potentially for userland one day).
43  *
44  * XXX: When we expose this to userland, we need to be make sure to NOT
45  * expose kernel pointer/address values embedded within.
46  *
47  * Values in flow_{l2,l3,l4} are stored in network byte order.  Pointers
48  * are defined using mach_vm_address_t because it's stable across user
49  * and kernel, and therefore keeps the structure size the same.
50  *
51  * Because this structure might be initialized on a per-packet allocation
52  * basis, it as well as some of its member sub-subtructures are allocated
53  * on a 16-bytes address boundary to allow 128-bit operations on platforms
54  * that support them.
55  *
56  * XXX: when adding new fields, try to leverage __pad ones first.
57  *
58  * TODO: we should consider embedding a flow_key structure here and
59  * use that to store the tuples.  That way we can leverage that for
60  * flow lookups without having to copy things back-and-forth.
61  */
62 struct __flow {
63 	union {
64 		/*
65 		 * The following is always zeroed out on each alloc.
66 		 */
67 		struct __flow_init {
68 			/*
69 			 * Layer 3
70 			 */
71 			struct __flow_l3 {
72 				union {
73 					struct __flow_l3_ipv4_addrs {
74 						struct in_addr _src;
75 						struct in_addr _dst;
76 					} _l3_ipv4;
77 					struct __flow_l3_ipv6_addrs {
78 						struct in6_addr _src;
79 						struct in6_addr _dst;
80 					} _l3_ipv6;
81 				};
82 				uint8_t  _l3_ip_ver;
83 				uint8_t  _l3_proto;
84 				uint8_t  _l3_hlen;
85 				unsigned _l3_is_frag : 1;
86 				unsigned _l3_is_first_frag : 1;
87 				unsigned _l3_reserved_flags : 6;
88 				uint32_t _l3_frag_id;
89 				mach_vm_address_t _l3_ptr;
90 			} __l3;
91 			/*
92 			 * AQM
93 			 */
94 			struct __flow_classq {
95 				uint32_t _fcq_hash;  /* classq-specific hash */
96 				uint32_t _fcq_flags; /* classq-specific flags */
97 			} __classq;
98 			/*
99 			 * Misc.
100 			 */
101 			uint32_t __ulen;      /* user data length */
102 			uint8_t  __ulp_encap; /* e.g. IPPROTO_QUIC */
103 			uint8_t  __pad[3];
104 			uint64_t __pad64[2];
105 			/*
106 			 * Flow Source.
107 			 */
108 			struct __flow_source {
109 				union {
110 					/* source identifier */
111 					uint64_t _fsrc_id_64[2];
112 					uint32_t _fsrc_id_32[4];
113 					uuid_t   _fsrc_id;
114 				} __attribute__((aligned(sizeof(uint64_t))));
115 				flowadv_idx_t _fsrc_fidx; /* flow adv. index */
116 				uint8_t       _fsrc_type; /* FLOWSRC_* mbuf.h */
117 				uint8_t       _fsrc_pad[3];
118 			} __source;
119 			/*
120 			 * Policy.
121 			 */
122 			struct __flow_policy {
123 				uint32_t _fpc_id; /* policy id of pkt sender */
124 				uint32_t _fpc_skip_id; /* skip policy id of pkt sender */
125 				union {
126 					/* process identifier */
127 					uint64_t _fpc_euuid_64[2];
128 					uint32_t _fpc_euuid_32[4];
129 					uuid_t   _fpc_euuid;
130 				} __attribute__((aligned(sizeof(uint64_t))));
131 			} __policy;
132 		} flow_init;
133 		uint64_t flow_init_data[16];
134 	} __attribute((aligned(16)));
135 #define flow_l3                 flow_init.__l3
136 #define flow_classq             flow_init.__classq
137 #define flow_ulen               flow_init.__ulen
138 #define flow_ulp_encap          flow_init.__ulp_encap
139 #define flow_source             flow_init.__source
140 #define flow_policy             flow_init.__policy
141 
142 #define flow_ipv4_addrs         flow_l3._l3_ipv4
143 #define flow_ipv4_src           flow_l3._l3_ipv4._src
144 #define flow_ipv4_dst           flow_l3._l3_ipv4._dst
145 #define flow_ipv6_addrs         flow_l3._l3_ipv6
146 #define flow_ipv6_src           flow_l3._l3_ipv6._src
147 #define flow_ipv6_dst           flow_l3._l3_ipv6._dst
148 #define flow_ip_ver             flow_l3._l3_ip_ver
149 #define flow_ip_proto           flow_l3._l3_proto
150 #define flow_ip_hlen            flow_l3._l3_hlen
151 #define flow_ip_hdr             flow_l3._l3_ptr
152 #define flow_ip_frag_id         flow_l3._l3_frag_id
153 #define flow_ip_is_frag         flow_l3._l3_is_frag
154 #define flow_ip_is_first_frag   flow_l3._l3_is_first_frag
155 
156 #define flow_classq_hash        flow_classq._fcq_hash
157 #define flow_classq_flags       flow_classq._fcq_flags
158 
159 #define flow_src_token          flow_source._fsrc_id_32[0]
160 #define flow_src_id             flow_source._fsrc_id
161 #define flow_src_fidx           flow_source._fsrc_fidx
162 #define flow_src_type           flow_source._fsrc_type
163 
164 #define flow_policy_id          flow_policy._fpc_id
165 #define flow_skip_policy_id     flow_policy._fpc_skip_id
166 #define flow_policy_euuid       flow_policy._fpc_euuid
167 
168 	/*
169 	 * Layer 4.
170 	 */
171 	union {
172 		struct __flow_l4 {
173 			union {
174 				struct __flow_l4_tcp {
175 					in_port_t _src;
176 					in_port_t _dst;
177 					uint32_t _seq;
178 					uint32_t _ack;
179 					union {
180 						struct {
181 #if BYTE_ORDER == LITTLE_ENDIAN
182 							uint8_t _tcp_res:4;
183 							uint8_t _off:4;
184 #else /* BYTE_ORDER == BIG_ENDIAN */
185 							uint8_t _off:4;
186 							uint8_t _tcp_res:4;
187 #endif /* BYTE_ORDER == BIG_ENDIAN */
188 							uint8_t _flags;
189 							uint16_t _win;
190 						};
191 						uint32_t _ofw;
192 					};
193 				} _l4_tcp;
194 				struct __flow_l4_udp {
195 					in_port_t _src;
196 					in_port_t _dst;
197 					uint32_t _ls;
198 				} _l4_udp;
199 				struct __flow_l4_esp {
200 					uint32_t _spi;
201 				} _l4_esp;
202 			};
203 			uint8_t _l4_hlen;
204 			uint8_t _l4_agg_fast;
205 			uint8_t _l4_pad[6];
206 			mach_vm_address_t _l4_ptr;
207 		} flow_l4;
208 		uint64_t flow_l4_data[4];
209 	} __attribute((aligned(sizeof(uint64_t))));
210 #define flow_tcp                flow_l4._l4_tcp
211 #define flow_tcp_src            flow_l4._l4_tcp._src
212 #define flow_tcp_dst            flow_l4._l4_tcp._dst
213 #define flow_tcp_seq            flow_l4._l4_tcp._seq
214 #define flow_tcp_ack            flow_l4._l4_tcp._ack
215 #define flow_tcp_off            flow_l4._l4_tcp._off
216 #define flow_tcp_flags          flow_l4._l4_tcp._flags
217 #define flow_tcp_win            flow_l4._l4_tcp._win
218 #define flow_tcp_hlen           flow_l4._l4_hlen
219 #define flow_tcp_hdr            flow_l4._l4_ptr
220 #define flow_tcp_agg_fast       flow_l4._l4_agg_fast
221 #define flow_udp                flow_l4._l4_udp
222 #define flow_udp_src            flow_l4._l4_udp._src
223 #define flow_udp_dst            flow_l4._l4_udp._dst
224 #define flow_udp_hlen           flow_l4._l4_hlen
225 #define flow_udp_hdr            flow_l4._l4_ptr
226 #define flow_esp_spi            flow_l4._l4_esp._spi
227 } __attribute((aligned(16)));
228 #endif /* BSD_KERNEL_PRIVATE */
229 
230 /*
231  * Maximum size of L2, L3 & L4 headers combined.
232  */
233 #define PKT_MAX_PROTO_HEADER_SIZE       256
234 
235 /* based on 2KB buflet size */
236 #define BUFLETS_MIN             1       /* Ethernet MTU (default) */
237 #define BUFLETS_9K_JUMBO        5       /* 9000 bytes MTU */
238 #define BUFLETS_GSO             46      /* 64KB GSO, Ethernet MTU */
239 
240 /*
241  * Common buflet structure shared by {__user,__kern}_buflet.
242  */
243 struct __buflet {
244 	union {
245 		/* for skmem batch alloc/free */
246 		uint64_t __buflet_next;
247 		/* address of next buflet in chain */
248 		const mach_vm_address_t __nbft_addr;
249 	};
250 	/* buffer data address */
251 	const mach_vm_address_t __baddr;
252 	/* index of buflet object in the owning buflet region */
253 	const obj_idx_t __bft_idx;
254 	/* buffer object index in buffer region */
255 	const obj_idx_t __bidx;
256 	/* object index in buflet region of next buflet(for buflet chaining) */
257 	const obj_idx_t __nbft_idx;
258 	const uint32_t  __dlim;         /* maximum length */
259 	uint32_t        __doff;         /* offset of data in buflet */
260 	uint32_t        __dlen;         /* length of data in buflet */
261 	const uint16_t  __flag;
262 #define BUFLET_FLAG_EXTERNAL    0x0001
263 #define BUFLET_FLAG_LARGE_BUF   0x0002 /* buflet holds large buffer */
264 } __attribute((packed));
265 
266 /*
267  * A buflet represents the smallest buffer fragment representing
268  * part of the packet.  The index refers to the position of the buflet
269  * in the pool, and the data length represents the actual payload
270  * size -- not the buflet size itself as it is fixed for all objects
271  * in the pool.
272  */
273 struct __user_buflet {
274 	/*
275 	 * Common area between user and kernel variants.
276 	 */
277 	struct __buflet buf_com;
278 #define buf_addr        buf_com.__baddr
279 #define buf_nbft_addr   buf_com.__nbft_addr
280 #define buf_idx         buf_com.__bidx
281 #define buf_nbft_idx    buf_com.__nbft_idx
282 #define buf_dlim        buf_com.__dlim
283 #define buf_dlen        buf_com.__dlen
284 #define buf_doff        buf_com.__doff
285 #define buf_flag        buf_com.__flag
286 #define buf_bft_idx_reg buf_com.__bft_idx
287 };
288 
289 #define BUFLET_HAS_LARGE_BUF(_buf)    \
290 	(((_buf)->buf_flag & BUFLET_FLAG_LARGE_BUF) != 0)
291 
292 #define BUF_BADDR(_buf, _addr)                                              \
293 	*__DECONST(mach_vm_address_t *, &(_buf)->buf_addr) =                \
294 	(mach_vm_address_t)(_addr)
295 
296 #define BUF_BIDX(_buf, _idx)                                                \
297 	*__DECONST(obj_idx_t *, &(_buf)->buf_idx) = (obj_idx_t)(_idx)
298 
299 #define BUF_NBFT_ADDR(_buf, _addr)                                          \
300 	*__DECONST(mach_vm_address_t *, &(_buf)->buf_nbft_addr) =           \
301 	(mach_vm_address_t)(_addr)
302 
303 #define BUF_NBFT_IDX(_buf, _idx)                                            \
304 	*__DECONST(obj_idx_t *, &(_buf)->buf_nbft_idx) = (obj_idx_t)(_idx)
305 
306 #define BUF_BFT_IDX_REG(_buf, _idx)    \
307 	*__DECONST(obj_idx_t *, &(_buf)->buf_bft_idx_reg) = (_idx)
308 
309 #define UBUF_LINK(_pubft, _ubft) do {                                   \
310 	ASSERT((_ubft) != NULL);                                        \
311 	BUF_NBFT_ADDR(_pubft, _ubft);                                   \
312 	BUF_NBFT_IDX(_pubft, (_ubft)->buf_bft_idx_reg);                 \
313 } while (0)
314 
315 #ifdef KERNEL
316 #define BUF_CTOR(_buf, _baddr, _bidx, _dlim, _dlen, _doff, _nbaddr, _nbidx, _bflag) do {  \
317 	_CASSERT(sizeof ((_buf)->buf_addr) == sizeof (mach_vm_address_t)); \
318 	_CASSERT(sizeof ((_buf)->buf_idx) == sizeof (obj_idx_t));       \
319 	_CASSERT(sizeof ((_buf)->buf_dlim) == sizeof (uint32_t));       \
320 	BUF_BADDR(_buf, _baddr);                                        \
321 	BUF_NBFT_ADDR(_buf, _nbaddr);                                   \
322 	BUF_BIDX(_buf, _bidx);                                          \
323 	BUF_NBFT_IDX(_buf, _nbidx);                                     \
324 	*(uint32_t *)(uintptr_t)&(_buf)->buf_dlim = (_dlim);            \
325 	(_buf)->buf_dlen = (_dlen);                                     \
326 	(_buf)->buf_doff = (_doff);                                     \
327 	*(uint16_t *)(uintptr_t)&(_buf)->buf_flag = (_bflag);           \
328 } while (0)
329 
330 #define BUF_INIT(_buf, _dlen, _doff) do {                               \
331 	(_buf)->buf_dlen = (_dlen);                                     \
332 	(_buf)->buf_doff = (_doff);                                     \
333 } while (0)
334 
335 #endif /* KERNEL */
336 
337 #ifdef KERNEL
338 #define BUF_IN_RANGE(_buf)                                              \
339 	((_buf)->buf_addr >= (mach_vm_address_t)(_buf)->buf_objaddr &&  \
340 	((uintptr_t)(_buf)->buf_addr + (_buf)->buf_dlim) <=             \
341 	((uintptr_t)(_buf)->buf_objaddr + (_buf)->buf_objlim) &&        \
342 	((_buf)->buf_doff + (_buf)->buf_dlen) <= (_buf)->buf_dlim)
343 #else /* !KERNEL */
344 #define BUF_IN_RANGE(_buf)                                              \
345 	(((_buf)->buf_doff + (_buf)->buf_dlen) <= (_buf)->buf_dlim)
346 #endif /* !KERNEL */
347 
348 /*
349  * Metadata preamble.  This structure is placed at begining of each
350  * __{user,kern}_{quantum,packet} object.  Each user metadata object has a
351  * unique red zone pattern, which is an XOR of the redzone cookie and
352  * offset of the metadata object in the object's region.  Due to the use
353  * of tagged pointer, we need the structure size to be multiples of 16.
354  * See SK_PTR_TAG() definition for details.
355  */
356 struct __metadata_preamble {
357 	union {
358 		uint64_t        _mdp_next;      /* for batch alloc/free (K) */
359 		uint64_t        mdp_redzone;    /* red zone cookie (U) */
360 	};
361 	const obj_idx_t         mdp_idx;        /* index within region (UK) */
362 	uint16_t                mdp_type;       /* nexus_meta_type_t (UK) */
363 	uint16_t                mdp_subtype;    /* nexus_meta_subtype_t (UK) */
364 };
365 
366 #define METADATA_PREAMBLE_SZ    (sizeof (struct __metadata_preamble))
367 
368 #define METADATA_PREAMBLE(_md)                  \
369 	((struct __metadata_preamble *)         \
370 	((mach_vm_address_t)(_md) - METADATA_PREAMBLE_SZ))
371 
372 #define METADATA_IDX(_md)                       \
373 	(METADATA_PREAMBLE(_md)->mdp_idx)
374 
375 #define METADATA_TYPE(_md)                      \
376 	(METADATA_PREAMBLE(_md)->mdp_type)
377 
378 #define METADATA_SUBTYPE(_md)                   \
379 	(METADATA_PREAMBLE(_md)->mdp_subtype)
380 
381 /*
382  * Common packet structure shared by {__user,__kern}_quantum.
383  */
384 struct __quantum {
385 	union {
386 		uuid_t          __uuid;         /* flow UUID */
387 		uint8_t         __val8[16];
388 		uint16_t        __val16[8];
389 		uint32_t        __val32[4];
390 		uint64_t        __val64[2];
391 	} __flow_id_u;
392 #define __q_flow_id             __flow_id_u.__uuid
393 #define __q_flow_id_val8        __flow_id_u.__val8
394 #define __q_flow_id_val16       __flow_id_u.__val16
395 #define __q_flow_id_val32       __flow_id_u.__val32
396 #define __q_flow_id_val64       __flow_id_u.__val64
397 
398 	uint32_t                __q_len;
399 
400 	/* QoS service class, see packet_svc_class_t */
401 	uint32_t                __q_svc_class;  /* PKT_SC_* values */
402 
403 	/*
404 	 * See notes on _QUM_{INTERNALIZE,EXTERNALIZE}() regarding
405 	 * portion of this structure above __flags that gets copied.
406 	 * Adding more user-mutable fields after __flags would also
407 	 * require adjusting those macros as well.
408 	 */
409 	volatile uint16_t       __q_flags;      /* QUMF_* flags */
410 	uint16_t                __q_pad[3];
411 } __attribute((aligned(sizeof(uint64_t))));
412 
413 /*
414  * Quantum.
415  *
416  * This structure is aligned for efficient copy and accesses.
417  * It is the user version of the __kernel_quantum structure.
418  *
419  * XXX: Do NOT store kernel pointer/address values here.
420  */
421 struct __user_quantum {
422 	/*
423 	 * Common area between user and kernel variants.
424 	 */
425 	struct __quantum qum_com;
426 #define qum_flow_id             qum_com.__q_flow_id
427 #define qum_flow_id_val8        qum_com.__q_flow_id_val8
428 #define qum_flow_id_val16       qum_com.__q_flow_id_val16
429 #define qum_flow_id_val32       qum_com.__q_flow_id_val32
430 #define qum_flow_id_val64       qum_com.__q_flow_id_val64
431 #define qum_len                 qum_com.__q_len
432 #define qum_qflags              qum_com.__q_flags
433 #define qum_svc_class           qum_com.__q_svc_class
434 
435 	/*
436 	 * Userland specific.
437 	 */
438 	struct __user_buflet    qum_buf[1];             /* 1 buflet */
439 	/*
440 	 * use count for packet.
441 	 */
442 	uint16_t qum_usecnt;
443 } __attribute((aligned(sizeof(uint64_t))));
444 
445 /*
446  * Valid values for (16-bit) qum_qflags.
447  */
448 #define QUM_F_FINALIZED         0x0001  /* has been finalized */
449 #define QUM_F_DROPPED           0x0002  /* has been dropped */
450 #define QUM_F_FLOW_CLASSIFIED   0x0010  /* flow has been classified */
451 #ifdef KERNEL
452 #define QUM_F_INTERNALIZED      0x1000  /* has been internalized */
453 #define QUM_F_KERNEL_ONLY       0x8000  /* kernel only; no user counterpart */
454 
455 /* invariant flags we want to keep */
456 #define QUM_F_SAVE_MASK         (QUM_F_KERNEL_ONLY)
457 /* kernel-only flags that's never externalized */
458 #define QUM_F_KERNEL_FLAGS      (QUM_F_INTERNALIZED|QUM_F_KERNEL_ONLY)
459 #endif /* KERNEL */
460 
461 #ifdef KERNEL
462 #define _KQUM_CTOR(_kqum, _flags, _len, _baddr, _bidx, _dlim, _qidx) do {    \
463 	(_kqum)->qum_flow_id_val64[0] = 0;                                   \
464 	(_kqum)->qum_flow_id_val64[1] = 0;                                   \
465 	(_kqum)->qum_qflags = (_flags);                                      \
466 	(_kqum)->qum_len = (_len);                                           \
467 	_CASSERT(sizeof(METADATA_IDX(_kqum)) == sizeof(obj_idx_t));          \
468 	*(obj_idx_t *)(uintptr_t)&METADATA_IDX(_kqum) = (_qidx);             \
469 	BUF_CTOR(&(_kqum)->qum_buf[0], (_baddr), (_bidx), (_dlim), 0, 0, 0,  \
470 	    OBJ_IDX_NONE, 0);                                                \
471 } while (0)
472 
473 #define _KQUM_INIT(_kqum, _flags, _len, _qidx) do {                          \
474 	(_kqum)->qum_flow_id_val64[0] = 0;                                   \
475 	(_kqum)->qum_flow_id_val64[1] = 0;                                   \
476 	(_kqum)->qum_qflags = (_flags);                                      \
477 	(_kqum)->qum_len = (_len);                                           \
478 	BUF_INIT(&(_kqum)->qum_buf[0], 0, 0);                                \
479 } while (0)
480 #endif /* KERNEL */
481 
482 /*
483  * Common packet structure shared by {__user,__kern}_packet.
484  */
485 struct __packet_com {
486 	/* Link layer (offset relevant to first buflet) */
487 	uint16_t __link_flags;                          /* PKT_LINKF_* flags */
488 
489 	/*
490 	 * Headroom/protocol header length
491 	 *
492 	 * Since the security model of Skywalk nexus is that we doesn't trust
493 	 * packets either from above (userspace) or below (driver/firmware),
494 	 * the only metadata field that nexus makes use of from external is the
495 	 * headroom. Based on headroom, the flowswitch starts demux routine on
496 	 * l2 header, if any. The l2_len is stored in this step. Then the flow
497 	 * extraction (l3+l4 flow) begins parsing from (headroom + l2_len).
498 	 *
499 	 * __headroom is the empty buffer space before any packet data,
500 	 * it is also the equivalent to the first header offset.
501 	 *
502 	 * __l2_len is l2 (link layer) protocol header length, if any.
503 	 */
504 	uint8_t __headroom;
505 	uint8_t __l2_len;
506 
507 	/*
508 	 * Checksum offload.
509 	 *
510 	 * Partial checksum does not require any header parsing and is
511 	 * therefore simpler to implement both in software and hardware.
512 	 *
513 	 * On transmit, PKT_CSUMF_PARTIAL indicates that a partial one's
514 	 * complement checksum to be computed on the span starting from
515 	 * pkt_csum_tx_start_off to the end of the packet, and have the
516 	 * resulted checksum value written at the location specified by
517 	 * pkt_csum_tx_stuff_off.
518 	 *
519 	 * The PKT_CSUMF_ZERO_INVERT flag is used on transmit to indicate
520 	 * that the value 0xffff (negative 0 in one's complement) must be
521 	 * substituted for the value of 0.  This is set for UDP packets,
522 	 * since otherwise the receiver may not validate the checksum
523 	 * (UDP/IPv4), or drop the packet altogether (UDP/IPv6).
524 	 *
525 	 * On receive, PKT_CSUMF_PARTIAL indicates that a partial one's
526 	 * complement checksum has been computed on the span beginning at
527 	 * pkt_csum_rx_start_off to the end of the packet, and that the
528 	 * computed value is now stored in pkt_csum_rx_value.
529 	 *
530 	 * All offsets are relative to the base of the first buflet.
531 	 */
532 	uint32_t __csum_flags;                          /* PKT_CSUMF_* flags */
533 	union {
534 		struct {
535 			uint16_t __csum_start_off;      /* start offset */
536 			uint16_t __csum_value;          /* checksum value */
537 		} __csum_rx;
538 		struct {
539 			uint16_t __csum_start_off;      /* start offset */
540 			uint16_t __csum_stuff_off;      /* stuff offset */
541 		} __csum_tx;
542 		uint32_t __csum_data;
543 	};
544 
545 	/* Compression generation count */
546 	uint32_t __comp_gencnt;
547 
548 	/*
549 	 * Trace ID for each sampled packet.
550 	 * Non-zero ID indicates that the packet is being actively traced.
551 	 */
552 	packet_trace_id_t __trace_id;
553 
554 	/* Aggregation type */
555 	uint8_t __aggr_type;                     /* PKT_AGGR_* values */
556 	uint8_t __seg_cnt;                       /* Number of LRO-packets */
557 
558 	uint16_t __proto_seg_sz;                 /* Protocol segment size */
559 
560 	/*
561 	 * See notes on _PKT_{INTERNALIZE,EXTERNALIZE}() regarding portion
562 	 * of this structure above __p_flags that gets copied.  Adding
563 	 * more user-mutable fields after __p_flags would also require
564 	 * adjusting those macros as well.
565 	 */
566 	union {
567 		volatile uint32_t __flags32[2];
568 		volatile uint64_t __flags;              /* PKT_F_* flags */
569 	};
570 } __attribute((aligned(sizeof(uint64_t))));
571 
572 struct __packet {
573 	union {
574 		uint64_t                __pkt_data[4];
575 		struct __packet_com     __pkt_com;
576 	};
577 #define __p_link_flags          __pkt_com.__link_flags
578 #define __p_headroom            __pkt_com.__headroom
579 #define __p_l2_len              __pkt_com.__l2_len
580 #define __p_csum_flags          __pkt_com.__csum_flags
581 #define __p_csum_rx             __pkt_com.__csum_rx
582 #define __p_csum_tx             __pkt_com.__csum_tx
583 #define __p_csum_data           __pkt_com.__csum_data
584 #define __p_comp_gencnt         __pkt_com.__comp_gencnt
585 #define __p_aggr_type           __pkt_com.__aggr_type
586 #define __p_seg_cnt             __pkt_com.__seg_cnt
587 #define __p_proto_seg_sz        __pkt_com.__proto_seg_sz
588 #define __p_trace_id            __pkt_com.__trace_id
589 #define __p_flags32             __pkt_com.__flags32
590 #define __p_flags               __pkt_com.__flags
591 };
592 
593 /* optional packet token types */
594 #define PKT_OPT_TOKEN_TYPE_OPAQUE       1 /* token has opaque data */
595 #define PKT_OPT_TOKEN_TYPE_PACKET_ID    2 /* token has packet_id */
596 
597 /* maximum token size */
598 #define PKT_OPT_MAX_TOKEN_SIZE          16
599 
600 struct __packet_opt_com {
601 	union {
602 		uint64_t        __token_data[2];
603 		uint8_t         __token[PKT_OPT_MAX_TOKEN_SIZE];
604 	};
605 	uint64_t        __expire_ts;
606 	uint64_t        __pkt_tx_time;
607 	uint16_t        __vlan_tag;
608 	uint16_t        __token_len;
609 	uint8_t         __token_type;
610 	uint8_t         __expiry_action;
611 	uint8_t         __app_type;
612 	uint8_t         __app_metadata;
613 } __attribute((aligned(sizeof(uint64_t))));
614 
615 struct __packet_opt {
616 	union {
617 		uint64_t                __pkt_opt_data[5];
618 		struct __packet_opt_com __pkt_opt_com;
619 	};
620 #define __po_token_type         __pkt_opt_com.__token_type
621 #define __po_token_len          __pkt_opt_com.__token_len
622 #define __po_vlan_tag           __pkt_opt_com.__vlan_tag
623 #define __po_token_data         __pkt_opt_com.__token_data
624 #define __po_token              __pkt_opt_com.__token
625 #define __po_expire_ts          __pkt_opt_com.__expire_ts
626 #define __po_expiry_action      __pkt_opt_com.__expiry_action
627 #define __po_app_type           __pkt_opt_com.__app_type
628 #define __po_app_metadata       __pkt_opt_com.__app_metadata
629 #define __po_pkt_tx_time        __pkt_opt_com.__pkt_tx_time
630 };
631 
632 /*
633  * Packet.
634  *
635  * This structure is aligned for efficient copy and accesses.
636  * It is the user version of the __kern_packet structure.
637  *
638  * XXX: Do NOT store kernel pointer/address values here.
639  */
640 struct __user_packet {
641 	struct __user_quantum   pkt_qum;
642 /*
643  * pkt_flow_id is the flow identifier used by user space stack to identfy a
644  * flow. This identifier is passed as a metadata on all packets generated by
645  * the user space stack. On RX flowswitch fills in this metadata on every
646  * packet and can be used by user space stack for flow classification purposes.
647  */
648 #define pkt_flow_id             pkt_qum.qum_flow_id
649 #define pkt_flow_id_64          pkt_qum.qum_flow_id_val64
650 #define pkt_qum_qflags          pkt_qum.qum_qflags
651 #define pkt_length              pkt_qum.qum_len
652 #define pkt_qum_buf             pkt_qum.qum_buf[0]
653 #define pkt_svc_class           pkt_qum.qum_svc_class
654 #ifdef KERNEL
655 /*
656  * pkt_flow_token is a globally unique flow identifier generated by the
657  * flowswitch for each flow. Flowswitch stamps every TX packet with this
658  * identifier. This is the flow identifier which would be visible to the AQM
659  * logic and the driver.
660  * pkt_flow_token uses the first 4 bytes of pkt_flow_id as the storage space.
661  * This is not a problem as pkt_flow_id is only for flowswitch consumption
662  * and is not required by any other module after the flowswitch TX processing
663  * stage.
664  */
665 #define pkt_flow_token          pkt_qum.qum_flow_id_val32[0]
666 #endif /* KERNEL */
667 
668 	/*
669 	 * Common area between user and kernel variants.
670 	 */
671 	struct __packet pkt_com;
672 #define pkt_link_flags          pkt_com.__p_link_flags
673 #define pkt_headroom            pkt_com.__p_headroom
674 #define pkt_l2_len              pkt_com.__p_l2_len
675 #define pkt_csum_flags          pkt_com.__p_csum_flags
676 #define pkt_csum_rx_start_off   pkt_com.__p_csum_rx.__csum_start_off
677 #define pkt_csum_rx_value       pkt_com.__p_csum_rx.__csum_value
678 #define pkt_csum_tx_start_off   pkt_com.__p_csum_tx.__csum_start_off
679 #define pkt_csum_tx_stuff_off   pkt_com.__p_csum_tx.__csum_stuff_off
680 #define pkt_csum_data           pkt_com.__p_csum_data
681 #define pkt_comp_gencnt         pkt_com.__p_comp_gencnt
682 #define pkt_aggr_type           pkt_com.__p_aggr_type
683 #define pkt_seg_cnt             pkt_com.__p_seg_cnt
684 #define pkt_proto_seg_sz        pkt_com.__p_proto_seg_sz
685 #define pkt_trace_id            pkt_com.__p_trace_id
686 #if BYTE_ORDER == LITTLE_ENDIAN
687 #define pkt_pflags32            pkt_com.__p_flags32[0]
688 #else /* BYTE_ORDER != LITTLE_ENDIAN */
689 #define pkt_pflags32            pkt_com.__p_flags32[1]
690 #endif /* BYTE_ORDER != LITTLE_ENDIAN */
691 #define pkt_pflags              pkt_com.__p_flags
692 
693 	/*
694 	 * Optional common metadata.
695 	 */
696 	struct __packet_opt pkt_com_opt;
697 
698 	/*
699 	 * Userland specific.
700 	 */
701 
702 	/*
703 	 * pkt_{bufs,max} aren't part of the common area, on purpose,
704 	 * since we selectively update them on internalize/externalize.
705 	 */
706 	const uint16_t  pkt_bufs_max;       /* maximum size of buflet chain */
707 	const uint16_t  pkt_bufs_cnt;       /* buflet chain size */
708 } __attribute((aligned(sizeof(uint64_t))));
709 
710 /* the size of __user_packet structure for n total buflets */
711 #define _USER_PACKET_SIZE(n) sizeof(struct __user_packet)
712 
713 /*
714  * Valid values for pkt_link_flags.
715  */
716 #define PKT_LINKF_BCAST         0x0001  /* send/received as link-level bcast */
717 #define PKT_LINKF_MCAST         0x0002  /* send/received as link-level mcast */
718 #define PKT_LINKF_ETHFCS        0x0004  /* has Ethernet FCS */
719 
720 /*
721  * XXX IMPORTANT - READ THIS XXX
722  *
723  * Valid values for (64-bit) pkt_pflags.
724  *
725  * The lower 32-bit values are equivalent to PKTF_* flags used by mbufs,
726  * hence the unused values are reserved.  Do not use define any of these
727  * values unless they correspond to PKTF_* flags.  Make sure to do the
728  * following when adding a value in the lower 32-bit range:
729  *
730  * a. If the flag is kernel-only, prefix it with 2 underscore characters,
731  *    then add a PKT_F_* alias under the KERNEL block conditional.  This
732  *    will help ensure that the libsyscall code doesn't mistakenly use it.
733  *
734  * b. In pp_init(), add compile-time assertion to ensure that the PKT_F_*
735  *    value matches the corresponding PKTF_* as defined in <sys/mbuf.h>.
736  *
737  * c. Add the new flag to PKT_F_USER_MASK depending on whether it's allowed
738  *    to be used by userland.  Flags not added to this mask will only be
739  *    used by the kernel.  We only internalize and externalize flags listed
740  *    in PKT_F_USER_MASK.
741  *
742  * d. Add the new flag to PKT_F_COMMON_MASK.
743  *
744  * When adding an upper 32-bit value, ensure (a) and (c) above are done.
745  *
746  * Legend:
747  *
748  * (K)        - Kernel-only
749  * (U+K)      - User and kernel
750  * (reserved) - Only to be used for mapping with mbuf PKTF_* flags
751  */
752 #define __PKT_F_FLOW_ID         0x0000000000000001ULL /* (K) */
753 #define __PKT_F_FLOW_ADV        0x0000000000000002ULL /* (K) */
754 /*                              0x0000000000000004ULL    (reserved) */
755 /*                              0x0000000000000008ULL    (reserved) */
756 /*                              0x0000000000000010ULL    (reserved) */
757 /*                              0x0000000000000020ULL    (reserved) */
758 /*                              0x0000000000000040ULL    (reserved) */
759 /*                              0x0000000000000080ULL    (reserved) */
760 /*                              0x0000000000000100ULL    (reserved) */
761 /*                              0x0000000000000200ULL    (reserved) */
762 #define PKT_F_WAKE_PKT          0x0000000000000400ULL /* (U+K) */
763 /*                              0x0000000000000800ULL    (reserved) */
764 /*                              0x0000000000001000ULL    (reserved) */
765 /*                              0x0000000000002000ULL    (reserved) */
766 /*                              0x0000000000004000ULL    (reserved) */
767 #define PKT_F_BACKGROUND        0x0000000000008000ULL /* (U+K) */
768 /*                              0x0000000000010000ULL    (reserved) */
769 /*                              0x0000000000020000ULL    (reserved) */
770 #define PKT_F_KEEPALIVE         0x0000000000040000ULL /* (U+K) */
771 #define PKT_F_REALTIME          0x0000000000080000ULL /* (U+K) */
772 /*                              0x0000000000100000ULL    (reserved) */
773 #define PKT_F_REXMT             0x0000000000200000ULL /* (U+K) */
774 /*                              0x0000000000400000ULL    (reserved) */
775 #define __PKT_F_TX_COMPL_TS_REQ 0x0000000000800000ULL /* (K) */
776 #define __PKT_F_TS_VALID        0x0000000001000000ULL /* (K) */
777 /*                              0x0000000002000000ULL    (reserved) */
778 #define __PKT_F_NEW_FLOW        0x0000000004000000ULL /* (K) */
779 #define __PKT_F_START_SEQ       0x0000000008000000ULL /* (K) */
780 #define PKT_F_LAST_PKT          0x0000000010000000ULL /* (U+K) */
781 /*                              0x0000000020000000ULL    (reserved) */
782 /*                              0x0000000040000000ULL    (reserved) */
783 /*                              0x0000000080000000ULL    (reserved) */
784 /*                              ---------------------    upper 32-bit below */
785 #define PKT_F_OPT_GROUP_START   0x0000000100000000ULL /* (U+K) */
786 #define PKT_F_OPT_GROUP_END     0x0000000200000000ULL /* (U+K) */
787 #define PKT_F_OPT_EXPIRE_TS     0x0000000400000000ULL /* (U+K) */
788 #define PKT_F_OPT_TOKEN         0x0000000800000000ULL /* (U+K) */
789 #define __PKT_F_FLOW_DATA       0x0000001000000000ULL /* (K) */
790 #define __PKT_F_TX_COMPL_DATA   0x0000002000000000ULL /* (K) */
791 #define __PKT_F_MBUF_DATA       0x0000004000000000ULL /* (K) */
792 #define PKT_F_TRUNCATED         0x0000008000000000ULL /* (U+K) */
793 #define __PKT_F_PKT_DATA        0x0000010000000000ULL /* (K) */
794 #define PKT_F_PROMISC           0x0000020000000000ULL /* (U+K) */
795 #define PKT_F_OPT_VLTAG         0x0000040000000000ULL /* (U+K) */
796 #define PKT_F_OPT_VLTAG_IN_PKT  0x0000080000000000ULL /* (U+K) */
797 #define __PKT_F_TX_PORT_DATA    0x0000100000000000ULL /* (K) */
798 #define PKT_F_OPT_EXP_ACTION    0x0000200000000000ULL /* (U+K) */
799 #define PKT_F_OPT_APP_METADATA  0x0000400000000000ULL /* (U+K) */
800 #define PKT_F_L4S               0x0000800000000000ULL /* (U+K) */
801 #define PKT_F_OPT_TX_TIMESTAMP  0x0001000000000000ULL /* (U+K) */
802 /*                              0x0002000000000000ULL */
803 /*                              0x0004000000000000ULL */
804 /*                              0x0008000000000000ULL */
805 /*                              0x0010000000000000ULL */
806 /*                              0x0020000000000000ULL */
807 /*                              0x0040000000000000ULL */
808 /*                              0x0080000000000000ULL */
809 #define __PKT_F_OPT_ALLOC       0x0100000000000000ULL /* (K) */
810 #define __PKT_F_FLOW_ALLOC      0x0200000000000000ULL /* (K) */
811 #define __PKT_F_TX_COMPL_ALLOC  0x0400000000000000ULL /* (K) */
812 /*                              0x0800000000000000ULL */
813 /*                              0x1000000000000000ULL */
814 /*                              0x2000000000000000ULL */
815 /*                              0x4000000000000000ULL */
816 /*                              0x8000000000000000ULL */
817 
818 /*
819  * Packet option flags.
820  */
821 #define PKT_F_OPT_DATA                                                  \
822 	(PKT_F_OPT_GROUP_START | PKT_F_OPT_GROUP_END |                  \
823 	PKT_F_OPT_EXPIRE_TS | PKT_F_OPT_TOKEN |                         \
824 	PKT_F_OPT_VLTAG | PKT_F_OPT_VLTAG_IN_PKT | PKT_F_OPT_EXP_ACTION | \
825 	PKT_F_OPT_APP_METADATA | PKT_F_OPT_TX_TIMESTAMP)
826 
827 #ifdef KERNEL
828 /*
829  * Flags exposed to user (and kernel).  See notes above.
830  */
831 #define PKT_F_USER_MASK                                                 \
832 	(PKT_F_BACKGROUND | PKT_F_REALTIME | PKT_F_REXMT |              \
833 	PKT_F_LAST_PKT | PKT_F_OPT_DATA | PKT_F_PROMISC |               \
834 	PKT_F_TRUNCATED | PKT_F_WAKE_PKT | PKT_F_L4S)
835 
836 /*
837  * Aliases for kernel-only flags.  See notes above.  The ones marked
838  * with (common) have corresponding PKTF_* definitions and are also
839  * included in PKT_F_COMMON_MASK below.
840  */
841 #define PKT_F_FLOW_ID           __PKT_F_FLOW_ID         /* (common) */
842 #define PKT_F_FLOW_ADV          __PKT_F_FLOW_ADV        /* (common) */
843 #define PKT_F_TX_COMPL_TS_REQ   __PKT_F_TX_COMPL_TS_REQ /* (common) */
844 #define PKT_F_TS_VALID          __PKT_F_TS_VALID        /* (common) */
845 #define PKT_F_NEW_FLOW          __PKT_F_NEW_FLOW        /* (common) */
846 #define PKT_F_START_SEQ         __PKT_F_START_SEQ       /* (common) */
847 #define PKT_F_FLOW_DATA         __PKT_F_FLOW_DATA
848 #define PKT_F_TX_COMPL_DATA     __PKT_F_TX_COMPL_DATA
849 #define PKT_F_MBUF_DATA         __PKT_F_MBUF_DATA
850 #define PKT_F_PKT_DATA          __PKT_F_PKT_DATA
851 #define PKT_F_OPT_ALLOC         __PKT_F_OPT_ALLOC
852 #define PKT_F_FLOW_ALLOC        __PKT_F_FLOW_ALLOC
853 #define PKT_F_TX_COMPL_ALLOC    __PKT_F_TX_COMPL_ALLOC
854 #define PKT_F_TX_PORT_DATA      __PKT_F_TX_PORT_DATA
855 
856 /*
857  * Flags related to mbuf attached to the packet.
858  */
859 #define PKT_F_MBUF_MASK         (PKT_F_MBUF_DATA | PKT_F_TRUNCATED)
860 
861 /*
862  * Flags related to packet attached to the packet.
863  */
864 #define PKT_F_PKT_MASK         (PKT_F_PKT_DATA | PKT_F_TRUNCATED)
865 
866 /*
867  * Invariant flags kept during _PKT_COPY().  At the moment we keep
868  * all except those related to the attached mbuf.
869  */
870 #define PKT_F_COPY_MASK         (~(PKT_F_MBUF_MASK | PKT_F_PKT_MASK))
871 
872 /*
873  * Lower 32-bit flags common to mbuf and __kern_packet.  See notes above.
874  * DO NOT add flags to this mask unless they have equivalent PKTF_* flags
875  * defined in <sys/mbuf.h>
876  */
877 #define PKT_F_COMMON_MASK                                               \
878 	(PKT_F_BACKGROUND | PKT_F_REALTIME | PKT_F_REXMT |              \
879 	PKT_F_LAST_PKT | PKT_F_FLOW_ID | PKT_F_FLOW_ADV |               \
880 	PKT_F_TX_COMPL_TS_REQ | PKT_F_TS_VALID | PKT_F_NEW_FLOW |       \
881 	PKT_F_START_SEQ | PKT_F_KEEPALIVE | PKT_F_WAKE_PKT)
882 
883 /*
884  * Flags retained across alloc/free.
885  */
886 #define PKT_F_INIT_MASK                                                 \
887 	(PKT_F_OPT_ALLOC | PKT_F_FLOW_ALLOC | PKT_F_TX_COMPL_ALLOC)
888 #endif /* KERNEL */
889 
890 /*
891  * 64-bit tagged pointer (limit tag to least significant byte).
892  * We use 2 bits to encode type, and another 2 bits for subtype.
893  */
894 #define SK_PTR_TYPE_MASK        ((uint64_t)0x3)         /* 00 11 */
895 #define SK_PTR_SUBTYPE_MASK     ((uint64_t)0xc)         /* 11 00 */
896 #define SK_PTR_TAG_MASK         ((uint64_t)0xf)         /* 11 11 */
897 
898 #define SK_PTR_TAG(_p)          ((uint64_t)(_p) & SK_PTR_TAG_MASK)
899 #define SK_PTR_ADDR_MASK        (~SK_PTR_TAG_MASK)
900 
901 #define SK_PTR_TYPE(_p)         ((uint64_t)(_p) & SK_PTR_TYPE_MASK)
902 #define SK_PTR_TYPE_ENC(_t)     ((uint64_t)(_t) & SK_PTR_TYPE_MASK)
903 
904 #define SK_PTR_SUBTYPE(_p)      (((uint64_t)(_p) & SK_PTR_SUBTYPE_MASK) >> 2)
905 #define SK_PTR_SUBTYPE_ENC(_s)  (((uint64_t)(_s) << 2) & SK_PTR_SUBTYPE_MASK)
906 
907 #define SK_PTR_ADDR(_p)         ((uint64_t)(_p) & SK_PTR_ADDR_MASK)
908 #define SK_PTR_ADDR_ENC(_p)     ((uint64_t)(_p) & SK_PTR_ADDR_MASK)
909 
910 #define SK_PTR_ENCODE(_p, _t, _s)       \
911 	(SK_PTR_ADDR_ENC(_p) | SK_PTR_TYPE_ENC(_t) | SK_PTR_SUBTYPE_ENC(_s))
912 
913 #define SK_PTR_ADDR_UQUM(_ph)   ((struct __user_quantum *)SK_PTR_ADDR(_ph))
914 #define SK_PTR_ADDR_UPKT(_ph)   ((struct __user_packet *)SK_PTR_ADDR(_ph))
915 
916 #ifdef KERNEL
917 __BEGIN_DECLS
918 /*
919  * Packets.
920  */
921 extern struct mbuf *kern_packet_get_mbuf(const kern_packet_t);
922 __END_DECLS
923 #else /* !KERNEL */
924 #if defined(LIBSYSCALL_INTERFACE)
925 __BEGIN_DECLS
926 extern void pkt_subtype_assert_fail(const packet_t, uint64_t, uint64_t);
927 extern void pkt_type_assert_fail(const packet_t, uint64_t);
928 __END_DECLS
929 #endif /* LIBSYSCALL_INTERFACE */
930 #endif /* !KERNEL */
931 #if defined(LIBSYSCALL_INTERFACE) || defined(BSD_KERNEL_PRIVATE)
932 #include <skywalk/packet_common.h>
933 #endif /* LIBSYSCALL_INTERFACE || BSD_KERNEL_PRIVATE */
934 #endif /* PRIVATE || BSD_KERNEL_PRIVATE */
935 #endif /* !_SKYWALK_OS_PACKET_PRIVATE_H_ */
936