xref: /freebsd/sys/sys/mbuf.h (revision ec4deee4e4f2aef1b97d9424f25d04e91fd7dc10)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)mbuf.h	8.5 (Berkeley) 2/19/95
33  * $FreeBSD$
34  */
35 
36 #ifndef _SYS_MBUF_H_
37 #define	_SYS_MBUF_H_
38 
39 /* XXX: These includes suck. Sorry! */
40 #include <sys/queue.h>
41 #ifdef _KERNEL
42 #include <sys/systm.h>
43 #include <sys/refcount.h>
44 #include <vm/uma.h>
45 #ifdef WITNESS
46 #include <sys/lock.h>
47 #endif
48 #endif
49 
50 #ifdef _KERNEL
51 #include <sys/sdt.h>
52 
53 #define	MBUF_PROBE1(probe, arg0)					\
54 	SDT_PROBE1(sdt, , , probe, arg0)
55 #define	MBUF_PROBE2(probe, arg0, arg1)					\
56 	SDT_PROBE2(sdt, , , probe, arg0, arg1)
57 #define	MBUF_PROBE3(probe, arg0, arg1, arg2)				\
58 	SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
59 #define	MBUF_PROBE4(probe, arg0, arg1, arg2, arg3)			\
60 	SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
61 #define	MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4)		\
62 	SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
63 
64 SDT_PROBE_DECLARE(sdt, , , m__init);
65 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
66 SDT_PROBE_DECLARE(sdt, , , m__get);
67 SDT_PROBE_DECLARE(sdt, , , m__getcl);
68 SDT_PROBE_DECLARE(sdt, , , m__getjcl);
69 SDT_PROBE_DECLARE(sdt, , , m__clget);
70 SDT_PROBE_DECLARE(sdt, , , m__cljget);
71 SDT_PROBE_DECLARE(sdt, , , m__cljset);
72 SDT_PROBE_DECLARE(sdt, , , m__free);
73 SDT_PROBE_DECLARE(sdt, , , m__freem);
74 
75 #endif /* _KERNEL */
76 
77 /*
78  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
79  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
80  * sys/param.h), which has no additional overhead and is used instead of the
81  * internal data area; this is done when at least MINCLSIZE of data must be
82  * stored.  Additionally, it is possible to allocate a separate buffer
83  * externally and attach it to the mbuf in a way similar to that of mbuf
84  * clusters.
85  *
86  * NB: These calculation do not take actual compiler-induced alignment and
87  * padding inside the complete struct mbuf into account.  Appropriate
88  * attention is required when changing members of struct mbuf.
89  *
90  * MLEN is data length in a normal mbuf.
91  * MHLEN is data length in an mbuf with pktheader.
92  * MINCLSIZE is a smallest amount of data that should be put into cluster.
93  *
94  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
95  * they are sensible.
96  */
97 struct mbuf;
98 #define	MHSIZE		offsetof(struct mbuf, m_dat)
99 #define	MPKTHSIZE	offsetof(struct mbuf, m_pktdat)
100 #define	MLEN		((int)(MSIZE - MHSIZE))
101 #define	MHLEN		((int)(MSIZE - MPKTHSIZE))
102 #define	MINCLSIZE	(MHLEN + 1)
103 #define	M_NODOM		255
104 
105 #ifdef _KERNEL
106 /*-
107  * Macro for type conversion: convert mbuf pointer to data pointer of correct
108  * type:
109  *
110  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
111  * mtodo(m, o) -- Same as above but with offset 'o' into data.
112  */
113 #define	mtod(m, t)	((t)((m)->m_data))
114 #define	mtodo(m, o)	((void *)(((m)->m_data) + (o)))
115 
116 /*
117  * Argument structure passed to UMA routines during mbuf and packet
118  * allocations.
119  */
120 struct mb_args {
121 	int	flags;	/* Flags for mbuf being allocated */
122 	short	type;	/* Type of mbuf being allocated */
123 };
124 #endif /* _KERNEL */
125 
126 /*
127  * Packet tag structure (see below for details).
128  */
129 struct m_tag {
130 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
131 	u_int16_t		m_tag_id;	/* Tag ID */
132 	u_int16_t		m_tag_len;	/* Length of data */
133 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
134 	void			(*m_tag_free)(struct m_tag *);
135 };
136 
137 /*
138  * Static network interface owned tag.
139  * Allocated through ifp->if_snd_tag_alloc().
140  */
141 struct m_snd_tag {
142 	struct ifnet *ifp;		/* network interface tag belongs to */
143 	volatile u_int refcount;
144 };
145 
146 /*
147  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
148  * Size ILP32: 48
149  *	 LP64: 56
150  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
151  * they are correct.
152  */
153 struct pkthdr {
154 	union {
155 		struct m_snd_tag *snd_tag;	/* send tag, if any */
156 		struct ifnet	*rcvif;		/* rcv interface */
157 	};
158 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
159 	int32_t		 len;		/* total packet length */
160 
161 	/* Layer crossing persistent information. */
162 	uint32_t	 flowid;	/* packet's 4-tuple system */
163 	uint32_t	 csum_flags;	/* checksum and offload features */
164 	uint16_t	 fibnum;	/* this packet should use this fib */
165 	uint8_t		 numa_domain;	/* NUMA domain of recvd pkt */
166 	uint8_t		 rsstype;	/* hash type */
167 	union {
168 		uint64_t	rcv_tstmp;	/* timestamp in ns */
169 		struct {
170 			uint8_t		 l2hlen;	/* layer 2 hdr len */
171 			uint8_t		 l3hlen;	/* layer 3 hdr len */
172 			uint8_t		 l4hlen;	/* layer 4 hdr len */
173 			uint8_t		 l5hlen;	/* layer 5 hdr len */
174 			uint32_t	 spare;
175 		};
176 	};
177 	union {
178 		uint8_t  eight[8];
179 		uint16_t sixteen[4];
180 		uint32_t thirtytwo[2];
181 		uint64_t sixtyfour[1];
182 		uintptr_t unintptr[1];
183 		void	*ptr;
184 	} PH_per;
185 
186 	/* Layer specific non-persistent local storage for reassembly, etc. */
187 	union {
188 		uint8_t  eight[8];
189 		uint16_t sixteen[4];
190 		uint32_t thirtytwo[2];
191 		uint64_t sixtyfour[1];
192 		uintptr_t unintptr[1];
193 		void 	*ptr;
194 	} PH_loc;
195 };
196 #define	ether_vtag	PH_per.sixteen[0]
197 #define	PH_vt		PH_per
198 #define	vt_nrecs	sixteen[0]	  /* mld and v6-ND */
199 #define	tso_segsz	PH_per.sixteen[1] /* inbound after LRO */
200 #define	lro_nsegs	tso_segsz	  /* inbound after LRO */
201 #define	csum_data	PH_per.thirtytwo[1] /* inbound from hardware up */
202 #define lro_len		PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
203 #define lro_csum	PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
204 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
205 
206 /*
207  * TLS records for TLS 1.0-1.2 can have the following header lengths:
208  * - 5 (AES-CBC with implicit IV)
209  * - 21 (AES-CBC with explicit IV)
210  * - 13 (AES-GCM with 8 byte explicit IV)
211  */
212 #define	MBUF_PEXT_HDR_LEN	23
213 
214 /*
215  * TLS records for TLS 1.0-1.2 can have the following maximum trailer
216  * lengths:
217  * - 16 (AES-GCM)
218  * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
219  * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
220  * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
221  */
222 #define	MBUF_PEXT_TRAIL_LEN	64
223 
224 #if defined(__LP64__)
225 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
226 #else
227 #define MBUF_PEXT_MAX_PGS (72 / sizeof(vm_paddr_t))
228 #endif
229 
230 #define	MBUF_PEXT_MAX_BYTES						\
231     (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
232 
233 struct ktls_session;
234 struct socket;
235 
236 /*
237  * Description of external storage mapped into mbuf; valid only if M_EXT is
238  * set.
239  * Size ILP32: 28
240  *	 LP64: 48
241  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
242  * they are correct.
243  */
244 typedef	void m_ext_free_t(struct mbuf *);
245 struct m_ext {
246 	union {
247 		/*
248 		 * If EXT_FLAG_EMBREF is set, then we use refcount in the
249 		 * mbuf, the 'ext_count' member.  Otherwise, we have a
250 		 * shadow copy and we use pointer 'ext_cnt'.  The original
251 		 * mbuf is responsible to carry the pointer to free routine
252 		 * and its arguments.  They aren't copied into shadows in
253 		 * mb_dupcl() to avoid dereferencing next cachelines.
254 		 */
255 		volatile u_int	 ext_count;
256 		volatile u_int	*ext_cnt;
257 	};
258 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
259 	uint32_t	 ext_type:8,	/* type of external storage */
260 			 ext_flags:24;	/* external storage mbuf flags */
261 	union {
262 		struct {
263 			/*
264 			 * Regular M_EXT mbuf:
265 			 * o ext_buf always points to the external buffer.
266 			 * o ext_free (below) and two optional arguments
267 			 *   ext_arg1 and ext_arg2 store the free context for
268 			 *   the external storage.  They are set only in the
269 			 *   refcount carrying mbuf, the one with
270 			 *   EXT_FLAG_EMBREF flag, with exclusion for
271 			 *   EXT_EXTREF type, where the free context is copied
272 			 *   into all mbufs that use same external storage.
273 			 */
274 			char 	*ext_buf;	/* start of buffer */
275 #define	m_ext_copylen	offsetof(struct m_ext, ext_arg2)
276 			void	*ext_arg2;
277 		};
278 		struct {
279 			/*
280 			 * Multi-page M_EXTPG mbuf:
281 			 * o extpg_pa - page vector.
282 			 * o extpg_trail and extpg_hdr - TLS trailer and
283 			 *   header.
284 			 * Uses ext_free and may also use ext_arg1.
285 			 */
286 			vm_paddr_t	extpg_pa[MBUF_PEXT_MAX_PGS];
287 			char		extpg_trail[MBUF_PEXT_TRAIL_LEN];
288 			char		extpg_hdr[MBUF_PEXT_HDR_LEN];
289 			/* Pretend these 3 fields are part of mbuf itself. */
290 #define	m_epg_pa	m_ext.extpg_pa
291 #define	m_epg_trail	m_ext.extpg_trail
292 #define	m_epg_hdr	m_ext.extpg_hdr
293 #define	m_epg_ext_copylen	offsetof(struct m_ext, ext_free)
294 		};
295 	};
296 	/*
297 	 * Free method and optional argument pointer, both
298 	 * used by M_EXT and M_EXTPG.
299 	 */
300 	m_ext_free_t	*ext_free;
301 	void		*ext_arg1;
302 };
303 
304 /*
305  * The core of the mbuf object along with some shortcut defines for practical
306  * purposes.
307  */
308 struct mbuf {
309 	/*
310 	 * Header present at the beginning of every mbuf.
311 	 * Size ILP32: 24
312 	 *      LP64: 32
313 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
314 	 * that they are correct.
315 	 */
316 	union {	/* next buffer in chain */
317 		struct mbuf		*m_next;
318 		SLIST_ENTRY(mbuf)	m_slist;
319 		STAILQ_ENTRY(mbuf)	m_stailq;
320 	};
321 	union {	/* next chain in queue/record */
322 		struct mbuf		*m_nextpkt;
323 		SLIST_ENTRY(mbuf)	m_slistpkt;
324 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
325 	};
326 	caddr_t		 m_data;	/* location of data */
327 	int32_t		 m_len;		/* amount of data in this mbuf */
328 	uint32_t	 m_type:8,	/* type of data in this mbuf */
329 			 m_flags:24;	/* flags; see below */
330 #if !defined(__LP64__)
331 	uint32_t	 m_pad;		/* pad for 64bit alignment */
332 #endif
333 
334 	/*
335 	 * A set of optional headers (packet header, external storage header)
336 	 * and internal data storage.  Historically, these arrays were sized
337 	 * to MHLEN (space left after a packet header) and MLEN (space left
338 	 * after only a regular mbuf header); they are now variable size in
339 	 * order to support future work on variable-size mbufs.
340 	 */
341 	union {
342 		struct {
343 			union {
344 				/* M_PKTHDR set. */
345 				struct pkthdr	m_pkthdr;
346 
347 				/* M_EXTPG set.
348 				 * Multi-page M_EXTPG mbuf has its meta data
349 				 * split between the below anonymous structure
350 				 * and m_ext.  It carries vector of pages,
351 				 * optional header and trailer char vectors
352 				 * and pointers to socket/TLS data.
353 				 */
354 #define	m_epg_startcopy		m_epg_npgs
355 #define	m_epg_endcopy		m_epg_stailq
356 				struct {
357 					/* Overall count of pages and count of
358 					 * pages with I/O pending. */
359 					uint8_t	m_epg_npgs;
360 					uint8_t	m_epg_nrdy;
361 					/* TLS header and trailer lengths.
362 					 * The data itself resides in m_ext. */
363 					uint8_t	m_epg_hdrlen;
364 					uint8_t	m_epg_trllen;
365 					/* Offset into 1st page and length of
366 					 * data in the last page. */
367 					uint16_t m_epg_1st_off;
368 					uint16_t m_epg_last_len;
369 					uint8_t	m_epg_flags;
370 #define	EPG_FLAG_ANON	0x1	/* Data can be encrypted in place. */
371 #define	EPG_FLAG_2FREE	0x2	/* Scheduled for free. */
372 					uint8_t	m_epg_record_type;
373 					uint8_t	__spare[2];
374 					int	m_epg_enc_cnt;
375 					struct ktls_session *m_epg_tls;
376 					struct socket	*m_epg_so;
377 					uint64_t	m_epg_seqno;
378 					STAILQ_ENTRY(mbuf) m_epg_stailq;
379 				};
380 			};
381 			union {
382 				/* M_EXT or M_EXTPG set. */
383 				struct m_ext	m_ext;
384 				/* M_PKTHDR set, neither M_EXT nor M_EXTPG. */
385 				char		m_pktdat[0];
386 			};
387 		};
388 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
389 	};
390 };
391 
392 #ifdef _KERNEL
393 static inline int
394 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff)
395 {
396 
397 	KASSERT(pgoff == 0 || pidx == 0,
398 	    ("page %d with non-zero offset %d in %p", pidx, pgoff, m));
399 
400 	if (pidx == m->m_epg_npgs - 1) {
401 		return (m->m_epg_last_len);
402 	} else {
403 		return (PAGE_SIZE - pgoff);
404 	}
405 }
406 
407 #ifdef INVARIANTS
408 #define	MCHECK(ex, msg)	KASSERT((ex),				\
409 	    ("Multi page mbuf %p with " #msg " at %s:%d",	\
410 	    m, __FILE__, __LINE__))
411 /*
412  * NB: This expects a non-empty buffer (npgs > 0 and
413  * last_pg_len > 0).
414  */
415 #define	MBUF_EXT_PGS_ASSERT_SANITY(m)	do {				\
416 	MCHECK(m->m_epg_npgs > 0, "no valid pages");		\
417 	MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa),		\
418 	    "too many pages");						\
419 	MCHECK(m->m_epg_nrdy <= m->m_epg_npgs,			\
420 	    "too many ready pages");					\
421 	MCHECK(m->m_epg_1st_off < PAGE_SIZE,			\
422 		"too large page offset");				\
423 	MCHECK(m->m_epg_last_len > 0, "zero last page length");	\
424 	MCHECK(m->m_epg_last_len <= PAGE_SIZE,			\
425 	    "too large last page length");				\
426 	if (m->m_epg_npgs == 1)					\
427 		MCHECK(m->m_epg_1st_off +			\
428 		    m->m_epg_last_len <=	 PAGE_SIZE,		\
429 		    "single page too large");				\
430 	MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr),		\
431 	    "too large header length");					\
432 	MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail),	\
433 	    "too large header length");					\
434 } while (0)
435 #else
436 #define	MBUF_EXT_PGS_ASSERT_SANITY(m)	do {} while (0);
437 #endif
438 #endif
439 
440 /*
441  * mbuf flags of global significance and layer crossing.
442  * Those of only protocol/layer specific significance are to be mapped
443  * to M_PROTO[1-11] and cleared at layer handoff boundaries.
444  * NB: Limited to the lower 24 bits.
445  */
446 #define	M_EXT		0x00000001 /* has associated external storage */
447 #define	M_PKTHDR	0x00000002 /* start of record */
448 #define	M_EOR		0x00000004 /* end of record */
449 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
450 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
451 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
452 #define	M_PROMISC	0x00000040 /* packet was not for us */
453 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
454 #define	M_EXTPG		0x00000100 /* has array of unmapped pages and TLS */
455 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
456 #define	M_TSTMP		0x00000400 /* rcv_tstmp field is valid */
457 #define	M_TSTMP_HPREC	0x00000800 /* rcv_tstmp is high-prec, typically
458 				      hw-stamped on port (useful for IEEE 1588
459 				      and 802.1AS) */
460 #define M_TSTMP_LRO	0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
461 
462 #define	M_PROTO1	0x00002000 /* protocol-specific */
463 #define	M_PROTO2	0x00004000 /* protocol-specific */
464 #define	M_PROTO3	0x00008000 /* protocol-specific */
465 #define	M_PROTO4	0x00010000 /* protocol-specific */
466 #define	M_PROTO5	0x00020000 /* protocol-specific */
467 #define	M_PROTO6	0x00040000 /* protocol-specific */
468 #define	M_PROTO7	0x00080000 /* protocol-specific */
469 #define	M_PROTO8	0x00100000 /* protocol-specific */
470 #define	M_PROTO9	0x00200000 /* protocol-specific */
471 #define	M_PROTO10	0x00400000 /* protocol-specific */
472 #define	M_PROTO11	0x00800000 /* protocol-specific */
473 
474 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
475 
476 /*
477  * Flags to purge when crossing layers.
478  */
479 #define	M_PROTOFLAGS \
480     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
481      M_PROTO9|M_PROTO10|M_PROTO11)
482 
483 /*
484  * Flags preserved when copying m_pkthdr.
485  */
486 #define M_COPYFLAGS \
487     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
488      M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
489 
490 /*
491  * Mbuf flag description for use with printf(9) %b identifier.
492  */
493 #define	M_FLAG_BITS \
494     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
495     "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
496 #define	M_FLAG_PROTOBITS \
497     "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
498     "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
499     "\27M_PROTO10\28M_PROTO11"
500 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
501 
502 /*
503  * Network interface cards are able to hash protocol fields (such as IPv4
504  * addresses and TCP port numbers) classify packets into flows.  These flows
505  * can then be used to maintain ordering while delivering packets to the OS
506  * via parallel input queues, as well as to provide a stateless affinity
507  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
508  * m_flag fields to indicate how the hash should be interpreted by the
509  * network stack.
510  *
511  * Most NICs support RSS, which provides ordering and explicit affinity, and
512  * use the hash m_flag bits to indicate what header fields were covered by
513  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
514  * RSS cards or configurations that provide an opaque flow identifier, allowing
515  * for ordering and distribution without explicit affinity.  Additionally,
516  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
517  * properties.
518  *
519  * The meaning of the IPV6_EX suffix:
520  * "o  Home address from the home address option in the IPv6 destination
521  *     options header.  If the extension header is not present, use the Source
522  *     IPv6 Address.
523  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
524  *     associated extension header.  If the extension header is not present,
525  *     use the Destination IPv6 Address."
526  * Quoted from:
527  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
528  */
529 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
530 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
531 /* Microsoft RSS standard hash types */
532 #define	M_HASHTYPE_NONE			0
533 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
534 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
535 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
536 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
537 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
538 							    * ext hdrs */
539 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
540 							    * ext hdrs */
541 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
542 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
543 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
544 							    * ext hdrs */
545 
546 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
547 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
548 						/* ordering+hash, not affinity*/
549 
550 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
551 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
552 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
553 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
554 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
555 
556 /*
557  * External mbuf storage buffer types.
558  */
559 #define	EXT_CLUSTER	1	/* mbuf cluster */
560 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
561 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
562 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
563 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
564 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
565 #define	EXT_MBUF	7	/* external mbuf reference */
566 #define	EXT_RXRING	8	/* data in NIC receive ring */
567 
568 #define	EXT_VENDOR1	224	/* for vendor-internal use */
569 #define	EXT_VENDOR2	225	/* for vendor-internal use */
570 #define	EXT_VENDOR3	226	/* for vendor-internal use */
571 #define	EXT_VENDOR4	227	/* for vendor-internal use */
572 
573 #define	EXT_EXP1	244	/* for experimental use */
574 #define	EXT_EXP2	245	/* for experimental use */
575 #define	EXT_EXP3	246	/* for experimental use */
576 #define	EXT_EXP4	247	/* for experimental use */
577 
578 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
579 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
580 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
581 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
582 
583 /*
584  * Flags for external mbuf buffer types.
585  * NB: limited to the lower 24 bits.
586  */
587 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
588 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
589 
590 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
591 
592 #define	EXT_FLAG_VENDOR1	0x010000	/* These flags are vendor */
593 #define	EXT_FLAG_VENDOR2	0x020000	/* or submodule specific, */
594 #define	EXT_FLAG_VENDOR3	0x040000	/* not used by mbuf code. */
595 #define	EXT_FLAG_VENDOR4	0x080000	/* Set/read by submodule. */
596 
597 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
598 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
599 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
600 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
601 
602 /*
603  * EXT flag description for use with printf(9) %b identifier.
604  */
605 #define	EXT_FLAG_BITS \
606     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
607     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
608     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
609     "\30EXT_FLAG_EXP4"
610 
611 /*
612  * Flags indicating checksum, segmentation and other offload work to be
613  * done, or already done, by hardware or lower layers.  It is split into
614  * separate inbound and outbound flags.
615  *
616  * Outbound flags that are set by upper protocol layers requesting lower
617  * layers, or ideally the hardware, to perform these offloading tasks.
618  * For outbound packets this field and its flags can be directly tested
619  * against ifnet if_hwassist.
620  */
621 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
622 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
623 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
624 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
625 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
626 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
627 
628 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
629 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
630 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
631 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
632 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
633 
634 /* Inbound checksum support where the checksum was verified by hardware. */
635 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
636 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
637 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
638 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
639 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
640 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
641 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
642 
643 #define	CSUM_SND_TAG		0x80000000	/* Packet header has send tag */
644 
645 /*
646  * CSUM flag description for use with printf(9) %b identifier.
647  */
648 #define	CSUM_BITS \
649     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
650     "\6CSUM_IP_ISCSI" \
651     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
652     "\16CSUM_IP6_ISCSI" \
653     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
654     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
655 
656 /* CSUM flags compatibility mappings. */
657 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
658 #define	CSUM_IP_VALID		CSUM_L3_VALID
659 #define	CSUM_DATA_VALID		CSUM_L4_VALID
660 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
661 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
662 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
663 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
664 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
665 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
666 #define	CSUM_TCP		CSUM_IP_TCP
667 #define	CSUM_UDP		CSUM_IP_UDP
668 #define	CSUM_SCTP		CSUM_IP_SCTP
669 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
670 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
671 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
672 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
673 
674 /*
675  * mbuf types describing the content of the mbuf (including external storage).
676  */
677 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
678 #define	MT_DATA		1	/* dynamic (data) allocation */
679 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
680 
681 #define	MT_VENDOR1	4	/* for vendor-internal use */
682 #define	MT_VENDOR2	5	/* for vendor-internal use */
683 #define	MT_VENDOR3	6	/* for vendor-internal use */
684 #define	MT_VENDOR4	7	/* for vendor-internal use */
685 
686 #define	MT_SONAME	8	/* socket name */
687 
688 #define	MT_EXP1		9	/* for experimental use */
689 #define	MT_EXP2		10	/* for experimental use */
690 #define	MT_EXP3		11	/* for experimental use */
691 #define	MT_EXP4		12	/* for experimental use */
692 
693 #define	MT_CONTROL	14	/* extra-data protocol message */
694 #define	MT_EXTCONTROL	15	/* control message with externalized contents */
695 #define	MT_OOBDATA	16	/* expedited data  */
696 
697 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
698 				   a non-initialized mbuf */
699 
700 /*
701  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
702  * !_KERNEL so that monitoring tools can look up the zones with
703  * libmemstat(3).
704  */
705 #define	MBUF_MEM_NAME		"mbuf"
706 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
707 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
708 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
709 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
710 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
711 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
712 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
713 #define	MBUF_EXTPGS_MEM_NAME	"mbuf_extpgs"
714 
715 #ifdef _KERNEL
716 
717 #ifdef WITNESS
718 #define	MBUF_CHECKSLEEP(how) do {					\
719 	if (how == M_WAITOK)						\
720 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
721 		    "Sleeping in \"%s\"", __func__);			\
722 } while (0)
723 #else
724 #define	MBUF_CHECKSLEEP(how)
725 #endif
726 
727 /*
728  * Network buffer allocation API
729  *
730  * The rest of it is defined in kern/kern_mbuf.c
731  */
732 extern uma_zone_t	zone_mbuf;
733 extern uma_zone_t	zone_clust;
734 extern uma_zone_t	zone_pack;
735 extern uma_zone_t	zone_jumbop;
736 extern uma_zone_t	zone_jumbo9;
737 extern uma_zone_t	zone_jumbo16;
738 extern uma_zone_t	zone_extpgs;
739 
740 void		 mb_dupcl(struct mbuf *, struct mbuf *);
741 void		 mb_free_ext(struct mbuf *);
742 void		 mb_free_extpg(struct mbuf *);
743 void		 mb_free_mext_pgs(struct mbuf *);
744 struct mbuf	*mb_alloc_ext_pgs(int, m_ext_free_t);
745 struct mbuf	*mb_alloc_ext_plus_pages(int, int);
746 struct mbuf	*mb_mapped_to_unmapped(struct mbuf *, int, int, int,
747 		    struct mbuf **);
748 int		 mb_unmapped_compress(struct mbuf *m);
749 struct mbuf 	*mb_unmapped_to_ext(struct mbuf *m);
750 void		 mb_free_notready(struct mbuf *m, int count);
751 void		 m_adj(struct mbuf *, int);
752 int		 m_apply(struct mbuf *, int, int,
753 		    int (*)(void *, void *, u_int), void *);
754 int		 m_append(struct mbuf *, int, c_caddr_t);
755 void		 m_cat(struct mbuf *, struct mbuf *);
756 void		 m_catpkt(struct mbuf *, struct mbuf *);
757 int		 m_clget(struct mbuf *m, int how);
758 void 		*m_cljget(struct mbuf *m, int how, int size);
759 struct mbuf	*m_collapse(struct mbuf *, int, int);
760 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
761 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
762 struct mbuf	*m_copym(struct mbuf *, int, int, int);
763 struct mbuf	*m_copypacket(struct mbuf *, int);
764 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
765 struct mbuf	*m_copyup(struct mbuf *, int, int);
766 struct mbuf	*m_defrag(struct mbuf *, int);
767 void		 m_demote_pkthdr(struct mbuf *);
768 void		 m_demote(struct mbuf *, int, int);
769 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
770 		    void (*)(char *, caddr_t, u_int));
771 void		 m_dispose_extcontrolm(struct mbuf *m);
772 struct mbuf	*m_dup(const struct mbuf *, int);
773 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
774 void		 m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
775 		    void *, void *, int, int);
776 u_int		 m_fixhdr(struct mbuf *);
777 struct mbuf	*m_fragment(struct mbuf *, int, int);
778 void		 m_freem(struct mbuf *);
779 struct mbuf	*m_get2(int, int, short, int);
780 struct mbuf	*m_getjcl(int, short, int, int);
781 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
782 struct mbuf	*m_getptr(struct mbuf *, int, int *);
783 u_int		 m_length(struct mbuf *, struct mbuf **);
784 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
785 int		 m_unmappedtouio(const struct mbuf *, int, struct uio *, int);
786 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
787 int		 m_pkthdr_init(struct mbuf *, int);
788 struct mbuf	*m_prepend(struct mbuf *, int, int);
789 void		 m_print(const struct mbuf *, int);
790 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
791 struct mbuf	*m_pullup(struct mbuf *, int);
792 int		 m_sanity(struct mbuf *, int);
793 struct mbuf	*m_split(struct mbuf *, int, int);
794 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
795 struct mbuf	*m_unshare(struct mbuf *, int);
796 void		 m_snd_tag_init(struct m_snd_tag *, struct ifnet *);
797 void		 m_snd_tag_destroy(struct m_snd_tag *);
798 
799 static __inline int
800 m_gettype(int size)
801 {
802 	int type;
803 
804 	switch (size) {
805 	case MSIZE:
806 		type = EXT_MBUF;
807 		break;
808 	case MCLBYTES:
809 		type = EXT_CLUSTER;
810 		break;
811 #if MJUMPAGESIZE != MCLBYTES
812 	case MJUMPAGESIZE:
813 		type = EXT_JUMBOP;
814 		break;
815 #endif
816 	case MJUM9BYTES:
817 		type = EXT_JUMBO9;
818 		break;
819 	case MJUM16BYTES:
820 		type = EXT_JUMBO16;
821 		break;
822 	default:
823 		panic("%s: invalid cluster size %d", __func__, size);
824 	}
825 
826 	return (type);
827 }
828 
829 /*
830  * Associated an external reference counted buffer with an mbuf.
831  */
832 static __inline void
833 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
834     m_ext_free_t freef, void *arg1, void *arg2)
835 {
836 
837 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
838 
839 	atomic_add_int(ref_cnt, 1);
840 	m->m_flags |= M_EXT;
841 	m->m_ext.ext_buf = buf;
842 	m->m_ext.ext_cnt = ref_cnt;
843 	m->m_data = m->m_ext.ext_buf;
844 	m->m_ext.ext_size = size;
845 	m->m_ext.ext_free = freef;
846 	m->m_ext.ext_arg1 = arg1;
847 	m->m_ext.ext_arg2 = arg2;
848 	m->m_ext.ext_type = EXT_EXTREF;
849 	m->m_ext.ext_flags = 0;
850 }
851 
852 static __inline uma_zone_t
853 m_getzone(int size)
854 {
855 	uma_zone_t zone;
856 
857 	switch (size) {
858 	case MCLBYTES:
859 		zone = zone_clust;
860 		break;
861 #if MJUMPAGESIZE != MCLBYTES
862 	case MJUMPAGESIZE:
863 		zone = zone_jumbop;
864 		break;
865 #endif
866 	case MJUM9BYTES:
867 		zone = zone_jumbo9;
868 		break;
869 	case MJUM16BYTES:
870 		zone = zone_jumbo16;
871 		break;
872 	default:
873 		panic("%s: invalid cluster size %d", __func__, size);
874 	}
875 
876 	return (zone);
877 }
878 
879 /*
880  * Initialize an mbuf with linear storage.
881  *
882  * Inline because the consumer text overhead will be roughly the same to
883  * initialize or call a function with this many parameters and M_PKTHDR
884  * should go away with constant propagation for !MGETHDR.
885  */
886 static __inline int
887 m_init(struct mbuf *m, int how, short type, int flags)
888 {
889 	int error;
890 
891 	m->m_next = NULL;
892 	m->m_nextpkt = NULL;
893 	m->m_data = m->m_dat;
894 	m->m_len = 0;
895 	m->m_flags = flags;
896 	m->m_type = type;
897 	if (flags & M_PKTHDR)
898 		error = m_pkthdr_init(m, how);
899 	else
900 		error = 0;
901 
902 	MBUF_PROBE5(m__init, m, how, type, flags, error);
903 	return (error);
904 }
905 
906 static __inline struct mbuf *
907 m_get(int how, short type)
908 {
909 	struct mbuf *m;
910 	struct mb_args args;
911 
912 	args.flags = 0;
913 	args.type = type;
914 	m = uma_zalloc_arg(zone_mbuf, &args, how);
915 	MBUF_PROBE3(m__get, how, type, m);
916 	return (m);
917 }
918 
919 static __inline struct mbuf *
920 m_gethdr(int how, short type)
921 {
922 	struct mbuf *m;
923 	struct mb_args args;
924 
925 	args.flags = M_PKTHDR;
926 	args.type = type;
927 	m = uma_zalloc_arg(zone_mbuf, &args, how);
928 	MBUF_PROBE3(m__gethdr, how, type, m);
929 	return (m);
930 }
931 
932 static __inline struct mbuf *
933 m_getcl(int how, short type, int flags)
934 {
935 	struct mbuf *m;
936 	struct mb_args args;
937 
938 	args.flags = flags;
939 	args.type = type;
940 	m = uma_zalloc_arg(zone_pack, &args, how);
941 	MBUF_PROBE4(m__getcl, how, type, flags, m);
942 	return (m);
943 }
944 
945 /*
946  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
947  * unreferenced cluster to an mbuf(9).  It is not possible to assert
948  * that, so care can be taken only by users of the API.
949  */
950 static __inline void
951 m_cljset(struct mbuf *m, void *cl, int type)
952 {
953 	int size;
954 
955 	switch (type) {
956 	case EXT_CLUSTER:
957 		size = MCLBYTES;
958 		break;
959 #if MJUMPAGESIZE != MCLBYTES
960 	case EXT_JUMBOP:
961 		size = MJUMPAGESIZE;
962 		break;
963 #endif
964 	case EXT_JUMBO9:
965 		size = MJUM9BYTES;
966 		break;
967 	case EXT_JUMBO16:
968 		size = MJUM16BYTES;
969 		break;
970 	default:
971 		panic("%s: unknown cluster type %d", __func__, type);
972 		break;
973 	}
974 
975 	m->m_data = m->m_ext.ext_buf = cl;
976 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
977 	m->m_ext.ext_size = size;
978 	m->m_ext.ext_type = type;
979 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
980 	m->m_ext.ext_count = 1;
981 	m->m_flags |= M_EXT;
982 	MBUF_PROBE3(m__cljset, m, cl, type);
983 }
984 
985 static __inline void
986 m_chtype(struct mbuf *m, short new_type)
987 {
988 
989 	m->m_type = new_type;
990 }
991 
992 static __inline void
993 m_clrprotoflags(struct mbuf *m)
994 {
995 
996 	while (m) {
997 		m->m_flags &= ~M_PROTOFLAGS;
998 		m = m->m_next;
999 	}
1000 }
1001 
1002 static __inline struct mbuf *
1003 m_last(struct mbuf *m)
1004 {
1005 
1006 	while (m->m_next)
1007 		m = m->m_next;
1008 	return (m);
1009 }
1010 
1011 static inline u_int
1012 m_extrefcnt(struct mbuf *m)
1013 {
1014 
1015 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
1016 
1017 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
1018 	    *m->m_ext.ext_cnt);
1019 }
1020 
1021 /*
1022  * mbuf, cluster, and external object allocation macros (for compatibility
1023  * purposes).
1024  */
1025 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
1026 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
1027 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
1028 #define	MCLGET(m, how)		m_clget((m), (how))
1029 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
1030     m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2),	\
1031     (flags), (type))
1032 #define	m_getm(m, len, how, type)					\
1033     m_getm2((m), (len), (how), (type), M_PKTHDR)
1034 
1035 /*
1036  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
1037  * be both the local data payload, or an external buffer area, depending on
1038  * whether M_EXT is set).
1039  */
1040 #define	M_WRITABLE(m)	(((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 &&	\
1041 			 (!(((m)->m_flags & M_EXT)) ||			\
1042 			 (m_extrefcnt(m) == 1)))
1043 
1044 /* Check if the supplied mbuf has a packet header, or else panic. */
1045 #define	M_ASSERTPKTHDR(m)						\
1046 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
1047 	    ("%s: no mbuf packet header!", __func__))
1048 
1049 /* Check if mbuf is multipage. */
1050 #define M_ASSERTEXTPG(m)						\
1051 	KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG,		\
1052 	    ("%s: m %p is not multipage!", __func__, m))
1053 
1054 /*
1055  * Ensure that the supplied mbuf is a valid, non-free mbuf.
1056  *
1057  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
1058  */
1059 #define	M_ASSERTVALID(m)						\
1060 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
1061 	    ("%s: attempted use of a free mbuf!", __func__))
1062 
1063 /*
1064  * Return the address of the start of the buffer associated with an mbuf,
1065  * handling external storage, packet-header mbufs, and regular data mbufs.
1066  */
1067 #define	M_START(m)							\
1068 	(((m)->m_flags & M_EXTPG) ? NULL :				\
1069 	 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
1070 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
1071 	 &(m)->m_dat[0])
1072 
1073 /*
1074  * Return the size of the buffer associated with an mbuf, handling external
1075  * storage, packet-header mbufs, and regular data mbufs.
1076  */
1077 #define	M_SIZE(m)							\
1078 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
1079 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
1080 	 MLEN)
1081 
1082 /*
1083  * Set the m_data pointer of a newly allocated mbuf to place an object of the
1084  * specified size at the end of the mbuf, longword aligned.
1085  *
1086  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1087  * separate macros, each asserting that it was called at the proper moment.
1088  * This required callers to themselves test the storage type and call the
1089  * right one.  Rather than require callers to be aware of those layout
1090  * decisions, we centralize here.
1091  */
1092 static __inline void
1093 m_align(struct mbuf *m, int len)
1094 {
1095 #ifdef INVARIANTS
1096 	const char *msg = "%s: not a virgin mbuf";
1097 #endif
1098 	int adjust;
1099 
1100 	KASSERT(m->m_data == M_START(m), (msg, __func__));
1101 
1102 	adjust = M_SIZE(m) - len;
1103 	m->m_data += adjust &~ (sizeof(long)-1);
1104 }
1105 
1106 #define	M_ALIGN(m, len)		m_align(m, len)
1107 #define	MH_ALIGN(m, len)	m_align(m, len)
1108 #define	MEXT_ALIGN(m, len)	m_align(m, len)
1109 
1110 /*
1111  * Compute the amount of space available before the current start of data in
1112  * an mbuf.
1113  *
1114  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1115  * of checking writability of the mbuf data area rests solely with the caller.
1116  *
1117  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1118  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1119  * read-only as well.
1120  */
1121 #define	M_LEADINGSPACE(m)						\
1122 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1123 
1124 /*
1125  * Compute the amount of space available after the end of data in an mbuf.
1126  *
1127  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1128  * of checking writability of the mbuf data area rests solely with the caller.
1129  *
1130  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1131  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1132  * read-only as well.
1133  */
1134 #define	M_TRAILINGSPACE(m)						\
1135 	(M_WRITABLE(m) ?						\
1136 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
1137 
1138 /*
1139  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
1140  * allocated, how specifies whether to wait.  If the allocation fails, the
1141  * original mbuf chain is freed and m is set to NULL.
1142  */
1143 #define	M_PREPEND(m, plen, how) do {					\
1144 	struct mbuf **_mmp = &(m);					\
1145 	struct mbuf *_mm = *_mmp;					\
1146 	int _mplen = (plen);						\
1147 	int __mhow = (how);						\
1148 									\
1149 	MBUF_CHECKSLEEP(how);						\
1150 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
1151 		_mm->m_data -= _mplen;					\
1152 		_mm->m_len += _mplen;					\
1153 	} else								\
1154 		_mm = m_prepend(_mm, _mplen, __mhow);			\
1155 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
1156 		_mm->m_pkthdr.len += _mplen;				\
1157 	*_mmp = _mm;							\
1158 } while (0)
1159 
1160 /*
1161  * Change mbuf to new type.  This is a relatively expensive operation and
1162  * should be avoided.
1163  */
1164 #define	MCHTYPE(m, t)	m_chtype((m), (t))
1165 
1166 /* Return the rcvif of a packet header. */
1167 static __inline struct ifnet *
1168 m_rcvif(struct mbuf *m)
1169 {
1170 
1171 	M_ASSERTPKTHDR(m);
1172 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1173 		return (NULL);
1174 	return (m->m_pkthdr.rcvif);
1175 }
1176 
1177 /* Length to m_copy to copy all. */
1178 #define	M_COPYALL	1000000000
1179 
1180 extern int		max_datalen;	/* MHLEN - max_hdr */
1181 extern int		max_hdr;	/* Largest link + protocol header */
1182 extern int		max_linkhdr;	/* Largest link-level header */
1183 extern int		max_protohdr;	/* Largest protocol header */
1184 extern int		nmbclusters;	/* Maximum number of clusters */
1185 extern bool		mb_use_ext_pgs;	/* Use ext_pgs for sendfile */
1186 
1187 /*-
1188  * Network packets may have annotations attached by affixing a list of
1189  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1190  * allocated semi-opaque data structures that have a fixed header
1191  * (struct m_tag) that specifies the size of the memory block and a
1192  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1193  * unsigned value used to identify a module or ABI.  By convention this value
1194  * is chosen as the date+time that the module is created, expressed as the
1195  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1196  * value is an ABI/module-specific value that identifies a particular
1197  * annotation and is private to the module.  For compatibility with systems
1198  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1199  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1200  * compatibility shim functions and several tag types are defined below.
1201  * Users that do not require compatibility should use a private cookie value
1202  * so that packet tag-related definitions can be maintained privately.
1203  *
1204  * Note that the packet tag returned by m_tag_alloc has the default memory
1205  * alignment implemented by malloc.  To reference private data one can use a
1206  * construct like:
1207  *
1208  *	struct m_tag *mtag = m_tag_alloc(...);
1209  *	struct foo *p = (struct foo *)(mtag+1);
1210  *
1211  * if the alignment of struct m_tag is sufficient for referencing members of
1212  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1213  * private data structure to insure proper alignment; e.g.,
1214  *
1215  *	struct foo {
1216  *		struct m_tag	tag;
1217  *		...
1218  *	};
1219  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1220  *	struct m_tag *mtag = &p->tag;
1221  */
1222 
1223 /*
1224  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1225  * tags are expected to ``vanish'' when they pass through a network
1226  * interface.  For most interfaces this happens normally as the tags are
1227  * reclaimed when the mbuf is free'd.  However in some special cases
1228  * reclaiming must be done manually.  An example is packets that pass through
1229  * the loopback interface.  Also, one must be careful to do this when
1230  * ``turning around'' packets (e.g., icmp_reflect).
1231  *
1232  * To mark a tag persistent bit-or this flag in when defining the tag id.
1233  * The tag will then be treated as described above.
1234  */
1235 #define	MTAG_PERSISTENT				0x800
1236 
1237 #define	PACKET_TAG_NONE				0  /* Nadda */
1238 
1239 /* Packet tags for use with PACKET_ABI_COMPAT. */
1240 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1241 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1242 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1243 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1244 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1245 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1246 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1247 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1248 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1249 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1250 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1251 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1252 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1253 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1254 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1255 #define	PACKET_TAG_DIVERT			17 /* divert info */
1256 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1257 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1258 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1259 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1260 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1261 #define	PACKET_TAG_CARP				28 /* CARP info */
1262 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1263 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1264 
1265 /* Specific cookies and tags. */
1266 
1267 /* Packet tag routines. */
1268 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1269 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1270 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1271 void		 m_tag_free_default(struct m_tag *);
1272 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1273 struct m_tag	*m_tag_copy(struct m_tag *, int);
1274 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1275 void		 m_tag_delete_nonpersistent(struct mbuf *);
1276 
1277 /*
1278  * Initialize the list of tags associated with an mbuf.
1279  */
1280 static __inline void
1281 m_tag_init(struct mbuf *m)
1282 {
1283 
1284 	SLIST_INIT(&m->m_pkthdr.tags);
1285 }
1286 
1287 /*
1288  * Set up the contents of a tag.  Note that this does not fill in the free
1289  * method; the caller is expected to do that.
1290  *
1291  * XXX probably should be called m_tag_init, but that was already taken.
1292  */
1293 static __inline void
1294 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1295 {
1296 
1297 	t->m_tag_id = type;
1298 	t->m_tag_len = len;
1299 	t->m_tag_cookie = cookie;
1300 }
1301 
1302 /*
1303  * Reclaim resources associated with a tag.
1304  */
1305 static __inline void
1306 m_tag_free(struct m_tag *t)
1307 {
1308 
1309 	(*t->m_tag_free)(t);
1310 }
1311 
1312 /*
1313  * Return the first tag associated with an mbuf.
1314  */
1315 static __inline struct m_tag *
1316 m_tag_first(struct mbuf *m)
1317 {
1318 
1319 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1320 }
1321 
1322 /*
1323  * Return the next tag in the list of tags associated with an mbuf.
1324  */
1325 static __inline struct m_tag *
1326 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1327 {
1328 
1329 	return (SLIST_NEXT(t, m_tag_link));
1330 }
1331 
1332 /*
1333  * Prepend a tag to the list of tags associated with an mbuf.
1334  */
1335 static __inline void
1336 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1337 {
1338 
1339 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1340 }
1341 
1342 /*
1343  * Unlink a tag from the list of tags associated with an mbuf.
1344  */
1345 static __inline void
1346 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1347 {
1348 
1349 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1350 }
1351 
1352 /* These are for OpenBSD compatibility. */
1353 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1354 
1355 static __inline struct m_tag *
1356 m_tag_get(int type, int length, int wait)
1357 {
1358 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1359 }
1360 
1361 static __inline struct m_tag *
1362 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1363 {
1364 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1365 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1366 }
1367 
1368 static inline struct m_snd_tag *
1369 m_snd_tag_ref(struct m_snd_tag *mst)
1370 {
1371 
1372 	refcount_acquire(&mst->refcount);
1373 	return (mst);
1374 }
1375 
1376 static inline void
1377 m_snd_tag_rele(struct m_snd_tag *mst)
1378 {
1379 
1380 	if (refcount_release(&mst->refcount))
1381 		m_snd_tag_destroy(mst);
1382 }
1383 
1384 static __inline struct mbuf *
1385 m_free(struct mbuf *m)
1386 {
1387 	struct mbuf *n = m->m_next;
1388 
1389 	MBUF_PROBE1(m__free, m);
1390 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1391 		m_tag_delete_chain(m, NULL);
1392 	if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1393 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
1394 	if (m->m_flags & M_EXTPG)
1395 		mb_free_extpg(m);
1396 	else if (m->m_flags & M_EXT)
1397 		mb_free_ext(m);
1398 	else if ((m->m_flags & M_NOFREE) == 0)
1399 		uma_zfree(zone_mbuf, m);
1400 	return (n);
1401 }
1402 
1403 static __inline int
1404 rt_m_getfib(struct mbuf *m)
1405 {
1406 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1407 	return (m->m_pkthdr.fibnum);
1408 }
1409 
1410 #define M_GETFIB(_m)   rt_m_getfib(_m)
1411 
1412 #define M_SETFIB(_m, _fib) do {						\
1413         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1414 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1415 } while (0)
1416 
1417 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1418 #define	MBUF_HASHFLAG_L2	(1 << 2)
1419 #define	MBUF_HASHFLAG_L3	(1 << 3)
1420 #define	MBUF_HASHFLAG_L4	(1 << 4)
1421 
1422 /* mbuf hashing helper routines */
1423 uint32_t	m_ether_tcpip_hash_init(void);
1424 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1425 
1426 #ifdef MBUF_PROFILING
1427  void m_profile(struct mbuf *m);
1428  #define M_PROFILE(m) m_profile(m)
1429 #else
1430  #define M_PROFILE(m)
1431 #endif
1432 
1433 struct mbufq {
1434 	STAILQ_HEAD(, mbuf)	mq_head;
1435 	int			mq_len;
1436 	int			mq_maxlen;
1437 };
1438 
1439 static inline void
1440 mbufq_init(struct mbufq *mq, int maxlen)
1441 {
1442 
1443 	STAILQ_INIT(&mq->mq_head);
1444 	mq->mq_maxlen = maxlen;
1445 	mq->mq_len = 0;
1446 }
1447 
1448 static inline struct mbuf *
1449 mbufq_flush(struct mbufq *mq)
1450 {
1451 	struct mbuf *m;
1452 
1453 	m = STAILQ_FIRST(&mq->mq_head);
1454 	STAILQ_INIT(&mq->mq_head);
1455 	mq->mq_len = 0;
1456 	return (m);
1457 }
1458 
1459 static inline void
1460 mbufq_drain(struct mbufq *mq)
1461 {
1462 	struct mbuf *m, *n;
1463 
1464 	n = mbufq_flush(mq);
1465 	while ((m = n) != NULL) {
1466 		n = STAILQ_NEXT(m, m_stailqpkt);
1467 		m_freem(m);
1468 	}
1469 }
1470 
1471 static inline struct mbuf *
1472 mbufq_first(const struct mbufq *mq)
1473 {
1474 
1475 	return (STAILQ_FIRST(&mq->mq_head));
1476 }
1477 
1478 static inline struct mbuf *
1479 mbufq_last(const struct mbufq *mq)
1480 {
1481 
1482 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1483 }
1484 
1485 static inline int
1486 mbufq_full(const struct mbufq *mq)
1487 {
1488 
1489 	return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1490 }
1491 
1492 static inline int
1493 mbufq_len(const struct mbufq *mq)
1494 {
1495 
1496 	return (mq->mq_len);
1497 }
1498 
1499 static inline int
1500 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1501 {
1502 
1503 	if (mbufq_full(mq))
1504 		return (ENOBUFS);
1505 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1506 	mq->mq_len++;
1507 	return (0);
1508 }
1509 
1510 static inline struct mbuf *
1511 mbufq_dequeue(struct mbufq *mq)
1512 {
1513 	struct mbuf *m;
1514 
1515 	m = STAILQ_FIRST(&mq->mq_head);
1516 	if (m) {
1517 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1518 		m->m_nextpkt = NULL;
1519 		mq->mq_len--;
1520 	}
1521 	return (m);
1522 }
1523 
1524 static inline void
1525 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1526 {
1527 
1528 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1529 	mq->mq_len++;
1530 }
1531 
1532 /*
1533  * Note: this doesn't enforce the maximum list size for dst.
1534  */
1535 static inline void
1536 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1537 {
1538 
1539 	mq_dst->mq_len += mq_src->mq_len;
1540 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1541 	mq_src->mq_len = 0;
1542 }
1543 
1544 #ifdef _SYS_TIMESPEC_H_
1545 static inline void
1546 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1547 {
1548 
1549 	KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1550 	KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0, ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1551 	ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1552 	ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1553 }
1554 #endif
1555 
1556 #ifdef DEBUGNET
1557 /* Invoked from the debugnet client code. */
1558 void	debugnet_mbuf_drain(void);
1559 void	debugnet_mbuf_start(void);
1560 void	debugnet_mbuf_finish(void);
1561 void	debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1562 #endif
1563 
1564 static inline bool
1565 mbuf_has_tls_session(struct mbuf *m)
1566 {
1567 
1568 	if (m->m_flags & M_EXTPG) {
1569 		if (m->m_epg_tls != NULL) {
1570 			return (true);
1571 		}
1572 	}
1573 	return (false);
1574 }
1575 
1576 #endif /* _KERNEL */
1577 #endif /* !_SYS_MBUF_H_ */
1578