xref: /freebsd/sys/sys/mbuf.h (revision 8c5a9161d16094d9db474fe78ddead1325246d05)
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__clget);
69 SDT_PROBE_DECLARE(sdt, , , m__cljget);
70 SDT_PROBE_DECLARE(sdt, , , m__cljset);
71 SDT_PROBE_DECLARE(sdt, , , m__free);
72 SDT_PROBE_DECLARE(sdt, , , m__freem);
73 
74 #endif /* _KERNEL */
75 
76 /*
77  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
78  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
79  * sys/param.h), which has no additional overhead and is used instead of the
80  * internal data area; this is done when at least MINCLSIZE of data must be
81  * stored.  Additionally, it is possible to allocate a separate buffer
82  * externally and attach it to the mbuf in a way similar to that of mbuf
83  * clusters.
84  *
85  * NB: These calculation do not take actual compiler-induced alignment and
86  * padding inside the complete struct mbuf into account.  Appropriate
87  * attention is required when changing members of struct mbuf.
88  *
89  * MLEN is data length in a normal mbuf.
90  * MHLEN is data length in an mbuf with pktheader.
91  * MINCLSIZE is a smallest amount of data that should be put into cluster.
92  *
93  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
94  * they are sensible.
95  */
96 struct mbuf;
97 #define	MHSIZE		offsetof(struct mbuf, m_dat)
98 #define	MPKTHSIZE	offsetof(struct mbuf, m_pktdat)
99 #define	MLEN		((int)(MSIZE - MHSIZE))
100 #define	MHLEN		((int)(MSIZE - MPKTHSIZE))
101 #define	MINCLSIZE	(MHLEN + 1)
102 #define	M_NODOM		255
103 
104 #ifdef _KERNEL
105 /*-
106  * Macro for type conversion: convert mbuf pointer to data pointer of correct
107  * type:
108  *
109  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
110  * mtodo(m, o) -- Same as above but with offset 'o' into data.
111  */
112 #define	mtod(m, t)	((t)((m)->m_data))
113 #define	mtodo(m, o)	((void *)(((m)->m_data) + (o)))
114 
115 /*
116  * Argument structure passed to UMA routines during mbuf and packet
117  * allocations.
118  */
119 struct mb_args {
120 	int	flags;	/* Flags for mbuf being allocated */
121 	short	type;	/* Type of mbuf being allocated */
122 };
123 #endif /* _KERNEL */
124 
125 /*
126  * Packet tag structure (see below for details).
127  */
128 struct m_tag {
129 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
130 	u_int16_t		m_tag_id;	/* Tag ID */
131 	u_int16_t		m_tag_len;	/* Length of data */
132 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
133 	void			(*m_tag_free)(struct m_tag *);
134 };
135 
136 /*
137  * Static network interface owned tag.
138  * Allocated through ifp->if_snd_tag_alloc().
139  */
140 struct m_snd_tag {
141 	struct ifnet *ifp;		/* network interface tag belongs to */
142 	volatile u_int refcount;
143 };
144 
145 /*
146  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
147  * Size ILP32: 48
148  *	 LP64: 56
149  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
150  * they are correct.
151  */
152 struct pkthdr {
153 	union {
154 		struct m_snd_tag *snd_tag;	/* send tag, if any */
155 		struct ifnet	*rcvif;		/* rcv interface */
156 	};
157 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
158 	int32_t		 len;		/* total packet length */
159 
160 	/* Layer crossing persistent information. */
161 	uint32_t	 flowid;	/* packet's 4-tuple system */
162 	uint32_t	 csum_flags;	/* checksum and offload features */
163 	uint16_t	 fibnum;	/* this packet should use this fib */
164 	uint8_t		 numa_domain;	/* NUMA domain of recvd pkt */
165 	uint8_t		 rsstype;	/* hash type */
166 	union {
167 		uint64_t	rcv_tstmp;	/* timestamp in ns */
168 		struct {
169 			uint8_t		 l2hlen;	/* layer 2 hdr len */
170 			uint8_t		 l3hlen;	/* layer 3 hdr len */
171 			uint8_t		 l4hlen;	/* layer 4 hdr len */
172 			uint8_t		 l5hlen;	/* layer 5 hdr len */
173 			uint32_t	 spare;
174 		};
175 	};
176 	union {
177 		uint8_t  eight[8];
178 		uint16_t sixteen[4];
179 		uint32_t thirtytwo[2];
180 		uint64_t sixtyfour[1];
181 		uintptr_t unintptr[1];
182 		void	*ptr;
183 	} PH_per;
184 
185 	/* Layer specific non-persistent local storage for reassembly, etc. */
186 	union {
187 		uint8_t  eight[8];
188 		uint16_t sixteen[4];
189 		uint32_t thirtytwo[2];
190 		uint64_t sixtyfour[1];
191 		uintptr_t unintptr[1];
192 		void 	*ptr;
193 	} PH_loc;
194 };
195 #define	ether_vtag	PH_per.sixteen[0]
196 #define	PH_vt		PH_per
197 #define	vt_nrecs	sixteen[0]
198 #define	tso_segsz	PH_per.sixteen[1]
199 #define	lro_nsegs	tso_segsz
200 #define	csum_phsum	PH_per.sixteen[2]
201 #define	csum_data	PH_per.thirtytwo[1]
202 #define pace_thoff	PH_loc.sixteen[0]
203 #define pace_tlen	PH_loc.sixteen[1]
204 #define pace_drphdrlen	PH_loc.sixteen[2]
205 #define pace_tos	PH_loc.eight[6]
206 #define pace_lock	PH_loc.eight[7]
207 
208 /*
209  * Description of external storage mapped into mbuf; valid only if M_EXT is
210  * set.
211  * Size ILP32: 28
212  *	 LP64: 48
213  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
214  * they are correct.
215  */
216 typedef	void m_ext_free_t(struct mbuf *);
217 struct m_ext {
218 	union {
219 		/*
220 		 * If EXT_FLAG_EMBREF is set, then we use refcount in the
221 		 * mbuf, the 'ext_count' member.  Otherwise, we have a
222 		 * shadow copy and we use pointer 'ext_cnt'.  The original
223 		 * mbuf is responsible to carry the pointer to free routine
224 		 * and its arguments.  They aren't copied into shadows in
225 		 * mb_dupcl() to avoid dereferencing next cachelines.
226 		 */
227 		volatile u_int	 ext_count;
228 		volatile u_int	*ext_cnt;
229 	};
230 	union {
231 		/*
232 		 * If ext_type == EXT_PGS, 'ext_pgs' points to a
233 		 * structure describing the buffer.  Otherwise,
234 		 * 'ext_buf' points to the start of the buffer.
235 		 */
236 		struct mbuf_ext_pgs *ext_pgs;
237 		char		*ext_buf;
238 	};
239 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
240 	uint32_t	 ext_type:8,	/* type of external storage */
241 			 ext_flags:24;	/* external storage mbuf flags */
242 	/*
243 	 * Fields below store the free context for the external storage.
244 	 * They are valid only in the refcount carrying mbuf, the one with
245 	 * EXT_FLAG_EMBREF flag, with exclusion for EXT_EXTREF type, where
246 	 * the free context is copied into all mbufs that use same external
247 	 * storage.
248 	 */
249 #define	m_ext_copylen	offsetof(struct m_ext, ext_free)
250 	m_ext_free_t	*ext_free;	/* free routine if not the usual */
251 	void		*ext_arg1;	/* optional argument pointer */
252 	void		*ext_arg2;	/* optional argument pointer */
253 };
254 
255 /*
256  * The core of the mbuf object along with some shortcut defines for practical
257  * purposes.
258  */
259 struct mbuf {
260 	/*
261 	 * Header present at the beginning of every mbuf.
262 	 * Size ILP32: 24
263 	 *      LP64: 32
264 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
265 	 * that they are correct.
266 	 */
267 	union {	/* next buffer in chain */
268 		struct mbuf		*m_next;
269 		SLIST_ENTRY(mbuf)	m_slist;
270 		STAILQ_ENTRY(mbuf)	m_stailq;
271 	};
272 	union {	/* next chain in queue/record */
273 		struct mbuf		*m_nextpkt;
274 		SLIST_ENTRY(mbuf)	m_slistpkt;
275 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
276 	};
277 	caddr_t		 m_data;	/* location of data */
278 	int32_t		 m_len;		/* amount of data in this mbuf */
279 	uint32_t	 m_type:8,	/* type of data in this mbuf */
280 			 m_flags:24;	/* flags; see below */
281 #if !defined(__LP64__)
282 	uint32_t	 m_pad;		/* pad for 64bit alignment */
283 #endif
284 
285 	/*
286 	 * A set of optional headers (packet header, external storage header)
287 	 * and internal data storage.  Historically, these arrays were sized
288 	 * to MHLEN (space left after a packet header) and MLEN (space left
289 	 * after only a regular mbuf header); they are now variable size in
290 	 * order to support future work on variable-size mbufs.
291 	 */
292 	union {
293 		struct {
294 			struct pkthdr	m_pkthdr;	/* M_PKTHDR set */
295 			union {
296 				struct m_ext	m_ext;	/* M_EXT set */
297 				char		m_pktdat[0];
298 			};
299 		};
300 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
301 	};
302 };
303 
304 struct socket;
305 
306 /*
307  * TLS records for TLS 1.0-1.2 can have the following header lengths:
308  * - 5 (AES-CBC with implicit IV)
309  * - 21 (AES-CBC with explicit IV)
310  * - 13 (AES-GCM with 8 byte explicit IV)
311  */
312 #define	MBUF_PEXT_HDR_LEN	24
313 
314 /*
315  * TLS records for TLS 1.0-1.2 can have the following maximum trailer
316  * lengths:
317  * - 16 (AES-GCM)
318  * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
319  * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
320  * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
321  */
322 #define	MBUF_PEXT_TRAIL_LEN	64
323 
324 #ifdef __LP64__
325 #define	MBUF_PEXT_MAX_PGS	(152 / sizeof(vm_paddr_t))
326 #else
327 #define	MBUF_PEXT_MAX_PGS	(156 / sizeof(vm_paddr_t))
328 #endif
329 
330 #define	MBUF_PEXT_MAX_BYTES						\
331     (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
332 
333 /*
334  * This struct is 256 bytes in size and is arranged so that the most
335  * common case (accessing the first 4 pages of a 16KB TLS record) will
336  * fit in a single 64 byte cacheline.
337  */
338 struct mbuf_ext_pgs {
339 	uint8_t		npgs;			/* Number of attached pages */
340 	uint8_t		nrdy;			/* Pages with I/O pending */
341 	uint8_t		hdr_len;		/* TLS header length */
342 	uint8_t		trail_len;		/* TLS trailer length */
343 	uint16_t	first_pg_off;		/* Offset into 1st page */
344 	uint16_t	last_pg_len;		/* Length of last page */
345 	vm_paddr_t	pa[MBUF_PEXT_MAX_PGS];	/* phys addrs of pages */
346 	char		hdr[MBUF_PEXT_HDR_LEN];	/* TLS header */
347 	void		*tls;			/* TLS session */
348 #if defined(__i386__) || \
349     (defined(__powerpc__) && !defined(__powerpc64__) && defined(BOOKE))
350 	/*
351 	 * i386 and Book-E PowerPC have 64-bit vm_paddr_t, so there is
352 	 * a 4 byte remainder from the space allocated for pa[].
353 	 */
354 	uint32_t	pad;
355 #endif
356 	union {
357 		char	trail[MBUF_PEXT_TRAIL_LEN]; /* TLS trailer */
358 		struct {
359 			struct socket *so;
360 			void	*mbuf;
361 			uint64_t seqno;
362 			STAILQ_ENTRY(mbuf_ext_pgs) stailq;
363 		};
364 	};
365 };
366 
367 #ifdef _KERNEL
368 static inline int
369 mbuf_ext_pg_len(struct mbuf_ext_pgs *ext_pgs, int pidx, int pgoff)
370 {
371 	KASSERT(pgoff == 0 || pidx == 0,
372 	    ("page %d with non-zero offset %d in %p", pidx, pgoff, ext_pgs));
373 	if (pidx == ext_pgs->npgs - 1) {
374 		return (ext_pgs->last_pg_len);
375 	} else {
376 		return (PAGE_SIZE - pgoff);
377 	}
378 }
379 
380 #ifdef INVARIANT_SUPPORT
381 void	mb_ext_pgs_check(struct mbuf_ext_pgs *ext_pgs);
382 #endif
383 #ifdef INVARIANTS
384 #define	MBUF_EXT_PGS_ASSERT_SANITY(ext_pgs)	mb_ext_pgs_check((ext_pgs))
385 #else
386 #define	MBUF_EXT_PGS_ASSERT_SANITY(ext_pgs)
387 #endif
388 #endif
389 
390 /*
391  * mbuf flags of global significance and layer crossing.
392  * Those of only protocol/layer specific significance are to be mapped
393  * to M_PROTO[1-12] and cleared at layer handoff boundaries.
394  * NB: Limited to the lower 24 bits.
395  */
396 #define	M_EXT		0x00000001 /* has associated external storage */
397 #define	M_PKTHDR	0x00000002 /* start of record */
398 #define	M_EOR		0x00000004 /* end of record */
399 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
400 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
401 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
402 #define	M_PROMISC	0x00000040 /* packet was not for us */
403 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
404 #define	M_NOMAP		0x00000100 /* mbuf data is unmapped */
405 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
406 #define	M_TSTMP		0x00000400 /* rcv_tstmp field is valid */
407 #define	M_TSTMP_HPREC	0x00000800 /* rcv_tstmp is high-prec, typically
408 				      hw-stamped on port (useful for IEEE 1588
409 				      and 802.1AS) */
410 
411 #define	M_PROTO1	0x00001000 /* protocol-specific */
412 #define	M_PROTO2	0x00002000 /* protocol-specific */
413 #define	M_PROTO3	0x00004000 /* protocol-specific */
414 #define	M_PROTO4	0x00008000 /* protocol-specific */
415 #define	M_PROTO5	0x00010000 /* protocol-specific */
416 #define	M_PROTO6	0x00020000 /* protocol-specific */
417 #define	M_PROTO7	0x00040000 /* protocol-specific */
418 #define	M_PROTO8	0x00080000 /* protocol-specific */
419 #define	M_PROTO9	0x00100000 /* protocol-specific */
420 #define	M_PROTO10	0x00200000 /* protocol-specific */
421 #define	M_PROTO11	0x00400000 /* protocol-specific */
422 #define	M_PROTO12	0x00800000 /* protocol-specific */
423 
424 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
425 
426 /*
427  * Flags to purge when crossing layers.
428  */
429 #define	M_PROTOFLAGS \
430     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
431      M_PROTO9|M_PROTO10|M_PROTO11|M_PROTO12)
432 
433 /*
434  * Flags preserved when copying m_pkthdr.
435  */
436 #define M_COPYFLAGS \
437     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
438      M_TSTMP_HPREC|M_PROTOFLAGS)
439 
440 /*
441  * Mbuf flag description for use with printf(9) %b identifier.
442  */
443 #define	M_FLAG_BITS \
444     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
445     "\7M_PROMISC\10M_VLANTAG\11M_NOMAP\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC"
446 #define	M_FLAG_PROTOBITS \
447     "\15M_PROTO1\16M_PROTO2\17M_PROTO3\20M_PROTO4\21M_PROTO5" \
448     "\22M_PROTO6\23M_PROTO7\24M_PROTO8\25M_PROTO9\26M_PROTO10" \
449     "\27M_PROTO11\30M_PROTO12"
450 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
451 
452 /*
453  * Network interface cards are able to hash protocol fields (such as IPv4
454  * addresses and TCP port numbers) classify packets into flows.  These flows
455  * can then be used to maintain ordering while delivering packets to the OS
456  * via parallel input queues, as well as to provide a stateless affinity
457  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
458  * m_flag fields to indicate how the hash should be interpreted by the
459  * network stack.
460  *
461  * Most NICs support RSS, which provides ordering and explicit affinity, and
462  * use the hash m_flag bits to indicate what header fields were covered by
463  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
464  * RSS cards or configurations that provide an opaque flow identifier, allowing
465  * for ordering and distribution without explicit affinity.  Additionally,
466  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
467  * properties.
468  *
469  * The meaning of the IPV6_EX suffix:
470  * "o  Home address from the home address option in the IPv6 destination
471  *     options header.  If the extension header is not present, use the Source
472  *     IPv6 Address.
473  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
474  *     associated extension header.  If the extension header is not present,
475  *     use the Destination IPv6 Address."
476  * Quoted from:
477  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
478  */
479 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
480 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
481 /* Microsoft RSS standard hash types */
482 #define	M_HASHTYPE_NONE			0
483 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
484 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
485 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
486 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
487 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
488 							    * ext hdrs */
489 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
490 							    * ext hdrs */
491 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
492 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
493 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
494 							    * ext hdrs */
495 
496 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
497 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
498 						/* ordering+hash, not affinity*/
499 
500 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
501 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
502 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
503 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
504 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
505 
506 /*
507  * External mbuf storage buffer types.
508  */
509 #define	EXT_CLUSTER	1	/* mbuf cluster */
510 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
511 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
512 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
513 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
514 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
515 #define	EXT_MBUF	7	/* external mbuf reference */
516 #define	EXT_RXRING	8	/* data in NIC receive ring */
517 #define	EXT_PGS		9	/* array of unmapped pages */
518 
519 #define	EXT_VENDOR1	224	/* for vendor-internal use */
520 #define	EXT_VENDOR2	225	/* for vendor-internal use */
521 #define	EXT_VENDOR3	226	/* for vendor-internal use */
522 #define	EXT_VENDOR4	227	/* for vendor-internal use */
523 
524 #define	EXT_EXP1	244	/* for experimental use */
525 #define	EXT_EXP2	245	/* for experimental use */
526 #define	EXT_EXP3	246	/* for experimental use */
527 #define	EXT_EXP4	247	/* for experimental use */
528 
529 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
530 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
531 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
532 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
533 
534 /*
535  * Flags for external mbuf buffer types.
536  * NB: limited to the lower 24 bits.
537  */
538 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
539 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
540 
541 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
542 
543 #define	EXT_FLAG_VENDOR1	0x010000	/* These flags are vendor */
544 #define	EXT_FLAG_VENDOR2	0x020000	/* or submodule specific, */
545 #define	EXT_FLAG_VENDOR3	0x040000	/* not used by mbuf code. */
546 #define	EXT_FLAG_VENDOR4	0x080000	/* Set/read by submodule. */
547 
548 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
549 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
550 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
551 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
552 
553 /*
554  * EXT flag description for use with printf(9) %b identifier.
555  */
556 #define	EXT_FLAG_BITS \
557     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
558     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
559     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
560     "\30EXT_FLAG_EXP4"
561 
562 #define MBUF_EXT_PGS_ASSERT(m)						\
563 	KASSERT((((m)->m_flags & M_EXT) != 0) &&			\
564 	    ((m)->m_ext.ext_type == EXT_PGS),				\
565 	    ("%s: m %p !M_EXT or !EXT_PGS", __func__, m))
566 
567 /*
568  * Flags indicating checksum, segmentation and other offload work to be
569  * done, or already done, by hardware or lower layers.  It is split into
570  * separate inbound and outbound flags.
571  *
572  * Outbound flags that are set by upper protocol layers requesting lower
573  * layers, or ideally the hardware, to perform these offloading tasks.
574  * For outbound packets this field and its flags can be directly tested
575  * against ifnet if_hwassist.
576  */
577 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
578 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
579 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
580 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
581 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
582 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
583 
584 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
585 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
586 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
587 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
588 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
589 
590 /* Inbound checksum support where the checksum was verified by hardware. */
591 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
592 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
593 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
594 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
595 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
596 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
597 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
598 
599 #define	CSUM_SND_TAG		0x80000000	/* Packet header has send tag */
600 
601 /*
602  * CSUM flag description for use with printf(9) %b identifier.
603  */
604 #define	CSUM_BITS \
605     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
606     "\6CSUM_IP_ISCSI" \
607     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
608     "\16CSUM_IP6_ISCSI" \
609     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
610     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
611 
612 /* CSUM flags compatibility mappings. */
613 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
614 #define	CSUM_IP_VALID		CSUM_L3_VALID
615 #define	CSUM_DATA_VALID		CSUM_L4_VALID
616 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
617 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
618 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
619 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
620 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
621 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
622 #define	CSUM_TCP		CSUM_IP_TCP
623 #define	CSUM_UDP		CSUM_IP_UDP
624 #define	CSUM_SCTP		CSUM_IP_SCTP
625 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
626 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
627 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
628 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
629 
630 /*
631  * mbuf types describing the content of the mbuf (including external storage).
632  */
633 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
634 #define	MT_DATA		1	/* dynamic (data) allocation */
635 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
636 
637 #define	MT_VENDOR1	4	/* for vendor-internal use */
638 #define	MT_VENDOR2	5	/* for vendor-internal use */
639 #define	MT_VENDOR3	6	/* for vendor-internal use */
640 #define	MT_VENDOR4	7	/* for vendor-internal use */
641 
642 #define	MT_SONAME	8	/* socket name */
643 
644 #define	MT_EXP1		9	/* for experimental use */
645 #define	MT_EXP2		10	/* for experimental use */
646 #define	MT_EXP3		11	/* for experimental use */
647 #define	MT_EXP4		12	/* for experimental use */
648 
649 #define	MT_CONTROL	14	/* extra-data protocol message */
650 #define	MT_EXTCONTROL	15	/* control message with externalized contents */
651 #define	MT_OOBDATA	16	/* expedited data  */
652 
653 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
654 				   a non-initialized mbuf */
655 
656 /*
657  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
658  * !_KERNEL so that monitoring tools can look up the zones with
659  * libmemstat(3).
660  */
661 #define	MBUF_MEM_NAME		"mbuf"
662 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
663 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
664 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
665 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
666 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
667 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
668 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
669 #define	MBUF_EXTPGS_MEM_NAME	"mbuf_extpgs"
670 
671 #ifdef _KERNEL
672 
673 #ifdef WITNESS
674 #define	MBUF_CHECKSLEEP(how) do {					\
675 	if (how == M_WAITOK)						\
676 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
677 		    "Sleeping in \"%s\"", __func__);			\
678 } while (0)
679 #else
680 #define	MBUF_CHECKSLEEP(how)
681 #endif
682 
683 /*
684  * Network buffer allocation API
685  *
686  * The rest of it is defined in kern/kern_mbuf.c
687  */
688 extern uma_zone_t	zone_mbuf;
689 extern uma_zone_t	zone_clust;
690 extern uma_zone_t	zone_pack;
691 extern uma_zone_t	zone_jumbop;
692 extern uma_zone_t	zone_jumbo9;
693 extern uma_zone_t	zone_jumbo16;
694 extern uma_zone_t	zone_extpgs;
695 
696 void		 mb_dupcl(struct mbuf *, struct mbuf *);
697 void		 mb_free_ext(struct mbuf *);
698 void		 mb_free_mext_pgs(struct mbuf *);
699 struct mbuf	*mb_alloc_ext_pgs(int, bool, m_ext_free_t);
700 int		 mb_unmapped_compress(struct mbuf *m);
701 struct mbuf 	*mb_unmapped_to_ext(struct mbuf *m);
702 void		 mb_free_notready(struct mbuf *m, int count);
703 void		 m_adj(struct mbuf *, int);
704 int		 m_apply(struct mbuf *, int, int,
705 		    int (*)(void *, void *, u_int), void *);
706 int		 m_append(struct mbuf *, int, c_caddr_t);
707 void		 m_cat(struct mbuf *, struct mbuf *);
708 void		 m_catpkt(struct mbuf *, struct mbuf *);
709 int		 m_clget(struct mbuf *m, int how);
710 void 		*m_cljget(struct mbuf *m, int how, int size);
711 struct mbuf	*m_collapse(struct mbuf *, int, int);
712 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
713 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
714 struct mbuf	*m_copym(struct mbuf *, int, int, int);
715 struct mbuf	*m_copypacket(struct mbuf *, int);
716 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
717 struct mbuf	*m_copyup(struct mbuf *, int, int);
718 struct mbuf	*m_defrag(struct mbuf *, int);
719 void		 m_demote_pkthdr(struct mbuf *);
720 void		 m_demote(struct mbuf *, int, int);
721 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
722 		    void (*)(char *, caddr_t, u_int));
723 void		 m_dispose_extcontrolm(struct mbuf *m);
724 struct mbuf	*m_dup(const struct mbuf *, int);
725 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
726 void		 m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
727 		    void *, void *, int, int);
728 u_int		 m_fixhdr(struct mbuf *);
729 struct mbuf	*m_fragment(struct mbuf *, int, int);
730 void		 m_freem(struct mbuf *);
731 struct mbuf	*m_get2(int, int, short, int);
732 struct mbuf	*m_getjcl(int, short, int, int);
733 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
734 struct mbuf	*m_getptr(struct mbuf *, int, int *);
735 u_int		 m_length(struct mbuf *, struct mbuf **);
736 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
737 int		 m_unmappedtouio(const struct mbuf *, int, struct uio *, int);
738 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
739 int		 m_pkthdr_init(struct mbuf *, int);
740 struct mbuf	*m_prepend(struct mbuf *, int, int);
741 void		 m_print(const struct mbuf *, int);
742 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
743 struct mbuf	*m_pullup(struct mbuf *, int);
744 int		 m_sanity(struct mbuf *, int);
745 struct mbuf	*m_split(struct mbuf *, int, int);
746 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
747 struct mbuf	*m_unshare(struct mbuf *, int);
748 void		 m_snd_tag_init(struct m_snd_tag *, struct ifnet *);
749 void		 m_snd_tag_destroy(struct m_snd_tag *);
750 
751 static __inline int
752 m_gettype(int size)
753 {
754 	int type;
755 
756 	switch (size) {
757 	case MSIZE:
758 		type = EXT_MBUF;
759 		break;
760 	case MCLBYTES:
761 		type = EXT_CLUSTER;
762 		break;
763 #if MJUMPAGESIZE != MCLBYTES
764 	case MJUMPAGESIZE:
765 		type = EXT_JUMBOP;
766 		break;
767 #endif
768 	case MJUM9BYTES:
769 		type = EXT_JUMBO9;
770 		break;
771 	case MJUM16BYTES:
772 		type = EXT_JUMBO16;
773 		break;
774 	default:
775 		panic("%s: invalid cluster size %d", __func__, size);
776 	}
777 
778 	return (type);
779 }
780 
781 /*
782  * Associated an external reference counted buffer with an mbuf.
783  */
784 static __inline void
785 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
786     m_ext_free_t freef, void *arg1, void *arg2)
787 {
788 
789 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
790 
791 	atomic_add_int(ref_cnt, 1);
792 	m->m_flags |= M_EXT;
793 	m->m_ext.ext_buf = buf;
794 	m->m_ext.ext_cnt = ref_cnt;
795 	m->m_data = m->m_ext.ext_buf;
796 	m->m_ext.ext_size = size;
797 	m->m_ext.ext_free = freef;
798 	m->m_ext.ext_arg1 = arg1;
799 	m->m_ext.ext_arg2 = arg2;
800 	m->m_ext.ext_type = EXT_EXTREF;
801 	m->m_ext.ext_flags = 0;
802 }
803 
804 static __inline uma_zone_t
805 m_getzone(int size)
806 {
807 	uma_zone_t zone;
808 
809 	switch (size) {
810 	case MCLBYTES:
811 		zone = zone_clust;
812 		break;
813 #if MJUMPAGESIZE != MCLBYTES
814 	case MJUMPAGESIZE:
815 		zone = zone_jumbop;
816 		break;
817 #endif
818 	case MJUM9BYTES:
819 		zone = zone_jumbo9;
820 		break;
821 	case MJUM16BYTES:
822 		zone = zone_jumbo16;
823 		break;
824 	default:
825 		panic("%s: invalid cluster size %d", __func__, size);
826 	}
827 
828 	return (zone);
829 }
830 
831 /*
832  * Initialize an mbuf with linear storage.
833  *
834  * Inline because the consumer text overhead will be roughly the same to
835  * initialize or call a function with this many parameters and M_PKTHDR
836  * should go away with constant propagation for !MGETHDR.
837  */
838 static __inline int
839 m_init(struct mbuf *m, int how, short type, int flags)
840 {
841 	int error;
842 
843 	m->m_next = NULL;
844 	m->m_nextpkt = NULL;
845 	m->m_data = m->m_dat;
846 	m->m_len = 0;
847 	m->m_flags = flags;
848 	m->m_type = type;
849 	if (flags & M_PKTHDR)
850 		error = m_pkthdr_init(m, how);
851 	else
852 		error = 0;
853 
854 	MBUF_PROBE5(m__init, m, how, type, flags, error);
855 	return (error);
856 }
857 
858 static __inline struct mbuf *
859 m_get(int how, short type)
860 {
861 	struct mbuf *m;
862 	struct mb_args args;
863 
864 	args.flags = 0;
865 	args.type = type;
866 	m = uma_zalloc_arg(zone_mbuf, &args, how);
867 	MBUF_PROBE3(m__get, how, type, m);
868 	return (m);
869 }
870 
871 static __inline struct mbuf *
872 m_gethdr(int how, short type)
873 {
874 	struct mbuf *m;
875 	struct mb_args args;
876 
877 	args.flags = M_PKTHDR;
878 	args.type = type;
879 	m = uma_zalloc_arg(zone_mbuf, &args, how);
880 	MBUF_PROBE3(m__gethdr, how, type, m);
881 	return (m);
882 }
883 
884 static __inline struct mbuf *
885 m_getcl(int how, short type, int flags)
886 {
887 	struct mbuf *m;
888 	struct mb_args args;
889 
890 	args.flags = flags;
891 	args.type = type;
892 	m = uma_zalloc_arg(zone_pack, &args, how);
893 	MBUF_PROBE4(m__getcl, how, type, flags, m);
894 	return (m);
895 }
896 
897 /*
898  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
899  * unreferenced cluster to an mbuf(9).  It is not possible to assert
900  * that, so care can be taken only by users of the API.
901  */
902 static __inline void
903 m_cljset(struct mbuf *m, void *cl, int type)
904 {
905 	int size;
906 
907 	switch (type) {
908 	case EXT_CLUSTER:
909 		size = MCLBYTES;
910 		break;
911 #if MJUMPAGESIZE != MCLBYTES
912 	case EXT_JUMBOP:
913 		size = MJUMPAGESIZE;
914 		break;
915 #endif
916 	case EXT_JUMBO9:
917 		size = MJUM9BYTES;
918 		break;
919 	case EXT_JUMBO16:
920 		size = MJUM16BYTES;
921 		break;
922 	default:
923 		panic("%s: unknown cluster type %d", __func__, type);
924 		break;
925 	}
926 
927 	m->m_data = m->m_ext.ext_buf = cl;
928 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
929 	m->m_ext.ext_size = size;
930 	m->m_ext.ext_type = type;
931 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
932 	m->m_ext.ext_count = 1;
933 	m->m_flags |= M_EXT;
934 	MBUF_PROBE3(m__cljset, m, cl, type);
935 }
936 
937 static __inline void
938 m_chtype(struct mbuf *m, short new_type)
939 {
940 
941 	m->m_type = new_type;
942 }
943 
944 static __inline void
945 m_clrprotoflags(struct mbuf *m)
946 {
947 
948 	while (m) {
949 		m->m_flags &= ~M_PROTOFLAGS;
950 		m = m->m_next;
951 	}
952 }
953 
954 static __inline struct mbuf *
955 m_last(struct mbuf *m)
956 {
957 
958 	while (m->m_next)
959 		m = m->m_next;
960 	return (m);
961 }
962 
963 static inline u_int
964 m_extrefcnt(struct mbuf *m)
965 {
966 
967 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
968 
969 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
970 	    *m->m_ext.ext_cnt);
971 }
972 
973 /*
974  * mbuf, cluster, and external object allocation macros (for compatibility
975  * purposes).
976  */
977 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
978 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
979 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
980 #define	MCLGET(m, how)		m_clget((m), (how))
981 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
982     m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2),	\
983     (flags), (type))
984 #define	m_getm(m, len, how, type)					\
985     m_getm2((m), (len), (how), (type), M_PKTHDR)
986 
987 /*
988  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
989  * be both the local data payload, or an external buffer area, depending on
990  * whether M_EXT is set).
991  */
992 #define	M_WRITABLE(m)	(((m)->m_flags & (M_RDONLY | M_NOMAP)) == 0 &&	\
993 			 (!(((m)->m_flags & M_EXT)) ||			\
994 			 (m_extrefcnt(m) == 1)))
995 
996 /* Check if the supplied mbuf has a packet header, or else panic. */
997 #define	M_ASSERTPKTHDR(m)						\
998 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
999 	    ("%s: no mbuf packet header!", __func__))
1000 
1001 /*
1002  * Ensure that the supplied mbuf is a valid, non-free mbuf.
1003  *
1004  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
1005  */
1006 #define	M_ASSERTVALID(m)						\
1007 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
1008 	    ("%s: attempted use of a free mbuf!", __func__))
1009 
1010 /*
1011  * Return the address of the start of the buffer associated with an mbuf,
1012  * handling external storage, packet-header mbufs, and regular data mbufs.
1013  */
1014 #define	M_START(m)							\
1015 	(((m)->m_flags & M_NOMAP) ? NULL :				\
1016 	 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
1017 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
1018 	 &(m)->m_dat[0])
1019 
1020 /*
1021  * Return the size of the buffer associated with an mbuf, handling external
1022  * storage, packet-header mbufs, and regular data mbufs.
1023  */
1024 #define	M_SIZE(m)							\
1025 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
1026 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
1027 	 MLEN)
1028 
1029 /*
1030  * Set the m_data pointer of a newly allocated mbuf to place an object of the
1031  * specified size at the end of the mbuf, longword aligned.
1032  *
1033  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1034  * separate macros, each asserting that it was called at the proper moment.
1035  * This required callers to themselves test the storage type and call the
1036  * right one.  Rather than require callers to be aware of those layout
1037  * decisions, we centralize here.
1038  */
1039 static __inline void
1040 m_align(struct mbuf *m, int len)
1041 {
1042 #ifdef INVARIANTS
1043 	const char *msg = "%s: not a virgin mbuf";
1044 #endif
1045 	int adjust;
1046 
1047 	KASSERT(m->m_data == M_START(m), (msg, __func__));
1048 
1049 	adjust = M_SIZE(m) - len;
1050 	m->m_data += adjust &~ (sizeof(long)-1);
1051 }
1052 
1053 #define	M_ALIGN(m, len)		m_align(m, len)
1054 #define	MH_ALIGN(m, len)	m_align(m, len)
1055 #define	MEXT_ALIGN(m, len)	m_align(m, len)
1056 
1057 /*
1058  * Compute the amount of space available before the current start of data in
1059  * an mbuf.
1060  *
1061  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1062  * of checking writability of the mbuf data area rests solely with the caller.
1063  *
1064  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1065  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1066  * read-only as well.
1067  */
1068 #define	M_LEADINGSPACE(m)						\
1069 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1070 
1071 /*
1072  * Compute the amount of space available after the end of data in an mbuf.
1073  *
1074  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1075  * of checking writability of the mbuf data area rests solely with the caller.
1076  *
1077  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1078  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1079  * read-only as well.
1080  */
1081 #define	M_TRAILINGSPACE(m)						\
1082 	(M_WRITABLE(m) ?						\
1083 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
1084 
1085 /*
1086  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
1087  * allocated, how specifies whether to wait.  If the allocation fails, the
1088  * original mbuf chain is freed and m is set to NULL.
1089  */
1090 #define	M_PREPEND(m, plen, how) do {					\
1091 	struct mbuf **_mmp = &(m);					\
1092 	struct mbuf *_mm = *_mmp;					\
1093 	int _mplen = (plen);						\
1094 	int __mhow = (how);						\
1095 									\
1096 	MBUF_CHECKSLEEP(how);						\
1097 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
1098 		_mm->m_data -= _mplen;					\
1099 		_mm->m_len += _mplen;					\
1100 	} else								\
1101 		_mm = m_prepend(_mm, _mplen, __mhow);			\
1102 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
1103 		_mm->m_pkthdr.len += _mplen;				\
1104 	*_mmp = _mm;							\
1105 } while (0)
1106 
1107 /*
1108  * Change mbuf to new type.  This is a relatively expensive operation and
1109  * should be avoided.
1110  */
1111 #define	MCHTYPE(m, t)	m_chtype((m), (t))
1112 
1113 /* Return the rcvif of a packet header. */
1114 static __inline struct ifnet *
1115 m_rcvif(struct mbuf *m)
1116 {
1117 
1118 	M_ASSERTPKTHDR(m);
1119 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1120 		return (NULL);
1121 	return (m->m_pkthdr.rcvif);
1122 }
1123 
1124 /* Length to m_copy to copy all. */
1125 #define	M_COPYALL	1000000000
1126 
1127 extern int		max_datalen;	/* MHLEN - max_hdr */
1128 extern int		max_hdr;	/* Largest link + protocol header */
1129 extern int		max_linkhdr;	/* Largest link-level header */
1130 extern int		max_protohdr;	/* Largest protocol header */
1131 extern int		nmbclusters;	/* Maximum number of clusters */
1132 extern bool		mb_use_ext_pgs;	/* Use ext_pgs for sendfile */
1133 
1134 /*-
1135  * Network packets may have annotations attached by affixing a list of
1136  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1137  * allocated semi-opaque data structures that have a fixed header
1138  * (struct m_tag) that specifies the size of the memory block and a
1139  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1140  * unsigned value used to identify a module or ABI.  By convention this value
1141  * is chosen as the date+time that the module is created, expressed as the
1142  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1143  * value is an ABI/module-specific value that identifies a particular
1144  * annotation and is private to the module.  For compatibility with systems
1145  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1146  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1147  * compatibility shim functions and several tag types are defined below.
1148  * Users that do not require compatibility should use a private cookie value
1149  * so that packet tag-related definitions can be maintained privately.
1150  *
1151  * Note that the packet tag returned by m_tag_alloc has the default memory
1152  * alignment implemented by malloc.  To reference private data one can use a
1153  * construct like:
1154  *
1155  *	struct m_tag *mtag = m_tag_alloc(...);
1156  *	struct foo *p = (struct foo *)(mtag+1);
1157  *
1158  * if the alignment of struct m_tag is sufficient for referencing members of
1159  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1160  * private data structure to insure proper alignment; e.g.,
1161  *
1162  *	struct foo {
1163  *		struct m_tag	tag;
1164  *		...
1165  *	};
1166  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1167  *	struct m_tag *mtag = &p->tag;
1168  */
1169 
1170 /*
1171  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1172  * tags are expected to ``vanish'' when they pass through a network
1173  * interface.  For most interfaces this happens normally as the tags are
1174  * reclaimed when the mbuf is free'd.  However in some special cases
1175  * reclaiming must be done manually.  An example is packets that pass through
1176  * the loopback interface.  Also, one must be careful to do this when
1177  * ``turning around'' packets (e.g., icmp_reflect).
1178  *
1179  * To mark a tag persistent bit-or this flag in when defining the tag id.
1180  * The tag will then be treated as described above.
1181  */
1182 #define	MTAG_PERSISTENT				0x800
1183 
1184 #define	PACKET_TAG_NONE				0  /* Nadda */
1185 
1186 /* Packet tags for use with PACKET_ABI_COMPAT. */
1187 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1188 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1189 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1190 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1191 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1192 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1193 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1194 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1195 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1196 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1197 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1198 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1199 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1200 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1201 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1202 #define	PACKET_TAG_DIVERT			17 /* divert info */
1203 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1204 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1205 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1206 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1207 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1208 #define	PACKET_TAG_CARP				28 /* CARP info */
1209 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1210 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1211 
1212 /* Specific cookies and tags. */
1213 
1214 /* Packet tag routines. */
1215 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1216 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1217 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1218 void		 m_tag_free_default(struct m_tag *);
1219 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1220 struct m_tag	*m_tag_copy(struct m_tag *, int);
1221 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1222 void		 m_tag_delete_nonpersistent(struct mbuf *);
1223 
1224 /*
1225  * Initialize the list of tags associated with an mbuf.
1226  */
1227 static __inline void
1228 m_tag_init(struct mbuf *m)
1229 {
1230 
1231 	SLIST_INIT(&m->m_pkthdr.tags);
1232 }
1233 
1234 /*
1235  * Set up the contents of a tag.  Note that this does not fill in the free
1236  * method; the caller is expected to do that.
1237  *
1238  * XXX probably should be called m_tag_init, but that was already taken.
1239  */
1240 static __inline void
1241 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1242 {
1243 
1244 	t->m_tag_id = type;
1245 	t->m_tag_len = len;
1246 	t->m_tag_cookie = cookie;
1247 }
1248 
1249 /*
1250  * Reclaim resources associated with a tag.
1251  */
1252 static __inline void
1253 m_tag_free(struct m_tag *t)
1254 {
1255 
1256 	(*t->m_tag_free)(t);
1257 }
1258 
1259 /*
1260  * Return the first tag associated with an mbuf.
1261  */
1262 static __inline struct m_tag *
1263 m_tag_first(struct mbuf *m)
1264 {
1265 
1266 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1267 }
1268 
1269 /*
1270  * Return the next tag in the list of tags associated with an mbuf.
1271  */
1272 static __inline struct m_tag *
1273 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1274 {
1275 
1276 	return (SLIST_NEXT(t, m_tag_link));
1277 }
1278 
1279 /*
1280  * Prepend a tag to the list of tags associated with an mbuf.
1281  */
1282 static __inline void
1283 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1284 {
1285 
1286 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1287 }
1288 
1289 /*
1290  * Unlink a tag from the list of tags associated with an mbuf.
1291  */
1292 static __inline void
1293 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1294 {
1295 
1296 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1297 }
1298 
1299 /* These are for OpenBSD compatibility. */
1300 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1301 
1302 static __inline struct m_tag *
1303 m_tag_get(int type, int length, int wait)
1304 {
1305 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1306 }
1307 
1308 static __inline struct m_tag *
1309 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1310 {
1311 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1312 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1313 }
1314 
1315 static inline struct m_snd_tag *
1316 m_snd_tag_ref(struct m_snd_tag *mst)
1317 {
1318 
1319 	refcount_acquire(&mst->refcount);
1320 	return (mst);
1321 }
1322 
1323 static inline void
1324 m_snd_tag_rele(struct m_snd_tag *mst)
1325 {
1326 
1327 	if (refcount_release(&mst->refcount))
1328 		m_snd_tag_destroy(mst);
1329 }
1330 
1331 static __inline struct mbuf *
1332 m_free(struct mbuf *m)
1333 {
1334 	struct mbuf *n = m->m_next;
1335 
1336 	MBUF_PROBE1(m__free, m);
1337 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1338 		m_tag_delete_chain(m, NULL);
1339 	if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1340 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
1341 	if (m->m_flags & M_EXT)
1342 		mb_free_ext(m);
1343 	else if ((m->m_flags & M_NOFREE) == 0)
1344 		uma_zfree(zone_mbuf, m);
1345 	return (n);
1346 }
1347 
1348 static __inline int
1349 rt_m_getfib(struct mbuf *m)
1350 {
1351 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1352 	return (m->m_pkthdr.fibnum);
1353 }
1354 
1355 #define M_GETFIB(_m)   rt_m_getfib(_m)
1356 
1357 #define M_SETFIB(_m, _fib) do {						\
1358         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1359 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1360 } while (0)
1361 
1362 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1363 #define	MBUF_HASHFLAG_L2	(1 << 2)
1364 #define	MBUF_HASHFLAG_L3	(1 << 3)
1365 #define	MBUF_HASHFLAG_L4	(1 << 4)
1366 
1367 /* mbuf hashing helper routines */
1368 uint32_t	m_ether_tcpip_hash_init(void);
1369 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1370 
1371 #ifdef MBUF_PROFILING
1372  void m_profile(struct mbuf *m);
1373  #define M_PROFILE(m) m_profile(m)
1374 #else
1375  #define M_PROFILE(m)
1376 #endif
1377 
1378 struct mbufq {
1379 	STAILQ_HEAD(, mbuf)	mq_head;
1380 	int			mq_len;
1381 	int			mq_maxlen;
1382 };
1383 
1384 static inline void
1385 mbufq_init(struct mbufq *mq, int maxlen)
1386 {
1387 
1388 	STAILQ_INIT(&mq->mq_head);
1389 	mq->mq_maxlen = maxlen;
1390 	mq->mq_len = 0;
1391 }
1392 
1393 static inline struct mbuf *
1394 mbufq_flush(struct mbufq *mq)
1395 {
1396 	struct mbuf *m;
1397 
1398 	m = STAILQ_FIRST(&mq->mq_head);
1399 	STAILQ_INIT(&mq->mq_head);
1400 	mq->mq_len = 0;
1401 	return (m);
1402 }
1403 
1404 static inline void
1405 mbufq_drain(struct mbufq *mq)
1406 {
1407 	struct mbuf *m, *n;
1408 
1409 	n = mbufq_flush(mq);
1410 	while ((m = n) != NULL) {
1411 		n = STAILQ_NEXT(m, m_stailqpkt);
1412 		m_freem(m);
1413 	}
1414 }
1415 
1416 static inline struct mbuf *
1417 mbufq_first(const struct mbufq *mq)
1418 {
1419 
1420 	return (STAILQ_FIRST(&mq->mq_head));
1421 }
1422 
1423 static inline struct mbuf *
1424 mbufq_last(const struct mbufq *mq)
1425 {
1426 
1427 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1428 }
1429 
1430 static inline int
1431 mbufq_full(const struct mbufq *mq)
1432 {
1433 
1434 	return (mq->mq_len >= mq->mq_maxlen);
1435 }
1436 
1437 static inline int
1438 mbufq_len(const struct mbufq *mq)
1439 {
1440 
1441 	return (mq->mq_len);
1442 }
1443 
1444 static inline int
1445 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1446 {
1447 
1448 	if (mbufq_full(mq))
1449 		return (ENOBUFS);
1450 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1451 	mq->mq_len++;
1452 	return (0);
1453 }
1454 
1455 static inline struct mbuf *
1456 mbufq_dequeue(struct mbufq *mq)
1457 {
1458 	struct mbuf *m;
1459 
1460 	m = STAILQ_FIRST(&mq->mq_head);
1461 	if (m) {
1462 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1463 		m->m_nextpkt = NULL;
1464 		mq->mq_len--;
1465 	}
1466 	return (m);
1467 }
1468 
1469 static inline void
1470 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1471 {
1472 
1473 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1474 	mq->mq_len++;
1475 }
1476 
1477 /*
1478  * Note: this doesn't enforce the maximum list size for dst.
1479  */
1480 static inline void
1481 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1482 {
1483 
1484 	mq_dst->mq_len += mq_src->mq_len;
1485 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1486 	mq_src->mq_len = 0;
1487 }
1488 
1489 #ifdef _SYS_TIMESPEC_H_
1490 static inline void
1491 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1492 {
1493 
1494 	KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1495 	KASSERT((m->m_flags & M_TSTMP) != 0, ("mbuf %p no M_TSTMP", m));
1496 	ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1497 	ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1498 }
1499 #endif
1500 
1501 #ifdef NETDUMP
1502 /* Invoked from the netdump client code. */
1503 void	netdump_mbuf_drain(void);
1504 void	netdump_mbuf_dump(void);
1505 void	netdump_mbuf_reinit(int nmbuf, int nclust, int clsize);
1506 #endif
1507 
1508 #endif /* _KERNEL */
1509 #endif /* !_SYS_MBUF_H_ */
1510