xref: /freebsd/sys/sys/mbuf.h (revision 5c831a5bd61576cacb48b39f8eeb47b92707a355)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1993
3  *	The Regents of the University of California.
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	@(#)mbuf.h	8.5 (Berkeley) 2/19/95
31  * $FreeBSD$
32  */
33 
34 #ifndef _SYS_MBUF_H_
35 #define	_SYS_MBUF_H_
36 
37 /* XXX: These includes suck. Sorry! */
38 #include <sys/queue.h>
39 #ifdef _KERNEL
40 #include <sys/systm.h>
41 #include <vm/uma.h>
42 #ifdef WITNESS
43 #include <sys/lock.h>
44 #endif
45 #endif
46 
47 #ifdef _KERNEL
48 #include <sys/sdt.h>
49 
50 #define	MBUF_PROBE1(probe, arg0)					\
51 	SDT_PROBE1(sdt, , , probe, arg0)
52 #define	MBUF_PROBE2(probe, arg0, arg1)					\
53 	SDT_PROBE2(sdt, , , probe, arg0, arg1)
54 #define	MBUF_PROBE3(probe, arg0, arg1, arg2)				\
55 	SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
56 #define	MBUF_PROBE4(probe, arg0, arg1, arg2, arg3)			\
57 	SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
58 #define	MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4)		\
59 	SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
60 
61 SDT_PROBE_DECLARE(sdt, , , m__init);
62 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
63 SDT_PROBE_DECLARE(sdt, , , m__get);
64 SDT_PROBE_DECLARE(sdt, , , m__getcl);
65 SDT_PROBE_DECLARE(sdt, , , m__clget);
66 SDT_PROBE_DECLARE(sdt, , , m__cljget);
67 SDT_PROBE_DECLARE(sdt, , , m__cljset);
68 SDT_PROBE_DECLARE(sdt, , , m__free);
69 SDT_PROBE_DECLARE(sdt, , , m__freem);
70 
71 #endif /* _KERNEL */
72 
73 /*
74  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
75  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
76  * sys/param.h), which has no additional overhead and is used instead of the
77  * internal data area; this is done when at least MINCLSIZE of data must be
78  * stored.  Additionally, it is possible to allocate a separate buffer
79  * externally and attach it to the mbuf in a way similar to that of mbuf
80  * clusters.
81  *
82  * NB: These calculation do not take actual compiler-induced alignment and
83  * padding inside the complete struct mbuf into account.  Appropriate
84  * attention is required when changing members of struct mbuf.
85  *
86  * MLEN is data length in a normal mbuf.
87  * MHLEN is data length in an mbuf with pktheader.
88  * MINCLSIZE is a smallest amount of data that should be put into cluster.
89  *
90  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
91  * they are sensible.
92  */
93 struct mbuf;
94 #define	MHSIZE		offsetof(struct mbuf, m_dat)
95 #define	MPKTHSIZE	offsetof(struct mbuf, m_pktdat)
96 #define	MLEN		((int)(MSIZE - MHSIZE))
97 #define	MHLEN		((int)(MSIZE - MPKTHSIZE))
98 #define	MINCLSIZE	(MHLEN + 1)
99 
100 #ifdef _KERNEL
101 /*-
102  * Macro for type conversion: convert mbuf pointer to data pointer of correct
103  * type:
104  *
105  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
106  * mtodo(m, o) -- Same as above but with offset 'o' into data.
107  */
108 #define	mtod(m, t)	((t)((m)->m_data))
109 #define	mtodo(m, o)	((void *)(((m)->m_data) + (o)))
110 
111 /*
112  * Argument structure passed to UMA routines during mbuf and packet
113  * allocations.
114  */
115 struct mb_args {
116 	int	flags;	/* Flags for mbuf being allocated */
117 	short	type;	/* Type of mbuf being allocated */
118 };
119 #endif /* _KERNEL */
120 
121 /*
122  * Packet tag structure (see below for details).
123  */
124 struct m_tag {
125 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
126 	u_int16_t		m_tag_id;	/* Tag ID */
127 	u_int16_t		m_tag_len;	/* Length of data */
128 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
129 	void			(*m_tag_free)(struct m_tag *);
130 };
131 
132 /*
133  * Static network interface owned tag.
134  * Allocated through ifp->if_snd_tag_alloc().
135  */
136 struct m_snd_tag {
137 	struct ifnet *ifp;		/* network interface tag belongs to */
138 };
139 
140 /*
141  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
142  * Size ILP32: 48
143  *	 LP64: 56
144  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
145  * they are correct.
146  */
147 struct pkthdr {
148 	union {
149 		struct m_snd_tag *snd_tag;	/* send tag, if any */
150 		struct ifnet	*rcvif;		/* rcv interface */
151 	};
152 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
153 	int32_t		 len;		/* total packet length */
154 
155 	/* Layer crossing persistent information. */
156 	uint32_t	 flowid;	/* packet's 4-tuple system */
157 	uint64_t	 csum_flags;	/* checksum and offload features */
158 	uint16_t	 fibnum;	/* this packet should use this fib */
159 	uint8_t		 cosqos;	/* class/quality of service */
160 	uint8_t		 rsstype;	/* hash type */
161 	uint8_t		 l2hlen;	/* layer 2 header length */
162 	uint8_t		 l3hlen;	/* layer 3 header length */
163 	uint8_t		 l4hlen;	/* layer 4 header length */
164 	uint8_t		 l5hlen;	/* layer 5 header length */
165 	union {
166 		uint8_t  eight[8];
167 		uint16_t sixteen[4];
168 		uint32_t thirtytwo[2];
169 		uint64_t sixtyfour[1];
170 		uintptr_t unintptr[1];
171 		void	*ptr;
172 	} PH_per;
173 
174 	/* Layer specific non-persistent local storage for reassembly, etc. */
175 	union {
176 		uint8_t  eight[8];
177 		uint16_t sixteen[4];
178 		uint32_t thirtytwo[2];
179 		uint64_t sixtyfour[1];
180 		uintptr_t unintptr[1];
181 		void 	*ptr;
182 	} PH_loc;
183 };
184 #define	ether_vtag	PH_per.sixteen[0]
185 #define	PH_vt		PH_per
186 #define	vt_nrecs	sixteen[0]
187 #define	tso_segsz	PH_per.sixteen[1]
188 #define	lro_nsegs	tso_segsz
189 #define	csum_phsum	PH_per.sixteen[2]
190 #define	csum_data	PH_per.thirtytwo[1]
191 
192 /*
193  * Description of external storage mapped into mbuf; valid only if M_EXT is
194  * set.
195  * Size ILP32: 28
196  *	 LP64: 48
197  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
198  * they are correct.
199  */
200 struct m_ext {
201 	union {
202 		volatile u_int	 ext_count;	/* value of ref count info */
203 		volatile u_int	*ext_cnt;	/* pointer to ref count info */
204 	};
205 	caddr_t		 ext_buf;	/* start of buffer */
206 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
207 	uint32_t	 ext_type:8,	/* type of external storage */
208 			 ext_flags:24;	/* external storage mbuf flags */
209 	void		(*ext_free)	/* free routine if not the usual */
210 			    (struct mbuf *, void *, void *);
211 	void		*ext_arg1;	/* optional argument pointer */
212 	void		*ext_arg2;	/* optional argument pointer */
213 };
214 
215 /*
216  * The core of the mbuf object along with some shortcut defines for practical
217  * purposes.
218  */
219 struct mbuf {
220 	/*
221 	 * Header present at the beginning of every mbuf.
222 	 * Size ILP32: 24
223 	 *      LP64: 32
224 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
225 	 * that they are correct.
226 	 */
227 	union {	/* next buffer in chain */
228 		struct mbuf		*m_next;
229 		SLIST_ENTRY(mbuf)	m_slist;
230 		STAILQ_ENTRY(mbuf)	m_stailq;
231 	};
232 	union {	/* next chain in queue/record */
233 		struct mbuf		*m_nextpkt;
234 		SLIST_ENTRY(mbuf)	m_slistpkt;
235 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
236 	};
237 	caddr_t		 m_data;	/* location of data */
238 	int32_t		 m_len;		/* amount of data in this mbuf */
239 	uint32_t	 m_type:8,	/* type of data in this mbuf */
240 			 m_flags:24;	/* flags; see below */
241 #if !defined(__LP64__)
242 	uint32_t	 m_pad;		/* pad for 64bit alignment */
243 #endif
244 
245 	/*
246 	 * A set of optional headers (packet header, external storage header)
247 	 * and internal data storage.  Historically, these arrays were sized
248 	 * to MHLEN (space left after a packet header) and MLEN (space left
249 	 * after only a regular mbuf header); they are now variable size in
250 	 * order to support future work on variable-size mbufs.
251 	 */
252 	union {
253 		struct {
254 			struct pkthdr	m_pkthdr;	/* M_PKTHDR set */
255 			union {
256 				struct m_ext	m_ext;	/* M_EXT set */
257 				char		m_pktdat[0];
258 			};
259 		};
260 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
261 	};
262 };
263 
264 /*
265  * mbuf flags of global significance and layer crossing.
266  * Those of only protocol/layer specific significance are to be mapped
267  * to M_PROTO[1-12] and cleared at layer handoff boundaries.
268  * NB: Limited to the lower 24 bits.
269  */
270 #define	M_EXT		0x00000001 /* has associated external storage */
271 #define	M_PKTHDR	0x00000002 /* start of record */
272 #define	M_EOR		0x00000004 /* end of record */
273 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
274 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
275 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
276 #define	M_PROMISC	0x00000040 /* packet was not for us */
277 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
278 #define	M_UNUSED_8	0x00000100 /* --available-- */
279 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
280 
281 #define	M_PROTO1	0x00001000 /* protocol-specific */
282 #define	M_PROTO2	0x00002000 /* protocol-specific */
283 #define	M_PROTO3	0x00004000 /* protocol-specific */
284 #define	M_PROTO4	0x00008000 /* protocol-specific */
285 #define	M_PROTO5	0x00010000 /* protocol-specific */
286 #define	M_PROTO6	0x00020000 /* protocol-specific */
287 #define	M_PROTO7	0x00040000 /* protocol-specific */
288 #define	M_PROTO8	0x00080000 /* protocol-specific */
289 #define	M_PROTO9	0x00100000 /* protocol-specific */
290 #define	M_PROTO10	0x00200000 /* protocol-specific */
291 #define	M_PROTO11	0x00400000 /* protocol-specific */
292 #define	M_PROTO12	0x00800000 /* protocol-specific */
293 
294 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
295 
296 /*
297  * Flags to purge when crossing layers.
298  */
299 #define	M_PROTOFLAGS \
300     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
301      M_PROTO9|M_PROTO10|M_PROTO11|M_PROTO12)
302 
303 /*
304  * Flags preserved when copying m_pkthdr.
305  */
306 #define M_COPYFLAGS \
307     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG| \
308      M_PROTOFLAGS)
309 
310 /*
311  * Mbuf flag description for use with printf(9) %b identifier.
312  */
313 #define	M_FLAG_BITS \
314     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
315     "\7M_PROMISC\10M_VLANTAG"
316 #define	M_FLAG_PROTOBITS \
317     "\15M_PROTO1\16M_PROTO2\17M_PROTO3\20M_PROTO4\21M_PROTO5" \
318     "\22M_PROTO6\23M_PROTO7\24M_PROTO8\25M_PROTO9\26M_PROTO10" \
319     "\27M_PROTO11\30M_PROTO12"
320 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
321 
322 /*
323  * Network interface cards are able to hash protocol fields (such as IPv4
324  * addresses and TCP port numbers) classify packets into flows.  These flows
325  * can then be used to maintain ordering while delivering packets to the OS
326  * via parallel input queues, as well as to provide a stateless affinity
327  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
328  * m_flag fields to indicate how the hash should be interpreted by the
329  * network stack.
330  *
331  * Most NICs support RSS, which provides ordering and explicit affinity, and
332  * use the hash m_flag bits to indicate what header fields were covered by
333  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
334  * RSS cards or configurations that provide an opaque flow identifier, allowing
335  * for ordering and distribution without explicit affinity.  Additionally,
336  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
337  * properties.
338  *
339  * The meaning of the IPV6_EX suffix:
340  * "o  Home address from the home address option in the IPv6 destination
341  *     options header.  If the extension header is not present, use the Source
342  *     IPv6 Address.
343  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
344  *     associated extension header.  If the extension header is not present,
345  *     use the Destination IPv6 Address."
346  * Quoted from:
347  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
348  */
349 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
350 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
351 /* Microsoft RSS standard hash types */
352 #define	M_HASHTYPE_NONE			0
353 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
354 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
355 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
356 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
357 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
358 							    * ext hdrs */
359 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
360 							    * ext hdrs */
361 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
362 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
363 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
364 							    * ext hdrs */
365 
366 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
367 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
368 						/* ordering+hash, not affinity*/
369 
370 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
371 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
372 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
373 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
374 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
375 
376 /*
377  * COS/QOS class and quality of service tags.
378  * It uses DSCP code points as base.
379  */
380 #define	QOS_DSCP_CS0		0x00
381 #define	QOS_DSCP_DEF		QOS_DSCP_CS0
382 #define	QOS_DSCP_CS1		0x20
383 #define	QOS_DSCP_AF11		0x28
384 #define	QOS_DSCP_AF12		0x30
385 #define	QOS_DSCP_AF13		0x38
386 #define	QOS_DSCP_CS2		0x40
387 #define	QOS_DSCP_AF21		0x48
388 #define	QOS_DSCP_AF22		0x50
389 #define	QOS_DSCP_AF23		0x58
390 #define	QOS_DSCP_CS3		0x60
391 #define	QOS_DSCP_AF31		0x68
392 #define	QOS_DSCP_AF32		0x70
393 #define	QOS_DSCP_AF33		0x78
394 #define	QOS_DSCP_CS4		0x80
395 #define	QOS_DSCP_AF41		0x88
396 #define	QOS_DSCP_AF42		0x90
397 #define	QOS_DSCP_AF43		0x98
398 #define	QOS_DSCP_CS5		0xa0
399 #define	QOS_DSCP_EF		0xb8
400 #define	QOS_DSCP_CS6		0xc0
401 #define	QOS_DSCP_CS7		0xe0
402 
403 /*
404  * External mbuf storage buffer types.
405  */
406 #define	EXT_CLUSTER	1	/* mbuf cluster */
407 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
408 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
409 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
410 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
411 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
412 #define	EXT_MBUF	7	/* external mbuf reference */
413 #define	EXT_SFBUF_NOCACHE 8	/* sendfile(2)'s sf_buf not to be cached */
414 
415 #define	EXT_VENDOR1	224	/* for vendor-internal use */
416 #define	EXT_VENDOR2	225	/* for vendor-internal use */
417 #define	EXT_VENDOR3	226	/* for vendor-internal use */
418 #define	EXT_VENDOR4	227	/* for vendor-internal use */
419 
420 #define	EXT_EXP1	244	/* for experimental use */
421 #define	EXT_EXP2	245	/* for experimental use */
422 #define	EXT_EXP3	246	/* for experimental use */
423 #define	EXT_EXP4	247	/* for experimental use */
424 
425 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
426 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
427 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
428 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
429 
430 /*
431  * Flags for external mbuf buffer types.
432  * NB: limited to the lower 24 bits.
433  */
434 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
435 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
436 
437 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
438 
439 #define	EXT_FLAG_VENDOR1	0x010000	/* for vendor-internal use */
440 #define	EXT_FLAG_VENDOR2	0x020000	/* for vendor-internal use */
441 #define	EXT_FLAG_VENDOR3	0x040000	/* for vendor-internal use */
442 #define	EXT_FLAG_VENDOR4	0x080000	/* for vendor-internal use */
443 
444 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
445 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
446 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
447 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
448 
449 /*
450  * EXT flag description for use with printf(9) %b identifier.
451  */
452 #define	EXT_FLAG_BITS \
453     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
454     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
455     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
456     "\30EXT_FLAG_EXP4"
457 
458 /*
459  * External reference/free functions.
460  */
461 void sf_ext_free(void *, void *);
462 void sf_ext_free_nocache(void *, void *);
463 
464 /*
465  * Flags indicating checksum, segmentation and other offload work to be
466  * done, or already done, by hardware or lower layers.  It is split into
467  * separate inbound and outbound flags.
468  *
469  * Outbound flags that are set by upper protocol layers requesting lower
470  * layers, or ideally the hardware, to perform these offloading tasks.
471  * For outbound packets this field and its flags can be directly tested
472  * against ifnet if_hwassist.
473  */
474 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
475 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
476 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
477 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
478 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
479 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
480 
481 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
482 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
483 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
484 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
485 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
486 
487 /* Inbound checksum support where the checksum was verified by hardware. */
488 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
489 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
490 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
491 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
492 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
493 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
494 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
495 
496 /*
497  * CSUM flag description for use with printf(9) %b identifier.
498  */
499 #define	CSUM_BITS \
500     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
501     "\6CSUM_IP_ISCSI" \
502     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
503     "\16CSUM_IP6_ISCSI" \
504     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
505     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED"
506 
507 /* CSUM flags compatibility mappings. */
508 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
509 #define	CSUM_IP_VALID		CSUM_L3_VALID
510 #define	CSUM_DATA_VALID		CSUM_L4_VALID
511 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
512 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
513 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
514 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
515 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
516 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
517 #define	CSUM_TCP		CSUM_IP_TCP
518 #define	CSUM_UDP		CSUM_IP_UDP
519 #define	CSUM_SCTP		CSUM_IP_SCTP
520 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
521 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
522 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
523 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
524 
525 /*
526  * mbuf types describing the content of the mbuf (including external storage).
527  */
528 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
529 #define	MT_DATA		1	/* dynamic (data) allocation */
530 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
531 
532 #define	MT_VENDOR1	4	/* for vendor-internal use */
533 #define	MT_VENDOR2	5	/* for vendor-internal use */
534 #define	MT_VENDOR3	6	/* for vendor-internal use */
535 #define	MT_VENDOR4	7	/* for vendor-internal use */
536 
537 #define	MT_SONAME	8	/* socket name */
538 
539 #define	MT_EXP1		9	/* for experimental use */
540 #define	MT_EXP2		10	/* for experimental use */
541 #define	MT_EXP3		11	/* for experimental use */
542 #define	MT_EXP4		12	/* for experimental use */
543 
544 #define	MT_CONTROL	14	/* extra-data protocol message */
545 #define	MT_OOBDATA	15	/* expedited data  */
546 #define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
547 
548 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
549 				   a non-initialized mbuf */
550 
551 /*
552  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
553  * !_KERNEL so that monitoring tools can look up the zones with
554  * libmemstat(3).
555  */
556 #define	MBUF_MEM_NAME		"mbuf"
557 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
558 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
559 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
560 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
561 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
562 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
563 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
564 
565 #ifdef _KERNEL
566 
567 #ifdef WITNESS
568 #define	MBUF_CHECKSLEEP(how) do {					\
569 	if (how == M_WAITOK)						\
570 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
571 		    "Sleeping in \"%s\"", __func__);			\
572 } while (0)
573 #else
574 #define	MBUF_CHECKSLEEP(how)
575 #endif
576 
577 /*
578  * Network buffer allocation API
579  *
580  * The rest of it is defined in kern/kern_mbuf.c
581  */
582 extern uma_zone_t	zone_mbuf;
583 extern uma_zone_t	zone_clust;
584 extern uma_zone_t	zone_pack;
585 extern uma_zone_t	zone_jumbop;
586 extern uma_zone_t	zone_jumbo9;
587 extern uma_zone_t	zone_jumbo16;
588 
589 void		 mb_dupcl(struct mbuf *, struct mbuf *);
590 void		 mb_free_ext(struct mbuf *);
591 void		 m_adj(struct mbuf *, int);
592 int		 m_apply(struct mbuf *, int, int,
593 		    int (*)(void *, void *, u_int), void *);
594 int		 m_append(struct mbuf *, int, c_caddr_t);
595 void		 m_cat(struct mbuf *, struct mbuf *);
596 void		 m_catpkt(struct mbuf *, struct mbuf *);
597 int		 m_clget(struct mbuf *m, int how);
598 void 		*m_cljget(struct mbuf *m, int how, int size);
599 struct mbuf	*m_collapse(struct mbuf *, int, int);
600 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
601 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
602 struct mbuf	*m_copym(struct mbuf *, int, int, int);
603 struct mbuf	*m_copypacket(struct mbuf *, int);
604 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
605 struct mbuf	*m_copyup(struct mbuf *, int, int);
606 struct mbuf	*m_defrag(struct mbuf *, int);
607 void		 m_demote_pkthdr(struct mbuf *);
608 void		 m_demote(struct mbuf *, int, int);
609 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
610 		    void (*)(char *, caddr_t, u_int));
611 struct mbuf	*m_dup(const struct mbuf *, int);
612 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
613 void		 m_extadd(struct mbuf *, caddr_t, u_int,
614 		    void (*)(struct mbuf *, void *, void *), void *, void *,
615 		    int, int);
616 u_int		 m_fixhdr(struct mbuf *);
617 struct mbuf	*m_fragment(struct mbuf *, int, int);
618 void		 m_freem(struct mbuf *);
619 struct mbuf	*m_get2(int, int, short, int);
620 struct mbuf	*m_getjcl(int, short, int, int);
621 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
622 struct mbuf	*m_getptr(struct mbuf *, int, int *);
623 u_int		 m_length(struct mbuf *, struct mbuf **);
624 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
625 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
626 int		 m_pkthdr_init(struct mbuf *, int);
627 struct mbuf	*m_prepend(struct mbuf *, int, int);
628 void		 m_print(const struct mbuf *, int);
629 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
630 struct mbuf	*m_pullup(struct mbuf *, int);
631 int		 m_sanity(struct mbuf *, int);
632 struct mbuf	*m_split(struct mbuf *, int, int);
633 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
634 struct mbuf	*m_unshare(struct mbuf *, int);
635 
636 static __inline int
637 m_gettype(int size)
638 {
639 	int type;
640 
641 	switch (size) {
642 	case MSIZE:
643 		type = EXT_MBUF;
644 		break;
645 	case MCLBYTES:
646 		type = EXT_CLUSTER;
647 		break;
648 #if MJUMPAGESIZE != MCLBYTES
649 	case MJUMPAGESIZE:
650 		type = EXT_JUMBOP;
651 		break;
652 #endif
653 	case MJUM9BYTES:
654 		type = EXT_JUMBO9;
655 		break;
656 	case MJUM16BYTES:
657 		type = EXT_JUMBO16;
658 		break;
659 	default:
660 		panic("%s: invalid cluster size %d", __func__, size);
661 	}
662 
663 	return (type);
664 }
665 
666 /*
667  * Associated an external reference counted buffer with an mbuf.
668  */
669 static __inline void
670 m_extaddref(struct mbuf *m, caddr_t buf, u_int size, u_int *ref_cnt,
671     void (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2)
672 {
673 
674 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
675 
676 	atomic_add_int(ref_cnt, 1);
677 	m->m_flags |= M_EXT;
678 	m->m_ext.ext_buf = buf;
679 	m->m_ext.ext_cnt = ref_cnt;
680 	m->m_data = m->m_ext.ext_buf;
681 	m->m_ext.ext_size = size;
682 	m->m_ext.ext_free = freef;
683 	m->m_ext.ext_arg1 = arg1;
684 	m->m_ext.ext_arg2 = arg2;
685 	m->m_ext.ext_type = EXT_EXTREF;
686 	m->m_ext.ext_flags = 0;
687 }
688 
689 static __inline uma_zone_t
690 m_getzone(int size)
691 {
692 	uma_zone_t zone;
693 
694 	switch (size) {
695 	case MCLBYTES:
696 		zone = zone_clust;
697 		break;
698 #if MJUMPAGESIZE != MCLBYTES
699 	case MJUMPAGESIZE:
700 		zone = zone_jumbop;
701 		break;
702 #endif
703 	case MJUM9BYTES:
704 		zone = zone_jumbo9;
705 		break;
706 	case MJUM16BYTES:
707 		zone = zone_jumbo16;
708 		break;
709 	default:
710 		panic("%s: invalid cluster size %d", __func__, size);
711 	}
712 
713 	return (zone);
714 }
715 
716 /*
717  * Initialize an mbuf with linear storage.
718  *
719  * Inline because the consumer text overhead will be roughly the same to
720  * initialize or call a function with this many parameters and M_PKTHDR
721  * should go away with constant propagation for !MGETHDR.
722  */
723 static __inline int
724 m_init(struct mbuf *m, int how, short type, int flags)
725 {
726 	int error;
727 
728 	m->m_next = NULL;
729 	m->m_nextpkt = NULL;
730 	m->m_data = m->m_dat;
731 	m->m_len = 0;
732 	m->m_flags = flags;
733 	m->m_type = type;
734 	if (flags & M_PKTHDR)
735 		error = m_pkthdr_init(m, how);
736 	else
737 		error = 0;
738 
739 	MBUF_PROBE5(m__init, m, how, type, flags, error);
740 	return (error);
741 }
742 
743 static __inline struct mbuf *
744 m_get(int how, short type)
745 {
746 	struct mbuf *m;
747 	struct mb_args args;
748 
749 	args.flags = 0;
750 	args.type = type;
751 	m = uma_zalloc_arg(zone_mbuf, &args, how);
752 	MBUF_PROBE3(m__get, how, type, m);
753 	return (m);
754 }
755 
756 static __inline struct mbuf *
757 m_gethdr(int how, short type)
758 {
759 	struct mbuf *m;
760 	struct mb_args args;
761 
762 	args.flags = M_PKTHDR;
763 	args.type = type;
764 	m = uma_zalloc_arg(zone_mbuf, &args, how);
765 	MBUF_PROBE3(m__gethdr, how, type, m);
766 	return (m);
767 }
768 
769 static __inline struct mbuf *
770 m_getcl(int how, short type, int flags)
771 {
772 	struct mbuf *m;
773 	struct mb_args args;
774 
775 	args.flags = flags;
776 	args.type = type;
777 	m = uma_zalloc_arg(zone_pack, &args, how);
778 	MBUF_PROBE4(m__getcl, how, type, flags, m);
779 	return (m);
780 }
781 
782 /*
783  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
784  * unreferenced cluster to an mbuf(9).  It is not possible to assert
785  * that, so care can be taken only by users of the API.
786  */
787 static __inline void
788 m_cljset(struct mbuf *m, void *cl, int type)
789 {
790 	int size;
791 
792 	switch (type) {
793 	case EXT_CLUSTER:
794 		size = MCLBYTES;
795 		break;
796 #if MJUMPAGESIZE != MCLBYTES
797 	case EXT_JUMBOP:
798 		size = MJUMPAGESIZE;
799 		break;
800 #endif
801 	case EXT_JUMBO9:
802 		size = MJUM9BYTES;
803 		break;
804 	case EXT_JUMBO16:
805 		size = MJUM16BYTES;
806 		break;
807 	default:
808 		panic("%s: unknown cluster type %d", __func__, type);
809 		break;
810 	}
811 
812 	m->m_data = m->m_ext.ext_buf = cl;
813 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
814 	m->m_ext.ext_size = size;
815 	m->m_ext.ext_type = type;
816 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
817 	m->m_ext.ext_count = 1;
818 	m->m_flags |= M_EXT;
819 	MBUF_PROBE3(m__cljset, m, cl, type);
820 }
821 
822 static __inline void
823 m_chtype(struct mbuf *m, short new_type)
824 {
825 
826 	m->m_type = new_type;
827 }
828 
829 static __inline void
830 m_clrprotoflags(struct mbuf *m)
831 {
832 
833 	while (m) {
834 		m->m_flags &= ~M_PROTOFLAGS;
835 		m = m->m_next;
836 	}
837 }
838 
839 static __inline struct mbuf *
840 m_last(struct mbuf *m)
841 {
842 
843 	while (m->m_next)
844 		m = m->m_next;
845 	return (m);
846 }
847 
848 static inline u_int
849 m_extrefcnt(struct mbuf *m)
850 {
851 
852 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
853 
854 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
855 	    *m->m_ext.ext_cnt);
856 }
857 
858 /*
859  * mbuf, cluster, and external object allocation macros (for compatibility
860  * purposes).
861  */
862 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
863 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
864 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
865 #define	MCLGET(m, how)		m_clget((m), (how))
866 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
867     m_extadd((m), (caddr_t)(buf), (size), (free), (arg1), (arg2),	\
868     (flags), (type))
869 #define	m_getm(m, len, how, type)					\
870     m_getm2((m), (len), (how), (type), M_PKTHDR)
871 
872 /*
873  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
874  * be both the local data payload, or an external buffer area, depending on
875  * whether M_EXT is set).
876  */
877 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
878 			 (!(((m)->m_flags & M_EXT)) ||			\
879 			 (m_extrefcnt(m) == 1)))
880 
881 /* Check if the supplied mbuf has a packet header, or else panic. */
882 #define	M_ASSERTPKTHDR(m)						\
883 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
884 	    ("%s: no mbuf packet header!", __func__))
885 
886 /*
887  * Ensure that the supplied mbuf is a valid, non-free mbuf.
888  *
889  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
890  */
891 #define	M_ASSERTVALID(m)						\
892 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
893 	    ("%s: attempted use of a free mbuf!", __func__))
894 
895 /*
896  * Return the address of the start of the buffer associated with an mbuf,
897  * handling external storage, packet-header mbufs, and regular data mbufs.
898  */
899 #define	M_START(m)							\
900 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
901 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
902 	 &(m)->m_dat[0])
903 
904 /*
905  * Return the size of the buffer associated with an mbuf, handling external
906  * storage, packet-header mbufs, and regular data mbufs.
907  */
908 #define	M_SIZE(m)							\
909 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
910 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
911 	 MLEN)
912 
913 /*
914  * Set the m_data pointer of a newly allocated mbuf to place an object of the
915  * specified size at the end of the mbuf, longword aligned.
916  *
917  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
918  * separate macros, each asserting that it was called at the proper moment.
919  * This required callers to themselves test the storage type and call the
920  * right one.  Rather than require callers to be aware of those layout
921  * decisions, we centralize here.
922  */
923 static __inline void
924 m_align(struct mbuf *m, int len)
925 {
926 #ifdef INVARIANTS
927 	const char *msg = "%s: not a virgin mbuf";
928 #endif
929 	int adjust;
930 
931 	KASSERT(m->m_data == M_START(m), (msg, __func__));
932 
933 	adjust = M_SIZE(m) - len;
934 	m->m_data += adjust &~ (sizeof(long)-1);
935 }
936 
937 #define	M_ALIGN(m, len)		m_align(m, len)
938 #define	MH_ALIGN(m, len)	m_align(m, len)
939 #define	MEXT_ALIGN(m, len)	m_align(m, len)
940 
941 /*
942  * Compute the amount of space available before the current start of data in
943  * an mbuf.
944  *
945  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
946  * of checking writability of the mbuf data area rests solely with the caller.
947  *
948  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
949  * for mbufs with external storage.  We now allow mbuf-embedded data to be
950  * read-only as well.
951  */
952 #define	M_LEADINGSPACE(m)						\
953 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
954 
955 /*
956  * Compute the amount of space available after the end of data in an mbuf.
957  *
958  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
959  * of checking writability of the mbuf data area rests solely with the caller.
960  *
961  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
962  * for mbufs with external storage.  We now allow mbuf-embedded data to be
963  * read-only as well.
964  */
965 #define	M_TRAILINGSPACE(m)						\
966 	(M_WRITABLE(m) ?						\
967 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
968 
969 /*
970  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
971  * allocated, how specifies whether to wait.  If the allocation fails, the
972  * original mbuf chain is freed and m is set to NULL.
973  */
974 #define	M_PREPEND(m, plen, how) do {					\
975 	struct mbuf **_mmp = &(m);					\
976 	struct mbuf *_mm = *_mmp;					\
977 	int _mplen = (plen);						\
978 	int __mhow = (how);						\
979 									\
980 	MBUF_CHECKSLEEP(how);						\
981 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
982 		_mm->m_data -= _mplen;					\
983 		_mm->m_len += _mplen;					\
984 	} else								\
985 		_mm = m_prepend(_mm, _mplen, __mhow);			\
986 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
987 		_mm->m_pkthdr.len += _mplen;				\
988 	*_mmp = _mm;							\
989 } while (0)
990 
991 /*
992  * Change mbuf to new type.  This is a relatively expensive operation and
993  * should be avoided.
994  */
995 #define	MCHTYPE(m, t)	m_chtype((m), (t))
996 
997 /* Length to m_copy to copy all. */
998 #define	M_COPYALL	1000000000
999 
1000 extern int		max_datalen;	/* MHLEN - max_hdr */
1001 extern int		max_hdr;	/* Largest link + protocol header */
1002 extern int		max_linkhdr;	/* Largest link-level header */
1003 extern int		max_protohdr;	/* Largest protocol header */
1004 extern int		nmbclusters;	/* Maximum number of clusters */
1005 
1006 /*-
1007  * Network packets may have annotations attached by affixing a list of
1008  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1009  * allocated semi-opaque data structures that have a fixed header
1010  * (struct m_tag) that specifies the size of the memory block and a
1011  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1012  * unsigned value used to identify a module or ABI.  By convention this value
1013  * is chosen as the date+time that the module is created, expressed as the
1014  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1015  * value is an ABI/module-specific value that identifies a particular
1016  * annotation and is private to the module.  For compatibility with systems
1017  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1018  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1019  * compatibility shim functions and several tag types are defined below.
1020  * Users that do not require compatibility should use a private cookie value
1021  * so that packet tag-related definitions can be maintained privately.
1022  *
1023  * Note that the packet tag returned by m_tag_alloc has the default memory
1024  * alignment implemented by malloc.  To reference private data one can use a
1025  * construct like:
1026  *
1027  *	struct m_tag *mtag = m_tag_alloc(...);
1028  *	struct foo *p = (struct foo *)(mtag+1);
1029  *
1030  * if the alignment of struct m_tag is sufficient for referencing members of
1031  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1032  * private data structure to insure proper alignment; e.g.,
1033  *
1034  *	struct foo {
1035  *		struct m_tag	tag;
1036  *		...
1037  *	};
1038  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1039  *	struct m_tag *mtag = &p->tag;
1040  */
1041 
1042 /*
1043  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1044  * tags are expected to ``vanish'' when they pass through a network
1045  * interface.  For most interfaces this happens normally as the tags are
1046  * reclaimed when the mbuf is free'd.  However in some special cases
1047  * reclaiming must be done manually.  An example is packets that pass through
1048  * the loopback interface.  Also, one must be careful to do this when
1049  * ``turning around'' packets (e.g., icmp_reflect).
1050  *
1051  * To mark a tag persistent bit-or this flag in when defining the tag id.
1052  * The tag will then be treated as described above.
1053  */
1054 #define	MTAG_PERSISTENT				0x800
1055 
1056 #define	PACKET_TAG_NONE				0  /* Nadda */
1057 
1058 /* Packet tags for use with PACKET_ABI_COMPAT. */
1059 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1060 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1061 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1062 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1063 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1064 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1065 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1066 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1067 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1068 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1069 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1070 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1071 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1072 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1073 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1074 #define	PACKET_TAG_DIVERT			17 /* divert info */
1075 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1076 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1077 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1078 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1079 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1080 #define	PACKET_TAG_CARP				28 /* CARP info */
1081 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1082 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1083 
1084 /* Specific cookies and tags. */
1085 
1086 /* Packet tag routines. */
1087 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1088 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1089 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1090 void		 m_tag_free_default(struct m_tag *);
1091 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1092 struct m_tag	*m_tag_copy(struct m_tag *, int);
1093 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1094 void		 m_tag_delete_nonpersistent(struct mbuf *);
1095 
1096 /*
1097  * Initialize the list of tags associated with an mbuf.
1098  */
1099 static __inline void
1100 m_tag_init(struct mbuf *m)
1101 {
1102 
1103 	SLIST_INIT(&m->m_pkthdr.tags);
1104 }
1105 
1106 /*
1107  * Set up the contents of a tag.  Note that this does not fill in the free
1108  * method; the caller is expected to do that.
1109  *
1110  * XXX probably should be called m_tag_init, but that was already taken.
1111  */
1112 static __inline void
1113 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1114 {
1115 
1116 	t->m_tag_id = type;
1117 	t->m_tag_len = len;
1118 	t->m_tag_cookie = cookie;
1119 }
1120 
1121 /*
1122  * Reclaim resources associated with a tag.
1123  */
1124 static __inline void
1125 m_tag_free(struct m_tag *t)
1126 {
1127 
1128 	(*t->m_tag_free)(t);
1129 }
1130 
1131 /*
1132  * Return the first tag associated with an mbuf.
1133  */
1134 static __inline struct m_tag *
1135 m_tag_first(struct mbuf *m)
1136 {
1137 
1138 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1139 }
1140 
1141 /*
1142  * Return the next tag in the list of tags associated with an mbuf.
1143  */
1144 static __inline struct m_tag *
1145 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1146 {
1147 
1148 	return (SLIST_NEXT(t, m_tag_link));
1149 }
1150 
1151 /*
1152  * Prepend a tag to the list of tags associated with an mbuf.
1153  */
1154 static __inline void
1155 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1156 {
1157 
1158 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1159 }
1160 
1161 /*
1162  * Unlink a tag from the list of tags associated with an mbuf.
1163  */
1164 static __inline void
1165 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1166 {
1167 
1168 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1169 }
1170 
1171 /* These are for OpenBSD compatibility. */
1172 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1173 
1174 static __inline struct m_tag *
1175 m_tag_get(int type, int length, int wait)
1176 {
1177 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1178 }
1179 
1180 static __inline struct m_tag *
1181 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1182 {
1183 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1184 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1185 }
1186 
1187 static __inline struct mbuf *
1188 m_free(struct mbuf *m)
1189 {
1190 	struct mbuf *n = m->m_next;
1191 
1192 	MBUF_PROBE1(m__free, m);
1193 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1194 		m_tag_delete_chain(m, NULL);
1195 	if (m->m_flags & M_EXT)
1196 		mb_free_ext(m);
1197 	else if ((m->m_flags & M_NOFREE) == 0)
1198 		uma_zfree(zone_mbuf, m);
1199 	return (n);
1200 }
1201 
1202 static __inline int
1203 rt_m_getfib(struct mbuf *m)
1204 {
1205 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1206 	return (m->m_pkthdr.fibnum);
1207 }
1208 
1209 #define M_GETFIB(_m)   rt_m_getfib(_m)
1210 
1211 #define M_SETFIB(_m, _fib) do {						\
1212         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1213 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1214 } while (0)
1215 
1216 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1217 #define	MBUF_HASHFLAG_L2	(1 << 2)
1218 #define	MBUF_HASHFLAG_L3	(1 << 3)
1219 #define	MBUF_HASHFLAG_L4	(1 << 4)
1220 
1221 /* mbuf hashing helper routines */
1222 uint32_t	m_ether_tcpip_hash_init(void);
1223 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1224 
1225 #ifdef MBUF_PROFILING
1226  void m_profile(struct mbuf *m);
1227  #define M_PROFILE(m) m_profile(m)
1228 #else
1229  #define M_PROFILE(m)
1230 #endif
1231 
1232 struct mbufq {
1233 	STAILQ_HEAD(, mbuf)	mq_head;
1234 	int			mq_len;
1235 	int			mq_maxlen;
1236 };
1237 
1238 static inline void
1239 mbufq_init(struct mbufq *mq, int maxlen)
1240 {
1241 
1242 	STAILQ_INIT(&mq->mq_head);
1243 	mq->mq_maxlen = maxlen;
1244 	mq->mq_len = 0;
1245 }
1246 
1247 static inline struct mbuf *
1248 mbufq_flush(struct mbufq *mq)
1249 {
1250 	struct mbuf *m;
1251 
1252 	m = STAILQ_FIRST(&mq->mq_head);
1253 	STAILQ_INIT(&mq->mq_head);
1254 	mq->mq_len = 0;
1255 	return (m);
1256 }
1257 
1258 static inline void
1259 mbufq_drain(struct mbufq *mq)
1260 {
1261 	struct mbuf *m, *n;
1262 
1263 	n = mbufq_flush(mq);
1264 	while ((m = n) != NULL) {
1265 		n = STAILQ_NEXT(m, m_stailqpkt);
1266 		m_freem(m);
1267 	}
1268 }
1269 
1270 static inline struct mbuf *
1271 mbufq_first(const struct mbufq *mq)
1272 {
1273 
1274 	return (STAILQ_FIRST(&mq->mq_head));
1275 }
1276 
1277 static inline struct mbuf *
1278 mbufq_last(const struct mbufq *mq)
1279 {
1280 
1281 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1282 }
1283 
1284 static inline int
1285 mbufq_full(const struct mbufq *mq)
1286 {
1287 
1288 	return (mq->mq_len >= mq->mq_maxlen);
1289 }
1290 
1291 static inline int
1292 mbufq_len(const struct mbufq *mq)
1293 {
1294 
1295 	return (mq->mq_len);
1296 }
1297 
1298 static inline int
1299 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1300 {
1301 
1302 	if (mbufq_full(mq))
1303 		return (ENOBUFS);
1304 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1305 	mq->mq_len++;
1306 	return (0);
1307 }
1308 
1309 static inline struct mbuf *
1310 mbufq_dequeue(struct mbufq *mq)
1311 {
1312 	struct mbuf *m;
1313 
1314 	m = STAILQ_FIRST(&mq->mq_head);
1315 	if (m) {
1316 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1317 		m->m_nextpkt = NULL;
1318 		mq->mq_len--;
1319 	}
1320 	return (m);
1321 }
1322 
1323 static inline void
1324 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1325 {
1326 
1327 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1328 	mq->mq_len++;
1329 }
1330 
1331 /*
1332  * Note: this doesn't enforce the maximum list size for dst.
1333  */
1334 static inline void
1335 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1336 {
1337 
1338 	mq_dst->mq_len += mq_src->mq_len;
1339 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1340 	mq_src->mq_len = 0;
1341 }
1342 
1343 #endif /* _KERNEL */
1344 #endif /* !_SYS_MBUF_H_ */
1345