xref: /freebsd/sys/sys/mbuf.h (revision 5dae51da3da0cc94d17bd67b308fad304ebec7e0)
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  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
134  * Size ILP32: 48
135  *	 LP64: 56
136  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
137  * they are correct.
138  */
139 struct pkthdr {
140 	struct ifnet	*rcvif;		/* rcv interface */
141 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
142 	int32_t		 len;		/* total packet length */
143 
144 	/* Layer crossing persistent information. */
145 	uint32_t	 flowid;	/* packet's 4-tuple system */
146 	uint64_t	 csum_flags;	/* checksum and offload features */
147 	uint16_t	 fibnum;	/* this packet should use this fib */
148 	uint8_t		 cosqos;	/* class/quality of service */
149 	uint8_t		 rsstype;	/* hash type */
150 	uint8_t		 l2hlen;	/* layer 2 header length */
151 	uint8_t		 l3hlen;	/* layer 3 header length */
152 	uint8_t		 l4hlen;	/* layer 4 header length */
153 	uint8_t		 l5hlen;	/* layer 5 header length */
154 	union {
155 		uint8_t  eight[8];
156 		uint16_t sixteen[4];
157 		uint32_t thirtytwo[2];
158 		uint64_t sixtyfour[1];
159 		uintptr_t unintptr[1];
160 		void	*ptr;
161 	} PH_per;
162 
163 	/* Layer specific non-persistent local storage for reassembly, etc. */
164 	union {
165 		uint8_t  eight[8];
166 		uint16_t sixteen[4];
167 		uint32_t thirtytwo[2];
168 		uint64_t sixtyfour[1];
169 		uintptr_t unintptr[1];
170 		void 	*ptr;
171 	} PH_loc;
172 };
173 #define	ether_vtag	PH_per.sixteen[0]
174 #define	PH_vt		PH_per
175 #define	vt_nrecs	sixteen[0]
176 #define	tso_segsz	PH_per.sixteen[1]
177 #define	lro_nsegs	tso_segsz
178 #define	csum_phsum	PH_per.sixteen[2]
179 #define	csum_data	PH_per.thirtytwo[1]
180 
181 /*
182  * Description of external storage mapped into mbuf; valid only if M_EXT is
183  * set.
184  * Size ILP32: 28
185  *	 LP64: 48
186  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
187  * they are correct.
188  */
189 struct m_ext {
190 	union {
191 		volatile u_int	 ext_count;	/* value of ref count info */
192 		volatile u_int	*ext_cnt;	/* pointer to ref count info */
193 	};
194 	caddr_t		 ext_buf;	/* start of buffer */
195 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
196 	uint32_t	 ext_type:8,	/* type of external storage */
197 			 ext_flags:24;	/* external storage mbuf flags */
198 	void		(*ext_free)	/* free routine if not the usual */
199 			    (struct mbuf *, void *, void *);
200 	void		*ext_arg1;	/* optional argument pointer */
201 	void		*ext_arg2;	/* optional argument pointer */
202 };
203 
204 /*
205  * The core of the mbuf object along with some shortcut defines for practical
206  * purposes.
207  */
208 struct mbuf {
209 	/*
210 	 * Header present at the beginning of every mbuf.
211 	 * Size ILP32: 24
212 	 *      LP64: 32
213 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
214 	 * that they are correct.
215 	 */
216 	union {	/* next buffer in chain */
217 		struct mbuf		*m_next;
218 		SLIST_ENTRY(mbuf)	m_slist;
219 		STAILQ_ENTRY(mbuf)	m_stailq;
220 	};
221 	union {	/* next chain in queue/record */
222 		struct mbuf		*m_nextpkt;
223 		SLIST_ENTRY(mbuf)	m_slistpkt;
224 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
225 	};
226 	caddr_t		 m_data;	/* location of data */
227 	int32_t		 m_len;		/* amount of data in this mbuf */
228 	uint32_t	 m_type:8,	/* type of data in this mbuf */
229 			 m_flags:24;	/* flags; see below */
230 #if !defined(__LP64__)
231 	uint32_t	 m_pad;		/* pad for 64bit alignment */
232 #endif
233 
234 	/*
235 	 * A set of optional headers (packet header, external storage header)
236 	 * and internal data storage.  Historically, these arrays were sized
237 	 * to MHLEN (space left after a packet header) and MLEN (space left
238 	 * after only a regular mbuf header); they are now variable size in
239 	 * order to support future work on variable-size mbufs.
240 	 */
241 	union {
242 		struct {
243 			struct pkthdr	m_pkthdr;	/* M_PKTHDR set */
244 			union {
245 				struct m_ext	m_ext;	/* M_EXT set */
246 				char		m_pktdat[0];
247 			};
248 		};
249 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
250 	};
251 };
252 
253 /*
254  * mbuf flags of global significance and layer crossing.
255  * Those of only protocol/layer specific significance are to be mapped
256  * to M_PROTO[1-12] and cleared at layer handoff boundaries.
257  * NB: Limited to the lower 24 bits.
258  */
259 #define	M_EXT		0x00000001 /* has associated external storage */
260 #define	M_PKTHDR	0x00000002 /* start of record */
261 #define	M_EOR		0x00000004 /* end of record */
262 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
263 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
264 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
265 #define	M_PROMISC	0x00000040 /* packet was not for us */
266 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
267 #define	M_UNUSED_8	0x00000100 /* --available-- */
268 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
269 
270 #define	M_PROTO1	0x00001000 /* protocol-specific */
271 #define	M_PROTO2	0x00002000 /* protocol-specific */
272 #define	M_PROTO3	0x00004000 /* protocol-specific */
273 #define	M_PROTO4	0x00008000 /* protocol-specific */
274 #define	M_PROTO5	0x00010000 /* protocol-specific */
275 #define	M_PROTO6	0x00020000 /* protocol-specific */
276 #define	M_PROTO7	0x00040000 /* protocol-specific */
277 #define	M_PROTO8	0x00080000 /* protocol-specific */
278 #define	M_PROTO9	0x00100000 /* protocol-specific */
279 #define	M_PROTO10	0x00200000 /* protocol-specific */
280 #define	M_PROTO11	0x00400000 /* protocol-specific */
281 #define	M_PROTO12	0x00800000 /* protocol-specific */
282 
283 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
284 
285 /*
286  * Flags to purge when crossing layers.
287  */
288 #define	M_PROTOFLAGS \
289     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
290      M_PROTO9|M_PROTO10|M_PROTO11|M_PROTO12)
291 
292 /*
293  * Flags preserved when copying m_pkthdr.
294  */
295 #define M_COPYFLAGS \
296     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG| \
297      M_PROTOFLAGS)
298 
299 /*
300  * Mbuf flag description for use with printf(9) %b identifier.
301  */
302 #define	M_FLAG_BITS \
303     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
304     "\7M_PROMISC\10M_VLANTAG"
305 #define	M_FLAG_PROTOBITS \
306     "\15M_PROTO1\16M_PROTO2\17M_PROTO3\20M_PROTO4\21M_PROTO5" \
307     "\22M_PROTO6\23M_PROTO7\24M_PROTO8\25M_PROTO9\26M_PROTO10" \
308     "\27M_PROTO11\30M_PROTO12"
309 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
310 
311 /*
312  * Network interface cards are able to hash protocol fields (such as IPv4
313  * addresses and TCP port numbers) classify packets into flows.  These flows
314  * can then be used to maintain ordering while delivering packets to the OS
315  * via parallel input queues, as well as to provide a stateless affinity
316  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
317  * m_flag fields to indicate how the hash should be interpreted by the
318  * network stack.
319  *
320  * Most NICs support RSS, which provides ordering and explicit affinity, and
321  * use the hash m_flag bits to indicate what header fields were covered by
322  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
323  * RSS cards or configurations that provide an opaque flow identifier, allowing
324  * for ordering and distribution without explicit affinity.  Additionally,
325  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
326  * properties.
327  */
328 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
329 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
330 /* Microsoft RSS standard hash types */
331 #define	M_HASHTYPE_NONE			0
332 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
333 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
334 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
335 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
336 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
337 							    * ext hdrs */
338 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
339 							    * ext hdrs */
340 /* Non-standard RSS hash types */
341 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
342 #define	M_HASHTYPE_RSS_UDP_IPV4_EX	M_HASHTYPE_HASH(8) /* IPv4 UDP 4-tuple +
343 							    * ext hdrs */
344 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
345 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
346 							    * ext hdrs */
347 
348 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
349 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
350 						/* ordering+hash, not affinity*/
351 
352 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
353 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
354 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
355 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
356 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
357 
358 /*
359  * COS/QOS class and quality of service tags.
360  * It uses DSCP code points as base.
361  */
362 #define	QOS_DSCP_CS0		0x00
363 #define	QOS_DSCP_DEF		QOS_DSCP_CS0
364 #define	QOS_DSCP_CS1		0x20
365 #define	QOS_DSCP_AF11		0x28
366 #define	QOS_DSCP_AF12		0x30
367 #define	QOS_DSCP_AF13		0x38
368 #define	QOS_DSCP_CS2		0x40
369 #define	QOS_DSCP_AF21		0x48
370 #define	QOS_DSCP_AF22		0x50
371 #define	QOS_DSCP_AF23		0x58
372 #define	QOS_DSCP_CS3		0x60
373 #define	QOS_DSCP_AF31		0x68
374 #define	QOS_DSCP_AF32		0x70
375 #define	QOS_DSCP_AF33		0x78
376 #define	QOS_DSCP_CS4		0x80
377 #define	QOS_DSCP_AF41		0x88
378 #define	QOS_DSCP_AF42		0x90
379 #define	QOS_DSCP_AF43		0x98
380 #define	QOS_DSCP_CS5		0xa0
381 #define	QOS_DSCP_EF		0xb8
382 #define	QOS_DSCP_CS6		0xc0
383 #define	QOS_DSCP_CS7		0xe0
384 
385 /*
386  * External mbuf storage buffer types.
387  */
388 #define	EXT_CLUSTER	1	/* mbuf cluster */
389 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
390 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
391 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
392 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
393 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
394 #define	EXT_MBUF	7	/* external mbuf reference */
395 #define	EXT_SFBUF_NOCACHE 8	/* sendfile(2)'s sf_buf not to be cached */
396 
397 #define	EXT_VENDOR1	224	/* for vendor-internal use */
398 #define	EXT_VENDOR2	225	/* for vendor-internal use */
399 #define	EXT_VENDOR3	226	/* for vendor-internal use */
400 #define	EXT_VENDOR4	227	/* for vendor-internal use */
401 
402 #define	EXT_EXP1	244	/* for experimental use */
403 #define	EXT_EXP2	245	/* for experimental use */
404 #define	EXT_EXP3	246	/* for experimental use */
405 #define	EXT_EXP4	247	/* for experimental use */
406 
407 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
408 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
409 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
410 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
411 
412 /*
413  * Flags for external mbuf buffer types.
414  * NB: limited to the lower 24 bits.
415  */
416 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
417 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
418 
419 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
420 
421 #define	EXT_FLAG_VENDOR1	0x010000	/* for vendor-internal use */
422 #define	EXT_FLAG_VENDOR2	0x020000	/* for vendor-internal use */
423 #define	EXT_FLAG_VENDOR3	0x040000	/* for vendor-internal use */
424 #define	EXT_FLAG_VENDOR4	0x080000	/* for vendor-internal use */
425 
426 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
427 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
428 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
429 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
430 
431 /*
432  * EXT flag description for use with printf(9) %b identifier.
433  */
434 #define	EXT_FLAG_BITS \
435     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
436     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
437     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
438     "\30EXT_FLAG_EXP4"
439 
440 /*
441  * External reference/free functions.
442  */
443 void sf_ext_free(void *, void *);
444 void sf_ext_free_nocache(void *, void *);
445 
446 /*
447  * Flags indicating checksum, segmentation and other offload work to be
448  * done, or already done, by hardware or lower layers.  It is split into
449  * separate inbound and outbound flags.
450  *
451  * Outbound flags that are set by upper protocol layers requesting lower
452  * layers, or ideally the hardware, to perform these offloading tasks.
453  * For outbound packets this field and its flags can be directly tested
454  * against ifnet if_hwassist.
455  */
456 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
457 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
458 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
459 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
460 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
461 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
462 
463 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
464 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
465 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
466 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
467 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
468 
469 /* Inbound checksum support where the checksum was verified by hardware. */
470 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
471 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
472 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
473 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
474 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
475 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
476 #define	CSUM_COALESED		0x40000000	/* contains merged segments */
477 
478 /*
479  * CSUM flag description for use with printf(9) %b identifier.
480  */
481 #define	CSUM_BITS \
482     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
483     "\6CSUM_IP_ISCSI" \
484     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
485     "\16CSUM_IP6_ISCSI" \
486     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
487     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESED"
488 
489 /* CSUM flags compatibility mappings. */
490 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
491 #define	CSUM_IP_VALID		CSUM_L3_VALID
492 #define	CSUM_DATA_VALID		CSUM_L4_VALID
493 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
494 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
495 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
496 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
497 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
498 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
499 #define	CSUM_TCP		CSUM_IP_TCP
500 #define	CSUM_UDP		CSUM_IP_UDP
501 #define	CSUM_SCTP		CSUM_IP_SCTP
502 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
503 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
504 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
505 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
506 
507 /*
508  * mbuf types describing the content of the mbuf (including external storage).
509  */
510 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
511 #define	MT_DATA		1	/* dynamic (data) allocation */
512 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
513 
514 #define	MT_VENDOR1	4	/* for vendor-internal use */
515 #define	MT_VENDOR2	5	/* for vendor-internal use */
516 #define	MT_VENDOR3	6	/* for vendor-internal use */
517 #define	MT_VENDOR4	7	/* for vendor-internal use */
518 
519 #define	MT_SONAME	8	/* socket name */
520 
521 #define	MT_EXP1		9	/* for experimental use */
522 #define	MT_EXP2		10	/* for experimental use */
523 #define	MT_EXP3		11	/* for experimental use */
524 #define	MT_EXP4		12	/* for experimental use */
525 
526 #define	MT_CONTROL	14	/* extra-data protocol message */
527 #define	MT_OOBDATA	15	/* expedited data  */
528 #define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
529 
530 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
531 				   a non-initialized mbuf */
532 
533 /*
534  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
535  * !_KERNEL so that monitoring tools can look up the zones with
536  * libmemstat(3).
537  */
538 #define	MBUF_MEM_NAME		"mbuf"
539 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
540 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
541 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
542 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
543 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
544 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
545 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
546 
547 #ifdef _KERNEL
548 
549 #ifdef WITNESS
550 #define	MBUF_CHECKSLEEP(how) do {					\
551 	if (how == M_WAITOK)						\
552 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
553 		    "Sleeping in \"%s\"", __func__);			\
554 } while (0)
555 #else
556 #define	MBUF_CHECKSLEEP(how)
557 #endif
558 
559 /*
560  * Network buffer allocation API
561  *
562  * The rest of it is defined in kern/kern_mbuf.c
563  */
564 extern uma_zone_t	zone_mbuf;
565 extern uma_zone_t	zone_clust;
566 extern uma_zone_t	zone_pack;
567 extern uma_zone_t	zone_jumbop;
568 extern uma_zone_t	zone_jumbo9;
569 extern uma_zone_t	zone_jumbo16;
570 
571 void		 mb_dupcl(struct mbuf *, struct mbuf *);
572 void		 mb_free_ext(struct mbuf *);
573 void		 m_adj(struct mbuf *, int);
574 int		 m_apply(struct mbuf *, int, int,
575 		    int (*)(void *, void *, u_int), void *);
576 int		 m_append(struct mbuf *, int, c_caddr_t);
577 void		 m_cat(struct mbuf *, struct mbuf *);
578 void		 m_catpkt(struct mbuf *, struct mbuf *);
579 int		 m_clget(struct mbuf *m, int how);
580 void 		*m_cljget(struct mbuf *m, int how, int size);
581 struct mbuf	*m_collapse(struct mbuf *, int, int);
582 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
583 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
584 struct mbuf	*m_copym(struct mbuf *, int, int, int);
585 struct mbuf	*m_copypacket(struct mbuf *, int);
586 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
587 struct mbuf	*m_copyup(struct mbuf *, int, int);
588 struct mbuf	*m_defrag(struct mbuf *, int);
589 void		 m_demote_pkthdr(struct mbuf *);
590 void		 m_demote(struct mbuf *, int, int);
591 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
592 		    void (*)(char *, caddr_t, u_int));
593 struct mbuf	*m_dup(const struct mbuf *, int);
594 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
595 void		 m_extadd(struct mbuf *, caddr_t, u_int,
596 		    void (*)(struct mbuf *, void *, void *), void *, void *,
597 		    int, int);
598 u_int		 m_fixhdr(struct mbuf *);
599 struct mbuf	*m_fragment(struct mbuf *, int, int);
600 void		 m_freem(struct mbuf *);
601 struct mbuf	*m_get2(int, int, short, int);
602 struct mbuf	*m_getjcl(int, short, int, int);
603 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
604 struct mbuf	*m_getptr(struct mbuf *, int, int *);
605 u_int		 m_length(struct mbuf *, struct mbuf **);
606 int		 m_mbuftouio(struct uio *, struct mbuf *, int);
607 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
608 int		 m_pkthdr_init(struct mbuf *, int);
609 struct mbuf	*m_prepend(struct mbuf *, int, int);
610 void		 m_print(const struct mbuf *, int);
611 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
612 struct mbuf	*m_pullup(struct mbuf *, int);
613 int		 m_sanity(struct mbuf *, int);
614 struct mbuf	*m_split(struct mbuf *, int, int);
615 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
616 struct mbuf	*m_unshare(struct mbuf *, int);
617 
618 static __inline int
619 m_gettype(int size)
620 {
621 	int type;
622 
623 	switch (size) {
624 	case MSIZE:
625 		type = EXT_MBUF;
626 		break;
627 	case MCLBYTES:
628 		type = EXT_CLUSTER;
629 		break;
630 #if MJUMPAGESIZE != MCLBYTES
631 	case MJUMPAGESIZE:
632 		type = EXT_JUMBOP;
633 		break;
634 #endif
635 	case MJUM9BYTES:
636 		type = EXT_JUMBO9;
637 		break;
638 	case MJUM16BYTES:
639 		type = EXT_JUMBO16;
640 		break;
641 	default:
642 		panic("%s: invalid cluster size %d", __func__, size);
643 	}
644 
645 	return (type);
646 }
647 
648 /*
649  * Associated an external reference counted buffer with an mbuf.
650  */
651 static __inline void
652 m_extaddref(struct mbuf *m, caddr_t buf, u_int size, u_int *ref_cnt,
653     void (*freef)(struct mbuf *, void *, void *), void *arg1, void *arg2)
654 {
655 
656 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
657 
658 	atomic_add_int(ref_cnt, 1);
659 	m->m_flags |= M_EXT;
660 	m->m_ext.ext_buf = buf;
661 	m->m_ext.ext_cnt = ref_cnt;
662 	m->m_data = m->m_ext.ext_buf;
663 	m->m_ext.ext_size = size;
664 	m->m_ext.ext_free = freef;
665 	m->m_ext.ext_arg1 = arg1;
666 	m->m_ext.ext_arg2 = arg2;
667 	m->m_ext.ext_type = EXT_EXTREF;
668 	m->m_ext.ext_flags = 0;
669 }
670 
671 static __inline uma_zone_t
672 m_getzone(int size)
673 {
674 	uma_zone_t zone;
675 
676 	switch (size) {
677 	case MCLBYTES:
678 		zone = zone_clust;
679 		break;
680 #if MJUMPAGESIZE != MCLBYTES
681 	case MJUMPAGESIZE:
682 		zone = zone_jumbop;
683 		break;
684 #endif
685 	case MJUM9BYTES:
686 		zone = zone_jumbo9;
687 		break;
688 	case MJUM16BYTES:
689 		zone = zone_jumbo16;
690 		break;
691 	default:
692 		panic("%s: invalid cluster size %d", __func__, size);
693 	}
694 
695 	return (zone);
696 }
697 
698 /*
699  * Initialize an mbuf with linear storage.
700  *
701  * Inline because the consumer text overhead will be roughly the same to
702  * initialize or call a function with this many parameters and M_PKTHDR
703  * should go away with constant propagation for !MGETHDR.
704  */
705 static __inline int
706 m_init(struct mbuf *m, int how, short type, int flags)
707 {
708 	int error;
709 
710 	m->m_next = NULL;
711 	m->m_nextpkt = NULL;
712 	m->m_data = m->m_dat;
713 	m->m_len = 0;
714 	m->m_flags = flags;
715 	m->m_type = type;
716 	if (flags & M_PKTHDR)
717 		error = m_pkthdr_init(m, how);
718 	else
719 		error = 0;
720 
721 	MBUF_PROBE5(m__init, m, how, type, flags, error);
722 	return (error);
723 }
724 
725 static __inline struct mbuf *
726 m_get(int how, short type)
727 {
728 	struct mbuf *m;
729 	struct mb_args args;
730 
731 	args.flags = 0;
732 	args.type = type;
733 	m = uma_zalloc_arg(zone_mbuf, &args, how);
734 	MBUF_PROBE3(m__get, how, type, m);
735 	return (m);
736 }
737 
738 static __inline struct mbuf *
739 m_gethdr(int how, short type)
740 {
741 	struct mbuf *m;
742 	struct mb_args args;
743 
744 	args.flags = M_PKTHDR;
745 	args.type = type;
746 	m = uma_zalloc_arg(zone_mbuf, &args, how);
747 	MBUF_PROBE3(m__gethdr, how, type, m);
748 	return (m);
749 }
750 
751 static __inline struct mbuf *
752 m_getcl(int how, short type, int flags)
753 {
754 	struct mbuf *m;
755 	struct mb_args args;
756 
757 	args.flags = flags;
758 	args.type = type;
759 	m = uma_zalloc_arg(zone_pack, &args, how);
760 	MBUF_PROBE4(m__getcl, how, type, flags, m);
761 	return (m);
762 }
763 
764 /*
765  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
766  * unreferenced cluster to an mbuf(9).  It is not possible to assert
767  * that, so care can be taken only by users of the API.
768  */
769 static __inline void
770 m_cljset(struct mbuf *m, void *cl, int type)
771 {
772 	int size;
773 
774 	switch (type) {
775 	case EXT_CLUSTER:
776 		size = MCLBYTES;
777 		break;
778 #if MJUMPAGESIZE != MCLBYTES
779 	case EXT_JUMBOP:
780 		size = MJUMPAGESIZE;
781 		break;
782 #endif
783 	case EXT_JUMBO9:
784 		size = MJUM9BYTES;
785 		break;
786 	case EXT_JUMBO16:
787 		size = MJUM16BYTES;
788 		break;
789 	default:
790 		panic("%s: unknown cluster type %d", __func__, type);
791 		break;
792 	}
793 
794 	m->m_data = m->m_ext.ext_buf = cl;
795 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
796 	m->m_ext.ext_size = size;
797 	m->m_ext.ext_type = type;
798 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
799 	m->m_ext.ext_count = 1;
800 	m->m_flags |= M_EXT;
801 	MBUF_PROBE3(m__cljset, m, cl, type);
802 }
803 
804 static __inline void
805 m_chtype(struct mbuf *m, short new_type)
806 {
807 
808 	m->m_type = new_type;
809 }
810 
811 static __inline void
812 m_clrprotoflags(struct mbuf *m)
813 {
814 
815 	while (m) {
816 		m->m_flags &= ~M_PROTOFLAGS;
817 		m = m->m_next;
818 	}
819 }
820 
821 static __inline struct mbuf *
822 m_last(struct mbuf *m)
823 {
824 
825 	while (m->m_next)
826 		m = m->m_next;
827 	return (m);
828 }
829 
830 static inline u_int
831 m_extrefcnt(struct mbuf *m)
832 {
833 
834 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
835 
836 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
837 	    *m->m_ext.ext_cnt);
838 }
839 
840 /*
841  * mbuf, cluster, and external object allocation macros (for compatibility
842  * purposes).
843  */
844 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
845 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
846 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
847 #define	MCLGET(m, how)		m_clget((m), (how))
848 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
849     m_extadd((m), (caddr_t)(buf), (size), (free), (arg1), (arg2),	\
850     (flags), (type))
851 #define	m_getm(m, len, how, type)					\
852     m_getm2((m), (len), (how), (type), M_PKTHDR)
853 
854 /*
855  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
856  * be both the local data payload, or an external buffer area, depending on
857  * whether M_EXT is set).
858  */
859 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
860 			 (!(((m)->m_flags & M_EXT)) ||			\
861 			 (m_extrefcnt(m) == 1)))
862 
863 /* Check if the supplied mbuf has a packet header, or else panic. */
864 #define	M_ASSERTPKTHDR(m)						\
865 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
866 	    ("%s: no mbuf packet header!", __func__))
867 
868 /*
869  * Ensure that the supplied mbuf is a valid, non-free mbuf.
870  *
871  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
872  */
873 #define	M_ASSERTVALID(m)						\
874 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
875 	    ("%s: attempted use of a free mbuf!", __func__))
876 
877 /*
878  * Return the address of the start of the buffer associated with an mbuf,
879  * handling external storage, packet-header mbufs, and regular data mbufs.
880  */
881 #define	M_START(m)							\
882 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
883 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
884 	 &(m)->m_dat[0])
885 
886 /*
887  * Return the size of the buffer associated with an mbuf, handling external
888  * storage, packet-header mbufs, and regular data mbufs.
889  */
890 #define	M_SIZE(m)							\
891 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
892 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
893 	 MLEN)
894 
895 /*
896  * Set the m_data pointer of a newly allocated mbuf to place an object of the
897  * specified size at the end of the mbuf, longword aligned.
898  *
899  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
900  * separate macros, each asserting that it was called at the proper moment.
901  * This required callers to themselves test the storage type and call the
902  * right one.  Rather than require callers to be aware of those layout
903  * decisions, we centralize here.
904  */
905 static __inline void
906 m_align(struct mbuf *m, int len)
907 {
908 #ifdef INVARIANTS
909 	const char *msg = "%s: not a virgin mbuf";
910 #endif
911 	int adjust;
912 
913 	KASSERT(m->m_data == M_START(m), (msg, __func__));
914 
915 	adjust = M_SIZE(m) - len;
916 	m->m_data += adjust &~ (sizeof(long)-1);
917 }
918 
919 #define	M_ALIGN(m, len)		m_align(m, len)
920 #define	MH_ALIGN(m, len)	m_align(m, len)
921 #define	MEXT_ALIGN(m, len)	m_align(m, len)
922 
923 /*
924  * Compute the amount of space available before the current start of data in
925  * an mbuf.
926  *
927  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
928  * of checking writability of the mbuf data area rests solely with the caller.
929  *
930  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
931  * for mbufs with external storage.  We now allow mbuf-embedded data to be
932  * read-only as well.
933  */
934 #define	M_LEADINGSPACE(m)						\
935 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
936 
937 /*
938  * Compute the amount of space available after the end of data in an mbuf.
939  *
940  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
941  * of checking writability of the mbuf data area rests solely with the caller.
942  *
943  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
944  * for mbufs with external storage.  We now allow mbuf-embedded data to be
945  * read-only as well.
946  */
947 #define	M_TRAILINGSPACE(m)						\
948 	(M_WRITABLE(m) ?						\
949 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
950 
951 /*
952  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
953  * allocated, how specifies whether to wait.  If the allocation fails, the
954  * original mbuf chain is freed and m is set to NULL.
955  */
956 #define	M_PREPEND(m, plen, how) do {					\
957 	struct mbuf **_mmp = &(m);					\
958 	struct mbuf *_mm = *_mmp;					\
959 	int _mplen = (plen);						\
960 	int __mhow = (how);						\
961 									\
962 	MBUF_CHECKSLEEP(how);						\
963 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
964 		_mm->m_data -= _mplen;					\
965 		_mm->m_len += _mplen;					\
966 	} else								\
967 		_mm = m_prepend(_mm, _mplen, __mhow);			\
968 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
969 		_mm->m_pkthdr.len += _mplen;				\
970 	*_mmp = _mm;							\
971 } while (0)
972 
973 /*
974  * Change mbuf to new type.  This is a relatively expensive operation and
975  * should be avoided.
976  */
977 #define	MCHTYPE(m, t)	m_chtype((m), (t))
978 
979 /* Length to m_copy to copy all. */
980 #define	M_COPYALL	1000000000
981 
982 extern int		max_datalen;	/* MHLEN - max_hdr */
983 extern int		max_hdr;	/* Largest link + protocol header */
984 extern int		max_linkhdr;	/* Largest link-level header */
985 extern int		max_protohdr;	/* Largest protocol header */
986 extern int		nmbclusters;	/* Maximum number of clusters */
987 
988 /*-
989  * Network packets may have annotations attached by affixing a list of
990  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
991  * allocated semi-opaque data structures that have a fixed header
992  * (struct m_tag) that specifies the size of the memory block and a
993  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
994  * unsigned value used to identify a module or ABI.  By convention this value
995  * is chosen as the date+time that the module is created, expressed as the
996  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
997  * value is an ABI/module-specific value that identifies a particular
998  * annotation and is private to the module.  For compatibility with systems
999  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1000  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1001  * compatibility shim functions and several tag types are defined below.
1002  * Users that do not require compatibility should use a private cookie value
1003  * so that packet tag-related definitions can be maintained privately.
1004  *
1005  * Note that the packet tag returned by m_tag_alloc has the default memory
1006  * alignment implemented by malloc.  To reference private data one can use a
1007  * construct like:
1008  *
1009  *	struct m_tag *mtag = m_tag_alloc(...);
1010  *	struct foo *p = (struct foo *)(mtag+1);
1011  *
1012  * if the alignment of struct m_tag is sufficient for referencing members of
1013  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1014  * private data structure to insure proper alignment; e.g.,
1015  *
1016  *	struct foo {
1017  *		struct m_tag	tag;
1018  *		...
1019  *	};
1020  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1021  *	struct m_tag *mtag = &p->tag;
1022  */
1023 
1024 /*
1025  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1026  * tags are expected to ``vanish'' when they pass through a network
1027  * interface.  For most interfaces this happens normally as the tags are
1028  * reclaimed when the mbuf is free'd.  However in some special cases
1029  * reclaiming must be done manually.  An example is packets that pass through
1030  * the loopback interface.  Also, one must be careful to do this when
1031  * ``turning around'' packets (e.g., icmp_reflect).
1032  *
1033  * To mark a tag persistent bit-or this flag in when defining the tag id.
1034  * The tag will then be treated as described above.
1035  */
1036 #define	MTAG_PERSISTENT				0x800
1037 
1038 #define	PACKET_TAG_NONE				0  /* Nadda */
1039 
1040 /* Packet tags for use with PACKET_ABI_COMPAT. */
1041 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1042 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1043 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1044 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1045 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1046 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1047 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1048 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1049 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1050 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1051 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1052 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1053 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1054 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1055 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1056 #define	PACKET_TAG_DIVERT			17 /* divert info */
1057 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1058 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1059 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1060 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1061 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1062 #define	PACKET_TAG_CARP				28 /* CARP info */
1063 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1064 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1065 
1066 /* Specific cookies and tags. */
1067 
1068 /* Packet tag routines. */
1069 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1070 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1071 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1072 void		 m_tag_free_default(struct m_tag *);
1073 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1074 struct m_tag	*m_tag_copy(struct m_tag *, int);
1075 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1076 void		 m_tag_delete_nonpersistent(struct mbuf *);
1077 
1078 /*
1079  * Initialize the list of tags associated with an mbuf.
1080  */
1081 static __inline void
1082 m_tag_init(struct mbuf *m)
1083 {
1084 
1085 	SLIST_INIT(&m->m_pkthdr.tags);
1086 }
1087 
1088 /*
1089  * Set up the contents of a tag.  Note that this does not fill in the free
1090  * method; the caller is expected to do that.
1091  *
1092  * XXX probably should be called m_tag_init, but that was already taken.
1093  */
1094 static __inline void
1095 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1096 {
1097 
1098 	t->m_tag_id = type;
1099 	t->m_tag_len = len;
1100 	t->m_tag_cookie = cookie;
1101 }
1102 
1103 /*
1104  * Reclaim resources associated with a tag.
1105  */
1106 static __inline void
1107 m_tag_free(struct m_tag *t)
1108 {
1109 
1110 	(*t->m_tag_free)(t);
1111 }
1112 
1113 /*
1114  * Return the first tag associated with an mbuf.
1115  */
1116 static __inline struct m_tag *
1117 m_tag_first(struct mbuf *m)
1118 {
1119 
1120 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1121 }
1122 
1123 /*
1124  * Return the next tag in the list of tags associated with an mbuf.
1125  */
1126 static __inline struct m_tag *
1127 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1128 {
1129 
1130 	return (SLIST_NEXT(t, m_tag_link));
1131 }
1132 
1133 /*
1134  * Prepend a tag to the list of tags associated with an mbuf.
1135  */
1136 static __inline void
1137 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1138 {
1139 
1140 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1141 }
1142 
1143 /*
1144  * Unlink a tag from the list of tags associated with an mbuf.
1145  */
1146 static __inline void
1147 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1148 {
1149 
1150 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1151 }
1152 
1153 /* These are for OpenBSD compatibility. */
1154 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1155 
1156 static __inline struct m_tag *
1157 m_tag_get(int type, int length, int wait)
1158 {
1159 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1160 }
1161 
1162 static __inline struct m_tag *
1163 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1164 {
1165 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1166 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1167 }
1168 
1169 static __inline struct mbuf *
1170 m_free(struct mbuf *m)
1171 {
1172 	struct mbuf *n = m->m_next;
1173 
1174 	MBUF_PROBE1(m__free, m);
1175 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1176 		m_tag_delete_chain(m, NULL);
1177 	if (m->m_flags & M_EXT)
1178 		mb_free_ext(m);
1179 	else if ((m->m_flags & M_NOFREE) == 0)
1180 		uma_zfree(zone_mbuf, m);
1181 	return (n);
1182 }
1183 
1184 static __inline int
1185 rt_m_getfib(struct mbuf *m)
1186 {
1187 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1188 	return (m->m_pkthdr.fibnum);
1189 }
1190 
1191 #define M_GETFIB(_m)   rt_m_getfib(_m)
1192 
1193 #define M_SETFIB(_m, _fib) do {						\
1194         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1195 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1196 } while (0)
1197 
1198 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1199 #define	MBUF_HASHFLAG_L2	(1 << 2)
1200 #define	MBUF_HASHFLAG_L3	(1 << 3)
1201 #define	MBUF_HASHFLAG_L4	(1 << 4)
1202 
1203 /* mbuf hashing helper routines */
1204 uint32_t	m_ether_tcpip_hash_init(void);
1205 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1206 
1207 #ifdef MBUF_PROFILING
1208  void m_profile(struct mbuf *m);
1209  #define M_PROFILE(m) m_profile(m)
1210 #else
1211  #define M_PROFILE(m)
1212 #endif
1213 
1214 struct mbufq {
1215 	STAILQ_HEAD(, mbuf)	mq_head;
1216 	int			mq_len;
1217 	int			mq_maxlen;
1218 };
1219 
1220 static inline void
1221 mbufq_init(struct mbufq *mq, int maxlen)
1222 {
1223 
1224 	STAILQ_INIT(&mq->mq_head);
1225 	mq->mq_maxlen = maxlen;
1226 	mq->mq_len = 0;
1227 }
1228 
1229 static inline struct mbuf *
1230 mbufq_flush(struct mbufq *mq)
1231 {
1232 	struct mbuf *m;
1233 
1234 	m = STAILQ_FIRST(&mq->mq_head);
1235 	STAILQ_INIT(&mq->mq_head);
1236 	mq->mq_len = 0;
1237 	return (m);
1238 }
1239 
1240 static inline void
1241 mbufq_drain(struct mbufq *mq)
1242 {
1243 	struct mbuf *m, *n;
1244 
1245 	n = mbufq_flush(mq);
1246 	while ((m = n) != NULL) {
1247 		n = STAILQ_NEXT(m, m_stailqpkt);
1248 		m_freem(m);
1249 	}
1250 }
1251 
1252 static inline struct mbuf *
1253 mbufq_first(const struct mbufq *mq)
1254 {
1255 
1256 	return (STAILQ_FIRST(&mq->mq_head));
1257 }
1258 
1259 static inline struct mbuf *
1260 mbufq_last(const struct mbufq *mq)
1261 {
1262 
1263 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1264 }
1265 
1266 static inline int
1267 mbufq_full(const struct mbufq *mq)
1268 {
1269 
1270 	return (mq->mq_len >= mq->mq_maxlen);
1271 }
1272 
1273 static inline int
1274 mbufq_len(const struct mbufq *mq)
1275 {
1276 
1277 	return (mq->mq_len);
1278 }
1279 
1280 static inline int
1281 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1282 {
1283 
1284 	if (mbufq_full(mq))
1285 		return (ENOBUFS);
1286 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1287 	mq->mq_len++;
1288 	return (0);
1289 }
1290 
1291 static inline struct mbuf *
1292 mbufq_dequeue(struct mbufq *mq)
1293 {
1294 	struct mbuf *m;
1295 
1296 	m = STAILQ_FIRST(&mq->mq_head);
1297 	if (m) {
1298 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1299 		m->m_nextpkt = NULL;
1300 		mq->mq_len--;
1301 	}
1302 	return (m);
1303 }
1304 
1305 static inline void
1306 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1307 {
1308 
1309 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1310 	mq->mq_len++;
1311 }
1312 #endif /* _KERNEL */
1313 #endif /* !_SYS_MBUF_H_ */
1314