xref: /freebsd/sys/sys/mbuf.h (revision 74036c4de983981ee2b23f91e285363293aa7c3c)
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 /*
48  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
49  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
50  * sys/param.h), which has no additional overhead and is used instead of the
51  * internal data area; this is done when at least MINCLSIZE of data must be
52  * stored.  Additionally, it is possible to allocate a separate buffer
53  * externally and attach it to the mbuf in a way similar to that of mbuf
54  * clusters.
55  *
56  * MLEN is data length in a normal mbuf.
57  * MHLEN is data length in an mbuf with pktheader.
58  * MINCLSIZE is a smallest amount of data that should be put into cluster.
59  */
60 #define	MLEN		((int)(MSIZE - sizeof(struct m_hdr)))
61 #define	MHLEN		((int)(MLEN - sizeof(struct pkthdr)))
62 #define	MINCLSIZE	(MHLEN + 1)
63 
64 #ifdef _KERNEL
65 /*-
66  * Macro for type conversion: convert mbuf pointer to data pointer of correct
67  * type:
68  *
69  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
70  */
71 #define	mtod(m, t)	((t)((m)->m_data))
72 
73 /*
74  * Argument structure passed to UMA routines during mbuf and packet
75  * allocations.
76  */
77 struct mb_args {
78 	int	flags;	/* Flags for mbuf being allocated */
79 	short	type;	/* Type of mbuf being allocated */
80 };
81 #endif /* _KERNEL */
82 
83 #if defined(__LP64__)
84 #define M_HDR_PAD    6
85 #else
86 #define M_HDR_PAD    2
87 #endif
88 
89 /*
90  * Header present at the beginning of every mbuf.
91  */
92 struct m_hdr {
93 	struct mbuf	*mh_next;	/* next buffer in chain */
94 	struct mbuf	*mh_nextpkt;	/* next chain in queue/record */
95 	caddr_t		 mh_data;	/* location of data */
96 	int		 mh_len;	/* amount of data in this mbuf */
97 	int		 mh_flags;	/* flags; see below */
98 	short		 mh_type;	/* type of data in this mbuf */
99 	uint8_t          pad[M_HDR_PAD];/* word align                  */
100 };
101 
102 /*
103  * Packet tag structure (see below for details).
104  */
105 struct m_tag {
106 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
107 	u_int16_t		m_tag_id;	/* Tag ID */
108 	u_int16_t		m_tag_len;	/* Length of data */
109 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
110 	void			(*m_tag_free)(struct m_tag *);
111 };
112 
113 /*
114  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
115  */
116 struct pkthdr {
117 	struct ifnet	*rcvif;		/* rcv interface */
118 	/* variables for ip and tcp reassembly */
119 	void		*header;	/* pointer to packet header */
120 	int		 len;		/* total packet length */
121 	uint32_t	 flowid;	/* packet's 4-tuple system
122 					 * flow identifier
123 					 */
124 	/* variables for hardware checksum */
125 	int		 csum_flags;	/* flags regarding checksum */
126 	int		 csum_data;	/* data field used by csum routines */
127 	u_int16_t	 tso_segsz;	/* TSO segment size */
128 	union {
129 		u_int16_t vt_vtag;	/* Ethernet 802.1p+q vlan tag */
130 		u_int16_t vt_nrecs;	/* # of IGMPv3 records in this chain */
131 	} PH_vt;
132 	u_int16_t	 fibnum;	/* this packet should use this fib */
133 	u_int16_t	 pad2;		/* align to 32 bits */
134 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
135 };
136 #define ether_vtag	PH_vt.vt_vtag
137 
138 /*
139  * Description of external storage mapped into mbuf; valid only if M_EXT is
140  * set.
141  */
142 struct m_ext {
143 	caddr_t		 ext_buf;	/* start of buffer */
144 	void		(*ext_free)	/* free routine if not the usual */
145 			    (void *, void *);
146 	void		*ext_arg1;	/* optional argument pointer */
147 	void		*ext_arg2;	/* optional argument pointer */
148 	u_int		 ext_size;	/* size of buffer, for ext_free */
149 	volatile u_int	*ref_cnt;	/* pointer to ref count info */
150 	int		 ext_type;	/* type of external storage */
151 };
152 
153 /*
154  * The core of the mbuf object along with some shortcut defines for practical
155  * purposes.
156  */
157 struct mbuf {
158 	struct m_hdr	m_hdr;
159 	union {
160 		struct {
161 			struct pkthdr	MH_pkthdr;	/* M_PKTHDR set */
162 			union {
163 				struct m_ext	MH_ext;	/* M_EXT set */
164 				char		MH_databuf[MHLEN];
165 			} MH_dat;
166 		} MH;
167 		char	M_databuf[MLEN];		/* !M_PKTHDR, !M_EXT */
168 	} M_dat;
169 };
170 #define	m_next		m_hdr.mh_next
171 #define	m_len		m_hdr.mh_len
172 #define	m_data		m_hdr.mh_data
173 #define	m_type		m_hdr.mh_type
174 #define	m_flags		m_hdr.mh_flags
175 #define	m_nextpkt	m_hdr.mh_nextpkt
176 #define	m_act		m_nextpkt
177 #define	m_pkthdr	M_dat.MH.MH_pkthdr
178 #define	m_ext		M_dat.MH.MH_dat.MH_ext
179 #define	m_pktdat	M_dat.MH.MH_dat.MH_databuf
180 #define	m_dat		M_dat.M_databuf
181 
182 /*
183  * mbuf flags.
184  */
185 #define	M_EXT		0x00000001 /* has associated external storage */
186 #define	M_PKTHDR	0x00000002 /* start of record */
187 #define	M_EOR		0x00000004 /* end of record */
188 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
189 #define	M_PROTO1	0x00000010 /* protocol-specific */
190 #define	M_PROTO2	0x00000020 /* protocol-specific */
191 #define	M_PROTO3	0x00000040 /* protocol-specific */
192 #define	M_PROTO4	0x00000080 /* protocol-specific */
193 #define	M_PROTO5	0x00000100 /* protocol-specific */
194 #define	M_BCAST		0x00000200 /* send/received as link-level broadcast */
195 #define	M_MCAST		0x00000400 /* send/received as link-level multicast */
196 #define	M_FRAG		0x00000800 /* packet is a fragment of a larger packet */
197 #define	M_FIRSTFRAG	0x00001000 /* packet is first fragment */
198 #define	M_LASTFRAG	0x00002000 /* packet is last fragment */
199 #define	M_SKIP_FIREWALL	0x00004000 /* skip firewall processing */
200 		     /*	0x00008000    free */
201 #define	M_VLANTAG	0x00010000 /* ether_vtag is valid */
202 #define	M_PROMISC	0x00020000 /* packet was not for us */
203 #define	M_NOFREE	0x00040000 /* do not free mbuf, embedded in cluster */
204 #define	M_PROTO6	0x00080000 /* protocol-specific */
205 #define	M_PROTO7	0x00100000 /* protocol-specific */
206 #define	M_PROTO8	0x00200000 /* protocol-specific */
207 #define	M_FLOWID	0x00400000 /* deprecated: flowid is valid */
208 #define	M_HASHTYPEBITS	0x0F000000 /* mask of bits holding flowid hash type */
209 
210 #define	M_NOTIFICATION	M_PROTO5    /* SCTP notification */
211 
212 /*
213  * Flags to purge when crossing layers.
214  */
215 #define	M_PROTOFLAGS \
216     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8)
217 
218 /*
219  * Network interface cards are able to hash protocol fields (such as IPv4
220  * addresses and TCP port numbers) classify packets into flows.  These flows
221  * can then be used to maintain ordering while delivering packets to the OS
222  * via parallel input queues, as well as to provide a stateless affinity
223  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
224  * m_flag fields to indicate how the hash should be interpreted by the
225  * network stack.
226  *
227  * Most NICs support RSS, which provides ordering and explicit affinity, and
228  * use the hash m_flag bits to indicate what header fields were covered by
229  * the hash.  M_HASHTYPE_OPAQUE can be set by non-RSS cards or configurations
230  * that provide an opaque flow identifier, allowing for ordering and
231  * distribution without explicit affinity.
232  */
233 #define	M_HASHTYPE_SHIFT		24
234 #define	M_HASHTYPE_NONE			0x0
235 #define	M_HASHTYPE_RSS_IPV4		0x1	/* IPv4 2-tuple */
236 #define	M_HASHTYPE_RSS_TCP_IPV4		0x2	/* TCPv4 4-tuple */
237 #define	M_HASHTYPE_RSS_IPV6		0x3	/* IPv6 2-tuple */
238 #define	M_HASHTYPE_RSS_TCP_IPV6		0x4	/* TCPv6 4-tuple */
239 #define	M_HASHTYPE_RSS_IPV6_EX		0x5	/* IPv6 2-tuple + ext hdrs */
240 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	0x6	/* TCPv6 4-tiple + ext hdrs */
241 #define	M_HASHTYPE_OPAQUE		0xf	/* ordering, not affinity */
242 
243 #define	M_HASHTYPE_CLEAR(m)	(m)->m_flags &= ~(M_HASHTYPEBITS)
244 #define	M_HASHTYPE_GET(m)	(((m)->m_flags & M_HASHTYPEBITS) >> \
245 				    M_HASHTYPE_SHIFT)
246 #define	M_HASHTYPE_SET(m, v)	do {					\
247 	(m)->m_flags &= ~M_HASHTYPEBITS;				\
248 	(m)->m_flags |= ((v) << M_HASHTYPE_SHIFT);			\
249 } while (0)
250 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
251 
252 /*
253  * Flags preserved when copying m_pkthdr.
254  */
255 #define	M_COPYFLAGS \
256     (M_PKTHDR|M_EOR|M_RDONLY|M_PROTOFLAGS|M_SKIP_FIREWALL|M_BCAST|M_MCAST|\
257      M_FRAG|M_FIRSTFRAG|M_LASTFRAG|M_VLANTAG|M_PROMISC|M_HASHTYPEBITS)
258 
259 /*
260  * External buffer types: identify ext_buf type.
261  */
262 #define	EXT_CLUSTER	1	/* mbuf cluster */
263 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_bufs */
264 #define	EXT_JUMBOP	3	/* jumbo cluster 4096 bytes */
265 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
266 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
267 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
268 #define	EXT_MBUF	7	/* external mbuf reference (M_IOVEC) */
269 #define	EXT_NET_DRV	100	/* custom ext_buf provided by net driver(s) */
270 #define	EXT_MOD_TYPE	200	/* custom module's ext_buf type */
271 #define	EXT_DISPOSABLE	300	/* can throw this buffer away w/page flipping */
272 #define	EXT_EXTREF	400	/* has externally maintained ref_cnt ptr */
273 
274 /*
275  * Flags indicating hw checksum support and sw checksum requirements.  This
276  * field can be directly tested against if_data.ifi_hwassist.
277  */
278 #define	CSUM_IP			0x0001		/* will csum IP */
279 #define	CSUM_TCP		0x0002		/* will csum TCP */
280 #define	CSUM_UDP		0x0004		/* will csum UDP */
281 #define	CSUM_FRAGMENT		0x0010		/* will do IP fragmentation */
282 #define	CSUM_TSO		0x0020		/* will do TSO */
283 #define	CSUM_SCTP		0x0040		/* will csum SCTP */
284 #define CSUM_SCTP_IPV6		0x0080		/* will csum IPv6/SCTP */
285 
286 #define	CSUM_IP_CHECKED		0x0100		/* did csum IP */
287 #define	CSUM_IP_VALID		0x0200		/*   ... the csum is valid */
288 #define	CSUM_DATA_VALID		0x0400		/* csum_data field is valid */
289 #define	CSUM_PSEUDO_HDR		0x0800		/* csum_data has pseudo hdr */
290 #define	CSUM_SCTP_VALID		0x1000		/* SCTP checksum is valid */
291 #define	CSUM_UDP_IPV6		0x2000		/* will csum IPv6/UDP */
292 #define	CSUM_TCP_IPV6		0x4000		/* will csum IPv6/TCP */
293 /*	CSUM_TSO_IPV6		0x8000		will do IPv6/TSO */
294 
295 /*	CSUM_FRAGMENT_IPV6	0x10000		will do IPv6 fragementation */
296 
297 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
298 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
299 
300 #define	CSUM_DELAY_DATA		(CSUM_TCP | CSUM_UDP)
301 #define	CSUM_DELAY_IP		(CSUM_IP)	/* Only v4, no v6 IP hdr csum */
302 
303 /*
304  * mbuf types.
305  */
306 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
307 #define	MT_DATA		1	/* dynamic (data) allocation */
308 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
309 #define	MT_SONAME	8	/* socket name */
310 #define	MT_CONTROL	14	/* extra-data protocol message */
311 #define	MT_OOBDATA	15	/* expedited data  */
312 #define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
313 
314 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
315 				   a non-initialized mbuf */
316 
317 #define MB_NOTAGS	0x1UL	/* no tags attached to mbuf */
318 
319 /*
320  * General mbuf allocator statistics structure.
321  *
322  * Many of these statistics are no longer used; we instead track many
323  * allocator statistics through UMA's built in statistics mechanism.
324  */
325 struct mbstat {
326 	u_long	m_mbufs;	/* XXX */
327 	u_long	m_mclusts;	/* XXX */
328 
329 	u_long	m_drain;	/* times drained protocols for space */
330 	u_long	m_mcfail;	/* XXX: times m_copym failed */
331 	u_long	m_mpfail;	/* XXX: times m_pullup failed */
332 	u_long	m_msize;	/* length of an mbuf */
333 	u_long	m_mclbytes;	/* length of an mbuf cluster */
334 	u_long	m_minclsize;	/* min length of data to allocate a cluster */
335 	u_long	m_mlen;		/* length of data in an mbuf */
336 	u_long	m_mhlen;	/* length of data in a header mbuf */
337 
338 	/* Number of mbtypes (gives # elems in mbtypes[] array) */
339 	short	m_numtypes;
340 
341 	/* XXX: Sendfile stats should eventually move to their own struct */
342 	u_long	sf_iocnt;	/* times sendfile had to do disk I/O */
343 	u_long	sf_allocfail;	/* times sfbuf allocation failed */
344 	u_long	sf_allocwait;	/* times sfbuf allocation had to wait */
345 };
346 
347 /*
348  * Compatibility with historic mbuf allocator.
349  */
350 #define	MBTOM(how)	(how)
351 #define	M_DONTWAIT	M_NOWAIT
352 #define	M_TRYWAIT	M_WAITOK
353 #define	M_WAIT		M_WAITOK
354 
355 /*
356  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
357  * !_KERNEL so that monitoring tools can look up the zones with
358  * libmemstat(3).
359  */
360 #define	MBUF_MEM_NAME		"mbuf"
361 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
362 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
363 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
364 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
365 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
366 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
367 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
368 
369 #ifdef _KERNEL
370 
371 #ifdef WITNESS
372 #define	MBUF_CHECKSLEEP(how) do {					\
373 	if (how == M_WAITOK)						\
374 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
375 		    "Sleeping in \"%s\"", __func__);			\
376 } while (0)
377 #else
378 #define	MBUF_CHECKSLEEP(how)
379 #endif
380 
381 /*
382  * Network buffer allocation API
383  *
384  * The rest of it is defined in kern/kern_mbuf.c
385  */
386 extern uma_zone_t	zone_mbuf;
387 extern uma_zone_t	zone_clust;
388 extern uma_zone_t	zone_pack;
389 extern uma_zone_t	zone_jumbop;
390 extern uma_zone_t	zone_jumbo9;
391 extern uma_zone_t	zone_jumbo16;
392 extern uma_zone_t	zone_ext_refcnt;
393 
394 void		 mb_free_ext(struct mbuf *);
395 int		 m_pkthdr_init(struct mbuf *, int);
396 
397 static __inline int
398 m_gettype(int size)
399 {
400 	int type;
401 
402 	switch (size) {
403 	case MSIZE:
404 		type = EXT_MBUF;
405 		break;
406 	case MCLBYTES:
407 		type = EXT_CLUSTER;
408 		break;
409 #if MJUMPAGESIZE != MCLBYTES
410 	case MJUMPAGESIZE:
411 		type = EXT_JUMBOP;
412 		break;
413 #endif
414 	case MJUM9BYTES:
415 		type = EXT_JUMBO9;
416 		break;
417 	case MJUM16BYTES:
418 		type = EXT_JUMBO16;
419 		break;
420 	default:
421 		panic("%s: invalid cluster size", __func__);
422 	}
423 
424 	return (type);
425 }
426 
427 static __inline uma_zone_t
428 m_getzone(int size)
429 {
430 	uma_zone_t zone;
431 
432 	switch (size) {
433 	case MCLBYTES:
434 		zone = zone_clust;
435 		break;
436 #if MJUMPAGESIZE != MCLBYTES
437 	case MJUMPAGESIZE:
438 		zone = zone_jumbop;
439 		break;
440 #endif
441 	case MJUM9BYTES:
442 		zone = zone_jumbo9;
443 		break;
444 	case MJUM16BYTES:
445 		zone = zone_jumbo16;
446 		break;
447 	default:
448 		panic("%s: invalid cluster size", __func__);
449 	}
450 
451 	return (zone);
452 }
453 
454 /*
455  * Initialize an mbuf with linear storage.
456  *
457  * Inline because the consumer text overhead will be roughly the same to
458  * initialize or call a function with this many parameters and M_PKTHDR
459  * should go away with constant propagation for !MGETHDR.
460  */
461 static __inline int
462 m_init(struct mbuf *m, uma_zone_t zone, int size, int how, short type,
463     int flags)
464 {
465 	int error;
466 
467 	m->m_next = NULL;
468 	m->m_nextpkt = NULL;
469 	m->m_data = m->m_dat;
470 	m->m_len = 0;
471 	m->m_flags = flags;
472 	m->m_type = type;
473 	if (flags & M_PKTHDR) {
474 		if ((error = m_pkthdr_init(m, how)) != 0)
475 			return (error);
476 	}
477 
478 	return (0);
479 }
480 
481 static __inline struct mbuf *
482 m_get(int how, short type)
483 {
484 	struct mb_args args;
485 
486 	args.flags = 0;
487 	args.type = type;
488 	return (uma_zalloc_arg(zone_mbuf, &args, how));
489 }
490 
491 /*
492  * XXX This should be deprecated, very little use.
493  */
494 static __inline struct mbuf *
495 m_getclr(int how, short type)
496 {
497 	struct mbuf *m;
498 	struct mb_args args;
499 
500 	args.flags = 0;
501 	args.type = type;
502 	m = uma_zalloc_arg(zone_mbuf, &args, how);
503 	if (m != NULL)
504 		bzero(m->m_data, MLEN);
505 	return (m);
506 }
507 
508 static __inline struct mbuf *
509 m_gethdr(int how, short type)
510 {
511 	struct mb_args args;
512 
513 	args.flags = M_PKTHDR;
514 	args.type = type;
515 	return (uma_zalloc_arg(zone_mbuf, &args, how));
516 }
517 
518 static __inline struct mbuf *
519 m_getcl(int how, short type, int flags)
520 {
521 	struct mb_args args;
522 
523 	args.flags = flags;
524 	args.type = type;
525 	return (uma_zalloc_arg(zone_pack, &args, how));
526 }
527 
528 static __inline void
529 m_free_fast(struct mbuf *m)
530 {
531 #ifdef INVARIANTS
532 	if (m->m_flags & M_PKTHDR)
533 		KASSERT(SLIST_EMPTY(&m->m_pkthdr.tags), ("doing fast free of mbuf with tags"));
534 #endif
535 
536 	uma_zfree_arg(zone_mbuf, m, (void *)MB_NOTAGS);
537 }
538 
539 static __inline struct mbuf *
540 m_free(struct mbuf *m)
541 {
542 	struct mbuf *n = m->m_next;
543 
544 	if (m->m_flags & M_EXT)
545 		mb_free_ext(m);
546 	else if ((m->m_flags & M_NOFREE) == 0)
547 		uma_zfree(zone_mbuf, m);
548 	return (n);
549 }
550 
551 static __inline void
552 m_clget(struct mbuf *m, int how)
553 {
554 
555 	if (m->m_flags & M_EXT)
556 		printf("%s: %p mbuf already has cluster\n", __func__, m);
557 	m->m_ext.ext_buf = (char *)NULL;
558 	uma_zalloc_arg(zone_clust, m, how);
559 	/*
560 	 * On a cluster allocation failure, drain the packet zone and retry,
561 	 * we might be able to loosen a few clusters up on the drain.
562 	 */
563 	if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) {
564 		zone_drain(zone_pack);
565 		uma_zalloc_arg(zone_clust, m, how);
566 	}
567 }
568 
569 /*
570  * m_cljget() is different from m_clget() as it can allocate clusters without
571  * attaching them to an mbuf.  In that case the return value is the pointer
572  * to the cluster of the requested size.  If an mbuf was specified, it gets
573  * the cluster attached to it and the return value can be safely ignored.
574  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
575  */
576 static __inline void *
577 m_cljget(struct mbuf *m, int how, int size)
578 {
579 	uma_zone_t zone;
580 
581 	if (m && m->m_flags & M_EXT)
582 		printf("%s: %p mbuf already has cluster\n", __func__, m);
583 	if (m != NULL)
584 		m->m_ext.ext_buf = NULL;
585 
586 	zone = m_getzone(size);
587 	return (uma_zalloc_arg(zone, m, how));
588 }
589 
590 static __inline void
591 m_cljset(struct mbuf *m, void *cl, int type)
592 {
593 	uma_zone_t zone;
594 	int size;
595 
596 	switch (type) {
597 	case EXT_CLUSTER:
598 		size = MCLBYTES;
599 		zone = zone_clust;
600 		break;
601 #if MJUMPAGESIZE != MCLBYTES
602 	case EXT_JUMBOP:
603 		size = MJUMPAGESIZE;
604 		zone = zone_jumbop;
605 		break;
606 #endif
607 	case EXT_JUMBO9:
608 		size = MJUM9BYTES;
609 		zone = zone_jumbo9;
610 		break;
611 	case EXT_JUMBO16:
612 		size = MJUM16BYTES;
613 		zone = zone_jumbo16;
614 		break;
615 	default:
616 		panic("%s: unknown cluster type", __func__);
617 		break;
618 	}
619 
620 	m->m_data = m->m_ext.ext_buf = cl;
621 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
622 	m->m_ext.ext_size = size;
623 	m->m_ext.ext_type = type;
624 	m->m_ext.ref_cnt = uma_find_refcnt(zone, cl);
625 	m->m_flags |= M_EXT;
626 
627 }
628 
629 static __inline void
630 m_chtype(struct mbuf *m, short new_type)
631 {
632 
633 	m->m_type = new_type;
634 }
635 
636 static __inline struct mbuf *
637 m_last(struct mbuf *m)
638 {
639 
640 	while (m->m_next)
641 		m = m->m_next;
642 	return (m);
643 }
644 
645 /*
646  * mbuf, cluster, and external object allocation macros (for compatibility
647  * purposes).
648  */
649 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
650 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
651 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
652 #define	MCLGET(m, how)		m_clget((m), (how))
653 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
654     (void )m_extadd((m), (caddr_t)(buf), (size), (free), (arg1), (arg2),\
655     (flags), (type), M_NOWAIT)
656 #define	m_getm(m, len, how, type)					\
657     m_getm2((m), (len), (how), (type), M_PKTHDR)
658 
659 /*
660  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
661  * be both the local data payload, or an external buffer area, depending on
662  * whether M_EXT is set).
663  */
664 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
665 			 (!(((m)->m_flags & M_EXT)) ||			\
666 			 (*((m)->m_ext.ref_cnt) == 1)) )		\
667 
668 /* Check if the supplied mbuf has a packet header, or else panic. */
669 #define	M_ASSERTPKTHDR(m)						\
670 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
671 	    ("%s: no mbuf packet header!", __func__))
672 
673 /*
674  * Ensure that the supplied mbuf is a valid, non-free mbuf.
675  *
676  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
677  */
678 #define	M_ASSERTVALID(m)						\
679 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
680 	    ("%s: attempted use of a free mbuf!", __func__))
681 
682 /*
683  * Set the m_data pointer of a newly-allocated mbuf (m_get/MGET) to place an
684  * object of the specified size at the end of the mbuf, longword aligned.
685  */
686 #define	M_ALIGN(m, len) do {						\
687 	KASSERT(!((m)->m_flags & (M_PKTHDR|M_EXT)),			\
688 		("%s: M_ALIGN not normal mbuf", __func__));		\
689 	KASSERT((m)->m_data == (m)->m_dat,				\
690 		("%s: M_ALIGN not a virgin mbuf", __func__));		\
691 	(m)->m_data += (MLEN - (len)) & ~(sizeof(long) - 1);		\
692 } while (0)
693 
694 /*
695  * As above, for mbufs allocated with m_gethdr/MGETHDR or initialized by
696  * M_DUP/MOVE_PKTHDR.
697  */
698 #define	MH_ALIGN(m, len) do {						\
699 	KASSERT((m)->m_flags & M_PKTHDR && !((m)->m_flags & M_EXT),	\
700 		("%s: MH_ALIGN not PKTHDR mbuf", __func__));		\
701 	KASSERT((m)->m_data == (m)->m_pktdat,				\
702 		("%s: MH_ALIGN not a virgin mbuf", __func__));		\
703 	(m)->m_data += (MHLEN - (len)) & ~(sizeof(long) - 1);		\
704 } while (0)
705 
706 /*
707  * As above, for mbuf with external storage.
708  */
709 #define	MEXT_ALIGN(m, len) do {						\
710 	KASSERT((m)->m_flags & M_EXT,					\
711 		("%s: MEXT_ALIGN not an M_EXT mbuf", __func__));	\
712 	KASSERT((m)->m_data == (m)->m_ext.ext_buf,			\
713 		("%s: MEXT_ALIGN not a virgin mbuf", __func__));	\
714 	(m)->m_data += ((m)->m_ext.ext_size - (len)) &			\
715 	    ~(sizeof(long) - 1); 					\
716 } while (0)
717 
718 /*
719  * Compute the amount of space available before the current start of data in
720  * an mbuf.
721  *
722  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
723  * of checking writability of the mbuf data area rests solely with the caller.
724  */
725 #define	M_LEADINGSPACE(m)						\
726 	((m)->m_flags & M_EXT ?						\
727 	    (M_WRITABLE(m) ? (m)->m_data - (m)->m_ext.ext_buf : 0):	\
728 	    (m)->m_flags & M_PKTHDR ? (m)->m_data - (m)->m_pktdat :	\
729 	    (m)->m_data - (m)->m_dat)
730 
731 /*
732  * Compute the amount of space available after the end of data in an mbuf.
733  *
734  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
735  * of checking writability of the mbuf data area rests solely with the caller.
736  */
737 #define	M_TRAILINGSPACE(m)						\
738 	((m)->m_flags & M_EXT ?						\
739 	    (M_WRITABLE(m) ? (m)->m_ext.ext_buf + (m)->m_ext.ext_size	\
740 		- ((m)->m_data + (m)->m_len) : 0) :			\
741 	    &(m)->m_dat[MLEN] - ((m)->m_data + (m)->m_len))
742 
743 /*
744  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
745  * allocated, how specifies whether to wait.  If the allocation fails, the
746  * original mbuf chain is freed and m is set to NULL.
747  */
748 #define	M_PREPEND(m, plen, how) do {					\
749 	struct mbuf **_mmp = &(m);					\
750 	struct mbuf *_mm = *_mmp;					\
751 	int _mplen = (plen);						\
752 	int __mhow = (how);						\
753 									\
754 	MBUF_CHECKSLEEP(how);						\
755 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
756 		_mm->m_data -= _mplen;					\
757 		_mm->m_len += _mplen;					\
758 	} else								\
759 		_mm = m_prepend(_mm, _mplen, __mhow);			\
760 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
761 		_mm->m_pkthdr.len += _mplen;				\
762 	*_mmp = _mm;							\
763 } while (0)
764 
765 /*
766  * Change mbuf to new type.  This is a relatively expensive operation and
767  * should be avoided.
768  */
769 #define	MCHTYPE(m, t)	m_chtype((m), (t))
770 
771 /* Length to m_copy to copy all. */
772 #define	M_COPYALL	1000000000
773 
774 /* Compatibility with 4.3. */
775 #define	m_copy(m, o, l)	m_copym((m), (o), (l), M_NOWAIT)
776 
777 extern int		max_datalen;	/* MHLEN - max_hdr */
778 extern int		max_hdr;	/* Largest link + protocol header */
779 extern int		max_linkhdr;	/* Largest link-level header */
780 extern int		max_protohdr;	/* Largest protocol header */
781 extern struct mbstat	mbstat;		/* General mbuf stats/infos */
782 extern int		nmbclusters;	/* Maximum number of clusters */
783 
784 struct uio;
785 
786 void		 m_adj(struct mbuf *, int);
787 void		 m_align(struct mbuf *, int);
788 int		 m_apply(struct mbuf *, int, int,
789 		    int (*)(void *, void *, u_int), void *);
790 int		 m_append(struct mbuf *, int, c_caddr_t);
791 void		 m_cat(struct mbuf *, struct mbuf *);
792 int		 m_extadd(struct mbuf *, caddr_t, u_int,
793 		    void (*)(void *, void *), void *, void *, int, int, int);
794 struct mbuf	*m_collapse(struct mbuf *, int, int);
795 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
796 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
797 struct mbuf	*m_copym(struct mbuf *, int, int, int);
798 struct mbuf	*m_copymdata(struct mbuf *, struct mbuf *,
799 		    int, int, int, int);
800 struct mbuf	*m_copypacket(struct mbuf *, int);
801 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
802 struct mbuf	*m_copyup(struct mbuf *, int, int);
803 struct mbuf	*m_defrag(struct mbuf *, int);
804 void		 m_demote(struct mbuf *, int);
805 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
806 		    void (*)(char *, caddr_t, u_int));
807 struct mbuf	*m_dup(struct mbuf *, int);
808 int		 m_dup_pkthdr(struct mbuf *, struct mbuf *, int);
809 u_int		 m_fixhdr(struct mbuf *);
810 struct mbuf	*m_fragment(struct mbuf *, int, int);
811 void		 m_freem(struct mbuf *);
812 struct mbuf	*m_get2(int, int, short, int);
813 struct mbuf	*m_getjcl(int, short, int, int);
814 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
815 struct mbuf	*m_getptr(struct mbuf *, int, int *);
816 u_int		 m_length(struct mbuf *, struct mbuf **);
817 int		 m_mbuftouio(struct uio *, struct mbuf *, int);
818 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
819 struct mbuf	*m_prepend(struct mbuf *, int, int);
820 void		 m_print(const struct mbuf *, int);
821 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
822 struct mbuf	*m_pullup(struct mbuf *, int);
823 int		 m_sanity(struct mbuf *, int);
824 struct mbuf	*m_split(struct mbuf *, int, int);
825 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
826 struct mbuf	*m_unshare(struct mbuf *, int);
827 
828 /*-
829  * Network packets may have annotations attached by affixing a list of
830  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
831  * allocated semi-opaque data structures that have a fixed header
832  * (struct m_tag) that specifies the size of the memory block and a
833  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
834  * unsigned value used to identify a module or ABI.  By convention this value
835  * is chosen as the date+time that the module is created, expressed as the
836  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
837  * value is an ABI/module-specific value that identifies a particular
838  * annotation and is private to the module.  For compatibility with systems
839  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
840  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
841  * compatibility shim functions and several tag types are defined below.
842  * Users that do not require compatibility should use a private cookie value
843  * so that packet tag-related definitions can be maintained privately.
844  *
845  * Note that the packet tag returned by m_tag_alloc has the default memory
846  * alignment implemented by malloc.  To reference private data one can use a
847  * construct like:
848  *
849  *	struct m_tag *mtag = m_tag_alloc(...);
850  *	struct foo *p = (struct foo *)(mtag+1);
851  *
852  * if the alignment of struct m_tag is sufficient for referencing members of
853  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
854  * private data structure to insure proper alignment; e.g.,
855  *
856  *	struct foo {
857  *		struct m_tag	tag;
858  *		...
859  *	};
860  *	struct foo *p = (struct foo *) m_tag_alloc(...);
861  *	struct m_tag *mtag = &p->tag;
862  */
863 
864 /*
865  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
866  * tags are expected to ``vanish'' when they pass through a network
867  * interface.  For most interfaces this happens normally as the tags are
868  * reclaimed when the mbuf is free'd.  However in some special cases
869  * reclaiming must be done manually.  An example is packets that pass through
870  * the loopback interface.  Also, one must be careful to do this when
871  * ``turning around'' packets (e.g., icmp_reflect).
872  *
873  * To mark a tag persistent bit-or this flag in when defining the tag id.
874  * The tag will then be treated as described above.
875  */
876 #define	MTAG_PERSISTENT				0x800
877 
878 #define	PACKET_TAG_NONE				0  /* Nadda */
879 
880 /* Packet tags for use with PACKET_ABI_COMPAT. */
881 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
882 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
883 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
884 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
885 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
886 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
887 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
888 #define	PACKET_TAG_GIF				8  /* GIF processing done */
889 #define	PACKET_TAG_GRE				9  /* GRE processing done */
890 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
891 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
892 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
893 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
894 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
895 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
896 #define	PACKET_TAG_DIVERT			17 /* divert info */
897 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
898 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
899 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
900 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
901 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
902 #define	PACKET_TAG_CARP				28 /* CARP info */
903 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
904 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
905 
906 /* Specific cookies and tags. */
907 
908 /* Packet tag routines. */
909 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
910 void		 m_tag_delete(struct mbuf *, struct m_tag *);
911 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
912 void		 m_tag_free_default(struct m_tag *);
913 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
914 struct m_tag	*m_tag_copy(struct m_tag *, int);
915 int		 m_tag_copy_chain(struct mbuf *, struct mbuf *, int);
916 void		 m_tag_delete_nonpersistent(struct mbuf *);
917 
918 /*
919  * Initialize the list of tags associated with an mbuf.
920  */
921 static __inline void
922 m_tag_init(struct mbuf *m)
923 {
924 
925 	SLIST_INIT(&m->m_pkthdr.tags);
926 }
927 
928 /*
929  * Set up the contents of a tag.  Note that this does not fill in the free
930  * method; the caller is expected to do that.
931  *
932  * XXX probably should be called m_tag_init, but that was already taken.
933  */
934 static __inline void
935 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
936 {
937 
938 	t->m_tag_id = type;
939 	t->m_tag_len = len;
940 	t->m_tag_cookie = cookie;
941 }
942 
943 /*
944  * Reclaim resources associated with a tag.
945  */
946 static __inline void
947 m_tag_free(struct m_tag *t)
948 {
949 
950 	(*t->m_tag_free)(t);
951 }
952 
953 /*
954  * Return the first tag associated with an mbuf.
955  */
956 static __inline struct m_tag *
957 m_tag_first(struct mbuf *m)
958 {
959 
960 	return (SLIST_FIRST(&m->m_pkthdr.tags));
961 }
962 
963 /*
964  * Return the next tag in the list of tags associated with an mbuf.
965  */
966 static __inline struct m_tag *
967 m_tag_next(struct mbuf *m, struct m_tag *t)
968 {
969 
970 	return (SLIST_NEXT(t, m_tag_link));
971 }
972 
973 /*
974  * Prepend a tag to the list of tags associated with an mbuf.
975  */
976 static __inline void
977 m_tag_prepend(struct mbuf *m, struct m_tag *t)
978 {
979 
980 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
981 }
982 
983 /*
984  * Unlink a tag from the list of tags associated with an mbuf.
985  */
986 static __inline void
987 m_tag_unlink(struct mbuf *m, struct m_tag *t)
988 {
989 
990 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
991 }
992 
993 /* These are for OpenBSD compatibility. */
994 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
995 
996 static __inline struct m_tag *
997 m_tag_get(int type, int length, int wait)
998 {
999 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1000 }
1001 
1002 static __inline struct m_tag *
1003 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1004 {
1005 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1006 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1007 }
1008 
1009 static int inline
1010 rt_m_getfib(struct mbuf *m)
1011 {
1012 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1013 	return (m->m_pkthdr.fibnum);
1014 }
1015 
1016 #define M_GETFIB(_m)   rt_m_getfib(_m)
1017 
1018 #define M_SETFIB(_m, _fib) do {						\
1019         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1020 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1021 } while (0)
1022 
1023 #endif /* _KERNEL */
1024 
1025 #ifdef MBUF_PROFILING
1026  void m_profile(struct mbuf *m);
1027  #define M_PROFILE(m) m_profile(m)
1028 #else
1029  #define M_PROFILE(m)
1030 #endif
1031 
1032 
1033 #endif /* !_SYS_MBUF_H_ */
1034