xref: /freebsd/sys/sys/mbuf.h (revision ea906c4152774dff300bb26fbfc1e4188351c89a)
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 #define	MLEN		(MSIZE - sizeof(struct m_hdr))	/* normal data len */
57 #define	MHLEN		(MLEN - sizeof(struct pkthdr))	/* data len w/pkthdr */
58 #define	MINCLSIZE	(MHLEN + 1)	/* smallest amount to put in cluster */
59 #define	M_MAXCOMPRESS	(MHLEN / 2)	/* max amount to copy for compression */
60 
61 #ifdef _KERNEL
62 /*-
63  * Macros for type conversion:
64  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
65  * dtom(x)	-- Convert data pointer within mbuf to mbuf pointer (XXX).
66  */
67 #define	mtod(m, t)	((t)((m)->m_data))
68 #define	dtom(x)		((struct mbuf *)((intptr_t)(x) & ~(MSIZE-1)))
69 
70 /*
71  * Argument structure passed to UMA routines during mbuf and packet
72  * allocations.
73  */
74 struct mb_args {
75 	int	flags;	/* Flags for mbuf being allocated */
76 	short	type;	/* Type of mbuf being allocated */
77 };
78 #endif /* _KERNEL */
79 
80 #if defined(__LP64__)
81 #define M_HDR_PAD    6
82 #else
83 #define M_HDR_PAD    2
84 #endif
85 
86 /*
87  * Header present at the beginning of every mbuf.
88  */
89 struct m_hdr {
90 	struct mbuf	*mh_next;	/* next buffer in chain */
91 	struct mbuf	*mh_nextpkt;	/* next chain in queue/record */
92 	caddr_t		 mh_data;	/* location of data */
93 	int		 mh_len;	/* amount of data in this mbuf */
94 	int		 mh_flags;	/* flags; see below */
95 	short		 mh_type;	/* type of data in this mbuf */
96 	uint8_t          pad[M_HDR_PAD];/* word align                  */
97 };
98 
99 /*
100  * Packet tag structure (see below for details).
101  */
102 struct m_tag {
103 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
104 	u_int16_t		m_tag_id;	/* Tag ID */
105 	u_int16_t		m_tag_len;	/* Length of data */
106 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
107 	void			(*m_tag_free)(struct m_tag *);
108 };
109 
110 /*
111  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
112  */
113 struct pkthdr {
114 	struct ifnet	*rcvif;		/* rcv interface */
115 	/* variables for ip and tcp reassembly */
116 	void		*header;	/* pointer to packet header */
117 	int		 len;		/* total packet length */
118 	/* variables for hardware checksum */
119 	int		 csum_flags;	/* flags regarding checksum */
120 	int		 csum_data;	/* data field used by csum routines */
121 	u_int16_t	 tso_segsz;	/* TSO segment size */
122 	u_int16_t	 ether_vtag;	/* Ethernet 802.1p+q vlan tag */
123 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
124 };
125 
126 /*
127  * Description of external storage mapped into mbuf; valid only if M_EXT is
128  * set.
129  */
130 struct m_ext {
131 	caddr_t		 ext_buf;	/* start of buffer */
132 	void		(*ext_free)	/* free routine if not the usual */
133 			    (void *, void *);
134 	void		*ext_arg1;	/* optional argument pointer */
135 	void		*ext_arg2;	/* optional argument pointer */
136 	u_int		 ext_size;	/* size of buffer, for ext_free */
137 	volatile u_int	*ref_cnt;	/* pointer to ref count info */
138 	int		 ext_type;	/* type of external storage */
139 };
140 
141 /*
142  * The core of the mbuf object along with some shortcut defines for practical
143  * purposes.
144  */
145 struct mbuf {
146 	struct m_hdr	m_hdr;
147 	union {
148 		struct {
149 			struct pkthdr	MH_pkthdr;	/* M_PKTHDR set */
150 			union {
151 				struct m_ext	MH_ext;	/* M_EXT set */
152 				char		MH_databuf[MHLEN];
153 			} MH_dat;
154 		} MH;
155 		char	M_databuf[MLEN];		/* !M_PKTHDR, !M_EXT */
156 	} M_dat;
157 };
158 #define	m_next		m_hdr.mh_next
159 #define	m_len		m_hdr.mh_len
160 #define	m_data		m_hdr.mh_data
161 #define	m_type		m_hdr.mh_type
162 #define	m_flags		m_hdr.mh_flags
163 #define	m_nextpkt	m_hdr.mh_nextpkt
164 #define	m_act		m_nextpkt
165 #define	m_pkthdr	M_dat.MH.MH_pkthdr
166 #define	m_ext		M_dat.MH.MH_dat.MH_ext
167 #define	m_pktdat	M_dat.MH.MH_dat.MH_databuf
168 #define	m_dat		M_dat.M_databuf
169 
170 /*
171  * mbuf flags.
172  */
173 #define	M_EXT		0x00000001 /* has associated external storage */
174 #define	M_PKTHDR	0x00000002 /* start of record */
175 #define	M_EOR		0x00000004 /* end of record */
176 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
177 #define	M_PROTO1	0x00000010 /* protocol-specific */
178 #define	M_PROTO2	0x00000020 /* protocol-specific */
179 #define	M_PROTO3	0x00000040 /* protocol-specific */
180 #define	M_PROTO4	0x00000080 /* protocol-specific */
181 #define	M_PROTO5	0x00000100 /* protocol-specific */
182 #define	M_BCAST		0x00000200 /* send/received as link-level broadcast */
183 #define	M_MCAST		0x00000400 /* send/received as link-level multicast */
184 #define	M_FRAG		0x00000800 /* packet is a fragment of a larger packet */
185 #define	M_FIRSTFRAG	0x00001000 /* packet is first fragment */
186 #define	M_LASTFRAG	0x00002000 /* packet is last fragment */
187 #define	M_SKIP_FIREWALL	0x00004000 /* skip firewall processing */
188 #define	M_FREELIST	0x00008000 /* mbuf is on the free list */
189 #define	M_VLANTAG	0x00010000 /* ether_vtag is valid */
190 #define	M_PROMISC	0x00020000 /* packet was not for us */
191 #define	M_NOFREE	0x00040000 /* do not free mbuf, embedded in cluster */
192 #define	M_PROTO6	0x00080000 /* protocol-specific */
193 #define	M_PROTO7	0x00100000 /* protocol-specific */
194 #define	M_PROTO8	0x00200000 /* protocol-specific */
195 /*
196  * For RELENG_{6,7} steal these flags for limited multiple routing table
197  * support. In RELENG_8 and beyond, use just one flag and a tag.
198  */
199 #define	M_FIB		0xF0000000 /* steal some bits to store fib number. */
200 
201 #define	M_NOTIFICATION	M_PROTO5    /* SCTP notification */
202 
203 /*
204  * Flags to purge when crossing layers.
205  */
206 #define	M_PROTOFLAGS \
207     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8)
208 
209 /*
210  * Flags preserved when copying m_pkthdr.
211  */
212 #define	M_COPYFLAGS \
213     (M_PKTHDR|M_EOR|M_RDONLY|M_PROTOFLAGS|M_SKIP_FIREWALL|M_BCAST|M_MCAST|\
214      M_FRAG|M_FIRSTFRAG|M_LASTFRAG|M_VLANTAG|M_PROMISC|M_FIB)
215 
216 /*
217  * External buffer types: identify ext_buf type.
218  */
219 #define	EXT_CLUSTER	1	/* mbuf cluster */
220 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_bufs */
221 #define	EXT_JUMBOP	3	/* jumbo cluster 4096 bytes */
222 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
223 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
224 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
225 #define	EXT_MBUF	7	/* external mbuf reference (M_IOVEC) */
226 #define	EXT_NET_DRV	100	/* custom ext_buf provided by net driver(s) */
227 #define	EXT_MOD_TYPE	200	/* custom module's ext_buf type */
228 #define	EXT_DISPOSABLE	300	/* can throw this buffer away w/page flipping */
229 #define	EXT_EXTREF	400	/* has externally maintained ref_cnt ptr */
230 
231 /*
232  * Flags indicating hw checksum support and sw checksum requirements.  This
233  * field can be directly tested against if_data.ifi_hwassist.
234  */
235 #define	CSUM_IP			0x0001		/* will csum IP */
236 #define	CSUM_TCP		0x0002		/* will csum TCP */
237 #define	CSUM_UDP		0x0004		/* will csum UDP */
238 #define	CSUM_IP_FRAGS		0x0008		/* will csum IP fragments */
239 #define	CSUM_FRAGMENT		0x0010		/* will do IP fragmentation */
240 #define	CSUM_TSO		0x0020		/* will do TSO */
241 
242 #define	CSUM_IP_CHECKED		0x0100		/* did csum IP */
243 #define	CSUM_IP_VALID		0x0200		/*   ... the csum is valid */
244 #define	CSUM_DATA_VALID		0x0400		/* csum_data field is valid */
245 #define	CSUM_PSEUDO_HDR		0x0800		/* csum_data has pseudo hdr */
246 
247 #define	CSUM_DELAY_DATA		(CSUM_TCP | CSUM_UDP)
248 #define	CSUM_DELAY_IP		(CSUM_IP)	/* XXX add ipv6 here too? */
249 
250 /*
251  * mbuf types.
252  */
253 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
254 #define	MT_DATA		1	/* dynamic (data) allocation */
255 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
256 #define	MT_SONAME	8	/* socket name */
257 #define	MT_CONTROL	14	/* extra-data protocol message */
258 #define	MT_OOBDATA	15	/* expedited data  */
259 #define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
260 
261 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
262 				   a non-initialized mbuf */
263 
264 #define MB_NOTAGS	0x1UL	/* no tags attached to mbuf */
265 
266 /*
267  * General mbuf allocator statistics structure.
268  *
269  * Many of these statistics are no longer used; we instead track many
270  * allocator statistics through UMA's built in statistics mechanism.
271  */
272 struct mbstat {
273 	u_long	m_mbufs;	/* XXX */
274 	u_long	m_mclusts;	/* XXX */
275 
276 	u_long	m_drain;	/* times drained protocols for space */
277 	u_long	m_mcfail;	/* XXX: times m_copym failed */
278 	u_long	m_mpfail;	/* XXX: times m_pullup failed */
279 	u_long	m_msize;	/* length of an mbuf */
280 	u_long	m_mclbytes;	/* length of an mbuf cluster */
281 	u_long	m_minclsize;	/* min length of data to allocate a cluster */
282 	u_long	m_mlen;		/* length of data in an mbuf */
283 	u_long	m_mhlen;	/* length of data in a header mbuf */
284 
285 	/* Number of mbtypes (gives # elems in mbtypes[] array) */
286 	short	m_numtypes;
287 
288 	/* XXX: Sendfile stats should eventually move to their own struct */
289 	u_long	sf_iocnt;	/* times sendfile had to do disk I/O */
290 	u_long	sf_allocfail;	/* times sfbuf allocation failed */
291 	u_long	sf_allocwait;	/* times sfbuf allocation had to wait */
292 };
293 
294 /*
295  * Flags specifying how an allocation should be made.
296  *
297  * The flag to use is as follows:
298  * - M_DONTWAIT or M_NOWAIT from an interrupt handler to not block allocation.
299  * - M_WAIT or M_WAITOK from wherever it is safe to block.
300  *
301  * M_DONTWAIT/M_NOWAIT means that we will not block the thread explicitly and
302  * if we cannot allocate immediately we may return NULL, whereas
303  * M_WAIT/M_WAITOK means that if we cannot allocate resources we
304  * will block until they are available, and thus never return NULL.
305  *
306  * XXX Eventually just phase this out to use M_WAITOK/M_NOWAIT.
307  */
308 #define	MBTOM(how)	(how)
309 #define	M_DONTWAIT	M_NOWAIT
310 #define	M_TRYWAIT	M_WAITOK
311 #define	M_WAIT		M_WAITOK
312 
313 /*
314  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
315  * !_KERNEL so that monitoring tools can look up the zones with
316  * libmemstat(3).
317  */
318 #define	MBUF_MEM_NAME		"mbuf"
319 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
320 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
321 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
322 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
323 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
324 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
325 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
326 
327 #ifdef _KERNEL
328 
329 #ifdef WITNESS
330 #define	MBUF_CHECKSLEEP(how) do {					\
331 	if (how == M_WAITOK)						\
332 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
333 		    "Sleeping in \"%s\"", __func__);			\
334 } while (0)
335 #else
336 #define	MBUF_CHECKSLEEP(how)
337 #endif
338 
339 /*
340  * Network buffer allocation API
341  *
342  * The rest of it is defined in kern/kern_mbuf.c
343  */
344 
345 extern uma_zone_t	zone_mbuf;
346 extern uma_zone_t	zone_clust;
347 extern uma_zone_t	zone_pack;
348 extern uma_zone_t	zone_jumbop;
349 extern uma_zone_t	zone_jumbo9;
350 extern uma_zone_t	zone_jumbo16;
351 extern uma_zone_t	zone_ext_refcnt;
352 
353 static __inline struct mbuf	*m_getcl(int how, short type, int flags);
354 static __inline struct mbuf	*m_get(int how, short type);
355 static __inline struct mbuf	*m_gethdr(int how, short type);
356 static __inline struct mbuf	*m_getjcl(int how, short type, int flags,
357 				    int size);
358 static __inline struct mbuf	*m_getclr(int how, short type);	/* XXX */
359 static __inline struct mbuf	*m_free(struct mbuf *m);
360 static __inline void		 m_clget(struct mbuf *m, int how);
361 static __inline void		*m_cljget(struct mbuf *m, int how, int size);
362 static __inline void		 m_chtype(struct mbuf *m, short new_type);
363 void				 mb_free_ext(struct mbuf *);
364 static __inline struct mbuf	*m_last(struct mbuf *m);
365 
366 static __inline int
367 m_gettype(int size)
368 {
369 	int type;
370 
371 	switch (size) {
372 	case MSIZE:
373 		type = EXT_MBUF;
374 		break;
375 	case MCLBYTES:
376 		type = EXT_CLUSTER;
377 		break;
378 #if MJUMPAGESIZE != MCLBYTES
379 	case MJUMPAGESIZE:
380 		type = EXT_JUMBOP;
381 		break;
382 #endif
383 	case MJUM9BYTES:
384 		type = EXT_JUMBO9;
385 		break;
386 	case MJUM16BYTES:
387 		type = EXT_JUMBO16;
388 		break;
389 	default:
390 		panic("%s: m_getjcl: invalid cluster size", __func__);
391 	}
392 
393 	return (type);
394 }
395 
396 static __inline uma_zone_t
397 m_getzone(int size)
398 {
399 	uma_zone_t zone;
400 
401 	switch (size) {
402 	case MSIZE:
403 		zone = zone_mbuf;
404 		break;
405 	case MCLBYTES:
406 		zone = zone_clust;
407 		break;
408 #if MJUMPAGESIZE != MCLBYTES
409 	case MJUMPAGESIZE:
410 		zone = zone_jumbop;
411 		break;
412 #endif
413 	case MJUM9BYTES:
414 		zone = zone_jumbo9;
415 		break;
416 	case MJUM16BYTES:
417 		zone = zone_jumbo16;
418 		break;
419 	default:
420 		panic("%s: m_getjcl: invalid cluster type", __func__);
421 	}
422 
423 	return (zone);
424 }
425 
426 static __inline struct mbuf *
427 m_get(int how, short type)
428 {
429 	struct mb_args args;
430 
431 	args.flags = 0;
432 	args.type = type;
433 	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
434 }
435 
436 /*
437  * XXX This should be deprecated, very little use.
438  */
439 static __inline struct mbuf *
440 m_getclr(int how, short type)
441 {
442 	struct mbuf *m;
443 	struct mb_args args;
444 
445 	args.flags = 0;
446 	args.type = type;
447 	m = uma_zalloc_arg(zone_mbuf, &args, how);
448 	if (m != NULL)
449 		bzero(m->m_data, MLEN);
450 	return (m);
451 }
452 
453 static __inline struct mbuf *
454 m_gethdr(int how, short type)
455 {
456 	struct mb_args args;
457 
458 	args.flags = M_PKTHDR;
459 	args.type = type;
460 	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
461 }
462 
463 static __inline struct mbuf *
464 m_getcl(int how, short type, int flags)
465 {
466 	struct mb_args args;
467 
468 	args.flags = flags;
469 	args.type = type;
470 	return ((struct mbuf *)(uma_zalloc_arg(zone_pack, &args, how)));
471 }
472 
473 /*
474  * m_getjcl() returns an mbuf with a cluster of the specified size attached.
475  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
476  *
477  * XXX: This is rather large, should be real function maybe.
478  */
479 static __inline struct mbuf *
480 m_getjcl(int how, short type, int flags, int size)
481 {
482 	struct mb_args args;
483 	struct mbuf *m, *n;
484 	uma_zone_t zone;
485 
486 	args.flags = flags;
487 	args.type = type;
488 
489 	m = uma_zalloc_arg(zone_mbuf, &args, how);
490 	if (m == NULL)
491 		return (NULL);
492 
493 	zone = m_getzone(size);
494 	n = uma_zalloc_arg(zone, m, how);
495 	if (n == NULL) {
496 		uma_zfree(zone_mbuf, m);
497 		return (NULL);
498 	}
499 	return (m);
500 }
501 
502 static __inline void
503 m_free_fast(struct mbuf *m)
504 {
505 #ifdef INVARIANTS
506 	if (m->m_flags & M_PKTHDR)
507 		KASSERT(SLIST_EMPTY(&m->m_pkthdr.tags), ("doing fast free of mbuf with tags"));
508 #endif
509 
510 	uma_zfree_arg(zone_mbuf, m, (void *)MB_NOTAGS);
511 }
512 
513 static __inline struct mbuf *
514 m_free(struct mbuf *m)
515 {
516 	struct mbuf *n = m->m_next;
517 
518 	if (m->m_flags & M_EXT)
519 		mb_free_ext(m);
520 	else if ((m->m_flags & M_NOFREE) == 0)
521 		uma_zfree(zone_mbuf, m);
522 	return (n);
523 }
524 
525 static __inline void
526 m_clget(struct mbuf *m, int how)
527 {
528 
529 	if (m->m_flags & M_EXT)
530 		printf("%s: %p mbuf already has cluster\n", __func__, m);
531 	m->m_ext.ext_buf = (char *)NULL;
532 	uma_zalloc_arg(zone_clust, m, how);
533 	/*
534 	 * On a cluster allocation failure, drain the packet zone and retry,
535 	 * we might be able to loosen a few clusters up on the drain.
536 	 */
537 	if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) {
538 		zone_drain(zone_pack);
539 		uma_zalloc_arg(zone_clust, m, how);
540 	}
541 }
542 
543 /*
544  * m_cljget() is different from m_clget() as it can allocate clusters without
545  * attaching them to an mbuf.  In that case the return value is the pointer
546  * to the cluster of the requested size.  If an mbuf was specified, it gets
547  * the cluster attached to it and the return value can be safely ignored.
548  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
549  */
550 static __inline void *
551 m_cljget(struct mbuf *m, int how, int size)
552 {
553 	uma_zone_t zone;
554 
555 	if (m && m->m_flags & M_EXT)
556 		printf("%s: %p mbuf already has cluster\n", __func__, m);
557 	if (m != NULL)
558 		m->m_ext.ext_buf = NULL;
559 
560 	zone = m_getzone(size);
561 	return (uma_zalloc_arg(zone, m, how));
562 }
563 
564 static __inline void
565 m_cljset(struct mbuf *m, void *cl, int type)
566 {
567 	uma_zone_t zone;
568 	int size;
569 
570 	switch (type) {
571 	case EXT_CLUSTER:
572 		size = MCLBYTES;
573 		zone = zone_clust;
574 		break;
575 #if MJUMPAGESIZE != MCLBYTES
576 	case EXT_JUMBOP:
577 		size = MJUMPAGESIZE;
578 		zone = zone_jumbop;
579 		break;
580 #endif
581 	case EXT_JUMBO9:
582 		size = MJUM9BYTES;
583 		zone = zone_jumbo9;
584 		break;
585 	case EXT_JUMBO16:
586 		size = MJUM16BYTES;
587 		zone = zone_jumbo16;
588 		break;
589 	default:
590 		panic("unknown cluster type");
591 		break;
592 	}
593 
594 	m->m_data = m->m_ext.ext_buf = cl;
595 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
596 	m->m_ext.ext_size = size;
597 	m->m_ext.ext_type = type;
598 	m->m_ext.ref_cnt = uma_find_refcnt(zone, cl);
599 	m->m_flags |= M_EXT;
600 
601 }
602 
603 static __inline void
604 m_chtype(struct mbuf *m, short new_type)
605 {
606 
607 	m->m_type = new_type;
608 }
609 
610 static __inline struct mbuf *
611 m_last(struct mbuf *m)
612 {
613 
614 	while (m->m_next)
615 		m = m->m_next;
616 	return (m);
617 }
618 
619 /*
620  * mbuf, cluster, and external object allocation macros (for compatibility
621  * purposes).
622  */
623 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
624 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
625 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
626 #define	MCLGET(m, how)		m_clget((m), (how))
627 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
628     m_extadd((m), (caddr_t)(buf), (size), (free),(arg1),(arg2),(flags), (type))
629 #define	m_getm(m, len, how, type)					\
630     m_getm2((m), (len), (how), (type), M_PKTHDR)
631 
632 /*
633  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
634  * be both the local data payload, or an external buffer area, depending on
635  * whether M_EXT is set).
636  */
637 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
638 			 (!(((m)->m_flags & M_EXT)) ||			\
639 			 (*((m)->m_ext.ref_cnt) == 1)) )		\
640 
641 /* Check if the supplied mbuf has a packet header, or else panic. */
642 #define	M_ASSERTPKTHDR(m)						\
643 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
644 	    ("%s: no mbuf packet header!", __func__))
645 
646 /*
647  * Ensure that the supplied mbuf is a valid, non-free mbuf.
648  *
649  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
650  */
651 #define	M_ASSERTVALID(m)						\
652 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
653 	    ("%s: attempted use of a free mbuf!", __func__))
654 
655 /*
656  * Set the m_data pointer of a newly-allocated mbuf (m_get/MGET) to place an
657  * object of the specified size at the end of the mbuf, longword aligned.
658  */
659 #define	M_ALIGN(m, len) do {						\
660 	KASSERT(!((m)->m_flags & (M_PKTHDR|M_EXT)),			\
661 		("%s: M_ALIGN not normal mbuf", __func__));		\
662 	KASSERT((m)->m_data == (m)->m_dat,				\
663 		("%s: M_ALIGN not a virgin mbuf", __func__));		\
664 	(m)->m_data += (MLEN - (len)) & ~(sizeof(long) - 1);		\
665 } while (0)
666 
667 /*
668  * As above, for mbufs allocated with m_gethdr/MGETHDR or initialized by
669  * M_DUP/MOVE_PKTHDR.
670  */
671 #define	MH_ALIGN(m, len) do {						\
672 	KASSERT((m)->m_flags & M_PKTHDR && !((m)->m_flags & M_EXT),	\
673 		("%s: MH_ALIGN not PKTHDR mbuf", __func__));		\
674 	KASSERT((m)->m_data == (m)->m_pktdat,				\
675 		("%s: MH_ALIGN not a virgin mbuf", __func__));		\
676 	(m)->m_data += (MHLEN - (len)) & ~(sizeof(long) - 1);		\
677 } while (0)
678 
679 /*
680  * Compute the amount of space available before the current start of data in
681  * an mbuf.
682  *
683  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
684  * of checking writability of the mbuf data area rests solely with the caller.
685  */
686 #define	M_LEADINGSPACE(m)						\
687 	((m)->m_flags & M_EXT ?						\
688 	    (M_WRITABLE(m) ? (m)->m_data - (m)->m_ext.ext_buf : 0):	\
689 	    (m)->m_flags & M_PKTHDR ? (m)->m_data - (m)->m_pktdat :	\
690 	    (m)->m_data - (m)->m_dat)
691 
692 /*
693  * Compute the amount of space available after the end of data in an mbuf.
694  *
695  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
696  * of checking writability of the mbuf data area rests solely with the caller.
697  */
698 #define	M_TRAILINGSPACE(m)						\
699 	((m)->m_flags & M_EXT ?						\
700 	    (M_WRITABLE(m) ? (m)->m_ext.ext_buf + (m)->m_ext.ext_size	\
701 		- ((m)->m_data + (m)->m_len) : 0) :			\
702 	    &(m)->m_dat[MLEN] - ((m)->m_data + (m)->m_len))
703 
704 /*
705  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
706  * allocated, how specifies whether to wait.  If the allocation fails, the
707  * original mbuf chain is freed and m is set to NULL.
708  */
709 #define	M_PREPEND(m, plen, how) do {					\
710 	struct mbuf **_mmp = &(m);					\
711 	struct mbuf *_mm = *_mmp;					\
712 	int _mplen = (plen);						\
713 	int __mhow = (how);						\
714 									\
715 	MBUF_CHECKSLEEP(how);						\
716 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
717 		_mm->m_data -= _mplen;					\
718 		_mm->m_len += _mplen;					\
719 	} else								\
720 		_mm = m_prepend(_mm, _mplen, __mhow);			\
721 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
722 		_mm->m_pkthdr.len += _mplen;				\
723 	*_mmp = _mm;							\
724 } while (0)
725 
726 /*
727  * Change mbuf to new type.  This is a relatively expensive operation and
728  * should be avoided.
729  */
730 #define	MCHTYPE(m, t)	m_chtype((m), (t))
731 
732 /* Length to m_copy to copy all. */
733 #define	M_COPYALL	1000000000
734 
735 /* Compatibility with 4.3. */
736 #define	m_copy(m, o, l)	m_copym((m), (o), (l), M_DONTWAIT)
737 
738 extern int		max_datalen;	/* MHLEN - max_hdr */
739 extern int		max_hdr;	/* Largest link + protocol header */
740 extern int		max_linkhdr;	/* Largest link-level header */
741 extern int		max_protohdr;	/* Largest protocol header */
742 extern struct mbstat	mbstat;		/* General mbuf stats/infos */
743 extern int		nmbclusters;	/* Maximum number of clusters */
744 
745 struct uio;
746 
747 void		 m_adj(struct mbuf *, int);
748 void		 m_align(struct mbuf *, int);
749 int		 m_apply(struct mbuf *, int, int,
750 		    int (*)(void *, void *, u_int), void *);
751 int		 m_append(struct mbuf *, int, c_caddr_t);
752 void		 m_cat(struct mbuf *, struct mbuf *);
753 void		 m_extadd(struct mbuf *, caddr_t, u_int,
754 		    void (*)(void *, void *), void *, void *, int, int);
755 struct mbuf	*m_collapse(struct mbuf *, int, int);
756 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
757 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
758 struct mbuf	*m_copym(struct mbuf *, int, int, int);
759 struct mbuf	*m_copymdata(struct mbuf *, struct mbuf *,
760 		    int, int, int, int);
761 struct mbuf	*m_copypacket(struct mbuf *, int);
762 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
763 struct mbuf	*m_copyup(struct mbuf *n, int len, int dstoff);
764 struct mbuf	*m_defrag(struct mbuf *, int);
765 void		 m_demote(struct mbuf *, int);
766 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
767 		    void (*)(char *, caddr_t, u_int));
768 struct mbuf	*m_dup(struct mbuf *, int);
769 int		 m_dup_pkthdr(struct mbuf *, struct mbuf *, int);
770 u_int		 m_fixhdr(struct mbuf *);
771 struct mbuf	*m_fragment(struct mbuf *, int, int);
772 void		 m_freem(struct mbuf *);
773 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
774 struct mbuf	*m_getptr(struct mbuf *, int, int *);
775 u_int		 m_length(struct mbuf *, struct mbuf **);
776 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
777 struct mbuf	*m_prepend(struct mbuf *, int, int);
778 void		 m_print(const struct mbuf *, int);
779 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
780 struct mbuf	*m_pullup(struct mbuf *, int);
781 int		m_sanity(struct mbuf *, int);
782 struct mbuf	*m_split(struct mbuf *, int, int);
783 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
784 struct mbuf	*m_unshare(struct mbuf *, int how);
785 
786 /*-
787  * Network packets may have annotations attached by affixing a list of
788  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
789  * allocated semi-opaque data structures that have a fixed header
790  * (struct m_tag) that specifies the size of the memory block and a
791  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
792  * unsigned value used to identify a module or ABI.  By convention this value
793  * is chosen as the date+time that the module is created, expressed as the
794  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
795  * value is an ABI/module-specific value that identifies a particular
796  * annotation and is private to the module.  For compatibility with systems
797  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
798  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
799  * compatibility shim functions and several tag types are defined below.
800  * Users that do not require compatibility should use a private cookie value
801  * so that packet tag-related definitions can be maintained privately.
802  *
803  * Note that the packet tag returned by m_tag_alloc has the default memory
804  * alignment implemented by malloc.  To reference private data one can use a
805  * construct like:
806  *
807  *	struct m_tag *mtag = m_tag_alloc(...);
808  *	struct foo *p = (struct foo *)(mtag+1);
809  *
810  * if the alignment of struct m_tag is sufficient for referencing members of
811  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
812  * private data structure to insure proper alignment; e.g.,
813  *
814  *	struct foo {
815  *		struct m_tag	tag;
816  *		...
817  *	};
818  *	struct foo *p = (struct foo *) m_tag_alloc(...);
819  *	struct m_tag *mtag = &p->tag;
820  */
821 
822 /*
823  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
824  * tags are expected to ``vanish'' when they pass through a network
825  * interface.  For most interfaces this happens normally as the tags are
826  * reclaimed when the mbuf is free'd.  However in some special cases
827  * reclaiming must be done manually.  An example is packets that pass through
828  * the loopback interface.  Also, one must be careful to do this when
829  * ``turning around'' packets (e.g., icmp_reflect).
830  *
831  * To mark a tag persistent bit-or this flag in when defining the tag id.
832  * The tag will then be treated as described above.
833  */
834 #define	MTAG_PERSISTENT				0x800
835 
836 #define	PACKET_TAG_NONE				0  /* Nadda */
837 
838 /* Packet tags for use with PACKET_ABI_COMPAT. */
839 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
840 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
841 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
842 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
843 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
844 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
845 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
846 #define	PACKET_TAG_GIF				8  /* GIF processing done */
847 #define	PACKET_TAG_GRE				9  /* GRE processing done */
848 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
849 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
850 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
851 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
852 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
853 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
854 #define	PACKET_TAG_DIVERT			17 /* divert info */
855 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
856 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
857 #define	PACKET_TAG_PF				21 /* PF + ALTQ information */
858 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
859 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
860 #define	PACKET_TAG_CARP                         28 /* CARP info */
861 
862 /* Specific cookies and tags. */
863 
864 /* Packet tag routines. */
865 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
866 void		 m_tag_delete(struct mbuf *, struct m_tag *);
867 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
868 void		 m_tag_free_default(struct m_tag *);
869 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
870 struct m_tag	*m_tag_copy(struct m_tag *, int);
871 int		 m_tag_copy_chain(struct mbuf *, struct mbuf *, int);
872 void		 m_tag_delete_nonpersistent(struct mbuf *);
873 
874 /*
875  * Initialize the list of tags associated with an mbuf.
876  */
877 static __inline void
878 m_tag_init(struct mbuf *m)
879 {
880 
881 	SLIST_INIT(&m->m_pkthdr.tags);
882 }
883 
884 /*
885  * Set up the contents of a tag.  Note that this does not fill in the free
886  * method; the caller is expected to do that.
887  *
888  * XXX probably should be called m_tag_init, but that was already taken.
889  */
890 static __inline void
891 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
892 {
893 
894 	t->m_tag_id = type;
895 	t->m_tag_len = len;
896 	t->m_tag_cookie = cookie;
897 }
898 
899 /*
900  * Reclaim resources associated with a tag.
901  */
902 static __inline void
903 m_tag_free(struct m_tag *t)
904 {
905 
906 	(*t->m_tag_free)(t);
907 }
908 
909 /*
910  * Return the first tag associated with an mbuf.
911  */
912 static __inline struct m_tag *
913 m_tag_first(struct mbuf *m)
914 {
915 
916 	return (SLIST_FIRST(&m->m_pkthdr.tags));
917 }
918 
919 /*
920  * Return the next tag in the list of tags associated with an mbuf.
921  */
922 static __inline struct m_tag *
923 m_tag_next(struct mbuf *m, struct m_tag *t)
924 {
925 
926 	return (SLIST_NEXT(t, m_tag_link));
927 }
928 
929 /*
930  * Prepend a tag to the list of tags associated with an mbuf.
931  */
932 static __inline void
933 m_tag_prepend(struct mbuf *m, struct m_tag *t)
934 {
935 
936 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
937 }
938 
939 /*
940  * Unlink a tag from the list of tags associated with an mbuf.
941  */
942 static __inline void
943 m_tag_unlink(struct mbuf *m, struct m_tag *t)
944 {
945 
946 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
947 }
948 
949 /* These are for OpenBSD compatibility. */
950 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
951 
952 static __inline struct m_tag *
953 m_tag_get(int type, int length, int wait)
954 {
955 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
956 }
957 
958 static __inline struct m_tag *
959 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
960 {
961 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
962 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
963 }
964 
965 /* XXX temporary FIB methods probably eventually use tags.*/
966 #define M_FIBSHIFT    28
967 #define M_FIBMASK	0x0F
968 
969 /* get the fib from an mbuf and if it is not set, return the default */
970 #define M_GETFIB(_m) \
971     ((((_m)->m_flags & M_FIB) >> M_FIBSHIFT) & M_FIBMASK)
972 
973 #define M_SETFIB(_m, _fib) do {						\
974 	_m->m_flags &= ~M_FIB;					   	\
975 	_m->m_flags |= (((_fib) << M_FIBSHIFT) & M_FIB);  \
976 } while (0)
977 
978 #endif /* _KERNEL */
979 
980 #ifdef MBUF_PROFILING
981  void m_profile(struct mbuf *m);
982  #define M_PROFILE(m) m_profile(m)
983 #else
984  #define M_PROFILE(m)
985 #endif
986 
987 
988 #endif /* !_SYS_MBUF_H_ */
989