xref: /freebsd/sys/sys/mbuf.h (revision db612abe8df3355d1eb23bb3b50fdd97bc21e979)
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 #define	M_NOTIFICATION	M_PROTO5    /* SCTP notification */
197 
198 /*
199  * Flags to purge when crossing layers.
200  */
201 #define	M_PROTOFLAGS \
202     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8)
203 
204 /*
205  * Flags preserved when copying m_pkthdr.
206  */
207 #define	M_COPYFLAGS \
208     (M_PKTHDR|M_EOR|M_RDONLY|M_PROTOFLAGS|M_SKIP_FIREWALL|M_BCAST|M_MCAST|\
209      M_FRAG|M_FIRSTFRAG|M_LASTFRAG|M_VLANTAG|M_PROMISC)
210 
211 /*
212  * External buffer types: identify ext_buf type.
213  */
214 #define	EXT_CLUSTER	1	/* mbuf cluster */
215 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_bufs */
216 #define	EXT_JUMBOP	3	/* jumbo cluster 4096 bytes */
217 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
218 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
219 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
220 #define	EXT_MBUF	7	/* external mbuf reference (M_IOVEC) */
221 #define	EXT_NET_DRV	100	/* custom ext_buf provided by net driver(s) */
222 #define	EXT_MOD_TYPE	200	/* custom module's ext_buf type */
223 #define	EXT_DISPOSABLE	300	/* can throw this buffer away w/page flipping */
224 #define	EXT_EXTREF	400	/* has externally maintained ref_cnt ptr */
225 
226 /*
227  * Flags indicating hw checksum support and sw checksum requirements.  This
228  * field can be directly tested against if_data.ifi_hwassist.
229  */
230 #define	CSUM_IP			0x0001		/* will csum IP */
231 #define	CSUM_TCP		0x0002		/* will csum TCP */
232 #define	CSUM_UDP		0x0004		/* will csum UDP */
233 #define	CSUM_IP_FRAGS		0x0008		/* will csum IP fragments */
234 #define	CSUM_FRAGMENT		0x0010		/* will do IP fragmentation */
235 #define	CSUM_TSO		0x0020		/* will do TSO */
236 
237 #define	CSUM_IP_CHECKED		0x0100		/* did csum IP */
238 #define	CSUM_IP_VALID		0x0200		/*   ... the csum is valid */
239 #define	CSUM_DATA_VALID		0x0400		/* csum_data field is valid */
240 #define	CSUM_PSEUDO_HDR		0x0800		/* csum_data has pseudo hdr */
241 
242 #define	CSUM_DELAY_DATA		(CSUM_TCP | CSUM_UDP)
243 #define	CSUM_DELAY_IP		(CSUM_IP)	/* XXX add ipv6 here too? */
244 
245 /*
246  * mbuf types.
247  */
248 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
249 #define	MT_DATA		1	/* dynamic (data) allocation */
250 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
251 #define	MT_SONAME	8	/* socket name */
252 #define	MT_CONTROL	14	/* extra-data protocol message */
253 #define	MT_OOBDATA	15	/* expedited data  */
254 #define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
255 
256 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
257 				   a non-initialized mbuf */
258 
259 #define MB_NOTAGS	0x1UL	/* no tags attached to mbuf */
260 
261 /*
262  * General mbuf allocator statistics structure.
263  *
264  * Many of these statistics are no longer used; we instead track many
265  * allocator statistics through UMA's built in statistics mechanism.
266  */
267 struct mbstat {
268 	u_long	m_mbufs;	/* XXX */
269 	u_long	m_mclusts;	/* XXX */
270 
271 	u_long	m_drain;	/* times drained protocols for space */
272 	u_long	m_mcfail;	/* XXX: times m_copym failed */
273 	u_long	m_mpfail;	/* XXX: times m_pullup failed */
274 	u_long	m_msize;	/* length of an mbuf */
275 	u_long	m_mclbytes;	/* length of an mbuf cluster */
276 	u_long	m_minclsize;	/* min length of data to allocate a cluster */
277 	u_long	m_mlen;		/* length of data in an mbuf */
278 	u_long	m_mhlen;	/* length of data in a header mbuf */
279 
280 	/* Number of mbtypes (gives # elems in mbtypes[] array: */
281 	short	m_numtypes;
282 
283 	/* XXX: Sendfile stats should eventually move to their own struct */
284 	u_long	sf_iocnt;	/* times sendfile had to do disk I/O */
285 	u_long	sf_allocfail;	/* times sfbuf allocation failed */
286 	u_long	sf_allocwait;	/* times sfbuf allocation had to wait */
287 };
288 
289 /*
290  * Flags specifying how an allocation should be made.
291  *
292  * The flag to use is as follows:
293  * - M_DONTWAIT or M_NOWAIT from an interrupt handler to not block allocation.
294  * - M_WAIT or M_WAITOK from wherever it is safe to block.
295  *
296  * M_DONTWAIT/M_NOWAIT means that we will not block the thread explicitly and
297  * if we cannot allocate immediately we may return NULL, whereas
298  * M_WAIT/M_WAITOK means that if we cannot allocate resources we
299  * will block until they are available, and thus never return NULL.
300  *
301  * XXX Eventually just phase this out to use M_WAITOK/M_NOWAIT.
302  */
303 #define	MBTOM(how)	(how)
304 #define	M_DONTWAIT	M_NOWAIT
305 #define	M_TRYWAIT	M_WAITOK
306 #define	M_WAIT		M_WAITOK
307 
308 /*
309  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
310  * !_KERNEL so that monitoring tools can look up the zones with
311  * libmemstat(3).
312  */
313 #define	MBUF_MEM_NAME		"mbuf"
314 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
315 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
316 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
317 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
318 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
319 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
320 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
321 
322 #ifdef _KERNEL
323 
324 #ifdef WITNESS
325 #define	MBUF_CHECKSLEEP(how) do {					\
326 	if (how == M_WAITOK)						\
327 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
328 		    "Sleeping in \"%s\"", __func__);			\
329 } while (0)
330 #else
331 #define	MBUF_CHECKSLEEP(how)
332 #endif
333 
334 /*
335  * Network buffer allocation API
336  *
337  * The rest of it is defined in kern/kern_mbuf.c
338  */
339 
340 extern uma_zone_t	zone_mbuf;
341 extern uma_zone_t	zone_clust;
342 extern uma_zone_t	zone_pack;
343 extern uma_zone_t	zone_jumbop;
344 extern uma_zone_t	zone_jumbo9;
345 extern uma_zone_t	zone_jumbo16;
346 extern uma_zone_t	zone_ext_refcnt;
347 
348 static __inline struct mbuf	*m_getcl(int how, short type, int flags);
349 static __inline struct mbuf	*m_get(int how, short type);
350 static __inline struct mbuf	*m_gethdr(int how, short type);
351 static __inline struct mbuf	*m_getjcl(int how, short type, int flags,
352 				    int size);
353 static __inline struct mbuf	*m_getclr(int how, short type);	/* XXX */
354 static __inline struct mbuf	*m_free(struct mbuf *m);
355 static __inline void		 m_clget(struct mbuf *m, int how);
356 static __inline void		*m_cljget(struct mbuf *m, int how, int size);
357 static __inline void		 m_chtype(struct mbuf *m, short new_type);
358 void				 mb_free_ext(struct mbuf *);
359 static __inline struct mbuf	*m_last(struct mbuf *m);
360 
361 static __inline int
362 m_gettype(int size)
363 {
364 	int type;
365 
366 	switch (size) {
367 	case MSIZE:
368 		type = EXT_MBUF;
369 		break;
370 	case MCLBYTES:
371 		type = EXT_CLUSTER;
372 		break;
373 #if MJUMPAGESIZE != MCLBYTES
374 	case MJUMPAGESIZE:
375 		type = EXT_JUMBOP;
376 		break;
377 #endif
378 	case MJUM9BYTES:
379 		type = EXT_JUMBO9;
380 		break;
381 	case MJUM16BYTES:
382 		type = EXT_JUMBO16;
383 		break;
384 	default:
385 		panic("%s: m_getjcl: invalid cluster size", __func__);
386 	}
387 
388 	return (type);
389 }
390 
391 static __inline uma_zone_t
392 m_getzone(int size)
393 {
394 	uma_zone_t zone;
395 
396 	switch (size) {
397 	case MSIZE:
398 		zone = zone_mbuf;
399 		break;
400 	case MCLBYTES:
401 		zone = zone_clust;
402 		break;
403 #if MJUMPAGESIZE != MCLBYTES
404 	case MJUMPAGESIZE:
405 		zone = zone_jumbop;
406 		break;
407 #endif
408 	case MJUM9BYTES:
409 		zone = zone_jumbo9;
410 		break;
411 	case MJUM16BYTES:
412 		zone = zone_jumbo16;
413 		break;
414 	default:
415 		panic("%s: m_getjcl: invalid cluster type", __func__);
416 	}
417 
418 	return (zone);
419 }
420 
421 static __inline struct mbuf *
422 m_get(int how, short type)
423 {
424 	struct mb_args args;
425 
426 	args.flags = 0;
427 	args.type = type;
428 	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
429 }
430 
431 /*
432  * XXX This should be deprecated, very little use.
433  */
434 static __inline struct mbuf *
435 m_getclr(int how, short type)
436 {
437 	struct mbuf *m;
438 	struct mb_args args;
439 
440 	args.flags = 0;
441 	args.type = type;
442 	m = uma_zalloc_arg(zone_mbuf, &args, how);
443 	if (m != NULL)
444 		bzero(m->m_data, MLEN);
445 	return (m);
446 }
447 
448 static __inline struct mbuf *
449 m_gethdr(int how, short type)
450 {
451 	struct mb_args args;
452 
453 	args.flags = M_PKTHDR;
454 	args.type = type;
455 	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
456 }
457 
458 static __inline struct mbuf *
459 m_getcl(int how, short type, int flags)
460 {
461 	struct mb_args args;
462 
463 	args.flags = flags;
464 	args.type = type;
465 	return ((struct mbuf *)(uma_zalloc_arg(zone_pack, &args, how)));
466 }
467 
468 /*
469  * m_getjcl() returns an mbuf with a cluster of the specified size attached.
470  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
471  *
472  * XXX: This is rather large, should be real function maybe.
473  */
474 static __inline struct mbuf *
475 m_getjcl(int how, short type, int flags, int size)
476 {
477 	struct mb_args args;
478 	struct mbuf *m, *n;
479 	uma_zone_t zone;
480 
481 	args.flags = flags;
482 	args.type = type;
483 
484 	m = uma_zalloc_arg(zone_mbuf, &args, how);
485 	if (m == NULL)
486 		return (NULL);
487 
488 	zone = m_getzone(size);
489 	n = uma_zalloc_arg(zone, m, how);
490 	if (n == NULL) {
491 		uma_zfree(zone_mbuf, m);
492 		return (NULL);
493 	}
494 	return (m);
495 }
496 
497 static __inline void
498 m_free_fast(struct mbuf *m)
499 {
500 #ifdef INVARIANTS
501 	if (m->m_flags & M_PKTHDR)
502 		KASSERT(SLIST_EMPTY(&m->m_pkthdr.tags), ("doing fast free of mbuf with tags"));
503 #endif
504 
505 	uma_zfree_arg(zone_mbuf, m, (void *)MB_NOTAGS);
506 }
507 
508 static __inline struct mbuf *
509 m_free(struct mbuf *m)
510 {
511 	struct mbuf *n = m->m_next;
512 
513 	if (m->m_flags & M_EXT)
514 		mb_free_ext(m);
515 	else if ((m->m_flags & M_NOFREE) == 0)
516 		uma_zfree(zone_mbuf, m);
517 	return (n);
518 }
519 
520 static __inline void
521 m_clget(struct mbuf *m, int how)
522 {
523 
524 	if (m->m_flags & M_EXT)
525 		printf("%s: %p mbuf already has cluster\n", __func__, m);
526 	m->m_ext.ext_buf = (char *)NULL;
527 	uma_zalloc_arg(zone_clust, m, how);
528 	/*
529 	 * On a cluster allocation failure, drain the packet zone and retry,
530 	 * we might be able to loosen a few clusters up on the drain.
531 	 */
532 	if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) {
533 		zone_drain(zone_pack);
534 		uma_zalloc_arg(zone_clust, m, how);
535 	}
536 }
537 
538 /*
539  * m_cljget() is different from m_clget() as it can allocate clusters without
540  * attaching them to an mbuf.  In that case the return value is the pointer
541  * to the cluster of the requested size.  If an mbuf was specified, it gets
542  * the cluster attached to it and the return value can be safely ignored.
543  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
544  */
545 static __inline void *
546 m_cljget(struct mbuf *m, int how, int size)
547 {
548 	uma_zone_t zone;
549 
550 	if (m && m->m_flags & M_EXT)
551 		printf("%s: %p mbuf already has cluster\n", __func__, m);
552 	if (m != NULL)
553 		m->m_ext.ext_buf = NULL;
554 
555 	zone = m_getzone(size);
556 	return (uma_zalloc_arg(zone, m, how));
557 }
558 
559 static __inline void
560 m_cljset(struct mbuf *m, void *cl, int type)
561 {
562 	uma_zone_t zone;
563 	int size;
564 
565 	switch (type) {
566 	case EXT_CLUSTER:
567 		size = MCLBYTES;
568 		zone = zone_clust;
569 		break;
570 #if MJUMPAGESIZE != MCLBYTES
571 	case EXT_JUMBOP:
572 		size = MJUMPAGESIZE;
573 		zone = zone_jumbop;
574 		break;
575 #endif
576 	case EXT_JUMBO9:
577 		size = MJUM9BYTES;
578 		zone = zone_jumbo9;
579 		break;
580 	case EXT_JUMBO16:
581 		size = MJUM16BYTES;
582 		zone = zone_jumbo16;
583 		break;
584 	default:
585 		panic("unknown cluster type");
586 		break;
587 	}
588 
589 	m->m_data = m->m_ext.ext_buf = cl;
590 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
591 	m->m_ext.ext_size = size;
592 	m->m_ext.ext_type = type;
593 	m->m_ext.ref_cnt = uma_find_refcnt(zone, cl);
594 	m->m_flags |= M_EXT;
595 
596 }
597 
598 static __inline void
599 m_chtype(struct mbuf *m, short new_type)
600 {
601 
602 	m->m_type = new_type;
603 }
604 
605 static __inline struct mbuf *
606 m_last(struct mbuf *m)
607 {
608 
609 	while (m->m_next)
610 		m = m->m_next;
611 	return (m);
612 }
613 
614 /*
615  * mbuf, cluster, and external object allocation macros (for compatibility
616  * purposes).
617  */
618 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
619 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
620 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
621 #define	MCLGET(m, how)		m_clget((m), (how))
622 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
623     m_extadd((m), (caddr_t)(buf), (size), (free),(arg1),(arg2),(flags), (type))
624 #define	m_getm(m, len, how, type)					\
625     m_getm2((m), (len), (how), (type), M_PKTHDR)
626 
627 /*
628  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
629  * be both the local data payload, or an external buffer area, depending on
630  * whether M_EXT is set).
631  */
632 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
633 			 (!(((m)->m_flags & M_EXT)) ||			\
634 			 (*((m)->m_ext.ref_cnt) == 1)) )		\
635 
636 /* Check if the supplied mbuf has a packet header, or else panic. */
637 #define	M_ASSERTPKTHDR(m)						\
638 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
639 	    ("%s: no mbuf packet header!", __func__))
640 
641 /*
642  * Ensure that the supplied mbuf is a valid, non-free mbuf.
643  *
644  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
645  */
646 #define	M_ASSERTVALID(m)						\
647 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
648 	    ("%s: attempted use of a free mbuf!", __func__))
649 
650 /*
651  * Set the m_data pointer of a newly-allocated mbuf (m_get/MGET) to place an
652  * object of the specified size at the end of the mbuf, longword aligned.
653  */
654 #define	M_ALIGN(m, len) do {						\
655 	KASSERT(!((m)->m_flags & (M_PKTHDR|M_EXT)),			\
656 		("%s: M_ALIGN not normal mbuf", __func__));		\
657 	KASSERT((m)->m_data == (m)->m_dat,				\
658 		("%s: M_ALIGN not a virgin mbuf", __func__));		\
659 	(m)->m_data += (MLEN - (len)) & ~(sizeof(long) - 1);		\
660 } while (0)
661 
662 /*
663  * As above, for mbufs allocated with m_gethdr/MGETHDR or initialized by
664  * M_DUP/MOVE_PKTHDR.
665  */
666 #define	MH_ALIGN(m, len) do {						\
667 	KASSERT((m)->m_flags & M_PKTHDR && !((m)->m_flags & M_EXT),	\
668 		("%s: MH_ALIGN not PKTHDR mbuf", __func__));		\
669 	KASSERT((m)->m_data == (m)->m_pktdat,				\
670 		("%s: MH_ALIGN not a virgin mbuf", __func__));		\
671 	(m)->m_data += (MHLEN - (len)) & ~(sizeof(long) - 1);		\
672 } while (0)
673 
674 /*
675  * Compute the amount of space available before the current start of data in
676  * an mbuf.
677  *
678  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
679  * of checking writability of the mbuf data area rests solely with the caller.
680  */
681 #define	M_LEADINGSPACE(m)						\
682 	((m)->m_flags & M_EXT ?						\
683 	    (M_WRITABLE(m) ? (m)->m_data - (m)->m_ext.ext_buf : 0):	\
684 	    (m)->m_flags & M_PKTHDR ? (m)->m_data - (m)->m_pktdat :	\
685 	    (m)->m_data - (m)->m_dat)
686 
687 /*
688  * Compute the amount of space available after the end of data in an mbuf.
689  *
690  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
691  * of checking writability of the mbuf data area rests solely with the caller.
692  */
693 #define	M_TRAILINGSPACE(m)						\
694 	((m)->m_flags & M_EXT ?						\
695 	    (M_WRITABLE(m) ? (m)->m_ext.ext_buf + (m)->m_ext.ext_size	\
696 		- ((m)->m_data + (m)->m_len) : 0) :			\
697 	    &(m)->m_dat[MLEN] - ((m)->m_data + (m)->m_len))
698 
699 /*
700  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
701  * allocated, how specifies whether to wait.  If the allocation fails, the
702  * original mbuf chain is freed and m is set to NULL.
703  */
704 #define	M_PREPEND(m, plen, how) do {					\
705 	struct mbuf **_mmp = &(m);					\
706 	struct mbuf *_mm = *_mmp;					\
707 	int _mplen = (plen);						\
708 	int __mhow = (how);						\
709 									\
710 	MBUF_CHECKSLEEP(how);						\
711 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
712 		_mm->m_data -= _mplen;					\
713 		_mm->m_len += _mplen;					\
714 	} else								\
715 		_mm = m_prepend(_mm, _mplen, __mhow);			\
716 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
717 		_mm->m_pkthdr.len += _mplen;				\
718 	*_mmp = _mm;							\
719 } while (0)
720 
721 /*
722  * Change mbuf to new type.  This is a relatively expensive operation and
723  * should be avoided.
724  */
725 #define	MCHTYPE(m, t)	m_chtype((m), (t))
726 
727 /* Length to m_copy to copy all. */
728 #define	M_COPYALL	1000000000
729 
730 /* Compatibility with 4.3. */
731 #define	m_copy(m, o, l)	m_copym((m), (o), (l), M_DONTWAIT)
732 
733 extern int		max_datalen;	/* MHLEN - max_hdr */
734 extern int		max_hdr;	/* Largest link + protocol header */
735 extern int		max_linkhdr;	/* Largest link-level header */
736 extern int		max_protohdr;	/* Largest protocol header */
737 extern struct mbstat	mbstat;		/* General mbuf stats/infos */
738 extern int		nmbclusters;	/* Maximum number of clusters */
739 
740 struct uio;
741 
742 void		 m_adj(struct mbuf *, int);
743 void		 m_align(struct mbuf *, int);
744 int		 m_apply(struct mbuf *, int, int,
745 		    int (*)(void *, void *, u_int), void *);
746 int		 m_append(struct mbuf *, int, c_caddr_t);
747 void		 m_cat(struct mbuf *, struct mbuf *);
748 void		 m_extadd(struct mbuf *, caddr_t, u_int,
749 		    void (*)(void *, void *), void *, void *, int, int);
750 struct mbuf	*m_collapse(struct mbuf *, int, int);
751 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
752 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
753 struct mbuf	*m_copym(struct mbuf *, int, int, int);
754 struct mbuf	*m_copymdata(struct mbuf *, struct mbuf *,
755 		    int, int, int, int);
756 struct mbuf	*m_copypacket(struct mbuf *, int);
757 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
758 struct mbuf	*m_copyup(struct mbuf *n, int len, int dstoff);
759 struct mbuf	*m_defrag(struct mbuf *, int);
760 void		 m_demote(struct mbuf *, int);
761 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
762 		    void (*)(char *, caddr_t, u_int));
763 struct mbuf	*m_dup(struct mbuf *, int);
764 int		 m_dup_pkthdr(struct mbuf *, struct mbuf *, int);
765 u_int		 m_fixhdr(struct mbuf *);
766 struct mbuf	*m_fragment(struct mbuf *, int, int);
767 void		 m_freem(struct mbuf *);
768 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
769 struct mbuf	*m_getptr(struct mbuf *, int, int *);
770 u_int		 m_length(struct mbuf *, struct mbuf **);
771 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
772 struct mbuf	*m_prepend(struct mbuf *, int, int);
773 void		 m_print(const struct mbuf *, int);
774 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
775 struct mbuf	*m_pullup(struct mbuf *, int);
776 int		m_sanity(struct mbuf *, int);
777 struct mbuf	*m_split(struct mbuf *, int, int);
778 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
779 struct mbuf	*m_unshare(struct mbuf *, int how);
780 
781 /*-
782  * Network packets may have annotations attached by affixing a list of
783  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
784  * allocated semi-opaque data structures that have a fixed header
785  * (struct m_tag) that specifies the size of the memory block and a
786  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
787  * unsigned value used to identify a module or ABI.  By convention this value
788  * is chosen as the date+time that the module is created, expressed as the
789  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
790  * value is an ABI/module-specific value that identifies a particular
791  * annotation and is private to the module.  For compatibility with systems
792  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
793  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
794  * compatibility shim functions and several tag types are defined below.
795  * Users that do not require compatibility should use a private cookie value
796  * so that packet tag-related definitions can be maintained privately.
797  *
798  * Note that the packet tag returned by m_tag_alloc has the default memory
799  * alignment implemented by malloc.  To reference private data one can use a
800  * construct like:
801  *
802  *	struct m_tag *mtag = m_tag_alloc(...);
803  *	struct foo *p = (struct foo *)(mtag+1);
804  *
805  * if the alignment of struct m_tag is sufficient for referencing members of
806  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
807  * private data structure to insure proper alignment; e.g.,
808  *
809  *	struct foo {
810  *		struct m_tag	tag;
811  *		...
812  *	};
813  *	struct foo *p = (struct foo *) m_tag_alloc(...);
814  *	struct m_tag *mtag = &p->tag;
815  */
816 
817 /*
818  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
819  * tags are expected to ``vanish'' when they pass through a network
820  * interface.  For most interfaces this happens normally as the tags are
821  * reclaimed when the mbuf is free'd.  However in some special cases
822  * reclaiming must be done manually.  An example is packets that pass through
823  * the loopback interface.  Also, one must be careful to do this when
824  * ``turning around'' packets (e.g., icmp_reflect).
825  *
826  * To mark a tag persistent bit-or this flag in when defining the tag id.
827  * The tag will then be treated as described above.
828  */
829 #define	MTAG_PERSISTENT				0x800
830 
831 #define	PACKET_TAG_NONE				0  /* Nadda */
832 
833 /* Packet tags for use with PACKET_ABI_COMPAT. */
834 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
835 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
836 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
837 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
838 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
839 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
840 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
841 #define	PACKET_TAG_GIF				8  /* GIF processing done */
842 #define	PACKET_TAG_GRE				9  /* GRE processing done */
843 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
844 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
845 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
846 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
847 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
848 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
849 #define	PACKET_TAG_DIVERT			17 /* divert info */
850 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
851 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
852 #define	PACKET_TAG_PF				21 /* PF + ALTQ information */
853 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
854 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
855 #define	PACKET_TAG_CARP                         28 /* CARP info */
856 
857 /* Specific cookies and tags. */
858 
859 /* Packet tag routines. */
860 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
861 void		 m_tag_delete(struct mbuf *, struct m_tag *);
862 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
863 void		 m_tag_free_default(struct m_tag *);
864 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
865 struct m_tag	*m_tag_copy(struct m_tag *, int);
866 int		 m_tag_copy_chain(struct mbuf *, struct mbuf *, int);
867 void		 m_tag_delete_nonpersistent(struct mbuf *);
868 
869 /*
870  * Initialize the list of tags associated with an mbuf.
871  */
872 static __inline void
873 m_tag_init(struct mbuf *m)
874 {
875 
876 	SLIST_INIT(&m->m_pkthdr.tags);
877 }
878 
879 /*
880  * Set up the contents of a tag.  Note that this does not fill in the free
881  * method; the caller is expected to do that.
882  *
883  * XXX probably should be called m_tag_init, but that was already taken.
884  */
885 static __inline void
886 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
887 {
888 
889 	t->m_tag_id = type;
890 	t->m_tag_len = len;
891 	t->m_tag_cookie = cookie;
892 }
893 
894 /*
895  * Reclaim resources associated with a tag.
896  */
897 static __inline void
898 m_tag_free(struct m_tag *t)
899 {
900 
901 	(*t->m_tag_free)(t);
902 }
903 
904 /*
905  * Return the first tag associated with an mbuf.
906  */
907 static __inline struct m_tag *
908 m_tag_first(struct mbuf *m)
909 {
910 
911 	return (SLIST_FIRST(&m->m_pkthdr.tags));
912 }
913 
914 /*
915  * Return the next tag in the list of tags associated with an mbuf.
916  */
917 static __inline struct m_tag *
918 m_tag_next(struct mbuf *m, struct m_tag *t)
919 {
920 
921 	return (SLIST_NEXT(t, m_tag_link));
922 }
923 
924 /*
925  * Prepend a tag to the list of tags associated with an mbuf.
926  */
927 static __inline void
928 m_tag_prepend(struct mbuf *m, struct m_tag *t)
929 {
930 
931 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
932 }
933 
934 /*
935  * Unlink a tag from the list of tags associated with an mbuf.
936  */
937 static __inline void
938 m_tag_unlink(struct mbuf *m, struct m_tag *t)
939 {
940 
941 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
942 }
943 
944 /* These are for OpenBSD compatibility. */
945 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
946 
947 static __inline struct m_tag *
948 m_tag_get(int type, int length, int wait)
949 {
950 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
951 }
952 
953 static __inline struct m_tag *
954 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
955 {
956 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
957 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
958 }
959 
960 #endif /* _KERNEL */
961 
962 #endif /* !_SYS_MBUF_H_ */
963