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 * Macro for type conversion: convert mbuf pointer to data pointer of correct
64 * type:
65 *
66 * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
67 */
68#define	mtod(m, t)	((t)((m)->m_data))
69
70/*
71 * Argument structure passed to UMA routines during mbuf and packet
72 * allocations.
73 */
74struct 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 */
89struct 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 */
102struct 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 */
113struct 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	uint32_t	 flowid;	/* packet's 4-tuple system
119					 * flow identifier
120					 */
121	/* variables for hardware checksum */
122	int		 csum_flags;	/* flags regarding checksum */
123	int		 csum_data;	/* data field used by csum routines */
124	u_int16_t	 tso_segsz;	/* TSO segment size */
125	union {
126		u_int16_t vt_vtag;	/* Ethernet 802.1p+q vlan tag */
127		u_int16_t vt_nrecs;	/* # of IGMPv3 records in this chain */
128	} PH_vt;
129	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
130};
131#define ether_vtag	PH_vt.vt_vtag
132
133/*
134 * Description of external storage mapped into mbuf; valid only if M_EXT is
135 * set.
136 */
137struct m_ext {
138	caddr_t		 ext_buf;	/* start of buffer */
139	void		(*ext_free)	/* free routine if not the usual */
140			    (void *, void *);
141	void		*ext_arg1;	/* optional argument pointer */
142	void		*ext_arg2;	/* optional argument pointer */
143	u_int		 ext_size;	/* size of buffer, for ext_free */
144	volatile u_int	*ref_cnt;	/* pointer to ref count info */
145	int		 ext_type;	/* type of external storage */
146};
147
148/*
149 * The core of the mbuf object along with some shortcut defines for practical
150 * purposes.
151 */
152struct mbuf {
153	struct m_hdr	m_hdr;
154	union {
155		struct {
156			struct pkthdr	MH_pkthdr;	/* M_PKTHDR set */
157			union {
158				struct m_ext	MH_ext;	/* M_EXT set */
159				char		MH_databuf[MHLEN];
160			} MH_dat;
161		} MH;
162		char	M_databuf[MLEN];		/* !M_PKTHDR, !M_EXT */
163	} M_dat;
164};
165#define	m_next		m_hdr.mh_next
166#define	m_len		m_hdr.mh_len
167#define	m_data		m_hdr.mh_data
168#define	m_type		m_hdr.mh_type
169#define	m_flags		m_hdr.mh_flags
170#define	m_nextpkt	m_hdr.mh_nextpkt
171#define	m_act		m_nextpkt
172#define	m_pkthdr	M_dat.MH.MH_pkthdr
173#define	m_ext		M_dat.MH.MH_dat.MH_ext
174#define	m_pktdat	M_dat.MH.MH_dat.MH_databuf
175#define	m_dat		M_dat.M_databuf
176
177/*
178 * mbuf flags.
179 */
180#define	M_EXT		0x00000001 /* has associated external storage */
181#define	M_PKTHDR	0x00000002 /* start of record */
182#define	M_EOR		0x00000004 /* end of record */
183#define	M_RDONLY	0x00000008 /* associated data is marked read-only */
184#define	M_PROTO1	0x00000010 /* protocol-specific */
185#define	M_PROTO2	0x00000020 /* protocol-specific */
186#define	M_PROTO3	0x00000040 /* protocol-specific */
187#define	M_PROTO4	0x00000080 /* protocol-specific */
188#define	M_PROTO5	0x00000100 /* protocol-specific */
189#define	M_BCAST		0x00000200 /* send/received as link-level broadcast */
190#define	M_MCAST		0x00000400 /* send/received as link-level multicast */
191#define	M_FRAG		0x00000800 /* packet is a fragment of a larger packet */
192#define	M_FIRSTFRAG	0x00001000 /* packet is first fragment */
193#define	M_LASTFRAG	0x00002000 /* packet is last fragment */
194#define	M_SKIP_FIREWALL	0x00004000 /* skip firewall processing */
195#define	M_FREELIST	0x00008000 /* mbuf is on the free list */
196#define	M_VLANTAG	0x00010000 /* ether_vtag is valid */
197#define	M_PROMISC	0x00020000 /* packet was not for us */
198#define	M_NOFREE	0x00040000 /* do not free mbuf, embedded in cluster */
199#define	M_PROTO6	0x00080000 /* protocol-specific */
200#define	M_PROTO7	0x00100000 /* protocol-specific */
201#define	M_PROTO8	0x00200000 /* protocol-specific */
202#define	M_FLOWID	0x00400000 /* deprecated: flowid is valid */
203#define	M_HASHTYPEBITS	0x0F000000 /* mask of bits holding flowid hash type */
204
205/*
206 * For RELENG_{6,7} steal these flags for limited multiple routing table
207 * support. In RELENG_8 and beyond, use just one flag and a tag.
208 */
209#define	M_FIB		0xF0000000 /* steal some bits to store fib number. */
210
211#define	M_NOTIFICATION	M_PROTO5    /* SCTP notification */
212
213/*
214 * Flags to purge when crossing layers.
215 */
216#define	M_PROTOFLAGS \
217    (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8)
218
219/*
220 * Network interface cards are able to hash protocol fields (such as IPv4
221 * addresses and TCP port numbers) classify packets into flows.  These flows
222 * can then be used to maintain ordering while delivering packets to the OS
223 * via parallel input queues, as well as to provide a stateless affinity
224 * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
225 * m_flag fields to indicate how the hash should be interpreted by the
226 * network stack.
227 *
228 * Most NICs support RSS, which provides ordering and explicit affinity, and
229 * use the hash m_flag bits to indicate what header fields were covered by
230 * the hash.  M_HASHTYPE_OPAQUE can be set by non-RSS cards or configurations
231 * that provide an opaque flow identifier, allowing for ordering and
232 * distribution without explicit affinity.
233 */
234#define	M_HASHTYPE_SHIFT		24
235#define	M_HASHTYPE_NONE			0x0
236#define	M_HASHTYPE_RSS_IPV4		0x1	/* IPv4 2-tuple */
237#define	M_HASHTYPE_RSS_TCP_IPV4		0x2	/* TCPv4 4-tuple */
238#define	M_HASHTYPE_RSS_IPV6		0x3	/* IPv6 2-tuple */
239#define	M_HASHTYPE_RSS_TCP_IPV6		0x4	/* TCPv6 4-tuple */
240#define	M_HASHTYPE_RSS_IPV6_EX		0x5	/* IPv6 2-tuple + ext hdrs */
241#define	M_HASHTYPE_RSS_TCP_IPV6_EX	0x6	/* TCPv6 4-tiple + ext hdrs */
242#define	M_HASHTYPE_OPAQUE		0xf	/* ordering, not affinity */
243
244#define	M_HASHTYPE_CLEAR(m)	(m)->m_flags &= ~(M_HASHTYPEBITS)
245#define	M_HASHTYPE_GET(m)	(((m)->m_flags & M_HASHTYPEBITS) >> \
246				    M_HASHTYPE_SHIFT)
247#define	M_HASHTYPE_SET(m, v)	do {					\
248	(m)->m_flags &= ~M_HASHTYPEBITS;				\
249	(m)->m_flags |= ((v) << M_HASHTYPE_SHIFT);			\
250} while (0)
251#define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
252
253/*
254 * Flags preserved when copying m_pkthdr.
255 */
256#define	M_COPYFLAGS \
257    (M_PKTHDR|M_EOR|M_RDONLY|M_PROTOFLAGS|M_SKIP_FIREWALL|M_BCAST|M_MCAST|\
258     M_FRAG|M_FIRSTFRAG|M_LASTFRAG|M_VLANTAG|M_PROMISC|M_FIB|M_HASHTYPEBITS)
259
260/*
261 * External buffer types: identify ext_buf type.
262 */
263#define	EXT_CLUSTER	1	/* mbuf cluster */
264#define	EXT_SFBUF	2	/* sendfile(2)'s sf_bufs */
265#define	EXT_JUMBOP	3	/* jumbo cluster 4096 bytes */
266#define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
267#define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
268#define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
269#define	EXT_MBUF	7	/* external mbuf reference (M_IOVEC) */
270#define	EXT_NET_DRV	100	/* custom ext_buf provided by net driver(s) */
271#define	EXT_MOD_TYPE	200	/* custom module's ext_buf type */
272#define	EXT_DISPOSABLE	300	/* can throw this buffer away w/page flipping */
273#define	EXT_EXTREF	400	/* has externally maintained ref_cnt ptr */
274
275/*
276 * Flags indicating hw checksum support and sw checksum requirements.  This
277 * field can be directly tested against if_data.ifi_hwassist.
278 */
279#define	CSUM_IP			0x0001		/* will csum IP */
280#define	CSUM_TCP		0x0002		/* will csum TCP */
281#define	CSUM_UDP		0x0004		/* will csum UDP */
282#define	CSUM_IP_FRAGS		0x0008		/* removed, left for compat */
283#define	CSUM_FRAGMENT		0x0010		/* will do IP fragmentation */
284#define	CSUM_TSO		0x0020		/* will do TSO */
285#define	CSUM_SCTP		0x0040		/* will csum SCTP */
286#define CSUM_SCTP_IPV6		0x0080		/* will csum IPv6/SCTP */
287
288#define	CSUM_IP_CHECKED		0x0100		/* did csum IP */
289#define	CSUM_IP_VALID		0x0200		/*   ... the csum is valid */
290#define	CSUM_DATA_VALID		0x0400		/* csum_data field is valid */
291#define	CSUM_PSEUDO_HDR		0x0800		/* csum_data has pseudo hdr */
292#define	CSUM_SCTP_VALID		0x1000		/* SCTP checksum is valid */
293#define	CSUM_UDP_IPV6		0x2000		/* will csum IPv6/UDP */
294#define	CSUM_TCP_IPV6		0x4000		/* will csum IPv6/TCP */
295/*	CSUM_TSO_IPV6		0x8000		will do IPv6/TSO */
296
297/*	CSUM_FRAGMENT_IPV6	0x10000		will do IPv6 fragementation */
298
299#define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
300#define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
301
302#define	CSUM_DELAY_DATA		(CSUM_TCP | CSUM_UDP)
303#define	CSUM_DELAY_IP		(CSUM_IP)	/* Only v4, no v6 IP hdr csum */
304
305/*
306 * mbuf types.
307 */
308#define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
309#define	MT_DATA		1	/* dynamic (data) allocation */
310#define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
311#define	MT_SONAME	8	/* socket name */
312#define	MT_CONTROL	14	/* extra-data protocol message */
313#define	MT_OOBDATA	15	/* expedited data  */
314#define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
315
316#define	MT_NOINIT	255	/* Not a type but a flag to allocate
317				   a non-initialized mbuf */
318
319#define MB_NOTAGS	0x1UL	/* no tags attached to mbuf */
320
321/*
322 * General mbuf allocator statistics structure.
323 *
324 * Many of these statistics are no longer used; we instead track many
325 * allocator statistics through UMA's built in statistics mechanism.
326 */
327struct mbstat {
328	u_long	m_mbufs;	/* XXX */
329	u_long	m_mclusts;	/* XXX */
330
331	u_long	m_drain;	/* times drained protocols for space */
332	u_long	m_mcfail;	/* XXX: times m_copym failed */
333	u_long	m_mpfail;	/* XXX: times m_pullup failed */
334	u_long	m_msize;	/* length of an mbuf */
335	u_long	m_mclbytes;	/* length of an mbuf cluster */
336	u_long	m_minclsize;	/* min length of data to allocate a cluster */
337	u_long	m_mlen;		/* length of data in an mbuf */
338	u_long	m_mhlen;	/* length of data in a header mbuf */
339
340	/* Number of mbtypes (gives # elems in mbtypes[] array) */
341	short	m_numtypes;
342
343	/* XXX: Sendfile stats should eventually move to their own struct */
344	u_long	sf_iocnt;	/* times sendfile had to do disk I/O */
345	u_long	sf_allocfail;	/* times sfbuf allocation failed */
346	u_long	sf_allocwait;	/* times sfbuf allocation had to wait */
347};
348
349/*
350 * Flags specifying how an allocation should be made.
351 *
352 * The flag to use is as follows:
353 * - M_NOWAIT (M_DONTWAIT) from an interrupt handler to not block allocation.
354 * - M_WAITOK (M_WAIT) from wherever it is safe to block.
355 *
356 * M_DONTWAIT/M_NOWAIT means that we will not block the thread explicitly and
357 * if we cannot allocate immediately we may return NULL, whereas
358 * M_WAIT/M_WAITOK means that if we cannot allocate resources we
359 * will block until they are available, and thus never return NULL.
360 *
361 * XXX Eventually just phase this out to use M_WAITOK/M_NOWAIT.
362 */
363#define	MBTOM(how)	(how)
364#define	M_DONTWAIT	M_NOWAIT
365#define	M_TRYWAIT	M_WAITOK
366#define	M_WAIT		M_WAITOK
367
368/*
369 * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
370 * !_KERNEL so that monitoring tools can look up the zones with
371 * libmemstat(3).
372 */
373#define	MBUF_MEM_NAME		"mbuf"
374#define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
375#define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
376#define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
377#define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
378#define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
379#define	MBUF_TAG_MEM_NAME	"mbuf_tag"
380#define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
381
382#ifdef _KERNEL
383
384#ifdef WITNESS
385#define	MBUF_CHECKSLEEP(how) do {					\
386	if (how == M_WAITOK)						\
387		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
388		    "Sleeping in \"%s\"", __func__);			\
389} while (0)
390#else
391#define	MBUF_CHECKSLEEP(how)
392#endif
393
394/*
395 * Network buffer allocation API
396 *
397 * The rest of it is defined in kern/kern_mbuf.c
398 */
399extern uma_zone_t	zone_mbuf;
400extern uma_zone_t	zone_clust;
401extern uma_zone_t	zone_pack;
402extern uma_zone_t	zone_jumbop;
403extern uma_zone_t	zone_jumbo9;
404extern uma_zone_t	zone_jumbo16;
405extern uma_zone_t	zone_ext_refcnt;
406
407static __inline struct mbuf	*m_getcl(int how, short type, int flags);
408static __inline struct mbuf	*m_get(int how, short type);
409static __inline struct mbuf	*m_gethdr(int how, short type);
410static __inline struct mbuf	*m_getjcl(int how, short type, int flags,
411				    int size);
412static __inline struct mbuf	*m_getclr(int how, short type);	/* XXX */
413static __inline int		 m_init(struct mbuf *m, uma_zone_t zone,
414				    int size, int how, short type, int flags);
415static __inline struct mbuf	*m_free(struct mbuf *m);
416static __inline void		 m_clget(struct mbuf *m, int how);
417static __inline void		*m_cljget(struct mbuf *m, int how, int size);
418static __inline void		 m_chtype(struct mbuf *m, short new_type);
419void				 mb_free_ext(struct mbuf *);
420static __inline struct mbuf	*m_last(struct mbuf *m);
421int				 m_pkthdr_init(struct mbuf *m, int how);
422
423static __inline int
424m_gettype(int size)
425{
426	int type;
427
428	switch (size) {
429	case MSIZE:
430		type = EXT_MBUF;
431		break;
432	case MCLBYTES:
433		type = EXT_CLUSTER;
434		break;
435#if MJUMPAGESIZE != MCLBYTES
436	case MJUMPAGESIZE:
437		type = EXT_JUMBOP;
438		break;
439#endif
440	case MJUM9BYTES:
441		type = EXT_JUMBO9;
442		break;
443	case MJUM16BYTES:
444		type = EXT_JUMBO16;
445		break;
446	default:
447		panic("%s: m_getjcl: invalid cluster size", __func__);
448	}
449
450	return (type);
451}
452
453/*
454 * Associated an external reference counted buffer with an mbuf.
455 */
456static __inline void
457m_extaddref(struct mbuf *m, caddr_t buf, u_int size, u_int *ref_cnt,
458    void (*freef)(void *, void *), void *arg1, void *arg2)
459{
460
461	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
462
463	atomic_add_int(ref_cnt, 1);
464	m->m_flags |= M_EXT;
465	m->m_ext.ext_buf = buf;
466	m->m_ext.ref_cnt = ref_cnt;
467	m->m_data = m->m_ext.ext_buf;
468	m->m_ext.ext_size = size;
469	m->m_ext.ext_free = freef;
470	m->m_ext.ext_arg1 = arg1;
471	m->m_ext.ext_arg2 = arg2;
472	m->m_ext.ext_type = EXT_EXTREF;
473}
474
475static __inline uma_zone_t
476m_getzone(int size)
477{
478	uma_zone_t zone;
479
480	switch (size) {
481	case MSIZE:
482		zone = zone_mbuf;
483		break;
484	case MCLBYTES:
485		zone = zone_clust;
486		break;
487#if MJUMPAGESIZE != MCLBYTES
488	case MJUMPAGESIZE:
489		zone = zone_jumbop;
490		break;
491#endif
492	case MJUM9BYTES:
493		zone = zone_jumbo9;
494		break;
495	case MJUM16BYTES:
496		zone = zone_jumbo16;
497		break;
498	default:
499		panic("%s: m_getjcl: invalid cluster type", __func__);
500	}
501
502	return (zone);
503}
504
505/*
506 * Initialize an mbuf with linear storage.
507 *
508 * Inline because the consumer text overhead will be roughly the same to
509 * initialize or call a function with this many parameters and M_PKTHDR
510 * should go away with constant propagation for !MGETHDR.
511 */
512static __inline int
513m_init(struct mbuf *m, uma_zone_t zone, int size, int how, short type,
514    int flags)
515{
516	int error;
517
518	m->m_next = NULL;
519	m->m_nextpkt = NULL;
520	m->m_data = m->m_dat;
521	m->m_len = 0;
522	m->m_flags = flags;
523	m->m_type = type;
524	if (flags & M_PKTHDR) {
525		if ((error = m_pkthdr_init(m, how)) != 0)
526			return (error);
527	}
528
529	return (0);
530}
531
532static __inline struct mbuf *
533m_get(int how, short type)
534{
535	struct mb_args args;
536
537	args.flags = 0;
538	args.type = type;
539	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
540}
541
542/*
543 * XXX This should be deprecated, very little use.
544 */
545static __inline struct mbuf *
546m_getclr(int how, short type)
547{
548	struct mbuf *m;
549	struct mb_args args;
550
551	args.flags = 0;
552	args.type = type;
553	m = uma_zalloc_arg(zone_mbuf, &args, how);
554	if (m != NULL)
555		bzero(m->m_data, MLEN);
556	return (m);
557}
558
559static __inline struct mbuf *
560m_gethdr(int how, short type)
561{
562	struct mb_args args;
563
564	args.flags = M_PKTHDR;
565	args.type = type;
566	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
567}
568
569static __inline struct mbuf *
570m_getcl(int how, short type, int flags)
571{
572	struct mb_args args;
573
574	args.flags = flags;
575	args.type = type;
576	return ((struct mbuf *)(uma_zalloc_arg(zone_pack, &args, how)));
577}
578
579/*
580 * m_getjcl() returns an mbuf with a cluster of the specified size attached.
581 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
582 *
583 * XXX: This is rather large, should be real function maybe.
584 */
585static __inline struct mbuf *
586m_getjcl(int how, short type, int flags, int size)
587{
588	struct mb_args args;
589	struct mbuf *m, *n;
590	uma_zone_t zone;
591
592	if (size == MCLBYTES)
593		return m_getcl(how, type, flags);
594
595	args.flags = flags;
596	args.type = type;
597
598	m = uma_zalloc_arg(zone_mbuf, &args, how);
599	if (m == NULL)
600		return (NULL);
601
602	zone = m_getzone(size);
603	n = uma_zalloc_arg(zone, m, how);
604	if (n == NULL) {
605		uma_zfree(zone_mbuf, m);
606		return (NULL);
607	}
608	return (m);
609}
610
611static __inline void
612m_free_fast(struct mbuf *m)
613{
614#ifdef INVARIANTS
615	if (m->m_flags & M_PKTHDR)
616		KASSERT(SLIST_EMPTY(&m->m_pkthdr.tags), ("doing fast free of mbuf with tags"));
617#endif
618
619	uma_zfree_arg(zone_mbuf, m, (void *)MB_NOTAGS);
620}
621
622static __inline struct mbuf *
623m_free(struct mbuf *m)
624{
625	struct mbuf *n = m->m_next;
626
627	if (m->m_flags & M_EXT)
628		mb_free_ext(m);
629	else if ((m->m_flags & M_NOFREE) == 0)
630		uma_zfree(zone_mbuf, m);
631	return (n);
632}
633
634static __inline void
635m_clget(struct mbuf *m, int how)
636{
637
638	if (m->m_flags & M_EXT)
639		printf("%s: %p mbuf already has cluster\n", __func__, m);
640	m->m_ext.ext_buf = (char *)NULL;
641	uma_zalloc_arg(zone_clust, m, how);
642	/*
643	 * On a cluster allocation failure, drain the packet zone and retry,
644	 * we might be able to loosen a few clusters up on the drain.
645	 */
646	if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) {
647		zone_drain(zone_pack);
648		uma_zalloc_arg(zone_clust, m, how);
649	}
650}
651
652/*
653 * m_cljget() is different from m_clget() as it can allocate clusters without
654 * attaching them to an mbuf.  In that case the return value is the pointer
655 * to the cluster of the requested size.  If an mbuf was specified, it gets
656 * the cluster attached to it and the return value can be safely ignored.
657 * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
658 */
659static __inline void *
660m_cljget(struct mbuf *m, int how, int size)
661{
662	uma_zone_t zone;
663
664	if (m && m->m_flags & M_EXT)
665		printf("%s: %p mbuf already has cluster\n", __func__, m);
666	if (m != NULL)
667		m->m_ext.ext_buf = NULL;
668
669	zone = m_getzone(size);
670	return (uma_zalloc_arg(zone, m, how));
671}
672
673static __inline void
674m_cljset(struct mbuf *m, void *cl, int type)
675{
676	uma_zone_t zone;
677	int size;
678
679	switch (type) {
680	case EXT_CLUSTER:
681		size = MCLBYTES;
682		zone = zone_clust;
683		break;
684#if MJUMPAGESIZE != MCLBYTES
685	case EXT_JUMBOP:
686		size = MJUMPAGESIZE;
687		zone = zone_jumbop;
688		break;
689#endif
690	case EXT_JUMBO9:
691		size = MJUM9BYTES;
692		zone = zone_jumbo9;
693		break;
694	case EXT_JUMBO16:
695		size = MJUM16BYTES;
696		zone = zone_jumbo16;
697		break;
698	default:
699		panic("unknown cluster type");
700		break;
701	}
702
703	m->m_data = m->m_ext.ext_buf = cl;
704	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
705	m->m_ext.ext_size = size;
706	m->m_ext.ext_type = type;
707	m->m_ext.ref_cnt = uma_find_refcnt(zone, cl);
708	m->m_flags |= M_EXT;
709
710}
711
712static __inline void
713m_chtype(struct mbuf *m, short new_type)
714{
715
716	m->m_type = new_type;
717}
718
719static __inline struct mbuf *
720m_last(struct mbuf *m)
721{
722
723	while (m->m_next)
724		m = m->m_next;
725	return (m);
726}
727
728extern void (*m_addr_chg_pf_p)(struct mbuf *m);
729
730static __inline void
731m_addr_changed(struct mbuf *m)
732{
733
734	if (m_addr_chg_pf_p)
735		m_addr_chg_pf_p(m);
736}
737
738/*
739 * mbuf, cluster, and external object allocation macros (for compatibility
740 * purposes).
741 */
742#define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
743#define	MGET(m, how, type)	((m) = m_get((how), (type)))
744#define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
745#define	MCLGET(m, how)		m_clget((m), (how))
746#define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
747    m_extadd((m), (caddr_t)(buf), (size), (free),(arg1),(arg2),(flags), (type))
748#define	m_getm(m, len, how, type)					\
749    m_getm2((m), (len), (how), (type), M_PKTHDR)
750
751/*
752 * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
753 * be both the local data payload, or an external buffer area, depending on
754 * whether M_EXT is set).
755 */
756#define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
757			 (!(((m)->m_flags & M_EXT)) ||			\
758			 (*((m)->m_ext.ref_cnt) == 1)) )		\
759
760/* Check if the supplied mbuf has a packet header, or else panic. */
761#define	M_ASSERTPKTHDR(m)						\
762	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
763	    ("%s: no mbuf packet header!", __func__))
764
765/*
766 * Ensure that the supplied mbuf is a valid, non-free mbuf.
767 *
768 * XXX: Broken at the moment.  Need some UMA magic to make it work again.
769 */
770#define	M_ASSERTVALID(m)						\
771	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
772	    ("%s: attempted use of a free mbuf!", __func__))
773
774/*
775 * Set the m_data pointer of a newly-allocated mbuf (m_get/MGET) to place an
776 * object of the specified size at the end of the mbuf, longword aligned.
777 */
778#define	M_ALIGN(m, len) do {						\
779	KASSERT(!((m)->m_flags & (M_PKTHDR|M_EXT)),			\
780		("%s: M_ALIGN not normal mbuf", __func__));		\
781	KASSERT((m)->m_data == (m)->m_dat,				\
782		("%s: M_ALIGN not a virgin mbuf", __func__));		\
783	(m)->m_data += (MLEN - (len)) & ~(sizeof(long) - 1);		\
784} while (0)
785
786/*
787 * As above, for mbufs allocated with m_gethdr/MGETHDR or initialized by
788 * M_DUP/MOVE_PKTHDR.
789 */
790#define	MH_ALIGN(m, len) do {						\
791	KASSERT((m)->m_flags & M_PKTHDR && !((m)->m_flags & M_EXT),	\
792		("%s: MH_ALIGN not PKTHDR mbuf", __func__));		\
793	KASSERT((m)->m_data == (m)->m_pktdat,				\
794		("%s: MH_ALIGN not a virgin mbuf", __func__));		\
795	(m)->m_data += (MHLEN - (len)) & ~(sizeof(long) - 1);		\
796} while (0)
797
798/*
799 * Compute the amount of space available before the current start of data in
800 * an mbuf.
801 *
802 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
803 * of checking writability of the mbuf data area rests solely with the caller.
804 */
805#define	M_LEADINGSPACE(m)						\
806	((m)->m_flags & M_EXT ?						\
807	    (M_WRITABLE(m) ? (m)->m_data - (m)->m_ext.ext_buf : 0):	\
808	    (m)->m_flags & M_PKTHDR ? (m)->m_data - (m)->m_pktdat :	\
809	    (m)->m_data - (m)->m_dat)
810
811/*
812 * Compute the amount of space available after the end of data in an mbuf.
813 *
814 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
815 * of checking writability of the mbuf data area rests solely with the caller.
816 */
817#define	M_TRAILINGSPACE(m)						\
818	((m)->m_flags & M_EXT ?						\
819	    (M_WRITABLE(m) ? (m)->m_ext.ext_buf + (m)->m_ext.ext_size	\
820		- ((m)->m_data + (m)->m_len) : 0) :			\
821	    &(m)->m_dat[MLEN] - ((m)->m_data + (m)->m_len))
822
823/*
824 * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
825 * allocated, how specifies whether to wait.  If the allocation fails, the
826 * original mbuf chain is freed and m is set to NULL.
827 */
828#define	M_PREPEND(m, plen, how) do {					\
829	struct mbuf **_mmp = &(m);					\
830	struct mbuf *_mm = *_mmp;					\
831	int _mplen = (plen);						\
832	int __mhow = (how);						\
833									\
834	MBUF_CHECKSLEEP(how);						\
835	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
836		_mm->m_data -= _mplen;					\
837		_mm->m_len += _mplen;					\
838	} else								\
839		_mm = m_prepend(_mm, _mplen, __mhow);			\
840	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
841		_mm->m_pkthdr.len += _mplen;				\
842	*_mmp = _mm;							\
843} while (0)
844
845/*
846 * Change mbuf to new type.  This is a relatively expensive operation and
847 * should be avoided.
848 */
849#define	MCHTYPE(m, t)	m_chtype((m), (t))
850
851/* Length to m_copy to copy all. */
852#define	M_COPYALL	1000000000
853
854/* Compatibility with 4.3. */
855#define	m_copy(m, o, l)	m_copym((m), (o), (l), M_DONTWAIT)
856
857extern int		max_datalen;	/* MHLEN - max_hdr */
858extern int		max_hdr;	/* Largest link + protocol header */
859extern int		max_linkhdr;	/* Largest link-level header */
860extern int		max_protohdr;	/* Largest protocol header */
861extern struct mbstat	mbstat;		/* General mbuf stats/infos */
862extern int		nmbclusters;	/* Maximum number of clusters */
863
864struct uio;
865
866void		 m_adj(struct mbuf *, int);
867void		 m_align(struct mbuf *, int);
868int		 m_apply(struct mbuf *, int, int,
869		    int (*)(void *, void *, u_int), void *);
870int		 m_append(struct mbuf *, int, c_caddr_t);
871void		 m_cat(struct mbuf *, struct mbuf *);
872void		 m_extadd(struct mbuf *, caddr_t, u_int,
873		    void (*)(void *, void *), void *, void *, int, int);
874struct mbuf	*m_collapse(struct mbuf *, int, int);
875void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
876void		 m_copydata(const struct mbuf *, int, int, caddr_t);
877struct mbuf	*m_copym(struct mbuf *, int, int, int);
878struct mbuf	*m_copymdata(struct mbuf *, struct mbuf *,
879		    int, int, int, int);
880struct mbuf	*m_copypacket(struct mbuf *, int);
881void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
882struct mbuf	*m_copyup(struct mbuf *n, int len, int dstoff);
883struct mbuf	*m_defrag(struct mbuf *, int);
884void		 m_demote(struct mbuf *, int);
885struct mbuf	*m_devget(char *, int, int, struct ifnet *,
886		    void (*)(char *, caddr_t, u_int));
887struct mbuf	*m_dup(struct mbuf *, int);
888int		 m_dup_pkthdr(struct mbuf *, struct mbuf *, int);
889u_int		 m_fixhdr(struct mbuf *);
890struct mbuf	*m_fragment(struct mbuf *, int, int);
891void		 m_freem(struct mbuf *);
892struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
893struct mbuf	*m_getptr(struct mbuf *, int, int *);
894u_int		 m_length(struct mbuf *, struct mbuf **);
895int		 m_mbuftouio(struct uio *, struct mbuf *, int);
896void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
897struct mbuf	*m_prepend(struct mbuf *, int, int);
898void		 m_print(const struct mbuf *, int);
899struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
900struct mbuf	*m_pullup(struct mbuf *, int);
901int		m_sanity(struct mbuf *, int);
902struct mbuf	*m_split(struct mbuf *, int, int);
903struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
904struct mbuf	*m_unshare(struct mbuf *, int how);
905
906/*-
907 * Network packets may have annotations attached by affixing a list of
908 * "packet tags" to the pkthdr structure.  Packet tags are dynamically
909 * allocated semi-opaque data structures that have a fixed header
910 * (struct m_tag) that specifies the size of the memory block and a
911 * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
912 * unsigned value used to identify a module or ABI.  By convention this value
913 * is chosen as the date+time that the module is created, expressed as the
914 * number of seconds since the epoch (e.g., using date -u +'%s').  The type
915 * value is an ABI/module-specific value that identifies a particular
916 * annotation and is private to the module.  For compatibility with systems
917 * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
918 * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
919 * compatibility shim functions and several tag types are defined below.
920 * Users that do not require compatibility should use a private cookie value
921 * so that packet tag-related definitions can be maintained privately.
922 *
923 * Note that the packet tag returned by m_tag_alloc has the default memory
924 * alignment implemented by malloc.  To reference private data one can use a
925 * construct like:
926 *
927 *	struct m_tag *mtag = m_tag_alloc(...);
928 *	struct foo *p = (struct foo *)(mtag+1);
929 *
930 * if the alignment of struct m_tag is sufficient for referencing members of
931 * struct foo.  Otherwise it is necessary to embed struct m_tag within the
932 * private data structure to insure proper alignment; e.g.,
933 *
934 *	struct foo {
935 *		struct m_tag	tag;
936 *		...
937 *	};
938 *	struct foo *p = (struct foo *) m_tag_alloc(...);
939 *	struct m_tag *mtag = &p->tag;
940 */
941
942/*
943 * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
944 * tags are expected to ``vanish'' when they pass through a network
945 * interface.  For most interfaces this happens normally as the tags are
946 * reclaimed when the mbuf is free'd.  However in some special cases
947 * reclaiming must be done manually.  An example is packets that pass through
948 * the loopback interface.  Also, one must be careful to do this when
949 * ``turning around'' packets (e.g., icmp_reflect).
950 *
951 * To mark a tag persistent bit-or this flag in when defining the tag id.
952 * The tag will then be treated as described above.
953 */
954#define	MTAG_PERSISTENT				0x800
955
956#define	PACKET_TAG_NONE				0  /* Nadda */
957
958/* Packet tags for use with PACKET_ABI_COMPAT. */
959#define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
960#define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
961#define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
962#define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
963#define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
964#define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
965#define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
966#define	PACKET_TAG_GIF				8  /* GIF processing done */
967#define	PACKET_TAG_GRE				9  /* GRE processing done */
968#define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
969#define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
970#define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
971#define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
972#define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
973#define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
974#define	PACKET_TAG_DIVERT			17 /* divert info */
975#define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
976#define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
977#define	PACKET_TAG_PF				21 /* PF + ALTQ information */
978#define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
979#define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
980#define	PACKET_TAG_CARP				28 /* CARP info */
981#define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
982#define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
983
984/* Specific cookies and tags. */
985
986/* Packet tag routines. */
987struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
988void		 m_tag_delete(struct mbuf *, struct m_tag *);
989void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
990void		 m_tag_free_default(struct m_tag *);
991struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
992struct m_tag	*m_tag_copy(struct m_tag *, int);
993int		 m_tag_copy_chain(struct mbuf *, struct mbuf *, int);
994void		 m_tag_delete_nonpersistent(struct mbuf *);
995
996/*
997 * Initialize the list of tags associated with an mbuf.
998 */
999static __inline void
1000m_tag_init(struct mbuf *m)
1001{
1002
1003	SLIST_INIT(&m->m_pkthdr.tags);
1004}
1005
1006/*
1007 * Set up the contents of a tag.  Note that this does not fill in the free
1008 * method; the caller is expected to do that.
1009 *
1010 * XXX probably should be called m_tag_init, but that was already taken.
1011 */
1012static __inline void
1013m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1014{
1015
1016	t->m_tag_id = type;
1017	t->m_tag_len = len;
1018	t->m_tag_cookie = cookie;
1019}
1020
1021/*
1022 * Reclaim resources associated with a tag.
1023 */
1024static __inline void
1025m_tag_free(struct m_tag *t)
1026{
1027
1028	(*t->m_tag_free)(t);
1029}
1030
1031/*
1032 * Return the first tag associated with an mbuf.
1033 */
1034static __inline struct m_tag *
1035m_tag_first(struct mbuf *m)
1036{
1037
1038	return (SLIST_FIRST(&m->m_pkthdr.tags));
1039}
1040
1041/*
1042 * Return the next tag in the list of tags associated with an mbuf.
1043 */
1044static __inline struct m_tag *
1045m_tag_next(struct mbuf *m, struct m_tag *t)
1046{
1047
1048	return (SLIST_NEXT(t, m_tag_link));
1049}
1050
1051/*
1052 * Prepend a tag to the list of tags associated with an mbuf.
1053 */
1054static __inline void
1055m_tag_prepend(struct mbuf *m, struct m_tag *t)
1056{
1057
1058	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1059}
1060
1061/*
1062 * Unlink a tag from the list of tags associated with an mbuf.
1063 */
1064static __inline void
1065m_tag_unlink(struct mbuf *m, struct m_tag *t)
1066{
1067
1068	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1069}
1070
1071/* These are for OpenBSD compatibility. */
1072#define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1073
1074static __inline struct m_tag *
1075m_tag_get(int type, int length, int wait)
1076{
1077	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1078}
1079
1080static __inline struct m_tag *
1081m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1082{
1083	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1084	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1085}
1086
1087/* XXX temporary FIB methods probably eventually use tags.*/
1088#define M_FIBSHIFT    28
1089#define M_FIBMASK	0x0F
1090
1091/* get the fib from an mbuf and if it is not set, return the default */
1092#define M_GETFIB(_m) \
1093    ((((_m)->m_flags & M_FIB) >> M_FIBSHIFT) & M_FIBMASK)
1094
1095#define M_SETFIB(_m, _fib) do {						\
1096	_m->m_flags &= ~M_FIB;					   	\
1097	_m->m_flags |= (((_fib) << M_FIBSHIFT) & M_FIB);  \
1098} while (0)
1099
1100#endif /* _KERNEL */
1101
1102#ifdef MBUF_PROFILING
1103 void m_profile(struct mbuf *m);
1104 #define M_PROFILE(m) m_profile(m)
1105#else
1106 #define M_PROFILE(m)
1107#endif
1108
1109
1110#endif /* !_SYS_MBUF_H_ */
1111