xref: /freebsd/sys/sys/mbuf.h (revision 10b59a9b4add0320d52c15ce057dd697261e7dfc)
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  */
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 	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  */
137 struct 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  */
152 struct 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		/* will csum IP fragments */
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 
287 #define	CSUM_IP_CHECKED		0x0100		/* did csum IP */
288 #define	CSUM_IP_VALID		0x0200		/*   ... the csum is valid */
289 #define	CSUM_DATA_VALID		0x0400		/* csum_data field is valid */
290 #define	CSUM_PSEUDO_HDR		0x0800		/* csum_data has pseudo hdr */
291 #define	CSUM_SCTP_VALID		0x1000		/* SCTP checksum is valid */
292 
293 #define	CSUM_DELAY_DATA		(CSUM_TCP | CSUM_UDP)
294 #define	CSUM_DELAY_IP		(CSUM_IP)	/* XXX add ipv6 here too? */
295 
296 /*
297  * mbuf types.
298  */
299 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
300 #define	MT_DATA		1	/* dynamic (data) allocation */
301 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
302 #define	MT_SONAME	8	/* socket name */
303 #define	MT_CONTROL	14	/* extra-data protocol message */
304 #define	MT_OOBDATA	15	/* expedited data  */
305 #define	MT_NTYPES	16	/* number of mbuf types for mbtypes[] */
306 
307 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
308 				   a non-initialized mbuf */
309 
310 #define MB_NOTAGS	0x1UL	/* no tags attached to mbuf */
311 
312 /*
313  * General mbuf allocator statistics structure.
314  *
315  * Many of these statistics are no longer used; we instead track many
316  * allocator statistics through UMA's built in statistics mechanism.
317  */
318 struct mbstat {
319 	u_long	m_mbufs;	/* XXX */
320 	u_long	m_mclusts;	/* XXX */
321 
322 	u_long	m_drain;	/* times drained protocols for space */
323 	u_long	m_mcfail;	/* XXX: times m_copym failed */
324 	u_long	m_mpfail;	/* XXX: times m_pullup failed */
325 	u_long	m_msize;	/* length of an mbuf */
326 	u_long	m_mclbytes;	/* length of an mbuf cluster */
327 	u_long	m_minclsize;	/* min length of data to allocate a cluster */
328 	u_long	m_mlen;		/* length of data in an mbuf */
329 	u_long	m_mhlen;	/* length of data in a header mbuf */
330 
331 	/* Number of mbtypes (gives # elems in mbtypes[] array) */
332 	short	m_numtypes;
333 
334 	/* XXX: Sendfile stats should eventually move to their own struct */
335 	u_long	sf_iocnt;	/* times sendfile had to do disk I/O */
336 	u_long	sf_allocfail;	/* times sfbuf allocation failed */
337 	u_long	sf_allocwait;	/* times sfbuf allocation had to wait */
338 };
339 
340 /*
341  * Flags specifying how an allocation should be made.
342  *
343  * The flag to use is as follows:
344  * - M_DONTWAIT or M_NOWAIT from an interrupt handler to not block allocation.
345  * - M_WAIT or M_WAITOK from wherever it is safe to block.
346  *
347  * M_DONTWAIT/M_NOWAIT means that we will not block the thread explicitly and
348  * if we cannot allocate immediately we may return NULL, whereas
349  * M_WAIT/M_WAITOK means that if we cannot allocate resources we
350  * will block until they are available, and thus never return NULL.
351  *
352  * XXX Eventually just phase this out to use M_WAITOK/M_NOWAIT.
353  */
354 #define	MBTOM(how)	(how)
355 #define	M_DONTWAIT	M_NOWAIT
356 #define	M_TRYWAIT	M_WAITOK
357 #define	M_WAIT		M_WAITOK
358 
359 /*
360  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
361  * !_KERNEL so that monitoring tools can look up the zones with
362  * libmemstat(3).
363  */
364 #define	MBUF_MEM_NAME		"mbuf"
365 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
366 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
367 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
368 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
369 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
370 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
371 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
372 
373 #ifdef _KERNEL
374 
375 #ifdef WITNESS
376 #define	MBUF_CHECKSLEEP(how) do {					\
377 	if (how == M_WAITOK)						\
378 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
379 		    "Sleeping in \"%s\"", __func__);			\
380 } while (0)
381 #else
382 #define	MBUF_CHECKSLEEP(how)
383 #endif
384 
385 /*
386  * Network buffer allocation API
387  *
388  * The rest of it is defined in kern/kern_mbuf.c
389  */
390 
391 extern uma_zone_t	zone_mbuf;
392 extern uma_zone_t	zone_clust;
393 extern uma_zone_t	zone_pack;
394 extern uma_zone_t	zone_jumbop;
395 extern uma_zone_t	zone_jumbo9;
396 extern uma_zone_t	zone_jumbo16;
397 extern uma_zone_t	zone_ext_refcnt;
398 
399 static __inline struct mbuf	*m_getcl(int how, short type, int flags);
400 static __inline struct mbuf	*m_get(int how, short type);
401 static __inline struct mbuf	*m_gethdr(int how, short type);
402 static __inline struct mbuf	*m_getjcl(int how, short type, int flags,
403 				    int size);
404 static __inline struct mbuf	*m_getclr(int how, short type);	/* XXX */
405 static __inline int		 m_init(struct mbuf *m, uma_zone_t zone,
406 				    int size, int how, short type, int flags);
407 static __inline struct mbuf	*m_free(struct mbuf *m);
408 static __inline void		 m_clget(struct mbuf *m, int how);
409 static __inline void		*m_cljget(struct mbuf *m, int how, int size);
410 static __inline void		 m_chtype(struct mbuf *m, short new_type);
411 void				 mb_free_ext(struct mbuf *);
412 static __inline struct mbuf	*m_last(struct mbuf *m);
413 int				 m_pkthdr_init(struct mbuf *m, int how);
414 
415 static __inline int
416 m_gettype(int size)
417 {
418 	int type;
419 
420 	switch (size) {
421 	case MSIZE:
422 		type = EXT_MBUF;
423 		break;
424 	case MCLBYTES:
425 		type = EXT_CLUSTER;
426 		break;
427 #if MJUMPAGESIZE != MCLBYTES
428 	case MJUMPAGESIZE:
429 		type = EXT_JUMBOP;
430 		break;
431 #endif
432 	case MJUM9BYTES:
433 		type = EXT_JUMBO9;
434 		break;
435 	case MJUM16BYTES:
436 		type = EXT_JUMBO16;
437 		break;
438 	default:
439 		panic("%s: m_getjcl: invalid cluster size", __func__);
440 	}
441 
442 	return (type);
443 }
444 
445 static __inline uma_zone_t
446 m_getzone(int size)
447 {
448 	uma_zone_t zone;
449 
450 	switch (size) {
451 	case MSIZE:
452 		zone = zone_mbuf;
453 		break;
454 	case MCLBYTES:
455 		zone = zone_clust;
456 		break;
457 #if MJUMPAGESIZE != MCLBYTES
458 	case MJUMPAGESIZE:
459 		zone = zone_jumbop;
460 		break;
461 #endif
462 	case MJUM9BYTES:
463 		zone = zone_jumbo9;
464 		break;
465 	case MJUM16BYTES:
466 		zone = zone_jumbo16;
467 		break;
468 	default:
469 		panic("%s: m_getjcl: invalid cluster type", __func__);
470 	}
471 
472 	return (zone);
473 }
474 
475 /*
476  * Initialize an mbuf with linear storage.
477  *
478  * Inline because the consumer text overhead will be roughly the same to
479  * initialize or call a function with this many parameters and M_PKTHDR
480  * should go away with constant propagation for !MGETHDR.
481  */
482 static __inline int
483 m_init(struct mbuf *m, uma_zone_t zone, int size, int how, short type,
484     int flags)
485 {
486 	int error;
487 
488 	m->m_next = NULL;
489 	m->m_nextpkt = NULL;
490 	m->m_data = m->m_dat;
491 	m->m_len = 0;
492 	m->m_flags = flags;
493 	m->m_type = type;
494 	if (flags & M_PKTHDR) {
495 		if ((error = m_pkthdr_init(m, how)) != 0)
496 			return (error);
497 	}
498 
499 	return (0);
500 }
501 
502 static __inline struct mbuf *
503 m_get(int how, short type)
504 {
505 	struct mb_args args;
506 
507 	args.flags = 0;
508 	args.type = type;
509 	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
510 }
511 
512 /*
513  * XXX This should be deprecated, very little use.
514  */
515 static __inline struct mbuf *
516 m_getclr(int how, short type)
517 {
518 	struct mbuf *m;
519 	struct mb_args args;
520 
521 	args.flags = 0;
522 	args.type = type;
523 	m = uma_zalloc_arg(zone_mbuf, &args, how);
524 	if (m != NULL)
525 		bzero(m->m_data, MLEN);
526 	return (m);
527 }
528 
529 static __inline struct mbuf *
530 m_gethdr(int how, short type)
531 {
532 	struct mb_args args;
533 
534 	args.flags = M_PKTHDR;
535 	args.type = type;
536 	return ((struct mbuf *)(uma_zalloc_arg(zone_mbuf, &args, how)));
537 }
538 
539 static __inline struct mbuf *
540 m_getcl(int how, short type, int flags)
541 {
542 	struct mb_args args;
543 
544 	args.flags = flags;
545 	args.type = type;
546 	return ((struct mbuf *)(uma_zalloc_arg(zone_pack, &args, how)));
547 }
548 
549 /*
550  * m_getjcl() returns an mbuf with a cluster of the specified size attached.
551  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
552  *
553  * XXX: This is rather large, should be real function maybe.
554  */
555 static __inline struct mbuf *
556 m_getjcl(int how, short type, int flags, int size)
557 {
558 	struct mb_args args;
559 	struct mbuf *m, *n;
560 	uma_zone_t zone;
561 
562 	if (size == MCLBYTES)
563 		return m_getcl(how, type, flags);
564 
565 	args.flags = flags;
566 	args.type = type;
567 
568 	m = uma_zalloc_arg(zone_mbuf, &args, how);
569 	if (m == NULL)
570 		return (NULL);
571 
572 	zone = m_getzone(size);
573 	n = uma_zalloc_arg(zone, m, how);
574 	if (n == NULL) {
575 		uma_zfree(zone_mbuf, m);
576 		return (NULL);
577 	}
578 	return (m);
579 }
580 
581 static __inline void
582 m_free_fast(struct mbuf *m)
583 {
584 #ifdef INVARIANTS
585 	if (m->m_flags & M_PKTHDR)
586 		KASSERT(SLIST_EMPTY(&m->m_pkthdr.tags), ("doing fast free of mbuf with tags"));
587 #endif
588 
589 	uma_zfree_arg(zone_mbuf, m, (void *)MB_NOTAGS);
590 }
591 
592 static __inline struct mbuf *
593 m_free(struct mbuf *m)
594 {
595 	struct mbuf *n = m->m_next;
596 
597 	if (m->m_flags & M_EXT)
598 		mb_free_ext(m);
599 	else if ((m->m_flags & M_NOFREE) == 0)
600 		uma_zfree(zone_mbuf, m);
601 	return (n);
602 }
603 
604 static __inline void
605 m_clget(struct mbuf *m, int how)
606 {
607 
608 	if (m->m_flags & M_EXT)
609 		printf("%s: %p mbuf already has cluster\n", __func__, m);
610 	m->m_ext.ext_buf = (char *)NULL;
611 	uma_zalloc_arg(zone_clust, m, how);
612 	/*
613 	 * On a cluster allocation failure, drain the packet zone and retry,
614 	 * we might be able to loosen a few clusters up on the drain.
615 	 */
616 	if ((how & M_NOWAIT) && (m->m_ext.ext_buf == NULL)) {
617 		zone_drain(zone_pack);
618 		uma_zalloc_arg(zone_clust, m, how);
619 	}
620 }
621 
622 /*
623  * m_cljget() is different from m_clget() as it can allocate clusters without
624  * attaching them to an mbuf.  In that case the return value is the pointer
625  * to the cluster of the requested size.  If an mbuf was specified, it gets
626  * the cluster attached to it and the return value can be safely ignored.
627  * For size it takes MCLBYTES, MJUMPAGESIZE, MJUM9BYTES, MJUM16BYTES.
628  */
629 static __inline void *
630 m_cljget(struct mbuf *m, int how, int size)
631 {
632 	uma_zone_t zone;
633 
634 	if (m && m->m_flags & M_EXT)
635 		printf("%s: %p mbuf already has cluster\n", __func__, m);
636 	if (m != NULL)
637 		m->m_ext.ext_buf = NULL;
638 
639 	zone = m_getzone(size);
640 	return (uma_zalloc_arg(zone, m, how));
641 }
642 
643 static __inline void
644 m_cljset(struct mbuf *m, void *cl, int type)
645 {
646 	uma_zone_t zone;
647 	int size;
648 
649 	switch (type) {
650 	case EXT_CLUSTER:
651 		size = MCLBYTES;
652 		zone = zone_clust;
653 		break;
654 #if MJUMPAGESIZE != MCLBYTES
655 	case EXT_JUMBOP:
656 		size = MJUMPAGESIZE;
657 		zone = zone_jumbop;
658 		break;
659 #endif
660 	case EXT_JUMBO9:
661 		size = MJUM9BYTES;
662 		zone = zone_jumbo9;
663 		break;
664 	case EXT_JUMBO16:
665 		size = MJUM16BYTES;
666 		zone = zone_jumbo16;
667 		break;
668 	default:
669 		panic("unknown cluster type");
670 		break;
671 	}
672 
673 	m->m_data = m->m_ext.ext_buf = cl;
674 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
675 	m->m_ext.ext_size = size;
676 	m->m_ext.ext_type = type;
677 	m->m_ext.ref_cnt = uma_find_refcnt(zone, cl);
678 	m->m_flags |= M_EXT;
679 
680 }
681 
682 static __inline void
683 m_chtype(struct mbuf *m, short new_type)
684 {
685 
686 	m->m_type = new_type;
687 }
688 
689 static __inline struct mbuf *
690 m_last(struct mbuf *m)
691 {
692 
693 	while (m->m_next)
694 		m = m->m_next;
695 	return (m);
696 }
697 
698 extern void (*m_addr_chg_pf_p)(struct mbuf *m);
699 
700 static __inline void
701 m_addr_changed(struct mbuf *m)
702 {
703 
704 	if (m_addr_chg_pf_p)
705 		m_addr_chg_pf_p(m);
706 }
707 
708 /*
709  * mbuf, cluster, and external object allocation macros (for compatibility
710  * purposes).
711  */
712 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
713 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
714 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
715 #define	MCLGET(m, how)		m_clget((m), (how))
716 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
717     m_extadd((m), (caddr_t)(buf), (size), (free),(arg1),(arg2),(flags), (type))
718 #define	m_getm(m, len, how, type)					\
719     m_getm2((m), (len), (how), (type), M_PKTHDR)
720 
721 /*
722  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
723  * be both the local data payload, or an external buffer area, depending on
724  * whether M_EXT is set).
725  */
726 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
727 			 (!(((m)->m_flags & M_EXT)) ||			\
728 			 (*((m)->m_ext.ref_cnt) == 1)) )		\
729 
730 /* Check if the supplied mbuf has a packet header, or else panic. */
731 #define	M_ASSERTPKTHDR(m)						\
732 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
733 	    ("%s: no mbuf packet header!", __func__))
734 
735 /*
736  * Ensure that the supplied mbuf is a valid, non-free mbuf.
737  *
738  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
739  */
740 #define	M_ASSERTVALID(m)						\
741 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
742 	    ("%s: attempted use of a free mbuf!", __func__))
743 
744 /*
745  * Set the m_data pointer of a newly-allocated mbuf (m_get/MGET) to place an
746  * object of the specified size at the end of the mbuf, longword aligned.
747  */
748 #define	M_ALIGN(m, len) do {						\
749 	KASSERT(!((m)->m_flags & (M_PKTHDR|M_EXT)),			\
750 		("%s: M_ALIGN not normal mbuf", __func__));		\
751 	KASSERT((m)->m_data == (m)->m_dat,				\
752 		("%s: M_ALIGN not a virgin mbuf", __func__));		\
753 	(m)->m_data += (MLEN - (len)) & ~(sizeof(long) - 1);		\
754 } while (0)
755 
756 /*
757  * As above, for mbufs allocated with m_gethdr/MGETHDR or initialized by
758  * M_DUP/MOVE_PKTHDR.
759  */
760 #define	MH_ALIGN(m, len) do {						\
761 	KASSERT((m)->m_flags & M_PKTHDR && !((m)->m_flags & M_EXT),	\
762 		("%s: MH_ALIGN not PKTHDR mbuf", __func__));		\
763 	KASSERT((m)->m_data == (m)->m_pktdat,				\
764 		("%s: MH_ALIGN not a virgin mbuf", __func__));		\
765 	(m)->m_data += (MHLEN - (len)) & ~(sizeof(long) - 1);		\
766 } while (0)
767 
768 /*
769  * Compute the amount of space available before the current start of data in
770  * an mbuf.
771  *
772  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
773  * of checking writability of the mbuf data area rests solely with the caller.
774  */
775 #define	M_LEADINGSPACE(m)						\
776 	((m)->m_flags & M_EXT ?						\
777 	    (M_WRITABLE(m) ? (m)->m_data - (m)->m_ext.ext_buf : 0):	\
778 	    (m)->m_flags & M_PKTHDR ? (m)->m_data - (m)->m_pktdat :	\
779 	    (m)->m_data - (m)->m_dat)
780 
781 /*
782  * Compute the amount of space available after the end of data in an mbuf.
783  *
784  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
785  * of checking writability of the mbuf data area rests solely with the caller.
786  */
787 #define	M_TRAILINGSPACE(m)						\
788 	((m)->m_flags & M_EXT ?						\
789 	    (M_WRITABLE(m) ? (m)->m_ext.ext_buf + (m)->m_ext.ext_size	\
790 		- ((m)->m_data + (m)->m_len) : 0) :			\
791 	    &(m)->m_dat[MLEN] - ((m)->m_data + (m)->m_len))
792 
793 /*
794  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
795  * allocated, how specifies whether to wait.  If the allocation fails, the
796  * original mbuf chain is freed and m is set to NULL.
797  */
798 #define	M_PREPEND(m, plen, how) do {					\
799 	struct mbuf **_mmp = &(m);					\
800 	struct mbuf *_mm = *_mmp;					\
801 	int _mplen = (plen);						\
802 	int __mhow = (how);						\
803 									\
804 	MBUF_CHECKSLEEP(how);						\
805 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
806 		_mm->m_data -= _mplen;					\
807 		_mm->m_len += _mplen;					\
808 	} else								\
809 		_mm = m_prepend(_mm, _mplen, __mhow);			\
810 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
811 		_mm->m_pkthdr.len += _mplen;				\
812 	*_mmp = _mm;							\
813 } while (0)
814 
815 /*
816  * Change mbuf to new type.  This is a relatively expensive operation and
817  * should be avoided.
818  */
819 #define	MCHTYPE(m, t)	m_chtype((m), (t))
820 
821 /* Length to m_copy to copy all. */
822 #define	M_COPYALL	1000000000
823 
824 /* Compatibility with 4.3. */
825 #define	m_copy(m, o, l)	m_copym((m), (o), (l), M_DONTWAIT)
826 
827 extern int		max_datalen;	/* MHLEN - max_hdr */
828 extern int		max_hdr;	/* Largest link + protocol header */
829 extern int		max_linkhdr;	/* Largest link-level header */
830 extern int		max_protohdr;	/* Largest protocol header */
831 extern struct mbstat	mbstat;		/* General mbuf stats/infos */
832 extern int		nmbclusters;	/* Maximum number of clusters */
833 
834 struct uio;
835 
836 void		 m_adj(struct mbuf *, int);
837 void		 m_align(struct mbuf *, int);
838 int		 m_apply(struct mbuf *, int, int,
839 		    int (*)(void *, void *, u_int), void *);
840 int		 m_append(struct mbuf *, int, c_caddr_t);
841 void		 m_cat(struct mbuf *, struct mbuf *);
842 void		 m_extadd(struct mbuf *, caddr_t, u_int,
843 		    void (*)(void *, void *), void *, void *, int, int);
844 struct mbuf	*m_collapse(struct mbuf *, int, int);
845 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
846 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
847 struct mbuf	*m_copym(struct mbuf *, int, int, int);
848 struct mbuf	*m_copymdata(struct mbuf *, struct mbuf *,
849 		    int, int, int, int);
850 struct mbuf	*m_copypacket(struct mbuf *, int);
851 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
852 struct mbuf	*m_copyup(struct mbuf *n, int len, int dstoff);
853 struct mbuf	*m_defrag(struct mbuf *, int);
854 void		 m_demote(struct mbuf *, int);
855 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
856 		    void (*)(char *, caddr_t, u_int));
857 struct mbuf	*m_dup(struct mbuf *, int);
858 int		 m_dup_pkthdr(struct mbuf *, struct mbuf *, int);
859 u_int		 m_fixhdr(struct mbuf *);
860 struct mbuf	*m_fragment(struct mbuf *, int, int);
861 void		 m_freem(struct mbuf *);
862 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
863 struct mbuf	*m_getptr(struct mbuf *, int, int *);
864 u_int		 m_length(struct mbuf *, struct mbuf **);
865 int		 m_mbuftouio(struct uio *, struct mbuf *, int);
866 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
867 struct mbuf	*m_prepend(struct mbuf *, int, int);
868 void		 m_print(const struct mbuf *, int);
869 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
870 struct mbuf	*m_pullup(struct mbuf *, int);
871 int		m_sanity(struct mbuf *, int);
872 struct mbuf	*m_split(struct mbuf *, int, int);
873 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
874 struct mbuf	*m_unshare(struct mbuf *, int how);
875 
876 /*-
877  * Network packets may have annotations attached by affixing a list of
878  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
879  * allocated semi-opaque data structures that have a fixed header
880  * (struct m_tag) that specifies the size of the memory block and a
881  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
882  * unsigned value used to identify a module or ABI.  By convention this value
883  * is chosen as the date+time that the module is created, expressed as the
884  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
885  * value is an ABI/module-specific value that identifies a particular
886  * annotation and is private to the module.  For compatibility with systems
887  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
888  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
889  * compatibility shim functions and several tag types are defined below.
890  * Users that do not require compatibility should use a private cookie value
891  * so that packet tag-related definitions can be maintained privately.
892  *
893  * Note that the packet tag returned by m_tag_alloc has the default memory
894  * alignment implemented by malloc.  To reference private data one can use a
895  * construct like:
896  *
897  *	struct m_tag *mtag = m_tag_alloc(...);
898  *	struct foo *p = (struct foo *)(mtag+1);
899  *
900  * if the alignment of struct m_tag is sufficient for referencing members of
901  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
902  * private data structure to insure proper alignment; e.g.,
903  *
904  *	struct foo {
905  *		struct m_tag	tag;
906  *		...
907  *	};
908  *	struct foo *p = (struct foo *) m_tag_alloc(...);
909  *	struct m_tag *mtag = &p->tag;
910  */
911 
912 /*
913  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
914  * tags are expected to ``vanish'' when they pass through a network
915  * interface.  For most interfaces this happens normally as the tags are
916  * reclaimed when the mbuf is free'd.  However in some special cases
917  * reclaiming must be done manually.  An example is packets that pass through
918  * the loopback interface.  Also, one must be careful to do this when
919  * ``turning around'' packets (e.g., icmp_reflect).
920  *
921  * To mark a tag persistent bit-or this flag in when defining the tag id.
922  * The tag will then be treated as described above.
923  */
924 #define	MTAG_PERSISTENT				0x800
925 
926 #define	PACKET_TAG_NONE				0  /* Nadda */
927 
928 /* Packet tags for use with PACKET_ABI_COMPAT. */
929 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
930 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
931 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
932 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
933 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
934 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
935 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
936 #define	PACKET_TAG_GIF				8  /* GIF processing done */
937 #define	PACKET_TAG_GRE				9  /* GRE processing done */
938 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
939 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
940 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
941 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
942 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
943 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
944 #define	PACKET_TAG_DIVERT			17 /* divert info */
945 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
946 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
947 #define	PACKET_TAG_PF				21 /* PF + ALTQ information */
948 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
949 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
950 #define	PACKET_TAG_CARP				28 /* CARP info */
951 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
952 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
953 
954 /* Specific cookies and tags. */
955 
956 /* Packet tag routines. */
957 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
958 void		 m_tag_delete(struct mbuf *, struct m_tag *);
959 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
960 void		 m_tag_free_default(struct m_tag *);
961 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
962 struct m_tag	*m_tag_copy(struct m_tag *, int);
963 int		 m_tag_copy_chain(struct mbuf *, struct mbuf *, int);
964 void		 m_tag_delete_nonpersistent(struct mbuf *);
965 
966 /*
967  * Initialize the list of tags associated with an mbuf.
968  */
969 static __inline void
970 m_tag_init(struct mbuf *m)
971 {
972 
973 	SLIST_INIT(&m->m_pkthdr.tags);
974 }
975 
976 /*
977  * Set up the contents of a tag.  Note that this does not fill in the free
978  * method; the caller is expected to do that.
979  *
980  * XXX probably should be called m_tag_init, but that was already taken.
981  */
982 static __inline void
983 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
984 {
985 
986 	t->m_tag_id = type;
987 	t->m_tag_len = len;
988 	t->m_tag_cookie = cookie;
989 }
990 
991 /*
992  * Reclaim resources associated with a tag.
993  */
994 static __inline void
995 m_tag_free(struct m_tag *t)
996 {
997 
998 	(*t->m_tag_free)(t);
999 }
1000 
1001 /*
1002  * Return the first tag associated with an mbuf.
1003  */
1004 static __inline struct m_tag *
1005 m_tag_first(struct mbuf *m)
1006 {
1007 
1008 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1009 }
1010 
1011 /*
1012  * Return the next tag in the list of tags associated with an mbuf.
1013  */
1014 static __inline struct m_tag *
1015 m_tag_next(struct mbuf *m, struct m_tag *t)
1016 {
1017 
1018 	return (SLIST_NEXT(t, m_tag_link));
1019 }
1020 
1021 /*
1022  * Prepend a tag to the list of tags associated with an mbuf.
1023  */
1024 static __inline void
1025 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1026 {
1027 
1028 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1029 }
1030 
1031 /*
1032  * Unlink a tag from the list of tags associated with an mbuf.
1033  */
1034 static __inline void
1035 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1036 {
1037 
1038 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1039 }
1040 
1041 /* These are for OpenBSD compatibility. */
1042 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1043 
1044 static __inline struct m_tag *
1045 m_tag_get(int type, int length, int wait)
1046 {
1047 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1048 }
1049 
1050 static __inline struct m_tag *
1051 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1052 {
1053 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1054 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1055 }
1056 
1057 /* XXX temporary FIB methods probably eventually use tags.*/
1058 #define M_FIBSHIFT    28
1059 #define M_FIBMASK	0x0F
1060 
1061 /* get the fib from an mbuf and if it is not set, return the default */
1062 #define M_GETFIB(_m) \
1063     ((((_m)->m_flags & M_FIB) >> M_FIBSHIFT) & M_FIBMASK)
1064 
1065 #define M_SETFIB(_m, _fib) do {						\
1066 	_m->m_flags &= ~M_FIB;					   	\
1067 	_m->m_flags |= (((_fib) << M_FIBSHIFT) & M_FIB);  \
1068 } while (0)
1069 
1070 #endif /* _KERNEL */
1071 
1072 #ifdef MBUF_PROFILING
1073  void m_profile(struct mbuf *m);
1074  #define M_PROFILE(m) m_profile(m)
1075 #else
1076  #define M_PROFILE(m)
1077 #endif
1078 
1079 
1080 #endif /* !_SYS_MBUF_H_ */
1081