xref: /freebsd/sys/sys/mbuf.h (revision 562894f0dc310f658284863ff329906e7737a0a0)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1993
5  *	The Regents of the University of California.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)mbuf.h	8.5 (Berkeley) 2/19/95
33  * $FreeBSD$
34  */
35 
36 #ifndef _SYS_MBUF_H_
37 #define	_SYS_MBUF_H_
38 
39 /* XXX: These includes suck. Sorry! */
40 #include <sys/queue.h>
41 #ifdef _KERNEL
42 #include <sys/systm.h>
43 #include <sys/refcount.h>
44 #include <vm/uma.h>
45 #ifdef WITNESS
46 #include <sys/lock.h>
47 #endif
48 #endif
49 
50 #ifdef _KERNEL
51 #include <sys/sdt.h>
52 
53 #define	MBUF_PROBE1(probe, arg0)					\
54 	SDT_PROBE1(sdt, , , probe, arg0)
55 #define	MBUF_PROBE2(probe, arg0, arg1)					\
56 	SDT_PROBE2(sdt, , , probe, arg0, arg1)
57 #define	MBUF_PROBE3(probe, arg0, arg1, arg2)				\
58 	SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
59 #define	MBUF_PROBE4(probe, arg0, arg1, arg2, arg3)			\
60 	SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
61 #define	MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4)		\
62 	SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
63 
64 SDT_PROBE_DECLARE(sdt, , , m__init);
65 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
66 SDT_PROBE_DECLARE(sdt, , , m__get);
67 SDT_PROBE_DECLARE(sdt, , , m__getcl);
68 SDT_PROBE_DECLARE(sdt, , , m__clget);
69 SDT_PROBE_DECLARE(sdt, , , m__cljget);
70 SDT_PROBE_DECLARE(sdt, , , m__cljset);
71 SDT_PROBE_DECLARE(sdt, , , m__free);
72 SDT_PROBE_DECLARE(sdt, , , m__freem);
73 
74 #endif /* _KERNEL */
75 
76 /*
77  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
78  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
79  * sys/param.h), which has no additional overhead and is used instead of the
80  * internal data area; this is done when at least MINCLSIZE of data must be
81  * stored.  Additionally, it is possible to allocate a separate buffer
82  * externally and attach it to the mbuf in a way similar to that of mbuf
83  * clusters.
84  *
85  * NB: These calculation do not take actual compiler-induced alignment and
86  * padding inside the complete struct mbuf into account.  Appropriate
87  * attention is required when changing members of struct mbuf.
88  *
89  * MLEN is data length in a normal mbuf.
90  * MHLEN is data length in an mbuf with pktheader.
91  * MINCLSIZE is a smallest amount of data that should be put into cluster.
92  *
93  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
94  * they are sensible.
95  */
96 struct mbuf;
97 #define	MHSIZE		offsetof(struct mbuf, m_dat)
98 #define	MPKTHSIZE	offsetof(struct mbuf, m_pktdat)
99 #define	MLEN		((int)(MSIZE - MHSIZE))
100 #define	MHLEN		((int)(MSIZE - MPKTHSIZE))
101 #define	MINCLSIZE	(MHLEN + 1)
102 #define	M_NODOM		255
103 
104 #ifdef _KERNEL
105 /*-
106  * Macro for type conversion: convert mbuf pointer to data pointer of correct
107  * type:
108  *
109  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
110  * mtodo(m, o) -- Same as above but with offset 'o' into data.
111  */
112 #define	mtod(m, t)	((t)((m)->m_data))
113 #define	mtodo(m, o)	((void *)(((m)->m_data) + (o)))
114 
115 /*
116  * Argument structure passed to UMA routines during mbuf and packet
117  * allocations.
118  */
119 struct mb_args {
120 	int	flags;	/* Flags for mbuf being allocated */
121 	short	type;	/* Type of mbuf being allocated */
122 };
123 #endif /* _KERNEL */
124 
125 /*
126  * Packet tag structure (see below for details).
127  */
128 struct m_tag {
129 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
130 	u_int16_t		m_tag_id;	/* Tag ID */
131 	u_int16_t		m_tag_len;	/* Length of data */
132 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
133 	void			(*m_tag_free)(struct m_tag *);
134 };
135 
136 /*
137  * Static network interface owned tag.
138  * Allocated through ifp->if_snd_tag_alloc().
139  */
140 struct m_snd_tag {
141 	struct ifnet *ifp;		/* network interface tag belongs to */
142 	volatile u_int refcount;
143 };
144 
145 /*
146  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
147  * Size ILP32: 48
148  *	 LP64: 56
149  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
150  * they are correct.
151  */
152 struct pkthdr {
153 	union {
154 		struct m_snd_tag *snd_tag;	/* send tag, if any */
155 		struct ifnet	*rcvif;		/* rcv interface */
156 	};
157 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
158 	int32_t		 len;		/* total packet length */
159 
160 	/* Layer crossing persistent information. */
161 	uint32_t	 flowid;	/* packet's 4-tuple system */
162 	uint32_t	 csum_flags;	/* checksum and offload features */
163 	uint16_t	 fibnum;	/* this packet should use this fib */
164 	uint8_t		 numa_domain;	/* NUMA domain of recvd pkt */
165 	uint8_t		 rsstype;	/* hash type */
166 	union {
167 		uint64_t	rcv_tstmp;	/* timestamp in ns */
168 		struct {
169 			uint8_t		 l2hlen;	/* layer 2 hdr len */
170 			uint8_t		 l3hlen;	/* layer 3 hdr len */
171 			uint8_t		 l4hlen;	/* layer 4 hdr len */
172 			uint8_t		 l5hlen;	/* layer 5 hdr len */
173 			uint32_t	 spare;
174 		};
175 	};
176 	union {
177 		uint8_t  eight[8];
178 		uint16_t sixteen[4];
179 		uint32_t thirtytwo[2];
180 		uint64_t sixtyfour[1];
181 		uintptr_t unintptr[1];
182 		void	*ptr;
183 	} PH_per;
184 
185 	/* Layer specific non-persistent local storage for reassembly, etc. */
186 	union {
187 		uint8_t  eight[8];
188 		uint16_t sixteen[4];
189 		uint32_t thirtytwo[2];
190 		uint64_t sixtyfour[1];
191 		uintptr_t unintptr[1];
192 		void 	*ptr;
193 	} PH_loc;
194 };
195 #define	ether_vtag	PH_per.sixteen[0]
196 #define	PH_vt		PH_per
197 #define	vt_nrecs	sixteen[0]	  /* mld and v6-ND */
198 #define	tso_segsz	PH_per.sixteen[1] /* inbound after LRO */
199 #define	lro_nsegs	tso_segsz	  /* inbound after LRO */
200 #define	csum_data	PH_per.thirtytwo[1] /* inbound from hardware up */
201 #define lro_len		PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
202 #define lro_csum	PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
203 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
204 
205 /*
206  * TLS records for TLS 1.0-1.2 can have the following header lengths:
207  * - 5 (AES-CBC with implicit IV)
208  * - 21 (AES-CBC with explicit IV)
209  * - 13 (AES-GCM with 8 byte explicit IV)
210  */
211 #define	MBUF_PEXT_HDR_LEN	23
212 
213 /*
214  * TLS records for TLS 1.0-1.2 can have the following maximum trailer
215  * lengths:
216  * - 16 (AES-GCM)
217  * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
218  * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
219  * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
220  */
221 #define	MBUF_PEXT_TRAIL_LEN	64
222 
223 #if defined(__LP64__)
224 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
225 #else
226 #define MBUF_PEXT_MAX_PGS (72 / sizeof(vm_paddr_t))
227 #endif
228 
229 #define	MBUF_PEXT_MAX_BYTES						\
230     (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
231 
232 #define MBUF_PEXT_FLAG_ANON	1	/* Data can be encrypted in place. */
233 
234 
235 struct mbuf_ext_pgs_data {
236 	vm_paddr_t	pa[MBUF_PEXT_MAX_PGS];		/* phys addrs of pgs */
237 	char		trail[MBUF_PEXT_TRAIL_LEN]; 	/* TLS trailer */
238 	char		hdr[MBUF_PEXT_HDR_LEN];		/* TLS header */
239 };
240 
241 struct ktls_session;
242 struct socket;
243 
244 /*
245  * Description of external storage mapped into mbuf; valid only if M_EXT is
246  * set.
247  * Size ILP32: 28
248  *	 LP64: 48
249  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
250  * they are correct.
251  */
252 typedef	void m_ext_free_t(struct mbuf *);
253 struct m_ext {
254 	union {
255 		/*
256 		 * If EXT_FLAG_EMBREF is set, then we use refcount in the
257 		 * mbuf, the 'ext_count' member.  Otherwise, we have a
258 		 * shadow copy and we use pointer 'ext_cnt'.  The original
259 		 * mbuf is responsible to carry the pointer to free routine
260 		 * and its arguments.  They aren't copied into shadows in
261 		 * mb_dupcl() to avoid dereferencing next cachelines.
262 		 */
263 		volatile u_int	 ext_count;
264 		volatile u_int	*ext_cnt;
265 	};
266 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
267 	uint32_t	 ext_type:8,	/* type of external storage */
268 			 ext_flags:24;	/* external storage mbuf flags */
269 	char		*ext_buf;	/* start of buffer */
270 	/*
271 	 * Fields below store the free context for the external storage.
272 	 * They are valid only in the refcount carrying mbuf, the one with
273 	 * EXT_FLAG_EMBREF flag, with exclusion for EXT_EXTREF type, where
274 	 * the free context is copied into all mbufs that use same external
275 	 * storage.
276 	 */
277 #define	m_ext_copylen	offsetof(struct m_ext, ext_free)
278 	m_ext_free_t	*ext_free;	/* free routine if not the usual */
279 	void		*ext_arg1;	/* optional argument pointer */
280 	union {
281 		void		*ext_arg2;	/* optional argument pointer */
282 		struct mbuf_ext_pgs_data ext_pgs;
283 	};
284 };
285 
286 struct mbuf_ext_pgs {
287 	uint8_t		npgs;			/* Number of attached pages */
288 	uint8_t		nrdy;			/* Pages with I/O pending */
289 	uint8_t		hdr_len;		/* TLS header length */
290 	uint8_t		trail_len;		/* TLS trailer length */
291 	uint16_t	first_pg_off;		/* Offset into 1st page */
292 	uint16_t	last_pg_len;		/* Length of last page */
293 	uint8_t		flags;			/* Flags */
294 	uint8_t		record_type;
295 	uint8_t		spare[2];
296 	int		enc_cnt;
297 	struct ktls_session *tls;		/* TLS session */
298 	struct socket	*so;
299 	uint64_t	seqno;
300 	struct mbuf	*mbuf;
301 	STAILQ_ENTRY(mbuf_ext_pgs) stailq;
302 #if !defined(__LP64__)
303 	uint8_t		pad[8];		/* pad to size of pkthdr */
304 #endif
305 	struct m_ext	m_ext;
306 };
307 
308 #define m_epg_hdr	m_ext.ext_pgs.hdr
309 #define m_epg_trail	m_ext.ext_pgs.trail
310 #define m_epg_pa	m_ext.ext_pgs.pa
311 
312 /*
313  * The core of the mbuf object along with some shortcut defines for practical
314  * purposes.
315  */
316 struct mbuf {
317 	/*
318 	 * Header present at the beginning of every mbuf.
319 	 * Size ILP32: 24
320 	 *      LP64: 32
321 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
322 	 * that they are correct.
323 	 */
324 	union {	/* next buffer in chain */
325 		struct mbuf		*m_next;
326 		SLIST_ENTRY(mbuf)	m_slist;
327 		STAILQ_ENTRY(mbuf)	m_stailq;
328 	};
329 	union {	/* next chain in queue/record */
330 		struct mbuf		*m_nextpkt;
331 		SLIST_ENTRY(mbuf)	m_slistpkt;
332 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
333 	};
334 	caddr_t		 m_data;	/* location of data */
335 	int32_t		 m_len;		/* amount of data in this mbuf */
336 	uint32_t	 m_type:8,	/* type of data in this mbuf */
337 			 m_flags:24;	/* flags; see below */
338 #if !defined(__LP64__)
339 	uint32_t	 m_pad;		/* pad for 64bit alignment */
340 #endif
341 
342 	/*
343 	 * A set of optional headers (packet header, external storage header)
344 	 * and internal data storage.  Historically, these arrays were sized
345 	 * to MHLEN (space left after a packet header) and MLEN (space left
346 	 * after only a regular mbuf header); they are now variable size in
347 	 * order to support future work on variable-size mbufs.
348 	 */
349 	union {
350 		union {
351 			struct {
352 				struct pkthdr	m_pkthdr; /* M_PKTHDR set */
353 				union {
354 					struct m_ext	m_ext;	/* M_EXT set */
355 					char		m_pktdat[0];
356 				};
357 			};
358 			struct mbuf_ext_pgs m_ext_pgs;
359 		};
360 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
361 	};
362 };
363 
364 #ifdef _KERNEL
365 static inline int
366 mbuf_ext_pg_len(struct mbuf_ext_pgs *ext_pgs, int pidx, int pgoff)
367 {
368 	KASSERT(pgoff == 0 || pidx == 0,
369 	    ("page %d with non-zero offset %d in %p", pidx, pgoff, ext_pgs));
370 	if (pidx == ext_pgs->npgs - 1) {
371 		return (ext_pgs->last_pg_len);
372 	} else {
373 		return (PAGE_SIZE - pgoff);
374 	}
375 }
376 
377 #ifdef INVARIANT_SUPPORT
378 void	mb_ext_pgs_check(struct mbuf_ext_pgs *ext_pgs);
379 #endif
380 #ifdef INVARIANTS
381 #define	MBUF_EXT_PGS_ASSERT_SANITY(ext_pgs)	mb_ext_pgs_check((ext_pgs))
382 #else
383 #define	MBUF_EXT_PGS_ASSERT_SANITY(ext_pgs)
384 #endif
385 #endif
386 
387 /*
388  * mbuf flags of global significance and layer crossing.
389  * Those of only protocol/layer specific significance are to be mapped
390  * to M_PROTO[1-11] and cleared at layer handoff boundaries.
391  * NB: Limited to the lower 24 bits.
392  */
393 #define	M_EXT		0x00000001 /* has associated external storage */
394 #define	M_PKTHDR	0x00000002 /* start of record */
395 #define	M_EOR		0x00000004 /* end of record */
396 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
397 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
398 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
399 #define	M_PROMISC	0x00000040 /* packet was not for us */
400 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
401 #define	M_NOMAP		0x00000100 /* mbuf data is unmapped */
402 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
403 #define	M_TSTMP		0x00000400 /* rcv_tstmp field is valid */
404 #define	M_TSTMP_HPREC	0x00000800 /* rcv_tstmp is high-prec, typically
405 				      hw-stamped on port (useful for IEEE 1588
406 				      and 802.1AS) */
407 #define M_TSTMP_LRO	0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
408 
409 #define	M_PROTO1	0x00002000 /* protocol-specific */
410 #define	M_PROTO2	0x00004000 /* protocol-specific */
411 #define	M_PROTO3	0x00008000 /* protocol-specific */
412 #define	M_PROTO4	0x00010000 /* protocol-specific */
413 #define	M_PROTO5	0x00020000 /* protocol-specific */
414 #define	M_PROTO6	0x00040000 /* protocol-specific */
415 #define	M_PROTO7	0x00080000 /* protocol-specific */
416 #define	M_PROTO8	0x00100000 /* protocol-specific */
417 #define	M_PROTO9	0x00200000 /* protocol-specific */
418 #define	M_PROTO10	0x00400000 /* protocol-specific */
419 #define	M_PROTO11	0x00800000 /* protocol-specific */
420 
421 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
422 
423 /*
424  * Flags to purge when crossing layers.
425  */
426 #define	M_PROTOFLAGS \
427     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
428      M_PROTO9|M_PROTO10|M_PROTO11)
429 
430 /*
431  * Flags preserved when copying m_pkthdr.
432  */
433 #define M_COPYFLAGS \
434     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
435      M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
436 
437 /*
438  * Mbuf flag description for use with printf(9) %b identifier.
439  */
440 #define	M_FLAG_BITS \
441     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
442     "\7M_PROMISC\10M_VLANTAG\11M_NOMAP\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
443 #define	M_FLAG_PROTOBITS \
444     "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
445     "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
446     "\27M_PROTO10\28M_PROTO11"
447 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
448 
449 /*
450  * Network interface cards are able to hash protocol fields (such as IPv4
451  * addresses and TCP port numbers) classify packets into flows.  These flows
452  * can then be used to maintain ordering while delivering packets to the OS
453  * via parallel input queues, as well as to provide a stateless affinity
454  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
455  * m_flag fields to indicate how the hash should be interpreted by the
456  * network stack.
457  *
458  * Most NICs support RSS, which provides ordering and explicit affinity, and
459  * use the hash m_flag bits to indicate what header fields were covered by
460  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
461  * RSS cards or configurations that provide an opaque flow identifier, allowing
462  * for ordering and distribution without explicit affinity.  Additionally,
463  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
464  * properties.
465  *
466  * The meaning of the IPV6_EX suffix:
467  * "o  Home address from the home address option in the IPv6 destination
468  *     options header.  If the extension header is not present, use the Source
469  *     IPv6 Address.
470  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
471  *     associated extension header.  If the extension header is not present,
472  *     use the Destination IPv6 Address."
473  * Quoted from:
474  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
475  */
476 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
477 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
478 /* Microsoft RSS standard hash types */
479 #define	M_HASHTYPE_NONE			0
480 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
481 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
482 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
483 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
484 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
485 							    * ext hdrs */
486 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
487 							    * ext hdrs */
488 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
489 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
490 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
491 							    * ext hdrs */
492 
493 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
494 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
495 						/* ordering+hash, not affinity*/
496 
497 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
498 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
499 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
500 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
501 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
502 
503 /*
504  * External mbuf storage buffer types.
505  */
506 #define	EXT_CLUSTER	1	/* mbuf cluster */
507 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
508 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
509 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
510 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
511 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
512 #define	EXT_MBUF	7	/* external mbuf reference */
513 #define	EXT_RXRING	8	/* data in NIC receive ring */
514 #define	EXT_PGS		9	/* array of unmapped pages */
515 
516 #define	EXT_VENDOR1	224	/* for vendor-internal use */
517 #define	EXT_VENDOR2	225	/* for vendor-internal use */
518 #define	EXT_VENDOR3	226	/* for vendor-internal use */
519 #define	EXT_VENDOR4	227	/* for vendor-internal use */
520 
521 #define	EXT_EXP1	244	/* for experimental use */
522 #define	EXT_EXP2	245	/* for experimental use */
523 #define	EXT_EXP3	246	/* for experimental use */
524 #define	EXT_EXP4	247	/* for experimental use */
525 
526 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
527 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
528 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
529 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
530 
531 /*
532  * Flags for external mbuf buffer types.
533  * NB: limited to the lower 24 bits.
534  */
535 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
536 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
537 
538 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
539 
540 #define	EXT_FLAG_VENDOR1	0x010000	/* These flags are vendor */
541 #define	EXT_FLAG_VENDOR2	0x020000	/* or submodule specific, */
542 #define	EXT_FLAG_VENDOR3	0x040000	/* not used by mbuf code. */
543 #define	EXT_FLAG_VENDOR4	0x080000	/* Set/read by submodule. */
544 
545 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
546 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
547 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
548 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
549 
550 /*
551  * EXT flag description for use with printf(9) %b identifier.
552  */
553 #define	EXT_FLAG_BITS \
554     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
555     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
556     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
557     "\30EXT_FLAG_EXP4"
558 
559 #define MBUF_EXT_PGS_ASSERT(m)						\
560 	KASSERT((((m)->m_flags & M_EXT) != 0) &&			\
561 	    ((m)->m_ext.ext_type == EXT_PGS),				\
562 	    ("%s: m %p !M_EXT or !EXT_PGS", __func__, m))
563 
564 /*
565  * Flags indicating checksum, segmentation and other offload work to be
566  * done, or already done, by hardware or lower layers.  It is split into
567  * separate inbound and outbound flags.
568  *
569  * Outbound flags that are set by upper protocol layers requesting lower
570  * layers, or ideally the hardware, to perform these offloading tasks.
571  * For outbound packets this field and its flags can be directly tested
572  * against ifnet if_hwassist.
573  */
574 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
575 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
576 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
577 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
578 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
579 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
580 
581 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
582 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
583 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
584 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
585 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
586 
587 /* Inbound checksum support where the checksum was verified by hardware. */
588 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
589 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
590 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
591 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
592 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
593 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
594 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
595 
596 #define	CSUM_SND_TAG		0x80000000	/* Packet header has send tag */
597 
598 /*
599  * CSUM flag description for use with printf(9) %b identifier.
600  */
601 #define	CSUM_BITS \
602     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
603     "\6CSUM_IP_ISCSI" \
604     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
605     "\16CSUM_IP6_ISCSI" \
606     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
607     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
608 
609 /* CSUM flags compatibility mappings. */
610 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
611 #define	CSUM_IP_VALID		CSUM_L3_VALID
612 #define	CSUM_DATA_VALID		CSUM_L4_VALID
613 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
614 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
615 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
616 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
617 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
618 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
619 #define	CSUM_TCP		CSUM_IP_TCP
620 #define	CSUM_UDP		CSUM_IP_UDP
621 #define	CSUM_SCTP		CSUM_IP_SCTP
622 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
623 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
624 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
625 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
626 
627 /*
628  * mbuf types describing the content of the mbuf (including external storage).
629  */
630 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
631 #define	MT_DATA		1	/* dynamic (data) allocation */
632 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
633 
634 #define	MT_VENDOR1	4	/* for vendor-internal use */
635 #define	MT_VENDOR2	5	/* for vendor-internal use */
636 #define	MT_VENDOR3	6	/* for vendor-internal use */
637 #define	MT_VENDOR4	7	/* for vendor-internal use */
638 
639 #define	MT_SONAME	8	/* socket name */
640 
641 #define	MT_EXP1		9	/* for experimental use */
642 #define	MT_EXP2		10	/* for experimental use */
643 #define	MT_EXP3		11	/* for experimental use */
644 #define	MT_EXP4		12	/* for experimental use */
645 
646 #define	MT_CONTROL	14	/* extra-data protocol message */
647 #define	MT_EXTCONTROL	15	/* control message with externalized contents */
648 #define	MT_OOBDATA	16	/* expedited data  */
649 
650 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
651 				   a non-initialized mbuf */
652 
653 /*
654  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
655  * !_KERNEL so that monitoring tools can look up the zones with
656  * libmemstat(3).
657  */
658 #define	MBUF_MEM_NAME		"mbuf"
659 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
660 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
661 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
662 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
663 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
664 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
665 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
666 #define	MBUF_EXTPGS_MEM_NAME	"mbuf_extpgs"
667 
668 #ifdef _KERNEL
669 
670 #ifdef WITNESS
671 #define	MBUF_CHECKSLEEP(how) do {					\
672 	if (how == M_WAITOK)						\
673 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
674 		    "Sleeping in \"%s\"", __func__);			\
675 } while (0)
676 #else
677 #define	MBUF_CHECKSLEEP(how)
678 #endif
679 
680 /*
681  * Network buffer allocation API
682  *
683  * The rest of it is defined in kern/kern_mbuf.c
684  */
685 extern uma_zone_t	zone_mbuf;
686 extern uma_zone_t	zone_clust;
687 extern uma_zone_t	zone_pack;
688 extern uma_zone_t	zone_jumbop;
689 extern uma_zone_t	zone_jumbo9;
690 extern uma_zone_t	zone_jumbo16;
691 extern uma_zone_t	zone_extpgs;
692 
693 void		 mb_dupcl(struct mbuf *, struct mbuf *);
694 void		 mb_free_ext(struct mbuf *);
695 void		 mb_free_mext_pgs(struct mbuf *);
696 struct mbuf	*mb_alloc_ext_pgs(int, m_ext_free_t);
697 int		 mb_unmapped_compress(struct mbuf *m);
698 struct mbuf 	*mb_unmapped_to_ext(struct mbuf *m);
699 void		 mb_free_notready(struct mbuf *m, int count);
700 void		 m_adj(struct mbuf *, int);
701 int		 m_apply(struct mbuf *, int, int,
702 		    int (*)(void *, void *, u_int), void *);
703 int		 m_append(struct mbuf *, int, c_caddr_t);
704 void		 m_cat(struct mbuf *, struct mbuf *);
705 void		 m_catpkt(struct mbuf *, struct mbuf *);
706 int		 m_clget(struct mbuf *m, int how);
707 void 		*m_cljget(struct mbuf *m, int how, int size);
708 struct mbuf	*m_collapse(struct mbuf *, int, int);
709 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
710 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
711 struct mbuf	*m_copym(struct mbuf *, int, int, int);
712 struct mbuf	*m_copypacket(struct mbuf *, int);
713 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
714 struct mbuf	*m_copyup(struct mbuf *, int, int);
715 struct mbuf	*m_defrag(struct mbuf *, int);
716 void		 m_demote_pkthdr(struct mbuf *);
717 void		 m_demote(struct mbuf *, int, int);
718 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
719 		    void (*)(char *, caddr_t, u_int));
720 void		 m_dispose_extcontrolm(struct mbuf *m);
721 struct mbuf	*m_dup(const struct mbuf *, int);
722 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
723 void		 m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
724 		    void *, void *, int, int);
725 u_int		 m_fixhdr(struct mbuf *);
726 struct mbuf	*m_fragment(struct mbuf *, int, int);
727 void		 m_freem(struct mbuf *);
728 struct mbuf	*m_get2(int, int, short, int);
729 struct mbuf	*m_getjcl(int, short, int, int);
730 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
731 struct mbuf	*m_getptr(struct mbuf *, int, int *);
732 u_int		 m_length(struct mbuf *, struct mbuf **);
733 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
734 int		 m_unmappedtouio(const struct mbuf *, int, struct uio *, int);
735 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
736 int		 m_pkthdr_init(struct mbuf *, int);
737 struct mbuf	*m_prepend(struct mbuf *, int, int);
738 void		 m_print(const struct mbuf *, int);
739 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
740 struct mbuf	*m_pullup(struct mbuf *, int);
741 int		 m_sanity(struct mbuf *, int);
742 struct mbuf	*m_split(struct mbuf *, int, int);
743 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
744 struct mbuf	*m_unshare(struct mbuf *, int);
745 void		 m_snd_tag_init(struct m_snd_tag *, struct ifnet *);
746 void		 m_snd_tag_destroy(struct m_snd_tag *);
747 
748 static __inline int
749 m_gettype(int size)
750 {
751 	int type;
752 
753 	switch (size) {
754 	case MSIZE:
755 		type = EXT_MBUF;
756 		break;
757 	case MCLBYTES:
758 		type = EXT_CLUSTER;
759 		break;
760 #if MJUMPAGESIZE != MCLBYTES
761 	case MJUMPAGESIZE:
762 		type = EXT_JUMBOP;
763 		break;
764 #endif
765 	case MJUM9BYTES:
766 		type = EXT_JUMBO9;
767 		break;
768 	case MJUM16BYTES:
769 		type = EXT_JUMBO16;
770 		break;
771 	default:
772 		panic("%s: invalid cluster size %d", __func__, size);
773 	}
774 
775 	return (type);
776 }
777 
778 /*
779  * Associated an external reference counted buffer with an mbuf.
780  */
781 static __inline void
782 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
783     m_ext_free_t freef, void *arg1, void *arg2)
784 {
785 
786 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
787 
788 	atomic_add_int(ref_cnt, 1);
789 	m->m_flags |= M_EXT;
790 	m->m_ext.ext_buf = buf;
791 	m->m_ext.ext_cnt = ref_cnt;
792 	m->m_data = m->m_ext.ext_buf;
793 	m->m_ext.ext_size = size;
794 	m->m_ext.ext_free = freef;
795 	m->m_ext.ext_arg1 = arg1;
796 	m->m_ext.ext_arg2 = arg2;
797 	m->m_ext.ext_type = EXT_EXTREF;
798 	m->m_ext.ext_flags = 0;
799 }
800 
801 static __inline uma_zone_t
802 m_getzone(int size)
803 {
804 	uma_zone_t zone;
805 
806 	switch (size) {
807 	case MCLBYTES:
808 		zone = zone_clust;
809 		break;
810 #if MJUMPAGESIZE != MCLBYTES
811 	case MJUMPAGESIZE:
812 		zone = zone_jumbop;
813 		break;
814 #endif
815 	case MJUM9BYTES:
816 		zone = zone_jumbo9;
817 		break;
818 	case MJUM16BYTES:
819 		zone = zone_jumbo16;
820 		break;
821 	default:
822 		panic("%s: invalid cluster size %d", __func__, size);
823 	}
824 
825 	return (zone);
826 }
827 
828 /*
829  * Initialize an mbuf with linear storage.
830  *
831  * Inline because the consumer text overhead will be roughly the same to
832  * initialize or call a function with this many parameters and M_PKTHDR
833  * should go away with constant propagation for !MGETHDR.
834  */
835 static __inline int
836 m_init(struct mbuf *m, int how, short type, int flags)
837 {
838 	int error;
839 
840 	m->m_next = NULL;
841 	m->m_nextpkt = NULL;
842 	m->m_data = m->m_dat;
843 	m->m_len = 0;
844 	m->m_flags = flags;
845 	m->m_type = type;
846 	if (flags & M_PKTHDR)
847 		error = m_pkthdr_init(m, how);
848 	else
849 		error = 0;
850 
851 	MBUF_PROBE5(m__init, m, how, type, flags, error);
852 	return (error);
853 }
854 
855 static __inline struct mbuf *
856 m_get(int how, short type)
857 {
858 	struct mbuf *m;
859 	struct mb_args args;
860 
861 	args.flags = 0;
862 	args.type = type;
863 	m = uma_zalloc_arg(zone_mbuf, &args, how);
864 	MBUF_PROBE3(m__get, how, type, m);
865 	return (m);
866 }
867 
868 static __inline struct mbuf *
869 m_gethdr(int how, short type)
870 {
871 	struct mbuf *m;
872 	struct mb_args args;
873 
874 	args.flags = M_PKTHDR;
875 	args.type = type;
876 	m = uma_zalloc_arg(zone_mbuf, &args, how);
877 	MBUF_PROBE3(m__gethdr, how, type, m);
878 	return (m);
879 }
880 
881 static __inline struct mbuf *
882 m_getcl(int how, short type, int flags)
883 {
884 	struct mbuf *m;
885 	struct mb_args args;
886 
887 	args.flags = flags;
888 	args.type = type;
889 	m = uma_zalloc_arg(zone_pack, &args, how);
890 	MBUF_PROBE4(m__getcl, how, type, flags, m);
891 	return (m);
892 }
893 
894 /*
895  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
896  * unreferenced cluster to an mbuf(9).  It is not possible to assert
897  * that, so care can be taken only by users of the API.
898  */
899 static __inline void
900 m_cljset(struct mbuf *m, void *cl, int type)
901 {
902 	int size;
903 
904 	switch (type) {
905 	case EXT_CLUSTER:
906 		size = MCLBYTES;
907 		break;
908 #if MJUMPAGESIZE != MCLBYTES
909 	case EXT_JUMBOP:
910 		size = MJUMPAGESIZE;
911 		break;
912 #endif
913 	case EXT_JUMBO9:
914 		size = MJUM9BYTES;
915 		break;
916 	case EXT_JUMBO16:
917 		size = MJUM16BYTES;
918 		break;
919 	default:
920 		panic("%s: unknown cluster type %d", __func__, type);
921 		break;
922 	}
923 
924 	m->m_data = m->m_ext.ext_buf = cl;
925 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
926 	m->m_ext.ext_size = size;
927 	m->m_ext.ext_type = type;
928 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
929 	m->m_ext.ext_count = 1;
930 	m->m_flags |= M_EXT;
931 	MBUF_PROBE3(m__cljset, m, cl, type);
932 }
933 
934 static __inline void
935 m_chtype(struct mbuf *m, short new_type)
936 {
937 
938 	m->m_type = new_type;
939 }
940 
941 static __inline void
942 m_clrprotoflags(struct mbuf *m)
943 {
944 
945 	while (m) {
946 		m->m_flags &= ~M_PROTOFLAGS;
947 		m = m->m_next;
948 	}
949 }
950 
951 static __inline struct mbuf *
952 m_last(struct mbuf *m)
953 {
954 
955 	while (m->m_next)
956 		m = m->m_next;
957 	return (m);
958 }
959 
960 static inline u_int
961 m_extrefcnt(struct mbuf *m)
962 {
963 
964 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
965 
966 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
967 	    *m->m_ext.ext_cnt);
968 }
969 
970 /*
971  * mbuf, cluster, and external object allocation macros (for compatibility
972  * purposes).
973  */
974 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
975 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
976 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
977 #define	MCLGET(m, how)		m_clget((m), (how))
978 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
979     m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2),	\
980     (flags), (type))
981 #define	m_getm(m, len, how, type)					\
982     m_getm2((m), (len), (how), (type), M_PKTHDR)
983 
984 /*
985  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
986  * be both the local data payload, or an external buffer area, depending on
987  * whether M_EXT is set).
988  */
989 #define	M_WRITABLE(m)	(((m)->m_flags & (M_RDONLY | M_NOMAP)) == 0 &&	\
990 			 (!(((m)->m_flags & M_EXT)) ||			\
991 			 (m_extrefcnt(m) == 1)))
992 
993 /* Check if the supplied mbuf has a packet header, or else panic. */
994 #define	M_ASSERTPKTHDR(m)						\
995 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
996 	    ("%s: no mbuf packet header!", __func__))
997 
998 /*
999  * Ensure that the supplied mbuf is a valid, non-free mbuf.
1000  *
1001  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
1002  */
1003 #define	M_ASSERTVALID(m)						\
1004 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
1005 	    ("%s: attempted use of a free mbuf!", __func__))
1006 
1007 /*
1008  * Return the address of the start of the buffer associated with an mbuf,
1009  * handling external storage, packet-header mbufs, and regular data mbufs.
1010  */
1011 #define	M_START(m)							\
1012 	(((m)->m_flags & M_NOMAP) ? NULL :				\
1013 	 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
1014 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
1015 	 &(m)->m_dat[0])
1016 
1017 /*
1018  * Return the size of the buffer associated with an mbuf, handling external
1019  * storage, packet-header mbufs, and regular data mbufs.
1020  */
1021 #define	M_SIZE(m)							\
1022 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
1023 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
1024 	 MLEN)
1025 
1026 /*
1027  * Set the m_data pointer of a newly allocated mbuf to place an object of the
1028  * specified size at the end of the mbuf, longword aligned.
1029  *
1030  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1031  * separate macros, each asserting that it was called at the proper moment.
1032  * This required callers to themselves test the storage type and call the
1033  * right one.  Rather than require callers to be aware of those layout
1034  * decisions, we centralize here.
1035  */
1036 static __inline void
1037 m_align(struct mbuf *m, int len)
1038 {
1039 #ifdef INVARIANTS
1040 	const char *msg = "%s: not a virgin mbuf";
1041 #endif
1042 	int adjust;
1043 
1044 	KASSERT(m->m_data == M_START(m), (msg, __func__));
1045 
1046 	adjust = M_SIZE(m) - len;
1047 	m->m_data += adjust &~ (sizeof(long)-1);
1048 }
1049 
1050 #define	M_ALIGN(m, len)		m_align(m, len)
1051 #define	MH_ALIGN(m, len)	m_align(m, len)
1052 #define	MEXT_ALIGN(m, len)	m_align(m, len)
1053 
1054 /*
1055  * Compute the amount of space available before the current start of data in
1056  * an mbuf.
1057  *
1058  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1059  * of checking writability of the mbuf data area rests solely with the caller.
1060  *
1061  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1062  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1063  * read-only as well.
1064  */
1065 #define	M_LEADINGSPACE(m)						\
1066 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1067 
1068 /*
1069  * Compute the amount of space available after the end of data in an mbuf.
1070  *
1071  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1072  * of checking writability of the mbuf data area rests solely with the caller.
1073  *
1074  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1075  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1076  * read-only as well.
1077  */
1078 #define	M_TRAILINGSPACE(m)						\
1079 	(M_WRITABLE(m) ?						\
1080 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
1081 
1082 /*
1083  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
1084  * allocated, how specifies whether to wait.  If the allocation fails, the
1085  * original mbuf chain is freed and m is set to NULL.
1086  */
1087 #define	M_PREPEND(m, plen, how) do {					\
1088 	struct mbuf **_mmp = &(m);					\
1089 	struct mbuf *_mm = *_mmp;					\
1090 	int _mplen = (plen);						\
1091 	int __mhow = (how);						\
1092 									\
1093 	MBUF_CHECKSLEEP(how);						\
1094 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
1095 		_mm->m_data -= _mplen;					\
1096 		_mm->m_len += _mplen;					\
1097 	} else								\
1098 		_mm = m_prepend(_mm, _mplen, __mhow);			\
1099 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
1100 		_mm->m_pkthdr.len += _mplen;				\
1101 	*_mmp = _mm;							\
1102 } while (0)
1103 
1104 /*
1105  * Change mbuf to new type.  This is a relatively expensive operation and
1106  * should be avoided.
1107  */
1108 #define	MCHTYPE(m, t)	m_chtype((m), (t))
1109 
1110 /* Return the rcvif of a packet header. */
1111 static __inline struct ifnet *
1112 m_rcvif(struct mbuf *m)
1113 {
1114 
1115 	M_ASSERTPKTHDR(m);
1116 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1117 		return (NULL);
1118 	return (m->m_pkthdr.rcvif);
1119 }
1120 
1121 /* Length to m_copy to copy all. */
1122 #define	M_COPYALL	1000000000
1123 
1124 extern int		max_datalen;	/* MHLEN - max_hdr */
1125 extern int		max_hdr;	/* Largest link + protocol header */
1126 extern int		max_linkhdr;	/* Largest link-level header */
1127 extern int		max_protohdr;	/* Largest protocol header */
1128 extern int		nmbclusters;	/* Maximum number of clusters */
1129 extern bool		mb_use_ext_pgs;	/* Use ext_pgs for sendfile */
1130 
1131 /*-
1132  * Network packets may have annotations attached by affixing a list of
1133  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1134  * allocated semi-opaque data structures that have a fixed header
1135  * (struct m_tag) that specifies the size of the memory block and a
1136  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1137  * unsigned value used to identify a module or ABI.  By convention this value
1138  * is chosen as the date+time that the module is created, expressed as the
1139  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1140  * value is an ABI/module-specific value that identifies a particular
1141  * annotation and is private to the module.  For compatibility with systems
1142  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1143  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1144  * compatibility shim functions and several tag types are defined below.
1145  * Users that do not require compatibility should use a private cookie value
1146  * so that packet tag-related definitions can be maintained privately.
1147  *
1148  * Note that the packet tag returned by m_tag_alloc has the default memory
1149  * alignment implemented by malloc.  To reference private data one can use a
1150  * construct like:
1151  *
1152  *	struct m_tag *mtag = m_tag_alloc(...);
1153  *	struct foo *p = (struct foo *)(mtag+1);
1154  *
1155  * if the alignment of struct m_tag is sufficient for referencing members of
1156  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1157  * private data structure to insure proper alignment; e.g.,
1158  *
1159  *	struct foo {
1160  *		struct m_tag	tag;
1161  *		...
1162  *	};
1163  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1164  *	struct m_tag *mtag = &p->tag;
1165  */
1166 
1167 /*
1168  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1169  * tags are expected to ``vanish'' when they pass through a network
1170  * interface.  For most interfaces this happens normally as the tags are
1171  * reclaimed when the mbuf is free'd.  However in some special cases
1172  * reclaiming must be done manually.  An example is packets that pass through
1173  * the loopback interface.  Also, one must be careful to do this when
1174  * ``turning around'' packets (e.g., icmp_reflect).
1175  *
1176  * To mark a tag persistent bit-or this flag in when defining the tag id.
1177  * The tag will then be treated as described above.
1178  */
1179 #define	MTAG_PERSISTENT				0x800
1180 
1181 #define	PACKET_TAG_NONE				0  /* Nadda */
1182 
1183 /* Packet tags for use with PACKET_ABI_COMPAT. */
1184 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1185 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1186 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1187 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1188 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1189 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1190 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1191 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1192 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1193 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1194 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1195 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1196 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1197 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1198 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1199 #define	PACKET_TAG_DIVERT			17 /* divert info */
1200 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1201 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1202 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1203 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1204 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1205 #define	PACKET_TAG_CARP				28 /* CARP info */
1206 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1207 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1208 
1209 /* Specific cookies and tags. */
1210 
1211 /* Packet tag routines. */
1212 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1213 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1214 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1215 void		 m_tag_free_default(struct m_tag *);
1216 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1217 struct m_tag	*m_tag_copy(struct m_tag *, int);
1218 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1219 void		 m_tag_delete_nonpersistent(struct mbuf *);
1220 
1221 /*
1222  * Initialize the list of tags associated with an mbuf.
1223  */
1224 static __inline void
1225 m_tag_init(struct mbuf *m)
1226 {
1227 
1228 	SLIST_INIT(&m->m_pkthdr.tags);
1229 }
1230 
1231 /*
1232  * Set up the contents of a tag.  Note that this does not fill in the free
1233  * method; the caller is expected to do that.
1234  *
1235  * XXX probably should be called m_tag_init, but that was already taken.
1236  */
1237 static __inline void
1238 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1239 {
1240 
1241 	t->m_tag_id = type;
1242 	t->m_tag_len = len;
1243 	t->m_tag_cookie = cookie;
1244 }
1245 
1246 /*
1247  * Reclaim resources associated with a tag.
1248  */
1249 static __inline void
1250 m_tag_free(struct m_tag *t)
1251 {
1252 
1253 	(*t->m_tag_free)(t);
1254 }
1255 
1256 /*
1257  * Return the first tag associated with an mbuf.
1258  */
1259 static __inline struct m_tag *
1260 m_tag_first(struct mbuf *m)
1261 {
1262 
1263 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1264 }
1265 
1266 /*
1267  * Return the next tag in the list of tags associated with an mbuf.
1268  */
1269 static __inline struct m_tag *
1270 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1271 {
1272 
1273 	return (SLIST_NEXT(t, m_tag_link));
1274 }
1275 
1276 /*
1277  * Prepend a tag to the list of tags associated with an mbuf.
1278  */
1279 static __inline void
1280 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1281 {
1282 
1283 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1284 }
1285 
1286 /*
1287  * Unlink a tag from the list of tags associated with an mbuf.
1288  */
1289 static __inline void
1290 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1291 {
1292 
1293 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1294 }
1295 
1296 /* These are for OpenBSD compatibility. */
1297 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1298 
1299 static __inline struct m_tag *
1300 m_tag_get(int type, int length, int wait)
1301 {
1302 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1303 }
1304 
1305 static __inline struct m_tag *
1306 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1307 {
1308 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1309 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1310 }
1311 
1312 static inline struct m_snd_tag *
1313 m_snd_tag_ref(struct m_snd_tag *mst)
1314 {
1315 
1316 	refcount_acquire(&mst->refcount);
1317 	return (mst);
1318 }
1319 
1320 static inline void
1321 m_snd_tag_rele(struct m_snd_tag *mst)
1322 {
1323 
1324 	if (refcount_release(&mst->refcount))
1325 		m_snd_tag_destroy(mst);
1326 }
1327 
1328 static __inline struct mbuf *
1329 m_free(struct mbuf *m)
1330 {
1331 	struct mbuf *n = m->m_next;
1332 
1333 	MBUF_PROBE1(m__free, m);
1334 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1335 		m_tag_delete_chain(m, NULL);
1336 	if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1337 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
1338 	if (m->m_flags & M_EXT)
1339 		mb_free_ext(m);
1340 	else if ((m->m_flags & M_NOFREE) == 0)
1341 		uma_zfree(zone_mbuf, m);
1342 	return (n);
1343 }
1344 
1345 static __inline int
1346 rt_m_getfib(struct mbuf *m)
1347 {
1348 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1349 	return (m->m_pkthdr.fibnum);
1350 }
1351 
1352 #define M_GETFIB(_m)   rt_m_getfib(_m)
1353 
1354 #define M_SETFIB(_m, _fib) do {						\
1355         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1356 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1357 } while (0)
1358 
1359 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1360 #define	MBUF_HASHFLAG_L2	(1 << 2)
1361 #define	MBUF_HASHFLAG_L3	(1 << 3)
1362 #define	MBUF_HASHFLAG_L4	(1 << 4)
1363 
1364 /* mbuf hashing helper routines */
1365 uint32_t	m_ether_tcpip_hash_init(void);
1366 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1367 
1368 #ifdef MBUF_PROFILING
1369  void m_profile(struct mbuf *m);
1370  #define M_PROFILE(m) m_profile(m)
1371 #else
1372  #define M_PROFILE(m)
1373 #endif
1374 
1375 struct mbufq {
1376 	STAILQ_HEAD(, mbuf)	mq_head;
1377 	int			mq_len;
1378 	int			mq_maxlen;
1379 };
1380 
1381 static inline void
1382 mbufq_init(struct mbufq *mq, int maxlen)
1383 {
1384 
1385 	STAILQ_INIT(&mq->mq_head);
1386 	mq->mq_maxlen = maxlen;
1387 	mq->mq_len = 0;
1388 }
1389 
1390 static inline struct mbuf *
1391 mbufq_flush(struct mbufq *mq)
1392 {
1393 	struct mbuf *m;
1394 
1395 	m = STAILQ_FIRST(&mq->mq_head);
1396 	STAILQ_INIT(&mq->mq_head);
1397 	mq->mq_len = 0;
1398 	return (m);
1399 }
1400 
1401 static inline void
1402 mbufq_drain(struct mbufq *mq)
1403 {
1404 	struct mbuf *m, *n;
1405 
1406 	n = mbufq_flush(mq);
1407 	while ((m = n) != NULL) {
1408 		n = STAILQ_NEXT(m, m_stailqpkt);
1409 		m_freem(m);
1410 	}
1411 }
1412 
1413 static inline struct mbuf *
1414 mbufq_first(const struct mbufq *mq)
1415 {
1416 
1417 	return (STAILQ_FIRST(&mq->mq_head));
1418 }
1419 
1420 static inline struct mbuf *
1421 mbufq_last(const struct mbufq *mq)
1422 {
1423 
1424 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1425 }
1426 
1427 static inline int
1428 mbufq_full(const struct mbufq *mq)
1429 {
1430 
1431 	return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1432 }
1433 
1434 static inline int
1435 mbufq_len(const struct mbufq *mq)
1436 {
1437 
1438 	return (mq->mq_len);
1439 }
1440 
1441 static inline int
1442 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1443 {
1444 
1445 	if (mbufq_full(mq))
1446 		return (ENOBUFS);
1447 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1448 	mq->mq_len++;
1449 	return (0);
1450 }
1451 
1452 static inline struct mbuf *
1453 mbufq_dequeue(struct mbufq *mq)
1454 {
1455 	struct mbuf *m;
1456 
1457 	m = STAILQ_FIRST(&mq->mq_head);
1458 	if (m) {
1459 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1460 		m->m_nextpkt = NULL;
1461 		mq->mq_len--;
1462 	}
1463 	return (m);
1464 }
1465 
1466 static inline void
1467 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1468 {
1469 
1470 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1471 	mq->mq_len++;
1472 }
1473 
1474 /*
1475  * Note: this doesn't enforce the maximum list size for dst.
1476  */
1477 static inline void
1478 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1479 {
1480 
1481 	mq_dst->mq_len += mq_src->mq_len;
1482 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1483 	mq_src->mq_len = 0;
1484 }
1485 
1486 #ifdef _SYS_TIMESPEC_H_
1487 static inline void
1488 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1489 {
1490 
1491 	KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1492 	KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0, ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1493 	ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1494 	ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1495 }
1496 #endif
1497 
1498 #ifdef DEBUGNET
1499 /* Invoked from the debugnet client code. */
1500 void	debugnet_mbuf_drain(void);
1501 void	debugnet_mbuf_start(void);
1502 void	debugnet_mbuf_finish(void);
1503 void	debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1504 #endif
1505 
1506 static inline bool
1507 mbuf_has_tls_session(struct mbuf *m)
1508 {
1509 
1510 	if (m->m_flags & M_NOMAP) {
1511 		MBUF_EXT_PGS_ASSERT(m);
1512 		if (m->m_ext_pgs.tls != NULL) {
1513 			return (true);
1514 		}
1515 	}
1516 	return (false);
1517 }
1518 
1519 #endif /* _KERNEL */
1520 #endif /* !_SYS_MBUF_H_ */
1521