xref: /freebsd/sys/sys/mbuf.h (revision c8b057f4a76947a67916e659428583334d54c1a0)
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]
198 #define	tso_segsz	PH_per.sixteen[1]
199 #define	lro_nsegs	tso_segsz
200 #define	csum_phsum	PH_per.sixteen[2]
201 #define	csum_data	PH_per.thirtytwo[1]
202 #define pace_thoff	PH_loc.sixteen[0]
203 #define pace_tlen	PH_loc.sixteen[1]
204 #define pace_drphdrlen	PH_loc.sixteen[2]
205 #define pace_tos	PH_loc.eight[6]
206 #define pace_lock	PH_loc.eight[7]
207 
208 /*
209  * Description of external storage mapped into mbuf; valid only if M_EXT is
210  * set.
211  * Size ILP32: 28
212  *	 LP64: 48
213  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
214  * they are correct.
215  */
216 typedef	void m_ext_free_t(struct mbuf *);
217 struct m_ext {
218 	union {
219 		/*
220 		 * If EXT_FLAG_EMBREF is set, then we use refcount in the
221 		 * mbuf, the 'ext_count' member.  Otherwise, we have a
222 		 * shadow copy and we use pointer 'ext_cnt'.  The original
223 		 * mbuf is responsible to carry the pointer to free routine
224 		 * and its arguments.  They aren't copied into shadows in
225 		 * mb_dupcl() to avoid dereferencing next cachelines.
226 		 */
227 		volatile u_int	 ext_count;
228 		volatile u_int	*ext_cnt;
229 	};
230 	char		*ext_buf;	/* start of buffer */
231 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
232 	uint32_t	 ext_type:8,	/* type of external storage */
233 			 ext_flags:24;	/* external storage mbuf flags */
234 	/*
235 	 * Fields below store the free context for the external storage.
236 	 * They are valid only in the refcount carrying mbuf, the one with
237 	 * EXT_FLAG_EMBREF flag, with exclusion for EXT_EXTREF type, where
238 	 * the free context is copied into all mbufs that use same external
239 	 * storage.
240 	 */
241 #define	m_ext_copylen	offsetof(struct m_ext, ext_free)
242 	m_ext_free_t	*ext_free;	/* free routine if not the usual */
243 	void		*ext_arg1;	/* optional argument pointer */
244 	void		*ext_arg2;	/* optional argument pointer */
245 };
246 
247 /*
248  * The core of the mbuf object along with some shortcut defines for practical
249  * purposes.
250  */
251 struct mbuf {
252 	/*
253 	 * Header present at the beginning of every mbuf.
254 	 * Size ILP32: 24
255 	 *      LP64: 32
256 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
257 	 * that they are correct.
258 	 */
259 	union {	/* next buffer in chain */
260 		struct mbuf		*m_next;
261 		SLIST_ENTRY(mbuf)	m_slist;
262 		STAILQ_ENTRY(mbuf)	m_stailq;
263 	};
264 	union {	/* next chain in queue/record */
265 		struct mbuf		*m_nextpkt;
266 		SLIST_ENTRY(mbuf)	m_slistpkt;
267 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
268 	};
269 	caddr_t		 m_data;	/* location of data */
270 	int32_t		 m_len;		/* amount of data in this mbuf */
271 	uint32_t	 m_type:8,	/* type of data in this mbuf */
272 			 m_flags:24;	/* flags; see below */
273 #if !defined(__LP64__)
274 	uint32_t	 m_pad;		/* pad for 64bit alignment */
275 #endif
276 
277 	/*
278 	 * A set of optional headers (packet header, external storage header)
279 	 * and internal data storage.  Historically, these arrays were sized
280 	 * to MHLEN (space left after a packet header) and MLEN (space left
281 	 * after only a regular mbuf header); they are now variable size in
282 	 * order to support future work on variable-size mbufs.
283 	 */
284 	union {
285 		struct {
286 			struct pkthdr	m_pkthdr;	/* M_PKTHDR set */
287 			union {
288 				struct m_ext	m_ext;	/* M_EXT set */
289 				char		m_pktdat[0];
290 			};
291 		};
292 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
293 	};
294 };
295 
296 /*
297  * mbuf flags of global significance and layer crossing.
298  * Those of only protocol/layer specific significance are to be mapped
299  * to M_PROTO[1-12] and cleared at layer handoff boundaries.
300  * NB: Limited to the lower 24 bits.
301  */
302 #define	M_EXT		0x00000001 /* has associated external storage */
303 #define	M_PKTHDR	0x00000002 /* start of record */
304 #define	M_EOR		0x00000004 /* end of record */
305 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
306 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
307 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
308 #define	M_PROMISC	0x00000040 /* packet was not for us */
309 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
310 #define	M_NOMAP		0x00000100 /* mbuf data is unmapped (soon from Drew) */
311 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
312 #define	M_TSTMP		0x00000400 /* rcv_tstmp field is valid */
313 #define	M_TSTMP_HPREC	0x00000800 /* rcv_tstmp is high-prec, typically
314 				      hw-stamped on port (useful for IEEE 1588
315 				      and 802.1AS) */
316 
317 #define	M_PROTO1	0x00001000 /* protocol-specific */
318 #define	M_PROTO2	0x00002000 /* protocol-specific */
319 #define	M_PROTO3	0x00004000 /* protocol-specific */
320 #define	M_PROTO4	0x00008000 /* protocol-specific */
321 #define	M_PROTO5	0x00010000 /* protocol-specific */
322 #define	M_PROTO6	0x00020000 /* protocol-specific */
323 #define	M_PROTO7	0x00040000 /* protocol-specific */
324 #define	M_PROTO8	0x00080000 /* protocol-specific */
325 #define	M_PROTO9	0x00100000 /* protocol-specific */
326 #define	M_PROTO10	0x00200000 /* protocol-specific */
327 #define	M_PROTO11	0x00400000 /* protocol-specific */
328 #define	M_PROTO12	0x00800000 /* protocol-specific */
329 
330 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
331 
332 /*
333  * Flags to purge when crossing layers.
334  */
335 #define	M_PROTOFLAGS \
336     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
337      M_PROTO9|M_PROTO10|M_PROTO11|M_PROTO12)
338 
339 /*
340  * Flags preserved when copying m_pkthdr.
341  */
342 #define M_COPYFLAGS \
343     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
344      M_TSTMP_HPREC|M_PROTOFLAGS)
345 
346 /*
347  * Mbuf flag description for use with printf(9) %b identifier.
348  */
349 #define	M_FLAG_BITS \
350     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
351     "\7M_PROMISC\10M_VLANTAG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC"
352 #define	M_FLAG_PROTOBITS \
353     "\15M_PROTO1\16M_PROTO2\17M_PROTO3\20M_PROTO4\21M_PROTO5" \
354     "\22M_PROTO6\23M_PROTO7\24M_PROTO8\25M_PROTO9\26M_PROTO10" \
355     "\27M_PROTO11\30M_PROTO12"
356 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
357 
358 /*
359  * Network interface cards are able to hash protocol fields (such as IPv4
360  * addresses and TCP port numbers) classify packets into flows.  These flows
361  * can then be used to maintain ordering while delivering packets to the OS
362  * via parallel input queues, as well as to provide a stateless affinity
363  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
364  * m_flag fields to indicate how the hash should be interpreted by the
365  * network stack.
366  *
367  * Most NICs support RSS, which provides ordering and explicit affinity, and
368  * use the hash m_flag bits to indicate what header fields were covered by
369  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
370  * RSS cards or configurations that provide an opaque flow identifier, allowing
371  * for ordering and distribution without explicit affinity.  Additionally,
372  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
373  * properties.
374  *
375  * The meaning of the IPV6_EX suffix:
376  * "o  Home address from the home address option in the IPv6 destination
377  *     options header.  If the extension header is not present, use the Source
378  *     IPv6 Address.
379  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
380  *     associated extension header.  If the extension header is not present,
381  *     use the Destination IPv6 Address."
382  * Quoted from:
383  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
384  */
385 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
386 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
387 /* Microsoft RSS standard hash types */
388 #define	M_HASHTYPE_NONE			0
389 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
390 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
391 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
392 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
393 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
394 							    * ext hdrs */
395 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
396 							    * ext hdrs */
397 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
398 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
399 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
400 							    * ext hdrs */
401 
402 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
403 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
404 						/* ordering+hash, not affinity*/
405 
406 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
407 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
408 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
409 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
410 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
411 
412 /*
413  * External mbuf storage buffer types.
414  */
415 #define	EXT_CLUSTER	1	/* mbuf cluster */
416 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
417 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
418 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
419 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
420 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
421 #define	EXT_MBUF	7	/* external mbuf reference */
422 #define	EXT_RXRING	8	/* data in NIC receive ring */
423 
424 #define	EXT_VENDOR1	224	/* for vendor-internal use */
425 #define	EXT_VENDOR2	225	/* for vendor-internal use */
426 #define	EXT_VENDOR3	226	/* for vendor-internal use */
427 #define	EXT_VENDOR4	227	/* for vendor-internal use */
428 
429 #define	EXT_EXP1	244	/* for experimental use */
430 #define	EXT_EXP2	245	/* for experimental use */
431 #define	EXT_EXP3	246	/* for experimental use */
432 #define	EXT_EXP4	247	/* for experimental use */
433 
434 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
435 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
436 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
437 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
438 
439 /*
440  * Flags for external mbuf buffer types.
441  * NB: limited to the lower 24 bits.
442  */
443 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
444 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
445 
446 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
447 
448 #define	EXT_FLAG_VENDOR1	0x010000	/* These flags are vendor */
449 #define	EXT_FLAG_VENDOR2	0x020000	/* or submodule specific, */
450 #define	EXT_FLAG_VENDOR3	0x040000	/* not used by mbuf code. */
451 #define	EXT_FLAG_VENDOR4	0x080000	/* Set/read by submodule. */
452 
453 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
454 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
455 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
456 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
457 
458 /*
459  * EXT flag description for use with printf(9) %b identifier.
460  */
461 #define	EXT_FLAG_BITS \
462     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
463     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
464     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
465     "\30EXT_FLAG_EXP4"
466 
467 /*
468  * Flags indicating checksum, segmentation and other offload work to be
469  * done, or already done, by hardware or lower layers.  It is split into
470  * separate inbound and outbound flags.
471  *
472  * Outbound flags that are set by upper protocol layers requesting lower
473  * layers, or ideally the hardware, to perform these offloading tasks.
474  * For outbound packets this field and its flags can be directly tested
475  * against ifnet if_hwassist.
476  */
477 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
478 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
479 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
480 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
481 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
482 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
483 
484 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
485 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
486 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
487 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
488 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
489 
490 /* Inbound checksum support where the checksum was verified by hardware. */
491 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
492 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
493 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
494 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
495 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
496 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
497 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
498 
499 #define	CSUM_SND_TAG		0x80000000	/* Packet header has send tag */
500 
501 /*
502  * CSUM flag description for use with printf(9) %b identifier.
503  */
504 #define	CSUM_BITS \
505     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
506     "\6CSUM_IP_ISCSI" \
507     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
508     "\16CSUM_IP6_ISCSI" \
509     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
510     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
511 
512 /* CSUM flags compatibility mappings. */
513 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
514 #define	CSUM_IP_VALID		CSUM_L3_VALID
515 #define	CSUM_DATA_VALID		CSUM_L4_VALID
516 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
517 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
518 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
519 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
520 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
521 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
522 #define	CSUM_TCP		CSUM_IP_TCP
523 #define	CSUM_UDP		CSUM_IP_UDP
524 #define	CSUM_SCTP		CSUM_IP_SCTP
525 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
526 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
527 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
528 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
529 
530 /*
531  * mbuf types describing the content of the mbuf (including external storage).
532  */
533 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
534 #define	MT_DATA		1	/* dynamic (data) allocation */
535 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
536 
537 #define	MT_VENDOR1	4	/* for vendor-internal use */
538 #define	MT_VENDOR2	5	/* for vendor-internal use */
539 #define	MT_VENDOR3	6	/* for vendor-internal use */
540 #define	MT_VENDOR4	7	/* for vendor-internal use */
541 
542 #define	MT_SONAME	8	/* socket name */
543 
544 #define	MT_EXP1		9	/* for experimental use */
545 #define	MT_EXP2		10	/* for experimental use */
546 #define	MT_EXP3		11	/* for experimental use */
547 #define	MT_EXP4		12	/* for experimental use */
548 
549 #define	MT_CONTROL	14	/* extra-data protocol message */
550 #define	MT_EXTCONTROL	15	/* control message with externalized contents */
551 #define	MT_OOBDATA	16	/* expedited data  */
552 
553 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
554 				   a non-initialized mbuf */
555 
556 /*
557  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
558  * !_KERNEL so that monitoring tools can look up the zones with
559  * libmemstat(3).
560  */
561 #define	MBUF_MEM_NAME		"mbuf"
562 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
563 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
564 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
565 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
566 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
567 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
568 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
569 
570 #ifdef _KERNEL
571 
572 #ifdef WITNESS
573 #define	MBUF_CHECKSLEEP(how) do {					\
574 	if (how == M_WAITOK)						\
575 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
576 		    "Sleeping in \"%s\"", __func__);			\
577 } while (0)
578 #else
579 #define	MBUF_CHECKSLEEP(how)
580 #endif
581 
582 /*
583  * Network buffer allocation API
584  *
585  * The rest of it is defined in kern/kern_mbuf.c
586  */
587 extern uma_zone_t	zone_mbuf;
588 extern uma_zone_t	zone_clust;
589 extern uma_zone_t	zone_pack;
590 extern uma_zone_t	zone_jumbop;
591 extern uma_zone_t	zone_jumbo9;
592 extern uma_zone_t	zone_jumbo16;
593 
594 void		 mb_dupcl(struct mbuf *, struct mbuf *);
595 void		 mb_free_ext(struct mbuf *);
596 void		 m_adj(struct mbuf *, int);
597 int		 m_apply(struct mbuf *, int, int,
598 		    int (*)(void *, void *, u_int), void *);
599 int		 m_append(struct mbuf *, int, c_caddr_t);
600 void		 m_cat(struct mbuf *, struct mbuf *);
601 void		 m_catpkt(struct mbuf *, struct mbuf *);
602 int		 m_clget(struct mbuf *m, int how);
603 void 		*m_cljget(struct mbuf *m, int how, int size);
604 struct mbuf	*m_collapse(struct mbuf *, int, int);
605 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
606 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
607 struct mbuf	*m_copym(struct mbuf *, int, int, int);
608 struct mbuf	*m_copypacket(struct mbuf *, int);
609 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
610 struct mbuf	*m_copyup(struct mbuf *, int, int);
611 struct mbuf	*m_defrag(struct mbuf *, int);
612 void		 m_demote_pkthdr(struct mbuf *);
613 void		 m_demote(struct mbuf *, int, int);
614 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
615 		    void (*)(char *, caddr_t, u_int));
616 void		 m_dispose_extcontrolm(struct mbuf *m);
617 struct mbuf	*m_dup(const struct mbuf *, int);
618 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
619 void		 m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
620 		    void *, void *, int, int);
621 u_int		 m_fixhdr(struct mbuf *);
622 struct mbuf	*m_fragment(struct mbuf *, int, int);
623 void		 m_freem(struct mbuf *);
624 struct mbuf	*m_get2(int, int, short, int);
625 struct mbuf	*m_getjcl(int, short, int, int);
626 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
627 struct mbuf	*m_getptr(struct mbuf *, int, int *);
628 u_int		 m_length(struct mbuf *, struct mbuf **);
629 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
630 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
631 int		 m_pkthdr_init(struct mbuf *, int);
632 struct mbuf	*m_prepend(struct mbuf *, int, int);
633 void		 m_print(const struct mbuf *, int);
634 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
635 struct mbuf	*m_pullup(struct mbuf *, int);
636 int		 m_sanity(struct mbuf *, int);
637 struct mbuf	*m_split(struct mbuf *, int, int);
638 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
639 struct mbuf	*m_unshare(struct mbuf *, int);
640 void		 m_snd_tag_init(struct m_snd_tag *, struct ifnet *);
641 void		 m_snd_tag_destroy(struct m_snd_tag *);
642 
643 static __inline int
644 m_gettype(int size)
645 {
646 	int type;
647 
648 	switch (size) {
649 	case MSIZE:
650 		type = EXT_MBUF;
651 		break;
652 	case MCLBYTES:
653 		type = EXT_CLUSTER;
654 		break;
655 #if MJUMPAGESIZE != MCLBYTES
656 	case MJUMPAGESIZE:
657 		type = EXT_JUMBOP;
658 		break;
659 #endif
660 	case MJUM9BYTES:
661 		type = EXT_JUMBO9;
662 		break;
663 	case MJUM16BYTES:
664 		type = EXT_JUMBO16;
665 		break;
666 	default:
667 		panic("%s: invalid cluster size %d", __func__, size);
668 	}
669 
670 	return (type);
671 }
672 
673 /*
674  * Associated an external reference counted buffer with an mbuf.
675  */
676 static __inline void
677 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
678     m_ext_free_t freef, void *arg1, void *arg2)
679 {
680 
681 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
682 
683 	atomic_add_int(ref_cnt, 1);
684 	m->m_flags |= M_EXT;
685 	m->m_ext.ext_buf = buf;
686 	m->m_ext.ext_cnt = ref_cnt;
687 	m->m_data = m->m_ext.ext_buf;
688 	m->m_ext.ext_size = size;
689 	m->m_ext.ext_free = freef;
690 	m->m_ext.ext_arg1 = arg1;
691 	m->m_ext.ext_arg2 = arg2;
692 	m->m_ext.ext_type = EXT_EXTREF;
693 	m->m_ext.ext_flags = 0;
694 }
695 
696 static __inline uma_zone_t
697 m_getzone(int size)
698 {
699 	uma_zone_t zone;
700 
701 	switch (size) {
702 	case MCLBYTES:
703 		zone = zone_clust;
704 		break;
705 #if MJUMPAGESIZE != MCLBYTES
706 	case MJUMPAGESIZE:
707 		zone = zone_jumbop;
708 		break;
709 #endif
710 	case MJUM9BYTES:
711 		zone = zone_jumbo9;
712 		break;
713 	case MJUM16BYTES:
714 		zone = zone_jumbo16;
715 		break;
716 	default:
717 		panic("%s: invalid cluster size %d", __func__, size);
718 	}
719 
720 	return (zone);
721 }
722 
723 /*
724  * Initialize an mbuf with linear storage.
725  *
726  * Inline because the consumer text overhead will be roughly the same to
727  * initialize or call a function with this many parameters and M_PKTHDR
728  * should go away with constant propagation for !MGETHDR.
729  */
730 static __inline int
731 m_init(struct mbuf *m, int how, short type, int flags)
732 {
733 	int error;
734 
735 	m->m_next = NULL;
736 	m->m_nextpkt = NULL;
737 	m->m_data = m->m_dat;
738 	m->m_len = 0;
739 	m->m_flags = flags;
740 	m->m_type = type;
741 	if (flags & M_PKTHDR)
742 		error = m_pkthdr_init(m, how);
743 	else
744 		error = 0;
745 
746 	MBUF_PROBE5(m__init, m, how, type, flags, error);
747 	return (error);
748 }
749 
750 static __inline struct mbuf *
751 m_get(int how, short type)
752 {
753 	struct mbuf *m;
754 	struct mb_args args;
755 
756 	args.flags = 0;
757 	args.type = type;
758 	m = uma_zalloc_arg(zone_mbuf, &args, how);
759 	MBUF_PROBE3(m__get, how, type, m);
760 	return (m);
761 }
762 
763 static __inline struct mbuf *
764 m_gethdr(int how, short type)
765 {
766 	struct mbuf *m;
767 	struct mb_args args;
768 
769 	args.flags = M_PKTHDR;
770 	args.type = type;
771 	m = uma_zalloc_arg(zone_mbuf, &args, how);
772 	MBUF_PROBE3(m__gethdr, how, type, m);
773 	return (m);
774 }
775 
776 static __inline struct mbuf *
777 m_getcl(int how, short type, int flags)
778 {
779 	struct mbuf *m;
780 	struct mb_args args;
781 
782 	args.flags = flags;
783 	args.type = type;
784 	m = uma_zalloc_arg(zone_pack, &args, how);
785 	MBUF_PROBE4(m__getcl, how, type, flags, m);
786 	return (m);
787 }
788 
789 /*
790  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
791  * unreferenced cluster to an mbuf(9).  It is not possible to assert
792  * that, so care can be taken only by users of the API.
793  */
794 static __inline void
795 m_cljset(struct mbuf *m, void *cl, int type)
796 {
797 	int size;
798 
799 	switch (type) {
800 	case EXT_CLUSTER:
801 		size = MCLBYTES;
802 		break;
803 #if MJUMPAGESIZE != MCLBYTES
804 	case EXT_JUMBOP:
805 		size = MJUMPAGESIZE;
806 		break;
807 #endif
808 	case EXT_JUMBO9:
809 		size = MJUM9BYTES;
810 		break;
811 	case EXT_JUMBO16:
812 		size = MJUM16BYTES;
813 		break;
814 	default:
815 		panic("%s: unknown cluster type %d", __func__, type);
816 		break;
817 	}
818 
819 	m->m_data = m->m_ext.ext_buf = cl;
820 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
821 	m->m_ext.ext_size = size;
822 	m->m_ext.ext_type = type;
823 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
824 	m->m_ext.ext_count = 1;
825 	m->m_flags |= M_EXT;
826 	MBUF_PROBE3(m__cljset, m, cl, type);
827 }
828 
829 static __inline void
830 m_chtype(struct mbuf *m, short new_type)
831 {
832 
833 	m->m_type = new_type;
834 }
835 
836 static __inline void
837 m_clrprotoflags(struct mbuf *m)
838 {
839 
840 	while (m) {
841 		m->m_flags &= ~M_PROTOFLAGS;
842 		m = m->m_next;
843 	}
844 }
845 
846 static __inline struct mbuf *
847 m_last(struct mbuf *m)
848 {
849 
850 	while (m->m_next)
851 		m = m->m_next;
852 	return (m);
853 }
854 
855 static inline u_int
856 m_extrefcnt(struct mbuf *m)
857 {
858 
859 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
860 
861 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
862 	    *m->m_ext.ext_cnt);
863 }
864 
865 /*
866  * mbuf, cluster, and external object allocation macros (for compatibility
867  * purposes).
868  */
869 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
870 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
871 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
872 #define	MCLGET(m, how)		m_clget((m), (how))
873 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
874     m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2),	\
875     (flags), (type))
876 #define	m_getm(m, len, how, type)					\
877     m_getm2((m), (len), (how), (type), M_PKTHDR)
878 
879 /*
880  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
881  * be both the local data payload, or an external buffer area, depending on
882  * whether M_EXT is set).
883  */
884 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
885 			 (!(((m)->m_flags & M_EXT)) ||			\
886 			 (m_extrefcnt(m) == 1)))
887 
888 /* Check if the supplied mbuf has a packet header, or else panic. */
889 #define	M_ASSERTPKTHDR(m)						\
890 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
891 	    ("%s: no mbuf packet header!", __func__))
892 
893 /*
894  * Ensure that the supplied mbuf is a valid, non-free mbuf.
895  *
896  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
897  */
898 #define	M_ASSERTVALID(m)						\
899 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
900 	    ("%s: attempted use of a free mbuf!", __func__))
901 
902 /*
903  * Return the address of the start of the buffer associated with an mbuf,
904  * handling external storage, packet-header mbufs, and regular data mbufs.
905  */
906 #define	M_START(m)							\
907 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
908 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
909 	 &(m)->m_dat[0])
910 
911 /*
912  * Return the size of the buffer associated with an mbuf, handling external
913  * storage, packet-header mbufs, and regular data mbufs.
914  */
915 #define	M_SIZE(m)							\
916 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
917 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
918 	 MLEN)
919 
920 /*
921  * Set the m_data pointer of a newly allocated mbuf to place an object of the
922  * specified size at the end of the mbuf, longword aligned.
923  *
924  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
925  * separate macros, each asserting that it was called at the proper moment.
926  * This required callers to themselves test the storage type and call the
927  * right one.  Rather than require callers to be aware of those layout
928  * decisions, we centralize here.
929  */
930 static __inline void
931 m_align(struct mbuf *m, int len)
932 {
933 #ifdef INVARIANTS
934 	const char *msg = "%s: not a virgin mbuf";
935 #endif
936 	int adjust;
937 
938 	KASSERT(m->m_data == M_START(m), (msg, __func__));
939 
940 	adjust = M_SIZE(m) - len;
941 	m->m_data += adjust &~ (sizeof(long)-1);
942 }
943 
944 #define	M_ALIGN(m, len)		m_align(m, len)
945 #define	MH_ALIGN(m, len)	m_align(m, len)
946 #define	MEXT_ALIGN(m, len)	m_align(m, len)
947 
948 /*
949  * Compute the amount of space available before the current start of data in
950  * an mbuf.
951  *
952  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
953  * of checking writability of the mbuf data area rests solely with the caller.
954  *
955  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
956  * for mbufs with external storage.  We now allow mbuf-embedded data to be
957  * read-only as well.
958  */
959 #define	M_LEADINGSPACE(m)						\
960 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
961 
962 /*
963  * Compute the amount of space available after the end of data in an mbuf.
964  *
965  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
966  * of checking writability of the mbuf data area rests solely with the caller.
967  *
968  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
969  * for mbufs with external storage.  We now allow mbuf-embedded data to be
970  * read-only as well.
971  */
972 #define	M_TRAILINGSPACE(m)						\
973 	(M_WRITABLE(m) ?						\
974 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
975 
976 /*
977  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
978  * allocated, how specifies whether to wait.  If the allocation fails, the
979  * original mbuf chain is freed and m is set to NULL.
980  */
981 #define	M_PREPEND(m, plen, how) do {					\
982 	struct mbuf **_mmp = &(m);					\
983 	struct mbuf *_mm = *_mmp;					\
984 	int _mplen = (plen);						\
985 	int __mhow = (how);						\
986 									\
987 	MBUF_CHECKSLEEP(how);						\
988 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
989 		_mm->m_data -= _mplen;					\
990 		_mm->m_len += _mplen;					\
991 	} else								\
992 		_mm = m_prepend(_mm, _mplen, __mhow);			\
993 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
994 		_mm->m_pkthdr.len += _mplen;				\
995 	*_mmp = _mm;							\
996 } while (0)
997 
998 /*
999  * Change mbuf to new type.  This is a relatively expensive operation and
1000  * should be avoided.
1001  */
1002 #define	MCHTYPE(m, t)	m_chtype((m), (t))
1003 
1004 /* Return the rcvif of a packet header. */
1005 static __inline struct ifnet *
1006 m_rcvif(struct mbuf *m)
1007 {
1008 
1009 	M_ASSERTPKTHDR(m);
1010 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1011 		return (NULL);
1012 	return (m->m_pkthdr.rcvif);
1013 }
1014 
1015 /* Length to m_copy to copy all. */
1016 #define	M_COPYALL	1000000000
1017 
1018 extern int		max_datalen;	/* MHLEN - max_hdr */
1019 extern int		max_hdr;	/* Largest link + protocol header */
1020 extern int		max_linkhdr;	/* Largest link-level header */
1021 extern int		max_protohdr;	/* Largest protocol header */
1022 extern int		nmbclusters;	/* Maximum number of clusters */
1023 
1024 /*-
1025  * Network packets may have annotations attached by affixing a list of
1026  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1027  * allocated semi-opaque data structures that have a fixed header
1028  * (struct m_tag) that specifies the size of the memory block and a
1029  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1030  * unsigned value used to identify a module or ABI.  By convention this value
1031  * is chosen as the date+time that the module is created, expressed as the
1032  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1033  * value is an ABI/module-specific value that identifies a particular
1034  * annotation and is private to the module.  For compatibility with systems
1035  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1036  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1037  * compatibility shim functions and several tag types are defined below.
1038  * Users that do not require compatibility should use a private cookie value
1039  * so that packet tag-related definitions can be maintained privately.
1040  *
1041  * Note that the packet tag returned by m_tag_alloc has the default memory
1042  * alignment implemented by malloc.  To reference private data one can use a
1043  * construct like:
1044  *
1045  *	struct m_tag *mtag = m_tag_alloc(...);
1046  *	struct foo *p = (struct foo *)(mtag+1);
1047  *
1048  * if the alignment of struct m_tag is sufficient for referencing members of
1049  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1050  * private data structure to insure proper alignment; e.g.,
1051  *
1052  *	struct foo {
1053  *		struct m_tag	tag;
1054  *		...
1055  *	};
1056  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1057  *	struct m_tag *mtag = &p->tag;
1058  */
1059 
1060 /*
1061  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1062  * tags are expected to ``vanish'' when they pass through a network
1063  * interface.  For most interfaces this happens normally as the tags are
1064  * reclaimed when the mbuf is free'd.  However in some special cases
1065  * reclaiming must be done manually.  An example is packets that pass through
1066  * the loopback interface.  Also, one must be careful to do this when
1067  * ``turning around'' packets (e.g., icmp_reflect).
1068  *
1069  * To mark a tag persistent bit-or this flag in when defining the tag id.
1070  * The tag will then be treated as described above.
1071  */
1072 #define	MTAG_PERSISTENT				0x800
1073 
1074 #define	PACKET_TAG_NONE				0  /* Nadda */
1075 
1076 /* Packet tags for use with PACKET_ABI_COMPAT. */
1077 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1078 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1079 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1080 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1081 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1082 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1083 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1084 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1085 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1086 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1087 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1088 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1089 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1090 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1091 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1092 #define	PACKET_TAG_DIVERT			17 /* divert info */
1093 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1094 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1095 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1096 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1097 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1098 #define	PACKET_TAG_CARP				28 /* CARP info */
1099 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1100 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1101 
1102 /* Specific cookies and tags. */
1103 
1104 /* Packet tag routines. */
1105 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1106 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1107 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1108 void		 m_tag_free_default(struct m_tag *);
1109 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1110 struct m_tag	*m_tag_copy(struct m_tag *, int);
1111 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1112 void		 m_tag_delete_nonpersistent(struct mbuf *);
1113 
1114 /*
1115  * Initialize the list of tags associated with an mbuf.
1116  */
1117 static __inline void
1118 m_tag_init(struct mbuf *m)
1119 {
1120 
1121 	SLIST_INIT(&m->m_pkthdr.tags);
1122 }
1123 
1124 /*
1125  * Set up the contents of a tag.  Note that this does not fill in the free
1126  * method; the caller is expected to do that.
1127  *
1128  * XXX probably should be called m_tag_init, but that was already taken.
1129  */
1130 static __inline void
1131 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1132 {
1133 
1134 	t->m_tag_id = type;
1135 	t->m_tag_len = len;
1136 	t->m_tag_cookie = cookie;
1137 }
1138 
1139 /*
1140  * Reclaim resources associated with a tag.
1141  */
1142 static __inline void
1143 m_tag_free(struct m_tag *t)
1144 {
1145 
1146 	(*t->m_tag_free)(t);
1147 }
1148 
1149 /*
1150  * Return the first tag associated with an mbuf.
1151  */
1152 static __inline struct m_tag *
1153 m_tag_first(struct mbuf *m)
1154 {
1155 
1156 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1157 }
1158 
1159 /*
1160  * Return the next tag in the list of tags associated with an mbuf.
1161  */
1162 static __inline struct m_tag *
1163 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1164 {
1165 
1166 	return (SLIST_NEXT(t, m_tag_link));
1167 }
1168 
1169 /*
1170  * Prepend a tag to the list of tags associated with an mbuf.
1171  */
1172 static __inline void
1173 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1174 {
1175 
1176 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1177 }
1178 
1179 /*
1180  * Unlink a tag from the list of tags associated with an mbuf.
1181  */
1182 static __inline void
1183 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1184 {
1185 
1186 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1187 }
1188 
1189 /* These are for OpenBSD compatibility. */
1190 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1191 
1192 static __inline struct m_tag *
1193 m_tag_get(int type, int length, int wait)
1194 {
1195 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1196 }
1197 
1198 static __inline struct m_tag *
1199 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1200 {
1201 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1202 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1203 }
1204 
1205 static inline struct m_snd_tag *
1206 m_snd_tag_ref(struct m_snd_tag *mst)
1207 {
1208 
1209 	refcount_acquire(&mst->refcount);
1210 	return (mst);
1211 }
1212 
1213 static inline void
1214 m_snd_tag_rele(struct m_snd_tag *mst)
1215 {
1216 
1217 	if (refcount_release(&mst->refcount))
1218 		m_snd_tag_destroy(mst);
1219 }
1220 
1221 static __inline struct mbuf *
1222 m_free(struct mbuf *m)
1223 {
1224 	struct mbuf *n = m->m_next;
1225 
1226 	MBUF_PROBE1(m__free, m);
1227 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1228 		m_tag_delete_chain(m, NULL);
1229 	if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1230 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
1231 	if (m->m_flags & M_EXT)
1232 		mb_free_ext(m);
1233 	else if ((m->m_flags & M_NOFREE) == 0)
1234 		uma_zfree(zone_mbuf, m);
1235 	return (n);
1236 }
1237 
1238 static __inline int
1239 rt_m_getfib(struct mbuf *m)
1240 {
1241 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1242 	return (m->m_pkthdr.fibnum);
1243 }
1244 
1245 #define M_GETFIB(_m)   rt_m_getfib(_m)
1246 
1247 #define M_SETFIB(_m, _fib) do {						\
1248         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1249 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1250 } while (0)
1251 
1252 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1253 #define	MBUF_HASHFLAG_L2	(1 << 2)
1254 #define	MBUF_HASHFLAG_L3	(1 << 3)
1255 #define	MBUF_HASHFLAG_L4	(1 << 4)
1256 
1257 /* mbuf hashing helper routines */
1258 uint32_t	m_ether_tcpip_hash_init(void);
1259 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1260 
1261 #ifdef MBUF_PROFILING
1262  void m_profile(struct mbuf *m);
1263  #define M_PROFILE(m) m_profile(m)
1264 #else
1265  #define M_PROFILE(m)
1266 #endif
1267 
1268 struct mbufq {
1269 	STAILQ_HEAD(, mbuf)	mq_head;
1270 	int			mq_len;
1271 	int			mq_maxlen;
1272 };
1273 
1274 static inline void
1275 mbufq_init(struct mbufq *mq, int maxlen)
1276 {
1277 
1278 	STAILQ_INIT(&mq->mq_head);
1279 	mq->mq_maxlen = maxlen;
1280 	mq->mq_len = 0;
1281 }
1282 
1283 static inline struct mbuf *
1284 mbufq_flush(struct mbufq *mq)
1285 {
1286 	struct mbuf *m;
1287 
1288 	m = STAILQ_FIRST(&mq->mq_head);
1289 	STAILQ_INIT(&mq->mq_head);
1290 	mq->mq_len = 0;
1291 	return (m);
1292 }
1293 
1294 static inline void
1295 mbufq_drain(struct mbufq *mq)
1296 {
1297 	struct mbuf *m, *n;
1298 
1299 	n = mbufq_flush(mq);
1300 	while ((m = n) != NULL) {
1301 		n = STAILQ_NEXT(m, m_stailqpkt);
1302 		m_freem(m);
1303 	}
1304 }
1305 
1306 static inline struct mbuf *
1307 mbufq_first(const struct mbufq *mq)
1308 {
1309 
1310 	return (STAILQ_FIRST(&mq->mq_head));
1311 }
1312 
1313 static inline struct mbuf *
1314 mbufq_last(const struct mbufq *mq)
1315 {
1316 
1317 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1318 }
1319 
1320 static inline int
1321 mbufq_full(const struct mbufq *mq)
1322 {
1323 
1324 	return (mq->mq_len >= mq->mq_maxlen);
1325 }
1326 
1327 static inline int
1328 mbufq_len(const struct mbufq *mq)
1329 {
1330 
1331 	return (mq->mq_len);
1332 }
1333 
1334 static inline int
1335 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1336 {
1337 
1338 	if (mbufq_full(mq))
1339 		return (ENOBUFS);
1340 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1341 	mq->mq_len++;
1342 	return (0);
1343 }
1344 
1345 static inline struct mbuf *
1346 mbufq_dequeue(struct mbufq *mq)
1347 {
1348 	struct mbuf *m;
1349 
1350 	m = STAILQ_FIRST(&mq->mq_head);
1351 	if (m) {
1352 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1353 		m->m_nextpkt = NULL;
1354 		mq->mq_len--;
1355 	}
1356 	return (m);
1357 }
1358 
1359 static inline void
1360 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1361 {
1362 
1363 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1364 	mq->mq_len++;
1365 }
1366 
1367 /*
1368  * Note: this doesn't enforce the maximum list size for dst.
1369  */
1370 static inline void
1371 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1372 {
1373 
1374 	mq_dst->mq_len += mq_src->mq_len;
1375 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1376 	mq_src->mq_len = 0;
1377 }
1378 
1379 #ifdef _SYS_TIMESPEC_H_
1380 static inline void
1381 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1382 {
1383 
1384 	KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1385 	KASSERT((m->m_flags & M_TSTMP) != 0, ("mbuf %p no M_TSTMP", m));
1386 	ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1387 	ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1388 }
1389 #endif
1390 
1391 #ifdef NETDUMP
1392 /* Invoked from the netdump client code. */
1393 void	netdump_mbuf_drain(void);
1394 void	netdump_mbuf_dump(void);
1395 void	netdump_mbuf_reinit(int nmbuf, int nclust, int clsize);
1396 #endif
1397 
1398 #endif /* _KERNEL */
1399 #endif /* !_SYS_MBUF_H_ */
1400