xref: /freebsd/sys/sys/mbuf.h (revision cddbc3b40812213ff00041f79174cac0be360a2a)
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 <vm/uma.h>
44 #ifdef WITNESS
45 #include <sys/lock.h>
46 #endif
47 #endif
48 
49 #ifdef _KERNEL
50 #include <sys/sdt.h>
51 
52 #define	MBUF_PROBE1(probe, arg0)					\
53 	SDT_PROBE1(sdt, , , probe, arg0)
54 #define	MBUF_PROBE2(probe, arg0, arg1)					\
55 	SDT_PROBE2(sdt, , , probe, arg0, arg1)
56 #define	MBUF_PROBE3(probe, arg0, arg1, arg2)				\
57 	SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
58 #define	MBUF_PROBE4(probe, arg0, arg1, arg2, arg3)			\
59 	SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
60 #define	MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4)		\
61 	SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
62 
63 SDT_PROBE_DECLARE(sdt, , , m__init);
64 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
65 SDT_PROBE_DECLARE(sdt, , , m__get);
66 SDT_PROBE_DECLARE(sdt, , , m__getcl);
67 SDT_PROBE_DECLARE(sdt, , , m__clget);
68 SDT_PROBE_DECLARE(sdt, , , m__cljget);
69 SDT_PROBE_DECLARE(sdt, , , m__cljset);
70 SDT_PROBE_DECLARE(sdt, , , m__free);
71 SDT_PROBE_DECLARE(sdt, , , m__freem);
72 
73 #endif /* _KERNEL */
74 
75 /*
76  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
77  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
78  * sys/param.h), which has no additional overhead and is used instead of the
79  * internal data area; this is done when at least MINCLSIZE of data must be
80  * stored.  Additionally, it is possible to allocate a separate buffer
81  * externally and attach it to the mbuf in a way similar to that of mbuf
82  * clusters.
83  *
84  * NB: These calculation do not take actual compiler-induced alignment and
85  * padding inside the complete struct mbuf into account.  Appropriate
86  * attention is required when changing members of struct mbuf.
87  *
88  * MLEN is data length in a normal mbuf.
89  * MHLEN is data length in an mbuf with pktheader.
90  * MINCLSIZE is a smallest amount of data that should be put into cluster.
91  *
92  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
93  * they are sensible.
94  */
95 struct mbuf;
96 #define	MHSIZE		offsetof(struct mbuf, m_dat)
97 #define	MPKTHSIZE	offsetof(struct mbuf, m_pktdat)
98 #define	MLEN		((int)(MSIZE - MHSIZE))
99 #define	MHLEN		((int)(MSIZE - MPKTHSIZE))
100 #define	MINCLSIZE	(MHLEN + 1)
101 #define	M_NODOM		255
102 
103 #ifdef _KERNEL
104 /*-
105  * Macro for type conversion: convert mbuf pointer to data pointer of correct
106  * type:
107  *
108  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
109  * mtodo(m, o) -- Same as above but with offset 'o' into data.
110  */
111 #define	mtod(m, t)	((t)((m)->m_data))
112 #define	mtodo(m, o)	((void *)(((m)->m_data) + (o)))
113 
114 /*
115  * Argument structure passed to UMA routines during mbuf and packet
116  * allocations.
117  */
118 struct mb_args {
119 	int	flags;	/* Flags for mbuf being allocated */
120 	short	type;	/* Type of mbuf being allocated */
121 };
122 #endif /* _KERNEL */
123 
124 /*
125  * Packet tag structure (see below for details).
126  */
127 struct m_tag {
128 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
129 	u_int16_t		m_tag_id;	/* Tag ID */
130 	u_int16_t		m_tag_len;	/* Length of data */
131 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
132 	void			(*m_tag_free)(struct m_tag *);
133 };
134 
135 /*
136  * Static network interface owned tag.
137  * Allocated through ifp->if_snd_tag_alloc().
138  */
139 struct m_snd_tag {
140 	struct ifnet *ifp;		/* network interface tag belongs to */
141 };
142 
143 /*
144  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
145  * Size ILP32: 48
146  *	 LP64: 56
147  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
148  * they are correct.
149  */
150 struct pkthdr {
151 	union {
152 		struct m_snd_tag *snd_tag;	/* send tag, if any */
153 		struct ifnet	*rcvif;		/* rcv interface */
154 	};
155 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
156 	int32_t		 len;		/* total packet length */
157 
158 	/* Layer crossing persistent information. */
159 	uint32_t	 flowid;	/* packet's 4-tuple system */
160 	uint32_t	 csum_flags;	/* checksum and offload features */
161 	uint16_t	 fibnum;	/* this packet should use this fib */
162 	uint8_t		 numa_domain;	/* NUMA domain of recvd pkt */
163 	uint8_t		 rsstype;	/* hash type */
164 	union {
165 		uint64_t	rcv_tstmp;	/* timestamp in ns */
166 		struct {
167 			uint8_t		 l2hlen;	/* layer 2 hdr len */
168 			uint8_t		 l3hlen;	/* layer 3 hdr len */
169 			uint8_t		 l4hlen;	/* layer 4 hdr len */
170 			uint8_t		 l5hlen;	/* layer 5 hdr len */
171 			uint32_t	 spare;
172 		};
173 	};
174 	union {
175 		uint8_t  eight[8];
176 		uint16_t sixteen[4];
177 		uint32_t thirtytwo[2];
178 		uint64_t sixtyfour[1];
179 		uintptr_t unintptr[1];
180 		void	*ptr;
181 	} PH_per;
182 
183 	/* Layer specific non-persistent local storage for reassembly, etc. */
184 	union {
185 		uint8_t  eight[8];
186 		uint16_t sixteen[4];
187 		uint32_t thirtytwo[2];
188 		uint64_t sixtyfour[1];
189 		uintptr_t unintptr[1];
190 		void 	*ptr;
191 	} PH_loc;
192 };
193 #define	ether_vtag	PH_per.sixteen[0]
194 #define	PH_vt		PH_per
195 #define	vt_nrecs	sixteen[0]
196 #define	tso_segsz	PH_per.sixteen[1]
197 #define	lro_nsegs	tso_segsz
198 #define	csum_phsum	PH_per.sixteen[2]
199 #define	csum_data	PH_per.thirtytwo[1]
200 #define pace_thoff	PH_loc.sixteen[0]
201 #define pace_tlen	PH_loc.sixteen[1]
202 #define pace_drphdrlen	PH_loc.sixteen[2]
203 #define pace_tos	PH_loc.eight[6]
204 #define pace_lock	PH_loc.eight[7]
205 
206 /*
207  * Description of external storage mapped into mbuf; valid only if M_EXT is
208  * set.
209  * Size ILP32: 28
210  *	 LP64: 48
211  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
212  * they are correct.
213  */
214 typedef	void m_ext_free_t(struct mbuf *);
215 struct m_ext {
216 	union {
217 		/*
218 		 * If EXT_FLAG_EMBREF is set, then we use refcount in the
219 		 * mbuf, the 'ext_count' member.  Otherwise, we have a
220 		 * shadow copy and we use pointer 'ext_cnt'.  The original
221 		 * mbuf is responsible to carry the pointer to free routine
222 		 * and its arguments.  They aren't copied into shadows in
223 		 * mb_dupcl() to avoid dereferencing next cachelines.
224 		 */
225 		volatile u_int	 ext_count;
226 		volatile u_int	*ext_cnt;
227 	};
228 	char		*ext_buf;	/* start of buffer */
229 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
230 	uint32_t	 ext_type:8,	/* type of external storage */
231 			 ext_flags:24;	/* external storage mbuf flags */
232 	/*
233 	 * Fields below store the free context for the external storage.
234 	 * They are valid only in the refcount carrying mbuf, the one with
235 	 * EXT_FLAG_EMBREF flag, with exclusion for EXT_EXTREF type, where
236 	 * the free context is copied into all mbufs that use same external
237 	 * storage.
238 	 */
239 #define	m_ext_copylen	offsetof(struct m_ext, ext_free)
240 	m_ext_free_t	*ext_free;	/* free routine if not the usual */
241 	void		*ext_arg1;	/* optional argument pointer */
242 	void		*ext_arg2;	/* optional argument pointer */
243 };
244 
245 /*
246  * The core of the mbuf object along with some shortcut defines for practical
247  * purposes.
248  */
249 struct mbuf {
250 	/*
251 	 * Header present at the beginning of every mbuf.
252 	 * Size ILP32: 24
253 	 *      LP64: 32
254 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
255 	 * that they are correct.
256 	 */
257 	union {	/* next buffer in chain */
258 		struct mbuf		*m_next;
259 		SLIST_ENTRY(mbuf)	m_slist;
260 		STAILQ_ENTRY(mbuf)	m_stailq;
261 	};
262 	union {	/* next chain in queue/record */
263 		struct mbuf		*m_nextpkt;
264 		SLIST_ENTRY(mbuf)	m_slistpkt;
265 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
266 	};
267 	caddr_t		 m_data;	/* location of data */
268 	int32_t		 m_len;		/* amount of data in this mbuf */
269 	uint32_t	 m_type:8,	/* type of data in this mbuf */
270 			 m_flags:24;	/* flags; see below */
271 #if !defined(__LP64__)
272 	uint32_t	 m_pad;		/* pad for 64bit alignment */
273 #endif
274 
275 	/*
276 	 * A set of optional headers (packet header, external storage header)
277 	 * and internal data storage.  Historically, these arrays were sized
278 	 * to MHLEN (space left after a packet header) and MLEN (space left
279 	 * after only a regular mbuf header); they are now variable size in
280 	 * order to support future work on variable-size mbufs.
281 	 */
282 	union {
283 		struct {
284 			struct pkthdr	m_pkthdr;	/* M_PKTHDR set */
285 			union {
286 				struct m_ext	m_ext;	/* M_EXT set */
287 				char		m_pktdat[0];
288 			};
289 		};
290 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
291 	};
292 };
293 
294 /*
295  * mbuf flags of global significance and layer crossing.
296  * Those of only protocol/layer specific significance are to be mapped
297  * to M_PROTO[1-12] and cleared at layer handoff boundaries.
298  * NB: Limited to the lower 24 bits.
299  */
300 #define	M_EXT		0x00000001 /* has associated external storage */
301 #define	M_PKTHDR	0x00000002 /* start of record */
302 #define	M_EOR		0x00000004 /* end of record */
303 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
304 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
305 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
306 #define	M_PROMISC	0x00000040 /* packet was not for us */
307 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
308 #define	M_NOMAP		0x00000100 /* mbuf data is unmapped (soon from Drew) */
309 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
310 #define	M_TSTMP		0x00000400 /* rcv_tstmp field is valid */
311 #define	M_TSTMP_HPREC	0x00000800 /* rcv_tstmp is high-prec, typically
312 				      hw-stamped on port (useful for IEEE 1588
313 				      and 802.1AS) */
314 
315 #define	M_PROTO1	0x00001000 /* protocol-specific */
316 #define	M_PROTO2	0x00002000 /* protocol-specific */
317 #define	M_PROTO3	0x00004000 /* protocol-specific */
318 #define	M_PROTO4	0x00008000 /* protocol-specific */
319 #define	M_PROTO5	0x00010000 /* protocol-specific */
320 #define	M_PROTO6	0x00020000 /* protocol-specific */
321 #define	M_PROTO7	0x00040000 /* protocol-specific */
322 #define	M_PROTO8	0x00080000 /* protocol-specific */
323 #define	M_PROTO9	0x00100000 /* protocol-specific */
324 #define	M_PROTO10	0x00200000 /* protocol-specific */
325 #define	M_PROTO11	0x00400000 /* protocol-specific */
326 #define	M_PROTO12	0x00800000 /* protocol-specific */
327 
328 #define MB_DTOR_SKIP	0x1	/* don't pollute the cache by touching a freed mbuf */
329 
330 /*
331  * Flags to purge when crossing layers.
332  */
333 #define	M_PROTOFLAGS \
334     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
335      M_PROTO9|M_PROTO10|M_PROTO11|M_PROTO12)
336 
337 /*
338  * Flags preserved when copying m_pkthdr.
339  */
340 #define M_COPYFLAGS \
341     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
342      M_TSTMP_HPREC|M_PROTOFLAGS)
343 
344 /*
345  * Mbuf flag description for use with printf(9) %b identifier.
346  */
347 #define	M_FLAG_BITS \
348     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
349     "\7M_PROMISC\10M_VLANTAG\13M_TSTMP\14M_TSTMP_HPREC"
350 #define	M_FLAG_PROTOBITS \
351     "\15M_PROTO1\16M_PROTO2\17M_PROTO3\20M_PROTO4\21M_PROTO5" \
352     "\22M_PROTO6\23M_PROTO7\24M_PROTO8\25M_PROTO9\26M_PROTO10" \
353     "\27M_PROTO11\30M_PROTO12"
354 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
355 
356 /*
357  * Network interface cards are able to hash protocol fields (such as IPv4
358  * addresses and TCP port numbers) classify packets into flows.  These flows
359  * can then be used to maintain ordering while delivering packets to the OS
360  * via parallel input queues, as well as to provide a stateless affinity
361  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
362  * m_flag fields to indicate how the hash should be interpreted by the
363  * network stack.
364  *
365  * Most NICs support RSS, which provides ordering and explicit affinity, and
366  * use the hash m_flag bits to indicate what header fields were covered by
367  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
368  * RSS cards or configurations that provide an opaque flow identifier, allowing
369  * for ordering and distribution without explicit affinity.  Additionally,
370  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
371  * properties.
372  *
373  * The meaning of the IPV6_EX suffix:
374  * "o  Home address from the home address option in the IPv6 destination
375  *     options header.  If the extension header is not present, use the Source
376  *     IPv6 Address.
377  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
378  *     associated extension header.  If the extension header is not present,
379  *     use the Destination IPv6 Address."
380  * Quoted from:
381  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
382  */
383 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
384 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
385 /* Microsoft RSS standard hash types */
386 #define	M_HASHTYPE_NONE			0
387 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
388 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
389 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
390 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
391 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
392 							    * ext hdrs */
393 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
394 							    * ext hdrs */
395 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
396 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
397 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
398 							    * ext hdrs */
399 
400 #define	M_HASHTYPE_OPAQUE		63	/* ordering, not affinity */
401 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
402 						/* ordering+hash, not affinity*/
403 
404 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
405 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype)
406 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
407 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
408 #define	M_HASHTYPE_ISHASH(m)	(M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP)
409 
410 /*
411  * External mbuf storage buffer types.
412  */
413 #define	EXT_CLUSTER	1	/* mbuf cluster */
414 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
415 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
416 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
417 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
418 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
419 #define	EXT_MBUF	7	/* external mbuf reference */
420 #define	EXT_RXRING	8	/* data in NIC receive ring */
421 
422 #define	EXT_VENDOR1	224	/* for vendor-internal use */
423 #define	EXT_VENDOR2	225	/* for vendor-internal use */
424 #define	EXT_VENDOR3	226	/* for vendor-internal use */
425 #define	EXT_VENDOR4	227	/* for vendor-internal use */
426 
427 #define	EXT_EXP1	244	/* for experimental use */
428 #define	EXT_EXP2	245	/* for experimental use */
429 #define	EXT_EXP3	246	/* for experimental use */
430 #define	EXT_EXP4	247	/* for experimental use */
431 
432 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
433 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
434 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
435 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
436 
437 /*
438  * Flags for external mbuf buffer types.
439  * NB: limited to the lower 24 bits.
440  */
441 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
442 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
443 
444 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
445 
446 #define	EXT_FLAG_VENDOR1	0x010000	/* These flags are vendor */
447 #define	EXT_FLAG_VENDOR2	0x020000	/* or submodule specific, */
448 #define	EXT_FLAG_VENDOR3	0x040000	/* not used by mbuf code. */
449 #define	EXT_FLAG_VENDOR4	0x080000	/* Set/read by submodule. */
450 
451 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
452 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
453 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
454 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
455 
456 /*
457  * EXT flag description for use with printf(9) %b identifier.
458  */
459 #define	EXT_FLAG_BITS \
460     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
461     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
462     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
463     "\30EXT_FLAG_EXP4"
464 
465 /*
466  * Flags indicating checksum, segmentation and other offload work to be
467  * done, or already done, by hardware or lower layers.  It is split into
468  * separate inbound and outbound flags.
469  *
470  * Outbound flags that are set by upper protocol layers requesting lower
471  * layers, or ideally the hardware, to perform these offloading tasks.
472  * For outbound packets this field and its flags can be directly tested
473  * against ifnet if_hwassist.
474  */
475 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
476 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
477 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
478 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
479 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
480 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
481 
482 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
483 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
484 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
485 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
486 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
487 
488 /* Inbound checksum support where the checksum was verified by hardware. */
489 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
490 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
491 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
492 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
493 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
494 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
495 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
496 
497 /*
498  * CSUM flag description for use with printf(9) %b identifier.
499  */
500 #define	CSUM_BITS \
501     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
502     "\6CSUM_IP_ISCSI" \
503     "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \
504     "\16CSUM_IP6_ISCSI" \
505     "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \
506     "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED"
507 
508 /* CSUM flags compatibility mappings. */
509 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
510 #define	CSUM_IP_VALID		CSUM_L3_VALID
511 #define	CSUM_DATA_VALID		CSUM_L4_VALID
512 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
513 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
514 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
515 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
516 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
517 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
518 #define	CSUM_TCP		CSUM_IP_TCP
519 #define	CSUM_UDP		CSUM_IP_UDP
520 #define	CSUM_SCTP		CSUM_IP_SCTP
521 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
522 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
523 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
524 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
525 
526 /*
527  * mbuf types describing the content of the mbuf (including external storage).
528  */
529 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
530 #define	MT_DATA		1	/* dynamic (data) allocation */
531 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
532 
533 #define	MT_VENDOR1	4	/* for vendor-internal use */
534 #define	MT_VENDOR2	5	/* for vendor-internal use */
535 #define	MT_VENDOR3	6	/* for vendor-internal use */
536 #define	MT_VENDOR4	7	/* for vendor-internal use */
537 
538 #define	MT_SONAME	8	/* socket name */
539 
540 #define	MT_EXP1		9	/* for experimental use */
541 #define	MT_EXP2		10	/* for experimental use */
542 #define	MT_EXP3		11	/* for experimental use */
543 #define	MT_EXP4		12	/* for experimental use */
544 
545 #define	MT_CONTROL	14	/* extra-data protocol message */
546 #define	MT_EXTCONTROL	15	/* control message with externalized contents */
547 #define	MT_OOBDATA	16	/* expedited data  */
548 
549 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
550 				   a non-initialized mbuf */
551 
552 /*
553  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
554  * !_KERNEL so that monitoring tools can look up the zones with
555  * libmemstat(3).
556  */
557 #define	MBUF_MEM_NAME		"mbuf"
558 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
559 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
560 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
561 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
562 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
563 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
564 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
565 
566 #ifdef _KERNEL
567 
568 #ifdef WITNESS
569 #define	MBUF_CHECKSLEEP(how) do {					\
570 	if (how == M_WAITOK)						\
571 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
572 		    "Sleeping in \"%s\"", __func__);			\
573 } while (0)
574 #else
575 #define	MBUF_CHECKSLEEP(how)
576 #endif
577 
578 /*
579  * Network buffer allocation API
580  *
581  * The rest of it is defined in kern/kern_mbuf.c
582  */
583 extern uma_zone_t	zone_mbuf;
584 extern uma_zone_t	zone_clust;
585 extern uma_zone_t	zone_pack;
586 extern uma_zone_t	zone_jumbop;
587 extern uma_zone_t	zone_jumbo9;
588 extern uma_zone_t	zone_jumbo16;
589 
590 void		 mb_dupcl(struct mbuf *, struct mbuf *);
591 void		 mb_free_ext(struct mbuf *);
592 void		 m_adj(struct mbuf *, int);
593 int		 m_apply(struct mbuf *, int, int,
594 		    int (*)(void *, void *, u_int), void *);
595 int		 m_append(struct mbuf *, int, c_caddr_t);
596 void		 m_cat(struct mbuf *, struct mbuf *);
597 void		 m_catpkt(struct mbuf *, struct mbuf *);
598 int		 m_clget(struct mbuf *m, int how);
599 void 		*m_cljget(struct mbuf *m, int how, int size);
600 struct mbuf	*m_collapse(struct mbuf *, int, int);
601 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
602 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
603 struct mbuf	*m_copym(struct mbuf *, int, int, int);
604 struct mbuf	*m_copypacket(struct mbuf *, int);
605 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
606 struct mbuf	*m_copyup(struct mbuf *, int, int);
607 struct mbuf	*m_defrag(struct mbuf *, int);
608 void		 m_demote_pkthdr(struct mbuf *);
609 void		 m_demote(struct mbuf *, int, int);
610 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
611 		    void (*)(char *, caddr_t, u_int));
612 void		 m_dispose_extcontrolm(struct mbuf *m);
613 struct mbuf	*m_dup(const struct mbuf *, int);
614 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
615 void		 m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
616 		    void *, void *, int, int);
617 u_int		 m_fixhdr(struct mbuf *);
618 struct mbuf	*m_fragment(struct mbuf *, int, int);
619 void		 m_freem(struct mbuf *);
620 struct mbuf	*m_get2(int, int, short, int);
621 struct mbuf	*m_getjcl(int, short, int, int);
622 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
623 struct mbuf	*m_getptr(struct mbuf *, int, int *);
624 u_int		 m_length(struct mbuf *, struct mbuf **);
625 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
626 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
627 int		 m_pkthdr_init(struct mbuf *, int);
628 struct mbuf	*m_prepend(struct mbuf *, int, int);
629 void		 m_print(const struct mbuf *, int);
630 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
631 struct mbuf	*m_pullup(struct mbuf *, int);
632 int		 m_sanity(struct mbuf *, int);
633 struct mbuf	*m_split(struct mbuf *, int, int);
634 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
635 struct mbuf	*m_unshare(struct mbuf *, int);
636 
637 static __inline int
638 m_gettype(int size)
639 {
640 	int type;
641 
642 	switch (size) {
643 	case MSIZE:
644 		type = EXT_MBUF;
645 		break;
646 	case MCLBYTES:
647 		type = EXT_CLUSTER;
648 		break;
649 #if MJUMPAGESIZE != MCLBYTES
650 	case MJUMPAGESIZE:
651 		type = EXT_JUMBOP;
652 		break;
653 #endif
654 	case MJUM9BYTES:
655 		type = EXT_JUMBO9;
656 		break;
657 	case MJUM16BYTES:
658 		type = EXT_JUMBO16;
659 		break;
660 	default:
661 		panic("%s: invalid cluster size %d", __func__, size);
662 	}
663 
664 	return (type);
665 }
666 
667 /*
668  * Associated an external reference counted buffer with an mbuf.
669  */
670 static __inline void
671 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
672     m_ext_free_t freef, void *arg1, void *arg2)
673 {
674 
675 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
676 
677 	atomic_add_int(ref_cnt, 1);
678 	m->m_flags |= M_EXT;
679 	m->m_ext.ext_buf = buf;
680 	m->m_ext.ext_cnt = ref_cnt;
681 	m->m_data = m->m_ext.ext_buf;
682 	m->m_ext.ext_size = size;
683 	m->m_ext.ext_free = freef;
684 	m->m_ext.ext_arg1 = arg1;
685 	m->m_ext.ext_arg2 = arg2;
686 	m->m_ext.ext_type = EXT_EXTREF;
687 	m->m_ext.ext_flags = 0;
688 }
689 
690 static __inline uma_zone_t
691 m_getzone(int size)
692 {
693 	uma_zone_t zone;
694 
695 	switch (size) {
696 	case MCLBYTES:
697 		zone = zone_clust;
698 		break;
699 #if MJUMPAGESIZE != MCLBYTES
700 	case MJUMPAGESIZE:
701 		zone = zone_jumbop;
702 		break;
703 #endif
704 	case MJUM9BYTES:
705 		zone = zone_jumbo9;
706 		break;
707 	case MJUM16BYTES:
708 		zone = zone_jumbo16;
709 		break;
710 	default:
711 		panic("%s: invalid cluster size %d", __func__, size);
712 	}
713 
714 	return (zone);
715 }
716 
717 /*
718  * Initialize an mbuf with linear storage.
719  *
720  * Inline because the consumer text overhead will be roughly the same to
721  * initialize or call a function with this many parameters and M_PKTHDR
722  * should go away with constant propagation for !MGETHDR.
723  */
724 static __inline int
725 m_init(struct mbuf *m, int how, short type, int flags)
726 {
727 	int error;
728 
729 	m->m_next = NULL;
730 	m->m_nextpkt = NULL;
731 	m->m_data = m->m_dat;
732 	m->m_len = 0;
733 	m->m_flags = flags;
734 	m->m_type = type;
735 	if (flags & M_PKTHDR)
736 		error = m_pkthdr_init(m, how);
737 	else
738 		error = 0;
739 
740 	MBUF_PROBE5(m__init, m, how, type, flags, error);
741 	return (error);
742 }
743 
744 static __inline struct mbuf *
745 m_get(int how, short type)
746 {
747 	struct mbuf *m;
748 	struct mb_args args;
749 
750 	args.flags = 0;
751 	args.type = type;
752 	m = uma_zalloc_arg(zone_mbuf, &args, how);
753 	MBUF_PROBE3(m__get, how, type, m);
754 	return (m);
755 }
756 
757 static __inline struct mbuf *
758 m_gethdr(int how, short type)
759 {
760 	struct mbuf *m;
761 	struct mb_args args;
762 
763 	args.flags = M_PKTHDR;
764 	args.type = type;
765 	m = uma_zalloc_arg(zone_mbuf, &args, how);
766 	MBUF_PROBE3(m__gethdr, how, type, m);
767 	return (m);
768 }
769 
770 static __inline struct mbuf *
771 m_getcl(int how, short type, int flags)
772 {
773 	struct mbuf *m;
774 	struct mb_args args;
775 
776 	args.flags = flags;
777 	args.type = type;
778 	m = uma_zalloc_arg(zone_pack, &args, how);
779 	MBUF_PROBE4(m__getcl, how, type, flags, m);
780 	return (m);
781 }
782 
783 /*
784  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
785  * unreferenced cluster to an mbuf(9).  It is not possible to assert
786  * that, so care can be taken only by users of the API.
787  */
788 static __inline void
789 m_cljset(struct mbuf *m, void *cl, int type)
790 {
791 	int size;
792 
793 	switch (type) {
794 	case EXT_CLUSTER:
795 		size = MCLBYTES;
796 		break;
797 #if MJUMPAGESIZE != MCLBYTES
798 	case EXT_JUMBOP:
799 		size = MJUMPAGESIZE;
800 		break;
801 #endif
802 	case EXT_JUMBO9:
803 		size = MJUM9BYTES;
804 		break;
805 	case EXT_JUMBO16:
806 		size = MJUM16BYTES;
807 		break;
808 	default:
809 		panic("%s: unknown cluster type %d", __func__, type);
810 		break;
811 	}
812 
813 	m->m_data = m->m_ext.ext_buf = cl;
814 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
815 	m->m_ext.ext_size = size;
816 	m->m_ext.ext_type = type;
817 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
818 	m->m_ext.ext_count = 1;
819 	m->m_flags |= M_EXT;
820 	MBUF_PROBE3(m__cljset, m, cl, type);
821 }
822 
823 static __inline void
824 m_chtype(struct mbuf *m, short new_type)
825 {
826 
827 	m->m_type = new_type;
828 }
829 
830 static __inline void
831 m_clrprotoflags(struct mbuf *m)
832 {
833 
834 	while (m) {
835 		m->m_flags &= ~M_PROTOFLAGS;
836 		m = m->m_next;
837 	}
838 }
839 
840 static __inline struct mbuf *
841 m_last(struct mbuf *m)
842 {
843 
844 	while (m->m_next)
845 		m = m->m_next;
846 	return (m);
847 }
848 
849 static inline u_int
850 m_extrefcnt(struct mbuf *m)
851 {
852 
853 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
854 
855 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
856 	    *m->m_ext.ext_cnt);
857 }
858 
859 /*
860  * mbuf, cluster, and external object allocation macros (for compatibility
861  * purposes).
862  */
863 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
864 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
865 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
866 #define	MCLGET(m, how)		m_clget((m), (how))
867 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
868     m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2),	\
869     (flags), (type))
870 #define	m_getm(m, len, how, type)					\
871     m_getm2((m), (len), (how), (type), M_PKTHDR)
872 
873 /*
874  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
875  * be both the local data payload, or an external buffer area, depending on
876  * whether M_EXT is set).
877  */
878 #define	M_WRITABLE(m)	(!((m)->m_flags & M_RDONLY) &&			\
879 			 (!(((m)->m_flags & M_EXT)) ||			\
880 			 (m_extrefcnt(m) == 1)))
881 
882 /* Check if the supplied mbuf has a packet header, or else panic. */
883 #define	M_ASSERTPKTHDR(m)						\
884 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
885 	    ("%s: no mbuf packet header!", __func__))
886 
887 /*
888  * Ensure that the supplied mbuf is a valid, non-free mbuf.
889  *
890  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
891  */
892 #define	M_ASSERTVALID(m)						\
893 	KASSERT((((struct mbuf *)m)->m_flags & 0) == 0,			\
894 	    ("%s: attempted use of a free mbuf!", __func__))
895 
896 /*
897  * Return the address of the start of the buffer associated with an mbuf,
898  * handling external storage, packet-header mbufs, and regular data mbufs.
899  */
900 #define	M_START(m)							\
901 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
902 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
903 	 &(m)->m_dat[0])
904 
905 /*
906  * Return the size of the buffer associated with an mbuf, handling external
907  * storage, packet-header mbufs, and regular data mbufs.
908  */
909 #define	M_SIZE(m)							\
910 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
911 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
912 	 MLEN)
913 
914 /*
915  * Set the m_data pointer of a newly allocated mbuf to place an object of the
916  * specified size at the end of the mbuf, longword aligned.
917  *
918  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
919  * separate macros, each asserting that it was called at the proper moment.
920  * This required callers to themselves test the storage type and call the
921  * right one.  Rather than require callers to be aware of those layout
922  * decisions, we centralize here.
923  */
924 static __inline void
925 m_align(struct mbuf *m, int len)
926 {
927 #ifdef INVARIANTS
928 	const char *msg = "%s: not a virgin mbuf";
929 #endif
930 	int adjust;
931 
932 	KASSERT(m->m_data == M_START(m), (msg, __func__));
933 
934 	adjust = M_SIZE(m) - len;
935 	m->m_data += adjust &~ (sizeof(long)-1);
936 }
937 
938 #define	M_ALIGN(m, len)		m_align(m, len)
939 #define	MH_ALIGN(m, len)	m_align(m, len)
940 #define	MEXT_ALIGN(m, len)	m_align(m, len)
941 
942 /*
943  * Compute the amount of space available before the current start of data in
944  * an mbuf.
945  *
946  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
947  * of checking writability of the mbuf data area rests solely with the caller.
948  *
949  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
950  * for mbufs with external storage.  We now allow mbuf-embedded data to be
951  * read-only as well.
952  */
953 #define	M_LEADINGSPACE(m)						\
954 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
955 
956 /*
957  * Compute the amount of space available after the end of data in an mbuf.
958  *
959  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
960  * of checking writability of the mbuf data area rests solely with the caller.
961  *
962  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
963  * for mbufs with external storage.  We now allow mbuf-embedded data to be
964  * read-only as well.
965  */
966 #define	M_TRAILINGSPACE(m)						\
967 	(M_WRITABLE(m) ?						\
968 	    ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
969 
970 /*
971  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
972  * allocated, how specifies whether to wait.  If the allocation fails, the
973  * original mbuf chain is freed and m is set to NULL.
974  */
975 #define	M_PREPEND(m, plen, how) do {					\
976 	struct mbuf **_mmp = &(m);					\
977 	struct mbuf *_mm = *_mmp;					\
978 	int _mplen = (plen);						\
979 	int __mhow = (how);						\
980 									\
981 	MBUF_CHECKSLEEP(how);						\
982 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
983 		_mm->m_data -= _mplen;					\
984 		_mm->m_len += _mplen;					\
985 	} else								\
986 		_mm = m_prepend(_mm, _mplen, __mhow);			\
987 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
988 		_mm->m_pkthdr.len += _mplen;				\
989 	*_mmp = _mm;							\
990 } while (0)
991 
992 /*
993  * Change mbuf to new type.  This is a relatively expensive operation and
994  * should be avoided.
995  */
996 #define	MCHTYPE(m, t)	m_chtype((m), (t))
997 
998 /* Length to m_copy to copy all. */
999 #define	M_COPYALL	1000000000
1000 
1001 extern int		max_datalen;	/* MHLEN - max_hdr */
1002 extern int		max_hdr;	/* Largest link + protocol header */
1003 extern int		max_linkhdr;	/* Largest link-level header */
1004 extern int		max_protohdr;	/* Largest protocol header */
1005 extern int		nmbclusters;	/* Maximum number of clusters */
1006 
1007 /*-
1008  * Network packets may have annotations attached by affixing a list of
1009  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1010  * allocated semi-opaque data structures that have a fixed header
1011  * (struct m_tag) that specifies the size of the memory block and a
1012  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1013  * unsigned value used to identify a module or ABI.  By convention this value
1014  * is chosen as the date+time that the module is created, expressed as the
1015  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1016  * value is an ABI/module-specific value that identifies a particular
1017  * annotation and is private to the module.  For compatibility with systems
1018  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1019  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1020  * compatibility shim functions and several tag types are defined below.
1021  * Users that do not require compatibility should use a private cookie value
1022  * so that packet tag-related definitions can be maintained privately.
1023  *
1024  * Note that the packet tag returned by m_tag_alloc has the default memory
1025  * alignment implemented by malloc.  To reference private data one can use a
1026  * construct like:
1027  *
1028  *	struct m_tag *mtag = m_tag_alloc(...);
1029  *	struct foo *p = (struct foo *)(mtag+1);
1030  *
1031  * if the alignment of struct m_tag is sufficient for referencing members of
1032  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1033  * private data structure to insure proper alignment; e.g.,
1034  *
1035  *	struct foo {
1036  *		struct m_tag	tag;
1037  *		...
1038  *	};
1039  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1040  *	struct m_tag *mtag = &p->tag;
1041  */
1042 
1043 /*
1044  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1045  * tags are expected to ``vanish'' when they pass through a network
1046  * interface.  For most interfaces this happens normally as the tags are
1047  * reclaimed when the mbuf is free'd.  However in some special cases
1048  * reclaiming must be done manually.  An example is packets that pass through
1049  * the loopback interface.  Also, one must be careful to do this when
1050  * ``turning around'' packets (e.g., icmp_reflect).
1051  *
1052  * To mark a tag persistent bit-or this flag in when defining the tag id.
1053  * The tag will then be treated as described above.
1054  */
1055 #define	MTAG_PERSISTENT				0x800
1056 
1057 #define	PACKET_TAG_NONE				0  /* Nadda */
1058 
1059 /* Packet tags for use with PACKET_ABI_COMPAT. */
1060 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1061 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1062 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1063 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1064 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1065 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1066 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1067 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1068 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1069 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1070 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1071 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1072 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1073 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1074 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1075 #define	PACKET_TAG_DIVERT			17 /* divert info */
1076 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1077 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1078 #define	PACKET_TAG_PF		(21 | MTAG_PERSISTENT) /* PF/ALTQ information */
1079 #define	PACKET_TAG_RTSOCKFAM			25 /* rtsock sa family */
1080 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1081 #define	PACKET_TAG_CARP				28 /* CARP info */
1082 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1083 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1084 
1085 /* Specific cookies and tags. */
1086 
1087 /* Packet tag routines. */
1088 struct m_tag	*m_tag_alloc(u_int32_t, int, int, int);
1089 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1090 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1091 void		 m_tag_free_default(struct m_tag *);
1092 struct m_tag	*m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *);
1093 struct m_tag	*m_tag_copy(struct m_tag *, int);
1094 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1095 void		 m_tag_delete_nonpersistent(struct mbuf *);
1096 
1097 /*
1098  * Initialize the list of tags associated with an mbuf.
1099  */
1100 static __inline void
1101 m_tag_init(struct mbuf *m)
1102 {
1103 
1104 	SLIST_INIT(&m->m_pkthdr.tags);
1105 }
1106 
1107 /*
1108  * Set up the contents of a tag.  Note that this does not fill in the free
1109  * method; the caller is expected to do that.
1110  *
1111  * XXX probably should be called m_tag_init, but that was already taken.
1112  */
1113 static __inline void
1114 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len)
1115 {
1116 
1117 	t->m_tag_id = type;
1118 	t->m_tag_len = len;
1119 	t->m_tag_cookie = cookie;
1120 }
1121 
1122 /*
1123  * Reclaim resources associated with a tag.
1124  */
1125 static __inline void
1126 m_tag_free(struct m_tag *t)
1127 {
1128 
1129 	(*t->m_tag_free)(t);
1130 }
1131 
1132 /*
1133  * Return the first tag associated with an mbuf.
1134  */
1135 static __inline struct m_tag *
1136 m_tag_first(struct mbuf *m)
1137 {
1138 
1139 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1140 }
1141 
1142 /*
1143  * Return the next tag in the list of tags associated with an mbuf.
1144  */
1145 static __inline struct m_tag *
1146 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1147 {
1148 
1149 	return (SLIST_NEXT(t, m_tag_link));
1150 }
1151 
1152 /*
1153  * Prepend a tag to the list of tags associated with an mbuf.
1154  */
1155 static __inline void
1156 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1157 {
1158 
1159 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1160 }
1161 
1162 /*
1163  * Unlink a tag from the list of tags associated with an mbuf.
1164  */
1165 static __inline void
1166 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1167 {
1168 
1169 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1170 }
1171 
1172 /* These are for OpenBSD compatibility. */
1173 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1174 
1175 static __inline struct m_tag *
1176 m_tag_get(int type, int length, int wait)
1177 {
1178 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1179 }
1180 
1181 static __inline struct m_tag *
1182 m_tag_find(struct mbuf *m, int type, struct m_tag *start)
1183 {
1184 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1185 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1186 }
1187 
1188 static __inline struct mbuf *
1189 m_free(struct mbuf *m)
1190 {
1191 	struct mbuf *n = m->m_next;
1192 
1193 	MBUF_PROBE1(m__free, m);
1194 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1195 		m_tag_delete_chain(m, NULL);
1196 	if (m->m_flags & M_EXT)
1197 		mb_free_ext(m);
1198 	else if ((m->m_flags & M_NOFREE) == 0)
1199 		uma_zfree(zone_mbuf, m);
1200 	return (n);
1201 }
1202 
1203 static __inline int
1204 rt_m_getfib(struct mbuf *m)
1205 {
1206 	KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1207 	return (m->m_pkthdr.fibnum);
1208 }
1209 
1210 #define M_GETFIB(_m)   rt_m_getfib(_m)
1211 
1212 #define M_SETFIB(_m, _fib) do {						\
1213         KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf."));	\
1214 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1215 } while (0)
1216 
1217 /* flags passed as first argument for "m_ether_tcpip_hash()" */
1218 #define	MBUF_HASHFLAG_L2	(1 << 2)
1219 #define	MBUF_HASHFLAG_L3	(1 << 3)
1220 #define	MBUF_HASHFLAG_L4	(1 << 4)
1221 
1222 /* mbuf hashing helper routines */
1223 uint32_t	m_ether_tcpip_hash_init(void);
1224 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t);
1225 
1226 #ifdef MBUF_PROFILING
1227  void m_profile(struct mbuf *m);
1228  #define M_PROFILE(m) m_profile(m)
1229 #else
1230  #define M_PROFILE(m)
1231 #endif
1232 
1233 struct mbufq {
1234 	STAILQ_HEAD(, mbuf)	mq_head;
1235 	int			mq_len;
1236 	int			mq_maxlen;
1237 };
1238 
1239 static inline void
1240 mbufq_init(struct mbufq *mq, int maxlen)
1241 {
1242 
1243 	STAILQ_INIT(&mq->mq_head);
1244 	mq->mq_maxlen = maxlen;
1245 	mq->mq_len = 0;
1246 }
1247 
1248 static inline struct mbuf *
1249 mbufq_flush(struct mbufq *mq)
1250 {
1251 	struct mbuf *m;
1252 
1253 	m = STAILQ_FIRST(&mq->mq_head);
1254 	STAILQ_INIT(&mq->mq_head);
1255 	mq->mq_len = 0;
1256 	return (m);
1257 }
1258 
1259 static inline void
1260 mbufq_drain(struct mbufq *mq)
1261 {
1262 	struct mbuf *m, *n;
1263 
1264 	n = mbufq_flush(mq);
1265 	while ((m = n) != NULL) {
1266 		n = STAILQ_NEXT(m, m_stailqpkt);
1267 		m_freem(m);
1268 	}
1269 }
1270 
1271 static inline struct mbuf *
1272 mbufq_first(const struct mbufq *mq)
1273 {
1274 
1275 	return (STAILQ_FIRST(&mq->mq_head));
1276 }
1277 
1278 static inline struct mbuf *
1279 mbufq_last(const struct mbufq *mq)
1280 {
1281 
1282 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1283 }
1284 
1285 static inline int
1286 mbufq_full(const struct mbufq *mq)
1287 {
1288 
1289 	return (mq->mq_len >= mq->mq_maxlen);
1290 }
1291 
1292 static inline int
1293 mbufq_len(const struct mbufq *mq)
1294 {
1295 
1296 	return (mq->mq_len);
1297 }
1298 
1299 static inline int
1300 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1301 {
1302 
1303 	if (mbufq_full(mq))
1304 		return (ENOBUFS);
1305 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1306 	mq->mq_len++;
1307 	return (0);
1308 }
1309 
1310 static inline struct mbuf *
1311 mbufq_dequeue(struct mbufq *mq)
1312 {
1313 	struct mbuf *m;
1314 
1315 	m = STAILQ_FIRST(&mq->mq_head);
1316 	if (m) {
1317 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1318 		m->m_nextpkt = NULL;
1319 		mq->mq_len--;
1320 	}
1321 	return (m);
1322 }
1323 
1324 static inline void
1325 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1326 {
1327 
1328 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1329 	mq->mq_len++;
1330 }
1331 
1332 /*
1333  * Note: this doesn't enforce the maximum list size for dst.
1334  */
1335 static inline void
1336 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1337 {
1338 
1339 	mq_dst->mq_len += mq_src->mq_len;
1340 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1341 	mq_src->mq_len = 0;
1342 }
1343 
1344 #ifdef _SYS_TIMESPEC_H_
1345 static inline void
1346 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1347 {
1348 
1349 	KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1350 	KASSERT((m->m_flags & M_TSTMP) != 0, ("mbuf %p no M_TSTMP", m));
1351 	ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1352 	ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1353 }
1354 #endif
1355 
1356 #ifdef NETDUMP
1357 /* Invoked from the netdump client code. */
1358 void	netdump_mbuf_drain(void);
1359 void	netdump_mbuf_dump(void);
1360 void	netdump_mbuf_reinit(int nmbuf, int nclust, int clsize);
1361 #endif
1362 
1363 #endif /* _KERNEL */
1364 #endif /* !_SYS_MBUF_H_ */
1365