xref: /freebsd/sys/sys/mbuf.h (revision 3b93d3597cc336d5637e12ade8642d589cb0bc8a)
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 
33 #ifndef _SYS_MBUF_H_
34 #define	_SYS_MBUF_H_
35 
36 /* XXX: These includes suck. Sorry! */
37 #include <sys/queue.h>
38 #ifdef _KERNEL
39 #include <sys/systm.h>
40 #include <sys/refcount.h>
41 #include <vm/uma.h>
42 
43 #include <sys/sdt.h>
44 
45 #define	MBUF_PROBE1(probe, arg0)					\
46 	SDT_PROBE1(sdt, , , probe, arg0)
47 #define	MBUF_PROBE2(probe, arg0, arg1)					\
48 	SDT_PROBE2(sdt, , , probe, arg0, arg1)
49 #define	MBUF_PROBE3(probe, arg0, arg1, arg2)				\
50 	SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
51 #define	MBUF_PROBE4(probe, arg0, arg1, arg2, arg3)			\
52 	SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
53 #define	MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4)		\
54 	SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
55 
56 SDT_PROBE_DECLARE(sdt, , , m__init);
57 SDT_PROBE_DECLARE(sdt, , , m__gethdr_raw);
58 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
59 SDT_PROBE_DECLARE(sdt, , , m__get_raw);
60 SDT_PROBE_DECLARE(sdt, , , m__get);
61 SDT_PROBE_DECLARE(sdt, , , m__getcl);
62 SDT_PROBE_DECLARE(sdt, , , m__getjcl);
63 SDT_PROBE_DECLARE(sdt, , , m__clget);
64 SDT_PROBE_DECLARE(sdt, , , m__cljget);
65 SDT_PROBE_DECLARE(sdt, , , m__cljset);
66 SDT_PROBE_DECLARE(sdt, , , m__free);
67 SDT_PROBE_DECLARE(sdt, , , m__freem);
68 SDT_PROBE_DECLARE(sdt, , , m__freemp);
69 
70 #endif /* _KERNEL */
71 
72 /*
73  * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
74  * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
75  * sys/param.h), which has no additional overhead and is used instead of the
76  * internal data area; this is done when at least MINCLSIZE of data must be
77  * stored.  Additionally, it is possible to allocate a separate buffer
78  * externally and attach it to the mbuf in a way similar to that of mbuf
79  * clusters.
80  *
81  * NB: These calculation do not take actual compiler-induced alignment and
82  * padding inside the complete struct mbuf into account.  Appropriate
83  * attention is required when changing members of struct mbuf.
84  *
85  * MLEN is data length in a normal mbuf.
86  * MHLEN is data length in an mbuf with pktheader.
87  * MINCLSIZE is a smallest amount of data that should be put into cluster.
88  *
89  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
90  * they are sensible.
91  */
92 struct mbuf;
93 #define	MHSIZE		offsetof(struct mbuf, m_dat)
94 #define	MPKTHSIZE	offsetof(struct mbuf, m_pktdat)
95 #define	MLEN		((int)(MSIZE - MHSIZE))
96 #define	MHLEN		((int)(MSIZE - MPKTHSIZE))
97 #define	MINCLSIZE	(MHLEN + 1)
98 #define	M_NODOM		255
99 
100 #ifdef _KERNEL
101 /*-
102  * Macro for type conversion: convert mbuf pointer to data pointer of correct
103  * type:
104  *
105  * mtod(m, t)	-- Convert mbuf pointer to data pointer of correct type.
106  * mtodo(m, o) -- Same as above but with offset 'o' into data.
107  */
108 #define	mtod(m, t)	((t)((m)->m_data))
109 #define	mtodo(m, o)	((void *)(((m)->m_data) + (o)))
110 
111 /*
112  * Argument structure passed to UMA routines during mbuf and packet
113  * allocations.
114  */
115 struct mb_args {
116 	int	flags;	/* Flags for mbuf being allocated */
117 	short	type;	/* Type of mbuf being allocated */
118 };
119 #endif /* _KERNEL */
120 
121 /*
122  * Packet tag structure (see below for details).
123  */
124 struct m_tag {
125 	SLIST_ENTRY(m_tag)	m_tag_link;	/* List of packet tags */
126 	u_int16_t		m_tag_id;	/* Tag ID */
127 	u_int16_t		m_tag_len;	/* Length of data */
128 	u_int32_t		m_tag_cookie;	/* ABI/Module ID */
129 	void			(*m_tag_free)(struct m_tag *);
130 };
131 
132 /*
133  * Static network interface owned tag.
134  * Allocated through ifp->if_snd_tag_alloc().
135  */
136 struct if_snd_tag_sw;
137 
138 struct m_snd_tag {
139 	struct ifnet *ifp;		/* network interface tag belongs to */
140 	const struct if_snd_tag_sw *sw;
141 	volatile u_int refcount;
142 };
143 
144 /*
145  * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
146  * Size ILP32: 56
147  *	 LP64: 64
148  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
149  * they are correct.
150  */
151 struct pkthdr {
152 	union {
153 		struct m_snd_tag *snd_tag;	/* send tag, if any */
154 		struct ifnet	*rcvif;		/* rcv interface */
155 		struct {
156 			uint16_t rcvidx;	/* rcv interface index ... */
157 			uint16_t rcvgen;	/* ... and generation count */
158 		};
159 	};
160 	union {
161 		struct ifnet	*leaf_rcvif;	/* leaf rcv interface */
162 		struct {
163 			uint16_t leaf_rcvidx;	/* leaf rcv interface index ... */
164 			uint16_t leaf_rcvgen;	/* ... and generation count */
165 		};
166 	};
167 	SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
168 	int32_t		 len;		/* total packet length */
169 
170 	/* Layer crossing persistent information. */
171 	uint32_t	 flowid;	/* packet's 4-tuple system */
172 	uint32_t	 csum_flags;	/* checksum and offload features */
173 	uint16_t	 fibnum;	/* this packet should use this fib */
174 	uint8_t		 numa_domain;	/* NUMA domain of recvd pkt */
175 	uint8_t		 rsstype;	/* hash type */
176 #if !defined(__LP64__)
177 	uint32_t	 pad;		/* pad for 64bit alignment */
178 #endif
179 	union {
180 		uint64_t	rcv_tstmp;	/* timestamp in ns */
181 		struct {
182 			uint8_t		 l2hlen;	/* layer 2 hdr len */
183 			uint8_t		 l3hlen;	/* layer 3 hdr len */
184 			uint8_t		 l4hlen;	/* layer 4 hdr len */
185 			uint8_t		 l5hlen;	/* layer 5 hdr len */
186 			uint8_t		 inner_l2hlen;
187 			uint8_t		 inner_l3hlen;
188 			uint8_t		 inner_l4hlen;
189 			uint8_t		 inner_l5hlen;
190 		};
191 	};
192 	union {
193 		uint8_t  eight[8];
194 		uint16_t sixteen[4];
195 		uint32_t thirtytwo[2];
196 		uint64_t sixtyfour[1];
197 		uintptr_t unintptr[1];
198 		void	*ptr;
199 	} PH_per;
200 
201 	/* Layer specific non-persistent local storage for reassembly, etc. */
202 	union {
203 		union {
204 			uint8_t  eight[8];
205 			uint16_t sixteen[4];
206 			uint32_t thirtytwo[2];
207 			uint64_t sixtyfour[1];
208 			uintptr_t unintptr[1];
209 			void 	*ptr;
210 		} PH_loc;
211 		/* Upon allocation: total packet memory consumption. */
212 		u_int	memlen;
213 	};
214 };
215 #define	ether_vtag	PH_per.sixteen[0]
216 #define tcp_tun_port	PH_per.sixteen[0] /* outbound */
217 #define	vt_nrecs	PH_per.sixteen[0]	  /* mld and v6-ND */
218 #define	tso_segsz	PH_per.sixteen[1] /* inbound after LRO */
219 #define	lro_nsegs	tso_segsz	  /* inbound after LRO */
220 #define	csum_data	PH_per.thirtytwo[1] /* inbound from hardware up */
221 #define	lro_tcp_d_len	PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
222 #define	lro_tcp_d_csum	PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
223 #define	lro_tcp_h_off	PH_loc.sixteen[2] /* inbound during LRO (no reassembly) */
224 #define	lro_etype	PH_loc.sixteen[3] /* inbound during LRO (no reassembly) */
225 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
226 
227 /*
228  * TLS records for TLS 1.0-1.2 can have the following header lengths:
229  * - 5 (AES-CBC with implicit IV)
230  * - 21 (AES-CBC with explicit IV)
231  * - 13 (AES-GCM with 8 byte explicit IV)
232  */
233 #define	MBUF_PEXT_HDR_LEN	23
234 
235 /*
236  * TLS records for TLS 1.0-1.2 can have the following maximum trailer
237  * lengths:
238  * - 16 (AES-GCM)
239  * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
240  * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
241  * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
242  */
243 #define	MBUF_PEXT_TRAIL_LEN	64
244 
245 #if defined(__LP64__)
246 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
247 #else
248 #define MBUF_PEXT_MAX_PGS (64 / sizeof(vm_paddr_t))
249 #endif
250 
251 #define	MBUF_PEXT_MAX_BYTES						\
252     (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
253 
254 struct ktls_session;
255 struct socket;
256 
257 /*
258  * Description of external storage mapped into mbuf; valid only if M_EXT is
259  * set.
260  * Size ILP32: 28
261  *	 LP64: 48
262  * Compile-time assertions in uipc_mbuf.c test these values to ensure that
263  * they are correct.
264  */
265 typedef	void m_ext_free_t(struct mbuf *);
266 struct m_ext {
267 	union {
268 		/*
269 		 * If EXT_FLAG_EMBREF is set, then we use refcount in the
270 		 * mbuf, the 'ext_count' member.  Otherwise, we have a
271 		 * shadow copy and we use pointer 'ext_cnt'.  The original
272 		 * mbuf is responsible to carry the pointer to free routine
273 		 * and its arguments.  They aren't copied into shadows in
274 		 * mb_dupcl() to avoid dereferencing next cachelines.
275 		 */
276 		volatile u_int	 ext_count;
277 		volatile u_int	*ext_cnt;
278 	};
279 	uint32_t	 ext_size;	/* size of buffer, for ext_free */
280 	uint32_t	 ext_type:8,	/* type of external storage */
281 			 ext_flags:24;	/* external storage mbuf flags */
282 	union {
283 		struct {
284 			/*
285 			 * Regular M_EXT mbuf:
286 			 * o ext_buf always points to the external buffer.
287 			 * o ext_free (below) and two optional arguments
288 			 *   ext_arg1 and ext_arg2 store the free context for
289 			 *   the external storage.  They are set only in the
290 			 *   refcount carrying mbuf, the one with
291 			 *   EXT_FLAG_EMBREF flag, with exclusion for
292 			 *   EXT_EXTREF type, where the free context is copied
293 			 *   into all mbufs that use same external storage.
294 			 */
295 			char 	*ext_buf;	/* start of buffer */
296 #define	m_ext_copylen	offsetof(struct m_ext, ext_arg2)
297 			void	*ext_arg2;
298 		};
299 		struct {
300 			/*
301 			 * Multi-page M_EXTPG mbuf:
302 			 * o extpg_pa - page vector.
303 			 * o extpg_trail and extpg_hdr - TLS trailer and
304 			 *   header.
305 			 * Uses ext_free and may also use ext_arg1.
306 			 */
307 			vm_paddr_t	extpg_pa[MBUF_PEXT_MAX_PGS];
308 			char		extpg_trail[MBUF_PEXT_TRAIL_LEN];
309 			char		extpg_hdr[MBUF_PEXT_HDR_LEN];
310 			/* Pretend these 3 fields are part of mbuf itself. */
311 #define	m_epg_pa	m_ext.extpg_pa
312 #define	m_epg_trail	m_ext.extpg_trail
313 #define	m_epg_hdr	m_ext.extpg_hdr
314 #define	m_epg_ext_copylen	offsetof(struct m_ext, ext_free)
315 		};
316 	};
317 	/*
318 	 * Free method and optional argument pointer, both
319 	 * used by M_EXT and M_EXTPG.
320 	 */
321 	m_ext_free_t	*ext_free;
322 	void		*ext_arg1;
323 };
324 
325 /*
326  * The core of the mbuf object along with some shortcut defines for practical
327  * purposes.
328  */
329 struct mbuf {
330 	/*
331 	 * Header present at the beginning of every mbuf.
332 	 * Size ILP32: 24
333 	 *      LP64: 32
334 	 * Compile-time assertions in uipc_mbuf.c test these values to ensure
335 	 * that they are correct.
336 	 */
337 	union {	/* next buffer in chain */
338 		struct mbuf		*m_next;
339 		SLIST_ENTRY(mbuf)	m_slist;
340 		STAILQ_ENTRY(mbuf)	m_stailq;
341 	};
342 	union {	/* next chain in queue/record */
343 		struct mbuf		*m_nextpkt;
344 		SLIST_ENTRY(mbuf)	m_slistpkt;
345 		STAILQ_ENTRY(mbuf)	m_stailqpkt;
346 	};
347 	caddr_t		 m_data;	/* location of data */
348 	int32_t		 m_len;		/* amount of data in this mbuf */
349 	uint32_t	 m_type:8,	/* type of data in this mbuf */
350 			 m_flags:24;	/* flags; see below */
351 #if !defined(__LP64__)
352 	uint32_t	 m_pad;		/* pad for 64bit alignment */
353 #endif
354 
355 	/*
356 	 * A set of optional headers (packet header, external storage header)
357 	 * and internal data storage.  Historically, these arrays were sized
358 	 * to MHLEN (space left after a packet header) and MLEN (space left
359 	 * after only a regular mbuf header); they are now variable size in
360 	 * order to support future work on variable-size mbufs.
361 	 */
362 	union {
363 		struct {
364 			union {
365 				/* M_PKTHDR set. */
366 				struct pkthdr	m_pkthdr;
367 
368 				/* M_EXTPG set.
369 				 * Multi-page M_EXTPG mbuf has its meta data
370 				 * split between the below anonymous structure
371 				 * and m_ext.  It carries vector of pages,
372 				 * optional header and trailer char vectors
373 				 * and pointers to socket/TLS data.
374 				 */
375 #define	m_epg_startcopy		m_epg_npgs
376 #define	m_epg_endcopy		m_epg_stailq
377 				struct {
378 					/* Overall count of pages and count of
379 					 * pages with I/O pending. */
380 					uint8_t	m_epg_npgs;
381 					uint8_t	m_epg_nrdy;
382 					/* TLS header and trailer lengths.
383 					 * The data itself resides in m_ext. */
384 					uint8_t	m_epg_hdrlen;
385 					uint8_t	m_epg_trllen;
386 					/* Offset into 1st page and length of
387 					 * data in the last page. */
388 					uint16_t m_epg_1st_off;
389 					uint16_t m_epg_last_len;
390 					uint8_t	m_epg_flags;
391 #define	EPG_FLAG_ANON	0x1	/* Data can be encrypted in place. */
392 #define	EPG_FLAG_2FREE	0x2	/* Scheduled for free. */
393 					uint8_t	m_epg_record_type;
394 					uint8_t	__spare[2];
395 					int	m_epg_enc_cnt;
396 					struct ktls_session *m_epg_tls;
397 					struct socket	*m_epg_so;
398 					uint64_t	m_epg_seqno;
399 					STAILQ_ENTRY(mbuf) m_epg_stailq;
400 				};
401 			};
402 			union {
403 				/* M_EXT or M_EXTPG set. */
404 				struct m_ext	m_ext;
405 				/* M_PKTHDR set, neither M_EXT nor M_EXTPG. */
406 				char		m_pktdat[0];
407 			};
408 		};
409 		char	m_dat[0];			/* !M_PKTHDR, !M_EXT */
410 	};
411 };
412 
413 #ifdef _KERNEL
414 static inline int
m_epg_pagelen(const struct mbuf * m,int pidx,int pgoff)415 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff)
416 {
417 
418 	KASSERT(pgoff == 0 || pidx == 0,
419 	    ("page %d with non-zero offset %d in %p", pidx, pgoff, m));
420 
421 	if (pidx == m->m_epg_npgs - 1) {
422 		return (m->m_epg_last_len);
423 	} else {
424 		return (PAGE_SIZE - pgoff);
425 	}
426 }
427 
428 #ifdef INVARIANTS
429 #define	MCHECK(ex, msg)	KASSERT((ex),				\
430 	    ("Multi page mbuf %p with " #msg " at %s:%d",	\
431 	    m, __FILE__, __LINE__))
432 /*
433  * NB: This expects a non-empty buffer (npgs > 0 and
434  * last_pg_len > 0).
435  */
436 #define	MBUF_EXT_PGS_ASSERT_SANITY(m)	do {				\
437 	MCHECK(m->m_epg_npgs > 0, "no valid pages");		\
438 	MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa),		\
439 	    "too many pages");						\
440 	MCHECK(m->m_epg_nrdy <= m->m_epg_npgs,			\
441 	    "too many ready pages");					\
442 	MCHECK(m->m_epg_1st_off < PAGE_SIZE,			\
443 		"too large page offset");				\
444 	MCHECK(m->m_epg_last_len > 0, "zero last page length");	\
445 	MCHECK(m->m_epg_last_len <= PAGE_SIZE,			\
446 	    "too large last page length");				\
447 	if (m->m_epg_npgs == 1)					\
448 		MCHECK(m->m_epg_1st_off +			\
449 		    m->m_epg_last_len <=	 PAGE_SIZE,		\
450 		    "single page too large");				\
451 	MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr),		\
452 	    "too large header length");					\
453 	MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail),	\
454 	    "too large header length");					\
455 } while (0)
456 #else
457 #define	MBUF_EXT_PGS_ASSERT_SANITY(m)	do {} while (0)
458 #endif
459 #endif
460 
461 /*
462  * mbuf flags of global significance and layer crossing.
463  * Those of only protocol/layer specific significance are to be mapped
464  * to M_PROTO[1-11] and cleared at layer handoff boundaries.
465  * NB: Limited to the lower 24 bits.
466  */
467 #define	M_EXT		0x00000001 /* has associated external storage */
468 #define	M_PKTHDR	0x00000002 /* start of record */
469 #define	M_EOR		0x00000004 /* end of record */
470 #define	M_RDONLY	0x00000008 /* associated data is marked read-only */
471 #define	M_BCAST		0x00000010 /* send/received as link-level broadcast */
472 #define	M_MCAST		0x00000020 /* send/received as link-level multicast */
473 #define	M_PROMISC	0x00000040 /* packet was not for us */
474 #define	M_VLANTAG	0x00000080 /* ether_vtag is valid */
475 #define	M_EXTPG		0x00000100 /* has array of unmapped pages and TLS */
476 #define	M_NOFREE	0x00000200 /* do not free mbuf, embedded in cluster */
477 #define	M_TSTMP		0x00000400 /* rcv_tstmp field is valid */
478 #define	M_TSTMP_HPREC	0x00000800 /* rcv_tstmp is high-prec, typically
479 				      hw-stamped on port (useful for IEEE 1588
480 				      and 802.1AS) */
481 #define M_TSTMP_LRO	0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
482 
483 #define	M_PROTO1	0x00002000 /* protocol-specific */
484 #define	M_PROTO2	0x00004000 /* protocol-specific */
485 #define	M_PROTO3	0x00008000 /* protocol-specific */
486 #define	M_PROTO4	0x00010000 /* protocol-specific */
487 #define	M_PROTO5	0x00020000 /* protocol-specific */
488 #define	M_PROTO6	0x00040000 /* protocol-specific */
489 #define	M_PROTO7	0x00080000 /* protocol-specific */
490 #define	M_PROTO8	0x00100000 /* protocol-specific */
491 #define	M_PROTO9	0x00200000 /* protocol-specific */
492 #define	M_PROTO10	0x00400000 /* protocol-specific */
493 #define	M_PROTO11	0x00800000 /* protocol-specific */
494 
495 /*
496  * Flags to purge when crossing layers.
497  */
498 #define	M_PROTOFLAGS \
499     (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
500      M_PROTO9|M_PROTO10|M_PROTO11)
501 
502 /*
503  * Flags preserved when copying m_pkthdr.
504  */
505 #define M_COPYFLAGS \
506     (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
507      M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
508 
509 /*
510  * Flags preserved during demote.
511  */
512 #define	M_DEMOTEFLAGS \
513     (M_EXT | M_RDONLY | M_NOFREE | M_EXTPG)
514 
515 /*
516  * Mbuf flag description for use with printf(9) %b identifier.
517  */
518 #define	M_FLAG_BITS \
519     "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
520     "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
521 #define	M_FLAG_PROTOBITS \
522     "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
523     "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
524     "\27M_PROTO10\28M_PROTO11"
525 #define	M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
526 
527 /*
528  * Network interface cards are able to hash protocol fields (such as IPv4
529  * addresses and TCP port numbers) classify packets into flows.  These flows
530  * can then be used to maintain ordering while delivering packets to the OS
531  * via parallel input queues, as well as to provide a stateless affinity
532  * model.  NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
533  * m_flag fields to indicate how the hash should be interpreted by the
534  * network stack.
535  *
536  * Most NICs support RSS, which provides ordering and explicit affinity, and
537  * use the hash m_flag bits to indicate what header fields were covered by
538  * the hash.  M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
539  * RSS cards or configurations that provide an opaque flow identifier, allowing
540  * for ordering and distribution without explicit affinity.  Additionally,
541  * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
542  * properties.
543  *
544  * The meaning of the IPV6_EX suffix:
545  * "o  Home address from the home address option in the IPv6 destination
546  *     options header.  If the extension header is not present, use the Source
547  *     IPv6 Address.
548  *  o  IPv6 address that is contained in the Routing-Header-Type-2 from the
549  *     associated extension header.  If the extension header is not present,
550  *     use the Destination IPv6 Address."
551  * Quoted from:
552  * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
553  */
554 #define	M_HASHTYPE_HASHPROP		0x80	/* has hash properties */
555 #define	M_HASHTYPE_INNER		0x40	/* calculated from inner headers */
556 #define	M_HASHTYPE_HASH(t)		(M_HASHTYPE_HASHPROP | (t))
557 /* Microsoft RSS standard hash types */
558 #define	M_HASHTYPE_NONE			0
559 #define	M_HASHTYPE_RSS_IPV4		M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
560 #define	M_HASHTYPE_RSS_TCP_IPV4		M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
561 #define	M_HASHTYPE_RSS_IPV6		M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
562 #define	M_HASHTYPE_RSS_TCP_IPV6		M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
563 #define	M_HASHTYPE_RSS_IPV6_EX		M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
564 							    * ext hdrs */
565 #define	M_HASHTYPE_RSS_TCP_IPV6_EX	M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
566 							    * ext hdrs */
567 #define	M_HASHTYPE_RSS_UDP_IPV4		M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
568 #define	M_HASHTYPE_RSS_UDP_IPV6		M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
569 #define	M_HASHTYPE_RSS_UDP_IPV6_EX	M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
570 							    * ext hdrs */
571 
572 #define	M_HASHTYPE_OPAQUE		0x3f	/* ordering, not affinity */
573 #define	M_HASHTYPE_OPAQUE_HASH		M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
574 						/* ordering+hash, not affinity*/
575 
576 #define	M_HASHTYPE_CLEAR(m)	((m)->m_pkthdr.rsstype = 0)
577 #define	M_HASHTYPE_GET(m)	((m)->m_pkthdr.rsstype & ~M_HASHTYPE_INNER)
578 #define	M_HASHTYPE_SET(m, v)	((m)->m_pkthdr.rsstype = (v))
579 #define	M_HASHTYPE_TEST(m, v)	(M_HASHTYPE_GET(m) == (v))
580 #define	M_HASHTYPE_ISHASH(m)	\
581     (((m)->m_pkthdr.rsstype & M_HASHTYPE_HASHPROP) != 0)
582 #define	M_HASHTYPE_ISHASH_TCP(m)				\
583     (((m)->m_pkthdr.rsstype & (M_HASHTYPE_RSS_TCP_IPV4 |		\
584 				M_HASHTYPE_RSS_TCP_IPV6 |		\
585 				M_HASHTYPE_RSS_TCP_IPV6_EX)) != 0)
586 #define	M_HASHTYPE_SETINNER(m)	do {			\
587 	(m)->m_pkthdr.rsstype |= M_HASHTYPE_INNER;	\
588     } while (0)
589 
590 /*
591  * External mbuf storage buffer types.
592  */
593 #define	EXT_CLUSTER	1	/* mbuf cluster */
594 #define	EXT_SFBUF	2	/* sendfile(2)'s sf_buf */
595 #define	EXT_JUMBOP	3	/* jumbo cluster page sized */
596 #define	EXT_JUMBO9	4	/* jumbo cluster 9216 bytes */
597 #define	EXT_JUMBO16	5	/* jumbo cluster 16184 bytes */
598 #define	EXT_PACKET	6	/* mbuf+cluster from packet zone */
599 #define	EXT_MBUF	7	/* external mbuf reference */
600 #define	EXT_RXRING	8	/* data in NIC receive ring */
601 #define	EXT_CTL		9	/* buffer from a ctl(4) backend */
602 
603 #define	EXT_VENDOR1	224	/* for vendor-internal use */
604 #define	EXT_VENDOR2	225	/* for vendor-internal use */
605 #define	EXT_VENDOR3	226	/* for vendor-internal use */
606 #define	EXT_VENDOR4	227	/* for vendor-internal use */
607 
608 #define	EXT_EXP1	244	/* for experimental use */
609 #define	EXT_EXP2	245	/* for experimental use */
610 #define	EXT_EXP3	246	/* for experimental use */
611 #define	EXT_EXP4	247	/* for experimental use */
612 
613 #define	EXT_NET_DRV	252	/* custom ext_buf provided by net driver(s) */
614 #define	EXT_MOD_TYPE	253	/* custom module's ext_buf type */
615 #define	EXT_DISPOSABLE	254	/* can throw this buffer away w/page flipping */
616 #define	EXT_EXTREF	255	/* has externally maintained ext_cnt ptr */
617 
618 /*
619  * Flags for external mbuf buffer types.
620  * NB: limited to the lower 24 bits.
621  */
622 #define	EXT_FLAG_EMBREF		0x000001	/* embedded ext_count */
623 #define	EXT_FLAG_EXTREF		0x000002	/* external ext_cnt, notyet */
624 #define	EXT_FLAG_SFBUF_ANON	0x000004	/* sendfile buffer is mutable */
625 
626 #define	EXT_FLAG_NOFREE		0x000010	/* don't free mbuf to pool, notyet */
627 
628 #define	EXT_FLAG_VENDOR1	0x010000	/* These flags are vendor */
629 #define	EXT_FLAG_VENDOR2	0x020000	/* or submodule specific, */
630 #define	EXT_FLAG_VENDOR3	0x040000	/* not used by mbuf code. */
631 #define	EXT_FLAG_VENDOR4	0x080000	/* Set/read by submodule. */
632 
633 #define	EXT_FLAG_EXP1		0x100000	/* for experimental use */
634 #define	EXT_FLAG_EXP2		0x200000	/* for experimental use */
635 #define	EXT_FLAG_EXP3		0x400000	/* for experimental use */
636 #define	EXT_FLAG_EXP4		0x800000	/* for experimental use */
637 
638 /*
639  * EXT flag description for use with printf(9) %b identifier.
640  */
641 #define	EXT_FLAG_BITS \
642     "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
643     "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
644     "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
645     "\30EXT_FLAG_EXP4"
646 
647 /*
648  * Flags indicating checksum, segmentation and other offload work to be
649  * done, or already done, by hardware or lower layers.
650  *
651  * Flags that are set by upper protocol layers requesting lower
652  * layers, or ideally the hardware, to perform these offloading tasks.
653  * Before passing packets to a network interface this field and its flags can
654  * be directly tested against ifnet if_hwassist.  Note that the flags
655  * CSUM_IP_SCTP, CSUM_IP_TCP, and CSUM_IP_UDP can appear on input processing
656  * of SCTP, TCP, and UDP.  In such a case the checksum will not be computed or
657  * validated by SCTP, TCP, or TCP, since the packet has not been on the wire.
658  *
659  * CSUM_INNER_<x> is the same as CSUM_<x> but it applies to the inner frame.
660  * The CSUM_ENCAP_<x> bits identify the outer encapsulation.
661  */
662 #define	CSUM_IP			0x00000001	/* IP header checksum offload */
663 #define	CSUM_IP_UDP		0x00000002	/* UDP checksum offload */
664 #define	CSUM_IP_TCP		0x00000004	/* TCP checksum offload */
665 #define	CSUM_IP_SCTP		0x00000008	/* SCTP checksum offload */
666 #define	CSUM_IP_TSO		0x00000010	/* TCP segmentation offload */
667 #define	CSUM_IP_ISCSI		0x00000020	/* iSCSI checksum offload */
668 
669 #define	CSUM_INNER_IP6_UDP	0x00000040
670 #define	CSUM_INNER_IP6_TCP	0x00000080
671 #define	CSUM_INNER_IP6_TSO	0x00000100
672 #define	CSUM_IP6_UDP		0x00000200	/* UDP checksum offload */
673 #define	CSUM_IP6_TCP		0x00000400	/* TCP checksum offload */
674 #define	CSUM_IP6_SCTP		0x00000800	/* SCTP checksum offload */
675 #define	CSUM_IP6_TSO		0x00001000	/* TCP segmentation offload */
676 #define	CSUM_IP6_ISCSI		0x00002000	/* iSCSI checksum offload */
677 
678 #define	CSUM_INNER_IP		0x00004000
679 #define	CSUM_INNER_IP_UDP	0x00008000
680 #define	CSUM_INNER_IP_TCP	0x00010000
681 #define	CSUM_INNER_IP_TSO	0x00020000
682 
683 #define	CSUM_ENCAP_VXLAN	0x00040000	/* VXLAN outer encapsulation */
684 #define	CSUM_ENCAP_GENEVE	0x00080000	/* GENEVE outer encapsulation */
685 
686 /* Flags used to indicate that the checksum was verified by hardware. */
687 #define	CSUM_INNER_L3_CALC	0x00100000
688 #define	CSUM_INNER_L3_VALID	0x00200000
689 #define	CSUM_INNER_L4_CALC	0x00400000
690 #define	CSUM_INNER_L4_VALID	0x00800000
691 #define	CSUM_L3_CALC		0x01000000	/* calculated layer 3 csum */
692 #define	CSUM_L3_VALID		0x02000000	/* checksum is correct */
693 #define	CSUM_L4_CALC		0x04000000	/* calculated layer 4 csum */
694 #define	CSUM_L4_VALID		0x08000000	/* checksum is correct */
695 #define	CSUM_L5_CALC		0x10000000	/* calculated layer 5 csum */
696 #define	CSUM_L5_VALID		0x20000000	/* checksum is correct */
697 #define	CSUM_COALESCED		0x40000000	/* contains merged segments */
698 
699 #define	CSUM_SND_TAG		0x80000000	/* Packet header has send tag */
700 
701 #define CSUM_FLAGS_TX (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP | CSUM_IP_SCTP | \
702     CSUM_IP_TSO | CSUM_IP_ISCSI | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | \
703     CSUM_INNER_IP6_TSO | CSUM_IP6_UDP | CSUM_IP6_TCP | CSUM_IP6_SCTP | \
704     CSUM_IP6_TSO | CSUM_IP6_ISCSI | CSUM_INNER_IP | CSUM_INNER_IP_UDP | \
705     CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN | \
706     CSUM_ENCAP_GENEVE | CSUM_SND_TAG)
707 
708 #define CSUM_FLAGS_RX (CSUM_INNER_L3_CALC | CSUM_INNER_L3_VALID | \
709     CSUM_INNER_L4_CALC | CSUM_INNER_L4_VALID | CSUM_L3_CALC | CSUM_L3_VALID | \
710     CSUM_L4_CALC | CSUM_L4_VALID | CSUM_L5_CALC | CSUM_L5_VALID | \
711     CSUM_COALESCED)
712 
713 /*
714  * CSUM flag description for use with printf(9) %b identifier.
715  */
716 #define	CSUM_BITS \
717     "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
718     "\6CSUM_IP_ISCSI\7CSUM_INNER_IP6_UDP\10CSUM_INNER_IP6_TCP" \
719     "\11CSUM_INNER_IP6_TSO\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP" \
720     "\15CSUM_IP6_TSO\16CSUM_IP6_ISCSI\17CSUM_INNER_IP\20CSUM_INNER_IP_UDP" \
721     "\21CSUM_INNER_IP_TCP\22CSUM_INNER_IP_TSO\23CSUM_ENCAP_VXLAN" \
722     "\24CSUM_ENCAP_GENEVE\25CSUM_INNER_L3_CALC\26CSUM_INNER_L3_VALID" \
723     "\27CSUM_INNER_L4_CALC\30CSUM_INNER_L4_VALID\31CSUM_L3_CALC" \
724     "\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID\35CSUM_L5_CALC" \
725     "\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
726 
727 /* CSUM flags compatibility mappings. */
728 #define	CSUM_IP_CHECKED		CSUM_L3_CALC
729 #define	CSUM_IP_VALID		CSUM_L3_VALID
730 #define	CSUM_DATA_VALID		CSUM_L4_VALID
731 #define	CSUM_PSEUDO_HDR		CSUM_L4_CALC
732 #define	CSUM_SCTP_VALID		CSUM_L4_VALID
733 #define	CSUM_DELAY_DATA		(CSUM_TCP|CSUM_UDP)
734 #define	CSUM_DELAY_IP		CSUM_IP		/* Only v4, no v6 IP hdr csum */
735 #define	CSUM_DELAY_DATA_IPV6	(CSUM_TCP_IPV6|CSUM_UDP_IPV6)
736 #define	CSUM_DATA_VALID_IPV6	CSUM_DATA_VALID
737 #define	CSUM_TCP		CSUM_IP_TCP
738 #define	CSUM_UDP		CSUM_IP_UDP
739 #define	CSUM_SCTP		CSUM_IP_SCTP
740 #define	CSUM_TSO		(CSUM_IP_TSO|CSUM_IP6_TSO)
741 #define	CSUM_INNER_TSO		(CSUM_INNER_IP_TSO|CSUM_INNER_IP6_TSO)
742 #define	CSUM_UDP_IPV6		CSUM_IP6_UDP
743 #define	CSUM_TCP_IPV6		CSUM_IP6_TCP
744 #define	CSUM_SCTP_IPV6		CSUM_IP6_SCTP
745 #define	CSUM_TLS_MASK		(CSUM_L5_CALC|CSUM_L5_VALID)
746 #define	CSUM_TLS_DECRYPTED	CSUM_L5_CALC
747 
748 /*
749  * mbuf types describing the content of the mbuf (including external storage).
750  */
751 #define	MT_NOTMBUF	0	/* USED INTERNALLY ONLY! Object is not mbuf */
752 #define	MT_DATA		1	/* dynamic (data) allocation */
753 #define	MT_HEADER	MT_DATA	/* packet header, use M_PKTHDR instead */
754 
755 #define	MT_VENDOR1	4	/* for vendor-internal use */
756 #define	MT_VENDOR2	5	/* for vendor-internal use */
757 #define	MT_VENDOR3	6	/* for vendor-internal use */
758 #define	MT_VENDOR4	7	/* for vendor-internal use */
759 
760 #define	MT_SONAME	8	/* socket name */
761 
762 #define	MT_EXP1		9	/* for experimental use */
763 #define	MT_EXP2		10	/* for experimental use */
764 #define	MT_EXP3		11	/* for experimental use */
765 #define	MT_EXP4		12	/* for experimental use */
766 
767 #define	MT_CONTROL	14	/* extra-data protocol message */
768 #define	MT_EXTCONTROL	15	/* control message with externalized contents */
769 #define	MT_OOBDATA	16	/* expedited data  */
770 
771 #define	MT_NOINIT	255	/* Not a type but a flag to allocate
772 				   a non-initialized mbuf */
773 
774 /*
775  * String names of mbuf-related UMA(9) and malloc(9) types.  Exposed to
776  * !_KERNEL so that monitoring tools can look up the zones with
777  * libmemstat(3).
778  */
779 #define	MBUF_MEM_NAME		"mbuf"
780 #define	MBUF_CLUSTER_MEM_NAME	"mbuf_cluster"
781 #define	MBUF_PACKET_MEM_NAME	"mbuf_packet"
782 #define	MBUF_JUMBOP_MEM_NAME	"mbuf_jumbo_page"
783 #define	MBUF_JUMBO9_MEM_NAME	"mbuf_jumbo_9k"
784 #define	MBUF_JUMBO16_MEM_NAME	"mbuf_jumbo_16k"
785 #define	MBUF_TAG_MEM_NAME	"mbuf_tag"
786 #define	MBUF_EXTREFCNT_MEM_NAME	"mbuf_ext_refcnt"
787 #define	MBUF_EXTPGS_MEM_NAME	"mbuf_extpgs"
788 
789 #ifdef _KERNEL
790 union if_snd_tag_alloc_params;
791 
792 #define	MBUF_CHECKSLEEP(how) do {					\
793 	if (how == M_WAITOK)						\
794 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,		\
795 		    "Sleeping in \"%s\"", __func__);			\
796 } while (0)
797 
798 /*
799  * Network buffer allocation API
800  *
801  * The rest of it is defined in kern/kern_mbuf.c
802  */
803 extern uma_zone_t	zone_mbuf;
804 extern uma_zone_t	zone_clust;
805 extern uma_zone_t	zone_pack;
806 extern uma_zone_t	zone_jumbop;
807 extern uma_zone_t	zone_jumbo9;
808 extern uma_zone_t	zone_jumbo16;
809 extern uma_zone_t	zone_extpgs;
810 
811 void		 mb_dupcl(struct mbuf *, struct mbuf *);
812 void		 mb_free_ext(struct mbuf *);
813 void		 mb_free_extpg(struct mbuf *);
814 void		 mb_free_mext_pgs(struct mbuf *);
815 struct mbuf	*mb_alloc_ext_pgs(int, m_ext_free_t, int);
816 struct mbuf	*mb_alloc_ext_plus_pages(int, int);
817 struct mbuf	*mb_mapped_to_unmapped(struct mbuf *, int, int, int,
818 		    struct mbuf **);
819 int		 mb_unmapped_compress(struct mbuf *m);
820 int		 mb_unmapped_to_ext(struct mbuf *m, struct mbuf **mres);
821 void		 mb_free_notready(struct mbuf *m, int count);
822 void		 m_adj(struct mbuf *, int);
823 void		 m_adj_decap(struct mbuf *, int);
824 int		 m_apply(struct mbuf *, int, int,
825 		    int (*)(void *, void *, u_int), void *);
826 int		 m_append(struct mbuf *, int, c_caddr_t);
827 void		 m_cat(struct mbuf *, struct mbuf *);
828 void		 m_catpkt(struct mbuf *, struct mbuf *);
829 int		 m_clget(struct mbuf *m, int how);
830 void 		*m_cljget(struct mbuf *m, int how, int size);
831 struct mbuf	*m_collapse(struct mbuf *, int, int);
832 void		 m_copyback(struct mbuf *, int, int, c_caddr_t);
833 void		 m_copydata(const struct mbuf *, int, int, caddr_t);
834 struct mbuf	*m_copym(struct mbuf *, int, int, int);
835 struct mbuf	*m_copypacket(struct mbuf *, int);
836 void		 m_copy_pkthdr(struct mbuf *, struct mbuf *);
837 struct mbuf	*m_copyup(struct mbuf *, int, int);
838 struct mbuf	*m_defrag(struct mbuf *, int);
839 void		 m_demote_pkthdr(struct mbuf *);
840 void		 m_demote(struct mbuf *, int, int);
841 struct mbuf	*m_devget(char *, int, int, struct ifnet *,
842 		    void (*)(char *, caddr_t, u_int));
843 void		 m_dispose_extcontrolm(struct mbuf *m);
844 struct mbuf	*m_dup(const struct mbuf *, int);
845 int		 m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
846 void		 m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
847 		    void *, void *, int, int);
848 u_int		 m_fixhdr(struct mbuf *);
849 struct mbuf	*m_fragment(struct mbuf *, int, int);
850 void		 m_freem(struct mbuf *);
851 void		 m_freemp(struct mbuf *);
852 void		 m_free_raw(struct mbuf *);
853 struct mbuf	*m_get2(int, int, short, int);
854 struct mbuf	*m_get3(int, int, short, int);
855 struct mbuf	*m_getjcl(int, short, int, int);
856 struct mbuf	*m_getm2(struct mbuf *, int, int, short, int);
857 struct mbuf	*m_getptr(struct mbuf *, int, int *);
858 u_int		 m_length(struct mbuf *, struct mbuf **);
859 int		 m_mbuftouio(struct uio *, const struct mbuf *, int);
860 void		 m_move_pkthdr(struct mbuf *, struct mbuf *);
861 int		 m_pkthdr_init(struct mbuf *, int);
862 struct mbuf	*m_prepend(struct mbuf *, int, int);
863 void		 m_print(const struct mbuf *, int);
864 struct mbuf	*m_pulldown(struct mbuf *, int, int, int *);
865 struct mbuf	*m_pullup(struct mbuf *, int);
866 int		 m_sanity(struct mbuf *, int);
867 struct mbuf	*m_split(struct mbuf *, int, int);
868 struct mbuf	*m_uiotombuf(struct uio *, int, int, int, int);
869 int		 m_unmapped_uiomove(const struct mbuf *, int, struct uio *,
870 		    int);
871 struct mbuf	*m_unshare(struct mbuf *, int);
872 int		 m_snd_tag_alloc(struct ifnet *,
873 		    union if_snd_tag_alloc_params *, struct m_snd_tag **);
874 void		 m_snd_tag_init(struct m_snd_tag *, struct ifnet *,
875 		    const struct if_snd_tag_sw *);
876 void		 m_snd_tag_destroy(struct m_snd_tag *);
877 void		 m_rcvif_serialize(struct mbuf *);
878 struct ifnet	*m_rcvif_restore(struct mbuf *);
879 
880 static __inline int
m_gettype(int size)881 m_gettype(int size)
882 {
883 	int type;
884 
885 	switch (size) {
886 	case MSIZE:
887 		type = EXT_MBUF;
888 		break;
889 	case MCLBYTES:
890 		type = EXT_CLUSTER;
891 		break;
892 #if MJUMPAGESIZE != MCLBYTES
893 	case MJUMPAGESIZE:
894 		type = EXT_JUMBOP;
895 		break;
896 #endif
897 	case MJUM9BYTES:
898 		type = EXT_JUMBO9;
899 		break;
900 	case MJUM16BYTES:
901 		type = EXT_JUMBO16;
902 		break;
903 	default:
904 		panic("%s: invalid cluster size %d", __func__, size);
905 	}
906 
907 	return (type);
908 }
909 
910 /*
911  * Associated an external reference counted buffer with an mbuf.
912  */
913 static __inline void
m_extaddref(struct mbuf * m,char * buf,u_int size,u_int * ref_cnt,m_ext_free_t freef,void * arg1,void * arg2)914 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
915     m_ext_free_t freef, void *arg1, void *arg2)
916 {
917 
918 	KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
919 
920 	atomic_add_int(ref_cnt, 1);
921 	m->m_flags |= M_EXT;
922 	m->m_ext.ext_buf = buf;
923 	m->m_ext.ext_cnt = ref_cnt;
924 	m->m_data = m->m_ext.ext_buf;
925 	m->m_ext.ext_size = size;
926 	m->m_ext.ext_free = freef;
927 	m->m_ext.ext_arg1 = arg1;
928 	m->m_ext.ext_arg2 = arg2;
929 	m->m_ext.ext_type = EXT_EXTREF;
930 	m->m_ext.ext_flags = 0;
931 }
932 
933 static __inline uma_zone_t
m_getzone(int size)934 m_getzone(int size)
935 {
936 	uma_zone_t zone;
937 
938 	switch (size) {
939 	case MCLBYTES:
940 		zone = zone_clust;
941 		break;
942 #if MJUMPAGESIZE != MCLBYTES
943 	case MJUMPAGESIZE:
944 		zone = zone_jumbop;
945 		break;
946 #endif
947 	case MJUM9BYTES:
948 		zone = zone_jumbo9;
949 		break;
950 	case MJUM16BYTES:
951 		zone = zone_jumbo16;
952 		break;
953 	default:
954 		panic("%s: invalid cluster size %d", __func__, size);
955 	}
956 
957 	return (zone);
958 }
959 
960 /*
961  * Initialize an mbuf with linear storage.
962  *
963  * Inline because the consumer text overhead will be roughly the same to
964  * initialize or call a function with this many parameters and M_PKTHDR
965  * should go away with constant propagation for !MGETHDR.
966  */
967 static __inline int
m_init(struct mbuf * m,int how,short type,int flags)968 m_init(struct mbuf *m, int how, short type, int flags)
969 {
970 	int error;
971 
972 	m->m_next = NULL;
973 	m->m_nextpkt = NULL;
974 	m->m_data = m->m_dat;
975 	m->m_len = 0;
976 	m->m_flags = flags;
977 	m->m_type = type;
978 	if (flags & M_PKTHDR)
979 		error = m_pkthdr_init(m, how);
980 	else
981 		error = 0;
982 
983 	MBUF_PROBE5(m__init, m, how, type, flags, error);
984 	return (error);
985 }
986 
987 static __inline struct mbuf *
m_get_raw(int how,short type)988 m_get_raw(int how, short type)
989 {
990 	struct mbuf *m;
991 	struct mb_args args;
992 
993 	args.flags = 0;
994 	args.type = type | MT_NOINIT;
995 	m = uma_zalloc_arg(zone_mbuf, &args, how);
996 	MBUF_PROBE3(m__get_raw, how, type, m);
997 	return (m);
998 }
999 
1000 static __inline struct mbuf *
m_get(int how,short type)1001 m_get(int how, short type)
1002 {
1003 	struct mbuf *m;
1004 	struct mb_args args;
1005 
1006 	args.flags = 0;
1007 	args.type = type;
1008 	m = uma_zalloc_arg(zone_mbuf, &args, how);
1009 	MBUF_PROBE3(m__get, how, type, m);
1010 	return (m);
1011 }
1012 
1013 static __inline struct mbuf *
m_gethdr_raw(int how,short type)1014 m_gethdr_raw(int how, short type)
1015 {
1016 	struct mbuf *m;
1017 	struct mb_args args;
1018 
1019 	args.flags = M_PKTHDR;
1020 	args.type = type | MT_NOINIT;
1021 	m = uma_zalloc_arg(zone_mbuf, &args, how);
1022 	MBUF_PROBE3(m__gethdr_raw, how, type, m);
1023 	return (m);
1024 }
1025 
1026 static __inline struct mbuf *
m_gethdr(int how,short type)1027 m_gethdr(int how, short type)
1028 {
1029 	struct mbuf *m;
1030 	struct mb_args args;
1031 
1032 	args.flags = M_PKTHDR;
1033 	args.type = type;
1034 	m = uma_zalloc_arg(zone_mbuf, &args, how);
1035 	MBUF_PROBE3(m__gethdr, how, type, m);
1036 	return (m);
1037 }
1038 
1039 static __inline struct mbuf *
m_getcl(int how,short type,int flags)1040 m_getcl(int how, short type, int flags)
1041 {
1042 	struct mbuf *m;
1043 	struct mb_args args;
1044 
1045 	args.flags = flags;
1046 	args.type = type;
1047 	m = uma_zalloc_arg(zone_pack, &args, how);
1048 	MBUF_PROBE4(m__getcl, how, type, flags, m);
1049 	return (m);
1050 }
1051 
1052 /*
1053  * XXX: m_cljset() is a dangerous API.  One must attach only a new,
1054  * unreferenced cluster to an mbuf(9).  It is not possible to assert
1055  * that, so care can be taken only by users of the API.
1056  */
1057 static __inline void
m_cljset(struct mbuf * m,void * cl,int type)1058 m_cljset(struct mbuf *m, void *cl, int type)
1059 {
1060 	int size;
1061 
1062 	switch (type) {
1063 	case EXT_CLUSTER:
1064 		size = MCLBYTES;
1065 		break;
1066 #if MJUMPAGESIZE != MCLBYTES
1067 	case EXT_JUMBOP:
1068 		size = MJUMPAGESIZE;
1069 		break;
1070 #endif
1071 	case EXT_JUMBO9:
1072 		size = MJUM9BYTES;
1073 		break;
1074 	case EXT_JUMBO16:
1075 		size = MJUM16BYTES;
1076 		break;
1077 	default:
1078 		panic("%s: unknown cluster type %d", __func__, type);
1079 		break;
1080 	}
1081 
1082 	m->m_data = m->m_ext.ext_buf = cl;
1083 	m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
1084 	m->m_ext.ext_size = size;
1085 	m->m_ext.ext_type = type;
1086 	m->m_ext.ext_flags = EXT_FLAG_EMBREF;
1087 	m->m_ext.ext_count = 1;
1088 	m->m_flags |= M_EXT;
1089 	MBUF_PROBE3(m__cljset, m, cl, type);
1090 }
1091 
1092 static __inline void
m_chtype(struct mbuf * m,short new_type)1093 m_chtype(struct mbuf *m, short new_type)
1094 {
1095 
1096 	m->m_type = new_type;
1097 }
1098 
1099 static __inline void
m_clrprotoflags(struct mbuf * m)1100 m_clrprotoflags(struct mbuf *m)
1101 {
1102 
1103 	while (m) {
1104 		m->m_flags &= ~M_PROTOFLAGS;
1105 		m = m->m_next;
1106 	}
1107 }
1108 
1109 static __inline struct mbuf *
m_last(struct mbuf * m)1110 m_last(struct mbuf *m)
1111 {
1112 
1113 	while (m->m_next)
1114 		m = m->m_next;
1115 	return (m);
1116 }
1117 
1118 static inline u_int
m_extrefcnt(struct mbuf * m)1119 m_extrefcnt(struct mbuf *m)
1120 {
1121 
1122 	KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing for %p", __func__, m));
1123 
1124 	return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
1125 	    *m->m_ext.ext_cnt);
1126 }
1127 
1128 /*
1129  * mbuf, cluster, and external object allocation macros (for compatibility
1130  * purposes).
1131  */
1132 #define	M_MOVE_PKTHDR(to, from)	m_move_pkthdr((to), (from))
1133 #define	MGET(m, how, type)	((m) = m_get((how), (type)))
1134 #define	MGETHDR(m, how, type)	((m) = m_gethdr((how), (type)))
1135 #define	MCLGET(m, how)		m_clget((m), (how))
1136 #define	MEXTADD(m, buf, size, free, arg1, arg2, flags, type)		\
1137     m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2),	\
1138     (flags), (type))
1139 #define	m_getm(m, len, how, type)					\
1140     m_getm2((m), (len), (how), (type), M_PKTHDR)
1141 
1142 /*
1143  * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
1144  * be both the local data payload, or an external buffer area, depending on
1145  * whether M_EXT is set).
1146  */
1147 #define	M_WRITABLE(m)	(((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 &&	\
1148 			 (!(((m)->m_flags & M_EXT)) ||			\
1149 			 (m_extrefcnt(m) == 1)))
1150 
1151 /* Check if the supplied mbuf has a packet header, or else panic. */
1152 #define	M_ASSERTPKTHDR(m)						\
1153 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR,			\
1154 	    ("%s: no mbuf %p packet header!", __func__, (m)))
1155 
1156 /* Check if the supplied mbuf has no send tag, or else panic. */
1157 #define	M_ASSERT_NO_SND_TAG(m)						\
1158 	KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR &&		\
1159 	       ((m)->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0,		\
1160 	    ("%s: receive mbuf %p has send tag!", __func__, (m)))
1161 
1162 /* Check if mbuf is multipage. */
1163 #define M_ASSERTEXTPG(m)						\
1164 	KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG,		\
1165 	    ("%s: m %p is not multipage!", __func__, m))
1166 
1167 /*
1168  * Ensure that the supplied mbuf is a valid, non-free mbuf.
1169  *
1170  * XXX: Broken at the moment.  Need some UMA magic to make it work again.
1171  */
1172 #define	M_ASSERTVALID(m)						\
1173 	KASSERT((((struct mbuf *)(m))->m_flags & 0) == 0,			\
1174 	    ("%s: attempted use of a free mbuf %p!", __func__, (m)))
1175 
1176 /* Check whether any mbuf in the chain is unmapped. */
1177 #ifdef INVARIANTS
1178 #define	M_ASSERTMAPPED(m) do {						\
1179 	for (struct mbuf *__m = (m); __m != NULL; __m = __m->m_next)	\
1180 		KASSERT((__m->m_flags & M_EXTPG) == 0,			\
1181 		    ("%s: chain %p contains an unmapped mbuf", __func__, (m)));\
1182 } while (0)
1183 #else
1184 #define	M_ASSERTMAPPED(m) do {} while (0)
1185 #endif
1186 
1187 /*
1188  * Return the address of the start of the buffer associated with an mbuf,
1189  * handling external storage, packet-header mbufs, and regular data mbufs.
1190  */
1191 #define	M_START(m)							\
1192 	(((m)->m_flags & M_EXTPG) ? NULL :				\
1193 	 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf :			\
1194 	 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] :		\
1195 	 &(m)->m_dat[0])
1196 
1197 /*
1198  * Return the size of the buffer associated with an mbuf, handling external
1199  * storage, packet-header mbufs, and regular data mbufs.
1200  */
1201 #define	M_SIZE(m)							\
1202 	(((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size :			\
1203 	 ((m)->m_flags & M_PKTHDR) ? MHLEN :				\
1204 	 MLEN)
1205 
1206 /*
1207  * Set the m_data pointer of a newly allocated mbuf to place an object of the
1208  * specified size at the end of the mbuf, longword aligned.
1209  *
1210  * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1211  * separate macros, each asserting that it was called at the proper moment.
1212  * This required callers to themselves test the storage type and call the
1213  * right one.  Rather than require callers to be aware of those layout
1214  * decisions, we centralize here.
1215  */
1216 static __inline void
m_align(struct mbuf * m,int len)1217 m_align(struct mbuf *m, int len)
1218 {
1219 	int adjust;
1220 	KASSERT(m->m_data == M_START(m),
1221 	    ("%s: not a virgin mbuf %p", __func__, m));
1222 
1223 	adjust = M_SIZE(m) - len;
1224 	m->m_data += adjust &~ (sizeof(long)-1);
1225 }
1226 
1227 #define	M_ALIGN(m, len)		m_align(m, len)
1228 #define	MH_ALIGN(m, len)	m_align(m, len)
1229 #define	MEXT_ALIGN(m, len)	m_align(m, len)
1230 
1231 /*
1232  * Compute the amount of space available before the current start of data in
1233  * an mbuf.
1234  *
1235  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1236  * of checking writability of the mbuf data area rests solely with the caller.
1237  *
1238  * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1239  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1240  * read-only as well.
1241  */
1242 #define	M_LEADINGSPACE(m)						\
1243 	(M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1244 
1245 /*
1246  * So M_TRAILINGROOM() is for when you want to know how much space
1247  * would be there if it was writable. This can be used to
1248  * detect changes in mbufs by knowing the value at one point
1249  * and then being able to compare it later to the current M_TRAILINGROOM().
1250  * The TRAILINGSPACE() macro is not suitable for this since an mbuf
1251  * at one point might not be writable and then later it becomes writable
1252  * even though the space at the back of it has not changed.
1253  */
1254 #define M_TRAILINGROOM(m) ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len))
1255 /*
1256  * Compute the amount of space available after the end of data in an mbuf.
1257  *
1258  * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1259  * of checking writability of the mbuf data area rests solely with the caller.
1260  *
1261  * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1262  * for mbufs with external storage.  We now allow mbuf-embedded data to be
1263  * read-only as well.
1264  */
1265 #define	M_TRAILINGSPACE(m) (M_WRITABLE(m) ? M_TRAILINGROOM(m) : 0)
1266 
1267 /*
1268  * Arrange to prepend space of size plen to mbuf m.  If a new mbuf must be
1269  * allocated, how specifies whether to wait.  If the allocation fails, the
1270  * original mbuf chain is freed and m is set to NULL.
1271  */
1272 #define	M_PREPEND(m, plen, how) do {					\
1273 	struct mbuf **_mmp = &(m);					\
1274 	struct mbuf *_mm = *_mmp;					\
1275 	int _mplen = (plen);						\
1276 	int __mhow = (how);						\
1277 									\
1278 	MBUF_CHECKSLEEP(how);						\
1279 	if (M_LEADINGSPACE(_mm) >= _mplen) {				\
1280 		_mm->m_data -= _mplen;					\
1281 		_mm->m_len += _mplen;					\
1282 	} else								\
1283 		_mm = m_prepend(_mm, _mplen, __mhow);			\
1284 	if (_mm != NULL && _mm->m_flags & M_PKTHDR)			\
1285 		_mm->m_pkthdr.len += _mplen;				\
1286 	*_mmp = _mm;							\
1287 } while (0)
1288 
1289 /*
1290  * Change mbuf to new type.  This is a relatively expensive operation and
1291  * should be avoided.
1292  */
1293 #define	MCHTYPE(m, t)	m_chtype((m), (t))
1294 
1295 /* Return the rcvif of a packet header. */
1296 static __inline struct ifnet *
m_rcvif(const struct mbuf * m)1297 m_rcvif(const struct mbuf *m)
1298 {
1299 
1300 	M_ASSERTPKTHDR(m);
1301 	if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1302 		return (NULL);
1303 	return (m->m_pkthdr.rcvif);
1304 }
1305 
1306 /* Length to m_copy to copy all. */
1307 #define	M_COPYALL	1000000000
1308 
1309 extern u_int		max_linkhdr;	/* Largest link-level header */
1310 extern u_int		max_hdr;	/* Largest link + protocol header */
1311 extern u_int		max_protohdr;	/* Largest protocol header */
1312 void max_linkhdr_grow(u_int);
1313 void max_protohdr_grow(u_int);
1314 
1315 extern int		nmbclusters;	/* Maximum number of clusters */
1316 extern bool		mb_use_ext_pgs;	/* Use ext_pgs for sendfile */
1317 
1318 /*-
1319  * Network packets may have annotations attached by affixing a list of
1320  * "packet tags" to the pkthdr structure.  Packet tags are dynamically
1321  * allocated semi-opaque data structures that have a fixed header
1322  * (struct m_tag) that specifies the size of the memory block and a
1323  * <cookie,type> pair that identifies it.  The cookie is a 32-bit unique
1324  * unsigned value used to identify a module or ABI.  By convention this value
1325  * is chosen as the date+time that the module is created, expressed as the
1326  * number of seconds since the epoch (e.g., using date -u +'%s').  The type
1327  * value is an ABI/module-specific value that identifies a particular
1328  * annotation and is private to the module.  For compatibility with systems
1329  * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1330  * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1331  * compatibility shim functions and several tag types are defined below.
1332  * Users that do not require compatibility should use a private cookie value
1333  * so that packet tag-related definitions can be maintained privately.
1334  *
1335  * Note that the packet tag returned by m_tag_alloc has the default memory
1336  * alignment implemented by malloc.  To reference private data one can use a
1337  * construct like:
1338  *
1339  *	struct m_tag *mtag = m_tag_alloc(...);
1340  *	struct foo *p = (struct foo *)(mtag+1);
1341  *
1342  * if the alignment of struct m_tag is sufficient for referencing members of
1343  * struct foo.  Otherwise it is necessary to embed struct m_tag within the
1344  * private data structure to insure proper alignment; e.g.,
1345  *
1346  *	struct foo {
1347  *		struct m_tag	tag;
1348  *		...
1349  *	};
1350  *	struct foo *p = (struct foo *) m_tag_alloc(...);
1351  *	struct m_tag *mtag = &p->tag;
1352  */
1353 
1354 /*
1355  * Persistent tags stay with an mbuf until the mbuf is reclaimed.  Otherwise
1356  * tags are expected to ``vanish'' when they pass through a network
1357  * interface.  For most interfaces this happens normally as the tags are
1358  * reclaimed when the mbuf is free'd.  However in some special cases
1359  * reclaiming must be done manually.  An example is packets that pass through
1360  * the loopback interface.  Also, one must be careful to do this when
1361  * ``turning around'' packets (e.g., icmp_reflect).
1362  *
1363  * To mark a tag persistent bit-or this flag in when defining the tag id.
1364  * The tag will then be treated as described above.
1365  */
1366 #define	MTAG_PERSISTENT				0x800
1367 
1368 #define	PACKET_TAG_NONE				0  /* Nadda */
1369 
1370 /* Packet tags for use with PACKET_ABI_COMPAT. */
1371 #define	PACKET_TAG_IPSEC_IN_DONE		1  /* IPsec applied, in */
1372 #define	PACKET_TAG_IPSEC_OUT_DONE		2  /* IPsec applied, out */
1373 #define	PACKET_TAG_IPSEC_IN_CRYPTO_DONE		3  /* NIC IPsec crypto done */
1374 #define	PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED	4  /* NIC IPsec crypto req'ed */
1375 #define	PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO	5  /* NIC notifies IPsec */
1376 #define	PACKET_TAG_IPSEC_PENDING_TDB		6  /* Reminder to do IPsec */
1377 #define	PACKET_TAG_BRIDGE			7  /* Bridge processing done */
1378 #define	PACKET_TAG_GIF				8  /* GIF processing done */
1379 #define	PACKET_TAG_GRE				9  /* GRE processing done */
1380 #define	PACKET_TAG_IN_PACKET_CHECKSUM		10 /* NIC checksumming done */
1381 #define	PACKET_TAG_ENCAP			11 /* Encap.  processing */
1382 #define	PACKET_TAG_IPSEC_SOCKET			12 /* IPSEC socket ref */
1383 #define	PACKET_TAG_IPSEC_HISTORY		13 /* IPSEC history */
1384 #define	PACKET_TAG_IPV6_INPUT			14 /* IPV6 input processing */
1385 #define	PACKET_TAG_DUMMYNET			15 /* dummynet info */
1386 #define	PACKET_TAG_DIVERT			17 /* divert info */
1387 #define	PACKET_TAG_IPFORWARD			18 /* ipforward info */
1388 #define	PACKET_TAG_MACLABEL	(19 | MTAG_PERSISTENT) /* MAC label */
1389 #define	PACKET_TAG_PF				21 /* PF/ALTQ information */
1390 /* was	PACKET_TAG_RTSOCKFAM			25    rtsock sa family */
1391 #define	PACKET_TAG_IPOPTIONS			27 /* Saved IP options */
1392 #define	PACKET_TAG_CARP				28 /* CARP info */
1393 #define	PACKET_TAG_IPSEC_NAT_T_PORTS		29 /* two uint16_t */
1394 #define	PACKET_TAG_ND_OUTGOING			30 /* ND outgoing */
1395 #define	PACKET_TAG_PF_REASSEMBLED		31
1396 #define	PACKET_TAG_IPSEC_ACCEL_OUT		32  /* IPSEC accel out */
1397 #define	PACKET_TAG_IPSEC_ACCEL_IN		33  /* IPSEC accel in */
1398 #define	PACKET_TAG_OVPN				34 /* if_ovpn */
1399 
1400 /* Specific cookies and tags. */
1401 
1402 /* Packet tag routines. */
1403 struct m_tag	*m_tag_alloc(uint32_t, uint16_t, int, int);
1404 void		 m_tag_delete(struct mbuf *, struct m_tag *);
1405 void		 m_tag_delete_chain(struct mbuf *, struct m_tag *);
1406 void		 m_tag_free_default(struct m_tag *);
1407 struct m_tag	*m_tag_locate(struct mbuf *, uint32_t, uint16_t,
1408     struct m_tag *);
1409 struct m_tag	*m_tag_copy(struct m_tag *, int);
1410 int		 m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1411 void		 m_tag_delete_nonpersistent(struct mbuf *);
1412 
1413 /*
1414  * Initialize the list of tags associated with an mbuf.
1415  */
1416 static __inline void
m_tag_init(struct mbuf * m)1417 m_tag_init(struct mbuf *m)
1418 {
1419 
1420 	SLIST_INIT(&m->m_pkthdr.tags);
1421 }
1422 
1423 /*
1424  * Set up the contents of a tag.  Note that this does not fill in the free
1425  * method; the caller is expected to do that.
1426  *
1427  * XXX probably should be called m_tag_init, but that was already taken.
1428  */
1429 static __inline void
m_tag_setup(struct m_tag * t,uint32_t cookie,uint16_t type,int len)1430 m_tag_setup(struct m_tag *t, uint32_t cookie, uint16_t type, int len)
1431 {
1432 
1433 	t->m_tag_id = type;
1434 	t->m_tag_len = len;
1435 	t->m_tag_cookie = cookie;
1436 }
1437 
1438 /*
1439  * Reclaim resources associated with a tag.
1440  */
1441 static __inline void
m_tag_free(struct m_tag * t)1442 m_tag_free(struct m_tag *t)
1443 {
1444 
1445 	(*t->m_tag_free)(t);
1446 }
1447 
1448 /*
1449  * Return the first tag associated with an mbuf.
1450  */
1451 static __inline struct m_tag *
m_tag_first(struct mbuf * m)1452 m_tag_first(struct mbuf *m)
1453 {
1454 
1455 	return (SLIST_FIRST(&m->m_pkthdr.tags));
1456 }
1457 
1458 /*
1459  * Return the next tag in the list of tags associated with an mbuf.
1460  */
1461 static __inline struct m_tag *
m_tag_next(struct mbuf * m __unused,struct m_tag * t)1462 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1463 {
1464 
1465 	return (SLIST_NEXT(t, m_tag_link));
1466 }
1467 
1468 /*
1469  * Prepend a tag to the list of tags associated with an mbuf.
1470  */
1471 static __inline void
m_tag_prepend(struct mbuf * m,struct m_tag * t)1472 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1473 {
1474 
1475 	SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1476 }
1477 
1478 /*
1479  * Unlink a tag from the list of tags associated with an mbuf.
1480  */
1481 static __inline void
m_tag_unlink(struct mbuf * m,struct m_tag * t)1482 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1483 {
1484 
1485 	SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1486 }
1487 
1488 /* These are for OpenBSD compatibility. */
1489 #define	MTAG_ABI_COMPAT		0		/* compatibility ABI */
1490 
1491 static __inline struct m_tag *
m_tag_get(uint16_t type,int length,int wait)1492 m_tag_get(uint16_t type, int length, int wait)
1493 {
1494 	return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1495 }
1496 
1497 static __inline struct m_tag *
m_tag_find(struct mbuf * m,uint16_t type,struct m_tag * start)1498 m_tag_find(struct mbuf *m, uint16_t type, struct m_tag *start)
1499 {
1500 	return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1501 	    m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1502 }
1503 
1504 static inline struct m_snd_tag *
m_snd_tag_ref(struct m_snd_tag * mst)1505 m_snd_tag_ref(struct m_snd_tag *mst)
1506 {
1507 
1508 	refcount_acquire(&mst->refcount);
1509 	return (mst);
1510 }
1511 
1512 static inline void
m_snd_tag_rele(struct m_snd_tag * mst)1513 m_snd_tag_rele(struct m_snd_tag *mst)
1514 {
1515 
1516 	if (refcount_release(&mst->refcount))
1517 		m_snd_tag_destroy(mst);
1518 }
1519 
1520 static __inline struct mbuf *
m_free(struct mbuf * m)1521 m_free(struct mbuf *m)
1522 {
1523 	struct mbuf *n = m->m_next;
1524 
1525 	MBUF_PROBE1(m__free, m);
1526 	if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1527 		m_tag_delete_chain(m, NULL);
1528 	if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1529 		m_snd_tag_rele(m->m_pkthdr.snd_tag);
1530 	if (m->m_flags & M_EXTPG)
1531 		mb_free_extpg(m);
1532 	else if (m->m_flags & M_EXT)
1533 		mb_free_ext(m);
1534 	else if ((m->m_flags & M_NOFREE) == 0)
1535 		uma_zfree(zone_mbuf, m);
1536 	return (n);
1537 }
1538 
1539 static __inline int
rt_m_getfib(struct mbuf * m)1540 rt_m_getfib(struct mbuf *m)
1541 {
1542 	KASSERT(m->m_flags & M_PKTHDR,
1543 	    ("%s: Attempt to get FIB from non header mbuf %p", __func__, m));
1544 	return (m->m_pkthdr.fibnum);
1545 }
1546 
1547 #define M_GETFIB(_m)   rt_m_getfib(_m)
1548 
1549 #define M_SETFIB(_m, _fib) do {						\
1550         KASSERT((_m)->m_flags & M_PKTHDR, \
1551 	    ("%s: Attempt to set FIB on non header mbuf %p", __func__, (_m))); \
1552 	((_m)->m_pkthdr.fibnum) = (_fib);				\
1553 } while (0)
1554 
1555 /* flags passed as first argument for "m_xxx_tcpip_hash()" */
1556 #define	MBUF_HASHFLAG_L2	(1 << 2)
1557 #define	MBUF_HASHFLAG_L3	(1 << 3)
1558 #define	MBUF_HASHFLAG_L4	(1 << 4)
1559 
1560 /* mbuf hashing helper routines */
1561 uint32_t	m_ether_tcpip_hash_init(void);
1562 uint32_t	m_ether_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1563 uint32_t	m_infiniband_tcpip_hash_init(void);
1564 uint32_t	m_infiniband_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1565 
1566 #ifdef MBUF_PROFILING
1567  void m_profile(struct mbuf *m);
1568  #define M_PROFILE(m) m_profile(m)
1569 #else
1570  #define M_PROFILE(m)
1571 #endif
1572 
1573 /*
1574  * Structure describing a packet queue: mbufs linked by m_stailqpkt.
1575  * Does accounting of number of packets and has a cap.
1576  */
1577 struct mbufq {
1578 	STAILQ_HEAD(, mbuf)	mq_head;
1579 	int			mq_len;
1580 	int			mq_maxlen;
1581 };
1582 
1583 static inline void
mbufq_init(struct mbufq * mq,int maxlen)1584 mbufq_init(struct mbufq *mq, int maxlen)
1585 {
1586 
1587 	STAILQ_INIT(&mq->mq_head);
1588 	mq->mq_maxlen = maxlen;
1589 	mq->mq_len = 0;
1590 }
1591 
1592 static inline struct mbuf *
mbufq_flush(struct mbufq * mq)1593 mbufq_flush(struct mbufq *mq)
1594 {
1595 	struct mbuf *m;
1596 
1597 	m = STAILQ_FIRST(&mq->mq_head);
1598 	STAILQ_INIT(&mq->mq_head);
1599 	mq->mq_len = 0;
1600 	return (m);
1601 }
1602 
1603 static inline void
mbufq_drain(struct mbufq * mq)1604 mbufq_drain(struct mbufq *mq)
1605 {
1606 	struct mbuf *m, *n;
1607 
1608 	n = mbufq_flush(mq);
1609 	while ((m = n) != NULL) {
1610 		n = STAILQ_NEXT(m, m_stailqpkt);
1611 		m_freem(m);
1612 	}
1613 }
1614 
1615 static inline struct mbuf *
mbufq_first(const struct mbufq * mq)1616 mbufq_first(const struct mbufq *mq)
1617 {
1618 
1619 	return (STAILQ_FIRST(&mq->mq_head));
1620 }
1621 
1622 static inline struct mbuf *
mbufq_last(const struct mbufq * mq)1623 mbufq_last(const struct mbufq *mq)
1624 {
1625 
1626 	return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1627 }
1628 
1629 static inline bool
mbufq_empty(const struct mbufq * mq)1630 mbufq_empty(const struct mbufq *mq)
1631 {
1632 	return (mq->mq_len == 0);
1633 }
1634 
1635 static inline int
mbufq_full(const struct mbufq * mq)1636 mbufq_full(const struct mbufq *mq)
1637 {
1638 
1639 	return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1640 }
1641 
1642 static inline int
mbufq_len(const struct mbufq * mq)1643 mbufq_len(const struct mbufq *mq)
1644 {
1645 
1646 	return (mq->mq_len);
1647 }
1648 
1649 static inline int
mbufq_enqueue(struct mbufq * mq,struct mbuf * m)1650 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1651 {
1652 
1653 	if (mbufq_full(mq))
1654 		return (ENOBUFS);
1655 	STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1656 	mq->mq_len++;
1657 	return (0);
1658 }
1659 
1660 static inline void
mbufq_remove(struct mbufq * mq,struct mbuf * m)1661 mbufq_remove(struct mbufq *mq, struct mbuf *m)
1662 {
1663 
1664 	STAILQ_REMOVE(&mq->mq_head, m, mbuf, m_stailqpkt);
1665 	mq->mq_len--;
1666 }
1667 
1668 static inline struct mbuf *
mbufq_dequeue(struct mbufq * mq)1669 mbufq_dequeue(struct mbufq *mq)
1670 {
1671 	struct mbuf *m;
1672 
1673 	m = STAILQ_FIRST(&mq->mq_head);
1674 	if (m) {
1675 		STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1676 		m->m_nextpkt = NULL;
1677 		mq->mq_len--;
1678 	}
1679 	return (m);
1680 }
1681 
1682 static inline void
mbufq_prepend(struct mbufq * mq,struct mbuf * m)1683 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1684 {
1685 
1686 	STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1687 	mq->mq_len++;
1688 }
1689 
1690 /*
1691  * Note: this doesn't enforce the maximum list size for dst.
1692  */
1693 static inline void
mbufq_concat(struct mbufq * mq_dst,struct mbufq * mq_src)1694 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1695 {
1696 
1697 	mq_dst->mq_len += mq_src->mq_len;
1698 	STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1699 	mq_src->mq_len = 0;
1700 }
1701 
1702 /*
1703  * Structure describing a chain of mbufs linked by m_stailq, also tracking
1704  * the pointer to the last.  Also does accounting of data length and memory
1705  * usage.
1706  * To be used as an argument to mbuf chain allocation and manipulation KPIs,
1707  * and can be allocated on the stack of a caller.  Kernel facilities may use
1708  * it internally as a most simple implementation of a stream data buffer.
1709  */
1710 struct mchain {
1711 	STAILQ_HEAD(, mbuf) mc_q;
1712 	u_int mc_len;
1713 	u_int mc_mlen;
1714 };
1715 
1716 #define	MCHAIN_INITIALIZER(mc)	\
1717 	(struct mchain){ .mc_q = STAILQ_HEAD_INITIALIZER((mc)->mc_q) }
1718 
1719 static inline struct mbuf *
mc_first(struct mchain * mc)1720 mc_first(struct mchain *mc)
1721 {
1722 	return (STAILQ_FIRST(&mc->mc_q));
1723 }
1724 
1725 static inline struct mbuf *
mc_last(struct mchain * mc)1726 mc_last(struct mchain *mc)
1727 {
1728 	return (STAILQ_LAST(&mc->mc_q, mbuf, m_stailq));
1729 }
1730 
1731 static inline bool
mc_empty(struct mchain * mc)1732 mc_empty(struct mchain *mc)
1733 {
1734 	return (STAILQ_EMPTY(&mc->mc_q));
1735 }
1736 
1737 /* Account addition of m to mc. */
1738 static inline void
mc_inc(struct mchain * mc,struct mbuf * m)1739 mc_inc(struct mchain *mc, struct mbuf *m)
1740 {
1741 	mc->mc_len += m->m_len;
1742 	mc->mc_mlen += MSIZE;
1743 	if (m->m_flags & M_EXT)
1744 		mc->mc_mlen += m->m_ext.ext_size;
1745 }
1746 
1747 /* Account removal of m from mc. */
1748 static inline void
mc_dec(struct mchain * mc,struct mbuf * m)1749 mc_dec(struct mchain *mc, struct mbuf *m)
1750 {
1751 	MPASS(mc->mc_len >= m->m_len);
1752 	mc->mc_len -= m->m_len;
1753 	MPASS(mc->mc_mlen >= MSIZE);
1754 	mc->mc_mlen -= MSIZE;
1755 	if (m->m_flags & M_EXT) {
1756 		MPASS(mc->mc_mlen >= m->m_ext.ext_size);
1757 		mc->mc_mlen -= m->m_ext.ext_size;
1758 	}
1759 }
1760 
1761 static inline void
mc_init(struct mchain * mc)1762 mc_init(struct mchain *mc)
1763 {
1764 	STAILQ_INIT(&mc->mc_q);
1765 	mc->mc_len = mc->mc_mlen = 0;
1766 }
1767 
1768 /*
1769  * Get mchain from a classic mbuf chain linked by m_next.  Two hacks here:
1770  * we use the fact that m_next is alias to m_stailq, we use internal queue(3)
1771  * fields.
1772  */
1773 static inline void
mc_init_m(struct mchain * mc,struct mbuf * m)1774 mc_init_m(struct mchain *mc, struct mbuf *m)
1775 {
1776 	struct mbuf *last;
1777 
1778 	STAILQ_FIRST(&mc->mc_q) = m;
1779 	mc->mc_len = mc->mc_mlen = 0;
1780 	STAILQ_FOREACH(m, &mc->mc_q, m_stailq) {
1781 		mc_inc(mc, m);
1782 		last = m;
1783 	}
1784 	mc->mc_q.stqh_last = &STAILQ_NEXT(last, m_stailq);
1785 }
1786 
1787 static inline void
mc_freem(struct mchain * mc)1788 mc_freem(struct mchain *mc)
1789 {
1790 	if (!mc_empty(mc))
1791 		m_freem(mc_first(mc));
1792 }
1793 
1794 static inline void
mc_prepend(struct mchain * mc,struct mbuf * m)1795 mc_prepend(struct mchain *mc, struct mbuf *m)
1796 {
1797 	STAILQ_INSERT_HEAD(&mc->mc_q, m, m_stailq);
1798 	mc_inc(mc, m);
1799 }
1800 
1801 static inline void
mc_append(struct mchain * mc,struct mbuf * m)1802 mc_append(struct mchain *mc, struct mbuf *m)
1803 {
1804 	STAILQ_INSERT_TAIL(&mc->mc_q, m, m_stailq);
1805 	mc_inc(mc, m);
1806 }
1807 
1808 static inline void
mc_concat(struct mchain * head,struct mchain * tail)1809 mc_concat(struct mchain *head, struct mchain *tail)
1810 {
1811 	STAILQ_CONCAT(&head->mc_q, &tail->mc_q);
1812 	head->mc_len += tail->mc_len;
1813 	head->mc_mlen += tail->mc_mlen;
1814 	tail->mc_len = tail->mc_mlen = 0;
1815 }
1816 
1817 /*
1818  * Note: STAILQ_REMOVE() is expensive. mc_remove_after() needs to be provided
1819  * as long as there consumers that would benefit from it.
1820  */
1821 static inline void
mc_remove(struct mchain * mc,struct mbuf * m)1822 mc_remove(struct mchain *mc, struct mbuf *m)
1823 {
1824 	STAILQ_REMOVE(&mc->mc_q, m, mbuf, m_stailq);
1825 	mc_dec(mc, m);
1826 }
1827 
1828 int mc_get(struct mchain *, u_int, int, short, int);
1829 int mc_split(struct mchain *, struct mchain *, u_int, int);
1830 int mc_uiotomc(struct mchain *, struct uio *, u_int, u_int, int, int);
1831 
1832 #ifdef _SYS_TIMESPEC_H_
1833 static inline void
mbuf_tstmp2timespec(struct mbuf * m,struct timespec * ts)1834 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1835 {
1836 
1837 	M_ASSERTPKTHDR(m);
1838 	KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1839 	    ("%s: mbuf %p no M_TSTMP or M_TSTMP_LRO", __func__, m));
1840 	ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1841 	ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1842 }
1843 #endif
1844 
1845 static inline void
mbuf_tstmp2timeval(struct mbuf * m,struct timeval * tv)1846 mbuf_tstmp2timeval(struct mbuf *m, struct timeval *tv)
1847 {
1848 
1849 	M_ASSERTPKTHDR(m);
1850 	KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1851 	    ("%s: mbuf %p no M_TSTMP or M_TSTMP_LRO", __func__, m));
1852 	tv->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1853 	tv->tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000) / 1000;
1854 }
1855 
1856 #ifdef DEBUGNET
1857 /* Invoked from the debugnet client code. */
1858 void	debugnet_mbuf_drain(void);
1859 void	debugnet_mbuf_start(void);
1860 void	debugnet_mbuf_finish(void);
1861 void	debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1862 #endif
1863 
1864 static inline bool
mbuf_has_tls_session(struct mbuf * m)1865 mbuf_has_tls_session(struct mbuf *m)
1866 {
1867 
1868 	if (m->m_flags & M_EXTPG) {
1869 		if (m->m_epg_tls != NULL) {
1870 			return (true);
1871 		}
1872 	}
1873 	return (false);
1874 }
1875 
1876 #endif /* _KERNEL */
1877 #endif /* !_SYS_MBUF_H_ */
1878