2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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.
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
32 * @(#)mbuf.h 8.5 (Berkeley) 2/19/95
38 /* XXX: These includes suck. Sorry! */
39 #include <sys/queue.h>
41 #include <sys/systm.h>
42 #include <sys/refcount.h>
47 #define MBUF_PROBE1(probe, arg0) \
48 SDT_PROBE1(sdt, , , probe, arg0)
49 #define MBUF_PROBE2(probe, arg0, arg1) \
50 SDT_PROBE2(sdt, , , probe, arg0, arg1)
51 #define MBUF_PROBE3(probe, arg0, arg1, arg2) \
52 SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
53 #define MBUF_PROBE4(probe, arg0, arg1, arg2, arg3) \
54 SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
55 #define MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4) \
56 SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
58 SDT_PROBE_DECLARE(sdt, , , m__init);
59 SDT_PROBE_DECLARE(sdt, , , m__gethdr_raw);
60 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
61 SDT_PROBE_DECLARE(sdt, , , m__get_raw);
62 SDT_PROBE_DECLARE(sdt, , , m__get);
63 SDT_PROBE_DECLARE(sdt, , , m__getcl);
64 SDT_PROBE_DECLARE(sdt, , , m__getjcl);
65 SDT_PROBE_DECLARE(sdt, , , m__clget);
66 SDT_PROBE_DECLARE(sdt, , , m__cljget);
67 SDT_PROBE_DECLARE(sdt, , , m__cljset);
68 SDT_PROBE_DECLARE(sdt, , , m__free);
69 SDT_PROBE_DECLARE(sdt, , , m__freem);
74 * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
75 * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
76 * sys/param.h), which has no additional overhead and is used instead of the
77 * internal data area; this is done when at least MINCLSIZE of data must be
78 * stored. Additionally, it is possible to allocate a separate buffer
79 * externally and attach it to the mbuf in a way similar to that of mbuf
82 * NB: These calculation do not take actual compiler-induced alignment and
83 * padding inside the complete struct mbuf into account. Appropriate
84 * attention is required when changing members of struct mbuf.
86 * MLEN is data length in a normal mbuf.
87 * MHLEN is data length in an mbuf with pktheader.
88 * MINCLSIZE is a smallest amount of data that should be put into cluster.
90 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
94 #define MHSIZE offsetof(struct mbuf, m_dat)
95 #define MPKTHSIZE offsetof(struct mbuf, m_pktdat)
96 #define MLEN ((int)(MSIZE - MHSIZE))
97 #define MHLEN ((int)(MSIZE - MPKTHSIZE))
98 #define MINCLSIZE (MHLEN + 1)
103 * Macro for type conversion: convert mbuf pointer to data pointer of correct
106 * mtod(m, t) -- Convert mbuf pointer to data pointer of correct type.
107 * mtodo(m, o) -- Same as above but with offset 'o' into data.
109 #define mtod(m, t) ((t)((m)->m_data))
110 #define mtodo(m, o) ((void *)(((m)->m_data) + (o)))
113 * Argument structure passed to UMA routines during mbuf and packet
117 int flags; /* Flags for mbuf being allocated */
118 short type; /* Type of mbuf being allocated */
123 * Packet tag structure (see below for details).
126 SLIST_ENTRY(m_tag) m_tag_link; /* List of packet tags */
127 u_int16_t m_tag_id; /* Tag ID */
128 u_int16_t m_tag_len; /* Length of data */
129 u_int32_t m_tag_cookie; /* ABI/Module ID */
130 void (*m_tag_free)(struct m_tag *);
134 * Static network interface owned tag.
135 * Allocated through ifp->if_snd_tag_alloc().
137 struct if_snd_tag_sw;
140 struct ifnet *ifp; /* network interface tag belongs to */
141 const struct if_snd_tag_sw *sw;
142 volatile u_int refcount;
146 * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
149 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
154 struct m_snd_tag *snd_tag; /* send tag, if any */
155 struct ifnet *rcvif; /* rcv interface */
157 uint16_t rcvidx; /* rcv interface index ... */
158 uint16_t rcvgen; /* ... and generation count */
162 struct ifnet *leaf_rcvif; /* leaf rcv interface */
164 uint16_t leaf_rcvidx; /* leaf rcv interface index ... */
165 uint16_t leaf_rcvgen; /* ... and generation count */
168 SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
169 int32_t len; /* total packet length */
171 /* Layer crossing persistent information. */
172 uint32_t flowid; /* packet's 4-tuple system */
173 uint32_t csum_flags; /* checksum and offload features */
174 uint16_t fibnum; /* this packet should use this fib */
175 uint8_t numa_domain; /* NUMA domain of recvd pkt */
176 uint8_t rsstype; /* hash type */
177 #if !defined(__LP64__)
178 uint32_t pad; /* pad for 64bit alignment */
181 uint64_t rcv_tstmp; /* timestamp in ns */
183 uint8_t l2hlen; /* layer 2 hdr len */
184 uint8_t l3hlen; /* layer 3 hdr len */
185 uint8_t l4hlen; /* layer 4 hdr len */
186 uint8_t l5hlen; /* layer 5 hdr len */
187 uint8_t inner_l2hlen;
188 uint8_t inner_l3hlen;
189 uint8_t inner_l4hlen;
190 uint8_t inner_l5hlen;
196 uint32_t thirtytwo[2];
197 uint64_t sixtyfour[1];
198 uintptr_t unintptr[1];
202 /* Layer specific non-persistent local storage for reassembly, etc. */
207 uint32_t thirtytwo[2];
208 uint64_t sixtyfour[1];
209 uintptr_t unintptr[1];
212 /* Upon allocation: total packet memory consumption. */
216 #define ether_vtag PH_per.sixteen[0]
217 #define tcp_tun_port PH_per.sixteen[0] /* outbound */
218 #define vt_nrecs PH_per.sixteen[0] /* mld and v6-ND */
219 #define tso_segsz PH_per.sixteen[1] /* inbound after LRO */
220 #define lro_nsegs tso_segsz /* inbound after LRO */
221 #define csum_data PH_per.thirtytwo[1] /* inbound from hardware up */
222 #define lro_tcp_d_len PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
223 #define lro_tcp_d_csum PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
224 #define lro_tcp_h_off PH_loc.sixteen[2] /* inbound during LRO (no reassembly) */
225 #define lro_etype PH_loc.sixteen[3] /* inbound during LRO (no reassembly) */
226 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
229 * TLS records for TLS 1.0-1.2 can have the following header lengths:
230 * - 5 (AES-CBC with implicit IV)
231 * - 21 (AES-CBC with explicit IV)
232 * - 13 (AES-GCM with 8 byte explicit IV)
234 #define MBUF_PEXT_HDR_LEN 23
237 * TLS records for TLS 1.0-1.2 can have the following maximum trailer
240 * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
241 * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
242 * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
244 #define MBUF_PEXT_TRAIL_LEN 64
246 #if defined(__LP64__)
247 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
249 #define MBUF_PEXT_MAX_PGS (64 / sizeof(vm_paddr_t))
252 #define MBUF_PEXT_MAX_BYTES \
253 (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
259 * Description of external storage mapped into mbuf; valid only if M_EXT is
263 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
266 typedef void m_ext_free_t(struct mbuf *);
270 * If EXT_FLAG_EMBREF is set, then we use refcount in the
271 * mbuf, the 'ext_count' member. Otherwise, we have a
272 * shadow copy and we use pointer 'ext_cnt'. The original
273 * mbuf is responsible to carry the pointer to free routine
274 * and its arguments. They aren't copied into shadows in
275 * mb_dupcl() to avoid dereferencing next cachelines.
277 volatile u_int ext_count;
278 volatile u_int *ext_cnt;
280 uint32_t ext_size; /* size of buffer, for ext_free */
281 uint32_t ext_type:8, /* type of external storage */
282 ext_flags:24; /* external storage mbuf flags */
286 * Regular M_EXT mbuf:
287 * o ext_buf always points to the external buffer.
288 * o ext_free (below) and two optional arguments
289 * ext_arg1 and ext_arg2 store the free context for
290 * the external storage. They are set only in the
291 * refcount carrying mbuf, the one with
292 * EXT_FLAG_EMBREF flag, with exclusion for
293 * EXT_EXTREF type, where the free context is copied
294 * into all mbufs that use same external storage.
296 char *ext_buf; /* start of buffer */
297 #define m_ext_copylen offsetof(struct m_ext, ext_arg2)
302 * Multi-page M_EXTPG mbuf:
303 * o extpg_pa - page vector.
304 * o extpg_trail and extpg_hdr - TLS trailer and
306 * Uses ext_free and may also use ext_arg1.
308 vm_paddr_t extpg_pa[MBUF_PEXT_MAX_PGS];
309 char extpg_trail[MBUF_PEXT_TRAIL_LEN];
310 char extpg_hdr[MBUF_PEXT_HDR_LEN];
311 /* Pretend these 3 fields are part of mbuf itself. */
312 #define m_epg_pa m_ext.extpg_pa
313 #define m_epg_trail m_ext.extpg_trail
314 #define m_epg_hdr m_ext.extpg_hdr
315 #define m_epg_ext_copylen offsetof(struct m_ext, ext_free)
319 * Free method and optional argument pointer, both
320 * used by M_EXT and M_EXTPG.
322 m_ext_free_t *ext_free;
327 * The core of the mbuf object along with some shortcut defines for practical
332 * Header present at the beginning of every mbuf.
335 * Compile-time assertions in uipc_mbuf.c test these values to ensure
336 * that they are correct.
338 union { /* next buffer in chain */
340 SLIST_ENTRY(mbuf) m_slist;
341 STAILQ_ENTRY(mbuf) m_stailq;
343 union { /* next chain in queue/record */
344 struct mbuf *m_nextpkt;
345 SLIST_ENTRY(mbuf) m_slistpkt;
346 STAILQ_ENTRY(mbuf) m_stailqpkt;
348 caddr_t m_data; /* location of data */
349 int32_t m_len; /* amount of data in this mbuf */
350 uint32_t m_type:8, /* type of data in this mbuf */
351 m_flags:24; /* flags; see below */
352 #if !defined(__LP64__)
353 uint32_t m_pad; /* pad for 64bit alignment */
357 * A set of optional headers (packet header, external storage header)
358 * and internal data storage. Historically, these arrays were sized
359 * to MHLEN (space left after a packet header) and MLEN (space left
360 * after only a regular mbuf header); they are now variable size in
361 * order to support future work on variable-size mbufs.
367 struct pkthdr m_pkthdr;
370 * Multi-page M_EXTPG mbuf has its meta data
371 * split between the below anonymous structure
372 * and m_ext. It carries vector of pages,
373 * optional header and trailer char vectors
374 * and pointers to socket/TLS data.
376 #define m_epg_startcopy m_epg_npgs
377 #define m_epg_endcopy m_epg_stailq
379 /* Overall count of pages and count of
380 * pages with I/O pending. */
383 /* TLS header and trailer lengths.
384 * The data itself resides in m_ext. */
385 uint8_t m_epg_hdrlen;
386 uint8_t m_epg_trllen;
387 /* Offset into 1st page and length of
388 * data in the last page. */
389 uint16_t m_epg_1st_off;
390 uint16_t m_epg_last_len;
392 #define EPG_FLAG_ANON 0x1 /* Data can be encrypted in place. */
393 #define EPG_FLAG_2FREE 0x2 /* Scheduled for free. */
394 uint8_t m_epg_record_type;
397 struct ktls_session *m_epg_tls;
398 struct socket *m_epg_so;
399 uint64_t m_epg_seqno;
400 STAILQ_ENTRY(mbuf) m_epg_stailq;
404 /* M_EXT or M_EXTPG set. */
406 /* M_PKTHDR set, neither M_EXT nor M_EXTPG. */
410 char m_dat[0]; /* !M_PKTHDR, !M_EXT */
416 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff)
419 KASSERT(pgoff == 0 || pidx == 0,
420 ("page %d with non-zero offset %d in %p", pidx, pgoff, m));
422 if (pidx == m->m_epg_npgs - 1) {
423 return (m->m_epg_last_len);
425 return (PAGE_SIZE - pgoff);
430 #define MCHECK(ex, msg) KASSERT((ex), \
431 ("Multi page mbuf %p with " #msg " at %s:%d", \
432 m, __FILE__, __LINE__))
434 * NB: This expects a non-empty buffer (npgs > 0 and
437 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do { \
438 MCHECK(m->m_epg_npgs > 0, "no valid pages"); \
439 MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa), \
441 MCHECK(m->m_epg_nrdy <= m->m_epg_npgs, \
442 "too many ready pages"); \
443 MCHECK(m->m_epg_1st_off < PAGE_SIZE, \
444 "too large page offset"); \
445 MCHECK(m->m_epg_last_len > 0, "zero last page length"); \
446 MCHECK(m->m_epg_last_len <= PAGE_SIZE, \
447 "too large last page length"); \
448 if (m->m_epg_npgs == 1) \
449 MCHECK(m->m_epg_1st_off + \
450 m->m_epg_last_len <= PAGE_SIZE, \
451 "single page too large"); \
452 MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr), \
453 "too large header length"); \
454 MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail), \
455 "too large header length"); \
458 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do {} while (0)
463 * mbuf flags of global significance and layer crossing.
464 * Those of only protocol/layer specific significance are to be mapped
465 * to M_PROTO[1-11] and cleared at layer handoff boundaries.
466 * NB: Limited to the lower 24 bits.
468 #define M_EXT 0x00000001 /* has associated external storage */
469 #define M_PKTHDR 0x00000002 /* start of record */
470 #define M_EOR 0x00000004 /* end of record */
471 #define M_RDONLY 0x00000008 /* associated data is marked read-only */
472 #define M_BCAST 0x00000010 /* send/received as link-level broadcast */
473 #define M_MCAST 0x00000020 /* send/received as link-level multicast */
474 #define M_PROMISC 0x00000040 /* packet was not for us */
475 #define M_VLANTAG 0x00000080 /* ether_vtag is valid */
476 #define M_EXTPG 0x00000100 /* has array of unmapped pages and TLS */
477 #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */
478 #define M_TSTMP 0x00000400 /* rcv_tstmp field is valid */
479 #define M_TSTMP_HPREC 0x00000800 /* rcv_tstmp is high-prec, typically
480 hw-stamped on port (useful for IEEE 1588
482 #define M_TSTMP_LRO 0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
484 #define M_PROTO1 0x00002000 /* protocol-specific */
485 #define M_PROTO2 0x00004000 /* protocol-specific */
486 #define M_PROTO3 0x00008000 /* protocol-specific */
487 #define M_PROTO4 0x00010000 /* protocol-specific */
488 #define M_PROTO5 0x00020000 /* protocol-specific */
489 #define M_PROTO6 0x00040000 /* protocol-specific */
490 #define M_PROTO7 0x00080000 /* protocol-specific */
491 #define M_PROTO8 0x00100000 /* protocol-specific */
492 #define M_PROTO9 0x00200000 /* protocol-specific */
493 #define M_PROTO10 0x00400000 /* protocol-specific */
494 #define M_PROTO11 0x00800000 /* protocol-specific */
497 * Flags to purge when crossing layers.
499 #define M_PROTOFLAGS \
500 (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
501 M_PROTO9|M_PROTO10|M_PROTO11)
504 * Flags preserved when copying m_pkthdr.
506 #define M_COPYFLAGS \
507 (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
508 M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
511 * Flags preserved during demote.
513 #define M_DEMOTEFLAGS \
514 (M_EXT | M_RDONLY | M_NOFREE | M_EXTPG)
517 * Mbuf flag description for use with printf(9) %b identifier.
519 #define M_FLAG_BITS \
520 "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
521 "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
522 #define M_FLAG_PROTOBITS \
523 "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
524 "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
525 "\27M_PROTO10\28M_PROTO11"
526 #define M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
529 * Network interface cards are able to hash protocol fields (such as IPv4
530 * addresses and TCP port numbers) classify packets into flows. These flows
531 * can then be used to maintain ordering while delivering packets to the OS
532 * via parallel input queues, as well as to provide a stateless affinity
533 * model. NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
534 * m_flag fields to indicate how the hash should be interpreted by the
537 * Most NICs support RSS, which provides ordering and explicit affinity, and
538 * use the hash m_flag bits to indicate what header fields were covered by
539 * the hash. M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
540 * RSS cards or configurations that provide an opaque flow identifier, allowing
541 * for ordering and distribution without explicit affinity. Additionally,
542 * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
545 * The meaning of the IPV6_EX suffix:
546 * "o Home address from the home address option in the IPv6 destination
547 * options header. If the extension header is not present, use the Source
549 * o IPv6 address that is contained in the Routing-Header-Type-2 from the
550 * associated extension header. If the extension header is not present,
551 * use the Destination IPv6 Address."
553 * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
555 #define M_HASHTYPE_HASHPROP 0x80 /* has hash properties */
556 #define M_HASHTYPE_INNER 0x40 /* calculated from inner headers */
557 #define M_HASHTYPE_HASH(t) (M_HASHTYPE_HASHPROP | (t))
558 /* Microsoft RSS standard hash types */
559 #define M_HASHTYPE_NONE 0
560 #define M_HASHTYPE_RSS_IPV4 M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
561 #define M_HASHTYPE_RSS_TCP_IPV4 M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
562 #define M_HASHTYPE_RSS_IPV6 M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
563 #define M_HASHTYPE_RSS_TCP_IPV6 M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
564 #define M_HASHTYPE_RSS_IPV6_EX M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
566 #define M_HASHTYPE_RSS_TCP_IPV6_EX M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
568 #define M_HASHTYPE_RSS_UDP_IPV4 M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
569 #define M_HASHTYPE_RSS_UDP_IPV6 M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
570 #define M_HASHTYPE_RSS_UDP_IPV6_EX M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
573 #define M_HASHTYPE_OPAQUE 0x3f /* ordering, not affinity */
574 #define M_HASHTYPE_OPAQUE_HASH M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
575 /* ordering+hash, not affinity*/
577 #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0)
578 #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype & ~M_HASHTYPE_INNER)
579 #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v))
580 #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v))
581 #define M_HASHTYPE_ISHASH(m) \
582 (((m)->m_pkthdr.rsstype & M_HASHTYPE_HASHPROP) != 0)
583 #define M_HASHTYPE_SETINNER(m) do { \
584 (m)->m_pkthdr.rsstype |= M_HASHTYPE_INNER; \
588 * External mbuf storage buffer types.
590 #define EXT_CLUSTER 1 /* mbuf cluster */
591 #define EXT_SFBUF 2 /* sendfile(2)'s sf_buf */
592 #define EXT_JUMBOP 3 /* jumbo cluster page sized */
593 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */
594 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */
595 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */
596 #define EXT_MBUF 7 /* external mbuf reference */
597 #define EXT_RXRING 8 /* data in NIC receive ring */
599 #define EXT_VENDOR1 224 /* for vendor-internal use */
600 #define EXT_VENDOR2 225 /* for vendor-internal use */
601 #define EXT_VENDOR3 226 /* for vendor-internal use */
602 #define EXT_VENDOR4 227 /* for vendor-internal use */
604 #define EXT_EXP1 244 /* for experimental use */
605 #define EXT_EXP2 245 /* for experimental use */
606 #define EXT_EXP3 246 /* for experimental use */
607 #define EXT_EXP4 247 /* for experimental use */
609 #define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */
610 #define EXT_MOD_TYPE 253 /* custom module's ext_buf type */
611 #define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */
612 #define EXT_EXTREF 255 /* has externally maintained ext_cnt ptr */
615 * Flags for external mbuf buffer types.
616 * NB: limited to the lower 24 bits.
618 #define EXT_FLAG_EMBREF 0x000001 /* embedded ext_count */
619 #define EXT_FLAG_EXTREF 0x000002 /* external ext_cnt, notyet */
621 #define EXT_FLAG_NOFREE 0x000010 /* don't free mbuf to pool, notyet */
623 #define EXT_FLAG_VENDOR1 0x010000 /* These flags are vendor */
624 #define EXT_FLAG_VENDOR2 0x020000 /* or submodule specific, */
625 #define EXT_FLAG_VENDOR3 0x040000 /* not used by mbuf code. */
626 #define EXT_FLAG_VENDOR4 0x080000 /* Set/read by submodule. */
628 #define EXT_FLAG_EXP1 0x100000 /* for experimental use */
629 #define EXT_FLAG_EXP2 0x200000 /* for experimental use */
630 #define EXT_FLAG_EXP3 0x400000 /* for experimental use */
631 #define EXT_FLAG_EXP4 0x800000 /* for experimental use */
634 * EXT flag description for use with printf(9) %b identifier.
636 #define EXT_FLAG_BITS \
637 "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
638 "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
639 "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
643 * Flags indicating checksum, segmentation and other offload work to be
644 * done, or already done, by hardware or lower layers. It is split into
645 * separate inbound and outbound flags.
647 * Outbound flags that are set by upper protocol layers requesting lower
648 * layers, or ideally the hardware, to perform these offloading tasks.
649 * For outbound packets this field and its flags can be directly tested
650 * against ifnet if_hwassist. Note that the outbound and the inbound flags do
651 * not collide right now but they could be allowed to (as long as the flags are
652 * scrubbed appropriately when the direction of an mbuf changes). CSUM_BITS
653 * would also have to split into CSUM_BITS_TX and CSUM_BITS_RX.
655 * CSUM_INNER_<x> is the same as CSUM_<x> but it applies to the inner frame.
656 * The CSUM_ENCAP_<x> bits identify the outer encapsulation.
658 #define CSUM_IP 0x00000001 /* IP header checksum offload */
659 #define CSUM_IP_UDP 0x00000002 /* UDP checksum offload */
660 #define CSUM_IP_TCP 0x00000004 /* TCP checksum offload */
661 #define CSUM_IP_SCTP 0x00000008 /* SCTP checksum offload */
662 #define CSUM_IP_TSO 0x00000010 /* TCP segmentation offload */
663 #define CSUM_IP_ISCSI 0x00000020 /* iSCSI checksum offload */
665 #define CSUM_INNER_IP6_UDP 0x00000040
666 #define CSUM_INNER_IP6_TCP 0x00000080
667 #define CSUM_INNER_IP6_TSO 0x00000100
668 #define CSUM_IP6_UDP 0x00000200 /* UDP checksum offload */
669 #define CSUM_IP6_TCP 0x00000400 /* TCP checksum offload */
670 #define CSUM_IP6_SCTP 0x00000800 /* SCTP checksum offload */
671 #define CSUM_IP6_TSO 0x00001000 /* TCP segmentation offload */
672 #define CSUM_IP6_ISCSI 0x00002000 /* iSCSI checksum offload */
674 #define CSUM_INNER_IP 0x00004000
675 #define CSUM_INNER_IP_UDP 0x00008000
676 #define CSUM_INNER_IP_TCP 0x00010000
677 #define CSUM_INNER_IP_TSO 0x00020000
679 #define CSUM_ENCAP_VXLAN 0x00040000 /* VXLAN outer encapsulation */
680 #define CSUM_ENCAP_RSVD1 0x00080000
682 /* Inbound checksum support where the checksum was verified by hardware. */
683 #define CSUM_INNER_L3_CALC 0x00100000
684 #define CSUM_INNER_L3_VALID 0x00200000
685 #define CSUM_INNER_L4_CALC 0x00400000
686 #define CSUM_INNER_L4_VALID 0x00800000
687 #define CSUM_L3_CALC 0x01000000 /* calculated layer 3 csum */
688 #define CSUM_L3_VALID 0x02000000 /* checksum is correct */
689 #define CSUM_L4_CALC 0x04000000 /* calculated layer 4 csum */
690 #define CSUM_L4_VALID 0x08000000 /* checksum is correct */
691 #define CSUM_L5_CALC 0x10000000 /* calculated layer 5 csum */
692 #define CSUM_L5_VALID 0x20000000 /* checksum is correct */
693 #define CSUM_COALESCED 0x40000000 /* contains merged segments */
695 #define CSUM_SND_TAG 0x80000000 /* Packet header has send tag */
697 #define CSUM_FLAGS_TX (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP | CSUM_IP_SCTP | \
698 CSUM_IP_TSO | CSUM_IP_ISCSI | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | \
699 CSUM_INNER_IP6_TSO | CSUM_IP6_UDP | CSUM_IP6_TCP | CSUM_IP6_SCTP | \
700 CSUM_IP6_TSO | CSUM_IP6_ISCSI | CSUM_INNER_IP | CSUM_INNER_IP_UDP | \
701 CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN | \
702 CSUM_ENCAP_RSVD1 | CSUM_SND_TAG)
704 #define CSUM_FLAGS_RX (CSUM_INNER_L3_CALC | CSUM_INNER_L3_VALID | \
705 CSUM_INNER_L4_CALC | CSUM_INNER_L4_VALID | CSUM_L3_CALC | CSUM_L3_VALID | \
706 CSUM_L4_CALC | CSUM_L4_VALID | CSUM_L5_CALC | CSUM_L5_VALID | \
710 * CSUM flag description for use with printf(9) %b identifier.
713 "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
714 "\6CSUM_IP_ISCSI\7CSUM_INNER_IP6_UDP\10CSUM_INNER_IP6_TCP" \
715 "\11CSUM_INNER_IP6_TSO\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP" \
716 "\15CSUM_IP6_TSO\16CSUM_IP6_ISCSI\17CSUM_INNER_IP\20CSUM_INNER_IP_UDP" \
717 "\21CSUM_INNER_IP_TCP\22CSUM_INNER_IP_TSO\23CSUM_ENCAP_VXLAN" \
718 "\24CSUM_ENCAP_RSVD1\25CSUM_INNER_L3_CALC\26CSUM_INNER_L3_VALID" \
719 "\27CSUM_INNER_L4_CALC\30CSUM_INNER_L4_VALID\31CSUM_L3_CALC" \
720 "\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID\35CSUM_L5_CALC" \
721 "\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
723 /* CSUM flags compatibility mappings. */
724 #define CSUM_IP_CHECKED CSUM_L3_CALC
725 #define CSUM_IP_VALID CSUM_L3_VALID
726 #define CSUM_DATA_VALID CSUM_L4_VALID
727 #define CSUM_PSEUDO_HDR CSUM_L4_CALC
728 #define CSUM_SCTP_VALID CSUM_L4_VALID
729 #define CSUM_DELAY_DATA (CSUM_TCP|CSUM_UDP)
730 #define CSUM_DELAY_IP CSUM_IP /* Only v4, no v6 IP hdr csum */
731 #define CSUM_DELAY_DATA_IPV6 (CSUM_TCP_IPV6|CSUM_UDP_IPV6)
732 #define CSUM_DATA_VALID_IPV6 CSUM_DATA_VALID
733 #define CSUM_TCP CSUM_IP_TCP
734 #define CSUM_UDP CSUM_IP_UDP
735 #define CSUM_SCTP CSUM_IP_SCTP
736 #define CSUM_TSO (CSUM_IP_TSO|CSUM_IP6_TSO)
737 #define CSUM_INNER_TSO (CSUM_INNER_IP_TSO|CSUM_INNER_IP6_TSO)
738 #define CSUM_UDP_IPV6 CSUM_IP6_UDP
739 #define CSUM_TCP_IPV6 CSUM_IP6_TCP
740 #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP
741 #define CSUM_TLS_MASK (CSUM_L5_CALC|CSUM_L5_VALID)
742 #define CSUM_TLS_DECRYPTED CSUM_L5_CALC
745 * mbuf types describing the content of the mbuf (including external storage).
747 #define MT_NOTMBUF 0 /* USED INTERNALLY ONLY! Object is not mbuf */
748 #define MT_DATA 1 /* dynamic (data) allocation */
749 #define MT_HEADER MT_DATA /* packet header, use M_PKTHDR instead */
751 #define MT_VENDOR1 4 /* for vendor-internal use */
752 #define MT_VENDOR2 5 /* for vendor-internal use */
753 #define MT_VENDOR3 6 /* for vendor-internal use */
754 #define MT_VENDOR4 7 /* for vendor-internal use */
756 #define MT_SONAME 8 /* socket name */
758 #define MT_EXP1 9 /* for experimental use */
759 #define MT_EXP2 10 /* for experimental use */
760 #define MT_EXP3 11 /* for experimental use */
761 #define MT_EXP4 12 /* for experimental use */
763 #define MT_CONTROL 14 /* extra-data protocol message */
764 #define MT_EXTCONTROL 15 /* control message with externalized contents */
765 #define MT_OOBDATA 16 /* expedited data */
767 #define MT_NOINIT 255 /* Not a type but a flag to allocate
768 a non-initialized mbuf */
771 * String names of mbuf-related UMA(9) and malloc(9) types. Exposed to
772 * !_KERNEL so that monitoring tools can look up the zones with
775 #define MBUF_MEM_NAME "mbuf"
776 #define MBUF_CLUSTER_MEM_NAME "mbuf_cluster"
777 #define MBUF_PACKET_MEM_NAME "mbuf_packet"
778 #define MBUF_JUMBOP_MEM_NAME "mbuf_jumbo_page"
779 #define MBUF_JUMBO9_MEM_NAME "mbuf_jumbo_9k"
780 #define MBUF_JUMBO16_MEM_NAME "mbuf_jumbo_16k"
781 #define MBUF_TAG_MEM_NAME "mbuf_tag"
782 #define MBUF_EXTREFCNT_MEM_NAME "mbuf_ext_refcnt"
783 #define MBUF_EXTPGS_MEM_NAME "mbuf_extpgs"
786 union if_snd_tag_alloc_params;
788 #define MBUF_CHECKSLEEP(how) do { \
789 if (how == M_WAITOK) \
790 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \
791 "Sleeping in \"%s\"", __func__); \
795 * Network buffer allocation API
797 * The rest of it is defined in kern/kern_mbuf.c
799 extern uma_zone_t zone_mbuf;
800 extern uma_zone_t zone_clust;
801 extern uma_zone_t zone_pack;
802 extern uma_zone_t zone_jumbop;
803 extern uma_zone_t zone_jumbo9;
804 extern uma_zone_t zone_jumbo16;
805 extern uma_zone_t zone_extpgs;
807 void mb_dupcl(struct mbuf *, struct mbuf *);
808 void mb_free_ext(struct mbuf *);
809 void mb_free_extpg(struct mbuf *);
810 void mb_free_mext_pgs(struct mbuf *);
811 struct mbuf *mb_alloc_ext_pgs(int, m_ext_free_t);
812 struct mbuf *mb_alloc_ext_plus_pages(int, int);
813 struct mbuf *mb_mapped_to_unmapped(struct mbuf *, int, int, int,
815 int mb_unmapped_compress(struct mbuf *m);
816 struct mbuf *mb_unmapped_to_ext(struct mbuf *m);
817 void mb_free_notready(struct mbuf *m, int count);
818 void m_adj(struct mbuf *, int);
819 void m_adj_decap(struct mbuf *, int);
820 int m_apply(struct mbuf *, int, int,
821 int (*)(void *, void *, u_int), void *);
822 int m_append(struct mbuf *, int, c_caddr_t);
823 void m_cat(struct mbuf *, struct mbuf *);
824 void m_catpkt(struct mbuf *, struct mbuf *);
825 int m_clget(struct mbuf *m, int how);
826 void *m_cljget(struct mbuf *m, int how, int size);
827 struct mbuf *m_collapse(struct mbuf *, int, int);
828 void m_copyback(struct mbuf *, int, int, c_caddr_t);
829 void m_copydata(const struct mbuf *, int, int, caddr_t);
830 struct mbuf *m_copym(struct mbuf *, int, int, int);
831 struct mbuf *m_copypacket(struct mbuf *, int);
832 void m_copy_pkthdr(struct mbuf *, struct mbuf *);
833 struct mbuf *m_copyup(struct mbuf *, int, int);
834 struct mbuf *m_defrag(struct mbuf *, int);
835 void m_demote_pkthdr(struct mbuf *);
836 void m_demote(struct mbuf *, int, int);
837 struct mbuf *m_devget(char *, int, int, struct ifnet *,
838 void (*)(char *, caddr_t, u_int));
839 void m_dispose_extcontrolm(struct mbuf *m);
840 struct mbuf *m_dup(const struct mbuf *, int);
841 int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
842 void m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
843 void *, void *, int, int);
844 u_int m_fixhdr(struct mbuf *);
845 struct mbuf *m_fragment(struct mbuf *, int, int);
846 void m_freem(struct mbuf *);
847 void m_free_raw(struct mbuf *);
848 struct mbuf *m_get2(int, int, short, int);
849 struct mbuf *m_get3(int, int, short, int);
850 struct mbuf *m_getjcl(int, short, int, int);
851 struct mbuf *m_getm2(struct mbuf *, int, int, short, int);
852 struct mbuf *m_getptr(struct mbuf *, int, int *);
853 u_int m_length(struct mbuf *, struct mbuf **);
854 int m_mbuftouio(struct uio *, const struct mbuf *, int);
855 void m_move_pkthdr(struct mbuf *, struct mbuf *);
856 int m_pkthdr_init(struct mbuf *, int);
857 struct mbuf *m_prepend(struct mbuf *, int, int);
858 void m_print(const struct mbuf *, int);
859 struct mbuf *m_pulldown(struct mbuf *, int, int, int *);
860 struct mbuf *m_pullup(struct mbuf *, int);
861 int m_sanity(struct mbuf *, int);
862 struct mbuf *m_split(struct mbuf *, int, int);
863 struct mbuf *m_uiotombuf(struct uio *, int, int, int, int);
864 int m_unmapped_uiomove(const struct mbuf *, int, struct uio *,
866 struct mbuf *m_unshare(struct mbuf *, int);
867 int m_snd_tag_alloc(struct ifnet *,
868 union if_snd_tag_alloc_params *, struct m_snd_tag **);
869 void m_snd_tag_init(struct m_snd_tag *, struct ifnet *,
870 const struct if_snd_tag_sw *);
871 void m_snd_tag_destroy(struct m_snd_tag *);
872 void m_rcvif_serialize(struct mbuf *);
873 struct ifnet *m_rcvif_restore(struct mbuf *);
897 panic("%s: invalid cluster size %d", __func__, size);
904 * Associated an external reference counted buffer with an mbuf.
907 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
908 m_ext_free_t freef, void *arg1, void *arg2)
911 KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
913 atomic_add_int(ref_cnt, 1);
915 m->m_ext.ext_buf = buf;
916 m->m_ext.ext_cnt = ref_cnt;
917 m->m_data = m->m_ext.ext_buf;
918 m->m_ext.ext_size = size;
919 m->m_ext.ext_free = freef;
920 m->m_ext.ext_arg1 = arg1;
921 m->m_ext.ext_arg2 = arg2;
922 m->m_ext.ext_type = EXT_EXTREF;
923 m->m_ext.ext_flags = 0;
926 static __inline uma_zone_t
945 panic("%s: invalid cluster size %d", __func__, size);
952 * Initialize an mbuf with linear storage.
954 * Inline because the consumer text overhead will be roughly the same to
955 * initialize or call a function with this many parameters and M_PKTHDR
956 * should go away with constant propagation for !MGETHDR.
959 m_init(struct mbuf *m, int how, short type, int flags)
965 m->m_data = m->m_dat;
969 if (flags & M_PKTHDR)
970 error = m_pkthdr_init(m, how);
974 MBUF_PROBE5(m__init, m, how, type, flags, error);
978 static __inline struct mbuf *
979 m_get_raw(int how, short type)
985 args.type = type | MT_NOINIT;
986 m = uma_zalloc_arg(zone_mbuf, &args, how);
987 MBUF_PROBE3(m__get_raw, how, type, m);
991 static __inline struct mbuf *
992 m_get(int how, short type)
999 m = uma_zalloc_arg(zone_mbuf, &args, how);
1000 MBUF_PROBE3(m__get, how, type, m);
1004 static __inline struct mbuf *
1005 m_gethdr_raw(int how, short type)
1008 struct mb_args args;
1010 args.flags = M_PKTHDR;
1011 args.type = type | MT_NOINIT;
1012 m = uma_zalloc_arg(zone_mbuf, &args, how);
1013 MBUF_PROBE3(m__gethdr_raw, how, type, m);
1017 static __inline struct mbuf *
1018 m_gethdr(int how, short type)
1021 struct mb_args args;
1023 args.flags = M_PKTHDR;
1025 m = uma_zalloc_arg(zone_mbuf, &args, how);
1026 MBUF_PROBE3(m__gethdr, how, type, m);
1030 static __inline struct mbuf *
1031 m_getcl(int how, short type, int flags)
1034 struct mb_args args;
1038 m = uma_zalloc_arg(zone_pack, &args, how);
1039 MBUF_PROBE4(m__getcl, how, type, flags, m);
1044 * XXX: m_cljset() is a dangerous API. One must attach only a new,
1045 * unreferenced cluster to an mbuf(9). It is not possible to assert
1046 * that, so care can be taken only by users of the API.
1048 static __inline void
1049 m_cljset(struct mbuf *m, void *cl, int type)
1058 size = MJUMPAGESIZE;
1067 panic("%s: unknown cluster type %d", __func__, type);
1071 m->m_data = m->m_ext.ext_buf = cl;
1072 m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
1073 m->m_ext.ext_size = size;
1074 m->m_ext.ext_type = type;
1075 m->m_ext.ext_flags = EXT_FLAG_EMBREF;
1076 m->m_ext.ext_count = 1;
1077 m->m_flags |= M_EXT;
1078 MBUF_PROBE3(m__cljset, m, cl, type);
1081 static __inline void
1082 m_chtype(struct mbuf *m, short new_type)
1085 m->m_type = new_type;
1088 static __inline void
1089 m_clrprotoflags(struct mbuf *m)
1093 m->m_flags &= ~M_PROTOFLAGS;
1098 static __inline struct mbuf *
1099 m_last(struct mbuf *m)
1108 m_extrefcnt(struct mbuf *m)
1111 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
1113 return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
1118 * mbuf, cluster, and external object allocation macros (for compatibility
1121 #define M_MOVE_PKTHDR(to, from) m_move_pkthdr((to), (from))
1122 #define MGET(m, how, type) ((m) = m_get((how), (type)))
1123 #define MGETHDR(m, how, type) ((m) = m_gethdr((how), (type)))
1124 #define MCLGET(m, how) m_clget((m), (how))
1125 #define MEXTADD(m, buf, size, free, arg1, arg2, flags, type) \
1126 m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2), \
1128 #define m_getm(m, len, how, type) \
1129 m_getm2((m), (len), (how), (type), M_PKTHDR)
1132 * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
1133 * be both the local data payload, or an external buffer area, depending on
1134 * whether M_EXT is set).
1136 #define M_WRITABLE(m) (((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 && \
1137 (!(((m)->m_flags & M_EXT)) || \
1138 (m_extrefcnt(m) == 1)))
1140 /* Check if the supplied mbuf has a packet header, or else panic. */
1141 #define M_ASSERTPKTHDR(m) \
1142 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR, \
1143 ("%s: no mbuf packet header!", __func__))
1145 /* Check if the supplied mbuf has no send tag, or else panic. */
1146 #define M_ASSERT_NO_SND_TAG(m) \
1147 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR && \
1148 ((m)->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0, \
1149 ("%s: receive mbuf has send tag!", __func__))
1151 /* Check if mbuf is multipage. */
1152 #define M_ASSERTEXTPG(m) \
1153 KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG, \
1154 ("%s: m %p is not multipage!", __func__, m))
1157 * Ensure that the supplied mbuf is a valid, non-free mbuf.
1159 * XXX: Broken at the moment. Need some UMA magic to make it work again.
1161 #define M_ASSERTVALID(m) \
1162 KASSERT((((struct mbuf *)m)->m_flags & 0) == 0, \
1163 ("%s: attempted use of a free mbuf!", __func__))
1165 /* Check whether any mbuf in the chain is unmapped. */
1167 #define M_ASSERTMAPPED(m) do { \
1168 for (struct mbuf *__m = (m); __m != NULL; __m = __m->m_next) \
1169 KASSERT((__m->m_flags & M_EXTPG) == 0, \
1170 ("%s: chain %p contains an unmapped mbuf", __func__, (m)));\
1173 #define M_ASSERTMAPPED(m) do {} while (0)
1177 * Return the address of the start of the buffer associated with an mbuf,
1178 * handling external storage, packet-header mbufs, and regular data mbufs.
1180 #define M_START(m) \
1181 (((m)->m_flags & M_EXTPG) ? NULL : \
1182 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf : \
1183 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] : \
1187 * Return the size of the buffer associated with an mbuf, handling external
1188 * storage, packet-header mbufs, and regular data mbufs.
1191 (((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size : \
1192 ((m)->m_flags & M_PKTHDR) ? MHLEN : \
1196 * Set the m_data pointer of a newly allocated mbuf to place an object of the
1197 * specified size at the end of the mbuf, longword aligned.
1199 * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1200 * separate macros, each asserting that it was called at the proper moment.
1201 * This required callers to themselves test the storage type and call the
1202 * right one. Rather than require callers to be aware of those layout
1203 * decisions, we centralize here.
1205 static __inline void
1206 m_align(struct mbuf *m, int len)
1209 const char *msg = "%s: not a virgin mbuf";
1213 KASSERT(m->m_data == M_START(m), (msg, __func__));
1215 adjust = M_SIZE(m) - len;
1216 m->m_data += adjust &~ (sizeof(long)-1);
1219 #define M_ALIGN(m, len) m_align(m, len)
1220 #define MH_ALIGN(m, len) m_align(m, len)
1221 #define MEXT_ALIGN(m, len) m_align(m, len)
1224 * Compute the amount of space available before the current start of data in
1227 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1228 * of checking writability of the mbuf data area rests solely with the caller.
1230 * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1231 * for mbufs with external storage. We now allow mbuf-embedded data to be
1232 * read-only as well.
1234 #define M_LEADINGSPACE(m) \
1235 (M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1238 * So M_TRAILINGROOM() is for when you want to know how much space
1239 * would be there if it was writable. This can be used to
1240 * detect changes in mbufs by knowing the value at one point
1241 * and then being able to compare it later to the current M_TRAILINGROOM().
1242 * The TRAILINGSPACE() macro is not suitable for this since an mbuf
1243 * at one point might not be writable and then later it becomes writable
1244 * even though the space at the back of it has not changed.
1246 #define M_TRAILINGROOM(m) ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len))
1248 * Compute the amount of space available after the end of data in an mbuf.
1250 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1251 * of checking writability of the mbuf data area rests solely with the caller.
1253 * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1254 * for mbufs with external storage. We now allow mbuf-embedded data to be
1255 * read-only as well.
1257 #define M_TRAILINGSPACE(m) (M_WRITABLE(m) ? M_TRAILINGROOM(m) : 0)
1260 * Arrange to prepend space of size plen to mbuf m. If a new mbuf must be
1261 * allocated, how specifies whether to wait. If the allocation fails, the
1262 * original mbuf chain is freed and m is set to NULL.
1264 #define M_PREPEND(m, plen, how) do { \
1265 struct mbuf **_mmp = &(m); \
1266 struct mbuf *_mm = *_mmp; \
1267 int _mplen = (plen); \
1268 int __mhow = (how); \
1270 MBUF_CHECKSLEEP(how); \
1271 if (M_LEADINGSPACE(_mm) >= _mplen) { \
1272 _mm->m_data -= _mplen; \
1273 _mm->m_len += _mplen; \
1275 _mm = m_prepend(_mm, _mplen, __mhow); \
1276 if (_mm != NULL && _mm->m_flags & M_PKTHDR) \
1277 _mm->m_pkthdr.len += _mplen; \
1282 * Change mbuf to new type. This is a relatively expensive operation and
1283 * should be avoided.
1285 #define MCHTYPE(m, t) m_chtype((m), (t))
1287 /* Return the rcvif of a packet header. */
1288 static __inline struct ifnet *
1289 m_rcvif(struct mbuf *m)
1293 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1295 return (m->m_pkthdr.rcvif);
1298 /* Length to m_copy to copy all. */
1299 #define M_COPYALL 1000000000
1301 extern u_int max_linkhdr; /* Largest link-level header */
1302 extern u_int max_hdr; /* Largest link + protocol header */
1303 extern u_int max_protohdr; /* Largest protocol header */
1304 void max_linkhdr_grow(u_int);
1305 void max_protohdr_grow(u_int);
1307 extern int nmbclusters; /* Maximum number of clusters */
1308 extern bool mb_use_ext_pgs; /* Use ext_pgs for sendfile */
1311 * Network packets may have annotations attached by affixing a list of
1312 * "packet tags" to the pkthdr structure. Packet tags are dynamically
1313 * allocated semi-opaque data structures that have a fixed header
1314 * (struct m_tag) that specifies the size of the memory block and a
1315 * <cookie,type> pair that identifies it. The cookie is a 32-bit unique
1316 * unsigned value used to identify a module or ABI. By convention this value
1317 * is chosen as the date+time that the module is created, expressed as the
1318 * number of seconds since the epoch (e.g., using date -u +'%s'). The type
1319 * value is an ABI/module-specific value that identifies a particular
1320 * annotation and is private to the module. For compatibility with systems
1321 * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1322 * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1323 * compatibility shim functions and several tag types are defined below.
1324 * Users that do not require compatibility should use a private cookie value
1325 * so that packet tag-related definitions can be maintained privately.
1327 * Note that the packet tag returned by m_tag_alloc has the default memory
1328 * alignment implemented by malloc. To reference private data one can use a
1331 * struct m_tag *mtag = m_tag_alloc(...);
1332 * struct foo *p = (struct foo *)(mtag+1);
1334 * if the alignment of struct m_tag is sufficient for referencing members of
1335 * struct foo. Otherwise it is necessary to embed struct m_tag within the
1336 * private data structure to insure proper alignment; e.g.,
1342 * struct foo *p = (struct foo *) m_tag_alloc(...);
1343 * struct m_tag *mtag = &p->tag;
1347 * Persistent tags stay with an mbuf until the mbuf is reclaimed. Otherwise
1348 * tags are expected to ``vanish'' when they pass through a network
1349 * interface. For most interfaces this happens normally as the tags are
1350 * reclaimed when the mbuf is free'd. However in some special cases
1351 * reclaiming must be done manually. An example is packets that pass through
1352 * the loopback interface. Also, one must be careful to do this when
1353 * ``turning around'' packets (e.g., icmp_reflect).
1355 * To mark a tag persistent bit-or this flag in when defining the tag id.
1356 * The tag will then be treated as described above.
1358 #define MTAG_PERSISTENT 0x800
1360 #define PACKET_TAG_NONE 0 /* Nadda */
1362 /* Packet tags for use with PACKET_ABI_COMPAT. */
1363 #define PACKET_TAG_IPSEC_IN_DONE 1 /* IPsec applied, in */
1364 #define PACKET_TAG_IPSEC_OUT_DONE 2 /* IPsec applied, out */
1365 #define PACKET_TAG_IPSEC_IN_CRYPTO_DONE 3 /* NIC IPsec crypto done */
1366 #define PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED 4 /* NIC IPsec crypto req'ed */
1367 #define PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO 5 /* NIC notifies IPsec */
1368 #define PACKET_TAG_IPSEC_PENDING_TDB 6 /* Reminder to do IPsec */
1369 #define PACKET_TAG_BRIDGE 7 /* Bridge processing done */
1370 #define PACKET_TAG_GIF 8 /* GIF processing done */
1371 #define PACKET_TAG_GRE 9 /* GRE processing done */
1372 #define PACKET_TAG_IN_PACKET_CHECKSUM 10 /* NIC checksumming done */
1373 #define PACKET_TAG_ENCAP 11 /* Encap. processing */
1374 #define PACKET_TAG_IPSEC_SOCKET 12 /* IPSEC socket ref */
1375 #define PACKET_TAG_IPSEC_HISTORY 13 /* IPSEC history */
1376 #define PACKET_TAG_IPV6_INPUT 14 /* IPV6 input processing */
1377 #define PACKET_TAG_DUMMYNET 15 /* dummynet info */
1378 #define PACKET_TAG_DIVERT 17 /* divert info */
1379 #define PACKET_TAG_IPFORWARD 18 /* ipforward info */
1380 #define PACKET_TAG_MACLABEL (19 | MTAG_PERSISTENT) /* MAC label */
1381 #define PACKET_TAG_PF 21 /* PF/ALTQ information */
1382 /* was PACKET_TAG_RTSOCKFAM 25 rtsock sa family */
1383 #define PACKET_TAG_IPOPTIONS 27 /* Saved IP options */
1384 #define PACKET_TAG_CARP 28 /* CARP info */
1385 #define PACKET_TAG_IPSEC_NAT_T_PORTS 29 /* two uint16_t */
1386 #define PACKET_TAG_ND_OUTGOING 30 /* ND outgoing */
1387 #define PACKET_TAG_PF_REASSEMBLED 31
1389 /* Specific cookies and tags. */
1391 /* Packet tag routines. */
1392 struct m_tag *m_tag_alloc(uint32_t, uint16_t, int, int);
1393 void m_tag_delete(struct mbuf *, struct m_tag *);
1394 void m_tag_delete_chain(struct mbuf *, struct m_tag *);
1395 void m_tag_free_default(struct m_tag *);
1396 struct m_tag *m_tag_locate(struct mbuf *, uint32_t, uint16_t,
1398 struct m_tag *m_tag_copy(struct m_tag *, int);
1399 int m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1400 void m_tag_delete_nonpersistent(struct mbuf *);
1403 * Initialize the list of tags associated with an mbuf.
1405 static __inline void
1406 m_tag_init(struct mbuf *m)
1409 SLIST_INIT(&m->m_pkthdr.tags);
1413 * Set up the contents of a tag. Note that this does not fill in the free
1414 * method; the caller is expected to do that.
1416 * XXX probably should be called m_tag_init, but that was already taken.
1418 static __inline void
1419 m_tag_setup(struct m_tag *t, uint32_t cookie, uint16_t type, int len)
1424 t->m_tag_cookie = cookie;
1428 * Reclaim resources associated with a tag.
1430 static __inline void
1431 m_tag_free(struct m_tag *t)
1434 (*t->m_tag_free)(t);
1438 * Return the first tag associated with an mbuf.
1440 static __inline struct m_tag *
1441 m_tag_first(struct mbuf *m)
1444 return (SLIST_FIRST(&m->m_pkthdr.tags));
1448 * Return the next tag in the list of tags associated with an mbuf.
1450 static __inline struct m_tag *
1451 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1454 return (SLIST_NEXT(t, m_tag_link));
1458 * Prepend a tag to the list of tags associated with an mbuf.
1460 static __inline void
1461 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1464 SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1468 * Unlink a tag from the list of tags associated with an mbuf.
1470 static __inline void
1471 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1474 SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1477 /* These are for OpenBSD compatibility. */
1478 #define MTAG_ABI_COMPAT 0 /* compatibility ABI */
1480 static __inline struct m_tag *
1481 m_tag_get(uint16_t type, int length, int wait)
1483 return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1486 static __inline struct m_tag *
1487 m_tag_find(struct mbuf *m, uint16_t type, struct m_tag *start)
1489 return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1490 m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1493 static inline struct m_snd_tag *
1494 m_snd_tag_ref(struct m_snd_tag *mst)
1497 refcount_acquire(&mst->refcount);
1502 m_snd_tag_rele(struct m_snd_tag *mst)
1505 if (refcount_release(&mst->refcount))
1506 m_snd_tag_destroy(mst);
1509 static __inline struct mbuf *
1510 m_free(struct mbuf *m)
1512 struct mbuf *n = m->m_next;
1514 MBUF_PROBE1(m__free, m);
1515 if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1516 m_tag_delete_chain(m, NULL);
1517 if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1518 m_snd_tag_rele(m->m_pkthdr.snd_tag);
1519 if (m->m_flags & M_EXTPG)
1521 else if (m->m_flags & M_EXT)
1523 else if ((m->m_flags & M_NOFREE) == 0)
1524 uma_zfree(zone_mbuf, m);
1529 rt_m_getfib(struct mbuf *m)
1531 KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1532 return (m->m_pkthdr.fibnum);
1535 #define M_GETFIB(_m) rt_m_getfib(_m)
1537 #define M_SETFIB(_m, _fib) do { \
1538 KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf.")); \
1539 ((_m)->m_pkthdr.fibnum) = (_fib); \
1542 /* flags passed as first argument for "m_xxx_tcpip_hash()" */
1543 #define MBUF_HASHFLAG_L2 (1 << 2)
1544 #define MBUF_HASHFLAG_L3 (1 << 3)
1545 #define MBUF_HASHFLAG_L4 (1 << 4)
1547 /* mbuf hashing helper routines */
1548 uint32_t m_ether_tcpip_hash_init(void);
1549 uint32_t m_ether_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1550 uint32_t m_infiniband_tcpip_hash_init(void);
1551 uint32_t m_infiniband_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1553 #ifdef MBUF_PROFILING
1554 void m_profile(struct mbuf *m);
1555 #define M_PROFILE(m) m_profile(m)
1557 #define M_PROFILE(m)
1561 STAILQ_HEAD(, mbuf) mq_head;
1567 mbufq_init(struct mbufq *mq, int maxlen)
1570 STAILQ_INIT(&mq->mq_head);
1571 mq->mq_maxlen = maxlen;
1575 static inline struct mbuf *
1576 mbufq_flush(struct mbufq *mq)
1580 m = STAILQ_FIRST(&mq->mq_head);
1581 STAILQ_INIT(&mq->mq_head);
1587 mbufq_drain(struct mbufq *mq)
1591 n = mbufq_flush(mq);
1592 while ((m = n) != NULL) {
1593 n = STAILQ_NEXT(m, m_stailqpkt);
1598 static inline struct mbuf *
1599 mbufq_first(const struct mbufq *mq)
1602 return (STAILQ_FIRST(&mq->mq_head));
1605 static inline struct mbuf *
1606 mbufq_last(const struct mbufq *mq)
1609 return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1613 mbufq_full(const struct mbufq *mq)
1616 return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1620 mbufq_len(const struct mbufq *mq)
1623 return (mq->mq_len);
1627 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1632 STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1637 static inline struct mbuf *
1638 mbufq_dequeue(struct mbufq *mq)
1642 m = STAILQ_FIRST(&mq->mq_head);
1644 STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1645 m->m_nextpkt = NULL;
1652 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1655 STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1660 * Note: this doesn't enforce the maximum list size for dst.
1663 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1666 mq_dst->mq_len += mq_src->mq_len;
1667 STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1671 #ifdef _SYS_TIMESPEC_H_
1673 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1676 KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1677 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1678 ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1679 ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1680 ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1685 mbuf_tstmp2timeval(struct mbuf *m, struct timeval *tv)
1688 KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1689 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1690 ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1691 tv->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1692 tv->tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000) / 1000;
1696 /* Invoked from the debugnet client code. */
1697 void debugnet_mbuf_drain(void);
1698 void debugnet_mbuf_start(void);
1699 void debugnet_mbuf_finish(void);
1700 void debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1704 mbuf_has_tls_session(struct mbuf *m)
1707 if (m->m_flags & M_EXTPG) {
1708 if (m->m_epg_tls != NULL) {
1715 #endif /* _KERNEL */
1716 #endif /* !_SYS_MBUF_H_ */