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
39 /* XXX: These includes suck. Sorry! */
40 #include <sys/queue.h>
42 #include <sys/systm.h>
43 #include <sys/refcount.h>
53 #define MBUF_PROBE1(probe, arg0) \
54 SDT_PROBE1(sdt, , , probe, arg0)
55 #define MBUF_PROBE2(probe, arg0, arg1) \
56 SDT_PROBE2(sdt, , , probe, arg0, arg1)
57 #define MBUF_PROBE3(probe, arg0, arg1, arg2) \
58 SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
59 #define MBUF_PROBE4(probe, arg0, arg1, arg2, arg3) \
60 SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
61 #define MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4) \
62 SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
64 SDT_PROBE_DECLARE(sdt, , , m__init);
65 SDT_PROBE_DECLARE(sdt, , , m__gethdr_raw);
66 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
67 SDT_PROBE_DECLARE(sdt, , , m__get_raw);
68 SDT_PROBE_DECLARE(sdt, , , m__get);
69 SDT_PROBE_DECLARE(sdt, , , m__getcl);
70 SDT_PROBE_DECLARE(sdt, , , m__getjcl);
71 SDT_PROBE_DECLARE(sdt, , , m__clget);
72 SDT_PROBE_DECLARE(sdt, , , m__cljget);
73 SDT_PROBE_DECLARE(sdt, , , m__cljset);
74 SDT_PROBE_DECLARE(sdt, , , m__free);
75 SDT_PROBE_DECLARE(sdt, , , m__freem);
80 * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
81 * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
82 * sys/param.h), which has no additional overhead and is used instead of the
83 * internal data area; this is done when at least MINCLSIZE of data must be
84 * stored. Additionally, it is possible to allocate a separate buffer
85 * externally and attach it to the mbuf in a way similar to that of mbuf
88 * NB: These calculation do not take actual compiler-induced alignment and
89 * padding inside the complete struct mbuf into account. Appropriate
90 * attention is required when changing members of struct mbuf.
92 * MLEN is data length in a normal mbuf.
93 * MHLEN is data length in an mbuf with pktheader.
94 * MINCLSIZE is a smallest amount of data that should be put into cluster.
96 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
100 #define MHSIZE offsetof(struct mbuf, m_dat)
101 #define MPKTHSIZE offsetof(struct mbuf, m_pktdat)
102 #define MLEN ((int)(MSIZE - MHSIZE))
103 #define MHLEN ((int)(MSIZE - MPKTHSIZE))
104 #define MINCLSIZE (MHLEN + 1)
109 * Macro for type conversion: convert mbuf pointer to data pointer of correct
112 * mtod(m, t) -- Convert mbuf pointer to data pointer of correct type.
113 * mtodo(m, o) -- Same as above but with offset 'o' into data.
115 #define mtod(m, t) ((t)((m)->m_data))
116 #define mtodo(m, o) ((void *)(((m)->m_data) + (o)))
119 * Argument structure passed to UMA routines during mbuf and packet
123 int flags; /* Flags for mbuf being allocated */
124 short type; /* Type of mbuf being allocated */
129 * Packet tag structure (see below for details).
132 SLIST_ENTRY(m_tag) m_tag_link; /* List of packet tags */
133 u_int16_t m_tag_id; /* Tag ID */
134 u_int16_t m_tag_len; /* Length of data */
135 u_int32_t m_tag_cookie; /* ABI/Module ID */
136 void (*m_tag_free)(struct m_tag *);
140 * Static network interface owned tag.
141 * Allocated through ifp->if_snd_tag_alloc().
143 struct if_snd_tag_sw;
146 struct ifnet *ifp; /* network interface tag belongs to */
147 const struct if_snd_tag_sw *sw;
148 volatile u_int refcount;
152 * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
155 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
160 struct m_snd_tag *snd_tag; /* send tag, if any */
161 struct ifnet *rcvif; /* rcv interface */
163 uint16_t rcvidx; /* rcv interface index ... */
164 uint16_t rcvgen; /* ... and generation count */
168 struct ifnet *leaf_rcvif; /* leaf rcv interface */
170 uint16_t leaf_rcvidx; /* leaf rcv interface index ... */
171 uint16_t leaf_rcvgen; /* ... and generation count */
174 SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
175 int32_t len; /* total packet length */
177 /* Layer crossing persistent information. */
178 uint32_t flowid; /* packet's 4-tuple system */
179 uint32_t csum_flags; /* checksum and offload features */
180 uint16_t fibnum; /* this packet should use this fib */
181 uint8_t numa_domain; /* NUMA domain of recvd pkt */
182 uint8_t rsstype; /* hash type */
183 #if !defined(__LP64__)
184 uint32_t pad; /* pad for 64bit alignment */
187 uint64_t rcv_tstmp; /* timestamp in ns */
189 uint8_t l2hlen; /* layer 2 hdr len */
190 uint8_t l3hlen; /* layer 3 hdr len */
191 uint8_t l4hlen; /* layer 4 hdr len */
192 uint8_t l5hlen; /* layer 5 hdr len */
193 uint8_t inner_l2hlen;
194 uint8_t inner_l3hlen;
195 uint8_t inner_l4hlen;
196 uint8_t inner_l5hlen;
202 uint32_t thirtytwo[2];
203 uint64_t sixtyfour[1];
204 uintptr_t unintptr[1];
208 /* Layer specific non-persistent local storage for reassembly, etc. */
213 uint32_t thirtytwo[2];
214 uint64_t sixtyfour[1];
215 uintptr_t unintptr[1];
218 /* Upon allocation: total packet memory consumption. */
222 #define ether_vtag PH_per.sixteen[0]
223 #define tcp_tun_port PH_per.sixteen[0] /* outbound */
224 #define vt_nrecs PH_per.sixteen[0] /* mld and v6-ND */
225 #define tso_segsz PH_per.sixteen[1] /* inbound after LRO */
226 #define lro_nsegs tso_segsz /* inbound after LRO */
227 #define csum_data PH_per.thirtytwo[1] /* inbound from hardware up */
228 #define lro_tcp_d_len PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
229 #define lro_tcp_d_csum PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
230 #define lro_tcp_h_off PH_loc.sixteen[2] /* inbound during LRO (no reassembly) */
231 #define lro_etype PH_loc.sixteen[3] /* inbound during LRO (no reassembly) */
232 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
235 * TLS records for TLS 1.0-1.2 can have the following header lengths:
236 * - 5 (AES-CBC with implicit IV)
237 * - 21 (AES-CBC with explicit IV)
238 * - 13 (AES-GCM with 8 byte explicit IV)
240 #define MBUF_PEXT_HDR_LEN 23
243 * TLS records for TLS 1.0-1.2 can have the following maximum trailer
246 * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
247 * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
248 * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
250 #define MBUF_PEXT_TRAIL_LEN 64
252 #if defined(__LP64__)
253 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
255 #define MBUF_PEXT_MAX_PGS (64 / sizeof(vm_paddr_t))
258 #define MBUF_PEXT_MAX_BYTES \
259 (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
265 * Description of external storage mapped into mbuf; valid only if M_EXT is
269 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
272 typedef void m_ext_free_t(struct mbuf *);
276 * If EXT_FLAG_EMBREF is set, then we use refcount in the
277 * mbuf, the 'ext_count' member. Otherwise, we have a
278 * shadow copy and we use pointer 'ext_cnt'. The original
279 * mbuf is responsible to carry the pointer to free routine
280 * and its arguments. They aren't copied into shadows in
281 * mb_dupcl() to avoid dereferencing next cachelines.
283 volatile u_int ext_count;
284 volatile u_int *ext_cnt;
286 uint32_t ext_size; /* size of buffer, for ext_free */
287 uint32_t ext_type:8, /* type of external storage */
288 ext_flags:24; /* external storage mbuf flags */
292 * Regular M_EXT mbuf:
293 * o ext_buf always points to the external buffer.
294 * o ext_free (below) and two optional arguments
295 * ext_arg1 and ext_arg2 store the free context for
296 * the external storage. They are set only in the
297 * refcount carrying mbuf, the one with
298 * EXT_FLAG_EMBREF flag, with exclusion for
299 * EXT_EXTREF type, where the free context is copied
300 * into all mbufs that use same external storage.
302 char *ext_buf; /* start of buffer */
303 #define m_ext_copylen offsetof(struct m_ext, ext_arg2)
308 * Multi-page M_EXTPG mbuf:
309 * o extpg_pa - page vector.
310 * o extpg_trail and extpg_hdr - TLS trailer and
312 * Uses ext_free and may also use ext_arg1.
314 vm_paddr_t extpg_pa[MBUF_PEXT_MAX_PGS];
315 char extpg_trail[MBUF_PEXT_TRAIL_LEN];
316 char extpg_hdr[MBUF_PEXT_HDR_LEN];
317 /* Pretend these 3 fields are part of mbuf itself. */
318 #define m_epg_pa m_ext.extpg_pa
319 #define m_epg_trail m_ext.extpg_trail
320 #define m_epg_hdr m_ext.extpg_hdr
321 #define m_epg_ext_copylen offsetof(struct m_ext, ext_free)
325 * Free method and optional argument pointer, both
326 * used by M_EXT and M_EXTPG.
328 m_ext_free_t *ext_free;
333 * The core of the mbuf object along with some shortcut defines for practical
338 * Header present at the beginning of every mbuf.
341 * Compile-time assertions in uipc_mbuf.c test these values to ensure
342 * that they are correct.
344 union { /* next buffer in chain */
346 SLIST_ENTRY(mbuf) m_slist;
347 STAILQ_ENTRY(mbuf) m_stailq;
349 union { /* next chain in queue/record */
350 struct mbuf *m_nextpkt;
351 SLIST_ENTRY(mbuf) m_slistpkt;
352 STAILQ_ENTRY(mbuf) m_stailqpkt;
354 caddr_t m_data; /* location of data */
355 int32_t m_len; /* amount of data in this mbuf */
356 uint32_t m_type:8, /* type of data in this mbuf */
357 m_flags:24; /* flags; see below */
358 #if !defined(__LP64__)
359 uint32_t m_pad; /* pad for 64bit alignment */
363 * A set of optional headers (packet header, external storage header)
364 * and internal data storage. Historically, these arrays were sized
365 * to MHLEN (space left after a packet header) and MLEN (space left
366 * after only a regular mbuf header); they are now variable size in
367 * order to support future work on variable-size mbufs.
373 struct pkthdr m_pkthdr;
376 * Multi-page M_EXTPG mbuf has its meta data
377 * split between the below anonymous structure
378 * and m_ext. It carries vector of pages,
379 * optional header and trailer char vectors
380 * and pointers to socket/TLS data.
382 #define m_epg_startcopy m_epg_npgs
383 #define m_epg_endcopy m_epg_stailq
385 /* Overall count of pages and count of
386 * pages with I/O pending. */
389 /* TLS header and trailer lengths.
390 * The data itself resides in m_ext. */
391 uint8_t m_epg_hdrlen;
392 uint8_t m_epg_trllen;
393 /* Offset into 1st page and length of
394 * data in the last page. */
395 uint16_t m_epg_1st_off;
396 uint16_t m_epg_last_len;
398 #define EPG_FLAG_ANON 0x1 /* Data can be encrypted in place. */
399 #define EPG_FLAG_2FREE 0x2 /* Scheduled for free. */
400 uint8_t m_epg_record_type;
403 struct ktls_session *m_epg_tls;
404 struct socket *m_epg_so;
405 uint64_t m_epg_seqno;
406 STAILQ_ENTRY(mbuf) m_epg_stailq;
410 /* M_EXT or M_EXTPG set. */
412 /* M_PKTHDR set, neither M_EXT nor M_EXTPG. */
416 char m_dat[0]; /* !M_PKTHDR, !M_EXT */
422 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff)
425 KASSERT(pgoff == 0 || pidx == 0,
426 ("page %d with non-zero offset %d in %p", pidx, pgoff, m));
428 if (pidx == m->m_epg_npgs - 1) {
429 return (m->m_epg_last_len);
431 return (PAGE_SIZE - pgoff);
436 #define MCHECK(ex, msg) KASSERT((ex), \
437 ("Multi page mbuf %p with " #msg " at %s:%d", \
438 m, __FILE__, __LINE__))
440 * NB: This expects a non-empty buffer (npgs > 0 and
443 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do { \
444 MCHECK(m->m_epg_npgs > 0, "no valid pages"); \
445 MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa), \
447 MCHECK(m->m_epg_nrdy <= m->m_epg_npgs, \
448 "too many ready pages"); \
449 MCHECK(m->m_epg_1st_off < PAGE_SIZE, \
450 "too large page offset"); \
451 MCHECK(m->m_epg_last_len > 0, "zero last page length"); \
452 MCHECK(m->m_epg_last_len <= PAGE_SIZE, \
453 "too large last page length"); \
454 if (m->m_epg_npgs == 1) \
455 MCHECK(m->m_epg_1st_off + \
456 m->m_epg_last_len <= PAGE_SIZE, \
457 "single page too large"); \
458 MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr), \
459 "too large header length"); \
460 MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail), \
461 "too large header length"); \
464 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do {} while (0)
469 * mbuf flags of global significance and layer crossing.
470 * Those of only protocol/layer specific significance are to be mapped
471 * to M_PROTO[1-11] and cleared at layer handoff boundaries.
472 * NB: Limited to the lower 24 bits.
474 #define M_EXT 0x00000001 /* has associated external storage */
475 #define M_PKTHDR 0x00000002 /* start of record */
476 #define M_EOR 0x00000004 /* end of record */
477 #define M_RDONLY 0x00000008 /* associated data is marked read-only */
478 #define M_BCAST 0x00000010 /* send/received as link-level broadcast */
479 #define M_MCAST 0x00000020 /* send/received as link-level multicast */
480 #define M_PROMISC 0x00000040 /* packet was not for us */
481 #define M_VLANTAG 0x00000080 /* ether_vtag is valid */
482 #define M_EXTPG 0x00000100 /* has array of unmapped pages and TLS */
483 #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */
484 #define M_TSTMP 0x00000400 /* rcv_tstmp field is valid */
485 #define M_TSTMP_HPREC 0x00000800 /* rcv_tstmp is high-prec, typically
486 hw-stamped on port (useful for IEEE 1588
488 #define M_TSTMP_LRO 0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
490 #define M_PROTO1 0x00002000 /* protocol-specific */
491 #define M_PROTO2 0x00004000 /* protocol-specific */
492 #define M_PROTO3 0x00008000 /* protocol-specific */
493 #define M_PROTO4 0x00010000 /* protocol-specific */
494 #define M_PROTO5 0x00020000 /* protocol-specific */
495 #define M_PROTO6 0x00040000 /* protocol-specific */
496 #define M_PROTO7 0x00080000 /* protocol-specific */
497 #define M_PROTO8 0x00100000 /* protocol-specific */
498 #define M_PROTO9 0x00200000 /* protocol-specific */
499 #define M_PROTO10 0x00400000 /* protocol-specific */
500 #define M_PROTO11 0x00800000 /* protocol-specific */
503 * Flags to purge when crossing layers.
505 #define M_PROTOFLAGS \
506 (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
507 M_PROTO9|M_PROTO10|M_PROTO11)
510 * Flags preserved when copying m_pkthdr.
512 #define M_COPYFLAGS \
513 (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
514 M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
517 * Flags preserved during demote.
519 #define M_DEMOTEFLAGS \
520 (M_EXT | M_RDONLY | M_NOFREE | M_EXTPG)
523 * Mbuf flag description for use with printf(9) %b identifier.
525 #define M_FLAG_BITS \
526 "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
527 "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
528 #define M_FLAG_PROTOBITS \
529 "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
530 "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
531 "\27M_PROTO10\28M_PROTO11"
532 #define M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
535 * Network interface cards are able to hash protocol fields (such as IPv4
536 * addresses and TCP port numbers) classify packets into flows. These flows
537 * can then be used to maintain ordering while delivering packets to the OS
538 * via parallel input queues, as well as to provide a stateless affinity
539 * model. NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
540 * m_flag fields to indicate how the hash should be interpreted by the
543 * Most NICs support RSS, which provides ordering and explicit affinity, and
544 * use the hash m_flag bits to indicate what header fields were covered by
545 * the hash. M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
546 * RSS cards or configurations that provide an opaque flow identifier, allowing
547 * for ordering and distribution without explicit affinity. Additionally,
548 * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
551 * The meaning of the IPV6_EX suffix:
552 * "o Home address from the home address option in the IPv6 destination
553 * options header. If the extension header is not present, use the Source
555 * o IPv6 address that is contained in the Routing-Header-Type-2 from the
556 * associated extension header. If the extension header is not present,
557 * use the Destination IPv6 Address."
559 * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
561 #define M_HASHTYPE_HASHPROP 0x80 /* has hash properties */
562 #define M_HASHTYPE_INNER 0x40 /* calculated from inner headers */
563 #define M_HASHTYPE_HASH(t) (M_HASHTYPE_HASHPROP | (t))
564 /* Microsoft RSS standard hash types */
565 #define M_HASHTYPE_NONE 0
566 #define M_HASHTYPE_RSS_IPV4 M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
567 #define M_HASHTYPE_RSS_TCP_IPV4 M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
568 #define M_HASHTYPE_RSS_IPV6 M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
569 #define M_HASHTYPE_RSS_TCP_IPV6 M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
570 #define M_HASHTYPE_RSS_IPV6_EX M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
572 #define M_HASHTYPE_RSS_TCP_IPV6_EX M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
574 #define M_HASHTYPE_RSS_UDP_IPV4 M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
575 #define M_HASHTYPE_RSS_UDP_IPV6 M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
576 #define M_HASHTYPE_RSS_UDP_IPV6_EX M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
579 #define M_HASHTYPE_OPAQUE 0x3f /* ordering, not affinity */
580 #define M_HASHTYPE_OPAQUE_HASH M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
581 /* ordering+hash, not affinity*/
583 #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0)
584 #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype & ~M_HASHTYPE_INNER)
585 #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v))
586 #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v))
587 #define M_HASHTYPE_ISHASH(m) \
588 (((m)->m_pkthdr.rsstype & M_HASHTYPE_HASHPROP) != 0)
589 #define M_HASHTYPE_SETINNER(m) do { \
590 (m)->m_pkthdr.rsstype |= M_HASHTYPE_INNER; \
594 * External mbuf storage buffer types.
596 #define EXT_CLUSTER 1 /* mbuf cluster */
597 #define EXT_SFBUF 2 /* sendfile(2)'s sf_buf */
598 #define EXT_JUMBOP 3 /* jumbo cluster page sized */
599 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */
600 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */
601 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */
602 #define EXT_MBUF 7 /* external mbuf reference */
603 #define EXT_RXRING 8 /* data in NIC receive ring */
605 #define EXT_VENDOR1 224 /* for vendor-internal use */
606 #define EXT_VENDOR2 225 /* for vendor-internal use */
607 #define EXT_VENDOR3 226 /* for vendor-internal use */
608 #define EXT_VENDOR4 227 /* for vendor-internal use */
610 #define EXT_EXP1 244 /* for experimental use */
611 #define EXT_EXP2 245 /* for experimental use */
612 #define EXT_EXP3 246 /* for experimental use */
613 #define EXT_EXP4 247 /* for experimental use */
615 #define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */
616 #define EXT_MOD_TYPE 253 /* custom module's ext_buf type */
617 #define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */
618 #define EXT_EXTREF 255 /* has externally maintained ext_cnt ptr */
621 * Flags for external mbuf buffer types.
622 * NB: limited to the lower 24 bits.
624 #define EXT_FLAG_EMBREF 0x000001 /* embedded ext_count */
625 #define EXT_FLAG_EXTREF 0x000002 /* external ext_cnt, notyet */
627 #define EXT_FLAG_NOFREE 0x000010 /* don't free mbuf to pool, notyet */
629 #define EXT_FLAG_VENDOR1 0x010000 /* These flags are vendor */
630 #define EXT_FLAG_VENDOR2 0x020000 /* or submodule specific, */
631 #define EXT_FLAG_VENDOR3 0x040000 /* not used by mbuf code. */
632 #define EXT_FLAG_VENDOR4 0x080000 /* Set/read by submodule. */
634 #define EXT_FLAG_EXP1 0x100000 /* for experimental use */
635 #define EXT_FLAG_EXP2 0x200000 /* for experimental use */
636 #define EXT_FLAG_EXP3 0x400000 /* for experimental use */
637 #define EXT_FLAG_EXP4 0x800000 /* for experimental use */
640 * EXT flag description for use with printf(9) %b identifier.
642 #define EXT_FLAG_BITS \
643 "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
644 "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
645 "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
649 * Flags indicating checksum, segmentation and other offload work to be
650 * done, or already done, by hardware or lower layers. It is split into
651 * separate inbound and outbound flags.
653 * Outbound flags that are set by upper protocol layers requesting lower
654 * layers, or ideally the hardware, to perform these offloading tasks.
655 * For outbound packets this field and its flags can be directly tested
656 * against ifnet if_hwassist. Note that the outbound and the inbound flags do
657 * not collide right now but they could be allowed to (as long as the flags are
658 * scrubbed appropriately when the direction of an mbuf changes). CSUM_BITS
659 * would also have to split into CSUM_BITS_TX and CSUM_BITS_RX.
661 * CSUM_INNER_<x> is the same as CSUM_<x> but it applies to the inner frame.
662 * The CSUM_ENCAP_<x> bits identify the outer encapsulation.
664 #define CSUM_IP 0x00000001 /* IP header checksum offload */
665 #define CSUM_IP_UDP 0x00000002 /* UDP checksum offload */
666 #define CSUM_IP_TCP 0x00000004 /* TCP checksum offload */
667 #define CSUM_IP_SCTP 0x00000008 /* SCTP checksum offload */
668 #define CSUM_IP_TSO 0x00000010 /* TCP segmentation offload */
669 #define CSUM_IP_ISCSI 0x00000020 /* iSCSI checksum offload */
671 #define CSUM_INNER_IP6_UDP 0x00000040
672 #define CSUM_INNER_IP6_TCP 0x00000080
673 #define CSUM_INNER_IP6_TSO 0x00000100
674 #define CSUM_IP6_UDP 0x00000200 /* UDP checksum offload */
675 #define CSUM_IP6_TCP 0x00000400 /* TCP checksum offload */
676 #define CSUM_IP6_SCTP 0x00000800 /* SCTP checksum offload */
677 #define CSUM_IP6_TSO 0x00001000 /* TCP segmentation offload */
678 #define CSUM_IP6_ISCSI 0x00002000 /* iSCSI checksum offload */
680 #define CSUM_INNER_IP 0x00004000
681 #define CSUM_INNER_IP_UDP 0x00008000
682 #define CSUM_INNER_IP_TCP 0x00010000
683 #define CSUM_INNER_IP_TSO 0x00020000
685 #define CSUM_ENCAP_VXLAN 0x00040000 /* VXLAN outer encapsulation */
686 #define CSUM_ENCAP_RSVD1 0x00080000
688 /* Inbound checksum support where the checksum was verified by hardware. */
689 #define CSUM_INNER_L3_CALC 0x00100000
690 #define CSUM_INNER_L3_VALID 0x00200000
691 #define CSUM_INNER_L4_CALC 0x00400000
692 #define CSUM_INNER_L4_VALID 0x00800000
693 #define CSUM_L3_CALC 0x01000000 /* calculated layer 3 csum */
694 #define CSUM_L3_VALID 0x02000000 /* checksum is correct */
695 #define CSUM_L4_CALC 0x04000000 /* calculated layer 4 csum */
696 #define CSUM_L4_VALID 0x08000000 /* checksum is correct */
697 #define CSUM_L5_CALC 0x10000000 /* calculated layer 5 csum */
698 #define CSUM_L5_VALID 0x20000000 /* checksum is correct */
699 #define CSUM_COALESCED 0x40000000 /* contains merged segments */
701 #define CSUM_SND_TAG 0x80000000 /* Packet header has send tag */
703 #define CSUM_FLAGS_TX (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP | CSUM_IP_SCTP | \
704 CSUM_IP_TSO | CSUM_IP_ISCSI | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | \
705 CSUM_INNER_IP6_TSO | CSUM_IP6_UDP | CSUM_IP6_TCP | CSUM_IP6_SCTP | \
706 CSUM_IP6_TSO | CSUM_IP6_ISCSI | CSUM_INNER_IP | CSUM_INNER_IP_UDP | \
707 CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN | \
708 CSUM_ENCAP_RSVD1 | CSUM_SND_TAG)
710 #define CSUM_FLAGS_RX (CSUM_INNER_L3_CALC | CSUM_INNER_L3_VALID | \
711 CSUM_INNER_L4_CALC | CSUM_INNER_L4_VALID | CSUM_L3_CALC | CSUM_L3_VALID | \
712 CSUM_L4_CALC | CSUM_L4_VALID | CSUM_L5_CALC | CSUM_L5_VALID | \
716 * CSUM flag description for use with printf(9) %b identifier.
719 "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
720 "\6CSUM_IP_ISCSI\7CSUM_INNER_IP6_UDP\10CSUM_INNER_IP6_TCP" \
721 "\11CSUM_INNER_IP6_TSO\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP" \
722 "\15CSUM_IP6_TSO\16CSUM_IP6_ISCSI\17CSUM_INNER_IP\20CSUM_INNER_IP_UDP" \
723 "\21CSUM_INNER_IP_TCP\22CSUM_INNER_IP_TSO\23CSUM_ENCAP_VXLAN" \
724 "\24CSUM_ENCAP_RSVD1\25CSUM_INNER_L3_CALC\26CSUM_INNER_L3_VALID" \
725 "\27CSUM_INNER_L4_CALC\30CSUM_INNER_L4_VALID\31CSUM_L3_CALC" \
726 "\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID\35CSUM_L5_CALC" \
727 "\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
729 /* CSUM flags compatibility mappings. */
730 #define CSUM_IP_CHECKED CSUM_L3_CALC
731 #define CSUM_IP_VALID CSUM_L3_VALID
732 #define CSUM_DATA_VALID CSUM_L4_VALID
733 #define CSUM_PSEUDO_HDR CSUM_L4_CALC
734 #define CSUM_SCTP_VALID CSUM_L4_VALID
735 #define CSUM_DELAY_DATA (CSUM_TCP|CSUM_UDP)
736 #define CSUM_DELAY_IP CSUM_IP /* Only v4, no v6 IP hdr csum */
737 #define CSUM_DELAY_DATA_IPV6 (CSUM_TCP_IPV6|CSUM_UDP_IPV6)
738 #define CSUM_DATA_VALID_IPV6 CSUM_DATA_VALID
739 #define CSUM_TCP CSUM_IP_TCP
740 #define CSUM_UDP CSUM_IP_UDP
741 #define CSUM_SCTP CSUM_IP_SCTP
742 #define CSUM_TSO (CSUM_IP_TSO|CSUM_IP6_TSO)
743 #define CSUM_INNER_TSO (CSUM_INNER_IP_TSO|CSUM_INNER_IP6_TSO)
744 #define CSUM_UDP_IPV6 CSUM_IP6_UDP
745 #define CSUM_TCP_IPV6 CSUM_IP6_TCP
746 #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP
747 #define CSUM_TLS_MASK (CSUM_L5_CALC|CSUM_L5_VALID)
748 #define CSUM_TLS_DECRYPTED CSUM_L5_CALC
751 * mbuf types describing the content of the mbuf (including external storage).
753 #define MT_NOTMBUF 0 /* USED INTERNALLY ONLY! Object is not mbuf */
754 #define MT_DATA 1 /* dynamic (data) allocation */
755 #define MT_HEADER MT_DATA /* packet header, use M_PKTHDR instead */
757 #define MT_VENDOR1 4 /* for vendor-internal use */
758 #define MT_VENDOR2 5 /* for vendor-internal use */
759 #define MT_VENDOR3 6 /* for vendor-internal use */
760 #define MT_VENDOR4 7 /* for vendor-internal use */
762 #define MT_SONAME 8 /* socket name */
764 #define MT_EXP1 9 /* for experimental use */
765 #define MT_EXP2 10 /* for experimental use */
766 #define MT_EXP3 11 /* for experimental use */
767 #define MT_EXP4 12 /* for experimental use */
769 #define MT_CONTROL 14 /* extra-data protocol message */
770 #define MT_EXTCONTROL 15 /* control message with externalized contents */
771 #define MT_OOBDATA 16 /* expedited data */
773 #define MT_NOINIT 255 /* Not a type but a flag to allocate
774 a non-initialized mbuf */
777 * String names of mbuf-related UMA(9) and malloc(9) types. Exposed to
778 * !_KERNEL so that monitoring tools can look up the zones with
781 #define MBUF_MEM_NAME "mbuf"
782 #define MBUF_CLUSTER_MEM_NAME "mbuf_cluster"
783 #define MBUF_PACKET_MEM_NAME "mbuf_packet"
784 #define MBUF_JUMBOP_MEM_NAME "mbuf_jumbo_page"
785 #define MBUF_JUMBO9_MEM_NAME "mbuf_jumbo_9k"
786 #define MBUF_JUMBO16_MEM_NAME "mbuf_jumbo_16k"
787 #define MBUF_TAG_MEM_NAME "mbuf_tag"
788 #define MBUF_EXTREFCNT_MEM_NAME "mbuf_ext_refcnt"
789 #define MBUF_EXTPGS_MEM_NAME "mbuf_extpgs"
792 union if_snd_tag_alloc_params;
795 #define MBUF_CHECKSLEEP(how) do { \
796 if (how == M_WAITOK) \
797 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \
798 "Sleeping in \"%s\"", __func__); \
801 #define MBUF_CHECKSLEEP(how) do {} while (0)
805 * Network buffer allocation API
807 * The rest of it is defined in kern/kern_mbuf.c
809 extern uma_zone_t zone_mbuf;
810 extern uma_zone_t zone_clust;
811 extern uma_zone_t zone_pack;
812 extern uma_zone_t zone_jumbop;
813 extern uma_zone_t zone_jumbo9;
814 extern uma_zone_t zone_jumbo16;
815 extern uma_zone_t zone_extpgs;
817 void mb_dupcl(struct mbuf *, struct mbuf *);
818 void mb_free_ext(struct mbuf *);
819 void mb_free_extpg(struct mbuf *);
820 void mb_free_mext_pgs(struct mbuf *);
821 struct mbuf *mb_alloc_ext_pgs(int, m_ext_free_t);
822 struct mbuf *mb_alloc_ext_plus_pages(int, int);
823 struct mbuf *mb_mapped_to_unmapped(struct mbuf *, int, int, int,
825 int mb_unmapped_compress(struct mbuf *m);
826 struct mbuf *mb_unmapped_to_ext(struct mbuf *m);
827 void mb_free_notready(struct mbuf *m, int count);
828 void m_adj(struct mbuf *, int);
829 void m_adj_decap(struct mbuf *, int);
830 int m_apply(struct mbuf *, int, int,
831 int (*)(void *, void *, u_int), void *);
832 int m_append(struct mbuf *, int, c_caddr_t);
833 void m_cat(struct mbuf *, struct mbuf *);
834 void m_catpkt(struct mbuf *, struct mbuf *);
835 int m_clget(struct mbuf *m, int how);
836 void *m_cljget(struct mbuf *m, int how, int size);
837 struct mbuf *m_collapse(struct mbuf *, int, int);
838 void m_copyback(struct mbuf *, int, int, c_caddr_t);
839 void m_copydata(const struct mbuf *, int, int, caddr_t);
840 struct mbuf *m_copym(struct mbuf *, int, int, int);
841 struct mbuf *m_copypacket(struct mbuf *, int);
842 void m_copy_pkthdr(struct mbuf *, struct mbuf *);
843 struct mbuf *m_copyup(struct mbuf *, int, int);
844 struct mbuf *m_defrag(struct mbuf *, int);
845 void m_demote_pkthdr(struct mbuf *);
846 void m_demote(struct mbuf *, int, int);
847 struct mbuf *m_devget(char *, int, int, struct ifnet *,
848 void (*)(char *, caddr_t, u_int));
849 void m_dispose_extcontrolm(struct mbuf *m);
850 struct mbuf *m_dup(const struct mbuf *, int);
851 int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
852 void m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
853 void *, void *, int, int);
854 u_int m_fixhdr(struct mbuf *);
855 struct mbuf *m_fragment(struct mbuf *, int, int);
856 void m_freem(struct mbuf *);
857 void m_free_raw(struct mbuf *);
858 struct mbuf *m_get2(int, int, short, int);
859 struct mbuf *m_get3(int, int, short, int);
860 struct mbuf *m_getjcl(int, short, int, int);
861 struct mbuf *m_getm2(struct mbuf *, int, int, short, int);
862 struct mbuf *m_getptr(struct mbuf *, int, int *);
863 u_int m_length(struct mbuf *, struct mbuf **);
864 int m_mbuftouio(struct uio *, const struct mbuf *, int);
865 void m_move_pkthdr(struct mbuf *, struct mbuf *);
866 int m_pkthdr_init(struct mbuf *, int);
867 struct mbuf *m_prepend(struct mbuf *, int, int);
868 void m_print(const struct mbuf *, int);
869 struct mbuf *m_pulldown(struct mbuf *, int, int, int *);
870 struct mbuf *m_pullup(struct mbuf *, int);
871 int m_sanity(struct mbuf *, int);
872 struct mbuf *m_split(struct mbuf *, int, int);
873 struct mbuf *m_uiotombuf(struct uio *, int, int, int, int);
874 int m_unmapped_uiomove(const struct mbuf *, int, struct uio *,
876 struct mbuf *m_unshare(struct mbuf *, int);
877 int m_snd_tag_alloc(struct ifnet *,
878 union if_snd_tag_alloc_params *, struct m_snd_tag **);
879 void m_snd_tag_init(struct m_snd_tag *, struct ifnet *,
880 const struct if_snd_tag_sw *);
881 void m_snd_tag_destroy(struct m_snd_tag *);
882 void m_rcvif_serialize(struct mbuf *);
883 struct ifnet *m_rcvif_restore(struct mbuf *);
897 #if MJUMPAGESIZE != MCLBYTES
909 panic("%s: invalid cluster size %d", __func__, size);
916 * Associated an external reference counted buffer with an mbuf.
919 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
920 m_ext_free_t freef, void *arg1, void *arg2)
923 KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
925 atomic_add_int(ref_cnt, 1);
927 m->m_ext.ext_buf = buf;
928 m->m_ext.ext_cnt = ref_cnt;
929 m->m_data = m->m_ext.ext_buf;
930 m->m_ext.ext_size = size;
931 m->m_ext.ext_free = freef;
932 m->m_ext.ext_arg1 = arg1;
933 m->m_ext.ext_arg2 = arg2;
934 m->m_ext.ext_type = EXT_EXTREF;
935 m->m_ext.ext_flags = 0;
938 static __inline uma_zone_t
947 #if MJUMPAGESIZE != MCLBYTES
959 panic("%s: invalid cluster size %d", __func__, size);
966 * Initialize an mbuf with linear storage.
968 * Inline because the consumer text overhead will be roughly the same to
969 * initialize or call a function with this many parameters and M_PKTHDR
970 * should go away with constant propagation for !MGETHDR.
973 m_init(struct mbuf *m, int how, short type, int flags)
979 m->m_data = m->m_dat;
983 if (flags & M_PKTHDR)
984 error = m_pkthdr_init(m, how);
988 MBUF_PROBE5(m__init, m, how, type, flags, error);
992 static __inline struct mbuf *
993 m_get_raw(int how, short type)
999 args.type = type | MT_NOINIT;
1000 m = uma_zalloc_arg(zone_mbuf, &args, how);
1001 MBUF_PROBE3(m__get_raw, how, type, m);
1005 static __inline struct mbuf *
1006 m_get(int how, short type)
1009 struct mb_args args;
1013 m = uma_zalloc_arg(zone_mbuf, &args, how);
1014 MBUF_PROBE3(m__get, how, type, m);
1018 static __inline struct mbuf *
1019 m_gethdr_raw(int how, short type)
1022 struct mb_args args;
1024 args.flags = M_PKTHDR;
1025 args.type = type | MT_NOINIT;
1026 m = uma_zalloc_arg(zone_mbuf, &args, how);
1027 MBUF_PROBE3(m__gethdr_raw, how, type, m);
1031 static __inline struct mbuf *
1032 m_gethdr(int how, short type)
1035 struct mb_args args;
1037 args.flags = M_PKTHDR;
1039 m = uma_zalloc_arg(zone_mbuf, &args, how);
1040 MBUF_PROBE3(m__gethdr, how, type, m);
1044 static __inline struct mbuf *
1045 m_getcl(int how, short type, int flags)
1048 struct mb_args args;
1052 m = uma_zalloc_arg(zone_pack, &args, how);
1053 MBUF_PROBE4(m__getcl, how, type, flags, m);
1058 * XXX: m_cljset() is a dangerous API. One must attach only a new,
1059 * unreferenced cluster to an mbuf(9). It is not possible to assert
1060 * that, so care can be taken only by users of the API.
1062 static __inline void
1063 m_cljset(struct mbuf *m, void *cl, int type)
1071 #if MJUMPAGESIZE != MCLBYTES
1073 size = MJUMPAGESIZE;
1083 panic("%s: unknown cluster type %d", __func__, type);
1087 m->m_data = m->m_ext.ext_buf = cl;
1088 m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
1089 m->m_ext.ext_size = size;
1090 m->m_ext.ext_type = type;
1091 m->m_ext.ext_flags = EXT_FLAG_EMBREF;
1092 m->m_ext.ext_count = 1;
1093 m->m_flags |= M_EXT;
1094 MBUF_PROBE3(m__cljset, m, cl, type);
1097 static __inline void
1098 m_chtype(struct mbuf *m, short new_type)
1101 m->m_type = new_type;
1104 static __inline void
1105 m_clrprotoflags(struct mbuf *m)
1109 m->m_flags &= ~M_PROTOFLAGS;
1114 static __inline struct mbuf *
1115 m_last(struct mbuf *m)
1124 m_extrefcnt(struct mbuf *m)
1127 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
1129 return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
1134 * mbuf, cluster, and external object allocation macros (for compatibility
1137 #define M_MOVE_PKTHDR(to, from) m_move_pkthdr((to), (from))
1138 #define MGET(m, how, type) ((m) = m_get((how), (type)))
1139 #define MGETHDR(m, how, type) ((m) = m_gethdr((how), (type)))
1140 #define MCLGET(m, how) m_clget((m), (how))
1141 #define MEXTADD(m, buf, size, free, arg1, arg2, flags, type) \
1142 m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2), \
1144 #define m_getm(m, len, how, type) \
1145 m_getm2((m), (len), (how), (type), M_PKTHDR)
1148 * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
1149 * be both the local data payload, or an external buffer area, depending on
1150 * whether M_EXT is set).
1152 #define M_WRITABLE(m) (((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 && \
1153 (!(((m)->m_flags & M_EXT)) || \
1154 (m_extrefcnt(m) == 1)))
1156 /* Check if the supplied mbuf has a packet header, or else panic. */
1157 #define M_ASSERTPKTHDR(m) \
1158 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR, \
1159 ("%s: no mbuf packet header!", __func__))
1161 /* Check if the supplied mbuf has no send tag, or else panic. */
1162 #define M_ASSERT_NO_SND_TAG(m) \
1163 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR && \
1164 ((m)->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0, \
1165 ("%s: receive mbuf has send tag!", __func__))
1167 /* Check if mbuf is multipage. */
1168 #define M_ASSERTEXTPG(m) \
1169 KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG, \
1170 ("%s: m %p is not multipage!", __func__, m))
1173 * Ensure that the supplied mbuf is a valid, non-free mbuf.
1175 * XXX: Broken at the moment. Need some UMA magic to make it work again.
1177 #define M_ASSERTVALID(m) \
1178 KASSERT((((struct mbuf *)m)->m_flags & 0) == 0, \
1179 ("%s: attempted use of a free mbuf!", __func__))
1181 /* Check whether any mbuf in the chain is unmapped. */
1183 #define M_ASSERTMAPPED(m) do { \
1184 for (struct mbuf *__m = (m); __m != NULL; __m = __m->m_next) \
1185 KASSERT((__m->m_flags & M_EXTPG) == 0, \
1186 ("%s: chain %p contains an unmapped mbuf", __func__, (m)));\
1189 #define M_ASSERTMAPPED(m) do {} while (0)
1193 * Return the address of the start of the buffer associated with an mbuf,
1194 * handling external storage, packet-header mbufs, and regular data mbufs.
1196 #define M_START(m) \
1197 (((m)->m_flags & M_EXTPG) ? NULL : \
1198 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf : \
1199 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] : \
1203 * Return the size of the buffer associated with an mbuf, handling external
1204 * storage, packet-header mbufs, and regular data mbufs.
1207 (((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size : \
1208 ((m)->m_flags & M_PKTHDR) ? MHLEN : \
1212 * Set the m_data pointer of a newly allocated mbuf to place an object of the
1213 * specified size at the end of the mbuf, longword aligned.
1215 * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1216 * separate macros, each asserting that it was called at the proper moment.
1217 * This required callers to themselves test the storage type and call the
1218 * right one. Rather than require callers to be aware of those layout
1219 * decisions, we centralize here.
1221 static __inline void
1222 m_align(struct mbuf *m, int len)
1225 const char *msg = "%s: not a virgin mbuf";
1229 KASSERT(m->m_data == M_START(m), (msg, __func__));
1231 adjust = M_SIZE(m) - len;
1232 m->m_data += adjust &~ (sizeof(long)-1);
1235 #define M_ALIGN(m, len) m_align(m, len)
1236 #define MH_ALIGN(m, len) m_align(m, len)
1237 #define MEXT_ALIGN(m, len) m_align(m, len)
1240 * Compute the amount of space available before the current start of data in
1243 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1244 * of checking writability of the mbuf data area rests solely with the caller.
1246 * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1247 * for mbufs with external storage. We now allow mbuf-embedded data to be
1248 * read-only as well.
1250 #define M_LEADINGSPACE(m) \
1251 (M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1254 * Compute the amount of space available after the end of data in an mbuf.
1256 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1257 * of checking writability of the mbuf data area rests solely with the caller.
1259 * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1260 * for mbufs with external storage. We now allow mbuf-embedded data to be
1261 * read-only as well.
1263 #define M_TRAILINGSPACE(m) \
1265 ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0)
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.
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); \
1278 MBUF_CHECKSLEEP(how); \
1279 if (M_LEADINGSPACE(_mm) >= _mplen) { \
1280 _mm->m_data -= _mplen; \
1281 _mm->m_len += _mplen; \
1283 _mm = m_prepend(_mm, _mplen, __mhow); \
1284 if (_mm != NULL && _mm->m_flags & M_PKTHDR) \
1285 _mm->m_pkthdr.len += _mplen; \
1290 * Change mbuf to new type. This is a relatively expensive operation and
1291 * should be avoided.
1293 #define MCHTYPE(m, t) m_chtype((m), (t))
1295 /* Return the rcvif of a packet header. */
1296 static __inline struct ifnet *
1297 m_rcvif(struct mbuf *m)
1301 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1303 return (m->m_pkthdr.rcvif);
1306 /* Length to m_copy to copy all. */
1307 #define M_COPYALL 1000000000
1309 extern int max_datalen; /* MHLEN - max_hdr */
1310 extern int max_hdr; /* Largest link + protocol header */
1311 extern int max_linkhdr; /* Largest link-level header */
1312 extern int max_protohdr; /* Largest protocol header */
1313 extern int nmbclusters; /* Maximum number of clusters */
1314 extern bool mb_use_ext_pgs; /* Use ext_pgs for sendfile */
1317 * Network packets may have annotations attached by affixing a list of
1318 * "packet tags" to the pkthdr structure. Packet tags are dynamically
1319 * allocated semi-opaque data structures that have a fixed header
1320 * (struct m_tag) that specifies the size of the memory block and a
1321 * <cookie,type> pair that identifies it. The cookie is a 32-bit unique
1322 * unsigned value used to identify a module or ABI. By convention this value
1323 * is chosen as the date+time that the module is created, expressed as the
1324 * number of seconds since the epoch (e.g., using date -u +'%s'). The type
1325 * value is an ABI/module-specific value that identifies a particular
1326 * annotation and is private to the module. For compatibility with systems
1327 * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1328 * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1329 * compatibility shim functions and several tag types are defined below.
1330 * Users that do not require compatibility should use a private cookie value
1331 * so that packet tag-related definitions can be maintained privately.
1333 * Note that the packet tag returned by m_tag_alloc has the default memory
1334 * alignment implemented by malloc. To reference private data one can use a
1337 * struct m_tag *mtag = m_tag_alloc(...);
1338 * struct foo *p = (struct foo *)(mtag+1);
1340 * if the alignment of struct m_tag is sufficient for referencing members of
1341 * struct foo. Otherwise it is necessary to embed struct m_tag within the
1342 * private data structure to insure proper alignment; e.g.,
1348 * struct foo *p = (struct foo *) m_tag_alloc(...);
1349 * struct m_tag *mtag = &p->tag;
1353 * Persistent tags stay with an mbuf until the mbuf is reclaimed. Otherwise
1354 * tags are expected to ``vanish'' when they pass through a network
1355 * interface. For most interfaces this happens normally as the tags are
1356 * reclaimed when the mbuf is free'd. However in some special cases
1357 * reclaiming must be done manually. An example is packets that pass through
1358 * the loopback interface. Also, one must be careful to do this when
1359 * ``turning around'' packets (e.g., icmp_reflect).
1361 * To mark a tag persistent bit-or this flag in when defining the tag id.
1362 * The tag will then be treated as described above.
1364 #define MTAG_PERSISTENT 0x800
1366 #define PACKET_TAG_NONE 0 /* Nadda */
1368 /* Packet tags for use with PACKET_ABI_COMPAT. */
1369 #define PACKET_TAG_IPSEC_IN_DONE 1 /* IPsec applied, in */
1370 #define PACKET_TAG_IPSEC_OUT_DONE 2 /* IPsec applied, out */
1371 #define PACKET_TAG_IPSEC_IN_CRYPTO_DONE 3 /* NIC IPsec crypto done */
1372 #define PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED 4 /* NIC IPsec crypto req'ed */
1373 #define PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO 5 /* NIC notifies IPsec */
1374 #define PACKET_TAG_IPSEC_PENDING_TDB 6 /* Reminder to do IPsec */
1375 #define PACKET_TAG_BRIDGE 7 /* Bridge processing done */
1376 #define PACKET_TAG_GIF 8 /* GIF processing done */
1377 #define PACKET_TAG_GRE 9 /* GRE processing done */
1378 #define PACKET_TAG_IN_PACKET_CHECKSUM 10 /* NIC checksumming done */
1379 #define PACKET_TAG_ENCAP 11 /* Encap. processing */
1380 #define PACKET_TAG_IPSEC_SOCKET 12 /* IPSEC socket ref */
1381 #define PACKET_TAG_IPSEC_HISTORY 13 /* IPSEC history */
1382 #define PACKET_TAG_IPV6_INPUT 14 /* IPV6 input processing */
1383 #define PACKET_TAG_DUMMYNET 15 /* dummynet info */
1384 #define PACKET_TAG_DIVERT 17 /* divert info */
1385 #define PACKET_TAG_IPFORWARD 18 /* ipforward info */
1386 #define PACKET_TAG_MACLABEL (19 | MTAG_PERSISTENT) /* MAC label */
1387 #define PACKET_TAG_PF 21 /* PF/ALTQ information */
1388 #define PACKET_TAG_RTSOCKFAM 25 /* rtsock sa family */
1389 #define PACKET_TAG_IPOPTIONS 27 /* Saved IP options */
1390 #define PACKET_TAG_CARP 28 /* CARP info */
1391 #define PACKET_TAG_IPSEC_NAT_T_PORTS 29 /* two uint16_t */
1392 #define PACKET_TAG_ND_OUTGOING 30 /* ND outgoing */
1394 /* Specific cookies and tags. */
1396 /* Packet tag routines. */
1397 struct m_tag *m_tag_alloc(uint32_t, uint16_t, int, int);
1398 void m_tag_delete(struct mbuf *, struct m_tag *);
1399 void m_tag_delete_chain(struct mbuf *, struct m_tag *);
1400 void m_tag_free_default(struct m_tag *);
1401 struct m_tag *m_tag_locate(struct mbuf *, uint32_t, uint16_t,
1403 struct m_tag *m_tag_copy(struct m_tag *, int);
1404 int m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1405 void m_tag_delete_nonpersistent(struct mbuf *);
1408 * Initialize the list of tags associated with an mbuf.
1410 static __inline void
1411 m_tag_init(struct mbuf *m)
1414 SLIST_INIT(&m->m_pkthdr.tags);
1418 * Set up the contents of a tag. Note that this does not fill in the free
1419 * method; the caller is expected to do that.
1421 * XXX probably should be called m_tag_init, but that was already taken.
1423 static __inline void
1424 m_tag_setup(struct m_tag *t, uint32_t cookie, uint16_t type, int len)
1429 t->m_tag_cookie = cookie;
1433 * Reclaim resources associated with a tag.
1435 static __inline void
1436 m_tag_free(struct m_tag *t)
1439 (*t->m_tag_free)(t);
1443 * Return the first tag associated with an mbuf.
1445 static __inline struct m_tag *
1446 m_tag_first(struct mbuf *m)
1449 return (SLIST_FIRST(&m->m_pkthdr.tags));
1453 * Return the next tag in the list of tags associated with an mbuf.
1455 static __inline struct m_tag *
1456 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1459 return (SLIST_NEXT(t, m_tag_link));
1463 * Prepend a tag to the list of tags associated with an mbuf.
1465 static __inline void
1466 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1469 SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1473 * Unlink a tag from the list of tags associated with an mbuf.
1475 static __inline void
1476 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1479 SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1482 /* These are for OpenBSD compatibility. */
1483 #define MTAG_ABI_COMPAT 0 /* compatibility ABI */
1485 static __inline struct m_tag *
1486 m_tag_get(uint16_t type, int length, int wait)
1488 return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1491 static __inline struct m_tag *
1492 m_tag_find(struct mbuf *m, uint16_t type, struct m_tag *start)
1494 return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1495 m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1498 static inline struct m_snd_tag *
1499 m_snd_tag_ref(struct m_snd_tag *mst)
1502 refcount_acquire(&mst->refcount);
1507 m_snd_tag_rele(struct m_snd_tag *mst)
1510 if (refcount_release(&mst->refcount))
1511 m_snd_tag_destroy(mst);
1514 static __inline struct mbuf *
1515 m_free(struct mbuf *m)
1517 struct mbuf *n = m->m_next;
1519 MBUF_PROBE1(m__free, m);
1520 if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1521 m_tag_delete_chain(m, NULL);
1522 if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1523 m_snd_tag_rele(m->m_pkthdr.snd_tag);
1524 if (m->m_flags & M_EXTPG)
1526 else if (m->m_flags & M_EXT)
1528 else if ((m->m_flags & M_NOFREE) == 0)
1529 uma_zfree(zone_mbuf, m);
1534 rt_m_getfib(struct mbuf *m)
1536 KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1537 return (m->m_pkthdr.fibnum);
1540 #define M_GETFIB(_m) rt_m_getfib(_m)
1542 #define M_SETFIB(_m, _fib) do { \
1543 KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf.")); \
1544 ((_m)->m_pkthdr.fibnum) = (_fib); \
1547 /* flags passed as first argument for "m_xxx_tcpip_hash()" */
1548 #define MBUF_HASHFLAG_L2 (1 << 2)
1549 #define MBUF_HASHFLAG_L3 (1 << 3)
1550 #define MBUF_HASHFLAG_L4 (1 << 4)
1552 /* mbuf hashing helper routines */
1553 uint32_t m_ether_tcpip_hash_init(void);
1554 uint32_t m_ether_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1555 uint32_t m_infiniband_tcpip_hash_init(void);
1556 uint32_t m_infiniband_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1558 #ifdef MBUF_PROFILING
1559 void m_profile(struct mbuf *m);
1560 #define M_PROFILE(m) m_profile(m)
1562 #define M_PROFILE(m)
1566 STAILQ_HEAD(, mbuf) mq_head;
1572 mbufq_init(struct mbufq *mq, int maxlen)
1575 STAILQ_INIT(&mq->mq_head);
1576 mq->mq_maxlen = maxlen;
1580 static inline struct mbuf *
1581 mbufq_flush(struct mbufq *mq)
1585 m = STAILQ_FIRST(&mq->mq_head);
1586 STAILQ_INIT(&mq->mq_head);
1592 mbufq_drain(struct mbufq *mq)
1596 n = mbufq_flush(mq);
1597 while ((m = n) != NULL) {
1598 n = STAILQ_NEXT(m, m_stailqpkt);
1603 static inline struct mbuf *
1604 mbufq_first(const struct mbufq *mq)
1607 return (STAILQ_FIRST(&mq->mq_head));
1610 static inline struct mbuf *
1611 mbufq_last(const struct mbufq *mq)
1614 return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1618 mbufq_full(const struct mbufq *mq)
1621 return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1625 mbufq_len(const struct mbufq *mq)
1628 return (mq->mq_len);
1632 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1637 STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1642 static inline struct mbuf *
1643 mbufq_dequeue(struct mbufq *mq)
1647 m = STAILQ_FIRST(&mq->mq_head);
1649 STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1650 m->m_nextpkt = NULL;
1657 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1660 STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1665 * Note: this doesn't enforce the maximum list size for dst.
1668 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1671 mq_dst->mq_len += mq_src->mq_len;
1672 STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1676 #ifdef _SYS_TIMESPEC_H_
1678 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1681 KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1682 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0, ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1683 ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1684 ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1689 /* Invoked from the debugnet client code. */
1690 void debugnet_mbuf_drain(void);
1691 void debugnet_mbuf_start(void);
1692 void debugnet_mbuf_finish(void);
1693 void debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1697 mbuf_has_tls_session(struct mbuf *m)
1700 if (m->m_flags & M_EXTPG) {
1701 if (m->m_epg_tls != NULL) {
1708 #endif /* _KERNEL */
1709 #endif /* !_SYS_MBUF_H_ */