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
36 /* XXX: These includes suck. Sorry! */
37 #include <sys/queue.h>
39 #include <sys/systm.h>
40 #include <sys/refcount.h>
45 #define MBUF_PROBE1(probe, arg0) \
46 SDT_PROBE1(sdt, , , probe, arg0)
47 #define MBUF_PROBE2(probe, arg0, arg1) \
48 SDT_PROBE2(sdt, , , probe, arg0, arg1)
49 #define MBUF_PROBE3(probe, arg0, arg1, arg2) \
50 SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
51 #define MBUF_PROBE4(probe, arg0, arg1, arg2, arg3) \
52 SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
53 #define MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4) \
54 SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
56 SDT_PROBE_DECLARE(sdt, , , m__init);
57 SDT_PROBE_DECLARE(sdt, , , m__gethdr_raw);
58 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
59 SDT_PROBE_DECLARE(sdt, , , m__get_raw);
60 SDT_PROBE_DECLARE(sdt, , , m__get);
61 SDT_PROBE_DECLARE(sdt, , , m__getcl);
62 SDT_PROBE_DECLARE(sdt, , , m__getjcl);
63 SDT_PROBE_DECLARE(sdt, , , m__clget);
64 SDT_PROBE_DECLARE(sdt, , , m__cljget);
65 SDT_PROBE_DECLARE(sdt, , , m__cljset);
66 SDT_PROBE_DECLARE(sdt, , , m__free);
67 SDT_PROBE_DECLARE(sdt, , , m__freem);
72 * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
73 * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
74 * sys/param.h), which has no additional overhead and is used instead of the
75 * internal data area; this is done when at least MINCLSIZE of data must be
76 * stored. Additionally, it is possible to allocate a separate buffer
77 * externally and attach it to the mbuf in a way similar to that of mbuf
80 * NB: These calculation do not take actual compiler-induced alignment and
81 * padding inside the complete struct mbuf into account. Appropriate
82 * attention is required when changing members of struct mbuf.
84 * MLEN is data length in a normal mbuf.
85 * MHLEN is data length in an mbuf with pktheader.
86 * MINCLSIZE is a smallest amount of data that should be put into cluster.
88 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
92 #define MHSIZE offsetof(struct mbuf, m_dat)
93 #define MPKTHSIZE offsetof(struct mbuf, m_pktdat)
94 #define MLEN ((int)(MSIZE - MHSIZE))
95 #define MHLEN ((int)(MSIZE - MPKTHSIZE))
96 #define MINCLSIZE (MHLEN + 1)
101 * Macro for type conversion: convert mbuf pointer to data pointer of correct
104 * mtod(m, t) -- Convert mbuf pointer to data pointer of correct type.
105 * mtodo(m, o) -- Same as above but with offset 'o' into data.
107 #define mtod(m, t) ((t)((m)->m_data))
108 #define mtodo(m, o) ((void *)(((m)->m_data) + (o)))
111 * Argument structure passed to UMA routines during mbuf and packet
115 int flags; /* Flags for mbuf being allocated */
116 short type; /* Type of mbuf being allocated */
121 * Packet tag structure (see below for details).
124 SLIST_ENTRY(m_tag) m_tag_link; /* List of packet tags */
125 u_int16_t m_tag_id; /* Tag ID */
126 u_int16_t m_tag_len; /* Length of data */
127 u_int32_t m_tag_cookie; /* ABI/Module ID */
128 void (*m_tag_free)(struct m_tag *);
132 * Static network interface owned tag.
133 * Allocated through ifp->if_snd_tag_alloc().
135 struct if_snd_tag_sw;
138 struct ifnet *ifp; /* network interface tag belongs to */
139 const struct if_snd_tag_sw *sw;
140 volatile u_int refcount;
144 * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
147 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
152 struct m_snd_tag *snd_tag; /* send tag, if any */
153 struct ifnet *rcvif; /* rcv interface */
155 uint16_t rcvidx; /* rcv interface index ... */
156 uint16_t rcvgen; /* ... and generation count */
160 struct ifnet *leaf_rcvif; /* leaf rcv interface */
162 uint16_t leaf_rcvidx; /* leaf rcv interface index ... */
163 uint16_t leaf_rcvgen; /* ... and generation count */
166 SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
167 int32_t len; /* total packet length */
169 /* Layer crossing persistent information. */
170 uint32_t flowid; /* packet's 4-tuple system */
171 uint32_t csum_flags; /* checksum and offload features */
172 uint16_t fibnum; /* this packet should use this fib */
173 uint8_t numa_domain; /* NUMA domain of recvd pkt */
174 uint8_t rsstype; /* hash type */
175 #if !defined(__LP64__)
176 uint32_t pad; /* pad for 64bit alignment */
179 uint64_t rcv_tstmp; /* timestamp in ns */
181 uint8_t l2hlen; /* layer 2 hdr len */
182 uint8_t l3hlen; /* layer 3 hdr len */
183 uint8_t l4hlen; /* layer 4 hdr len */
184 uint8_t l5hlen; /* layer 5 hdr len */
185 uint8_t inner_l2hlen;
186 uint8_t inner_l3hlen;
187 uint8_t inner_l4hlen;
188 uint8_t inner_l5hlen;
194 uint32_t thirtytwo[2];
195 uint64_t sixtyfour[1];
196 uintptr_t unintptr[1];
200 /* Layer specific non-persistent local storage for reassembly, etc. */
205 uint32_t thirtytwo[2];
206 uint64_t sixtyfour[1];
207 uintptr_t unintptr[1];
210 /* Upon allocation: total packet memory consumption. */
214 #define ether_vtag PH_per.sixteen[0]
215 #define tcp_tun_port PH_per.sixteen[0] /* outbound */
216 #define vt_nrecs PH_per.sixteen[0] /* mld and v6-ND */
217 #define tso_segsz PH_per.sixteen[1] /* inbound after LRO */
218 #define lro_nsegs tso_segsz /* inbound after LRO */
219 #define csum_data PH_per.thirtytwo[1] /* inbound from hardware up */
220 #define lro_tcp_d_len PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
221 #define lro_tcp_d_csum PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
222 #define lro_tcp_h_off PH_loc.sixteen[2] /* inbound during LRO (no reassembly) */
223 #define lro_etype PH_loc.sixteen[3] /* inbound during LRO (no reassembly) */
224 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
227 * TLS records for TLS 1.0-1.2 can have the following header lengths:
228 * - 5 (AES-CBC with implicit IV)
229 * - 21 (AES-CBC with explicit IV)
230 * - 13 (AES-GCM with 8 byte explicit IV)
232 #define MBUF_PEXT_HDR_LEN 23
235 * TLS records for TLS 1.0-1.2 can have the following maximum trailer
238 * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
239 * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
240 * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
242 #define MBUF_PEXT_TRAIL_LEN 64
244 #if defined(__LP64__)
245 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
247 #define MBUF_PEXT_MAX_PGS (64 / sizeof(vm_paddr_t))
250 #define MBUF_PEXT_MAX_BYTES \
251 (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
257 * Description of external storage mapped into mbuf; valid only if M_EXT is
261 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
264 typedef void m_ext_free_t(struct mbuf *);
268 * If EXT_FLAG_EMBREF is set, then we use refcount in the
269 * mbuf, the 'ext_count' member. Otherwise, we have a
270 * shadow copy and we use pointer 'ext_cnt'. The original
271 * mbuf is responsible to carry the pointer to free routine
272 * and its arguments. They aren't copied into shadows in
273 * mb_dupcl() to avoid dereferencing next cachelines.
275 volatile u_int ext_count;
276 volatile u_int *ext_cnt;
278 uint32_t ext_size; /* size of buffer, for ext_free */
279 uint32_t ext_type:8, /* type of external storage */
280 ext_flags:24; /* external storage mbuf flags */
284 * Regular M_EXT mbuf:
285 * o ext_buf always points to the external buffer.
286 * o ext_free (below) and two optional arguments
287 * ext_arg1 and ext_arg2 store the free context for
288 * the external storage. They are set only in the
289 * refcount carrying mbuf, the one with
290 * EXT_FLAG_EMBREF flag, with exclusion for
291 * EXT_EXTREF type, where the free context is copied
292 * into all mbufs that use same external storage.
294 char *ext_buf; /* start of buffer */
295 #define m_ext_copylen offsetof(struct m_ext, ext_arg2)
300 * Multi-page M_EXTPG mbuf:
301 * o extpg_pa - page vector.
302 * o extpg_trail and extpg_hdr - TLS trailer and
304 * Uses ext_free and may also use ext_arg1.
306 vm_paddr_t extpg_pa[MBUF_PEXT_MAX_PGS];
307 char extpg_trail[MBUF_PEXT_TRAIL_LEN];
308 char extpg_hdr[MBUF_PEXT_HDR_LEN];
309 /* Pretend these 3 fields are part of mbuf itself. */
310 #define m_epg_pa m_ext.extpg_pa
311 #define m_epg_trail m_ext.extpg_trail
312 #define m_epg_hdr m_ext.extpg_hdr
313 #define m_epg_ext_copylen offsetof(struct m_ext, ext_free)
317 * Free method and optional argument pointer, both
318 * used by M_EXT and M_EXTPG.
320 m_ext_free_t *ext_free;
325 * The core of the mbuf object along with some shortcut defines for practical
330 * Header present at the beginning of every mbuf.
333 * Compile-time assertions in uipc_mbuf.c test these values to ensure
334 * that they are correct.
336 union { /* next buffer in chain */
338 SLIST_ENTRY(mbuf) m_slist;
339 STAILQ_ENTRY(mbuf) m_stailq;
341 union { /* next chain in queue/record */
342 struct mbuf *m_nextpkt;
343 SLIST_ENTRY(mbuf) m_slistpkt;
344 STAILQ_ENTRY(mbuf) m_stailqpkt;
346 caddr_t m_data; /* location of data */
347 int32_t m_len; /* amount of data in this mbuf */
348 uint32_t m_type:8, /* type of data in this mbuf */
349 m_flags:24; /* flags; see below */
350 #if !defined(__LP64__)
351 uint32_t m_pad; /* pad for 64bit alignment */
355 * A set of optional headers (packet header, external storage header)
356 * and internal data storage. Historically, these arrays were sized
357 * to MHLEN (space left after a packet header) and MLEN (space left
358 * after only a regular mbuf header); they are now variable size in
359 * order to support future work on variable-size mbufs.
365 struct pkthdr m_pkthdr;
368 * Multi-page M_EXTPG mbuf has its meta data
369 * split between the below anonymous structure
370 * and m_ext. It carries vector of pages,
371 * optional header and trailer char vectors
372 * and pointers to socket/TLS data.
374 #define m_epg_startcopy m_epg_npgs
375 #define m_epg_endcopy m_epg_stailq
377 /* Overall count of pages and count of
378 * pages with I/O pending. */
381 /* TLS header and trailer lengths.
382 * The data itself resides in m_ext. */
383 uint8_t m_epg_hdrlen;
384 uint8_t m_epg_trllen;
385 /* Offset into 1st page and length of
386 * data in the last page. */
387 uint16_t m_epg_1st_off;
388 uint16_t m_epg_last_len;
390 #define EPG_FLAG_ANON 0x1 /* Data can be encrypted in place. */
391 #define EPG_FLAG_2FREE 0x2 /* Scheduled for free. */
392 uint8_t m_epg_record_type;
395 struct ktls_session *m_epg_tls;
396 struct socket *m_epg_so;
397 uint64_t m_epg_seqno;
398 STAILQ_ENTRY(mbuf) m_epg_stailq;
402 /* M_EXT or M_EXTPG set. */
404 /* M_PKTHDR set, neither M_EXT nor M_EXTPG. */
408 char m_dat[0]; /* !M_PKTHDR, !M_EXT */
414 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff)
417 KASSERT(pgoff == 0 || pidx == 0,
418 ("page %d with non-zero offset %d in %p", pidx, pgoff, m));
420 if (pidx == m->m_epg_npgs - 1) {
421 return (m->m_epg_last_len);
423 return (PAGE_SIZE - pgoff);
428 #define MCHECK(ex, msg) KASSERT((ex), \
429 ("Multi page mbuf %p with " #msg " at %s:%d", \
430 m, __FILE__, __LINE__))
432 * NB: This expects a non-empty buffer (npgs > 0 and
435 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do { \
436 MCHECK(m->m_epg_npgs > 0, "no valid pages"); \
437 MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa), \
439 MCHECK(m->m_epg_nrdy <= m->m_epg_npgs, \
440 "too many ready pages"); \
441 MCHECK(m->m_epg_1st_off < PAGE_SIZE, \
442 "too large page offset"); \
443 MCHECK(m->m_epg_last_len > 0, "zero last page length"); \
444 MCHECK(m->m_epg_last_len <= PAGE_SIZE, \
445 "too large last page length"); \
446 if (m->m_epg_npgs == 1) \
447 MCHECK(m->m_epg_1st_off + \
448 m->m_epg_last_len <= PAGE_SIZE, \
449 "single page too large"); \
450 MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr), \
451 "too large header length"); \
452 MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail), \
453 "too large header length"); \
456 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do {} while (0)
461 * mbuf flags of global significance and layer crossing.
462 * Those of only protocol/layer specific significance are to be mapped
463 * to M_PROTO[1-11] and cleared at layer handoff boundaries.
464 * NB: Limited to the lower 24 bits.
466 #define M_EXT 0x00000001 /* has associated external storage */
467 #define M_PKTHDR 0x00000002 /* start of record */
468 #define M_EOR 0x00000004 /* end of record */
469 #define M_RDONLY 0x00000008 /* associated data is marked read-only */
470 #define M_BCAST 0x00000010 /* send/received as link-level broadcast */
471 #define M_MCAST 0x00000020 /* send/received as link-level multicast */
472 #define M_PROMISC 0x00000040 /* packet was not for us */
473 #define M_VLANTAG 0x00000080 /* ether_vtag is valid */
474 #define M_EXTPG 0x00000100 /* has array of unmapped pages and TLS */
475 #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */
476 #define M_TSTMP 0x00000400 /* rcv_tstmp field is valid */
477 #define M_TSTMP_HPREC 0x00000800 /* rcv_tstmp is high-prec, typically
478 hw-stamped on port (useful for IEEE 1588
480 #define M_TSTMP_LRO 0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
482 #define M_PROTO1 0x00002000 /* protocol-specific */
483 #define M_PROTO2 0x00004000 /* protocol-specific */
484 #define M_PROTO3 0x00008000 /* protocol-specific */
485 #define M_PROTO4 0x00010000 /* protocol-specific */
486 #define M_PROTO5 0x00020000 /* protocol-specific */
487 #define M_PROTO6 0x00040000 /* protocol-specific */
488 #define M_PROTO7 0x00080000 /* protocol-specific */
489 #define M_PROTO8 0x00100000 /* protocol-specific */
490 #define M_PROTO9 0x00200000 /* protocol-specific */
491 #define M_PROTO10 0x00400000 /* protocol-specific */
492 #define M_PROTO11 0x00800000 /* protocol-specific */
495 * Flags to purge when crossing layers.
497 #define M_PROTOFLAGS \
498 (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
499 M_PROTO9|M_PROTO10|M_PROTO11)
502 * Flags preserved when copying m_pkthdr.
504 #define M_COPYFLAGS \
505 (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
506 M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
509 * Flags preserved during demote.
511 #define M_DEMOTEFLAGS \
512 (M_EXT | M_RDONLY | M_NOFREE | M_EXTPG)
515 * Mbuf flag description for use with printf(9) %b identifier.
517 #define M_FLAG_BITS \
518 "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
519 "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
520 #define M_FLAG_PROTOBITS \
521 "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
522 "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
523 "\27M_PROTO10\28M_PROTO11"
524 #define M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
527 * Network interface cards are able to hash protocol fields (such as IPv4
528 * addresses and TCP port numbers) classify packets into flows. These flows
529 * can then be used to maintain ordering while delivering packets to the OS
530 * via parallel input queues, as well as to provide a stateless affinity
531 * model. NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
532 * m_flag fields to indicate how the hash should be interpreted by the
535 * Most NICs support RSS, which provides ordering and explicit affinity, and
536 * use the hash m_flag bits to indicate what header fields were covered by
537 * the hash. M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
538 * RSS cards or configurations that provide an opaque flow identifier, allowing
539 * for ordering and distribution without explicit affinity. Additionally,
540 * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
543 * The meaning of the IPV6_EX suffix:
544 * "o Home address from the home address option in the IPv6 destination
545 * options header. If the extension header is not present, use the Source
547 * o IPv6 address that is contained in the Routing-Header-Type-2 from the
548 * associated extension header. If the extension header is not present,
549 * use the Destination IPv6 Address."
551 * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
553 #define M_HASHTYPE_HASHPROP 0x80 /* has hash properties */
554 #define M_HASHTYPE_INNER 0x40 /* calculated from inner headers */
555 #define M_HASHTYPE_HASH(t) (M_HASHTYPE_HASHPROP | (t))
556 /* Microsoft RSS standard hash types */
557 #define M_HASHTYPE_NONE 0
558 #define M_HASHTYPE_RSS_IPV4 M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
559 #define M_HASHTYPE_RSS_TCP_IPV4 M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
560 #define M_HASHTYPE_RSS_IPV6 M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
561 #define M_HASHTYPE_RSS_TCP_IPV6 M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
562 #define M_HASHTYPE_RSS_IPV6_EX M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
564 #define M_HASHTYPE_RSS_TCP_IPV6_EX M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
566 #define M_HASHTYPE_RSS_UDP_IPV4 M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
567 #define M_HASHTYPE_RSS_UDP_IPV6 M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
568 #define M_HASHTYPE_RSS_UDP_IPV6_EX M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
571 #define M_HASHTYPE_OPAQUE 0x3f /* ordering, not affinity */
572 #define M_HASHTYPE_OPAQUE_HASH M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
573 /* ordering+hash, not affinity*/
575 #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0)
576 #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype & ~M_HASHTYPE_INNER)
577 #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v))
578 #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v))
579 #define M_HASHTYPE_ISHASH(m) \
580 (((m)->m_pkthdr.rsstype & M_HASHTYPE_HASHPROP) != 0)
581 #define M_HASHTYPE_SETINNER(m) do { \
582 (m)->m_pkthdr.rsstype |= M_HASHTYPE_INNER; \
586 * External mbuf storage buffer types.
588 #define EXT_CLUSTER 1 /* mbuf cluster */
589 #define EXT_SFBUF 2 /* sendfile(2)'s sf_buf */
590 #define EXT_JUMBOP 3 /* jumbo cluster page sized */
591 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */
592 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */
593 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */
594 #define EXT_MBUF 7 /* external mbuf reference */
595 #define EXT_RXRING 8 /* data in NIC receive ring */
597 #define EXT_VENDOR1 224 /* for vendor-internal use */
598 #define EXT_VENDOR2 225 /* for vendor-internal use */
599 #define EXT_VENDOR3 226 /* for vendor-internal use */
600 #define EXT_VENDOR4 227 /* for vendor-internal use */
602 #define EXT_EXP1 244 /* for experimental use */
603 #define EXT_EXP2 245 /* for experimental use */
604 #define EXT_EXP3 246 /* for experimental use */
605 #define EXT_EXP4 247 /* for experimental use */
607 #define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */
608 #define EXT_MOD_TYPE 253 /* custom module's ext_buf type */
609 #define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */
610 #define EXT_EXTREF 255 /* has externally maintained ext_cnt ptr */
613 * Flags for external mbuf buffer types.
614 * NB: limited to the lower 24 bits.
616 #define EXT_FLAG_EMBREF 0x000001 /* embedded ext_count */
617 #define EXT_FLAG_EXTREF 0x000002 /* external ext_cnt, notyet */
619 #define EXT_FLAG_NOFREE 0x000010 /* don't free mbuf to pool, notyet */
621 #define EXT_FLAG_VENDOR1 0x010000 /* These flags are vendor */
622 #define EXT_FLAG_VENDOR2 0x020000 /* or submodule specific, */
623 #define EXT_FLAG_VENDOR3 0x040000 /* not used by mbuf code. */
624 #define EXT_FLAG_VENDOR4 0x080000 /* Set/read by submodule. */
626 #define EXT_FLAG_EXP1 0x100000 /* for experimental use */
627 #define EXT_FLAG_EXP2 0x200000 /* for experimental use */
628 #define EXT_FLAG_EXP3 0x400000 /* for experimental use */
629 #define EXT_FLAG_EXP4 0x800000 /* for experimental use */
632 * EXT flag description for use with printf(9) %b identifier.
634 #define EXT_FLAG_BITS \
635 "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
636 "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
637 "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
641 * Flags indicating checksum, segmentation and other offload work to be
642 * done, or already done, by hardware or lower layers. It is split into
643 * separate inbound and outbound flags.
645 * Outbound flags that are set by upper protocol layers requesting lower
646 * layers, or ideally the hardware, to perform these offloading tasks.
647 * For outbound packets this field and its flags can be directly tested
648 * against ifnet if_hwassist. Note that the outbound and the inbound flags do
649 * not collide right now but they could be allowed to (as long as the flags are
650 * scrubbed appropriately when the direction of an mbuf changes). CSUM_BITS
651 * would also have to split into CSUM_BITS_TX and CSUM_BITS_RX.
653 * CSUM_INNER_<x> is the same as CSUM_<x> but it applies to the inner frame.
654 * The CSUM_ENCAP_<x> bits identify the outer encapsulation.
656 #define CSUM_IP 0x00000001 /* IP header checksum offload */
657 #define CSUM_IP_UDP 0x00000002 /* UDP checksum offload */
658 #define CSUM_IP_TCP 0x00000004 /* TCP checksum offload */
659 #define CSUM_IP_SCTP 0x00000008 /* SCTP checksum offload */
660 #define CSUM_IP_TSO 0x00000010 /* TCP segmentation offload */
661 #define CSUM_IP_ISCSI 0x00000020 /* iSCSI checksum offload */
663 #define CSUM_INNER_IP6_UDP 0x00000040
664 #define CSUM_INNER_IP6_TCP 0x00000080
665 #define CSUM_INNER_IP6_TSO 0x00000100
666 #define CSUM_IP6_UDP 0x00000200 /* UDP checksum offload */
667 #define CSUM_IP6_TCP 0x00000400 /* TCP checksum offload */
668 #define CSUM_IP6_SCTP 0x00000800 /* SCTP checksum offload */
669 #define CSUM_IP6_TSO 0x00001000 /* TCP segmentation offload */
670 #define CSUM_IP6_ISCSI 0x00002000 /* iSCSI checksum offload */
672 #define CSUM_INNER_IP 0x00004000
673 #define CSUM_INNER_IP_UDP 0x00008000
674 #define CSUM_INNER_IP_TCP 0x00010000
675 #define CSUM_INNER_IP_TSO 0x00020000
677 #define CSUM_ENCAP_VXLAN 0x00040000 /* VXLAN outer encapsulation */
678 #define CSUM_ENCAP_RSVD1 0x00080000
680 /* Inbound checksum support where the checksum was verified by hardware. */
681 #define CSUM_INNER_L3_CALC 0x00100000
682 #define CSUM_INNER_L3_VALID 0x00200000
683 #define CSUM_INNER_L4_CALC 0x00400000
684 #define CSUM_INNER_L4_VALID 0x00800000
685 #define CSUM_L3_CALC 0x01000000 /* calculated layer 3 csum */
686 #define CSUM_L3_VALID 0x02000000 /* checksum is correct */
687 #define CSUM_L4_CALC 0x04000000 /* calculated layer 4 csum */
688 #define CSUM_L4_VALID 0x08000000 /* checksum is correct */
689 #define CSUM_L5_CALC 0x10000000 /* calculated layer 5 csum */
690 #define CSUM_L5_VALID 0x20000000 /* checksum is correct */
691 #define CSUM_COALESCED 0x40000000 /* contains merged segments */
693 #define CSUM_SND_TAG 0x80000000 /* Packet header has send tag */
695 #define CSUM_FLAGS_TX (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP | CSUM_IP_SCTP | \
696 CSUM_IP_TSO | CSUM_IP_ISCSI | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | \
697 CSUM_INNER_IP6_TSO | CSUM_IP6_UDP | CSUM_IP6_TCP | CSUM_IP6_SCTP | \
698 CSUM_IP6_TSO | CSUM_IP6_ISCSI | CSUM_INNER_IP | CSUM_INNER_IP_UDP | \
699 CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN | \
700 CSUM_ENCAP_RSVD1 | CSUM_SND_TAG)
702 #define CSUM_FLAGS_RX (CSUM_INNER_L3_CALC | CSUM_INNER_L3_VALID | \
703 CSUM_INNER_L4_CALC | CSUM_INNER_L4_VALID | CSUM_L3_CALC | CSUM_L3_VALID | \
704 CSUM_L4_CALC | CSUM_L4_VALID | CSUM_L5_CALC | CSUM_L5_VALID | \
708 * CSUM flag description for use with printf(9) %b identifier.
711 "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
712 "\6CSUM_IP_ISCSI\7CSUM_INNER_IP6_UDP\10CSUM_INNER_IP6_TCP" \
713 "\11CSUM_INNER_IP6_TSO\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP" \
714 "\15CSUM_IP6_TSO\16CSUM_IP6_ISCSI\17CSUM_INNER_IP\20CSUM_INNER_IP_UDP" \
715 "\21CSUM_INNER_IP_TCP\22CSUM_INNER_IP_TSO\23CSUM_ENCAP_VXLAN" \
716 "\24CSUM_ENCAP_RSVD1\25CSUM_INNER_L3_CALC\26CSUM_INNER_L3_VALID" \
717 "\27CSUM_INNER_L4_CALC\30CSUM_INNER_L4_VALID\31CSUM_L3_CALC" \
718 "\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID\35CSUM_L5_CALC" \
719 "\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
721 /* CSUM flags compatibility mappings. */
722 #define CSUM_IP_CHECKED CSUM_L3_CALC
723 #define CSUM_IP_VALID CSUM_L3_VALID
724 #define CSUM_DATA_VALID CSUM_L4_VALID
725 #define CSUM_PSEUDO_HDR CSUM_L4_CALC
726 #define CSUM_SCTP_VALID CSUM_L4_VALID
727 #define CSUM_DELAY_DATA (CSUM_TCP|CSUM_UDP)
728 #define CSUM_DELAY_IP CSUM_IP /* Only v4, no v6 IP hdr csum */
729 #define CSUM_DELAY_DATA_IPV6 (CSUM_TCP_IPV6|CSUM_UDP_IPV6)
730 #define CSUM_DATA_VALID_IPV6 CSUM_DATA_VALID
731 #define CSUM_TCP CSUM_IP_TCP
732 #define CSUM_UDP CSUM_IP_UDP
733 #define CSUM_SCTP CSUM_IP_SCTP
734 #define CSUM_TSO (CSUM_IP_TSO|CSUM_IP6_TSO)
735 #define CSUM_INNER_TSO (CSUM_INNER_IP_TSO|CSUM_INNER_IP6_TSO)
736 #define CSUM_UDP_IPV6 CSUM_IP6_UDP
737 #define CSUM_TCP_IPV6 CSUM_IP6_TCP
738 #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP
739 #define CSUM_TLS_MASK (CSUM_L5_CALC|CSUM_L5_VALID)
740 #define CSUM_TLS_DECRYPTED CSUM_L5_CALC
743 * mbuf types describing the content of the mbuf (including external storage).
745 #define MT_NOTMBUF 0 /* USED INTERNALLY ONLY! Object is not mbuf */
746 #define MT_DATA 1 /* dynamic (data) allocation */
747 #define MT_HEADER MT_DATA /* packet header, use M_PKTHDR instead */
749 #define MT_VENDOR1 4 /* for vendor-internal use */
750 #define MT_VENDOR2 5 /* for vendor-internal use */
751 #define MT_VENDOR3 6 /* for vendor-internal use */
752 #define MT_VENDOR4 7 /* for vendor-internal use */
754 #define MT_SONAME 8 /* socket name */
756 #define MT_EXP1 9 /* for experimental use */
757 #define MT_EXP2 10 /* for experimental use */
758 #define MT_EXP3 11 /* for experimental use */
759 #define MT_EXP4 12 /* for experimental use */
761 #define MT_CONTROL 14 /* extra-data protocol message */
762 #define MT_EXTCONTROL 15 /* control message with externalized contents */
763 #define MT_OOBDATA 16 /* expedited data */
765 #define MT_NOINIT 255 /* Not a type but a flag to allocate
766 a non-initialized mbuf */
769 * String names of mbuf-related UMA(9) and malloc(9) types. Exposed to
770 * !_KERNEL so that monitoring tools can look up the zones with
773 #define MBUF_MEM_NAME "mbuf"
774 #define MBUF_CLUSTER_MEM_NAME "mbuf_cluster"
775 #define MBUF_PACKET_MEM_NAME "mbuf_packet"
776 #define MBUF_JUMBOP_MEM_NAME "mbuf_jumbo_page"
777 #define MBUF_JUMBO9_MEM_NAME "mbuf_jumbo_9k"
778 #define MBUF_JUMBO16_MEM_NAME "mbuf_jumbo_16k"
779 #define MBUF_TAG_MEM_NAME "mbuf_tag"
780 #define MBUF_EXTREFCNT_MEM_NAME "mbuf_ext_refcnt"
781 #define MBUF_EXTPGS_MEM_NAME "mbuf_extpgs"
784 union if_snd_tag_alloc_params;
786 #define MBUF_CHECKSLEEP(how) do { \
787 if (how == M_WAITOK) \
788 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \
789 "Sleeping in \"%s\"", __func__); \
793 * Network buffer allocation API
795 * The rest of it is defined in kern/kern_mbuf.c
797 extern uma_zone_t zone_mbuf;
798 extern uma_zone_t zone_clust;
799 extern uma_zone_t zone_pack;
800 extern uma_zone_t zone_jumbop;
801 extern uma_zone_t zone_jumbo9;
802 extern uma_zone_t zone_jumbo16;
803 extern uma_zone_t zone_extpgs;
805 void mb_dupcl(struct mbuf *, struct mbuf *);
806 void mb_free_ext(struct mbuf *);
807 void mb_free_extpg(struct mbuf *);
808 void mb_free_mext_pgs(struct mbuf *);
809 struct mbuf *mb_alloc_ext_pgs(int, m_ext_free_t);
810 struct mbuf *mb_alloc_ext_plus_pages(int, int);
811 struct mbuf *mb_mapped_to_unmapped(struct mbuf *, int, int, int,
813 int mb_unmapped_compress(struct mbuf *m);
814 struct mbuf *mb_unmapped_to_ext(struct mbuf *m);
815 void mb_free_notready(struct mbuf *m, int count);
816 void m_adj(struct mbuf *, int);
817 void m_adj_decap(struct mbuf *, int);
818 int m_apply(struct mbuf *, int, int,
819 int (*)(void *, void *, u_int), void *);
820 int m_append(struct mbuf *, int, c_caddr_t);
821 void m_cat(struct mbuf *, struct mbuf *);
822 void m_catpkt(struct mbuf *, struct mbuf *);
823 int m_clget(struct mbuf *m, int how);
824 void *m_cljget(struct mbuf *m, int how, int size);
825 struct mbuf *m_collapse(struct mbuf *, int, int);
826 void m_copyback(struct mbuf *, int, int, c_caddr_t);
827 void m_copydata(const struct mbuf *, int, int, caddr_t);
828 struct mbuf *m_copym(struct mbuf *, int, int, int);
829 struct mbuf *m_copypacket(struct mbuf *, int);
830 void m_copy_pkthdr(struct mbuf *, struct mbuf *);
831 struct mbuf *m_copyup(struct mbuf *, int, int);
832 struct mbuf *m_defrag(struct mbuf *, int);
833 void m_demote_pkthdr(struct mbuf *);
834 void m_demote(struct mbuf *, int, int);
835 struct mbuf *m_devget(char *, int, int, struct ifnet *,
836 void (*)(char *, caddr_t, u_int));
837 void m_dispose_extcontrolm(struct mbuf *m);
838 struct mbuf *m_dup(const struct mbuf *, int);
839 int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
840 void m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
841 void *, void *, int, int);
842 u_int m_fixhdr(struct mbuf *);
843 struct mbuf *m_fragment(struct mbuf *, int, int);
844 void m_freem(struct mbuf *);
845 void m_free_raw(struct mbuf *);
846 struct mbuf *m_get2(int, int, short, int);
847 struct mbuf *m_get3(int, int, short, int);
848 struct mbuf *m_getjcl(int, short, int, int);
849 struct mbuf *m_getm2(struct mbuf *, int, int, short, int);
850 struct mbuf *m_getptr(struct mbuf *, int, int *);
851 u_int m_length(struct mbuf *, struct mbuf **);
852 int m_mbuftouio(struct uio *, const struct mbuf *, int);
853 void m_move_pkthdr(struct mbuf *, struct mbuf *);
854 int m_pkthdr_init(struct mbuf *, int);
855 struct mbuf *m_prepend(struct mbuf *, int, int);
856 void m_print(const struct mbuf *, int);
857 struct mbuf *m_pulldown(struct mbuf *, int, int, int *);
858 struct mbuf *m_pullup(struct mbuf *, int);
859 int m_sanity(struct mbuf *, int);
860 struct mbuf *m_split(struct mbuf *, int, int);
861 struct mbuf *m_uiotombuf(struct uio *, int, int, int, int);
862 int m_unmapped_uiomove(const struct mbuf *, int, struct uio *,
864 struct mbuf *m_unshare(struct mbuf *, int);
865 int m_snd_tag_alloc(struct ifnet *,
866 union if_snd_tag_alloc_params *, struct m_snd_tag **);
867 void m_snd_tag_init(struct m_snd_tag *, struct ifnet *,
868 const struct if_snd_tag_sw *);
869 void m_snd_tag_destroy(struct m_snd_tag *);
870 void m_rcvif_serialize(struct mbuf *);
871 struct ifnet *m_rcvif_restore(struct mbuf *);
895 panic("%s: invalid cluster size %d", __func__, size);
902 * Associated an external reference counted buffer with an mbuf.
905 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
906 m_ext_free_t freef, void *arg1, void *arg2)
909 KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
911 atomic_add_int(ref_cnt, 1);
913 m->m_ext.ext_buf = buf;
914 m->m_ext.ext_cnt = ref_cnt;
915 m->m_data = m->m_ext.ext_buf;
916 m->m_ext.ext_size = size;
917 m->m_ext.ext_free = freef;
918 m->m_ext.ext_arg1 = arg1;
919 m->m_ext.ext_arg2 = arg2;
920 m->m_ext.ext_type = EXT_EXTREF;
921 m->m_ext.ext_flags = 0;
924 static __inline uma_zone_t
943 panic("%s: invalid cluster size %d", __func__, size);
950 * Initialize an mbuf with linear storage.
952 * Inline because the consumer text overhead will be roughly the same to
953 * initialize or call a function with this many parameters and M_PKTHDR
954 * should go away with constant propagation for !MGETHDR.
957 m_init(struct mbuf *m, int how, short type, int flags)
963 m->m_data = m->m_dat;
967 if (flags & M_PKTHDR)
968 error = m_pkthdr_init(m, how);
972 MBUF_PROBE5(m__init, m, how, type, flags, error);
976 static __inline struct mbuf *
977 m_get_raw(int how, short type)
983 args.type = type | MT_NOINIT;
984 m = uma_zalloc_arg(zone_mbuf, &args, how);
985 MBUF_PROBE3(m__get_raw, how, type, m);
989 static __inline struct mbuf *
990 m_get(int how, short type)
997 m = uma_zalloc_arg(zone_mbuf, &args, how);
998 MBUF_PROBE3(m__get, how, type, m);
1002 static __inline struct mbuf *
1003 m_gethdr_raw(int how, short type)
1006 struct mb_args args;
1008 args.flags = M_PKTHDR;
1009 args.type = type | MT_NOINIT;
1010 m = uma_zalloc_arg(zone_mbuf, &args, how);
1011 MBUF_PROBE3(m__gethdr_raw, how, type, m);
1015 static __inline struct mbuf *
1016 m_gethdr(int how, short type)
1019 struct mb_args args;
1021 args.flags = M_PKTHDR;
1023 m = uma_zalloc_arg(zone_mbuf, &args, how);
1024 MBUF_PROBE3(m__gethdr, how, type, m);
1028 static __inline struct mbuf *
1029 m_getcl(int how, short type, int flags)
1032 struct mb_args args;
1036 m = uma_zalloc_arg(zone_pack, &args, how);
1037 MBUF_PROBE4(m__getcl, how, type, flags, m);
1042 * XXX: m_cljset() is a dangerous API. One must attach only a new,
1043 * unreferenced cluster to an mbuf(9). It is not possible to assert
1044 * that, so care can be taken only by users of the API.
1046 static __inline void
1047 m_cljset(struct mbuf *m, void *cl, int type)
1056 size = MJUMPAGESIZE;
1065 panic("%s: unknown cluster type %d", __func__, type);
1069 m->m_data = m->m_ext.ext_buf = cl;
1070 m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
1071 m->m_ext.ext_size = size;
1072 m->m_ext.ext_type = type;
1073 m->m_ext.ext_flags = EXT_FLAG_EMBREF;
1074 m->m_ext.ext_count = 1;
1075 m->m_flags |= M_EXT;
1076 MBUF_PROBE3(m__cljset, m, cl, type);
1079 static __inline void
1080 m_chtype(struct mbuf *m, short new_type)
1083 m->m_type = new_type;
1086 static __inline void
1087 m_clrprotoflags(struct mbuf *m)
1091 m->m_flags &= ~M_PROTOFLAGS;
1096 static __inline struct mbuf *
1097 m_last(struct mbuf *m)
1106 m_extrefcnt(struct mbuf *m)
1109 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__));
1111 return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
1116 * mbuf, cluster, and external object allocation macros (for compatibility
1119 #define M_MOVE_PKTHDR(to, from) m_move_pkthdr((to), (from))
1120 #define MGET(m, how, type) ((m) = m_get((how), (type)))
1121 #define MGETHDR(m, how, type) ((m) = m_gethdr((how), (type)))
1122 #define MCLGET(m, how) m_clget((m), (how))
1123 #define MEXTADD(m, buf, size, free, arg1, arg2, flags, type) \
1124 m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2), \
1126 #define m_getm(m, len, how, type) \
1127 m_getm2((m), (len), (how), (type), M_PKTHDR)
1130 * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
1131 * be both the local data payload, or an external buffer area, depending on
1132 * whether M_EXT is set).
1134 #define M_WRITABLE(m) (((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 && \
1135 (!(((m)->m_flags & M_EXT)) || \
1136 (m_extrefcnt(m) == 1)))
1138 /* Check if the supplied mbuf has a packet header, or else panic. */
1139 #define M_ASSERTPKTHDR(m) \
1140 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR, \
1141 ("%s: no mbuf packet header!", __func__))
1143 /* Check if the supplied mbuf has no send tag, or else panic. */
1144 #define M_ASSERT_NO_SND_TAG(m) \
1145 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR && \
1146 ((m)->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0, \
1147 ("%s: receive mbuf has send tag!", __func__))
1149 /* Check if mbuf is multipage. */
1150 #define M_ASSERTEXTPG(m) \
1151 KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG, \
1152 ("%s: m %p is not multipage!", __func__, m))
1155 * Ensure that the supplied mbuf is a valid, non-free mbuf.
1157 * XXX: Broken at the moment. Need some UMA magic to make it work again.
1159 #define M_ASSERTVALID(m) \
1160 KASSERT((((struct mbuf *)m)->m_flags & 0) == 0, \
1161 ("%s: attempted use of a free mbuf!", __func__))
1163 /* Check whether any mbuf in the chain is unmapped. */
1165 #define M_ASSERTMAPPED(m) do { \
1166 for (struct mbuf *__m = (m); __m != NULL; __m = __m->m_next) \
1167 KASSERT((__m->m_flags & M_EXTPG) == 0, \
1168 ("%s: chain %p contains an unmapped mbuf", __func__, (m)));\
1171 #define M_ASSERTMAPPED(m) do {} while (0)
1175 * Return the address of the start of the buffer associated with an mbuf,
1176 * handling external storage, packet-header mbufs, and regular data mbufs.
1178 #define M_START(m) \
1179 (((m)->m_flags & M_EXTPG) ? NULL : \
1180 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf : \
1181 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] : \
1185 * Return the size of the buffer associated with an mbuf, handling external
1186 * storage, packet-header mbufs, and regular data mbufs.
1189 (((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size : \
1190 ((m)->m_flags & M_PKTHDR) ? MHLEN : \
1194 * Set the m_data pointer of a newly allocated mbuf to place an object of the
1195 * specified size at the end of the mbuf, longword aligned.
1197 * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1198 * separate macros, each asserting that it was called at the proper moment.
1199 * This required callers to themselves test the storage type and call the
1200 * right one. Rather than require callers to be aware of those layout
1201 * decisions, we centralize here.
1203 static __inline void
1204 m_align(struct mbuf *m, int len)
1207 const char *msg = "%s: not a virgin mbuf";
1211 KASSERT(m->m_data == M_START(m), (msg, __func__));
1213 adjust = M_SIZE(m) - len;
1214 m->m_data += adjust &~ (sizeof(long)-1);
1217 #define M_ALIGN(m, len) m_align(m, len)
1218 #define MH_ALIGN(m, len) m_align(m, len)
1219 #define MEXT_ALIGN(m, len) m_align(m, len)
1222 * Compute the amount of space available before the current start of data in
1225 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1226 * of checking writability of the mbuf data area rests solely with the caller.
1228 * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1229 * for mbufs with external storage. We now allow mbuf-embedded data to be
1230 * read-only as well.
1232 #define M_LEADINGSPACE(m) \
1233 (M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1236 * So M_TRAILINGROOM() is for when you want to know how much space
1237 * would be there if it was writable. This can be used to
1238 * detect changes in mbufs by knowing the value at one point
1239 * and then being able to compare it later to the current M_TRAILINGROOM().
1240 * The TRAILINGSPACE() macro is not suitable for this since an mbuf
1241 * at one point might not be writable and then later it becomes writable
1242 * even though the space at the back of it has not changed.
1244 #define M_TRAILINGROOM(m) ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len))
1246 * Compute the amount of space available after the end of data in an mbuf.
1248 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1249 * of checking writability of the mbuf data area rests solely with the caller.
1251 * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1252 * for mbufs with external storage. We now allow mbuf-embedded data to be
1253 * read-only as well.
1255 #define M_TRAILINGSPACE(m) (M_WRITABLE(m) ? M_TRAILINGROOM(m) : 0)
1258 * Arrange to prepend space of size plen to mbuf m. If a new mbuf must be
1259 * allocated, how specifies whether to wait. If the allocation fails, the
1260 * original mbuf chain is freed and m is set to NULL.
1262 #define M_PREPEND(m, plen, how) do { \
1263 struct mbuf **_mmp = &(m); \
1264 struct mbuf *_mm = *_mmp; \
1265 int _mplen = (plen); \
1266 int __mhow = (how); \
1268 MBUF_CHECKSLEEP(how); \
1269 if (M_LEADINGSPACE(_mm) >= _mplen) { \
1270 _mm->m_data -= _mplen; \
1271 _mm->m_len += _mplen; \
1273 _mm = m_prepend(_mm, _mplen, __mhow); \
1274 if (_mm != NULL && _mm->m_flags & M_PKTHDR) \
1275 _mm->m_pkthdr.len += _mplen; \
1280 * Change mbuf to new type. This is a relatively expensive operation and
1281 * should be avoided.
1283 #define MCHTYPE(m, t) m_chtype((m), (t))
1285 /* Return the rcvif of a packet header. */
1286 static __inline struct ifnet *
1287 m_rcvif(struct mbuf *m)
1291 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1293 return (m->m_pkthdr.rcvif);
1296 /* Length to m_copy to copy all. */
1297 #define M_COPYALL 1000000000
1299 extern u_int max_linkhdr; /* Largest link-level header */
1300 extern u_int max_hdr; /* Largest link + protocol header */
1301 extern u_int max_protohdr; /* Largest protocol header */
1302 void max_linkhdr_grow(u_int);
1303 void max_protohdr_grow(u_int);
1305 extern int nmbclusters; /* Maximum number of clusters */
1306 extern bool mb_use_ext_pgs; /* Use ext_pgs for sendfile */
1309 * Network packets may have annotations attached by affixing a list of
1310 * "packet tags" to the pkthdr structure. Packet tags are dynamically
1311 * allocated semi-opaque data structures that have a fixed header
1312 * (struct m_tag) that specifies the size of the memory block and a
1313 * <cookie,type> pair that identifies it. The cookie is a 32-bit unique
1314 * unsigned value used to identify a module or ABI. By convention this value
1315 * is chosen as the date+time that the module is created, expressed as the
1316 * number of seconds since the epoch (e.g., using date -u +'%s'). The type
1317 * value is an ABI/module-specific value that identifies a particular
1318 * annotation and is private to the module. For compatibility with systems
1319 * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1320 * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1321 * compatibility shim functions and several tag types are defined below.
1322 * Users that do not require compatibility should use a private cookie value
1323 * so that packet tag-related definitions can be maintained privately.
1325 * Note that the packet tag returned by m_tag_alloc has the default memory
1326 * alignment implemented by malloc. To reference private data one can use a
1329 * struct m_tag *mtag = m_tag_alloc(...);
1330 * struct foo *p = (struct foo *)(mtag+1);
1332 * if the alignment of struct m_tag is sufficient for referencing members of
1333 * struct foo. Otherwise it is necessary to embed struct m_tag within the
1334 * private data structure to insure proper alignment; e.g.,
1340 * struct foo *p = (struct foo *) m_tag_alloc(...);
1341 * struct m_tag *mtag = &p->tag;
1345 * Persistent tags stay with an mbuf until the mbuf is reclaimed. Otherwise
1346 * tags are expected to ``vanish'' when they pass through a network
1347 * interface. For most interfaces this happens normally as the tags are
1348 * reclaimed when the mbuf is free'd. However in some special cases
1349 * reclaiming must be done manually. An example is packets that pass through
1350 * the loopback interface. Also, one must be careful to do this when
1351 * ``turning around'' packets (e.g., icmp_reflect).
1353 * To mark a tag persistent bit-or this flag in when defining the tag id.
1354 * The tag will then be treated as described above.
1356 #define MTAG_PERSISTENT 0x800
1358 #define PACKET_TAG_NONE 0 /* Nadda */
1360 /* Packet tags for use with PACKET_ABI_COMPAT. */
1361 #define PACKET_TAG_IPSEC_IN_DONE 1 /* IPsec applied, in */
1362 #define PACKET_TAG_IPSEC_OUT_DONE 2 /* IPsec applied, out */
1363 #define PACKET_TAG_IPSEC_IN_CRYPTO_DONE 3 /* NIC IPsec crypto done */
1364 #define PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED 4 /* NIC IPsec crypto req'ed */
1365 #define PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO 5 /* NIC notifies IPsec */
1366 #define PACKET_TAG_IPSEC_PENDING_TDB 6 /* Reminder to do IPsec */
1367 #define PACKET_TAG_BRIDGE 7 /* Bridge processing done */
1368 #define PACKET_TAG_GIF 8 /* GIF processing done */
1369 #define PACKET_TAG_GRE 9 /* GRE processing done */
1370 #define PACKET_TAG_IN_PACKET_CHECKSUM 10 /* NIC checksumming done */
1371 #define PACKET_TAG_ENCAP 11 /* Encap. processing */
1372 #define PACKET_TAG_IPSEC_SOCKET 12 /* IPSEC socket ref */
1373 #define PACKET_TAG_IPSEC_HISTORY 13 /* IPSEC history */
1374 #define PACKET_TAG_IPV6_INPUT 14 /* IPV6 input processing */
1375 #define PACKET_TAG_DUMMYNET 15 /* dummynet info */
1376 #define PACKET_TAG_DIVERT 17 /* divert info */
1377 #define PACKET_TAG_IPFORWARD 18 /* ipforward info */
1378 #define PACKET_TAG_MACLABEL (19 | MTAG_PERSISTENT) /* MAC label */
1379 #define PACKET_TAG_PF 21 /* PF/ALTQ information */
1380 /* was PACKET_TAG_RTSOCKFAM 25 rtsock sa family */
1381 #define PACKET_TAG_IPOPTIONS 27 /* Saved IP options */
1382 #define PACKET_TAG_CARP 28 /* CARP info */
1383 #define PACKET_TAG_IPSEC_NAT_T_PORTS 29 /* two uint16_t */
1384 #define PACKET_TAG_ND_OUTGOING 30 /* ND outgoing */
1385 #define PACKET_TAG_PF_REASSEMBLED 31
1387 /* Specific cookies and tags. */
1389 /* Packet tag routines. */
1390 struct m_tag *m_tag_alloc(uint32_t, uint16_t, int, int);
1391 void m_tag_delete(struct mbuf *, struct m_tag *);
1392 void m_tag_delete_chain(struct mbuf *, struct m_tag *);
1393 void m_tag_free_default(struct m_tag *);
1394 struct m_tag *m_tag_locate(struct mbuf *, uint32_t, uint16_t,
1396 struct m_tag *m_tag_copy(struct m_tag *, int);
1397 int m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1398 void m_tag_delete_nonpersistent(struct mbuf *);
1401 * Initialize the list of tags associated with an mbuf.
1403 static __inline void
1404 m_tag_init(struct mbuf *m)
1407 SLIST_INIT(&m->m_pkthdr.tags);
1411 * Set up the contents of a tag. Note that this does not fill in the free
1412 * method; the caller is expected to do that.
1414 * XXX probably should be called m_tag_init, but that was already taken.
1416 static __inline void
1417 m_tag_setup(struct m_tag *t, uint32_t cookie, uint16_t type, int len)
1422 t->m_tag_cookie = cookie;
1426 * Reclaim resources associated with a tag.
1428 static __inline void
1429 m_tag_free(struct m_tag *t)
1432 (*t->m_tag_free)(t);
1436 * Return the first tag associated with an mbuf.
1438 static __inline struct m_tag *
1439 m_tag_first(struct mbuf *m)
1442 return (SLIST_FIRST(&m->m_pkthdr.tags));
1446 * Return the next tag in the list of tags associated with an mbuf.
1448 static __inline struct m_tag *
1449 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1452 return (SLIST_NEXT(t, m_tag_link));
1456 * Prepend a tag to the list of tags associated with an mbuf.
1458 static __inline void
1459 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1462 SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1466 * Unlink a tag from the list of tags associated with an mbuf.
1468 static __inline void
1469 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1472 SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1475 /* These are for OpenBSD compatibility. */
1476 #define MTAG_ABI_COMPAT 0 /* compatibility ABI */
1478 static __inline struct m_tag *
1479 m_tag_get(uint16_t type, int length, int wait)
1481 return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1484 static __inline struct m_tag *
1485 m_tag_find(struct mbuf *m, uint16_t type, struct m_tag *start)
1487 return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1488 m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1491 static inline struct m_snd_tag *
1492 m_snd_tag_ref(struct m_snd_tag *mst)
1495 refcount_acquire(&mst->refcount);
1500 m_snd_tag_rele(struct m_snd_tag *mst)
1503 if (refcount_release(&mst->refcount))
1504 m_snd_tag_destroy(mst);
1507 static __inline struct mbuf *
1508 m_free(struct mbuf *m)
1510 struct mbuf *n = m->m_next;
1512 MBUF_PROBE1(m__free, m);
1513 if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1514 m_tag_delete_chain(m, NULL);
1515 if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1516 m_snd_tag_rele(m->m_pkthdr.snd_tag);
1517 if (m->m_flags & M_EXTPG)
1519 else if (m->m_flags & M_EXT)
1521 else if ((m->m_flags & M_NOFREE) == 0)
1522 uma_zfree(zone_mbuf, m);
1527 rt_m_getfib(struct mbuf *m)
1529 KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf."));
1530 return (m->m_pkthdr.fibnum);
1533 #define M_GETFIB(_m) rt_m_getfib(_m)
1535 #define M_SETFIB(_m, _fib) do { \
1536 KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf.")); \
1537 ((_m)->m_pkthdr.fibnum) = (_fib); \
1540 /* flags passed as first argument for "m_xxx_tcpip_hash()" */
1541 #define MBUF_HASHFLAG_L2 (1 << 2)
1542 #define MBUF_HASHFLAG_L3 (1 << 3)
1543 #define MBUF_HASHFLAG_L4 (1 << 4)
1545 /* mbuf hashing helper routines */
1546 uint32_t m_ether_tcpip_hash_init(void);
1547 uint32_t m_ether_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1548 uint32_t m_infiniband_tcpip_hash_init(void);
1549 uint32_t m_infiniband_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1551 #ifdef MBUF_PROFILING
1552 void m_profile(struct mbuf *m);
1553 #define M_PROFILE(m) m_profile(m)
1555 #define M_PROFILE(m)
1559 STAILQ_HEAD(, mbuf) mq_head;
1565 mbufq_init(struct mbufq *mq, int maxlen)
1568 STAILQ_INIT(&mq->mq_head);
1569 mq->mq_maxlen = maxlen;
1573 static inline struct mbuf *
1574 mbufq_flush(struct mbufq *mq)
1578 m = STAILQ_FIRST(&mq->mq_head);
1579 STAILQ_INIT(&mq->mq_head);
1585 mbufq_drain(struct mbufq *mq)
1589 n = mbufq_flush(mq);
1590 while ((m = n) != NULL) {
1591 n = STAILQ_NEXT(m, m_stailqpkt);
1596 static inline struct mbuf *
1597 mbufq_first(const struct mbufq *mq)
1600 return (STAILQ_FIRST(&mq->mq_head));
1603 static inline struct mbuf *
1604 mbufq_last(const struct mbufq *mq)
1607 return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1611 mbufq_full(const struct mbufq *mq)
1614 return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1618 mbufq_len(const struct mbufq *mq)
1621 return (mq->mq_len);
1625 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1630 STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1635 static inline struct mbuf *
1636 mbufq_dequeue(struct mbufq *mq)
1640 m = STAILQ_FIRST(&mq->mq_head);
1642 STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1643 m->m_nextpkt = NULL;
1650 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1653 STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1658 * Note: this doesn't enforce the maximum list size for dst.
1661 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1664 mq_dst->mq_len += mq_src->mq_len;
1665 STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1669 #ifdef _SYS_TIMESPEC_H_
1671 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1674 KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1675 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1676 ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1677 ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1678 ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1683 mbuf_tstmp2timeval(struct mbuf *m, struct timeval *tv)
1686 KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m));
1687 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1688 ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m));
1689 tv->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1690 tv->tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000) / 1000;
1694 /* Invoked from the debugnet client code. */
1695 void debugnet_mbuf_drain(void);
1696 void debugnet_mbuf_start(void);
1697 void debugnet_mbuf_finish(void);
1698 void debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1702 mbuf_has_tls_session(struct mbuf *m)
1705 if (m->m_flags & M_EXTPG) {
1706 if (m->m_epg_tls != NULL) {
1713 #endif /* _KERNEL */
1714 #endif /* !_SYS_MBUF_H_ */