1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 #ifndef _SYS_MBUF_H_
34 #define _SYS_MBUF_H_
35
36 /* XXX: These includes suck. Sorry! */
37 #include <sys/queue.h>
38 #ifdef _KERNEL
39 #include <sys/systm.h>
40 #include <sys/refcount.h>
41 #include <vm/uma.h>
42
43 #include <sys/sdt.h>
44
45 #define MBUF_PROBE1(probe, arg0) \
46 SDT_PROBE1(sdt, , , probe, arg0)
47 #define MBUF_PROBE2(probe, arg0, arg1) \
48 SDT_PROBE2(sdt, , , probe, arg0, arg1)
49 #define MBUF_PROBE3(probe, arg0, arg1, arg2) \
50 SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2)
51 #define MBUF_PROBE4(probe, arg0, arg1, arg2, arg3) \
52 SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3)
53 #define MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4) \
54 SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4)
55
56 SDT_PROBE_DECLARE(sdt, , , m__init);
57 SDT_PROBE_DECLARE(sdt, , , m__gethdr_raw);
58 SDT_PROBE_DECLARE(sdt, , , m__gethdr);
59 SDT_PROBE_DECLARE(sdt, , , m__get_raw);
60 SDT_PROBE_DECLARE(sdt, , , m__get);
61 SDT_PROBE_DECLARE(sdt, , , m__getcl);
62 SDT_PROBE_DECLARE(sdt, , , m__getjcl);
63 SDT_PROBE_DECLARE(sdt, , , m__clget);
64 SDT_PROBE_DECLARE(sdt, , , m__cljget);
65 SDT_PROBE_DECLARE(sdt, , , m__cljset);
66 SDT_PROBE_DECLARE(sdt, , , m__free);
67 SDT_PROBE_DECLARE(sdt, , , m__freem);
68 SDT_PROBE_DECLARE(sdt, , , m__freemp);
69
70 #endif /* _KERNEL */
71
72 /*
73 * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead.
74 * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in
75 * sys/param.h), which has no additional overhead and is used instead of the
76 * internal data area; this is done when at least MINCLSIZE of data must be
77 * stored. Additionally, it is possible to allocate a separate buffer
78 * externally and attach it to the mbuf in a way similar to that of mbuf
79 * clusters.
80 *
81 * NB: These calculation do not take actual compiler-induced alignment and
82 * padding inside the complete struct mbuf into account. Appropriate
83 * attention is required when changing members of struct mbuf.
84 *
85 * MLEN is data length in a normal mbuf.
86 * MHLEN is data length in an mbuf with pktheader.
87 * MINCLSIZE is a smallest amount of data that should be put into cluster.
88 *
89 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
90 * they are sensible.
91 */
92 struct mbuf;
93 #define MHSIZE offsetof(struct mbuf, m_dat)
94 #define MPKTHSIZE offsetof(struct mbuf, m_pktdat)
95 #define MLEN ((int)(MSIZE - MHSIZE))
96 #define MHLEN ((int)(MSIZE - MPKTHSIZE))
97 #define MINCLSIZE (MHLEN + 1)
98 #define M_NODOM 255
99
100 #ifdef _KERNEL
101 /*-
102 * Macro for type conversion: convert mbuf pointer to data pointer of correct
103 * type:
104 *
105 * mtod(m, t) -- Convert mbuf pointer to data pointer of correct type.
106 * mtodo(m, o) -- Same as above but with offset 'o' into data.
107 */
108 #define mtod(m, t) ((t)((m)->m_data))
109 #define mtodo(m, o) ((void *)(((m)->m_data) + (o)))
110
111 /*
112 * Argument structure passed to UMA routines during mbuf and packet
113 * allocations.
114 */
115 struct mb_args {
116 int flags; /* Flags for mbuf being allocated */
117 short type; /* Type of mbuf being allocated */
118 };
119 #endif /* _KERNEL */
120
121 /*
122 * Packet tag structure (see below for details).
123 */
124 struct m_tag {
125 SLIST_ENTRY(m_tag) m_tag_link; /* List of packet tags */
126 u_int16_t m_tag_id; /* Tag ID */
127 u_int16_t m_tag_len; /* Length of data */
128 u_int32_t m_tag_cookie; /* ABI/Module ID */
129 void (*m_tag_free)(struct m_tag *);
130 };
131
132 /*
133 * Static network interface owned tag.
134 * Allocated through ifp->if_snd_tag_alloc().
135 */
136 struct if_snd_tag_sw;
137
138 struct m_snd_tag {
139 struct ifnet *ifp; /* network interface tag belongs to */
140 const struct if_snd_tag_sw *sw;
141 volatile u_int refcount;
142 };
143
144 /*
145 * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set.
146 * Size ILP32: 56
147 * LP64: 64
148 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
149 * they are correct.
150 */
151 struct pkthdr {
152 union {
153 struct m_snd_tag *snd_tag; /* send tag, if any */
154 struct ifnet *rcvif; /* rcv interface */
155 struct {
156 uint16_t rcvidx; /* rcv interface index ... */
157 uint16_t rcvgen; /* ... and generation count */
158 };
159 };
160 union {
161 struct ifnet *leaf_rcvif; /* leaf rcv interface */
162 struct {
163 uint16_t leaf_rcvidx; /* leaf rcv interface index ... */
164 uint16_t leaf_rcvgen; /* ... and generation count */
165 };
166 };
167 SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */
168 int32_t len; /* total packet length */
169
170 /* Layer crossing persistent information. */
171 uint32_t flowid; /* packet's 4-tuple system */
172 uint32_t csum_flags; /* checksum and offload features */
173 uint16_t fibnum; /* this packet should use this fib */
174 uint8_t numa_domain; /* NUMA domain of recvd pkt */
175 uint8_t rsstype; /* hash type */
176 #if !defined(__LP64__)
177 uint32_t pad; /* pad for 64bit alignment */
178 #endif
179 union {
180 uint64_t rcv_tstmp; /* timestamp in ns */
181 struct {
182 uint8_t l2hlen; /* layer 2 hdr len */
183 uint8_t l3hlen; /* layer 3 hdr len */
184 uint8_t l4hlen; /* layer 4 hdr len */
185 uint8_t l5hlen; /* layer 5 hdr len */
186 uint8_t inner_l2hlen;
187 uint8_t inner_l3hlen;
188 uint8_t inner_l4hlen;
189 uint8_t inner_l5hlen;
190 };
191 };
192 union {
193 uint8_t eight[8];
194 uint16_t sixteen[4];
195 uint32_t thirtytwo[2];
196 uint64_t sixtyfour[1];
197 uintptr_t unintptr[1];
198 void *ptr;
199 } PH_per;
200
201 /* Layer specific non-persistent local storage for reassembly, etc. */
202 union {
203 union {
204 uint8_t eight[8];
205 uint16_t sixteen[4];
206 uint32_t thirtytwo[2];
207 uint64_t sixtyfour[1];
208 uintptr_t unintptr[1];
209 void *ptr;
210 } PH_loc;
211 /* Upon allocation: total packet memory consumption. */
212 u_int memlen;
213 };
214 };
215 #define ether_vtag PH_per.sixteen[0]
216 #define tcp_tun_port PH_per.sixteen[0] /* outbound */
217 #define vt_nrecs PH_per.sixteen[0] /* mld and v6-ND */
218 #define tso_segsz PH_per.sixteen[1] /* inbound after LRO */
219 #define lro_nsegs tso_segsz /* inbound after LRO */
220 #define csum_data PH_per.thirtytwo[1] /* inbound from hardware up */
221 #define lro_tcp_d_len PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */
222 #define lro_tcp_d_csum PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */
223 #define lro_tcp_h_off PH_loc.sixteen[2] /* inbound during LRO (no reassembly) */
224 #define lro_etype PH_loc.sixteen[3] /* inbound during LRO (no reassembly) */
225 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */
226
227 /*
228 * TLS records for TLS 1.0-1.2 can have the following header lengths:
229 * - 5 (AES-CBC with implicit IV)
230 * - 21 (AES-CBC with explicit IV)
231 * - 13 (AES-GCM with 8 byte explicit IV)
232 */
233 #define MBUF_PEXT_HDR_LEN 23
234
235 /*
236 * TLS records for TLS 1.0-1.2 can have the following maximum trailer
237 * lengths:
238 * - 16 (AES-GCM)
239 * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding)
240 * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding)
241 * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding)
242 */
243 #define MBUF_PEXT_TRAIL_LEN 64
244
245 #if defined(__LP64__)
246 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t))
247 #else
248 #define MBUF_PEXT_MAX_PGS (64 / sizeof(vm_paddr_t))
249 #endif
250
251 #define MBUF_PEXT_MAX_BYTES \
252 (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN)
253
254 struct ktls_session;
255 struct socket;
256
257 /*
258 * Description of external storage mapped into mbuf; valid only if M_EXT is
259 * set.
260 * Size ILP32: 28
261 * LP64: 48
262 * Compile-time assertions in uipc_mbuf.c test these values to ensure that
263 * they are correct.
264 */
265 typedef void m_ext_free_t(struct mbuf *);
266 struct m_ext {
267 union {
268 /*
269 * If EXT_FLAG_EMBREF is set, then we use refcount in the
270 * mbuf, the 'ext_count' member. Otherwise, we have a
271 * shadow copy and we use pointer 'ext_cnt'. The original
272 * mbuf is responsible to carry the pointer to free routine
273 * and its arguments. They aren't copied into shadows in
274 * mb_dupcl() to avoid dereferencing next cachelines.
275 */
276 volatile u_int ext_count;
277 volatile u_int *ext_cnt;
278 };
279 uint32_t ext_size; /* size of buffer, for ext_free */
280 uint32_t ext_type:8, /* type of external storage */
281 ext_flags:24; /* external storage mbuf flags */
282 union {
283 struct {
284 /*
285 * Regular M_EXT mbuf:
286 * o ext_buf always points to the external buffer.
287 * o ext_free (below) and two optional arguments
288 * ext_arg1 and ext_arg2 store the free context for
289 * the external storage. They are set only in the
290 * refcount carrying mbuf, the one with
291 * EXT_FLAG_EMBREF flag, with exclusion for
292 * EXT_EXTREF type, where the free context is copied
293 * into all mbufs that use same external storage.
294 */
295 char *ext_buf; /* start of buffer */
296 #define m_ext_copylen offsetof(struct m_ext, ext_arg2)
297 void *ext_arg2;
298 };
299 struct {
300 /*
301 * Multi-page M_EXTPG mbuf:
302 * o extpg_pa - page vector.
303 * o extpg_trail and extpg_hdr - TLS trailer and
304 * header.
305 * Uses ext_free and may also use ext_arg1.
306 */
307 vm_paddr_t extpg_pa[MBUF_PEXT_MAX_PGS];
308 char extpg_trail[MBUF_PEXT_TRAIL_LEN];
309 char extpg_hdr[MBUF_PEXT_HDR_LEN];
310 /* Pretend these 3 fields are part of mbuf itself. */
311 #define m_epg_pa m_ext.extpg_pa
312 #define m_epg_trail m_ext.extpg_trail
313 #define m_epg_hdr m_ext.extpg_hdr
314 #define m_epg_ext_copylen offsetof(struct m_ext, ext_free)
315 };
316 };
317 /*
318 * Free method and optional argument pointer, both
319 * used by M_EXT and M_EXTPG.
320 */
321 m_ext_free_t *ext_free;
322 void *ext_arg1;
323 };
324
325 /*
326 * The core of the mbuf object along with some shortcut defines for practical
327 * purposes.
328 */
329 struct mbuf {
330 /*
331 * Header present at the beginning of every mbuf.
332 * Size ILP32: 24
333 * LP64: 32
334 * Compile-time assertions in uipc_mbuf.c test these values to ensure
335 * that they are correct.
336 */
337 union { /* next buffer in chain */
338 struct mbuf *m_next;
339 SLIST_ENTRY(mbuf) m_slist;
340 STAILQ_ENTRY(mbuf) m_stailq;
341 };
342 union { /* next chain in queue/record */
343 struct mbuf *m_nextpkt;
344 SLIST_ENTRY(mbuf) m_slistpkt;
345 STAILQ_ENTRY(mbuf) m_stailqpkt;
346 };
347 caddr_t m_data; /* location of data */
348 int32_t m_len; /* amount of data in this mbuf */
349 uint32_t m_type:8, /* type of data in this mbuf */
350 m_flags:24; /* flags; see below */
351 #if !defined(__LP64__)
352 uint32_t m_pad; /* pad for 64bit alignment */
353 #endif
354
355 /*
356 * A set of optional headers (packet header, external storage header)
357 * and internal data storage. Historically, these arrays were sized
358 * to MHLEN (space left after a packet header) and MLEN (space left
359 * after only a regular mbuf header); they are now variable size in
360 * order to support future work on variable-size mbufs.
361 */
362 union {
363 struct {
364 union {
365 /* M_PKTHDR set. */
366 struct pkthdr m_pkthdr;
367
368 /* M_EXTPG set.
369 * Multi-page M_EXTPG mbuf has its meta data
370 * split between the below anonymous structure
371 * and m_ext. It carries vector of pages,
372 * optional header and trailer char vectors
373 * and pointers to socket/TLS data.
374 */
375 #define m_epg_startcopy m_epg_npgs
376 #define m_epg_endcopy m_epg_stailq
377 struct {
378 /* Overall count of pages and count of
379 * pages with I/O pending. */
380 uint8_t m_epg_npgs;
381 uint8_t m_epg_nrdy;
382 /* TLS header and trailer lengths.
383 * The data itself resides in m_ext. */
384 uint8_t m_epg_hdrlen;
385 uint8_t m_epg_trllen;
386 /* Offset into 1st page and length of
387 * data in the last page. */
388 uint16_t m_epg_1st_off;
389 uint16_t m_epg_last_len;
390 uint8_t m_epg_flags;
391 #define EPG_FLAG_ANON 0x1 /* Data can be encrypted in place. */
392 #define EPG_FLAG_2FREE 0x2 /* Scheduled for free. */
393 uint8_t m_epg_record_type;
394 uint8_t __spare[2];
395 int m_epg_enc_cnt;
396 struct ktls_session *m_epg_tls;
397 struct socket *m_epg_so;
398 uint64_t m_epg_seqno;
399 STAILQ_ENTRY(mbuf) m_epg_stailq;
400 };
401 };
402 union {
403 /* M_EXT or M_EXTPG set. */
404 struct m_ext m_ext;
405 /* M_PKTHDR set, neither M_EXT nor M_EXTPG. */
406 char m_pktdat[0];
407 };
408 };
409 char m_dat[0]; /* !M_PKTHDR, !M_EXT */
410 };
411 };
412
413 #ifdef _KERNEL
414 static inline int
m_epg_pagelen(const struct mbuf * m,int pidx,int pgoff)415 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff)
416 {
417
418 KASSERT(pgoff == 0 || pidx == 0,
419 ("page %d with non-zero offset %d in %p", pidx, pgoff, m));
420
421 if (pidx == m->m_epg_npgs - 1) {
422 return (m->m_epg_last_len);
423 } else {
424 return (PAGE_SIZE - pgoff);
425 }
426 }
427
428 #ifdef INVARIANTS
429 #define MCHECK(ex, msg) KASSERT((ex), \
430 ("Multi page mbuf %p with " #msg " at %s:%d", \
431 m, __FILE__, __LINE__))
432 /*
433 * NB: This expects a non-empty buffer (npgs > 0 and
434 * last_pg_len > 0).
435 */
436 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do { \
437 MCHECK(m->m_epg_npgs > 0, "no valid pages"); \
438 MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa), \
439 "too many pages"); \
440 MCHECK(m->m_epg_nrdy <= m->m_epg_npgs, \
441 "too many ready pages"); \
442 MCHECK(m->m_epg_1st_off < PAGE_SIZE, \
443 "too large page offset"); \
444 MCHECK(m->m_epg_last_len > 0, "zero last page length"); \
445 MCHECK(m->m_epg_last_len <= PAGE_SIZE, \
446 "too large last page length"); \
447 if (m->m_epg_npgs == 1) \
448 MCHECK(m->m_epg_1st_off + \
449 m->m_epg_last_len <= PAGE_SIZE, \
450 "single page too large"); \
451 MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr), \
452 "too large header length"); \
453 MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail), \
454 "too large header length"); \
455 } while (0)
456 #else
457 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do {} while (0)
458 #endif
459 #endif
460
461 /*
462 * mbuf flags of global significance and layer crossing.
463 * Those of only protocol/layer specific significance are to be mapped
464 * to M_PROTO[1-11] and cleared at layer handoff boundaries.
465 * NB: Limited to the lower 24 bits.
466 */
467 #define M_EXT 0x00000001 /* has associated external storage */
468 #define M_PKTHDR 0x00000002 /* start of record */
469 #define M_EOR 0x00000004 /* end of record */
470 #define M_RDONLY 0x00000008 /* associated data is marked read-only */
471 #define M_BCAST 0x00000010 /* send/received as link-level broadcast */
472 #define M_MCAST 0x00000020 /* send/received as link-level multicast */
473 #define M_PROMISC 0x00000040 /* packet was not for us */
474 #define M_VLANTAG 0x00000080 /* ether_vtag is valid */
475 #define M_EXTPG 0x00000100 /* has array of unmapped pages and TLS */
476 #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */
477 #define M_TSTMP 0x00000400 /* rcv_tstmp field is valid */
478 #define M_TSTMP_HPREC 0x00000800 /* rcv_tstmp is high-prec, typically
479 hw-stamped on port (useful for IEEE 1588
480 and 802.1AS) */
481 #define M_TSTMP_LRO 0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */
482
483 #define M_PROTO1 0x00002000 /* protocol-specific */
484 #define M_PROTO2 0x00004000 /* protocol-specific */
485 #define M_PROTO3 0x00008000 /* protocol-specific */
486 #define M_PROTO4 0x00010000 /* protocol-specific */
487 #define M_PROTO5 0x00020000 /* protocol-specific */
488 #define M_PROTO6 0x00040000 /* protocol-specific */
489 #define M_PROTO7 0x00080000 /* protocol-specific */
490 #define M_PROTO8 0x00100000 /* protocol-specific */
491 #define M_PROTO9 0x00200000 /* protocol-specific */
492 #define M_PROTO10 0x00400000 /* protocol-specific */
493 #define M_PROTO11 0x00800000 /* protocol-specific */
494
495 /*
496 * Flags to purge when crossing layers.
497 */
498 #define M_PROTOFLAGS \
499 (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\
500 M_PROTO9|M_PROTO10|M_PROTO11)
501
502 /*
503 * Flags preserved when copying m_pkthdr.
504 */
505 #define M_COPYFLAGS \
506 (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \
507 M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS)
508
509 /*
510 * Flags preserved during demote.
511 */
512 #define M_DEMOTEFLAGS \
513 (M_EXT | M_RDONLY | M_NOFREE | M_EXTPG)
514
515 /*
516 * Mbuf flag description for use with printf(9) %b identifier.
517 */
518 #define M_FLAG_BITS \
519 "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \
520 "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO"
521 #define M_FLAG_PROTOBITS \
522 "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \
523 "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \
524 "\27M_PROTO10\28M_PROTO11"
525 #define M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS)
526
527 /*
528 * Network interface cards are able to hash protocol fields (such as IPv4
529 * addresses and TCP port numbers) classify packets into flows. These flows
530 * can then be used to maintain ordering while delivering packets to the OS
531 * via parallel input queues, as well as to provide a stateless affinity
532 * model. NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set
533 * m_flag fields to indicate how the hash should be interpreted by the
534 * network stack.
535 *
536 * Most NICs support RSS, which provides ordering and explicit affinity, and
537 * use the hash m_flag bits to indicate what header fields were covered by
538 * the hash. M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non-
539 * RSS cards or configurations that provide an opaque flow identifier, allowing
540 * for ordering and distribution without explicit affinity. Additionally,
541 * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash
542 * properties.
543 *
544 * The meaning of the IPV6_EX suffix:
545 * "o Home address from the home address option in the IPv6 destination
546 * options header. If the extension header is not present, use the Source
547 * IPv6 Address.
548 * o IPv6 address that is contained in the Routing-Header-Type-2 from the
549 * associated extension header. If the extension header is not present,
550 * use the Destination IPv6 Address."
551 * Quoted from:
552 * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex
553 */
554 #define M_HASHTYPE_HASHPROP 0x80 /* has hash properties */
555 #define M_HASHTYPE_INNER 0x40 /* calculated from inner headers */
556 #define M_HASHTYPE_HASH(t) (M_HASHTYPE_HASHPROP | (t))
557 /* Microsoft RSS standard hash types */
558 #define M_HASHTYPE_NONE 0
559 #define M_HASHTYPE_RSS_IPV4 M_HASHTYPE_HASH(1) /* IPv4 2-tuple */
560 #define M_HASHTYPE_RSS_TCP_IPV4 M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */
561 #define M_HASHTYPE_RSS_IPV6 M_HASHTYPE_HASH(3) /* IPv6 2-tuple */
562 #define M_HASHTYPE_RSS_TCP_IPV6 M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */
563 #define M_HASHTYPE_RSS_IPV6_EX M_HASHTYPE_HASH(5) /* IPv6 2-tuple +
564 * ext hdrs */
565 #define M_HASHTYPE_RSS_TCP_IPV6_EX M_HASHTYPE_HASH(6) /* TCPv6 4-tuple +
566 * ext hdrs */
567 #define M_HASHTYPE_RSS_UDP_IPV4 M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/
568 #define M_HASHTYPE_RSS_UDP_IPV6 M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/
569 #define M_HASHTYPE_RSS_UDP_IPV6_EX M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple +
570 * ext hdrs */
571
572 #define M_HASHTYPE_OPAQUE 0x3f /* ordering, not affinity */
573 #define M_HASHTYPE_OPAQUE_HASH M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE)
574 /* ordering+hash, not affinity*/
575
576 #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0)
577 #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype & ~M_HASHTYPE_INNER)
578 #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v))
579 #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v))
580 #define M_HASHTYPE_ISHASH(m) \
581 (((m)->m_pkthdr.rsstype & M_HASHTYPE_HASHPROP) != 0)
582 #define M_HASHTYPE_SETINNER(m) do { \
583 (m)->m_pkthdr.rsstype |= M_HASHTYPE_INNER; \
584 } while (0)
585
586 /*
587 * External mbuf storage buffer types.
588 */
589 #define EXT_CLUSTER 1 /* mbuf cluster */
590 #define EXT_SFBUF 2 /* sendfile(2)'s sf_buf */
591 #define EXT_JUMBOP 3 /* jumbo cluster page sized */
592 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */
593 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */
594 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */
595 #define EXT_MBUF 7 /* external mbuf reference */
596 #define EXT_RXRING 8 /* data in NIC receive ring */
597 #define EXT_CTL 9 /* buffer from a ctl(4) backend */
598
599 #define EXT_VENDOR1 224 /* for vendor-internal use */
600 #define EXT_VENDOR2 225 /* for vendor-internal use */
601 #define EXT_VENDOR3 226 /* for vendor-internal use */
602 #define EXT_VENDOR4 227 /* for vendor-internal use */
603
604 #define EXT_EXP1 244 /* for experimental use */
605 #define EXT_EXP2 245 /* for experimental use */
606 #define EXT_EXP3 246 /* for experimental use */
607 #define EXT_EXP4 247 /* for experimental use */
608
609 #define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */
610 #define EXT_MOD_TYPE 253 /* custom module's ext_buf type */
611 #define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */
612 #define EXT_EXTREF 255 /* has externally maintained ext_cnt ptr */
613
614 /*
615 * Flags for external mbuf buffer types.
616 * NB: limited to the lower 24 bits.
617 */
618 #define EXT_FLAG_EMBREF 0x000001 /* embedded ext_count */
619 #define EXT_FLAG_EXTREF 0x000002 /* external ext_cnt, notyet */
620
621 #define EXT_FLAG_NOFREE 0x000010 /* don't free mbuf to pool, notyet */
622
623 #define EXT_FLAG_VENDOR1 0x010000 /* These flags are vendor */
624 #define EXT_FLAG_VENDOR2 0x020000 /* or submodule specific, */
625 #define EXT_FLAG_VENDOR3 0x040000 /* not used by mbuf code. */
626 #define EXT_FLAG_VENDOR4 0x080000 /* Set/read by submodule. */
627
628 #define EXT_FLAG_EXP1 0x100000 /* for experimental use */
629 #define EXT_FLAG_EXP2 0x200000 /* for experimental use */
630 #define EXT_FLAG_EXP3 0x400000 /* for experimental use */
631 #define EXT_FLAG_EXP4 0x800000 /* for experimental use */
632
633 /*
634 * EXT flag description for use with printf(9) %b identifier.
635 */
636 #define EXT_FLAG_BITS \
637 "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \
638 "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \
639 "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \
640 "\30EXT_FLAG_EXP4"
641
642 /*
643 * Flags indicating checksum, segmentation and other offload work to be
644 * done, or already done, by hardware or lower layers.
645 *
646 * Flags that are set by upper protocol layers requesting lower
647 * layers, or ideally the hardware, to perform these offloading tasks.
648 * Before passing packets to a network interface this field and its flags can
649 * be directly tested against ifnet if_hwassist. Note that the flags
650 * CSUM_IP_SCTP, CSUM_IP_TCP, and CSUM_IP_UDP can appear on input processing
651 * of SCTP, TCP, and UDP. In such a case the checksum will not be computed or
652 * validated by SCTP, TCP, or TCP, since the packet has not been on the wire.
653 *
654 * CSUM_INNER_<x> is the same as CSUM_<x> but it applies to the inner frame.
655 * The CSUM_ENCAP_<x> bits identify the outer encapsulation.
656 */
657 #define CSUM_IP 0x00000001 /* IP header checksum offload */
658 #define CSUM_IP_UDP 0x00000002 /* UDP checksum offload */
659 #define CSUM_IP_TCP 0x00000004 /* TCP checksum offload */
660 #define CSUM_IP_SCTP 0x00000008 /* SCTP checksum offload */
661 #define CSUM_IP_TSO 0x00000010 /* TCP segmentation offload */
662 #define CSUM_IP_ISCSI 0x00000020 /* iSCSI checksum offload */
663
664 #define CSUM_INNER_IP6_UDP 0x00000040
665 #define CSUM_INNER_IP6_TCP 0x00000080
666 #define CSUM_INNER_IP6_TSO 0x00000100
667 #define CSUM_IP6_UDP 0x00000200 /* UDP checksum offload */
668 #define CSUM_IP6_TCP 0x00000400 /* TCP checksum offload */
669 #define CSUM_IP6_SCTP 0x00000800 /* SCTP checksum offload */
670 #define CSUM_IP6_TSO 0x00001000 /* TCP segmentation offload */
671 #define CSUM_IP6_ISCSI 0x00002000 /* iSCSI checksum offload */
672
673 #define CSUM_INNER_IP 0x00004000
674 #define CSUM_INNER_IP_UDP 0x00008000
675 #define CSUM_INNER_IP_TCP 0x00010000
676 #define CSUM_INNER_IP_TSO 0x00020000
677
678 #define CSUM_ENCAP_VXLAN 0x00040000 /* VXLAN outer encapsulation */
679 #define CSUM_ENCAP_RSVD1 0x00080000
680
681 /* Flags used to indicate that the checksum was verified by hardware. */
682 #define CSUM_INNER_L3_CALC 0x00100000
683 #define CSUM_INNER_L3_VALID 0x00200000
684 #define CSUM_INNER_L4_CALC 0x00400000
685 #define CSUM_INNER_L4_VALID 0x00800000
686 #define CSUM_L3_CALC 0x01000000 /* calculated layer 3 csum */
687 #define CSUM_L3_VALID 0x02000000 /* checksum is correct */
688 #define CSUM_L4_CALC 0x04000000 /* calculated layer 4 csum */
689 #define CSUM_L4_VALID 0x08000000 /* checksum is correct */
690 #define CSUM_L5_CALC 0x10000000 /* calculated layer 5 csum */
691 #define CSUM_L5_VALID 0x20000000 /* checksum is correct */
692 #define CSUM_COALESCED 0x40000000 /* contains merged segments */
693
694 #define CSUM_SND_TAG 0x80000000 /* Packet header has send tag */
695
696 #define CSUM_FLAGS_TX (CSUM_IP | CSUM_IP_UDP | CSUM_IP_TCP | CSUM_IP_SCTP | \
697 CSUM_IP_TSO | CSUM_IP_ISCSI | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | \
698 CSUM_INNER_IP6_TSO | CSUM_IP6_UDP | CSUM_IP6_TCP | CSUM_IP6_SCTP | \
699 CSUM_IP6_TSO | CSUM_IP6_ISCSI | CSUM_INNER_IP | CSUM_INNER_IP_UDP | \
700 CSUM_INNER_IP_TCP | CSUM_INNER_IP_TSO | CSUM_ENCAP_VXLAN | \
701 CSUM_ENCAP_RSVD1 | CSUM_SND_TAG)
702
703 #define CSUM_FLAGS_RX (CSUM_INNER_L3_CALC | CSUM_INNER_L3_VALID | \
704 CSUM_INNER_L4_CALC | CSUM_INNER_L4_VALID | CSUM_L3_CALC | CSUM_L3_VALID | \
705 CSUM_L4_CALC | CSUM_L4_VALID | CSUM_L5_CALC | CSUM_L5_VALID | \
706 CSUM_COALESCED)
707
708 /*
709 * CSUM flag description for use with printf(9) %b identifier.
710 */
711 #define CSUM_BITS \
712 "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \
713 "\6CSUM_IP_ISCSI\7CSUM_INNER_IP6_UDP\10CSUM_INNER_IP6_TCP" \
714 "\11CSUM_INNER_IP6_TSO\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP" \
715 "\15CSUM_IP6_TSO\16CSUM_IP6_ISCSI\17CSUM_INNER_IP\20CSUM_INNER_IP_UDP" \
716 "\21CSUM_INNER_IP_TCP\22CSUM_INNER_IP_TSO\23CSUM_ENCAP_VXLAN" \
717 "\24CSUM_ENCAP_RSVD1\25CSUM_INNER_L3_CALC\26CSUM_INNER_L3_VALID" \
718 "\27CSUM_INNER_L4_CALC\30CSUM_INNER_L4_VALID\31CSUM_L3_CALC" \
719 "\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID\35CSUM_L5_CALC" \
720 "\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG"
721
722 /* CSUM flags compatibility mappings. */
723 #define CSUM_IP_CHECKED CSUM_L3_CALC
724 #define CSUM_IP_VALID CSUM_L3_VALID
725 #define CSUM_DATA_VALID CSUM_L4_VALID
726 #define CSUM_PSEUDO_HDR CSUM_L4_CALC
727 #define CSUM_SCTP_VALID CSUM_L4_VALID
728 #define CSUM_DELAY_DATA (CSUM_TCP|CSUM_UDP)
729 #define CSUM_DELAY_IP CSUM_IP /* Only v4, no v6 IP hdr csum */
730 #define CSUM_DELAY_DATA_IPV6 (CSUM_TCP_IPV6|CSUM_UDP_IPV6)
731 #define CSUM_DATA_VALID_IPV6 CSUM_DATA_VALID
732 #define CSUM_TCP CSUM_IP_TCP
733 #define CSUM_UDP CSUM_IP_UDP
734 #define CSUM_SCTP CSUM_IP_SCTP
735 #define CSUM_TSO (CSUM_IP_TSO|CSUM_IP6_TSO)
736 #define CSUM_INNER_TSO (CSUM_INNER_IP_TSO|CSUM_INNER_IP6_TSO)
737 #define CSUM_UDP_IPV6 CSUM_IP6_UDP
738 #define CSUM_TCP_IPV6 CSUM_IP6_TCP
739 #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP
740 #define CSUM_TLS_MASK (CSUM_L5_CALC|CSUM_L5_VALID)
741 #define CSUM_TLS_DECRYPTED CSUM_L5_CALC
742
743 /*
744 * mbuf types describing the content of the mbuf (including external storage).
745 */
746 #define MT_NOTMBUF 0 /* USED INTERNALLY ONLY! Object is not mbuf */
747 #define MT_DATA 1 /* dynamic (data) allocation */
748 #define MT_HEADER MT_DATA /* packet header, use M_PKTHDR instead */
749
750 #define MT_VENDOR1 4 /* for vendor-internal use */
751 #define MT_VENDOR2 5 /* for vendor-internal use */
752 #define MT_VENDOR3 6 /* for vendor-internal use */
753 #define MT_VENDOR4 7 /* for vendor-internal use */
754
755 #define MT_SONAME 8 /* socket name */
756
757 #define MT_EXP1 9 /* for experimental use */
758 #define MT_EXP2 10 /* for experimental use */
759 #define MT_EXP3 11 /* for experimental use */
760 #define MT_EXP4 12 /* for experimental use */
761
762 #define MT_CONTROL 14 /* extra-data protocol message */
763 #define MT_EXTCONTROL 15 /* control message with externalized contents */
764 #define MT_OOBDATA 16 /* expedited data */
765
766 #define MT_NOINIT 255 /* Not a type but a flag to allocate
767 a non-initialized mbuf */
768
769 /*
770 * String names of mbuf-related UMA(9) and malloc(9) types. Exposed to
771 * !_KERNEL so that monitoring tools can look up the zones with
772 * libmemstat(3).
773 */
774 #define MBUF_MEM_NAME "mbuf"
775 #define MBUF_CLUSTER_MEM_NAME "mbuf_cluster"
776 #define MBUF_PACKET_MEM_NAME "mbuf_packet"
777 #define MBUF_JUMBOP_MEM_NAME "mbuf_jumbo_page"
778 #define MBUF_JUMBO9_MEM_NAME "mbuf_jumbo_9k"
779 #define MBUF_JUMBO16_MEM_NAME "mbuf_jumbo_16k"
780 #define MBUF_TAG_MEM_NAME "mbuf_tag"
781 #define MBUF_EXTREFCNT_MEM_NAME "mbuf_ext_refcnt"
782 #define MBUF_EXTPGS_MEM_NAME "mbuf_extpgs"
783
784 #ifdef _KERNEL
785 union if_snd_tag_alloc_params;
786
787 #define MBUF_CHECKSLEEP(how) do { \
788 if (how == M_WAITOK) \
789 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \
790 "Sleeping in \"%s\"", __func__); \
791 } while (0)
792
793 /*
794 * Network buffer allocation API
795 *
796 * The rest of it is defined in kern/kern_mbuf.c
797 */
798 extern uma_zone_t zone_mbuf;
799 extern uma_zone_t zone_clust;
800 extern uma_zone_t zone_pack;
801 extern uma_zone_t zone_jumbop;
802 extern uma_zone_t zone_jumbo9;
803 extern uma_zone_t zone_jumbo16;
804 extern uma_zone_t zone_extpgs;
805
806 void mb_dupcl(struct mbuf *, struct mbuf *);
807 void mb_free_ext(struct mbuf *);
808 void mb_free_extpg(struct mbuf *);
809 void mb_free_mext_pgs(struct mbuf *);
810 struct mbuf *mb_alloc_ext_pgs(int, m_ext_free_t, int);
811 struct mbuf *mb_alloc_ext_plus_pages(int, int);
812 struct mbuf *mb_mapped_to_unmapped(struct mbuf *, int, int, int,
813 struct mbuf **);
814 int mb_unmapped_compress(struct mbuf *m);
815 int mb_unmapped_to_ext(struct mbuf *m, struct mbuf **mres);
816 void mb_free_notready(struct mbuf *m, int count);
817 void m_adj(struct mbuf *, int);
818 void m_adj_decap(struct mbuf *, int);
819 int m_apply(struct mbuf *, int, int,
820 int (*)(void *, void *, u_int), void *);
821 int m_append(struct mbuf *, int, c_caddr_t);
822 void m_cat(struct mbuf *, struct mbuf *);
823 void m_catpkt(struct mbuf *, struct mbuf *);
824 int m_clget(struct mbuf *m, int how);
825 void *m_cljget(struct mbuf *m, int how, int size);
826 struct mbuf *m_collapse(struct mbuf *, int, int);
827 void m_copyback(struct mbuf *, int, int, c_caddr_t);
828 void m_copydata(const struct mbuf *, int, int, caddr_t);
829 struct mbuf *m_copym(struct mbuf *, int, int, int);
830 struct mbuf *m_copypacket(struct mbuf *, int);
831 void m_copy_pkthdr(struct mbuf *, struct mbuf *);
832 struct mbuf *m_copyup(struct mbuf *, int, int);
833 struct mbuf *m_defrag(struct mbuf *, int);
834 void m_demote_pkthdr(struct mbuf *);
835 void m_demote(struct mbuf *, int, int);
836 struct mbuf *m_devget(char *, int, int, struct ifnet *,
837 void (*)(char *, caddr_t, u_int));
838 void m_dispose_extcontrolm(struct mbuf *m);
839 struct mbuf *m_dup(const struct mbuf *, int);
840 int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int);
841 void m_extadd(struct mbuf *, char *, u_int, m_ext_free_t,
842 void *, void *, int, int);
843 u_int m_fixhdr(struct mbuf *);
844 struct mbuf *m_fragment(struct mbuf *, int, int);
845 void m_freem(struct mbuf *);
846 void m_freemp(struct mbuf *);
847 void m_free_raw(struct mbuf *);
848 struct mbuf *m_get2(int, int, short, int);
849 struct mbuf *m_get3(int, int, short, int);
850 struct mbuf *m_getjcl(int, short, int, int);
851 struct mbuf *m_getm2(struct mbuf *, int, int, short, int);
852 struct mbuf *m_getptr(struct mbuf *, int, int *);
853 u_int m_length(struct mbuf *, struct mbuf **);
854 int m_mbuftouio(struct uio *, const struct mbuf *, int);
855 void m_move_pkthdr(struct mbuf *, struct mbuf *);
856 int m_pkthdr_init(struct mbuf *, int);
857 struct mbuf *m_prepend(struct mbuf *, int, int);
858 void m_print(const struct mbuf *, int);
859 struct mbuf *m_pulldown(struct mbuf *, int, int, int *);
860 struct mbuf *m_pullup(struct mbuf *, int);
861 int m_sanity(struct mbuf *, int);
862 struct mbuf *m_split(struct mbuf *, int, int);
863 struct mbuf *m_uiotombuf(struct uio *, int, int, int, int);
864 int m_unmapped_uiomove(const struct mbuf *, int, struct uio *,
865 int);
866 struct mbuf *m_unshare(struct mbuf *, int);
867 int m_snd_tag_alloc(struct ifnet *,
868 union if_snd_tag_alloc_params *, struct m_snd_tag **);
869 void m_snd_tag_init(struct m_snd_tag *, struct ifnet *,
870 const struct if_snd_tag_sw *);
871 void m_snd_tag_destroy(struct m_snd_tag *);
872 void m_rcvif_serialize(struct mbuf *);
873 struct ifnet *m_rcvif_restore(struct mbuf *);
874
875 static __inline int
m_gettype(int size)876 m_gettype(int size)
877 {
878 int type;
879
880 switch (size) {
881 case MSIZE:
882 type = EXT_MBUF;
883 break;
884 case MCLBYTES:
885 type = EXT_CLUSTER;
886 break;
887 #if MJUMPAGESIZE != MCLBYTES
888 case MJUMPAGESIZE:
889 type = EXT_JUMBOP;
890 break;
891 #endif
892 case MJUM9BYTES:
893 type = EXT_JUMBO9;
894 break;
895 case MJUM16BYTES:
896 type = EXT_JUMBO16;
897 break;
898 default:
899 panic("%s: invalid cluster size %d", __func__, size);
900 }
901
902 return (type);
903 }
904
905 /*
906 * Associated an external reference counted buffer with an mbuf.
907 */
908 static __inline void
m_extaddref(struct mbuf * m,char * buf,u_int size,u_int * ref_cnt,m_ext_free_t freef,void * arg1,void * arg2)909 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt,
910 m_ext_free_t freef, void *arg1, void *arg2)
911 {
912
913 KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__));
914
915 atomic_add_int(ref_cnt, 1);
916 m->m_flags |= M_EXT;
917 m->m_ext.ext_buf = buf;
918 m->m_ext.ext_cnt = ref_cnt;
919 m->m_data = m->m_ext.ext_buf;
920 m->m_ext.ext_size = size;
921 m->m_ext.ext_free = freef;
922 m->m_ext.ext_arg1 = arg1;
923 m->m_ext.ext_arg2 = arg2;
924 m->m_ext.ext_type = EXT_EXTREF;
925 m->m_ext.ext_flags = 0;
926 }
927
928 static __inline uma_zone_t
m_getzone(int size)929 m_getzone(int size)
930 {
931 uma_zone_t zone;
932
933 switch (size) {
934 case MCLBYTES:
935 zone = zone_clust;
936 break;
937 #if MJUMPAGESIZE != MCLBYTES
938 case MJUMPAGESIZE:
939 zone = zone_jumbop;
940 break;
941 #endif
942 case MJUM9BYTES:
943 zone = zone_jumbo9;
944 break;
945 case MJUM16BYTES:
946 zone = zone_jumbo16;
947 break;
948 default:
949 panic("%s: invalid cluster size %d", __func__, size);
950 }
951
952 return (zone);
953 }
954
955 /*
956 * Initialize an mbuf with linear storage.
957 *
958 * Inline because the consumer text overhead will be roughly the same to
959 * initialize or call a function with this many parameters and M_PKTHDR
960 * should go away with constant propagation for !MGETHDR.
961 */
962 static __inline int
m_init(struct mbuf * m,int how,short type,int flags)963 m_init(struct mbuf *m, int how, short type, int flags)
964 {
965 int error;
966
967 m->m_next = NULL;
968 m->m_nextpkt = NULL;
969 m->m_data = m->m_dat;
970 m->m_len = 0;
971 m->m_flags = flags;
972 m->m_type = type;
973 if (flags & M_PKTHDR)
974 error = m_pkthdr_init(m, how);
975 else
976 error = 0;
977
978 MBUF_PROBE5(m__init, m, how, type, flags, error);
979 return (error);
980 }
981
982 static __inline struct mbuf *
m_get_raw(int how,short type)983 m_get_raw(int how, short type)
984 {
985 struct mbuf *m;
986 struct mb_args args;
987
988 args.flags = 0;
989 args.type = type | MT_NOINIT;
990 m = uma_zalloc_arg(zone_mbuf, &args, how);
991 MBUF_PROBE3(m__get_raw, how, type, m);
992 return (m);
993 }
994
995 static __inline struct mbuf *
m_get(int how,short type)996 m_get(int how, short type)
997 {
998 struct mbuf *m;
999 struct mb_args args;
1000
1001 args.flags = 0;
1002 args.type = type;
1003 m = uma_zalloc_arg(zone_mbuf, &args, how);
1004 MBUF_PROBE3(m__get, how, type, m);
1005 return (m);
1006 }
1007
1008 static __inline struct mbuf *
m_gethdr_raw(int how,short type)1009 m_gethdr_raw(int how, short type)
1010 {
1011 struct mbuf *m;
1012 struct mb_args args;
1013
1014 args.flags = M_PKTHDR;
1015 args.type = type | MT_NOINIT;
1016 m = uma_zalloc_arg(zone_mbuf, &args, how);
1017 MBUF_PROBE3(m__gethdr_raw, how, type, m);
1018 return (m);
1019 }
1020
1021 static __inline struct mbuf *
m_gethdr(int how,short type)1022 m_gethdr(int how, short type)
1023 {
1024 struct mbuf *m;
1025 struct mb_args args;
1026
1027 args.flags = M_PKTHDR;
1028 args.type = type;
1029 m = uma_zalloc_arg(zone_mbuf, &args, how);
1030 MBUF_PROBE3(m__gethdr, how, type, m);
1031 return (m);
1032 }
1033
1034 static __inline struct mbuf *
m_getcl(int how,short type,int flags)1035 m_getcl(int how, short type, int flags)
1036 {
1037 struct mbuf *m;
1038 struct mb_args args;
1039
1040 args.flags = flags;
1041 args.type = type;
1042 m = uma_zalloc_arg(zone_pack, &args, how);
1043 MBUF_PROBE4(m__getcl, how, type, flags, m);
1044 return (m);
1045 }
1046
1047 /*
1048 * XXX: m_cljset() is a dangerous API. One must attach only a new,
1049 * unreferenced cluster to an mbuf(9). It is not possible to assert
1050 * that, so care can be taken only by users of the API.
1051 */
1052 static __inline void
m_cljset(struct mbuf * m,void * cl,int type)1053 m_cljset(struct mbuf *m, void *cl, int type)
1054 {
1055 int size;
1056
1057 switch (type) {
1058 case EXT_CLUSTER:
1059 size = MCLBYTES;
1060 break;
1061 #if MJUMPAGESIZE != MCLBYTES
1062 case EXT_JUMBOP:
1063 size = MJUMPAGESIZE;
1064 break;
1065 #endif
1066 case EXT_JUMBO9:
1067 size = MJUM9BYTES;
1068 break;
1069 case EXT_JUMBO16:
1070 size = MJUM16BYTES;
1071 break;
1072 default:
1073 panic("%s: unknown cluster type %d", __func__, type);
1074 break;
1075 }
1076
1077 m->m_data = m->m_ext.ext_buf = cl;
1078 m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL;
1079 m->m_ext.ext_size = size;
1080 m->m_ext.ext_type = type;
1081 m->m_ext.ext_flags = EXT_FLAG_EMBREF;
1082 m->m_ext.ext_count = 1;
1083 m->m_flags |= M_EXT;
1084 MBUF_PROBE3(m__cljset, m, cl, type);
1085 }
1086
1087 static __inline void
m_chtype(struct mbuf * m,short new_type)1088 m_chtype(struct mbuf *m, short new_type)
1089 {
1090
1091 m->m_type = new_type;
1092 }
1093
1094 static __inline void
m_clrprotoflags(struct mbuf * m)1095 m_clrprotoflags(struct mbuf *m)
1096 {
1097
1098 while (m) {
1099 m->m_flags &= ~M_PROTOFLAGS;
1100 m = m->m_next;
1101 }
1102 }
1103
1104 static __inline struct mbuf *
m_last(struct mbuf * m)1105 m_last(struct mbuf *m)
1106 {
1107
1108 while (m->m_next)
1109 m = m->m_next;
1110 return (m);
1111 }
1112
1113 static inline u_int
m_extrefcnt(struct mbuf * m)1114 m_extrefcnt(struct mbuf *m)
1115 {
1116
1117 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing for %p", __func__, m));
1118
1119 return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count :
1120 *m->m_ext.ext_cnt);
1121 }
1122
1123 /*
1124 * mbuf, cluster, and external object allocation macros (for compatibility
1125 * purposes).
1126 */
1127 #define M_MOVE_PKTHDR(to, from) m_move_pkthdr((to), (from))
1128 #define MGET(m, how, type) ((m) = m_get((how), (type)))
1129 #define MGETHDR(m, how, type) ((m) = m_gethdr((how), (type)))
1130 #define MCLGET(m, how) m_clget((m), (how))
1131 #define MEXTADD(m, buf, size, free, arg1, arg2, flags, type) \
1132 m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2), \
1133 (flags), (type))
1134 #define m_getm(m, len, how, type) \
1135 m_getm2((m), (len), (how), (type), M_PKTHDR)
1136
1137 /*
1138 * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can
1139 * be both the local data payload, or an external buffer area, depending on
1140 * whether M_EXT is set).
1141 */
1142 #define M_WRITABLE(m) (((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 && \
1143 (!(((m)->m_flags & M_EXT)) || \
1144 (m_extrefcnt(m) == 1)))
1145
1146 /* Check if the supplied mbuf has a packet header, or else panic. */
1147 #define M_ASSERTPKTHDR(m) \
1148 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR, \
1149 ("%s: no mbuf %p packet header!", __func__, (m)))
1150
1151 /* Check if the supplied mbuf has no send tag, or else panic. */
1152 #define M_ASSERT_NO_SND_TAG(m) \
1153 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR && \
1154 ((m)->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0, \
1155 ("%s: receive mbuf %p has send tag!", __func__, (m)))
1156
1157 /* Check if mbuf is multipage. */
1158 #define M_ASSERTEXTPG(m) \
1159 KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG, \
1160 ("%s: m %p is not multipage!", __func__, m))
1161
1162 /*
1163 * Ensure that the supplied mbuf is a valid, non-free mbuf.
1164 *
1165 * XXX: Broken at the moment. Need some UMA magic to make it work again.
1166 */
1167 #define M_ASSERTVALID(m) \
1168 KASSERT((((struct mbuf *)m)->m_flags & 0) == 0, \
1169 ("%s: attempted use of a free mbuf %p!", __func__, (m)))
1170
1171 /* Check whether any mbuf in the chain is unmapped. */
1172 #ifdef INVARIANTS
1173 #define M_ASSERTMAPPED(m) do { \
1174 for (struct mbuf *__m = (m); __m != NULL; __m = __m->m_next) \
1175 KASSERT((__m->m_flags & M_EXTPG) == 0, \
1176 ("%s: chain %p contains an unmapped mbuf", __func__, (m)));\
1177 } while (0)
1178 #else
1179 #define M_ASSERTMAPPED(m) do {} while (0)
1180 #endif
1181
1182 /*
1183 * Return the address of the start of the buffer associated with an mbuf,
1184 * handling external storage, packet-header mbufs, and regular data mbufs.
1185 */
1186 #define M_START(m) \
1187 (((m)->m_flags & M_EXTPG) ? NULL : \
1188 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf : \
1189 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] : \
1190 &(m)->m_dat[0])
1191
1192 /*
1193 * Return the size of the buffer associated with an mbuf, handling external
1194 * storage, packet-header mbufs, and regular data mbufs.
1195 */
1196 #define M_SIZE(m) \
1197 (((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size : \
1198 ((m)->m_flags & M_PKTHDR) ? MHLEN : \
1199 MLEN)
1200
1201 /*
1202 * Set the m_data pointer of a newly allocated mbuf to place an object of the
1203 * specified size at the end of the mbuf, longword aligned.
1204 *
1205 * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as
1206 * separate macros, each asserting that it was called at the proper moment.
1207 * This required callers to themselves test the storage type and call the
1208 * right one. Rather than require callers to be aware of those layout
1209 * decisions, we centralize here.
1210 */
1211 static __inline void
m_align(struct mbuf * m,int len)1212 m_align(struct mbuf *m, int len)
1213 {
1214 int adjust;
1215 KASSERT(m->m_data == M_START(m),
1216 ("%s: not a virgin mbuf %p", __func__, m));
1217
1218 adjust = M_SIZE(m) - len;
1219 m->m_data += adjust &~ (sizeof(long)-1);
1220 }
1221
1222 #define M_ALIGN(m, len) m_align(m, len)
1223 #define MH_ALIGN(m, len) m_align(m, len)
1224 #define MEXT_ALIGN(m, len) m_align(m, len)
1225
1226 /*
1227 * Compute the amount of space available before the current start of data in
1228 * an mbuf.
1229 *
1230 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1231 * of checking writability of the mbuf data area rests solely with the caller.
1232 *
1233 * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE()
1234 * for mbufs with external storage. We now allow mbuf-embedded data to be
1235 * read-only as well.
1236 */
1237 #define M_LEADINGSPACE(m) \
1238 (M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0)
1239
1240 /*
1241 * So M_TRAILINGROOM() is for when you want to know how much space
1242 * would be there if it was writable. This can be used to
1243 * detect changes in mbufs by knowing the value at one point
1244 * and then being able to compare it later to the current M_TRAILINGROOM().
1245 * The TRAILINGSPACE() macro is not suitable for this since an mbuf
1246 * at one point might not be writable and then later it becomes writable
1247 * even though the space at the back of it has not changed.
1248 */
1249 #define M_TRAILINGROOM(m) ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len))
1250 /*
1251 * Compute the amount of space available after the end of data in an mbuf.
1252 *
1253 * The M_WRITABLE() is a temporary, conservative safety measure: the burden
1254 * of checking writability of the mbuf data area rests solely with the caller.
1255 *
1256 * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE()
1257 * for mbufs with external storage. We now allow mbuf-embedded data to be
1258 * read-only as well.
1259 */
1260 #define M_TRAILINGSPACE(m) (M_WRITABLE(m) ? M_TRAILINGROOM(m) : 0)
1261
1262 /*
1263 * Arrange to prepend space of size plen to mbuf m. If a new mbuf must be
1264 * allocated, how specifies whether to wait. If the allocation fails, the
1265 * original mbuf chain is freed and m is set to NULL.
1266 */
1267 #define M_PREPEND(m, plen, how) do { \
1268 struct mbuf **_mmp = &(m); \
1269 struct mbuf *_mm = *_mmp; \
1270 int _mplen = (plen); \
1271 int __mhow = (how); \
1272 \
1273 MBUF_CHECKSLEEP(how); \
1274 if (M_LEADINGSPACE(_mm) >= _mplen) { \
1275 _mm->m_data -= _mplen; \
1276 _mm->m_len += _mplen; \
1277 } else \
1278 _mm = m_prepend(_mm, _mplen, __mhow); \
1279 if (_mm != NULL && _mm->m_flags & M_PKTHDR) \
1280 _mm->m_pkthdr.len += _mplen; \
1281 *_mmp = _mm; \
1282 } while (0)
1283
1284 /*
1285 * Change mbuf to new type. This is a relatively expensive operation and
1286 * should be avoided.
1287 */
1288 #define MCHTYPE(m, t) m_chtype((m), (t))
1289
1290 /* Return the rcvif of a packet header. */
1291 static __inline struct ifnet *
m_rcvif(struct mbuf * m)1292 m_rcvif(struct mbuf *m)
1293 {
1294
1295 M_ASSERTPKTHDR(m);
1296 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1297 return (NULL);
1298 return (m->m_pkthdr.rcvif);
1299 }
1300
1301 /* Length to m_copy to copy all. */
1302 #define M_COPYALL 1000000000
1303
1304 extern u_int max_linkhdr; /* Largest link-level header */
1305 extern u_int max_hdr; /* Largest link + protocol header */
1306 extern u_int max_protohdr; /* Largest protocol header */
1307 void max_linkhdr_grow(u_int);
1308 void max_protohdr_grow(u_int);
1309
1310 extern int nmbclusters; /* Maximum number of clusters */
1311 extern bool mb_use_ext_pgs; /* Use ext_pgs for sendfile */
1312
1313 /*-
1314 * Network packets may have annotations attached by affixing a list of
1315 * "packet tags" to the pkthdr structure. Packet tags are dynamically
1316 * allocated semi-opaque data structures that have a fixed header
1317 * (struct m_tag) that specifies the size of the memory block and a
1318 * <cookie,type> pair that identifies it. The cookie is a 32-bit unique
1319 * unsigned value used to identify a module or ABI. By convention this value
1320 * is chosen as the date+time that the module is created, expressed as the
1321 * number of seconds since the epoch (e.g., using date -u +'%s'). The type
1322 * value is an ABI/module-specific value that identifies a particular
1323 * annotation and is private to the module. For compatibility with systems
1324 * like OpenBSD that define packet tags w/o an ABI/module cookie, the value
1325 * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find
1326 * compatibility shim functions and several tag types are defined below.
1327 * Users that do not require compatibility should use a private cookie value
1328 * so that packet tag-related definitions can be maintained privately.
1329 *
1330 * Note that the packet tag returned by m_tag_alloc has the default memory
1331 * alignment implemented by malloc. To reference private data one can use a
1332 * construct like:
1333 *
1334 * struct m_tag *mtag = m_tag_alloc(...);
1335 * struct foo *p = (struct foo *)(mtag+1);
1336 *
1337 * if the alignment of struct m_tag is sufficient for referencing members of
1338 * struct foo. Otherwise it is necessary to embed struct m_tag within the
1339 * private data structure to insure proper alignment; e.g.,
1340 *
1341 * struct foo {
1342 * struct m_tag tag;
1343 * ...
1344 * };
1345 * struct foo *p = (struct foo *) m_tag_alloc(...);
1346 * struct m_tag *mtag = &p->tag;
1347 */
1348
1349 /*
1350 * Persistent tags stay with an mbuf until the mbuf is reclaimed. Otherwise
1351 * tags are expected to ``vanish'' when they pass through a network
1352 * interface. For most interfaces this happens normally as the tags are
1353 * reclaimed when the mbuf is free'd. However in some special cases
1354 * reclaiming must be done manually. An example is packets that pass through
1355 * the loopback interface. Also, one must be careful to do this when
1356 * ``turning around'' packets (e.g., icmp_reflect).
1357 *
1358 * To mark a tag persistent bit-or this flag in when defining the tag id.
1359 * The tag will then be treated as described above.
1360 */
1361 #define MTAG_PERSISTENT 0x800
1362
1363 #define PACKET_TAG_NONE 0 /* Nadda */
1364
1365 /* Packet tags for use with PACKET_ABI_COMPAT. */
1366 #define PACKET_TAG_IPSEC_IN_DONE 1 /* IPsec applied, in */
1367 #define PACKET_TAG_IPSEC_OUT_DONE 2 /* IPsec applied, out */
1368 #define PACKET_TAG_IPSEC_IN_CRYPTO_DONE 3 /* NIC IPsec crypto done */
1369 #define PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED 4 /* NIC IPsec crypto req'ed */
1370 #define PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO 5 /* NIC notifies IPsec */
1371 #define PACKET_TAG_IPSEC_PENDING_TDB 6 /* Reminder to do IPsec */
1372 #define PACKET_TAG_BRIDGE 7 /* Bridge processing done */
1373 #define PACKET_TAG_GIF 8 /* GIF processing done */
1374 #define PACKET_TAG_GRE 9 /* GRE processing done */
1375 #define PACKET_TAG_IN_PACKET_CHECKSUM 10 /* NIC checksumming done */
1376 #define PACKET_TAG_ENCAP 11 /* Encap. processing */
1377 #define PACKET_TAG_IPSEC_SOCKET 12 /* IPSEC socket ref */
1378 #define PACKET_TAG_IPSEC_HISTORY 13 /* IPSEC history */
1379 #define PACKET_TAG_IPV6_INPUT 14 /* IPV6 input processing */
1380 #define PACKET_TAG_DUMMYNET 15 /* dummynet info */
1381 #define PACKET_TAG_DIVERT 17 /* divert info */
1382 #define PACKET_TAG_IPFORWARD 18 /* ipforward info */
1383 #define PACKET_TAG_MACLABEL (19 | MTAG_PERSISTENT) /* MAC label */
1384 #define PACKET_TAG_PF 21 /* PF/ALTQ information */
1385 /* was PACKET_TAG_RTSOCKFAM 25 rtsock sa family */
1386 #define PACKET_TAG_IPOPTIONS 27 /* Saved IP options */
1387 #define PACKET_TAG_CARP 28 /* CARP info */
1388 #define PACKET_TAG_IPSEC_NAT_T_PORTS 29 /* two uint16_t */
1389 #define PACKET_TAG_ND_OUTGOING 30 /* ND outgoing */
1390 #define PACKET_TAG_PF_REASSEMBLED 31
1391 #define PACKET_TAG_IPSEC_ACCEL_OUT 32 /* IPSEC accel out */
1392 #define PACKET_TAG_IPSEC_ACCEL_IN 33 /* IPSEC accel in */
1393 #define PACKET_TAG_OVPN 34 /* if_ovpn */
1394
1395 /* Specific cookies and tags. */
1396
1397 /* Packet tag routines. */
1398 struct m_tag *m_tag_alloc(uint32_t, uint16_t, int, int);
1399 void m_tag_delete(struct mbuf *, struct m_tag *);
1400 void m_tag_delete_chain(struct mbuf *, struct m_tag *);
1401 void m_tag_free_default(struct m_tag *);
1402 struct m_tag *m_tag_locate(struct mbuf *, uint32_t, uint16_t,
1403 struct m_tag *);
1404 struct m_tag *m_tag_copy(struct m_tag *, int);
1405 int m_tag_copy_chain(struct mbuf *, const struct mbuf *, int);
1406 void m_tag_delete_nonpersistent(struct mbuf *);
1407
1408 /*
1409 * Initialize the list of tags associated with an mbuf.
1410 */
1411 static __inline void
m_tag_init(struct mbuf * m)1412 m_tag_init(struct mbuf *m)
1413 {
1414
1415 SLIST_INIT(&m->m_pkthdr.tags);
1416 }
1417
1418 /*
1419 * Set up the contents of a tag. Note that this does not fill in the free
1420 * method; the caller is expected to do that.
1421 *
1422 * XXX probably should be called m_tag_init, but that was already taken.
1423 */
1424 static __inline void
m_tag_setup(struct m_tag * t,uint32_t cookie,uint16_t type,int len)1425 m_tag_setup(struct m_tag *t, uint32_t cookie, uint16_t type, int len)
1426 {
1427
1428 t->m_tag_id = type;
1429 t->m_tag_len = len;
1430 t->m_tag_cookie = cookie;
1431 }
1432
1433 /*
1434 * Reclaim resources associated with a tag.
1435 */
1436 static __inline void
m_tag_free(struct m_tag * t)1437 m_tag_free(struct m_tag *t)
1438 {
1439
1440 (*t->m_tag_free)(t);
1441 }
1442
1443 /*
1444 * Return the first tag associated with an mbuf.
1445 */
1446 static __inline struct m_tag *
m_tag_first(struct mbuf * m)1447 m_tag_first(struct mbuf *m)
1448 {
1449
1450 return (SLIST_FIRST(&m->m_pkthdr.tags));
1451 }
1452
1453 /*
1454 * Return the next tag in the list of tags associated with an mbuf.
1455 */
1456 static __inline struct m_tag *
m_tag_next(struct mbuf * m __unused,struct m_tag * t)1457 m_tag_next(struct mbuf *m __unused, struct m_tag *t)
1458 {
1459
1460 return (SLIST_NEXT(t, m_tag_link));
1461 }
1462
1463 /*
1464 * Prepend a tag to the list of tags associated with an mbuf.
1465 */
1466 static __inline void
m_tag_prepend(struct mbuf * m,struct m_tag * t)1467 m_tag_prepend(struct mbuf *m, struct m_tag *t)
1468 {
1469
1470 SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link);
1471 }
1472
1473 /*
1474 * Unlink a tag from the list of tags associated with an mbuf.
1475 */
1476 static __inline void
m_tag_unlink(struct mbuf * m,struct m_tag * t)1477 m_tag_unlink(struct mbuf *m, struct m_tag *t)
1478 {
1479
1480 SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link);
1481 }
1482
1483 /* These are for OpenBSD compatibility. */
1484 #define MTAG_ABI_COMPAT 0 /* compatibility ABI */
1485
1486 static __inline struct m_tag *
m_tag_get(uint16_t type,int length,int wait)1487 m_tag_get(uint16_t type, int length, int wait)
1488 {
1489 return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait));
1490 }
1491
1492 static __inline struct m_tag *
m_tag_find(struct mbuf * m,uint16_t type,struct m_tag * start)1493 m_tag_find(struct mbuf *m, uint16_t type, struct m_tag *start)
1494 {
1495 return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL :
1496 m_tag_locate(m, MTAG_ABI_COMPAT, type, start));
1497 }
1498
1499 static inline struct m_snd_tag *
m_snd_tag_ref(struct m_snd_tag * mst)1500 m_snd_tag_ref(struct m_snd_tag *mst)
1501 {
1502
1503 refcount_acquire(&mst->refcount);
1504 return (mst);
1505 }
1506
1507 static inline void
m_snd_tag_rele(struct m_snd_tag * mst)1508 m_snd_tag_rele(struct m_snd_tag *mst)
1509 {
1510
1511 if (refcount_release(&mst->refcount))
1512 m_snd_tag_destroy(mst);
1513 }
1514
1515 static __inline struct mbuf *
m_free(struct mbuf * m)1516 m_free(struct mbuf *m)
1517 {
1518 struct mbuf *n = m->m_next;
1519
1520 MBUF_PROBE1(m__free, m);
1521 if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE))
1522 m_tag_delete_chain(m, NULL);
1523 if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG)
1524 m_snd_tag_rele(m->m_pkthdr.snd_tag);
1525 if (m->m_flags & M_EXTPG)
1526 mb_free_extpg(m);
1527 else if (m->m_flags & M_EXT)
1528 mb_free_ext(m);
1529 else if ((m->m_flags & M_NOFREE) == 0)
1530 uma_zfree(zone_mbuf, m);
1531 return (n);
1532 }
1533
1534 static __inline int
rt_m_getfib(struct mbuf * m)1535 rt_m_getfib(struct mbuf *m)
1536 {
1537 KASSERT(m->m_flags & M_PKTHDR,
1538 ("%s: Attempt to get FIB from non header mbuf %p", __func__, m));
1539 return (m->m_pkthdr.fibnum);
1540 }
1541
1542 #define M_GETFIB(_m) rt_m_getfib(_m)
1543
1544 #define M_SETFIB(_m, _fib) do { \
1545 KASSERT((_m)->m_flags & M_PKTHDR, \
1546 ("%s: Attempt to set FIB on non header mbuf %p", __func__, (_m))); \
1547 ((_m)->m_pkthdr.fibnum) = (_fib); \
1548 } while (0)
1549
1550 /* flags passed as first argument for "m_xxx_tcpip_hash()" */
1551 #define MBUF_HASHFLAG_L2 (1 << 2)
1552 #define MBUF_HASHFLAG_L3 (1 << 3)
1553 #define MBUF_HASHFLAG_L4 (1 << 4)
1554
1555 /* mbuf hashing helper routines */
1556 uint32_t m_ether_tcpip_hash_init(void);
1557 uint32_t m_ether_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1558 uint32_t m_infiniband_tcpip_hash_init(void);
1559 uint32_t m_infiniband_tcpip_hash(const uint32_t, const struct mbuf *, uint32_t);
1560
1561 #ifdef MBUF_PROFILING
1562 void m_profile(struct mbuf *m);
1563 #define M_PROFILE(m) m_profile(m)
1564 #else
1565 #define M_PROFILE(m)
1566 #endif
1567
1568 /*
1569 * Structure describing a packet queue: mbufs linked by m_stailqpkt.
1570 * Does accounting of number of packets and has a cap.
1571 */
1572 struct mbufq {
1573 STAILQ_HEAD(, mbuf) mq_head;
1574 int mq_len;
1575 int mq_maxlen;
1576 };
1577
1578 static inline void
mbufq_init(struct mbufq * mq,int maxlen)1579 mbufq_init(struct mbufq *mq, int maxlen)
1580 {
1581
1582 STAILQ_INIT(&mq->mq_head);
1583 mq->mq_maxlen = maxlen;
1584 mq->mq_len = 0;
1585 }
1586
1587 static inline struct mbuf *
mbufq_flush(struct mbufq * mq)1588 mbufq_flush(struct mbufq *mq)
1589 {
1590 struct mbuf *m;
1591
1592 m = STAILQ_FIRST(&mq->mq_head);
1593 STAILQ_INIT(&mq->mq_head);
1594 mq->mq_len = 0;
1595 return (m);
1596 }
1597
1598 static inline void
mbufq_drain(struct mbufq * mq)1599 mbufq_drain(struct mbufq *mq)
1600 {
1601 struct mbuf *m, *n;
1602
1603 n = mbufq_flush(mq);
1604 while ((m = n) != NULL) {
1605 n = STAILQ_NEXT(m, m_stailqpkt);
1606 m_freem(m);
1607 }
1608 }
1609
1610 static inline struct mbuf *
mbufq_first(const struct mbufq * mq)1611 mbufq_first(const struct mbufq *mq)
1612 {
1613
1614 return (STAILQ_FIRST(&mq->mq_head));
1615 }
1616
1617 static inline struct mbuf *
mbufq_last(const struct mbufq * mq)1618 mbufq_last(const struct mbufq *mq)
1619 {
1620
1621 return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt));
1622 }
1623
1624 static inline bool
mbufq_empty(const struct mbufq * mq)1625 mbufq_empty(const struct mbufq *mq)
1626 {
1627 return (mq->mq_len == 0);
1628 }
1629
1630 static inline int
mbufq_full(const struct mbufq * mq)1631 mbufq_full(const struct mbufq *mq)
1632 {
1633
1634 return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen);
1635 }
1636
1637 static inline int
mbufq_len(const struct mbufq * mq)1638 mbufq_len(const struct mbufq *mq)
1639 {
1640
1641 return (mq->mq_len);
1642 }
1643
1644 static inline int
mbufq_enqueue(struct mbufq * mq,struct mbuf * m)1645 mbufq_enqueue(struct mbufq *mq, struct mbuf *m)
1646 {
1647
1648 if (mbufq_full(mq))
1649 return (ENOBUFS);
1650 STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt);
1651 mq->mq_len++;
1652 return (0);
1653 }
1654
1655 static inline struct mbuf *
mbufq_dequeue(struct mbufq * mq)1656 mbufq_dequeue(struct mbufq *mq)
1657 {
1658 struct mbuf *m;
1659
1660 m = STAILQ_FIRST(&mq->mq_head);
1661 if (m) {
1662 STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt);
1663 m->m_nextpkt = NULL;
1664 mq->mq_len--;
1665 }
1666 return (m);
1667 }
1668
1669 static inline void
mbufq_prepend(struct mbufq * mq,struct mbuf * m)1670 mbufq_prepend(struct mbufq *mq, struct mbuf *m)
1671 {
1672
1673 STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt);
1674 mq->mq_len++;
1675 }
1676
1677 /*
1678 * Note: this doesn't enforce the maximum list size for dst.
1679 */
1680 static inline void
mbufq_concat(struct mbufq * mq_dst,struct mbufq * mq_src)1681 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src)
1682 {
1683
1684 mq_dst->mq_len += mq_src->mq_len;
1685 STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head);
1686 mq_src->mq_len = 0;
1687 }
1688
1689 /*
1690 * Structure describing a chain of mbufs linked by m_stailq, also tracking
1691 * the pointer to the last. Also does accounting of data length and memory
1692 * usage.
1693 * To be used as an argument to mbuf chain allocation and manipulation KPIs,
1694 * and can be allocated on the stack of a caller. Kernel facilities may use
1695 * it internally as a most simple implementation of a stream data buffer.
1696 */
1697 struct mchain {
1698 STAILQ_HEAD(, mbuf) mc_q;
1699 u_int mc_len;
1700 u_int mc_mlen;
1701 };
1702
1703 #define MCHAIN_INITIALIZER(mc) \
1704 (struct mchain){ .mc_q = STAILQ_HEAD_INITIALIZER((mc)->mc_q) }
1705
1706 static inline struct mbuf *
mc_first(struct mchain * mc)1707 mc_first(struct mchain *mc)
1708 {
1709 return (STAILQ_FIRST(&mc->mc_q));
1710 }
1711
1712 static inline struct mbuf *
mc_last(struct mchain * mc)1713 mc_last(struct mchain *mc)
1714 {
1715 return (STAILQ_LAST(&mc->mc_q, mbuf, m_stailq));
1716 }
1717
1718 static inline bool
mc_empty(struct mchain * mc)1719 mc_empty(struct mchain *mc)
1720 {
1721 return (STAILQ_EMPTY(&mc->mc_q));
1722 }
1723
1724 /* Account addition of m to mc. */
1725 static inline void
mc_inc(struct mchain * mc,struct mbuf * m)1726 mc_inc(struct mchain *mc, struct mbuf *m)
1727 {
1728 mc->mc_len += m->m_len;
1729 mc->mc_mlen += MSIZE;
1730 if (m->m_flags & M_EXT)
1731 mc->mc_mlen += m->m_ext.ext_size;
1732 }
1733
1734 /* Account removal of m from mc. */
1735 static inline void
mc_dec(struct mchain * mc,struct mbuf * m)1736 mc_dec(struct mchain *mc, struct mbuf *m)
1737 {
1738 MPASS(mc->mc_len >= m->m_len);
1739 mc->mc_len -= m->m_len;
1740 MPASS(mc->mc_mlen >= MSIZE);
1741 mc->mc_mlen -= MSIZE;
1742 if (m->m_flags & M_EXT) {
1743 MPASS(mc->mc_mlen >= m->m_ext.ext_size);
1744 mc->mc_mlen -= m->m_ext.ext_size;
1745 }
1746 }
1747
1748 /*
1749 * Get mchain from a classic mbuf chain linked by m_next. Two hacks here:
1750 * we use the fact that m_next is alias to m_stailq, we use internal queue(3)
1751 * fields.
1752 */
1753 static inline void
mc_init_m(struct mchain * mc,struct mbuf * m)1754 mc_init_m(struct mchain *mc, struct mbuf *m)
1755 {
1756 struct mbuf *last;
1757
1758 STAILQ_FIRST(&mc->mc_q) = m;
1759 mc->mc_len = mc->mc_mlen = 0;
1760 STAILQ_FOREACH(m, &mc->mc_q, m_stailq) {
1761 mc_inc(mc, m);
1762 last = m;
1763 }
1764 mc->mc_q.stqh_last = &STAILQ_NEXT(last, m_stailq);
1765 }
1766
1767 static inline void
mc_freem(struct mchain * mc)1768 mc_freem(struct mchain *mc)
1769 {
1770 if (!mc_empty(mc))
1771 m_freem(mc_first(mc));
1772 }
1773
1774 static inline void
mc_prepend(struct mchain * mc,struct mbuf * m)1775 mc_prepend(struct mchain *mc, struct mbuf *m)
1776 {
1777 STAILQ_INSERT_HEAD(&mc->mc_q, m, m_stailq);
1778 mc_inc(mc, m);
1779 }
1780
1781 static inline void
mc_append(struct mchain * mc,struct mbuf * m)1782 mc_append(struct mchain *mc, struct mbuf *m)
1783 {
1784 STAILQ_INSERT_TAIL(&mc->mc_q, m, m_stailq);
1785 mc_inc(mc, m);
1786 }
1787
1788 static inline void
mc_concat(struct mchain * head,struct mchain * tail)1789 mc_concat(struct mchain *head, struct mchain *tail)
1790 {
1791 STAILQ_CONCAT(&head->mc_q, &tail->mc_q);
1792 head->mc_len += tail->mc_len;
1793 head->mc_mlen += tail->mc_mlen;
1794 tail->mc_len = tail->mc_mlen = 0;
1795 }
1796
1797 /*
1798 * Note: STAILQ_REMOVE() is expensive. mc_remove_after() needs to be provided
1799 * as long as there consumers that would benefit from it.
1800 */
1801 static inline void
mc_remove(struct mchain * mc,struct mbuf * m)1802 mc_remove(struct mchain *mc, struct mbuf *m)
1803 {
1804 STAILQ_REMOVE(&mc->mc_q, m, mbuf, m_stailq);
1805 mc_dec(mc, m);
1806 }
1807
1808 int mc_get(struct mchain *, u_int, int, short, int);
1809 int mc_split(struct mchain *, struct mchain *, u_int, int);
1810 int mc_uiotomc(struct mchain *, struct uio *, u_int, u_int, int, int);
1811
1812 #ifdef _SYS_TIMESPEC_H_
1813 static inline void
mbuf_tstmp2timespec(struct mbuf * m,struct timespec * ts)1814 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts)
1815 {
1816
1817 M_ASSERTPKTHDR(m);
1818 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1819 ("%s: mbuf %p no M_TSTMP or M_TSTMP_LRO", __func__, m));
1820 ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1821 ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000;
1822 }
1823 #endif
1824
1825 static inline void
mbuf_tstmp2timeval(struct mbuf * m,struct timeval * tv)1826 mbuf_tstmp2timeval(struct mbuf *m, struct timeval *tv)
1827 {
1828
1829 M_ASSERTPKTHDR(m);
1830 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0,
1831 ("%s: mbuf %p no M_TSTMP or M_TSTMP_LRO", __func__, m));
1832 tv->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000;
1833 tv->tv_usec = (m->m_pkthdr.rcv_tstmp % 1000000000) / 1000;
1834 }
1835
1836 #ifdef DEBUGNET
1837 /* Invoked from the debugnet client code. */
1838 void debugnet_mbuf_drain(void);
1839 void debugnet_mbuf_start(void);
1840 void debugnet_mbuf_finish(void);
1841 void debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize);
1842 #endif
1843
1844 static inline bool
mbuf_has_tls_session(struct mbuf * m)1845 mbuf_has_tls_session(struct mbuf *m)
1846 {
1847
1848 if (m->m_flags & M_EXTPG) {
1849 if (m->m_epg_tls != NULL) {
1850 return (true);
1851 }
1852 }
1853 return (false);
1854 }
1855
1856 #endif /* _KERNEL */
1857 #endif /* !_SYS_MBUF_H_ */
1858