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 * @(#)mbuf.h 8.5 (Berkeley) 2/19/95 33 * $FreeBSD$ 34 */ 35 36 #ifndef _SYS_MBUF_H_ 37 #define _SYS_MBUF_H_ 38 39 /* XXX: These includes suck. Sorry! */ 40 #include <sys/queue.h> 41 #ifdef _KERNEL 42 #include <sys/systm.h> 43 #include <sys/refcount.h> 44 #include <vm/uma.h> 45 #ifdef WITNESS 46 #include <sys/lock.h> 47 #endif 48 #endif 49 50 #ifdef _KERNEL 51 #include <sys/sdt.h> 52 53 #define MBUF_PROBE1(probe, arg0) \ 54 SDT_PROBE1(sdt, , , probe, arg0) 55 #define MBUF_PROBE2(probe, arg0, arg1) \ 56 SDT_PROBE2(sdt, , , probe, arg0, arg1) 57 #define MBUF_PROBE3(probe, arg0, arg1, arg2) \ 58 SDT_PROBE3(sdt, , , probe, arg0, arg1, arg2) 59 #define MBUF_PROBE4(probe, arg0, arg1, arg2, arg3) \ 60 SDT_PROBE4(sdt, , , probe, arg0, arg1, arg2, arg3) 61 #define MBUF_PROBE5(probe, arg0, arg1, arg2, arg3, arg4) \ 62 SDT_PROBE5(sdt, , , probe, arg0, arg1, arg2, arg3, arg4) 63 64 SDT_PROBE_DECLARE(sdt, , , m__init); 65 SDT_PROBE_DECLARE(sdt, , , m__gethdr); 66 SDT_PROBE_DECLARE(sdt, , , m__get); 67 SDT_PROBE_DECLARE(sdt, , , m__getcl); 68 SDT_PROBE_DECLARE(sdt, , , m__clget); 69 SDT_PROBE_DECLARE(sdt, , , m__cljget); 70 SDT_PROBE_DECLARE(sdt, , , m__cljset); 71 SDT_PROBE_DECLARE(sdt, , , m__free); 72 SDT_PROBE_DECLARE(sdt, , , m__freem); 73 74 #endif /* _KERNEL */ 75 76 /* 77 * Mbufs are of a single size, MSIZE (sys/param.h), which includes overhead. 78 * An mbuf may add a single "mbuf cluster" of size MCLBYTES (also in 79 * sys/param.h), which has no additional overhead and is used instead of the 80 * internal data area; this is done when at least MINCLSIZE of data must be 81 * stored. Additionally, it is possible to allocate a separate buffer 82 * externally and attach it to the mbuf in a way similar to that of mbuf 83 * clusters. 84 * 85 * NB: These calculation do not take actual compiler-induced alignment and 86 * padding inside the complete struct mbuf into account. Appropriate 87 * attention is required when changing members of struct mbuf. 88 * 89 * MLEN is data length in a normal mbuf. 90 * MHLEN is data length in an mbuf with pktheader. 91 * MINCLSIZE is a smallest amount of data that should be put into cluster. 92 * 93 * Compile-time assertions in uipc_mbuf.c test these values to ensure that 94 * they are sensible. 95 */ 96 struct mbuf; 97 #define MHSIZE offsetof(struct mbuf, m_dat) 98 #define MPKTHSIZE offsetof(struct mbuf, m_pktdat) 99 #define MLEN ((int)(MSIZE - MHSIZE)) 100 #define MHLEN ((int)(MSIZE - MPKTHSIZE)) 101 #define MINCLSIZE (MHLEN + 1) 102 #define M_NODOM 255 103 104 #ifdef _KERNEL 105 /*- 106 * Macro for type conversion: convert mbuf pointer to data pointer of correct 107 * type: 108 * 109 * mtod(m, t) -- Convert mbuf pointer to data pointer of correct type. 110 * mtodo(m, o) -- Same as above but with offset 'o' into data. 111 */ 112 #define mtod(m, t) ((t)((m)->m_data)) 113 #define mtodo(m, o) ((void *)(((m)->m_data) + (o))) 114 115 /* 116 * Argument structure passed to UMA routines during mbuf and packet 117 * allocations. 118 */ 119 struct mb_args { 120 int flags; /* Flags for mbuf being allocated */ 121 short type; /* Type of mbuf being allocated */ 122 }; 123 #endif /* _KERNEL */ 124 125 /* 126 * Packet tag structure (see below for details). 127 */ 128 struct m_tag { 129 SLIST_ENTRY(m_tag) m_tag_link; /* List of packet tags */ 130 u_int16_t m_tag_id; /* Tag ID */ 131 u_int16_t m_tag_len; /* Length of data */ 132 u_int32_t m_tag_cookie; /* ABI/Module ID */ 133 void (*m_tag_free)(struct m_tag *); 134 }; 135 136 /* 137 * Static network interface owned tag. 138 * Allocated through ifp->if_snd_tag_alloc(). 139 */ 140 struct m_snd_tag { 141 struct ifnet *ifp; /* network interface tag belongs to */ 142 volatile u_int refcount; 143 }; 144 145 /* 146 * Record/packet header in first mbuf of chain; valid only if M_PKTHDR is set. 147 * Size ILP32: 48 148 * LP64: 56 149 * Compile-time assertions in uipc_mbuf.c test these values to ensure that 150 * they are correct. 151 */ 152 struct pkthdr { 153 union { 154 struct m_snd_tag *snd_tag; /* send tag, if any */ 155 struct ifnet *rcvif; /* rcv interface */ 156 }; 157 SLIST_HEAD(packet_tags, m_tag) tags; /* list of packet tags */ 158 int32_t len; /* total packet length */ 159 160 /* Layer crossing persistent information. */ 161 uint32_t flowid; /* packet's 4-tuple system */ 162 uint32_t csum_flags; /* checksum and offload features */ 163 uint16_t fibnum; /* this packet should use this fib */ 164 uint8_t numa_domain; /* NUMA domain of recvd pkt */ 165 uint8_t rsstype; /* hash type */ 166 union { 167 uint64_t rcv_tstmp; /* timestamp in ns */ 168 struct { 169 uint8_t l2hlen; /* layer 2 hdr len */ 170 uint8_t l3hlen; /* layer 3 hdr len */ 171 uint8_t l4hlen; /* layer 4 hdr len */ 172 uint8_t l5hlen; /* layer 5 hdr len */ 173 uint32_t spare; 174 }; 175 }; 176 union { 177 uint8_t eight[8]; 178 uint16_t sixteen[4]; 179 uint32_t thirtytwo[2]; 180 uint64_t sixtyfour[1]; 181 uintptr_t unintptr[1]; 182 void *ptr; 183 } PH_per; 184 185 /* Layer specific non-persistent local storage for reassembly, etc. */ 186 union { 187 uint8_t eight[8]; 188 uint16_t sixteen[4]; 189 uint32_t thirtytwo[2]; 190 uint64_t sixtyfour[1]; 191 uintptr_t unintptr[1]; 192 void *ptr; 193 } PH_loc; 194 }; 195 #define ether_vtag PH_per.sixteen[0] 196 #define PH_vt PH_per 197 #define vt_nrecs sixteen[0] /* mld and v6-ND */ 198 #define tso_segsz PH_per.sixteen[1] /* inbound after LRO */ 199 #define lro_nsegs tso_segsz /* inbound after LRO */ 200 #define csum_data PH_per.thirtytwo[1] /* inbound from hardware up */ 201 #define lro_len PH_loc.sixteen[0] /* inbound during LRO (no reassembly) */ 202 #define lro_csum PH_loc.sixteen[1] /* inbound during LRO (no reassembly) */ 203 /* Note PH_loc is used during IP reassembly (all 8 bytes as a ptr) */ 204 205 /* 206 * TLS records for TLS 1.0-1.2 can have the following header lengths: 207 * - 5 (AES-CBC with implicit IV) 208 * - 21 (AES-CBC with explicit IV) 209 * - 13 (AES-GCM with 8 byte explicit IV) 210 */ 211 #define MBUF_PEXT_HDR_LEN 23 212 213 /* 214 * TLS records for TLS 1.0-1.2 can have the following maximum trailer 215 * lengths: 216 * - 16 (AES-GCM) 217 * - 36 (AES-CBC with SHA1 and up to 16 bytes of padding) 218 * - 48 (AES-CBC with SHA2-256 and up to 16 bytes of padding) 219 * - 64 (AES-CBC with SHA2-384 and up to 16 bytes of padding) 220 */ 221 #define MBUF_PEXT_TRAIL_LEN 64 222 223 #if defined(__LP64__) 224 #define MBUF_PEXT_MAX_PGS (40 / sizeof(vm_paddr_t)) 225 #else 226 #define MBUF_PEXT_MAX_PGS (72 / sizeof(vm_paddr_t)) 227 #endif 228 229 #define MBUF_PEXT_MAX_BYTES \ 230 (MBUF_PEXT_MAX_PGS * PAGE_SIZE + MBUF_PEXT_HDR_LEN + MBUF_PEXT_TRAIL_LEN) 231 232 struct ktls_session; 233 struct socket; 234 235 /* 236 * Description of external storage mapped into mbuf; valid only if M_EXT is 237 * set. 238 * Size ILP32: 28 239 * LP64: 48 240 * Compile-time assertions in uipc_mbuf.c test these values to ensure that 241 * they are correct. 242 */ 243 typedef void m_ext_free_t(struct mbuf *); 244 struct m_ext { 245 union { 246 /* 247 * If EXT_FLAG_EMBREF is set, then we use refcount in the 248 * mbuf, the 'ext_count' member. Otherwise, we have a 249 * shadow copy and we use pointer 'ext_cnt'. The original 250 * mbuf is responsible to carry the pointer to free routine 251 * and its arguments. They aren't copied into shadows in 252 * mb_dupcl() to avoid dereferencing next cachelines. 253 */ 254 volatile u_int ext_count; 255 volatile u_int *ext_cnt; 256 }; 257 uint32_t ext_size; /* size of buffer, for ext_free */ 258 uint32_t ext_type:8, /* type of external storage */ 259 ext_flags:24; /* external storage mbuf flags */ 260 union { 261 struct { 262 /* 263 * Regular M_EXT mbuf: 264 * o ext_buf always points to the external buffer. 265 * o ext_free (below) and two optional arguments 266 * ext_arg1 and ext_arg2 store the free context for 267 * the external storage. They are set only in the 268 * refcount carrying mbuf, the one with 269 * EXT_FLAG_EMBREF flag, with exclusion for 270 * EXT_EXTREF type, where the free context is copied 271 * into all mbufs that use same external storage. 272 */ 273 char *ext_buf; /* start of buffer */ 274 #define m_ext_copylen offsetof(struct m_ext, ext_arg2) 275 void *ext_arg2; 276 }; 277 struct { 278 /* 279 * Multi-page M_EXTPG mbuf: 280 * o extpg_pa - page vector. 281 * o extpg_trail and extpg_hdr - TLS trailer and 282 * header. 283 * Uses ext_free and may also use ext_arg1. 284 */ 285 vm_paddr_t extpg_pa[MBUF_PEXT_MAX_PGS]; 286 char extpg_trail[MBUF_PEXT_TRAIL_LEN]; 287 char extpg_hdr[MBUF_PEXT_HDR_LEN]; 288 /* Pretend these 3 fields are part of mbuf itself. */ 289 #define m_epg_pa m_ext.extpg_pa 290 #define m_epg_trail m_ext.extpg_trail 291 #define m_epg_hdr m_ext.extpg_hdr 292 #define m_epg_ext_copylen offsetof(struct m_ext, ext_free) 293 }; 294 }; 295 /* 296 * Free method and optional argument pointer, both 297 * used by M_EXT and M_EXTPG. 298 */ 299 m_ext_free_t *ext_free; 300 void *ext_arg1; 301 }; 302 303 /* 304 * The core of the mbuf object along with some shortcut defines for practical 305 * purposes. 306 */ 307 struct mbuf { 308 /* 309 * Header present at the beginning of every mbuf. 310 * Size ILP32: 24 311 * LP64: 32 312 * Compile-time assertions in uipc_mbuf.c test these values to ensure 313 * that they are correct. 314 */ 315 union { /* next buffer in chain */ 316 struct mbuf *m_next; 317 SLIST_ENTRY(mbuf) m_slist; 318 STAILQ_ENTRY(mbuf) m_stailq; 319 }; 320 union { /* next chain in queue/record */ 321 struct mbuf *m_nextpkt; 322 SLIST_ENTRY(mbuf) m_slistpkt; 323 STAILQ_ENTRY(mbuf) m_stailqpkt; 324 }; 325 caddr_t m_data; /* location of data */ 326 int32_t m_len; /* amount of data in this mbuf */ 327 uint32_t m_type:8, /* type of data in this mbuf */ 328 m_flags:24; /* flags; see below */ 329 #if !defined(__LP64__) 330 uint32_t m_pad; /* pad for 64bit alignment */ 331 #endif 332 333 /* 334 * A set of optional headers (packet header, external storage header) 335 * and internal data storage. Historically, these arrays were sized 336 * to MHLEN (space left after a packet header) and MLEN (space left 337 * after only a regular mbuf header); they are now variable size in 338 * order to support future work on variable-size mbufs. 339 */ 340 union { 341 struct { 342 union { 343 /* M_PKTHDR set. */ 344 struct pkthdr m_pkthdr; 345 346 /* M_EXTPG set. 347 * Multi-page M_EXTPG mbuf has its meta data 348 * split between the below anonymous structure 349 * and m_ext. It carries vector of pages, 350 * optional header and trailer char vectors 351 * and pointers to socket/TLS data. 352 */ 353 #define m_epg_startcopy m_epg_npgs 354 #define m_epg_endcopy m_epg_stailq 355 struct { 356 /* Overall count of pages and count of 357 * pages with I/O pending. */ 358 uint8_t m_epg_npgs; 359 uint8_t m_epg_nrdy; 360 /* TLS header and trailer lengths. 361 * The data itself resides in m_ext. */ 362 uint8_t m_epg_hdrlen; 363 uint8_t m_epg_trllen; 364 /* Offset into 1st page and length of 365 * data in the last page. */ 366 uint16_t m_epg_1st_off; 367 uint16_t m_epg_last_len; 368 uint8_t m_epg_flags; 369 #define EPG_FLAG_ANON 0x1 /* Data can be encrypted in place. */ 370 #define EPG_FLAG_2FREE 0x2 /* Scheduled for free. */ 371 uint8_t m_epg_record_type; 372 uint8_t __spare[2]; 373 int m_epg_enc_cnt; 374 struct ktls_session *m_epg_tls; 375 struct socket *m_epg_so; 376 uint64_t m_epg_seqno; 377 STAILQ_ENTRY(mbuf) m_epg_stailq; 378 }; 379 }; 380 union { 381 /* M_EXT or M_EXTPG set. */ 382 struct m_ext m_ext; 383 /* M_PKTHDR set, neither M_EXT nor M_EXTPG. */ 384 char m_pktdat[0]; 385 }; 386 }; 387 char m_dat[0]; /* !M_PKTHDR, !M_EXT */ 388 }; 389 }; 390 391 #ifdef _KERNEL 392 static inline int 393 m_epg_pagelen(const struct mbuf *m, int pidx, int pgoff) 394 { 395 396 KASSERT(pgoff == 0 || pidx == 0, 397 ("page %d with non-zero offset %d in %p", pidx, pgoff, m)); 398 399 if (pidx == m->m_epg_npgs - 1) { 400 return (m->m_epg_last_len); 401 } else { 402 return (PAGE_SIZE - pgoff); 403 } 404 } 405 406 #ifdef INVARIANTS 407 #define MCHECK(ex, msg) KASSERT((ex), \ 408 ("Multi page mbuf %p with " #msg " at %s:%d", \ 409 m, __FILE__, __LINE__)) 410 /* 411 * NB: This expects a non-empty buffer (npgs > 0 and 412 * last_pg_len > 0). 413 */ 414 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do { \ 415 MCHECK(m->m_epg_npgs > 0, "no valid pages"); \ 416 MCHECK(m->m_epg_npgs <= nitems(m->m_epg_pa), \ 417 "too many pages"); \ 418 MCHECK(m->m_epg_nrdy <= m->m_epg_npgs, \ 419 "too many ready pages"); \ 420 MCHECK(m->m_epg_1st_off < PAGE_SIZE, \ 421 "too large page offset"); \ 422 MCHECK(m->m_epg_last_len > 0, "zero last page length"); \ 423 MCHECK(m->m_epg_last_len <= PAGE_SIZE, \ 424 "too large last page length"); \ 425 if (m->m_epg_npgs == 1) \ 426 MCHECK(m->m_epg_1st_off + \ 427 m->m_epg_last_len <= PAGE_SIZE, \ 428 "single page too large"); \ 429 MCHECK(m->m_epg_hdrlen <= sizeof(m->m_epg_hdr), \ 430 "too large header length"); \ 431 MCHECK(m->m_epg_trllen <= sizeof(m->m_epg_trail), \ 432 "too large header length"); \ 433 } while (0) 434 #else 435 #define MBUF_EXT_PGS_ASSERT_SANITY(m) do {} while (0); 436 #endif 437 #endif 438 439 /* 440 * mbuf flags of global significance and layer crossing. 441 * Those of only protocol/layer specific significance are to be mapped 442 * to M_PROTO[1-11] and cleared at layer handoff boundaries. 443 * NB: Limited to the lower 24 bits. 444 */ 445 #define M_EXT 0x00000001 /* has associated external storage */ 446 #define M_PKTHDR 0x00000002 /* start of record */ 447 #define M_EOR 0x00000004 /* end of record */ 448 #define M_RDONLY 0x00000008 /* associated data is marked read-only */ 449 #define M_BCAST 0x00000010 /* send/received as link-level broadcast */ 450 #define M_MCAST 0x00000020 /* send/received as link-level multicast */ 451 #define M_PROMISC 0x00000040 /* packet was not for us */ 452 #define M_VLANTAG 0x00000080 /* ether_vtag is valid */ 453 #define M_EXTPG 0x00000100 /* has array of unmapped pages and TLS */ 454 #define M_NOFREE 0x00000200 /* do not free mbuf, embedded in cluster */ 455 #define M_TSTMP 0x00000400 /* rcv_tstmp field is valid */ 456 #define M_TSTMP_HPREC 0x00000800 /* rcv_tstmp is high-prec, typically 457 hw-stamped on port (useful for IEEE 1588 458 and 802.1AS) */ 459 #define M_TSTMP_LRO 0x00001000 /* Time LRO pushed in pkt is valid in (PH_loc) */ 460 461 #define M_PROTO1 0x00002000 /* protocol-specific */ 462 #define M_PROTO2 0x00004000 /* protocol-specific */ 463 #define M_PROTO3 0x00008000 /* protocol-specific */ 464 #define M_PROTO4 0x00010000 /* protocol-specific */ 465 #define M_PROTO5 0x00020000 /* protocol-specific */ 466 #define M_PROTO6 0x00040000 /* protocol-specific */ 467 #define M_PROTO7 0x00080000 /* protocol-specific */ 468 #define M_PROTO8 0x00100000 /* protocol-specific */ 469 #define M_PROTO9 0x00200000 /* protocol-specific */ 470 #define M_PROTO10 0x00400000 /* protocol-specific */ 471 #define M_PROTO11 0x00800000 /* protocol-specific */ 472 473 #define MB_DTOR_SKIP 0x1 /* don't pollute the cache by touching a freed mbuf */ 474 475 /* 476 * Flags to purge when crossing layers. 477 */ 478 #define M_PROTOFLAGS \ 479 (M_PROTO1|M_PROTO2|M_PROTO3|M_PROTO4|M_PROTO5|M_PROTO6|M_PROTO7|M_PROTO8|\ 480 M_PROTO9|M_PROTO10|M_PROTO11) 481 482 /* 483 * Flags preserved when copying m_pkthdr. 484 */ 485 #define M_COPYFLAGS \ 486 (M_PKTHDR|M_EOR|M_RDONLY|M_BCAST|M_MCAST|M_PROMISC|M_VLANTAG|M_TSTMP| \ 487 M_TSTMP_HPREC|M_TSTMP_LRO|M_PROTOFLAGS) 488 489 /* 490 * Mbuf flag description for use with printf(9) %b identifier. 491 */ 492 #define M_FLAG_BITS \ 493 "\20\1M_EXT\2M_PKTHDR\3M_EOR\4M_RDONLY\5M_BCAST\6M_MCAST" \ 494 "\7M_PROMISC\10M_VLANTAG\11M_EXTPG\12M_NOFREE\13M_TSTMP\14M_TSTMP_HPREC\15M_TSTMP_LRO" 495 #define M_FLAG_PROTOBITS \ 496 "\16M_PROTO1\17M_PROTO2\20M_PROTO3\21M_PROTO4" \ 497 "\22M_PROTO5\23M_PROTO6\24M_PROTO7\25M_PROTO8\26M_PROTO9" \ 498 "\27M_PROTO10\28M_PROTO11" 499 #define M_FLAG_PRINTF (M_FLAG_BITS M_FLAG_PROTOBITS) 500 501 /* 502 * Network interface cards are able to hash protocol fields (such as IPv4 503 * addresses and TCP port numbers) classify packets into flows. These flows 504 * can then be used to maintain ordering while delivering packets to the OS 505 * via parallel input queues, as well as to provide a stateless affinity 506 * model. NIC drivers can pass up the hash via m->m_pkthdr.flowid, and set 507 * m_flag fields to indicate how the hash should be interpreted by the 508 * network stack. 509 * 510 * Most NICs support RSS, which provides ordering and explicit affinity, and 511 * use the hash m_flag bits to indicate what header fields were covered by 512 * the hash. M_HASHTYPE_OPAQUE and M_HASHTYPE_OPAQUE_HASH can be set by non- 513 * RSS cards or configurations that provide an opaque flow identifier, allowing 514 * for ordering and distribution without explicit affinity. Additionally, 515 * M_HASHTYPE_OPAQUE_HASH indicates that the flow identifier has hash 516 * properties. 517 * 518 * The meaning of the IPV6_EX suffix: 519 * "o Home address from the home address option in the IPv6 destination 520 * options header. If the extension header is not present, use the Source 521 * IPv6 Address. 522 * o IPv6 address that is contained in the Routing-Header-Type-2 from the 523 * associated extension header. If the extension header is not present, 524 * use the Destination IPv6 Address." 525 * Quoted from: 526 * https://docs.microsoft.com/en-us/windows-hardware/drivers/network/rss-hashing-types#ndishashipv6ex 527 */ 528 #define M_HASHTYPE_HASHPROP 0x80 /* has hash properties */ 529 #define M_HASHTYPE_HASH(t) (M_HASHTYPE_HASHPROP | (t)) 530 /* Microsoft RSS standard hash types */ 531 #define M_HASHTYPE_NONE 0 532 #define M_HASHTYPE_RSS_IPV4 M_HASHTYPE_HASH(1) /* IPv4 2-tuple */ 533 #define M_HASHTYPE_RSS_TCP_IPV4 M_HASHTYPE_HASH(2) /* TCPv4 4-tuple */ 534 #define M_HASHTYPE_RSS_IPV6 M_HASHTYPE_HASH(3) /* IPv6 2-tuple */ 535 #define M_HASHTYPE_RSS_TCP_IPV6 M_HASHTYPE_HASH(4) /* TCPv6 4-tuple */ 536 #define M_HASHTYPE_RSS_IPV6_EX M_HASHTYPE_HASH(5) /* IPv6 2-tuple + 537 * ext hdrs */ 538 #define M_HASHTYPE_RSS_TCP_IPV6_EX M_HASHTYPE_HASH(6) /* TCPv6 4-tuple + 539 * ext hdrs */ 540 #define M_HASHTYPE_RSS_UDP_IPV4 M_HASHTYPE_HASH(7) /* IPv4 UDP 4-tuple*/ 541 #define M_HASHTYPE_RSS_UDP_IPV6 M_HASHTYPE_HASH(9) /* IPv6 UDP 4-tuple*/ 542 #define M_HASHTYPE_RSS_UDP_IPV6_EX M_HASHTYPE_HASH(10)/* IPv6 UDP 4-tuple + 543 * ext hdrs */ 544 545 #define M_HASHTYPE_OPAQUE 63 /* ordering, not affinity */ 546 #define M_HASHTYPE_OPAQUE_HASH M_HASHTYPE_HASH(M_HASHTYPE_OPAQUE) 547 /* ordering+hash, not affinity*/ 548 549 #define M_HASHTYPE_CLEAR(m) ((m)->m_pkthdr.rsstype = 0) 550 #define M_HASHTYPE_GET(m) ((m)->m_pkthdr.rsstype) 551 #define M_HASHTYPE_SET(m, v) ((m)->m_pkthdr.rsstype = (v)) 552 #define M_HASHTYPE_TEST(m, v) (M_HASHTYPE_GET(m) == (v)) 553 #define M_HASHTYPE_ISHASH(m) (M_HASHTYPE_GET(m) & M_HASHTYPE_HASHPROP) 554 555 /* 556 * External mbuf storage buffer types. 557 */ 558 #define EXT_CLUSTER 1 /* mbuf cluster */ 559 #define EXT_SFBUF 2 /* sendfile(2)'s sf_buf */ 560 #define EXT_JUMBOP 3 /* jumbo cluster page sized */ 561 #define EXT_JUMBO9 4 /* jumbo cluster 9216 bytes */ 562 #define EXT_JUMBO16 5 /* jumbo cluster 16184 bytes */ 563 #define EXT_PACKET 6 /* mbuf+cluster from packet zone */ 564 #define EXT_MBUF 7 /* external mbuf reference */ 565 #define EXT_RXRING 8 /* data in NIC receive ring */ 566 567 #define EXT_VENDOR1 224 /* for vendor-internal use */ 568 #define EXT_VENDOR2 225 /* for vendor-internal use */ 569 #define EXT_VENDOR3 226 /* for vendor-internal use */ 570 #define EXT_VENDOR4 227 /* for vendor-internal use */ 571 572 #define EXT_EXP1 244 /* for experimental use */ 573 #define EXT_EXP2 245 /* for experimental use */ 574 #define EXT_EXP3 246 /* for experimental use */ 575 #define EXT_EXP4 247 /* for experimental use */ 576 577 #define EXT_NET_DRV 252 /* custom ext_buf provided by net driver(s) */ 578 #define EXT_MOD_TYPE 253 /* custom module's ext_buf type */ 579 #define EXT_DISPOSABLE 254 /* can throw this buffer away w/page flipping */ 580 #define EXT_EXTREF 255 /* has externally maintained ext_cnt ptr */ 581 582 /* 583 * Flags for external mbuf buffer types. 584 * NB: limited to the lower 24 bits. 585 */ 586 #define EXT_FLAG_EMBREF 0x000001 /* embedded ext_count */ 587 #define EXT_FLAG_EXTREF 0x000002 /* external ext_cnt, notyet */ 588 589 #define EXT_FLAG_NOFREE 0x000010 /* don't free mbuf to pool, notyet */ 590 591 #define EXT_FLAG_VENDOR1 0x010000 /* These flags are vendor */ 592 #define EXT_FLAG_VENDOR2 0x020000 /* or submodule specific, */ 593 #define EXT_FLAG_VENDOR3 0x040000 /* not used by mbuf code. */ 594 #define EXT_FLAG_VENDOR4 0x080000 /* Set/read by submodule. */ 595 596 #define EXT_FLAG_EXP1 0x100000 /* for experimental use */ 597 #define EXT_FLAG_EXP2 0x200000 /* for experimental use */ 598 #define EXT_FLAG_EXP3 0x400000 /* for experimental use */ 599 #define EXT_FLAG_EXP4 0x800000 /* for experimental use */ 600 601 /* 602 * EXT flag description for use with printf(9) %b identifier. 603 */ 604 #define EXT_FLAG_BITS \ 605 "\20\1EXT_FLAG_EMBREF\2EXT_FLAG_EXTREF\5EXT_FLAG_NOFREE" \ 606 "\21EXT_FLAG_VENDOR1\22EXT_FLAG_VENDOR2\23EXT_FLAG_VENDOR3" \ 607 "\24EXT_FLAG_VENDOR4\25EXT_FLAG_EXP1\26EXT_FLAG_EXP2\27EXT_FLAG_EXP3" \ 608 "\30EXT_FLAG_EXP4" 609 610 /* 611 * Flags indicating checksum, segmentation and other offload work to be 612 * done, or already done, by hardware or lower layers. It is split into 613 * separate inbound and outbound flags. 614 * 615 * Outbound flags that are set by upper protocol layers requesting lower 616 * layers, or ideally the hardware, to perform these offloading tasks. 617 * For outbound packets this field and its flags can be directly tested 618 * against ifnet if_hwassist. 619 */ 620 #define CSUM_IP 0x00000001 /* IP header checksum offload */ 621 #define CSUM_IP_UDP 0x00000002 /* UDP checksum offload */ 622 #define CSUM_IP_TCP 0x00000004 /* TCP checksum offload */ 623 #define CSUM_IP_SCTP 0x00000008 /* SCTP checksum offload */ 624 #define CSUM_IP_TSO 0x00000010 /* TCP segmentation offload */ 625 #define CSUM_IP_ISCSI 0x00000020 /* iSCSI checksum offload */ 626 627 #define CSUM_IP6_UDP 0x00000200 /* UDP checksum offload */ 628 #define CSUM_IP6_TCP 0x00000400 /* TCP checksum offload */ 629 #define CSUM_IP6_SCTP 0x00000800 /* SCTP checksum offload */ 630 #define CSUM_IP6_TSO 0x00001000 /* TCP segmentation offload */ 631 #define CSUM_IP6_ISCSI 0x00002000 /* iSCSI checksum offload */ 632 633 /* Inbound checksum support where the checksum was verified by hardware. */ 634 #define CSUM_L3_CALC 0x01000000 /* calculated layer 3 csum */ 635 #define CSUM_L3_VALID 0x02000000 /* checksum is correct */ 636 #define CSUM_L4_CALC 0x04000000 /* calculated layer 4 csum */ 637 #define CSUM_L4_VALID 0x08000000 /* checksum is correct */ 638 #define CSUM_L5_CALC 0x10000000 /* calculated layer 5 csum */ 639 #define CSUM_L5_VALID 0x20000000 /* checksum is correct */ 640 #define CSUM_COALESCED 0x40000000 /* contains merged segments */ 641 642 #define CSUM_SND_TAG 0x80000000 /* Packet header has send tag */ 643 644 /* 645 * CSUM flag description for use with printf(9) %b identifier. 646 */ 647 #define CSUM_BITS \ 648 "\20\1CSUM_IP\2CSUM_IP_UDP\3CSUM_IP_TCP\4CSUM_IP_SCTP\5CSUM_IP_TSO" \ 649 "\6CSUM_IP_ISCSI" \ 650 "\12CSUM_IP6_UDP\13CSUM_IP6_TCP\14CSUM_IP6_SCTP\15CSUM_IP6_TSO" \ 651 "\16CSUM_IP6_ISCSI" \ 652 "\31CSUM_L3_CALC\32CSUM_L3_VALID\33CSUM_L4_CALC\34CSUM_L4_VALID" \ 653 "\35CSUM_L5_CALC\36CSUM_L5_VALID\37CSUM_COALESCED\40CSUM_SND_TAG" 654 655 /* CSUM flags compatibility mappings. */ 656 #define CSUM_IP_CHECKED CSUM_L3_CALC 657 #define CSUM_IP_VALID CSUM_L3_VALID 658 #define CSUM_DATA_VALID CSUM_L4_VALID 659 #define CSUM_PSEUDO_HDR CSUM_L4_CALC 660 #define CSUM_SCTP_VALID CSUM_L4_VALID 661 #define CSUM_DELAY_DATA (CSUM_TCP|CSUM_UDP) 662 #define CSUM_DELAY_IP CSUM_IP /* Only v4, no v6 IP hdr csum */ 663 #define CSUM_DELAY_DATA_IPV6 (CSUM_TCP_IPV6|CSUM_UDP_IPV6) 664 #define CSUM_DATA_VALID_IPV6 CSUM_DATA_VALID 665 #define CSUM_TCP CSUM_IP_TCP 666 #define CSUM_UDP CSUM_IP_UDP 667 #define CSUM_SCTP CSUM_IP_SCTP 668 #define CSUM_TSO (CSUM_IP_TSO|CSUM_IP6_TSO) 669 #define CSUM_UDP_IPV6 CSUM_IP6_UDP 670 #define CSUM_TCP_IPV6 CSUM_IP6_TCP 671 #define CSUM_SCTP_IPV6 CSUM_IP6_SCTP 672 673 /* 674 * mbuf types describing the content of the mbuf (including external storage). 675 */ 676 #define MT_NOTMBUF 0 /* USED INTERNALLY ONLY! Object is not mbuf */ 677 #define MT_DATA 1 /* dynamic (data) allocation */ 678 #define MT_HEADER MT_DATA /* packet header, use M_PKTHDR instead */ 679 680 #define MT_VENDOR1 4 /* for vendor-internal use */ 681 #define MT_VENDOR2 5 /* for vendor-internal use */ 682 #define MT_VENDOR3 6 /* for vendor-internal use */ 683 #define MT_VENDOR4 7 /* for vendor-internal use */ 684 685 #define MT_SONAME 8 /* socket name */ 686 687 #define MT_EXP1 9 /* for experimental use */ 688 #define MT_EXP2 10 /* for experimental use */ 689 #define MT_EXP3 11 /* for experimental use */ 690 #define MT_EXP4 12 /* for experimental use */ 691 692 #define MT_CONTROL 14 /* extra-data protocol message */ 693 #define MT_EXTCONTROL 15 /* control message with externalized contents */ 694 #define MT_OOBDATA 16 /* expedited data */ 695 696 #define MT_NOINIT 255 /* Not a type but a flag to allocate 697 a non-initialized mbuf */ 698 699 /* 700 * String names of mbuf-related UMA(9) and malloc(9) types. Exposed to 701 * !_KERNEL so that monitoring tools can look up the zones with 702 * libmemstat(3). 703 */ 704 #define MBUF_MEM_NAME "mbuf" 705 #define MBUF_CLUSTER_MEM_NAME "mbuf_cluster" 706 #define MBUF_PACKET_MEM_NAME "mbuf_packet" 707 #define MBUF_JUMBOP_MEM_NAME "mbuf_jumbo_page" 708 #define MBUF_JUMBO9_MEM_NAME "mbuf_jumbo_9k" 709 #define MBUF_JUMBO16_MEM_NAME "mbuf_jumbo_16k" 710 #define MBUF_TAG_MEM_NAME "mbuf_tag" 711 #define MBUF_EXTREFCNT_MEM_NAME "mbuf_ext_refcnt" 712 #define MBUF_EXTPGS_MEM_NAME "mbuf_extpgs" 713 714 #ifdef _KERNEL 715 716 #ifdef WITNESS 717 #define MBUF_CHECKSLEEP(how) do { \ 718 if (how == M_WAITOK) \ 719 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, \ 720 "Sleeping in \"%s\"", __func__); \ 721 } while (0) 722 #else 723 #define MBUF_CHECKSLEEP(how) 724 #endif 725 726 /* 727 * Network buffer allocation API 728 * 729 * The rest of it is defined in kern/kern_mbuf.c 730 */ 731 extern uma_zone_t zone_mbuf; 732 extern uma_zone_t zone_clust; 733 extern uma_zone_t zone_pack; 734 extern uma_zone_t zone_jumbop; 735 extern uma_zone_t zone_jumbo9; 736 extern uma_zone_t zone_jumbo16; 737 extern uma_zone_t zone_extpgs; 738 739 void mb_dupcl(struct mbuf *, struct mbuf *); 740 void mb_free_ext(struct mbuf *); 741 void mb_free_extpg(struct mbuf *); 742 void mb_free_mext_pgs(struct mbuf *); 743 struct mbuf *mb_alloc_ext_pgs(int, m_ext_free_t); 744 struct mbuf *mb_alloc_ext_plus_pages(int, int); 745 struct mbuf *mb_mapped_to_unmapped(struct mbuf *, int, int, int, 746 struct mbuf **); 747 int mb_unmapped_compress(struct mbuf *m); 748 struct mbuf *mb_unmapped_to_ext(struct mbuf *m); 749 void mb_free_notready(struct mbuf *m, int count); 750 void m_adj(struct mbuf *, int); 751 int m_apply(struct mbuf *, int, int, 752 int (*)(void *, void *, u_int), void *); 753 int m_append(struct mbuf *, int, c_caddr_t); 754 void m_cat(struct mbuf *, struct mbuf *); 755 void m_catpkt(struct mbuf *, struct mbuf *); 756 int m_clget(struct mbuf *m, int how); 757 void *m_cljget(struct mbuf *m, int how, int size); 758 struct mbuf *m_collapse(struct mbuf *, int, int); 759 void m_copyback(struct mbuf *, int, int, c_caddr_t); 760 void m_copydata(const struct mbuf *, int, int, caddr_t); 761 struct mbuf *m_copym(struct mbuf *, int, int, int); 762 struct mbuf *m_copypacket(struct mbuf *, int); 763 void m_copy_pkthdr(struct mbuf *, struct mbuf *); 764 struct mbuf *m_copyup(struct mbuf *, int, int); 765 struct mbuf *m_defrag(struct mbuf *, int); 766 void m_demote_pkthdr(struct mbuf *); 767 void m_demote(struct mbuf *, int, int); 768 struct mbuf *m_devget(char *, int, int, struct ifnet *, 769 void (*)(char *, caddr_t, u_int)); 770 void m_dispose_extcontrolm(struct mbuf *m); 771 struct mbuf *m_dup(const struct mbuf *, int); 772 int m_dup_pkthdr(struct mbuf *, const struct mbuf *, int); 773 void m_extadd(struct mbuf *, char *, u_int, m_ext_free_t, 774 void *, void *, int, int); 775 u_int m_fixhdr(struct mbuf *); 776 struct mbuf *m_fragment(struct mbuf *, int, int); 777 void m_freem(struct mbuf *); 778 struct mbuf *m_get2(int, int, short, int); 779 struct mbuf *m_getjcl(int, short, int, int); 780 struct mbuf *m_getm2(struct mbuf *, int, int, short, int); 781 struct mbuf *m_getptr(struct mbuf *, int, int *); 782 u_int m_length(struct mbuf *, struct mbuf **); 783 int m_mbuftouio(struct uio *, const struct mbuf *, int); 784 int m_unmappedtouio(const struct mbuf *, int, struct uio *, int); 785 void m_move_pkthdr(struct mbuf *, struct mbuf *); 786 int m_pkthdr_init(struct mbuf *, int); 787 struct mbuf *m_prepend(struct mbuf *, int, int); 788 void m_print(const struct mbuf *, int); 789 struct mbuf *m_pulldown(struct mbuf *, int, int, int *); 790 struct mbuf *m_pullup(struct mbuf *, int); 791 int m_sanity(struct mbuf *, int); 792 struct mbuf *m_split(struct mbuf *, int, int); 793 struct mbuf *m_uiotombuf(struct uio *, int, int, int, int); 794 struct mbuf *m_unshare(struct mbuf *, int); 795 void m_snd_tag_init(struct m_snd_tag *, struct ifnet *); 796 void m_snd_tag_destroy(struct m_snd_tag *); 797 798 static __inline int 799 m_gettype(int size) 800 { 801 int type; 802 803 switch (size) { 804 case MSIZE: 805 type = EXT_MBUF; 806 break; 807 case MCLBYTES: 808 type = EXT_CLUSTER; 809 break; 810 #if MJUMPAGESIZE != MCLBYTES 811 case MJUMPAGESIZE: 812 type = EXT_JUMBOP; 813 break; 814 #endif 815 case MJUM9BYTES: 816 type = EXT_JUMBO9; 817 break; 818 case MJUM16BYTES: 819 type = EXT_JUMBO16; 820 break; 821 default: 822 panic("%s: invalid cluster size %d", __func__, size); 823 } 824 825 return (type); 826 } 827 828 /* 829 * Associated an external reference counted buffer with an mbuf. 830 */ 831 static __inline void 832 m_extaddref(struct mbuf *m, char *buf, u_int size, u_int *ref_cnt, 833 m_ext_free_t freef, void *arg1, void *arg2) 834 { 835 836 KASSERT(ref_cnt != NULL, ("%s: ref_cnt not provided", __func__)); 837 838 atomic_add_int(ref_cnt, 1); 839 m->m_flags |= M_EXT; 840 m->m_ext.ext_buf = buf; 841 m->m_ext.ext_cnt = ref_cnt; 842 m->m_data = m->m_ext.ext_buf; 843 m->m_ext.ext_size = size; 844 m->m_ext.ext_free = freef; 845 m->m_ext.ext_arg1 = arg1; 846 m->m_ext.ext_arg2 = arg2; 847 m->m_ext.ext_type = EXT_EXTREF; 848 m->m_ext.ext_flags = 0; 849 } 850 851 static __inline uma_zone_t 852 m_getzone(int size) 853 { 854 uma_zone_t zone; 855 856 switch (size) { 857 case MCLBYTES: 858 zone = zone_clust; 859 break; 860 #if MJUMPAGESIZE != MCLBYTES 861 case MJUMPAGESIZE: 862 zone = zone_jumbop; 863 break; 864 #endif 865 case MJUM9BYTES: 866 zone = zone_jumbo9; 867 break; 868 case MJUM16BYTES: 869 zone = zone_jumbo16; 870 break; 871 default: 872 panic("%s: invalid cluster size %d", __func__, size); 873 } 874 875 return (zone); 876 } 877 878 /* 879 * Initialize an mbuf with linear storage. 880 * 881 * Inline because the consumer text overhead will be roughly the same to 882 * initialize or call a function with this many parameters and M_PKTHDR 883 * should go away with constant propagation for !MGETHDR. 884 */ 885 static __inline int 886 m_init(struct mbuf *m, int how, short type, int flags) 887 { 888 int error; 889 890 m->m_next = NULL; 891 m->m_nextpkt = NULL; 892 m->m_data = m->m_dat; 893 m->m_len = 0; 894 m->m_flags = flags; 895 m->m_type = type; 896 if (flags & M_PKTHDR) 897 error = m_pkthdr_init(m, how); 898 else 899 error = 0; 900 901 MBUF_PROBE5(m__init, m, how, type, flags, error); 902 return (error); 903 } 904 905 static __inline struct mbuf * 906 m_get(int how, short type) 907 { 908 struct mbuf *m; 909 struct mb_args args; 910 911 args.flags = 0; 912 args.type = type; 913 m = uma_zalloc_arg(zone_mbuf, &args, how); 914 MBUF_PROBE3(m__get, how, type, m); 915 return (m); 916 } 917 918 static __inline struct mbuf * 919 m_gethdr(int how, short type) 920 { 921 struct mbuf *m; 922 struct mb_args args; 923 924 args.flags = M_PKTHDR; 925 args.type = type; 926 m = uma_zalloc_arg(zone_mbuf, &args, how); 927 MBUF_PROBE3(m__gethdr, how, type, m); 928 return (m); 929 } 930 931 static __inline struct mbuf * 932 m_getcl(int how, short type, int flags) 933 { 934 struct mbuf *m; 935 struct mb_args args; 936 937 args.flags = flags; 938 args.type = type; 939 m = uma_zalloc_arg(zone_pack, &args, how); 940 MBUF_PROBE4(m__getcl, how, type, flags, m); 941 return (m); 942 } 943 944 /* 945 * XXX: m_cljset() is a dangerous API. One must attach only a new, 946 * unreferenced cluster to an mbuf(9). It is not possible to assert 947 * that, so care can be taken only by users of the API. 948 */ 949 static __inline void 950 m_cljset(struct mbuf *m, void *cl, int type) 951 { 952 int size; 953 954 switch (type) { 955 case EXT_CLUSTER: 956 size = MCLBYTES; 957 break; 958 #if MJUMPAGESIZE != MCLBYTES 959 case EXT_JUMBOP: 960 size = MJUMPAGESIZE; 961 break; 962 #endif 963 case EXT_JUMBO9: 964 size = MJUM9BYTES; 965 break; 966 case EXT_JUMBO16: 967 size = MJUM16BYTES; 968 break; 969 default: 970 panic("%s: unknown cluster type %d", __func__, type); 971 break; 972 } 973 974 m->m_data = m->m_ext.ext_buf = cl; 975 m->m_ext.ext_free = m->m_ext.ext_arg1 = m->m_ext.ext_arg2 = NULL; 976 m->m_ext.ext_size = size; 977 m->m_ext.ext_type = type; 978 m->m_ext.ext_flags = EXT_FLAG_EMBREF; 979 m->m_ext.ext_count = 1; 980 m->m_flags |= M_EXT; 981 MBUF_PROBE3(m__cljset, m, cl, type); 982 } 983 984 static __inline void 985 m_chtype(struct mbuf *m, short new_type) 986 { 987 988 m->m_type = new_type; 989 } 990 991 static __inline void 992 m_clrprotoflags(struct mbuf *m) 993 { 994 995 while (m) { 996 m->m_flags &= ~M_PROTOFLAGS; 997 m = m->m_next; 998 } 999 } 1000 1001 static __inline struct mbuf * 1002 m_last(struct mbuf *m) 1003 { 1004 1005 while (m->m_next) 1006 m = m->m_next; 1007 return (m); 1008 } 1009 1010 static inline u_int 1011 m_extrefcnt(struct mbuf *m) 1012 { 1013 1014 KASSERT(m->m_flags & M_EXT, ("%s: M_EXT missing", __func__)); 1015 1016 return ((m->m_ext.ext_flags & EXT_FLAG_EMBREF) ? m->m_ext.ext_count : 1017 *m->m_ext.ext_cnt); 1018 } 1019 1020 /* 1021 * mbuf, cluster, and external object allocation macros (for compatibility 1022 * purposes). 1023 */ 1024 #define M_MOVE_PKTHDR(to, from) m_move_pkthdr((to), (from)) 1025 #define MGET(m, how, type) ((m) = m_get((how), (type))) 1026 #define MGETHDR(m, how, type) ((m) = m_gethdr((how), (type))) 1027 #define MCLGET(m, how) m_clget((m), (how)) 1028 #define MEXTADD(m, buf, size, free, arg1, arg2, flags, type) \ 1029 m_extadd((m), (char *)(buf), (size), (free), (arg1), (arg2), \ 1030 (flags), (type)) 1031 #define m_getm(m, len, how, type) \ 1032 m_getm2((m), (len), (how), (type), M_PKTHDR) 1033 1034 /* 1035 * Evaluate TRUE if it's safe to write to the mbuf m's data region (this can 1036 * be both the local data payload, or an external buffer area, depending on 1037 * whether M_EXT is set). 1038 */ 1039 #define M_WRITABLE(m) (((m)->m_flags & (M_RDONLY | M_EXTPG)) == 0 && \ 1040 (!(((m)->m_flags & M_EXT)) || \ 1041 (m_extrefcnt(m) == 1))) 1042 1043 /* Check if the supplied mbuf has a packet header, or else panic. */ 1044 #define M_ASSERTPKTHDR(m) \ 1045 KASSERT((m) != NULL && (m)->m_flags & M_PKTHDR, \ 1046 ("%s: no mbuf packet header!", __func__)) 1047 1048 /* Check if mbuf is multipage. */ 1049 #define M_ASSERTEXTPG(m) \ 1050 KASSERT(((m)->m_flags & (M_EXTPG|M_PKTHDR)) == M_EXTPG, \ 1051 ("%s: m %p is not multipage!", __func__, m)) 1052 1053 /* 1054 * Ensure that the supplied mbuf is a valid, non-free mbuf. 1055 * 1056 * XXX: Broken at the moment. Need some UMA magic to make it work again. 1057 */ 1058 #define M_ASSERTVALID(m) \ 1059 KASSERT((((struct mbuf *)m)->m_flags & 0) == 0, \ 1060 ("%s: attempted use of a free mbuf!", __func__)) 1061 1062 /* 1063 * Return the address of the start of the buffer associated with an mbuf, 1064 * handling external storage, packet-header mbufs, and regular data mbufs. 1065 */ 1066 #define M_START(m) \ 1067 (((m)->m_flags & M_EXTPG) ? NULL : \ 1068 ((m)->m_flags & M_EXT) ? (m)->m_ext.ext_buf : \ 1069 ((m)->m_flags & M_PKTHDR) ? &(m)->m_pktdat[0] : \ 1070 &(m)->m_dat[0]) 1071 1072 /* 1073 * Return the size of the buffer associated with an mbuf, handling external 1074 * storage, packet-header mbufs, and regular data mbufs. 1075 */ 1076 #define M_SIZE(m) \ 1077 (((m)->m_flags & M_EXT) ? (m)->m_ext.ext_size : \ 1078 ((m)->m_flags & M_PKTHDR) ? MHLEN : \ 1079 MLEN) 1080 1081 /* 1082 * Set the m_data pointer of a newly allocated mbuf to place an object of the 1083 * specified size at the end of the mbuf, longword aligned. 1084 * 1085 * NB: Historically, we had M_ALIGN(), MH_ALIGN(), and MEXT_ALIGN() as 1086 * separate macros, each asserting that it was called at the proper moment. 1087 * This required callers to themselves test the storage type and call the 1088 * right one. Rather than require callers to be aware of those layout 1089 * decisions, we centralize here. 1090 */ 1091 static __inline void 1092 m_align(struct mbuf *m, int len) 1093 { 1094 #ifdef INVARIANTS 1095 const char *msg = "%s: not a virgin mbuf"; 1096 #endif 1097 int adjust; 1098 1099 KASSERT(m->m_data == M_START(m), (msg, __func__)); 1100 1101 adjust = M_SIZE(m) - len; 1102 m->m_data += adjust &~ (sizeof(long)-1); 1103 } 1104 1105 #define M_ALIGN(m, len) m_align(m, len) 1106 #define MH_ALIGN(m, len) m_align(m, len) 1107 #define MEXT_ALIGN(m, len) m_align(m, len) 1108 1109 /* 1110 * Compute the amount of space available before the current start of data in 1111 * an mbuf. 1112 * 1113 * The M_WRITABLE() is a temporary, conservative safety measure: the burden 1114 * of checking writability of the mbuf data area rests solely with the caller. 1115 * 1116 * NB: In previous versions, M_LEADINGSPACE() would only check M_WRITABLE() 1117 * for mbufs with external storage. We now allow mbuf-embedded data to be 1118 * read-only as well. 1119 */ 1120 #define M_LEADINGSPACE(m) \ 1121 (M_WRITABLE(m) ? ((m)->m_data - M_START(m)) : 0) 1122 1123 /* 1124 * Compute the amount of space available after the end of data in an mbuf. 1125 * 1126 * The M_WRITABLE() is a temporary, conservative safety measure: the burden 1127 * of checking writability of the mbuf data area rests solely with the caller. 1128 * 1129 * NB: In previous versions, M_TRAILINGSPACE() would only check M_WRITABLE() 1130 * for mbufs with external storage. We now allow mbuf-embedded data to be 1131 * read-only as well. 1132 */ 1133 #define M_TRAILINGSPACE(m) \ 1134 (M_WRITABLE(m) ? \ 1135 ((M_START(m) + M_SIZE(m)) - ((m)->m_data + (m)->m_len)) : 0) 1136 1137 /* 1138 * Arrange to prepend space of size plen to mbuf m. If a new mbuf must be 1139 * allocated, how specifies whether to wait. If the allocation fails, the 1140 * original mbuf chain is freed and m is set to NULL. 1141 */ 1142 #define M_PREPEND(m, plen, how) do { \ 1143 struct mbuf **_mmp = &(m); \ 1144 struct mbuf *_mm = *_mmp; \ 1145 int _mplen = (plen); \ 1146 int __mhow = (how); \ 1147 \ 1148 MBUF_CHECKSLEEP(how); \ 1149 if (M_LEADINGSPACE(_mm) >= _mplen) { \ 1150 _mm->m_data -= _mplen; \ 1151 _mm->m_len += _mplen; \ 1152 } else \ 1153 _mm = m_prepend(_mm, _mplen, __mhow); \ 1154 if (_mm != NULL && _mm->m_flags & M_PKTHDR) \ 1155 _mm->m_pkthdr.len += _mplen; \ 1156 *_mmp = _mm; \ 1157 } while (0) 1158 1159 /* 1160 * Change mbuf to new type. This is a relatively expensive operation and 1161 * should be avoided. 1162 */ 1163 #define MCHTYPE(m, t) m_chtype((m), (t)) 1164 1165 /* Return the rcvif of a packet header. */ 1166 static __inline struct ifnet * 1167 m_rcvif(struct mbuf *m) 1168 { 1169 1170 M_ASSERTPKTHDR(m); 1171 if (m->m_pkthdr.csum_flags & CSUM_SND_TAG) 1172 return (NULL); 1173 return (m->m_pkthdr.rcvif); 1174 } 1175 1176 /* Length to m_copy to copy all. */ 1177 #define M_COPYALL 1000000000 1178 1179 extern int max_datalen; /* MHLEN - max_hdr */ 1180 extern int max_hdr; /* Largest link + protocol header */ 1181 extern int max_linkhdr; /* Largest link-level header */ 1182 extern int max_protohdr; /* Largest protocol header */ 1183 extern int nmbclusters; /* Maximum number of clusters */ 1184 extern bool mb_use_ext_pgs; /* Use ext_pgs for sendfile */ 1185 1186 /*- 1187 * Network packets may have annotations attached by affixing a list of 1188 * "packet tags" to the pkthdr structure. Packet tags are dynamically 1189 * allocated semi-opaque data structures that have a fixed header 1190 * (struct m_tag) that specifies the size of the memory block and a 1191 * <cookie,type> pair that identifies it. The cookie is a 32-bit unique 1192 * unsigned value used to identify a module or ABI. By convention this value 1193 * is chosen as the date+time that the module is created, expressed as the 1194 * number of seconds since the epoch (e.g., using date -u +'%s'). The type 1195 * value is an ABI/module-specific value that identifies a particular 1196 * annotation and is private to the module. For compatibility with systems 1197 * like OpenBSD that define packet tags w/o an ABI/module cookie, the value 1198 * PACKET_ABI_COMPAT is used to implement m_tag_get and m_tag_find 1199 * compatibility shim functions and several tag types are defined below. 1200 * Users that do not require compatibility should use a private cookie value 1201 * so that packet tag-related definitions can be maintained privately. 1202 * 1203 * Note that the packet tag returned by m_tag_alloc has the default memory 1204 * alignment implemented by malloc. To reference private data one can use a 1205 * construct like: 1206 * 1207 * struct m_tag *mtag = m_tag_alloc(...); 1208 * struct foo *p = (struct foo *)(mtag+1); 1209 * 1210 * if the alignment of struct m_tag is sufficient for referencing members of 1211 * struct foo. Otherwise it is necessary to embed struct m_tag within the 1212 * private data structure to insure proper alignment; e.g., 1213 * 1214 * struct foo { 1215 * struct m_tag tag; 1216 * ... 1217 * }; 1218 * struct foo *p = (struct foo *) m_tag_alloc(...); 1219 * struct m_tag *mtag = &p->tag; 1220 */ 1221 1222 /* 1223 * Persistent tags stay with an mbuf until the mbuf is reclaimed. Otherwise 1224 * tags are expected to ``vanish'' when they pass through a network 1225 * interface. For most interfaces this happens normally as the tags are 1226 * reclaimed when the mbuf is free'd. However in some special cases 1227 * reclaiming must be done manually. An example is packets that pass through 1228 * the loopback interface. Also, one must be careful to do this when 1229 * ``turning around'' packets (e.g., icmp_reflect). 1230 * 1231 * To mark a tag persistent bit-or this flag in when defining the tag id. 1232 * The tag will then be treated as described above. 1233 */ 1234 #define MTAG_PERSISTENT 0x800 1235 1236 #define PACKET_TAG_NONE 0 /* Nadda */ 1237 1238 /* Packet tags for use with PACKET_ABI_COMPAT. */ 1239 #define PACKET_TAG_IPSEC_IN_DONE 1 /* IPsec applied, in */ 1240 #define PACKET_TAG_IPSEC_OUT_DONE 2 /* IPsec applied, out */ 1241 #define PACKET_TAG_IPSEC_IN_CRYPTO_DONE 3 /* NIC IPsec crypto done */ 1242 #define PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED 4 /* NIC IPsec crypto req'ed */ 1243 #define PACKET_TAG_IPSEC_IN_COULD_DO_CRYPTO 5 /* NIC notifies IPsec */ 1244 #define PACKET_TAG_IPSEC_PENDING_TDB 6 /* Reminder to do IPsec */ 1245 #define PACKET_TAG_BRIDGE 7 /* Bridge processing done */ 1246 #define PACKET_TAG_GIF 8 /* GIF processing done */ 1247 #define PACKET_TAG_GRE 9 /* GRE processing done */ 1248 #define PACKET_TAG_IN_PACKET_CHECKSUM 10 /* NIC checksumming done */ 1249 #define PACKET_TAG_ENCAP 11 /* Encap. processing */ 1250 #define PACKET_TAG_IPSEC_SOCKET 12 /* IPSEC socket ref */ 1251 #define PACKET_TAG_IPSEC_HISTORY 13 /* IPSEC history */ 1252 #define PACKET_TAG_IPV6_INPUT 14 /* IPV6 input processing */ 1253 #define PACKET_TAG_DUMMYNET 15 /* dummynet info */ 1254 #define PACKET_TAG_DIVERT 17 /* divert info */ 1255 #define PACKET_TAG_IPFORWARD 18 /* ipforward info */ 1256 #define PACKET_TAG_MACLABEL (19 | MTAG_PERSISTENT) /* MAC label */ 1257 #define PACKET_TAG_PF (21 | MTAG_PERSISTENT) /* PF/ALTQ information */ 1258 #define PACKET_TAG_RTSOCKFAM 25 /* rtsock sa family */ 1259 #define PACKET_TAG_IPOPTIONS 27 /* Saved IP options */ 1260 #define PACKET_TAG_CARP 28 /* CARP info */ 1261 #define PACKET_TAG_IPSEC_NAT_T_PORTS 29 /* two uint16_t */ 1262 #define PACKET_TAG_ND_OUTGOING 30 /* ND outgoing */ 1263 1264 /* Specific cookies and tags. */ 1265 1266 /* Packet tag routines. */ 1267 struct m_tag *m_tag_alloc(u_int32_t, int, int, int); 1268 void m_tag_delete(struct mbuf *, struct m_tag *); 1269 void m_tag_delete_chain(struct mbuf *, struct m_tag *); 1270 void m_tag_free_default(struct m_tag *); 1271 struct m_tag *m_tag_locate(struct mbuf *, u_int32_t, int, struct m_tag *); 1272 struct m_tag *m_tag_copy(struct m_tag *, int); 1273 int m_tag_copy_chain(struct mbuf *, const struct mbuf *, int); 1274 void m_tag_delete_nonpersistent(struct mbuf *); 1275 1276 /* 1277 * Initialize the list of tags associated with an mbuf. 1278 */ 1279 static __inline void 1280 m_tag_init(struct mbuf *m) 1281 { 1282 1283 SLIST_INIT(&m->m_pkthdr.tags); 1284 } 1285 1286 /* 1287 * Set up the contents of a tag. Note that this does not fill in the free 1288 * method; the caller is expected to do that. 1289 * 1290 * XXX probably should be called m_tag_init, but that was already taken. 1291 */ 1292 static __inline void 1293 m_tag_setup(struct m_tag *t, u_int32_t cookie, int type, int len) 1294 { 1295 1296 t->m_tag_id = type; 1297 t->m_tag_len = len; 1298 t->m_tag_cookie = cookie; 1299 } 1300 1301 /* 1302 * Reclaim resources associated with a tag. 1303 */ 1304 static __inline void 1305 m_tag_free(struct m_tag *t) 1306 { 1307 1308 (*t->m_tag_free)(t); 1309 } 1310 1311 /* 1312 * Return the first tag associated with an mbuf. 1313 */ 1314 static __inline struct m_tag * 1315 m_tag_first(struct mbuf *m) 1316 { 1317 1318 return (SLIST_FIRST(&m->m_pkthdr.tags)); 1319 } 1320 1321 /* 1322 * Return the next tag in the list of tags associated with an mbuf. 1323 */ 1324 static __inline struct m_tag * 1325 m_tag_next(struct mbuf *m __unused, struct m_tag *t) 1326 { 1327 1328 return (SLIST_NEXT(t, m_tag_link)); 1329 } 1330 1331 /* 1332 * Prepend a tag to the list of tags associated with an mbuf. 1333 */ 1334 static __inline void 1335 m_tag_prepend(struct mbuf *m, struct m_tag *t) 1336 { 1337 1338 SLIST_INSERT_HEAD(&m->m_pkthdr.tags, t, m_tag_link); 1339 } 1340 1341 /* 1342 * Unlink a tag from the list of tags associated with an mbuf. 1343 */ 1344 static __inline void 1345 m_tag_unlink(struct mbuf *m, struct m_tag *t) 1346 { 1347 1348 SLIST_REMOVE(&m->m_pkthdr.tags, t, m_tag, m_tag_link); 1349 } 1350 1351 /* These are for OpenBSD compatibility. */ 1352 #define MTAG_ABI_COMPAT 0 /* compatibility ABI */ 1353 1354 static __inline struct m_tag * 1355 m_tag_get(int type, int length, int wait) 1356 { 1357 return (m_tag_alloc(MTAG_ABI_COMPAT, type, length, wait)); 1358 } 1359 1360 static __inline struct m_tag * 1361 m_tag_find(struct mbuf *m, int type, struct m_tag *start) 1362 { 1363 return (SLIST_EMPTY(&m->m_pkthdr.tags) ? (struct m_tag *)NULL : 1364 m_tag_locate(m, MTAG_ABI_COMPAT, type, start)); 1365 } 1366 1367 static inline struct m_snd_tag * 1368 m_snd_tag_ref(struct m_snd_tag *mst) 1369 { 1370 1371 refcount_acquire(&mst->refcount); 1372 return (mst); 1373 } 1374 1375 static inline void 1376 m_snd_tag_rele(struct m_snd_tag *mst) 1377 { 1378 1379 if (refcount_release(&mst->refcount)) 1380 m_snd_tag_destroy(mst); 1381 } 1382 1383 static __inline struct mbuf * 1384 m_free(struct mbuf *m) 1385 { 1386 struct mbuf *n = m->m_next; 1387 1388 MBUF_PROBE1(m__free, m); 1389 if ((m->m_flags & (M_PKTHDR|M_NOFREE)) == (M_PKTHDR|M_NOFREE)) 1390 m_tag_delete_chain(m, NULL); 1391 if (m->m_flags & M_PKTHDR && m->m_pkthdr.csum_flags & CSUM_SND_TAG) 1392 m_snd_tag_rele(m->m_pkthdr.snd_tag); 1393 if (m->m_flags & M_EXTPG) 1394 mb_free_extpg(m); 1395 else if (m->m_flags & M_EXT) 1396 mb_free_ext(m); 1397 else if ((m->m_flags & M_NOFREE) == 0) 1398 uma_zfree(zone_mbuf, m); 1399 return (n); 1400 } 1401 1402 static __inline int 1403 rt_m_getfib(struct mbuf *m) 1404 { 1405 KASSERT(m->m_flags & M_PKTHDR , ("Attempt to get FIB from non header mbuf.")); 1406 return (m->m_pkthdr.fibnum); 1407 } 1408 1409 #define M_GETFIB(_m) rt_m_getfib(_m) 1410 1411 #define M_SETFIB(_m, _fib) do { \ 1412 KASSERT((_m)->m_flags & M_PKTHDR, ("Attempt to set FIB on non header mbuf.")); \ 1413 ((_m)->m_pkthdr.fibnum) = (_fib); \ 1414 } while (0) 1415 1416 /* flags passed as first argument for "m_ether_tcpip_hash()" */ 1417 #define MBUF_HASHFLAG_L2 (1 << 2) 1418 #define MBUF_HASHFLAG_L3 (1 << 3) 1419 #define MBUF_HASHFLAG_L4 (1 << 4) 1420 1421 /* mbuf hashing helper routines */ 1422 uint32_t m_ether_tcpip_hash_init(void); 1423 uint32_t m_ether_tcpip_hash(const uint32_t, const struct mbuf *, const uint32_t); 1424 1425 #ifdef MBUF_PROFILING 1426 void m_profile(struct mbuf *m); 1427 #define M_PROFILE(m) m_profile(m) 1428 #else 1429 #define M_PROFILE(m) 1430 #endif 1431 1432 struct mbufq { 1433 STAILQ_HEAD(, mbuf) mq_head; 1434 int mq_len; 1435 int mq_maxlen; 1436 }; 1437 1438 static inline void 1439 mbufq_init(struct mbufq *mq, int maxlen) 1440 { 1441 1442 STAILQ_INIT(&mq->mq_head); 1443 mq->mq_maxlen = maxlen; 1444 mq->mq_len = 0; 1445 } 1446 1447 static inline struct mbuf * 1448 mbufq_flush(struct mbufq *mq) 1449 { 1450 struct mbuf *m; 1451 1452 m = STAILQ_FIRST(&mq->mq_head); 1453 STAILQ_INIT(&mq->mq_head); 1454 mq->mq_len = 0; 1455 return (m); 1456 } 1457 1458 static inline void 1459 mbufq_drain(struct mbufq *mq) 1460 { 1461 struct mbuf *m, *n; 1462 1463 n = mbufq_flush(mq); 1464 while ((m = n) != NULL) { 1465 n = STAILQ_NEXT(m, m_stailqpkt); 1466 m_freem(m); 1467 } 1468 } 1469 1470 static inline struct mbuf * 1471 mbufq_first(const struct mbufq *mq) 1472 { 1473 1474 return (STAILQ_FIRST(&mq->mq_head)); 1475 } 1476 1477 static inline struct mbuf * 1478 mbufq_last(const struct mbufq *mq) 1479 { 1480 1481 return (STAILQ_LAST(&mq->mq_head, mbuf, m_stailqpkt)); 1482 } 1483 1484 static inline int 1485 mbufq_full(const struct mbufq *mq) 1486 { 1487 1488 return (mq->mq_maxlen > 0 && mq->mq_len >= mq->mq_maxlen); 1489 } 1490 1491 static inline int 1492 mbufq_len(const struct mbufq *mq) 1493 { 1494 1495 return (mq->mq_len); 1496 } 1497 1498 static inline int 1499 mbufq_enqueue(struct mbufq *mq, struct mbuf *m) 1500 { 1501 1502 if (mbufq_full(mq)) 1503 return (ENOBUFS); 1504 STAILQ_INSERT_TAIL(&mq->mq_head, m, m_stailqpkt); 1505 mq->mq_len++; 1506 return (0); 1507 } 1508 1509 static inline struct mbuf * 1510 mbufq_dequeue(struct mbufq *mq) 1511 { 1512 struct mbuf *m; 1513 1514 m = STAILQ_FIRST(&mq->mq_head); 1515 if (m) { 1516 STAILQ_REMOVE_HEAD(&mq->mq_head, m_stailqpkt); 1517 m->m_nextpkt = NULL; 1518 mq->mq_len--; 1519 } 1520 return (m); 1521 } 1522 1523 static inline void 1524 mbufq_prepend(struct mbufq *mq, struct mbuf *m) 1525 { 1526 1527 STAILQ_INSERT_HEAD(&mq->mq_head, m, m_stailqpkt); 1528 mq->mq_len++; 1529 } 1530 1531 /* 1532 * Note: this doesn't enforce the maximum list size for dst. 1533 */ 1534 static inline void 1535 mbufq_concat(struct mbufq *mq_dst, struct mbufq *mq_src) 1536 { 1537 1538 mq_dst->mq_len += mq_src->mq_len; 1539 STAILQ_CONCAT(&mq_dst->mq_head, &mq_src->mq_head); 1540 mq_src->mq_len = 0; 1541 } 1542 1543 #ifdef _SYS_TIMESPEC_H_ 1544 static inline void 1545 mbuf_tstmp2timespec(struct mbuf *m, struct timespec *ts) 1546 { 1547 1548 KASSERT((m->m_flags & M_PKTHDR) != 0, ("mbuf %p no M_PKTHDR", m)); 1549 KASSERT((m->m_flags & (M_TSTMP|M_TSTMP_LRO)) != 0, ("mbuf %p no M_TSTMP or M_TSTMP_LRO", m)); 1550 ts->tv_sec = m->m_pkthdr.rcv_tstmp / 1000000000; 1551 ts->tv_nsec = m->m_pkthdr.rcv_tstmp % 1000000000; 1552 } 1553 #endif 1554 1555 #ifdef DEBUGNET 1556 /* Invoked from the debugnet client code. */ 1557 void debugnet_mbuf_drain(void); 1558 void debugnet_mbuf_start(void); 1559 void debugnet_mbuf_finish(void); 1560 void debugnet_mbuf_reinit(int nmbuf, int nclust, int clsize); 1561 #endif 1562 1563 static inline bool 1564 mbuf_has_tls_session(struct mbuf *m) 1565 { 1566 1567 if (m->m_flags & M_EXTPG) { 1568 if (m->m_epg_tls != NULL) { 1569 return (true); 1570 } 1571 } 1572 return (false); 1573 } 1574 1575 #endif /* _KERNEL */ 1576 #endif /* !_SYS_MBUF_H_ */ 1577