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