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