uipc_mbuf.c revision 177599
1139804Simp/*- 21541Srgrimes * Copyright (c) 1982, 1986, 1988, 1991, 1993 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * Redistribution and use in source and binary forms, with or without 61541Srgrimes * modification, are permitted provided that the following conditions 71541Srgrimes * are met: 81541Srgrimes * 1. Redistributions of source code must retain the above copyright 91541Srgrimes * notice, this list of conditions and the following disclaimer. 101541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 111541Srgrimes * notice, this list of conditions and the following disclaimer in the 121541Srgrimes * documentation and/or other materials provided with the distribution. 131541Srgrimes * 4. Neither the name of the University nor the names of its contributors 141541Srgrimes * may be used to endorse or promote products derived from this software 151541Srgrimes * without specific prior written permission. 161541Srgrimes * 171541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 181541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 191541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 201541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 211541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 221541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 231541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 241541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 251541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 261541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 271541Srgrimes * SUCH DAMAGE. 281541Srgrimes * 291541Srgrimes * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 301541Srgrimes */ 311541Srgrimes 32116182Sobrien#include <sys/cdefs.h> 33116182Sobrien__FBSDID("$FreeBSD: head/sys/kern/uipc_mbuf.c 177599 2008-03-25 09:39:02Z ru $"); 34116182Sobrien 35101007Srwatson#include "opt_mac.h" 3677572Sobrien#include "opt_param.h" 37113490Ssilby#include "opt_mbuf_stress_test.h" 38101007Srwatson 391541Srgrimes#include <sys/param.h> 401541Srgrimes#include <sys/systm.h> 4176166Smarkm#include <sys/kernel.h> 42125296Ssilby#include <sys/limits.h> 4376166Smarkm#include <sys/lock.h> 4432036Sbde#include <sys/malloc.h> 451541Srgrimes#include <sys/mbuf.h> 4623081Swollman#include <sys/sysctl.h> 471541Srgrimes#include <sys/domain.h> 481541Srgrimes#include <sys/protosw.h> 49125296Ssilby#include <sys/uio.h> 5076166Smarkm 51163606Srwatson#include <security/mac/mac_framework.h> 52163606Srwatson 539759Sbdeint max_linkhdr; 549759Sbdeint max_protohdr; 559759Sbdeint max_hdr; 569759Sbdeint max_datalen; 57116455Ssilby#ifdef MBUF_STRESS_TEST 58112777Ssilbyint m_defragpackets; 59112777Ssilbyint m_defragbytes; 60112777Ssilbyint m_defraguseless; 61112777Ssilbyint m_defragfailure; 62113490Ssilbyint m_defragrandomfailures; 63113490Ssilby#endif 641541Srgrimes 6566475Sbmilekic/* 6666475Sbmilekic * sysctl(8) exported objects 6766475Sbmilekic */ 68155820SandreSYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD, 69155820Sandre &max_linkhdr, 0, "Size of largest link layer header"); 70155820SandreSYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD, 71155820Sandre &max_protohdr, 0, "Size of largest protocol layer header"); 72155820SandreSYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD, 73155820Sandre &max_hdr, 0, "Size of largest link plus protocol header"); 74155820SandreSYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD, 75155820Sandre &max_datalen, 0, "Minimum space left in mbuf after max_hdr"); 76116455Ssilby#ifdef MBUF_STRESS_TEST 77112777SsilbySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD, 78112777Ssilby &m_defragpackets, 0, ""); 79112777SsilbySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD, 80112777Ssilby &m_defragbytes, 0, ""); 81112777SsilbySYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD, 82112777Ssilby &m_defraguseless, 0, ""); 83112777SsilbySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD, 84112777Ssilby &m_defragfailure, 0, ""); 85113490SsilbySYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW, 86113490Ssilby &m_defragrandomfailures, 0, ""); 87113490Ssilby#endif 8875112Sbmilekic 891541Srgrimes/* 90129906Sbmilekic * Allocate a given length worth of mbufs and/or clusters (whatever fits 91129906Sbmilekic * best) and return a pointer to the top of the allocated chain. If an 92129906Sbmilekic * existing mbuf chain is provided, then we will append the new chain 93129906Sbmilekic * to the existing one but still return the top of the newly allocated 94129906Sbmilekic * chain. 95129906Sbmilekic */ 96129906Sbmilekicstruct mbuf * 97163915Sandrem_getm2(struct mbuf *m, int len, int how, short type, int flags) 98129906Sbmilekic{ 99163915Sandre struct mbuf *mb, *nm = NULL, *mtail = NULL; 100129906Sbmilekic 101163915Sandre KASSERT(len >= 0, ("%s: len is < 0", __func__)); 102129906Sbmilekic 103163915Sandre /* Validate flags. */ 104163915Sandre flags &= (M_PKTHDR | M_EOR); 105129906Sbmilekic 106163915Sandre /* Packet header mbuf must be first in chain. */ 107163915Sandre if ((flags & M_PKTHDR) && m != NULL) 108163915Sandre flags &= ~M_PKTHDR; 109129906Sbmilekic 110163915Sandre /* Loop and append maximum sized mbufs to the chain tail. */ 111163915Sandre while (len > 0) { 112163915Sandre if (len > MCLBYTES) 113163915Sandre mb = m_getjcl(how, type, (flags & M_PKTHDR), 114163915Sandre MJUMPAGESIZE); 115163915Sandre else if (len >= MINCLSIZE) 116163915Sandre mb = m_getcl(how, type, (flags & M_PKTHDR)); 117163915Sandre else if (flags & M_PKTHDR) 118163915Sandre mb = m_gethdr(how, type); 119129906Sbmilekic else 120163915Sandre mb = m_get(how, type); 121163915Sandre 122163915Sandre /* Fail the whole operation if one mbuf can't be allocated. */ 123163915Sandre if (mb == NULL) { 124163915Sandre if (nm != NULL) 125163915Sandre m_freem(nm); 126163915Sandre return (NULL); 127163915Sandre } 128163915Sandre 129163915Sandre /* Book keeping. */ 130163915Sandre len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size : 131163915Sandre ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN); 132163915Sandre if (mtail != NULL) 133163915Sandre mtail->m_next = mb; 134163915Sandre else 135163915Sandre nm = mb; 136163915Sandre mtail = mb; 137163915Sandre flags &= ~M_PKTHDR; /* Only valid on the first mbuf. */ 138129906Sbmilekic } 139163915Sandre if (flags & M_EOR) 140163915Sandre mtail->m_flags |= M_EOR; /* Only valid on the last mbuf. */ 141129906Sbmilekic 142163915Sandre /* If mbuf was supplied, append new chain to the end of it. */ 143163915Sandre if (m != NULL) { 144163915Sandre for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next) 145163915Sandre ; 146163915Sandre mtail->m_next = nm; 147163915Sandre mtail->m_flags &= ~M_EOR; 148163915Sandre } else 149163915Sandre m = nm; 150163915Sandre 151163915Sandre return (m); 152129906Sbmilekic} 153129906Sbmilekic 154129906Sbmilekic/* 155129906Sbmilekic * Free an entire chain of mbufs and associated external buffers, if 156129906Sbmilekic * applicable. 157129906Sbmilekic */ 158129906Sbmilekicvoid 159129906Sbmilekicm_freem(struct mbuf *mb) 160129906Sbmilekic{ 161129906Sbmilekic 162129906Sbmilekic while (mb != NULL) 163129906Sbmilekic mb = m_free(mb); 164129906Sbmilekic} 165129906Sbmilekic 166129906Sbmilekic/*- 167129906Sbmilekic * Configure a provided mbuf to refer to the provided external storage 168129906Sbmilekic * buffer and setup a reference count for said buffer. If the setting 169129906Sbmilekic * up of the reference count fails, the M_EXT bit will not be set. If 170129906Sbmilekic * successfull, the M_EXT bit is set in the mbuf's flags. 171129906Sbmilekic * 172129906Sbmilekic * Arguments: 173129906Sbmilekic * mb The existing mbuf to which to attach the provided buffer. 174129906Sbmilekic * buf The address of the provided external storage buffer. 175129906Sbmilekic * size The size of the provided buffer. 176129906Sbmilekic * freef A pointer to a routine that is responsible for freeing the 177129906Sbmilekic * provided external storage buffer. 178129906Sbmilekic * args A pointer to an argument structure (of any type) to be passed 179129906Sbmilekic * to the provided freef routine (may be NULL). 180129906Sbmilekic * flags Any other flags to be passed to the provided mbuf. 181129906Sbmilekic * type The type that the external storage buffer should be 182129906Sbmilekic * labeled with. 183129906Sbmilekic * 184129906Sbmilekic * Returns: 185129906Sbmilekic * Nothing. 186129906Sbmilekic */ 187129906Sbmilekicvoid 188129906Sbmilekicm_extadd(struct mbuf *mb, caddr_t buf, u_int size, 189175872Sphk void (*freef)(void *, void *), void *arg1, void *arg2, int flags, int type) 190129906Sbmilekic{ 191151976Sandre KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__)); 192129906Sbmilekic 193151976Sandre if (type != EXT_EXTREF) 194151976Sandre mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT); 195129906Sbmilekic if (mb->m_ext.ref_cnt != NULL) { 196129906Sbmilekic *(mb->m_ext.ref_cnt) = 1; 197129906Sbmilekic mb->m_flags |= (M_EXT | flags); 198129906Sbmilekic mb->m_ext.ext_buf = buf; 199129906Sbmilekic mb->m_data = mb->m_ext.ext_buf; 200129906Sbmilekic mb->m_ext.ext_size = size; 201129906Sbmilekic mb->m_ext.ext_free = freef; 202175872Sphk mb->m_ext.ext_arg1 = arg1; 203175872Sphk mb->m_ext.ext_arg2 = arg2; 204129906Sbmilekic mb->m_ext.ext_type = type; 205129906Sbmilekic } 206129906Sbmilekic} 207129906Sbmilekic 208129906Sbmilekic/* 209129906Sbmilekic * Non-directly-exported function to clean up after mbufs with M_EXT 210151976Sandre * storage attached to them if the reference count hits 1. 211129906Sbmilekic */ 212129906Sbmilekicvoid 213129906Sbmilekicmb_free_ext(struct mbuf *m) 214129906Sbmilekic{ 215172463Skmacy int skipmbuf; 216172463Skmacy 217151976Sandre KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 218151976Sandre KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 219129906Sbmilekic 220172463Skmacy 221172463Skmacy /* 222172463Skmacy * check if the header is embedded in the cluster 223172463Skmacy */ 224172463Skmacy skipmbuf = (m->m_flags & M_NOFREE); 225172463Skmacy 226151976Sandre /* Free attached storage if this mbuf is the only reference to it. */ 227151976Sandre if (*(m->m_ext.ref_cnt) == 1 || 228162515Srrs atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) { 229151976Sandre switch (m->m_ext.ext_type) { 230152101Sandre case EXT_PACKET: /* The packet zone is special. */ 231152035Sandre if (*(m->m_ext.ref_cnt) == 0) 232152035Sandre *(m->m_ext.ref_cnt) = 1; 233151976Sandre uma_zfree(zone_pack, m); 234151976Sandre return; /* Job done. */ 235152101Sandre case EXT_CLUSTER: 236152101Sandre uma_zfree(zone_clust, m->m_ext.ext_buf); 237130289Sbmilekic break; 238155780Sandre case EXT_JUMBOP: 239155780Sandre uma_zfree(zone_jumbop, m->m_ext.ext_buf); 240153232Sandre break; 241151976Sandre case EXT_JUMBO9: 242151976Sandre uma_zfree(zone_jumbo9, m->m_ext.ext_buf); 243151976Sandre break; 244151976Sandre case EXT_JUMBO16: 245151976Sandre uma_zfree(zone_jumbo16, m->m_ext.ext_buf); 246151976Sandre break; 247151976Sandre case EXT_SFBUF: 248151976Sandre case EXT_NET_DRV: 249151976Sandre case EXT_MOD_TYPE: 250151976Sandre case EXT_DISPOSABLE: 251151976Sandre *(m->m_ext.ref_cnt) = 0; 252151976Sandre uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *, 253151976Sandre m->m_ext.ref_cnt)); 254151976Sandre /* FALLTHROUGH */ 255151976Sandre case EXT_EXTREF: 256151976Sandre KASSERT(m->m_ext.ext_free != NULL, 257151976Sandre ("%s: ext_free not set", __func__)); 258175872Sphk (*(m->m_ext.ext_free))(m->m_ext.ext_arg1, 259175872Sphk m->m_ext.ext_arg2); 260151976Sandre break; 261151976Sandre default: 262151976Sandre KASSERT(m->m_ext.ext_type == 0, 263151976Sandre ("%s: unknown ext_type", __func__)); 264141668Sbmilekic } 265141668Sbmilekic } 266172463Skmacy if (skipmbuf) 267172463Skmacy return; 268172463Skmacy 269151976Sandre /* 270151976Sandre * Free this mbuf back to the mbuf zone with all m_ext 271151976Sandre * information purged. 272151976Sandre */ 273151976Sandre m->m_ext.ext_buf = NULL; 274151976Sandre m->m_ext.ext_free = NULL; 275175872Sphk m->m_ext.ext_arg1 = NULL; 276175872Sphk m->m_ext.ext_arg2 = NULL; 277151976Sandre m->m_ext.ref_cnt = NULL; 278151976Sandre m->m_ext.ext_size = 0; 279151976Sandre m->m_ext.ext_type = 0; 280151976Sandre m->m_flags &= ~M_EXT; 281130357Sbmilekic uma_zfree(zone_mbuf, m); 282129906Sbmilekic} 283129906Sbmilekic 284129906Sbmilekic/* 285151976Sandre * Attach the the cluster from *m to *n, set up m_ext in *n 286151976Sandre * and bump the refcount of the cluster. 287151976Sandre */ 288151976Sandrestatic void 289151976Sandremb_dupcl(struct mbuf *n, struct mbuf *m) 290151976Sandre{ 291151976Sandre KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 292151976Sandre KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 293151976Sandre KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 294151976Sandre 295151976Sandre if (*(m->m_ext.ref_cnt) == 1) 296151976Sandre *(m->m_ext.ref_cnt) += 1; 297151976Sandre else 298151976Sandre atomic_add_int(m->m_ext.ref_cnt, 1); 299151976Sandre n->m_ext.ext_buf = m->m_ext.ext_buf; 300151976Sandre n->m_ext.ext_free = m->m_ext.ext_free; 301175872Sphk n->m_ext.ext_arg1 = m->m_ext.ext_arg1; 302175872Sphk n->m_ext.ext_arg2 = m->m_ext.ext_arg2; 303151976Sandre n->m_ext.ext_size = m->m_ext.ext_size; 304151976Sandre n->m_ext.ref_cnt = m->m_ext.ref_cnt; 305151976Sandre n->m_ext.ext_type = m->m_ext.ext_type; 306151976Sandre n->m_flags |= M_EXT; 307151976Sandre} 308151976Sandre 309151976Sandre/* 310149598Sandre * Clean up mbuf (chain) from any tags and packet headers. 311149647Sandre * If "all" is set then the first mbuf in the chain will be 312149647Sandre * cleaned too. 313149598Sandre */ 314149598Sandrevoid 315149647Sandrem_demote(struct mbuf *m0, int all) 316149598Sandre{ 317149598Sandre struct mbuf *m; 318149598Sandre 319149598Sandre for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { 320149598Sandre if (m->m_flags & M_PKTHDR) { 321149598Sandre m_tag_delete_chain(m, NULL); 322149598Sandre m->m_flags &= ~M_PKTHDR; 323149598Sandre bzero(&m->m_pkthdr, sizeof(struct pkthdr)); 324149598Sandre } 325149643Sandre if (m->m_type == MT_HEADER) 326149598Sandre m->m_type = MT_DATA; 327149643Sandre if (m != m0 && m->m_nextpkt != NULL) 328149598Sandre m->m_nextpkt = NULL; 329149598Sandre m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST); 330149598Sandre } 331149598Sandre} 332149598Sandre 333149598Sandre/* 334149648Sandre * Sanity checks on mbuf (chain) for use in KASSERT() and general 335149648Sandre * debugging. 336149648Sandre * Returns 0 or panics when bad and 1 on all tests passed. 337149648Sandre * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they 338149648Sandre * blow up later. 339149599Sandre */ 340149599Sandreint 341149599Sandrem_sanity(struct mbuf *m0, int sanitize) 342149599Sandre{ 343149599Sandre struct mbuf *m; 344149599Sandre caddr_t a, b; 345149599Sandre int pktlen = 0; 346149599Sandre 347168734Skmacy#ifdef INVARIANTS 348168734Skmacy#define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) 349168734Skmacy#else 350168734Skmacy#define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) 351168734Skmacy#endif 352149599Sandre 353149648Sandre for (m = m0; m != NULL; m = m->m_next) { 354149599Sandre /* 355149599Sandre * Basic pointer checks. If any of these fails then some 356149599Sandre * unrelated kernel memory before or after us is trashed. 357149599Sandre * No way to recover from that. 358149599Sandre */ 359149648Sandre a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf : 360149648Sandre ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) : 361149599Sandre (caddr_t)(&m->m_dat)) ); 362149599Sandre b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size : 363149648Sandre ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN))); 364149599Sandre if ((caddr_t)m->m_data < a) 365149599Sandre M_SANITY_ACTION("m_data outside mbuf data range left"); 366149599Sandre if ((caddr_t)m->m_data > b) 367149599Sandre M_SANITY_ACTION("m_data outside mbuf data range right"); 368149599Sandre if ((caddr_t)m->m_data + m->m_len > b) 369149599Sandre M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); 370149648Sandre if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) { 371149599Sandre if ((caddr_t)m->m_pkthdr.header < a || 372149599Sandre (caddr_t)m->m_pkthdr.header > b) 373149599Sandre M_SANITY_ACTION("m_pkthdr.header outside mbuf data range"); 374149599Sandre } 375149599Sandre 376149599Sandre /* m->m_nextpkt may only be set on first mbuf in chain. */ 377149648Sandre if (m != m0 && m->m_nextpkt != NULL) { 378149599Sandre if (sanitize) { 379149599Sandre m_freem(m->m_nextpkt); 380149599Sandre m->m_nextpkt = (struct mbuf *)0xDEADC0DE; 381149599Sandre } else 382149599Sandre M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); 383149599Sandre } 384149599Sandre 385149599Sandre /* packet length (not mbuf length!) calculation */ 386149599Sandre if (m0->m_flags & M_PKTHDR) 387149599Sandre pktlen += m->m_len; 388149599Sandre 389149599Sandre /* m_tags may only be attached to first mbuf in chain. */ 390149599Sandre if (m != m0 && m->m_flags & M_PKTHDR && 391149599Sandre !SLIST_EMPTY(&m->m_pkthdr.tags)) { 392149599Sandre if (sanitize) { 393149599Sandre m_tag_delete_chain(m, NULL); 394149599Sandre /* put in 0xDEADC0DE perhaps? */ 395149648Sandre } else 396149599Sandre M_SANITY_ACTION("m_tags on in-chain mbuf"); 397149599Sandre } 398149599Sandre 399149599Sandre /* M_PKTHDR may only be set on first mbuf in chain */ 400149599Sandre if (m != m0 && m->m_flags & M_PKTHDR) { 401149599Sandre if (sanitize) { 402149599Sandre bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); 403149599Sandre m->m_flags &= ~M_PKTHDR; 404149599Sandre /* put in 0xDEADCODE and leave hdr flag in */ 405149599Sandre } else 406149599Sandre M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); 407149599Sandre } 408149599Sandre } 409149648Sandre m = m0; 410149648Sandre if (pktlen && pktlen != m->m_pkthdr.len) { 411149599Sandre if (sanitize) 412149648Sandre m->m_pkthdr.len = 0; 413149599Sandre else 414149599Sandre M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); 415149599Sandre } 416149648Sandre return 1; 417149648Sandre 418149599Sandre#undef M_SANITY_ACTION 419149599Sandre} 420149599Sandre 421149599Sandre 422149599Sandre/* 423108466Ssam * "Move" mbuf pkthdr from "from" to "to". 424100960Srwatson * "from" must have M_PKTHDR set, and "to" must be empty. 425100960Srwatson */ 426100960Srwatsonvoid 427108466Ssamm_move_pkthdr(struct mbuf *to, struct mbuf *from) 428100960Srwatson{ 429100960Srwatson 430100960Srwatson#if 0 431108466Ssam /* see below for why these are not enabled */ 432113255Sdes M_ASSERTPKTHDR(to); 433113487Srwatson /* Note: with MAC, this may not be a good assertion. */ 434108466Ssam KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 435108466Ssam ("m_move_pkthdr: to has tags")); 436100960Srwatson#endif 437101007Srwatson#ifdef MAC 438113487Srwatson /* 439113487Srwatson * XXXMAC: It could be this should also occur for non-MAC? 440113487Srwatson */ 441101007Srwatson if (to->m_flags & M_PKTHDR) 442113487Srwatson m_tag_delete_chain(to, NULL); 443101007Srwatson#endif 444143302Ssam to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 445143302Ssam if ((to->m_flags & M_EXT) == 0) 446143302Ssam to->m_data = to->m_pktdat; 447108466Ssam to->m_pkthdr = from->m_pkthdr; /* especially tags */ 448108466Ssam SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 449108466Ssam from->m_flags &= ~M_PKTHDR; 450108466Ssam} 451108466Ssam 452108466Ssam/* 453108466Ssam * Duplicate "from"'s mbuf pkthdr in "to". 454108466Ssam * "from" must have M_PKTHDR set, and "to" must be empty. 455108466Ssam * In particular, this does a deep copy of the packet tags. 456108466Ssam */ 457108466Ssamint 458108466Ssamm_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 459108466Ssam{ 460108466Ssam 461108466Ssam#if 0 462108466Ssam /* 463108466Ssam * The mbuf allocator only initializes the pkthdr 464108466Ssam * when the mbuf is allocated with MGETHDR. Many users 465108466Ssam * (e.g. m_copy*, m_prepend) use MGET and then 466108466Ssam * smash the pkthdr as needed causing these 467108466Ssam * assertions to trip. For now just disable them. 468108466Ssam */ 469113255Sdes M_ASSERTPKTHDR(to); 470113487Srwatson /* Note: with MAC, this may not be a good assertion. */ 471108466Ssam KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 472108466Ssam#endif 473132488Salfred MBUF_CHECKSLEEP(how); 474108466Ssam#ifdef MAC 475108466Ssam if (to->m_flags & M_PKTHDR) 476113487Srwatson m_tag_delete_chain(to, NULL); 477108466Ssam#endif 478112733Ssilby to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 479112733Ssilby if ((to->m_flags & M_EXT) == 0) 480112733Ssilby to->m_data = to->m_pktdat; 481100960Srwatson to->m_pkthdr = from->m_pkthdr; 482108466Ssam SLIST_INIT(&to->m_pkthdr.tags); 483113480Srwatson return (m_tag_copy_chain(to, from, MBTOM(how))); 484100960Srwatson} 485100960Srwatson 486100960Srwatson/* 4871541Srgrimes * Lesser-used path for M_PREPEND: 4881541Srgrimes * allocate new mbuf to prepend to chain, 4891541Srgrimes * copy junk along. 4901541Srgrimes */ 4911541Srgrimesstruct mbuf * 49272356Sbmilekicm_prepend(struct mbuf *m, int len, int how) 4931541Srgrimes{ 4941541Srgrimes struct mbuf *mn; 4951541Srgrimes 496117770Ssilby if (m->m_flags & M_PKTHDR) 497117770Ssilby MGETHDR(mn, how, m->m_type); 498117770Ssilby else 499117770Ssilby MGET(mn, how, m->m_type); 50072356Sbmilekic if (mn == NULL) { 5011541Srgrimes m_freem(m); 50272356Sbmilekic return (NULL); 5031541Srgrimes } 504113487Srwatson if (m->m_flags & M_PKTHDR) 505108466Ssam M_MOVE_PKTHDR(mn, m); 5061541Srgrimes mn->m_next = m; 5071541Srgrimes m = mn; 508165447Srrs if(m->m_flags & M_PKTHDR) { 509165447Srrs if (len < MHLEN) 510165447Srrs MH_ALIGN(m, len); 511165447Srrs } else { 512165447Srrs if (len < MLEN) 513165447Srrs M_ALIGN(m, len); 514165447Srrs } 5151541Srgrimes m->m_len = len; 5161541Srgrimes return (m); 5171541Srgrimes} 5181541Srgrimes 5191541Srgrimes/* 5201541Srgrimes * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 5211541Srgrimes * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 522177599Sru * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller. 52354002Sarchie * Note that the copy is read-only, because clusters are not copied, 52454002Sarchie * only their reference counts are incremented. 5251541Srgrimes */ 5261541Srgrimesstruct mbuf * 52772356Sbmilekicm_copym(struct mbuf *m, int off0, int len, int wait) 5281541Srgrimes{ 52972356Sbmilekic struct mbuf *n, **np; 53072356Sbmilekic int off = off0; 5311541Srgrimes struct mbuf *top; 5321541Srgrimes int copyhdr = 0; 5331541Srgrimes 53452201Salfred KASSERT(off >= 0, ("m_copym, negative off %d", off)); 53552201Salfred KASSERT(len >= 0, ("m_copym, negative len %d", len)); 536132488Salfred MBUF_CHECKSLEEP(wait); 5371541Srgrimes if (off == 0 && m->m_flags & M_PKTHDR) 5381541Srgrimes copyhdr = 1; 5391541Srgrimes while (off > 0) { 54052201Salfred KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 5411541Srgrimes if (off < m->m_len) 5421541Srgrimes break; 5431541Srgrimes off -= m->m_len; 5441541Srgrimes m = m->m_next; 5451541Srgrimes } 5461541Srgrimes np = ⊤ 5471541Srgrimes top = 0; 5481541Srgrimes while (len > 0) { 54972356Sbmilekic if (m == NULL) { 55052201Salfred KASSERT(len == M_COPYALL, 55152201Salfred ("m_copym, length > size of mbuf chain")); 5521541Srgrimes break; 5531541Srgrimes } 554117770Ssilby if (copyhdr) 555117770Ssilby MGETHDR(n, wait, m->m_type); 556117770Ssilby else 557117770Ssilby MGET(n, wait, m->m_type); 5581541Srgrimes *np = n; 55972356Sbmilekic if (n == NULL) 5601541Srgrimes goto nospace; 5611541Srgrimes if (copyhdr) { 562108466Ssam if (!m_dup_pkthdr(n, m, wait)) 563108466Ssam goto nospace; 5641541Srgrimes if (len == M_COPYALL) 5651541Srgrimes n->m_pkthdr.len -= off0; 5661541Srgrimes else 5671541Srgrimes n->m_pkthdr.len = len; 5681541Srgrimes copyhdr = 0; 5691541Srgrimes } 5701541Srgrimes n->m_len = min(len, m->m_len - off); 5711541Srgrimes if (m->m_flags & M_EXT) { 5721541Srgrimes n->m_data = m->m_data + off; 573151976Sandre mb_dupcl(n, m); 5741541Srgrimes } else 5751541Srgrimes bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 576103569Sbmilekic (u_int)n->m_len); 5771541Srgrimes if (len != M_COPYALL) 5781541Srgrimes len -= n->m_len; 5791541Srgrimes off = 0; 5801541Srgrimes m = m->m_next; 5811541Srgrimes np = &n->m_next; 5821541Srgrimes } 58378592Sbmilekic if (top == NULL) 58478592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 58578592Sbmilekic 5861541Srgrimes return (top); 5871541Srgrimesnospace: 5881541Srgrimes m_freem(top); 58978592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 59072356Sbmilekic return (NULL); 5911541Srgrimes} 5921541Srgrimes 5931541Srgrimes/* 594149602Sandre * Returns mbuf chain with new head for the prepending case. 595149602Sandre * Copies from mbuf (chain) n from off for len to mbuf (chain) m 596149602Sandre * either prepending or appending the data. 597149602Sandre * The resulting mbuf (chain) m is fully writeable. 598149602Sandre * m is destination (is made writeable) 599149602Sandre * n is source, off is offset in source, len is len from offset 600149602Sandre * dir, 0 append, 1 prepend 601149602Sandre * how, wait or nowait 602149602Sandre */ 603149602Sandre 604149602Sandrestatic int 605149602Sandrem_bcopyxxx(void *s, void *t, u_int len) 606149602Sandre{ 607149602Sandre bcopy(s, t, (size_t)len); 608149602Sandre return 0; 609149602Sandre} 610149602Sandre 611149602Sandrestruct mbuf * 612149602Sandrem_copymdata(struct mbuf *m, struct mbuf *n, int off, int len, 613149602Sandre int prep, int how) 614149602Sandre{ 615151976Sandre struct mbuf *mm, *x, *z, *prev = NULL; 616149602Sandre caddr_t p; 617151976Sandre int i, nlen = 0; 618149602Sandre caddr_t buf[MLEN]; 619149602Sandre 620149602Sandre KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source")); 621149602Sandre KASSERT(off >= 0, ("m_copymdata, negative off %d", off)); 622149602Sandre KASSERT(len >= 0, ("m_copymdata, negative len %d", len)); 623149602Sandre KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep)); 624149602Sandre 625151976Sandre mm = m; 626151976Sandre if (!prep) { 627151976Sandre while(mm->m_next) { 628151976Sandre prev = mm; 629151976Sandre mm = mm->m_next; 630149602Sandre } 631149602Sandre } 632149602Sandre for (z = n; z != NULL; z = z->m_next) 633149602Sandre nlen += z->m_len; 634149602Sandre if (len == M_COPYALL) 635149602Sandre len = nlen - off; 636149602Sandre if (off + len > nlen || len < 1) 637149602Sandre return NULL; 638149602Sandre 639151976Sandre if (!M_WRITABLE(mm)) { 640151976Sandre /* XXX: Use proper m_xxx function instead. */ 641151976Sandre x = m_getcl(how, MT_DATA, mm->m_flags); 642151976Sandre if (x == NULL) 643151976Sandre return NULL; 644151976Sandre bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size); 645151976Sandre p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf); 646151976Sandre x->m_data = p; 647151976Sandre mm->m_next = NULL; 648151976Sandre if (mm != m) 649151976Sandre prev->m_next = x; 650151976Sandre m_free(mm); 651151976Sandre mm = x; 652151976Sandre } 653151976Sandre 654149602Sandre /* 655149602Sandre * Append/prepend the data. Allocating mbufs as necessary. 656149602Sandre */ 657149602Sandre /* Shortcut if enough free space in first/last mbuf. */ 658149602Sandre if (!prep && M_TRAILINGSPACE(mm) >= len) { 659149602Sandre m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) + 660149602Sandre mm->m_len); 661149602Sandre mm->m_len += len; 662149602Sandre mm->m_pkthdr.len += len; 663149602Sandre return m; 664149602Sandre } 665149602Sandre if (prep && M_LEADINGSPACE(mm) >= len) { 666149602Sandre mm->m_data = mtod(mm, caddr_t) - len; 667149602Sandre m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t)); 668149602Sandre mm->m_len += len; 669149602Sandre mm->m_pkthdr.len += len; 670149602Sandre return mm; 671149602Sandre } 672149602Sandre 673149602Sandre /* Expand first/last mbuf to cluster if possible. */ 674149602Sandre if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) { 675149602Sandre bcopy(mm->m_data, &buf, mm->m_len); 676149602Sandre m_clget(mm, how); 677149602Sandre if (!(mm->m_flags & M_EXT)) 678149602Sandre return NULL; 679149602Sandre bcopy(&buf, mm->m_ext.ext_buf, mm->m_len); 680149602Sandre mm->m_data = mm->m_ext.ext_buf; 681149602Sandre mm->m_pkthdr.header = NULL; 682149602Sandre } 683149602Sandre if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) { 684149602Sandre bcopy(mm->m_data, &buf, mm->m_len); 685149602Sandre m_clget(mm, how); 686149602Sandre if (!(mm->m_flags & M_EXT)) 687149602Sandre return NULL; 688149602Sandre bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf + 689149602Sandre mm->m_ext.ext_size - mm->m_len, mm->m_len); 690149602Sandre mm->m_data = (caddr_t)mm->m_ext.ext_buf + 691149602Sandre mm->m_ext.ext_size - mm->m_len; 692149602Sandre mm->m_pkthdr.header = NULL; 693149602Sandre } 694149602Sandre 695149602Sandre /* Append/prepend as many mbuf (clusters) as necessary to fit len. */ 696149602Sandre if (!prep && len > M_TRAILINGSPACE(mm)) { 697149602Sandre if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA)) 698149602Sandre return NULL; 699149602Sandre } 700149602Sandre if (prep && len > M_LEADINGSPACE(mm)) { 701149602Sandre if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA))) 702149602Sandre return NULL; 703149602Sandre i = 0; 704149602Sandre for (x = z; x != NULL; x = x->m_next) { 705149602Sandre i += x->m_flags & M_EXT ? x->m_ext.ext_size : 706149602Sandre (x->m_flags & M_PKTHDR ? MHLEN : MLEN); 707149602Sandre if (!x->m_next) 708149602Sandre break; 709149602Sandre } 710149602Sandre z->m_data += i - len; 711149602Sandre m_move_pkthdr(mm, z); 712149602Sandre x->m_next = mm; 713149602Sandre mm = z; 714149602Sandre } 715149602Sandre 716149602Sandre /* Seek to start position in source mbuf. Optimization for long chains. */ 717149602Sandre while (off > 0) { 718149602Sandre if (off < n->m_len) 719149602Sandre break; 720149602Sandre off -= n->m_len; 721149602Sandre n = n->m_next; 722149602Sandre } 723149602Sandre 724149602Sandre /* Copy data into target mbuf. */ 725149602Sandre z = mm; 726149602Sandre while (len > 0) { 727149602Sandre KASSERT(z != NULL, ("m_copymdata, falling off target edge")); 728149602Sandre i = M_TRAILINGSPACE(z); 729149602Sandre m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len); 730149602Sandre z->m_len += i; 731149602Sandre /* fixup pkthdr.len if necessary */ 732149602Sandre if ((prep ? mm : m)->m_flags & M_PKTHDR) 733149602Sandre (prep ? mm : m)->m_pkthdr.len += i; 734149602Sandre off += i; 735149602Sandre len -= i; 736149602Sandre z = z->m_next; 737149602Sandre } 738149602Sandre return (prep ? mm : m); 739149602Sandre} 740149602Sandre 741149602Sandre/* 74215689Swollman * Copy an entire packet, including header (which must be present). 74315689Swollman * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 74454002Sarchie * Note that the copy is read-only, because clusters are not copied, 74554002Sarchie * only their reference counts are incremented. 74672750Sluigi * Preserve alignment of the first mbuf so if the creator has left 74772750Sluigi * some room at the beginning (e.g. for inserting protocol headers) 74872750Sluigi * the copies still have the room available. 74915689Swollman */ 75015689Swollmanstruct mbuf * 75172356Sbmilekicm_copypacket(struct mbuf *m, int how) 75215689Swollman{ 75315689Swollman struct mbuf *top, *n, *o; 75415689Swollman 755132488Salfred MBUF_CHECKSLEEP(how); 75615689Swollman MGET(n, how, m->m_type); 75715689Swollman top = n; 75872356Sbmilekic if (n == NULL) 75915689Swollman goto nospace; 76015689Swollman 761108466Ssam if (!m_dup_pkthdr(n, m, how)) 762108466Ssam goto nospace; 76315689Swollman n->m_len = m->m_len; 76415689Swollman if (m->m_flags & M_EXT) { 76515689Swollman n->m_data = m->m_data; 766151976Sandre mb_dupcl(n, m); 76715689Swollman } else { 76872750Sluigi n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 76915689Swollman bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 77015689Swollman } 77115689Swollman 77215689Swollman m = m->m_next; 77315689Swollman while (m) { 77415689Swollman MGET(o, how, m->m_type); 77572356Sbmilekic if (o == NULL) 77615689Swollman goto nospace; 77715689Swollman 77815689Swollman n->m_next = o; 77915689Swollman n = n->m_next; 78015689Swollman 78115689Swollman n->m_len = m->m_len; 78215689Swollman if (m->m_flags & M_EXT) { 78315689Swollman n->m_data = m->m_data; 784151976Sandre mb_dupcl(n, m); 78515689Swollman } else { 78615689Swollman bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 78715689Swollman } 78815689Swollman 78915689Swollman m = m->m_next; 79015689Swollman } 79115689Swollman return top; 79215689Swollmannospace: 79315689Swollman m_freem(top); 79478592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 79572356Sbmilekic return (NULL); 79615689Swollman} 79715689Swollman 79815689Swollman/* 7991541Srgrimes * Copy data from an mbuf chain starting "off" bytes from the beginning, 8001541Srgrimes * continuing for "len" bytes, into the indicated buffer. 8011541Srgrimes */ 8021549Srgrimesvoid 80381907Sjulianm_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 8041541Srgrimes{ 805103569Sbmilekic u_int count; 8061541Srgrimes 80752201Salfred KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 80852201Salfred KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 8091541Srgrimes while (off > 0) { 81052201Salfred KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 8111541Srgrimes if (off < m->m_len) 8121541Srgrimes break; 8131541Srgrimes off -= m->m_len; 8141541Srgrimes m = m->m_next; 8151541Srgrimes } 8161541Srgrimes while (len > 0) { 81752201Salfred KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 8181541Srgrimes count = min(m->m_len - off, len); 8191541Srgrimes bcopy(mtod(m, caddr_t) + off, cp, count); 8201541Srgrimes len -= count; 8211541Srgrimes cp += count; 8221541Srgrimes off = 0; 8231541Srgrimes m = m->m_next; 8241541Srgrimes } 8251541Srgrimes} 8261541Srgrimes 8271541Srgrimes/* 82854002Sarchie * Copy a packet header mbuf chain into a completely new chain, including 82954002Sarchie * copying any mbuf clusters. Use this instead of m_copypacket() when 83054002Sarchie * you need a writable copy of an mbuf chain. 83154002Sarchie */ 83254002Sarchiestruct mbuf * 83372356Sbmilekicm_dup(struct mbuf *m, int how) 83454002Sarchie{ 83554002Sarchie struct mbuf **p, *top = NULL; 83654002Sarchie int remain, moff, nsize; 83754002Sarchie 838132488Salfred MBUF_CHECKSLEEP(how); 83954002Sarchie /* Sanity check */ 84054002Sarchie if (m == NULL) 84172356Sbmilekic return (NULL); 842113255Sdes M_ASSERTPKTHDR(m); 84354002Sarchie 84454002Sarchie /* While there's more data, get a new mbuf, tack it on, and fill it */ 84554002Sarchie remain = m->m_pkthdr.len; 84654002Sarchie moff = 0; 84754002Sarchie p = ⊤ 84854002Sarchie while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 84954002Sarchie struct mbuf *n; 85054002Sarchie 85154002Sarchie /* Get the next new mbuf */ 852129906Sbmilekic if (remain >= MINCLSIZE) { 853129906Sbmilekic n = m_getcl(how, m->m_type, 0); 854129906Sbmilekic nsize = MCLBYTES; 855129906Sbmilekic } else { 856129906Sbmilekic n = m_get(how, m->m_type); 857129906Sbmilekic nsize = MLEN; 858129906Sbmilekic } 85954002Sarchie if (n == NULL) 86054002Sarchie goto nospace; 861129906Sbmilekic 862129906Sbmilekic if (top == NULL) { /* First one, must be PKTHDR */ 863129906Sbmilekic if (!m_dup_pkthdr(n, m, how)) { 864129906Sbmilekic m_free(n); 865108466Ssam goto nospace; 866129906Sbmilekic } 867153428Semaste if ((n->m_flags & M_EXT) == 0) 868153428Semaste nsize = MHLEN; 86954002Sarchie } 87054002Sarchie n->m_len = 0; 87154002Sarchie 87254002Sarchie /* Link it into the new chain */ 87354002Sarchie *p = n; 87454002Sarchie p = &n->m_next; 87554002Sarchie 87654002Sarchie /* Copy data from original mbuf(s) into new mbuf */ 87754002Sarchie while (n->m_len < nsize && m != NULL) { 87854002Sarchie int chunk = min(nsize - n->m_len, m->m_len - moff); 87954002Sarchie 88054002Sarchie bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 88154002Sarchie moff += chunk; 88254002Sarchie n->m_len += chunk; 88354002Sarchie remain -= chunk; 88454002Sarchie if (moff == m->m_len) { 88554002Sarchie m = m->m_next; 88654002Sarchie moff = 0; 88754002Sarchie } 88854002Sarchie } 88954002Sarchie 89054002Sarchie /* Check correct total mbuf length */ 89154002Sarchie KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 89287594Sobrien ("%s: bogus m_pkthdr.len", __func__)); 89354002Sarchie } 89454002Sarchie return (top); 89554002Sarchie 89654002Sarchienospace: 89754002Sarchie m_freem(top); 89878592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 89972356Sbmilekic return (NULL); 90054002Sarchie} 90154002Sarchie 90254002Sarchie/* 9031541Srgrimes * Concatenate mbuf chain n to m. 9041541Srgrimes * Both chains must be of the same type (e.g. MT_DATA). 9051541Srgrimes * Any m_pkthdr is not updated. 9061541Srgrimes */ 9071549Srgrimesvoid 90872356Sbmilekicm_cat(struct mbuf *m, struct mbuf *n) 9091541Srgrimes{ 9101541Srgrimes while (m->m_next) 9111541Srgrimes m = m->m_next; 9121541Srgrimes while (n) { 9131541Srgrimes if (m->m_flags & M_EXT || 9141541Srgrimes m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 9151541Srgrimes /* just join the two chains */ 9161541Srgrimes m->m_next = n; 9171541Srgrimes return; 9181541Srgrimes } 9191541Srgrimes /* splat the data from one into the other */ 9201541Srgrimes bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 9211541Srgrimes (u_int)n->m_len); 9221541Srgrimes m->m_len += n->m_len; 9231541Srgrimes n = m_free(n); 9241541Srgrimes } 9251541Srgrimes} 9261541Srgrimes 9271549Srgrimesvoid 92872356Sbmilekicm_adj(struct mbuf *mp, int req_len) 9291541Srgrimes{ 93072356Sbmilekic int len = req_len; 93172356Sbmilekic struct mbuf *m; 93272356Sbmilekic int count; 9331541Srgrimes 9341541Srgrimes if ((m = mp) == NULL) 9351541Srgrimes return; 9361541Srgrimes if (len >= 0) { 9371541Srgrimes /* 9381541Srgrimes * Trim from head. 9391541Srgrimes */ 9401541Srgrimes while (m != NULL && len > 0) { 9411541Srgrimes if (m->m_len <= len) { 9421541Srgrimes len -= m->m_len; 9431541Srgrimes m->m_len = 0; 9441541Srgrimes m = m->m_next; 9451541Srgrimes } else { 9461541Srgrimes m->m_len -= len; 9471541Srgrimes m->m_data += len; 9481541Srgrimes len = 0; 9491541Srgrimes } 9501541Srgrimes } 9511541Srgrimes m = mp; 9521541Srgrimes if (mp->m_flags & M_PKTHDR) 9531541Srgrimes m->m_pkthdr.len -= (req_len - len); 9541541Srgrimes } else { 9551541Srgrimes /* 9561541Srgrimes * Trim from tail. Scan the mbuf chain, 9571541Srgrimes * calculating its length and finding the last mbuf. 9581541Srgrimes * If the adjustment only affects this mbuf, then just 9591541Srgrimes * adjust and return. Otherwise, rescan and truncate 9601541Srgrimes * after the remaining size. 9611541Srgrimes */ 9621541Srgrimes len = -len; 9631541Srgrimes count = 0; 9641541Srgrimes for (;;) { 9651541Srgrimes count += m->m_len; 9661541Srgrimes if (m->m_next == (struct mbuf *)0) 9671541Srgrimes break; 9681541Srgrimes m = m->m_next; 9691541Srgrimes } 9701541Srgrimes if (m->m_len >= len) { 9711541Srgrimes m->m_len -= len; 9721541Srgrimes if (mp->m_flags & M_PKTHDR) 9731541Srgrimes mp->m_pkthdr.len -= len; 9741541Srgrimes return; 9751541Srgrimes } 9761541Srgrimes count -= len; 9771541Srgrimes if (count < 0) 9781541Srgrimes count = 0; 9791541Srgrimes /* 9801541Srgrimes * Correct length for chain is "count". 9811541Srgrimes * Find the mbuf with last data, adjust its length, 9821541Srgrimes * and toss data from remaining mbufs on chain. 9831541Srgrimes */ 9841541Srgrimes m = mp; 9851541Srgrimes if (m->m_flags & M_PKTHDR) 9861541Srgrimes m->m_pkthdr.len = count; 9871541Srgrimes for (; m; m = m->m_next) { 9881541Srgrimes if (m->m_len >= count) { 9891541Srgrimes m->m_len = count; 990142350Ssam if (m->m_next != NULL) { 991142350Ssam m_freem(m->m_next); 992142350Ssam m->m_next = NULL; 993142350Ssam } 9941541Srgrimes break; 9951541Srgrimes } 9961541Srgrimes count -= m->m_len; 9971541Srgrimes } 9981541Srgrimes } 9991541Srgrimes} 10001541Srgrimes 10011541Srgrimes/* 10021541Srgrimes * Rearange an mbuf chain so that len bytes are contiguous 10031541Srgrimes * and in the data area of an mbuf (so that mtod and dtom 10041541Srgrimes * will work for a structure of size len). Returns the resulting 10051541Srgrimes * mbuf chain on success, frees it and returns null on failure. 10061541Srgrimes * If there is room, it will add up to max_protohdr-len extra bytes to the 10071541Srgrimes * contiguous region in an attempt to avoid being called next time. 10081541Srgrimes */ 10091541Srgrimesstruct mbuf * 101072356Sbmilekicm_pullup(struct mbuf *n, int len) 10111541Srgrimes{ 101272356Sbmilekic struct mbuf *m; 101372356Sbmilekic int count; 10141541Srgrimes int space; 10151541Srgrimes 10161541Srgrimes /* 10171541Srgrimes * If first mbuf has no cluster, and has room for len bytes 10181541Srgrimes * without shifting current data, pullup into it, 10191541Srgrimes * otherwise allocate a new mbuf to prepend to the chain. 10201541Srgrimes */ 10211541Srgrimes if ((n->m_flags & M_EXT) == 0 && 10221541Srgrimes n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 10231541Srgrimes if (n->m_len >= len) 10241541Srgrimes return (n); 10251541Srgrimes m = n; 10261541Srgrimes n = n->m_next; 10271541Srgrimes len -= m->m_len; 10281541Srgrimes } else { 10291541Srgrimes if (len > MHLEN) 10301541Srgrimes goto bad; 1031111119Simp MGET(m, M_DONTWAIT, n->m_type); 103272356Sbmilekic if (m == NULL) 10331541Srgrimes goto bad; 10341541Srgrimes m->m_len = 0; 1035108466Ssam if (n->m_flags & M_PKTHDR) 1036108466Ssam M_MOVE_PKTHDR(m, n); 10371541Srgrimes } 10381541Srgrimes space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 10391541Srgrimes do { 10401541Srgrimes count = min(min(max(len, max_protohdr), space), n->m_len); 10411541Srgrimes bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1042103569Sbmilekic (u_int)count); 10431541Srgrimes len -= count; 10441541Srgrimes m->m_len += count; 10451541Srgrimes n->m_len -= count; 10461541Srgrimes space -= count; 10471541Srgrimes if (n->m_len) 10481541Srgrimes n->m_data += count; 10491541Srgrimes else 10501541Srgrimes n = m_free(n); 10511541Srgrimes } while (len > 0 && n); 10521541Srgrimes if (len > 0) { 10531541Srgrimes (void) m_free(m); 10541541Srgrimes goto bad; 10551541Srgrimes } 10561541Srgrimes m->m_next = n; 10571541Srgrimes return (m); 10581541Srgrimesbad: 10591541Srgrimes m_freem(n); 106078592Sbmilekic mbstat.m_mpfail++; /* XXX: No consistency. */ 106172356Sbmilekic return (NULL); 10621541Srgrimes} 10631541Srgrimes 10641541Srgrimes/* 1065143761Sjmg * Like m_pullup(), except a new mbuf is always allocated, and we allow 1066143761Sjmg * the amount of empty space before the data in the new mbuf to be specified 1067143761Sjmg * (in the event that the caller expects to prepend later). 1068143761Sjmg */ 1069143761Sjmgint MSFail; 1070143761Sjmg 1071143761Sjmgstruct mbuf * 1072143761Sjmgm_copyup(struct mbuf *n, int len, int dstoff) 1073143761Sjmg{ 1074143761Sjmg struct mbuf *m; 1075143761Sjmg int count, space; 1076143761Sjmg 1077143761Sjmg if (len > (MHLEN - dstoff)) 1078143761Sjmg goto bad; 1079143761Sjmg MGET(m, M_DONTWAIT, n->m_type); 1080143761Sjmg if (m == NULL) 1081143761Sjmg goto bad; 1082143761Sjmg m->m_len = 0; 1083143761Sjmg if (n->m_flags & M_PKTHDR) 1084143761Sjmg M_MOVE_PKTHDR(m, n); 1085143761Sjmg m->m_data += dstoff; 1086143761Sjmg space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1087143761Sjmg do { 1088143761Sjmg count = min(min(max(len, max_protohdr), space), n->m_len); 1089143761Sjmg memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 1090143761Sjmg (unsigned)count); 1091143761Sjmg len -= count; 1092143761Sjmg m->m_len += count; 1093143761Sjmg n->m_len -= count; 1094143761Sjmg space -= count; 1095143761Sjmg if (n->m_len) 1096143761Sjmg n->m_data += count; 1097143761Sjmg else 1098143761Sjmg n = m_free(n); 1099143761Sjmg } while (len > 0 && n); 1100143761Sjmg if (len > 0) { 1101143761Sjmg (void) m_free(m); 1102143761Sjmg goto bad; 1103143761Sjmg } 1104143761Sjmg m->m_next = n; 1105143761Sjmg return (m); 1106143761Sjmg bad: 1107143761Sjmg m_freem(n); 1108143761Sjmg MSFail++; 1109143761Sjmg return (NULL); 1110143761Sjmg} 1111143761Sjmg 1112143761Sjmg/* 11131541Srgrimes * Partition an mbuf chain in two pieces, returning the tail -- 11141541Srgrimes * all but the first len0 bytes. In case of failure, it returns NULL and 11151541Srgrimes * attempts to restore the chain to its original state. 111697681Sarchie * 111797681Sarchie * Note that the resulting mbufs might be read-only, because the new 111897681Sarchie * mbuf can end up sharing an mbuf cluster with the original mbuf if 111997681Sarchie * the "breaking point" happens to lie within a cluster mbuf. Use the 112097681Sarchie * M_WRITABLE() macro to check for this case. 11211541Srgrimes */ 11221541Srgrimesstruct mbuf * 112372356Sbmilekicm_split(struct mbuf *m0, int len0, int wait) 11241541Srgrimes{ 112572356Sbmilekic struct mbuf *m, *n; 1126103569Sbmilekic u_int len = len0, remain; 11271541Srgrimes 1128132488Salfred MBUF_CHECKSLEEP(wait); 11291541Srgrimes for (m = m0; m && len > m->m_len; m = m->m_next) 11301541Srgrimes len -= m->m_len; 113172356Sbmilekic if (m == NULL) 113272356Sbmilekic return (NULL); 11331541Srgrimes remain = m->m_len - len; 11341541Srgrimes if (m0->m_flags & M_PKTHDR) { 11351541Srgrimes MGETHDR(n, wait, m0->m_type); 113672356Sbmilekic if (n == NULL) 113772356Sbmilekic return (NULL); 11381541Srgrimes n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 11391541Srgrimes n->m_pkthdr.len = m0->m_pkthdr.len - len0; 11401541Srgrimes m0->m_pkthdr.len = len0; 11411541Srgrimes if (m->m_flags & M_EXT) 11421541Srgrimes goto extpacket; 11431541Srgrimes if (remain > MHLEN) { 11441541Srgrimes /* m can't be the lead packet */ 11451541Srgrimes MH_ALIGN(n, 0); 11461541Srgrimes n->m_next = m_split(m, len, wait); 114772356Sbmilekic if (n->m_next == NULL) { 11481541Srgrimes (void) m_free(n); 114972356Sbmilekic return (NULL); 115094471Shsu } else { 115194471Shsu n->m_len = 0; 11521541Srgrimes return (n); 115394471Shsu } 11541541Srgrimes } else 11551541Srgrimes MH_ALIGN(n, remain); 11561541Srgrimes } else if (remain == 0) { 11571541Srgrimes n = m->m_next; 115872356Sbmilekic m->m_next = NULL; 11591541Srgrimes return (n); 11601541Srgrimes } else { 11611541Srgrimes MGET(n, wait, m->m_type); 116272356Sbmilekic if (n == NULL) 116372356Sbmilekic return (NULL); 11641541Srgrimes M_ALIGN(n, remain); 11651541Srgrimes } 11661541Srgrimesextpacket: 11671541Srgrimes if (m->m_flags & M_EXT) { 11681541Srgrimes n->m_data = m->m_data + len; 1169151976Sandre mb_dupcl(n, m); 11701541Srgrimes } else { 11711541Srgrimes bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 11721541Srgrimes } 11731541Srgrimes n->m_len = remain; 11741541Srgrimes m->m_len = len; 11751541Srgrimes n->m_next = m->m_next; 117672356Sbmilekic m->m_next = NULL; 11771541Srgrimes return (n); 11781541Srgrimes} 11791541Srgrimes/* 11801541Srgrimes * Routine to copy from device local memory into mbufs. 118178508Sbmilekic * Note that `off' argument is offset into first mbuf of target chain from 118278508Sbmilekic * which to begin copying the data to. 11831541Srgrimes */ 11841541Srgrimesstruct mbuf * 118578508Sbmilekicm_devget(char *buf, int totlen, int off, struct ifnet *ifp, 1186169624Srwatson void (*copy)(char *from, caddr_t to, u_int len)) 11871541Srgrimes{ 118872356Sbmilekic struct mbuf *m; 1189129906Sbmilekic struct mbuf *top = NULL, **mp = ⊤ 119078508Sbmilekic int len; 11911541Srgrimes 119278508Sbmilekic if (off < 0 || off > MHLEN) 119378508Sbmilekic return (NULL); 119478508Sbmilekic 1195129906Sbmilekic while (totlen > 0) { 1196129906Sbmilekic if (top == NULL) { /* First one, must be PKTHDR */ 1197129906Sbmilekic if (totlen + off >= MINCLSIZE) { 1198129906Sbmilekic m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 1199129906Sbmilekic len = MCLBYTES; 1200129906Sbmilekic } else { 1201129906Sbmilekic m = m_gethdr(M_DONTWAIT, MT_DATA); 1202129906Sbmilekic len = MHLEN; 12031541Srgrimes 1204129906Sbmilekic /* Place initial small packet/header at end of mbuf */ 1205129906Sbmilekic if (m && totlen + off + max_linkhdr <= MLEN) { 1206129906Sbmilekic m->m_data += max_linkhdr; 1207129906Sbmilekic len -= max_linkhdr; 1208129906Sbmilekic } 1209129906Sbmilekic } 1210129906Sbmilekic if (m == NULL) 1211129906Sbmilekic return NULL; 1212129906Sbmilekic m->m_pkthdr.rcvif = ifp; 1213129906Sbmilekic m->m_pkthdr.len = totlen; 1214129906Sbmilekic } else { 1215129906Sbmilekic if (totlen + off >= MINCLSIZE) { 1216129906Sbmilekic m = m_getcl(M_DONTWAIT, MT_DATA, 0); 1217129906Sbmilekic len = MCLBYTES; 1218129906Sbmilekic } else { 1219129906Sbmilekic m = m_get(M_DONTWAIT, MT_DATA); 1220129906Sbmilekic len = MLEN; 1221129906Sbmilekic } 122272356Sbmilekic if (m == NULL) { 12231541Srgrimes m_freem(top); 1224129906Sbmilekic return NULL; 12251541Srgrimes } 12261541Srgrimes } 122778508Sbmilekic if (off) { 122878508Sbmilekic m->m_data += off; 122978508Sbmilekic len -= off; 123078508Sbmilekic off = 0; 123178508Sbmilekic } 123278508Sbmilekic m->m_len = len = min(totlen, len); 12331541Srgrimes if (copy) 1234103569Sbmilekic copy(buf, mtod(m, caddr_t), (u_int)len); 12351541Srgrimes else 1236103569Sbmilekic bcopy(buf, mtod(m, caddr_t), (u_int)len); 123778508Sbmilekic buf += len; 12381541Srgrimes *mp = m; 12391541Srgrimes mp = &m->m_next; 12401541Srgrimes totlen -= len; 12411541Srgrimes } 12421541Srgrimes return (top); 12431541Srgrimes} 12443352Sphk 12453352Sphk/* 12463352Sphk * Copy data from a buffer back into the indicated mbuf chain, 12473352Sphk * starting "off" bytes from the beginning, extending the mbuf 12483352Sphk * chain if necessary. 12493352Sphk */ 12503352Sphkvoid 1251128402Sluigim_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 12523352Sphk{ 125372356Sbmilekic int mlen; 125472356Sbmilekic struct mbuf *m = m0, *n; 12553352Sphk int totlen = 0; 12563352Sphk 125772356Sbmilekic if (m0 == NULL) 12583352Sphk return; 12593352Sphk while (off > (mlen = m->m_len)) { 12603352Sphk off -= mlen; 12613352Sphk totlen += mlen; 126272356Sbmilekic if (m->m_next == NULL) { 1263129906Sbmilekic n = m_get(M_DONTWAIT, m->m_type); 126472356Sbmilekic if (n == NULL) 12653352Sphk goto out; 1266129906Sbmilekic bzero(mtod(n, caddr_t), MLEN); 12673352Sphk n->m_len = min(MLEN, len + off); 12683352Sphk m->m_next = n; 12693352Sphk } 12703352Sphk m = m->m_next; 12713352Sphk } 12723352Sphk while (len > 0) { 12733352Sphk mlen = min (m->m_len - off, len); 1274103569Sbmilekic bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 12753352Sphk cp += mlen; 12763352Sphk len -= mlen; 12773352Sphk mlen += off; 12783352Sphk off = 0; 12793352Sphk totlen += mlen; 12803352Sphk if (len == 0) 12813352Sphk break; 128272356Sbmilekic if (m->m_next == NULL) { 1283111119Simp n = m_get(M_DONTWAIT, m->m_type); 128472356Sbmilekic if (n == NULL) 12853352Sphk break; 12863352Sphk n->m_len = min(MLEN, len); 12873352Sphk m->m_next = n; 12883352Sphk } 12893352Sphk m = m->m_next; 12903352Sphk } 12913352Sphkout: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 12923352Sphk m->m_pkthdr.len = totlen; 12933352Sphk} 129452756Sphk 1295123557Sbms/* 1296138541Ssam * Append the specified data to the indicated mbuf chain, 1297138541Ssam * Extend the mbuf chain if the new data does not fit in 1298138541Ssam * existing space. 1299138541Ssam * 1300138541Ssam * Return 1 if able to complete the job; otherwise 0. 1301138541Ssam */ 1302138541Ssamint 1303138541Ssamm_append(struct mbuf *m0, int len, c_caddr_t cp) 1304138541Ssam{ 1305138541Ssam struct mbuf *m, *n; 1306138541Ssam int remainder, space; 1307138541Ssam 1308138541Ssam for (m = m0; m->m_next != NULL; m = m->m_next) 1309138541Ssam ; 1310138541Ssam remainder = len; 1311138541Ssam space = M_TRAILINGSPACE(m); 1312138541Ssam if (space > 0) { 1313138541Ssam /* 1314138541Ssam * Copy into available space. 1315138541Ssam */ 1316138541Ssam if (space > remainder) 1317138541Ssam space = remainder; 1318138541Ssam bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1319138541Ssam m->m_len += space; 1320138541Ssam cp += space, remainder -= space; 1321138541Ssam } 1322138541Ssam while (remainder > 0) { 1323138541Ssam /* 1324138541Ssam * Allocate a new mbuf; could check space 1325138541Ssam * and allocate a cluster instead. 1326138541Ssam */ 1327138541Ssam n = m_get(M_DONTWAIT, m->m_type); 1328138541Ssam if (n == NULL) 1329138541Ssam break; 1330138541Ssam n->m_len = min(MLEN, remainder); 1331138894Ssam bcopy(cp, mtod(n, caddr_t), n->m_len); 1332138894Ssam cp += n->m_len, remainder -= n->m_len; 1333138541Ssam m->m_next = n; 1334138541Ssam m = n; 1335138541Ssam } 1336138541Ssam if (m0->m_flags & M_PKTHDR) 1337138541Ssam m0->m_pkthdr.len += len - remainder; 1338138541Ssam return (remainder == 0); 1339138541Ssam} 1340138541Ssam 1341138541Ssam/* 1342123557Sbms * Apply function f to the data in an mbuf chain starting "off" bytes from 1343123557Sbms * the beginning, continuing for "len" bytes. 1344123557Sbms */ 1345123557Sbmsint 1346123557Sbmsm_apply(struct mbuf *m, int off, int len, 1347123564Sbms int (*f)(void *, void *, u_int), void *arg) 1348123557Sbms{ 1349123564Sbms u_int count; 1350123557Sbms int rval; 1351123557Sbms 1352123557Sbms KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1353123557Sbms KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1354123557Sbms while (off > 0) { 1355123557Sbms KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1356123557Sbms if (off < m->m_len) 1357123557Sbms break; 1358123557Sbms off -= m->m_len; 1359123557Sbms m = m->m_next; 1360123557Sbms } 1361123557Sbms while (len > 0) { 1362123557Sbms KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1363123557Sbms count = min(m->m_len - off, len); 1364123557Sbms rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1365123557Sbms if (rval) 1366123557Sbms return (rval); 1367123557Sbms len -= count; 1368123557Sbms off = 0; 1369123557Sbms m = m->m_next; 1370123557Sbms } 1371123557Sbms return (0); 1372123557Sbms} 1373123557Sbms 1374123557Sbms/* 1375123557Sbms * Return a pointer to mbuf/offset of location in mbuf chain. 1376123557Sbms */ 1377123557Sbmsstruct mbuf * 1378123557Sbmsm_getptr(struct mbuf *m, int loc, int *off) 1379123557Sbms{ 1380123557Sbms 1381123557Sbms while (loc >= 0) { 1382123564Sbms /* Normal end of search. */ 1383123557Sbms if (m->m_len > loc) { 1384123557Sbms *off = loc; 1385123557Sbms return (m); 1386123557Sbms } else { 1387123557Sbms loc -= m->m_len; 1388123557Sbms if (m->m_next == NULL) { 1389123557Sbms if (loc == 0) { 1390123564Sbms /* Point at the end of valid data. */ 1391123557Sbms *off = m->m_len; 1392123557Sbms return (m); 1393123564Sbms } 1394123564Sbms return (NULL); 1395123564Sbms } 1396123564Sbms m = m->m_next; 1397123557Sbms } 1398123557Sbms } 1399123557Sbms return (NULL); 1400123557Sbms} 1401123557Sbms 140252756Sphkvoid 1403135904Sjmgm_print(const struct mbuf *m, int maxlen) 140452756Sphk{ 140552756Sphk int len; 1406135904Sjmg int pdata; 140754906Seivind const struct mbuf *m2; 140852756Sphk 1409135904Sjmg if (m->m_flags & M_PKTHDR) 1410135904Sjmg len = m->m_pkthdr.len; 1411135904Sjmg else 1412135904Sjmg len = -1; 141352756Sphk m2 = m; 1414135904Sjmg while (m2 != NULL && (len == -1 || len)) { 1415135904Sjmg pdata = m2->m_len; 1416135904Sjmg if (maxlen != -1 && pdata > maxlen) 1417135904Sjmg pdata = maxlen; 1418135904Sjmg printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1419135904Sjmg m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1420135904Sjmg "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1421135904Sjmg "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1422135904Sjmg if (pdata) 1423156700Sjmg printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); 1424135904Sjmg if (len != -1) 1425135904Sjmg len -= m2->m_len; 142652756Sphk m2 = m2->m_next; 142752756Sphk } 1428135904Sjmg if (len > 0) 1429135904Sjmg printf("%d bytes unaccounted for.\n", len); 143052756Sphk return; 143152756Sphk} 1432103540Sphk 1433103569Sbmilekicu_int 1434103540Sphkm_fixhdr(struct mbuf *m0) 1435103540Sphk{ 1436103569Sbmilekic u_int len; 1437103540Sphk 1438103544Sphk len = m_length(m0, NULL); 1439103544Sphk m0->m_pkthdr.len = len; 1440103544Sphk return (len); 1441103544Sphk} 1442103544Sphk 1443103569Sbmilekicu_int 1444103544Sphkm_length(struct mbuf *m0, struct mbuf **last) 1445103544Sphk{ 1446103544Sphk struct mbuf *m; 1447103569Sbmilekic u_int len; 1448103544Sphk 1449103544Sphk len = 0; 1450103544Sphk for (m = m0; m != NULL; m = m->m_next) { 1451103540Sphk len += m->m_len; 1452103544Sphk if (m->m_next == NULL) 1453103544Sphk break; 1454103540Sphk } 1455103544Sphk if (last != NULL) 1456103544Sphk *last = m; 1457103544Sphk return (len); 1458103540Sphk} 1459112777Ssilby 1460112777Ssilby/* 1461112777Ssilby * Defragment a mbuf chain, returning the shortest possible 1462112777Ssilby * chain of mbufs and clusters. If allocation fails and 1463112777Ssilby * this cannot be completed, NULL will be returned, but 1464112777Ssilby * the passed in chain will be unchanged. Upon success, 1465112777Ssilby * the original chain will be freed, and the new chain 1466112777Ssilby * will be returned. 1467112777Ssilby * 1468112777Ssilby * If a non-packet header is passed in, the original 1469112777Ssilby * mbuf (chain?) will be returned unharmed. 1470112777Ssilby */ 1471112777Ssilbystruct mbuf * 1472112777Ssilbym_defrag(struct mbuf *m0, int how) 1473112777Ssilby{ 1474125472Ssilby struct mbuf *m_new = NULL, *m_final = NULL; 1475125472Ssilby int progress = 0, length; 1476112777Ssilby 1477132488Salfred MBUF_CHECKSLEEP(how); 1478112777Ssilby if (!(m0->m_flags & M_PKTHDR)) 1479112777Ssilby return (m0); 1480112777Ssilby 1481117770Ssilby m_fixhdr(m0); /* Needed sanity check */ 1482117770Ssilby 1483113490Ssilby#ifdef MBUF_STRESS_TEST 1484113490Ssilby if (m_defragrandomfailures) { 1485113490Ssilby int temp = arc4random() & 0xff; 1486113490Ssilby if (temp == 0xba) 1487113490Ssilby goto nospace; 1488113490Ssilby } 1489113490Ssilby#endif 1490112777Ssilby 1491112777Ssilby if (m0->m_pkthdr.len > MHLEN) 1492112777Ssilby m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1493112777Ssilby else 1494112777Ssilby m_final = m_gethdr(how, MT_DATA); 1495112777Ssilby 1496112777Ssilby if (m_final == NULL) 1497112777Ssilby goto nospace; 1498112777Ssilby 1499123740Speter if (m_dup_pkthdr(m_final, m0, how) == 0) 1500112777Ssilby goto nospace; 1501112777Ssilby 1502112777Ssilby m_new = m_final; 1503112777Ssilby 1504112777Ssilby while (progress < m0->m_pkthdr.len) { 1505112777Ssilby length = m0->m_pkthdr.len - progress; 1506112777Ssilby if (length > MCLBYTES) 1507112777Ssilby length = MCLBYTES; 1508112777Ssilby 1509112777Ssilby if (m_new == NULL) { 1510112777Ssilby if (length > MLEN) 1511112777Ssilby m_new = m_getcl(how, MT_DATA, 0); 1512112777Ssilby else 1513112777Ssilby m_new = m_get(how, MT_DATA); 1514112777Ssilby if (m_new == NULL) 1515112777Ssilby goto nospace; 1516112777Ssilby } 1517112777Ssilby 1518112777Ssilby m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1519112777Ssilby progress += length; 1520112777Ssilby m_new->m_len = length; 1521112777Ssilby if (m_new != m_final) 1522112777Ssilby m_cat(m_final, m_new); 1523112777Ssilby m_new = NULL; 1524112777Ssilby } 1525116455Ssilby#ifdef MBUF_STRESS_TEST 1526112777Ssilby if (m0->m_next == NULL) 1527112777Ssilby m_defraguseless++; 1528116455Ssilby#endif 1529112777Ssilby m_freem(m0); 1530112777Ssilby m0 = m_final; 1531116455Ssilby#ifdef MBUF_STRESS_TEST 1532112777Ssilby m_defragpackets++; 1533112777Ssilby m_defragbytes += m0->m_pkthdr.len; 1534116455Ssilby#endif 1535112777Ssilby return (m0); 1536112777Ssilbynospace: 1537116455Ssilby#ifdef MBUF_STRESS_TEST 1538112777Ssilby m_defragfailure++; 1539116455Ssilby#endif 1540112777Ssilby if (m_final) 1541112777Ssilby m_freem(m_final); 1542112777Ssilby return (NULL); 1543112777Ssilby} 1544119644Ssilby 1545175414Ssam/* 1546175414Ssam * Defragment an mbuf chain, returning at most maxfrags separate 1547175414Ssam * mbufs+clusters. If this is not possible NULL is returned and 1548175414Ssam * the original mbuf chain is left in it's present (potentially 1549175414Ssam * modified) state. We use two techniques: collapsing consecutive 1550175414Ssam * mbufs and replacing consecutive mbufs by a cluster. 1551175414Ssam * 1552175414Ssam * NB: this should really be named m_defrag but that name is taken 1553175414Ssam */ 1554175414Ssamstruct mbuf * 1555175414Ssamm_collapse(struct mbuf *m0, int how, int maxfrags) 1556175414Ssam{ 1557175414Ssam struct mbuf *m, *n, *n2, **prev; 1558175414Ssam u_int curfrags; 1559175414Ssam 1560175414Ssam /* 1561175414Ssam * Calculate the current number of frags. 1562175414Ssam */ 1563175414Ssam curfrags = 0; 1564175414Ssam for (m = m0; m != NULL; m = m->m_next) 1565175414Ssam curfrags++; 1566175414Ssam /* 1567175414Ssam * First, try to collapse mbufs. Note that we always collapse 1568175414Ssam * towards the front so we don't need to deal with moving the 1569175414Ssam * pkthdr. This may be suboptimal if the first mbuf has much 1570175414Ssam * less data than the following. 1571175414Ssam */ 1572175414Ssam m = m0; 1573175414Ssamagain: 1574175414Ssam for (;;) { 1575175414Ssam n = m->m_next; 1576175414Ssam if (n == NULL) 1577175414Ssam break; 1578175414Ssam if ((m->m_flags & M_RDONLY) == 0 && 1579175414Ssam n->m_len < M_TRAILINGSPACE(m)) { 1580175414Ssam bcopy(mtod(n, void *), mtod(m, char *) + m->m_len, 1581175414Ssam n->m_len); 1582175414Ssam m->m_len += n->m_len; 1583175414Ssam m->m_next = n->m_next; 1584175414Ssam m_free(n); 1585175414Ssam if (--curfrags <= maxfrags) 1586175414Ssam return m0; 1587175414Ssam } else 1588175414Ssam m = n; 1589175414Ssam } 1590175414Ssam KASSERT(maxfrags > 1, 1591175414Ssam ("maxfrags %u, but normal collapse failed", maxfrags)); 1592175414Ssam /* 1593175414Ssam * Collapse consecutive mbufs to a cluster. 1594175414Ssam */ 1595175414Ssam prev = &m0->m_next; /* NB: not the first mbuf */ 1596175414Ssam while ((n = *prev) != NULL) { 1597175414Ssam if ((n2 = n->m_next) != NULL && 1598175414Ssam n->m_len + n2->m_len < MCLBYTES) { 1599175414Ssam m = m_getcl(how, MT_DATA, 0); 1600175414Ssam if (m == NULL) 1601175414Ssam goto bad; 1602175414Ssam bcopy(mtod(n, void *), mtod(m, void *), n->m_len); 1603175414Ssam bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len, 1604175414Ssam n2->m_len); 1605175414Ssam m->m_len = n->m_len + n2->m_len; 1606175414Ssam m->m_next = n2->m_next; 1607175414Ssam *prev = m; 1608175414Ssam m_free(n); 1609175414Ssam m_free(n2); 1610175414Ssam if (--curfrags <= maxfrags) /* +1 cl -2 mbufs */ 1611175414Ssam return m0; 1612175414Ssam /* 1613175414Ssam * Still not there, try the normal collapse 1614175414Ssam * again before we allocate another cluster. 1615175414Ssam */ 1616175414Ssam goto again; 1617175414Ssam } 1618175414Ssam prev = &n->m_next; 1619175414Ssam } 1620175414Ssam /* 1621175414Ssam * No place where we can collapse to a cluster; punt. 1622175414Ssam * This can occur if, for example, you request 2 frags 1623175414Ssam * but the packet requires that both be clusters (we 1624175414Ssam * never reallocate the first mbuf to avoid moving the 1625175414Ssam * packet header). 1626175414Ssam */ 1627175414Ssambad: 1628175414Ssam return NULL; 1629175414Ssam} 1630175414Ssam 1631119644Ssilby#ifdef MBUF_STRESS_TEST 1632119644Ssilby 1633119644Ssilby/* 1634119644Ssilby * Fragment an mbuf chain. There's no reason you'd ever want to do 1635119644Ssilby * this in normal usage, but it's great for stress testing various 1636119644Ssilby * mbuf consumers. 1637119644Ssilby * 1638119644Ssilby * If fragmentation is not possible, the original chain will be 1639119644Ssilby * returned. 1640119644Ssilby * 1641119644Ssilby * Possible length values: 1642119644Ssilby * 0 no fragmentation will occur 1643119644Ssilby * > 0 each fragment will be of the specified length 1644119644Ssilby * -1 each fragment will be the same random value in length 1645119644Ssilby * -2 each fragment's length will be entirely random 1646119644Ssilby * (Random values range from 1 to 256) 1647119644Ssilby */ 1648119644Ssilbystruct mbuf * 1649119644Ssilbym_fragment(struct mbuf *m0, int how, int length) 1650119644Ssilby{ 1651125472Ssilby struct mbuf *m_new = NULL, *m_final = NULL; 1652125472Ssilby int progress = 0; 1653119644Ssilby 1654119644Ssilby if (!(m0->m_flags & M_PKTHDR)) 1655119644Ssilby return (m0); 1656119644Ssilby 1657119644Ssilby if ((length == 0) || (length < -2)) 1658119644Ssilby return (m0); 1659119644Ssilby 1660119644Ssilby m_fixhdr(m0); /* Needed sanity check */ 1661119644Ssilby 1662119644Ssilby m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1663119644Ssilby 1664119644Ssilby if (m_final == NULL) 1665119644Ssilby goto nospace; 1666119644Ssilby 1667123823Ssilby if (m_dup_pkthdr(m_final, m0, how) == 0) 1668119644Ssilby goto nospace; 1669119644Ssilby 1670119644Ssilby m_new = m_final; 1671119644Ssilby 1672119644Ssilby if (length == -1) 1673119644Ssilby length = 1 + (arc4random() & 255); 1674119644Ssilby 1675119644Ssilby while (progress < m0->m_pkthdr.len) { 1676119644Ssilby int fraglen; 1677119644Ssilby 1678119644Ssilby if (length > 0) 1679119644Ssilby fraglen = length; 1680119644Ssilby else 1681119644Ssilby fraglen = 1 + (arc4random() & 255); 1682119644Ssilby if (fraglen > m0->m_pkthdr.len - progress) 1683119644Ssilby fraglen = m0->m_pkthdr.len - progress; 1684119644Ssilby 1685119644Ssilby if (fraglen > MCLBYTES) 1686119644Ssilby fraglen = MCLBYTES; 1687119644Ssilby 1688119644Ssilby if (m_new == NULL) { 1689119644Ssilby m_new = m_getcl(how, MT_DATA, 0); 1690119644Ssilby if (m_new == NULL) 1691119644Ssilby goto nospace; 1692119644Ssilby } 1693119644Ssilby 1694119644Ssilby m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1695119644Ssilby progress += fraglen; 1696119644Ssilby m_new->m_len = fraglen; 1697119644Ssilby if (m_new != m_final) 1698119644Ssilby m_cat(m_final, m_new); 1699119644Ssilby m_new = NULL; 1700119644Ssilby } 1701119644Ssilby m_freem(m0); 1702119644Ssilby m0 = m_final; 1703119644Ssilby return (m0); 1704119644Ssilbynospace: 1705119644Ssilby if (m_final) 1706119644Ssilby m_freem(m_final); 1707119644Ssilby /* Return the original chain on failure */ 1708119644Ssilby return (m0); 1709119644Ssilby} 1710119644Ssilby 1711119644Ssilby#endif 1712125296Ssilby 1713163915Sandre/* 1714163915Sandre * Copy the contents of uio into a properly sized mbuf chain. 1715163915Sandre */ 1716125296Ssilbystruct mbuf * 1717163915Sandrem_uiotombuf(struct uio *uio, int how, int len, int align, int flags) 1718125296Ssilby{ 1719163915Sandre struct mbuf *m, *mb; 1720163915Sandre int error, length, total; 1721163915Sandre int progress = 0; 1722125296Ssilby 1723163915Sandre /* 1724163915Sandre * len can be zero or an arbitrary large value bound by 1725163915Sandre * the total data supplied by the uio. 1726163915Sandre */ 1727125296Ssilby if (len > 0) 1728125296Ssilby total = min(uio->uio_resid, len); 1729125296Ssilby else 1730125296Ssilby total = uio->uio_resid; 1731163915Sandre 1732163915Sandre /* 1733163915Sandre * The smallest unit returned by m_getm2() is a single mbuf 1734163915Sandre * with pkthdr. We can't align past it. Align align itself. 1735163915Sandre */ 1736163915Sandre if (align) 1737163915Sandre align &= ~(sizeof(long) - 1); 1738145883Semax if (align >= MHLEN) 1739163915Sandre return (NULL); 1740163915Sandre 1741166171Sandre /* 1742166171Sandre * Give us the full allocation or nothing. 1743166171Sandre * If len is zero return the smallest empty mbuf. 1744166171Sandre */ 1745166171Sandre m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); 1746163915Sandre if (m == NULL) 1747163915Sandre return (NULL); 1748163915Sandre m->m_data += align; 1749163915Sandre 1750163915Sandre /* Fill all mbufs with uio data and update header information. */ 1751163915Sandre for (mb = m; mb != NULL; mb = mb->m_next) { 1752163915Sandre length = min(M_TRAILINGSPACE(mb), total - progress); 1753163915Sandre 1754163915Sandre error = uiomove(mtod(mb, void *), length, uio); 1755163915Sandre if (error) { 1756163915Sandre m_freem(m); 1757163915Sandre return (NULL); 1758125296Ssilby } 1759163915Sandre 1760163915Sandre mb->m_len = length; 1761125296Ssilby progress += length; 1762163915Sandre if (flags & M_PKTHDR) 1763163915Sandre m->m_pkthdr.len += length; 1764125296Ssilby } 1765163915Sandre KASSERT(progress == total, ("%s: progress != total", __func__)); 1766163915Sandre 1767163915Sandre return (m); 1768125296Ssilby} 1769148552Ssam 1770148552Ssam/* 1771148552Ssam * Set the m_data pointer of a newly-allocated mbuf 1772148552Ssam * to place an object of the specified size at the 1773148552Ssam * end of the mbuf, longword aligned. 1774148552Ssam */ 1775148552Ssamvoid 1776148552Ssamm_align(struct mbuf *m, int len) 1777148552Ssam{ 1778148552Ssam int adjust; 1779148552Ssam 1780148552Ssam if (m->m_flags & M_EXT) 1781148552Ssam adjust = m->m_ext.ext_size - len; 1782148552Ssam else if (m->m_flags & M_PKTHDR) 1783148552Ssam adjust = MHLEN - len; 1784148552Ssam else 1785148552Ssam adjust = MLEN - len; 1786148552Ssam m->m_data += adjust &~ (sizeof(long)-1); 1787148552Ssam} 1788156756Ssam 1789156756Ssam/* 1790156756Ssam * Create a writable copy of the mbuf chain. While doing this 1791156756Ssam * we compact the chain with a goal of producing a chain with 1792156756Ssam * at most two mbufs. The second mbuf in this chain is likely 1793156756Ssam * to be a cluster. The primary purpose of this work is to create 1794156756Ssam * a writable packet for encryption, compression, etc. The 1795156756Ssam * secondary goal is to linearize the data so the data can be 1796156756Ssam * passed to crypto hardware in the most efficient manner possible. 1797156756Ssam */ 1798156756Ssamstruct mbuf * 1799156756Ssamm_unshare(struct mbuf *m0, int how) 1800156756Ssam{ 1801156756Ssam struct mbuf *m, *mprev; 1802156756Ssam struct mbuf *n, *mfirst, *mlast; 1803156756Ssam int len, off; 1804156756Ssam 1805156756Ssam mprev = NULL; 1806156756Ssam for (m = m0; m != NULL; m = mprev->m_next) { 1807156756Ssam /* 1808156756Ssam * Regular mbufs are ignored unless there's a cluster 1809156756Ssam * in front of it that we can use to coalesce. We do 1810156756Ssam * the latter mainly so later clusters can be coalesced 1811156756Ssam * also w/o having to handle them specially (i.e. convert 1812156756Ssam * mbuf+cluster -> cluster). This optimization is heavily 1813156756Ssam * influenced by the assumption that we're running over 1814156756Ssam * Ethernet where MCLBYTES is large enough that the max 1815156756Ssam * packet size will permit lots of coalescing into a 1816156756Ssam * single cluster. This in turn permits efficient 1817156756Ssam * crypto operations, especially when using hardware. 1818156756Ssam */ 1819156756Ssam if ((m->m_flags & M_EXT) == 0) { 1820156756Ssam if (mprev && (mprev->m_flags & M_EXT) && 1821156756Ssam m->m_len <= M_TRAILINGSPACE(mprev)) { 1822156756Ssam /* XXX: this ignores mbuf types */ 1823156756Ssam memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1824156756Ssam mtod(m, caddr_t), m->m_len); 1825156756Ssam mprev->m_len += m->m_len; 1826156756Ssam mprev->m_next = m->m_next; /* unlink from chain */ 1827156756Ssam m_free(m); /* reclaim mbuf */ 1828156756Ssam#if 0 1829156756Ssam newipsecstat.ips_mbcoalesced++; 1830156756Ssam#endif 1831156756Ssam } else { 1832156756Ssam mprev = m; 1833156756Ssam } 1834156756Ssam continue; 1835156756Ssam } 1836156756Ssam /* 1837156756Ssam * Writable mbufs are left alone (for now). 1838156756Ssam */ 1839156756Ssam if (M_WRITABLE(m)) { 1840156756Ssam mprev = m; 1841156756Ssam continue; 1842156756Ssam } 1843156756Ssam 1844156756Ssam /* 1845156756Ssam * Not writable, replace with a copy or coalesce with 1846156756Ssam * the previous mbuf if possible (since we have to copy 1847156756Ssam * it anyway, we try to reduce the number of mbufs and 1848156756Ssam * clusters so that future work is easier). 1849156756Ssam */ 1850156756Ssam KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 1851156756Ssam /* NB: we only coalesce into a cluster or larger */ 1852156756Ssam if (mprev != NULL && (mprev->m_flags & M_EXT) && 1853156756Ssam m->m_len <= M_TRAILINGSPACE(mprev)) { 1854156756Ssam /* XXX: this ignores mbuf types */ 1855156756Ssam memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1856156756Ssam mtod(m, caddr_t), m->m_len); 1857156756Ssam mprev->m_len += m->m_len; 1858156756Ssam mprev->m_next = m->m_next; /* unlink from chain */ 1859156756Ssam m_free(m); /* reclaim mbuf */ 1860156756Ssam#if 0 1861156756Ssam newipsecstat.ips_clcoalesced++; 1862156756Ssam#endif 1863156756Ssam continue; 1864156756Ssam } 1865156756Ssam 1866156756Ssam /* 1867156756Ssam * Allocate new space to hold the copy... 1868156756Ssam */ 1869156756Ssam /* XXX why can M_PKTHDR be set past the first mbuf? */ 1870156756Ssam if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 1871156756Ssam /* 1872156756Ssam * NB: if a packet header is present we must 1873156756Ssam * allocate the mbuf separately from any cluster 1874156756Ssam * because M_MOVE_PKTHDR will smash the data 1875156756Ssam * pointer and drop the M_EXT marker. 1876156756Ssam */ 1877156756Ssam MGETHDR(n, how, m->m_type); 1878156756Ssam if (n == NULL) { 1879156756Ssam m_freem(m0); 1880156756Ssam return (NULL); 1881156756Ssam } 1882156756Ssam M_MOVE_PKTHDR(n, m); 1883156756Ssam MCLGET(n, how); 1884156756Ssam if ((n->m_flags & M_EXT) == 0) { 1885156756Ssam m_free(n); 1886156756Ssam m_freem(m0); 1887156756Ssam return (NULL); 1888156756Ssam } 1889156756Ssam } else { 1890156756Ssam n = m_getcl(how, m->m_type, m->m_flags); 1891156756Ssam if (n == NULL) { 1892156756Ssam m_freem(m0); 1893156756Ssam return (NULL); 1894156756Ssam } 1895156756Ssam } 1896156756Ssam /* 1897156756Ssam * ... and copy the data. We deal with jumbo mbufs 1898156756Ssam * (i.e. m_len > MCLBYTES) by splitting them into 1899156756Ssam * clusters. We could just malloc a buffer and make 1900156756Ssam * it external but too many device drivers don't know 1901156756Ssam * how to break up the non-contiguous memory when 1902156756Ssam * doing DMA. 1903156756Ssam */ 1904156756Ssam len = m->m_len; 1905156756Ssam off = 0; 1906156756Ssam mfirst = n; 1907156756Ssam mlast = NULL; 1908156756Ssam for (;;) { 1909156756Ssam int cc = min(len, MCLBYTES); 1910156756Ssam memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 1911156756Ssam n->m_len = cc; 1912156756Ssam if (mlast != NULL) 1913156756Ssam mlast->m_next = n; 1914156756Ssam mlast = n; 1915156756Ssam#if 0 1916156756Ssam newipsecstat.ips_clcopied++; 1917156756Ssam#endif 1918156756Ssam 1919156756Ssam len -= cc; 1920156756Ssam if (len <= 0) 1921156756Ssam break; 1922156756Ssam off += cc; 1923156756Ssam 1924156756Ssam n = m_getcl(how, m->m_type, m->m_flags); 1925156756Ssam if (n == NULL) { 1926156756Ssam m_freem(mfirst); 1927156756Ssam m_freem(m0); 1928156756Ssam return (NULL); 1929156756Ssam } 1930156756Ssam } 1931156756Ssam n->m_next = m->m_next; 1932156756Ssam if (mprev == NULL) 1933156756Ssam m0 = mfirst; /* new head of chain */ 1934156756Ssam else 1935156756Ssam mprev->m_next = mfirst; /* replace old mbuf */ 1936156756Ssam m_free(m); /* release old mbuf */ 1937156756Ssam mprev = mfirst; 1938156756Ssam } 1939156756Ssam return (m0); 1940156756Ssam} 1941