uipc_mbuf.c revision 169624
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 169624 2007-05-16 20:41:08Z rwatson $"); 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, 189129906Sbmilekic void (*freef)(void *, void *), void *args, 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; 202129906Sbmilekic mb->m_ext.ext_args = args; 203129906Sbmilekic mb->m_ext.ext_type = type; 204129906Sbmilekic } 205129906Sbmilekic} 206129906Sbmilekic 207129906Sbmilekic/* 208129906Sbmilekic * Non-directly-exported function to clean up after mbufs with M_EXT 209151976Sandre * storage attached to them if the reference count hits 1. 210129906Sbmilekic */ 211129906Sbmilekicvoid 212129906Sbmilekicmb_free_ext(struct mbuf *m) 213129906Sbmilekic{ 214151976Sandre KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 215151976Sandre KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 216129906Sbmilekic 217151976Sandre /* Free attached storage if this mbuf is the only reference to it. */ 218151976Sandre if (*(m->m_ext.ref_cnt) == 1 || 219162515Srrs atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) { 220151976Sandre switch (m->m_ext.ext_type) { 221152101Sandre case EXT_PACKET: /* The packet zone is special. */ 222152035Sandre if (*(m->m_ext.ref_cnt) == 0) 223152035Sandre *(m->m_ext.ref_cnt) = 1; 224151976Sandre uma_zfree(zone_pack, m); 225151976Sandre return; /* Job done. */ 226152101Sandre case EXT_CLUSTER: 227152101Sandre uma_zfree(zone_clust, m->m_ext.ext_buf); 228130289Sbmilekic break; 229155780Sandre case EXT_JUMBOP: 230155780Sandre uma_zfree(zone_jumbop, m->m_ext.ext_buf); 231153232Sandre break; 232151976Sandre case EXT_JUMBO9: 233151976Sandre uma_zfree(zone_jumbo9, m->m_ext.ext_buf); 234151976Sandre break; 235151976Sandre case EXT_JUMBO16: 236151976Sandre uma_zfree(zone_jumbo16, m->m_ext.ext_buf); 237151976Sandre break; 238151976Sandre case EXT_SFBUF: 239151976Sandre case EXT_NET_DRV: 240151976Sandre case EXT_MOD_TYPE: 241151976Sandre case EXT_DISPOSABLE: 242151976Sandre *(m->m_ext.ref_cnt) = 0; 243151976Sandre uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *, 244151976Sandre m->m_ext.ref_cnt)); 245151976Sandre /* FALLTHROUGH */ 246151976Sandre case EXT_EXTREF: 247151976Sandre KASSERT(m->m_ext.ext_free != NULL, 248151976Sandre ("%s: ext_free not set", __func__)); 249141668Sbmilekic (*(m->m_ext.ext_free))(m->m_ext.ext_buf, 250141668Sbmilekic m->m_ext.ext_args); 251151976Sandre break; 252151976Sandre default: 253151976Sandre KASSERT(m->m_ext.ext_type == 0, 254151976Sandre ("%s: unknown ext_type", __func__)); 255141668Sbmilekic } 256141668Sbmilekic } 257151976Sandre /* 258151976Sandre * Free this mbuf back to the mbuf zone with all m_ext 259151976Sandre * information purged. 260151976Sandre */ 261151976Sandre m->m_ext.ext_buf = NULL; 262151976Sandre m->m_ext.ext_free = NULL; 263151976Sandre m->m_ext.ext_args = NULL; 264151976Sandre m->m_ext.ref_cnt = NULL; 265151976Sandre m->m_ext.ext_size = 0; 266151976Sandre m->m_ext.ext_type = 0; 267151976Sandre m->m_flags &= ~M_EXT; 268130357Sbmilekic uma_zfree(zone_mbuf, m); 269129906Sbmilekic} 270129906Sbmilekic 271129906Sbmilekic/* 272151976Sandre * Attach the the cluster from *m to *n, set up m_ext in *n 273151976Sandre * and bump the refcount of the cluster. 274151976Sandre */ 275151976Sandrestatic void 276151976Sandremb_dupcl(struct mbuf *n, struct mbuf *m) 277151976Sandre{ 278151976Sandre KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__)); 279151976Sandre KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__)); 280151976Sandre KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__)); 281151976Sandre 282151976Sandre if (*(m->m_ext.ref_cnt) == 1) 283151976Sandre *(m->m_ext.ref_cnt) += 1; 284151976Sandre else 285151976Sandre atomic_add_int(m->m_ext.ref_cnt, 1); 286151976Sandre n->m_ext.ext_buf = m->m_ext.ext_buf; 287151976Sandre n->m_ext.ext_free = m->m_ext.ext_free; 288151976Sandre n->m_ext.ext_args = m->m_ext.ext_args; 289151976Sandre n->m_ext.ext_size = m->m_ext.ext_size; 290151976Sandre n->m_ext.ref_cnt = m->m_ext.ref_cnt; 291151976Sandre n->m_ext.ext_type = m->m_ext.ext_type; 292151976Sandre n->m_flags |= M_EXT; 293151976Sandre} 294151976Sandre 295151976Sandre/* 296149598Sandre * Clean up mbuf (chain) from any tags and packet headers. 297149647Sandre * If "all" is set then the first mbuf in the chain will be 298149647Sandre * cleaned too. 299149598Sandre */ 300149598Sandrevoid 301149647Sandrem_demote(struct mbuf *m0, int all) 302149598Sandre{ 303149598Sandre struct mbuf *m; 304149598Sandre 305149598Sandre for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) { 306149598Sandre if (m->m_flags & M_PKTHDR) { 307149598Sandre m_tag_delete_chain(m, NULL); 308149598Sandre m->m_flags &= ~M_PKTHDR; 309149598Sandre bzero(&m->m_pkthdr, sizeof(struct pkthdr)); 310149598Sandre } 311149643Sandre if (m->m_type == MT_HEADER) 312149598Sandre m->m_type = MT_DATA; 313149643Sandre if (m != m0 && m->m_nextpkt != NULL) 314149598Sandre m->m_nextpkt = NULL; 315149598Sandre m->m_flags = m->m_flags & (M_EXT|M_EOR|M_RDONLY|M_FREELIST); 316149598Sandre } 317149598Sandre} 318149598Sandre 319149598Sandre/* 320149648Sandre * Sanity checks on mbuf (chain) for use in KASSERT() and general 321149648Sandre * debugging. 322149648Sandre * Returns 0 or panics when bad and 1 on all tests passed. 323149648Sandre * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they 324149648Sandre * blow up later. 325149599Sandre */ 326149599Sandreint 327149599Sandrem_sanity(struct mbuf *m0, int sanitize) 328149599Sandre{ 329149599Sandre struct mbuf *m; 330149599Sandre caddr_t a, b; 331149599Sandre int pktlen = 0; 332149599Sandre 333168734Skmacy#ifdef INVARIANTS 334168734Skmacy#define M_SANITY_ACTION(s) panic("mbuf %p: " s, m) 335168734Skmacy#else 336168734Skmacy#define M_SANITY_ACTION(s) printf("mbuf %p: " s, m) 337168734Skmacy#endif 338149599Sandre 339149648Sandre for (m = m0; m != NULL; m = m->m_next) { 340149599Sandre /* 341149599Sandre * Basic pointer checks. If any of these fails then some 342149599Sandre * unrelated kernel memory before or after us is trashed. 343149599Sandre * No way to recover from that. 344149599Sandre */ 345149648Sandre a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf : 346149648Sandre ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) : 347149599Sandre (caddr_t)(&m->m_dat)) ); 348149599Sandre b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size : 349149648Sandre ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN))); 350149599Sandre if ((caddr_t)m->m_data < a) 351149599Sandre M_SANITY_ACTION("m_data outside mbuf data range left"); 352149599Sandre if ((caddr_t)m->m_data > b) 353149599Sandre M_SANITY_ACTION("m_data outside mbuf data range right"); 354149599Sandre if ((caddr_t)m->m_data + m->m_len > b) 355149599Sandre M_SANITY_ACTION("m_data + m_len exeeds mbuf space"); 356149648Sandre if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) { 357149599Sandre if ((caddr_t)m->m_pkthdr.header < a || 358149599Sandre (caddr_t)m->m_pkthdr.header > b) 359149599Sandre M_SANITY_ACTION("m_pkthdr.header outside mbuf data range"); 360149599Sandre } 361149599Sandre 362149599Sandre /* m->m_nextpkt may only be set on first mbuf in chain. */ 363149648Sandre if (m != m0 && m->m_nextpkt != NULL) { 364149599Sandre if (sanitize) { 365149599Sandre m_freem(m->m_nextpkt); 366149599Sandre m->m_nextpkt = (struct mbuf *)0xDEADC0DE; 367149599Sandre } else 368149599Sandre M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf"); 369149599Sandre } 370149599Sandre 371149599Sandre /* packet length (not mbuf length!) calculation */ 372149599Sandre if (m0->m_flags & M_PKTHDR) 373149599Sandre pktlen += m->m_len; 374149599Sandre 375149599Sandre /* m_tags may only be attached to first mbuf in chain. */ 376149599Sandre if (m != m0 && m->m_flags & M_PKTHDR && 377149599Sandre !SLIST_EMPTY(&m->m_pkthdr.tags)) { 378149599Sandre if (sanitize) { 379149599Sandre m_tag_delete_chain(m, NULL); 380149599Sandre /* put in 0xDEADC0DE perhaps? */ 381149648Sandre } else 382149599Sandre M_SANITY_ACTION("m_tags on in-chain mbuf"); 383149599Sandre } 384149599Sandre 385149599Sandre /* M_PKTHDR may only be set on first mbuf in chain */ 386149599Sandre if (m != m0 && m->m_flags & M_PKTHDR) { 387149599Sandre if (sanitize) { 388149599Sandre bzero(&m->m_pkthdr, sizeof(m->m_pkthdr)); 389149599Sandre m->m_flags &= ~M_PKTHDR; 390149599Sandre /* put in 0xDEADCODE and leave hdr flag in */ 391149599Sandre } else 392149599Sandre M_SANITY_ACTION("M_PKTHDR on in-chain mbuf"); 393149599Sandre } 394149599Sandre } 395149648Sandre m = m0; 396149648Sandre if (pktlen && pktlen != m->m_pkthdr.len) { 397149599Sandre if (sanitize) 398149648Sandre m->m_pkthdr.len = 0; 399149599Sandre else 400149599Sandre M_SANITY_ACTION("m_pkthdr.len != mbuf chain length"); 401149599Sandre } 402149648Sandre return 1; 403149648Sandre 404149599Sandre#undef M_SANITY_ACTION 405149599Sandre} 406149599Sandre 407149599Sandre 408149599Sandre/* 409108466Ssam * "Move" mbuf pkthdr from "from" to "to". 410100960Srwatson * "from" must have M_PKTHDR set, and "to" must be empty. 411100960Srwatson */ 412100960Srwatsonvoid 413108466Ssamm_move_pkthdr(struct mbuf *to, struct mbuf *from) 414100960Srwatson{ 415100960Srwatson 416100960Srwatson#if 0 417108466Ssam /* see below for why these are not enabled */ 418113255Sdes M_ASSERTPKTHDR(to); 419113487Srwatson /* Note: with MAC, this may not be a good assertion. */ 420108466Ssam KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), 421108466Ssam ("m_move_pkthdr: to has tags")); 422100960Srwatson#endif 423101007Srwatson#ifdef MAC 424113487Srwatson /* 425113487Srwatson * XXXMAC: It could be this should also occur for non-MAC? 426113487Srwatson */ 427101007Srwatson if (to->m_flags & M_PKTHDR) 428113487Srwatson m_tag_delete_chain(to, NULL); 429101007Srwatson#endif 430143302Ssam to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 431143302Ssam if ((to->m_flags & M_EXT) == 0) 432143302Ssam to->m_data = to->m_pktdat; 433108466Ssam to->m_pkthdr = from->m_pkthdr; /* especially tags */ 434108466Ssam SLIST_INIT(&from->m_pkthdr.tags); /* purge tags from src */ 435108466Ssam from->m_flags &= ~M_PKTHDR; 436108466Ssam} 437108466Ssam 438108466Ssam/* 439108466Ssam * Duplicate "from"'s mbuf pkthdr in "to". 440108466Ssam * "from" must have M_PKTHDR set, and "to" must be empty. 441108466Ssam * In particular, this does a deep copy of the packet tags. 442108466Ssam */ 443108466Ssamint 444108466Ssamm_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how) 445108466Ssam{ 446108466Ssam 447108466Ssam#if 0 448108466Ssam /* 449108466Ssam * The mbuf allocator only initializes the pkthdr 450108466Ssam * when the mbuf is allocated with MGETHDR. Many users 451108466Ssam * (e.g. m_copy*, m_prepend) use MGET and then 452108466Ssam * smash the pkthdr as needed causing these 453108466Ssam * assertions to trip. For now just disable them. 454108466Ssam */ 455113255Sdes M_ASSERTPKTHDR(to); 456113487Srwatson /* Note: with MAC, this may not be a good assertion. */ 457108466Ssam KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags")); 458108466Ssam#endif 459132488Salfred MBUF_CHECKSLEEP(how); 460108466Ssam#ifdef MAC 461108466Ssam if (to->m_flags & M_PKTHDR) 462113487Srwatson m_tag_delete_chain(to, NULL); 463108466Ssam#endif 464112733Ssilby to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT); 465112733Ssilby if ((to->m_flags & M_EXT) == 0) 466112733Ssilby to->m_data = to->m_pktdat; 467100960Srwatson to->m_pkthdr = from->m_pkthdr; 468108466Ssam SLIST_INIT(&to->m_pkthdr.tags); 469113480Srwatson return (m_tag_copy_chain(to, from, MBTOM(how))); 470100960Srwatson} 471100960Srwatson 472100960Srwatson/* 4731541Srgrimes * Lesser-used path for M_PREPEND: 4741541Srgrimes * allocate new mbuf to prepend to chain, 4751541Srgrimes * copy junk along. 4761541Srgrimes */ 4771541Srgrimesstruct mbuf * 47872356Sbmilekicm_prepend(struct mbuf *m, int len, int how) 4791541Srgrimes{ 4801541Srgrimes struct mbuf *mn; 4811541Srgrimes 482117770Ssilby if (m->m_flags & M_PKTHDR) 483117770Ssilby MGETHDR(mn, how, m->m_type); 484117770Ssilby else 485117770Ssilby MGET(mn, how, m->m_type); 48672356Sbmilekic if (mn == NULL) { 4871541Srgrimes m_freem(m); 48872356Sbmilekic return (NULL); 4891541Srgrimes } 490113487Srwatson if (m->m_flags & M_PKTHDR) 491108466Ssam M_MOVE_PKTHDR(mn, m); 4921541Srgrimes mn->m_next = m; 4931541Srgrimes m = mn; 494165447Srrs if(m->m_flags & M_PKTHDR) { 495165447Srrs if (len < MHLEN) 496165447Srrs MH_ALIGN(m, len); 497165447Srrs } else { 498165447Srrs if (len < MLEN) 499165447Srrs M_ALIGN(m, len); 500165447Srrs } 5011541Srgrimes m->m_len = len; 5021541Srgrimes return (m); 5031541Srgrimes} 5041541Srgrimes 5051541Srgrimes/* 5061541Srgrimes * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 5071541Srgrimes * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 508111119Simp * The wait parameter is a choice of M_TRYWAIT/M_DONTWAIT from caller. 50954002Sarchie * Note that the copy is read-only, because clusters are not copied, 51054002Sarchie * only their reference counts are incremented. 5111541Srgrimes */ 5121541Srgrimesstruct mbuf * 51372356Sbmilekicm_copym(struct mbuf *m, int off0, int len, int wait) 5141541Srgrimes{ 51572356Sbmilekic struct mbuf *n, **np; 51672356Sbmilekic int off = off0; 5171541Srgrimes struct mbuf *top; 5181541Srgrimes int copyhdr = 0; 5191541Srgrimes 52052201Salfred KASSERT(off >= 0, ("m_copym, negative off %d", off)); 52152201Salfred KASSERT(len >= 0, ("m_copym, negative len %d", len)); 522132488Salfred MBUF_CHECKSLEEP(wait); 5231541Srgrimes if (off == 0 && m->m_flags & M_PKTHDR) 5241541Srgrimes copyhdr = 1; 5251541Srgrimes while (off > 0) { 52652201Salfred KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 5271541Srgrimes if (off < m->m_len) 5281541Srgrimes break; 5291541Srgrimes off -= m->m_len; 5301541Srgrimes m = m->m_next; 5311541Srgrimes } 5321541Srgrimes np = ⊤ 5331541Srgrimes top = 0; 5341541Srgrimes while (len > 0) { 53572356Sbmilekic if (m == NULL) { 53652201Salfred KASSERT(len == M_COPYALL, 53752201Salfred ("m_copym, length > size of mbuf chain")); 5381541Srgrimes break; 5391541Srgrimes } 540117770Ssilby if (copyhdr) 541117770Ssilby MGETHDR(n, wait, m->m_type); 542117770Ssilby else 543117770Ssilby MGET(n, wait, m->m_type); 5441541Srgrimes *np = n; 54572356Sbmilekic if (n == NULL) 5461541Srgrimes goto nospace; 5471541Srgrimes if (copyhdr) { 548108466Ssam if (!m_dup_pkthdr(n, m, wait)) 549108466Ssam goto nospace; 5501541Srgrimes if (len == M_COPYALL) 5511541Srgrimes n->m_pkthdr.len -= off0; 5521541Srgrimes else 5531541Srgrimes n->m_pkthdr.len = len; 5541541Srgrimes copyhdr = 0; 5551541Srgrimes } 5561541Srgrimes n->m_len = min(len, m->m_len - off); 5571541Srgrimes if (m->m_flags & M_EXT) { 5581541Srgrimes n->m_data = m->m_data + off; 559151976Sandre mb_dupcl(n, m); 5601541Srgrimes } else 5611541Srgrimes bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 562103569Sbmilekic (u_int)n->m_len); 5631541Srgrimes if (len != M_COPYALL) 5641541Srgrimes len -= n->m_len; 5651541Srgrimes off = 0; 5661541Srgrimes m = m->m_next; 5671541Srgrimes np = &n->m_next; 5681541Srgrimes } 56978592Sbmilekic if (top == NULL) 57078592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 57178592Sbmilekic 5721541Srgrimes return (top); 5731541Srgrimesnospace: 5741541Srgrimes m_freem(top); 57578592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 57672356Sbmilekic return (NULL); 5771541Srgrimes} 5781541Srgrimes 5791541Srgrimes/* 580149602Sandre * Returns mbuf chain with new head for the prepending case. 581149602Sandre * Copies from mbuf (chain) n from off for len to mbuf (chain) m 582149602Sandre * either prepending or appending the data. 583149602Sandre * The resulting mbuf (chain) m is fully writeable. 584149602Sandre * m is destination (is made writeable) 585149602Sandre * n is source, off is offset in source, len is len from offset 586149602Sandre * dir, 0 append, 1 prepend 587149602Sandre * how, wait or nowait 588149602Sandre */ 589149602Sandre 590149602Sandrestatic int 591149602Sandrem_bcopyxxx(void *s, void *t, u_int len) 592149602Sandre{ 593149602Sandre bcopy(s, t, (size_t)len); 594149602Sandre return 0; 595149602Sandre} 596149602Sandre 597149602Sandrestruct mbuf * 598149602Sandrem_copymdata(struct mbuf *m, struct mbuf *n, int off, int len, 599149602Sandre int prep, int how) 600149602Sandre{ 601151976Sandre struct mbuf *mm, *x, *z, *prev = NULL; 602149602Sandre caddr_t p; 603151976Sandre int i, nlen = 0; 604149602Sandre caddr_t buf[MLEN]; 605149602Sandre 606149602Sandre KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source")); 607149602Sandre KASSERT(off >= 0, ("m_copymdata, negative off %d", off)); 608149602Sandre KASSERT(len >= 0, ("m_copymdata, negative len %d", len)); 609149602Sandre KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep)); 610149602Sandre 611151976Sandre mm = m; 612151976Sandre if (!prep) { 613151976Sandre while(mm->m_next) { 614151976Sandre prev = mm; 615151976Sandre mm = mm->m_next; 616149602Sandre } 617149602Sandre } 618149602Sandre for (z = n; z != NULL; z = z->m_next) 619149602Sandre nlen += z->m_len; 620149602Sandre if (len == M_COPYALL) 621149602Sandre len = nlen - off; 622149602Sandre if (off + len > nlen || len < 1) 623149602Sandre return NULL; 624149602Sandre 625151976Sandre if (!M_WRITABLE(mm)) { 626151976Sandre /* XXX: Use proper m_xxx function instead. */ 627151976Sandre x = m_getcl(how, MT_DATA, mm->m_flags); 628151976Sandre if (x == NULL) 629151976Sandre return NULL; 630151976Sandre bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size); 631151976Sandre p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf); 632151976Sandre x->m_data = p; 633151976Sandre mm->m_next = NULL; 634151976Sandre if (mm != m) 635151976Sandre prev->m_next = x; 636151976Sandre m_free(mm); 637151976Sandre mm = x; 638151976Sandre } 639151976Sandre 640149602Sandre /* 641149602Sandre * Append/prepend the data. Allocating mbufs as necessary. 642149602Sandre */ 643149602Sandre /* Shortcut if enough free space in first/last mbuf. */ 644149602Sandre if (!prep && M_TRAILINGSPACE(mm) >= len) { 645149602Sandre m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) + 646149602Sandre mm->m_len); 647149602Sandre mm->m_len += len; 648149602Sandre mm->m_pkthdr.len += len; 649149602Sandre return m; 650149602Sandre } 651149602Sandre if (prep && M_LEADINGSPACE(mm) >= len) { 652149602Sandre mm->m_data = mtod(mm, caddr_t) - len; 653149602Sandre m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t)); 654149602Sandre mm->m_len += len; 655149602Sandre mm->m_pkthdr.len += len; 656149602Sandre return mm; 657149602Sandre } 658149602Sandre 659149602Sandre /* Expand first/last mbuf to cluster if possible. */ 660149602Sandre if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) { 661149602Sandre bcopy(mm->m_data, &buf, mm->m_len); 662149602Sandre m_clget(mm, how); 663149602Sandre if (!(mm->m_flags & M_EXT)) 664149602Sandre return NULL; 665149602Sandre bcopy(&buf, mm->m_ext.ext_buf, mm->m_len); 666149602Sandre mm->m_data = mm->m_ext.ext_buf; 667149602Sandre mm->m_pkthdr.header = NULL; 668149602Sandre } 669149602Sandre if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) { 670149602Sandre bcopy(mm->m_data, &buf, mm->m_len); 671149602Sandre m_clget(mm, how); 672149602Sandre if (!(mm->m_flags & M_EXT)) 673149602Sandre return NULL; 674149602Sandre bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf + 675149602Sandre mm->m_ext.ext_size - mm->m_len, mm->m_len); 676149602Sandre mm->m_data = (caddr_t)mm->m_ext.ext_buf + 677149602Sandre mm->m_ext.ext_size - mm->m_len; 678149602Sandre mm->m_pkthdr.header = NULL; 679149602Sandre } 680149602Sandre 681149602Sandre /* Append/prepend as many mbuf (clusters) as necessary to fit len. */ 682149602Sandre if (!prep && len > M_TRAILINGSPACE(mm)) { 683149602Sandre if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA)) 684149602Sandre return NULL; 685149602Sandre } 686149602Sandre if (prep && len > M_LEADINGSPACE(mm)) { 687149602Sandre if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA))) 688149602Sandre return NULL; 689149602Sandre i = 0; 690149602Sandre for (x = z; x != NULL; x = x->m_next) { 691149602Sandre i += x->m_flags & M_EXT ? x->m_ext.ext_size : 692149602Sandre (x->m_flags & M_PKTHDR ? MHLEN : MLEN); 693149602Sandre if (!x->m_next) 694149602Sandre break; 695149602Sandre } 696149602Sandre z->m_data += i - len; 697149602Sandre m_move_pkthdr(mm, z); 698149602Sandre x->m_next = mm; 699149602Sandre mm = z; 700149602Sandre } 701149602Sandre 702149602Sandre /* Seek to start position in source mbuf. Optimization for long chains. */ 703149602Sandre while (off > 0) { 704149602Sandre if (off < n->m_len) 705149602Sandre break; 706149602Sandre off -= n->m_len; 707149602Sandre n = n->m_next; 708149602Sandre } 709149602Sandre 710149602Sandre /* Copy data into target mbuf. */ 711149602Sandre z = mm; 712149602Sandre while (len > 0) { 713149602Sandre KASSERT(z != NULL, ("m_copymdata, falling off target edge")); 714149602Sandre i = M_TRAILINGSPACE(z); 715149602Sandre m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len); 716149602Sandre z->m_len += i; 717149602Sandre /* fixup pkthdr.len if necessary */ 718149602Sandre if ((prep ? mm : m)->m_flags & M_PKTHDR) 719149602Sandre (prep ? mm : m)->m_pkthdr.len += i; 720149602Sandre off += i; 721149602Sandre len -= i; 722149602Sandre z = z->m_next; 723149602Sandre } 724149602Sandre return (prep ? mm : m); 725149602Sandre} 726149602Sandre 727149602Sandre/* 72815689Swollman * Copy an entire packet, including header (which must be present). 72915689Swollman * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 73054002Sarchie * Note that the copy is read-only, because clusters are not copied, 73154002Sarchie * only their reference counts are incremented. 73272750Sluigi * Preserve alignment of the first mbuf so if the creator has left 73372750Sluigi * some room at the beginning (e.g. for inserting protocol headers) 73472750Sluigi * the copies still have the room available. 73515689Swollman */ 73615689Swollmanstruct mbuf * 73772356Sbmilekicm_copypacket(struct mbuf *m, int how) 73815689Swollman{ 73915689Swollman struct mbuf *top, *n, *o; 74015689Swollman 741132488Salfred MBUF_CHECKSLEEP(how); 74215689Swollman MGET(n, how, m->m_type); 74315689Swollman top = n; 74472356Sbmilekic if (n == NULL) 74515689Swollman goto nospace; 74615689Swollman 747108466Ssam if (!m_dup_pkthdr(n, m, how)) 748108466Ssam goto nospace; 74915689Swollman n->m_len = m->m_len; 75015689Swollman if (m->m_flags & M_EXT) { 75115689Swollman n->m_data = m->m_data; 752151976Sandre mb_dupcl(n, m); 75315689Swollman } else { 75472750Sluigi n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat ); 75515689Swollman bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 75615689Swollman } 75715689Swollman 75815689Swollman m = m->m_next; 75915689Swollman while (m) { 76015689Swollman MGET(o, how, m->m_type); 76172356Sbmilekic if (o == NULL) 76215689Swollman goto nospace; 76315689Swollman 76415689Swollman n->m_next = o; 76515689Swollman n = n->m_next; 76615689Swollman 76715689Swollman n->m_len = m->m_len; 76815689Swollman if (m->m_flags & M_EXT) { 76915689Swollman n->m_data = m->m_data; 770151976Sandre mb_dupcl(n, m); 77115689Swollman } else { 77215689Swollman bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 77315689Swollman } 77415689Swollman 77515689Swollman m = m->m_next; 77615689Swollman } 77715689Swollman return top; 77815689Swollmannospace: 77915689Swollman m_freem(top); 78078592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 78172356Sbmilekic return (NULL); 78215689Swollman} 78315689Swollman 78415689Swollman/* 7851541Srgrimes * Copy data from an mbuf chain starting "off" bytes from the beginning, 7861541Srgrimes * continuing for "len" bytes, into the indicated buffer. 7871541Srgrimes */ 7881549Srgrimesvoid 78981907Sjulianm_copydata(const struct mbuf *m, int off, int len, caddr_t cp) 7901541Srgrimes{ 791103569Sbmilekic u_int count; 7921541Srgrimes 79352201Salfred KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 79452201Salfred KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 7951541Srgrimes while (off > 0) { 79652201Salfred KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 7971541Srgrimes if (off < m->m_len) 7981541Srgrimes break; 7991541Srgrimes off -= m->m_len; 8001541Srgrimes m = m->m_next; 8011541Srgrimes } 8021541Srgrimes while (len > 0) { 80352201Salfred KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 8041541Srgrimes count = min(m->m_len - off, len); 8051541Srgrimes bcopy(mtod(m, caddr_t) + off, cp, count); 8061541Srgrimes len -= count; 8071541Srgrimes cp += count; 8081541Srgrimes off = 0; 8091541Srgrimes m = m->m_next; 8101541Srgrimes } 8111541Srgrimes} 8121541Srgrimes 8131541Srgrimes/* 81454002Sarchie * Copy a packet header mbuf chain into a completely new chain, including 81554002Sarchie * copying any mbuf clusters. Use this instead of m_copypacket() when 81654002Sarchie * you need a writable copy of an mbuf chain. 81754002Sarchie */ 81854002Sarchiestruct mbuf * 81972356Sbmilekicm_dup(struct mbuf *m, int how) 82054002Sarchie{ 82154002Sarchie struct mbuf **p, *top = NULL; 82254002Sarchie int remain, moff, nsize; 82354002Sarchie 824132488Salfred MBUF_CHECKSLEEP(how); 82554002Sarchie /* Sanity check */ 82654002Sarchie if (m == NULL) 82772356Sbmilekic return (NULL); 828113255Sdes M_ASSERTPKTHDR(m); 82954002Sarchie 83054002Sarchie /* While there's more data, get a new mbuf, tack it on, and fill it */ 83154002Sarchie remain = m->m_pkthdr.len; 83254002Sarchie moff = 0; 83354002Sarchie p = ⊤ 83454002Sarchie while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 83554002Sarchie struct mbuf *n; 83654002Sarchie 83754002Sarchie /* Get the next new mbuf */ 838129906Sbmilekic if (remain >= MINCLSIZE) { 839129906Sbmilekic n = m_getcl(how, m->m_type, 0); 840129906Sbmilekic nsize = MCLBYTES; 841129906Sbmilekic } else { 842129906Sbmilekic n = m_get(how, m->m_type); 843129906Sbmilekic nsize = MLEN; 844129906Sbmilekic } 84554002Sarchie if (n == NULL) 84654002Sarchie goto nospace; 847129906Sbmilekic 848129906Sbmilekic if (top == NULL) { /* First one, must be PKTHDR */ 849129906Sbmilekic if (!m_dup_pkthdr(n, m, how)) { 850129906Sbmilekic m_free(n); 851108466Ssam goto nospace; 852129906Sbmilekic } 853153428Semaste if ((n->m_flags & M_EXT) == 0) 854153428Semaste nsize = MHLEN; 85554002Sarchie } 85654002Sarchie n->m_len = 0; 85754002Sarchie 85854002Sarchie /* Link it into the new chain */ 85954002Sarchie *p = n; 86054002Sarchie p = &n->m_next; 86154002Sarchie 86254002Sarchie /* Copy data from original mbuf(s) into new mbuf */ 86354002Sarchie while (n->m_len < nsize && m != NULL) { 86454002Sarchie int chunk = min(nsize - n->m_len, m->m_len - moff); 86554002Sarchie 86654002Sarchie bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 86754002Sarchie moff += chunk; 86854002Sarchie n->m_len += chunk; 86954002Sarchie remain -= chunk; 87054002Sarchie if (moff == m->m_len) { 87154002Sarchie m = m->m_next; 87254002Sarchie moff = 0; 87354002Sarchie } 87454002Sarchie } 87554002Sarchie 87654002Sarchie /* Check correct total mbuf length */ 87754002Sarchie KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 87887594Sobrien ("%s: bogus m_pkthdr.len", __func__)); 87954002Sarchie } 88054002Sarchie return (top); 88154002Sarchie 88254002Sarchienospace: 88354002Sarchie m_freem(top); 88478592Sbmilekic mbstat.m_mcfail++; /* XXX: No consistency. */ 88572356Sbmilekic return (NULL); 88654002Sarchie} 88754002Sarchie 88854002Sarchie/* 8891541Srgrimes * Concatenate mbuf chain n to m. 8901541Srgrimes * Both chains must be of the same type (e.g. MT_DATA). 8911541Srgrimes * Any m_pkthdr is not updated. 8921541Srgrimes */ 8931549Srgrimesvoid 89472356Sbmilekicm_cat(struct mbuf *m, struct mbuf *n) 8951541Srgrimes{ 8961541Srgrimes while (m->m_next) 8971541Srgrimes m = m->m_next; 8981541Srgrimes while (n) { 8991541Srgrimes if (m->m_flags & M_EXT || 9001541Srgrimes m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 9011541Srgrimes /* just join the two chains */ 9021541Srgrimes m->m_next = n; 9031541Srgrimes return; 9041541Srgrimes } 9051541Srgrimes /* splat the data from one into the other */ 9061541Srgrimes bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 9071541Srgrimes (u_int)n->m_len); 9081541Srgrimes m->m_len += n->m_len; 9091541Srgrimes n = m_free(n); 9101541Srgrimes } 9111541Srgrimes} 9121541Srgrimes 9131549Srgrimesvoid 91472356Sbmilekicm_adj(struct mbuf *mp, int req_len) 9151541Srgrimes{ 91672356Sbmilekic int len = req_len; 91772356Sbmilekic struct mbuf *m; 91872356Sbmilekic int count; 9191541Srgrimes 9201541Srgrimes if ((m = mp) == NULL) 9211541Srgrimes return; 9221541Srgrimes if (len >= 0) { 9231541Srgrimes /* 9241541Srgrimes * Trim from head. 9251541Srgrimes */ 9261541Srgrimes while (m != NULL && len > 0) { 9271541Srgrimes if (m->m_len <= len) { 9281541Srgrimes len -= m->m_len; 9291541Srgrimes m->m_len = 0; 9301541Srgrimes m = m->m_next; 9311541Srgrimes } else { 9321541Srgrimes m->m_len -= len; 9331541Srgrimes m->m_data += len; 9341541Srgrimes len = 0; 9351541Srgrimes } 9361541Srgrimes } 9371541Srgrimes m = mp; 9381541Srgrimes if (mp->m_flags & M_PKTHDR) 9391541Srgrimes m->m_pkthdr.len -= (req_len - len); 9401541Srgrimes } else { 9411541Srgrimes /* 9421541Srgrimes * Trim from tail. Scan the mbuf chain, 9431541Srgrimes * calculating its length and finding the last mbuf. 9441541Srgrimes * If the adjustment only affects this mbuf, then just 9451541Srgrimes * adjust and return. Otherwise, rescan and truncate 9461541Srgrimes * after the remaining size. 9471541Srgrimes */ 9481541Srgrimes len = -len; 9491541Srgrimes count = 0; 9501541Srgrimes for (;;) { 9511541Srgrimes count += m->m_len; 9521541Srgrimes if (m->m_next == (struct mbuf *)0) 9531541Srgrimes break; 9541541Srgrimes m = m->m_next; 9551541Srgrimes } 9561541Srgrimes if (m->m_len >= len) { 9571541Srgrimes m->m_len -= len; 9581541Srgrimes if (mp->m_flags & M_PKTHDR) 9591541Srgrimes mp->m_pkthdr.len -= len; 9601541Srgrimes return; 9611541Srgrimes } 9621541Srgrimes count -= len; 9631541Srgrimes if (count < 0) 9641541Srgrimes count = 0; 9651541Srgrimes /* 9661541Srgrimes * Correct length for chain is "count". 9671541Srgrimes * Find the mbuf with last data, adjust its length, 9681541Srgrimes * and toss data from remaining mbufs on chain. 9691541Srgrimes */ 9701541Srgrimes m = mp; 9711541Srgrimes if (m->m_flags & M_PKTHDR) 9721541Srgrimes m->m_pkthdr.len = count; 9731541Srgrimes for (; m; m = m->m_next) { 9741541Srgrimes if (m->m_len >= count) { 9751541Srgrimes m->m_len = count; 976142350Ssam if (m->m_next != NULL) { 977142350Ssam m_freem(m->m_next); 978142350Ssam m->m_next = NULL; 979142350Ssam } 9801541Srgrimes break; 9811541Srgrimes } 9821541Srgrimes count -= m->m_len; 9831541Srgrimes } 9841541Srgrimes } 9851541Srgrimes} 9861541Srgrimes 9871541Srgrimes/* 9881541Srgrimes * Rearange an mbuf chain so that len bytes are contiguous 9891541Srgrimes * and in the data area of an mbuf (so that mtod and dtom 9901541Srgrimes * will work for a structure of size len). Returns the resulting 9911541Srgrimes * mbuf chain on success, frees it and returns null on failure. 9921541Srgrimes * If there is room, it will add up to max_protohdr-len extra bytes to the 9931541Srgrimes * contiguous region in an attempt to avoid being called next time. 9941541Srgrimes */ 9951541Srgrimesstruct mbuf * 99672356Sbmilekicm_pullup(struct mbuf *n, int len) 9971541Srgrimes{ 99872356Sbmilekic struct mbuf *m; 99972356Sbmilekic int count; 10001541Srgrimes int space; 10011541Srgrimes 10021541Srgrimes /* 10031541Srgrimes * If first mbuf has no cluster, and has room for len bytes 10041541Srgrimes * without shifting current data, pullup into it, 10051541Srgrimes * otherwise allocate a new mbuf to prepend to the chain. 10061541Srgrimes */ 10071541Srgrimes if ((n->m_flags & M_EXT) == 0 && 10081541Srgrimes n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 10091541Srgrimes if (n->m_len >= len) 10101541Srgrimes return (n); 10111541Srgrimes m = n; 10121541Srgrimes n = n->m_next; 10131541Srgrimes len -= m->m_len; 10141541Srgrimes } else { 10151541Srgrimes if (len > MHLEN) 10161541Srgrimes goto bad; 1017111119Simp MGET(m, M_DONTWAIT, n->m_type); 101872356Sbmilekic if (m == NULL) 10191541Srgrimes goto bad; 10201541Srgrimes m->m_len = 0; 1021108466Ssam if (n->m_flags & M_PKTHDR) 1022108466Ssam M_MOVE_PKTHDR(m, n); 10231541Srgrimes } 10241541Srgrimes space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 10251541Srgrimes do { 10261541Srgrimes count = min(min(max(len, max_protohdr), space), n->m_len); 10271541Srgrimes bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 1028103569Sbmilekic (u_int)count); 10291541Srgrimes len -= count; 10301541Srgrimes m->m_len += count; 10311541Srgrimes n->m_len -= count; 10321541Srgrimes space -= count; 10331541Srgrimes if (n->m_len) 10341541Srgrimes n->m_data += count; 10351541Srgrimes else 10361541Srgrimes n = m_free(n); 10371541Srgrimes } while (len > 0 && n); 10381541Srgrimes if (len > 0) { 10391541Srgrimes (void) m_free(m); 10401541Srgrimes goto bad; 10411541Srgrimes } 10421541Srgrimes m->m_next = n; 10431541Srgrimes return (m); 10441541Srgrimesbad: 10451541Srgrimes m_freem(n); 104678592Sbmilekic mbstat.m_mpfail++; /* XXX: No consistency. */ 104772356Sbmilekic return (NULL); 10481541Srgrimes} 10491541Srgrimes 10501541Srgrimes/* 1051143761Sjmg * Like m_pullup(), except a new mbuf is always allocated, and we allow 1052143761Sjmg * the amount of empty space before the data in the new mbuf to be specified 1053143761Sjmg * (in the event that the caller expects to prepend later). 1054143761Sjmg */ 1055143761Sjmgint MSFail; 1056143761Sjmg 1057143761Sjmgstruct mbuf * 1058143761Sjmgm_copyup(struct mbuf *n, int len, int dstoff) 1059143761Sjmg{ 1060143761Sjmg struct mbuf *m; 1061143761Sjmg int count, space; 1062143761Sjmg 1063143761Sjmg if (len > (MHLEN - dstoff)) 1064143761Sjmg goto bad; 1065143761Sjmg MGET(m, M_DONTWAIT, n->m_type); 1066143761Sjmg if (m == NULL) 1067143761Sjmg goto bad; 1068143761Sjmg m->m_len = 0; 1069143761Sjmg if (n->m_flags & M_PKTHDR) 1070143761Sjmg M_MOVE_PKTHDR(m, n); 1071143761Sjmg m->m_data += dstoff; 1072143761Sjmg space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 1073143761Sjmg do { 1074143761Sjmg count = min(min(max(len, max_protohdr), space), n->m_len); 1075143761Sjmg memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t), 1076143761Sjmg (unsigned)count); 1077143761Sjmg len -= count; 1078143761Sjmg m->m_len += count; 1079143761Sjmg n->m_len -= count; 1080143761Sjmg space -= count; 1081143761Sjmg if (n->m_len) 1082143761Sjmg n->m_data += count; 1083143761Sjmg else 1084143761Sjmg n = m_free(n); 1085143761Sjmg } while (len > 0 && n); 1086143761Sjmg if (len > 0) { 1087143761Sjmg (void) m_free(m); 1088143761Sjmg goto bad; 1089143761Sjmg } 1090143761Sjmg m->m_next = n; 1091143761Sjmg return (m); 1092143761Sjmg bad: 1093143761Sjmg m_freem(n); 1094143761Sjmg MSFail++; 1095143761Sjmg return (NULL); 1096143761Sjmg} 1097143761Sjmg 1098143761Sjmg/* 10991541Srgrimes * Partition an mbuf chain in two pieces, returning the tail -- 11001541Srgrimes * all but the first len0 bytes. In case of failure, it returns NULL and 11011541Srgrimes * attempts to restore the chain to its original state. 110297681Sarchie * 110397681Sarchie * Note that the resulting mbufs might be read-only, because the new 110497681Sarchie * mbuf can end up sharing an mbuf cluster with the original mbuf if 110597681Sarchie * the "breaking point" happens to lie within a cluster mbuf. Use the 110697681Sarchie * M_WRITABLE() macro to check for this case. 11071541Srgrimes */ 11081541Srgrimesstruct mbuf * 110972356Sbmilekicm_split(struct mbuf *m0, int len0, int wait) 11101541Srgrimes{ 111172356Sbmilekic struct mbuf *m, *n; 1112103569Sbmilekic u_int len = len0, remain; 11131541Srgrimes 1114132488Salfred MBUF_CHECKSLEEP(wait); 11151541Srgrimes for (m = m0; m && len > m->m_len; m = m->m_next) 11161541Srgrimes len -= m->m_len; 111772356Sbmilekic if (m == NULL) 111872356Sbmilekic return (NULL); 11191541Srgrimes remain = m->m_len - len; 11201541Srgrimes if (m0->m_flags & M_PKTHDR) { 11211541Srgrimes MGETHDR(n, wait, m0->m_type); 112272356Sbmilekic if (n == NULL) 112372356Sbmilekic return (NULL); 11241541Srgrimes n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 11251541Srgrimes n->m_pkthdr.len = m0->m_pkthdr.len - len0; 11261541Srgrimes m0->m_pkthdr.len = len0; 11271541Srgrimes if (m->m_flags & M_EXT) 11281541Srgrimes goto extpacket; 11291541Srgrimes if (remain > MHLEN) { 11301541Srgrimes /* m can't be the lead packet */ 11311541Srgrimes MH_ALIGN(n, 0); 11321541Srgrimes n->m_next = m_split(m, len, wait); 113372356Sbmilekic if (n->m_next == NULL) { 11341541Srgrimes (void) m_free(n); 113572356Sbmilekic return (NULL); 113694471Shsu } else { 113794471Shsu n->m_len = 0; 11381541Srgrimes return (n); 113994471Shsu } 11401541Srgrimes } else 11411541Srgrimes MH_ALIGN(n, remain); 11421541Srgrimes } else if (remain == 0) { 11431541Srgrimes n = m->m_next; 114472356Sbmilekic m->m_next = NULL; 11451541Srgrimes return (n); 11461541Srgrimes } else { 11471541Srgrimes MGET(n, wait, m->m_type); 114872356Sbmilekic if (n == NULL) 114972356Sbmilekic return (NULL); 11501541Srgrimes M_ALIGN(n, remain); 11511541Srgrimes } 11521541Srgrimesextpacket: 11531541Srgrimes if (m->m_flags & M_EXT) { 11541541Srgrimes n->m_data = m->m_data + len; 1155151976Sandre mb_dupcl(n, m); 11561541Srgrimes } else { 11571541Srgrimes bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 11581541Srgrimes } 11591541Srgrimes n->m_len = remain; 11601541Srgrimes m->m_len = len; 11611541Srgrimes n->m_next = m->m_next; 116272356Sbmilekic m->m_next = NULL; 11631541Srgrimes return (n); 11641541Srgrimes} 11651541Srgrimes/* 11661541Srgrimes * Routine to copy from device local memory into mbufs. 116778508Sbmilekic * Note that `off' argument is offset into first mbuf of target chain from 116878508Sbmilekic * which to begin copying the data to. 11691541Srgrimes */ 11701541Srgrimesstruct mbuf * 117178508Sbmilekicm_devget(char *buf, int totlen, int off, struct ifnet *ifp, 1172169624Srwatson void (*copy)(char *from, caddr_t to, u_int len)) 11731541Srgrimes{ 117472356Sbmilekic struct mbuf *m; 1175129906Sbmilekic struct mbuf *top = NULL, **mp = ⊤ 117678508Sbmilekic int len; 11771541Srgrimes 117878508Sbmilekic if (off < 0 || off > MHLEN) 117978508Sbmilekic return (NULL); 118078508Sbmilekic 1181129906Sbmilekic while (totlen > 0) { 1182129906Sbmilekic if (top == NULL) { /* First one, must be PKTHDR */ 1183129906Sbmilekic if (totlen + off >= MINCLSIZE) { 1184129906Sbmilekic m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 1185129906Sbmilekic len = MCLBYTES; 1186129906Sbmilekic } else { 1187129906Sbmilekic m = m_gethdr(M_DONTWAIT, MT_DATA); 1188129906Sbmilekic len = MHLEN; 11891541Srgrimes 1190129906Sbmilekic /* Place initial small packet/header at end of mbuf */ 1191129906Sbmilekic if (m && totlen + off + max_linkhdr <= MLEN) { 1192129906Sbmilekic m->m_data += max_linkhdr; 1193129906Sbmilekic len -= max_linkhdr; 1194129906Sbmilekic } 1195129906Sbmilekic } 1196129906Sbmilekic if (m == NULL) 1197129906Sbmilekic return NULL; 1198129906Sbmilekic m->m_pkthdr.rcvif = ifp; 1199129906Sbmilekic m->m_pkthdr.len = totlen; 1200129906Sbmilekic } else { 1201129906Sbmilekic if (totlen + off >= MINCLSIZE) { 1202129906Sbmilekic m = m_getcl(M_DONTWAIT, MT_DATA, 0); 1203129906Sbmilekic len = MCLBYTES; 1204129906Sbmilekic } else { 1205129906Sbmilekic m = m_get(M_DONTWAIT, MT_DATA); 1206129906Sbmilekic len = MLEN; 1207129906Sbmilekic } 120872356Sbmilekic if (m == NULL) { 12091541Srgrimes m_freem(top); 1210129906Sbmilekic return NULL; 12111541Srgrimes } 12121541Srgrimes } 121378508Sbmilekic if (off) { 121478508Sbmilekic m->m_data += off; 121578508Sbmilekic len -= off; 121678508Sbmilekic off = 0; 121778508Sbmilekic } 121878508Sbmilekic m->m_len = len = min(totlen, len); 12191541Srgrimes if (copy) 1220103569Sbmilekic copy(buf, mtod(m, caddr_t), (u_int)len); 12211541Srgrimes else 1222103569Sbmilekic bcopy(buf, mtod(m, caddr_t), (u_int)len); 122378508Sbmilekic buf += len; 12241541Srgrimes *mp = m; 12251541Srgrimes mp = &m->m_next; 12261541Srgrimes totlen -= len; 12271541Srgrimes } 12281541Srgrimes return (top); 12291541Srgrimes} 12303352Sphk 12313352Sphk/* 12323352Sphk * Copy data from a buffer back into the indicated mbuf chain, 12333352Sphk * starting "off" bytes from the beginning, extending the mbuf 12343352Sphk * chain if necessary. 12353352Sphk */ 12363352Sphkvoid 1237128402Sluigim_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp) 12383352Sphk{ 123972356Sbmilekic int mlen; 124072356Sbmilekic struct mbuf *m = m0, *n; 12413352Sphk int totlen = 0; 12423352Sphk 124372356Sbmilekic if (m0 == NULL) 12443352Sphk return; 12453352Sphk while (off > (mlen = m->m_len)) { 12463352Sphk off -= mlen; 12473352Sphk totlen += mlen; 124872356Sbmilekic if (m->m_next == NULL) { 1249129906Sbmilekic n = m_get(M_DONTWAIT, m->m_type); 125072356Sbmilekic if (n == NULL) 12513352Sphk goto out; 1252129906Sbmilekic bzero(mtod(n, caddr_t), MLEN); 12533352Sphk n->m_len = min(MLEN, len + off); 12543352Sphk m->m_next = n; 12553352Sphk } 12563352Sphk m = m->m_next; 12573352Sphk } 12583352Sphk while (len > 0) { 12593352Sphk mlen = min (m->m_len - off, len); 1260103569Sbmilekic bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen); 12613352Sphk cp += mlen; 12623352Sphk len -= mlen; 12633352Sphk mlen += off; 12643352Sphk off = 0; 12653352Sphk totlen += mlen; 12663352Sphk if (len == 0) 12673352Sphk break; 126872356Sbmilekic if (m->m_next == NULL) { 1269111119Simp n = m_get(M_DONTWAIT, m->m_type); 127072356Sbmilekic if (n == NULL) 12713352Sphk break; 12723352Sphk n->m_len = min(MLEN, len); 12733352Sphk m->m_next = n; 12743352Sphk } 12753352Sphk m = m->m_next; 12763352Sphk } 12773352Sphkout: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 12783352Sphk m->m_pkthdr.len = totlen; 12793352Sphk} 128052756Sphk 1281123557Sbms/* 1282138541Ssam * Append the specified data to the indicated mbuf chain, 1283138541Ssam * Extend the mbuf chain if the new data does not fit in 1284138541Ssam * existing space. 1285138541Ssam * 1286138541Ssam * Return 1 if able to complete the job; otherwise 0. 1287138541Ssam */ 1288138541Ssamint 1289138541Ssamm_append(struct mbuf *m0, int len, c_caddr_t cp) 1290138541Ssam{ 1291138541Ssam struct mbuf *m, *n; 1292138541Ssam int remainder, space; 1293138541Ssam 1294138541Ssam for (m = m0; m->m_next != NULL; m = m->m_next) 1295138541Ssam ; 1296138541Ssam remainder = len; 1297138541Ssam space = M_TRAILINGSPACE(m); 1298138541Ssam if (space > 0) { 1299138541Ssam /* 1300138541Ssam * Copy into available space. 1301138541Ssam */ 1302138541Ssam if (space > remainder) 1303138541Ssam space = remainder; 1304138541Ssam bcopy(cp, mtod(m, caddr_t) + m->m_len, space); 1305138541Ssam m->m_len += space; 1306138541Ssam cp += space, remainder -= space; 1307138541Ssam } 1308138541Ssam while (remainder > 0) { 1309138541Ssam /* 1310138541Ssam * Allocate a new mbuf; could check space 1311138541Ssam * and allocate a cluster instead. 1312138541Ssam */ 1313138541Ssam n = m_get(M_DONTWAIT, m->m_type); 1314138541Ssam if (n == NULL) 1315138541Ssam break; 1316138541Ssam n->m_len = min(MLEN, remainder); 1317138894Ssam bcopy(cp, mtod(n, caddr_t), n->m_len); 1318138894Ssam cp += n->m_len, remainder -= n->m_len; 1319138541Ssam m->m_next = n; 1320138541Ssam m = n; 1321138541Ssam } 1322138541Ssam if (m0->m_flags & M_PKTHDR) 1323138541Ssam m0->m_pkthdr.len += len - remainder; 1324138541Ssam return (remainder == 0); 1325138541Ssam} 1326138541Ssam 1327138541Ssam/* 1328123557Sbms * Apply function f to the data in an mbuf chain starting "off" bytes from 1329123557Sbms * the beginning, continuing for "len" bytes. 1330123557Sbms */ 1331123557Sbmsint 1332123557Sbmsm_apply(struct mbuf *m, int off, int len, 1333123564Sbms int (*f)(void *, void *, u_int), void *arg) 1334123557Sbms{ 1335123564Sbms u_int count; 1336123557Sbms int rval; 1337123557Sbms 1338123557Sbms KASSERT(off >= 0, ("m_apply, negative off %d", off)); 1339123557Sbms KASSERT(len >= 0, ("m_apply, negative len %d", len)); 1340123557Sbms while (off > 0) { 1341123557Sbms KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1342123557Sbms if (off < m->m_len) 1343123557Sbms break; 1344123557Sbms off -= m->m_len; 1345123557Sbms m = m->m_next; 1346123557Sbms } 1347123557Sbms while (len > 0) { 1348123557Sbms KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain")); 1349123557Sbms count = min(m->m_len - off, len); 1350123557Sbms rval = (*f)(arg, mtod(m, caddr_t) + off, count); 1351123557Sbms if (rval) 1352123557Sbms return (rval); 1353123557Sbms len -= count; 1354123557Sbms off = 0; 1355123557Sbms m = m->m_next; 1356123557Sbms } 1357123557Sbms return (0); 1358123557Sbms} 1359123557Sbms 1360123557Sbms/* 1361123557Sbms * Return a pointer to mbuf/offset of location in mbuf chain. 1362123557Sbms */ 1363123557Sbmsstruct mbuf * 1364123557Sbmsm_getptr(struct mbuf *m, int loc, int *off) 1365123557Sbms{ 1366123557Sbms 1367123557Sbms while (loc >= 0) { 1368123564Sbms /* Normal end of search. */ 1369123557Sbms if (m->m_len > loc) { 1370123557Sbms *off = loc; 1371123557Sbms return (m); 1372123557Sbms } else { 1373123557Sbms loc -= m->m_len; 1374123557Sbms if (m->m_next == NULL) { 1375123557Sbms if (loc == 0) { 1376123564Sbms /* Point at the end of valid data. */ 1377123557Sbms *off = m->m_len; 1378123557Sbms return (m); 1379123564Sbms } 1380123564Sbms return (NULL); 1381123564Sbms } 1382123564Sbms m = m->m_next; 1383123557Sbms } 1384123557Sbms } 1385123557Sbms return (NULL); 1386123557Sbms} 1387123557Sbms 138852756Sphkvoid 1389135904Sjmgm_print(const struct mbuf *m, int maxlen) 139052756Sphk{ 139152756Sphk int len; 1392135904Sjmg int pdata; 139354906Seivind const struct mbuf *m2; 139452756Sphk 1395135904Sjmg if (m->m_flags & M_PKTHDR) 1396135904Sjmg len = m->m_pkthdr.len; 1397135904Sjmg else 1398135904Sjmg len = -1; 139952756Sphk m2 = m; 1400135904Sjmg while (m2 != NULL && (len == -1 || len)) { 1401135904Sjmg pdata = m2->m_len; 1402135904Sjmg if (maxlen != -1 && pdata > maxlen) 1403135904Sjmg pdata = maxlen; 1404135904Sjmg printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len, 1405135904Sjmg m2->m_next, m2->m_flags, "\20\20freelist\17skipfw" 1406135904Sjmg "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly" 1407135904Sjmg "\3eor\2pkthdr\1ext", pdata ? "" : "\n"); 1408135904Sjmg if (pdata) 1409156700Sjmg printf(", %*D\n", pdata, (u_char *)m2->m_data, "-"); 1410135904Sjmg if (len != -1) 1411135904Sjmg len -= m2->m_len; 141252756Sphk m2 = m2->m_next; 141352756Sphk } 1414135904Sjmg if (len > 0) 1415135904Sjmg printf("%d bytes unaccounted for.\n", len); 141652756Sphk return; 141752756Sphk} 1418103540Sphk 1419103569Sbmilekicu_int 1420103540Sphkm_fixhdr(struct mbuf *m0) 1421103540Sphk{ 1422103569Sbmilekic u_int len; 1423103540Sphk 1424103544Sphk len = m_length(m0, NULL); 1425103544Sphk m0->m_pkthdr.len = len; 1426103544Sphk return (len); 1427103544Sphk} 1428103544Sphk 1429103569Sbmilekicu_int 1430103544Sphkm_length(struct mbuf *m0, struct mbuf **last) 1431103544Sphk{ 1432103544Sphk struct mbuf *m; 1433103569Sbmilekic u_int len; 1434103544Sphk 1435103544Sphk len = 0; 1436103544Sphk for (m = m0; m != NULL; m = m->m_next) { 1437103540Sphk len += m->m_len; 1438103544Sphk if (m->m_next == NULL) 1439103544Sphk break; 1440103540Sphk } 1441103544Sphk if (last != NULL) 1442103544Sphk *last = m; 1443103544Sphk return (len); 1444103540Sphk} 1445112777Ssilby 1446112777Ssilby/* 1447112777Ssilby * Defragment a mbuf chain, returning the shortest possible 1448112777Ssilby * chain of mbufs and clusters. If allocation fails and 1449112777Ssilby * this cannot be completed, NULL will be returned, but 1450112777Ssilby * the passed in chain will be unchanged. Upon success, 1451112777Ssilby * the original chain will be freed, and the new chain 1452112777Ssilby * will be returned. 1453112777Ssilby * 1454112777Ssilby * If a non-packet header is passed in, the original 1455112777Ssilby * mbuf (chain?) will be returned unharmed. 1456112777Ssilby */ 1457112777Ssilbystruct mbuf * 1458112777Ssilbym_defrag(struct mbuf *m0, int how) 1459112777Ssilby{ 1460125472Ssilby struct mbuf *m_new = NULL, *m_final = NULL; 1461125472Ssilby int progress = 0, length; 1462112777Ssilby 1463132488Salfred MBUF_CHECKSLEEP(how); 1464112777Ssilby if (!(m0->m_flags & M_PKTHDR)) 1465112777Ssilby return (m0); 1466112777Ssilby 1467117770Ssilby m_fixhdr(m0); /* Needed sanity check */ 1468117770Ssilby 1469113490Ssilby#ifdef MBUF_STRESS_TEST 1470113490Ssilby if (m_defragrandomfailures) { 1471113490Ssilby int temp = arc4random() & 0xff; 1472113490Ssilby if (temp == 0xba) 1473113490Ssilby goto nospace; 1474113490Ssilby } 1475113490Ssilby#endif 1476112777Ssilby 1477112777Ssilby if (m0->m_pkthdr.len > MHLEN) 1478112777Ssilby m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1479112777Ssilby else 1480112777Ssilby m_final = m_gethdr(how, MT_DATA); 1481112777Ssilby 1482112777Ssilby if (m_final == NULL) 1483112777Ssilby goto nospace; 1484112777Ssilby 1485123740Speter if (m_dup_pkthdr(m_final, m0, how) == 0) 1486112777Ssilby goto nospace; 1487112777Ssilby 1488112777Ssilby m_new = m_final; 1489112777Ssilby 1490112777Ssilby while (progress < m0->m_pkthdr.len) { 1491112777Ssilby length = m0->m_pkthdr.len - progress; 1492112777Ssilby if (length > MCLBYTES) 1493112777Ssilby length = MCLBYTES; 1494112777Ssilby 1495112777Ssilby if (m_new == NULL) { 1496112777Ssilby if (length > MLEN) 1497112777Ssilby m_new = m_getcl(how, MT_DATA, 0); 1498112777Ssilby else 1499112777Ssilby m_new = m_get(how, MT_DATA); 1500112777Ssilby if (m_new == NULL) 1501112777Ssilby goto nospace; 1502112777Ssilby } 1503112777Ssilby 1504112777Ssilby m_copydata(m0, progress, length, mtod(m_new, caddr_t)); 1505112777Ssilby progress += length; 1506112777Ssilby m_new->m_len = length; 1507112777Ssilby if (m_new != m_final) 1508112777Ssilby m_cat(m_final, m_new); 1509112777Ssilby m_new = NULL; 1510112777Ssilby } 1511116455Ssilby#ifdef MBUF_STRESS_TEST 1512112777Ssilby if (m0->m_next == NULL) 1513112777Ssilby m_defraguseless++; 1514116455Ssilby#endif 1515112777Ssilby m_freem(m0); 1516112777Ssilby m0 = m_final; 1517116455Ssilby#ifdef MBUF_STRESS_TEST 1518112777Ssilby m_defragpackets++; 1519112777Ssilby m_defragbytes += m0->m_pkthdr.len; 1520116455Ssilby#endif 1521112777Ssilby return (m0); 1522112777Ssilbynospace: 1523116455Ssilby#ifdef MBUF_STRESS_TEST 1524112777Ssilby m_defragfailure++; 1525116455Ssilby#endif 1526112777Ssilby if (m_final) 1527112777Ssilby m_freem(m_final); 1528112777Ssilby return (NULL); 1529112777Ssilby} 1530119644Ssilby 1531119644Ssilby#ifdef MBUF_STRESS_TEST 1532119644Ssilby 1533119644Ssilby/* 1534119644Ssilby * Fragment an mbuf chain. There's no reason you'd ever want to do 1535119644Ssilby * this in normal usage, but it's great for stress testing various 1536119644Ssilby * mbuf consumers. 1537119644Ssilby * 1538119644Ssilby * If fragmentation is not possible, the original chain will be 1539119644Ssilby * returned. 1540119644Ssilby * 1541119644Ssilby * Possible length values: 1542119644Ssilby * 0 no fragmentation will occur 1543119644Ssilby * > 0 each fragment will be of the specified length 1544119644Ssilby * -1 each fragment will be the same random value in length 1545119644Ssilby * -2 each fragment's length will be entirely random 1546119644Ssilby * (Random values range from 1 to 256) 1547119644Ssilby */ 1548119644Ssilbystruct mbuf * 1549119644Ssilbym_fragment(struct mbuf *m0, int how, int length) 1550119644Ssilby{ 1551125472Ssilby struct mbuf *m_new = NULL, *m_final = NULL; 1552125472Ssilby int progress = 0; 1553119644Ssilby 1554119644Ssilby if (!(m0->m_flags & M_PKTHDR)) 1555119644Ssilby return (m0); 1556119644Ssilby 1557119644Ssilby if ((length == 0) || (length < -2)) 1558119644Ssilby return (m0); 1559119644Ssilby 1560119644Ssilby m_fixhdr(m0); /* Needed sanity check */ 1561119644Ssilby 1562119644Ssilby m_final = m_getcl(how, MT_DATA, M_PKTHDR); 1563119644Ssilby 1564119644Ssilby if (m_final == NULL) 1565119644Ssilby goto nospace; 1566119644Ssilby 1567123823Ssilby if (m_dup_pkthdr(m_final, m0, how) == 0) 1568119644Ssilby goto nospace; 1569119644Ssilby 1570119644Ssilby m_new = m_final; 1571119644Ssilby 1572119644Ssilby if (length == -1) 1573119644Ssilby length = 1 + (arc4random() & 255); 1574119644Ssilby 1575119644Ssilby while (progress < m0->m_pkthdr.len) { 1576119644Ssilby int fraglen; 1577119644Ssilby 1578119644Ssilby if (length > 0) 1579119644Ssilby fraglen = length; 1580119644Ssilby else 1581119644Ssilby fraglen = 1 + (arc4random() & 255); 1582119644Ssilby if (fraglen > m0->m_pkthdr.len - progress) 1583119644Ssilby fraglen = m0->m_pkthdr.len - progress; 1584119644Ssilby 1585119644Ssilby if (fraglen > MCLBYTES) 1586119644Ssilby fraglen = MCLBYTES; 1587119644Ssilby 1588119644Ssilby if (m_new == NULL) { 1589119644Ssilby m_new = m_getcl(how, MT_DATA, 0); 1590119644Ssilby if (m_new == NULL) 1591119644Ssilby goto nospace; 1592119644Ssilby } 1593119644Ssilby 1594119644Ssilby m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t)); 1595119644Ssilby progress += fraglen; 1596119644Ssilby m_new->m_len = fraglen; 1597119644Ssilby if (m_new != m_final) 1598119644Ssilby m_cat(m_final, m_new); 1599119644Ssilby m_new = NULL; 1600119644Ssilby } 1601119644Ssilby m_freem(m0); 1602119644Ssilby m0 = m_final; 1603119644Ssilby return (m0); 1604119644Ssilbynospace: 1605119644Ssilby if (m_final) 1606119644Ssilby m_freem(m_final); 1607119644Ssilby /* Return the original chain on failure */ 1608119644Ssilby return (m0); 1609119644Ssilby} 1610119644Ssilby 1611119644Ssilby#endif 1612125296Ssilby 1613163915Sandre/* 1614163915Sandre * Copy the contents of uio into a properly sized mbuf chain. 1615163915Sandre */ 1616125296Ssilbystruct mbuf * 1617163915Sandrem_uiotombuf(struct uio *uio, int how, int len, int align, int flags) 1618125296Ssilby{ 1619163915Sandre struct mbuf *m, *mb; 1620163915Sandre int error, length, total; 1621163915Sandre int progress = 0; 1622125296Ssilby 1623163915Sandre /* 1624163915Sandre * len can be zero or an arbitrary large value bound by 1625163915Sandre * the total data supplied by the uio. 1626163915Sandre */ 1627125296Ssilby if (len > 0) 1628125296Ssilby total = min(uio->uio_resid, len); 1629125296Ssilby else 1630125296Ssilby total = uio->uio_resid; 1631163915Sandre 1632163915Sandre /* 1633163915Sandre * The smallest unit returned by m_getm2() is a single mbuf 1634163915Sandre * with pkthdr. We can't align past it. Align align itself. 1635163915Sandre */ 1636163915Sandre if (align) 1637163915Sandre align &= ~(sizeof(long) - 1); 1638145883Semax if (align >= MHLEN) 1639163915Sandre return (NULL); 1640163915Sandre 1641166171Sandre /* 1642166171Sandre * Give us the full allocation or nothing. 1643166171Sandre * If len is zero return the smallest empty mbuf. 1644166171Sandre */ 1645166171Sandre m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags); 1646163915Sandre if (m == NULL) 1647163915Sandre return (NULL); 1648163915Sandre m->m_data += align; 1649163915Sandre 1650163915Sandre /* Fill all mbufs with uio data and update header information. */ 1651163915Sandre for (mb = m; mb != NULL; mb = mb->m_next) { 1652163915Sandre length = min(M_TRAILINGSPACE(mb), total - progress); 1653163915Sandre 1654163915Sandre error = uiomove(mtod(mb, void *), length, uio); 1655163915Sandre if (error) { 1656163915Sandre m_freem(m); 1657163915Sandre return (NULL); 1658125296Ssilby } 1659163915Sandre 1660163915Sandre mb->m_len = length; 1661125296Ssilby progress += length; 1662163915Sandre if (flags & M_PKTHDR) 1663163915Sandre m->m_pkthdr.len += length; 1664125296Ssilby } 1665163915Sandre KASSERT(progress == total, ("%s: progress != total", __func__)); 1666163915Sandre 1667163915Sandre return (m); 1668125296Ssilby} 1669148552Ssam 1670148552Ssam/* 1671148552Ssam * Set the m_data pointer of a newly-allocated mbuf 1672148552Ssam * to place an object of the specified size at the 1673148552Ssam * end of the mbuf, longword aligned. 1674148552Ssam */ 1675148552Ssamvoid 1676148552Ssamm_align(struct mbuf *m, int len) 1677148552Ssam{ 1678148552Ssam int adjust; 1679148552Ssam 1680148552Ssam if (m->m_flags & M_EXT) 1681148552Ssam adjust = m->m_ext.ext_size - len; 1682148552Ssam else if (m->m_flags & M_PKTHDR) 1683148552Ssam adjust = MHLEN - len; 1684148552Ssam else 1685148552Ssam adjust = MLEN - len; 1686148552Ssam m->m_data += adjust &~ (sizeof(long)-1); 1687148552Ssam} 1688156756Ssam 1689156756Ssam/* 1690156756Ssam * Create a writable copy of the mbuf chain. While doing this 1691156756Ssam * we compact the chain with a goal of producing a chain with 1692156756Ssam * at most two mbufs. The second mbuf in this chain is likely 1693156756Ssam * to be a cluster. The primary purpose of this work is to create 1694156756Ssam * a writable packet for encryption, compression, etc. The 1695156756Ssam * secondary goal is to linearize the data so the data can be 1696156756Ssam * passed to crypto hardware in the most efficient manner possible. 1697156756Ssam */ 1698156756Ssamstruct mbuf * 1699156756Ssamm_unshare(struct mbuf *m0, int how) 1700156756Ssam{ 1701156756Ssam struct mbuf *m, *mprev; 1702156756Ssam struct mbuf *n, *mfirst, *mlast; 1703156756Ssam int len, off; 1704156756Ssam 1705156756Ssam mprev = NULL; 1706156756Ssam for (m = m0; m != NULL; m = mprev->m_next) { 1707156756Ssam /* 1708156756Ssam * Regular mbufs are ignored unless there's a cluster 1709156756Ssam * in front of it that we can use to coalesce. We do 1710156756Ssam * the latter mainly so later clusters can be coalesced 1711156756Ssam * also w/o having to handle them specially (i.e. convert 1712156756Ssam * mbuf+cluster -> cluster). This optimization is heavily 1713156756Ssam * influenced by the assumption that we're running over 1714156756Ssam * Ethernet where MCLBYTES is large enough that the max 1715156756Ssam * packet size will permit lots of coalescing into a 1716156756Ssam * single cluster. This in turn permits efficient 1717156756Ssam * crypto operations, especially when using hardware. 1718156756Ssam */ 1719156756Ssam if ((m->m_flags & M_EXT) == 0) { 1720156756Ssam if (mprev && (mprev->m_flags & M_EXT) && 1721156756Ssam m->m_len <= M_TRAILINGSPACE(mprev)) { 1722156756Ssam /* XXX: this ignores mbuf types */ 1723156756Ssam memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1724156756Ssam mtod(m, caddr_t), m->m_len); 1725156756Ssam mprev->m_len += m->m_len; 1726156756Ssam mprev->m_next = m->m_next; /* unlink from chain */ 1727156756Ssam m_free(m); /* reclaim mbuf */ 1728156756Ssam#if 0 1729156756Ssam newipsecstat.ips_mbcoalesced++; 1730156756Ssam#endif 1731156756Ssam } else { 1732156756Ssam mprev = m; 1733156756Ssam } 1734156756Ssam continue; 1735156756Ssam } 1736156756Ssam /* 1737156756Ssam * Writable mbufs are left alone (for now). 1738156756Ssam */ 1739156756Ssam if (M_WRITABLE(m)) { 1740156756Ssam mprev = m; 1741156756Ssam continue; 1742156756Ssam } 1743156756Ssam 1744156756Ssam /* 1745156756Ssam * Not writable, replace with a copy or coalesce with 1746156756Ssam * the previous mbuf if possible (since we have to copy 1747156756Ssam * it anyway, we try to reduce the number of mbufs and 1748156756Ssam * clusters so that future work is easier). 1749156756Ssam */ 1750156756Ssam KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags)); 1751156756Ssam /* NB: we only coalesce into a cluster or larger */ 1752156756Ssam if (mprev != NULL && (mprev->m_flags & M_EXT) && 1753156756Ssam m->m_len <= M_TRAILINGSPACE(mprev)) { 1754156756Ssam /* XXX: this ignores mbuf types */ 1755156756Ssam memcpy(mtod(mprev, caddr_t) + mprev->m_len, 1756156756Ssam mtod(m, caddr_t), m->m_len); 1757156756Ssam mprev->m_len += m->m_len; 1758156756Ssam mprev->m_next = m->m_next; /* unlink from chain */ 1759156756Ssam m_free(m); /* reclaim mbuf */ 1760156756Ssam#if 0 1761156756Ssam newipsecstat.ips_clcoalesced++; 1762156756Ssam#endif 1763156756Ssam continue; 1764156756Ssam } 1765156756Ssam 1766156756Ssam /* 1767156756Ssam * Allocate new space to hold the copy... 1768156756Ssam */ 1769156756Ssam /* XXX why can M_PKTHDR be set past the first mbuf? */ 1770156756Ssam if (mprev == NULL && (m->m_flags & M_PKTHDR)) { 1771156756Ssam /* 1772156756Ssam * NB: if a packet header is present we must 1773156756Ssam * allocate the mbuf separately from any cluster 1774156756Ssam * because M_MOVE_PKTHDR will smash the data 1775156756Ssam * pointer and drop the M_EXT marker. 1776156756Ssam */ 1777156756Ssam MGETHDR(n, how, m->m_type); 1778156756Ssam if (n == NULL) { 1779156756Ssam m_freem(m0); 1780156756Ssam return (NULL); 1781156756Ssam } 1782156756Ssam M_MOVE_PKTHDR(n, m); 1783156756Ssam MCLGET(n, how); 1784156756Ssam if ((n->m_flags & M_EXT) == 0) { 1785156756Ssam m_free(n); 1786156756Ssam m_freem(m0); 1787156756Ssam return (NULL); 1788156756Ssam } 1789156756Ssam } else { 1790156756Ssam n = m_getcl(how, m->m_type, m->m_flags); 1791156756Ssam if (n == NULL) { 1792156756Ssam m_freem(m0); 1793156756Ssam return (NULL); 1794156756Ssam } 1795156756Ssam } 1796156756Ssam /* 1797156756Ssam * ... and copy the data. We deal with jumbo mbufs 1798156756Ssam * (i.e. m_len > MCLBYTES) by splitting them into 1799156756Ssam * clusters. We could just malloc a buffer and make 1800156756Ssam * it external but too many device drivers don't know 1801156756Ssam * how to break up the non-contiguous memory when 1802156756Ssam * doing DMA. 1803156756Ssam */ 1804156756Ssam len = m->m_len; 1805156756Ssam off = 0; 1806156756Ssam mfirst = n; 1807156756Ssam mlast = NULL; 1808156756Ssam for (;;) { 1809156756Ssam int cc = min(len, MCLBYTES); 1810156756Ssam memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc); 1811156756Ssam n->m_len = cc; 1812156756Ssam if (mlast != NULL) 1813156756Ssam mlast->m_next = n; 1814156756Ssam mlast = n; 1815156756Ssam#if 0 1816156756Ssam newipsecstat.ips_clcopied++; 1817156756Ssam#endif 1818156756Ssam 1819156756Ssam len -= cc; 1820156756Ssam if (len <= 0) 1821156756Ssam break; 1822156756Ssam off += cc; 1823156756Ssam 1824156756Ssam n = m_getcl(how, m->m_type, m->m_flags); 1825156756Ssam if (n == NULL) { 1826156756Ssam m_freem(mfirst); 1827156756Ssam m_freem(m0); 1828156756Ssam return (NULL); 1829156756Ssam } 1830156756Ssam } 1831156756Ssam n->m_next = m->m_next; 1832156756Ssam if (mprev == NULL) 1833156756Ssam m0 = mfirst; /* new head of chain */ 1834156756Ssam else 1835156756Ssam mprev->m_next = mfirst; /* replace old mbuf */ 1836156756Ssam m_free(m); /* release old mbuf */ 1837156756Ssam mprev = mfirst; 1838156756Ssam } 1839156756Ssam return (m0); 1840156756Ssam} 1841