uipc_mbuf.c revision 64837
11541Srgrimes/* 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 * 3. All advertising materials mentioning features or use of this software 141541Srgrimes * must display the following acknowledgement: 151541Srgrimes * This product includes software developed by the University of 161541Srgrimes * California, Berkeley and its contributors. 171541Srgrimes * 4. Neither the name of the University nor the names of its contributors 181541Srgrimes * may be used to endorse or promote products derived from this software 191541Srgrimes * without specific prior written permission. 201541Srgrimes * 211541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 221541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 231541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 241541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 251541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 261541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 271541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 281541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 291541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 301541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 311541Srgrimes * SUCH DAMAGE. 321541Srgrimes * 331541Srgrimes * @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94 3450477Speter * $FreeBSD: head/sys/kern/uipc_mbuf.c 64837 2000-08-19 08:32:59Z dwmalone $ 351541Srgrimes */ 361541Srgrimes 3748579Smsmith#include "opt_param.h" 381541Srgrimes#include <sys/param.h> 391541Srgrimes#include <sys/systm.h> 4032036Sbde#include <sys/malloc.h> 411541Srgrimes#include <sys/mbuf.h> 421541Srgrimes#include <sys/kernel.h> 4323081Swollman#include <sys/sysctl.h> 441541Srgrimes#include <sys/domain.h> 451541Srgrimes#include <sys/protosw.h> 461541Srgrimes 471541Srgrimes#include <vm/vm.h> 489759Sbde#include <vm/vm_kern.h> 4912662Sdg#include <vm/vm_extern.h> 501541Srgrimes 5154478Sgreen#ifdef INVARIANTS 5254478Sgreen#include <machine/cpu.h> 5354478Sgreen#endif 5454478Sgreen 5510653Sdgstatic void mbinit __P((void *)); 5610358SjulianSYSINIT(mbuf, SI_SUB_MBUF, SI_ORDER_FIRST, mbinit, NULL) 5710358Sjulian 589759Sbdestruct mbuf *mbutl; 599759Sbdestruct mbstat mbstat; 6063203Salfredu_long mbtypes[MT_NTYPES]; 6115689Swollmanstruct mbuf *mmbfree; 629759Sbdeunion mcluster *mclfree; 6364837Sdwmaloneunion mext_refcnt *mext_refcnt_free; 649759Sbdeint max_linkhdr; 659759Sbdeint max_protohdr; 669759Sbdeint max_hdr; 679759Sbdeint max_datalen; 6848579Smsmithint nmbclusters; 6948579Smsmithint nmbufs; 7054584Sgreenu_int m_mballoc_wid = 0; 7154584Sgreenu_int m_clalloc_wid = 0; 721541Srgrimes 7344078SdfrSYSCTL_DECL(_kern_ipc); 7423081SwollmanSYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RW, 7523081Swollman &max_linkhdr, 0, ""); 7623081SwollmanSYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RW, 7723081Swollman &max_protohdr, 0, ""); 7823081SwollmanSYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RW, &max_hdr, 0, ""); 7923081SwollmanSYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RW, 8023081Swollman &max_datalen, 0, ""); 8154478SgreenSYSCTL_INT(_kern_ipc, OID_AUTO, mbuf_wait, CTLFLAG_RW, 8254478Sgreen &mbuf_wait, 0, ""); 8364048SalfredSYSCTL_STRUCT(_kern_ipc, KIPC_MBSTAT, mbstat, CTLFLAG_RD, &mbstat, mbstat, ""); 8463203SalfredSYSCTL_OPAQUE(_kern_ipc, OID_AUTO, mbtypes, CTLFLAG_RD, mbtypes, 8563203Salfred sizeof(mbtypes), "LU", ""); 8648579SmsmithSYSCTL_INT(_kern_ipc, KIPC_NMBCLUSTERS, nmbclusters, CTLFLAG_RD, 8755171Smsmith &nmbclusters, 0, "Maximum number of mbuf clusters available"); 8855171SmsmithSYSCTL_INT(_kern_ipc, OID_AUTO, nmbufs, CTLFLAG_RD, &nmbufs, 0, 8955171Smsmith "Maximum number of mbufs available"); 9048579Smsmith#ifndef NMBCLUSTERS 9148579Smsmith#define NMBCLUSTERS (512 + MAXUSERS * 16) 9248579Smsmith#endif 9348579SmsmithTUNABLE_INT_DECL("kern.ipc.nmbclusters", NMBCLUSTERS, nmbclusters); 9455171SmsmithTUNABLE_INT_DECL("kern.ipc.nmbufs", NMBCLUSTERS * 4, nmbufs); 9523081Swollman 9612819Sphkstatic void m_reclaim __P((void)); 9712819Sphk 9864837Sdwmalone#define NCL_INIT 2 9915744Sphk#define NMB_INIT 16 10064837Sdwmalone#define REF_INIT (NMBCLUSTERS * 2) 10115744Sphk 10210358Sjulian/* ARGSUSED*/ 10310358Sjulianstatic void 10412569Sbdembinit(dummy) 10512569Sbde void *dummy; 1061541Srgrimes{ 1071541Srgrimes int s; 1081541Srgrimes 10964837Sdwmalone mmbfree = NULL; 11064837Sdwmalone mclfree = NULL; 11164837Sdwmalone mext_refcnt_free = NULL; 11264837Sdwmalone 11323081Swollman mbstat.m_msize = MSIZE; 11423081Swollman mbstat.m_mclbytes = MCLBYTES; 11523081Swollman mbstat.m_minclsize = MINCLSIZE; 11623081Swollman mbstat.m_mlen = MLEN; 11723081Swollman mbstat.m_mhlen = MHLEN; 11823081Swollman 1191541Srgrimes s = splimp(); 12064837Sdwmalone if (m_alloc_ref(REF_INIT) == 0) 12164837Sdwmalone goto bad; 12215689Swollman if (m_mballoc(NMB_INIT, M_DONTWAIT) == 0) 12315689Swollman goto bad; 12422671Swollman#if MCLBYTES <= PAGE_SIZE 1251541Srgrimes if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0) 1261541Srgrimes goto bad; 12722671Swollman#else 12822671Swollman /* It's OK to call contigmalloc in this context. */ 12932036Sbde if (m_clalloc(16, M_WAIT) == 0) 13022671Swollman goto bad; 13122671Swollman#endif 1321541Srgrimes splx(s); 1331541Srgrimes return; 1341541Srgrimesbad: 13564837Sdwmalone panic("mbinit: failed to initialize mbuf subsystem!"); 1361541Srgrimes} 1371541Srgrimes 1381541Srgrimes/* 13964837Sdwmalone * Allocate at least nmb reference count structs and place them 14064837Sdwmalone * on the ref cnt free list. 14164837Sdwmalone * Must be called at splimp. 14264837Sdwmalone */ 14364837Sdwmaloneint 14464837Sdwmalonem_alloc_ref(nmb) 14564837Sdwmalone u_int nmb; 14664837Sdwmalone{ 14764837Sdwmalone caddr_t p; 14864837Sdwmalone u_int nbytes; 14964837Sdwmalone int i; 15064837Sdwmalone 15164837Sdwmalone /* 15264837Sdwmalone * XXX: 15364837Sdwmalone * We don't cap the amount of memory that can be used 15464837Sdwmalone * by the reference counters, like we do for mbufs and 15564837Sdwmalone * mbuf clusters. The reason is that we don't really expect 15664837Sdwmalone * to have to be allocating too many of these guys with m_alloc_ref(), 15764837Sdwmalone * and if we are, we're probably not out of the woods anyway, 15864837Sdwmalone * so leave this way for now. 15964837Sdwmalone */ 16064837Sdwmalone 16164837Sdwmalone if (mb_map_full) 16264837Sdwmalone return (0); 16364837Sdwmalone 16464837Sdwmalone nbytes = round_page(nmb * sizeof(union mext_refcnt)); 16564837Sdwmalone if ((p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT)) == NULL) 16664837Sdwmalone return (0); 16764837Sdwmalone nmb = nbytes / sizeof(union mext_refcnt); 16864837Sdwmalone 16964837Sdwmalone for (i = 0; i < nmb; i++) { 17064837Sdwmalone ((union mext_refcnt *)p)->next_ref = mext_refcnt_free; 17164837Sdwmalone mext_refcnt_free = (union mext_refcnt *)p; 17264837Sdwmalone p += sizeof(union mext_refcnt); 17364837Sdwmalone mbstat.m_refree++; 17464837Sdwmalone } 17564837Sdwmalone mbstat.m_refcnt += nmb; 17664837Sdwmalone 17764837Sdwmalone return (1); 17864837Sdwmalone} 17964837Sdwmalone 18064837Sdwmalone/* 18115689Swollman * Allocate at least nmb mbufs and place on mbuf free list. 18215689Swollman * Must be called at splimp. 18315689Swollman */ 18415689Swollman/* ARGSUSED */ 18515689Swollmanint 18632036Sbdem_mballoc(nmb, how) 18715689Swollman register int nmb; 18832036Sbde int how; 18915689Swollman{ 19015689Swollman register caddr_t p; 19115689Swollman register int i; 19215689Swollman int nbytes; 19315689Swollman 19454478Sgreen /* 19555171Smsmith * If we've hit the mbuf limit, stop allocating from mb_map, 19655171Smsmith * (or trying to) in order to avoid dipping into the section of 19755171Smsmith * mb_map which we've "reserved" for clusters. 19855171Smsmith */ 19955171Smsmith if ((nmb + mbstat.m_mbufs) > nmbufs) 20055171Smsmith return (0); 20155171Smsmith 20255171Smsmith /* 20354478Sgreen * Once we run out of map space, it will be impossible to get 20454478Sgreen * any more (nothing is ever freed back to the map) 20554478Sgreen * -- however you are not dead as m_reclaim might 20654478Sgreen * still be able to free a substantial amount of space. 20754478Sgreen * 20854478Sgreen * XXX Furthermore, we can also work with "recycled" mbufs (when 20954478Sgreen * we're calling with M_WAIT the sleep procedure will be woken 21054478Sgreen * up when an mbuf is freed. See m_mballoc_wait()). 21115689Swollman */ 21215689Swollman if (mb_map_full) 21315689Swollman return (0); 21415689Swollman 21515689Swollman nbytes = round_page(nmb * MSIZE); 21622899Swollman p = (caddr_t)kmem_malloc(mb_map, nbytes, M_NOWAIT); 21732036Sbde if (p == 0 && how == M_WAIT) { 21822899Swollman mbstat.m_wait++; 21922899Swollman p = (caddr_t)kmem_malloc(mb_map, nbytes, M_WAITOK); 22022899Swollman } 22122899Swollman 22215689Swollman /* 22332036Sbde * Either the map is now full, or `how' is M_NOWAIT and there 22415689Swollman * are no pages left. 22515689Swollman */ 22615689Swollman if (p == NULL) 22715689Swollman return (0); 22815689Swollman 22915689Swollman nmb = nbytes / MSIZE; 23015689Swollman for (i = 0; i < nmb; i++) { 23115689Swollman ((struct mbuf *)p)->m_next = mmbfree; 23215689Swollman mmbfree = (struct mbuf *)p; 23315689Swollman p += MSIZE; 23415689Swollman } 23515689Swollman mbstat.m_mbufs += nmb; 23663203Salfred mbtypes[MT_FREE] += nmb; 23715689Swollman return (1); 23815689Swollman} 23915689Swollman 24054478Sgreen/* 24154478Sgreen * Once the mb_map has been exhausted and if the call to the allocation macros 24254478Sgreen * (or, in some cases, functions) is with M_WAIT, then it is necessary to rely 24354478Sgreen * solely on reclaimed mbufs. Here we wait for an mbuf to be freed for a 24454478Sgreen * designated (mbuf_wait) time. 24554478Sgreen */ 24654478Sgreenstruct mbuf * 24754478Sgreenm_mballoc_wait(int caller, int type) 24854478Sgreen{ 24954478Sgreen struct mbuf *p; 25054478Sgreen int s; 25154478Sgreen 25254478Sgreen m_mballoc_wid++; 25354478Sgreen if ((tsleep(&m_mballoc_wid, PVM, "mballc", mbuf_wait)) == EWOULDBLOCK) 25454478Sgreen m_mballoc_wid--; 25554478Sgreen 25654478Sgreen /* 25754478Sgreen * Now that we (think) that we've got something, we will redo an 25854478Sgreen * MGET, but avoid getting into another instance of m_mballoc_wait() 25954478Sgreen * XXX: We retry to fetch _even_ if the sleep timed out. This is left 26054478Sgreen * this way, purposely, in the [unlikely] case that an mbuf was 26154478Sgreen * freed but the sleep was not awakened in time. 26254478Sgreen */ 26354478Sgreen p = NULL; 26454478Sgreen switch (caller) { 26554478Sgreen case MGET_C: 26654478Sgreen MGET(p, M_DONTWAIT, type); 26754478Sgreen break; 26854478Sgreen case MGETHDR_C: 26954478Sgreen MGETHDR(p, M_DONTWAIT, type); 27054478Sgreen break; 27154478Sgreen default: 27254478Sgreen panic("m_mballoc_wait: invalid caller (%d)", caller); 27354478Sgreen } 27454478Sgreen 27554478Sgreen s = splimp(); 27654478Sgreen if (p != NULL) { /* We waited and got something... */ 27754478Sgreen mbstat.m_wait++; 27854478Sgreen /* Wake up another if we have more free. */ 27954478Sgreen if (mmbfree != NULL) 28054787Sgreen MMBWAKEUP(); 28154478Sgreen } 28254478Sgreen splx(s); 28354478Sgreen return (p); 28454478Sgreen} 28554478Sgreen 28622671Swollman#if MCLBYTES > PAGE_SIZE 28722899Swollmanstatic int i_want_my_mcl; 28822671Swollman 28922899Swollmanstatic void 29022671Swollmankproc_mclalloc(void) 29122671Swollman{ 29222671Swollman int status; 29322671Swollman 29422671Swollman while (1) { 29522671Swollman tsleep(&i_want_my_mcl, PVM, "mclalloc", 0); 29622671Swollman 29722671Swollman for (; i_want_my_mcl; i_want_my_mcl--) { 29832036Sbde if (m_clalloc(1, M_WAIT) == 0) 29922671Swollman printf("m_clalloc failed even in process context!\n"); 30022671Swollman } 30122671Swollman } 30222671Swollman} 30322671Swollman 30422671Swollmanstatic struct proc *mclallocproc; 30522671Swollmanstatic struct kproc_desc mclalloc_kp = { 30622671Swollman "mclalloc", 30722671Swollman kproc_mclalloc, 30822671Swollman &mclallocproc 30922671Swollman}; 31048391SpeterSYSINIT(mclallocproc, SI_SUB_KTHREAD_UPDATE, SI_ORDER_ANY, kproc_start, 31122671Swollman &mclalloc_kp); 31222671Swollman#endif 31322671Swollman 31415689Swollman/* 3151541Srgrimes * Allocate some number of mbuf clusters 3161541Srgrimes * and place on cluster free list. 3171541Srgrimes * Must be called at splimp. 3181541Srgrimes */ 3191541Srgrimes/* ARGSUSED */ 3201549Srgrimesint 32132036Sbdem_clalloc(ncl, how) 3221541Srgrimes register int ncl; 32332036Sbde int how; 3241541Srgrimes{ 3251541Srgrimes register caddr_t p; 3261541Srgrimes register int i; 3271541Srgrimes int npg; 3281541Srgrimes 3297066Sdg /* 33055171Smsmith * If we've hit the mcluster number limit, stop allocating from 33155171Smsmith * mb_map, (or trying to) in order to avoid dipping into the section 33255171Smsmith * of mb_map which we've "reserved" for mbufs. 33355171Smsmith */ 33455171Smsmith if ((ncl + mbstat.m_clusters) > nmbclusters) { 33555171Smsmith mbstat.m_drops++; 33655171Smsmith return (0); 33755171Smsmith } 33855171Smsmith 33955171Smsmith /* 3407066Sdg * Once we run out of map space, it will be impossible 3417066Sdg * to get any more (nothing is ever freed back to the 34254478Sgreen * map). From this point on, we solely rely on freed 34354478Sgreen * mclusters. 3447066Sdg */ 34522899Swollman if (mb_map_full) { 34622899Swollman mbstat.m_drops++; 3477066Sdg return (0); 34822899Swollman } 3497066Sdg 35022671Swollman#if MCLBYTES > PAGE_SIZE 35132036Sbde if (how != M_WAIT) { 35222671Swollman i_want_my_mcl += ncl; 35322671Swollman wakeup(&i_want_my_mcl); 35422899Swollman mbstat.m_wait++; 35522671Swollman p = 0; 35622671Swollman } else { 35722671Swollman p = contigmalloc1(MCLBYTES * ncl, M_DEVBUF, M_WAITOK, 0ul, 35822671Swollman ~0ul, PAGE_SIZE, 0, mb_map); 35922671Swollman } 36022671Swollman#else 36115543Sphk npg = ncl; 36221737Sdg p = (caddr_t)kmem_malloc(mb_map, ctob(npg), 36332036Sbde how != M_WAIT ? M_NOWAIT : M_WAITOK); 36422671Swollman ncl = ncl * PAGE_SIZE / MCLBYTES; 36522671Swollman#endif 3667066Sdg /* 36732036Sbde * Either the map is now full, or `how' is M_NOWAIT and there 3687066Sdg * are no pages left. 3697066Sdg */ 37022899Swollman if (p == NULL) { 37122899Swollman mbstat.m_drops++; 3721541Srgrimes return (0); 37322899Swollman } 3747066Sdg 3751541Srgrimes for (i = 0; i < ncl; i++) { 3761541Srgrimes ((union mcluster *)p)->mcl_next = mclfree; 3771541Srgrimes mclfree = (union mcluster *)p; 3781541Srgrimes p += MCLBYTES; 3791541Srgrimes mbstat.m_clfree++; 3801541Srgrimes } 3811541Srgrimes mbstat.m_clusters += ncl; 3821541Srgrimes return (1); 3831541Srgrimes} 3841541Srgrimes 3851541Srgrimes/* 38654478Sgreen * Once the mb_map submap has been exhausted and the allocation is called with 38754478Sgreen * M_WAIT, we rely on the mclfree union pointers. If nothing is free, we will 38854478Sgreen * sleep for a designated amount of time (mbuf_wait) or until we're woken up 38954478Sgreen * due to sudden mcluster availability. 39054478Sgreen */ 39154478Sgreencaddr_t 39254478Sgreenm_clalloc_wait(void) 39354478Sgreen{ 39454478Sgreen caddr_t p; 39554478Sgreen int s; 39654478Sgreen 39754478Sgreen#ifdef __i386__ 39854478Sgreen /* If in interrupt context, and INVARIANTS, maintain sanity and die. */ 39954478Sgreen KASSERT(intr_nesting_level == 0, ("CLALLOC: CANNOT WAIT IN INTERRUPT")); 40054478Sgreen#endif 40154478Sgreen 40254478Sgreen /* Sleep until something's available or until we expire. */ 40354478Sgreen m_clalloc_wid++; 40454478Sgreen if ((tsleep(&m_clalloc_wid, PVM, "mclalc", mbuf_wait)) == EWOULDBLOCK) 40554478Sgreen m_clalloc_wid--; 40654478Sgreen 40754478Sgreen /* 40854478Sgreen * Now that we (think) that we've got something, we will redo and 40954478Sgreen * MGET, but avoid getting into another instance of m_clalloc_wait() 41054478Sgreen */ 41154478Sgreen p = NULL; 41264837Sdwmalone _MCLALLOC(p, M_DONTWAIT); 41354478Sgreen 41454478Sgreen s = splimp(); 41554478Sgreen if (p != NULL) { /* We waited and got something... */ 41654478Sgreen mbstat.m_wait++; 41754478Sgreen /* Wake up another if we have more free. */ 41854478Sgreen if (mclfree != NULL) 41954787Sgreen MCLWAKEUP(); 42054478Sgreen } 42154478Sgreen 42254478Sgreen splx(s); 42354478Sgreen return (p); 42454478Sgreen} 42554478Sgreen 42654478Sgreen/* 42745615Sdes * When MGET fails, ask protocols to free space when short of memory, 4281541Srgrimes * then re-attempt to allocate an mbuf. 4291541Srgrimes */ 4301541Srgrimesstruct mbuf * 4311541Srgrimesm_retry(i, t) 4321541Srgrimes int i, t; 4331541Srgrimes{ 4341541Srgrimes register struct mbuf *m; 4351541Srgrimes 43637878Sdg /* 43737878Sdg * Must only do the reclaim if not in an interrupt context. 43837878Sdg */ 43954478Sgreen if (i == M_WAIT) { 44054478Sgreen#ifdef __i386__ 44154478Sgreen KASSERT(intr_nesting_level == 0, 44254478Sgreen ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 44354478Sgreen#endif 44437878Sdg m_reclaim(); 44554478Sgreen } 44654478Sgreen 44754478Sgreen /* 44854478Sgreen * Both m_mballoc_wait and m_retry must be nulled because 44954478Sgreen * when the MGET macro is run from here, we deffinately do _not_ 45054478Sgreen * want to enter an instance of m_mballoc_wait() or m_retry() (again!) 45154478Sgreen */ 45254478Sgreen#define m_mballoc_wait(caller,type) (struct mbuf *)0 4531541Srgrimes#define m_retry(i, t) (struct mbuf *)0 4541541Srgrimes MGET(m, i, t); 4551541Srgrimes#undef m_retry 45654478Sgreen#undef m_mballoc_wait 45754478Sgreen 45854478Sgreen if (m != NULL) 4596669Sdg mbstat.m_wait++; 46054478Sgreen else 46154478Sgreen mbstat.m_drops++; 46254478Sgreen 4631541Srgrimes return (m); 4641541Srgrimes} 4651541Srgrimes 4661541Srgrimes/* 4671541Srgrimes * As above; retry an MGETHDR. 4681541Srgrimes */ 4691541Srgrimesstruct mbuf * 4701541Srgrimesm_retryhdr(i, t) 4711541Srgrimes int i, t; 4721541Srgrimes{ 4731541Srgrimes register struct mbuf *m; 4741541Srgrimes 47537878Sdg /* 47637878Sdg * Must only do the reclaim if not in an interrupt context. 47737878Sdg */ 47854478Sgreen if (i == M_WAIT) { 47954478Sgreen#ifdef __i386__ 48054478Sgreen KASSERT(intr_nesting_level == 0, 48154478Sgreen ("MBALLOC: CANNOT WAIT IN INTERRUPT")); 48254478Sgreen#endif 48337878Sdg m_reclaim(); 48454478Sgreen } 48554478Sgreen 48654478Sgreen#define m_mballoc_wait(caller,type) (struct mbuf *)0 4871541Srgrimes#define m_retryhdr(i, t) (struct mbuf *)0 4881541Srgrimes MGETHDR(m, i, t); 4891541Srgrimes#undef m_retryhdr 49054478Sgreen#undef m_mballoc_wait 49154478Sgreen 49254478Sgreen if (m != NULL) 4936669Sdg mbstat.m_wait++; 49454478Sgreen else 49554478Sgreen mbstat.m_drops++; 49654478Sgreen 4971541Srgrimes return (m); 4981541Srgrimes} 4991541Srgrimes 50012819Sphkstatic void 5011541Srgrimesm_reclaim() 5021541Srgrimes{ 5031541Srgrimes register struct domain *dp; 5041541Srgrimes register struct protosw *pr; 5051541Srgrimes int s = splimp(); 5061541Srgrimes 5071541Srgrimes for (dp = domains; dp; dp = dp->dom_next) 5081541Srgrimes for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++) 5091541Srgrimes if (pr->pr_drain) 5101541Srgrimes (*pr->pr_drain)(); 5111541Srgrimes splx(s); 5121541Srgrimes mbstat.m_drain++; 5131541Srgrimes} 5141541Srgrimes 5151541Srgrimes/* 5161541Srgrimes * Space allocation routines. 5171541Srgrimes * These are also available as macros 5181541Srgrimes * for critical paths. 5191541Srgrimes */ 5201541Srgrimesstruct mbuf * 52132036Sbdem_get(how, type) 52232036Sbde int how, type; 5231541Srgrimes{ 5241541Srgrimes register struct mbuf *m; 5251541Srgrimes 52632036Sbde MGET(m, how, type); 5271541Srgrimes return (m); 5281541Srgrimes} 5291541Srgrimes 5301541Srgrimesstruct mbuf * 53132036Sbdem_gethdr(how, type) 53232036Sbde int how, type; 5331541Srgrimes{ 5341541Srgrimes register struct mbuf *m; 5351541Srgrimes 53632036Sbde MGETHDR(m, how, type); 5371541Srgrimes return (m); 5381541Srgrimes} 5391541Srgrimes 5401541Srgrimesstruct mbuf * 54132036Sbdem_getclr(how, type) 54232036Sbde int how, type; 5431541Srgrimes{ 5441541Srgrimes register struct mbuf *m; 5451541Srgrimes 54632036Sbde MGET(m, how, type); 5471541Srgrimes if (m == 0) 5481541Srgrimes return (0); 5491541Srgrimes bzero(mtod(m, caddr_t), MLEN); 5501541Srgrimes return (m); 5511541Srgrimes} 5521541Srgrimes 5531541Srgrimesstruct mbuf * 5541541Srgrimesm_free(m) 5551541Srgrimes struct mbuf *m; 5561541Srgrimes{ 5571541Srgrimes register struct mbuf *n; 5581541Srgrimes 5591541Srgrimes MFREE(m, n); 5601541Srgrimes return (n); 5611541Srgrimes} 5621541Srgrimes 5631541Srgrimesvoid 5641541Srgrimesm_freem(m) 5651541Srgrimes register struct mbuf *m; 5661541Srgrimes{ 5671541Srgrimes register struct mbuf *n; 5681541Srgrimes 5691541Srgrimes if (m == NULL) 5701541Srgrimes return; 5711541Srgrimes do { 57262587Sitojun /* 57362587Sitojun * we do need to check non-first mbuf, since some of existing 57462587Sitojun * code does not call M_PREPEND properly. 57562587Sitojun * (example: call to bpf_mtap from drivers) 57662587Sitojun */ 57762587Sitojun if ((m->m_flags & M_PKTHDR) != 0 && m->m_pkthdr.aux) { 57862587Sitojun m_freem(m->m_pkthdr.aux); 57962587Sitojun m->m_pkthdr.aux = NULL; 58062587Sitojun } 5811541Srgrimes MFREE(m, n); 5823308Sphk m = n; 5833308Sphk } while (m); 5841541Srgrimes} 5851541Srgrimes 5861541Srgrimes/* 5871541Srgrimes * Mbuffer utility routines. 5881541Srgrimes */ 5891541Srgrimes 5901541Srgrimes/* 5911541Srgrimes * Lesser-used path for M_PREPEND: 5921541Srgrimes * allocate new mbuf to prepend to chain, 5931541Srgrimes * copy junk along. 5941541Srgrimes */ 5951541Srgrimesstruct mbuf * 5961541Srgrimesm_prepend(m, len, how) 5971541Srgrimes register struct mbuf *m; 5981541Srgrimes int len, how; 5991541Srgrimes{ 6001541Srgrimes struct mbuf *mn; 6011541Srgrimes 6021541Srgrimes MGET(mn, how, m->m_type); 6031541Srgrimes if (mn == (struct mbuf *)NULL) { 6041541Srgrimes m_freem(m); 6051541Srgrimes return ((struct mbuf *)NULL); 6061541Srgrimes } 6071541Srgrimes if (m->m_flags & M_PKTHDR) { 6081541Srgrimes M_COPY_PKTHDR(mn, m); 6091541Srgrimes m->m_flags &= ~M_PKTHDR; 6101541Srgrimes } 6111541Srgrimes mn->m_next = m; 6121541Srgrimes m = mn; 6131541Srgrimes if (len < MHLEN) 6141541Srgrimes MH_ALIGN(m, len); 6151541Srgrimes m->m_len = len; 6161541Srgrimes return (m); 6171541Srgrimes} 6181541Srgrimes 6191541Srgrimes/* 6201541Srgrimes * Make a copy of an mbuf chain starting "off0" bytes from the beginning, 6211541Srgrimes * continuing for "len" bytes. If len is M_COPYALL, copy to end of mbuf. 6221541Srgrimes * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller. 62354002Sarchie * Note that the copy is read-only, because clusters are not copied, 62454002Sarchie * only their reference counts are incremented. 6251541Srgrimes */ 62623081Swollman#define MCFail (mbstat.m_mcfail) 6271541Srgrimes 6281541Srgrimesstruct mbuf * 6291541Srgrimesm_copym(m, off0, len, wait) 6301541Srgrimes register struct mbuf *m; 6311541Srgrimes int off0, wait; 6321541Srgrimes register int len; 6331541Srgrimes{ 6341541Srgrimes register struct mbuf *n, **np; 6351541Srgrimes register int off = off0; 6361541Srgrimes struct mbuf *top; 6371541Srgrimes int copyhdr = 0; 6381541Srgrimes 63952201Salfred KASSERT(off >= 0, ("m_copym, negative off %d", off)); 64052201Salfred KASSERT(len >= 0, ("m_copym, negative len %d", len)); 6411541Srgrimes if (off == 0 && m->m_flags & M_PKTHDR) 6421541Srgrimes copyhdr = 1; 6431541Srgrimes while (off > 0) { 64452201Salfred KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain")); 6451541Srgrimes if (off < m->m_len) 6461541Srgrimes break; 6471541Srgrimes off -= m->m_len; 6481541Srgrimes m = m->m_next; 6491541Srgrimes } 6501541Srgrimes np = ⊤ 6511541Srgrimes top = 0; 6521541Srgrimes while (len > 0) { 6531541Srgrimes if (m == 0) { 65452201Salfred KASSERT(len == M_COPYALL, 65552201Salfred ("m_copym, length > size of mbuf chain")); 6561541Srgrimes break; 6571541Srgrimes } 6581541Srgrimes MGET(n, wait, m->m_type); 6591541Srgrimes *np = n; 6601541Srgrimes if (n == 0) 6611541Srgrimes goto nospace; 6621541Srgrimes if (copyhdr) { 6631541Srgrimes M_COPY_PKTHDR(n, m); 6641541Srgrimes if (len == M_COPYALL) 6651541Srgrimes n->m_pkthdr.len -= off0; 6661541Srgrimes else 6671541Srgrimes n->m_pkthdr.len = len; 6681541Srgrimes copyhdr = 0; 6691541Srgrimes } 6701541Srgrimes n->m_len = min(len, m->m_len - off); 6711541Srgrimes if (m->m_flags & M_EXT) { 6721541Srgrimes n->m_data = m->m_data + off; 6731541Srgrimes n->m_ext = m->m_ext; 6741541Srgrimes n->m_flags |= M_EXT; 67564837Sdwmalone MEXT_ADD_REF(m); 6761541Srgrimes } else 6771541Srgrimes bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), 6781541Srgrimes (unsigned)n->m_len); 6791541Srgrimes if (len != M_COPYALL) 6801541Srgrimes len -= n->m_len; 6811541Srgrimes off = 0; 6821541Srgrimes m = m->m_next; 6831541Srgrimes np = &n->m_next; 6841541Srgrimes } 6851541Srgrimes if (top == 0) 6861541Srgrimes MCFail++; 6871541Srgrimes return (top); 6881541Srgrimesnospace: 6891541Srgrimes m_freem(top); 6901541Srgrimes MCFail++; 6911541Srgrimes return (0); 6921541Srgrimes} 6931541Srgrimes 6941541Srgrimes/* 69515689Swollman * Copy an entire packet, including header (which must be present). 69615689Swollman * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'. 69754002Sarchie * Note that the copy is read-only, because clusters are not copied, 69854002Sarchie * only their reference counts are incremented. 69915689Swollman */ 70015689Swollmanstruct mbuf * 70115689Swollmanm_copypacket(m, how) 70215689Swollman struct mbuf *m; 70315689Swollman int how; 70415689Swollman{ 70515689Swollman struct mbuf *top, *n, *o; 70615689Swollman 70715689Swollman MGET(n, how, m->m_type); 70815689Swollman top = n; 70915689Swollman if (!n) 71015689Swollman goto nospace; 71115689Swollman 71215689Swollman M_COPY_PKTHDR(n, m); 71315689Swollman n->m_len = m->m_len; 71415689Swollman if (m->m_flags & M_EXT) { 71515689Swollman n->m_data = m->m_data; 71615689Swollman n->m_ext = m->m_ext; 71715689Swollman n->m_flags |= M_EXT; 71864837Sdwmalone MEXT_ADD_REF(m); 71915689Swollman } else { 72015689Swollman bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 72115689Swollman } 72215689Swollman 72315689Swollman m = m->m_next; 72415689Swollman while (m) { 72515689Swollman MGET(o, how, m->m_type); 72615689Swollman if (!o) 72715689Swollman goto nospace; 72815689Swollman 72915689Swollman n->m_next = o; 73015689Swollman n = n->m_next; 73115689Swollman 73215689Swollman n->m_len = m->m_len; 73315689Swollman if (m->m_flags & M_EXT) { 73415689Swollman n->m_data = m->m_data; 73515689Swollman n->m_ext = m->m_ext; 73615689Swollman n->m_flags |= M_EXT; 73764837Sdwmalone MEXT_ADD_REF(m); 73815689Swollman } else { 73915689Swollman bcopy(mtod(m, char *), mtod(n, char *), n->m_len); 74015689Swollman } 74115689Swollman 74215689Swollman m = m->m_next; 74315689Swollman } 74415689Swollman return top; 74515689Swollmannospace: 74615689Swollman m_freem(top); 74715689Swollman MCFail++; 74815689Swollman return 0; 74915689Swollman} 75015689Swollman 75115689Swollman/* 7521541Srgrimes * Copy data from an mbuf chain starting "off" bytes from the beginning, 7531541Srgrimes * continuing for "len" bytes, into the indicated buffer. 7541541Srgrimes */ 7551549Srgrimesvoid 7561541Srgrimesm_copydata(m, off, len, cp) 7571541Srgrimes register struct mbuf *m; 7581541Srgrimes register int off; 7591541Srgrimes register int len; 7601541Srgrimes caddr_t cp; 7611541Srgrimes{ 7621541Srgrimes register unsigned count; 7631541Srgrimes 76452201Salfred KASSERT(off >= 0, ("m_copydata, negative off %d", off)); 76552201Salfred KASSERT(len >= 0, ("m_copydata, negative len %d", len)); 7661541Srgrimes while (off > 0) { 76752201Salfred KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain")); 7681541Srgrimes if (off < m->m_len) 7691541Srgrimes break; 7701541Srgrimes off -= m->m_len; 7711541Srgrimes m = m->m_next; 7721541Srgrimes } 7731541Srgrimes while (len > 0) { 77452201Salfred KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain")); 7751541Srgrimes count = min(m->m_len - off, len); 7761541Srgrimes bcopy(mtod(m, caddr_t) + off, cp, count); 7771541Srgrimes len -= count; 7781541Srgrimes cp += count; 7791541Srgrimes off = 0; 7801541Srgrimes m = m->m_next; 7811541Srgrimes } 7821541Srgrimes} 7831541Srgrimes 7841541Srgrimes/* 78554002Sarchie * Copy a packet header mbuf chain into a completely new chain, including 78654002Sarchie * copying any mbuf clusters. Use this instead of m_copypacket() when 78754002Sarchie * you need a writable copy of an mbuf chain. 78854002Sarchie */ 78954002Sarchiestruct mbuf * 79054002Sarchiem_dup(m, how) 79154002Sarchie struct mbuf *m; 79254002Sarchie int how; 79354002Sarchie{ 79454002Sarchie struct mbuf **p, *top = NULL; 79554002Sarchie int remain, moff, nsize; 79654002Sarchie 79754002Sarchie /* Sanity check */ 79854002Sarchie if (m == NULL) 79954002Sarchie return (0); 80054002Sarchie KASSERT((m->m_flags & M_PKTHDR) != 0, ("%s: !PKTHDR", __FUNCTION__)); 80154002Sarchie 80254002Sarchie /* While there's more data, get a new mbuf, tack it on, and fill it */ 80354002Sarchie remain = m->m_pkthdr.len; 80454002Sarchie moff = 0; 80554002Sarchie p = ⊤ 80654002Sarchie while (remain > 0 || top == NULL) { /* allow m->m_pkthdr.len == 0 */ 80754002Sarchie struct mbuf *n; 80854002Sarchie 80954002Sarchie /* Get the next new mbuf */ 81054002Sarchie MGET(n, how, m->m_type); 81154002Sarchie if (n == NULL) 81254002Sarchie goto nospace; 81354002Sarchie if (top == NULL) { /* first one, must be PKTHDR */ 81454002Sarchie M_COPY_PKTHDR(n, m); 81554002Sarchie nsize = MHLEN; 81654002Sarchie } else /* not the first one */ 81754002Sarchie nsize = MLEN; 81854002Sarchie if (remain >= MINCLSIZE) { 81954002Sarchie MCLGET(n, how); 82054002Sarchie if ((n->m_flags & M_EXT) == 0) { 82154002Sarchie (void)m_free(n); 82254002Sarchie goto nospace; 82354002Sarchie } 82454002Sarchie nsize = MCLBYTES; 82554002Sarchie } 82654002Sarchie n->m_len = 0; 82754002Sarchie 82854002Sarchie /* Link it into the new chain */ 82954002Sarchie *p = n; 83054002Sarchie p = &n->m_next; 83154002Sarchie 83254002Sarchie /* Copy data from original mbuf(s) into new mbuf */ 83354002Sarchie while (n->m_len < nsize && m != NULL) { 83454002Sarchie int chunk = min(nsize - n->m_len, m->m_len - moff); 83554002Sarchie 83654002Sarchie bcopy(m->m_data + moff, n->m_data + n->m_len, chunk); 83754002Sarchie moff += chunk; 83854002Sarchie n->m_len += chunk; 83954002Sarchie remain -= chunk; 84054002Sarchie if (moff == m->m_len) { 84154002Sarchie m = m->m_next; 84254002Sarchie moff = 0; 84354002Sarchie } 84454002Sarchie } 84554002Sarchie 84654002Sarchie /* Check correct total mbuf length */ 84754002Sarchie KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL), 84854002Sarchie ("%s: bogus m_pkthdr.len", __FUNCTION__)); 84954002Sarchie } 85054002Sarchie return (top); 85154002Sarchie 85254002Sarchienospace: 85354002Sarchie m_freem(top); 85454002Sarchie MCFail++; 85554002Sarchie return (0); 85654002Sarchie} 85754002Sarchie 85854002Sarchie/* 8591541Srgrimes * Concatenate mbuf chain n to m. 8601541Srgrimes * Both chains must be of the same type (e.g. MT_DATA). 8611541Srgrimes * Any m_pkthdr is not updated. 8621541Srgrimes */ 8631549Srgrimesvoid 8641541Srgrimesm_cat(m, n) 8651541Srgrimes register struct mbuf *m, *n; 8661541Srgrimes{ 8671541Srgrimes while (m->m_next) 8681541Srgrimes m = m->m_next; 8691541Srgrimes while (n) { 8701541Srgrimes if (m->m_flags & M_EXT || 8711541Srgrimes m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) { 8721541Srgrimes /* just join the two chains */ 8731541Srgrimes m->m_next = n; 8741541Srgrimes return; 8751541Srgrimes } 8761541Srgrimes /* splat the data from one into the other */ 8771541Srgrimes bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 8781541Srgrimes (u_int)n->m_len); 8791541Srgrimes m->m_len += n->m_len; 8801541Srgrimes n = m_free(n); 8811541Srgrimes } 8821541Srgrimes} 8831541Srgrimes 8841549Srgrimesvoid 8851541Srgrimesm_adj(mp, req_len) 8861541Srgrimes struct mbuf *mp; 8871541Srgrimes int req_len; 8881541Srgrimes{ 8891541Srgrimes register int len = req_len; 8901541Srgrimes register struct mbuf *m; 89133678Sbde register int count; 8921541Srgrimes 8931541Srgrimes if ((m = mp) == NULL) 8941541Srgrimes return; 8951541Srgrimes if (len >= 0) { 8961541Srgrimes /* 8971541Srgrimes * Trim from head. 8981541Srgrimes */ 8991541Srgrimes while (m != NULL && len > 0) { 9001541Srgrimes if (m->m_len <= len) { 9011541Srgrimes len -= m->m_len; 9021541Srgrimes m->m_len = 0; 9031541Srgrimes m = m->m_next; 9041541Srgrimes } else { 9051541Srgrimes m->m_len -= len; 9061541Srgrimes m->m_data += len; 9071541Srgrimes len = 0; 9081541Srgrimes } 9091541Srgrimes } 9101541Srgrimes m = mp; 9111541Srgrimes if (mp->m_flags & M_PKTHDR) 9121541Srgrimes m->m_pkthdr.len -= (req_len - len); 9131541Srgrimes } else { 9141541Srgrimes /* 9151541Srgrimes * Trim from tail. Scan the mbuf chain, 9161541Srgrimes * calculating its length and finding the last mbuf. 9171541Srgrimes * If the adjustment only affects this mbuf, then just 9181541Srgrimes * adjust and return. Otherwise, rescan and truncate 9191541Srgrimes * after the remaining size. 9201541Srgrimes */ 9211541Srgrimes len = -len; 9221541Srgrimes count = 0; 9231541Srgrimes for (;;) { 9241541Srgrimes count += m->m_len; 9251541Srgrimes if (m->m_next == (struct mbuf *)0) 9261541Srgrimes break; 9271541Srgrimes m = m->m_next; 9281541Srgrimes } 9291541Srgrimes if (m->m_len >= len) { 9301541Srgrimes m->m_len -= len; 9311541Srgrimes if (mp->m_flags & M_PKTHDR) 9321541Srgrimes mp->m_pkthdr.len -= len; 9331541Srgrimes return; 9341541Srgrimes } 9351541Srgrimes count -= len; 9361541Srgrimes if (count < 0) 9371541Srgrimes count = 0; 9381541Srgrimes /* 9391541Srgrimes * Correct length for chain is "count". 9401541Srgrimes * Find the mbuf with last data, adjust its length, 9411541Srgrimes * and toss data from remaining mbufs on chain. 9421541Srgrimes */ 9431541Srgrimes m = mp; 9441541Srgrimes if (m->m_flags & M_PKTHDR) 9451541Srgrimes m->m_pkthdr.len = count; 9461541Srgrimes for (; m; m = m->m_next) { 9471541Srgrimes if (m->m_len >= count) { 9481541Srgrimes m->m_len = count; 9491541Srgrimes break; 9501541Srgrimes } 9511541Srgrimes count -= m->m_len; 9521541Srgrimes } 9533308Sphk while (m->m_next) 9543308Sphk (m = m->m_next) ->m_len = 0; 9551541Srgrimes } 9561541Srgrimes} 9571541Srgrimes 9581541Srgrimes/* 9591541Srgrimes * Rearange an mbuf chain so that len bytes are contiguous 9601541Srgrimes * and in the data area of an mbuf (so that mtod and dtom 9611541Srgrimes * will work for a structure of size len). Returns the resulting 9621541Srgrimes * mbuf chain on success, frees it and returns null on failure. 9631541Srgrimes * If there is room, it will add up to max_protohdr-len extra bytes to the 9641541Srgrimes * contiguous region in an attempt to avoid being called next time. 9651541Srgrimes */ 96623081Swollman#define MPFail (mbstat.m_mpfail) 9671541Srgrimes 9681541Srgrimesstruct mbuf * 9691541Srgrimesm_pullup(n, len) 9701541Srgrimes register struct mbuf *n; 9711541Srgrimes int len; 9721541Srgrimes{ 9731541Srgrimes register struct mbuf *m; 9741541Srgrimes register int count; 9751541Srgrimes int space; 9761541Srgrimes 9771541Srgrimes /* 9781541Srgrimes * If first mbuf has no cluster, and has room for len bytes 9791541Srgrimes * without shifting current data, pullup into it, 9801541Srgrimes * otherwise allocate a new mbuf to prepend to the chain. 9811541Srgrimes */ 9821541Srgrimes if ((n->m_flags & M_EXT) == 0 && 9831541Srgrimes n->m_data + len < &n->m_dat[MLEN] && n->m_next) { 9841541Srgrimes if (n->m_len >= len) 9851541Srgrimes return (n); 9861541Srgrimes m = n; 9871541Srgrimes n = n->m_next; 9881541Srgrimes len -= m->m_len; 9891541Srgrimes } else { 9901541Srgrimes if (len > MHLEN) 9911541Srgrimes goto bad; 9921541Srgrimes MGET(m, M_DONTWAIT, n->m_type); 9931541Srgrimes if (m == 0) 9941541Srgrimes goto bad; 9951541Srgrimes m->m_len = 0; 9961541Srgrimes if (n->m_flags & M_PKTHDR) { 9971541Srgrimes M_COPY_PKTHDR(m, n); 9981541Srgrimes n->m_flags &= ~M_PKTHDR; 9991541Srgrimes } 10001541Srgrimes } 10011541Srgrimes space = &m->m_dat[MLEN] - (m->m_data + m->m_len); 10021541Srgrimes do { 10031541Srgrimes count = min(min(max(len, max_protohdr), space), n->m_len); 10041541Srgrimes bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len, 10051541Srgrimes (unsigned)count); 10061541Srgrimes len -= count; 10071541Srgrimes m->m_len += count; 10081541Srgrimes n->m_len -= count; 10091541Srgrimes space -= count; 10101541Srgrimes if (n->m_len) 10111541Srgrimes n->m_data += count; 10121541Srgrimes else 10131541Srgrimes n = m_free(n); 10141541Srgrimes } while (len > 0 && n); 10151541Srgrimes if (len > 0) { 10161541Srgrimes (void) m_free(m); 10171541Srgrimes goto bad; 10181541Srgrimes } 10191541Srgrimes m->m_next = n; 10201541Srgrimes return (m); 10211541Srgrimesbad: 10221541Srgrimes m_freem(n); 10231541Srgrimes MPFail++; 10241541Srgrimes return (0); 10251541Srgrimes} 10261541Srgrimes 10271541Srgrimes/* 10281541Srgrimes * Partition an mbuf chain in two pieces, returning the tail -- 10291541Srgrimes * all but the first len0 bytes. In case of failure, it returns NULL and 10301541Srgrimes * attempts to restore the chain to its original state. 10311541Srgrimes */ 10321541Srgrimesstruct mbuf * 10331541Srgrimesm_split(m0, len0, wait) 10341541Srgrimes register struct mbuf *m0; 10351541Srgrimes int len0, wait; 10361541Srgrimes{ 10371541Srgrimes register struct mbuf *m, *n; 10381541Srgrimes unsigned len = len0, remain; 10391541Srgrimes 10401541Srgrimes for (m = m0; m && len > m->m_len; m = m->m_next) 10411541Srgrimes len -= m->m_len; 10421541Srgrimes if (m == 0) 10431541Srgrimes return (0); 10441541Srgrimes remain = m->m_len - len; 10451541Srgrimes if (m0->m_flags & M_PKTHDR) { 10461541Srgrimes MGETHDR(n, wait, m0->m_type); 10471541Srgrimes if (n == 0) 10481541Srgrimes return (0); 10491541Srgrimes n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif; 10501541Srgrimes n->m_pkthdr.len = m0->m_pkthdr.len - len0; 10511541Srgrimes m0->m_pkthdr.len = len0; 10521541Srgrimes if (m->m_flags & M_EXT) 10531541Srgrimes goto extpacket; 10541541Srgrimes if (remain > MHLEN) { 10551541Srgrimes /* m can't be the lead packet */ 10561541Srgrimes MH_ALIGN(n, 0); 10571541Srgrimes n->m_next = m_split(m, len, wait); 10581541Srgrimes if (n->m_next == 0) { 10591541Srgrimes (void) m_free(n); 10601541Srgrimes return (0); 10611541Srgrimes } else 10621541Srgrimes return (n); 10631541Srgrimes } else 10641541Srgrimes MH_ALIGN(n, remain); 10651541Srgrimes } else if (remain == 0) { 10661541Srgrimes n = m->m_next; 10671541Srgrimes m->m_next = 0; 10681541Srgrimes return (n); 10691541Srgrimes } else { 10701541Srgrimes MGET(n, wait, m->m_type); 10711541Srgrimes if (n == 0) 10721541Srgrimes return (0); 10731541Srgrimes M_ALIGN(n, remain); 10741541Srgrimes } 10751541Srgrimesextpacket: 10761541Srgrimes if (m->m_flags & M_EXT) { 10771541Srgrimes n->m_flags |= M_EXT; 10781541Srgrimes n->m_ext = m->m_ext; 107964837Sdwmalone MEXT_ADD_REF(m); 10801541Srgrimes m->m_ext.ext_size = 0; /* For Accounting XXXXXX danger */ 10811541Srgrimes n->m_data = m->m_data + len; 10821541Srgrimes } else { 10831541Srgrimes bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain); 10841541Srgrimes } 10851541Srgrimes n->m_len = remain; 10861541Srgrimes m->m_len = len; 10871541Srgrimes n->m_next = m->m_next; 10881541Srgrimes m->m_next = 0; 10891541Srgrimes return (n); 10901541Srgrimes} 10911541Srgrimes/* 10921541Srgrimes * Routine to copy from device local memory into mbufs. 10931541Srgrimes */ 10941541Srgrimesstruct mbuf * 10951541Srgrimesm_devget(buf, totlen, off0, ifp, copy) 10961541Srgrimes char *buf; 10971541Srgrimes int totlen, off0; 10981541Srgrimes struct ifnet *ifp; 109912577Sbde void (*copy) __P((char *from, caddr_t to, u_int len)); 11001541Srgrimes{ 11011541Srgrimes register struct mbuf *m; 11021541Srgrimes struct mbuf *top = 0, **mp = ⊤ 11031541Srgrimes register int off = off0, len; 11041541Srgrimes register char *cp; 11051541Srgrimes char *epkt; 11061541Srgrimes 11071541Srgrimes cp = buf; 11081541Srgrimes epkt = cp + totlen; 11091541Srgrimes if (off) { 11101541Srgrimes cp += off + 2 * sizeof(u_short); 11111541Srgrimes totlen -= 2 * sizeof(u_short); 11121541Srgrimes } 11131541Srgrimes MGETHDR(m, M_DONTWAIT, MT_DATA); 11141541Srgrimes if (m == 0) 11151541Srgrimes return (0); 11161541Srgrimes m->m_pkthdr.rcvif = ifp; 11171541Srgrimes m->m_pkthdr.len = totlen; 11181541Srgrimes m->m_len = MHLEN; 11191541Srgrimes 11201541Srgrimes while (totlen > 0) { 11211541Srgrimes if (top) { 11221541Srgrimes MGET(m, M_DONTWAIT, MT_DATA); 11231541Srgrimes if (m == 0) { 11241541Srgrimes m_freem(top); 11251541Srgrimes return (0); 11261541Srgrimes } 11271541Srgrimes m->m_len = MLEN; 11281541Srgrimes } 11291541Srgrimes len = min(totlen, epkt - cp); 11301541Srgrimes if (len >= MINCLSIZE) { 11311541Srgrimes MCLGET(m, M_DONTWAIT); 11321541Srgrimes if (m->m_flags & M_EXT) 11331541Srgrimes m->m_len = len = min(len, MCLBYTES); 11341541Srgrimes else 11351541Srgrimes len = m->m_len; 11361541Srgrimes } else { 11371541Srgrimes /* 11381541Srgrimes * Place initial small packet/header at end of mbuf. 11391541Srgrimes */ 11401541Srgrimes if (len < m->m_len) { 11411541Srgrimes if (top == 0 && len + max_linkhdr <= m->m_len) 11421541Srgrimes m->m_data += max_linkhdr; 11431541Srgrimes m->m_len = len; 11441541Srgrimes } else 11451541Srgrimes len = m->m_len; 11461541Srgrimes } 11471541Srgrimes if (copy) 11481541Srgrimes copy(cp, mtod(m, caddr_t), (unsigned)len); 11491541Srgrimes else 11501541Srgrimes bcopy(cp, mtod(m, caddr_t), (unsigned)len); 11511541Srgrimes cp += len; 11521541Srgrimes *mp = m; 11531541Srgrimes mp = &m->m_next; 11541541Srgrimes totlen -= len; 11551541Srgrimes if (cp == epkt) 11561541Srgrimes cp = buf; 11571541Srgrimes } 11581541Srgrimes return (top); 11591541Srgrimes} 11603352Sphk 11613352Sphk/* 11623352Sphk * Copy data from a buffer back into the indicated mbuf chain, 11633352Sphk * starting "off" bytes from the beginning, extending the mbuf 11643352Sphk * chain if necessary. 11653352Sphk */ 11663352Sphkvoid 11673352Sphkm_copyback(m0, off, len, cp) 11683352Sphk struct mbuf *m0; 11693352Sphk register int off; 11703352Sphk register int len; 11713352Sphk caddr_t cp; 11723352Sphk{ 11733352Sphk register int mlen; 11743352Sphk register struct mbuf *m = m0, *n; 11753352Sphk int totlen = 0; 11763352Sphk 11773352Sphk if (m0 == 0) 11783352Sphk return; 11793352Sphk while (off > (mlen = m->m_len)) { 11803352Sphk off -= mlen; 11813352Sphk totlen += mlen; 11823352Sphk if (m->m_next == 0) { 11833352Sphk n = m_getclr(M_DONTWAIT, m->m_type); 11843352Sphk if (n == 0) 11853352Sphk goto out; 11863352Sphk n->m_len = min(MLEN, len + off); 11873352Sphk m->m_next = n; 11883352Sphk } 11893352Sphk m = m->m_next; 11903352Sphk } 11913352Sphk while (len > 0) { 11923352Sphk mlen = min (m->m_len - off, len); 11933352Sphk bcopy(cp, off + mtod(m, caddr_t), (unsigned)mlen); 11943352Sphk cp += mlen; 11953352Sphk len -= mlen; 11963352Sphk mlen += off; 11973352Sphk off = 0; 11983352Sphk totlen += mlen; 11993352Sphk if (len == 0) 12003352Sphk break; 12013352Sphk if (m->m_next == 0) { 12023352Sphk n = m_get(M_DONTWAIT, m->m_type); 12033352Sphk if (n == 0) 12043352Sphk break; 12053352Sphk n->m_len = min(MLEN, len); 12063352Sphk m->m_next = n; 12073352Sphk } 12083352Sphk m = m->m_next; 12093352Sphk } 12103352Sphkout: if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen)) 12113352Sphk m->m_pkthdr.len = totlen; 12123352Sphk} 121352756Sphk 121452756Sphkvoid 121552756Sphkm_print(const struct mbuf *m) 121652756Sphk{ 121752756Sphk int len; 121854906Seivind const struct mbuf *m2; 121952756Sphk 122052756Sphk len = m->m_pkthdr.len; 122152756Sphk m2 = m; 122252756Sphk while (len) { 122352756Sphk printf("%p %*D\n", m2, m2->m_len, (u_char *)m2->m_data, "-"); 122452756Sphk len -= m2->m_len; 122552756Sphk m2 = m2->m_next; 122652756Sphk } 122752756Sphk return; 122852756Sphk} 1229