The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/uipc_mbuf.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1988, 1991, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * Redistribution and use in source and binary forms, with or without
    6  * modification, are permitted provided that the following conditions
    7  * are met:
    8  * 1. Redistributions of source code must retain the above copyright
    9  *    notice, this list of conditions and the following disclaimer.
   10  * 2. Redistributions in binary form must reproduce the above copyright
   11  *    notice, this list of conditions and the following disclaimer in the
   12  *    documentation and/or other materials provided with the distribution.
   13  * 4. Neither the name of the University nor the names of its contributors
   14  *    may be used to endorse or promote products derived from this software
   15  *    without specific prior written permission.
   16  *
   17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   27  * SUCH DAMAGE.
   28  *
   29  *      @(#)uipc_mbuf.c 8.2 (Berkeley) 1/4/94
   30  */
   31 
   32 #include <sys/cdefs.h>
   33 __FBSDID("$FreeBSD: releng/8.3/sys/kern/uipc_mbuf.c 231759 2012-02-15 14:23:01Z ken $");
   34 
   35 #include "opt_param.h"
   36 #include "opt_mbuf_stress_test.h"
   37 #include "opt_mbuf_profiling.h"
   38 
   39 #include <sys/param.h>
   40 #include <sys/systm.h>
   41 #include <sys/kernel.h>
   42 #include <sys/limits.h>
   43 #include <sys/lock.h>
   44 #include <sys/malloc.h>
   45 #include <sys/mbuf.h>
   46 #include <sys/sysctl.h>
   47 #include <sys/domain.h>
   48 #include <sys/protosw.h>
   49 #include <sys/uio.h>
   50 
   51 int     max_linkhdr;
   52 int     max_protohdr;
   53 int     max_hdr;
   54 int     max_datalen;
   55 #ifdef MBUF_STRESS_TEST
   56 int     m_defragpackets;
   57 int     m_defragbytes;
   58 int     m_defraguseless;
   59 int     m_defragfailure;
   60 int     m_defragrandomfailures;
   61 #endif
   62 
   63 /*
   64  * sysctl(8) exported objects
   65  */
   66 SYSCTL_INT(_kern_ipc, KIPC_MAX_LINKHDR, max_linkhdr, CTLFLAG_RD,
   67            &max_linkhdr, 0, "Size of largest link layer header");
   68 SYSCTL_INT(_kern_ipc, KIPC_MAX_PROTOHDR, max_protohdr, CTLFLAG_RD,
   69            &max_protohdr, 0, "Size of largest protocol layer header");
   70 SYSCTL_INT(_kern_ipc, KIPC_MAX_HDR, max_hdr, CTLFLAG_RD,
   71            &max_hdr, 0, "Size of largest link plus protocol header");
   72 SYSCTL_INT(_kern_ipc, KIPC_MAX_DATALEN, max_datalen, CTLFLAG_RD,
   73            &max_datalen, 0, "Minimum space left in mbuf after max_hdr");
   74 #ifdef MBUF_STRESS_TEST
   75 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragpackets, CTLFLAG_RD,
   76            &m_defragpackets, 0, "");
   77 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragbytes, CTLFLAG_RD,
   78            &m_defragbytes, 0, "");
   79 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defraguseless, CTLFLAG_RD,
   80            &m_defraguseless, 0, "");
   81 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragfailure, CTLFLAG_RD,
   82            &m_defragfailure, 0, "");
   83 SYSCTL_INT(_kern_ipc, OID_AUTO, m_defragrandomfailures, CTLFLAG_RW,
   84            &m_defragrandomfailures, 0, "");
   85 #endif
   86 
   87 /*
   88  * Allocate a given length worth of mbufs and/or clusters (whatever fits
   89  * best) and return a pointer to the top of the allocated chain.  If an
   90  * existing mbuf chain is provided, then we will append the new chain
   91  * to the existing one but still return the top of the newly allocated
   92  * chain.
   93  */
   94 struct mbuf *
   95 m_getm2(struct mbuf *m, int len, int how, short type, int flags)
   96 {
   97         struct mbuf *mb, *nm = NULL, *mtail = NULL;
   98 
   99         KASSERT(len >= 0, ("%s: len is < 0", __func__));
  100 
  101         /* Validate flags. */
  102         flags &= (M_PKTHDR | M_EOR);
  103 
  104         /* Packet header mbuf must be first in chain. */
  105         if ((flags & M_PKTHDR) && m != NULL)
  106                 flags &= ~M_PKTHDR;
  107 
  108         /* Loop and append maximum sized mbufs to the chain tail. */
  109         while (len > 0) {
  110                 if (len > MCLBYTES)
  111                         mb = m_getjcl(how, type, (flags & M_PKTHDR),
  112                             MJUMPAGESIZE);
  113                 else if (len >= MINCLSIZE)
  114                         mb = m_getcl(how, type, (flags & M_PKTHDR));
  115                 else if (flags & M_PKTHDR)
  116                         mb = m_gethdr(how, type);
  117                 else
  118                         mb = m_get(how, type);
  119 
  120                 /* Fail the whole operation if one mbuf can't be allocated. */
  121                 if (mb == NULL) {
  122                         if (nm != NULL)
  123                                 m_freem(nm);
  124                         return (NULL);
  125                 }
  126 
  127                 /* Book keeping. */
  128                 len -= (mb->m_flags & M_EXT) ? mb->m_ext.ext_size :
  129                         ((mb->m_flags & M_PKTHDR) ? MHLEN : MLEN);
  130                 if (mtail != NULL)
  131                         mtail->m_next = mb;
  132                 else
  133                         nm = mb;
  134                 mtail = mb;
  135                 flags &= ~M_PKTHDR;     /* Only valid on the first mbuf. */
  136         }
  137         if (flags & M_EOR)
  138                 mtail->m_flags |= M_EOR;  /* Only valid on the last mbuf. */
  139 
  140         /* If mbuf was supplied, append new chain to the end of it. */
  141         if (m != NULL) {
  142                 for (mtail = m; mtail->m_next != NULL; mtail = mtail->m_next)
  143                         ;
  144                 mtail->m_next = nm;
  145                 mtail->m_flags &= ~M_EOR;
  146         } else
  147                 m = nm;
  148 
  149         return (m);
  150 }
  151 
  152 /*
  153  * Free an entire chain of mbufs and associated external buffers, if
  154  * applicable.
  155  */
  156 void
  157 m_freem(struct mbuf *mb)
  158 {
  159 
  160         while (mb != NULL)
  161                 mb = m_free(mb);
  162 }
  163 
  164 /*-
  165  * Configure a provided mbuf to refer to the provided external storage
  166  * buffer and setup a reference count for said buffer.  If the setting
  167  * up of the reference count fails, the M_EXT bit will not be set.  If
  168  * successfull, the M_EXT bit is set in the mbuf's flags.
  169  *
  170  * Arguments:
  171  *    mb     The existing mbuf to which to attach the provided buffer.
  172  *    buf    The address of the provided external storage buffer.
  173  *    size   The size of the provided buffer.
  174  *    freef  A pointer to a routine that is responsible for freeing the
  175  *           provided external storage buffer.
  176  *    args   A pointer to an argument structure (of any type) to be passed
  177  *           to the provided freef routine (may be NULL).
  178  *    flags  Any other flags to be passed to the provided mbuf.
  179  *    type   The type that the external storage buffer should be
  180  *           labeled with.
  181  *
  182  * Returns:
  183  *    Nothing.
  184  */
  185 void
  186 m_extadd(struct mbuf *mb, caddr_t buf, u_int size,
  187     void (*freef)(void *, void *), void *arg1, void *arg2, int flags, int type)
  188 {
  189         KASSERT(type != EXT_CLUSTER, ("%s: EXT_CLUSTER not allowed", __func__));
  190 
  191         if (type != EXT_EXTREF)
  192                 mb->m_ext.ref_cnt = (u_int *)uma_zalloc(zone_ext_refcnt, M_NOWAIT);
  193         if (mb->m_ext.ref_cnt != NULL) {
  194                 *(mb->m_ext.ref_cnt) = 1;
  195                 mb->m_flags |= (M_EXT | flags);
  196                 mb->m_ext.ext_buf = buf;
  197                 mb->m_data = mb->m_ext.ext_buf;
  198                 mb->m_ext.ext_size = size;
  199                 mb->m_ext.ext_free = freef;
  200                 mb->m_ext.ext_arg1 = arg1;
  201                 mb->m_ext.ext_arg2 = arg2;
  202                 mb->m_ext.ext_type = type;
  203         }
  204 }
  205 
  206 /*
  207  * Non-directly-exported function to clean up after mbufs with M_EXT
  208  * storage attached to them if the reference count hits 1.
  209  */
  210 void
  211 mb_free_ext(struct mbuf *m)
  212 {
  213         int skipmbuf;
  214         
  215         KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
  216         KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
  217 
  218 
  219         /*
  220          * check if the header is embedded in the cluster
  221          */     
  222         skipmbuf = (m->m_flags & M_NOFREE);
  223         
  224         /* Free attached storage if this mbuf is the only reference to it. */
  225         if (*(m->m_ext.ref_cnt) == 1 ||
  226             atomic_fetchadd_int(m->m_ext.ref_cnt, -1) == 1) {
  227                 switch (m->m_ext.ext_type) {
  228                 case EXT_PACKET:        /* The packet zone is special. */
  229                         if (*(m->m_ext.ref_cnt) == 0)
  230                                 *(m->m_ext.ref_cnt) = 1;
  231                         uma_zfree(zone_pack, m);
  232                         return;         /* Job done. */
  233                 case EXT_CLUSTER:
  234                         uma_zfree(zone_clust, m->m_ext.ext_buf);
  235                         break;
  236                 case EXT_JUMBOP:
  237                         uma_zfree(zone_jumbop, m->m_ext.ext_buf);
  238                         break;
  239                 case EXT_JUMBO9:
  240                         uma_zfree(zone_jumbo9, m->m_ext.ext_buf);
  241                         break;
  242                 case EXT_JUMBO16:
  243                         uma_zfree(zone_jumbo16, m->m_ext.ext_buf);
  244                         break;
  245                 case EXT_SFBUF:
  246                 case EXT_NET_DRV:
  247                 case EXT_MOD_TYPE:
  248                 case EXT_DISPOSABLE:
  249                         *(m->m_ext.ref_cnt) = 0;
  250                         uma_zfree(zone_ext_refcnt, __DEVOLATILE(u_int *,
  251                                 m->m_ext.ref_cnt));
  252                         /* FALLTHROUGH */
  253                 case EXT_EXTREF:
  254                         KASSERT(m->m_ext.ext_free != NULL,
  255                                 ("%s: ext_free not set", __func__));
  256                         (*(m->m_ext.ext_free))(m->m_ext.ext_arg1,
  257                             m->m_ext.ext_arg2);
  258                         break;
  259                 default:
  260                         KASSERT(m->m_ext.ext_type == 0,
  261                                 ("%s: unknown ext_type", __func__));
  262                 }
  263         }
  264         if (skipmbuf)
  265                 return;
  266         
  267         /*
  268          * Free this mbuf back to the mbuf zone with all m_ext
  269          * information purged.
  270          */
  271         m->m_ext.ext_buf = NULL;
  272         m->m_ext.ext_free = NULL;
  273         m->m_ext.ext_arg1 = NULL;
  274         m->m_ext.ext_arg2 = NULL;
  275         m->m_ext.ref_cnt = NULL;
  276         m->m_ext.ext_size = 0;
  277         m->m_ext.ext_type = 0;
  278         m->m_flags &= ~M_EXT;
  279         uma_zfree(zone_mbuf, m);
  280 }
  281 
  282 /*
  283  * Attach the cluster from *m to *n, set up m_ext in *n
  284  * and bump the refcount of the cluster.
  285  */
  286 static void
  287 mb_dupcl(struct mbuf *n, struct mbuf *m)
  288 {
  289         KASSERT((m->m_flags & M_EXT) == M_EXT, ("%s: M_EXT not set", __func__));
  290         KASSERT(m->m_ext.ref_cnt != NULL, ("%s: ref_cnt not set", __func__));
  291         KASSERT((n->m_flags & M_EXT) == 0, ("%s: M_EXT set", __func__));
  292 
  293         if (*(m->m_ext.ref_cnt) == 1)
  294                 *(m->m_ext.ref_cnt) += 1;
  295         else
  296                 atomic_add_int(m->m_ext.ref_cnt, 1);
  297         n->m_ext.ext_buf = m->m_ext.ext_buf;
  298         n->m_ext.ext_free = m->m_ext.ext_free;
  299         n->m_ext.ext_arg1 = m->m_ext.ext_arg1;
  300         n->m_ext.ext_arg2 = m->m_ext.ext_arg2;
  301         n->m_ext.ext_size = m->m_ext.ext_size;
  302         n->m_ext.ref_cnt = m->m_ext.ref_cnt;
  303         n->m_ext.ext_type = m->m_ext.ext_type;
  304         n->m_flags |= M_EXT;
  305         n->m_flags |= m->m_flags & M_RDONLY;
  306 }
  307 
  308 /*
  309  * Clean up mbuf (chain) from any tags and packet headers.
  310  * If "all" is set then the first mbuf in the chain will be
  311  * cleaned too.
  312  */
  313 void
  314 m_demote(struct mbuf *m0, int all)
  315 {
  316         struct mbuf *m;
  317 
  318         for (m = all ? m0 : m0->m_next; m != NULL; m = m->m_next) {
  319                 if (m->m_flags & M_PKTHDR) {
  320                         m_tag_delete_chain(m, NULL);
  321                         m->m_flags &= ~M_PKTHDR;
  322                         bzero(&m->m_pkthdr, sizeof(struct pkthdr));
  323                 }
  324                 if (m != m0 && m->m_nextpkt != NULL) {
  325                         KASSERT(m->m_nextpkt == NULL,
  326                             ("%s: m_nextpkt not NULL", __func__));
  327                         m_freem(m->m_nextpkt);
  328                         m->m_nextpkt = NULL;
  329                 }
  330                 m->m_flags = m->m_flags & (M_EXT|M_RDONLY|M_FREELIST|M_NOFREE);
  331         }
  332 }
  333 
  334 /*
  335  * Sanity checks on mbuf (chain) for use in KASSERT() and general
  336  * debugging.
  337  * Returns 0 or panics when bad and 1 on all tests passed.
  338  * Sanitize, 0 to run M_SANITY_ACTION, 1 to garble things so they
  339  * blow up later.
  340  */
  341 int
  342 m_sanity(struct mbuf *m0, int sanitize)
  343 {
  344         struct mbuf *m;
  345         caddr_t a, b;
  346         int pktlen = 0;
  347 
  348 #ifdef INVARIANTS
  349 #define M_SANITY_ACTION(s)      panic("mbuf %p: " s, m)
  350 #else 
  351 #define M_SANITY_ACTION(s)      printf("mbuf %p: " s, m)
  352 #endif
  353 
  354         for (m = m0; m != NULL; m = m->m_next) {
  355                 /*
  356                  * Basic pointer checks.  If any of these fails then some
  357                  * unrelated kernel memory before or after us is trashed.
  358                  * No way to recover from that.
  359                  */
  360                 a = ((m->m_flags & M_EXT) ? m->m_ext.ext_buf :
  361                         ((m->m_flags & M_PKTHDR) ? (caddr_t)(&m->m_pktdat) :
  362                          (caddr_t)(&m->m_dat)) );
  363                 b = (caddr_t)(a + (m->m_flags & M_EXT ? m->m_ext.ext_size :
  364                         ((m->m_flags & M_PKTHDR) ? MHLEN : MLEN)));
  365                 if ((caddr_t)m->m_data < a)
  366                         M_SANITY_ACTION("m_data outside mbuf data range left");
  367                 if ((caddr_t)m->m_data > b)
  368                         M_SANITY_ACTION("m_data outside mbuf data range right");
  369                 if ((caddr_t)m->m_data + m->m_len > b)
  370                         M_SANITY_ACTION("m_data + m_len exeeds mbuf space");
  371                 if ((m->m_flags & M_PKTHDR) && m->m_pkthdr.header) {
  372                         if ((caddr_t)m->m_pkthdr.header < a ||
  373                             (caddr_t)m->m_pkthdr.header > b)
  374                                 M_SANITY_ACTION("m_pkthdr.header outside mbuf data range");
  375                 }
  376 
  377                 /* m->m_nextpkt may only be set on first mbuf in chain. */
  378                 if (m != m0 && m->m_nextpkt != NULL) {
  379                         if (sanitize) {
  380                                 m_freem(m->m_nextpkt);
  381                                 m->m_nextpkt = (struct mbuf *)0xDEADC0DE;
  382                         } else
  383                                 M_SANITY_ACTION("m->m_nextpkt on in-chain mbuf");
  384                 }
  385 
  386                 /* packet length (not mbuf length!) calculation */
  387                 if (m0->m_flags & M_PKTHDR)
  388                         pktlen += m->m_len;
  389 
  390                 /* m_tags may only be attached to first mbuf in chain. */
  391                 if (m != m0 && m->m_flags & M_PKTHDR &&
  392                     !SLIST_EMPTY(&m->m_pkthdr.tags)) {
  393                         if (sanitize) {
  394                                 m_tag_delete_chain(m, NULL);
  395                                 /* put in 0xDEADC0DE perhaps? */
  396                         } else
  397                                 M_SANITY_ACTION("m_tags on in-chain mbuf");
  398                 }
  399 
  400                 /* M_PKTHDR may only be set on first mbuf in chain */
  401                 if (m != m0 && m->m_flags & M_PKTHDR) {
  402                         if (sanitize) {
  403                                 bzero(&m->m_pkthdr, sizeof(m->m_pkthdr));
  404                                 m->m_flags &= ~M_PKTHDR;
  405                                 /* put in 0xDEADCODE and leave hdr flag in */
  406                         } else
  407                                 M_SANITY_ACTION("M_PKTHDR on in-chain mbuf");
  408                 }
  409         }
  410         m = m0;
  411         if (pktlen && pktlen != m->m_pkthdr.len) {
  412                 if (sanitize)
  413                         m->m_pkthdr.len = 0;
  414                 else
  415                         M_SANITY_ACTION("m_pkthdr.len != mbuf chain length");
  416         }
  417         return 1;
  418 
  419 #undef  M_SANITY_ACTION
  420 }
  421 
  422 
  423 /*
  424  * "Move" mbuf pkthdr from "from" to "to".
  425  * "from" must have M_PKTHDR set, and "to" must be empty.
  426  */
  427 void
  428 m_move_pkthdr(struct mbuf *to, struct mbuf *from)
  429 {
  430 
  431 #if 0
  432         /* see below for why these are not enabled */
  433         M_ASSERTPKTHDR(to);
  434         /* Note: with MAC, this may not be a good assertion. */
  435         KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags),
  436             ("m_move_pkthdr: to has tags"));
  437 #endif
  438 #ifdef MAC
  439         /*
  440          * XXXMAC: It could be this should also occur for non-MAC?
  441          */
  442         if (to->m_flags & M_PKTHDR)
  443                 m_tag_delete_chain(to, NULL);
  444 #endif
  445         to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
  446         if ((to->m_flags & M_EXT) == 0)
  447                 to->m_data = to->m_pktdat;
  448         to->m_pkthdr = from->m_pkthdr;          /* especially tags */
  449         SLIST_INIT(&from->m_pkthdr.tags);       /* purge tags from src */
  450         from->m_flags &= ~M_PKTHDR;
  451 }
  452 
  453 /*
  454  * Duplicate "from"'s mbuf pkthdr in "to".
  455  * "from" must have M_PKTHDR set, and "to" must be empty.
  456  * In particular, this does a deep copy of the packet tags.
  457  */
  458 int
  459 m_dup_pkthdr(struct mbuf *to, struct mbuf *from, int how)
  460 {
  461 
  462 #if 0
  463         /*
  464          * The mbuf allocator only initializes the pkthdr
  465          * when the mbuf is allocated with MGETHDR. Many users
  466          * (e.g. m_copy*, m_prepend) use MGET and then
  467          * smash the pkthdr as needed causing these
  468          * assertions to trip.  For now just disable them.
  469          */
  470         M_ASSERTPKTHDR(to);
  471         /* Note: with MAC, this may not be a good assertion. */
  472         KASSERT(SLIST_EMPTY(&to->m_pkthdr.tags), ("m_dup_pkthdr: to has tags"));
  473 #endif
  474         MBUF_CHECKSLEEP(how);
  475 #ifdef MAC
  476         if (to->m_flags & M_PKTHDR)
  477                 m_tag_delete_chain(to, NULL);
  478 #endif
  479         to->m_flags = (from->m_flags & M_COPYFLAGS) | (to->m_flags & M_EXT);
  480         if ((to->m_flags & M_EXT) == 0)
  481                 to->m_data = to->m_pktdat;
  482         to->m_pkthdr = from->m_pkthdr;
  483         SLIST_INIT(&to->m_pkthdr.tags);
  484         return (m_tag_copy_chain(to, from, MBTOM(how)));
  485 }
  486 
  487 /*
  488  * Lesser-used path for M_PREPEND:
  489  * allocate new mbuf to prepend to chain,
  490  * copy junk along.
  491  */
  492 struct mbuf *
  493 m_prepend(struct mbuf *m, int len, int how)
  494 {
  495         struct mbuf *mn;
  496 
  497         if (m->m_flags & M_PKTHDR)
  498                 MGETHDR(mn, how, m->m_type);
  499         else
  500                 MGET(mn, how, m->m_type);
  501         if (mn == NULL) {
  502                 m_freem(m);
  503                 return (NULL);
  504         }
  505         if (m->m_flags & M_PKTHDR)
  506                 M_MOVE_PKTHDR(mn, m);
  507         mn->m_next = m;
  508         m = mn;
  509         if(m->m_flags & M_PKTHDR) {
  510                 if (len < MHLEN)
  511                         MH_ALIGN(m, len);
  512         } else {
  513                 if (len < MLEN) 
  514                         M_ALIGN(m, len);
  515         }
  516         m->m_len = len;
  517         return (m);
  518 }
  519 
  520 /*
  521  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
  522  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
  523  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
  524  * Note that the copy is read-only, because clusters are not copied,
  525  * only their reference counts are incremented.
  526  */
  527 struct mbuf *
  528 m_copym(struct mbuf *m, int off0, int len, int wait)
  529 {
  530         struct mbuf *n, **np;
  531         int off = off0;
  532         struct mbuf *top;
  533         int copyhdr = 0;
  534 
  535         KASSERT(off >= 0, ("m_copym, negative off %d", off));
  536         KASSERT(len >= 0, ("m_copym, negative len %d", len));
  537         MBUF_CHECKSLEEP(wait);
  538         if (off == 0 && m->m_flags & M_PKTHDR)
  539                 copyhdr = 1;
  540         while (off > 0) {
  541                 KASSERT(m != NULL, ("m_copym, offset > size of mbuf chain"));
  542                 if (off < m->m_len)
  543                         break;
  544                 off -= m->m_len;
  545                 m = m->m_next;
  546         }
  547         np = &top;
  548         top = 0;
  549         while (len > 0) {
  550                 if (m == NULL) {
  551                         KASSERT(len == M_COPYALL, 
  552                             ("m_copym, length > size of mbuf chain"));
  553                         break;
  554                 }
  555                 if (copyhdr)
  556                         MGETHDR(n, wait, m->m_type);
  557                 else
  558                         MGET(n, wait, m->m_type);
  559                 *np = n;
  560                 if (n == NULL)
  561                         goto nospace;
  562                 if (copyhdr) {
  563                         if (!m_dup_pkthdr(n, m, wait))
  564                                 goto nospace;
  565                         if (len == M_COPYALL)
  566                                 n->m_pkthdr.len -= off0;
  567                         else
  568                                 n->m_pkthdr.len = len;
  569                         copyhdr = 0;
  570                 }
  571                 n->m_len = min(len, m->m_len - off);
  572                 if (m->m_flags & M_EXT) {
  573                         n->m_data = m->m_data + off;
  574                         mb_dupcl(n, m);
  575                 } else
  576                         bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
  577                             (u_int)n->m_len);
  578                 if (len != M_COPYALL)
  579                         len -= n->m_len;
  580                 off = 0;
  581                 m = m->m_next;
  582                 np = &n->m_next;
  583         }
  584         if (top == NULL)
  585                 mbstat.m_mcfail++;      /* XXX: No consistency. */
  586 
  587         return (top);
  588 nospace:
  589         m_freem(top);
  590         mbstat.m_mcfail++;      /* XXX: No consistency. */
  591         return (NULL);
  592 }
  593 
  594 /*
  595  * Returns mbuf chain with new head for the prepending case.
  596  * Copies from mbuf (chain) n from off for len to mbuf (chain) m
  597  * either prepending or appending the data.
  598  * The resulting mbuf (chain) m is fully writeable.
  599  * m is destination (is made writeable)
  600  * n is source, off is offset in source, len is len from offset
  601  * dir, 0 append, 1 prepend
  602  * how, wait or nowait
  603  */
  604 
  605 static int
  606 m_bcopyxxx(void *s, void *t, u_int len)
  607 {
  608         bcopy(s, t, (size_t)len);
  609         return 0;
  610 }
  611 
  612 struct mbuf *
  613 m_copymdata(struct mbuf *m, struct mbuf *n, int off, int len,
  614     int prep, int how)
  615 {
  616         struct mbuf *mm, *x, *z, *prev = NULL;
  617         caddr_t p;
  618         int i, nlen = 0;
  619         caddr_t buf[MLEN];
  620 
  621         KASSERT(m != NULL && n != NULL, ("m_copymdata, no target or source"));
  622         KASSERT(off >= 0, ("m_copymdata, negative off %d", off));
  623         KASSERT(len >= 0, ("m_copymdata, negative len %d", len));
  624         KASSERT(prep == 0 || prep == 1, ("m_copymdata, unknown direction %d", prep));
  625 
  626         mm = m;
  627         if (!prep) {
  628                 while(mm->m_next) {
  629                         prev = mm;
  630                         mm = mm->m_next;
  631                 }
  632         }
  633         for (z = n; z != NULL; z = z->m_next)
  634                 nlen += z->m_len;
  635         if (len == M_COPYALL)
  636                 len = nlen - off;
  637         if (off + len > nlen || len < 1)
  638                 return NULL;
  639 
  640         if (!M_WRITABLE(mm)) {
  641                 /* XXX: Use proper m_xxx function instead. */
  642                 x = m_getcl(how, MT_DATA, mm->m_flags);
  643                 if (x == NULL)
  644                         return NULL;
  645                 bcopy(mm->m_ext.ext_buf, x->m_ext.ext_buf, x->m_ext.ext_size);
  646                 p = x->m_ext.ext_buf + (mm->m_data - mm->m_ext.ext_buf);
  647                 x->m_data = p;
  648                 mm->m_next = NULL;
  649                 if (mm != m)
  650                         prev->m_next = x;
  651                 m_free(mm);
  652                 mm = x;
  653         }
  654 
  655         /*
  656          * Append/prepend the data.  Allocating mbufs as necessary.
  657          */
  658         /* Shortcut if enough free space in first/last mbuf. */
  659         if (!prep && M_TRAILINGSPACE(mm) >= len) {
  660                 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t) +
  661                          mm->m_len);
  662                 mm->m_len += len;
  663                 mm->m_pkthdr.len += len;
  664                 return m;
  665         }
  666         if (prep && M_LEADINGSPACE(mm) >= len) {
  667                 mm->m_data = mtod(mm, caddr_t) - len;
  668                 m_apply(n, off, len, m_bcopyxxx, mtod(mm, caddr_t));
  669                 mm->m_len += len;
  670                 mm->m_pkthdr.len += len;
  671                 return mm;
  672         }
  673 
  674         /* Expand first/last mbuf to cluster if possible. */
  675         if (!prep && !(mm->m_flags & M_EXT) && len > M_TRAILINGSPACE(mm)) {
  676                 bcopy(mm->m_data, &buf, mm->m_len);
  677                 m_clget(mm, how);
  678                 if (!(mm->m_flags & M_EXT))
  679                         return NULL;
  680                 bcopy(&buf, mm->m_ext.ext_buf, mm->m_len);
  681                 mm->m_data = mm->m_ext.ext_buf;
  682                 mm->m_pkthdr.header = NULL;
  683         }
  684         if (prep && !(mm->m_flags & M_EXT) && len > M_LEADINGSPACE(mm)) {
  685                 bcopy(mm->m_data, &buf, mm->m_len);
  686                 m_clget(mm, how);
  687                 if (!(mm->m_flags & M_EXT))
  688                         return NULL;
  689                 bcopy(&buf, (caddr_t *)mm->m_ext.ext_buf +
  690                        mm->m_ext.ext_size - mm->m_len, mm->m_len);
  691                 mm->m_data = (caddr_t)mm->m_ext.ext_buf +
  692                               mm->m_ext.ext_size - mm->m_len;
  693                 mm->m_pkthdr.header = NULL;
  694         }
  695 
  696         /* Append/prepend as many mbuf (clusters) as necessary to fit len. */
  697         if (!prep && len > M_TRAILINGSPACE(mm)) {
  698                 if (!m_getm(mm, len - M_TRAILINGSPACE(mm), how, MT_DATA))
  699                         return NULL;
  700         }
  701         if (prep && len > M_LEADINGSPACE(mm)) {
  702                 if (!(z = m_getm(NULL, len - M_LEADINGSPACE(mm), how, MT_DATA)))
  703                         return NULL;
  704                 i = 0;
  705                 for (x = z; x != NULL; x = x->m_next) {
  706                         i += x->m_flags & M_EXT ? x->m_ext.ext_size :
  707                               (x->m_flags & M_PKTHDR ? MHLEN : MLEN);
  708                         if (!x->m_next)
  709                                 break;
  710                 }
  711                 z->m_data += i - len;
  712                 m_move_pkthdr(mm, z);
  713                 x->m_next = mm;
  714                 mm = z;
  715         }
  716 
  717         /* Seek to start position in source mbuf. Optimization for long chains. */
  718         while (off > 0) {
  719                 if (off < n->m_len)
  720                         break;
  721                 off -= n->m_len;
  722                 n = n->m_next;
  723         }
  724 
  725         /* Copy data into target mbuf. */
  726         z = mm;
  727         while (len > 0) {
  728                 KASSERT(z != NULL, ("m_copymdata, falling off target edge"));
  729                 i = M_TRAILINGSPACE(z);
  730                 m_apply(n, off, i, m_bcopyxxx, mtod(z, caddr_t) + z->m_len);
  731                 z->m_len += i;
  732                 /* fixup pkthdr.len if necessary */
  733                 if ((prep ? mm : m)->m_flags & M_PKTHDR)
  734                         (prep ? mm : m)->m_pkthdr.len += i;
  735                 off += i;
  736                 len -= i;
  737                 z = z->m_next;
  738         }
  739         return (prep ? mm : m);
  740 }
  741 
  742 /*
  743  * Copy an entire packet, including header (which must be present).
  744  * An optimization of the common case `m_copym(m, 0, M_COPYALL, how)'.
  745  * Note that the copy is read-only, because clusters are not copied,
  746  * only their reference counts are incremented.
  747  * Preserve alignment of the first mbuf so if the creator has left
  748  * some room at the beginning (e.g. for inserting protocol headers)
  749  * the copies still have the room available.
  750  */
  751 struct mbuf *
  752 m_copypacket(struct mbuf *m, int how)
  753 {
  754         struct mbuf *top, *n, *o;
  755 
  756         MBUF_CHECKSLEEP(how);
  757         MGET(n, how, m->m_type);
  758         top = n;
  759         if (n == NULL)
  760                 goto nospace;
  761 
  762         if (!m_dup_pkthdr(n, m, how))
  763                 goto nospace;
  764         n->m_len = m->m_len;
  765         if (m->m_flags & M_EXT) {
  766                 n->m_data = m->m_data;
  767                 mb_dupcl(n, m);
  768         } else {
  769                 n->m_data = n->m_pktdat + (m->m_data - m->m_pktdat );
  770                 bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
  771         }
  772 
  773         m = m->m_next;
  774         while (m) {
  775                 MGET(o, how, m->m_type);
  776                 if (o == NULL)
  777                         goto nospace;
  778 
  779                 n->m_next = o;
  780                 n = n->m_next;
  781 
  782                 n->m_len = m->m_len;
  783                 if (m->m_flags & M_EXT) {
  784                         n->m_data = m->m_data;
  785                         mb_dupcl(n, m);
  786                 } else {
  787                         bcopy(mtod(m, char *), mtod(n, char *), n->m_len);
  788                 }
  789 
  790                 m = m->m_next;
  791         }
  792         return top;
  793 nospace:
  794         m_freem(top);
  795         mbstat.m_mcfail++;      /* XXX: No consistency. */ 
  796         return (NULL);
  797 }
  798 
  799 /*
  800  * Copy data from an mbuf chain starting "off" bytes from the beginning,
  801  * continuing for "len" bytes, into the indicated buffer.
  802  */
  803 void
  804 m_copydata(const struct mbuf *m, int off, int len, caddr_t cp)
  805 {
  806         u_int count;
  807 
  808         KASSERT(off >= 0, ("m_copydata, negative off %d", off));
  809         KASSERT(len >= 0, ("m_copydata, negative len %d", len));
  810         while (off > 0) {
  811                 KASSERT(m != NULL, ("m_copydata, offset > size of mbuf chain"));
  812                 if (off < m->m_len)
  813                         break;
  814                 off -= m->m_len;
  815                 m = m->m_next;
  816         }
  817         while (len > 0) {
  818                 KASSERT(m != NULL, ("m_copydata, length > size of mbuf chain"));
  819                 count = min(m->m_len - off, len);
  820                 bcopy(mtod(m, caddr_t) + off, cp, count);
  821                 len -= count;
  822                 cp += count;
  823                 off = 0;
  824                 m = m->m_next;
  825         }
  826 }
  827 
  828 /*
  829  * Copy a packet header mbuf chain into a completely new chain, including
  830  * copying any mbuf clusters.  Use this instead of m_copypacket() when
  831  * you need a writable copy of an mbuf chain.
  832  */
  833 struct mbuf *
  834 m_dup(struct mbuf *m, int how)
  835 {
  836         struct mbuf **p, *top = NULL;
  837         int remain, moff, nsize;
  838 
  839         MBUF_CHECKSLEEP(how);
  840         /* Sanity check */
  841         if (m == NULL)
  842                 return (NULL);
  843         M_ASSERTPKTHDR(m);
  844 
  845         /* While there's more data, get a new mbuf, tack it on, and fill it */
  846         remain = m->m_pkthdr.len;
  847         moff = 0;
  848         p = &top;
  849         while (remain > 0 || top == NULL) {     /* allow m->m_pkthdr.len == 0 */
  850                 struct mbuf *n;
  851 
  852                 /* Get the next new mbuf */
  853                 if (remain >= MINCLSIZE) {
  854                         n = m_getcl(how, m->m_type, 0);
  855                         nsize = MCLBYTES;
  856                 } else {
  857                         n = m_get(how, m->m_type);
  858                         nsize = MLEN;
  859                 }
  860                 if (n == NULL)
  861                         goto nospace;
  862 
  863                 if (top == NULL) {              /* First one, must be PKTHDR */
  864                         if (!m_dup_pkthdr(n, m, how)) {
  865                                 m_free(n);
  866                                 goto nospace;
  867                         }
  868                         if ((n->m_flags & M_EXT) == 0)
  869                                 nsize = MHLEN;
  870                 }
  871                 n->m_len = 0;
  872 
  873                 /* Link it into the new chain */
  874                 *p = n;
  875                 p = &n->m_next;
  876 
  877                 /* Copy data from original mbuf(s) into new mbuf */
  878                 while (n->m_len < nsize && m != NULL) {
  879                         int chunk = min(nsize - n->m_len, m->m_len - moff);
  880 
  881                         bcopy(m->m_data + moff, n->m_data + n->m_len, chunk);
  882                         moff += chunk;
  883                         n->m_len += chunk;
  884                         remain -= chunk;
  885                         if (moff == m->m_len) {
  886                                 m = m->m_next;
  887                                 moff = 0;
  888                         }
  889                 }
  890 
  891                 /* Check correct total mbuf length */
  892                 KASSERT((remain > 0 && m != NULL) || (remain == 0 && m == NULL),
  893                         ("%s: bogus m_pkthdr.len", __func__));
  894         }
  895         return (top);
  896 
  897 nospace:
  898         m_freem(top);
  899         mbstat.m_mcfail++;      /* XXX: No consistency. */
  900         return (NULL);
  901 }
  902 
  903 /*
  904  * Concatenate mbuf chain n to m.
  905  * Both chains must be of the same type (e.g. MT_DATA).
  906  * Any m_pkthdr is not updated.
  907  */
  908 void
  909 m_cat(struct mbuf *m, struct mbuf *n)
  910 {
  911         while (m->m_next)
  912                 m = m->m_next;
  913         while (n) {
  914                 if (m->m_flags & M_EXT ||
  915                     m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
  916                         /* just join the two chains */
  917                         m->m_next = n;
  918                         return;
  919                 }
  920                 /* splat the data from one into the other */
  921                 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
  922                     (u_int)n->m_len);
  923                 m->m_len += n->m_len;
  924                 n = m_free(n);
  925         }
  926 }
  927 
  928 void
  929 m_adj(struct mbuf *mp, int req_len)
  930 {
  931         int len = req_len;
  932         struct mbuf *m;
  933         int count;
  934 
  935         if ((m = mp) == NULL)
  936                 return;
  937         if (len >= 0) {
  938                 /*
  939                  * Trim from head.
  940                  */
  941                 while (m != NULL && len > 0) {
  942                         if (m->m_len <= len) {
  943                                 len -= m->m_len;
  944                                 m->m_len = 0;
  945                                 m = m->m_next;
  946                         } else {
  947                                 m->m_len -= len;
  948                                 m->m_data += len;
  949                                 len = 0;
  950                         }
  951                 }
  952                 if (mp->m_flags & M_PKTHDR)
  953                         mp->m_pkthdr.len -= (req_len - len);
  954         } else {
  955                 /*
  956                  * Trim from tail.  Scan the mbuf chain,
  957                  * calculating its length and finding the last mbuf.
  958                  * If the adjustment only affects this mbuf, then just
  959                  * adjust and return.  Otherwise, rescan and truncate
  960                  * after the remaining size.
  961                  */
  962                 len = -len;
  963                 count = 0;
  964                 for (;;) {
  965                         count += m->m_len;
  966                         if (m->m_next == (struct mbuf *)0)
  967                                 break;
  968                         m = m->m_next;
  969                 }
  970                 if (m->m_len >= len) {
  971                         m->m_len -= len;
  972                         if (mp->m_flags & M_PKTHDR)
  973                                 mp->m_pkthdr.len -= len;
  974                         return;
  975                 }
  976                 count -= len;
  977                 if (count < 0)
  978                         count = 0;
  979                 /*
  980                  * Correct length for chain is "count".
  981                  * Find the mbuf with last data, adjust its length,
  982                  * and toss data from remaining mbufs on chain.
  983                  */
  984                 m = mp;
  985                 if (m->m_flags & M_PKTHDR)
  986                         m->m_pkthdr.len = count;
  987                 for (; m; m = m->m_next) {
  988                         if (m->m_len >= count) {
  989                                 m->m_len = count;
  990                                 if (m->m_next != NULL) {
  991                                         m_freem(m->m_next);
  992                                         m->m_next = NULL;
  993                                 }
  994                                 break;
  995                         }
  996                         count -= m->m_len;
  997                 }
  998         }
  999 }
 1000 
 1001 /*
 1002  * Rearange an mbuf chain so that len bytes are contiguous
 1003  * and in the data area of an mbuf (so that mtod and dtom
 1004  * will work for a structure of size len).  Returns the resulting
 1005  * mbuf chain on success, frees it and returns null on failure.
 1006  * If there is room, it will add up to max_protohdr-len extra bytes to the
 1007  * contiguous region in an attempt to avoid being called next time.
 1008  */
 1009 struct mbuf *
 1010 m_pullup(struct mbuf *n, int len)
 1011 {
 1012         struct mbuf *m;
 1013         int count;
 1014         int space;
 1015 
 1016         /*
 1017          * If first mbuf has no cluster, and has room for len bytes
 1018          * without shifting current data, pullup into it,
 1019          * otherwise allocate a new mbuf to prepend to the chain.
 1020          */
 1021         if ((n->m_flags & M_EXT) == 0 &&
 1022             n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
 1023                 if (n->m_len >= len)
 1024                         return (n);
 1025                 m = n;
 1026                 n = n->m_next;
 1027                 len -= m->m_len;
 1028         } else {
 1029                 if (len > MHLEN)
 1030                         goto bad;
 1031                 MGET(m, M_DONTWAIT, n->m_type);
 1032                 if (m == NULL)
 1033                         goto bad;
 1034                 m->m_len = 0;
 1035                 if (n->m_flags & M_PKTHDR)
 1036                         M_MOVE_PKTHDR(m, n);
 1037         }
 1038         space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
 1039         do {
 1040                 count = min(min(max(len, max_protohdr), space), n->m_len);
 1041                 bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
 1042                   (u_int)count);
 1043                 len -= count;
 1044                 m->m_len += count;
 1045                 n->m_len -= count;
 1046                 space -= count;
 1047                 if (n->m_len)
 1048                         n->m_data += count;
 1049                 else
 1050                         n = m_free(n);
 1051         } while (len > 0 && n);
 1052         if (len > 0) {
 1053                 (void) m_free(m);
 1054                 goto bad;
 1055         }
 1056         m->m_next = n;
 1057         return (m);
 1058 bad:
 1059         m_freem(n);
 1060         mbstat.m_mpfail++;      /* XXX: No consistency. */
 1061         return (NULL);
 1062 }
 1063 
 1064 /*
 1065  * Like m_pullup(), except a new mbuf is always allocated, and we allow
 1066  * the amount of empty space before the data in the new mbuf to be specified
 1067  * (in the event that the caller expects to prepend later).
 1068  */
 1069 int MSFail;
 1070 
 1071 struct mbuf *
 1072 m_copyup(struct mbuf *n, int len, int dstoff)
 1073 {
 1074         struct mbuf *m;
 1075         int count, space;
 1076 
 1077         if (len > (MHLEN - dstoff))
 1078                 goto bad;
 1079         MGET(m, M_DONTWAIT, n->m_type);
 1080         if (m == NULL)
 1081                 goto bad;
 1082         m->m_len = 0;
 1083         if (n->m_flags & M_PKTHDR)
 1084                 M_MOVE_PKTHDR(m, n);
 1085         m->m_data += dstoff;
 1086         space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
 1087         do {
 1088                 count = min(min(max(len, max_protohdr), space), n->m_len);
 1089                 memcpy(mtod(m, caddr_t) + m->m_len, mtod(n, caddr_t),
 1090                     (unsigned)count);
 1091                 len -= count;
 1092                 m->m_len += count;
 1093                 n->m_len -= count;
 1094                 space -= count;
 1095                 if (n->m_len)
 1096                         n->m_data += count;
 1097                 else
 1098                         n = m_free(n);
 1099         } while (len > 0 && n);
 1100         if (len > 0) {
 1101                 (void) m_free(m);
 1102                 goto bad;
 1103         }
 1104         m->m_next = n;
 1105         return (m);
 1106  bad:
 1107         m_freem(n);
 1108         MSFail++;
 1109         return (NULL);
 1110 }
 1111 
 1112 /*
 1113  * Partition an mbuf chain in two pieces, returning the tail --
 1114  * all but the first len0 bytes.  In case of failure, it returns NULL and
 1115  * attempts to restore the chain to its original state.
 1116  *
 1117  * Note that the resulting mbufs might be read-only, because the new
 1118  * mbuf can end up sharing an mbuf cluster with the original mbuf if
 1119  * the "breaking point" happens to lie within a cluster mbuf. Use the
 1120  * M_WRITABLE() macro to check for this case.
 1121  */
 1122 struct mbuf *
 1123 m_split(struct mbuf *m0, int len0, int wait)
 1124 {
 1125         struct mbuf *m, *n;
 1126         u_int len = len0, remain;
 1127 
 1128         MBUF_CHECKSLEEP(wait);
 1129         for (m = m0; m && len > m->m_len; m = m->m_next)
 1130                 len -= m->m_len;
 1131         if (m == NULL)
 1132                 return (NULL);
 1133         remain = m->m_len - len;
 1134         if (m0->m_flags & M_PKTHDR) {
 1135                 MGETHDR(n, wait, m0->m_type);
 1136                 if (n == NULL)
 1137                         return (NULL);
 1138                 n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
 1139                 n->m_pkthdr.len = m0->m_pkthdr.len - len0;
 1140                 m0->m_pkthdr.len = len0;
 1141                 if (m->m_flags & M_EXT)
 1142                         goto extpacket;
 1143                 if (remain > MHLEN) {
 1144                         /* m can't be the lead packet */
 1145                         MH_ALIGN(n, 0);
 1146                         n->m_next = m_split(m, len, wait);
 1147                         if (n->m_next == NULL) {
 1148                                 (void) m_free(n);
 1149                                 return (NULL);
 1150                         } else {
 1151                                 n->m_len = 0;
 1152                                 return (n);
 1153                         }
 1154                 } else
 1155                         MH_ALIGN(n, remain);
 1156         } else if (remain == 0) {
 1157                 n = m->m_next;
 1158                 m->m_next = NULL;
 1159                 return (n);
 1160         } else {
 1161                 MGET(n, wait, m->m_type);
 1162                 if (n == NULL)
 1163                         return (NULL);
 1164                 M_ALIGN(n, remain);
 1165         }
 1166 extpacket:
 1167         if (m->m_flags & M_EXT) {
 1168                 n->m_data = m->m_data + len;
 1169                 mb_dupcl(n, m);
 1170         } else {
 1171                 bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
 1172         }
 1173         n->m_len = remain;
 1174         m->m_len = len;
 1175         n->m_next = m->m_next;
 1176         m->m_next = NULL;
 1177         return (n);
 1178 }
 1179 /*
 1180  * Routine to copy from device local memory into mbufs.
 1181  * Note that `off' argument is offset into first mbuf of target chain from
 1182  * which to begin copying the data to.
 1183  */
 1184 struct mbuf *
 1185 m_devget(char *buf, int totlen, int off, struct ifnet *ifp,
 1186     void (*copy)(char *from, caddr_t to, u_int len))
 1187 {
 1188         struct mbuf *m;
 1189         struct mbuf *top = NULL, **mp = &top;
 1190         int len;
 1191 
 1192         if (off < 0 || off > MHLEN)
 1193                 return (NULL);
 1194 
 1195         while (totlen > 0) {
 1196                 if (top == NULL) {      /* First one, must be PKTHDR */
 1197                         if (totlen + off >= MINCLSIZE) {
 1198                                 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
 1199                                 len = MCLBYTES;
 1200                         } else {
 1201                                 m = m_gethdr(M_DONTWAIT, MT_DATA);
 1202                                 len = MHLEN;
 1203 
 1204                                 /* Place initial small packet/header at end of mbuf */
 1205                                 if (m && totlen + off + max_linkhdr <= MLEN) {
 1206                                         m->m_data += max_linkhdr;
 1207                                         len -= max_linkhdr;
 1208                                 }
 1209                         }
 1210                         if (m == NULL)
 1211                                 return NULL;
 1212                         m->m_pkthdr.rcvif = ifp;
 1213                         m->m_pkthdr.len = totlen;
 1214                 } else {
 1215                         if (totlen + off >= MINCLSIZE) {
 1216                                 m = m_getcl(M_DONTWAIT, MT_DATA, 0);
 1217                                 len = MCLBYTES;
 1218                         } else {
 1219                                 m = m_get(M_DONTWAIT, MT_DATA);
 1220                                 len = MLEN;
 1221                         }
 1222                         if (m == NULL) {
 1223                                 m_freem(top);
 1224                                 return NULL;
 1225                         }
 1226                 }
 1227                 if (off) {
 1228                         m->m_data += off;
 1229                         len -= off;
 1230                         off = 0;
 1231                 }
 1232                 m->m_len = len = min(totlen, len);
 1233                 if (copy)
 1234                         copy(buf, mtod(m, caddr_t), (u_int)len);
 1235                 else
 1236                         bcopy(buf, mtod(m, caddr_t), (u_int)len);
 1237                 buf += len;
 1238                 *mp = m;
 1239                 mp = &m->m_next;
 1240                 totlen -= len;
 1241         }
 1242         return (top);
 1243 }
 1244 
 1245 /*
 1246  * Copy data from a buffer back into the indicated mbuf chain,
 1247  * starting "off" bytes from the beginning, extending the mbuf
 1248  * chain if necessary.
 1249  */
 1250 void
 1251 m_copyback(struct mbuf *m0, int off, int len, c_caddr_t cp)
 1252 {
 1253         int mlen;
 1254         struct mbuf *m = m0, *n;
 1255         int totlen = 0;
 1256 
 1257         if (m0 == NULL)
 1258                 return;
 1259         while (off > (mlen = m->m_len)) {
 1260                 off -= mlen;
 1261                 totlen += mlen;
 1262                 if (m->m_next == NULL) {
 1263                         n = m_get(M_DONTWAIT, m->m_type);
 1264                         if (n == NULL)
 1265                                 goto out;
 1266                         bzero(mtod(n, caddr_t), MLEN);
 1267                         n->m_len = min(MLEN, len + off);
 1268                         m->m_next = n;
 1269                 }
 1270                 m = m->m_next;
 1271         }
 1272         while (len > 0) {
 1273                 if (m->m_next == NULL && (len > m->m_len - off)) {
 1274                         m->m_len += min(len - (m->m_len - off),
 1275                             M_TRAILINGSPACE(m));
 1276                 }
 1277                 mlen = min (m->m_len - off, len);
 1278                 bcopy(cp, off + mtod(m, caddr_t), (u_int)mlen);
 1279                 cp += mlen;
 1280                 len -= mlen;
 1281                 mlen += off;
 1282                 off = 0;
 1283                 totlen += mlen;
 1284                 if (len == 0)
 1285                         break;
 1286                 if (m->m_next == NULL) {
 1287                         n = m_get(M_DONTWAIT, m->m_type);
 1288                         if (n == NULL)
 1289                                 break;
 1290                         n->m_len = min(MLEN, len);
 1291                         m->m_next = n;
 1292                 }
 1293                 m = m->m_next;
 1294         }
 1295 out:    if (((m = m0)->m_flags & M_PKTHDR) && (m->m_pkthdr.len < totlen))
 1296                 m->m_pkthdr.len = totlen;
 1297 }
 1298 
 1299 /*
 1300  * Append the specified data to the indicated mbuf chain,
 1301  * Extend the mbuf chain if the new data does not fit in
 1302  * existing space.
 1303  *
 1304  * Return 1 if able to complete the job; otherwise 0.
 1305  */
 1306 int
 1307 m_append(struct mbuf *m0, int len, c_caddr_t cp)
 1308 {
 1309         struct mbuf *m, *n;
 1310         int remainder, space;
 1311 
 1312         for (m = m0; m->m_next != NULL; m = m->m_next)
 1313                 ;
 1314         remainder = len;
 1315         space = M_TRAILINGSPACE(m);
 1316         if (space > 0) {
 1317                 /*
 1318                  * Copy into available space.
 1319                  */
 1320                 if (space > remainder)
 1321                         space = remainder;
 1322                 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
 1323                 m->m_len += space;
 1324                 cp += space, remainder -= space;
 1325         }
 1326         while (remainder > 0) {
 1327                 /*
 1328                  * Allocate a new mbuf; could check space
 1329                  * and allocate a cluster instead.
 1330                  */
 1331                 n = m_get(M_DONTWAIT, m->m_type);
 1332                 if (n == NULL)
 1333                         break;
 1334                 n->m_len = min(MLEN, remainder);
 1335                 bcopy(cp, mtod(n, caddr_t), n->m_len);
 1336                 cp += n->m_len, remainder -= n->m_len;
 1337                 m->m_next = n;
 1338                 m = n;
 1339         }
 1340         if (m0->m_flags & M_PKTHDR)
 1341                 m0->m_pkthdr.len += len - remainder;
 1342         return (remainder == 0);
 1343 }
 1344 
 1345 /*
 1346  * Apply function f to the data in an mbuf chain starting "off" bytes from
 1347  * the beginning, continuing for "len" bytes.
 1348  */
 1349 int
 1350 m_apply(struct mbuf *m, int off, int len,
 1351     int (*f)(void *, void *, u_int), void *arg)
 1352 {
 1353         u_int count;
 1354         int rval;
 1355 
 1356         KASSERT(off >= 0, ("m_apply, negative off %d", off));
 1357         KASSERT(len >= 0, ("m_apply, negative len %d", len));
 1358         while (off > 0) {
 1359                 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
 1360                 if (off < m->m_len)
 1361                         break;
 1362                 off -= m->m_len;
 1363                 m = m->m_next;
 1364         }
 1365         while (len > 0) {
 1366                 KASSERT(m != NULL, ("m_apply, offset > size of mbuf chain"));
 1367                 count = min(m->m_len - off, len);
 1368                 rval = (*f)(arg, mtod(m, caddr_t) + off, count);
 1369                 if (rval)
 1370                         return (rval);
 1371                 len -= count;
 1372                 off = 0;
 1373                 m = m->m_next;
 1374         }
 1375         return (0);
 1376 }
 1377 
 1378 /*
 1379  * Return a pointer to mbuf/offset of location in mbuf chain.
 1380  */
 1381 struct mbuf *
 1382 m_getptr(struct mbuf *m, int loc, int *off)
 1383 {
 1384 
 1385         while (loc >= 0) {
 1386                 /* Normal end of search. */
 1387                 if (m->m_len > loc) {
 1388                         *off = loc;
 1389                         return (m);
 1390                 } else {
 1391                         loc -= m->m_len;
 1392                         if (m->m_next == NULL) {
 1393                                 if (loc == 0) {
 1394                                         /* Point at the end of valid data. */
 1395                                         *off = m->m_len;
 1396                                         return (m);
 1397                                 }
 1398                                 return (NULL);
 1399                         }
 1400                         m = m->m_next;
 1401                 }
 1402         }
 1403         return (NULL);
 1404 }
 1405 
 1406 void
 1407 m_print(const struct mbuf *m, int maxlen)
 1408 {
 1409         int len;
 1410         int pdata;
 1411         const struct mbuf *m2;
 1412 
 1413         if (m == NULL) {
 1414                 printf("mbuf: %p\n", m);
 1415                 return;
 1416         }
 1417 
 1418         if (m->m_flags & M_PKTHDR)
 1419                 len = m->m_pkthdr.len;
 1420         else
 1421                 len = -1;
 1422         m2 = m;
 1423         while (m2 != NULL && (len == -1 || len)) {
 1424                 pdata = m2->m_len;
 1425                 if (maxlen != -1 && pdata > maxlen)
 1426                         pdata = maxlen;
 1427                 printf("mbuf: %p len: %d, next: %p, %b%s", m2, m2->m_len,
 1428                     m2->m_next, m2->m_flags, "\2\20freelist\17skipfw"
 1429                     "\11proto5\10proto4\7proto3\6proto2\5proto1\4rdonly"
 1430                     "\3eor\2pkthdr\1ext", pdata ? "" : "\n");
 1431                 if (pdata)
 1432                         printf(", %*D\n", pdata, (u_char *)m2->m_data, "-");
 1433                 if (len != -1)
 1434                         len -= m2->m_len;
 1435                 m2 = m2->m_next;
 1436         }
 1437         if (len > 0)
 1438                 printf("%d bytes unaccounted for.\n", len);
 1439         return;
 1440 }
 1441 
 1442 u_int
 1443 m_fixhdr(struct mbuf *m0)
 1444 {
 1445         u_int len;
 1446 
 1447         len = m_length(m0, NULL);
 1448         m0->m_pkthdr.len = len;
 1449         return (len);
 1450 }
 1451 
 1452 u_int
 1453 m_length(struct mbuf *m0, struct mbuf **last)
 1454 {
 1455         struct mbuf *m;
 1456         u_int len;
 1457 
 1458         len = 0;
 1459         for (m = m0; m != NULL; m = m->m_next) {
 1460                 len += m->m_len;
 1461                 if (m->m_next == NULL)
 1462                         break;
 1463         }
 1464         if (last != NULL)
 1465                 *last = m;
 1466         return (len);
 1467 }
 1468 
 1469 /*
 1470  * Defragment a mbuf chain, returning the shortest possible
 1471  * chain of mbufs and clusters.  If allocation fails and
 1472  * this cannot be completed, NULL will be returned, but
 1473  * the passed in chain will be unchanged.  Upon success,
 1474  * the original chain will be freed, and the new chain
 1475  * will be returned.
 1476  *
 1477  * If a non-packet header is passed in, the original
 1478  * mbuf (chain?) will be returned unharmed.
 1479  */
 1480 struct mbuf *
 1481 m_defrag(struct mbuf *m0, int how)
 1482 {
 1483         struct mbuf *m_new = NULL, *m_final = NULL;
 1484         int progress = 0, length;
 1485 
 1486         MBUF_CHECKSLEEP(how);
 1487         if (!(m0->m_flags & M_PKTHDR))
 1488                 return (m0);
 1489 
 1490         m_fixhdr(m0); /* Needed sanity check */
 1491 
 1492 #ifdef MBUF_STRESS_TEST
 1493         if (m_defragrandomfailures) {
 1494                 int temp = arc4random() & 0xff;
 1495                 if (temp == 0xba)
 1496                         goto nospace;
 1497         }
 1498 #endif
 1499         
 1500         if (m0->m_pkthdr.len > MHLEN)
 1501                 m_final = m_getcl(how, MT_DATA, M_PKTHDR);
 1502         else
 1503                 m_final = m_gethdr(how, MT_DATA);
 1504 
 1505         if (m_final == NULL)
 1506                 goto nospace;
 1507 
 1508         if (m_dup_pkthdr(m_final, m0, how) == 0)
 1509                 goto nospace;
 1510 
 1511         m_new = m_final;
 1512 
 1513         while (progress < m0->m_pkthdr.len) {
 1514                 length = m0->m_pkthdr.len - progress;
 1515                 if (length > MCLBYTES)
 1516                         length = MCLBYTES;
 1517 
 1518                 if (m_new == NULL) {
 1519                         if (length > MLEN)
 1520                                 m_new = m_getcl(how, MT_DATA, 0);
 1521                         else
 1522                                 m_new = m_get(how, MT_DATA);
 1523                         if (m_new == NULL)
 1524                                 goto nospace;
 1525                 }
 1526 
 1527                 m_copydata(m0, progress, length, mtod(m_new, caddr_t));
 1528                 progress += length;
 1529                 m_new->m_len = length;
 1530                 if (m_new != m_final)
 1531                         m_cat(m_final, m_new);
 1532                 m_new = NULL;
 1533         }
 1534 #ifdef MBUF_STRESS_TEST
 1535         if (m0->m_next == NULL)
 1536                 m_defraguseless++;
 1537 #endif
 1538         m_freem(m0);
 1539         m0 = m_final;
 1540 #ifdef MBUF_STRESS_TEST
 1541         m_defragpackets++;
 1542         m_defragbytes += m0->m_pkthdr.len;
 1543 #endif
 1544         return (m0);
 1545 nospace:
 1546 #ifdef MBUF_STRESS_TEST
 1547         m_defragfailure++;
 1548 #endif
 1549         if (m_final)
 1550                 m_freem(m_final);
 1551         return (NULL);
 1552 }
 1553 
 1554 /*
 1555  * Defragment an mbuf chain, returning at most maxfrags separate
 1556  * mbufs+clusters.  If this is not possible NULL is returned and
 1557  * the original mbuf chain is left in it's present (potentially
 1558  * modified) state.  We use two techniques: collapsing consecutive
 1559  * mbufs and replacing consecutive mbufs by a cluster.
 1560  *
 1561  * NB: this should really be named m_defrag but that name is taken
 1562  */
 1563 struct mbuf *
 1564 m_collapse(struct mbuf *m0, int how, int maxfrags)
 1565 {
 1566         struct mbuf *m, *n, *n2, **prev;
 1567         u_int curfrags;
 1568 
 1569         /*
 1570          * Calculate the current number of frags.
 1571          */
 1572         curfrags = 0;
 1573         for (m = m0; m != NULL; m = m->m_next)
 1574                 curfrags++;
 1575         /*
 1576          * First, try to collapse mbufs.  Note that we always collapse
 1577          * towards the front so we don't need to deal with moving the
 1578          * pkthdr.  This may be suboptimal if the first mbuf has much
 1579          * less data than the following.
 1580          */
 1581         m = m0;
 1582 again:
 1583         for (;;) {
 1584                 n = m->m_next;
 1585                 if (n == NULL)
 1586                         break;
 1587                 if ((m->m_flags & M_RDONLY) == 0 &&
 1588                     n->m_len < M_TRAILINGSPACE(m)) {
 1589                         bcopy(mtod(n, void *), mtod(m, char *) + m->m_len,
 1590                                 n->m_len);
 1591                         m->m_len += n->m_len;
 1592                         m->m_next = n->m_next;
 1593                         m_free(n);
 1594                         if (--curfrags <= maxfrags)
 1595                                 return m0;
 1596                 } else
 1597                         m = n;
 1598         }
 1599         KASSERT(maxfrags > 1,
 1600                 ("maxfrags %u, but normal collapse failed", maxfrags));
 1601         /*
 1602          * Collapse consecutive mbufs to a cluster.
 1603          */
 1604         prev = &m0->m_next;             /* NB: not the first mbuf */
 1605         while ((n = *prev) != NULL) {
 1606                 if ((n2 = n->m_next) != NULL &&
 1607                     n->m_len + n2->m_len < MCLBYTES) {
 1608                         m = m_getcl(how, MT_DATA, 0);
 1609                         if (m == NULL)
 1610                                 goto bad;
 1611                         bcopy(mtod(n, void *), mtod(m, void *), n->m_len);
 1612                         bcopy(mtod(n2, void *), mtod(m, char *) + n->m_len,
 1613                                 n2->m_len);
 1614                         m->m_len = n->m_len + n2->m_len;
 1615                         m->m_next = n2->m_next;
 1616                         *prev = m;
 1617                         m_free(n);
 1618                         m_free(n2);
 1619                         if (--curfrags <= maxfrags)     /* +1 cl -2 mbufs */
 1620                                 return m0;
 1621                         /*
 1622                          * Still not there, try the normal collapse
 1623                          * again before we allocate another cluster.
 1624                          */
 1625                         goto again;
 1626                 }
 1627                 prev = &n->m_next;
 1628         }
 1629         /*
 1630          * No place where we can collapse to a cluster; punt.
 1631          * This can occur if, for example, you request 2 frags
 1632          * but the packet requires that both be clusters (we
 1633          * never reallocate the first mbuf to avoid moving the
 1634          * packet header).
 1635          */
 1636 bad:
 1637         return NULL;
 1638 }
 1639 
 1640 #ifdef MBUF_STRESS_TEST
 1641 
 1642 /*
 1643  * Fragment an mbuf chain.  There's no reason you'd ever want to do
 1644  * this in normal usage, but it's great for stress testing various
 1645  * mbuf consumers.
 1646  *
 1647  * If fragmentation is not possible, the original chain will be
 1648  * returned.
 1649  *
 1650  * Possible length values:
 1651  * 0     no fragmentation will occur
 1652  * > 0  each fragment will be of the specified length
 1653  * -1   each fragment will be the same random value in length
 1654  * -2   each fragment's length will be entirely random
 1655  * (Random values range from 1 to 256)
 1656  */
 1657 struct mbuf *
 1658 m_fragment(struct mbuf *m0, int how, int length)
 1659 {
 1660         struct mbuf *m_new = NULL, *m_final = NULL;
 1661         int progress = 0;
 1662 
 1663         if (!(m0->m_flags & M_PKTHDR))
 1664                 return (m0);
 1665         
 1666         if ((length == 0) || (length < -2))
 1667                 return (m0);
 1668 
 1669         m_fixhdr(m0); /* Needed sanity check */
 1670 
 1671         m_final = m_getcl(how, MT_DATA, M_PKTHDR);
 1672 
 1673         if (m_final == NULL)
 1674                 goto nospace;
 1675 
 1676         if (m_dup_pkthdr(m_final, m0, how) == 0)
 1677                 goto nospace;
 1678 
 1679         m_new = m_final;
 1680 
 1681         if (length == -1)
 1682                 length = 1 + (arc4random() & 255);
 1683 
 1684         while (progress < m0->m_pkthdr.len) {
 1685                 int fraglen;
 1686 
 1687                 if (length > 0)
 1688                         fraglen = length;
 1689                 else
 1690                         fraglen = 1 + (arc4random() & 255);
 1691                 if (fraglen > m0->m_pkthdr.len - progress)
 1692                         fraglen = m0->m_pkthdr.len - progress;
 1693 
 1694                 if (fraglen > MCLBYTES)
 1695                         fraglen = MCLBYTES;
 1696 
 1697                 if (m_new == NULL) {
 1698                         m_new = m_getcl(how, MT_DATA, 0);
 1699                         if (m_new == NULL)
 1700                                 goto nospace;
 1701                 }
 1702 
 1703                 m_copydata(m0, progress, fraglen, mtod(m_new, caddr_t));
 1704                 progress += fraglen;
 1705                 m_new->m_len = fraglen;
 1706                 if (m_new != m_final)
 1707                         m_cat(m_final, m_new);
 1708                 m_new = NULL;
 1709         }
 1710         m_freem(m0);
 1711         m0 = m_final;
 1712         return (m0);
 1713 nospace:
 1714         if (m_final)
 1715                 m_freem(m_final);
 1716         /* Return the original chain on failure */
 1717         return (m0);
 1718 }
 1719 
 1720 #endif
 1721 
 1722 /*
 1723  * Copy the contents of uio into a properly sized mbuf chain.
 1724  */
 1725 struct mbuf *
 1726 m_uiotombuf(struct uio *uio, int how, int len, int align, int flags)
 1727 {
 1728         struct mbuf *m, *mb;
 1729         int error, length, total;
 1730         int progress = 0;
 1731 
 1732         /*
 1733          * len can be zero or an arbitrary large value bound by
 1734          * the total data supplied by the uio.
 1735          */
 1736         if (len > 0)
 1737                 total = min(uio->uio_resid, len);
 1738         else
 1739                 total = uio->uio_resid;
 1740 
 1741         /*
 1742          * The smallest unit returned by m_getm2() is a single mbuf
 1743          * with pkthdr.  We can't align past it.
 1744          */
 1745         if (align >= MHLEN)
 1746                 return (NULL);
 1747 
 1748         /*
 1749          * Give us the full allocation or nothing.
 1750          * If len is zero return the smallest empty mbuf.
 1751          */
 1752         m = m_getm2(NULL, max(total + align, 1), how, MT_DATA, flags);
 1753         if (m == NULL)
 1754                 return (NULL);
 1755         m->m_data += align;
 1756 
 1757         /* Fill all mbufs with uio data and update header information. */
 1758         for (mb = m; mb != NULL; mb = mb->m_next) {
 1759                 length = min(M_TRAILINGSPACE(mb), total - progress);
 1760 
 1761                 error = uiomove(mtod(mb, void *), length, uio);
 1762                 if (error) {
 1763                         m_freem(m);
 1764                         return (NULL);
 1765                 }
 1766 
 1767                 mb->m_len = length;
 1768                 progress += length;
 1769                 if (flags & M_PKTHDR)
 1770                         m->m_pkthdr.len += length;
 1771         }
 1772         KASSERT(progress == total, ("%s: progress != total", __func__));
 1773 
 1774         return (m);
 1775 }
 1776 
 1777 /*
 1778  * Copy an mbuf chain into a uio limited by len if set.
 1779  */
 1780 int
 1781 m_mbuftouio(struct uio *uio, struct mbuf *m, int len)
 1782 {
 1783         int error, length, total;
 1784         int progress = 0;
 1785 
 1786         if (len > 0)
 1787                 total = min(uio->uio_resid, len);
 1788         else
 1789                 total = uio->uio_resid;
 1790 
 1791         /* Fill the uio with data from the mbufs. */
 1792         for (; m != NULL; m = m->m_next) {
 1793                 length = min(m->m_len, total - progress);
 1794 
 1795                 error = uiomove(mtod(m, void *), length, uio);
 1796                 if (error)
 1797                         return (error);
 1798 
 1799                 progress += length;
 1800         }
 1801 
 1802         return (0);
 1803 }
 1804 
 1805 /*
 1806  * Set the m_data pointer of a newly-allocated mbuf
 1807  * to place an object of the specified size at the
 1808  * end of the mbuf, longword aligned.
 1809  */
 1810 void
 1811 m_align(struct mbuf *m, int len)
 1812 {
 1813         int adjust;
 1814 
 1815         if (m->m_flags & M_EXT)
 1816                 adjust = m->m_ext.ext_size - len;
 1817         else if (m->m_flags & M_PKTHDR)
 1818                 adjust = MHLEN - len;
 1819         else
 1820                 adjust = MLEN - len;
 1821         m->m_data += adjust &~ (sizeof(long)-1);
 1822 }
 1823 
 1824 /*
 1825  * Create a writable copy of the mbuf chain.  While doing this
 1826  * we compact the chain with a goal of producing a chain with
 1827  * at most two mbufs.  The second mbuf in this chain is likely
 1828  * to be a cluster.  The primary purpose of this work is to create
 1829  * a writable packet for encryption, compression, etc.  The
 1830  * secondary goal is to linearize the data so the data can be
 1831  * passed to crypto hardware in the most efficient manner possible.
 1832  */
 1833 struct mbuf *
 1834 m_unshare(struct mbuf *m0, int how)
 1835 {
 1836         struct mbuf *m, *mprev;
 1837         struct mbuf *n, *mfirst, *mlast;
 1838         int len, off;
 1839 
 1840         mprev = NULL;
 1841         for (m = m0; m != NULL; m = mprev->m_next) {
 1842                 /*
 1843                  * Regular mbufs are ignored unless there's a cluster
 1844                  * in front of it that we can use to coalesce.  We do
 1845                  * the latter mainly so later clusters can be coalesced
 1846                  * also w/o having to handle them specially (i.e. convert
 1847                  * mbuf+cluster -> cluster).  This optimization is heavily
 1848                  * influenced by the assumption that we're running over
 1849                  * Ethernet where MCLBYTES is large enough that the max
 1850                  * packet size will permit lots of coalescing into a
 1851                  * single cluster.  This in turn permits efficient
 1852                  * crypto operations, especially when using hardware.
 1853                  */
 1854                 if ((m->m_flags & M_EXT) == 0) {
 1855                         if (mprev && (mprev->m_flags & M_EXT) &&
 1856                             m->m_len <= M_TRAILINGSPACE(mprev)) {
 1857                                 /* XXX: this ignores mbuf types */
 1858                                 memcpy(mtod(mprev, caddr_t) + mprev->m_len,
 1859                                        mtod(m, caddr_t), m->m_len);
 1860                                 mprev->m_len += m->m_len;
 1861                                 mprev->m_next = m->m_next;      /* unlink from chain */
 1862                                 m_free(m);                      /* reclaim mbuf */
 1863 #if 0
 1864                                 newipsecstat.ips_mbcoalesced++;
 1865 #endif
 1866                         } else {
 1867                                 mprev = m;
 1868                         }
 1869                         continue;
 1870                 }
 1871                 /*
 1872                  * Writable mbufs are left alone (for now).
 1873                  */
 1874                 if (M_WRITABLE(m)) {
 1875                         mprev = m;
 1876                         continue;
 1877                 }
 1878 
 1879                 /*
 1880                  * Not writable, replace with a copy or coalesce with
 1881                  * the previous mbuf if possible (since we have to copy
 1882                  * it anyway, we try to reduce the number of mbufs and
 1883                  * clusters so that future work is easier).
 1884                  */
 1885                 KASSERT(m->m_flags & M_EXT, ("m_flags 0x%x", m->m_flags));
 1886                 /* NB: we only coalesce into a cluster or larger */
 1887                 if (mprev != NULL && (mprev->m_flags & M_EXT) &&
 1888                     m->m_len <= M_TRAILINGSPACE(mprev)) {
 1889                         /* XXX: this ignores mbuf types */
 1890                         memcpy(mtod(mprev, caddr_t) + mprev->m_len,
 1891                                mtod(m, caddr_t), m->m_len);
 1892                         mprev->m_len += m->m_len;
 1893                         mprev->m_next = m->m_next;      /* unlink from chain */
 1894                         m_free(m);                      /* reclaim mbuf */
 1895 #if 0
 1896                         newipsecstat.ips_clcoalesced++;
 1897 #endif
 1898                         continue;
 1899                 }
 1900 
 1901                 /*
 1902                  * Allocate new space to hold the copy...
 1903                  */
 1904                 /* XXX why can M_PKTHDR be set past the first mbuf? */
 1905                 if (mprev == NULL && (m->m_flags & M_PKTHDR)) {
 1906                         /*
 1907                          * NB: if a packet header is present we must
 1908                          * allocate the mbuf separately from any cluster
 1909                          * because M_MOVE_PKTHDR will smash the data
 1910                          * pointer and drop the M_EXT marker.
 1911                          */
 1912                         MGETHDR(n, how, m->m_type);
 1913                         if (n == NULL) {
 1914                                 m_freem(m0);
 1915                                 return (NULL);
 1916                         }
 1917                         M_MOVE_PKTHDR(n, m);
 1918                         MCLGET(n, how);
 1919                         if ((n->m_flags & M_EXT) == 0) {
 1920                                 m_free(n);
 1921                                 m_freem(m0);
 1922                                 return (NULL);
 1923                         }
 1924                 } else {
 1925                         n = m_getcl(how, m->m_type, m->m_flags);
 1926                         if (n == NULL) {
 1927                                 m_freem(m0);
 1928                                 return (NULL);
 1929                         }
 1930                 }
 1931                 /*
 1932                  * ... and copy the data.  We deal with jumbo mbufs
 1933                  * (i.e. m_len > MCLBYTES) by splitting them into
 1934                  * clusters.  We could just malloc a buffer and make
 1935                  * it external but too many device drivers don't know
 1936                  * how to break up the non-contiguous memory when
 1937                  * doing DMA.
 1938                  */
 1939                 len = m->m_len;
 1940                 off = 0;
 1941                 mfirst = n;
 1942                 mlast = NULL;
 1943                 for (;;) {
 1944                         int cc = min(len, MCLBYTES);
 1945                         memcpy(mtod(n, caddr_t), mtod(m, caddr_t) + off, cc);
 1946                         n->m_len = cc;
 1947                         if (mlast != NULL)
 1948                                 mlast->m_next = n;
 1949                         mlast = n;      
 1950 #if 0
 1951                         newipsecstat.ips_clcopied++;
 1952 #endif
 1953 
 1954                         len -= cc;
 1955                         if (len <= 0)
 1956                                 break;
 1957                         off += cc;
 1958 
 1959                         n = m_getcl(how, m->m_type, m->m_flags);
 1960                         if (n == NULL) {
 1961                                 m_freem(mfirst);
 1962                                 m_freem(m0);
 1963                                 return (NULL);
 1964                         }
 1965                 }
 1966                 n->m_next = m->m_next; 
 1967                 if (mprev == NULL)
 1968                         m0 = mfirst;            /* new head of chain */
 1969                 else
 1970                         mprev->m_next = mfirst; /* replace old mbuf */
 1971                 m_free(m);                      /* release old mbuf */
 1972                 mprev = mfirst;
 1973         }
 1974         return (m0);
 1975 }
 1976 
 1977 #ifdef MBUF_PROFILING
 1978 
 1979 #define MP_BUCKETS 32 /* don't just change this as things may overflow.*/
 1980 struct mbufprofile {
 1981         uintmax_t wasted[MP_BUCKETS];
 1982         uintmax_t used[MP_BUCKETS];
 1983         uintmax_t segments[MP_BUCKETS];
 1984 } mbprof;
 1985 
 1986 #define MP_MAXDIGITS 21 /* strlen("16,000,000,000,000,000,000") == 21 */
 1987 #define MP_NUMLINES 6
 1988 #define MP_NUMSPERLINE 16
 1989 #define MP_EXTRABYTES 64        /* > strlen("used:\nwasted:\nsegments:\n") */
 1990 /* work out max space needed and add a bit of spare space too */
 1991 #define MP_MAXLINE ((MP_MAXDIGITS+1) * MP_NUMSPERLINE)
 1992 #define MP_BUFSIZE ((MP_MAXLINE * MP_NUMLINES) + 1 + MP_EXTRABYTES)
 1993 
 1994 char mbprofbuf[MP_BUFSIZE];
 1995 
 1996 void
 1997 m_profile(struct mbuf *m)
 1998 {
 1999         int segments = 0;
 2000         int used = 0;
 2001         int wasted = 0;
 2002         
 2003         while (m) {
 2004                 segments++;
 2005                 used += m->m_len;
 2006                 if (m->m_flags & M_EXT) {
 2007                         wasted += MHLEN - sizeof(m->m_ext) +
 2008                             m->m_ext.ext_size - m->m_len;
 2009                 } else {
 2010                         if (m->m_flags & M_PKTHDR)
 2011                                 wasted += MHLEN - m->m_len;
 2012                         else
 2013                                 wasted += MLEN - m->m_len;
 2014                 }
 2015                 m = m->m_next;
 2016         }
 2017         /* be paranoid.. it helps */
 2018         if (segments > MP_BUCKETS - 1)
 2019                 segments = MP_BUCKETS - 1;
 2020         if (used > 100000)
 2021                 used = 100000;
 2022         if (wasted > 100000)
 2023                 wasted = 100000;
 2024         /* store in the appropriate bucket */
 2025         /* don't bother locking. if it's slightly off, so what? */
 2026         mbprof.segments[segments]++;
 2027         mbprof.used[fls(used)]++;
 2028         mbprof.wasted[fls(wasted)]++;
 2029 }
 2030 
 2031 static void
 2032 mbprof_textify(void)
 2033 {
 2034         int offset;
 2035         char *c;
 2036         u_int64_t *p;
 2037         
 2038 
 2039         p = &mbprof.wasted[0];
 2040         c = mbprofbuf;
 2041         offset = snprintf(c, MP_MAXLINE + 10, 
 2042             "wasted:\n"
 2043             "%ju %ju %ju %ju %ju %ju %ju %ju "
 2044             "%ju %ju %ju %ju %ju %ju %ju %ju\n",
 2045             p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
 2046             p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
 2047 #ifdef BIG_ARRAY
 2048         p = &mbprof.wasted[16];
 2049         c += offset;
 2050         offset = snprintf(c, MP_MAXLINE, 
 2051             "%ju %ju %ju %ju %ju %ju %ju %ju "
 2052             "%ju %ju %ju %ju %ju %ju %ju %ju\n",
 2053             p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
 2054             p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
 2055 #endif
 2056         p = &mbprof.used[0];
 2057         c += offset;
 2058         offset = snprintf(c, MP_MAXLINE + 10, 
 2059             "used:\n"
 2060             "%ju %ju %ju %ju %ju %ju %ju %ju "
 2061             "%ju %ju %ju %ju %ju %ju %ju %ju\n",
 2062             p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
 2063             p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
 2064 #ifdef BIG_ARRAY
 2065         p = &mbprof.used[16];
 2066         c += offset;
 2067         offset = snprintf(c, MP_MAXLINE, 
 2068             "%ju %ju %ju %ju %ju %ju %ju %ju "
 2069             "%ju %ju %ju %ju %ju %ju %ju %ju\n",
 2070             p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
 2071             p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
 2072 #endif
 2073         p = &mbprof.segments[0];
 2074         c += offset;
 2075         offset = snprintf(c, MP_MAXLINE + 10, 
 2076             "segments:\n"
 2077             "%ju %ju %ju %ju %ju %ju %ju %ju "
 2078             "%ju %ju %ju %ju %ju %ju %ju %ju\n",
 2079             p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
 2080             p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
 2081 #ifdef BIG_ARRAY
 2082         p = &mbprof.segments[16];
 2083         c += offset;
 2084         offset = snprintf(c, MP_MAXLINE, 
 2085             "%ju %ju %ju %ju %ju %ju %ju %ju "
 2086             "%ju %ju %ju %ju %ju %ju %ju %jju",
 2087             p[0], p[1], p[2], p[3], p[4], p[5], p[6], p[7],
 2088             p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]);
 2089 #endif
 2090 }
 2091 
 2092 static int
 2093 mbprof_handler(SYSCTL_HANDLER_ARGS)
 2094 {
 2095         int error;
 2096 
 2097         mbprof_textify();
 2098         error = SYSCTL_OUT(req, mbprofbuf, strlen(mbprofbuf) + 1);
 2099         return (error);
 2100 }
 2101 
 2102 static int
 2103 mbprof_clr_handler(SYSCTL_HANDLER_ARGS)
 2104 {
 2105         int clear, error;
 2106  
 2107         clear = 0;
 2108         error = sysctl_handle_int(oidp, &clear, 0, req);
 2109         if (error || !req->newptr)
 2110                 return (error);
 2111  
 2112         if (clear) {
 2113                 bzero(&mbprof, sizeof(mbprof));
 2114         }
 2115  
 2116         return (error);
 2117 }
 2118 
 2119 
 2120 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofile, CTLTYPE_STRING|CTLFLAG_RD,
 2121             NULL, 0, mbprof_handler, "A", "mbuf profiling statistics");
 2122 
 2123 SYSCTL_PROC(_kern_ipc, OID_AUTO, mbufprofileclr, CTLTYPE_INT|CTLFLAG_RW,
 2124             NULL, 0, mbprof_clr_handler, "I", "clear mbuf profiling statistics");
 2125 #endif
 2126 

Cache object: 38af126416ee8f6e5dcb175c6551cb2b


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.