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_sockbuf.c

Version: -  FREEBSD  -  FREEBSD11  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  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  * SPDX-License-Identifier: BSD-3-Clause
    3  *
    4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
    5  *      The Regents of the University of California.  All rights reserved.
    6  *
    7  * Redistribution and use in source and binary forms, with or without
    8  * modification, are permitted provided that the following conditions
    9  * are met:
   10  * 1. Redistributions of source code must retain the above copyright
   11  *    notice, this list of conditions and the following disclaimer.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice, this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  * 3. Neither the name of the University nor the names of its contributors
   16  *    may be used to endorse or promote products derived from this software
   17  *    without specific prior written permission.
   18  *
   19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   29  * SUCH DAMAGE.
   30  *
   31  *      @(#)uipc_socket2.c      8.1 (Berkeley) 6/10/93
   32  */
   33 
   34 #include <sys/cdefs.h>
   35 __FBSDID("$FreeBSD: head/sys/kern/uipc_sockbuf.c 337328 2018-08-04 20:26:54Z markj $");
   36 
   37 #include "opt_param.h"
   38 
   39 #include <sys/param.h>
   40 #include <sys/aio.h> /* for aio_swake proto */
   41 #include <sys/kernel.h>
   42 #include <sys/lock.h>
   43 #include <sys/malloc.h>
   44 #include <sys/mbuf.h>
   45 #include <sys/mutex.h>
   46 #include <sys/proc.h>
   47 #include <sys/protosw.h>
   48 #include <sys/resourcevar.h>
   49 #include <sys/signalvar.h>
   50 #include <sys/socket.h>
   51 #include <sys/socketvar.h>
   52 #include <sys/sx.h>
   53 #include <sys/sysctl.h>
   54 
   55 /*
   56  * Function pointer set by the AIO routines so that the socket buffer code
   57  * can call back into the AIO module if it is loaded.
   58  */
   59 void    (*aio_swake)(struct socket *, struct sockbuf *);
   60 
   61 /*
   62  * Primitive routines for operating on socket buffers
   63  */
   64 
   65 u_long  sb_max = SB_MAX;
   66 u_long sb_max_adj =
   67        (quad_t)SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
   68 
   69 static  u_long sb_efficiency = 8;       /* parameter for sbreserve() */
   70 
   71 static struct mbuf      *sbcut_internal(struct sockbuf *sb, int len);
   72 static void     sbflush_internal(struct sockbuf *sb);
   73 
   74 /*
   75  * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY.
   76  */
   77 static void
   78 sbm_clrprotoflags(struct mbuf *m, int flags)
   79 {
   80         int mask;
   81 
   82         mask = ~M_PROTOFLAGS;
   83         if (flags & PRUS_NOTREADY)
   84                 mask |= M_NOTREADY;
   85         while (m) {
   86                 m->m_flags &= mask;
   87                 m = m->m_next;
   88         }
   89 }
   90 
   91 /*
   92  * Mark ready "count" mbufs starting with "m".
   93  */
   94 int
   95 sbready(struct sockbuf *sb, struct mbuf *m, int count)
   96 {
   97         u_int blocker;
   98 
   99         SOCKBUF_LOCK_ASSERT(sb);
  100         KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb));
  101 
  102         blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0;
  103 
  104         for (int i = 0; i < count; i++, m = m->m_next) {
  105                 KASSERT(m->m_flags & M_NOTREADY,
  106                     ("%s: m %p !M_NOTREADY", __func__, m));
  107                 m->m_flags &= ~(M_NOTREADY | blocker);
  108                 if (blocker)
  109                         sb->sb_acc += m->m_len;
  110         }
  111 
  112         if (!blocker)
  113                 return (EINPROGRESS);
  114 
  115         /* This one was blocking all the queue. */
  116         for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) {
  117                 KASSERT(m->m_flags & M_BLOCKED,
  118                     ("%s: m %p !M_BLOCKED", __func__, m));
  119                 m->m_flags &= ~M_BLOCKED;
  120                 sb->sb_acc += m->m_len;
  121         }
  122 
  123         sb->sb_fnrdy = m;
  124 
  125         return (0);
  126 }
  127 
  128 /*
  129  * Adjust sockbuf state reflecting allocation of m.
  130  */
  131 void
  132 sballoc(struct sockbuf *sb, struct mbuf *m)
  133 {
  134 
  135         SOCKBUF_LOCK_ASSERT(sb);
  136 
  137         sb->sb_ccc += m->m_len;
  138 
  139         if (sb->sb_fnrdy == NULL) {
  140                 if (m->m_flags & M_NOTREADY)
  141                         sb->sb_fnrdy = m;
  142                 else
  143                         sb->sb_acc += m->m_len;
  144         } else
  145                 m->m_flags |= M_BLOCKED;
  146 
  147         if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
  148                 sb->sb_ctl += m->m_len;
  149 
  150         sb->sb_mbcnt += MSIZE;
  151         sb->sb_mcnt += 1;
  152 
  153         if (m->m_flags & M_EXT) {
  154                 sb->sb_mbcnt += m->m_ext.ext_size;
  155                 sb->sb_ccnt += 1;
  156         }
  157 }
  158 
  159 /*
  160  * Adjust sockbuf state reflecting freeing of m.
  161  */
  162 void
  163 sbfree(struct sockbuf *sb, struct mbuf *m)
  164 {
  165 
  166 #if 0   /* XXX: not yet: soclose() call path comes here w/o lock. */
  167         SOCKBUF_LOCK_ASSERT(sb);
  168 #endif
  169 
  170         sb->sb_ccc -= m->m_len;
  171 
  172         if (!(m->m_flags & M_NOTAVAIL))
  173                 sb->sb_acc -= m->m_len;
  174 
  175         if (m == sb->sb_fnrdy) {
  176                 struct mbuf *n;
  177 
  178                 KASSERT(m->m_flags & M_NOTREADY,
  179                     ("%s: m %p !M_NOTREADY", __func__, m));
  180 
  181                 n = m->m_next;
  182                 while (n != NULL && !(n->m_flags & M_NOTREADY)) {
  183                         n->m_flags &= ~M_BLOCKED;
  184                         sb->sb_acc += n->m_len;
  185                         n = n->m_next;
  186                 }
  187                 sb->sb_fnrdy = n;
  188         }
  189 
  190         if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
  191                 sb->sb_ctl -= m->m_len;
  192 
  193         sb->sb_mbcnt -= MSIZE;
  194         sb->sb_mcnt -= 1;
  195         if (m->m_flags & M_EXT) {
  196                 sb->sb_mbcnt -= m->m_ext.ext_size;
  197                 sb->sb_ccnt -= 1;
  198         }
  199 
  200         if (sb->sb_sndptr == m) {
  201                 sb->sb_sndptr = NULL;
  202                 sb->sb_sndptroff = 0;
  203         }
  204         if (sb->sb_sndptroff != 0)
  205                 sb->sb_sndptroff -= m->m_len;
  206 }
  207 
  208 /*
  209  * Socantsendmore indicates that no more data will be sent on the socket; it
  210  * would normally be applied to a socket when the user informs the system
  211  * that no more data is to be sent, by the protocol code (in case
  212  * PRU_SHUTDOWN).  Socantrcvmore indicates that no more data will be
  213  * received, and will normally be applied to the socket by a protocol when it
  214  * detects that the peer will send no more data.  Data queued for reading in
  215  * the socket may yet be read.
  216  */
  217 void
  218 socantsendmore_locked(struct socket *so)
  219 {
  220 
  221         SOCKBUF_LOCK_ASSERT(&so->so_snd);
  222 
  223         so->so_snd.sb_state |= SBS_CANTSENDMORE;
  224         sowwakeup_locked(so);
  225         mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
  226 }
  227 
  228 void
  229 socantsendmore(struct socket *so)
  230 {
  231 
  232         SOCKBUF_LOCK(&so->so_snd);
  233         socantsendmore_locked(so);
  234         mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
  235 }
  236 
  237 void
  238 socantrcvmore_locked(struct socket *so)
  239 {
  240 
  241         SOCKBUF_LOCK_ASSERT(&so->so_rcv);
  242 
  243         so->so_rcv.sb_state |= SBS_CANTRCVMORE;
  244         sorwakeup_locked(so);
  245         mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
  246 }
  247 
  248 void
  249 socantrcvmore(struct socket *so)
  250 {
  251 
  252         SOCKBUF_LOCK(&so->so_rcv);
  253         socantrcvmore_locked(so);
  254         mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
  255 }
  256 
  257 /*
  258  * Wait for data to arrive at/drain from a socket buffer.
  259  */
  260 int
  261 sbwait(struct sockbuf *sb)
  262 {
  263 
  264         SOCKBUF_LOCK_ASSERT(sb);
  265 
  266         sb->sb_flags |= SB_WAIT;
  267         return (msleep_sbt(&sb->sb_acc, &sb->sb_mtx,
  268             (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
  269             sb->sb_timeo, 0, 0));
  270 }
  271 
  272 int
  273 sblock(struct sockbuf *sb, int flags)
  274 {
  275 
  276         KASSERT((flags & SBL_VALID) == flags,
  277             ("sblock: flags invalid (0x%x)", flags));
  278 
  279         if (flags & SBL_WAIT) {
  280                 if ((sb->sb_flags & SB_NOINTR) ||
  281                     (flags & SBL_NOINTR)) {
  282                         sx_xlock(&sb->sb_sx);
  283                         return (0);
  284                 }
  285                 return (sx_xlock_sig(&sb->sb_sx));
  286         } else {
  287                 if (sx_try_xlock(&sb->sb_sx) == 0)
  288                         return (EWOULDBLOCK);
  289                 return (0);
  290         }
  291 }
  292 
  293 void
  294 sbunlock(struct sockbuf *sb)
  295 {
  296 
  297         sx_xunlock(&sb->sb_sx);
  298 }
  299 
  300 /*
  301  * Wakeup processes waiting on a socket buffer.  Do asynchronous notification
  302  * via SIGIO if the socket has the SS_ASYNC flag set.
  303  *
  304  * Called with the socket buffer lock held; will release the lock by the end
  305  * of the function.  This allows the caller to acquire the socket buffer lock
  306  * while testing for the need for various sorts of wakeup and hold it through
  307  * to the point where it's no longer required.  We currently hold the lock
  308  * through calls out to other subsystems (with the exception of kqueue), and
  309  * then release it to avoid lock order issues.  It's not clear that's
  310  * correct.
  311  */
  312 void
  313 sowakeup(struct socket *so, struct sockbuf *sb)
  314 {
  315         int ret;
  316 
  317         SOCKBUF_LOCK_ASSERT(sb);
  318 
  319         selwakeuppri(sb->sb_sel, PSOCK);
  320         if (!SEL_WAITING(sb->sb_sel))
  321                 sb->sb_flags &= ~SB_SEL;
  322         if (sb->sb_flags & SB_WAIT) {
  323                 sb->sb_flags &= ~SB_WAIT;
  324                 wakeup(&sb->sb_acc);
  325         }
  326         KNOTE_LOCKED(&sb->sb_sel->si_note, 0);
  327         if (sb->sb_upcall != NULL) {
  328                 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT);
  329                 if (ret == SU_ISCONNECTED) {
  330                         KASSERT(sb == &so->so_rcv,
  331                             ("SO_SND upcall returned SU_ISCONNECTED"));
  332                         soupcall_clear(so, SO_RCV);
  333                 }
  334         } else
  335                 ret = SU_OK;
  336         if (sb->sb_flags & SB_AIO)
  337                 sowakeup_aio(so, sb);
  338         SOCKBUF_UNLOCK(sb);
  339         if (ret == SU_ISCONNECTED)
  340                 soisconnected(so);
  341         if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
  342                 pgsigio(&so->so_sigio, SIGIO, 0);
  343         mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
  344 }
  345 
  346 /*
  347  * Socket buffer (struct sockbuf) utility routines.
  348  *
  349  * Each socket contains two socket buffers: one for sending data and one for
  350  * receiving data.  Each buffer contains a queue of mbufs, information about
  351  * the number of mbufs and amount of data in the queue, and other fields
  352  * allowing select() statements and notification on data availability to be
  353  * implemented.
  354  *
  355  * Data stored in a socket buffer is maintained as a list of records.  Each
  356  * record is a list of mbufs chained together with the m_next field.  Records
  357  * are chained together with the m_nextpkt field. The upper level routine
  358  * soreceive() expects the following conventions to be observed when placing
  359  * information in the receive buffer:
  360  *
  361  * 1. If the protocol requires each message be preceded by the sender's name,
  362  *    then a record containing that name must be present before any
  363  *    associated data (mbuf's must be of type MT_SONAME).
  364  * 2. If the protocol supports the exchange of ``access rights'' (really just
  365  *    additional data associated with the message), and there are ``rights''
  366  *    to be received, then a record containing this data should be present
  367  *    (mbuf's must be of type MT_RIGHTS).
  368  * 3. If a name or rights record exists, then it must be followed by a data
  369  *    record, perhaps of zero length.
  370  *
  371  * Before using a new socket structure it is first necessary to reserve
  372  * buffer space to the socket, by calling sbreserve().  This should commit
  373  * some of the available buffer space in the system buffer pool for the
  374  * socket (currently, it does nothing but enforce limits).  The space should
  375  * be released by calling sbrelease() when the socket is destroyed.
  376  */
  377 int
  378 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
  379 {
  380         struct thread *td = curthread;
  381 
  382         SOCKBUF_LOCK(&so->so_snd);
  383         SOCKBUF_LOCK(&so->so_rcv);
  384         if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
  385                 goto bad;
  386         if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
  387                 goto bad2;
  388         if (so->so_rcv.sb_lowat == 0)
  389                 so->so_rcv.sb_lowat = 1;
  390         if (so->so_snd.sb_lowat == 0)
  391                 so->so_snd.sb_lowat = MCLBYTES;
  392         if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
  393                 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
  394         SOCKBUF_UNLOCK(&so->so_rcv);
  395         SOCKBUF_UNLOCK(&so->so_snd);
  396         return (0);
  397 bad2:
  398         sbrelease_locked(&so->so_snd, so);
  399 bad:
  400         SOCKBUF_UNLOCK(&so->so_rcv);
  401         SOCKBUF_UNLOCK(&so->so_snd);
  402         return (ENOBUFS);
  403 }
  404 
  405 static int
  406 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
  407 {
  408         int error = 0;
  409         u_long tmp_sb_max = sb_max;
  410 
  411         error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
  412         if (error || !req->newptr)
  413                 return (error);
  414         if (tmp_sb_max < MSIZE + MCLBYTES)
  415                 return (EINVAL);
  416         sb_max = tmp_sb_max;
  417         sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
  418         return (0);
  419 }
  420         
  421 /*
  422  * Allot mbufs to a sockbuf.  Attempt to scale mbmax so that mbcnt doesn't
  423  * become limiting if buffering efficiency is near the normal case.
  424  */
  425 int
  426 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
  427     struct thread *td)
  428 {
  429         rlim_t sbsize_limit;
  430 
  431         SOCKBUF_LOCK_ASSERT(sb);
  432 
  433         /*
  434          * When a thread is passed, we take into account the thread's socket
  435          * buffer size limit.  The caller will generally pass curthread, but
  436          * in the TCP input path, NULL will be passed to indicate that no
  437          * appropriate thread resource limits are available.  In that case,
  438          * we don't apply a process limit.
  439          */
  440         if (cc > sb_max_adj)
  441                 return (0);
  442         if (td != NULL) {
  443                 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE);
  444         } else
  445                 sbsize_limit = RLIM_INFINITY;
  446         if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
  447             sbsize_limit))
  448                 return (0);
  449         sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
  450         if (sb->sb_lowat > sb->sb_hiwat)
  451                 sb->sb_lowat = sb->sb_hiwat;
  452         return (1);
  453 }
  454 
  455 int
  456 sbsetopt(struct socket *so, int cmd, u_long cc)
  457 {
  458         struct sockbuf *sb;
  459         short *flags;
  460         u_int *hiwat, *lowat;
  461         int error;
  462 
  463         sb = NULL;
  464         SOCK_LOCK(so);
  465         if (SOLISTENING(so)) {
  466                 switch (cmd) {
  467                         case SO_SNDLOWAT:
  468                         case SO_SNDBUF:
  469                                 lowat = &so->sol_sbsnd_lowat;
  470                                 hiwat = &so->sol_sbsnd_hiwat;
  471                                 flags = &so->sol_sbsnd_flags;
  472                                 break;
  473                         case SO_RCVLOWAT:
  474                         case SO_RCVBUF:
  475                                 lowat = &so->sol_sbrcv_lowat;
  476                                 hiwat = &so->sol_sbrcv_hiwat;
  477                                 flags = &so->sol_sbrcv_flags;
  478                                 break;
  479                 }
  480         } else {
  481                 switch (cmd) {
  482                         case SO_SNDLOWAT:
  483                         case SO_SNDBUF:
  484                                 sb = &so->so_snd;
  485                                 break;
  486                         case SO_RCVLOWAT:
  487                         case SO_RCVBUF:
  488                                 sb = &so->so_rcv;
  489                                 break;
  490                 }
  491                 flags = &sb->sb_flags;
  492                 hiwat = &sb->sb_hiwat;
  493                 lowat = &sb->sb_lowat;
  494                 SOCKBUF_LOCK(sb);
  495         }
  496 
  497         error = 0;
  498         switch (cmd) {
  499         case SO_SNDBUF:
  500         case SO_RCVBUF:
  501                 if (SOLISTENING(so)) {
  502                         if (cc > sb_max_adj) {
  503                                 error = ENOBUFS;
  504                                 break;
  505                         }
  506                         *hiwat = cc;
  507                         if (*lowat > *hiwat)
  508                                 *lowat = *hiwat;
  509                 } else {
  510                         if (!sbreserve_locked(sb, cc, so, curthread))
  511                                 error = ENOBUFS;
  512                 }
  513                 if (error == 0)
  514                         *flags &= ~SB_AUTOSIZE;
  515                 break;
  516         case SO_SNDLOWAT:
  517         case SO_RCVLOWAT:
  518                 /*
  519                  * Make sure the low-water is never greater than the
  520                  * high-water.
  521                  */
  522                 *lowat = (cc > *hiwat) ? *hiwat : cc;
  523                 break;
  524         }
  525 
  526         if (!SOLISTENING(so))
  527                 SOCKBUF_UNLOCK(sb);
  528         SOCK_UNLOCK(so);
  529         return (error);
  530 }
  531 
  532 /*
  533  * Free mbufs held by a socket, and reserved mbuf space.
  534  */
  535 void
  536 sbrelease_internal(struct sockbuf *sb, struct socket *so)
  537 {
  538 
  539         sbflush_internal(sb);
  540         (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
  541             RLIM_INFINITY);
  542         sb->sb_mbmax = 0;
  543 }
  544 
  545 void
  546 sbrelease_locked(struct sockbuf *sb, struct socket *so)
  547 {
  548 
  549         SOCKBUF_LOCK_ASSERT(sb);
  550 
  551         sbrelease_internal(sb, so);
  552 }
  553 
  554 void
  555 sbrelease(struct sockbuf *sb, struct socket *so)
  556 {
  557 
  558         SOCKBUF_LOCK(sb);
  559         sbrelease_locked(sb, so);
  560         SOCKBUF_UNLOCK(sb);
  561 }
  562 
  563 void
  564 sbdestroy(struct sockbuf *sb, struct socket *so)
  565 {
  566 
  567         sbrelease_internal(sb, so);
  568 }
  569 
  570 /*
  571  * Routines to add and remove data from an mbuf queue.
  572  *
  573  * The routines sbappend() or sbappendrecord() are normally called to append
  574  * new mbufs to a socket buffer, after checking that adequate space is
  575  * available, comparing the function sbspace() with the amount of data to be
  576  * added.  sbappendrecord() differs from sbappend() in that data supplied is
  577  * treated as the beginning of a new record.  To place a sender's address,
  578  * optional access rights, and data in a socket receive buffer,
  579  * sbappendaddr() should be used.  To place access rights and data in a
  580  * socket receive buffer, sbappendrights() should be used.  In either case,
  581  * the new data begins a new record.  Note that unlike sbappend() and
  582  * sbappendrecord(), these routines check for the caller that there will be
  583  * enough space to store the data.  Each fails if there is not enough space,
  584  * or if it cannot find mbufs to store additional information in.
  585  *
  586  * Reliable protocols may use the socket send buffer to hold data awaiting
  587  * acknowledgement.  Data is normally copied from a socket send buffer in a
  588  * protocol with m_copy for output to a peer, and then removing the data from
  589  * the socket buffer with sbdrop() or sbdroprecord() when the data is
  590  * acknowledged by the peer.
  591  */
  592 #ifdef SOCKBUF_DEBUG
  593 void
  594 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
  595 {
  596         struct mbuf *m = sb->sb_mb;
  597 
  598         SOCKBUF_LOCK_ASSERT(sb);
  599 
  600         while (m && m->m_nextpkt)
  601                 m = m->m_nextpkt;
  602 
  603         if (m != sb->sb_lastrecord) {
  604                 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
  605                         __func__, sb->sb_mb, sb->sb_lastrecord, m);
  606                 printf("packet chain:\n");
  607                 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
  608                         printf("\t%p\n", m);
  609                 panic("%s from %s:%u", __func__, file, line);
  610         }
  611 }
  612 
  613 void
  614 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
  615 {
  616         struct mbuf *m = sb->sb_mb;
  617         struct mbuf *n;
  618 
  619         SOCKBUF_LOCK_ASSERT(sb);
  620 
  621         while (m && m->m_nextpkt)
  622                 m = m->m_nextpkt;
  623 
  624         while (m && m->m_next)
  625                 m = m->m_next;
  626 
  627         if (m != sb->sb_mbtail) {
  628                 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
  629                         __func__, sb->sb_mb, sb->sb_mbtail, m);
  630                 printf("packet tree:\n");
  631                 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
  632                         printf("\t");
  633                         for (n = m; n != NULL; n = n->m_next)
  634                                 printf("%p ", n);
  635                         printf("\n");
  636                 }
  637                 panic("%s from %s:%u", __func__, file, line);
  638         }
  639 }
  640 #endif /* SOCKBUF_DEBUG */
  641 
  642 #define SBLINKRECORD(sb, m0) do {                                       \
  643         SOCKBUF_LOCK_ASSERT(sb);                                        \
  644         if ((sb)->sb_lastrecord != NULL)                                \
  645                 (sb)->sb_lastrecord->m_nextpkt = (m0);                  \
  646         else                                                            \
  647                 (sb)->sb_mb = (m0);                                     \
  648         (sb)->sb_lastrecord = (m0);                                     \
  649 } while (/*CONSTCOND*/0)
  650 
  651 /*
  652  * Append mbuf chain m to the last record in the socket buffer sb.  The
  653  * additional space associated the mbuf chain is recorded in sb.  Empty mbufs
  654  * are discarded and mbufs are compacted where possible.
  655  */
  656 void
  657 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags)
  658 {
  659         struct mbuf *n;
  660 
  661         SOCKBUF_LOCK_ASSERT(sb);
  662 
  663         if (m == NULL)
  664                 return;
  665         sbm_clrprotoflags(m, flags);
  666         SBLASTRECORDCHK(sb);
  667         n = sb->sb_mb;
  668         if (n) {
  669                 while (n->m_nextpkt)
  670                         n = n->m_nextpkt;
  671                 do {
  672                         if (n->m_flags & M_EOR) {
  673                                 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
  674                                 return;
  675                         }
  676                 } while (n->m_next && (n = n->m_next));
  677         } else {
  678                 /*
  679                  * XXX Would like to simply use sb_mbtail here, but
  680                  * XXX I need to verify that I won't miss an EOR that
  681                  * XXX way.
  682                  */
  683                 if ((n = sb->sb_lastrecord) != NULL) {
  684                         do {
  685                                 if (n->m_flags & M_EOR) {
  686                                         sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
  687                                         return;
  688                                 }
  689                         } while (n->m_next && (n = n->m_next));
  690                 } else {
  691                         /*
  692                          * If this is the first record in the socket buffer,
  693                          * it's also the last record.
  694                          */
  695                         sb->sb_lastrecord = m;
  696                 }
  697         }
  698         sbcompress(sb, m, n);
  699         SBLASTRECORDCHK(sb);
  700 }
  701 
  702 /*
  703  * Append mbuf chain m to the last record in the socket buffer sb.  The
  704  * additional space associated the mbuf chain is recorded in sb.  Empty mbufs
  705  * are discarded and mbufs are compacted where possible.
  706  */
  707 void
  708 sbappend(struct sockbuf *sb, struct mbuf *m, int flags)
  709 {
  710 
  711         SOCKBUF_LOCK(sb);
  712         sbappend_locked(sb, m, flags);
  713         SOCKBUF_UNLOCK(sb);
  714 }
  715 
  716 /*
  717  * This version of sbappend() should only be used when the caller absolutely
  718  * knows that there will never be more than one record in the socket buffer,
  719  * that is, a stream protocol (such as TCP).
  720  */
  721 void
  722 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags)
  723 {
  724         SOCKBUF_LOCK_ASSERT(sb);
  725 
  726         KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
  727         KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
  728 
  729         SBLASTMBUFCHK(sb);
  730 
  731         /* Remove all packet headers and mbuf tags to get a pure data chain. */
  732         m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0);
  733 
  734         sbcompress(sb, m, sb->sb_mbtail);
  735 
  736         sb->sb_lastrecord = sb->sb_mb;
  737         SBLASTRECORDCHK(sb);
  738 }
  739 
  740 /*
  741  * This version of sbappend() should only be used when the caller absolutely
  742  * knows that there will never be more than one record in the socket buffer,
  743  * that is, a stream protocol (such as TCP).
  744  */
  745 void
  746 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags)
  747 {
  748 
  749         SOCKBUF_LOCK(sb);
  750         sbappendstream_locked(sb, m, flags);
  751         SOCKBUF_UNLOCK(sb);
  752 }
  753 
  754 #ifdef SOCKBUF_DEBUG
  755 void
  756 sbcheck(struct sockbuf *sb, const char *file, int line)
  757 {
  758         struct mbuf *m, *n, *fnrdy;
  759         u_long acc, ccc, mbcnt;
  760 
  761         SOCKBUF_LOCK_ASSERT(sb);
  762 
  763         acc = ccc = mbcnt = 0;
  764         fnrdy = NULL;
  765 
  766         for (m = sb->sb_mb; m; m = n) {
  767             n = m->m_nextpkt;
  768             for (; m; m = m->m_next) {
  769                 if (m->m_len == 0) {
  770                         printf("sb %p empty mbuf %p\n", sb, m);
  771                         goto fail;
  772                 }
  773                 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) {
  774                         if (m != sb->sb_fnrdy) {
  775                                 printf("sb %p: fnrdy %p != m %p\n",
  776                                     sb, sb->sb_fnrdy, m);
  777                                 goto fail;
  778                         }
  779                         fnrdy = m;
  780                 }
  781                 if (fnrdy) {
  782                         if (!(m->m_flags & M_NOTAVAIL)) {
  783                                 printf("sb %p: fnrdy %p, m %p is avail\n",
  784                                     sb, sb->sb_fnrdy, m);
  785                                 goto fail;
  786                         }
  787                 } else
  788                         acc += m->m_len;
  789                 ccc += m->m_len;
  790                 mbcnt += MSIZE;
  791                 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
  792                         mbcnt += m->m_ext.ext_size;
  793             }
  794         }
  795         if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) {
  796                 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n",
  797                     acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt);
  798                 goto fail;
  799         }
  800         return;
  801 fail:
  802         panic("%s from %s:%u", __func__, file, line);
  803 }
  804 #endif
  805 
  806 /*
  807  * As above, except the mbuf chain begins a new record.
  808  */
  809 void
  810 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
  811 {
  812         struct mbuf *m;
  813 
  814         SOCKBUF_LOCK_ASSERT(sb);
  815 
  816         if (m0 == NULL)
  817                 return;
  818         m_clrprotoflags(m0);
  819         /*
  820          * Put the first mbuf on the queue.  Note this permits zero length
  821          * records.
  822          */
  823         sballoc(sb, m0);
  824         SBLASTRECORDCHK(sb);
  825         SBLINKRECORD(sb, m0);
  826         sb->sb_mbtail = m0;
  827         m = m0->m_next;
  828         m0->m_next = 0;
  829         if (m && (m0->m_flags & M_EOR)) {
  830                 m0->m_flags &= ~M_EOR;
  831                 m->m_flags |= M_EOR;
  832         }
  833         /* always call sbcompress() so it can do SBLASTMBUFCHK() */
  834         sbcompress(sb, m, m0);
  835 }
  836 
  837 /*
  838  * As above, except the mbuf chain begins a new record.
  839  */
  840 void
  841 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
  842 {
  843 
  844         SOCKBUF_LOCK(sb);
  845         sbappendrecord_locked(sb, m0);
  846         SOCKBUF_UNLOCK(sb);
  847 }
  848 
  849 /* Helper routine that appends data, control, and address to a sockbuf. */
  850 static int
  851 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa,
  852     struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last)
  853 {
  854         struct mbuf *m, *n, *nlast;
  855 #if MSIZE <= 256
  856         if (asa->sa_len > MLEN)
  857                 return (0);
  858 #endif
  859         m = m_get(M_NOWAIT, MT_SONAME);
  860         if (m == NULL)
  861                 return (0);
  862         m->m_len = asa->sa_len;
  863         bcopy(asa, mtod(m, caddr_t), asa->sa_len);
  864         if (m0) {
  865                 m_clrprotoflags(m0);
  866                 m_tag_delete_chain(m0, NULL);
  867                 /*
  868                  * Clear some persistent info from pkthdr.
  869                  * We don't use m_demote(), because some netgraph consumers
  870                  * expect M_PKTHDR presence.
  871                  */
  872                 m0->m_pkthdr.rcvif = NULL;
  873                 m0->m_pkthdr.flowid = 0;
  874                 m0->m_pkthdr.csum_flags = 0;
  875                 m0->m_pkthdr.fibnum = 0;
  876                 m0->m_pkthdr.rsstype = 0;
  877         }
  878         if (ctrl_last)
  879                 ctrl_last->m_next = m0; /* concatenate data to control */
  880         else
  881                 control = m0;
  882         m->m_next = control;
  883         for (n = m; n->m_next != NULL; n = n->m_next)
  884                 sballoc(sb, n);
  885         sballoc(sb, n);
  886         nlast = n;
  887         SBLINKRECORD(sb, m);
  888 
  889         sb->sb_mbtail = nlast;
  890         SBLASTMBUFCHK(sb);
  891 
  892         SBLASTRECORDCHK(sb);
  893         return (1);
  894 }
  895 
  896 /*
  897  * Append address and data, and optionally, control (ancillary) data to the
  898  * receive queue of a socket.  If present, m0 must include a packet header
  899  * with total length.  Returns 0 if no space in sockbuf or insufficient
  900  * mbufs.
  901  */
  902 int
  903 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
  904     struct mbuf *m0, struct mbuf *control)
  905 {
  906         struct mbuf *ctrl_last;
  907         int space = asa->sa_len;
  908 
  909         SOCKBUF_LOCK_ASSERT(sb);
  910 
  911         if (m0 && (m0->m_flags & M_PKTHDR) == 0)
  912                 panic("sbappendaddr_locked");
  913         if (m0)
  914                 space += m0->m_pkthdr.len;
  915         space += m_length(control, &ctrl_last);
  916 
  917         if (space > sbspace(sb))
  918                 return (0);
  919         return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
  920 }
  921 
  922 /*
  923  * Append address and data, and optionally, control (ancillary) data to the
  924  * receive queue of a socket.  If present, m0 must include a packet header
  925  * with total length.  Returns 0 if insufficient mbufs.  Does not validate space
  926  * on the receiving sockbuf.
  927  */
  928 int
  929 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa,
  930     struct mbuf *m0, struct mbuf *control)
  931 {
  932         struct mbuf *ctrl_last;
  933 
  934         SOCKBUF_LOCK_ASSERT(sb);
  935 
  936         ctrl_last = (control == NULL) ? NULL : m_last(control);
  937         return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
  938 }
  939 
  940 /*
  941  * Append address and data, and optionally, control (ancillary) data to the
  942  * receive queue of a socket.  If present, m0 must include a packet header
  943  * with total length.  Returns 0 if no space in sockbuf or insufficient
  944  * mbufs.
  945  */
  946 int
  947 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
  948     struct mbuf *m0, struct mbuf *control)
  949 {
  950         int retval;
  951 
  952         SOCKBUF_LOCK(sb);
  953         retval = sbappendaddr_locked(sb, asa, m0, control);
  954         SOCKBUF_UNLOCK(sb);
  955         return (retval);
  956 }
  957 
  958 void
  959 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
  960     struct mbuf *control)
  961 {
  962         struct mbuf *m, *mlast;
  963 
  964         m_clrprotoflags(m0);
  965         m_last(control)->m_next = m0;
  966 
  967         SBLASTRECORDCHK(sb);
  968 
  969         for (m = control; m->m_next; m = m->m_next)
  970                 sballoc(sb, m);
  971         sballoc(sb, m);
  972         mlast = m;
  973         SBLINKRECORD(sb, control);
  974 
  975         sb->sb_mbtail = mlast;
  976         SBLASTMBUFCHK(sb);
  977 
  978         SBLASTRECORDCHK(sb);
  979 }
  980 
  981 void
  982 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
  983 {
  984 
  985         SOCKBUF_LOCK(sb);
  986         sbappendcontrol_locked(sb, m0, control);
  987         SOCKBUF_UNLOCK(sb);
  988 }
  989 
  990 /*
  991  * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
  992  * (n).  If (n) is NULL, the buffer is presumed empty.
  993  *
  994  * When the data is compressed, mbufs in the chain may be handled in one of
  995  * three ways:
  996  *
  997  * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
  998  *     record boundary, and no change in data type).
  999  *
 1000  * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
 1001  *     an mbuf already in the socket buffer.  This can occur if an
 1002  *     appropriate mbuf exists, there is room, both mbufs are not marked as
 1003  *     not ready, and no merging of data types will occur.
 1004  *
 1005  * (3) The mbuf may be appended to the end of the existing mbuf chain.
 1006  *
 1007  * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
 1008  * end-of-record.
 1009  */
 1010 void
 1011 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
 1012 {
 1013         int eor = 0;
 1014         struct mbuf *o;
 1015 
 1016         SOCKBUF_LOCK_ASSERT(sb);
 1017 
 1018         while (m) {
 1019                 eor |= m->m_flags & M_EOR;
 1020                 if (m->m_len == 0 &&
 1021                     (eor == 0 ||
 1022                      (((o = m->m_next) || (o = n)) &&
 1023                       o->m_type == m->m_type))) {
 1024                         if (sb->sb_lastrecord == m)
 1025                                 sb->sb_lastrecord = m->m_next;
 1026                         m = m_free(m);
 1027                         continue;
 1028                 }
 1029                 if (n && (n->m_flags & M_EOR) == 0 &&
 1030                     M_WRITABLE(n) &&
 1031                     ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
 1032                     !(m->m_flags & M_NOTREADY) &&
 1033                     !(n->m_flags & M_NOTREADY) &&
 1034                     m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
 1035                     m->m_len <= M_TRAILINGSPACE(n) &&
 1036                     n->m_type == m->m_type) {
 1037                         bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
 1038                             (unsigned)m->m_len);
 1039                         n->m_len += m->m_len;
 1040                         sb->sb_ccc += m->m_len;
 1041                         if (sb->sb_fnrdy == NULL)
 1042                                 sb->sb_acc += m->m_len;
 1043                         if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
 1044                                 /* XXX: Probably don't need.*/
 1045                                 sb->sb_ctl += m->m_len;
 1046                         m = m_free(m);
 1047                         continue;
 1048                 }
 1049                 if (n)
 1050                         n->m_next = m;
 1051                 else
 1052                         sb->sb_mb = m;
 1053                 sb->sb_mbtail = m;
 1054                 sballoc(sb, m);
 1055                 n = m;
 1056                 m->m_flags &= ~M_EOR;
 1057                 m = m->m_next;
 1058                 n->m_next = 0;
 1059         }
 1060         if (eor) {
 1061                 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
 1062                 n->m_flags |= eor;
 1063         }
 1064         SBLASTMBUFCHK(sb);
 1065 }
 1066 
 1067 /*
 1068  * Free all mbufs in a sockbuf.  Check that all resources are reclaimed.
 1069  */
 1070 static void
 1071 sbflush_internal(struct sockbuf *sb)
 1072 {
 1073 
 1074         while (sb->sb_mbcnt) {
 1075                 /*
 1076                  * Don't call sbcut(sb, 0) if the leading mbuf is non-empty:
 1077                  * we would loop forever. Panic instead.
 1078                  */
 1079                 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len))
 1080                         break;
 1081                 m_freem(sbcut_internal(sb, (int)sb->sb_ccc));
 1082         }
 1083         KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
 1084             ("%s: ccc %u mb %p mbcnt %u", __func__,
 1085             sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
 1086 }
 1087 
 1088 void
 1089 sbflush_locked(struct sockbuf *sb)
 1090 {
 1091 
 1092         SOCKBUF_LOCK_ASSERT(sb);
 1093         sbflush_internal(sb);
 1094 }
 1095 
 1096 void
 1097 sbflush(struct sockbuf *sb)
 1098 {
 1099 
 1100         SOCKBUF_LOCK(sb);
 1101         sbflush_locked(sb);
 1102         SOCKBUF_UNLOCK(sb);
 1103 }
 1104 
 1105 /*
 1106  * Cut data from (the front of) a sockbuf.
 1107  */
 1108 static struct mbuf *
 1109 sbcut_internal(struct sockbuf *sb, int len)
 1110 {
 1111         struct mbuf *m, *next, *mfree;
 1112 
 1113         KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0",
 1114             __func__, len));
 1115         KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u",
 1116             __func__, len, sb->sb_ccc));
 1117 
 1118         next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
 1119         mfree = NULL;
 1120 
 1121         while (len > 0) {
 1122                 if (m == NULL) {
 1123                         KASSERT(next, ("%s: no next, len %d", __func__, len));
 1124                         m = next;
 1125                         next = m->m_nextpkt;
 1126                 }
 1127                 if (m->m_len > len) {
 1128                         KASSERT(!(m->m_flags & M_NOTAVAIL),
 1129                             ("%s: m %p M_NOTAVAIL", __func__, m));
 1130                         m->m_len -= len;
 1131                         m->m_data += len;
 1132                         sb->sb_ccc -= len;
 1133                         sb->sb_acc -= len;
 1134                         if (sb->sb_sndptroff != 0)
 1135                                 sb->sb_sndptroff -= len;
 1136                         if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
 1137                                 sb->sb_ctl -= len;
 1138                         break;
 1139                 }
 1140                 len -= m->m_len;
 1141                 sbfree(sb, m);
 1142                 /*
 1143                  * Do not put M_NOTREADY buffers to the free list, they
 1144                  * are referenced from outside.
 1145                  */
 1146                 if (m->m_flags & M_NOTREADY)
 1147                         m = m->m_next;
 1148                 else {
 1149                         struct mbuf *n;
 1150 
 1151                         n = m->m_next;
 1152                         m->m_next = mfree;
 1153                         mfree = m;
 1154                         m = n;
 1155                 }
 1156         }
 1157         /*
 1158          * Free any zero-length mbufs from the buffer.
 1159          * For SOCK_DGRAM sockets such mbufs represent empty records.
 1160          * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer,
 1161          * when sosend_generic() needs to send only control data.
 1162          */
 1163         while (m && m->m_len == 0) {
 1164                 struct mbuf *n;
 1165 
 1166                 sbfree(sb, m);
 1167                 n = m->m_next;
 1168                 m->m_next = mfree;
 1169                 mfree = m;
 1170                 m = n;
 1171         }
 1172         if (m) {
 1173                 sb->sb_mb = m;
 1174                 m->m_nextpkt = next;
 1175         } else
 1176                 sb->sb_mb = next;
 1177         /*
 1178          * First part is an inline SB_EMPTY_FIXUP().  Second part makes sure
 1179          * sb_lastrecord is up-to-date if we dropped part of the last record.
 1180          */
 1181         m = sb->sb_mb;
 1182         if (m == NULL) {
 1183                 sb->sb_mbtail = NULL;
 1184                 sb->sb_lastrecord = NULL;
 1185         } else if (m->m_nextpkt == NULL) {
 1186                 sb->sb_lastrecord = m;
 1187         }
 1188 
 1189         return (mfree);
 1190 }
 1191 
 1192 /*
 1193  * Drop data from (the front of) a sockbuf.
 1194  */
 1195 void
 1196 sbdrop_locked(struct sockbuf *sb, int len)
 1197 {
 1198 
 1199         SOCKBUF_LOCK_ASSERT(sb);
 1200         m_freem(sbcut_internal(sb, len));
 1201 }
 1202 
 1203 /*
 1204  * Drop data from (the front of) a sockbuf,
 1205  * and return it to caller.
 1206  */
 1207 struct mbuf *
 1208 sbcut_locked(struct sockbuf *sb, int len)
 1209 {
 1210 
 1211         SOCKBUF_LOCK_ASSERT(sb);
 1212         return (sbcut_internal(sb, len));
 1213 }
 1214 
 1215 void
 1216 sbdrop(struct sockbuf *sb, int len)
 1217 {
 1218         struct mbuf *mfree;
 1219 
 1220         SOCKBUF_LOCK(sb);
 1221         mfree = sbcut_internal(sb, len);
 1222         SOCKBUF_UNLOCK(sb);
 1223 
 1224         m_freem(mfree);
 1225 }
 1226 
 1227 /*
 1228  * Maintain a pointer and offset pair into the socket buffer mbuf chain to
 1229  * avoid traversal of the entire socket buffer for larger offsets.
 1230  */
 1231 struct mbuf *
 1232 sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff)
 1233 {
 1234         struct mbuf *m, *ret;
 1235 
 1236         KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
 1237         KASSERT(off + len <= sb->sb_acc, ("%s: beyond sb", __func__));
 1238         KASSERT(sb->sb_sndptroff <= sb->sb_acc, ("%s: sndptroff broken", __func__));
 1239 
 1240         /*
 1241          * Is off below stored offset? Happens on retransmits.
 1242          * Just return, we can't help here.
 1243          */
 1244         if (sb->sb_sndptroff > off) {
 1245                 *moff = off;
 1246                 return (sb->sb_mb);
 1247         }
 1248 
 1249         /* Return closest mbuf in chain for current offset. */
 1250         *moff = off - sb->sb_sndptroff;
 1251         m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb;
 1252         if (*moff == m->m_len) {
 1253                 *moff = 0;
 1254                 sb->sb_sndptroff += m->m_len;
 1255                 m = ret = m->m_next;
 1256                 KASSERT(ret->m_len > 0,
 1257                     ("mbuf %p in sockbuf %p chain has no valid data", ret, sb));
 1258         }
 1259 
 1260         /* Advance by len to be as close as possible for the next transmit. */
 1261         for (off = off - sb->sb_sndptroff + len - 1;
 1262              off > 0 && m != NULL && off >= m->m_len;
 1263              m = m->m_next) {
 1264                 sb->sb_sndptroff += m->m_len;
 1265                 off -= m->m_len;
 1266         }
 1267         if (off > 0 && m == NULL)
 1268                 panic("%s: sockbuf %p and mbuf %p clashing", __func__, sb, ret);
 1269         sb->sb_sndptr = m;
 1270 
 1271         return (ret);
 1272 }
 1273 
 1274 struct mbuf *
 1275 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff)
 1276 {
 1277         struct mbuf *m;
 1278 
 1279         KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
 1280         if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
 1281                 *moff = off;
 1282                 if (sb->sb_sndptr == NULL) {
 1283                         sb->sb_sndptr = sb->sb_mb;
 1284                         sb->sb_sndptroff = 0;
 1285                 }
 1286                 return (sb->sb_mb);
 1287         } else {
 1288                 m = sb->sb_sndptr;
 1289                 off -= sb->sb_sndptroff;
 1290         }
 1291         *moff = off;
 1292         return (m);
 1293 }
 1294 
 1295 void
 1296 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len)
 1297 {
 1298         /*
 1299          * A small copy was done, advance forward the sb_sbsndptr to cover
 1300          * it.
 1301          */
 1302         struct mbuf *m;
 1303 
 1304         if (mb != sb->sb_sndptr) {
 1305                 /* Did not copyout at the same mbuf */
 1306                 return;
 1307         }
 1308         m = mb;
 1309         while (m && (len > 0)) {
 1310                 if (len >= m->m_len) {
 1311                         len -= m->m_len;
 1312                         if (m->m_next) {
 1313                                 sb->sb_sndptroff += m->m_len;
 1314                                 sb->sb_sndptr = m->m_next;
 1315                         }
 1316                         m = m->m_next;
 1317                 } else {
 1318                         len = 0;
 1319                 }
 1320         }
 1321 }
 1322 
 1323 /*
 1324  * Return the first mbuf and the mbuf data offset for the provided
 1325  * send offset without changing the "sb_sndptroff" field.
 1326  */
 1327 struct mbuf *
 1328 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff)
 1329 {
 1330         struct mbuf *m;
 1331 
 1332         KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
 1333 
 1334         /*
 1335          * If the "off" is below the stored offset, which happens on
 1336          * retransmits, just use "sb_mb":
 1337          */
 1338         if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
 1339                 m = sb->sb_mb;
 1340         } else {
 1341                 m = sb->sb_sndptr;
 1342                 off -= sb->sb_sndptroff;
 1343         }
 1344         while (off > 0 && m != NULL) {
 1345                 if (off < m->m_len)
 1346                         break;
 1347                 off -= m->m_len;
 1348                 m = m->m_next;
 1349         }
 1350         *moff = off;
 1351         return (m);
 1352 }
 1353 
 1354 /*
 1355  * Drop a record off the front of a sockbuf and move the next record to the
 1356  * front.
 1357  */
 1358 void
 1359 sbdroprecord_locked(struct sockbuf *sb)
 1360 {
 1361         struct mbuf *m;
 1362 
 1363         SOCKBUF_LOCK_ASSERT(sb);
 1364 
 1365         m = sb->sb_mb;
 1366         if (m) {
 1367                 sb->sb_mb = m->m_nextpkt;
 1368                 do {
 1369                         sbfree(sb, m);
 1370                         m = m_free(m);
 1371                 } while (m);
 1372         }
 1373         SB_EMPTY_FIXUP(sb);
 1374 }
 1375 
 1376 /*
 1377  * Drop a record off the front of a sockbuf and move the next record to the
 1378  * front.
 1379  */
 1380 void
 1381 sbdroprecord(struct sockbuf *sb)
 1382 {
 1383 
 1384         SOCKBUF_LOCK(sb);
 1385         sbdroprecord_locked(sb);
 1386         SOCKBUF_UNLOCK(sb);
 1387 }
 1388 
 1389 /*
 1390  * Create a "control" mbuf containing the specified data with the specified
 1391  * type for presentation on a socket buffer.
 1392  */
 1393 struct mbuf *
 1394 sbcreatecontrol(caddr_t p, int size, int type, int level)
 1395 {
 1396         struct cmsghdr *cp;
 1397         struct mbuf *m;
 1398 
 1399         if (CMSG_SPACE((u_int)size) > MCLBYTES)
 1400                 return ((struct mbuf *) NULL);
 1401         if (CMSG_SPACE((u_int)size) > MLEN)
 1402                 m = m_getcl(M_NOWAIT, MT_CONTROL, 0);
 1403         else
 1404                 m = m_get(M_NOWAIT, MT_CONTROL);
 1405         if (m == NULL)
 1406                 return ((struct mbuf *) NULL);
 1407         cp = mtod(m, struct cmsghdr *);
 1408         m->m_len = 0;
 1409         KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
 1410             ("sbcreatecontrol: short mbuf"));
 1411         /*
 1412          * Don't leave the padding between the msg header and the
 1413          * cmsg data and the padding after the cmsg data un-initialized.
 1414          */
 1415         bzero(cp, CMSG_SPACE((u_int)size));
 1416         if (p != NULL)
 1417                 (void)memcpy(CMSG_DATA(cp), p, size);
 1418         m->m_len = CMSG_SPACE(size);
 1419         cp->cmsg_len = CMSG_LEN(size);
 1420         cp->cmsg_level = level;
 1421         cp->cmsg_type = type;
 1422         return (m);
 1423 }
 1424 
 1425 /*
 1426  * This does the same for socket buffers that sotoxsocket does for sockets:
 1427  * generate an user-format data structure describing the socket buffer.  Note
 1428  * that the xsockbuf structure, since it is always embedded in a socket, does
 1429  * not include a self pointer nor a length.  We make this entry point public
 1430  * in case some other mechanism needs it.
 1431  */
 1432 void
 1433 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
 1434 {
 1435 
 1436         xsb->sb_cc = sb->sb_ccc;
 1437         xsb->sb_hiwat = sb->sb_hiwat;
 1438         xsb->sb_mbcnt = sb->sb_mbcnt;
 1439         xsb->sb_mcnt = sb->sb_mcnt;     
 1440         xsb->sb_ccnt = sb->sb_ccnt;
 1441         xsb->sb_mbmax = sb->sb_mbmax;
 1442         xsb->sb_lowat = sb->sb_lowat;
 1443         xsb->sb_flags = sb->sb_flags;
 1444         xsb->sb_timeo = sb->sb_timeo;
 1445 }
 1446 
 1447 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
 1448 static int dummy;
 1449 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
 1450 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
 1451     &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
 1452 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
 1453     &sb_efficiency, 0, "Socket buffer size waste factor");

Cache object: 3650f998b29d92ec5ca66192e22497d1


[ 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.