2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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.
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
31 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #include "opt_param.h"
39 #include <sys/param.h>
40 #include <sys/aio.h> /* for aio_swake proto */
41 #include <sys/kernel.h>
43 #include <sys/malloc.h>
45 #include <sys/mutex.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>
53 #include <sys/sysctl.h>
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.
59 void (*aio_swake)(struct socket *, struct sockbuf *);
62 * Primitive routines for operating on socket buffers
65 u_long sb_max = SB_MAX;
67 (quad_t)SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
69 static u_long sb_efficiency = 8; /* parameter for sbreserve() */
71 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len);
72 static void sbflush_internal(struct sockbuf *sb);
75 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY.
78 sbm_clrprotoflags(struct mbuf *m, int flags)
83 if (flags & PRUS_NOTREADY)
92 * Mark ready "count" mbufs starting with "m".
95 sbready(struct sockbuf *sb, struct mbuf *m, int count)
99 SOCKBUF_LOCK_ASSERT(sb);
100 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb));
102 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0;
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);
109 sb->sb_acc += m->m_len;
113 return (EINPROGRESS);
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;
129 * Adjust sockbuf state reflecting allocation of m.
132 sballoc(struct sockbuf *sb, struct mbuf *m)
135 SOCKBUF_LOCK_ASSERT(sb);
137 sb->sb_ccc += m->m_len;
139 if (sb->sb_fnrdy == NULL) {
140 if (m->m_flags & M_NOTREADY)
143 sb->sb_acc += m->m_len;
145 m->m_flags |= M_BLOCKED;
147 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
148 sb->sb_ctl += m->m_len;
150 sb->sb_mbcnt += MSIZE;
153 if (m->m_flags & M_EXT) {
154 sb->sb_mbcnt += m->m_ext.ext_size;
160 * Adjust sockbuf state reflecting freeing of m.
163 sbfree(struct sockbuf *sb, struct mbuf *m)
166 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
167 SOCKBUF_LOCK_ASSERT(sb);
170 sb->sb_ccc -= m->m_len;
172 if (!(m->m_flags & M_NOTAVAIL))
173 sb->sb_acc -= m->m_len;
175 if (m == sb->sb_fnrdy) {
178 KASSERT(m->m_flags & M_NOTREADY,
179 ("%s: m %p !M_NOTREADY", __func__, m));
182 while (n != NULL && !(n->m_flags & M_NOTREADY)) {
183 n->m_flags &= ~M_BLOCKED;
184 sb->sb_acc += n->m_len;
190 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
191 sb->sb_ctl -= m->m_len;
193 sb->sb_mbcnt -= MSIZE;
195 if (m->m_flags & M_EXT) {
196 sb->sb_mbcnt -= m->m_ext.ext_size;
200 if (sb->sb_sndptr == m) {
201 sb->sb_sndptr = NULL;
202 sb->sb_sndptroff = 0;
204 if (sb->sb_sndptroff != 0)
205 sb->sb_sndptroff -= m->m_len;
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.
218 socantsendmore_locked(struct socket *so)
221 SOCKBUF_LOCK_ASSERT(&so->so_snd);
223 so->so_snd.sb_state |= SBS_CANTSENDMORE;
224 sowwakeup_locked(so);
225 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
229 socantsendmore(struct socket *so)
232 SOCKBUF_LOCK(&so->so_snd);
233 socantsendmore_locked(so);
234 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
238 socantrcvmore_locked(struct socket *so)
241 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
243 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
244 sorwakeup_locked(so);
245 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
249 socantrcvmore(struct socket *so)
252 SOCKBUF_LOCK(&so->so_rcv);
253 socantrcvmore_locked(so);
254 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
258 * Wait for data to arrive at/drain from a socket buffer.
261 sbwait(struct sockbuf *sb)
264 SOCKBUF_LOCK_ASSERT(sb);
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));
273 sblock(struct sockbuf *sb, int flags)
276 KASSERT((flags & SBL_VALID) == flags,
277 ("sblock: flags invalid (0x%x)", flags));
279 if (flags & SBL_WAIT) {
280 if ((sb->sb_flags & SB_NOINTR) ||
281 (flags & SBL_NOINTR)) {
282 sx_xlock(&sb->sb_sx);
285 return (sx_xlock_sig(&sb->sb_sx));
287 if (sx_try_xlock(&sb->sb_sx) == 0)
288 return (EWOULDBLOCK);
294 sbunlock(struct sockbuf *sb)
297 sx_xunlock(&sb->sb_sx);
301 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
302 * via SIGIO if the socket has the SS_ASYNC flag set.
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
313 sowakeup(struct socket *so, struct sockbuf *sb)
317 SOCKBUF_LOCK_ASSERT(sb);
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;
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);
336 if (sb->sb_flags & SB_AIO)
337 sowakeup_aio(so, sb);
339 if (ret == SU_ISCONNECTED)
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);
347 * Socket buffer (struct sockbuf) utility routines.
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
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:
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.
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.
378 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
380 struct thread *td = curthread;
382 SOCKBUF_LOCK(&so->so_snd);
383 SOCKBUF_LOCK(&so->so_rcv);
384 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
386 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
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);
398 sbrelease_locked(&so->so_snd, so);
400 SOCKBUF_UNLOCK(&so->so_rcv);
401 SOCKBUF_UNLOCK(&so->so_snd);
406 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
409 u_long tmp_sb_max = sb_max;
411 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
412 if (error || !req->newptr)
414 if (tmp_sb_max < MSIZE + MCLBYTES)
417 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
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.
426 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
431 SOCKBUF_LOCK_ASSERT(sb);
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.
443 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE);
445 sbsize_limit = RLIM_INFINITY;
446 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
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;
456 sbsetopt(struct socket *so, int cmd, u_long cc)
460 u_int *hiwat, *lowat;
464 if (SOLISTENING(so)) {
468 lowat = &so->sol_sbsnd_lowat;
469 hiwat = &so->sol_sbsnd_hiwat;
470 flags = &so->sol_sbsnd_flags;
474 lowat = &so->sol_sbrcv_lowat;
475 hiwat = &so->sol_sbrcv_hiwat;
476 flags = &so->sol_sbrcv_flags;
490 flags = &sb->sb_flags;
491 hiwat = &sb->sb_hiwat;
492 lowat = &sb->sb_lowat;
500 if (SOLISTENING(so)) {
501 if (cc > sb_max_adj) {
509 if (!sbreserve_locked(sb, cc, so, curthread))
513 *flags &= ~SB_AUTOSIZE;
518 * Make sure the low-water is never greater than the
521 *lowat = (cc > *hiwat) ? *hiwat : cc;
525 if (!SOLISTENING(so))
532 * Free mbufs held by a socket, and reserved mbuf space.
535 sbrelease_internal(struct sockbuf *sb, struct socket *so)
538 sbflush_internal(sb);
539 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
545 sbrelease_locked(struct sockbuf *sb, struct socket *so)
548 SOCKBUF_LOCK_ASSERT(sb);
550 sbrelease_internal(sb, so);
554 sbrelease(struct sockbuf *sb, struct socket *so)
558 sbrelease_locked(sb, so);
563 sbdestroy(struct sockbuf *sb, struct socket *so)
566 sbrelease_internal(sb, so);
570 * Routines to add and remove data from an mbuf queue.
572 * The routines sbappend() or sbappendrecord() are normally called to append
573 * new mbufs to a socket buffer, after checking that adequate space is
574 * available, comparing the function sbspace() with the amount of data to be
575 * added. sbappendrecord() differs from sbappend() in that data supplied is
576 * treated as the beginning of a new record. To place a sender's address,
577 * optional access rights, and data in a socket receive buffer,
578 * sbappendaddr() should be used. To place access rights and data in a
579 * socket receive buffer, sbappendrights() should be used. In either case,
580 * the new data begins a new record. Note that unlike sbappend() and
581 * sbappendrecord(), these routines check for the caller that there will be
582 * enough space to store the data. Each fails if there is not enough space,
583 * or if it cannot find mbufs to store additional information in.
585 * Reliable protocols may use the socket send buffer to hold data awaiting
586 * acknowledgement. Data is normally copied from a socket send buffer in a
587 * protocol with m_copy for output to a peer, and then removing the data from
588 * the socket buffer with sbdrop() or sbdroprecord() when the data is
589 * acknowledged by the peer.
593 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
595 struct mbuf *m = sb->sb_mb;
597 SOCKBUF_LOCK_ASSERT(sb);
599 while (m && m->m_nextpkt)
602 if (m != sb->sb_lastrecord) {
603 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
604 __func__, sb->sb_mb, sb->sb_lastrecord, m);
605 printf("packet chain:\n");
606 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
608 panic("%s from %s:%u", __func__, file, line);
613 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
615 struct mbuf *m = sb->sb_mb;
618 SOCKBUF_LOCK_ASSERT(sb);
620 while (m && m->m_nextpkt)
623 while (m && m->m_next)
626 if (m != sb->sb_mbtail) {
627 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
628 __func__, sb->sb_mb, sb->sb_mbtail, m);
629 printf("packet tree:\n");
630 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
632 for (n = m; n != NULL; n = n->m_next)
636 panic("%s from %s:%u", __func__, file, line);
639 #endif /* SOCKBUF_DEBUG */
641 #define SBLINKRECORD(sb, m0) do { \
642 SOCKBUF_LOCK_ASSERT(sb); \
643 if ((sb)->sb_lastrecord != NULL) \
644 (sb)->sb_lastrecord->m_nextpkt = (m0); \
646 (sb)->sb_mb = (m0); \
647 (sb)->sb_lastrecord = (m0); \
648 } while (/*CONSTCOND*/0)
651 * Append mbuf chain m to the last record in the socket buffer sb. The
652 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
653 * are discarded and mbufs are compacted where possible.
656 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags)
660 SOCKBUF_LOCK_ASSERT(sb);
664 sbm_clrprotoflags(m, flags);
671 if (n->m_flags & M_EOR) {
672 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
675 } while (n->m_next && (n = n->m_next));
678 * XXX Would like to simply use sb_mbtail here, but
679 * XXX I need to verify that I won't miss an EOR that
682 if ((n = sb->sb_lastrecord) != NULL) {
684 if (n->m_flags & M_EOR) {
685 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
688 } while (n->m_next && (n = n->m_next));
691 * If this is the first record in the socket buffer,
692 * it's also the last record.
694 sb->sb_lastrecord = m;
697 sbcompress(sb, m, n);
702 * Append mbuf chain m to the last record in the socket buffer sb. The
703 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
704 * are discarded and mbufs are compacted where possible.
707 sbappend(struct sockbuf *sb, struct mbuf *m, int flags)
711 sbappend_locked(sb, m, flags);
716 * This version of sbappend() should only be used when the caller absolutely
717 * knows that there will never be more than one record in the socket buffer,
718 * that is, a stream protocol (such as TCP).
721 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags)
723 SOCKBUF_LOCK_ASSERT(sb);
725 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
726 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
730 /* Remove all packet headers and mbuf tags to get a pure data chain. */
731 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0);
733 sbcompress(sb, m, sb->sb_mbtail);
735 sb->sb_lastrecord = sb->sb_mb;
740 * This version of sbappend() should only be used when the caller absolutely
741 * knows that there will never be more than one record in the socket buffer,
742 * that is, a stream protocol (such as TCP).
745 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags)
749 sbappendstream_locked(sb, m, flags);
755 sbcheck(struct sockbuf *sb, const char *file, int line)
757 struct mbuf *m, *n, *fnrdy;
758 u_long acc, ccc, mbcnt;
760 SOCKBUF_LOCK_ASSERT(sb);
762 acc = ccc = mbcnt = 0;
765 for (m = sb->sb_mb; m; m = n) {
767 for (; m; m = m->m_next) {
769 printf("sb %p empty mbuf %p\n", sb, m);
772 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) {
773 if (m != sb->sb_fnrdy) {
774 printf("sb %p: fnrdy %p != m %p\n",
775 sb, sb->sb_fnrdy, m);
781 if (!(m->m_flags & M_NOTAVAIL)) {
782 printf("sb %p: fnrdy %p, m %p is avail\n",
783 sb, sb->sb_fnrdy, m);
790 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
791 mbcnt += m->m_ext.ext_size;
794 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) {
795 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n",
796 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt);
801 panic("%s from %s:%u", __func__, file, line);
806 * As above, except the mbuf chain begins a new record.
809 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
813 SOCKBUF_LOCK_ASSERT(sb);
819 * Put the first mbuf on the queue. Note this permits zero length
824 SBLINKRECORD(sb, m0);
828 if (m && (m0->m_flags & M_EOR)) {
829 m0->m_flags &= ~M_EOR;
832 /* always call sbcompress() so it can do SBLASTMBUFCHK() */
833 sbcompress(sb, m, m0);
837 * As above, except the mbuf chain begins a new record.
840 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
844 sbappendrecord_locked(sb, m0);
848 /* Helper routine that appends data, control, and address to a sockbuf. */
850 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa,
851 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last)
853 struct mbuf *m, *n, *nlast;
855 if (asa->sa_len > MLEN)
858 m = m_get(M_NOWAIT, MT_SONAME);
861 m->m_len = asa->sa_len;
862 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
865 m_tag_delete_chain(m0, NULL);
867 * Clear some persistent info from pkthdr.
868 * We don't use m_demote(), because some netgraph consumers
869 * expect M_PKTHDR presence.
871 m0->m_pkthdr.rcvif = NULL;
872 m0->m_pkthdr.flowid = 0;
873 m0->m_pkthdr.csum_flags = 0;
874 m0->m_pkthdr.fibnum = 0;
875 m0->m_pkthdr.rsstype = 0;
878 ctrl_last->m_next = m0; /* concatenate data to control */
882 for (n = m; n->m_next != NULL; n = n->m_next)
888 sb->sb_mbtail = nlast;
896 * Append address and data, and optionally, control (ancillary) data to the
897 * receive queue of a socket. If present, m0 must include a packet header
898 * with total length. Returns 0 if no space in sockbuf or insufficient
902 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
903 struct mbuf *m0, struct mbuf *control)
905 struct mbuf *ctrl_last;
906 int space = asa->sa_len;
908 SOCKBUF_LOCK_ASSERT(sb);
910 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
911 panic("sbappendaddr_locked");
913 space += m0->m_pkthdr.len;
914 space += m_length(control, &ctrl_last);
916 if (space > sbspace(sb))
918 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
922 * Append address and data, and optionally, control (ancillary) data to the
923 * receive queue of a socket. If present, m0 must include a packet header
924 * with total length. Returns 0 if insufficient mbufs. Does not validate space
925 * on the receiving sockbuf.
928 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa,
929 struct mbuf *m0, struct mbuf *control)
931 struct mbuf *ctrl_last;
933 SOCKBUF_LOCK_ASSERT(sb);
935 ctrl_last = (control == NULL) ? NULL : m_last(control);
936 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
940 * Append address and data, and optionally, control (ancillary) data to the
941 * receive queue of a socket. If present, m0 must include a packet header
942 * with total length. Returns 0 if no space in sockbuf or insufficient
946 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
947 struct mbuf *m0, struct mbuf *control)
952 retval = sbappendaddr_locked(sb, asa, m0, control);
958 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
959 struct mbuf *control)
961 struct mbuf *m, *n, *mlast;
964 SOCKBUF_LOCK_ASSERT(sb);
967 panic("sbappendcontrol_locked");
968 space = m_length(control, &n) + m_length(m0, NULL);
970 if (space > sbspace(sb))
973 n->m_next = m0; /* concatenate data to control */
977 for (m = control; m->m_next; m = m->m_next)
981 SBLINKRECORD(sb, control);
983 sb->sb_mbtail = mlast;
991 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
996 retval = sbappendcontrol_locked(sb, m0, control);
1002 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
1003 * (n). If (n) is NULL, the buffer is presumed empty.
1005 * When the data is compressed, mbufs in the chain may be handled in one of
1008 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
1009 * record boundary, and no change in data type).
1011 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
1012 * an mbuf already in the socket buffer. This can occur if an
1013 * appropriate mbuf exists, there is room, both mbufs are not marked as
1014 * not ready, and no merging of data types will occur.
1016 * (3) The mbuf may be appended to the end of the existing mbuf chain.
1018 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
1022 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1027 SOCKBUF_LOCK_ASSERT(sb);
1030 eor |= m->m_flags & M_EOR;
1031 if (m->m_len == 0 &&
1033 (((o = m->m_next) || (o = n)) &&
1034 o->m_type == m->m_type))) {
1035 if (sb->sb_lastrecord == m)
1036 sb->sb_lastrecord = m->m_next;
1040 if (n && (n->m_flags & M_EOR) == 0 &&
1042 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1043 !(m->m_flags & M_NOTREADY) &&
1044 !(n->m_flags & M_NOTREADY) &&
1045 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1046 m->m_len <= M_TRAILINGSPACE(n) &&
1047 n->m_type == m->m_type) {
1048 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
1049 (unsigned)m->m_len);
1050 n->m_len += m->m_len;
1051 sb->sb_ccc += m->m_len;
1052 if (sb->sb_fnrdy == NULL)
1053 sb->sb_acc += m->m_len;
1054 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1055 /* XXX: Probably don't need.*/
1056 sb->sb_ctl += m->m_len;
1067 m->m_flags &= ~M_EOR;
1072 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
1079 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
1082 sbflush_internal(struct sockbuf *sb)
1085 while (sb->sb_mbcnt) {
1087 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty:
1088 * we would loop forever. Panic instead.
1090 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len))
1092 m_freem(sbcut_internal(sb, (int)sb->sb_ccc));
1094 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
1095 ("%s: ccc %u mb %p mbcnt %u", __func__,
1096 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
1100 sbflush_locked(struct sockbuf *sb)
1103 SOCKBUF_LOCK_ASSERT(sb);
1104 sbflush_internal(sb);
1108 sbflush(struct sockbuf *sb)
1117 * Cut data from (the front of) a sockbuf.
1119 static struct mbuf *
1120 sbcut_internal(struct sockbuf *sb, int len)
1122 struct mbuf *m, *next, *mfree;
1124 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0",
1126 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u",
1127 __func__, len, sb->sb_ccc));
1129 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1134 KASSERT(next, ("%s: no next, len %d", __func__, len));
1136 next = m->m_nextpkt;
1138 if (m->m_len > len) {
1139 KASSERT(!(m->m_flags & M_NOTAVAIL),
1140 ("%s: m %p M_NOTAVAIL", __func__, m));
1145 if (sb->sb_sndptroff != 0)
1146 sb->sb_sndptroff -= len;
1147 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1154 * Do not put M_NOTREADY buffers to the free list, they
1155 * are referenced from outside.
1157 if (m->m_flags & M_NOTREADY)
1169 * Free any zero-length mbufs from the buffer.
1170 * For SOCK_DGRAM sockets such mbufs represent empty records.
1171 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer,
1172 * when sosend_generic() needs to send only control data.
1174 while (m && m->m_len == 0) {
1185 m->m_nextpkt = next;
1189 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
1190 * sb_lastrecord is up-to-date if we dropped part of the last record.
1194 sb->sb_mbtail = NULL;
1195 sb->sb_lastrecord = NULL;
1196 } else if (m->m_nextpkt == NULL) {
1197 sb->sb_lastrecord = m;
1204 * Drop data from (the front of) a sockbuf.
1207 sbdrop_locked(struct sockbuf *sb, int len)
1210 SOCKBUF_LOCK_ASSERT(sb);
1211 m_freem(sbcut_internal(sb, len));
1215 * Drop data from (the front of) a sockbuf,
1216 * and return it to caller.
1219 sbcut_locked(struct sockbuf *sb, int len)
1222 SOCKBUF_LOCK_ASSERT(sb);
1223 return (sbcut_internal(sb, len));
1227 sbdrop(struct sockbuf *sb, int len)
1232 mfree = sbcut_internal(sb, len);
1239 * Maintain a pointer and offset pair into the socket buffer mbuf chain to
1240 * avoid traversal of the entire socket buffer for larger offsets.
1243 sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff)
1245 struct mbuf *m, *ret;
1247 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1248 KASSERT(off + len <= sb->sb_acc, ("%s: beyond sb", __func__));
1249 KASSERT(sb->sb_sndptroff <= sb->sb_acc, ("%s: sndptroff broken", __func__));
1252 * Is off below stored offset? Happens on retransmits.
1253 * Just return, we can't help here.
1255 if (sb->sb_sndptroff > off) {
1260 /* Return closest mbuf in chain for current offset. */
1261 *moff = off - sb->sb_sndptroff;
1262 m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb;
1263 if (*moff == m->m_len) {
1265 sb->sb_sndptroff += m->m_len;
1266 m = ret = m->m_next;
1267 KASSERT(ret->m_len > 0,
1268 ("mbuf %p in sockbuf %p chain has no valid data", ret, sb));
1271 /* Advance by len to be as close as possible for the next transmit. */
1272 for (off = off - sb->sb_sndptroff + len - 1;
1273 off > 0 && m != NULL && off >= m->m_len;
1275 sb->sb_sndptroff += m->m_len;
1278 if (off > 0 && m == NULL)
1279 panic("%s: sockbuf %p and mbuf %p clashing", __func__, sb, ret);
1286 * Return the first mbuf and the mbuf data offset for the provided
1287 * send offset without changing the "sb_sndptroff" field.
1290 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff)
1294 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1297 * If the "off" is below the stored offset, which happens on
1298 * retransmits, just use "sb_mb":
1300 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1304 off -= sb->sb_sndptroff;
1306 while (off > 0 && m != NULL) {
1317 * Drop a record off the front of a sockbuf and move the next record to the
1321 sbdroprecord_locked(struct sockbuf *sb)
1325 SOCKBUF_LOCK_ASSERT(sb);
1329 sb->sb_mb = m->m_nextpkt;
1339 * Drop a record off the front of a sockbuf and move the next record to the
1343 sbdroprecord(struct sockbuf *sb)
1347 sbdroprecord_locked(sb);
1352 * Create a "control" mbuf containing the specified data with the specified
1353 * type for presentation on a socket buffer.
1356 sbcreatecontrol(caddr_t p, int size, int type, int level)
1361 if (CMSG_SPACE((u_int)size) > MCLBYTES)
1362 return ((struct mbuf *) NULL);
1363 if (CMSG_SPACE((u_int)size) > MLEN)
1364 m = m_getcl(M_NOWAIT, MT_CONTROL, 0);
1366 m = m_get(M_NOWAIT, MT_CONTROL);
1368 return ((struct mbuf *) NULL);
1369 cp = mtod(m, struct cmsghdr *);
1371 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1372 ("sbcreatecontrol: short mbuf"));
1374 * Don't leave the padding between the msg header and the
1375 * cmsg data and the padding after the cmsg data un-initialized.
1377 bzero(cp, CMSG_SPACE((u_int)size));
1379 (void)memcpy(CMSG_DATA(cp), p, size);
1380 m->m_len = CMSG_SPACE(size);
1381 cp->cmsg_len = CMSG_LEN(size);
1382 cp->cmsg_level = level;
1383 cp->cmsg_type = type;
1388 * This does the same for socket buffers that sotoxsocket does for sockets:
1389 * generate an user-format data structure describing the socket buffer. Note
1390 * that the xsockbuf structure, since it is always embedded in a socket, does
1391 * not include a self pointer nor a length. We make this entry point public
1392 * in case some other mechanism needs it.
1395 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1398 xsb->sb_cc = sb->sb_ccc;
1399 xsb->sb_hiwat = sb->sb_hiwat;
1400 xsb->sb_mbcnt = sb->sb_mbcnt;
1401 xsb->sb_mcnt = sb->sb_mcnt;
1402 xsb->sb_ccnt = sb->sb_ccnt;
1403 xsb->sb_mbmax = sb->sb_mbmax;
1404 xsb->sb_lowat = sb->sb_lowat;
1405 xsb->sb_flags = sb->sb_flags;
1406 xsb->sb_timeo = sb->sb_timeo;
1409 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1411 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1412 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1413 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1414 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1415 &sb_efficiency, 0, "Socket buffer size waste factor");