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 #include "opt_kern_tls.h"
36 #include "opt_param.h"
38 #include <sys/param.h>
39 #include <sys/aio.h> /* for aio_swake proto */
40 #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 #define BUF_MAX_ADJ(_sz) (((u_quad_t)(_sz)) * MCLBYTES / (MSIZE + MCLBYTES))
67 u_long sb_max = SB_MAX;
68 u_long sb_max_adj = BUF_MAX_ADJ(SB_MAX);
70 static u_long sb_efficiency = 8; /* parameter for sbreserve() */
73 static void sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m,
76 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len);
77 static void sbflush_internal(struct sockbuf *sb);
80 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY.
83 sbm_clrprotoflags(struct mbuf *m, int flags)
88 if (flags & PRUS_NOTREADY)
97 * Compress M_NOTREADY mbufs after they have been readied by sbready().
99 * sbcompress() skips M_NOTREADY mbufs since the data is not available to
100 * be copied at the time of sbcompress(). This function combines small
101 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first
102 * mbuf sbready() marked ready, and 'end' is the first mbuf still not
106 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end)
111 SOCKBUF_LOCK_ASSERT(sb);
113 if ((sb->sb_flags & SB_NOCOALESCE) != 0)
116 for (m = m0; m != end; m = m->m_next) {
117 MPASS((m->m_flags & M_NOTREADY) == 0);
119 * NB: In sbcompress(), 'n' is the last mbuf in the
120 * socket buffer and 'm' is the new mbuf being copied
121 * into the trailing space of 'n'. Here, the roles
122 * are reversed and 'n' is the next mbuf after 'm'
123 * that is being copied into the trailing space of
128 /* Try to coalesce adjacent ktls mbuf hdr/trailers. */
129 if ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 &&
130 (m->m_flags & M_EXTPG) &&
131 (n->m_flags & M_EXTPG) &&
132 !mbuf_has_tls_session(m) &&
133 !mbuf_has_tls_session(n)) {
134 int hdr_len, trail_len;
136 hdr_len = n->m_epg_hdrlen;
137 trail_len = m->m_epg_trllen;
138 if (trail_len != 0 && hdr_len != 0 &&
139 trail_len + hdr_len <= MBUF_PEXT_TRAIL_LEN) {
140 /* copy n's header to m's trailer */
141 memcpy(&m->m_epg_trail[trail_len],
142 n->m_epg_hdr, hdr_len);
143 m->m_epg_trllen += hdr_len;
151 /* Compress small unmapped mbufs into plain mbufs. */
152 if ((m->m_flags & M_EXTPG) && m->m_len <= MLEN &&
153 !mbuf_has_tls_session(m)) {
154 ext_size = m->m_ext.ext_size;
155 if (mb_unmapped_compress(m) == 0) {
156 sb->sb_mbcnt -= ext_size;
161 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 &&
163 (m->m_flags & M_EXTPG) == 0 &&
164 !mbuf_has_tls_session(n) &&
165 !mbuf_has_tls_session(m) &&
166 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
167 n->m_len <= M_TRAILINGSPACE(m) &&
168 m->m_type == n->m_type) {
169 KASSERT(sb->sb_lastrecord != n,
170 ("%s: merging start of record (%p) into previous mbuf (%p)",
172 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len));
173 m->m_len += n->m_len;
174 m->m_next = n->m_next;
175 m->m_flags |= n->m_flags & M_EOR;
176 if (sb->sb_mbtail == n)
179 sb->sb_mbcnt -= MSIZE;
181 if (n->m_flags & M_EXT) {
182 sb->sb_mbcnt -= n->m_ext.ext_size;
194 * Mark ready "count" units of I/O starting with "m". Most mbufs
195 * count as a single unit of I/O except for M_EXTPG mbufs which
196 * are backed by multiple pages.
199 sbready(struct sockbuf *sb, struct mbuf *m0, int count)
204 SOCKBUF_LOCK_ASSERT(sb);
205 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb));
206 KASSERT(count > 0, ("%s: invalid count %d", __func__, count));
209 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0;
212 KASSERT(m->m_flags & M_NOTREADY,
213 ("%s: m %p !M_NOTREADY", __func__, m));
214 if ((m->m_flags & M_EXTPG) != 0 && m->m_epg_npgs != 0) {
215 if (count < m->m_epg_nrdy) {
216 m->m_epg_nrdy -= count;
220 count -= m->m_epg_nrdy;
225 m->m_flags &= ~(M_NOTREADY | blocker);
227 sb->sb_acc += m->m_len;
232 * If the first mbuf is still not fully ready because only
233 * some of its backing pages were readied, no further progress
237 MPASS(m->m_flags & M_NOTREADY);
238 return (EINPROGRESS);
242 sbready_compress(sb, m0, m);
243 return (EINPROGRESS);
246 /* This one was blocking all the queue. */
247 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) {
248 KASSERT(m->m_flags & M_BLOCKED,
249 ("%s: m %p !M_BLOCKED", __func__, m));
250 m->m_flags &= ~M_BLOCKED;
251 sb->sb_acc += m->m_len;
255 sbready_compress(sb, m0, m);
261 * Adjust sockbuf state reflecting allocation of m.
264 sballoc(struct sockbuf *sb, struct mbuf *m)
267 SOCKBUF_LOCK_ASSERT(sb);
269 sb->sb_ccc += m->m_len;
271 if (sb->sb_fnrdy == NULL) {
272 if (m->m_flags & M_NOTREADY)
275 sb->sb_acc += m->m_len;
277 m->m_flags |= M_BLOCKED;
279 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
280 sb->sb_ctl += m->m_len;
282 sb->sb_mbcnt += MSIZE;
285 if (m->m_flags & M_EXT) {
286 sb->sb_mbcnt += m->m_ext.ext_size;
292 * Adjust sockbuf state reflecting freeing of m.
295 sbfree(struct sockbuf *sb, struct mbuf *m)
298 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
299 SOCKBUF_LOCK_ASSERT(sb);
302 sb->sb_ccc -= m->m_len;
304 if (!(m->m_flags & M_NOTAVAIL))
305 sb->sb_acc -= m->m_len;
307 if (m == sb->sb_fnrdy) {
310 KASSERT(m->m_flags & M_NOTREADY,
311 ("%s: m %p !M_NOTREADY", __func__, m));
314 while (n != NULL && !(n->m_flags & M_NOTREADY)) {
315 n->m_flags &= ~M_BLOCKED;
316 sb->sb_acc += n->m_len;
322 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
323 sb->sb_ctl -= m->m_len;
325 sb->sb_mbcnt -= MSIZE;
327 if (m->m_flags & M_EXT) {
328 sb->sb_mbcnt -= m->m_ext.ext_size;
332 if (sb->sb_sndptr == m) {
333 sb->sb_sndptr = NULL;
334 sb->sb_sndptroff = 0;
336 if (sb->sb_sndptroff != 0)
337 sb->sb_sndptroff -= m->m_len;
342 * Similar to sballoc/sbfree but does not adjust state associated with
343 * the sb_mb chain such as sb_fnrdy or sb_sndptr*. Also assumes mbufs
347 sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m)
350 SOCKBUF_LOCK_ASSERT(sb);
352 sb->sb_ccc += m->m_len;
353 sb->sb_tlscc += m->m_len;
355 sb->sb_mbcnt += MSIZE;
358 if (m->m_flags & M_EXT) {
359 sb->sb_mbcnt += m->m_ext.ext_size;
365 sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m)
368 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
369 SOCKBUF_LOCK_ASSERT(sb);
372 sb->sb_ccc -= m->m_len;
373 sb->sb_tlscc -= m->m_len;
375 sb->sb_mbcnt -= MSIZE;
378 if (m->m_flags & M_EXT) {
379 sb->sb_mbcnt -= m->m_ext.ext_size;
386 * Socantsendmore indicates that no more data will be sent on the socket; it
387 * would normally be applied to a socket when the user informs the system
388 * that no more data is to be sent, by the protocol code (in case
389 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
390 * received, and will normally be applied to the socket by a protocol when it
391 * detects that the peer will send no more data. Data queued for reading in
392 * the socket may yet be read.
395 socantsendmore_locked(struct socket *so)
398 SOCKBUF_LOCK_ASSERT(&so->so_snd);
400 so->so_snd.sb_state |= SBS_CANTSENDMORE;
401 sowwakeup_locked(so);
402 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
406 socantsendmore(struct socket *so)
409 SOCKBUF_LOCK(&so->so_snd);
410 socantsendmore_locked(so);
411 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
415 socantrcvmore_locked(struct socket *so)
418 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
420 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
422 if (so->so_rcv.sb_flags & SB_TLS_RX)
423 ktls_check_rx(&so->so_rcv);
425 sorwakeup_locked(so);
426 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
430 socantrcvmore(struct socket *so)
433 SOCKBUF_LOCK(&so->so_rcv);
434 socantrcvmore_locked(so);
435 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
439 soroverflow_locked(struct socket *so)
442 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
444 if (so->so_options & SO_RERROR) {
445 so->so_rerror = ENOBUFS;
446 sorwakeup_locked(so);
448 SOCKBUF_UNLOCK(&so->so_rcv);
450 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
454 soroverflow(struct socket *so)
457 SOCKBUF_LOCK(&so->so_rcv);
458 soroverflow_locked(so);
459 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
463 * Wait for data to arrive at/drain from a socket buffer.
466 sbwait(struct sockbuf *sb)
469 SOCKBUF_LOCK_ASSERT(sb);
471 sb->sb_flags |= SB_WAIT;
472 return (msleep_sbt(&sb->sb_acc, SOCKBUF_MTX(sb),
473 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
474 sb->sb_timeo, 0, 0));
478 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
479 * via SIGIO if the socket has the SS_ASYNC flag set.
481 * Called with the socket buffer lock held; will release the lock by the end
482 * of the function. This allows the caller to acquire the socket buffer lock
483 * while testing for the need for various sorts of wakeup and hold it through
484 * to the point where it's no longer required. We currently hold the lock
485 * through calls out to other subsystems (with the exception of kqueue), and
486 * then release it to avoid lock order issues. It's not clear that's
490 sowakeup(struct socket *so, struct sockbuf *sb)
494 SOCKBUF_LOCK_ASSERT(sb);
496 selwakeuppri(sb->sb_sel, PSOCK);
497 if (!SEL_WAITING(sb->sb_sel))
498 sb->sb_flags &= ~SB_SEL;
499 if (sb->sb_flags & SB_WAIT) {
500 sb->sb_flags &= ~SB_WAIT;
503 KNOTE_LOCKED(&sb->sb_sel->si_note, 0);
504 if (sb->sb_upcall != NULL) {
505 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT);
506 if (ret == SU_ISCONNECTED) {
507 KASSERT(sb == &so->so_rcv,
508 ("SO_SND upcall returned SU_ISCONNECTED"));
509 soupcall_clear(so, SO_RCV);
513 if (sb->sb_flags & SB_AIO)
514 sowakeup_aio(so, sb);
516 if (ret == SU_ISCONNECTED)
518 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
519 pgsigio(&so->so_sigio, SIGIO, 0);
520 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
524 * Socket buffer (struct sockbuf) utility routines.
526 * Each socket contains two socket buffers: one for sending data and one for
527 * receiving data. Each buffer contains a queue of mbufs, information about
528 * the number of mbufs and amount of data in the queue, and other fields
529 * allowing select() statements and notification on data availability to be
532 * Data stored in a socket buffer is maintained as a list of records. Each
533 * record is a list of mbufs chained together with the m_next field. Records
534 * are chained together with the m_nextpkt field. The upper level routine
535 * soreceive() expects the following conventions to be observed when placing
536 * information in the receive buffer:
538 * 1. If the protocol requires each message be preceded by the sender's name,
539 * then a record containing that name must be present before any
540 * associated data (mbuf's must be of type MT_SONAME).
541 * 2. If the protocol supports the exchange of ``access rights'' (really just
542 * additional data associated with the message), and there are ``rights''
543 * to be received, then a record containing this data should be present
544 * (mbuf's must be of type MT_RIGHTS).
545 * 3. If a name or rights record exists, then it must be followed by a data
546 * record, perhaps of zero length.
548 * Before using a new socket structure it is first necessary to reserve
549 * buffer space to the socket, by calling sbreserve(). This should commit
550 * some of the available buffer space in the system buffer pool for the
551 * socket (currently, it does nothing but enforce limits). The space should
552 * be released by calling sbrelease() when the socket is destroyed.
555 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
557 struct thread *td = curthread;
559 SOCKBUF_LOCK(&so->so_snd);
560 SOCKBUF_LOCK(&so->so_rcv);
561 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
563 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
565 if (so->so_rcv.sb_lowat == 0)
566 so->so_rcv.sb_lowat = 1;
567 if (so->so_snd.sb_lowat == 0)
568 so->so_snd.sb_lowat = MCLBYTES;
569 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
570 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
571 SOCKBUF_UNLOCK(&so->so_rcv);
572 SOCKBUF_UNLOCK(&so->so_snd);
575 sbrelease_locked(&so->so_snd, so);
577 SOCKBUF_UNLOCK(&so->so_rcv);
578 SOCKBUF_UNLOCK(&so->so_snd);
583 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
586 u_long tmp_sb_max = sb_max;
588 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
589 if (error || !req->newptr)
591 if (tmp_sb_max < MSIZE + MCLBYTES)
594 sb_max_adj = BUF_MAX_ADJ(sb_max);
599 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
600 * become limiting if buffering efficiency is near the normal case.
603 sbreserve_locked_limit(struct sockbuf *sb, u_long cc, struct socket *so,
604 u_long buf_max, struct thread *td)
608 SOCKBUF_LOCK_ASSERT(sb);
611 * When a thread is passed, we take into account the thread's socket
612 * buffer size limit. The caller will generally pass curthread, but
613 * in the TCP input path, NULL will be passed to indicate that no
614 * appropriate thread resource limits are available. In that case,
615 * we don't apply a process limit.
617 if (cc > BUF_MAX_ADJ(buf_max))
620 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE);
622 sbsize_limit = RLIM_INFINITY;
623 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
626 sb->sb_mbmax = min(cc * sb_efficiency, buf_max);
627 if (sb->sb_lowat > sb->sb_hiwat)
628 sb->sb_lowat = sb->sb_hiwat;
633 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
636 return (sbreserve_locked_limit(sb, cc, so, sb_max, td));
640 sbsetopt(struct socket *so, int cmd, u_long cc)
644 u_int *hiwat, *lowat;
649 if (SOLISTENING(so)) {
653 lowat = &so->sol_sbsnd_lowat;
654 hiwat = &so->sol_sbsnd_hiwat;
655 flags = &so->sol_sbsnd_flags;
659 lowat = &so->sol_sbrcv_lowat;
660 hiwat = &so->sol_sbrcv_hiwat;
661 flags = &so->sol_sbrcv_flags;
675 flags = &sb->sb_flags;
676 hiwat = &sb->sb_hiwat;
677 lowat = &sb->sb_lowat;
685 if (SOLISTENING(so)) {
686 if (cc > sb_max_adj) {
694 u_long limit = sogetmaxbuf(so);
695 if (!sbreserve_locked_limit(sb, cc, so, limit, curthread))
699 *flags &= ~SB_AUTOSIZE;
704 * Make sure the low-water is never greater than the
707 *lowat = (cc > *hiwat) ? *hiwat : cc;
711 if (!SOLISTENING(so))
718 * Free mbufs held by a socket, and reserved mbuf space.
721 sbrelease_internal(struct sockbuf *sb, struct socket *so)
724 sbflush_internal(sb);
725 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
731 sbrelease_locked(struct sockbuf *sb, struct socket *so)
734 SOCKBUF_LOCK_ASSERT(sb);
736 sbrelease_internal(sb, so);
740 sbrelease(struct sockbuf *sb, struct socket *so)
744 sbrelease_locked(sb, so);
749 sbdestroy(struct sockbuf *sb, struct socket *so)
752 sbrelease_internal(sb, so);
754 if (sb->sb_tls_info != NULL)
755 ktls_free(sb->sb_tls_info);
756 sb->sb_tls_info = NULL;
761 * Routines to add and remove data from an mbuf queue.
763 * The routines sbappend() or sbappendrecord() are normally called to append
764 * new mbufs to a socket buffer, after checking that adequate space is
765 * available, comparing the function sbspace() with the amount of data to be
766 * added. sbappendrecord() differs from sbappend() in that data supplied is
767 * treated as the beginning of a new record. To place a sender's address,
768 * optional access rights, and data in a socket receive buffer,
769 * sbappendaddr() should be used. To place access rights and data in a
770 * socket receive buffer, sbappendrights() should be used. In either case,
771 * the new data begins a new record. Note that unlike sbappend() and
772 * sbappendrecord(), these routines check for the caller that there will be
773 * enough space to store the data. Each fails if there is not enough space,
774 * or if it cannot find mbufs to store additional information in.
776 * Reliable protocols may use the socket send buffer to hold data awaiting
777 * acknowledgement. Data is normally copied from a socket send buffer in a
778 * protocol with m_copy for output to a peer, and then removing the data from
779 * the socket buffer with sbdrop() or sbdroprecord() when the data is
780 * acknowledged by the peer.
784 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
786 struct mbuf *m = sb->sb_mb;
788 SOCKBUF_LOCK_ASSERT(sb);
790 while (m && m->m_nextpkt)
793 if (m != sb->sb_lastrecord) {
794 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
795 __func__, sb->sb_mb, sb->sb_lastrecord, m);
796 printf("packet chain:\n");
797 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
799 panic("%s from %s:%u", __func__, file, line);
804 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
806 struct mbuf *m = sb->sb_mb;
809 SOCKBUF_LOCK_ASSERT(sb);
811 while (m && m->m_nextpkt)
814 while (m && m->m_next)
817 if (m != sb->sb_mbtail) {
818 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
819 __func__, sb->sb_mb, sb->sb_mbtail, m);
820 printf("packet tree:\n");
821 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
823 for (n = m; n != NULL; n = n->m_next)
827 panic("%s from %s:%u", __func__, file, line);
832 while (m && m->m_next)
835 if (m != sb->sb_mtlstail) {
836 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n",
837 __func__, sb->sb_mtls, sb->sb_mtlstail, m);
838 printf("TLS packet tree:\n");
840 for (m = sb->sb_mtls; m != NULL; m = m->m_next) {
844 panic("%s from %s:%u", __func__, file, line);
848 #endif /* SOCKBUF_DEBUG */
850 #define SBLINKRECORD(sb, m0) do { \
851 SOCKBUF_LOCK_ASSERT(sb); \
852 if ((sb)->sb_lastrecord != NULL) \
853 (sb)->sb_lastrecord->m_nextpkt = (m0); \
855 (sb)->sb_mb = (m0); \
856 (sb)->sb_lastrecord = (m0); \
857 } while (/*CONSTCOND*/0)
860 * Append mbuf chain m to the last record in the socket buffer sb. The
861 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
862 * are discarded and mbufs are compacted where possible.
865 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags)
869 SOCKBUF_LOCK_ASSERT(sb);
873 sbm_clrprotoflags(m, flags);
880 if (n->m_flags & M_EOR) {
881 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
884 } while (n->m_next && (n = n->m_next));
887 * XXX Would like to simply use sb_mbtail here, but
888 * XXX I need to verify that I won't miss an EOR that
891 if ((n = sb->sb_lastrecord) != NULL) {
893 if (n->m_flags & M_EOR) {
894 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
897 } while (n->m_next && (n = n->m_next));
900 * If this is the first record in the socket buffer,
901 * it's also the last record.
903 sb->sb_lastrecord = m;
906 sbcompress(sb, m, n);
911 * Append mbuf chain m to the last record in the socket buffer sb. The
912 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
913 * are discarded and mbufs are compacted where possible.
916 sbappend(struct sockbuf *sb, struct mbuf *m, int flags)
920 sbappend_locked(sb, m, flags);
926 * Append an mbuf containing encrypted TLS data. The data
927 * is marked M_NOTREADY until it has been decrypted and
928 * stored as a TLS record.
931 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m)
937 /* Remove all packet headers and mbuf tags to get a pure data chain. */
940 for (n = m; n != NULL; n = n->m_next)
941 n->m_flags |= M_NOTREADY;
942 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail);
948 * This version of sbappend() should only be used when the caller absolutely
949 * knows that there will never be more than one record in the socket buffer,
950 * that is, a stream protocol (such as TCP).
953 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags)
955 SOCKBUF_LOCK_ASSERT(sb);
957 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
961 * Decrypted TLS records are appended as records via
962 * sbappendrecord(). TCP passes encrypted TLS records to this
963 * function which must be scheduled for decryption.
965 if (sb->sb_flags & SB_TLS_RX) {
966 sbappend_ktls_rx(sb, m);
971 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
976 if (sb->sb_tls_info != NULL)
980 /* Remove all packet headers and mbuf tags to get a pure data chain. */
981 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0);
983 sbcompress(sb, m, sb->sb_mbtail);
985 sb->sb_lastrecord = sb->sb_mb;
990 * This version of sbappend() should only be used when the caller absolutely
991 * knows that there will never be more than one record in the socket buffer,
992 * that is, a stream protocol (such as TCP).
995 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags)
999 sbappendstream_locked(sb, m, flags);
1003 #ifdef SOCKBUF_DEBUG
1005 sbcheck(struct sockbuf *sb, const char *file, int line)
1007 struct mbuf *m, *n, *fnrdy;
1008 u_long acc, ccc, mbcnt;
1013 SOCKBUF_LOCK_ASSERT(sb);
1015 acc = ccc = mbcnt = 0;
1018 for (m = sb->sb_mb; m; m = n) {
1020 for (; m; m = m->m_next) {
1021 if (m->m_len == 0) {
1022 printf("sb %p empty mbuf %p\n", sb, m);
1025 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) {
1026 if (m != sb->sb_fnrdy) {
1027 printf("sb %p: fnrdy %p != m %p\n",
1028 sb, sb->sb_fnrdy, m);
1034 if (!(m->m_flags & M_NOTAVAIL)) {
1035 printf("sb %p: fnrdy %p, m %p is avail\n",
1036 sb, sb->sb_fnrdy, m);
1043 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
1044 mbcnt += m->m_ext.ext_size;
1049 * Account for mbufs "detached" by ktls_detach_record() while
1050 * they are decrypted by ktls_decrypt(). tlsdcc gives a count
1051 * of the detached bytes that are included in ccc. The mbufs
1052 * and clusters are not included in the socket buffer
1055 ccc += sb->sb_tlsdcc;
1058 for (m = sb->sb_mtls; m; m = m->m_next) {
1059 if (m->m_nextpkt != NULL) {
1060 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m);
1063 if ((m->m_flags & M_NOTREADY) == 0) {
1064 printf("sb %p TLS mbuf %p ready\n", sb, m);
1070 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
1071 mbcnt += m->m_ext.ext_size;
1074 if (sb->sb_tlscc != tlscc) {
1075 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc,
1080 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) {
1081 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n",
1082 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt);
1084 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc,
1091 panic("%s from %s:%u", __func__, file, line);
1096 * As above, except the mbuf chain begins a new record.
1099 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
1103 SOCKBUF_LOCK_ASSERT(sb);
1107 m_clrprotoflags(m0);
1109 * Put the first mbuf on the queue. Note this permits zero length
1113 SBLASTRECORDCHK(sb);
1114 SBLINKRECORD(sb, m0);
1118 if (m && (m0->m_flags & M_EOR)) {
1119 m0->m_flags &= ~M_EOR;
1120 m->m_flags |= M_EOR;
1122 /* always call sbcompress() so it can do SBLASTMBUFCHK() */
1123 sbcompress(sb, m, m0);
1127 * As above, except the mbuf chain begins a new record.
1130 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
1134 sbappendrecord_locked(sb, m0);
1138 /* Helper routine that appends data, control, and address to a sockbuf. */
1140 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa,
1141 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last)
1143 struct mbuf *m, *n, *nlast;
1145 if (asa->sa_len > MLEN)
1148 m = m_get(M_NOWAIT, MT_SONAME);
1151 m->m_len = asa->sa_len;
1152 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
1154 M_ASSERT_NO_SND_TAG(m0);
1155 m_clrprotoflags(m0);
1156 m_tag_delete_chain(m0, NULL);
1158 * Clear some persistent info from pkthdr.
1159 * We don't use m_demote(), because some netgraph consumers
1160 * expect M_PKTHDR presence.
1162 m0->m_pkthdr.rcvif = NULL;
1163 m0->m_pkthdr.flowid = 0;
1164 m0->m_pkthdr.csum_flags = 0;
1165 m0->m_pkthdr.fibnum = 0;
1166 m0->m_pkthdr.rsstype = 0;
1169 ctrl_last->m_next = m0; /* concatenate data to control */
1172 m->m_next = control;
1173 for (n = m; n->m_next != NULL; n = n->m_next)
1177 SBLINKRECORD(sb, m);
1179 sb->sb_mbtail = nlast;
1182 SBLASTRECORDCHK(sb);
1187 * Append address and data, and optionally, control (ancillary) data to the
1188 * receive queue of a socket. If present, m0 must include a packet header
1189 * with total length. Returns 0 if no space in sockbuf or insufficient
1193 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
1194 struct mbuf *m0, struct mbuf *control)
1196 struct mbuf *ctrl_last;
1197 int space = asa->sa_len;
1199 SOCKBUF_LOCK_ASSERT(sb);
1201 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
1202 panic("sbappendaddr_locked");
1204 space += m0->m_pkthdr.len;
1205 space += m_length(control, &ctrl_last);
1207 if (space > sbspace(sb))
1209 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
1213 * Append address and data, and optionally, control (ancillary) data to the
1214 * receive queue of a socket. If present, m0 must include a packet header
1215 * with total length. Returns 0 if insufficient mbufs. Does not validate space
1216 * on the receiving sockbuf.
1219 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa,
1220 struct mbuf *m0, struct mbuf *control)
1222 struct mbuf *ctrl_last;
1224 SOCKBUF_LOCK_ASSERT(sb);
1226 ctrl_last = (control == NULL) ? NULL : m_last(control);
1227 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
1231 * Append address and data, and optionally, control (ancillary) data to the
1232 * receive queue of a socket. If present, m0 must include a packet header
1233 * with total length. Returns 0 if no space in sockbuf or insufficient
1237 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
1238 struct mbuf *m0, struct mbuf *control)
1243 retval = sbappendaddr_locked(sb, asa, m0, control);
1249 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
1250 struct mbuf *control, int flags)
1252 struct mbuf *m, *mlast;
1254 sbm_clrprotoflags(m0, flags);
1255 m_last(control)->m_next = m0;
1257 SBLASTRECORDCHK(sb);
1259 for (m = control; m->m_next; m = m->m_next)
1263 SBLINKRECORD(sb, control);
1265 sb->sb_mbtail = mlast;
1268 SBLASTRECORDCHK(sb);
1272 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control,
1277 sbappendcontrol_locked(sb, m0, control, flags);
1282 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
1283 * (n). If (n) is NULL, the buffer is presumed empty.
1285 * When the data is compressed, mbufs in the chain may be handled in one of
1288 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
1289 * record boundary, and no change in data type).
1291 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
1292 * an mbuf already in the socket buffer. This can occur if an
1293 * appropriate mbuf exists, there is room, both mbufs are not marked as
1294 * not ready, and no merging of data types will occur.
1296 * (3) The mbuf may be appended to the end of the existing mbuf chain.
1298 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
1302 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1307 SOCKBUF_LOCK_ASSERT(sb);
1310 eor |= m->m_flags & M_EOR;
1311 if (m->m_len == 0 &&
1313 (((o = m->m_next) || (o = n)) &&
1314 o->m_type == m->m_type))) {
1315 if (sb->sb_lastrecord == m)
1316 sb->sb_lastrecord = m->m_next;
1320 if (n && (n->m_flags & M_EOR) == 0 &&
1322 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1323 !(m->m_flags & M_NOTREADY) &&
1324 !(n->m_flags & (M_NOTREADY | M_EXTPG)) &&
1325 !mbuf_has_tls_session(m) &&
1326 !mbuf_has_tls_session(n) &&
1327 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1328 m->m_len <= M_TRAILINGSPACE(n) &&
1329 n->m_type == m->m_type) {
1330 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len));
1331 n->m_len += m->m_len;
1332 sb->sb_ccc += m->m_len;
1333 if (sb->sb_fnrdy == NULL)
1334 sb->sb_acc += m->m_len;
1335 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1336 /* XXX: Probably don't need.*/
1337 sb->sb_ctl += m->m_len;
1341 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) &&
1342 (m->m_flags & M_NOTREADY) == 0 &&
1343 !mbuf_has_tls_session(m))
1344 (void)mb_unmapped_compress(m);
1352 m->m_flags &= ~M_EOR;
1357 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
1365 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs
1366 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also
1367 * a bit simpler (no EOR markers, always MT_DATA, etc.).
1370 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1373 SOCKBUF_LOCK_ASSERT(sb);
1376 KASSERT((m->m_flags & M_EOR) == 0,
1377 ("TLS RX mbuf %p with EOR", m));
1378 KASSERT(m->m_type == MT_DATA,
1379 ("TLS RX mbuf %p is not MT_DATA", m));
1380 KASSERT((m->m_flags & M_NOTREADY) != 0,
1381 ("TLS RX mbuf %p ready", m));
1382 KASSERT((m->m_flags & M_EXTPG) == 0,
1383 ("TLS RX mbuf %p unmapped", m));
1385 if (m->m_len == 0) {
1391 * Even though both 'n' and 'm' are NOTREADY, it's ok
1392 * to coalesce the data.
1396 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1397 !(n->m_flags & (M_EXTPG)) &&
1398 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1399 m->m_len <= M_TRAILINGSPACE(n)) {
1400 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len));
1401 n->m_len += m->m_len;
1402 sb->sb_ccc += m->m_len;
1403 sb->sb_tlscc += m->m_len;
1411 sb->sb_mtlstail = m;
1412 sballoc_ktls_rx(sb, m);
1422 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
1425 sbflush_internal(struct sockbuf *sb)
1428 while (sb->sb_mbcnt || sb->sb_tlsdcc) {
1430 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty:
1431 * we would loop forever. Panic instead.
1433 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len))
1435 m_freem(sbcut_internal(sb, (int)sb->sb_ccc));
1437 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
1438 ("%s: ccc %u mb %p mbcnt %u", __func__,
1439 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
1443 sbflush_locked(struct sockbuf *sb)
1446 SOCKBUF_LOCK_ASSERT(sb);
1447 sbflush_internal(sb);
1451 sbflush(struct sockbuf *sb)
1460 * Cut data from (the front of) a sockbuf.
1462 static struct mbuf *
1463 sbcut_internal(struct sockbuf *sb, int len)
1465 struct mbuf *m, *next, *mfree;
1468 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0",
1470 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u",
1471 __func__, len, sb->sb_ccc));
1473 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1480 if (next == NULL && !is_tls) {
1481 if (sb->sb_tlsdcc != 0) {
1482 MPASS(len >= sb->sb_tlsdcc);
1483 len -= sb->sb_tlsdcc;
1484 sb->sb_ccc -= sb->sb_tlsdcc;
1493 KASSERT(next, ("%s: no next, len %d", __func__, len));
1495 next = m->m_nextpkt;
1497 if (m->m_len > len) {
1498 KASSERT(!(m->m_flags & M_NOTAVAIL),
1499 ("%s: m %p M_NOTAVAIL", __func__, m));
1504 if (sb->sb_sndptroff != 0)
1505 sb->sb_sndptroff -= len;
1506 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1513 sbfree_ktls_rx(sb, m);
1518 * Do not put M_NOTREADY buffers to the free list, they
1519 * are referenced from outside.
1521 if (m->m_flags & M_NOTREADY && !is_tls)
1533 * Free any zero-length mbufs from the buffer.
1534 * For SOCK_DGRAM sockets such mbufs represent empty records.
1535 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer,
1536 * when sosend_generic() needs to send only control data.
1538 while (m && m->m_len == 0) {
1552 sb->sb_mtlstail = NULL;
1557 m->m_nextpkt = next;
1561 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
1562 * sb_lastrecord is up-to-date if we dropped part of the last record.
1566 sb->sb_mbtail = NULL;
1567 sb->sb_lastrecord = NULL;
1568 } else if (m->m_nextpkt == NULL) {
1569 sb->sb_lastrecord = m;
1576 * Drop data from (the front of) a sockbuf.
1579 sbdrop_locked(struct sockbuf *sb, int len)
1582 SOCKBUF_LOCK_ASSERT(sb);
1583 m_freem(sbcut_internal(sb, len));
1587 * Drop data from (the front of) a sockbuf,
1588 * and return it to caller.
1591 sbcut_locked(struct sockbuf *sb, int len)
1594 SOCKBUF_LOCK_ASSERT(sb);
1595 return (sbcut_internal(sb, len));
1599 sbdrop(struct sockbuf *sb, int len)
1604 mfree = sbcut_internal(sb, len);
1611 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff)
1615 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1616 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1618 if (sb->sb_sndptr == NULL) {
1619 sb->sb_sndptr = sb->sb_mb;
1620 sb->sb_sndptroff = 0;
1625 off -= sb->sb_sndptroff;
1632 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len)
1635 * A small copy was done, advance forward the sb_sbsndptr to cover
1640 if (mb != sb->sb_sndptr) {
1641 /* Did not copyout at the same mbuf */
1645 while (m && (len > 0)) {
1646 if (len >= m->m_len) {
1649 sb->sb_sndptroff += m->m_len;
1650 sb->sb_sndptr = m->m_next;
1660 * Return the first mbuf and the mbuf data offset for the provided
1661 * send offset without changing the "sb_sndptroff" field.
1664 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff)
1668 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1671 * If the "off" is below the stored offset, which happens on
1672 * retransmits, just use "sb_mb":
1674 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1678 off -= sb->sb_sndptroff;
1680 while (off > 0 && m != NULL) {
1691 * Drop a record off the front of a sockbuf and move the next record to the
1695 sbdroprecord_locked(struct sockbuf *sb)
1699 SOCKBUF_LOCK_ASSERT(sb);
1703 sb->sb_mb = m->m_nextpkt;
1713 * Drop a record off the front of a sockbuf and move the next record to the
1717 sbdroprecord(struct sockbuf *sb)
1721 sbdroprecord_locked(sb);
1726 * Create a "control" mbuf containing the specified data with the specified
1727 * type for presentation on a socket buffer.
1730 sbcreatecontrol_how(void *p, int size, int type, int level, int wait)
1735 MBUF_CHECKSLEEP(wait);
1736 if (CMSG_SPACE((u_int)size) > MCLBYTES)
1737 return ((struct mbuf *) NULL);
1738 if (CMSG_SPACE((u_int)size) > MLEN)
1739 m = m_getcl(wait, MT_CONTROL, 0);
1741 m = m_get(wait, MT_CONTROL);
1743 return ((struct mbuf *) NULL);
1744 cp = mtod(m, struct cmsghdr *);
1746 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1747 ("sbcreatecontrol: short mbuf"));
1749 * Don't leave the padding between the msg header and the
1750 * cmsg data and the padding after the cmsg data un-initialized.
1752 bzero(cp, CMSG_SPACE((u_int)size));
1754 (void)memcpy(CMSG_DATA(cp), p, size);
1755 m->m_len = CMSG_SPACE(size);
1756 cp->cmsg_len = CMSG_LEN(size);
1757 cp->cmsg_level = level;
1758 cp->cmsg_type = type;
1763 sbcreatecontrol(caddr_t p, int size, int type, int level)
1766 return (sbcreatecontrol_how(p, size, type, level, M_NOWAIT));
1770 * This does the same for socket buffers that sotoxsocket does for sockets:
1771 * generate an user-format data structure describing the socket buffer. Note
1772 * that the xsockbuf structure, since it is always embedded in a socket, does
1773 * not include a self pointer nor a length. We make this entry point public
1774 * in case some other mechanism needs it.
1777 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1780 xsb->sb_cc = sb->sb_ccc;
1781 xsb->sb_hiwat = sb->sb_hiwat;
1782 xsb->sb_mbcnt = sb->sb_mbcnt;
1783 xsb->sb_mcnt = sb->sb_mcnt;
1784 xsb->sb_ccnt = sb->sb_ccnt;
1785 xsb->sb_mbmax = sb->sb_mbmax;
1786 xsb->sb_lowat = sb->sb_lowat;
1787 xsb->sb_flags = sb->sb_flags;
1788 xsb->sb_timeo = sb->sb_timeo;
1791 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1793 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, "");
1794 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf,
1795 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0,
1796 sysctl_handle_sb_max, "LU",
1797 "Maximum socket buffer size");
1798 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1799 &sb_efficiency, 0, "Socket buffer size waste factor");