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_kern_tls.h"
38 #include "opt_param.h"
40 #include <sys/param.h>
41 #include <sys/aio.h> /* for aio_swake proto */
42 #include <sys/kernel.h>
45 #include <sys/malloc.h>
47 #include <sys/mutex.h>
49 #include <sys/protosw.h>
50 #include <sys/resourcevar.h>
51 #include <sys/signalvar.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
55 #include <sys/sysctl.h>
58 * Function pointer set by the AIO routines so that the socket buffer code
59 * can call back into the AIO module if it is loaded.
61 void (*aio_swake)(struct socket *, struct sockbuf *);
64 * Primitive routines for operating on socket buffers
67 #define BUF_MAX_ADJ(_sz) (((u_quad_t)(_sz)) * MCLBYTES / (MSIZE + MCLBYTES))
69 u_long sb_max = SB_MAX;
70 u_long sb_max_adj = BUF_MAX_ADJ(SB_MAX);
72 static u_long sb_efficiency = 8; /* parameter for sbreserve() */
75 static void sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m,
78 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len);
79 static void sbflush_internal(struct sockbuf *sb);
82 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY.
85 sbm_clrprotoflags(struct mbuf *m, int flags)
90 if (flags & PRUS_NOTREADY)
99 * Compress M_NOTREADY mbufs after they have been readied by sbready().
101 * sbcompress() skips M_NOTREADY mbufs since the data is not available to
102 * be copied at the time of sbcompress(). This function combines small
103 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first
104 * mbuf sbready() marked ready, and 'end' is the first mbuf still not
108 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end)
113 SOCKBUF_LOCK_ASSERT(sb);
115 if ((sb->sb_flags & SB_NOCOALESCE) != 0)
118 for (m = m0; m != end; m = m->m_next) {
119 MPASS((m->m_flags & M_NOTREADY) == 0);
121 * NB: In sbcompress(), 'n' is the last mbuf in the
122 * socket buffer and 'm' is the new mbuf being copied
123 * into the trailing space of 'n'. Here, the roles
124 * are reversed and 'n' is the next mbuf after 'm'
125 * that is being copied into the trailing space of
130 /* Try to coalesce adjacent ktls mbuf hdr/trailers. */
131 if ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 &&
132 (m->m_flags & M_EXTPG) &&
133 (n->m_flags & M_EXTPG) &&
134 !mbuf_has_tls_session(m) &&
135 !mbuf_has_tls_session(n)) {
136 int hdr_len, trail_len;
138 hdr_len = n->m_epg_hdrlen;
139 trail_len = m->m_epg_trllen;
140 if (trail_len != 0 && hdr_len != 0 &&
141 trail_len + hdr_len <= MBUF_PEXT_TRAIL_LEN) {
142 /* copy n's header to m's trailer */
143 memcpy(&m->m_epg_trail[trail_len],
144 n->m_epg_hdr, hdr_len);
145 m->m_epg_trllen += hdr_len;
153 /* Compress small unmapped mbufs into plain mbufs. */
154 if ((m->m_flags & M_EXTPG) && m->m_len <= MLEN &&
155 !mbuf_has_tls_session(m)) {
156 ext_size = m->m_ext.ext_size;
157 if (mb_unmapped_compress(m) == 0) {
158 sb->sb_mbcnt -= ext_size;
163 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 &&
165 (m->m_flags & M_EXTPG) == 0 &&
166 !mbuf_has_tls_session(n) &&
167 !mbuf_has_tls_session(m) &&
168 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
169 n->m_len <= M_TRAILINGSPACE(m) &&
170 m->m_type == n->m_type) {
171 KASSERT(sb->sb_lastrecord != n,
172 ("%s: merging start of record (%p) into previous mbuf (%p)",
174 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len));
175 m->m_len += n->m_len;
176 m->m_next = n->m_next;
177 m->m_flags |= n->m_flags & M_EOR;
178 if (sb->sb_mbtail == n)
181 sb->sb_mbcnt -= MSIZE;
183 if (n->m_flags & M_EXT) {
184 sb->sb_mbcnt -= n->m_ext.ext_size;
196 * Mark ready "count" units of I/O starting with "m". Most mbufs
197 * count as a single unit of I/O except for M_EXTPG mbufs which
198 * are backed by multiple pages.
201 sbready(struct sockbuf *sb, struct mbuf *m0, int count)
206 SOCKBUF_LOCK_ASSERT(sb);
207 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb));
208 KASSERT(count > 0, ("%s: invalid count %d", __func__, count));
211 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0;
214 KASSERT(m->m_flags & M_NOTREADY,
215 ("%s: m %p !M_NOTREADY", __func__, m));
216 if ((m->m_flags & M_EXTPG) != 0 && m->m_epg_npgs != 0) {
217 if (count < m->m_epg_nrdy) {
218 m->m_epg_nrdy -= count;
222 count -= m->m_epg_nrdy;
227 m->m_flags &= ~(M_NOTREADY | blocker);
229 sb->sb_acc += m->m_len;
234 * If the first mbuf is still not fully ready because only
235 * some of its backing pages were readied, no further progress
239 MPASS(m->m_flags & M_NOTREADY);
240 return (EINPROGRESS);
244 sbready_compress(sb, m0, m);
245 return (EINPROGRESS);
248 /* This one was blocking all the queue. */
249 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) {
250 KASSERT(m->m_flags & M_BLOCKED,
251 ("%s: m %p !M_BLOCKED", __func__, m));
252 m->m_flags &= ~M_BLOCKED;
253 sb->sb_acc += m->m_len;
257 sbready_compress(sb, m0, m);
263 * Adjust sockbuf state reflecting allocation of m.
266 sballoc(struct sockbuf *sb, struct mbuf *m)
269 SOCKBUF_LOCK_ASSERT(sb);
271 sb->sb_ccc += m->m_len;
273 if (sb->sb_fnrdy == NULL) {
274 if (m->m_flags & M_NOTREADY)
277 sb->sb_acc += m->m_len;
279 m->m_flags |= M_BLOCKED;
281 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
282 sb->sb_ctl += m->m_len;
284 sb->sb_mbcnt += MSIZE;
287 if (m->m_flags & M_EXT) {
288 sb->sb_mbcnt += m->m_ext.ext_size;
294 * Adjust sockbuf state reflecting freeing of m.
297 sbfree(struct sockbuf *sb, struct mbuf *m)
300 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
301 SOCKBUF_LOCK_ASSERT(sb);
304 sb->sb_ccc -= m->m_len;
306 if (!(m->m_flags & M_NOTAVAIL))
307 sb->sb_acc -= m->m_len;
309 if (m == sb->sb_fnrdy) {
312 KASSERT(m->m_flags & M_NOTREADY,
313 ("%s: m %p !M_NOTREADY", __func__, m));
316 while (n != NULL && !(n->m_flags & M_NOTREADY)) {
317 n->m_flags &= ~M_BLOCKED;
318 sb->sb_acc += n->m_len;
324 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
325 sb->sb_ctl -= m->m_len;
327 sb->sb_mbcnt -= MSIZE;
329 if (m->m_flags & M_EXT) {
330 sb->sb_mbcnt -= m->m_ext.ext_size;
334 if (sb->sb_sndptr == m) {
335 sb->sb_sndptr = NULL;
336 sb->sb_sndptroff = 0;
338 if (sb->sb_sndptroff != 0)
339 sb->sb_sndptroff -= m->m_len;
344 * Similar to sballoc/sbfree but does not adjust state associated with
345 * the sb_mb chain such as sb_fnrdy or sb_sndptr*. Also assumes mbufs
349 sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m)
352 SOCKBUF_LOCK_ASSERT(sb);
354 sb->sb_ccc += m->m_len;
355 sb->sb_tlscc += m->m_len;
357 sb->sb_mbcnt += MSIZE;
360 if (m->m_flags & M_EXT) {
361 sb->sb_mbcnt += m->m_ext.ext_size;
367 sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m)
370 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
371 SOCKBUF_LOCK_ASSERT(sb);
374 sb->sb_ccc -= m->m_len;
375 sb->sb_tlscc -= m->m_len;
377 sb->sb_mbcnt -= MSIZE;
380 if (m->m_flags & M_EXT) {
381 sb->sb_mbcnt -= m->m_ext.ext_size;
388 * Socantsendmore indicates that no more data will be sent on the socket; it
389 * would normally be applied to a socket when the user informs the system
390 * that no more data is to be sent, by the protocol code (in case
391 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
392 * received, and will normally be applied to the socket by a protocol when it
393 * detects that the peer will send no more data. Data queued for reading in
394 * the socket may yet be read.
397 socantsendmore_locked(struct socket *so)
400 SOCKBUF_LOCK_ASSERT(&so->so_snd);
402 so->so_snd.sb_state |= SBS_CANTSENDMORE;
403 sowwakeup_locked(so);
404 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
408 socantsendmore(struct socket *so)
411 SOCKBUF_LOCK(&so->so_snd);
412 socantsendmore_locked(so);
413 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
417 socantrcvmore_locked(struct socket *so)
420 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
422 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
424 if (so->so_rcv.sb_flags & SB_TLS_RX)
425 ktls_check_rx(&so->so_rcv);
427 sorwakeup_locked(so);
428 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
432 socantrcvmore(struct socket *so)
435 SOCKBUF_LOCK(&so->so_rcv);
436 socantrcvmore_locked(so);
437 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
441 soroverflow_locked(struct socket *so)
444 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
446 if (so->so_options & SO_RERROR) {
447 so->so_rerror = ENOBUFS;
448 sorwakeup_locked(so);
450 SOCKBUF_UNLOCK(&so->so_rcv);
452 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
456 soroverflow(struct socket *so)
459 SOCKBUF_LOCK(&so->so_rcv);
460 soroverflow_locked(so);
461 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
465 * Wait for data to arrive at/drain from a socket buffer.
468 sbwait(struct sockbuf *sb)
471 SOCKBUF_LOCK_ASSERT(sb);
473 sb->sb_flags |= SB_WAIT;
474 return (msleep_sbt(&sb->sb_acc, SOCKBUF_MTX(sb),
475 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
476 sb->sb_timeo, 0, 0));
480 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
481 * via SIGIO if the socket has the SS_ASYNC flag set.
483 * Called with the socket buffer lock held; will release the lock by the end
484 * of the function. This allows the caller to acquire the socket buffer lock
485 * while testing for the need for various sorts of wakeup and hold it through
486 * to the point where it's no longer required. We currently hold the lock
487 * through calls out to other subsystems (with the exception of kqueue), and
488 * then release it to avoid lock order issues. It's not clear that's
492 sowakeup(struct socket *so, struct sockbuf *sb)
496 SOCKBUF_LOCK_ASSERT(sb);
498 selwakeuppri(sb->sb_sel, PSOCK);
499 if (!SEL_WAITING(sb->sb_sel))
500 sb->sb_flags &= ~SB_SEL;
501 if (sb->sb_flags & SB_WAIT) {
502 sb->sb_flags &= ~SB_WAIT;
505 KNOTE_LOCKED(&sb->sb_sel->si_note, 0);
506 if (sb->sb_upcall != NULL) {
507 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT);
508 if (ret == SU_ISCONNECTED) {
509 KASSERT(sb == &so->so_rcv,
510 ("SO_SND upcall returned SU_ISCONNECTED"));
511 soupcall_clear(so, SO_RCV);
515 if (sb->sb_flags & SB_AIO)
516 sowakeup_aio(so, sb);
518 if (ret == SU_ISCONNECTED)
520 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
521 pgsigio(&so->so_sigio, SIGIO, 0);
522 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
526 * Socket buffer (struct sockbuf) utility routines.
528 * Each socket contains two socket buffers: one for sending data and one for
529 * receiving data. Each buffer contains a queue of mbufs, information about
530 * the number of mbufs and amount of data in the queue, and other fields
531 * allowing select() statements and notification on data availability to be
534 * Data stored in a socket buffer is maintained as a list of records. Each
535 * record is a list of mbufs chained together with the m_next field. Records
536 * are chained together with the m_nextpkt field. The upper level routine
537 * soreceive() expects the following conventions to be observed when placing
538 * information in the receive buffer:
540 * 1. If the protocol requires each message be preceded by the sender's name,
541 * then a record containing that name must be present before any
542 * associated data (mbuf's must be of type MT_SONAME).
543 * 2. If the protocol supports the exchange of ``access rights'' (really just
544 * additional data associated with the message), and there are ``rights''
545 * to be received, then a record containing this data should be present
546 * (mbuf's must be of type MT_RIGHTS).
547 * 3. If a name or rights record exists, then it must be followed by a data
548 * record, perhaps of zero length.
550 * Before using a new socket structure it is first necessary to reserve
551 * buffer space to the socket, by calling sbreserve(). This should commit
552 * some of the available buffer space in the system buffer pool for the
553 * socket (currently, it does nothing but enforce limits). The space should
554 * be released by calling sbrelease() when the socket is destroyed.
557 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
559 struct thread *td = curthread;
561 SOCKBUF_LOCK(&so->so_snd);
562 SOCKBUF_LOCK(&so->so_rcv);
563 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
565 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
567 if (so->so_rcv.sb_lowat == 0)
568 so->so_rcv.sb_lowat = 1;
569 if (so->so_snd.sb_lowat == 0)
570 so->so_snd.sb_lowat = MCLBYTES;
571 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
572 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
573 SOCKBUF_UNLOCK(&so->so_rcv);
574 SOCKBUF_UNLOCK(&so->so_snd);
577 sbrelease_locked(&so->so_snd, so);
579 SOCKBUF_UNLOCK(&so->so_rcv);
580 SOCKBUF_UNLOCK(&so->so_snd);
585 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
588 u_long tmp_sb_max = sb_max;
590 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
591 if (error || !req->newptr)
593 if (tmp_sb_max < MSIZE + MCLBYTES)
596 sb_max_adj = BUF_MAX_ADJ(sb_max);
601 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
602 * become limiting if buffering efficiency is near the normal case.
605 sbreserve_locked_limit(struct sockbuf *sb, u_long cc, struct socket *so,
606 u_long buf_max, struct thread *td)
610 SOCKBUF_LOCK_ASSERT(sb);
613 * When a thread is passed, we take into account the thread's socket
614 * buffer size limit. The caller will generally pass curthread, but
615 * in the TCP input path, NULL will be passed to indicate that no
616 * appropriate thread resource limits are available. In that case,
617 * we don't apply a process limit.
619 if (cc > BUF_MAX_ADJ(buf_max))
622 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE);
624 sbsize_limit = RLIM_INFINITY;
625 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
628 sb->sb_mbmax = min(cc * sb_efficiency, buf_max);
629 if (sb->sb_lowat > sb->sb_hiwat)
630 sb->sb_lowat = sb->sb_hiwat;
635 sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
638 return (sbreserve_locked_limit(sb, cc, so, sb_max, td));
642 sbsetopt(struct socket *so, int cmd, u_long cc)
646 u_int *hiwat, *lowat;
651 if (SOLISTENING(so)) {
655 lowat = &so->sol_sbsnd_lowat;
656 hiwat = &so->sol_sbsnd_hiwat;
657 flags = &so->sol_sbsnd_flags;
661 lowat = &so->sol_sbrcv_lowat;
662 hiwat = &so->sol_sbrcv_hiwat;
663 flags = &so->sol_sbrcv_flags;
677 flags = &sb->sb_flags;
678 hiwat = &sb->sb_hiwat;
679 lowat = &sb->sb_lowat;
687 if (SOLISTENING(so)) {
688 if (cc > sb_max_adj) {
696 u_long limit = sogetmaxbuf(so);
697 if (!sbreserve_locked_limit(sb, cc, so, limit, curthread))
701 *flags &= ~SB_AUTOSIZE;
706 * Make sure the low-water is never greater than the
709 *lowat = (cc > *hiwat) ? *hiwat : cc;
713 if (!SOLISTENING(so))
720 * Free mbufs held by a socket, and reserved mbuf space.
723 sbrelease_internal(struct sockbuf *sb, struct socket *so)
726 sbflush_internal(sb);
727 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
733 sbrelease_locked(struct sockbuf *sb, struct socket *so)
736 SOCKBUF_LOCK_ASSERT(sb);
738 sbrelease_internal(sb, so);
742 sbrelease(struct sockbuf *sb, struct socket *so)
746 sbrelease_locked(sb, so);
751 sbdestroy(struct sockbuf *sb, struct socket *so)
754 sbrelease_internal(sb, so);
756 if (sb->sb_tls_info != NULL)
757 ktls_free(sb->sb_tls_info);
758 sb->sb_tls_info = NULL;
763 * Routines to add and remove data from an mbuf queue.
765 * The routines sbappend() or sbappendrecord() are normally called to append
766 * new mbufs to a socket buffer, after checking that adequate space is
767 * available, comparing the function sbspace() with the amount of data to be
768 * added. sbappendrecord() differs from sbappend() in that data supplied is
769 * treated as the beginning of a new record. To place a sender's address,
770 * optional access rights, and data in a socket receive buffer,
771 * sbappendaddr() should be used. To place access rights and data in a
772 * socket receive buffer, sbappendrights() should be used. In either case,
773 * the new data begins a new record. Note that unlike sbappend() and
774 * sbappendrecord(), these routines check for the caller that there will be
775 * enough space to store the data. Each fails if there is not enough space,
776 * or if it cannot find mbufs to store additional information in.
778 * Reliable protocols may use the socket send buffer to hold data awaiting
779 * acknowledgement. Data is normally copied from a socket send buffer in a
780 * protocol with m_copy for output to a peer, and then removing the data from
781 * the socket buffer with sbdrop() or sbdroprecord() when the data is
782 * acknowledged by the peer.
786 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
788 struct mbuf *m = sb->sb_mb;
790 SOCKBUF_LOCK_ASSERT(sb);
792 while (m && m->m_nextpkt)
795 if (m != sb->sb_lastrecord) {
796 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
797 __func__, sb->sb_mb, sb->sb_lastrecord, m);
798 printf("packet chain:\n");
799 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
801 panic("%s from %s:%u", __func__, file, line);
806 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
808 struct mbuf *m = sb->sb_mb;
811 SOCKBUF_LOCK_ASSERT(sb);
813 while (m && m->m_nextpkt)
816 while (m && m->m_next)
819 if (m != sb->sb_mbtail) {
820 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
821 __func__, sb->sb_mb, sb->sb_mbtail, m);
822 printf("packet tree:\n");
823 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
825 for (n = m; n != NULL; n = n->m_next)
829 panic("%s from %s:%u", __func__, file, line);
834 while (m && m->m_next)
837 if (m != sb->sb_mtlstail) {
838 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n",
839 __func__, sb->sb_mtls, sb->sb_mtlstail, m);
840 printf("TLS packet tree:\n");
842 for (m = sb->sb_mtls; m != NULL; m = m->m_next) {
846 panic("%s from %s:%u", __func__, file, line);
850 #endif /* SOCKBUF_DEBUG */
852 #define SBLINKRECORD(sb, m0) do { \
853 SOCKBUF_LOCK_ASSERT(sb); \
854 if ((sb)->sb_lastrecord != NULL) \
855 (sb)->sb_lastrecord->m_nextpkt = (m0); \
857 (sb)->sb_mb = (m0); \
858 (sb)->sb_lastrecord = (m0); \
859 } while (/*CONSTCOND*/0)
862 * Append mbuf chain m to the last record in the socket buffer sb. The
863 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
864 * are discarded and mbufs are compacted where possible.
867 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags)
871 SOCKBUF_LOCK_ASSERT(sb);
875 sbm_clrprotoflags(m, flags);
882 if (n->m_flags & M_EOR) {
883 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
886 } while (n->m_next && (n = n->m_next));
889 * XXX Would like to simply use sb_mbtail here, but
890 * XXX I need to verify that I won't miss an EOR that
893 if ((n = sb->sb_lastrecord) != NULL) {
895 if (n->m_flags & M_EOR) {
896 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
899 } while (n->m_next && (n = n->m_next));
902 * If this is the first record in the socket buffer,
903 * it's also the last record.
905 sb->sb_lastrecord = m;
908 sbcompress(sb, m, n);
913 * Append mbuf chain m to the last record in the socket buffer sb. The
914 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
915 * are discarded and mbufs are compacted where possible.
918 sbappend(struct sockbuf *sb, struct mbuf *m, int flags)
922 sbappend_locked(sb, m, flags);
928 * Append an mbuf containing encrypted TLS data. The data
929 * is marked M_NOTREADY until it has been decrypted and
930 * stored as a TLS record.
933 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m)
939 /* Remove all packet headers and mbuf tags to get a pure data chain. */
942 for (n = m; n != NULL; n = n->m_next)
943 n->m_flags |= M_NOTREADY;
944 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail);
950 * This version of sbappend() should only be used when the caller absolutely
951 * knows that there will never be more than one record in the socket buffer,
952 * that is, a stream protocol (such as TCP).
955 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags)
957 SOCKBUF_LOCK_ASSERT(sb);
959 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
963 * Decrypted TLS records are appended as records via
964 * sbappendrecord(). TCP passes encrypted TLS records to this
965 * function which must be scheduled for decryption.
967 if (sb->sb_flags & SB_TLS_RX) {
968 sbappend_ktls_rx(sb, m);
973 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
978 if (sb->sb_tls_info != NULL)
982 /* Remove all packet headers and mbuf tags to get a pure data chain. */
983 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0);
985 sbcompress(sb, m, sb->sb_mbtail);
987 sb->sb_lastrecord = sb->sb_mb;
992 * This version of sbappend() should only be used when the caller absolutely
993 * knows that there will never be more than one record in the socket buffer,
994 * that is, a stream protocol (such as TCP).
997 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags)
1001 sbappendstream_locked(sb, m, flags);
1005 #ifdef SOCKBUF_DEBUG
1007 sbcheck(struct sockbuf *sb, const char *file, int line)
1009 struct mbuf *m, *n, *fnrdy;
1010 u_long acc, ccc, mbcnt;
1015 SOCKBUF_LOCK_ASSERT(sb);
1017 acc = ccc = mbcnt = 0;
1020 for (m = sb->sb_mb; m; m = n) {
1022 for (; m; m = m->m_next) {
1023 if (m->m_len == 0) {
1024 printf("sb %p empty mbuf %p\n", sb, m);
1027 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) {
1028 if (m != sb->sb_fnrdy) {
1029 printf("sb %p: fnrdy %p != m %p\n",
1030 sb, sb->sb_fnrdy, m);
1036 if (!(m->m_flags & M_NOTAVAIL)) {
1037 printf("sb %p: fnrdy %p, m %p is avail\n",
1038 sb, sb->sb_fnrdy, m);
1045 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
1046 mbcnt += m->m_ext.ext_size;
1051 * Account for mbufs "detached" by ktls_detach_record() while
1052 * they are decrypted by ktls_decrypt(). tlsdcc gives a count
1053 * of the detached bytes that are included in ccc. The mbufs
1054 * and clusters are not included in the socket buffer
1057 ccc += sb->sb_tlsdcc;
1060 for (m = sb->sb_mtls; m; m = m->m_next) {
1061 if (m->m_nextpkt != NULL) {
1062 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m);
1065 if ((m->m_flags & M_NOTREADY) == 0) {
1066 printf("sb %p TLS mbuf %p ready\n", sb, m);
1072 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
1073 mbcnt += m->m_ext.ext_size;
1076 if (sb->sb_tlscc != tlscc) {
1077 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc,
1082 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) {
1083 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n",
1084 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt);
1086 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc,
1093 panic("%s from %s:%u", __func__, file, line);
1098 * As above, except the mbuf chain begins a new record.
1101 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
1105 SOCKBUF_LOCK_ASSERT(sb);
1109 m_clrprotoflags(m0);
1111 * Put the first mbuf on the queue. Note this permits zero length
1115 SBLASTRECORDCHK(sb);
1116 SBLINKRECORD(sb, m0);
1120 if (m && (m0->m_flags & M_EOR)) {
1121 m0->m_flags &= ~M_EOR;
1122 m->m_flags |= M_EOR;
1124 /* always call sbcompress() so it can do SBLASTMBUFCHK() */
1125 sbcompress(sb, m, m0);
1129 * As above, except the mbuf chain begins a new record.
1132 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
1136 sbappendrecord_locked(sb, m0);
1140 /* Helper routine that appends data, control, and address to a sockbuf. */
1142 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa,
1143 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last)
1145 struct mbuf *m, *n, *nlast;
1147 if (asa->sa_len > MLEN)
1150 m = m_get(M_NOWAIT, MT_SONAME);
1153 m->m_len = asa->sa_len;
1154 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
1156 M_ASSERT_NO_SND_TAG(m0);
1157 m_clrprotoflags(m0);
1158 m_tag_delete_chain(m0, NULL);
1160 * Clear some persistent info from pkthdr.
1161 * We don't use m_demote(), because some netgraph consumers
1162 * expect M_PKTHDR presence.
1164 m0->m_pkthdr.rcvif = NULL;
1165 m0->m_pkthdr.flowid = 0;
1166 m0->m_pkthdr.csum_flags = 0;
1167 m0->m_pkthdr.fibnum = 0;
1168 m0->m_pkthdr.rsstype = 0;
1171 ctrl_last->m_next = m0; /* concatenate data to control */
1174 m->m_next = control;
1175 for (n = m; n->m_next != NULL; n = n->m_next)
1179 SBLINKRECORD(sb, m);
1181 sb->sb_mbtail = nlast;
1184 SBLASTRECORDCHK(sb);
1189 * Append address and data, and optionally, control (ancillary) data to the
1190 * receive queue of a socket. If present, m0 must include a packet header
1191 * with total length. Returns 0 if no space in sockbuf or insufficient
1195 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
1196 struct mbuf *m0, struct mbuf *control)
1198 struct mbuf *ctrl_last;
1199 int space = asa->sa_len;
1201 SOCKBUF_LOCK_ASSERT(sb);
1203 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
1204 panic("sbappendaddr_locked");
1206 space += m0->m_pkthdr.len;
1207 space += m_length(control, &ctrl_last);
1209 if (space > sbspace(sb))
1211 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
1215 * Append address and data, and optionally, control (ancillary) data to the
1216 * receive queue of a socket. If present, m0 must include a packet header
1217 * with total length. Returns 0 if insufficient mbufs. Does not validate space
1218 * on the receiving sockbuf.
1221 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa,
1222 struct mbuf *m0, struct mbuf *control)
1224 struct mbuf *ctrl_last;
1226 SOCKBUF_LOCK_ASSERT(sb);
1228 ctrl_last = (control == NULL) ? NULL : m_last(control);
1229 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
1233 * Append address and data, and optionally, control (ancillary) data to the
1234 * receive queue of a socket. If present, m0 must include a packet header
1235 * with total length. Returns 0 if no space in sockbuf or insufficient
1239 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
1240 struct mbuf *m0, struct mbuf *control)
1245 retval = sbappendaddr_locked(sb, asa, m0, control);
1251 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
1252 struct mbuf *control, int flags)
1254 struct mbuf *m, *mlast;
1256 sbm_clrprotoflags(m0, flags);
1257 m_last(control)->m_next = m0;
1259 SBLASTRECORDCHK(sb);
1261 for (m = control; m->m_next; m = m->m_next)
1265 SBLINKRECORD(sb, control);
1267 sb->sb_mbtail = mlast;
1270 SBLASTRECORDCHK(sb);
1274 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control,
1279 sbappendcontrol_locked(sb, m0, control, flags);
1284 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
1285 * (n). If (n) is NULL, the buffer is presumed empty.
1287 * When the data is compressed, mbufs in the chain may be handled in one of
1290 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
1291 * record boundary, and no change in data type).
1293 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
1294 * an mbuf already in the socket buffer. This can occur if an
1295 * appropriate mbuf exists, there is room, both mbufs are not marked as
1296 * not ready, and no merging of data types will occur.
1298 * (3) The mbuf may be appended to the end of the existing mbuf chain.
1300 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
1304 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1309 SOCKBUF_LOCK_ASSERT(sb);
1312 eor |= m->m_flags & M_EOR;
1313 if (m->m_len == 0 &&
1315 (((o = m->m_next) || (o = n)) &&
1316 o->m_type == m->m_type))) {
1317 if (sb->sb_lastrecord == m)
1318 sb->sb_lastrecord = m->m_next;
1322 if (n && (n->m_flags & M_EOR) == 0 &&
1324 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1325 !(m->m_flags & M_NOTREADY) &&
1326 !(n->m_flags & (M_NOTREADY | M_EXTPG)) &&
1327 !mbuf_has_tls_session(m) &&
1328 !mbuf_has_tls_session(n) &&
1329 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1330 m->m_len <= M_TRAILINGSPACE(n) &&
1331 n->m_type == m->m_type) {
1332 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len));
1333 n->m_len += m->m_len;
1334 sb->sb_ccc += m->m_len;
1335 if (sb->sb_fnrdy == NULL)
1336 sb->sb_acc += m->m_len;
1337 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1338 /* XXX: Probably don't need.*/
1339 sb->sb_ctl += m->m_len;
1343 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) &&
1344 (m->m_flags & M_NOTREADY) == 0 &&
1345 !mbuf_has_tls_session(m))
1346 (void)mb_unmapped_compress(m);
1354 m->m_flags &= ~M_EOR;
1359 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
1367 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs
1368 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also
1369 * a bit simpler (no EOR markers, always MT_DATA, etc.).
1372 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1375 SOCKBUF_LOCK_ASSERT(sb);
1378 KASSERT((m->m_flags & M_EOR) == 0,
1379 ("TLS RX mbuf %p with EOR", m));
1380 KASSERT(m->m_type == MT_DATA,
1381 ("TLS RX mbuf %p is not MT_DATA", m));
1382 KASSERT((m->m_flags & M_NOTREADY) != 0,
1383 ("TLS RX mbuf %p ready", m));
1384 KASSERT((m->m_flags & M_EXTPG) == 0,
1385 ("TLS RX mbuf %p unmapped", m));
1387 if (m->m_len == 0) {
1393 * Even though both 'n' and 'm' are NOTREADY, it's ok
1394 * to coalesce the data.
1398 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1399 !(n->m_flags & (M_EXTPG)) &&
1400 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1401 m->m_len <= M_TRAILINGSPACE(n)) {
1402 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len));
1403 n->m_len += m->m_len;
1404 sb->sb_ccc += m->m_len;
1405 sb->sb_tlscc += m->m_len;
1413 sb->sb_mtlstail = m;
1414 sballoc_ktls_rx(sb, m);
1424 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
1427 sbflush_internal(struct sockbuf *sb)
1430 while (sb->sb_mbcnt || sb->sb_tlsdcc) {
1432 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty:
1433 * we would loop forever. Panic instead.
1435 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len))
1437 m_freem(sbcut_internal(sb, (int)sb->sb_ccc));
1439 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
1440 ("%s: ccc %u mb %p mbcnt %u", __func__,
1441 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
1445 sbflush_locked(struct sockbuf *sb)
1448 SOCKBUF_LOCK_ASSERT(sb);
1449 sbflush_internal(sb);
1453 sbflush(struct sockbuf *sb)
1462 * Cut data from (the front of) a sockbuf.
1464 static struct mbuf *
1465 sbcut_internal(struct sockbuf *sb, int len)
1467 struct mbuf *m, *next, *mfree;
1470 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0",
1472 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u",
1473 __func__, len, sb->sb_ccc));
1475 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1482 if (next == NULL && !is_tls) {
1483 if (sb->sb_tlsdcc != 0) {
1484 MPASS(len >= sb->sb_tlsdcc);
1485 len -= sb->sb_tlsdcc;
1486 sb->sb_ccc -= sb->sb_tlsdcc;
1495 KASSERT(next, ("%s: no next, len %d", __func__, len));
1497 next = m->m_nextpkt;
1499 if (m->m_len > len) {
1500 KASSERT(!(m->m_flags & M_NOTAVAIL),
1501 ("%s: m %p M_NOTAVAIL", __func__, m));
1506 if (sb->sb_sndptroff != 0)
1507 sb->sb_sndptroff -= len;
1508 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1515 sbfree_ktls_rx(sb, m);
1520 * Do not put M_NOTREADY buffers to the free list, they
1521 * are referenced from outside.
1523 if (m->m_flags & M_NOTREADY && !is_tls)
1535 * Free any zero-length mbufs from the buffer.
1536 * For SOCK_DGRAM sockets such mbufs represent empty records.
1537 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer,
1538 * when sosend_generic() needs to send only control data.
1540 while (m && m->m_len == 0) {
1554 sb->sb_mtlstail = NULL;
1559 m->m_nextpkt = next;
1563 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
1564 * sb_lastrecord is up-to-date if we dropped part of the last record.
1568 sb->sb_mbtail = NULL;
1569 sb->sb_lastrecord = NULL;
1570 } else if (m->m_nextpkt == NULL) {
1571 sb->sb_lastrecord = m;
1578 * Drop data from (the front of) a sockbuf.
1581 sbdrop_locked(struct sockbuf *sb, int len)
1584 SOCKBUF_LOCK_ASSERT(sb);
1585 m_freem(sbcut_internal(sb, len));
1589 * Drop data from (the front of) a sockbuf,
1590 * and return it to caller.
1593 sbcut_locked(struct sockbuf *sb, int len)
1596 SOCKBUF_LOCK_ASSERT(sb);
1597 return (sbcut_internal(sb, len));
1601 sbdrop(struct sockbuf *sb, int len)
1606 mfree = sbcut_internal(sb, len);
1613 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff)
1617 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1618 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1620 if (sb->sb_sndptr == NULL) {
1621 sb->sb_sndptr = sb->sb_mb;
1622 sb->sb_sndptroff = 0;
1627 off -= sb->sb_sndptroff;
1634 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len)
1637 * A small copy was done, advance forward the sb_sbsndptr to cover
1642 if (mb != sb->sb_sndptr) {
1643 /* Did not copyout at the same mbuf */
1647 while (m && (len > 0)) {
1648 if (len >= m->m_len) {
1651 sb->sb_sndptroff += m->m_len;
1652 sb->sb_sndptr = m->m_next;
1662 * Return the first mbuf and the mbuf data offset for the provided
1663 * send offset without changing the "sb_sndptroff" field.
1666 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff)
1670 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1673 * If the "off" is below the stored offset, which happens on
1674 * retransmits, just use "sb_mb":
1676 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1680 off -= sb->sb_sndptroff;
1682 while (off > 0 && m != NULL) {
1693 * Drop a record off the front of a sockbuf and move the next record to the
1697 sbdroprecord_locked(struct sockbuf *sb)
1701 SOCKBUF_LOCK_ASSERT(sb);
1705 sb->sb_mb = m->m_nextpkt;
1715 * Drop a record off the front of a sockbuf and move the next record to the
1719 sbdroprecord(struct sockbuf *sb)
1723 sbdroprecord_locked(sb);
1728 * Create a "control" mbuf containing the specified data with the specified
1729 * type for presentation on a socket buffer.
1732 sbcreatecontrol_how(void *p, int size, int type, int level, int wait)
1737 MBUF_CHECKSLEEP(wait);
1738 if (CMSG_SPACE((u_int)size) > MCLBYTES)
1739 return ((struct mbuf *) NULL);
1740 if (CMSG_SPACE((u_int)size) > MLEN)
1741 m = m_getcl(wait, MT_CONTROL, 0);
1743 m = m_get(wait, MT_CONTROL);
1745 return ((struct mbuf *) NULL);
1746 cp = mtod(m, struct cmsghdr *);
1748 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1749 ("sbcreatecontrol: short mbuf"));
1751 * Don't leave the padding between the msg header and the
1752 * cmsg data and the padding after the cmsg data un-initialized.
1754 bzero(cp, CMSG_SPACE((u_int)size));
1756 (void)memcpy(CMSG_DATA(cp), p, size);
1757 m->m_len = CMSG_SPACE(size);
1758 cp->cmsg_len = CMSG_LEN(size);
1759 cp->cmsg_level = level;
1760 cp->cmsg_type = type;
1765 sbcreatecontrol(caddr_t p, int size, int type, int level)
1768 return (sbcreatecontrol_how(p, size, type, level, M_NOWAIT));
1772 * This does the same for socket buffers that sotoxsocket does for sockets:
1773 * generate an user-format data structure describing the socket buffer. Note
1774 * that the xsockbuf structure, since it is always embedded in a socket, does
1775 * not include a self pointer nor a length. We make this entry point public
1776 * in case some other mechanism needs it.
1779 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1782 xsb->sb_cc = sb->sb_ccc;
1783 xsb->sb_hiwat = sb->sb_hiwat;
1784 xsb->sb_mbcnt = sb->sb_mbcnt;
1785 xsb->sb_mcnt = sb->sb_mcnt;
1786 xsb->sb_ccnt = sb->sb_ccnt;
1787 xsb->sb_mbmax = sb->sb_mbmax;
1788 xsb->sb_lowat = sb->sb_lowat;
1789 xsb->sb_flags = sb->sb_flags;
1790 xsb->sb_timeo = sb->sb_timeo;
1793 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1795 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, "");
1796 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf,
1797 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0,
1798 sysctl_handle_sb_max, "LU",
1799 "Maximum socket buffer size");
1800 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1801 &sb_efficiency, 0, "Socket buffer size waste factor");