38
39#include "opt_mac.h"
40#include "opt_param.h"
41
42#include <sys/param.h>
43#include <sys/aio.h> /* for aio_swake proto */
44#include <sys/domain.h>
45#include <sys/event.h>
46#include <sys/file.h> /* for maxfiles */
47#include <sys/kernel.h>
48#include <sys/lock.h>
49#include <sys/mac.h>
50#include <sys/malloc.h>
51#include <sys/mbuf.h>
52#include <sys/mutex.h>
53#include <sys/proc.h>
54#include <sys/protosw.h>
55#include <sys/resourcevar.h>
56#include <sys/signalvar.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/stat.h>
60#include <sys/sysctl.h>
61#include <sys/systm.h>
62
63int maxsockets;
64
65void (*aio_swake)(struct socket *, struct sockbuf *);
66
67/*
68 * Primitive routines for operating on sockets and socket buffers
69 */
70
71u_long sb_max = SB_MAX;
72static u_long sb_max_adj =
73 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
74
75static u_long sb_efficiency = 8; /* parameter for sbreserve() */
76
77/*
78 * Procedures to manipulate state flags of socket
79 * and do appropriate wakeups. Normal sequence from the
80 * active (originating) side is that soisconnecting() is
81 * called during processing of connect() call,
82 * resulting in an eventual call to soisconnected() if/when the
83 * connection is established. When the connection is torn down
84 * soisdisconnecting() is called during processing of disconnect() call,
85 * and soisdisconnected() is called when the connection to the peer
86 * is totally severed. The semantics of these routines are such that
87 * connectionless protocols can call soisconnected() and soisdisconnected()
88 * only, bypassing the in-progress calls when setting up a ``connection''
89 * takes no time.
90 *
91 * From the passive side, a socket is created with
92 * two queues of sockets: so_incomp for connections in progress
93 * and so_comp for connections already made and awaiting user acceptance.
94 * As a protocol is preparing incoming connections, it creates a socket
95 * structure queued on so_incomp by calling sonewconn(). When the connection
96 * is established, soisconnected() is called, and transfers the
97 * socket structure to so_comp, making it available to accept().
98 *
99 * If a socket is closed with sockets on either
100 * so_incomp or so_comp, these sockets are dropped.
101 *
102 * If higher level protocols are implemented in
103 * the kernel, the wakeups done here will sometimes
104 * cause software-interrupt process scheduling.
105 */
106
107void
108soisconnecting(so)
109 register struct socket *so;
110{
111
112 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
113 so->so_state |= SS_ISCONNECTING;
114}
115
116void
117soisconnected(so)
118 struct socket *so;
119{
120 struct socket *head = so->so_head;
121
122 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
123 so->so_state |= SS_ISCONNECTED;
124 if (head && (so->so_state & SS_INCOMP)) {
125 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
126 so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
127 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
128 so->so_rcv.sb_flags |= SB_UPCALL;
129 so->so_options &= ~SO_ACCEPTFILTER;
130 so->so_upcall(so, so->so_upcallarg, M_TRYWAIT);
131 return;
132 }
133 TAILQ_REMOVE(&head->so_incomp, so, so_list);
134 head->so_incqlen--;
135 so->so_state &= ~SS_INCOMP;
136 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
137 head->so_qlen++;
138 so->so_state |= SS_COMP;
139 sorwakeup(head);
140 wakeup_one(&head->so_timeo);
141 } else {
142 wakeup(&so->so_timeo);
143 sorwakeup(so);
144 sowwakeup(so);
145 }
146}
147
148void
149soisdisconnecting(so)
150 register struct socket *so;
151{
152
153 so->so_state &= ~SS_ISCONNECTING;
154 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
155 wakeup(&so->so_timeo);
156 sowwakeup(so);
157 sorwakeup(so);
158}
159
160void
161soisdisconnected(so)
162 register struct socket *so;
163{
164
165 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
166 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
167 wakeup(&so->so_timeo);
168 sbdrop(&so->so_snd, so->so_snd.sb_cc);
169 sowwakeup(so);
170 sorwakeup(so);
171}
172
173/*
174 * When an attempt at a new connection is noted on a socket
175 * which accepts connections, sonewconn is called. If the
176 * connection is possible (subject to space constraints, etc.)
177 * then we allocate a new structure, propoerly linked into the
178 * data structure of the original socket, and return this.
179 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
180 *
181 * note: the ref count on the socket is 0 on return
182 */
183struct socket *
184sonewconn(head, connstatus)
185 register struct socket *head;
186 int connstatus;
187{
188 register struct socket *so;
189
190 if (head->so_qlen > 3 * head->so_qlimit / 2)
191 return ((struct socket *)0);
192 so = soalloc(M_NOWAIT);
193 if (so == NULL)
194 return ((struct socket *)0);
195 if ((head->so_options & SO_ACCEPTFILTER) != 0)
196 connstatus = 0;
197 so->so_head = head;
198 so->so_type = head->so_type;
199 so->so_options = head->so_options &~ SO_ACCEPTCONN;
200 so->so_linger = head->so_linger;
201 so->so_state = head->so_state | SS_NOFDREF;
202 so->so_proto = head->so_proto;
203 so->so_timeo = head->so_timeo;
204 so->so_cred = crhold(head->so_cred);
205#ifdef MAC
206 mac_create_socket_from_socket(head, so);
207#endif
208 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
209 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
210 sodealloc(so);
211 return ((struct socket *)0);
212 }
213
214 if (connstatus) {
215 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
216 so->so_state |= SS_COMP;
217 head->so_qlen++;
218 } else {
219 if (head->so_incqlen > head->so_qlimit) {
220 struct socket *sp;
221 sp = TAILQ_FIRST(&head->so_incomp);
222 (void) soabort(sp);
223 }
224 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
225 so->so_state |= SS_INCOMP;
226 head->so_incqlen++;
227 }
228 if (connstatus) {
229 sorwakeup(head);
230 wakeup(&head->so_timeo);
231 so->so_state |= connstatus;
232 }
233 return (so);
234}
235
236/*
237 * Socantsendmore indicates that no more data will be sent on the
238 * socket; it would normally be applied to a socket when the user
239 * informs the system that no more data is to be sent, by the protocol
240 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
241 * will be received, and will normally be applied to the socket by a
242 * protocol when it detects that the peer will send no more data.
243 * Data queued for reading in the socket may yet be read.
244 */
245
246void
247socantsendmore(so)
248 struct socket *so;
249{
250
251 so->so_state |= SS_CANTSENDMORE;
252 sowwakeup(so);
253}
254
255void
256socantrcvmore(so)
257 struct socket *so;
258{
259
260 so->so_state |= SS_CANTRCVMORE;
261 sorwakeup(so);
262}
263
264/*
265 * Wait for data to arrive at/drain from a socket buffer.
266 */
267int
268sbwait(sb)
269 struct sockbuf *sb;
270{
271
272 sb->sb_flags |= SB_WAIT;
273 return (tsleep(&sb->sb_cc,
274 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
275 sb->sb_timeo));
276}
277
278/*
279 * Lock a sockbuf already known to be locked;
280 * return any error returned from sleep (EINTR).
281 */
282int
283sb_lock(sb)
284 register struct sockbuf *sb;
285{
286 int error;
287
288 while (sb->sb_flags & SB_LOCK) {
289 sb->sb_flags |= SB_WANT;
290 error = tsleep(&sb->sb_flags,
291 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
292 "sblock", 0);
293 if (error)
294 return (error);
295 }
296 sb->sb_flags |= SB_LOCK;
297 return (0);
298}
299
300/*
301 * Wakeup processes waiting on a socket buffer.
302 * Do asynchronous notification via SIGIO
303 * if the socket has the SS_ASYNC flag set.
304 */
305void
306sowakeup(so, sb)
307 register struct socket *so;
308 register struct sockbuf *sb;
309{
310
311 selwakeuppri(&sb->sb_sel, PSOCK);
312 sb->sb_flags &= ~SB_SEL;
313 if (sb->sb_flags & SB_WAIT) {
314 sb->sb_flags &= ~SB_WAIT;
315 wakeup(&sb->sb_cc);
316 }
317 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
318 pgsigio(&so->so_sigio, SIGIO, 0);
319 if (sb->sb_flags & SB_UPCALL)
320 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
321 if (sb->sb_flags & SB_AIO)
322 aio_swake(so, sb);
323 KNOTE(&sb->sb_sel.si_note, 0);
324}
325
326/*
327 * Socket buffer (struct sockbuf) utility routines.
328 *
329 * Each socket contains two socket buffers: one for sending data and
330 * one for receiving data. Each buffer contains a queue of mbufs,
331 * information about the number of mbufs and amount of data in the
332 * queue, and other fields allowing select() statements and notification
333 * on data availability to be implemented.
334 *
335 * Data stored in a socket buffer is maintained as a list of records.
336 * Each record is a list of mbufs chained together with the m_next
337 * field. Records are chained together with the m_nextpkt field. The upper
338 * level routine soreceive() expects the following conventions to be
339 * observed when placing information in the receive buffer:
340 *
341 * 1. If the protocol requires each message be preceded by the sender's
342 * name, then a record containing that name must be present before
343 * any associated data (mbuf's must be of type MT_SONAME).
344 * 2. If the protocol supports the exchange of ``access rights'' (really
345 * just additional data associated with the message), and there are
346 * ``rights'' to be received, then a record containing this data
347 * should be present (mbuf's must be of type MT_RIGHTS).
348 * 3. If a name or rights record exists, then it must be followed by
349 * a data record, perhaps of zero length.
350 *
351 * Before using a new socket structure it is first necessary to reserve
352 * buffer space to the socket, by calling sbreserve(). This should commit
353 * some of the available buffer space in the system buffer pool for the
354 * socket (currently, it does nothing but enforce limits). The space
355 * should be released by calling sbrelease() when the socket is destroyed.
356 */
357
358int
359soreserve(so, sndcc, rcvcc)
360 register struct socket *so;
361 u_long sndcc, rcvcc;
362{
363 struct thread *td = curthread;
364
365 if (sbreserve(&so->so_snd, sndcc, so, td) == 0)
366 goto bad;
367 if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0)
368 goto bad2;
369 if (so->so_rcv.sb_lowat == 0)
370 so->so_rcv.sb_lowat = 1;
371 if (so->so_snd.sb_lowat == 0)
372 so->so_snd.sb_lowat = MCLBYTES;
373 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
374 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
375 return (0);
376bad2:
377 sbrelease(&so->so_snd, so);
378bad:
379 return (ENOBUFS);
380}
381
382static int
383sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
384{
385 int error = 0;
386 u_long old_sb_max = sb_max;
387
388 error = SYSCTL_OUT(req, arg1, sizeof(u_long));
389 if (error || !req->newptr)
390 return (error);
391 error = SYSCTL_IN(req, arg1, sizeof(u_long));
392 if (error)
393 return (error);
394 if (sb_max < MSIZE + MCLBYTES) {
395 sb_max = old_sb_max;
396 return (EINVAL);
397 }
398 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
399 return (0);
400}
401
402/*
403 * Allot mbufs to a sockbuf.
404 * Attempt to scale mbmax so that mbcnt doesn't become limiting
405 * if buffering efficiency is near the normal case.
406 */
407int
408sbreserve(sb, cc, so, td)
409 struct sockbuf *sb;
410 u_long cc;
411 struct socket *so;
412 struct thread *td;
413{
414 rlim_t sbsize_limit;
415
416 /*
417 * td will only be NULL when we're in an interrupt
418 * (e.g. in tcp_input())
419 */
420 if (cc > sb_max_adj)
421 return (0);
422 if (td != NULL) {
423 PROC_LOCK(td->td_proc);
424 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
425 PROC_UNLOCK(td->td_proc);
426 } else
427 sbsize_limit = RLIM_INFINITY;
428 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
429 sbsize_limit))
430 return (0);
431 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
432 if (sb->sb_lowat > sb->sb_hiwat)
433 sb->sb_lowat = sb->sb_hiwat;
434 return (1);
435}
436
437/*
438 * Free mbufs held by a socket, and reserved mbuf space.
439 */
440void
441sbrelease(sb, so)
442 struct sockbuf *sb;
443 struct socket *so;
444{
445
446 sbflush(sb);
447 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
448 RLIM_INFINITY);
449 sb->sb_mbmax = 0;
450}
451
452/*
453 * Routines to add and remove
454 * data from an mbuf queue.
455 *
456 * The routines sbappend() or sbappendrecord() are normally called to
457 * append new mbufs to a socket buffer, after checking that adequate
458 * space is available, comparing the function sbspace() with the amount
459 * of data to be added. sbappendrecord() differs from sbappend() in
460 * that data supplied is treated as the beginning of a new record.
461 * To place a sender's address, optional access rights, and data in a
462 * socket receive buffer, sbappendaddr() should be used. To place
463 * access rights and data in a socket receive buffer, sbappendrights()
464 * should be used. In either case, the new data begins a new record.
465 * Note that unlike sbappend() and sbappendrecord(), these routines check
466 * for the caller that there will be enough space to store the data.
467 * Each fails if there is not enough space, or if it cannot find mbufs
468 * to store additional information in.
469 *
470 * Reliable protocols may use the socket send buffer to hold data
471 * awaiting acknowledgement. Data is normally copied from a socket
472 * send buffer in a protocol with m_copy for output to a peer,
473 * and then removing the data from the socket buffer with sbdrop()
474 * or sbdroprecord() when the data is acknowledged by the peer.
475 */
476
477#ifdef SOCKBUF_DEBUG
478void
479sblastrecordchk(struct sockbuf *sb, const char *file, int line)
480{
481 struct mbuf *m = sb->sb_mb;
482
483 while (m && m->m_nextpkt)
484 m = m->m_nextpkt;
485
486 if (m != sb->sb_lastrecord) {
487 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
488 __func__, sb->sb_mb, sb->sb_lastrecord, m);
489 printf("packet chain:\n");
490 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
491 printf("\t%p\n", m);
492 panic("%s from %s:%u", __func__, file, line);
493 }
494}
495
496void
497sblastmbufchk(struct sockbuf *sb, const char *file, int line)
498{
499 struct mbuf *m = sb->sb_mb;
500 struct mbuf *n;
501
502 while (m && m->m_nextpkt)
503 m = m->m_nextpkt;
504
505 while (m && m->m_next)
506 m = m->m_next;
507
508 if (m != sb->sb_mbtail) {
509 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
510 __func__, sb->sb_mb, sb->sb_mbtail, m);
511 printf("packet tree:\n");
512 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
513 printf("\t");
514 for (n = m; n != NULL; n = n->m_next)
515 printf("%p ", n);
516 printf("\n");
517 }
518 panic("%s from %s:%u", __func__, file, line);
519 }
520}
521#endif /* SOCKBUF_DEBUG */
522
523#define SBLINKRECORD(sb, m0) do { \
524 if ((sb)->sb_lastrecord != NULL) \
525 (sb)->sb_lastrecord->m_nextpkt = (m0); \
526 else \
527 (sb)->sb_mb = (m0); \
528 (sb)->sb_lastrecord = (m0); \
529} while (/*CONSTCOND*/0)
530
531/*
532 * Append mbuf chain m to the last record in the
533 * socket buffer sb. The additional space associated
534 * the mbuf chain is recorded in sb. Empty mbufs are
535 * discarded and mbufs are compacted where possible.
536 */
537void
538sbappend(sb, m)
539 struct sockbuf *sb;
540 struct mbuf *m;
541{
542 register struct mbuf *n;
543
544 if (m == 0)
545 return;
546 SBLASTRECORDCHK(sb);
547 n = sb->sb_mb;
548 if (n) {
549 while (n->m_nextpkt)
550 n = n->m_nextpkt;
551 do {
552 if (n->m_flags & M_EOR) {
553 sbappendrecord(sb, m); /* XXXXXX!!!! */
554 return;
555 }
556 } while (n->m_next && (n = n->m_next));
557 } else {
558 /*
559 * XXX Would like to simply use sb_mbtail here, but
560 * XXX I need to verify that I won't miss an EOR that
561 * XXX way.
562 */
563 if ((n = sb->sb_lastrecord) != NULL) {
564 do {
565 if (n->m_flags & M_EOR) {
566 sbappendrecord(sb, m); /* XXXXXX!!!! */
567 return;
568 }
569 } while (n->m_next && (n = n->m_next));
570 } else {
571 /*
572 * If this is the first record in the socket buffer,
573 * it's also the last record.
574 */
575 sb->sb_lastrecord = m;
576 }
577 }
578 sbcompress(sb, m, n);
579 SBLASTRECORDCHK(sb);
580}
581
582/*
583 * This version of sbappend() should only be used when the caller
584 * absolutely knows that there will never be more than one record
585 * in the socket buffer, that is, a stream protocol (such as TCP).
586 */
587void
588sbappendstream(struct sockbuf *sb, struct mbuf *m)
589{
590
591 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
592 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
593
594 SBLASTMBUFCHK(sb);
595
596#ifdef MBUFTRACE
597 m_claim(m, sb->sb_mowner);
598#endif
599
600 sbcompress(sb, m, sb->sb_mbtail);
601
602 sb->sb_lastrecord = sb->sb_mb;
603 SBLASTRECORDCHK(sb);
604}
605
606#ifdef SOCKBUF_DEBUG
607void
608sbcheck(sb)
609 struct sockbuf *sb;
610{
611 struct mbuf *m;
612 struct mbuf *n = 0;
613 u_long len = 0, mbcnt = 0;
614
615 for (m = sb->sb_mb; m; m = n) {
616 n = m->m_nextpkt;
617 for (; m; m = m->m_next) {
618 len += m->m_len;
619 mbcnt += MSIZE;
620 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
621 mbcnt += m->m_ext.ext_size;
622 }
623 }
624 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
625 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
626 mbcnt, sb->sb_mbcnt);
627 panic("sbcheck");
628 }
629}
630#endif
631
632/*
633 * As above, except the mbuf chain
634 * begins a new record.
635 */
636void
637sbappendrecord(sb, m0)
638 register struct sockbuf *sb;
639 register struct mbuf *m0;
640{
641 register struct mbuf *m;
642
643 if (m0 == 0)
644 return;
645 m = sb->sb_mb;
646 if (m)
647 while (m->m_nextpkt)
648 m = m->m_nextpkt;
649 /*
650 * Put the first mbuf on the queue.
651 * Note this permits zero length records.
652 */
653 sballoc(sb, m0);
654 SBLASTRECORDCHK(sb);
655 SBLINKRECORD(sb, m0);
656 if (m)
657 m->m_nextpkt = m0;
658 else
659 sb->sb_mb = m0;
660 m = m0->m_next;
661 m0->m_next = 0;
662 if (m && (m0->m_flags & M_EOR)) {
663 m0->m_flags &= ~M_EOR;
664 m->m_flags |= M_EOR;
665 }
666 sbcompress(sb, m, m0);
667}
668
669/*
670 * As above except that OOB data
671 * is inserted at the beginning of the sockbuf,
672 * but after any other OOB data.
673 */
674void
675sbinsertoob(sb, m0)
676 register struct sockbuf *sb;
677 register struct mbuf *m0;
678{
679 register struct mbuf *m;
680 register struct mbuf **mp;
681
682 if (m0 == 0)
683 return;
684 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
685 m = *mp;
686 again:
687 switch (m->m_type) {
688
689 case MT_OOBDATA:
690 continue; /* WANT next train */
691
692 case MT_CONTROL:
693 m = m->m_next;
694 if (m)
695 goto again; /* inspect THIS train further */
696 }
697 break;
698 }
699 /*
700 * Put the first mbuf on the queue.
701 * Note this permits zero length records.
702 */
703 sballoc(sb, m0);
704 m0->m_nextpkt = *mp;
705 *mp = m0;
706 m = m0->m_next;
707 m0->m_next = 0;
708 if (m && (m0->m_flags & M_EOR)) {
709 m0->m_flags &= ~M_EOR;
710 m->m_flags |= M_EOR;
711 }
712 sbcompress(sb, m, m0);
713}
714
715/*
716 * Append address and data, and optionally, control (ancillary) data
717 * to the receive queue of a socket. If present,
718 * m0 must include a packet header with total length.
719 * Returns 0 if no space in sockbuf or insufficient mbufs.
720 */
721int
722sbappendaddr(sb, asa, m0, control)
723 struct sockbuf *sb;
724 struct sockaddr *asa;
725 struct mbuf *m0, *control;
726{
727 struct mbuf *m, *n, *nlast;
728 int space = asa->sa_len;
729
730 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
731 panic("sbappendaddr");
732 if (m0)
733 space += m0->m_pkthdr.len;
734 space += m_length(control, &n);
735 if (space > sbspace(sb))
736 return (0);
737#if MSIZE <= 256
738 if (asa->sa_len > MLEN)
739 return (0);
740#endif
741 MGET(m, M_DONTWAIT, MT_SONAME);
742 if (m == 0)
743 return (0);
744 m->m_len = asa->sa_len;
745 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
746 if (n)
747 n->m_next = m0; /* concatenate data to control */
748 else
749 control = m0;
750 m->m_next = control;
751 for (n = m; n->m_next != NULL; n = n->m_next)
752 sballoc(sb, n);
753 sballoc(sb, n);
754 nlast = n;
755 SBLINKRECORD(sb, m);
756
757 sb->sb_mbtail = nlast;
758 SBLASTMBUFCHK(sb);
759
760 SBLASTRECORDCHK(sb);
761 return (1);
762}
763
764int
765sbappendcontrol(sb, m0, control)
766 struct sockbuf *sb;
767 struct mbuf *control, *m0;
768{
769 struct mbuf *m, *n, *mlast;
770 int space;
771
772 if (control == 0)
773 panic("sbappendcontrol");
774 space = m_length(control, &n) + m_length(m0, NULL);
775 if (space > sbspace(sb))
776 return (0);
777 n->m_next = m0; /* concatenate data to control */
778
779 SBLASTRECORDCHK(sb);
780
781 for (m = control; m->m_next; m = m->m_next)
782 sballoc(sb, m);
783 sballoc(sb, m);
784 mlast = m;
785 SBLINKRECORD(sb, control);
786
787 sb->sb_mbtail = mlast;
788 SBLASTMBUFCHK(sb);
789
790 SBLASTRECORDCHK(sb);
791 return (1);
792}
793
794/*
795 * Compress mbuf chain m into the socket
796 * buffer sb following mbuf n. If n
797 * is null, the buffer is presumed empty.
798 */
799void
800sbcompress(sb, m, n)
801 register struct sockbuf *sb;
802 register struct mbuf *m, *n;
803{
804 register int eor = 0;
805 register struct mbuf *o;
806
807 while (m) {
808 eor |= m->m_flags & M_EOR;
809 if (m->m_len == 0 &&
810 (eor == 0 ||
811 (((o = m->m_next) || (o = n)) &&
812 o->m_type == m->m_type))) {
813 if (sb->sb_lastrecord == m)
814 sb->sb_lastrecord = m->m_next;
815 m = m_free(m);
816 continue;
817 }
818 if (n && (n->m_flags & M_EOR) == 0 &&
819 M_WRITABLE(n) &&
820 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
821 m->m_len <= M_TRAILINGSPACE(n) &&
822 n->m_type == m->m_type) {
823 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
824 (unsigned)m->m_len);
825 n->m_len += m->m_len;
826 sb->sb_cc += m->m_len;
827 if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
828 m->m_type != MT_OOBDATA)
829 /* XXX: Probably don't need.*/
830 sb->sb_ctl += m->m_len;
831 m = m_free(m);
832 continue;
833 }
834 if (n)
835 n->m_next = m;
836 else
837 sb->sb_mb = m;
838 sb->sb_mbtail = m;
839 sballoc(sb, m);
840 n = m;
841 m->m_flags &= ~M_EOR;
842 m = m->m_next;
843 n->m_next = 0;
844 }
845 if (eor) {
846 if (n)
847 n->m_flags |= eor;
848 else
849 printf("semi-panic: sbcompress\n");
850 }
851 SBLASTMBUFCHK(sb);
852}
853
854/*
855 * Free all mbufs in a sockbuf.
856 * Check that all resources are reclaimed.
857 */
858void
859sbflush(sb)
860 register struct sockbuf *sb;
861{
862
863 if (sb->sb_flags & SB_LOCK)
864 panic("sbflush: locked");
865 while (sb->sb_mbcnt) {
866 /*
867 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
868 * we would loop forever. Panic instead.
869 */
870 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
871 break;
872 sbdrop(sb, (int)sb->sb_cc);
873 }
874 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
875 panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
876}
877
878/*
879 * Drop data from (the front of) a sockbuf.
880 */
881void
882sbdrop(sb, len)
883 register struct sockbuf *sb;
884 register int len;
885{
886 register struct mbuf *m;
887 struct mbuf *next;
888
889 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
890 while (len > 0) {
891 if (m == 0) {
892 if (next == 0)
893 panic("sbdrop");
894 m = next;
895 next = m->m_nextpkt;
896 continue;
897 }
898 if (m->m_len > len) {
899 m->m_len -= len;
900 m->m_data += len;
901 sb->sb_cc -= len;
902 if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
903 m->m_type != MT_OOBDATA)
904 sb->sb_ctl -= len;
905 break;
906 }
907 len -= m->m_len;
908 sbfree(sb, m);
909 m = m_free(m);
910 }
911 while (m && m->m_len == 0) {
912 sbfree(sb, m);
913 m = m_free(m);
914 }
915 if (m) {
916 sb->sb_mb = m;
917 m->m_nextpkt = next;
918 } else
919 sb->sb_mb = next;
920 /*
921 * First part is an inline SB_EMPTY_FIXUP(). Second part
922 * makes sure sb_lastrecord is up-to-date if we dropped
923 * part of the last record.
924 */
925 m = sb->sb_mb;
926 if (m == NULL) {
927 sb->sb_mbtail = NULL;
928 sb->sb_lastrecord = NULL;
929 } else if (m->m_nextpkt == NULL) {
930 sb->sb_lastrecord = m;
931 }
932}
933
934/*
935 * Drop a record off the front of a sockbuf
936 * and move the next record to the front.
937 */
938void
939sbdroprecord(sb)
940 register struct sockbuf *sb;
941{
942 register struct mbuf *m;
943
944 m = sb->sb_mb;
945 if (m) {
946 sb->sb_mb = m->m_nextpkt;
947 do {
948 sbfree(sb, m);
949 m = m_free(m);
950 } while (m);
951 }
952 SB_EMPTY_FIXUP(sb);
953}
954
955/*
956 * Create a "control" mbuf containing the specified data
957 * with the specified type for presentation on a socket buffer.
958 */
959struct mbuf *
960sbcreatecontrol(p, size, type, level)
961 caddr_t p;
962 register int size;
963 int type, level;
964{
965 register struct cmsghdr *cp;
966 struct mbuf *m;
967
968 if (CMSG_SPACE((u_int)size) > MCLBYTES)
969 return ((struct mbuf *) NULL);
970 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
971 return ((struct mbuf *) NULL);
972 if (CMSG_SPACE((u_int)size) > MLEN) {
973 MCLGET(m, M_DONTWAIT);
974 if ((m->m_flags & M_EXT) == 0) {
975 m_free(m);
976 return ((struct mbuf *) NULL);
977 }
978 }
979 cp = mtod(m, struct cmsghdr *);
980 m->m_len = 0;
981 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
982 ("sbcreatecontrol: short mbuf"));
983 if (p != NULL)
984 (void)memcpy(CMSG_DATA(cp), p, size);
985 m->m_len = CMSG_SPACE(size);
986 cp->cmsg_len = CMSG_LEN(size);
987 cp->cmsg_level = level;
988 cp->cmsg_type = type;
989 return (m);
990}
991
992/*
993 * Some routines that return EOPNOTSUPP for entry points that are not
994 * supported by a protocol. Fill in as needed.
995 */
996int
997pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
998{
999 return EOPNOTSUPP;
1000}
1001
1002int
1003pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
1004{
1005 return EOPNOTSUPP;
1006}
1007
1008int
1009pru_connect2_notsupp(struct socket *so1, struct socket *so2)
1010{
1011 return EOPNOTSUPP;
1012}
1013
1014int
1015pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
1016 struct ifnet *ifp, struct thread *td)
1017{
1018 return EOPNOTSUPP;
1019}
1020
1021int
1022pru_listen_notsupp(struct socket *so, struct thread *td)
1023{
1024 return EOPNOTSUPP;
1025}
1026
1027int
1028pru_rcvd_notsupp(struct socket *so, int flags)
1029{
1030 return EOPNOTSUPP;
1031}
1032
1033int
1034pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
1035{
1036 return EOPNOTSUPP;
1037}
1038
1039/*
1040 * This isn't really a ``null'' operation, but it's the default one
1041 * and doesn't do anything destructive.
1042 */
1043int
1044pru_sense_null(struct socket *so, struct stat *sb)
1045{
1046 sb->st_blksize = so->so_snd.sb_hiwat;
1047 return 0;
1048}
1049
1050/*
1051 * For protocol types that don't keep cached copies of labels in their
1052 * pcbs, provide a null sosetlabel that does a NOOP.
1053 */
1054void
1055pru_sosetlabel_null(struct socket *so)
1056{
1057
1058}
1059
1060/*
1061 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
1062 */
1063struct sockaddr *
| 38
39#include "opt_mac.h"
40#include "opt_param.h"
41
42#include <sys/param.h>
43#include <sys/aio.h> /* for aio_swake proto */
44#include <sys/domain.h>
45#include <sys/event.h>
46#include <sys/file.h> /* for maxfiles */
47#include <sys/kernel.h>
48#include <sys/lock.h>
49#include <sys/mac.h>
50#include <sys/malloc.h>
51#include <sys/mbuf.h>
52#include <sys/mutex.h>
53#include <sys/proc.h>
54#include <sys/protosw.h>
55#include <sys/resourcevar.h>
56#include <sys/signalvar.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/stat.h>
60#include <sys/sysctl.h>
61#include <sys/systm.h>
62
63int maxsockets;
64
65void (*aio_swake)(struct socket *, struct sockbuf *);
66
67/*
68 * Primitive routines for operating on sockets and socket buffers
69 */
70
71u_long sb_max = SB_MAX;
72static u_long sb_max_adj =
73 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
74
75static u_long sb_efficiency = 8; /* parameter for sbreserve() */
76
77/*
78 * Procedures to manipulate state flags of socket
79 * and do appropriate wakeups. Normal sequence from the
80 * active (originating) side is that soisconnecting() is
81 * called during processing of connect() call,
82 * resulting in an eventual call to soisconnected() if/when the
83 * connection is established. When the connection is torn down
84 * soisdisconnecting() is called during processing of disconnect() call,
85 * and soisdisconnected() is called when the connection to the peer
86 * is totally severed. The semantics of these routines are such that
87 * connectionless protocols can call soisconnected() and soisdisconnected()
88 * only, bypassing the in-progress calls when setting up a ``connection''
89 * takes no time.
90 *
91 * From the passive side, a socket is created with
92 * two queues of sockets: so_incomp for connections in progress
93 * and so_comp for connections already made and awaiting user acceptance.
94 * As a protocol is preparing incoming connections, it creates a socket
95 * structure queued on so_incomp by calling sonewconn(). When the connection
96 * is established, soisconnected() is called, and transfers the
97 * socket structure to so_comp, making it available to accept().
98 *
99 * If a socket is closed with sockets on either
100 * so_incomp or so_comp, these sockets are dropped.
101 *
102 * If higher level protocols are implemented in
103 * the kernel, the wakeups done here will sometimes
104 * cause software-interrupt process scheduling.
105 */
106
107void
108soisconnecting(so)
109 register struct socket *so;
110{
111
112 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
113 so->so_state |= SS_ISCONNECTING;
114}
115
116void
117soisconnected(so)
118 struct socket *so;
119{
120 struct socket *head = so->so_head;
121
122 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
123 so->so_state |= SS_ISCONNECTED;
124 if (head && (so->so_state & SS_INCOMP)) {
125 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
126 so->so_upcall = head->so_accf->so_accept_filter->accf_callback;
127 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
128 so->so_rcv.sb_flags |= SB_UPCALL;
129 so->so_options &= ~SO_ACCEPTFILTER;
130 so->so_upcall(so, so->so_upcallarg, M_TRYWAIT);
131 return;
132 }
133 TAILQ_REMOVE(&head->so_incomp, so, so_list);
134 head->so_incqlen--;
135 so->so_state &= ~SS_INCOMP;
136 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
137 head->so_qlen++;
138 so->so_state |= SS_COMP;
139 sorwakeup(head);
140 wakeup_one(&head->so_timeo);
141 } else {
142 wakeup(&so->so_timeo);
143 sorwakeup(so);
144 sowwakeup(so);
145 }
146}
147
148void
149soisdisconnecting(so)
150 register struct socket *so;
151{
152
153 so->so_state &= ~SS_ISCONNECTING;
154 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
155 wakeup(&so->so_timeo);
156 sowwakeup(so);
157 sorwakeup(so);
158}
159
160void
161soisdisconnected(so)
162 register struct socket *so;
163{
164
165 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
166 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
167 wakeup(&so->so_timeo);
168 sbdrop(&so->so_snd, so->so_snd.sb_cc);
169 sowwakeup(so);
170 sorwakeup(so);
171}
172
173/*
174 * When an attempt at a new connection is noted on a socket
175 * which accepts connections, sonewconn is called. If the
176 * connection is possible (subject to space constraints, etc.)
177 * then we allocate a new structure, propoerly linked into the
178 * data structure of the original socket, and return this.
179 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
180 *
181 * note: the ref count on the socket is 0 on return
182 */
183struct socket *
184sonewconn(head, connstatus)
185 register struct socket *head;
186 int connstatus;
187{
188 register struct socket *so;
189
190 if (head->so_qlen > 3 * head->so_qlimit / 2)
191 return ((struct socket *)0);
192 so = soalloc(M_NOWAIT);
193 if (so == NULL)
194 return ((struct socket *)0);
195 if ((head->so_options & SO_ACCEPTFILTER) != 0)
196 connstatus = 0;
197 so->so_head = head;
198 so->so_type = head->so_type;
199 so->so_options = head->so_options &~ SO_ACCEPTCONN;
200 so->so_linger = head->so_linger;
201 so->so_state = head->so_state | SS_NOFDREF;
202 so->so_proto = head->so_proto;
203 so->so_timeo = head->so_timeo;
204 so->so_cred = crhold(head->so_cred);
205#ifdef MAC
206 mac_create_socket_from_socket(head, so);
207#endif
208 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
209 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
210 sodealloc(so);
211 return ((struct socket *)0);
212 }
213
214 if (connstatus) {
215 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
216 so->so_state |= SS_COMP;
217 head->so_qlen++;
218 } else {
219 if (head->so_incqlen > head->so_qlimit) {
220 struct socket *sp;
221 sp = TAILQ_FIRST(&head->so_incomp);
222 (void) soabort(sp);
223 }
224 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
225 so->so_state |= SS_INCOMP;
226 head->so_incqlen++;
227 }
228 if (connstatus) {
229 sorwakeup(head);
230 wakeup(&head->so_timeo);
231 so->so_state |= connstatus;
232 }
233 return (so);
234}
235
236/*
237 * Socantsendmore indicates that no more data will be sent on the
238 * socket; it would normally be applied to a socket when the user
239 * informs the system that no more data is to be sent, by the protocol
240 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
241 * will be received, and will normally be applied to the socket by a
242 * protocol when it detects that the peer will send no more data.
243 * Data queued for reading in the socket may yet be read.
244 */
245
246void
247socantsendmore(so)
248 struct socket *so;
249{
250
251 so->so_state |= SS_CANTSENDMORE;
252 sowwakeup(so);
253}
254
255void
256socantrcvmore(so)
257 struct socket *so;
258{
259
260 so->so_state |= SS_CANTRCVMORE;
261 sorwakeup(so);
262}
263
264/*
265 * Wait for data to arrive at/drain from a socket buffer.
266 */
267int
268sbwait(sb)
269 struct sockbuf *sb;
270{
271
272 sb->sb_flags |= SB_WAIT;
273 return (tsleep(&sb->sb_cc,
274 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
275 sb->sb_timeo));
276}
277
278/*
279 * Lock a sockbuf already known to be locked;
280 * return any error returned from sleep (EINTR).
281 */
282int
283sb_lock(sb)
284 register struct sockbuf *sb;
285{
286 int error;
287
288 while (sb->sb_flags & SB_LOCK) {
289 sb->sb_flags |= SB_WANT;
290 error = tsleep(&sb->sb_flags,
291 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
292 "sblock", 0);
293 if (error)
294 return (error);
295 }
296 sb->sb_flags |= SB_LOCK;
297 return (0);
298}
299
300/*
301 * Wakeup processes waiting on a socket buffer.
302 * Do asynchronous notification via SIGIO
303 * if the socket has the SS_ASYNC flag set.
304 */
305void
306sowakeup(so, sb)
307 register struct socket *so;
308 register struct sockbuf *sb;
309{
310
311 selwakeuppri(&sb->sb_sel, PSOCK);
312 sb->sb_flags &= ~SB_SEL;
313 if (sb->sb_flags & SB_WAIT) {
314 sb->sb_flags &= ~SB_WAIT;
315 wakeup(&sb->sb_cc);
316 }
317 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
318 pgsigio(&so->so_sigio, SIGIO, 0);
319 if (sb->sb_flags & SB_UPCALL)
320 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
321 if (sb->sb_flags & SB_AIO)
322 aio_swake(so, sb);
323 KNOTE(&sb->sb_sel.si_note, 0);
324}
325
326/*
327 * Socket buffer (struct sockbuf) utility routines.
328 *
329 * Each socket contains two socket buffers: one for sending data and
330 * one for receiving data. Each buffer contains a queue of mbufs,
331 * information about the number of mbufs and amount of data in the
332 * queue, and other fields allowing select() statements and notification
333 * on data availability to be implemented.
334 *
335 * Data stored in a socket buffer is maintained as a list of records.
336 * Each record is a list of mbufs chained together with the m_next
337 * field. Records are chained together with the m_nextpkt field. The upper
338 * level routine soreceive() expects the following conventions to be
339 * observed when placing information in the receive buffer:
340 *
341 * 1. If the protocol requires each message be preceded by the sender's
342 * name, then a record containing that name must be present before
343 * any associated data (mbuf's must be of type MT_SONAME).
344 * 2. If the protocol supports the exchange of ``access rights'' (really
345 * just additional data associated with the message), and there are
346 * ``rights'' to be received, then a record containing this data
347 * should be present (mbuf's must be of type MT_RIGHTS).
348 * 3. If a name or rights record exists, then it must be followed by
349 * a data record, perhaps of zero length.
350 *
351 * Before using a new socket structure it is first necessary to reserve
352 * buffer space to the socket, by calling sbreserve(). This should commit
353 * some of the available buffer space in the system buffer pool for the
354 * socket (currently, it does nothing but enforce limits). The space
355 * should be released by calling sbrelease() when the socket is destroyed.
356 */
357
358int
359soreserve(so, sndcc, rcvcc)
360 register struct socket *so;
361 u_long sndcc, rcvcc;
362{
363 struct thread *td = curthread;
364
365 if (sbreserve(&so->so_snd, sndcc, so, td) == 0)
366 goto bad;
367 if (sbreserve(&so->so_rcv, rcvcc, so, td) == 0)
368 goto bad2;
369 if (so->so_rcv.sb_lowat == 0)
370 so->so_rcv.sb_lowat = 1;
371 if (so->so_snd.sb_lowat == 0)
372 so->so_snd.sb_lowat = MCLBYTES;
373 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
374 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
375 return (0);
376bad2:
377 sbrelease(&so->so_snd, so);
378bad:
379 return (ENOBUFS);
380}
381
382static int
383sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
384{
385 int error = 0;
386 u_long old_sb_max = sb_max;
387
388 error = SYSCTL_OUT(req, arg1, sizeof(u_long));
389 if (error || !req->newptr)
390 return (error);
391 error = SYSCTL_IN(req, arg1, sizeof(u_long));
392 if (error)
393 return (error);
394 if (sb_max < MSIZE + MCLBYTES) {
395 sb_max = old_sb_max;
396 return (EINVAL);
397 }
398 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
399 return (0);
400}
401
402/*
403 * Allot mbufs to a sockbuf.
404 * Attempt to scale mbmax so that mbcnt doesn't become limiting
405 * if buffering efficiency is near the normal case.
406 */
407int
408sbreserve(sb, cc, so, td)
409 struct sockbuf *sb;
410 u_long cc;
411 struct socket *so;
412 struct thread *td;
413{
414 rlim_t sbsize_limit;
415
416 /*
417 * td will only be NULL when we're in an interrupt
418 * (e.g. in tcp_input())
419 */
420 if (cc > sb_max_adj)
421 return (0);
422 if (td != NULL) {
423 PROC_LOCK(td->td_proc);
424 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
425 PROC_UNLOCK(td->td_proc);
426 } else
427 sbsize_limit = RLIM_INFINITY;
428 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
429 sbsize_limit))
430 return (0);
431 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
432 if (sb->sb_lowat > sb->sb_hiwat)
433 sb->sb_lowat = sb->sb_hiwat;
434 return (1);
435}
436
437/*
438 * Free mbufs held by a socket, and reserved mbuf space.
439 */
440void
441sbrelease(sb, so)
442 struct sockbuf *sb;
443 struct socket *so;
444{
445
446 sbflush(sb);
447 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
448 RLIM_INFINITY);
449 sb->sb_mbmax = 0;
450}
451
452/*
453 * Routines to add and remove
454 * data from an mbuf queue.
455 *
456 * The routines sbappend() or sbappendrecord() are normally called to
457 * append new mbufs to a socket buffer, after checking that adequate
458 * space is available, comparing the function sbspace() with the amount
459 * of data to be added. sbappendrecord() differs from sbappend() in
460 * that data supplied is treated as the beginning of a new record.
461 * To place a sender's address, optional access rights, and data in a
462 * socket receive buffer, sbappendaddr() should be used. To place
463 * access rights and data in a socket receive buffer, sbappendrights()
464 * should be used. In either case, the new data begins a new record.
465 * Note that unlike sbappend() and sbappendrecord(), these routines check
466 * for the caller that there will be enough space to store the data.
467 * Each fails if there is not enough space, or if it cannot find mbufs
468 * to store additional information in.
469 *
470 * Reliable protocols may use the socket send buffer to hold data
471 * awaiting acknowledgement. Data is normally copied from a socket
472 * send buffer in a protocol with m_copy for output to a peer,
473 * and then removing the data from the socket buffer with sbdrop()
474 * or sbdroprecord() when the data is acknowledged by the peer.
475 */
476
477#ifdef SOCKBUF_DEBUG
478void
479sblastrecordchk(struct sockbuf *sb, const char *file, int line)
480{
481 struct mbuf *m = sb->sb_mb;
482
483 while (m && m->m_nextpkt)
484 m = m->m_nextpkt;
485
486 if (m != sb->sb_lastrecord) {
487 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
488 __func__, sb->sb_mb, sb->sb_lastrecord, m);
489 printf("packet chain:\n");
490 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
491 printf("\t%p\n", m);
492 panic("%s from %s:%u", __func__, file, line);
493 }
494}
495
496void
497sblastmbufchk(struct sockbuf *sb, const char *file, int line)
498{
499 struct mbuf *m = sb->sb_mb;
500 struct mbuf *n;
501
502 while (m && m->m_nextpkt)
503 m = m->m_nextpkt;
504
505 while (m && m->m_next)
506 m = m->m_next;
507
508 if (m != sb->sb_mbtail) {
509 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
510 __func__, sb->sb_mb, sb->sb_mbtail, m);
511 printf("packet tree:\n");
512 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
513 printf("\t");
514 for (n = m; n != NULL; n = n->m_next)
515 printf("%p ", n);
516 printf("\n");
517 }
518 panic("%s from %s:%u", __func__, file, line);
519 }
520}
521#endif /* SOCKBUF_DEBUG */
522
523#define SBLINKRECORD(sb, m0) do { \
524 if ((sb)->sb_lastrecord != NULL) \
525 (sb)->sb_lastrecord->m_nextpkt = (m0); \
526 else \
527 (sb)->sb_mb = (m0); \
528 (sb)->sb_lastrecord = (m0); \
529} while (/*CONSTCOND*/0)
530
531/*
532 * Append mbuf chain m to the last record in the
533 * socket buffer sb. The additional space associated
534 * the mbuf chain is recorded in sb. Empty mbufs are
535 * discarded and mbufs are compacted where possible.
536 */
537void
538sbappend(sb, m)
539 struct sockbuf *sb;
540 struct mbuf *m;
541{
542 register struct mbuf *n;
543
544 if (m == 0)
545 return;
546 SBLASTRECORDCHK(sb);
547 n = sb->sb_mb;
548 if (n) {
549 while (n->m_nextpkt)
550 n = n->m_nextpkt;
551 do {
552 if (n->m_flags & M_EOR) {
553 sbappendrecord(sb, m); /* XXXXXX!!!! */
554 return;
555 }
556 } while (n->m_next && (n = n->m_next));
557 } else {
558 /*
559 * XXX Would like to simply use sb_mbtail here, but
560 * XXX I need to verify that I won't miss an EOR that
561 * XXX way.
562 */
563 if ((n = sb->sb_lastrecord) != NULL) {
564 do {
565 if (n->m_flags & M_EOR) {
566 sbappendrecord(sb, m); /* XXXXXX!!!! */
567 return;
568 }
569 } while (n->m_next && (n = n->m_next));
570 } else {
571 /*
572 * If this is the first record in the socket buffer,
573 * it's also the last record.
574 */
575 sb->sb_lastrecord = m;
576 }
577 }
578 sbcompress(sb, m, n);
579 SBLASTRECORDCHK(sb);
580}
581
582/*
583 * This version of sbappend() should only be used when the caller
584 * absolutely knows that there will never be more than one record
585 * in the socket buffer, that is, a stream protocol (such as TCP).
586 */
587void
588sbappendstream(struct sockbuf *sb, struct mbuf *m)
589{
590
591 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
592 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
593
594 SBLASTMBUFCHK(sb);
595
596#ifdef MBUFTRACE
597 m_claim(m, sb->sb_mowner);
598#endif
599
600 sbcompress(sb, m, sb->sb_mbtail);
601
602 sb->sb_lastrecord = sb->sb_mb;
603 SBLASTRECORDCHK(sb);
604}
605
606#ifdef SOCKBUF_DEBUG
607void
608sbcheck(sb)
609 struct sockbuf *sb;
610{
611 struct mbuf *m;
612 struct mbuf *n = 0;
613 u_long len = 0, mbcnt = 0;
614
615 for (m = sb->sb_mb; m; m = n) {
616 n = m->m_nextpkt;
617 for (; m; m = m->m_next) {
618 len += m->m_len;
619 mbcnt += MSIZE;
620 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
621 mbcnt += m->m_ext.ext_size;
622 }
623 }
624 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
625 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
626 mbcnt, sb->sb_mbcnt);
627 panic("sbcheck");
628 }
629}
630#endif
631
632/*
633 * As above, except the mbuf chain
634 * begins a new record.
635 */
636void
637sbappendrecord(sb, m0)
638 register struct sockbuf *sb;
639 register struct mbuf *m0;
640{
641 register struct mbuf *m;
642
643 if (m0 == 0)
644 return;
645 m = sb->sb_mb;
646 if (m)
647 while (m->m_nextpkt)
648 m = m->m_nextpkt;
649 /*
650 * Put the first mbuf on the queue.
651 * Note this permits zero length records.
652 */
653 sballoc(sb, m0);
654 SBLASTRECORDCHK(sb);
655 SBLINKRECORD(sb, m0);
656 if (m)
657 m->m_nextpkt = m0;
658 else
659 sb->sb_mb = m0;
660 m = m0->m_next;
661 m0->m_next = 0;
662 if (m && (m0->m_flags & M_EOR)) {
663 m0->m_flags &= ~M_EOR;
664 m->m_flags |= M_EOR;
665 }
666 sbcompress(sb, m, m0);
667}
668
669/*
670 * As above except that OOB data
671 * is inserted at the beginning of the sockbuf,
672 * but after any other OOB data.
673 */
674void
675sbinsertoob(sb, m0)
676 register struct sockbuf *sb;
677 register struct mbuf *m0;
678{
679 register struct mbuf *m;
680 register struct mbuf **mp;
681
682 if (m0 == 0)
683 return;
684 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
685 m = *mp;
686 again:
687 switch (m->m_type) {
688
689 case MT_OOBDATA:
690 continue; /* WANT next train */
691
692 case MT_CONTROL:
693 m = m->m_next;
694 if (m)
695 goto again; /* inspect THIS train further */
696 }
697 break;
698 }
699 /*
700 * Put the first mbuf on the queue.
701 * Note this permits zero length records.
702 */
703 sballoc(sb, m0);
704 m0->m_nextpkt = *mp;
705 *mp = m0;
706 m = m0->m_next;
707 m0->m_next = 0;
708 if (m && (m0->m_flags & M_EOR)) {
709 m0->m_flags &= ~M_EOR;
710 m->m_flags |= M_EOR;
711 }
712 sbcompress(sb, m, m0);
713}
714
715/*
716 * Append address and data, and optionally, control (ancillary) data
717 * to the receive queue of a socket. If present,
718 * m0 must include a packet header with total length.
719 * Returns 0 if no space in sockbuf or insufficient mbufs.
720 */
721int
722sbappendaddr(sb, asa, m0, control)
723 struct sockbuf *sb;
724 struct sockaddr *asa;
725 struct mbuf *m0, *control;
726{
727 struct mbuf *m, *n, *nlast;
728 int space = asa->sa_len;
729
730 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
731 panic("sbappendaddr");
732 if (m0)
733 space += m0->m_pkthdr.len;
734 space += m_length(control, &n);
735 if (space > sbspace(sb))
736 return (0);
737#if MSIZE <= 256
738 if (asa->sa_len > MLEN)
739 return (0);
740#endif
741 MGET(m, M_DONTWAIT, MT_SONAME);
742 if (m == 0)
743 return (0);
744 m->m_len = asa->sa_len;
745 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
746 if (n)
747 n->m_next = m0; /* concatenate data to control */
748 else
749 control = m0;
750 m->m_next = control;
751 for (n = m; n->m_next != NULL; n = n->m_next)
752 sballoc(sb, n);
753 sballoc(sb, n);
754 nlast = n;
755 SBLINKRECORD(sb, m);
756
757 sb->sb_mbtail = nlast;
758 SBLASTMBUFCHK(sb);
759
760 SBLASTRECORDCHK(sb);
761 return (1);
762}
763
764int
765sbappendcontrol(sb, m0, control)
766 struct sockbuf *sb;
767 struct mbuf *control, *m0;
768{
769 struct mbuf *m, *n, *mlast;
770 int space;
771
772 if (control == 0)
773 panic("sbappendcontrol");
774 space = m_length(control, &n) + m_length(m0, NULL);
775 if (space > sbspace(sb))
776 return (0);
777 n->m_next = m0; /* concatenate data to control */
778
779 SBLASTRECORDCHK(sb);
780
781 for (m = control; m->m_next; m = m->m_next)
782 sballoc(sb, m);
783 sballoc(sb, m);
784 mlast = m;
785 SBLINKRECORD(sb, control);
786
787 sb->sb_mbtail = mlast;
788 SBLASTMBUFCHK(sb);
789
790 SBLASTRECORDCHK(sb);
791 return (1);
792}
793
794/*
795 * Compress mbuf chain m into the socket
796 * buffer sb following mbuf n. If n
797 * is null, the buffer is presumed empty.
798 */
799void
800sbcompress(sb, m, n)
801 register struct sockbuf *sb;
802 register struct mbuf *m, *n;
803{
804 register int eor = 0;
805 register struct mbuf *o;
806
807 while (m) {
808 eor |= m->m_flags & M_EOR;
809 if (m->m_len == 0 &&
810 (eor == 0 ||
811 (((o = m->m_next) || (o = n)) &&
812 o->m_type == m->m_type))) {
813 if (sb->sb_lastrecord == m)
814 sb->sb_lastrecord = m->m_next;
815 m = m_free(m);
816 continue;
817 }
818 if (n && (n->m_flags & M_EOR) == 0 &&
819 M_WRITABLE(n) &&
820 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
821 m->m_len <= M_TRAILINGSPACE(n) &&
822 n->m_type == m->m_type) {
823 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
824 (unsigned)m->m_len);
825 n->m_len += m->m_len;
826 sb->sb_cc += m->m_len;
827 if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
828 m->m_type != MT_OOBDATA)
829 /* XXX: Probably don't need.*/
830 sb->sb_ctl += m->m_len;
831 m = m_free(m);
832 continue;
833 }
834 if (n)
835 n->m_next = m;
836 else
837 sb->sb_mb = m;
838 sb->sb_mbtail = m;
839 sballoc(sb, m);
840 n = m;
841 m->m_flags &= ~M_EOR;
842 m = m->m_next;
843 n->m_next = 0;
844 }
845 if (eor) {
846 if (n)
847 n->m_flags |= eor;
848 else
849 printf("semi-panic: sbcompress\n");
850 }
851 SBLASTMBUFCHK(sb);
852}
853
854/*
855 * Free all mbufs in a sockbuf.
856 * Check that all resources are reclaimed.
857 */
858void
859sbflush(sb)
860 register struct sockbuf *sb;
861{
862
863 if (sb->sb_flags & SB_LOCK)
864 panic("sbflush: locked");
865 while (sb->sb_mbcnt) {
866 /*
867 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
868 * we would loop forever. Panic instead.
869 */
870 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
871 break;
872 sbdrop(sb, (int)sb->sb_cc);
873 }
874 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
875 panic("sbflush: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
876}
877
878/*
879 * Drop data from (the front of) a sockbuf.
880 */
881void
882sbdrop(sb, len)
883 register struct sockbuf *sb;
884 register int len;
885{
886 register struct mbuf *m;
887 struct mbuf *next;
888
889 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
890 while (len > 0) {
891 if (m == 0) {
892 if (next == 0)
893 panic("sbdrop");
894 m = next;
895 next = m->m_nextpkt;
896 continue;
897 }
898 if (m->m_len > len) {
899 m->m_len -= len;
900 m->m_data += len;
901 sb->sb_cc -= len;
902 if (m->m_type != MT_DATA && m->m_type != MT_HEADER &&
903 m->m_type != MT_OOBDATA)
904 sb->sb_ctl -= len;
905 break;
906 }
907 len -= m->m_len;
908 sbfree(sb, m);
909 m = m_free(m);
910 }
911 while (m && m->m_len == 0) {
912 sbfree(sb, m);
913 m = m_free(m);
914 }
915 if (m) {
916 sb->sb_mb = m;
917 m->m_nextpkt = next;
918 } else
919 sb->sb_mb = next;
920 /*
921 * First part is an inline SB_EMPTY_FIXUP(). Second part
922 * makes sure sb_lastrecord is up-to-date if we dropped
923 * part of the last record.
924 */
925 m = sb->sb_mb;
926 if (m == NULL) {
927 sb->sb_mbtail = NULL;
928 sb->sb_lastrecord = NULL;
929 } else if (m->m_nextpkt == NULL) {
930 sb->sb_lastrecord = m;
931 }
932}
933
934/*
935 * Drop a record off the front of a sockbuf
936 * and move the next record to the front.
937 */
938void
939sbdroprecord(sb)
940 register struct sockbuf *sb;
941{
942 register struct mbuf *m;
943
944 m = sb->sb_mb;
945 if (m) {
946 sb->sb_mb = m->m_nextpkt;
947 do {
948 sbfree(sb, m);
949 m = m_free(m);
950 } while (m);
951 }
952 SB_EMPTY_FIXUP(sb);
953}
954
955/*
956 * Create a "control" mbuf containing the specified data
957 * with the specified type for presentation on a socket buffer.
958 */
959struct mbuf *
960sbcreatecontrol(p, size, type, level)
961 caddr_t p;
962 register int size;
963 int type, level;
964{
965 register struct cmsghdr *cp;
966 struct mbuf *m;
967
968 if (CMSG_SPACE((u_int)size) > MCLBYTES)
969 return ((struct mbuf *) NULL);
970 if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
971 return ((struct mbuf *) NULL);
972 if (CMSG_SPACE((u_int)size) > MLEN) {
973 MCLGET(m, M_DONTWAIT);
974 if ((m->m_flags & M_EXT) == 0) {
975 m_free(m);
976 return ((struct mbuf *) NULL);
977 }
978 }
979 cp = mtod(m, struct cmsghdr *);
980 m->m_len = 0;
981 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
982 ("sbcreatecontrol: short mbuf"));
983 if (p != NULL)
984 (void)memcpy(CMSG_DATA(cp), p, size);
985 m->m_len = CMSG_SPACE(size);
986 cp->cmsg_len = CMSG_LEN(size);
987 cp->cmsg_level = level;
988 cp->cmsg_type = type;
989 return (m);
990}
991
992/*
993 * Some routines that return EOPNOTSUPP for entry points that are not
994 * supported by a protocol. Fill in as needed.
995 */
996int
997pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
998{
999 return EOPNOTSUPP;
1000}
1001
1002int
1003pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
1004{
1005 return EOPNOTSUPP;
1006}
1007
1008int
1009pru_connect2_notsupp(struct socket *so1, struct socket *so2)
1010{
1011 return EOPNOTSUPP;
1012}
1013
1014int
1015pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
1016 struct ifnet *ifp, struct thread *td)
1017{
1018 return EOPNOTSUPP;
1019}
1020
1021int
1022pru_listen_notsupp(struct socket *so, struct thread *td)
1023{
1024 return EOPNOTSUPP;
1025}
1026
1027int
1028pru_rcvd_notsupp(struct socket *so, int flags)
1029{
1030 return EOPNOTSUPP;
1031}
1032
1033int
1034pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
1035{
1036 return EOPNOTSUPP;
1037}
1038
1039/*
1040 * This isn't really a ``null'' operation, but it's the default one
1041 * and doesn't do anything destructive.
1042 */
1043int
1044pru_sense_null(struct socket *so, struct stat *sb)
1045{
1046 sb->st_blksize = so->so_snd.sb_hiwat;
1047 return 0;
1048}
1049
1050/*
1051 * For protocol types that don't keep cached copies of labels in their
1052 * pcbs, provide a null sosetlabel that does a NOOP.
1053 */
1054void
1055pru_sosetlabel_null(struct socket *so)
1056{
1057
1058}
1059
1060/*
1061 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
1062 */
1063struct sockaddr *
|
1072 if (sa2)
1073 bcopy(sa, sa2, sa->sa_len);
1074 return sa2;
1075}
1076
1077/*
1078 * Create an external-format (``xsocket'') structure using the information
1079 * in the kernel-format socket structure pointed to by so. This is done
1080 * to reduce the spew of irrelevant information over this interface,
1081 * to isolate user code from changes in the kernel structure, and
1082 * potentially to provide information-hiding if we decide that
1083 * some of this information should be hidden from users.
1084 */
1085void
1086sotoxsocket(struct socket *so, struct xsocket *xso)
1087{
1088 xso->xso_len = sizeof *xso;
1089 xso->xso_so = so;
1090 xso->so_type = so->so_type;
1091 xso->so_options = so->so_options;
1092 xso->so_linger = so->so_linger;
1093 xso->so_state = so->so_state;
1094 xso->so_pcb = so->so_pcb;
1095 xso->xso_protocol = so->so_proto->pr_protocol;
1096 xso->xso_family = so->so_proto->pr_domain->dom_family;
1097 xso->so_qlen = so->so_qlen;
1098 xso->so_incqlen = so->so_incqlen;
1099 xso->so_qlimit = so->so_qlimit;
1100 xso->so_timeo = so->so_timeo;
1101 xso->so_error = so->so_error;
1102 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
1103 xso->so_oobmark = so->so_oobmark;
1104 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
1105 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
1106 xso->so_uid = so->so_cred->cr_uid;
1107}
1108
1109/*
1110 * This does the same for sockbufs. Note that the xsockbuf structure,
1111 * since it is always embedded in a socket, does not include a self
1112 * pointer nor a length. We make this entry point public in case
1113 * some other mechanism needs it.
1114 */
1115void
1116sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1117{
1118 xsb->sb_cc = sb->sb_cc;
1119 xsb->sb_hiwat = sb->sb_hiwat;
1120 xsb->sb_mbcnt = sb->sb_mbcnt;
1121 xsb->sb_mbmax = sb->sb_mbmax;
1122 xsb->sb_lowat = sb->sb_lowat;
1123 xsb->sb_flags = sb->sb_flags;
1124 xsb->sb_timeo = sb->sb_timeo;
1125}
1126
1127/*
1128 * Here is the definition of some of the basic objects in the kern.ipc
1129 * branch of the MIB.
1130 */
1131SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
1132
1133/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1134static int dummy;
1135SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1136SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1137 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1138SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RDTUN,
1139 &maxsockets, 0, "Maximum number of sockets avaliable");
1140SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1141 &sb_efficiency, 0, "");
1142
1143/*
1144 * Initialise maxsockets
1145 */
1146static void init_maxsockets(void *ignored)
1147{
1148 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1149 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1150}
1151SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
| 1069 if (sa2)
1070 bcopy(sa, sa2, sa->sa_len);
1071 return sa2;
1072}
1073
1074/*
1075 * Create an external-format (``xsocket'') structure using the information
1076 * in the kernel-format socket structure pointed to by so. This is done
1077 * to reduce the spew of irrelevant information over this interface,
1078 * to isolate user code from changes in the kernel structure, and
1079 * potentially to provide information-hiding if we decide that
1080 * some of this information should be hidden from users.
1081 */
1082void
1083sotoxsocket(struct socket *so, struct xsocket *xso)
1084{
1085 xso->xso_len = sizeof *xso;
1086 xso->xso_so = so;
1087 xso->so_type = so->so_type;
1088 xso->so_options = so->so_options;
1089 xso->so_linger = so->so_linger;
1090 xso->so_state = so->so_state;
1091 xso->so_pcb = so->so_pcb;
1092 xso->xso_protocol = so->so_proto->pr_protocol;
1093 xso->xso_family = so->so_proto->pr_domain->dom_family;
1094 xso->so_qlen = so->so_qlen;
1095 xso->so_incqlen = so->so_incqlen;
1096 xso->so_qlimit = so->so_qlimit;
1097 xso->so_timeo = so->so_timeo;
1098 xso->so_error = so->so_error;
1099 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
1100 xso->so_oobmark = so->so_oobmark;
1101 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
1102 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
1103 xso->so_uid = so->so_cred->cr_uid;
1104}
1105
1106/*
1107 * This does the same for sockbufs. Note that the xsockbuf structure,
1108 * since it is always embedded in a socket, does not include a self
1109 * pointer nor a length. We make this entry point public in case
1110 * some other mechanism needs it.
1111 */
1112void
1113sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1114{
1115 xsb->sb_cc = sb->sb_cc;
1116 xsb->sb_hiwat = sb->sb_hiwat;
1117 xsb->sb_mbcnt = sb->sb_mbcnt;
1118 xsb->sb_mbmax = sb->sb_mbmax;
1119 xsb->sb_lowat = sb->sb_lowat;
1120 xsb->sb_flags = sb->sb_flags;
1121 xsb->sb_timeo = sb->sb_timeo;
1122}
1123
1124/*
1125 * Here is the definition of some of the basic objects in the kern.ipc
1126 * branch of the MIB.
1127 */
1128SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
1129
1130/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1131static int dummy;
1132SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1133SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1134 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1135SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RDTUN,
1136 &maxsockets, 0, "Maximum number of sockets avaliable");
1137SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1138 &sb_efficiency, 0, "");
1139
1140/*
1141 * Initialise maxsockets
1142 */
1143static void init_maxsockets(void *ignored)
1144{
1145 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1146 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1147}
1148SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
|