34
35#include "opt_mac.h"
36#include "opt_param.h"
37
38#include <sys/param.h>
39#include <sys/aio.h> /* for aio_swake proto */
40#include <sys/domain.h>
41#include <sys/event.h>
42#include <sys/file.h> /* for maxfiles */
43#include <sys/kernel.h>
44#include <sys/lock.h>
45#include <sys/mac.h>
46#include <sys/malloc.h>
47#include <sys/mbuf.h>
48#include <sys/mutex.h>
49#include <sys/proc.h>
50#include <sys/protosw.h>
51#include <sys/resourcevar.h>
52#include <sys/signalvar.h>
53#include <sys/socket.h>
54#include <sys/socketvar.h>
55#include <sys/stat.h>
56#include <sys/sysctl.h>
57#include <sys/systm.h>
58
59int maxsockets;
60
61void (*aio_swake)(struct socket *, struct sockbuf *);
62
63/*
64 * Primitive routines for operating on sockets and socket buffers
65 */
66
67u_long sb_max = SB_MAX;
68static u_long sb_max_adj =
69 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
70
71static u_long sb_efficiency = 8; /* parameter for sbreserve() */
72
73#ifdef REGRESSION
74static int regression_sonewconn_earlytest = 1;
75SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
76 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
77#endif
78
79/*
80 * Procedures to manipulate state flags of socket
81 * and do appropriate wakeups. Normal sequence from the
82 * active (originating) side is that soisconnecting() is
83 * called during processing of connect() call,
84 * resulting in an eventual call to soisconnected() if/when the
85 * connection is established. When the connection is torn down
86 * soisdisconnecting() is called during processing of disconnect() call,
87 * and soisdisconnected() is called when the connection to the peer
88 * is totally severed. The semantics of these routines are such that
89 * connectionless protocols can call soisconnected() and soisdisconnected()
90 * only, bypassing the in-progress calls when setting up a ``connection''
91 * takes no time.
92 *
93 * From the passive side, a socket is created with
94 * two queues of sockets: so_incomp for connections in progress
95 * and so_comp for connections already made and awaiting user acceptance.
96 * As a protocol is preparing incoming connections, it creates a socket
97 * structure queued on so_incomp by calling sonewconn(). When the connection
98 * is established, soisconnected() is called, and transfers the
99 * socket structure to so_comp, making it available to accept().
100 *
101 * If a socket is closed with sockets on either
102 * so_incomp or so_comp, these sockets are dropped.
103 *
104 * If higher level protocols are implemented in
105 * the kernel, the wakeups done here will sometimes
106 * cause software-interrupt process scheduling.
107 */
108
109void
110soisconnecting(so)
111 register struct socket *so;
112{
113
114 SOCK_LOCK(so);
115 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
116 so->so_state |= SS_ISCONNECTING;
117 SOCK_UNLOCK(so);
118}
119
120void
121soisconnected(so)
122 struct socket *so;
123{
124 struct socket *head;
125
126 ACCEPT_LOCK();
127 SOCK_LOCK(so);
128 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
129 so->so_state |= SS_ISCONNECTED;
130 head = so->so_head;
131 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
132 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
133 SOCK_UNLOCK(so);
134 TAILQ_REMOVE(&head->so_incomp, so, so_list);
135 head->so_incqlen--;
136 so->so_qstate &= ~SQ_INCOMP;
137 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
138 head->so_qlen++;
139 so->so_qstate |= SQ_COMP;
140 ACCEPT_UNLOCK();
141 sorwakeup(head);
142 wakeup_one(&head->so_timeo);
143 } else {
144 ACCEPT_UNLOCK();
145 so->so_upcall =
146 head->so_accf->so_accept_filter->accf_callback;
147 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
148 so->so_rcv.sb_flags |= SB_UPCALL;
149 so->so_options &= ~SO_ACCEPTFILTER;
150 SOCK_UNLOCK(so);
151 so->so_upcall(so, so->so_upcallarg, M_DONTWAIT);
152 }
153 return;
154 }
155 SOCK_UNLOCK(so);
156 ACCEPT_UNLOCK();
157 wakeup(&so->so_timeo);
158 sorwakeup(so);
159 sowwakeup(so);
160}
161
162void
163soisdisconnecting(so)
164 register struct socket *so;
165{
166
167 /*
168 * XXXRW: This code assumes that SOCK_LOCK(so) and
169 * SOCKBUF_LOCK(&so->so_rcv) are the same.
170 */
171 SOCKBUF_LOCK(&so->so_rcv);
172 so->so_state &= ~SS_ISCONNECTING;
173 so->so_state |= SS_ISDISCONNECTING;
174 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
175 sorwakeup_locked(so);
176 SOCKBUF_LOCK(&so->so_snd);
177 so->so_snd.sb_state |= SBS_CANTSENDMORE;
178 sowwakeup_locked(so);
179 wakeup(&so->so_timeo);
180}
181
182void
183soisdisconnected(so)
184 register struct socket *so;
185{
186
187 /*
188 * XXXRW: This code assumes that SOCK_LOCK(so) and
189 * SOCKBUF_LOCK(&so->so_rcv) are the same.
190 */
191 SOCKBUF_LOCK(&so->so_rcv);
192 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
193 so->so_state |= SS_ISDISCONNECTED;
194 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
195 sorwakeup_locked(so);
196 SOCKBUF_LOCK(&so->so_snd);
197 so->so_snd.sb_state |= SBS_CANTSENDMORE;
198 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
199 sowwakeup_locked(so);
200 wakeup(&so->so_timeo);
201}
202
203/*
204 * When an attempt at a new connection is noted on a socket
205 * which accepts connections, sonewconn is called. If the
206 * connection is possible (subject to space constraints, etc.)
207 * then we allocate a new structure, propoerly linked into the
208 * data structure of the original socket, and return this.
209 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
210 *
211 * note: the ref count on the socket is 0 on return
212 */
213struct socket *
214sonewconn(head, connstatus)
215 register struct socket *head;
216 int connstatus;
217{
218 register struct socket *so;
219 int over;
220
221 ACCEPT_LOCK();
222 over = (head->so_qlen > 3 * head->so_qlimit / 2);
223 ACCEPT_UNLOCK();
224#ifdef REGRESSION
225 if (regression_sonewconn_earlytest && over)
226#else
227 if (over)
228#endif
229 return (NULL);
230 so = soalloc(M_NOWAIT);
231 if (so == NULL)
232 return (NULL);
233 if ((head->so_options & SO_ACCEPTFILTER) != 0)
234 connstatus = 0;
235 so->so_head = head;
236 so->so_type = head->so_type;
237 so->so_options = head->so_options &~ SO_ACCEPTCONN;
238 so->so_linger = head->so_linger;
239 so->so_state = head->so_state | SS_NOFDREF;
240 so->so_proto = head->so_proto;
241 so->so_timeo = head->so_timeo;
242 so->so_cred = crhold(head->so_cred);
243#ifdef MAC
244 SOCK_LOCK(head);
245 mac_create_socket_from_socket(head, so);
246 SOCK_UNLOCK(head);
247#endif
248 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
249 NULL, NULL, NULL);
250 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
251 NULL, NULL, NULL);
252 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
253 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
254 sodealloc(so);
255 return (NULL);
256 }
257 so->so_state |= connstatus;
258 ACCEPT_LOCK();
259 if (connstatus) {
260 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
261 so->so_qstate |= SQ_COMP;
262 head->so_qlen++;
263 } else {
264 /*
265 * Keep removing sockets from the head until there's room for
266 * us to insert on the tail. In pre-locking revisions, this
267 * was a simple if(), but as we could be racing with other
268 * threads and soabort() requires dropping locks, we must
269 * loop waiting for the condition to be true.
270 */
271 while (head->so_incqlen > head->so_qlimit) {
272 struct socket *sp;
273 sp = TAILQ_FIRST(&head->so_incomp);
274 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
275 head->so_incqlen--;
276 sp->so_qstate &= ~SQ_INCOMP;
277 sp->so_head = NULL;
278 ACCEPT_UNLOCK();
279 soabort(sp);
280 ACCEPT_LOCK();
281 }
282 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
283 so->so_qstate |= SQ_INCOMP;
284 head->so_incqlen++;
285 }
286 ACCEPT_UNLOCK();
287 if (connstatus) {
288 sorwakeup(head);
289 wakeup_one(&head->so_timeo);
290 }
291 return (so);
292}
293
294/*
295 * Socantsendmore indicates that no more data will be sent on the
296 * socket; it would normally be applied to a socket when the user
297 * informs the system that no more data is to be sent, by the protocol
298 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
299 * will be received, and will normally be applied to the socket by a
300 * protocol when it detects that the peer will send no more data.
301 * Data queued for reading in the socket may yet be read.
302 */
303void
304socantsendmore_locked(so)
305 struct socket *so;
306{
307
308 SOCKBUF_LOCK_ASSERT(&so->so_snd);
309
310 so->so_snd.sb_state |= SBS_CANTSENDMORE;
311 sowwakeup_locked(so);
312 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
313}
314
315void
316socantsendmore(so)
317 struct socket *so;
318{
319
320 SOCKBUF_LOCK(&so->so_snd);
321 socantsendmore_locked(so);
322 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
323}
324
325void
326socantrcvmore_locked(so)
327 struct socket *so;
328{
329
330 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
331
332 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
333 sorwakeup_locked(so);
334 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
335}
336
337void
338socantrcvmore(so)
339 struct socket *so;
340{
341
342 SOCKBUF_LOCK(&so->so_rcv);
343 socantrcvmore_locked(so);
344 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
345}
346
347/*
348 * Wait for data to arrive at/drain from a socket buffer.
349 */
350int
351sbwait(sb)
352 struct sockbuf *sb;
353{
354
355 SOCKBUF_LOCK_ASSERT(sb);
356
357 sb->sb_flags |= SB_WAIT;
358 return (msleep(&sb->sb_cc, &sb->sb_mtx,
359 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
360 sb->sb_timeo));
361}
362
363/*
364 * Lock a sockbuf already known to be locked;
365 * return any error returned from sleep (EINTR).
366 */
367int
368sb_lock(sb)
369 register struct sockbuf *sb;
370{
371 int error;
372
373 SOCKBUF_LOCK_ASSERT(sb);
374
375 while (sb->sb_flags & SB_LOCK) {
376 sb->sb_flags |= SB_WANT;
377 error = msleep(&sb->sb_flags, &sb->sb_mtx,
378 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
379 "sblock", 0);
380 if (error)
381 return (error);
382 }
383 sb->sb_flags |= SB_LOCK;
384 return (0);
385}
386
387/*
388 * Wakeup processes waiting on a socket buffer. Do asynchronous
389 * notification via SIGIO if the socket has the SS_ASYNC flag set.
390 *
391 * Called with the socket buffer lock held; will release the lock by the end
392 * of the function. This allows the caller to acquire the socket buffer lock
393 * while testing for the need for various sorts of wakeup and hold it through
394 * to the point where it's no longer required. We currently hold the lock
395 * through calls out to other subsystems (with the exception of kqueue), and
396 * then release it to avoid lock order issues. It's not clear that's
397 * correct.
398 */
399void
400sowakeup(so, sb)
401 register struct socket *so;
402 register struct sockbuf *sb;
403{
404
405 SOCKBUF_LOCK_ASSERT(sb);
406
407 selwakeuppri(&sb->sb_sel, PSOCK);
408 sb->sb_flags &= ~SB_SEL;
409 if (sb->sb_flags & SB_WAIT) {
410 sb->sb_flags &= ~SB_WAIT;
411 wakeup(&sb->sb_cc);
412 }
413 KNOTE_LOCKED(&sb->sb_sel.si_note, 0);
414 SOCKBUF_UNLOCK(sb);
415 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
416 pgsigio(&so->so_sigio, SIGIO, 0);
417 if (sb->sb_flags & SB_UPCALL)
418 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
419 if (sb->sb_flags & SB_AIO)
420 aio_swake(so, sb);
421 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
422}
423
424/*
425 * Socket buffer (struct sockbuf) utility routines.
426 *
427 * Each socket contains two socket buffers: one for sending data and
428 * one for receiving data. Each buffer contains a queue of mbufs,
429 * information about the number of mbufs and amount of data in the
430 * queue, and other fields allowing select() statements and notification
431 * on data availability to be implemented.
432 *
433 * Data stored in a socket buffer is maintained as a list of records.
434 * Each record is a list of mbufs chained together with the m_next
435 * field. Records are chained together with the m_nextpkt field. The upper
436 * level routine soreceive() expects the following conventions to be
437 * observed when placing information in the receive buffer:
438 *
439 * 1. If the protocol requires each message be preceded by the sender's
440 * name, then a record containing that name must be present before
441 * any associated data (mbuf's must be of type MT_SONAME).
442 * 2. If the protocol supports the exchange of ``access rights'' (really
443 * just additional data associated with the message), and there are
444 * ``rights'' to be received, then a record containing this data
445 * should be present (mbuf's must be of type MT_RIGHTS).
446 * 3. If a name or rights record exists, then it must be followed by
447 * a data record, perhaps of zero length.
448 *
449 * Before using a new socket structure it is first necessary to reserve
450 * buffer space to the socket, by calling sbreserve(). This should commit
451 * some of the available buffer space in the system buffer pool for the
452 * socket (currently, it does nothing but enforce limits). The space
453 * should be released by calling sbrelease() when the socket is destroyed.
454 */
455
456int
457soreserve(so, sndcc, rcvcc)
458 register struct socket *so;
459 u_long sndcc, rcvcc;
460{
461 struct thread *td = curthread;
462
463 SOCKBUF_LOCK(&so->so_snd);
464 SOCKBUF_LOCK(&so->so_rcv);
465 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
466 goto bad;
467 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
468 goto bad2;
469 if (so->so_rcv.sb_lowat == 0)
470 so->so_rcv.sb_lowat = 1;
471 if (so->so_snd.sb_lowat == 0)
472 so->so_snd.sb_lowat = MCLBYTES;
473 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
474 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
475 SOCKBUF_UNLOCK(&so->so_rcv);
476 SOCKBUF_UNLOCK(&so->so_snd);
477 return (0);
478bad2:
479 sbrelease_locked(&so->so_snd, so);
480bad:
481 SOCKBUF_UNLOCK(&so->so_rcv);
482 SOCKBUF_UNLOCK(&so->so_snd);
483 return (ENOBUFS);
484}
485
486static int
487sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
488{
489 int error = 0;
490 u_long old_sb_max = sb_max;
491
492 error = SYSCTL_OUT(req, arg1, sizeof(u_long));
493 if (error || !req->newptr)
494 return (error);
495 error = SYSCTL_IN(req, arg1, sizeof(u_long));
496 if (error)
497 return (error);
498 if (sb_max < MSIZE + MCLBYTES) {
499 sb_max = old_sb_max;
500 return (EINVAL);
501 }
502 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
503 return (0);
504}
505
506/*
507 * Allot mbufs to a sockbuf.
508 * Attempt to scale mbmax so that mbcnt doesn't become limiting
509 * if buffering efficiency is near the normal case.
510 */
511int
512sbreserve_locked(sb, cc, so, td)
513 struct sockbuf *sb;
514 u_long cc;
515 struct socket *so;
516 struct thread *td;
517{
518 rlim_t sbsize_limit;
519
520 SOCKBUF_LOCK_ASSERT(sb);
521
522 /*
523 * td will only be NULL when we're in an interrupt
524 * (e.g. in tcp_input())
525 */
526 if (cc > sb_max_adj)
527 return (0);
528 if (td != NULL) {
529 PROC_LOCK(td->td_proc);
530 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
531 PROC_UNLOCK(td->td_proc);
532 } else
533 sbsize_limit = RLIM_INFINITY;
534 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
535 sbsize_limit))
536 return (0);
537 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
538 if (sb->sb_lowat > sb->sb_hiwat)
539 sb->sb_lowat = sb->sb_hiwat;
540 return (1);
541}
542
543int
544sbreserve(sb, cc, so, td)
545 struct sockbuf *sb;
546 u_long cc;
547 struct socket *so;
548 struct thread *td;
549{
550 int error;
551
552 SOCKBUF_LOCK(sb);
553 error = sbreserve_locked(sb, cc, so, td);
554 SOCKBUF_UNLOCK(sb);
555 return (error);
556}
557
558/*
559 * Free mbufs held by a socket, and reserved mbuf space.
560 */
561void
562sbrelease_locked(sb, so)
563 struct sockbuf *sb;
564 struct socket *so;
565{
566
567 SOCKBUF_LOCK_ASSERT(sb);
568
569 sbflush_locked(sb);
570 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
571 RLIM_INFINITY);
572 sb->sb_mbmax = 0;
573}
574
575void
576sbrelease(sb, so)
577 struct sockbuf *sb;
578 struct socket *so;
579{
580
581 SOCKBUF_LOCK(sb);
582 sbrelease_locked(sb, so);
583 SOCKBUF_UNLOCK(sb);
584}
585/*
586 * Routines to add and remove
587 * data from an mbuf queue.
588 *
589 * The routines sbappend() or sbappendrecord() are normally called to
590 * append new mbufs to a socket buffer, after checking that adequate
591 * space is available, comparing the function sbspace() with the amount
592 * of data to be added. sbappendrecord() differs from sbappend() in
593 * that data supplied is treated as the beginning of a new record.
594 * To place a sender's address, optional access rights, and data in a
595 * socket receive buffer, sbappendaddr() should be used. To place
596 * access rights and data in a socket receive buffer, sbappendrights()
597 * should be used. In either case, the new data begins a new record.
598 * Note that unlike sbappend() and sbappendrecord(), these routines check
599 * for the caller that there will be enough space to store the data.
600 * Each fails if there is not enough space, or if it cannot find mbufs
601 * to store additional information in.
602 *
603 * Reliable protocols may use the socket send buffer to hold data
604 * awaiting acknowledgement. Data is normally copied from a socket
605 * send buffer in a protocol with m_copy for output to a peer,
606 * and then removing the data from the socket buffer with sbdrop()
607 * or sbdroprecord() when the data is acknowledged by the peer.
608 */
609
610#ifdef SOCKBUF_DEBUG
611void
612sblastrecordchk(struct sockbuf *sb, const char *file, int line)
613{
614 struct mbuf *m = sb->sb_mb;
615
616 SOCKBUF_LOCK_ASSERT(sb);
617
618 while (m && m->m_nextpkt)
619 m = m->m_nextpkt;
620
621 if (m != sb->sb_lastrecord) {
622 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
623 __func__, sb->sb_mb, sb->sb_lastrecord, m);
624 printf("packet chain:\n");
625 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
626 printf("\t%p\n", m);
627 panic("%s from %s:%u", __func__, file, line);
628 }
629}
630
631void
632sblastmbufchk(struct sockbuf *sb, const char *file, int line)
633{
634 struct mbuf *m = sb->sb_mb;
635 struct mbuf *n;
636
637 SOCKBUF_LOCK_ASSERT(sb);
638
639 while (m && m->m_nextpkt)
640 m = m->m_nextpkt;
641
642 while (m && m->m_next)
643 m = m->m_next;
644
645 if (m != sb->sb_mbtail) {
646 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
647 __func__, sb->sb_mb, sb->sb_mbtail, m);
648 printf("packet tree:\n");
649 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
650 printf("\t");
651 for (n = m; n != NULL; n = n->m_next)
652 printf("%p ", n);
653 printf("\n");
654 }
655 panic("%s from %s:%u", __func__, file, line);
656 }
657}
658#endif /* SOCKBUF_DEBUG */
659
660#define SBLINKRECORD(sb, m0) do { \
661 SOCKBUF_LOCK_ASSERT(sb); \
662 if ((sb)->sb_lastrecord != NULL) \
663 (sb)->sb_lastrecord->m_nextpkt = (m0); \
664 else \
665 (sb)->sb_mb = (m0); \
666 (sb)->sb_lastrecord = (m0); \
667} while (/*CONSTCOND*/0)
668
669/*
670 * Append mbuf chain m to the last record in the
671 * socket buffer sb. The additional space associated
672 * the mbuf chain is recorded in sb. Empty mbufs are
673 * discarded and mbufs are compacted where possible.
674 */
675void
676sbappend_locked(sb, m)
677 struct sockbuf *sb;
678 struct mbuf *m;
679{
680 register struct mbuf *n;
681
682 SOCKBUF_LOCK_ASSERT(sb);
683
684 if (m == 0)
685 return;
686
687 SBLASTRECORDCHK(sb);
688 n = sb->sb_mb;
689 if (n) {
690 while (n->m_nextpkt)
691 n = n->m_nextpkt;
692 do {
693 if (n->m_flags & M_EOR) {
694 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
695 return;
696 }
697 } while (n->m_next && (n = n->m_next));
698 } else {
699 /*
700 * XXX Would like to simply use sb_mbtail here, but
701 * XXX I need to verify that I won't miss an EOR that
702 * XXX way.
703 */
704 if ((n = sb->sb_lastrecord) != NULL) {
705 do {
706 if (n->m_flags & M_EOR) {
707 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
708 return;
709 }
710 } while (n->m_next && (n = n->m_next));
711 } else {
712 /*
713 * If this is the first record in the socket buffer,
714 * it's also the last record.
715 */
716 sb->sb_lastrecord = m;
717 }
718 }
719 sbcompress(sb, m, n);
720 SBLASTRECORDCHK(sb);
721}
722
723/*
724 * Append mbuf chain m to the last record in the
725 * socket buffer sb. The additional space associated
726 * the mbuf chain is recorded in sb. Empty mbufs are
727 * discarded and mbufs are compacted where possible.
728 */
729void
730sbappend(sb, m)
731 struct sockbuf *sb;
732 struct mbuf *m;
733{
734
735 SOCKBUF_LOCK(sb);
736 sbappend_locked(sb, m);
737 SOCKBUF_UNLOCK(sb);
738}
739
740/*
741 * This version of sbappend() should only be used when the caller
742 * absolutely knows that there will never be more than one record
743 * in the socket buffer, that is, a stream protocol (such as TCP).
744 */
745void
746sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
747{
748 SOCKBUF_LOCK_ASSERT(sb);
749
750 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
751 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
752
753 SBLASTMBUFCHK(sb);
754
755 sbcompress(sb, m, sb->sb_mbtail);
756
757 sb->sb_lastrecord = sb->sb_mb;
758 SBLASTRECORDCHK(sb);
759}
760
761/*
762 * This version of sbappend() should only be used when the caller
763 * absolutely knows that there will never be more than one record
764 * in the socket buffer, that is, a stream protocol (such as TCP).
765 */
766void
767sbappendstream(struct sockbuf *sb, struct mbuf *m)
768{
769
770 SOCKBUF_LOCK(sb);
771 sbappendstream_locked(sb, m);
772 SOCKBUF_UNLOCK(sb);
773}
774
775#ifdef SOCKBUF_DEBUG
776void
777sbcheck(sb)
778 struct sockbuf *sb;
779{
780 struct mbuf *m;
781 struct mbuf *n = 0;
782 u_long len = 0, mbcnt = 0;
783
784 SOCKBUF_LOCK_ASSERT(sb);
785
786 for (m = sb->sb_mb; m; m = n) {
787 n = m->m_nextpkt;
788 for (; m; m = m->m_next) {
789 len += m->m_len;
790 mbcnt += MSIZE;
791 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
792 mbcnt += m->m_ext.ext_size;
793 }
794 }
795 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
796 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
797 mbcnt, sb->sb_mbcnt);
798 panic("sbcheck");
799 }
800}
801#endif
802
803/*
804 * As above, except the mbuf chain
805 * begins a new record.
806 */
807void
808sbappendrecord_locked(sb, m0)
809 register struct sockbuf *sb;
810 register struct mbuf *m0;
811{
812 register struct mbuf *m;
813
814 SOCKBUF_LOCK_ASSERT(sb);
815
816 if (m0 == 0)
817 return;
818 m = sb->sb_mb;
819 if (m)
820 while (m->m_nextpkt)
821 m = m->m_nextpkt;
822 /*
823 * Put the first mbuf on the queue.
824 * Note this permits zero length records.
825 */
826 sballoc(sb, m0);
827 SBLASTRECORDCHK(sb);
828 SBLINKRECORD(sb, m0);
829 if (m)
830 m->m_nextpkt = m0;
831 else
832 sb->sb_mb = m0;
833 m = m0->m_next;
834 m0->m_next = 0;
835 if (m && (m0->m_flags & M_EOR)) {
836 m0->m_flags &= ~M_EOR;
837 m->m_flags |= M_EOR;
838 }
839 sbcompress(sb, m, m0);
840}
841
842/*
843 * As above, except the mbuf chain
844 * begins a new record.
845 */
846void
847sbappendrecord(sb, m0)
848 register struct sockbuf *sb;
849 register struct mbuf *m0;
850{
851
852 SOCKBUF_LOCK(sb);
853 sbappendrecord_locked(sb, m0);
854 SOCKBUF_UNLOCK(sb);
855}
856
857/*
858 * As above except that OOB data
859 * is inserted at the beginning of the sockbuf,
860 * but after any other OOB data.
861 */
862void
863sbinsertoob_locked(sb, m0)
864 register struct sockbuf *sb;
865 register struct mbuf *m0;
866{
867 register struct mbuf *m;
868 register struct mbuf **mp;
869
870 SOCKBUF_LOCK_ASSERT(sb);
871
872 if (m0 == 0)
873 return;
874 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
875 m = *mp;
876 again:
877 switch (m->m_type) {
878
879 case MT_OOBDATA:
880 continue; /* WANT next train */
881
882 case MT_CONTROL:
883 m = m->m_next;
884 if (m)
885 goto again; /* inspect THIS train further */
886 }
887 break;
888 }
889 /*
890 * Put the first mbuf on the queue.
891 * Note this permits zero length records.
892 */
893 sballoc(sb, m0);
894 m0->m_nextpkt = *mp;
895 *mp = m0;
896 m = m0->m_next;
897 m0->m_next = 0;
898 if (m && (m0->m_flags & M_EOR)) {
899 m0->m_flags &= ~M_EOR;
900 m->m_flags |= M_EOR;
901 }
902 sbcompress(sb, m, m0);
903}
904
905/*
906 * As above except that OOB data
907 * is inserted at the beginning of the sockbuf,
908 * but after any other OOB data.
909 */
910void
911sbinsertoob(sb, m0)
912 register struct sockbuf *sb;
913 register struct mbuf *m0;
914{
915
916 SOCKBUF_LOCK(sb);
917 sbinsertoob_locked(sb, m0);
918 SOCKBUF_UNLOCK(sb);
919}
920
921/*
922 * Append address and data, and optionally, control (ancillary) data
923 * to the receive queue of a socket. If present,
924 * m0 must include a packet header with total length.
925 * Returns 0 if no space in sockbuf or insufficient mbufs.
926 */
927int
928sbappendaddr_locked(sb, asa, m0, control)
929 struct sockbuf *sb;
930 const struct sockaddr *asa;
931 struct mbuf *m0, *control;
932{
933 struct mbuf *m, *n, *nlast;
934 int space = asa->sa_len;
935
936 SOCKBUF_LOCK_ASSERT(sb);
937
938 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
939 panic("sbappendaddr_locked");
940 if (m0)
941 space += m0->m_pkthdr.len;
942 space += m_length(control, &n);
943
944 if (space > sbspace(sb))
945 return (0);
946#if MSIZE <= 256
947 if (asa->sa_len > MLEN)
948 return (0);
949#endif
950 MGET(m, M_DONTWAIT, MT_SONAME);
951 if (m == 0)
952 return (0);
953 m->m_len = asa->sa_len;
954 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
955 if (n)
956 n->m_next = m0; /* concatenate data to control */
957 else
958 control = m0;
959 m->m_next = control;
960 for (n = m; n->m_next != NULL; n = n->m_next)
961 sballoc(sb, n);
962 sballoc(sb, n);
963 nlast = n;
964 SBLINKRECORD(sb, m);
965
966 sb->sb_mbtail = nlast;
967 SBLASTMBUFCHK(sb);
968
969 SBLASTRECORDCHK(sb);
970 return (1);
971}
972
973/*
974 * Append address and data, and optionally, control (ancillary) data
975 * to the receive queue of a socket. If present,
976 * m0 must include a packet header with total length.
977 * Returns 0 if no space in sockbuf or insufficient mbufs.
978 */
979int
980sbappendaddr(sb, asa, m0, control)
981 struct sockbuf *sb;
982 const struct sockaddr *asa;
983 struct mbuf *m0, *control;
984{
985 int retval;
986
987 SOCKBUF_LOCK(sb);
988 retval = sbappendaddr_locked(sb, asa, m0, control);
989 SOCKBUF_UNLOCK(sb);
990 return (retval);
991}
992
993int
994sbappendcontrol_locked(sb, m0, control)
995 struct sockbuf *sb;
996 struct mbuf *control, *m0;
997{
998 struct mbuf *m, *n, *mlast;
999 int space;
1000
1001 SOCKBUF_LOCK_ASSERT(sb);
1002
1003 if (control == 0)
1004 panic("sbappendcontrol_locked");
1005 space = m_length(control, &n) + m_length(m0, NULL);
1006
1007 if (space > sbspace(sb))
1008 return (0);
1009 n->m_next = m0; /* concatenate data to control */
1010
1011 SBLASTRECORDCHK(sb);
1012
1013 for (m = control; m->m_next; m = m->m_next)
1014 sballoc(sb, m);
1015 sballoc(sb, m);
1016 mlast = m;
1017 SBLINKRECORD(sb, control);
1018
1019 sb->sb_mbtail = mlast;
1020 SBLASTMBUFCHK(sb);
1021
1022 SBLASTRECORDCHK(sb);
1023 return (1);
1024}
1025
1026int
1027sbappendcontrol(sb, m0, control)
1028 struct sockbuf *sb;
1029 struct mbuf *control, *m0;
1030{
1031 int retval;
1032
1033 SOCKBUF_LOCK(sb);
1034 retval = sbappendcontrol_locked(sb, m0, control);
1035 SOCKBUF_UNLOCK(sb);
1036 return (retval);
1037}
1038
1039/*
1040 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
1041 * (n). If (n) is NULL, the buffer is presumed empty.
1042 *
1043 * When the data is compressed, mbufs in the chain may be handled in one of
1044 * three ways:
1045 *
1046 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
1047 * record boundary, and no change in data type).
1048 *
1049 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
1050 * an mbuf already in the socket buffer. This can occur if an
1051 * appropriate mbuf exists, there is room, and no merging of data types
1052 * will occur.
1053 *
1054 * (3) The mbuf may be appended to the end of the existing mbuf chain.
1055 *
1056 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
1057 * end-of-record.
1058 */
1059void
1060sbcompress(sb, m, n)
1061 register struct sockbuf *sb;
1062 register struct mbuf *m, *n;
1063{
1064 register int eor = 0;
1065 register struct mbuf *o;
1066
1067 SOCKBUF_LOCK_ASSERT(sb);
1068
1069 while (m) {
1070 eor |= m->m_flags & M_EOR;
1071 if (m->m_len == 0 &&
1072 (eor == 0 ||
1073 (((o = m->m_next) || (o = n)) &&
1074 o->m_type == m->m_type))) {
1075 if (sb->sb_lastrecord == m)
1076 sb->sb_lastrecord = m->m_next;
1077 m = m_free(m);
1078 continue;
1079 }
1080 if (n && (n->m_flags & M_EOR) == 0 &&
1081 M_WRITABLE(n) &&
1082 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1083 m->m_len <= M_TRAILINGSPACE(n) &&
1084 n->m_type == m->m_type) {
1085 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
1086 (unsigned)m->m_len);
1087 n->m_len += m->m_len;
1088 sb->sb_cc += m->m_len;
1089 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1090 /* XXX: Probably don't need.*/
1091 sb->sb_ctl += m->m_len;
1092 m = m_free(m);
1093 continue;
1094 }
1095 if (n)
1096 n->m_next = m;
1097 else
1098 sb->sb_mb = m;
1099 sb->sb_mbtail = m;
1100 sballoc(sb, m);
1101 n = m;
1102 m->m_flags &= ~M_EOR;
1103 m = m->m_next;
1104 n->m_next = 0;
1105 }
1106 if (eor) {
1107 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
1108 n->m_flags |= eor;
1109 }
1110 SBLASTMBUFCHK(sb);
1111}
1112
1113/*
1114 * Free all mbufs in a sockbuf.
1115 * Check that all resources are reclaimed.
1116 */
1117void
1118sbflush_locked(sb)
1119 register struct sockbuf *sb;
1120{
1121
1122 SOCKBUF_LOCK_ASSERT(sb);
1123
1124 if (sb->sb_flags & SB_LOCK)
1125 panic("sbflush_locked: locked");
1126 while (sb->sb_mbcnt) {
1127 /*
1128 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
1129 * we would loop forever. Panic instead.
1130 */
1131 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
1132 break;
1133 sbdrop_locked(sb, (int)sb->sb_cc);
1134 }
1135 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
1136 panic("sbflush_locked: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
1137}
1138
1139void
1140sbflush(sb)
1141 register struct sockbuf *sb;
1142{
1143
1144 SOCKBUF_LOCK(sb);
1145 sbflush_locked(sb);
1146 SOCKBUF_UNLOCK(sb);
1147}
1148
1149/*
1150 * Drop data from (the front of) a sockbuf.
1151 */
1152void
1153sbdrop_locked(sb, len)
1154 register struct sockbuf *sb;
1155 register int len;
1156{
1157 register struct mbuf *m;
1158 struct mbuf *next;
1159
1160 SOCKBUF_LOCK_ASSERT(sb);
1161
1162 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1163 while (len > 0) {
1164 if (m == 0) {
1165 if (next == 0)
1166 panic("sbdrop");
1167 m = next;
1168 next = m->m_nextpkt;
1169 continue;
1170 }
1171 if (m->m_len > len) {
1172 m->m_len -= len;
1173 m->m_data += len;
1174 sb->sb_cc -= len;
1175 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1176 sb->sb_ctl -= len;
1177 break;
1178 }
1179 len -= m->m_len;
1180 sbfree(sb, m);
1181 m = m_free(m);
1182 }
1183 while (m && m->m_len == 0) {
1184 sbfree(sb, m);
1185 m = m_free(m);
1186 }
1187 if (m) {
1188 sb->sb_mb = m;
1189 m->m_nextpkt = next;
1190 } else
1191 sb->sb_mb = next;
1192 /*
1193 * First part is an inline SB_EMPTY_FIXUP(). Second part
1194 * makes sure sb_lastrecord is up-to-date if we dropped
1195 * part of the last record.
1196 */
1197 m = sb->sb_mb;
1198 if (m == NULL) {
1199 sb->sb_mbtail = NULL;
1200 sb->sb_lastrecord = NULL;
1201 } else if (m->m_nextpkt == NULL) {
1202 sb->sb_lastrecord = m;
1203 }
1204}
1205
1206/*
1207 * Drop data from (the front of) a sockbuf.
1208 */
1209void
1210sbdrop(sb, len)
1211 register struct sockbuf *sb;
1212 register int len;
1213{
1214
1215 SOCKBUF_LOCK(sb);
1216 sbdrop_locked(sb, len);
1217 SOCKBUF_UNLOCK(sb);
1218}
1219
1220/*
1221 * Drop a record off the front of a sockbuf
1222 * and move the next record to the front.
1223 */
1224void
1225sbdroprecord_locked(sb)
1226 register struct sockbuf *sb;
1227{
1228 register struct mbuf *m;
1229
1230 SOCKBUF_LOCK_ASSERT(sb);
1231
1232 m = sb->sb_mb;
1233 if (m) {
1234 sb->sb_mb = m->m_nextpkt;
1235 do {
1236 sbfree(sb, m);
1237 m = m_free(m);
1238 } while (m);
1239 }
1240 SB_EMPTY_FIXUP(sb);
1241}
1242
1243/*
1244 * Drop a record off the front of a sockbuf
1245 * and move the next record to the front.
1246 */
1247void
1248sbdroprecord(sb)
1249 register struct sockbuf *sb;
1250{
1251
1252 SOCKBUF_LOCK(sb);
1253 sbdroprecord_locked(sb);
1254 SOCKBUF_UNLOCK(sb);
1255}
1256
1257/*
1258 * Create a "control" mbuf containing the specified data
1259 * with the specified type for presentation on a socket buffer.
1260 */
1261struct mbuf *
1262sbcreatecontrol(p, size, type, level)
1263 caddr_t p;
1264 register int size;
1265 int type, level;
1266{
1267 register struct cmsghdr *cp;
1268 struct mbuf *m;
1269
1270 if (CMSG_SPACE((u_int)size) > MCLBYTES)
1271 return ((struct mbuf *) NULL);
1272 if (CMSG_SPACE((u_int)size) > MLEN)
1273 m = m_getcl(M_DONTWAIT, MT_CONTROL, 0);
1274 else
1275 m = m_get(M_DONTWAIT, MT_CONTROL);
1276 if (m == NULL)
1277 return ((struct mbuf *) NULL);
1278 cp = mtod(m, struct cmsghdr *);
1279 m->m_len = 0;
1280 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1281 ("sbcreatecontrol: short mbuf"));
1282 if (p != NULL)
1283 (void)memcpy(CMSG_DATA(cp), p, size);
1284 m->m_len = CMSG_SPACE(size);
1285 cp->cmsg_len = CMSG_LEN(size);
1286 cp->cmsg_level = level;
1287 cp->cmsg_type = type;
1288 return (m);
1289}
1290
1291/*
1292 * Some routines that return EOPNOTSUPP for entry points that are not
1293 * supported by a protocol. Fill in as needed.
1294 */
| 34
35#include "opt_mac.h"
36#include "opt_param.h"
37
38#include <sys/param.h>
39#include <sys/aio.h> /* for aio_swake proto */
40#include <sys/domain.h>
41#include <sys/event.h>
42#include <sys/file.h> /* for maxfiles */
43#include <sys/kernel.h>
44#include <sys/lock.h>
45#include <sys/mac.h>
46#include <sys/malloc.h>
47#include <sys/mbuf.h>
48#include <sys/mutex.h>
49#include <sys/proc.h>
50#include <sys/protosw.h>
51#include <sys/resourcevar.h>
52#include <sys/signalvar.h>
53#include <sys/socket.h>
54#include <sys/socketvar.h>
55#include <sys/stat.h>
56#include <sys/sysctl.h>
57#include <sys/systm.h>
58
59int maxsockets;
60
61void (*aio_swake)(struct socket *, struct sockbuf *);
62
63/*
64 * Primitive routines for operating on sockets and socket buffers
65 */
66
67u_long sb_max = SB_MAX;
68static u_long sb_max_adj =
69 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
70
71static u_long sb_efficiency = 8; /* parameter for sbreserve() */
72
73#ifdef REGRESSION
74static int regression_sonewconn_earlytest = 1;
75SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
76 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
77#endif
78
79/*
80 * Procedures to manipulate state flags of socket
81 * and do appropriate wakeups. Normal sequence from the
82 * active (originating) side is that soisconnecting() is
83 * called during processing of connect() call,
84 * resulting in an eventual call to soisconnected() if/when the
85 * connection is established. When the connection is torn down
86 * soisdisconnecting() is called during processing of disconnect() call,
87 * and soisdisconnected() is called when the connection to the peer
88 * is totally severed. The semantics of these routines are such that
89 * connectionless protocols can call soisconnected() and soisdisconnected()
90 * only, bypassing the in-progress calls when setting up a ``connection''
91 * takes no time.
92 *
93 * From the passive side, a socket is created with
94 * two queues of sockets: so_incomp for connections in progress
95 * and so_comp for connections already made and awaiting user acceptance.
96 * As a protocol is preparing incoming connections, it creates a socket
97 * structure queued on so_incomp by calling sonewconn(). When the connection
98 * is established, soisconnected() is called, and transfers the
99 * socket structure to so_comp, making it available to accept().
100 *
101 * If a socket is closed with sockets on either
102 * so_incomp or so_comp, these sockets are dropped.
103 *
104 * If higher level protocols are implemented in
105 * the kernel, the wakeups done here will sometimes
106 * cause software-interrupt process scheduling.
107 */
108
109void
110soisconnecting(so)
111 register struct socket *so;
112{
113
114 SOCK_LOCK(so);
115 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
116 so->so_state |= SS_ISCONNECTING;
117 SOCK_UNLOCK(so);
118}
119
120void
121soisconnected(so)
122 struct socket *so;
123{
124 struct socket *head;
125
126 ACCEPT_LOCK();
127 SOCK_LOCK(so);
128 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
129 so->so_state |= SS_ISCONNECTED;
130 head = so->so_head;
131 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
132 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
133 SOCK_UNLOCK(so);
134 TAILQ_REMOVE(&head->so_incomp, so, so_list);
135 head->so_incqlen--;
136 so->so_qstate &= ~SQ_INCOMP;
137 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
138 head->so_qlen++;
139 so->so_qstate |= SQ_COMP;
140 ACCEPT_UNLOCK();
141 sorwakeup(head);
142 wakeup_one(&head->so_timeo);
143 } else {
144 ACCEPT_UNLOCK();
145 so->so_upcall =
146 head->so_accf->so_accept_filter->accf_callback;
147 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
148 so->so_rcv.sb_flags |= SB_UPCALL;
149 so->so_options &= ~SO_ACCEPTFILTER;
150 SOCK_UNLOCK(so);
151 so->so_upcall(so, so->so_upcallarg, M_DONTWAIT);
152 }
153 return;
154 }
155 SOCK_UNLOCK(so);
156 ACCEPT_UNLOCK();
157 wakeup(&so->so_timeo);
158 sorwakeup(so);
159 sowwakeup(so);
160}
161
162void
163soisdisconnecting(so)
164 register struct socket *so;
165{
166
167 /*
168 * XXXRW: This code assumes that SOCK_LOCK(so) and
169 * SOCKBUF_LOCK(&so->so_rcv) are the same.
170 */
171 SOCKBUF_LOCK(&so->so_rcv);
172 so->so_state &= ~SS_ISCONNECTING;
173 so->so_state |= SS_ISDISCONNECTING;
174 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
175 sorwakeup_locked(so);
176 SOCKBUF_LOCK(&so->so_snd);
177 so->so_snd.sb_state |= SBS_CANTSENDMORE;
178 sowwakeup_locked(so);
179 wakeup(&so->so_timeo);
180}
181
182void
183soisdisconnected(so)
184 register struct socket *so;
185{
186
187 /*
188 * XXXRW: This code assumes that SOCK_LOCK(so) and
189 * SOCKBUF_LOCK(&so->so_rcv) are the same.
190 */
191 SOCKBUF_LOCK(&so->so_rcv);
192 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
193 so->so_state |= SS_ISDISCONNECTED;
194 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
195 sorwakeup_locked(so);
196 SOCKBUF_LOCK(&so->so_snd);
197 so->so_snd.sb_state |= SBS_CANTSENDMORE;
198 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
199 sowwakeup_locked(so);
200 wakeup(&so->so_timeo);
201}
202
203/*
204 * When an attempt at a new connection is noted on a socket
205 * which accepts connections, sonewconn is called. If the
206 * connection is possible (subject to space constraints, etc.)
207 * then we allocate a new structure, propoerly linked into the
208 * data structure of the original socket, and return this.
209 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
210 *
211 * note: the ref count on the socket is 0 on return
212 */
213struct socket *
214sonewconn(head, connstatus)
215 register struct socket *head;
216 int connstatus;
217{
218 register struct socket *so;
219 int over;
220
221 ACCEPT_LOCK();
222 over = (head->so_qlen > 3 * head->so_qlimit / 2);
223 ACCEPT_UNLOCK();
224#ifdef REGRESSION
225 if (regression_sonewconn_earlytest && over)
226#else
227 if (over)
228#endif
229 return (NULL);
230 so = soalloc(M_NOWAIT);
231 if (so == NULL)
232 return (NULL);
233 if ((head->so_options & SO_ACCEPTFILTER) != 0)
234 connstatus = 0;
235 so->so_head = head;
236 so->so_type = head->so_type;
237 so->so_options = head->so_options &~ SO_ACCEPTCONN;
238 so->so_linger = head->so_linger;
239 so->so_state = head->so_state | SS_NOFDREF;
240 so->so_proto = head->so_proto;
241 so->so_timeo = head->so_timeo;
242 so->so_cred = crhold(head->so_cred);
243#ifdef MAC
244 SOCK_LOCK(head);
245 mac_create_socket_from_socket(head, so);
246 SOCK_UNLOCK(head);
247#endif
248 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
249 NULL, NULL, NULL);
250 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
251 NULL, NULL, NULL);
252 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
253 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
254 sodealloc(so);
255 return (NULL);
256 }
257 so->so_state |= connstatus;
258 ACCEPT_LOCK();
259 if (connstatus) {
260 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
261 so->so_qstate |= SQ_COMP;
262 head->so_qlen++;
263 } else {
264 /*
265 * Keep removing sockets from the head until there's room for
266 * us to insert on the tail. In pre-locking revisions, this
267 * was a simple if(), but as we could be racing with other
268 * threads and soabort() requires dropping locks, we must
269 * loop waiting for the condition to be true.
270 */
271 while (head->so_incqlen > head->so_qlimit) {
272 struct socket *sp;
273 sp = TAILQ_FIRST(&head->so_incomp);
274 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
275 head->so_incqlen--;
276 sp->so_qstate &= ~SQ_INCOMP;
277 sp->so_head = NULL;
278 ACCEPT_UNLOCK();
279 soabort(sp);
280 ACCEPT_LOCK();
281 }
282 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
283 so->so_qstate |= SQ_INCOMP;
284 head->so_incqlen++;
285 }
286 ACCEPT_UNLOCK();
287 if (connstatus) {
288 sorwakeup(head);
289 wakeup_one(&head->so_timeo);
290 }
291 return (so);
292}
293
294/*
295 * Socantsendmore indicates that no more data will be sent on the
296 * socket; it would normally be applied to a socket when the user
297 * informs the system that no more data is to be sent, by the protocol
298 * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
299 * will be received, and will normally be applied to the socket by a
300 * protocol when it detects that the peer will send no more data.
301 * Data queued for reading in the socket may yet be read.
302 */
303void
304socantsendmore_locked(so)
305 struct socket *so;
306{
307
308 SOCKBUF_LOCK_ASSERT(&so->so_snd);
309
310 so->so_snd.sb_state |= SBS_CANTSENDMORE;
311 sowwakeup_locked(so);
312 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
313}
314
315void
316socantsendmore(so)
317 struct socket *so;
318{
319
320 SOCKBUF_LOCK(&so->so_snd);
321 socantsendmore_locked(so);
322 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
323}
324
325void
326socantrcvmore_locked(so)
327 struct socket *so;
328{
329
330 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
331
332 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
333 sorwakeup_locked(so);
334 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
335}
336
337void
338socantrcvmore(so)
339 struct socket *so;
340{
341
342 SOCKBUF_LOCK(&so->so_rcv);
343 socantrcvmore_locked(so);
344 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
345}
346
347/*
348 * Wait for data to arrive at/drain from a socket buffer.
349 */
350int
351sbwait(sb)
352 struct sockbuf *sb;
353{
354
355 SOCKBUF_LOCK_ASSERT(sb);
356
357 sb->sb_flags |= SB_WAIT;
358 return (msleep(&sb->sb_cc, &sb->sb_mtx,
359 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
360 sb->sb_timeo));
361}
362
363/*
364 * Lock a sockbuf already known to be locked;
365 * return any error returned from sleep (EINTR).
366 */
367int
368sb_lock(sb)
369 register struct sockbuf *sb;
370{
371 int error;
372
373 SOCKBUF_LOCK_ASSERT(sb);
374
375 while (sb->sb_flags & SB_LOCK) {
376 sb->sb_flags |= SB_WANT;
377 error = msleep(&sb->sb_flags, &sb->sb_mtx,
378 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
379 "sblock", 0);
380 if (error)
381 return (error);
382 }
383 sb->sb_flags |= SB_LOCK;
384 return (0);
385}
386
387/*
388 * Wakeup processes waiting on a socket buffer. Do asynchronous
389 * notification via SIGIO if the socket has the SS_ASYNC flag set.
390 *
391 * Called with the socket buffer lock held; will release the lock by the end
392 * of the function. This allows the caller to acquire the socket buffer lock
393 * while testing for the need for various sorts of wakeup and hold it through
394 * to the point where it's no longer required. We currently hold the lock
395 * through calls out to other subsystems (with the exception of kqueue), and
396 * then release it to avoid lock order issues. It's not clear that's
397 * correct.
398 */
399void
400sowakeup(so, sb)
401 register struct socket *so;
402 register struct sockbuf *sb;
403{
404
405 SOCKBUF_LOCK_ASSERT(sb);
406
407 selwakeuppri(&sb->sb_sel, PSOCK);
408 sb->sb_flags &= ~SB_SEL;
409 if (sb->sb_flags & SB_WAIT) {
410 sb->sb_flags &= ~SB_WAIT;
411 wakeup(&sb->sb_cc);
412 }
413 KNOTE_LOCKED(&sb->sb_sel.si_note, 0);
414 SOCKBUF_UNLOCK(sb);
415 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
416 pgsigio(&so->so_sigio, SIGIO, 0);
417 if (sb->sb_flags & SB_UPCALL)
418 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
419 if (sb->sb_flags & SB_AIO)
420 aio_swake(so, sb);
421 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
422}
423
424/*
425 * Socket buffer (struct sockbuf) utility routines.
426 *
427 * Each socket contains two socket buffers: one for sending data and
428 * one for receiving data. Each buffer contains a queue of mbufs,
429 * information about the number of mbufs and amount of data in the
430 * queue, and other fields allowing select() statements and notification
431 * on data availability to be implemented.
432 *
433 * Data stored in a socket buffer is maintained as a list of records.
434 * Each record is a list of mbufs chained together with the m_next
435 * field. Records are chained together with the m_nextpkt field. The upper
436 * level routine soreceive() expects the following conventions to be
437 * observed when placing information in the receive buffer:
438 *
439 * 1. If the protocol requires each message be preceded by the sender's
440 * name, then a record containing that name must be present before
441 * any associated data (mbuf's must be of type MT_SONAME).
442 * 2. If the protocol supports the exchange of ``access rights'' (really
443 * just additional data associated with the message), and there are
444 * ``rights'' to be received, then a record containing this data
445 * should be present (mbuf's must be of type MT_RIGHTS).
446 * 3. If a name or rights record exists, then it must be followed by
447 * a data record, perhaps of zero length.
448 *
449 * Before using a new socket structure it is first necessary to reserve
450 * buffer space to the socket, by calling sbreserve(). This should commit
451 * some of the available buffer space in the system buffer pool for the
452 * socket (currently, it does nothing but enforce limits). The space
453 * should be released by calling sbrelease() when the socket is destroyed.
454 */
455
456int
457soreserve(so, sndcc, rcvcc)
458 register struct socket *so;
459 u_long sndcc, rcvcc;
460{
461 struct thread *td = curthread;
462
463 SOCKBUF_LOCK(&so->so_snd);
464 SOCKBUF_LOCK(&so->so_rcv);
465 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
466 goto bad;
467 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
468 goto bad2;
469 if (so->so_rcv.sb_lowat == 0)
470 so->so_rcv.sb_lowat = 1;
471 if (so->so_snd.sb_lowat == 0)
472 so->so_snd.sb_lowat = MCLBYTES;
473 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
474 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
475 SOCKBUF_UNLOCK(&so->so_rcv);
476 SOCKBUF_UNLOCK(&so->so_snd);
477 return (0);
478bad2:
479 sbrelease_locked(&so->so_snd, so);
480bad:
481 SOCKBUF_UNLOCK(&so->so_rcv);
482 SOCKBUF_UNLOCK(&so->so_snd);
483 return (ENOBUFS);
484}
485
486static int
487sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
488{
489 int error = 0;
490 u_long old_sb_max = sb_max;
491
492 error = SYSCTL_OUT(req, arg1, sizeof(u_long));
493 if (error || !req->newptr)
494 return (error);
495 error = SYSCTL_IN(req, arg1, sizeof(u_long));
496 if (error)
497 return (error);
498 if (sb_max < MSIZE + MCLBYTES) {
499 sb_max = old_sb_max;
500 return (EINVAL);
501 }
502 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
503 return (0);
504}
505
506/*
507 * Allot mbufs to a sockbuf.
508 * Attempt to scale mbmax so that mbcnt doesn't become limiting
509 * if buffering efficiency is near the normal case.
510 */
511int
512sbreserve_locked(sb, cc, so, td)
513 struct sockbuf *sb;
514 u_long cc;
515 struct socket *so;
516 struct thread *td;
517{
518 rlim_t sbsize_limit;
519
520 SOCKBUF_LOCK_ASSERT(sb);
521
522 /*
523 * td will only be NULL when we're in an interrupt
524 * (e.g. in tcp_input())
525 */
526 if (cc > sb_max_adj)
527 return (0);
528 if (td != NULL) {
529 PROC_LOCK(td->td_proc);
530 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
531 PROC_UNLOCK(td->td_proc);
532 } else
533 sbsize_limit = RLIM_INFINITY;
534 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
535 sbsize_limit))
536 return (0);
537 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
538 if (sb->sb_lowat > sb->sb_hiwat)
539 sb->sb_lowat = sb->sb_hiwat;
540 return (1);
541}
542
543int
544sbreserve(sb, cc, so, td)
545 struct sockbuf *sb;
546 u_long cc;
547 struct socket *so;
548 struct thread *td;
549{
550 int error;
551
552 SOCKBUF_LOCK(sb);
553 error = sbreserve_locked(sb, cc, so, td);
554 SOCKBUF_UNLOCK(sb);
555 return (error);
556}
557
558/*
559 * Free mbufs held by a socket, and reserved mbuf space.
560 */
561void
562sbrelease_locked(sb, so)
563 struct sockbuf *sb;
564 struct socket *so;
565{
566
567 SOCKBUF_LOCK_ASSERT(sb);
568
569 sbflush_locked(sb);
570 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
571 RLIM_INFINITY);
572 sb->sb_mbmax = 0;
573}
574
575void
576sbrelease(sb, so)
577 struct sockbuf *sb;
578 struct socket *so;
579{
580
581 SOCKBUF_LOCK(sb);
582 sbrelease_locked(sb, so);
583 SOCKBUF_UNLOCK(sb);
584}
585/*
586 * Routines to add and remove
587 * data from an mbuf queue.
588 *
589 * The routines sbappend() or sbappendrecord() are normally called to
590 * append new mbufs to a socket buffer, after checking that adequate
591 * space is available, comparing the function sbspace() with the amount
592 * of data to be added. sbappendrecord() differs from sbappend() in
593 * that data supplied is treated as the beginning of a new record.
594 * To place a sender's address, optional access rights, and data in a
595 * socket receive buffer, sbappendaddr() should be used. To place
596 * access rights and data in a socket receive buffer, sbappendrights()
597 * should be used. In either case, the new data begins a new record.
598 * Note that unlike sbappend() and sbappendrecord(), these routines check
599 * for the caller that there will be enough space to store the data.
600 * Each fails if there is not enough space, or if it cannot find mbufs
601 * to store additional information in.
602 *
603 * Reliable protocols may use the socket send buffer to hold data
604 * awaiting acknowledgement. Data is normally copied from a socket
605 * send buffer in a protocol with m_copy for output to a peer,
606 * and then removing the data from the socket buffer with sbdrop()
607 * or sbdroprecord() when the data is acknowledged by the peer.
608 */
609
610#ifdef SOCKBUF_DEBUG
611void
612sblastrecordchk(struct sockbuf *sb, const char *file, int line)
613{
614 struct mbuf *m = sb->sb_mb;
615
616 SOCKBUF_LOCK_ASSERT(sb);
617
618 while (m && m->m_nextpkt)
619 m = m->m_nextpkt;
620
621 if (m != sb->sb_lastrecord) {
622 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
623 __func__, sb->sb_mb, sb->sb_lastrecord, m);
624 printf("packet chain:\n");
625 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
626 printf("\t%p\n", m);
627 panic("%s from %s:%u", __func__, file, line);
628 }
629}
630
631void
632sblastmbufchk(struct sockbuf *sb, const char *file, int line)
633{
634 struct mbuf *m = sb->sb_mb;
635 struct mbuf *n;
636
637 SOCKBUF_LOCK_ASSERT(sb);
638
639 while (m && m->m_nextpkt)
640 m = m->m_nextpkt;
641
642 while (m && m->m_next)
643 m = m->m_next;
644
645 if (m != sb->sb_mbtail) {
646 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
647 __func__, sb->sb_mb, sb->sb_mbtail, m);
648 printf("packet tree:\n");
649 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
650 printf("\t");
651 for (n = m; n != NULL; n = n->m_next)
652 printf("%p ", n);
653 printf("\n");
654 }
655 panic("%s from %s:%u", __func__, file, line);
656 }
657}
658#endif /* SOCKBUF_DEBUG */
659
660#define SBLINKRECORD(sb, m0) do { \
661 SOCKBUF_LOCK_ASSERT(sb); \
662 if ((sb)->sb_lastrecord != NULL) \
663 (sb)->sb_lastrecord->m_nextpkt = (m0); \
664 else \
665 (sb)->sb_mb = (m0); \
666 (sb)->sb_lastrecord = (m0); \
667} while (/*CONSTCOND*/0)
668
669/*
670 * Append mbuf chain m to the last record in the
671 * socket buffer sb. The additional space associated
672 * the mbuf chain is recorded in sb. Empty mbufs are
673 * discarded and mbufs are compacted where possible.
674 */
675void
676sbappend_locked(sb, m)
677 struct sockbuf *sb;
678 struct mbuf *m;
679{
680 register struct mbuf *n;
681
682 SOCKBUF_LOCK_ASSERT(sb);
683
684 if (m == 0)
685 return;
686
687 SBLASTRECORDCHK(sb);
688 n = sb->sb_mb;
689 if (n) {
690 while (n->m_nextpkt)
691 n = n->m_nextpkt;
692 do {
693 if (n->m_flags & M_EOR) {
694 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
695 return;
696 }
697 } while (n->m_next && (n = n->m_next));
698 } else {
699 /*
700 * XXX Would like to simply use sb_mbtail here, but
701 * XXX I need to verify that I won't miss an EOR that
702 * XXX way.
703 */
704 if ((n = sb->sb_lastrecord) != NULL) {
705 do {
706 if (n->m_flags & M_EOR) {
707 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
708 return;
709 }
710 } while (n->m_next && (n = n->m_next));
711 } else {
712 /*
713 * If this is the first record in the socket buffer,
714 * it's also the last record.
715 */
716 sb->sb_lastrecord = m;
717 }
718 }
719 sbcompress(sb, m, n);
720 SBLASTRECORDCHK(sb);
721}
722
723/*
724 * Append mbuf chain m to the last record in the
725 * socket buffer sb. The additional space associated
726 * the mbuf chain is recorded in sb. Empty mbufs are
727 * discarded and mbufs are compacted where possible.
728 */
729void
730sbappend(sb, m)
731 struct sockbuf *sb;
732 struct mbuf *m;
733{
734
735 SOCKBUF_LOCK(sb);
736 sbappend_locked(sb, m);
737 SOCKBUF_UNLOCK(sb);
738}
739
740/*
741 * This version of sbappend() should only be used when the caller
742 * absolutely knows that there will never be more than one record
743 * in the socket buffer, that is, a stream protocol (such as TCP).
744 */
745void
746sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
747{
748 SOCKBUF_LOCK_ASSERT(sb);
749
750 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
751 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
752
753 SBLASTMBUFCHK(sb);
754
755 sbcompress(sb, m, sb->sb_mbtail);
756
757 sb->sb_lastrecord = sb->sb_mb;
758 SBLASTRECORDCHK(sb);
759}
760
761/*
762 * This version of sbappend() should only be used when the caller
763 * absolutely knows that there will never be more than one record
764 * in the socket buffer, that is, a stream protocol (such as TCP).
765 */
766void
767sbappendstream(struct sockbuf *sb, struct mbuf *m)
768{
769
770 SOCKBUF_LOCK(sb);
771 sbappendstream_locked(sb, m);
772 SOCKBUF_UNLOCK(sb);
773}
774
775#ifdef SOCKBUF_DEBUG
776void
777sbcheck(sb)
778 struct sockbuf *sb;
779{
780 struct mbuf *m;
781 struct mbuf *n = 0;
782 u_long len = 0, mbcnt = 0;
783
784 SOCKBUF_LOCK_ASSERT(sb);
785
786 for (m = sb->sb_mb; m; m = n) {
787 n = m->m_nextpkt;
788 for (; m; m = m->m_next) {
789 len += m->m_len;
790 mbcnt += MSIZE;
791 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
792 mbcnt += m->m_ext.ext_size;
793 }
794 }
795 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
796 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
797 mbcnt, sb->sb_mbcnt);
798 panic("sbcheck");
799 }
800}
801#endif
802
803/*
804 * As above, except the mbuf chain
805 * begins a new record.
806 */
807void
808sbappendrecord_locked(sb, m0)
809 register struct sockbuf *sb;
810 register struct mbuf *m0;
811{
812 register struct mbuf *m;
813
814 SOCKBUF_LOCK_ASSERT(sb);
815
816 if (m0 == 0)
817 return;
818 m = sb->sb_mb;
819 if (m)
820 while (m->m_nextpkt)
821 m = m->m_nextpkt;
822 /*
823 * Put the first mbuf on the queue.
824 * Note this permits zero length records.
825 */
826 sballoc(sb, m0);
827 SBLASTRECORDCHK(sb);
828 SBLINKRECORD(sb, m0);
829 if (m)
830 m->m_nextpkt = m0;
831 else
832 sb->sb_mb = m0;
833 m = m0->m_next;
834 m0->m_next = 0;
835 if (m && (m0->m_flags & M_EOR)) {
836 m0->m_flags &= ~M_EOR;
837 m->m_flags |= M_EOR;
838 }
839 sbcompress(sb, m, m0);
840}
841
842/*
843 * As above, except the mbuf chain
844 * begins a new record.
845 */
846void
847sbappendrecord(sb, m0)
848 register struct sockbuf *sb;
849 register struct mbuf *m0;
850{
851
852 SOCKBUF_LOCK(sb);
853 sbappendrecord_locked(sb, m0);
854 SOCKBUF_UNLOCK(sb);
855}
856
857/*
858 * As above except that OOB data
859 * is inserted at the beginning of the sockbuf,
860 * but after any other OOB data.
861 */
862void
863sbinsertoob_locked(sb, m0)
864 register struct sockbuf *sb;
865 register struct mbuf *m0;
866{
867 register struct mbuf *m;
868 register struct mbuf **mp;
869
870 SOCKBUF_LOCK_ASSERT(sb);
871
872 if (m0 == 0)
873 return;
874 for (mp = &sb->sb_mb; *mp ; mp = &((*mp)->m_nextpkt)) {
875 m = *mp;
876 again:
877 switch (m->m_type) {
878
879 case MT_OOBDATA:
880 continue; /* WANT next train */
881
882 case MT_CONTROL:
883 m = m->m_next;
884 if (m)
885 goto again; /* inspect THIS train further */
886 }
887 break;
888 }
889 /*
890 * Put the first mbuf on the queue.
891 * Note this permits zero length records.
892 */
893 sballoc(sb, m0);
894 m0->m_nextpkt = *mp;
895 *mp = m0;
896 m = m0->m_next;
897 m0->m_next = 0;
898 if (m && (m0->m_flags & M_EOR)) {
899 m0->m_flags &= ~M_EOR;
900 m->m_flags |= M_EOR;
901 }
902 sbcompress(sb, m, m0);
903}
904
905/*
906 * As above except that OOB data
907 * is inserted at the beginning of the sockbuf,
908 * but after any other OOB data.
909 */
910void
911sbinsertoob(sb, m0)
912 register struct sockbuf *sb;
913 register struct mbuf *m0;
914{
915
916 SOCKBUF_LOCK(sb);
917 sbinsertoob_locked(sb, m0);
918 SOCKBUF_UNLOCK(sb);
919}
920
921/*
922 * Append address and data, and optionally, control (ancillary) data
923 * to the receive queue of a socket. If present,
924 * m0 must include a packet header with total length.
925 * Returns 0 if no space in sockbuf or insufficient mbufs.
926 */
927int
928sbappendaddr_locked(sb, asa, m0, control)
929 struct sockbuf *sb;
930 const struct sockaddr *asa;
931 struct mbuf *m0, *control;
932{
933 struct mbuf *m, *n, *nlast;
934 int space = asa->sa_len;
935
936 SOCKBUF_LOCK_ASSERT(sb);
937
938 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
939 panic("sbappendaddr_locked");
940 if (m0)
941 space += m0->m_pkthdr.len;
942 space += m_length(control, &n);
943
944 if (space > sbspace(sb))
945 return (0);
946#if MSIZE <= 256
947 if (asa->sa_len > MLEN)
948 return (0);
949#endif
950 MGET(m, M_DONTWAIT, MT_SONAME);
951 if (m == 0)
952 return (0);
953 m->m_len = asa->sa_len;
954 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
955 if (n)
956 n->m_next = m0; /* concatenate data to control */
957 else
958 control = m0;
959 m->m_next = control;
960 for (n = m; n->m_next != NULL; n = n->m_next)
961 sballoc(sb, n);
962 sballoc(sb, n);
963 nlast = n;
964 SBLINKRECORD(sb, m);
965
966 sb->sb_mbtail = nlast;
967 SBLASTMBUFCHK(sb);
968
969 SBLASTRECORDCHK(sb);
970 return (1);
971}
972
973/*
974 * Append address and data, and optionally, control (ancillary) data
975 * to the receive queue of a socket. If present,
976 * m0 must include a packet header with total length.
977 * Returns 0 if no space in sockbuf or insufficient mbufs.
978 */
979int
980sbappendaddr(sb, asa, m0, control)
981 struct sockbuf *sb;
982 const struct sockaddr *asa;
983 struct mbuf *m0, *control;
984{
985 int retval;
986
987 SOCKBUF_LOCK(sb);
988 retval = sbappendaddr_locked(sb, asa, m0, control);
989 SOCKBUF_UNLOCK(sb);
990 return (retval);
991}
992
993int
994sbappendcontrol_locked(sb, m0, control)
995 struct sockbuf *sb;
996 struct mbuf *control, *m0;
997{
998 struct mbuf *m, *n, *mlast;
999 int space;
1000
1001 SOCKBUF_LOCK_ASSERT(sb);
1002
1003 if (control == 0)
1004 panic("sbappendcontrol_locked");
1005 space = m_length(control, &n) + m_length(m0, NULL);
1006
1007 if (space > sbspace(sb))
1008 return (0);
1009 n->m_next = m0; /* concatenate data to control */
1010
1011 SBLASTRECORDCHK(sb);
1012
1013 for (m = control; m->m_next; m = m->m_next)
1014 sballoc(sb, m);
1015 sballoc(sb, m);
1016 mlast = m;
1017 SBLINKRECORD(sb, control);
1018
1019 sb->sb_mbtail = mlast;
1020 SBLASTMBUFCHK(sb);
1021
1022 SBLASTRECORDCHK(sb);
1023 return (1);
1024}
1025
1026int
1027sbappendcontrol(sb, m0, control)
1028 struct sockbuf *sb;
1029 struct mbuf *control, *m0;
1030{
1031 int retval;
1032
1033 SOCKBUF_LOCK(sb);
1034 retval = sbappendcontrol_locked(sb, m0, control);
1035 SOCKBUF_UNLOCK(sb);
1036 return (retval);
1037}
1038
1039/*
1040 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
1041 * (n). If (n) is NULL, the buffer is presumed empty.
1042 *
1043 * When the data is compressed, mbufs in the chain may be handled in one of
1044 * three ways:
1045 *
1046 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
1047 * record boundary, and no change in data type).
1048 *
1049 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
1050 * an mbuf already in the socket buffer. This can occur if an
1051 * appropriate mbuf exists, there is room, and no merging of data types
1052 * will occur.
1053 *
1054 * (3) The mbuf may be appended to the end of the existing mbuf chain.
1055 *
1056 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
1057 * end-of-record.
1058 */
1059void
1060sbcompress(sb, m, n)
1061 register struct sockbuf *sb;
1062 register struct mbuf *m, *n;
1063{
1064 register int eor = 0;
1065 register struct mbuf *o;
1066
1067 SOCKBUF_LOCK_ASSERT(sb);
1068
1069 while (m) {
1070 eor |= m->m_flags & M_EOR;
1071 if (m->m_len == 0 &&
1072 (eor == 0 ||
1073 (((o = m->m_next) || (o = n)) &&
1074 o->m_type == m->m_type))) {
1075 if (sb->sb_lastrecord == m)
1076 sb->sb_lastrecord = m->m_next;
1077 m = m_free(m);
1078 continue;
1079 }
1080 if (n && (n->m_flags & M_EOR) == 0 &&
1081 M_WRITABLE(n) &&
1082 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1083 m->m_len <= M_TRAILINGSPACE(n) &&
1084 n->m_type == m->m_type) {
1085 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
1086 (unsigned)m->m_len);
1087 n->m_len += m->m_len;
1088 sb->sb_cc += m->m_len;
1089 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1090 /* XXX: Probably don't need.*/
1091 sb->sb_ctl += m->m_len;
1092 m = m_free(m);
1093 continue;
1094 }
1095 if (n)
1096 n->m_next = m;
1097 else
1098 sb->sb_mb = m;
1099 sb->sb_mbtail = m;
1100 sballoc(sb, m);
1101 n = m;
1102 m->m_flags &= ~M_EOR;
1103 m = m->m_next;
1104 n->m_next = 0;
1105 }
1106 if (eor) {
1107 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
1108 n->m_flags |= eor;
1109 }
1110 SBLASTMBUFCHK(sb);
1111}
1112
1113/*
1114 * Free all mbufs in a sockbuf.
1115 * Check that all resources are reclaimed.
1116 */
1117void
1118sbflush_locked(sb)
1119 register struct sockbuf *sb;
1120{
1121
1122 SOCKBUF_LOCK_ASSERT(sb);
1123
1124 if (sb->sb_flags & SB_LOCK)
1125 panic("sbflush_locked: locked");
1126 while (sb->sb_mbcnt) {
1127 /*
1128 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
1129 * we would loop forever. Panic instead.
1130 */
1131 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
1132 break;
1133 sbdrop_locked(sb, (int)sb->sb_cc);
1134 }
1135 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
1136 panic("sbflush_locked: cc %u || mb %p || mbcnt %u", sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
1137}
1138
1139void
1140sbflush(sb)
1141 register struct sockbuf *sb;
1142{
1143
1144 SOCKBUF_LOCK(sb);
1145 sbflush_locked(sb);
1146 SOCKBUF_UNLOCK(sb);
1147}
1148
1149/*
1150 * Drop data from (the front of) a sockbuf.
1151 */
1152void
1153sbdrop_locked(sb, len)
1154 register struct sockbuf *sb;
1155 register int len;
1156{
1157 register struct mbuf *m;
1158 struct mbuf *next;
1159
1160 SOCKBUF_LOCK_ASSERT(sb);
1161
1162 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1163 while (len > 0) {
1164 if (m == 0) {
1165 if (next == 0)
1166 panic("sbdrop");
1167 m = next;
1168 next = m->m_nextpkt;
1169 continue;
1170 }
1171 if (m->m_len > len) {
1172 m->m_len -= len;
1173 m->m_data += len;
1174 sb->sb_cc -= len;
1175 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1176 sb->sb_ctl -= len;
1177 break;
1178 }
1179 len -= m->m_len;
1180 sbfree(sb, m);
1181 m = m_free(m);
1182 }
1183 while (m && m->m_len == 0) {
1184 sbfree(sb, m);
1185 m = m_free(m);
1186 }
1187 if (m) {
1188 sb->sb_mb = m;
1189 m->m_nextpkt = next;
1190 } else
1191 sb->sb_mb = next;
1192 /*
1193 * First part is an inline SB_EMPTY_FIXUP(). Second part
1194 * makes sure sb_lastrecord is up-to-date if we dropped
1195 * part of the last record.
1196 */
1197 m = sb->sb_mb;
1198 if (m == NULL) {
1199 sb->sb_mbtail = NULL;
1200 sb->sb_lastrecord = NULL;
1201 } else if (m->m_nextpkt == NULL) {
1202 sb->sb_lastrecord = m;
1203 }
1204}
1205
1206/*
1207 * Drop data from (the front of) a sockbuf.
1208 */
1209void
1210sbdrop(sb, len)
1211 register struct sockbuf *sb;
1212 register int len;
1213{
1214
1215 SOCKBUF_LOCK(sb);
1216 sbdrop_locked(sb, len);
1217 SOCKBUF_UNLOCK(sb);
1218}
1219
1220/*
1221 * Drop a record off the front of a sockbuf
1222 * and move the next record to the front.
1223 */
1224void
1225sbdroprecord_locked(sb)
1226 register struct sockbuf *sb;
1227{
1228 register struct mbuf *m;
1229
1230 SOCKBUF_LOCK_ASSERT(sb);
1231
1232 m = sb->sb_mb;
1233 if (m) {
1234 sb->sb_mb = m->m_nextpkt;
1235 do {
1236 sbfree(sb, m);
1237 m = m_free(m);
1238 } while (m);
1239 }
1240 SB_EMPTY_FIXUP(sb);
1241}
1242
1243/*
1244 * Drop a record off the front of a sockbuf
1245 * and move the next record to the front.
1246 */
1247void
1248sbdroprecord(sb)
1249 register struct sockbuf *sb;
1250{
1251
1252 SOCKBUF_LOCK(sb);
1253 sbdroprecord_locked(sb);
1254 SOCKBUF_UNLOCK(sb);
1255}
1256
1257/*
1258 * Create a "control" mbuf containing the specified data
1259 * with the specified type for presentation on a socket buffer.
1260 */
1261struct mbuf *
1262sbcreatecontrol(p, size, type, level)
1263 caddr_t p;
1264 register int size;
1265 int type, level;
1266{
1267 register struct cmsghdr *cp;
1268 struct mbuf *m;
1269
1270 if (CMSG_SPACE((u_int)size) > MCLBYTES)
1271 return ((struct mbuf *) NULL);
1272 if (CMSG_SPACE((u_int)size) > MLEN)
1273 m = m_getcl(M_DONTWAIT, MT_CONTROL, 0);
1274 else
1275 m = m_get(M_DONTWAIT, MT_CONTROL);
1276 if (m == NULL)
1277 return ((struct mbuf *) NULL);
1278 cp = mtod(m, struct cmsghdr *);
1279 m->m_len = 0;
1280 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1281 ("sbcreatecontrol: short mbuf"));
1282 if (p != NULL)
1283 (void)memcpy(CMSG_DATA(cp), p, size);
1284 m->m_len = CMSG_SPACE(size);
1285 cp->cmsg_len = CMSG_LEN(size);
1286 cp->cmsg_level = level;
1287 cp->cmsg_type = type;
1288 return (m);
1289}
1290
1291/*
1292 * Some routines that return EOPNOTSUPP for entry points that are not
1293 * supported by a protocol. Fill in as needed.
1294 */
|
1299}
1300
1301int
1302pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
1303{
1304 return EOPNOTSUPP;
1305}
1306
1307int
1308pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
1309{
1310 return EOPNOTSUPP;
1311}
1312
1313int
1314pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
1315{
1316 return EOPNOTSUPP;
1317}
1318
1319int
1320pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
1321{
1322 return EOPNOTSUPP;
1323}
1324
1325int
1326pru_connect2_notsupp(struct socket *so1, struct socket *so2)
1327{
1328 return EOPNOTSUPP;
1329}
1330
1331int
1332pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
1333 struct ifnet *ifp, struct thread *td)
1334{
1335 return EOPNOTSUPP;
1336}
1337
1338int
1339pru_detach_notsupp(struct socket *so)
1340{
1341 return EOPNOTSUPP;
1342}
1343
1344int
1345pru_disconnect_notsupp(struct socket *so)
1346{
1347 return EOPNOTSUPP;
1348}
1349
1350int
1351pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
1352{
1353 return EOPNOTSUPP;
1354}
1355
1356int
1357pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
1358{
1359 return EOPNOTSUPP;
1360}
1361
1362int
1363pru_rcvd_notsupp(struct socket *so, int flags)
1364{
1365 return EOPNOTSUPP;
1366}
1367
1368int
1369pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
1370{
1371 return EOPNOTSUPP;
1372}
1373
1374int
1375pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
1376 struct sockaddr *addr, struct mbuf *control, struct thread *td)
1377{
1378 return EOPNOTSUPP;
1379}
1380
1381/*
1382 * This isn't really a ``null'' operation, but it's the default one
1383 * and doesn't do anything destructive.
1384 */
1385int
1386pru_sense_null(struct socket *so, struct stat *sb)
1387{
1388 sb->st_blksize = so->so_snd.sb_hiwat;
1389 return 0;
1390}
1391
1392int
1393pru_shutdown_notsupp(struct socket *so)
1394{
1395 return EOPNOTSUPP;
1396}
1397
1398int
1399pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
1400{
1401 return EOPNOTSUPP;
1402}
1403
1404int
1405pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1406 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1407{
1408 return EOPNOTSUPP;
1409}
1410
1411int
1412pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
1413 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp,
1414 int *flagsp)
1415{
1416 return EOPNOTSUPP;
1417}
1418
1419int
1420pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
1421 struct thread *td)
1422{
1423 return EOPNOTSUPP;
1424}
1425
1426/*
1427 * For protocol types that don't keep cached copies of labels in their
1428 * pcbs, provide a null sosetlabel that does a NOOP.
1429 */
1430void
1431pru_sosetlabel_null(struct socket *so)
1432{
1433
1434}
1435
1436/*
1437 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
1438 */
1439struct sockaddr *
1440sodupsockaddr(const struct sockaddr *sa, int mflags)
1441{
1442 struct sockaddr *sa2;
1443
1444 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
1445 if (sa2)
1446 bcopy(sa, sa2, sa->sa_len);
1447 return sa2;
1448}
1449
1450/*
1451 * Create an external-format (``xsocket'') structure using the information
1452 * in the kernel-format socket structure pointed to by so. This is done
1453 * to reduce the spew of irrelevant information over this interface,
1454 * to isolate user code from changes in the kernel structure, and
1455 * potentially to provide information-hiding if we decide that
1456 * some of this information should be hidden from users.
1457 */
1458void
1459sotoxsocket(struct socket *so, struct xsocket *xso)
1460{
1461 xso->xso_len = sizeof *xso;
1462 xso->xso_so = so;
1463 xso->so_type = so->so_type;
1464 xso->so_options = so->so_options;
1465 xso->so_linger = so->so_linger;
1466 xso->so_state = so->so_state;
1467 xso->so_pcb = so->so_pcb;
1468 xso->xso_protocol = so->so_proto->pr_protocol;
1469 xso->xso_family = so->so_proto->pr_domain->dom_family;
1470 xso->so_qlen = so->so_qlen;
1471 xso->so_incqlen = so->so_incqlen;
1472 xso->so_qlimit = so->so_qlimit;
1473 xso->so_timeo = so->so_timeo;
1474 xso->so_error = so->so_error;
1475 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
1476 xso->so_oobmark = so->so_oobmark;
1477 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
1478 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
1479 xso->so_uid = so->so_cred->cr_uid;
1480}
1481
1482/*
1483 * This does the same for sockbufs. Note that the xsockbuf structure,
1484 * since it is always embedded in a socket, does not include a self
1485 * pointer nor a length. We make this entry point public in case
1486 * some other mechanism needs it.
1487 */
1488void
1489sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1490{
1491 xsb->sb_cc = sb->sb_cc;
1492 xsb->sb_hiwat = sb->sb_hiwat;
1493 xsb->sb_mbcnt = sb->sb_mbcnt;
1494 xsb->sb_mbmax = sb->sb_mbmax;
1495 xsb->sb_lowat = sb->sb_lowat;
1496 xsb->sb_flags = sb->sb_flags;
1497 xsb->sb_timeo = sb->sb_timeo;
1498}
1499
1500/*
1501 * Here is the definition of some of the basic objects in the kern.ipc
1502 * branch of the MIB.
1503 */
1504SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
1505
1506/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1507static int dummy;
1508SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1509SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1510 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1511SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RDTUN,
1512 &maxsockets, 0, "Maximum number of sockets avaliable");
1513SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1514 &sb_efficiency, 0, "");
1515
1516/*
1517 * Initialise maxsockets
1518 */
1519static void init_maxsockets(void *ignored)
1520{
1521 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1522 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1523}
1524SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
| 1299}
1300
1301int
1302pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
1303{
1304 return EOPNOTSUPP;
1305}
1306
1307int
1308pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
1309{
1310 return EOPNOTSUPP;
1311}
1312
1313int
1314pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
1315{
1316 return EOPNOTSUPP;
1317}
1318
1319int
1320pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
1321{
1322 return EOPNOTSUPP;
1323}
1324
1325int
1326pru_connect2_notsupp(struct socket *so1, struct socket *so2)
1327{
1328 return EOPNOTSUPP;
1329}
1330
1331int
1332pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
1333 struct ifnet *ifp, struct thread *td)
1334{
1335 return EOPNOTSUPP;
1336}
1337
1338int
1339pru_detach_notsupp(struct socket *so)
1340{
1341 return EOPNOTSUPP;
1342}
1343
1344int
1345pru_disconnect_notsupp(struct socket *so)
1346{
1347 return EOPNOTSUPP;
1348}
1349
1350int
1351pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
1352{
1353 return EOPNOTSUPP;
1354}
1355
1356int
1357pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
1358{
1359 return EOPNOTSUPP;
1360}
1361
1362int
1363pru_rcvd_notsupp(struct socket *so, int flags)
1364{
1365 return EOPNOTSUPP;
1366}
1367
1368int
1369pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
1370{
1371 return EOPNOTSUPP;
1372}
1373
1374int
1375pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
1376 struct sockaddr *addr, struct mbuf *control, struct thread *td)
1377{
1378 return EOPNOTSUPP;
1379}
1380
1381/*
1382 * This isn't really a ``null'' operation, but it's the default one
1383 * and doesn't do anything destructive.
1384 */
1385int
1386pru_sense_null(struct socket *so, struct stat *sb)
1387{
1388 sb->st_blksize = so->so_snd.sb_hiwat;
1389 return 0;
1390}
1391
1392int
1393pru_shutdown_notsupp(struct socket *so)
1394{
1395 return EOPNOTSUPP;
1396}
1397
1398int
1399pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
1400{
1401 return EOPNOTSUPP;
1402}
1403
1404int
1405pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1406 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1407{
1408 return EOPNOTSUPP;
1409}
1410
1411int
1412pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
1413 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp,
1414 int *flagsp)
1415{
1416 return EOPNOTSUPP;
1417}
1418
1419int
1420pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
1421 struct thread *td)
1422{
1423 return EOPNOTSUPP;
1424}
1425
1426/*
1427 * For protocol types that don't keep cached copies of labels in their
1428 * pcbs, provide a null sosetlabel that does a NOOP.
1429 */
1430void
1431pru_sosetlabel_null(struct socket *so)
1432{
1433
1434}
1435
1436/*
1437 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
1438 */
1439struct sockaddr *
1440sodupsockaddr(const struct sockaddr *sa, int mflags)
1441{
1442 struct sockaddr *sa2;
1443
1444 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
1445 if (sa2)
1446 bcopy(sa, sa2, sa->sa_len);
1447 return sa2;
1448}
1449
1450/*
1451 * Create an external-format (``xsocket'') structure using the information
1452 * in the kernel-format socket structure pointed to by so. This is done
1453 * to reduce the spew of irrelevant information over this interface,
1454 * to isolate user code from changes in the kernel structure, and
1455 * potentially to provide information-hiding if we decide that
1456 * some of this information should be hidden from users.
1457 */
1458void
1459sotoxsocket(struct socket *so, struct xsocket *xso)
1460{
1461 xso->xso_len = sizeof *xso;
1462 xso->xso_so = so;
1463 xso->so_type = so->so_type;
1464 xso->so_options = so->so_options;
1465 xso->so_linger = so->so_linger;
1466 xso->so_state = so->so_state;
1467 xso->so_pcb = so->so_pcb;
1468 xso->xso_protocol = so->so_proto->pr_protocol;
1469 xso->xso_family = so->so_proto->pr_domain->dom_family;
1470 xso->so_qlen = so->so_qlen;
1471 xso->so_incqlen = so->so_incqlen;
1472 xso->so_qlimit = so->so_qlimit;
1473 xso->so_timeo = so->so_timeo;
1474 xso->so_error = so->so_error;
1475 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
1476 xso->so_oobmark = so->so_oobmark;
1477 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
1478 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
1479 xso->so_uid = so->so_cred->cr_uid;
1480}
1481
1482/*
1483 * This does the same for sockbufs. Note that the xsockbuf structure,
1484 * since it is always embedded in a socket, does not include a self
1485 * pointer nor a length. We make this entry point public in case
1486 * some other mechanism needs it.
1487 */
1488void
1489sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1490{
1491 xsb->sb_cc = sb->sb_cc;
1492 xsb->sb_hiwat = sb->sb_hiwat;
1493 xsb->sb_mbcnt = sb->sb_mbcnt;
1494 xsb->sb_mbmax = sb->sb_mbmax;
1495 xsb->sb_lowat = sb->sb_lowat;
1496 xsb->sb_flags = sb->sb_flags;
1497 xsb->sb_timeo = sb->sb_timeo;
1498}
1499
1500/*
1501 * Here is the definition of some of the basic objects in the kern.ipc
1502 * branch of the MIB.
1503 */
1504SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
1505
1506/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1507static int dummy;
1508SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1509SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1510 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1511SYSCTL_INT(_kern_ipc, OID_AUTO, maxsockets, CTLFLAG_RDTUN,
1512 &maxsockets, 0, "Maximum number of sockets avaliable");
1513SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1514 &sb_efficiency, 0, "");
1515
1516/*
1517 * Initialise maxsockets
1518 */
1519static void init_maxsockets(void *ignored)
1520{
1521 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
1522 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
1523}
1524SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
|