?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
accept: Save foreign address earlier, if protocol supports it
[dragonfly.git] / sys / kern / uipc_socket.c
blob118a7415453c241e3d203082d85854826e446e90
1 /*
2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2004 The DragonFly Project. All rights reserved.
4 *
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34 /*
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993
36 * The Regents of the University of California. All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
67 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
68 */
69
70 #include "opt_inet.h"
71 #include "opt_sctp.h"
72
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/fcntl.h>
76 #include <sys/malloc.h>
77 #include <sys/mbuf.h>
78 #include <sys/domain.h>
79 #include <sys/file.h> /* for struct knote */
80 #include <sys/kernel.h>
81 #include <sys/event.h>
82 #include <sys/proc.h>
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/socketops.h>
87 #include <sys/resourcevar.h>
88 #include <sys/signalvar.h>
89 #include <sys/sysctl.h>
90 #include <sys/uio.h>
91 #include <sys/jail.h>
92 #include <vm/vm_zone.h>
93 #include <vm/pmap.h>
94 #include <net/netmsg2.h>
95
96 #include <sys/thread2.h>
97 #include <sys/socketvar2.h>
98
99 #include <machine/limits.h>
100
101 extern int tcp_sosnd_agglim;
102 extern int tcp_sosnd_async;
103
104 #ifdef INET
105 static int do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
106 #endif /* INET */
107
108 static void filt_sordetach(struct knote *kn);
109 static int filt_soread(struct knote *kn, long hint);
110 static void filt_sowdetach(struct knote *kn);
111 static int filt_sowrite(struct knote *kn, long hint);
112 static int filt_solisten(struct knote *kn, long hint);
113
114 static void sodiscard(struct socket *so);
115 static int soclose_sync(struct socket *so, int fflag);
116 static void soclose_fast(struct socket *so);
117
118 static struct filterops solisten_filtops =
119 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
120 static struct filterops soread_filtops =
121 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
122 static struct filterops sowrite_filtops =
123 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
124 static struct filterops soexcept_filtops =
125 { FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
126
127 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
128 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
129 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
130
131
132 static int somaxconn = SOMAXCONN;
133 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
134 &somaxconn, 0, "Maximum pending socket connection queue size");
135
136 static int use_soclose_fast = 1;
137 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
138 &use_soclose_fast, 0, "Fast socket close");
139
140 int use_soaccept_pred_fast = 1;
141 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
142 &use_soaccept_pred_fast, 0, "Fast socket accept predication");
143
144 /*
145 * Socket operation routines.
146 * These routines are called by the routines in
147 * sys_socket.c or from a system process, and
148 * implement the semantics of socket operations by
149 * switching out to the protocol specific routines.
150 */
151
152 /*
153 * Get a socket structure, and initialize it.
154 * Note that it would probably be better to allocate socket
155 * and PCB at the same time, but I'm not convinced that all
156 * the protocols can be easily modified to do this.
157 */
158 struct socket *
159 soalloc(int waitok)
160 {
161 struct socket *so;
162 unsigned waitmask;
163
164 waitmask = waitok ? M_WAITOK : M_NOWAIT;
165 so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
166 if (so) {
167 /* XXX race condition for reentrant kernel */
168 TAILQ_INIT(&so->so_aiojobq);
169 TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
170 TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
171 lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
172 lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
173 so->so_state = SS_NOFDREF;
174 so->so_refs = 1;
175 }
176 return so;
177 }
178
179 int
180 socreate(int dom, struct socket **aso, int type,
181 int proto, struct thread *td)
182 {
183 struct proc *p = td->td_proc;
184 struct protosw *prp;
185 struct socket *so;
186 struct pru_attach_info ai;
187 int error;
188
189 if (proto)
190 prp = pffindproto(dom, proto, type);
191 else
192 prp = pffindtype(dom, type);
193
194 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
195 return (EPROTONOSUPPORT);
196
197 if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
198 prp->pr_domain->dom_family != PF_LOCAL &&
199 prp->pr_domain->dom_family != PF_INET &&
200 prp->pr_domain->dom_family != PF_INET6 &&
201 prp->pr_domain->dom_family != PF_ROUTE) {
202 return (EPROTONOSUPPORT);
203 }
204
205 if (prp->pr_type != type)
206 return (EPROTOTYPE);
207 so = soalloc(p != 0);
208 if (so == NULL)
209 return (ENOBUFS);
210
211 /*
212 * Callers of socreate() presumably will connect up a descriptor
213 * and call soclose() if they cannot. This represents our so_refs
214 * (which should be 1) from soalloc().
215 */
216 soclrstate(so, SS_NOFDREF);
217
218 /*
219 * Set a default port for protocol processing. No action will occur
220 * on the socket on this port until an inpcb is attached to it and
221 * is able to match incoming packets, or until the socket becomes
222 * available to userland.
223 *
224 * We normally default the socket to the protocol thread on cpu 0.
225 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
226 * thread and all pr_*()/pru_*() calls are executed synchronously.
227 */
228 if (prp->pr_flags & PR_SYNC_PORT)
229 so->so_port = &netisr_sync_port;
230 else
231 so->so_port = cpu_portfn(0);
232
233 TAILQ_INIT(&so->so_incomp);
234 TAILQ_INIT(&so->so_comp);
235 so->so_type = type;
236 so->so_cred = crhold(p->p_ucred);
237 so->so_proto = prp;
238 ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
239 ai.p_ucred = p->p_ucred;
240 ai.fd_rdir = p->p_fd->fd_rdir;
241
242 /*
243 * Auto-sizing of socket buffers is managed by the protocols and
244 * the appropriate flags must be set in the pru_attach function.
245 */
246 error = so_pru_attach(so, proto, &ai);
247 if (error) {
248 sosetstate(so, SS_NOFDREF);
249 sofree(so); /* from soalloc */
250 return error;
251 }
252
253 /*
254 * NOTE: Returns referenced socket.
255 */
256 *aso = so;
257 return (0);
258 }
259
260 int
261 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
262 {
263 int error;
264
265 error = so_pru_bind(so, nam, td);
266 return (error);
267 }
268
269 static void
270 sodealloc(struct socket *so)
271 {
272 if (so->so_rcv.ssb_hiwat)
273 (void)chgsbsize(so->so_cred->cr_uidinfo,
274 &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
275 if (so->so_snd.ssb_hiwat)
276 (void)chgsbsize(so->so_cred->cr_uidinfo,
277 &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
278 #ifdef INET
279 /* remove accept filter if present */
280 if (so->so_accf != NULL)
281 do_setopt_accept_filter(so, NULL);
282 #endif /* INET */
283 crfree(so->so_cred);
284 if (so->so_faddr != NULL)
285 kfree(so->so_faddr, M_SONAME);
286 kfree(so, M_SOCKET);
287 }
288
289 int
290 solisten(struct socket *so, int backlog, struct thread *td)
291 {
292 int error;
293 #ifdef SCTP
294 short oldopt, oldqlimit;
295 #endif /* SCTP */
296
297 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
298 return (EINVAL);
299
300 #ifdef SCTP
301 oldopt = so->so_options;
302 oldqlimit = so->so_qlimit;
303 #endif /* SCTP */
304
305 lwkt_gettoken(&so->so_rcv.ssb_token);
306 if (TAILQ_EMPTY(&so->so_comp))
307 so->so_options |= SO_ACCEPTCONN;
308 lwkt_reltoken(&so->so_rcv.ssb_token);
309 if (backlog < 0 || backlog > somaxconn)
310 backlog = somaxconn;
311 so->so_qlimit = backlog;
312 /* SCTP needs to look at tweak both the inbound backlog parameter AND
313 * the so_options (UDP model both connect's and gets inbound
314 * connections .. implicitly).
315 */
316 error = so_pru_listen(so, td);
317 if (error) {
318 #ifdef SCTP
319 /* Restore the params */
320 so->so_options = oldopt;
321 so->so_qlimit = oldqlimit;
322 #endif /* SCTP */
323 return (error);
324 }
325 return (0);
326 }
327
328 /*
329 * Destroy a disconnected socket. This routine is a NOP if entities
330 * still have a reference on the socket:
331 *
332 * so_pcb - The protocol stack still has a reference
333 * SS_NOFDREF - There is no longer a file pointer reference
334 */
335 void
336 sofree(struct socket *so)
337 {
338 struct socket *head;
339
340 /*
341 * This is a bit hackish at the moment. We need to interlock
342 * any accept queue we are on before we potentially lose the
343 * last reference to avoid races against a re-reference from
344 * someone operating on the queue.
345 */
346 while ((head = so->so_head) != NULL) {
347 lwkt_getpooltoken(head);
348 if (so->so_head == head)
349 break;
350 lwkt_relpooltoken(head);
351 }
352
353 /*
354 * Arbitrage the last free.
355 */
356 KKASSERT(so->so_refs > 0);
357 if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
358 if (head)
359 lwkt_relpooltoken(head);
360 return;
361 }
362
363 KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
364 KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
365
366 /*
367 * We're done, remove ourselves from the accept queue we are
368 * on, if we are on one.
369 */
370 if (head != NULL) {
371 if (so->so_state & SS_INCOMP) {
372 TAILQ_REMOVE(&head->so_incomp, so, so_list);
373 head->so_incqlen--;
374 } else if (so->so_state & SS_COMP) {
375 /*
376 * We must not decommission a socket that's
377 * on the accept(2) queue. If we do, then
378 * accept(2) may hang after select(2) indicated
379 * that the listening socket was ready.
380 */
381 lwkt_relpooltoken(head);
382 return;
383 } else {
384 panic("sofree: not queued");
385 }
386 soclrstate(so, SS_INCOMP);
387 so->so_head = NULL;
388 lwkt_relpooltoken(head);
389 }
390 ssb_release(&so->so_snd, so);
391 sorflush(so);
392 sodealloc(so);
393 }
394
395 /*
396 * Close a socket on last file table reference removal.
397 * Initiate disconnect if connected.
398 * Free socket when disconnect complete.
399 */
400 int
401 soclose(struct socket *so, int fflag)
402 {
403 int error;
404
405 funsetown(&so->so_sigio);
406 if (!use_soclose_fast ||
407 (so->so_proto->pr_flags & PR_SYNC_PORT) ||
408 (so->so_options & SO_LINGER)) {
409 error = soclose_sync(so, fflag);
410 } else {
411 soclose_fast(so);
412 error = 0;
413 }
414 return error;
415 }
416
417 static void
418 sodiscard(struct socket *so)
419 {
420 lwkt_getpooltoken(so);
421 if (so->so_options & SO_ACCEPTCONN) {
422 struct socket *sp;
423
424 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
425 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
426 soclrstate(sp, SS_INCOMP);
427 sp->so_head = NULL;
428 so->so_incqlen--;
429 soaborta(sp);
430 }
431 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
432 TAILQ_REMOVE(&so->so_comp, sp, so_list);
433 soclrstate(sp, SS_COMP);
434 sp->so_head = NULL;
435 so->so_qlen--;
436 soaborta(sp);
437 }
438 }
439 lwkt_relpooltoken(so);
440
441 if (so->so_state & SS_NOFDREF)
442 panic("soclose: NOFDREF");
443 sosetstate(so, SS_NOFDREF); /* take ref */
444 }
445
446 static int
447 soclose_sync(struct socket *so, int fflag)
448 {
449 int error = 0;
450
451 if (so->so_pcb == NULL)
452 goto discard;
453 if (so->so_state & SS_ISCONNECTED) {
454 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
455 error = sodisconnect(so);
456 if (error)
457 goto drop;
458 }
459 if (so->so_options & SO_LINGER) {
460 if ((so->so_state & SS_ISDISCONNECTING) &&
461 (fflag & FNONBLOCK))
462 goto drop;
463 while (so->so_state & SS_ISCONNECTED) {
464 error = tsleep(&so->so_timeo, PCATCH,
465 "soclos", so->so_linger * hz);
466 if (error)
467 break;
468 }
469 }
470 }
471 drop:
472 if (so->so_pcb) {
473 int error2;
474
475 error2 = so_pru_detach(so);
476 if (error == 0)
477 error = error2;
478 }
479 discard:
480 sodiscard(so);
481 so_pru_sync(so); /* unpend async sending */
482 sofree(so); /* dispose of ref */
483
484 return (error);
485 }
486
487 static void
488 soclose_sofree_async_handler(netmsg_t msg)
489 {
490 sofree(msg->base.nm_so);
491 }
492
493 static void
494 soclose_sofree_async(struct socket *so)
495 {
496 struct netmsg_base *base = &so->so_clomsg;
497
498 netmsg_init(base, so, &netisr_apanic_rport, 0,
499 soclose_sofree_async_handler);
500 lwkt_sendmsg(so->so_port, &base->lmsg);
501 }
502
503 static void
504 soclose_disconn_async_handler(netmsg_t msg)
505 {
506 struct socket *so = msg->base.nm_so;
507
508 if ((so->so_state & SS_ISCONNECTED) &&
509 (so->so_state & SS_ISDISCONNECTING) == 0)
510 so_pru_disconnect_direct(so);
511
512 if (so->so_pcb)
513 so_pru_detach_direct(so);
514
515 sodiscard(so);
516 sofree(so);
517 }
518
519 static void
520 soclose_disconn_async(struct socket *so)
521 {
522 struct netmsg_base *base = &so->so_clomsg;
523
524 netmsg_init(base, so, &netisr_apanic_rport, 0,
525 soclose_disconn_async_handler);
526 lwkt_sendmsg(so->so_port, &base->lmsg);
527 }
528
529 static void
530 soclose_detach_async_handler(netmsg_t msg)
531 {
532 struct socket *so = msg->base.nm_so;
533
534 if (so->so_pcb)
535 so_pru_detach_direct(so);
536
537 sodiscard(so);
538 sofree(so);
539 }
540
541 static void
542 soclose_detach_async(struct socket *so)
543 {
544 struct netmsg_base *base = &so->so_clomsg;
545
546 netmsg_init(base, so, &netisr_apanic_rport, 0,
547 soclose_detach_async_handler);
548 lwkt_sendmsg(so->so_port, &base->lmsg);
549 }
550
551 static void
552 soclose_fast(struct socket *so)
553 {
554 if (so->so_pcb == NULL)
555 goto discard;
556
557 if ((so->so_state & SS_ISCONNECTED) &&
558 (so->so_state & SS_ISDISCONNECTING) == 0) {
559 soclose_disconn_async(so);
560 return;
561 }
562
563 if (so->so_pcb) {
564 soclose_detach_async(so);
565 return;
566 }
567
568 discard:
569 sodiscard(so);
570 soclose_sofree_async(so);
571 }
572
573 /*
574 * Abort and destroy a socket. Only one abort can be in progress
575 * at any given moment.
576 */
577 void
578 soabort(struct socket *so)
579 {
580 soreference(so);
581 so_pru_abort(so);
582 }
583
584 void
585 soaborta(struct socket *so)
586 {
587 soreference(so);
588 so_pru_aborta(so);
589 }
590
591 void
592 soabort_oncpu(struct socket *so)
593 {
594 soreference(so);
595 so_pru_abort_oncpu(so);
596 }
597
598 /*
599 * so is passed in ref'd, which becomes owned by
600 * the cleared SS_NOFDREF flag.
601 */
602 void
603 soaccept_generic(struct socket *so)
604 {
605 if ((so->so_state & SS_NOFDREF) == 0)
606 panic("soaccept: !NOFDREF");
607 soclrstate(so, SS_NOFDREF); /* owned by lack of SS_NOFDREF */
608 }
609
610 int
611 soaccept(struct socket *so, struct sockaddr **nam)
612 {
613 int error;
614
615 soaccept_generic(so);
616 error = so_pru_accept_direct(so, nam);
617 return (error);
618 }
619
620 int
621 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
622 {
623 int error;
624
625 if (so->so_options & SO_ACCEPTCONN)
626 return (EOPNOTSUPP);
627 /*
628 * If protocol is connection-based, can only connect once.
629 * Otherwise, if connected, try to disconnect first.
630 * This allows user to disconnect by connecting to, e.g.,
631 * a null address.
632 */
633 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
634 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
635 (error = sodisconnect(so)))) {
636 error = EISCONN;
637 } else {
638 /*
639 * Prevent accumulated error from previous connection
640 * from biting us.
641 */
642 so->so_error = 0;
643 error = so_pru_connect(so, nam, td);
644 }
645 return (error);
646 }
647
648 int
649 soconnect2(struct socket *so1, struct socket *so2)
650 {
651 int error;
652
653 error = so_pru_connect2(so1, so2);
654 return (error);
655 }
656
657 int
658 sodisconnect(struct socket *so)
659 {
660 int error;
661
662 if ((so->so_state & SS_ISCONNECTED) == 0) {
663 error = ENOTCONN;
664 goto bad;
665 }
666 if (so->so_state & SS_ISDISCONNECTING) {
667 error = EALREADY;
668 goto bad;
669 }
670 error = so_pru_disconnect(so);
671 bad:
672 return (error);
673 }
674
675 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
676 /*
677 * Send on a socket.
678 * If send must go all at once and message is larger than
679 * send buffering, then hard error.
680 * Lock against other senders.
681 * If must go all at once and not enough room now, then
682 * inform user that this would block and do nothing.
683 * Otherwise, if nonblocking, send as much as possible.
684 * The data to be sent is described by "uio" if nonzero,
685 * otherwise by the mbuf chain "top" (which must be null
686 * if uio is not). Data provided in mbuf chain must be small
687 * enough to send all at once.
688 *
689 * Returns nonzero on error, timeout or signal; callers
690 * must check for short counts if EINTR/ERESTART are returned.
691 * Data and control buffers are freed on return.
692 */
693 int
694 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
695 struct mbuf *top, struct mbuf *control, int flags,
696 struct thread *td)
697 {
698 struct mbuf **mp;
699 struct mbuf *m;
700 size_t resid;
701 int space, len;
702 int clen = 0, error, dontroute, mlen;
703 int atomic = sosendallatonce(so) || top;
704 int pru_flags;
705
706 if (uio) {
707 resid = uio->uio_resid;
708 } else {
709 resid = (size_t)top->m_pkthdr.len;
710 #ifdef INVARIANTS
711 len = 0;
712 for (m = top; m; m = m->m_next)
713 len += m->m_len;
714 KKASSERT(top->m_pkthdr.len == len);
715 #endif
716 }
717
718 /*
719 * WARNING! resid is unsigned, space and len are signed. space
720 * can wind up negative if the sockbuf is overcommitted.
721 *
722 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
723 * type sockets since that's an error.
724 */
725 if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
726 error = EINVAL;
727 goto out;
728 }
729
730 dontroute =
731 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
732 (so->so_proto->pr_flags & PR_ATOMIC);
733 if (td->td_lwp != NULL)
734 td->td_lwp->lwp_ru.ru_msgsnd++;
735 if (control)
736 clen = control->m_len;
737 #define gotoerr(errcode) { error = errcode; goto release; }
738
739 restart:
740 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
741 if (error)
742 goto out;
743
744 do {
745 if (so->so_state & SS_CANTSENDMORE)
746 gotoerr(EPIPE);
747 if (so->so_error) {
748 error = so->so_error;
749 so->so_error = 0;
750 goto release;
751 }
752 if ((so->so_state & SS_ISCONNECTED) == 0) {
753 /*
754 * `sendto' and `sendmsg' is allowed on a connection-
755 * based socket if it supports implied connect.
756 * Return ENOTCONN if not connected and no address is
757 * supplied.
758 */
759 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
760 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
761 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
762 !(resid == 0 && clen != 0))
763 gotoerr(ENOTCONN);
764 } else if (addr == 0)
765 gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
766 ENOTCONN : EDESTADDRREQ);
767 }
768 if ((atomic && resid > so->so_snd.ssb_hiwat) ||
769 clen > so->so_snd.ssb_hiwat) {
770 gotoerr(EMSGSIZE);
771 }
772 space = ssb_space(&so->so_snd);
773 if (flags & MSG_OOB)
774 space += 1024;
775 if ((space < 0 || (size_t)space < resid + clen) && uio &&
776 (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
777 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
778 gotoerr(EWOULDBLOCK);
779 ssb_unlock(&so->so_snd);
780 error = ssb_wait(&so->so_snd);
781 if (error)
782 goto out;
783 goto restart;
784 }
785 mp = &top;
786 space -= clen;
787 do {
788 if (uio == NULL) {
789 /*
790 * Data is prepackaged in "top".
791 */
792 resid = 0;
793 if (flags & MSG_EOR)
794 top->m_flags |= M_EOR;
795 } else do {
796 if (resid > INT_MAX)
797 resid = INT_MAX;
798 m = m_getl((int)resid, MB_WAIT, MT_DATA,
799 top == NULL ? M_PKTHDR : 0, &mlen);
800 if (top == NULL) {
801 m->m_pkthdr.len = 0;
802 m->m_pkthdr.rcvif = NULL;
803 }
804 len = imin((int)szmin(mlen, resid), space);
805 if (resid < MINCLSIZE) {
806 /*
807 * For datagram protocols, leave room
808 * for protocol headers in first mbuf.
809 */
810 if (atomic && top == 0 && len < mlen)
811 MH_ALIGN(m, len);
812 }
813 space -= len;
814 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
815 resid = uio->uio_resid;
816 m->m_len = len;
817 *mp = m;
818 top->m_pkthdr.len += len;
819 if (error)
820 goto release;
821 mp = &m->m_next;
822 if (resid == 0) {
823 if (flags & MSG_EOR)
824 top->m_flags |= M_EOR;
825 break;
826 }
827 } while (space > 0 && atomic);
828 if (dontroute)
829 so->so_options |= SO_DONTROUTE;
830 if (flags & MSG_OOB) {
831 pru_flags = PRUS_OOB;
832 } else if ((flags & MSG_EOF) &&
833 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
834 (resid == 0)) {
835 /*
836 * If the user set MSG_EOF, the protocol
837 * understands this flag and nothing left to
838 * send then use PRU_SEND_EOF instead of PRU_SEND.
839 */
840 pru_flags = PRUS_EOF;
841 } else if (resid > 0 && space > 0) {
842 /* If there is more to send, set PRUS_MORETOCOME */
843 pru_flags = PRUS_MORETOCOME;
844 } else {
845 pru_flags = 0;
846 }
847 /*
848 * XXX all the SS_CANTSENDMORE checks previously
849 * done could be out of date. We could have recieved
850 * a reset packet in an interrupt or maybe we slept
851 * while doing page faults in uiomove() etc. We could
852 * probably recheck again inside the splnet() protection
853 * here, but there are probably other places that this
854 * also happens. We must rethink this.
855 */
856 error = so_pru_send(so, pru_flags, top, addr, control, td);
857 if (dontroute)
858 so->so_options &= ~SO_DONTROUTE;
859 clen = 0;
860 control = 0;
861 top = NULL;
862 mp = &top;
863 if (error)
864 goto release;
865 } while (resid && space > 0);
866 } while (resid);
867
868 release:
869 ssb_unlock(&so->so_snd);
870 out:
871 if (top)
872 m_freem(top);
873 if (control)
874 m_freem(control);
875 return (error);
876 }
877
878 /*
879 * A specialization of sosend() for UDP based on protocol-specific knowledge:
880 * so->so_proto->pr_flags has the PR_ATOMIC field set. This means that
881 * sosendallatonce() returns true,
882 * the "atomic" variable is true,
883 * and sosendudp() blocks until space is available for the entire send.
884 * so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
885 * PR_IMPLOPCL flags set.
886 * UDP has no out-of-band data.
887 * UDP has no control data.
888 * UDP does not support MSG_EOR.
889 */
890 int
891 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
892 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
893 {
894 boolean_t dontroute; /* temporary SO_DONTROUTE setting */
895 size_t resid;
896 int error;
897 int space;
898
899 if (td->td_lwp != NULL)
900 td->td_lwp->lwp_ru.ru_msgsnd++;
901 if (control)
902 m_freem(control);
903
904 KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
905 resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
906
907 restart:
908 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
909 if (error)
910 goto out;
911
912 if (so->so_state & SS_CANTSENDMORE)
913 gotoerr(EPIPE);
914 if (so->so_error) {
915 error = so->so_error;
916 so->so_error = 0;
917 goto release;
918 }
919 if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
920 gotoerr(EDESTADDRREQ);
921 if (resid > so->so_snd.ssb_hiwat)
922 gotoerr(EMSGSIZE);
923 space = ssb_space(&so->so_snd);
924 if (uio && (space < 0 || (size_t)space < resid)) {
925 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
926 gotoerr(EWOULDBLOCK);
927 ssb_unlock(&so->so_snd);
928 error = ssb_wait(&so->so_snd);
929 if (error)
930 goto out;
931 goto restart;
932 }
933
934 if (uio) {
935 top = m_uiomove(uio);
936 if (top == NULL)
937 goto release;
938 }
939
940 dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
941 if (dontroute)
942 so->so_options |= SO_DONTROUTE;
943
944 error = so_pru_send(so, 0, top, addr, NULL, td);
945 top = NULL; /* sent or freed in lower layer */
946
947 if (dontroute)
948 so->so_options &= ~SO_DONTROUTE;
949
950 release:
951 ssb_unlock(&so->so_snd);
952 out:
953 if (top)
954 m_freem(top);
955 return (error);
956 }
957
958 int
959 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
960 struct mbuf *top, struct mbuf *control, int flags,
961 struct thread *td)
962 {
963 struct mbuf **mp;
964 struct mbuf *m;
965 size_t resid;
966 int space, len;
967 int error, mlen;
968 int allatonce;
969 int pru_flags;
970
971 if (uio) {
972 KKASSERT(top == NULL);
973 allatonce = 0;
974 resid = uio->uio_resid;
975 } else {
976 allatonce = 1;
977 resid = (size_t)top->m_pkthdr.len;
978 #ifdef INVARIANTS
979 len = 0;
980 for (m = top; m; m = m->m_next)
981 len += m->m_len;
982 KKASSERT(top->m_pkthdr.len == len);
983 #endif
984 }
985
986 /*
987 * WARNING! resid is unsigned, space and len are signed. space
988 * can wind up negative if the sockbuf is overcommitted.
989 *
990 * Also check to make sure that MSG_EOR isn't used on TCP
991 */
992 if (flags & MSG_EOR) {
993 error = EINVAL;
994 goto out;
995 }
996
997 if (control) {
998 /* TCP doesn't do control messages (rights, creds, etc) */
999 if (control->m_len) {
1000 error = EINVAL;
1001 goto out;
1002 }
1003 m_freem(control); /* empty control, just free it */
1004 control = NULL;
1005 }
1006
1007 if (td->td_lwp != NULL)
1008 td->td_lwp->lwp_ru.ru_msgsnd++;
1009
1010 #define gotoerr(errcode) { error = errcode; goto release; }
1011
1012 restart:
1013 error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1014 if (error)
1015 goto out;
1016
1017 do {
1018 if (so->so_state & SS_CANTSENDMORE)
1019 gotoerr(EPIPE);
1020 if (so->so_error) {
1021 error = so->so_error;
1022 so->so_error = 0;
1023 goto release;
1024 }
1025 if ((so->so_state & SS_ISCONNECTED) == 0 &&
1026 (so->so_state & SS_ISCONFIRMING) == 0)
1027 gotoerr(ENOTCONN);
1028 if (allatonce && resid > so->so_snd.ssb_hiwat)
1029 gotoerr(EMSGSIZE);
1030
1031 space = ssb_space(&so->so_snd);
1032 if (flags & MSG_OOB)
1033 space += 1024;
1034 if ((space < 0 || (size_t)space < resid) && !allatonce &&
1035 space < so->so_snd.ssb_lowat) {
1036 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1037 gotoerr(EWOULDBLOCK);
1038 ssb_unlock(&so->so_snd);
1039 error = ssb_wait(&so->so_snd);
1040 if (error)
1041 goto out;
1042 goto restart;
1043 }
1044 mp = &top;
1045 do {
1046 int cnt = 0, async = 0;
1047
1048 if (uio == NULL) {
1049 /*
1050 * Data is prepackaged in "top".
1051 */
1052 resid = 0;
1053 } else do {
1054 if (resid > INT_MAX)
1055 resid = INT_MAX;
1056 m = m_getl((int)resid, MB_WAIT, MT_DATA,
1057 top == NULL ? M_PKTHDR : 0, &mlen);
1058 if (top == NULL) {
1059 m->m_pkthdr.len = 0;
1060 m->m_pkthdr.rcvif = NULL;
1061 }
1062 len = imin((int)szmin(mlen, resid), space);
1063 space -= len;
1064 error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1065 resid = uio->uio_resid;
1066 m->m_len = len;
1067 *mp = m;
1068 top->m_pkthdr.len += len;
1069 if (error)
1070 goto release;
1071 mp = &m->m_next;
1072 if (resid == 0)
1073 break;
1074 ++cnt;
1075 } while (space > 0 && cnt < tcp_sosnd_agglim);
1076
1077 if (tcp_sosnd_async)
1078 async = 1;
1079
1080 if (flags & MSG_OOB) {
1081 pru_flags = PRUS_OOB;
1082 async = 0;
1083 } else if ((flags & MSG_EOF) && resid == 0) {
1084 pru_flags = PRUS_EOF;
1085 } else if (resid > 0 && space > 0) {
1086 /* If there is more to send, set PRUS_MORETOCOME */
1087 pru_flags = PRUS_MORETOCOME;
1088 async = 1;
1089 } else {
1090 pru_flags = 0;
1091 }
1092
1093 if (flags & MSG_SYNC)
1094 async = 0;
1095
1096 /*
1097 * XXX all the SS_CANTSENDMORE checks previously
1098 * done could be out of date. We could have recieved
1099 * a reset packet in an interrupt or maybe we slept
1100 * while doing page faults in uiomove() etc. We could
1101 * probably recheck again inside the splnet() protection
1102 * here, but there are probably other places that this
1103 * also happens. We must rethink this.
1104 */
1105 if (!async) {
1106 error = so_pru_send(so, pru_flags, top,
1107 NULL, NULL, td);
1108 } else {
1109 so_pru_send_async(so, pru_flags, top,
1110 NULL, NULL, td);
1111 error = 0;
1112 }
1113
1114 top = NULL;
1115 mp = &top;
1116 if (error)
1117 goto release;
1118 } while (resid && space > 0);
1119 } while (resid);
1120
1121 release:
1122 ssb_unlock(&so->so_snd);
1123 out:
1124 if (top)
1125 m_freem(top);
1126 if (control)
1127 m_freem(control);
1128 return (error);
1129 }
1130
1131 /*
1132 * Implement receive operations on a socket.
1133 *
1134 * We depend on the way that records are added to the signalsockbuf
1135 * by sbappend*. In particular, each record (mbufs linked through m_next)
1136 * must begin with an address if the protocol so specifies,
1137 * followed by an optional mbuf or mbufs containing ancillary data,
1138 * and then zero or more mbufs of data.
1139 *
1140 * Although the signalsockbuf is locked, new data may still be appended.
1141 * A token inside the ssb_lock deals with MP issues and still allows
1142 * the network to access the socket if we block in a uio.
1143 *
1144 * The caller may receive the data as a single mbuf chain by supplying
1145 * an mbuf **mp0 for use in returning the chain. The uio is then used
1146 * only for the count in uio_resid.
1147 */
1148 int
1149 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1150 struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1151 {
1152 struct mbuf *m, *n;
1153 struct mbuf *free_chain = NULL;
1154 int flags, len, error, offset;
1155 struct protosw *pr = so->so_proto;
1156 int moff, type = 0;
1157 size_t resid, orig_resid;
1158
1159 if (uio)
1160 resid = uio->uio_resid;
1161 else
1162 resid = (size_t)(sio->sb_climit - sio->sb_cc);
1163 orig_resid = resid;
1164
1165 if (psa)
1166 *psa = NULL;
1167 if (controlp)
1168 *controlp = NULL;
1169 if (flagsp)
1170 flags = *flagsp &~ MSG_EOR;
1171 else
1172 flags = 0;
1173 if (flags & MSG_OOB) {
1174 m = m_get(MB_WAIT, MT_DATA);
1175 if (m == NULL)
1176 return (ENOBUFS);
1177 error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1178 if (error)
1179 goto bad;
1180 if (sio) {
1181 do {
1182 sbappend(sio, m);
1183 KKASSERT(resid >= (size_t)m->m_len);
1184 resid -= (size_t)m->m_len;
1185 } while (resid > 0 && m);
1186 } else {
1187 do {
1188 uio->uio_resid = resid;
1189 error = uiomove(mtod(m, caddr_t),
1190 (int)szmin(resid, m->m_len),
1191 uio);
1192 resid = uio->uio_resid;
1193 m = m_free(m);
1194 } while (uio->uio_resid && error == 0 && m);
1195 }
1196 bad:
1197 if (m)
1198 m_freem(m);
1199 return (error);
1200 }
1201 if ((so->so_state & SS_ISCONFIRMING) && resid)
1202 so_pru_rcvd(so, 0);
1203
1204 /*
1205 * The token interlocks against the protocol thread while
1206 * ssb_lock is a blocking lock against other userland entities.
1207 */
1208 lwkt_gettoken(&so->so_rcv.ssb_token);
1209 restart:
1210 error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1211 if (error)
1212 goto done;
1213
1214 m = so->so_rcv.ssb_mb;
1215 /*
1216 * If we have less data than requested, block awaiting more
1217 * (subject to any timeout) if:
1218 * 1. the current count is less than the low water mark, or
1219 * 2. MSG_WAITALL is set, and it is possible to do the entire
1220 * receive operation at once if we block (resid <= hiwat).
1221 * 3. MSG_DONTWAIT is not set
1222 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1223 * we have to do the receive in sections, and thus risk returning
1224 * a short count if a timeout or signal occurs after we start.
1225 */
1226 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1227 (size_t)so->so_rcv.ssb_cc < resid) &&
1228 (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1229 ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1230 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1231 KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1232 if (so->so_error) {
1233 if (m)
1234 goto dontblock;
1235 error = so->so_error;
1236 if ((flags & MSG_PEEK) == 0)
1237 so->so_error = 0;
1238 goto release;
1239 }
1240 if (so->so_state & SS_CANTRCVMORE) {
1241 if (m)
1242 goto dontblock;
1243 else
1244 goto release;
1245 }
1246 for (; m; m = m->m_next) {
1247 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1248 m = so->so_rcv.ssb_mb;
1249 goto dontblock;
1250 }
1251 }
1252 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1253 (pr->pr_flags & PR_CONNREQUIRED)) {
1254 error = ENOTCONN;
1255 goto release;
1256 }
1257 if (resid == 0)
1258 goto release;
1259 if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1260 error = EWOULDBLOCK;
1261 goto release;
1262 }
1263 ssb_unlock(&so->so_rcv);
1264 error = ssb_wait(&so->so_rcv);
1265 if (error)
1266 goto done;
1267 goto restart;
1268 }
1269 dontblock:
1270 if (uio && uio->uio_td && uio->uio_td->td_proc)
1271 uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1272
1273 /*
1274 * note: m should be == sb_mb here. Cache the next record while
1275 * cleaning up. Note that calling m_free*() will break out critical
1276 * section.
1277 */
1278 KKASSERT(m == so->so_rcv.ssb_mb);
1279
1280 /*
1281 * Skip any address mbufs prepending the record.
1282 */
1283 if (pr->pr_flags & PR_ADDR) {
1284 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1285 orig_resid = 0;
1286 if (psa)
1287 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1288 if (flags & MSG_PEEK)
1289 m = m->m_next;
1290 else
1291 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1292 }
1293
1294 /*
1295 * Skip any control mbufs prepending the record.
1296 */
1297 #ifdef SCTP
1298 if (pr->pr_flags & PR_ADDR_OPT) {
1299 /*
1300 * For SCTP we may be getting a
1301 * whole message OR a partial delivery.
1302 */
1303 if (m && m->m_type == MT_SONAME) {
1304 orig_resid = 0;
1305 if (psa)
1306 *psa = dup_sockaddr(mtod(m, struct sockaddr *));
1307 if (flags & MSG_PEEK)
1308 m = m->m_next;
1309 else
1310 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1311 }
1312 }
1313 #endif /* SCTP */
1314 while (m && m->m_type == MT_CONTROL && error == 0) {
1315 if (flags & MSG_PEEK) {
1316 if (controlp)
1317 *controlp = m_copy(m, 0, m->m_len);
1318 m = m->m_next; /* XXX race */
1319 } else {
1320 if (controlp) {
1321 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1322 if (pr->pr_domain->dom_externalize &&
1323 mtod(m, struct cmsghdr *)->cmsg_type ==
1324 SCM_RIGHTS)
1325 error = (*pr->pr_domain->dom_externalize)(m);
1326 *controlp = m;
1327 m = n;
1328 } else {
1329 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1330 }
1331 }
1332 if (controlp && *controlp) {
1333 orig_resid = 0;
1334 controlp = &(*controlp)->m_next;
1335 }
1336 }
1337
1338 /*
1339 * flag OOB data.
1340 */
1341 if (m) {
1342 type = m->m_type;
1343 if (type == MT_OOBDATA)
1344 flags |= MSG_OOB;
1345 }
1346
1347 /*
1348 * Copy to the UIO or mbuf return chain (*mp).
1349 */
1350 moff = 0;
1351 offset = 0;
1352 while (m && resid > 0 && error == 0) {
1353 if (m->m_type == MT_OOBDATA) {
1354 if (type != MT_OOBDATA)
1355 break;
1356 } else if (type == MT_OOBDATA)
1357 break;
1358 else
1359 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1360 ("receive 3"));
1361 soclrstate(so, SS_RCVATMARK);
1362 len = (resid > INT_MAX) ? INT_MAX : resid;
1363 if (so->so_oobmark && len > so->so_oobmark - offset)
1364 len = so->so_oobmark - offset;
1365 if (len > m->m_len - moff)
1366 len = m->m_len - moff;
1367
1368 /*
1369 * Copy out to the UIO or pass the mbufs back to the SIO.
1370 * The SIO is dealt with when we eat the mbuf, but deal
1371 * with the resid here either way.
1372 */
1373 if (uio) {
1374 uio->uio_resid = resid;
1375 error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1376 resid = uio->uio_resid;
1377 if (error)
1378 goto release;
1379 } else {
1380 resid -= (size_t)len;
1381 }
1382
1383 /*
1384 * Eat the entire mbuf or just a piece of it
1385 */
1386 if (len == m->m_len - moff) {
1387 if (m->m_flags & M_EOR)
1388 flags |= MSG_EOR;
1389 #ifdef SCTP
1390 if (m->m_flags & M_NOTIFICATION)
1391 flags |= MSG_NOTIFICATION;
1392 #endif /* SCTP */
1393 if (flags & MSG_PEEK) {
1394 m = m->m_next;
1395 moff = 0;
1396 } else {
1397 if (sio) {
1398 n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1399 sbappend(sio, m);
1400 m = n;
1401 } else {
1402 m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1403 }
1404 }
1405 } else {
1406 if (flags & MSG_PEEK) {
1407 moff += len;
1408 } else {
1409 if (sio) {
1410 n = m_copym(m, 0, len, MB_WAIT);
1411 if (n)
1412 sbappend(sio, n);
1413 }
1414 m->m_data += len;
1415 m->m_len -= len;
1416 so->so_rcv.ssb_cc -= len;
1417 }
1418 }
1419 if (so->so_oobmark) {
1420 if ((flags & MSG_PEEK) == 0) {
1421 so->so_oobmark -= len;
1422 if (so->so_oobmark == 0) {
1423 sosetstate(so, SS_RCVATMARK);
1424 break;
1425 }
1426 } else {
1427 offset += len;
1428 if (offset == so->so_oobmark)
1429 break;
1430 }
1431 }
1432 if (flags & MSG_EOR)
1433 break;
1434 /*
1435 * If the MSG_WAITALL flag is set (for non-atomic socket),
1436 * we must not quit until resid == 0 or an error
1437 * termination. If a signal/timeout occurs, return
1438 * with a short count but without error.
1439 * Keep signalsockbuf locked against other readers.
1440 */
1441 while ((flags & MSG_WAITALL) && m == NULL &&
1442 resid > 0 && !sosendallatonce(so) &&
1443 so->so_rcv.ssb_mb == NULL) {
1444 if (so->so_error || so->so_state & SS_CANTRCVMORE)
1445 break;
1446 /*
1447 * The window might have closed to zero, make
1448 * sure we send an ack now that we've drained
1449 * the buffer or we might end up blocking until
1450 * the idle takes over (5 seconds).
1451 */
1452 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1453 so_pru_rcvd(so, flags);
1454 error = ssb_wait(&so->so_rcv);
1455 if (error) {
1456 ssb_unlock(&so->so_rcv);
1457 error = 0;
1458 goto done;
1459 }
1460 m = so->so_rcv.ssb_mb;
1461 }
1462 }
1463
1464 /*
1465 * If an atomic read was requested but unread data still remains
1466 * in the record, set MSG_TRUNC.
1467 */
1468 if (m && pr->pr_flags & PR_ATOMIC)
1469 flags |= MSG_TRUNC;
1470
1471 /*
1472 * Cleanup. If an atomic read was requested drop any unread data.
1473 */
1474 if ((flags & MSG_PEEK) == 0) {
1475 if (m && (pr->pr_flags & PR_ATOMIC))
1476 sbdroprecord(&so->so_rcv.sb);
1477 if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1478 so_pru_rcvd(so, flags);
1479 }
1480
1481 if (orig_resid == resid && orig_resid &&
1482 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1483 ssb_unlock(&so->so_rcv);
1484 goto restart;
1485 }
1486
1487 if (flagsp)
1488 *flagsp |= flags;
1489 release:
1490 ssb_unlock(&so->so_rcv);
1491 done:
1492 lwkt_reltoken(&so->so_rcv.ssb_token);
1493 if (free_chain)
1494 m_freem(free_chain);
1495 return (error);
1496 }
1497
1498 /*
1499 * Shut a socket down. Note that we do not get a frontend lock as we
1500 * want to be able to shut the socket down even if another thread is
1501 * blocked in a read(), thus waking it up.
1502 */
1503 int
1504 soshutdown(struct socket *so, int how)
1505 {
1506 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1507 return (EINVAL);
1508
1509 if (how != SHUT_WR) {
1510 /*ssb_lock(&so->so_rcv, M_WAITOK);*/
1511 sorflush(so);
1512 /*ssb_unlock(&so->so_rcv);*/
1513 }
1514 if (how != SHUT_RD)
1515 return (so_pru_shutdown(so));
1516 return (0);
1517 }
1518
1519 void
1520 sorflush(struct socket *so)
1521 {
1522 struct signalsockbuf *ssb = &so->so_rcv;
1523 struct protosw *pr = so->so_proto;
1524 struct signalsockbuf asb;
1525
1526 atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1527
1528 lwkt_gettoken(&ssb->ssb_token);
1529 socantrcvmore(so);
1530 asb = *ssb;
1531
1532 /*
1533 * Can't just blow up the ssb structure here
1534 */
1535 bzero(&ssb->sb, sizeof(ssb->sb));
1536 ssb->ssb_timeo = 0;
1537 ssb->ssb_lowat = 0;
1538 ssb->ssb_hiwat = 0;
1539 ssb->ssb_mbmax = 0;
1540 atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1541
1542 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1543 (*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1544 ssb_release(&asb, so);
1545
1546 lwkt_reltoken(&ssb->ssb_token);
1547 }
1548
1549 #ifdef INET
1550 static int
1551 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1552 {
1553 struct accept_filter_arg *afap = NULL;
1554 struct accept_filter *afp;
1555 struct so_accf *af = so->so_accf;
1556 int error = 0;
1557
1558 /* do not set/remove accept filters on non listen sockets */
1559 if ((so->so_options & SO_ACCEPTCONN) == 0) {
1560 error = EINVAL;
1561 goto out;
1562 }
1563
1564 /* removing the filter */
1565 if (sopt == NULL) {
1566 if (af != NULL) {
1567 if (af->so_accept_filter != NULL &&
1568 af->so_accept_filter->accf_destroy != NULL) {
1569 af->so_accept_filter->accf_destroy(so);
1570 }
1571 if (af->so_accept_filter_str != NULL) {
1572 FREE(af->so_accept_filter_str, M_ACCF);
1573 }
1574 FREE(af, M_ACCF);
1575 so->so_accf = NULL;
1576 }
1577 so->so_options &= ~SO_ACCEPTFILTER;
1578 return (0);
1579 }
1580 /* adding a filter */
1581 /* must remove previous filter first */
1582 if (af != NULL) {
1583 error = EINVAL;
1584 goto out;
1585 }
1586 /* don't put large objects on the kernel stack */
1587 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1588 error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1589 afap->af_name[sizeof(afap->af_name)-1] = '\0';
1590 afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1591 if (error)
1592 goto out;
1593 afp = accept_filt_get(afap->af_name);
1594 if (afp == NULL) {
1595 error = ENOENT;
1596 goto out;
1597 }
1598 MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1599 if (afp->accf_create != NULL) {
1600 if (afap->af_name[0] != '\0') {
1601 int len = strlen(afap->af_name) + 1;
1602
1603 MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1604 strcpy(af->so_accept_filter_str, afap->af_name);
1605 }
1606 af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1607 if (af->so_accept_filter_arg == NULL) {
1608 FREE(af->so_accept_filter_str, M_ACCF);
1609 FREE(af, M_ACCF);
1610 so->so_accf = NULL;
1611 error = EINVAL;
1612 goto out;
1613 }
1614 }
1615 af->so_accept_filter = afp;
1616 so->so_accf = af;
1617 so->so_options |= SO_ACCEPTFILTER;
1618 out:
1619 if (afap != NULL)
1620 FREE(afap, M_TEMP);
1621 return (error);
1622 }
1623 #endif /* INET */
1624
1625 /*
1626 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1627 * an additional variant to handle the case where the option value needs
1628 * to be some kind of integer, but not a specific size.
1629 * In addition to their use here, these functions are also called by the
1630 * protocol-level pr_ctloutput() routines.
1631 */
1632 int
1633 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1634 {
1635 return soopt_to_kbuf(sopt, buf, len, minlen);
1636 }
1637
1638 int
1639 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1640 {
1641 size_t valsize;
1642
1643 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1644 KKASSERT(kva_p(buf));
1645
1646 /*
1647 * If the user gives us more than we wanted, we ignore it,
1648 * but if we don't get the minimum length the caller
1649 * wants, we return EINVAL. On success, sopt->sopt_valsize
1650 * is set to however much we actually retrieved.
1651 */
1652 if ((valsize = sopt->sopt_valsize) < minlen)
1653 return EINVAL;
1654 if (valsize > len)
1655 sopt->sopt_valsize = valsize = len;
1656
1657 bcopy(sopt->sopt_val, buf, valsize);
1658 return 0;
1659 }
1660
1661
1662 int
1663 sosetopt(struct socket *so, struct sockopt *sopt)
1664 {
1665 int error, optval;
1666 struct linger l;
1667 struct timeval tv;
1668 u_long val;
1669 struct signalsockbuf *sotmp;
1670
1671 error = 0;
1672 sopt->sopt_dir = SOPT_SET;
1673 if (sopt->sopt_level != SOL_SOCKET) {
1674 if (so->so_proto && so->so_proto->pr_ctloutput) {
1675 return (so_pr_ctloutput(so, sopt));
1676 }
1677 error = ENOPROTOOPT;
1678 } else {
1679 switch (sopt->sopt_name) {
1680 #ifdef INET
1681 case SO_ACCEPTFILTER:
1682 error = do_setopt_accept_filter(so, sopt);
1683 if (error)
1684 goto bad;
1685 break;
1686 #endif /* INET */
1687 case SO_LINGER:
1688 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1689 if (error)
1690 goto bad;
1691
1692 so->so_linger = l.l_linger;
1693 if (l.l_onoff)
1694 so->so_options |= SO_LINGER;
1695 else
1696 so->so_options &= ~SO_LINGER;
1697 break;
1698
1699 case SO_DEBUG:
1700 case SO_KEEPALIVE:
1701 case SO_DONTROUTE:
1702 case SO_USELOOPBACK:
1703 case SO_BROADCAST:
1704 case SO_REUSEADDR:
1705 case SO_REUSEPORT:
1706 case SO_OOBINLINE:
1707 case SO_TIMESTAMP:
1708 error = sooptcopyin(sopt, &optval, sizeof optval,
1709 sizeof optval);
1710 if (error)
1711 goto bad;
1712 if (optval)
1713 so->so_options |= sopt->sopt_name;
1714 else
1715 so->so_options &= ~sopt->sopt_name;
1716 break;
1717
1718 case SO_SNDBUF:
1719 case SO_RCVBUF:
1720 case SO_SNDLOWAT:
1721 case SO_RCVLOWAT:
1722 error = sooptcopyin(sopt, &optval, sizeof optval,
1723 sizeof optval);
1724 if (error)
1725 goto bad;
1726
1727 /*
1728 * Values < 1 make no sense for any of these
1729 * options, so disallow them.
1730 */
1731 if (optval < 1) {
1732 error = EINVAL;
1733 goto bad;
1734 }
1735
1736 switch (sopt->sopt_name) {
1737 case SO_SNDBUF:
1738 case SO_RCVBUF:
1739 if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1740 &so->so_snd : &so->so_rcv, (u_long)optval,
1741 so,
1742 &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1743 error = ENOBUFS;
1744 goto bad;
1745 }
1746 sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1747 &so->so_snd : &so->so_rcv;
1748 atomic_clear_int(&sotmp->ssb_flags,
1749 SSB_AUTOSIZE);
1750 break;
1751
1752 /*
1753 * Make sure the low-water is never greater than
1754 * the high-water.
1755 */
1756 case SO_SNDLOWAT:
1757 so->so_snd.ssb_lowat =
1758 (optval > so->so_snd.ssb_hiwat) ?
1759 so->so_snd.ssb_hiwat : optval;
1760 atomic_clear_int(&so->so_snd.ssb_flags,
1761 SSB_AUTOLOWAT);
1762 break;
1763 case SO_RCVLOWAT:
1764 so->so_rcv.ssb_lowat =
1765 (optval > so->so_rcv.ssb_hiwat) ?
1766 so->so_rcv.ssb_hiwat : optval;
1767 atomic_clear_int(&so->so_rcv.ssb_flags,
1768 SSB_AUTOLOWAT);
1769 break;
1770 }
1771 break;
1772
1773 case SO_SNDTIMEO:
1774 case SO_RCVTIMEO:
1775 error = sooptcopyin(sopt, &tv, sizeof tv,
1776 sizeof tv);
1777 if (error)
1778 goto bad;
1779
1780 /* assert(hz > 0); */
1781 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1782 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1783 error = EDOM;
1784 goto bad;
1785 }
1786 /* assert(tick > 0); */
1787 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
1788 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1789 if (val > INT_MAX) {
1790 error = EDOM;
1791 goto bad;
1792 }
1793 if (val == 0 && tv.tv_usec != 0)
1794 val = 1;
1795
1796 switch (sopt->sopt_name) {
1797 case SO_SNDTIMEO:
1798 so->so_snd.ssb_timeo = val;
1799 break;
1800 case SO_RCVTIMEO:
1801 so->so_rcv.ssb_timeo = val;
1802 break;
1803 }
1804 break;
1805 default:
1806 error = ENOPROTOOPT;
1807 break;
1808 }
1809 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1810 (void) so_pr_ctloutput(so, sopt);
1811 }
1812 }
1813 bad:
1814 return (error);
1815 }
1816
1817 /* Helper routine for getsockopt */
1818 int
1819 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1820 {
1821 soopt_from_kbuf(sopt, buf, len);
1822 return 0;
1823 }
1824
1825 void
1826 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1827 {
1828 size_t valsize;
1829
1830 if (len == 0) {
1831 sopt->sopt_valsize = 0;
1832 return;
1833 }
1834
1835 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1836 KKASSERT(kva_p(buf));
1837
1838 /*
1839 * Documented get behavior is that we always return a value,
1840 * possibly truncated to fit in the user's buffer.
1841 * Traditional behavior is that we always tell the user
1842 * precisely how much we copied, rather than something useful
1843 * like the total amount we had available for her.
1844 * Note that this interface is not idempotent; the entire answer must
1845 * generated ahead of time.
1846 */
1847 valsize = szmin(len, sopt->sopt_valsize);
1848 sopt->sopt_valsize = valsize;
1849 if (sopt->sopt_val != 0) {
1850 bcopy(buf, sopt->sopt_val, valsize);
1851 }
1852 }
1853
1854 int
1855 sogetopt(struct socket *so, struct sockopt *sopt)
1856 {
1857 int error, optval;
1858 long optval_l;
1859 struct linger l;
1860 struct timeval tv;
1861 #ifdef INET
1862 struct accept_filter_arg *afap;
1863 #endif
1864
1865 error = 0;
1866 sopt->sopt_dir = SOPT_GET;
1867 if (sopt->sopt_level != SOL_SOCKET) {
1868 if (so->so_proto && so->so_proto->pr_ctloutput) {
1869 return (so_pr_ctloutput(so, sopt));
1870 } else
1871 return (ENOPROTOOPT);
1872 } else {
1873 switch (sopt->sopt_name) {
1874 #ifdef INET
1875 case SO_ACCEPTFILTER:
1876 if ((so->so_options & SO_ACCEPTCONN) == 0)
1877 return (EINVAL);
1878 MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1879 M_TEMP, M_WAITOK | M_ZERO);
1880 if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1881 strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1882 if (so->so_accf->so_accept_filter_str != NULL)
1883 strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1884 }
1885 error = sooptcopyout(sopt, afap, sizeof(*afap));
1886 FREE(afap, M_TEMP);
1887 break;
1888 #endif /* INET */
1889
1890 case SO_LINGER:
1891 l.l_onoff = so->so_options & SO_LINGER;
1892 l.l_linger = so->so_linger;
1893 error = sooptcopyout(sopt, &l, sizeof l);
1894 break;
1895
1896 case SO_USELOOPBACK:
1897 case SO_DONTROUTE:
1898 case SO_DEBUG:
1899 case SO_KEEPALIVE:
1900 case SO_REUSEADDR:
1901 case SO_REUSEPORT:
1902 case SO_BROADCAST:
1903 case SO_OOBINLINE:
1904 case SO_TIMESTAMP:
1905 optval = so->so_options & sopt->sopt_name;
1906 integer:
1907 error = sooptcopyout(sopt, &optval, sizeof optval);
1908 break;
1909
1910 case SO_TYPE:
1911 optval = so->so_type;
1912 goto integer;
1913
1914 case SO_ERROR:
1915 optval = so->so_error;
1916 so->so_error = 0;
1917 goto integer;
1918
1919 case SO_SNDBUF:
1920 optval = so->so_snd.ssb_hiwat;
1921 goto integer;
1922
1923 case SO_RCVBUF:
1924 optval = so->so_rcv.ssb_hiwat;
1925 goto integer;
1926
1927 case SO_SNDLOWAT:
1928 optval = so->so_snd.ssb_lowat;
1929 goto integer;
1930
1931 case SO_RCVLOWAT:
1932 optval = so->so_rcv.ssb_lowat;
1933 goto integer;
1934
1935 case SO_SNDTIMEO:
1936 case SO_RCVTIMEO:
1937 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1938 so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1939
1940 tv.tv_sec = optval / hz;
1941 tv.tv_usec = (optval % hz) * ustick;
1942 error = sooptcopyout(sopt, &tv, sizeof tv);
1943 break;
1944
1945 case SO_SNDSPACE:
1946 optval_l = ssb_space(&so->so_snd);
1947 error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
1948 break;
1949
1950 default:
1951 error = ENOPROTOOPT;
1952 break;
1953 }
1954 return (error);
1955 }
1956 }
1957
1958 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1959 int
1960 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1961 {
1962 struct mbuf *m, *m_prev;
1963 int sopt_size = sopt->sopt_valsize, msize;
1964
1965 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1966 0, &msize);
1967 if (m == NULL)
1968 return (ENOBUFS);
1969 m->m_len = min(msize, sopt_size);
1970 sopt_size -= m->m_len;
1971 *mp = m;
1972 m_prev = m;
1973
1974 while (sopt_size > 0) {
1975 m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1976 MT_DATA, 0, &msize);
1977 if (m == NULL) {
1978 m_freem(*mp);
1979 return (ENOBUFS);
1980 }
1981 m->m_len = min(msize, sopt_size);
1982 sopt_size -= m->m_len;
1983 m_prev->m_next = m;
1984 m_prev = m;
1985 }
1986 return (0);
1987 }
1988
1989 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1990 int
1991 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1992 {
1993 soopt_to_mbuf(sopt, m);
1994 return 0;
1995 }
1996
1997 void
1998 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1999 {
2000 size_t valsize;
2001 void *val;
2002
2003 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2004 KKASSERT(kva_p(m));
2005 if (sopt->sopt_val == NULL)
2006 return;
2007 val = sopt->sopt_val;
2008 valsize = sopt->sopt_valsize;
2009 while (m != NULL && valsize >= m->m_len) {
2010 bcopy(val, mtod(m, char *), m->m_len);
2011 valsize -= m->m_len;
2012 val = (caddr_t)val + m->m_len;
2013 m = m->m_next;
2014 }
2015 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2016 panic("ip6_sooptmcopyin");
2017 }
2018
2019 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2020 int
2021 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2022 {
2023 return soopt_from_mbuf(sopt, m);
2024 }
2025
2026 int
2027 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2028 {
2029 struct mbuf *m0 = m;
2030 size_t valsize = 0;
2031 size_t maxsize;
2032 void *val;
2033
2034 KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2035 KKASSERT(kva_p(m));
2036 if (sopt->sopt_val == NULL)
2037 return 0;
2038 val = sopt->sopt_val;
2039 maxsize = sopt->sopt_valsize;
2040 while (m != NULL && maxsize >= m->m_len) {
2041 bcopy(mtod(m, char *), val, m->m_len);
2042 maxsize -= m->m_len;
2043 val = (caddr_t)val + m->m_len;
2044 valsize += m->m_len;
2045 m = m->m_next;
2046 }
2047 if (m != NULL) {
2048 /* enough soopt buffer should be given from user-land */
2049 m_freem(m0);
2050 return (EINVAL);
2051 }
2052 sopt->sopt_valsize = valsize;
2053 return 0;
2054 }
2055
2056 void
2057 sohasoutofband(struct socket *so)
2058 {
2059 if (so->so_sigio != NULL)
2060 pgsigio(so->so_sigio, SIGURG, 0);
2061 KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2062 }
2063
2064 int
2065 sokqfilter(struct file *fp, struct knote *kn)
2066 {
2067 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2068 struct signalsockbuf *ssb;
2069
2070 switch (kn->kn_filter) {
2071 case EVFILT_READ:
2072 if (so->so_options & SO_ACCEPTCONN)
2073 kn->kn_fop = &solisten_filtops;
2074 else
2075 kn->kn_fop = &soread_filtops;
2076 ssb = &so->so_rcv;
2077 break;
2078 case EVFILT_WRITE:
2079 kn->kn_fop = &sowrite_filtops;
2080 ssb = &so->so_snd;
2081 break;
2082 case EVFILT_EXCEPT:
2083 kn->kn_fop = &soexcept_filtops;
2084 ssb = &so->so_rcv;
2085 break;
2086 default:
2087 return (EOPNOTSUPP);
2088 }
2089
2090 knote_insert(&ssb->ssb_kq.ki_note, kn);
2091 atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2092 return (0);
2093 }
2094
2095 static void
2096 filt_sordetach(struct knote *kn)
2097 {
2098 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2099
2100 knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2101 if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2102 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2103 }
2104
2105 /*ARGSUSED*/
2106 static int
2107 filt_soread(struct knote *kn, long hint)
2108 {
2109 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2110
2111 if (kn->kn_sfflags & NOTE_OOB) {
2112 if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2113 kn->kn_fflags |= NOTE_OOB;
2114 return (1);
2115 }
2116 return (0);
2117 }
2118 kn->kn_data = so->so_rcv.ssb_cc;
2119
2120 if (so->so_state & SS_CANTRCVMORE) {
2121 /*
2122 * Only set NODATA if all data has been exhausted.
2123 */
2124 if (kn->kn_data == 0)
2125 kn->kn_flags |= EV_NODATA;
2126 kn->kn_flags |= EV_EOF;
2127 kn->kn_fflags = so->so_error;
2128 return (1);
2129 }
2130 if (so->so_error) /* temporary udp error */
2131 return (1);
2132 if (kn->kn_sfflags & NOTE_LOWAT)
2133 return (kn->kn_data >= kn->kn_sdata);
2134 return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2135 !TAILQ_EMPTY(&so->so_comp));
2136 }
2137
2138 static void
2139 filt_sowdetach(struct knote *kn)
2140 {
2141 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2142
2143 knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2144 if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2145 atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2146 }
2147
2148 /*ARGSUSED*/
2149 static int
2150 filt_sowrite(struct knote *kn, long hint)
2151 {
2152 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2153
2154 kn->kn_data = ssb_space(&so->so_snd);
2155 if (so->so_state & SS_CANTSENDMORE) {
2156 kn->kn_flags |= (EV_EOF | EV_NODATA);
2157 kn->kn_fflags = so->so_error;
2158 return (1);
2159 }
2160 if (so->so_error) /* temporary udp error */
2161 return (1);
2162 if (((so->so_state & SS_ISCONNECTED) == 0) &&
2163 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2164 return (0);
2165 if (kn->kn_sfflags & NOTE_LOWAT)
2166 return (kn->kn_data >= kn->kn_sdata);
2167 return (kn->kn_data >= so->so_snd.ssb_lowat);
2168 }
2169
2170 /*ARGSUSED*/
2171 static int
2172 filt_solisten(struct knote *kn, long hint)
2173 {
2174 struct socket *so = (struct socket *)kn->kn_fp->f_data;
2175
2176 kn->kn_data = so->so_qlen;
2177 return (! TAILQ_EMPTY(&so->so_comp));
2178 }
DragonFly BSD