1 /*-
2 * Copyright (c) 2004 The FreeBSD Foundation
3 * Copyright (c) 2004-2005 Robert N. M. Watson
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD: releng/6.1/sys/kern/uipc_socket.c 158179 2006-04-30 16:44:43Z cvs2svn $");
36
37 #include "opt_inet.h"
38 #include "opt_mac.h"
39 #include "opt_zero.h"
40 #include "opt_compat.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/fcntl.h>
45 #include <sys/limits.h>
46 #include <sys/lock.h>
47 #include <sys/mac.h>
48 #include <sys/malloc.h>
49 #include <sys/mbuf.h>
50 #include <sys/mutex.h>
51 #include <sys/domain.h>
52 #include <sys/file.h> /* for struct knote */
53 #include <sys/kernel.h>
54 #include <sys/event.h>
55 #include <sys/poll.h>
56 #include <sys/proc.h>
57 #include <sys/protosw.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/resourcevar.h>
61 #include <sys/signalvar.h>
62 #include <sys/sysctl.h>
63 #include <sys/uio.h>
64 #include <sys/jail.h>
65
66 #include <vm/uma.h>
67
68 #ifdef COMPAT_IA32
69 #include <sys/mount.h>
70 #include <compat/freebsd32/freebsd32.h>
71
72 extern struct sysentvec ia32_freebsd_sysvec;
73 #endif
74
75 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
76 int flags);
77
78 static void filt_sordetach(struct knote *kn);
79 static int filt_soread(struct knote *kn, long hint);
80 static void filt_sowdetach(struct knote *kn);
81 static int filt_sowrite(struct knote *kn, long hint);
82 static int filt_solisten(struct knote *kn, long hint);
83
84 static struct filterops solisten_filtops =
85 { 1, NULL, filt_sordetach, filt_solisten };
86 static struct filterops soread_filtops =
87 { 1, NULL, filt_sordetach, filt_soread };
88 static struct filterops sowrite_filtops =
89 { 1, NULL, filt_sowdetach, filt_sowrite };
90
91 uma_zone_t socket_zone;
92 so_gen_t so_gencnt; /* generation count for sockets */
93
94 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
95 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
96
97 SYSCTL_DECL(_kern_ipc);
98
99 static int somaxconn = SOMAXCONN;
100 static int somaxconn_sysctl(SYSCTL_HANDLER_ARGS);
101 /* XXX: we dont have SYSCTL_USHORT */
102 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
103 0, sizeof(int), somaxconn_sysctl, "I", "Maximum pending socket connection "
104 "queue size");
105 static int numopensockets;
106 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
107 &numopensockets, 0, "Number of open sockets");
108 #ifdef ZERO_COPY_SOCKETS
109 /* These aren't static because they're used in other files. */
110 int so_zero_copy_send = 1;
111 int so_zero_copy_receive = 1;
112 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
113 "Zero copy controls");
114 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
115 &so_zero_copy_receive, 0, "Enable zero copy receive");
116 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
117 &so_zero_copy_send, 0, "Enable zero copy send");
118 #endif /* ZERO_COPY_SOCKETS */
119
120 /*
121 * accept_mtx locks down per-socket fields relating to accept queues. See
122 * socketvar.h for an annotation of the protected fields of struct socket.
123 */
124 struct mtx accept_mtx;
125 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
126
127 /*
128 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
129 * so_gencnt field.
130 */
131 static struct mtx so_global_mtx;
132 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
133
134 /*
135 * Socket operation routines.
136 * These routines are called by the routines in
137 * sys_socket.c or from a system process, and
138 * implement the semantics of socket operations by
139 * switching out to the protocol specific routines.
140 */
141
142 /*
143 * Get a socket structure from our zone, and initialize it.
144 * Note that it would probably be better to allocate socket
145 * and PCB at the same time, but I'm not convinced that all
146 * the protocols can be easily modified to do this.
147 *
148 * soalloc() returns a socket with a ref count of 0.
149 */
150 struct socket *
151 soalloc(int mflags)
152 {
153 struct socket *so;
154
155 so = uma_zalloc(socket_zone, mflags | M_ZERO);
156 if (so != NULL) {
157 #ifdef MAC
158 if (mac_init_socket(so, mflags) != 0) {
159 uma_zfree(socket_zone, so);
160 return (NULL);
161 }
162 #endif
163 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
164 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
165 TAILQ_INIT(&so->so_aiojobq);
166 mtx_lock(&so_global_mtx);
167 so->so_gencnt = ++so_gencnt;
168 ++numopensockets;
169 mtx_unlock(&so_global_mtx);
170 }
171 return (so);
172 }
173
174 /*
175 * socreate returns a socket with a ref count of 1. The socket should be
176 * closed with soclose().
177 */
178 int
179 socreate(dom, aso, type, proto, cred, td)
180 int dom;
181 struct socket **aso;
182 int type;
183 int proto;
184 struct ucred *cred;
185 struct thread *td;
186 {
187 struct protosw *prp;
188 struct socket *so;
189 int error;
190
191 if (proto)
192 prp = pffindproto(dom, proto, type);
193 else
194 prp = pffindtype(dom, type);
195
196 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
197 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
198 return (EPROTONOSUPPORT);
199
200 if (jailed(cred) && jail_socket_unixiproute_only &&
201 prp->pr_domain->dom_family != PF_LOCAL &&
202 prp->pr_domain->dom_family != PF_INET &&
203 prp->pr_domain->dom_family != PF_ROUTE) {
204 return (EPROTONOSUPPORT);
205 }
206
207 if (prp->pr_type != type)
208 return (EPROTOTYPE);
209 so = soalloc(M_WAITOK);
210 if (so == NULL)
211 return (ENOBUFS);
212
213 TAILQ_INIT(&so->so_incomp);
214 TAILQ_INIT(&so->so_comp);
215 so->so_type = type;
216 so->so_cred = crhold(cred);
217 so->so_proto = prp;
218 #ifdef MAC
219 mac_create_socket(cred, so);
220 #endif
221 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
222 NULL, NULL, NULL);
223 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
224 NULL, NULL, NULL);
225 so->so_count = 1;
226 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
227 if (error) {
228 ACCEPT_LOCK();
229 SOCK_LOCK(so);
230 so->so_state |= SS_NOFDREF;
231 sorele(so);
232 return (error);
233 }
234 *aso = so;
235 return (0);
236 }
237
238 int
239 sobind(so, nam, td)
240 struct socket *so;
241 struct sockaddr *nam;
242 struct thread *td;
243 {
244
245 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
246 }
247
248 void
249 sodealloc(struct socket *so)
250 {
251
252 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
253 mtx_lock(&so_global_mtx);
254 so->so_gencnt = ++so_gencnt;
255 mtx_unlock(&so_global_mtx);
256 if (so->so_rcv.sb_hiwat)
257 (void)chgsbsize(so->so_cred->cr_uidinfo,
258 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
259 if (so->so_snd.sb_hiwat)
260 (void)chgsbsize(so->so_cred->cr_uidinfo,
261 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
262 #ifdef INET
263 /* remove acccept filter if one is present. */
264 if (so->so_accf != NULL)
265 do_setopt_accept_filter(so, NULL);
266 #endif
267 #ifdef MAC
268 mac_destroy_socket(so);
269 #endif
270 crfree(so->so_cred);
271 SOCKBUF_LOCK_DESTROY(&so->so_snd);
272 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
273 uma_zfree(socket_zone, so);
274 mtx_lock(&so_global_mtx);
275 --numopensockets;
276 mtx_unlock(&so_global_mtx);
277 }
278
279 /*
280 * solisten() transitions a socket from a non-listening state to a listening
281 * state, but can also be used to update the listen queue depth on an
282 * existing listen socket. The protocol will call back into the sockets
283 * layer using solisten_proto_check() and solisten_proto() to check and set
284 * socket-layer listen state. Call backs are used so that the protocol can
285 * acquire both protocol and socket layer locks in whatever order is required
286 * by the protocol.
287 *
288 * Protocol implementors are advised to hold the socket lock across the
289 * socket-layer test and set to avoid races at the socket layer.
290 */
291 int
292 solisten(so, backlog, td)
293 struct socket *so;
294 int backlog;
295 struct thread *td;
296 {
297 int error;
298
299 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td);
300 if (error)
301 return (error);
302
303 /*
304 * XXXRW: The following state adjustment should occur in
305 * solisten_proto(), but we don't currently pass the backlog request
306 * to the protocol via pru_listen().
307 */
308 if (backlog < 0 || backlog > somaxconn)
309 backlog = somaxconn;
310 so->so_qlimit = backlog;
311 return (0);
312 }
313
314 int
315 solisten_proto_check(so)
316 struct socket *so;
317 {
318
319 SOCK_LOCK_ASSERT(so);
320
321 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
322 SS_ISDISCONNECTING))
323 return (EINVAL);
324 return (0);
325 }
326
327 void
328 solisten_proto(so)
329 struct socket *so;
330 {
331
332 SOCK_LOCK_ASSERT(so);
333
334 so->so_options |= SO_ACCEPTCONN;
335 }
336
337 /*
338 * Attempt to free a socket. This should really be sotryfree().
339 *
340 * We free the socket if the protocol is no longer interested in the socket,
341 * there's no file descriptor reference, and the refcount is 0. While the
342 * calling macro sotryfree() tests the refcount, sofree() has to test it
343 * again as it's possible to race with an accept()ing thread if the socket is
344 * in an listen queue of a listen socket, as being in the listen queue
345 * doesn't elevate the reference count. sofree() acquires the accept mutex
346 * early for this test in order to avoid that race.
347 */
348 void
349 sofree(so)
350 struct socket *so;
351 {
352 struct socket *head;
353
354 ACCEPT_LOCK_ASSERT();
355 SOCK_LOCK_ASSERT(so);
356
357 if (so->so_pcb != NULL || (so->so_state & SS_NOFDREF) == 0 ||
358 so->so_count != 0) {
359 SOCK_UNLOCK(so);
360 ACCEPT_UNLOCK();
361 return;
362 }
363
364 head = so->so_head;
365 if (head != NULL) {
366 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
367 (so->so_qstate & SQ_INCOMP) != 0,
368 ("sofree: so_head != NULL, but neither SQ_COMP nor "
369 "SQ_INCOMP"));
370 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
371 (so->so_qstate & SQ_INCOMP) == 0,
372 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
373 /*
374 * accept(2) is responsible draining the completed
375 * connection queue and freeing those sockets, so
376 * we just return here if this socket is currently
377 * on the completed connection queue. Otherwise,
378 * accept(2) may hang after select(2) has indicating
379 * that a listening socket was ready. If it's an
380 * incomplete connection, we remove it from the queue
381 * and free it; otherwise, it won't be released until
382 * the listening socket is closed.
383 */
384 if ((so->so_qstate & SQ_COMP) != 0) {
385 SOCK_UNLOCK(so);
386 ACCEPT_UNLOCK();
387 return;
388 }
389 TAILQ_REMOVE(&head->so_incomp, so, so_list);
390 head->so_incqlen--;
391 so->so_qstate &= ~SQ_INCOMP;
392 so->so_head = NULL;
393 }
394 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
395 (so->so_qstate & SQ_INCOMP) == 0,
396 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
397 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
398 SOCK_UNLOCK(so);
399 ACCEPT_UNLOCK();
400 SOCKBUF_LOCK(&so->so_snd);
401 so->so_snd.sb_flags |= SB_NOINTR;
402 (void)sblock(&so->so_snd, M_WAITOK);
403 /*
404 * socantsendmore_locked() drops the socket buffer mutex so that it
405 * can safely perform wakeups. Re-acquire the mutex before
406 * continuing.
407 */
408 socantsendmore_locked(so);
409 SOCKBUF_LOCK(&so->so_snd);
410 sbunlock(&so->so_snd);
411 sbrelease_locked(&so->so_snd, so);
412 SOCKBUF_UNLOCK(&so->so_snd);
413 sorflush(so);
414 knlist_destroy(&so->so_rcv.sb_sel.si_note);
415 knlist_destroy(&so->so_snd.sb_sel.si_note);
416 sodealloc(so);
417 }
418
419 /*
420 * Close a socket on last file table reference removal.
421 * Initiate disconnect if connected.
422 * Free socket when disconnect complete.
423 *
424 * This function will sorele() the socket. Note that soclose() may be
425 * called prior to the ref count reaching zero. The actual socket
426 * structure will not be freed until the ref count reaches zero.
427 */
428 int
429 soclose(so)
430 struct socket *so;
431 {
432 int error = 0;
433
434 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
435
436 funsetown(&so->so_sigio);
437 if (so->so_options & SO_ACCEPTCONN) {
438 struct socket *sp;
439 ACCEPT_LOCK();
440 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
441 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
442 so->so_incqlen--;
443 sp->so_qstate &= ~SQ_INCOMP;
444 sp->so_head = NULL;
445 ACCEPT_UNLOCK();
446 (void) soabort(sp);
447 ACCEPT_LOCK();
448 }
449 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
450 TAILQ_REMOVE(&so->so_comp, sp, so_list);
451 so->so_qlen--;
452 sp->so_qstate &= ~SQ_COMP;
453 sp->so_head = NULL;
454 ACCEPT_UNLOCK();
455 (void) soabort(sp);
456 ACCEPT_LOCK();
457 }
458 ACCEPT_UNLOCK();
459 }
460 if (so->so_pcb == NULL)
461 goto discard;
462 if (so->so_state & SS_ISCONNECTED) {
463 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
464 error = sodisconnect(so);
465 if (error)
466 goto drop;
467 }
468 if (so->so_options & SO_LINGER) {
469 if ((so->so_state & SS_ISDISCONNECTING) &&
470 (so->so_state & SS_NBIO))
471 goto drop;
472 while (so->so_state & SS_ISCONNECTED) {
473 error = tsleep(&so->so_timeo,
474 PSOCK | PCATCH, "soclos", so->so_linger * hz);
475 if (error)
476 break;
477 }
478 }
479 }
480 drop:
481 if (so->so_pcb != NULL) {
482 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
483 if (error == 0)
484 error = error2;
485 }
486 discard:
487 ACCEPT_LOCK();
488 SOCK_LOCK(so);
489 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
490 so->so_state |= SS_NOFDREF;
491 sorele(so);
492 return (error);
493 }
494
495 /*
496 * soabort() must not be called with any socket locks held, as it calls
497 * into the protocol, which will call back into the socket code causing
498 * it to acquire additional socket locks that may cause recursion or lock
499 * order reversals.
500 */
501 int
502 soabort(so)
503 struct socket *so;
504 {
505 int error;
506
507 error = (*so->so_proto->pr_usrreqs->pru_abort)(so);
508 if (error) {
509 ACCEPT_LOCK();
510 SOCK_LOCK(so);
511 sotryfree(so); /* note: does not decrement the ref count */
512 return error;
513 }
514 return (0);
515 }
516
517 int
518 soaccept(so, nam)
519 struct socket *so;
520 struct sockaddr **nam;
521 {
522 int error;
523
524 SOCK_LOCK(so);
525 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
526 so->so_state &= ~SS_NOFDREF;
527 SOCK_UNLOCK(so);
528 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
529 return (error);
530 }
531
532 int
533 soconnect(so, nam, td)
534 struct socket *so;
535 struct sockaddr *nam;
536 struct thread *td;
537 {
538 int error;
539
540 if (so->so_options & SO_ACCEPTCONN)
541 return (EOPNOTSUPP);
542 /*
543 * If protocol is connection-based, can only connect once.
544 * Otherwise, if connected, try to disconnect first.
545 * This allows user to disconnect by connecting to, e.g.,
546 * a null address.
547 */
548 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
549 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
550 (error = sodisconnect(so)))) {
551 error = EISCONN;
552 } else {
553 /*
554 * Prevent accumulated error from previous connection
555 * from biting us.
556 */
557 so->so_error = 0;
558 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
559 }
560
561 return (error);
562 }
563
564 int
565 soconnect2(so1, so2)
566 struct socket *so1;
567 struct socket *so2;
568 {
569
570 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
571 }
572
573 int
574 sodisconnect(so)
575 struct socket *so;
576 {
577 int error;
578
579 if ((so->so_state & SS_ISCONNECTED) == 0)
580 return (ENOTCONN);
581 if (so->so_state & SS_ISDISCONNECTING)
582 return (EALREADY);
583 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
584 return (error);
585 }
586
587 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
588 /*
589 * Send on a socket.
590 * If send must go all at once and message is larger than
591 * send buffering, then hard error.
592 * Lock against other senders.
593 * If must go all at once and not enough room now, then
594 * inform user that this would block and do nothing.
595 * Otherwise, if nonblocking, send as much as possible.
596 * The data to be sent is described by "uio" if nonzero,
597 * otherwise by the mbuf chain "top" (which must be null
598 * if uio is not). Data provided in mbuf chain must be small
599 * enough to send all at once.
600 *
601 * Returns nonzero on error, timeout or signal; callers
602 * must check for short counts if EINTR/ERESTART are returned.
603 * Data and control buffers are freed on return.
604 */
605
606 #ifdef ZERO_COPY_SOCKETS
607 struct so_zerocopy_stats{
608 int size_ok;
609 int align_ok;
610 int found_ifp;
611 };
612 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
613 #include <netinet/in.h>
614 #include <net/route.h>
615 #include <netinet/in_pcb.h>
616 #include <vm/vm.h>
617 #include <vm/vm_page.h>
618 #include <vm/vm_object.h>
619 #endif /*ZERO_COPY_SOCKETS*/
620
621 int
622 sosend(so, addr, uio, top, control, flags, td)
623 struct socket *so;
624 struct sockaddr *addr;
625 struct uio *uio;
626 struct mbuf *top;
627 struct mbuf *control;
628 int flags;
629 struct thread *td;
630 {
631 struct mbuf **mp;
632 struct mbuf *m;
633 long space, len = 0, resid;
634 int clen = 0, error, dontroute;
635 int atomic = sosendallatonce(so) || top;
636 #ifdef ZERO_COPY_SOCKETS
637 int cow_send;
638 #endif /* ZERO_COPY_SOCKETS */
639
640 if (uio != NULL)
641 resid = uio->uio_resid;
642 else
643 resid = top->m_pkthdr.len;
644 /*
645 * In theory resid should be unsigned.
646 * However, space must be signed, as it might be less than 0
647 * if we over-committed, and we must use a signed comparison
648 * of space and resid. On the other hand, a negative resid
649 * causes us to loop sending 0-length segments to the protocol.
650 *
651 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
652 * type sockets since that's an error.
653 */
654 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
655 error = EINVAL;
656 goto out;
657 }
658
659 dontroute =
660 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
661 (so->so_proto->pr_flags & PR_ATOMIC);
662 if (td != NULL)
663 td->td_proc->p_stats->p_ru.ru_msgsnd++;
664 if (control != NULL)
665 clen = control->m_len;
666 #define snderr(errno) { error = (errno); goto release; }
667
668 SOCKBUF_LOCK(&so->so_snd);
669 restart:
670 SOCKBUF_LOCK_ASSERT(&so->so_snd);
671 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
672 if (error)
673 goto out_locked;
674 do {
675 SOCKBUF_LOCK_ASSERT(&so->so_snd);
676 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
677 snderr(EPIPE);
678 if (so->so_error) {
679 error = so->so_error;
680 so->so_error = 0;
681 goto release;
682 }
683 if ((so->so_state & SS_ISCONNECTED) == 0) {
684 /*
685 * `sendto' and `sendmsg' is allowed on a connection-
686 * based socket if it supports implied connect.
687 * Return ENOTCONN if not connected and no address is
688 * supplied.
689 */
690 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
691 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
692 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
693 !(resid == 0 && clen != 0))
694 snderr(ENOTCONN);
695 } else if (addr == NULL)
696 snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
697 ENOTCONN : EDESTADDRREQ);
698 }
699 space = sbspace(&so->so_snd);
700 if (flags & MSG_OOB)
701 space += 1024;
702 if ((atomic && resid > so->so_snd.sb_hiwat) ||
703 clen > so->so_snd.sb_hiwat)
704 snderr(EMSGSIZE);
705 if (space < resid + clen &&
706 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
707 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO))
708 snderr(EWOULDBLOCK);
709 sbunlock(&so->so_snd);
710 error = sbwait(&so->so_snd);
711 if (error)
712 goto out_locked;
713 goto restart;
714 }
715 SOCKBUF_UNLOCK(&so->so_snd);
716 mp = ⊤
717 space -= clen;
718 do {
719 if (uio == NULL) {
720 /*
721 * Data is prepackaged in "top".
722 */
723 resid = 0;
724 if (flags & MSG_EOR)
725 top->m_flags |= M_EOR;
726 } else do {
727 #ifdef ZERO_COPY_SOCKETS
728 cow_send = 0;
729 #endif /* ZERO_COPY_SOCKETS */
730 if (resid >= MINCLSIZE) {
731 #ifdef ZERO_COPY_SOCKETS
732 if (top == NULL) {
733 MGETHDR(m, M_TRYWAIT, MT_DATA);
734 if (m == NULL) {
735 error = ENOBUFS;
736 SOCKBUF_LOCK(&so->so_snd);
737 goto release;
738 }
739 m->m_pkthdr.len = 0;
740 m->m_pkthdr.rcvif = NULL;
741 } else {
742 MGET(m, M_TRYWAIT, MT_DATA);
743 if (m == NULL) {
744 error = ENOBUFS;
745 SOCKBUF_LOCK(&so->so_snd);
746 goto release;
747 }
748 }
749 if (so_zero_copy_send &&
750 resid>=PAGE_SIZE &&
751 space>=PAGE_SIZE &&
752 uio->uio_iov->iov_len>=PAGE_SIZE) {
753 so_zerocp_stats.size_ok++;
754 so_zerocp_stats.align_ok++;
755 cow_send = socow_setup(m, uio);
756 len = cow_send;
757 }
758 if (!cow_send) {
759 MCLGET(m, M_TRYWAIT);
760 if ((m->m_flags & M_EXT) == 0) {
761 m_free(m);
762 m = NULL;
763 } else {
764 len = min(min(MCLBYTES, resid), space);
765 }
766 }
767 #else /* ZERO_COPY_SOCKETS */
768 if (top == NULL) {
769 m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
770 m->m_pkthdr.len = 0;
771 m->m_pkthdr.rcvif = NULL;
772 } else
773 m = m_getcl(M_TRYWAIT, MT_DATA, 0);
774 len = min(min(MCLBYTES, resid), space);
775 #endif /* ZERO_COPY_SOCKETS */
776 } else {
777 if (top == NULL) {
778 m = m_gethdr(M_TRYWAIT, MT_DATA);
779 m->m_pkthdr.len = 0;
780 m->m_pkthdr.rcvif = NULL;
781
782 len = min(min(MHLEN, resid), space);
783 /*
784 * For datagram protocols, leave room
785 * for protocol headers in first mbuf.
786 */
787 if (atomic && m && len < MHLEN)
788 MH_ALIGN(m, len);
789 } else {
790 m = m_get(M_TRYWAIT, MT_DATA);
791 len = min(min(MLEN, resid), space);
792 }
793 }
794 if (m == NULL) {
795 error = ENOBUFS;
796 SOCKBUF_LOCK(&so->so_snd);
797 goto release;
798 }
799
800 space -= len;
801 #ifdef ZERO_COPY_SOCKETS
802 if (cow_send)
803 error = 0;
804 else
805 #endif /* ZERO_COPY_SOCKETS */
806 error = uiomove(mtod(m, void *), (int)len, uio);
807 resid = uio->uio_resid;
808 m->m_len = len;
809 *mp = m;
810 top->m_pkthdr.len += len;
811 if (error) {
812 SOCKBUF_LOCK(&so->so_snd);
813 goto release;
814 }
815 mp = &m->m_next;
816 if (resid <= 0) {
817 if (flags & MSG_EOR)
818 top->m_flags |= M_EOR;
819 break;
820 }
821 } while (space > 0 && atomic);
822 if (dontroute) {
823 SOCK_LOCK(so);
824 so->so_options |= SO_DONTROUTE;
825 SOCK_UNLOCK(so);
826 }
827 /*
828 * XXX all the SBS_CANTSENDMORE checks previously
829 * done could be out of date. We could have recieved
830 * a reset packet in an interrupt or maybe we slept
831 * while doing page faults in uiomove() etc. We could
832 * probably recheck again inside the locking protection
833 * here, but there are probably other places that this
834 * also happens. We must rethink this.
835 */
836 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
837 (flags & MSG_OOB) ? PRUS_OOB :
838 /*
839 * If the user set MSG_EOF, the protocol
840 * understands this flag and nothing left to
841 * send then use PRU_SEND_EOF instead of PRU_SEND.
842 */
843 ((flags & MSG_EOF) &&
844 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
845 (resid <= 0)) ?
846 PRUS_EOF :
847 /* If there is more to send set PRUS_MORETOCOME */
848 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
849 top, addr, control, td);
850 if (dontroute) {
851 SOCK_LOCK(so);
852 so->so_options &= ~SO_DONTROUTE;
853 SOCK_UNLOCK(so);
854 }
855 clen = 0;
856 control = NULL;
857 top = NULL;
858 mp = ⊤
859 if (error) {
860 SOCKBUF_LOCK(&so->so_snd);
861 goto release;
862 }
863 } while (resid && space > 0);
864 SOCKBUF_LOCK(&so->so_snd);
865 } while (resid);
866
867 release:
868 SOCKBUF_LOCK_ASSERT(&so->so_snd);
869 sbunlock(&so->so_snd);
870 out_locked:
871 SOCKBUF_LOCK_ASSERT(&so->so_snd);
872 SOCKBUF_UNLOCK(&so->so_snd);
873 out:
874 if (top != NULL)
875 m_freem(top);
876 if (control != NULL)
877 m_freem(control);
878 return (error);
879 }
880
881 /*
882 * The part of soreceive() that implements reading non-inline out-of-band
883 * data from a socket. For more complete comments, see soreceive(), from
884 * which this code originated.
885 *
886 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
887 * unable to return an mbuf chain to the caller.
888 */
889 static int
890 soreceive_rcvoob(so, uio, flags)
891 struct socket *so;
892 struct uio *uio;
893 int flags;
894 {
895 struct protosw *pr = so->so_proto;
896 struct mbuf *m;
897 int error;
898
899 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
900
901 m = m_get(M_TRYWAIT, MT_DATA);
902 if (m == NULL)
903 return (ENOBUFS);
904 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
905 if (error)
906 goto bad;
907 do {
908 #ifdef ZERO_COPY_SOCKETS
909 if (so_zero_copy_receive) {
910 int disposable;
911
912 if ((m->m_flags & M_EXT)
913 && (m->m_ext.ext_type == EXT_DISPOSABLE))
914 disposable = 1;
915 else
916 disposable = 0;
917
918 error = uiomoveco(mtod(m, void *),
919 min(uio->uio_resid, m->m_len),
920 uio, disposable);
921 } else
922 #endif /* ZERO_COPY_SOCKETS */
923 error = uiomove(mtod(m, void *),
924 (int) min(uio->uio_resid, m->m_len), uio);
925 m = m_free(m);
926 } while (uio->uio_resid && error == 0 && m);
927 bad:
928 if (m != NULL)
929 m_freem(m);
930 return (error);
931 }
932
933 /*
934 * Following replacement or removal of the first mbuf on the first mbuf chain
935 * of a socket buffer, push necessary state changes back into the socket
936 * buffer so that other consumers see the values consistently. 'nextrecord'
937 * is the callers locally stored value of the original value of
938 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
939 * NOTE: 'nextrecord' may be NULL.
940 */
941 static __inline void
942 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
943 {
944
945 SOCKBUF_LOCK_ASSERT(sb);
946 /*
947 * First, update for the new value of nextrecord. If necessary, make
948 * it the first record.
949 */
950 if (sb->sb_mb != NULL)
951 sb->sb_mb->m_nextpkt = nextrecord;
952 else
953 sb->sb_mb = nextrecord;
954
955 /*
956 * Now update any dependent socket buffer fields to reflect the new
957 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
958 * addition of a second clause that takes care of the case where
959 * sb_mb has been updated, but remains the last record.
960 */
961 if (sb->sb_mb == NULL) {
962 sb->sb_mbtail = NULL;
963 sb->sb_lastrecord = NULL;
964 } else if (sb->sb_mb->m_nextpkt == NULL)
965 sb->sb_lastrecord = sb->sb_mb;
966 }
967
968
969 /*
970 * Implement receive operations on a socket.
971 * We depend on the way that records are added to the sockbuf
972 * by sbappend*. In particular, each record (mbufs linked through m_next)
973 * must begin with an address if the protocol so specifies,
974 * followed by an optional mbuf or mbufs containing ancillary data,
975 * and then zero or more mbufs of data.
976 * In order to avoid blocking network interrupts for the entire time here,
977 * we splx() while doing the actual copy to user space.
978 * Although the sockbuf is locked, new data may still be appended,
979 * and thus we must maintain consistency of the sockbuf during that time.
980 *
981 * The caller may receive the data as a single mbuf chain by supplying
982 * an mbuf **mp0 for use in returning the chain. The uio is then used
983 * only for the count in uio_resid.
984 */
985 int
986 soreceive(so, psa, uio, mp0, controlp, flagsp)
987 struct socket *so;
988 struct sockaddr **psa;
989 struct uio *uio;
990 struct mbuf **mp0;
991 struct mbuf **controlp;
992 int *flagsp;
993 {
994 struct mbuf *m, **mp;
995 int flags, len, error, offset;
996 struct protosw *pr = so->so_proto;
997 struct mbuf *nextrecord;
998 int moff, type = 0;
999 int orig_resid = uio->uio_resid;
1000
1001 mp = mp0;
1002 if (psa != NULL)
1003 *psa = NULL;
1004 if (controlp != NULL)
1005 *controlp = NULL;
1006 if (flagsp != NULL)
1007 flags = *flagsp &~ MSG_EOR;
1008 else
1009 flags = 0;
1010 if (flags & MSG_OOB)
1011 return (soreceive_rcvoob(so, uio, flags));
1012 if (mp != NULL)
1013 *mp = NULL;
1014 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1015 && uio->uio_resid)
1016 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1017
1018 SOCKBUF_LOCK(&so->so_rcv);
1019 restart:
1020 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1021 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1022 if (error)
1023 goto out;
1024
1025 m = so->so_rcv.sb_mb;
1026 /*
1027 * If we have less data than requested, block awaiting more
1028 * (subject to any timeout) if:
1029 * 1. the current count is less than the low water mark, or
1030 * 2. MSG_WAITALL is set, and it is possible to do the entire
1031 * receive operation at once if we block (resid <= hiwat).
1032 * 3. MSG_DONTWAIT is not set
1033 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1034 * we have to do the receive in sections, and thus risk returning
1035 * a short count if a timeout or signal occurs after we start.
1036 */
1037 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1038 so->so_rcv.sb_cc < uio->uio_resid) &&
1039 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1040 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1041 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1042 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1043 ("receive: m == %p so->so_rcv.sb_cc == %u",
1044 m, so->so_rcv.sb_cc));
1045 if (so->so_error) {
1046 if (m != NULL)
1047 goto dontblock;
1048 error = so->so_error;
1049 if ((flags & MSG_PEEK) == 0)
1050 so->so_error = 0;
1051 goto release;
1052 }
1053 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1054 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1055 if (m)
1056 goto dontblock;
1057 else
1058 goto release;
1059 }
1060 for (; m != NULL; m = m->m_next)
1061 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1062 m = so->so_rcv.sb_mb;
1063 goto dontblock;
1064 }
1065 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1066 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1067 error = ENOTCONN;
1068 goto release;
1069 }
1070 if (uio->uio_resid == 0)
1071 goto release;
1072 if ((so->so_state & SS_NBIO) ||
1073 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1074 error = EWOULDBLOCK;
1075 goto release;
1076 }
1077 SBLASTRECORDCHK(&so->so_rcv);
1078 SBLASTMBUFCHK(&so->so_rcv);
1079 sbunlock(&so->so_rcv);
1080 error = sbwait(&so->so_rcv);
1081 if (error)
1082 goto out;
1083 goto restart;
1084 }
1085 dontblock:
1086 /*
1087 * From this point onward, we maintain 'nextrecord' as a cache of the
1088 * pointer to the next record in the socket buffer. We must keep the
1089 * various socket buffer pointers and local stack versions of the
1090 * pointers in sync, pushing out modifications before dropping the
1091 * socket buffer mutex, and re-reading them when picking it up.
1092 *
1093 * Otherwise, we will race with the network stack appending new data
1094 * or records onto the socket buffer by using inconsistent/stale
1095 * versions of the field, possibly resulting in socket buffer
1096 * corruption.
1097 *
1098 * By holding the high-level sblock(), we prevent simultaneous
1099 * readers from pulling off the front of the socket buffer.
1100 */
1101 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1102 if (uio->uio_td)
1103 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
1104 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1105 SBLASTRECORDCHK(&so->so_rcv);
1106 SBLASTMBUFCHK(&so->so_rcv);
1107 nextrecord = m->m_nextpkt;
1108 if (pr->pr_flags & PR_ADDR) {
1109 KASSERT(m->m_type == MT_SONAME,
1110 ("m->m_type == %d", m->m_type));
1111 orig_resid = 0;
1112 if (psa != NULL)
1113 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1114 M_NOWAIT);
1115 if (flags & MSG_PEEK) {
1116 m = m->m_next;
1117 } else {
1118 sbfree(&so->so_rcv, m);
1119 so->so_rcv.sb_mb = m_free(m);
1120 m = so->so_rcv.sb_mb;
1121 sockbuf_pushsync(&so->so_rcv, nextrecord);
1122 }
1123 }
1124
1125 /*
1126 * Process one or more MT_CONTROL mbufs present before any data mbufs
1127 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1128 * just copy the data; if !MSG_PEEK, we call into the protocol to
1129 * perform externalization (or freeing if controlp == NULL).
1130 */
1131 if (m != NULL && m->m_type == MT_CONTROL) {
1132 struct mbuf *cm = NULL, *cmn;
1133 struct mbuf **cme = &cm;
1134
1135 do {
1136 if (flags & MSG_PEEK) {
1137 if (controlp != NULL) {
1138 *controlp = m_copy(m, 0, m->m_len);
1139 controlp = &(*controlp)->m_next;
1140 }
1141 m = m->m_next;
1142 } else {
1143 sbfree(&so->so_rcv, m);
1144 so->so_rcv.sb_mb = m->m_next;
1145 m->m_next = NULL;
1146 *cme = m;
1147 cme = &(*cme)->m_next;
1148 m = so->so_rcv.sb_mb;
1149 }
1150 } while (m != NULL && m->m_type == MT_CONTROL);
1151 if ((flags & MSG_PEEK) == 0)
1152 sockbuf_pushsync(&so->so_rcv, nextrecord);
1153 while (cm != NULL) {
1154 cmn = cm->m_next;
1155 cm->m_next = NULL;
1156 if (pr->pr_domain->dom_externalize != NULL) {
1157 SOCKBUF_UNLOCK(&so->so_rcv);
1158 error = (*pr->pr_domain->dom_externalize)
1159 (cm, controlp);
1160 SOCKBUF_LOCK(&so->so_rcv);
1161 } else if (controlp != NULL)
1162 *controlp = cm;
1163 else
1164 m_freem(cm);
1165 if (controlp != NULL) {
1166 orig_resid = 0;
1167 while (*controlp != NULL)
1168 controlp = &(*controlp)->m_next;
1169 }
1170 cm = cmn;
1171 }
1172 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1173 orig_resid = 0;
1174 }
1175 if (m != NULL) {
1176 if ((flags & MSG_PEEK) == 0) {
1177 KASSERT(m->m_nextpkt == nextrecord,
1178 ("soreceive: post-control, nextrecord !sync"));
1179 if (nextrecord == NULL) {
1180 KASSERT(so->so_rcv.sb_mb == m,
1181 ("soreceive: post-control, sb_mb!=m"));
1182 KASSERT(so->so_rcv.sb_lastrecord == m,
1183 ("soreceive: post-control, lastrecord!=m"));
1184 }
1185 }
1186 type = m->m_type;
1187 if (type == MT_OOBDATA)
1188 flags |= MSG_OOB;
1189 } else {
1190 if ((flags & MSG_PEEK) == 0) {
1191 KASSERT(so->so_rcv.sb_mb == nextrecord,
1192 ("soreceive: sb_mb != nextrecord"));
1193 if (so->so_rcv.sb_mb == NULL) {
1194 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1195 ("soreceive: sb_lastercord != NULL"));
1196 }
1197 }
1198 }
1199 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1200 SBLASTRECORDCHK(&so->so_rcv);
1201 SBLASTMBUFCHK(&so->so_rcv);
1202
1203 /*
1204 * Now continue to read any data mbufs off of the head of the socket
1205 * buffer until the read request is satisfied. Note that 'type' is
1206 * used to store the type of any mbuf reads that have happened so far
1207 * such that soreceive() can stop reading if the type changes, which
1208 * causes soreceive() to return only one of regular data and inline
1209 * out-of-band data in a single socket receive operation.
1210 */
1211 moff = 0;
1212 offset = 0;
1213 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1214 /*
1215 * If the type of mbuf has changed since the last mbuf
1216 * examined ('type'), end the receive operation.
1217 */
1218 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1219 if (m->m_type == MT_OOBDATA) {
1220 if (type != MT_OOBDATA)
1221 break;
1222 } else if (type == MT_OOBDATA)
1223 break;
1224 else
1225 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1226 ("m->m_type == %d", m->m_type));
1227 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1228 len = uio->uio_resid;
1229 if (so->so_oobmark && len > so->so_oobmark - offset)
1230 len = so->so_oobmark - offset;
1231 if (len > m->m_len - moff)
1232 len = m->m_len - moff;
1233 /*
1234 * If mp is set, just pass back the mbufs.
1235 * Otherwise copy them out via the uio, then free.
1236 * Sockbuf must be consistent here (points to current mbuf,
1237 * it points to next record) when we drop priority;
1238 * we must note any additions to the sockbuf when we
1239 * block interrupts again.
1240 */
1241 if (mp == NULL) {
1242 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1243 SBLASTRECORDCHK(&so->so_rcv);
1244 SBLASTMBUFCHK(&so->so_rcv);
1245 SOCKBUF_UNLOCK(&so->so_rcv);
1246 #ifdef ZERO_COPY_SOCKETS
1247 if (so_zero_copy_receive) {
1248 int disposable;
1249
1250 if ((m->m_flags & M_EXT)
1251 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1252 disposable = 1;
1253 else
1254 disposable = 0;
1255
1256 error = uiomoveco(mtod(m, char *) + moff,
1257 (int)len, uio,
1258 disposable);
1259 } else
1260 #endif /* ZERO_COPY_SOCKETS */
1261 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1262 SOCKBUF_LOCK(&so->so_rcv);
1263 if (error)
1264 goto release;
1265 } else
1266 uio->uio_resid -= len;
1267 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1268 if (len == m->m_len - moff) {
1269 if (m->m_flags & M_EOR)
1270 flags |= MSG_EOR;
1271 if (flags & MSG_PEEK) {
1272 m = m->m_next;
1273 moff = 0;
1274 } else {
1275 nextrecord = m->m_nextpkt;
1276 sbfree(&so->so_rcv, m);
1277 if (mp != NULL) {
1278 *mp = m;
1279 mp = &m->m_next;
1280 so->so_rcv.sb_mb = m = m->m_next;
1281 *mp = NULL;
1282 } else {
1283 so->so_rcv.sb_mb = m_free(m);
1284 m = so->so_rcv.sb_mb;
1285 }
1286 if (m != NULL) {
1287 m->m_nextpkt = nextrecord;
1288 if (nextrecord == NULL)
1289 so->so_rcv.sb_lastrecord = m;
1290 } else {
1291 so->so_rcv.sb_mb = nextrecord;
1292 SB_EMPTY_FIXUP(&so->so_rcv);
1293 }
1294 SBLASTRECORDCHK(&so->so_rcv);
1295 SBLASTMBUFCHK(&so->so_rcv);
1296 }
1297 } else {
1298 if (flags & MSG_PEEK)
1299 moff += len;
1300 else {
1301 if (mp != NULL) {
1302 int copy_flag;
1303
1304 if (flags & MSG_DONTWAIT)
1305 copy_flag = M_DONTWAIT;
1306 else
1307 copy_flag = M_TRYWAIT;
1308 if (copy_flag == M_TRYWAIT)
1309 SOCKBUF_UNLOCK(&so->so_rcv);
1310 *mp = m_copym(m, 0, len, copy_flag);
1311 if (copy_flag == M_TRYWAIT)
1312 SOCKBUF_LOCK(&so->so_rcv);
1313 if (*mp == NULL) {
1314 /*
1315 * m_copym() couldn't allocate an mbuf.
1316 * Adjust uio_resid back (it was adjusted
1317 * down by len bytes, which we didn't end
1318 * up "copying" over).
1319 */
1320 uio->uio_resid += len;
1321 break;
1322 }
1323 }
1324 m->m_data += len;
1325 m->m_len -= len;
1326 so->so_rcv.sb_cc -= len;
1327 }
1328 }
1329 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1330 if (so->so_oobmark) {
1331 if ((flags & MSG_PEEK) == 0) {
1332 so->so_oobmark -= len;
1333 if (so->so_oobmark == 0) {
1334 so->so_rcv.sb_state |= SBS_RCVATMARK;
1335 break;
1336 }
1337 } else {
1338 offset += len;
1339 if (offset == so->so_oobmark)
1340 break;
1341 }
1342 }
1343 if (flags & MSG_EOR)
1344 break;
1345 /*
1346 * If the MSG_WAITALL flag is set (for non-atomic socket),
1347 * we must not quit until "uio->uio_resid == 0" or an error
1348 * termination. If a signal/timeout occurs, return
1349 * with a short count but without error.
1350 * Keep sockbuf locked against other readers.
1351 */
1352 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1353 !sosendallatonce(so) && nextrecord == NULL) {
1354 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1355 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1356 break;
1357 /*
1358 * Notify the protocol that some data has been
1359 * drained before blocking.
1360 */
1361 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb != NULL) {
1362 SOCKBUF_UNLOCK(&so->so_rcv);
1363 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1364 SOCKBUF_LOCK(&so->so_rcv);
1365 }
1366 SBLASTRECORDCHK(&so->so_rcv);
1367 SBLASTMBUFCHK(&so->so_rcv);
1368 error = sbwait(&so->so_rcv);
1369 if (error)
1370 goto release;
1371 m = so->so_rcv.sb_mb;
1372 if (m != NULL)
1373 nextrecord = m->m_nextpkt;
1374 }
1375 }
1376
1377 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1378 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1379 flags |= MSG_TRUNC;
1380 if ((flags & MSG_PEEK) == 0)
1381 (void) sbdroprecord_locked(&so->so_rcv);
1382 }
1383 if ((flags & MSG_PEEK) == 0) {
1384 if (m == NULL) {
1385 /*
1386 * First part is an inline SB_EMPTY_FIXUP(). Second
1387 * part makes sure sb_lastrecord is up-to-date if
1388 * there is still data in the socket buffer.
1389 */
1390 so->so_rcv.sb_mb = nextrecord;
1391 if (so->so_rcv.sb_mb == NULL) {
1392 so->so_rcv.sb_mbtail = NULL;
1393 so->so_rcv.sb_lastrecord = NULL;
1394 } else if (nextrecord->m_nextpkt == NULL)
1395 so->so_rcv.sb_lastrecord = nextrecord;
1396 }
1397 SBLASTRECORDCHK(&so->so_rcv);
1398 SBLASTMBUFCHK(&so->so_rcv);
1399 /*
1400 * If soreceive() is being done from the socket callback, then
1401 * don't need to generate ACK to peer to update window, since
1402 * ACK will be generated on return to TCP.
1403 */
1404 if (!(flags & MSG_SOCALLBCK) &&
1405 (pr->pr_flags & PR_WANTRCVD) && so->so_pcb) {
1406 SOCKBUF_UNLOCK(&so->so_rcv);
1407 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1408 SOCKBUF_LOCK(&so->so_rcv);
1409 }
1410 }
1411 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1412 if (orig_resid == uio->uio_resid && orig_resid &&
1413 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1414 sbunlock(&so->so_rcv);
1415 goto restart;
1416 }
1417
1418 if (flagsp != NULL)
1419 *flagsp |= flags;
1420 release:
1421 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1422 sbunlock(&so->so_rcv);
1423 out:
1424 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1425 SOCKBUF_UNLOCK(&so->so_rcv);
1426 return (error);
1427 }
1428
1429 int
1430 soshutdown(so, how)
1431 struct socket *so;
1432 int how;
1433 {
1434 struct protosw *pr = so->so_proto;
1435
1436 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1437 return (EINVAL);
1438
1439 if (how != SHUT_WR)
1440 sorflush(so);
1441 if (how != SHUT_RD)
1442 return ((*pr->pr_usrreqs->pru_shutdown)(so));
1443 return (0);
1444 }
1445
1446 void
1447 sorflush(so)
1448 struct socket *so;
1449 {
1450 struct sockbuf *sb = &so->so_rcv;
1451 struct protosw *pr = so->so_proto;
1452 struct sockbuf asb;
1453
1454 /*
1455 * XXXRW: This is quite ugly. Previously, this code made a copy of
1456 * the socket buffer, then zero'd the original to clear the buffer
1457 * fields. However, with mutexes in the socket buffer, this causes
1458 * problems. We only clear the zeroable bits of the original;
1459 * however, we have to initialize and destroy the mutex in the copy
1460 * so that dom_dispose() and sbrelease() can lock t as needed.
1461 */
1462 SOCKBUF_LOCK(sb);
1463 sb->sb_flags |= SB_NOINTR;
1464 (void) sblock(sb, M_WAITOK);
1465 /*
1466 * socantrcvmore_locked() drops the socket buffer mutex so that it
1467 * can safely perform wakeups. Re-acquire the mutex before
1468 * continuing.
1469 */
1470 socantrcvmore_locked(so);
1471 SOCKBUF_LOCK(sb);
1472 sbunlock(sb);
1473 /*
1474 * Invalidate/clear most of the sockbuf structure, but leave
1475 * selinfo and mutex data unchanged.
1476 */
1477 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1478 bcopy(&sb->sb_startzero, &asb.sb_startzero,
1479 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1480 bzero(&sb->sb_startzero,
1481 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1482 SOCKBUF_UNLOCK(sb);
1483
1484 SOCKBUF_LOCK_INIT(&asb, "so_rcv");
1485 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1486 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1487 sbrelease(&asb, so);
1488 SOCKBUF_LOCK_DESTROY(&asb);
1489 }
1490
1491 /*
1492 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1493 * an additional variant to handle the case where the option value needs
1494 * to be some kind of integer, but not a specific size.
1495 * In addition to their use here, these functions are also called by the
1496 * protocol-level pr_ctloutput() routines.
1497 */
1498 int
1499 sooptcopyin(sopt, buf, len, minlen)
1500 struct sockopt *sopt;
1501 void *buf;
1502 size_t len;
1503 size_t minlen;
1504 {
1505 size_t valsize;
1506
1507 /*
1508 * If the user gives us more than we wanted, we ignore it,
1509 * but if we don't get the minimum length the caller
1510 * wants, we return EINVAL. On success, sopt->sopt_valsize
1511 * is set to however much we actually retrieved.
1512 */
1513 if ((valsize = sopt->sopt_valsize) < minlen)
1514 return EINVAL;
1515 if (valsize > len)
1516 sopt->sopt_valsize = valsize = len;
1517
1518 if (sopt->sopt_td != NULL)
1519 return (copyin(sopt->sopt_val, buf, valsize));
1520
1521 bcopy(sopt->sopt_val, buf, valsize);
1522 return 0;
1523 }
1524
1525 /*
1526 * Kernel version of setsockopt(2)/
1527 * XXX: optlen is size_t, not socklen_t
1528 */
1529 int
1530 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1531 size_t optlen)
1532 {
1533 struct sockopt sopt;
1534
1535 sopt.sopt_level = level;
1536 sopt.sopt_name = optname;
1537 sopt.sopt_dir = SOPT_SET;
1538 sopt.sopt_val = optval;
1539 sopt.sopt_valsize = optlen;
1540 sopt.sopt_td = NULL;
1541 return (sosetopt(so, &sopt));
1542 }
1543
1544 int
1545 sosetopt(so, sopt)
1546 struct socket *so;
1547 struct sockopt *sopt;
1548 {
1549 int error, optval;
1550 struct linger l;
1551 struct timeval tv;
1552 u_long val;
1553 #ifdef MAC
1554 struct mac extmac;
1555 #endif
1556
1557 error = 0;
1558 if (sopt->sopt_level != SOL_SOCKET) {
1559 if (so->so_proto && so->so_proto->pr_ctloutput)
1560 return ((*so->so_proto->pr_ctloutput)
1561 (so, sopt));
1562 error = ENOPROTOOPT;
1563 } else {
1564 switch (sopt->sopt_name) {
1565 #ifdef INET
1566 case SO_ACCEPTFILTER:
1567 error = do_setopt_accept_filter(so, sopt);
1568 if (error)
1569 goto bad;
1570 break;
1571 #endif
1572 case SO_LINGER:
1573 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1574 if (error)
1575 goto bad;
1576
1577 SOCK_LOCK(so);
1578 so->so_linger = l.l_linger;
1579 if (l.l_onoff)
1580 so->so_options |= SO_LINGER;
1581 else
1582 so->so_options &= ~SO_LINGER;
1583 SOCK_UNLOCK(so);
1584 break;
1585
1586 case SO_DEBUG:
1587 case SO_KEEPALIVE:
1588 case SO_DONTROUTE:
1589 case SO_USELOOPBACK:
1590 case SO_BROADCAST:
1591 case SO_REUSEADDR:
1592 case SO_REUSEPORT:
1593 case SO_OOBINLINE:
1594 case SO_TIMESTAMP:
1595 case SO_BINTIME:
1596 case SO_NOSIGPIPE:
1597 error = sooptcopyin(sopt, &optval, sizeof optval,
1598 sizeof optval);
1599 if (error)
1600 goto bad;
1601 SOCK_LOCK(so);
1602 if (optval)
1603 so->so_options |= sopt->sopt_name;
1604 else
1605 so->so_options &= ~sopt->sopt_name;
1606 SOCK_UNLOCK(so);
1607 break;
1608
1609 case SO_SNDBUF:
1610 case SO_RCVBUF:
1611 case SO_SNDLOWAT:
1612 case SO_RCVLOWAT:
1613 error = sooptcopyin(sopt, &optval, sizeof optval,
1614 sizeof optval);
1615 if (error)
1616 goto bad;
1617
1618 /*
1619 * Values < 1 make no sense for any of these
1620 * options, so disallow them.
1621 */
1622 if (optval < 1) {
1623 error = EINVAL;
1624 goto bad;
1625 }
1626
1627 switch (sopt->sopt_name) {
1628 case SO_SNDBUF:
1629 case SO_RCVBUF:
1630 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1631 &so->so_snd : &so->so_rcv, (u_long)optval,
1632 so, curthread) == 0) {
1633 error = ENOBUFS;
1634 goto bad;
1635 }
1636 break;
1637
1638 /*
1639 * Make sure the low-water is never greater than
1640 * the high-water.
1641 */
1642 case SO_SNDLOWAT:
1643 SOCKBUF_LOCK(&so->so_snd);
1644 so->so_snd.sb_lowat =
1645 (optval > so->so_snd.sb_hiwat) ?
1646 so->so_snd.sb_hiwat : optval;
1647 SOCKBUF_UNLOCK(&so->so_snd);
1648 break;
1649 case SO_RCVLOWAT:
1650 SOCKBUF_LOCK(&so->so_rcv);
1651 so->so_rcv.sb_lowat =
1652 (optval > so->so_rcv.sb_hiwat) ?
1653 so->so_rcv.sb_hiwat : optval;
1654 SOCKBUF_UNLOCK(&so->so_rcv);
1655 break;
1656 }
1657 break;
1658
1659 case SO_SNDTIMEO:
1660 case SO_RCVTIMEO:
1661 #ifdef COMPAT_IA32
1662 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
1663 struct timeval32 tv32;
1664
1665 error = sooptcopyin(sopt, &tv32, sizeof tv32,
1666 sizeof tv32);
1667 CP(tv32, tv, tv_sec);
1668 CP(tv32, tv, tv_usec);
1669 } else
1670 #endif
1671 error = sooptcopyin(sopt, &tv, sizeof tv,
1672 sizeof tv);
1673 if (error)
1674 goto bad;
1675
1676 /* assert(hz > 0); */
1677 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1678 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1679 error = EDOM;
1680 goto bad;
1681 }
1682 /* assert(tick > 0); */
1683 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
1684 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1685 if (val > INT_MAX) {
1686 error = EDOM;
1687 goto bad;
1688 }
1689 if (val == 0 && tv.tv_usec != 0)
1690 val = 1;
1691
1692 switch (sopt->sopt_name) {
1693 case SO_SNDTIMEO:
1694 so->so_snd.sb_timeo = val;
1695 break;
1696 case SO_RCVTIMEO:
1697 so->so_rcv.sb_timeo = val;
1698 break;
1699 }
1700 break;
1701
1702 case SO_LABEL:
1703 #ifdef MAC
1704 error = sooptcopyin(sopt, &extmac, sizeof extmac,
1705 sizeof extmac);
1706 if (error)
1707 goto bad;
1708 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
1709 so, &extmac);
1710 #else
1711 error = EOPNOTSUPP;
1712 #endif
1713 break;
1714
1715 default:
1716 error = ENOPROTOOPT;
1717 break;
1718 }
1719 if (error == 0 && so->so_proto != NULL &&
1720 so->so_proto->pr_ctloutput != NULL) {
1721 (void) ((*so->so_proto->pr_ctloutput)
1722 (so, sopt));
1723 }
1724 }
1725 bad:
1726 return (error);
1727 }
1728
1729 /* Helper routine for getsockopt */
1730 int
1731 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1732 {
1733 int error;
1734 size_t valsize;
1735
1736 error = 0;
1737
1738 /*
1739 * Documented get behavior is that we always return a value,
1740 * possibly truncated to fit in the user's buffer.
1741 * Traditional behavior is that we always tell the user
1742 * precisely how much we copied, rather than something useful
1743 * like the total amount we had available for her.
1744 * Note that this interface is not idempotent; the entire answer must
1745 * generated ahead of time.
1746 */
1747 valsize = min(len, sopt->sopt_valsize);
1748 sopt->sopt_valsize = valsize;
1749 if (sopt->sopt_val != NULL) {
1750 if (sopt->sopt_td != NULL)
1751 error = copyout(buf, sopt->sopt_val, valsize);
1752 else
1753 bcopy(buf, sopt->sopt_val, valsize);
1754 }
1755 return error;
1756 }
1757
1758 int
1759 sogetopt(so, sopt)
1760 struct socket *so;
1761 struct sockopt *sopt;
1762 {
1763 int error, optval;
1764 struct linger l;
1765 struct timeval tv;
1766 #ifdef MAC
1767 struct mac extmac;
1768 #endif
1769
1770 error = 0;
1771 if (sopt->sopt_level != SOL_SOCKET) {
1772 if (so->so_proto && so->so_proto->pr_ctloutput) {
1773 return ((*so->so_proto->pr_ctloutput)
1774 (so, sopt));
1775 } else
1776 return (ENOPROTOOPT);
1777 } else {
1778 switch (sopt->sopt_name) {
1779 #ifdef INET
1780 case SO_ACCEPTFILTER:
1781 error = do_getopt_accept_filter(so, sopt);
1782 break;
1783 #endif
1784 case SO_LINGER:
1785 SOCK_LOCK(so);
1786 l.l_onoff = so->so_options & SO_LINGER;
1787 l.l_linger = so->so_linger;
1788 SOCK_UNLOCK(so);
1789 error = sooptcopyout(sopt, &l, sizeof l);
1790 break;
1791
1792 case SO_USELOOPBACK:
1793 case SO_DONTROUTE:
1794 case SO_DEBUG:
1795 case SO_KEEPALIVE:
1796 case SO_REUSEADDR:
1797 case SO_REUSEPORT:
1798 case SO_BROADCAST:
1799 case SO_OOBINLINE:
1800 case SO_ACCEPTCONN:
1801 case SO_TIMESTAMP:
1802 case SO_BINTIME:
1803 case SO_NOSIGPIPE:
1804 optval = so->so_options & sopt->sopt_name;
1805 integer:
1806 error = sooptcopyout(sopt, &optval, sizeof optval);
1807 break;
1808
1809 case SO_TYPE:
1810 optval = so->so_type;
1811 goto integer;
1812
1813 case SO_ERROR:
1814 optval = so->so_error;
1815 so->so_error = 0;
1816 goto integer;
1817
1818 case SO_SNDBUF:
1819 optval = so->so_snd.sb_hiwat;
1820 goto integer;
1821
1822 case SO_RCVBUF:
1823 optval = so->so_rcv.sb_hiwat;
1824 goto integer;
1825
1826 case SO_SNDLOWAT:
1827 optval = so->so_snd.sb_lowat;
1828 goto integer;
1829
1830 case SO_RCVLOWAT:
1831 optval = so->so_rcv.sb_lowat;
1832 goto integer;
1833
1834 case SO_SNDTIMEO:
1835 case SO_RCVTIMEO:
1836 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1837 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1838
1839 tv.tv_sec = optval / hz;
1840 tv.tv_usec = (optval % hz) * tick;
1841 #ifdef COMPAT_IA32
1842 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
1843 struct timeval32 tv32;
1844
1845 CP(tv, tv32, tv_sec);
1846 CP(tv, tv32, tv_usec);
1847 error = sooptcopyout(sopt, &tv32, sizeof tv32);
1848 } else
1849 #endif
1850 error = sooptcopyout(sopt, &tv, sizeof tv);
1851 break;
1852
1853 case SO_LABEL:
1854 #ifdef MAC
1855 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
1856 sizeof(extmac));
1857 if (error)
1858 return (error);
1859 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
1860 so, &extmac);
1861 if (error)
1862 return (error);
1863 error = sooptcopyout(sopt, &extmac, sizeof extmac);
1864 #else
1865 error = EOPNOTSUPP;
1866 #endif
1867 break;
1868
1869 case SO_PEERLABEL:
1870 #ifdef MAC
1871 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
1872 sizeof(extmac));
1873 if (error)
1874 return (error);
1875 error = mac_getsockopt_peerlabel(
1876 sopt->sopt_td->td_ucred, so, &extmac);
1877 if (error)
1878 return (error);
1879 error = sooptcopyout(sopt, &extmac, sizeof extmac);
1880 #else
1881 error = EOPNOTSUPP;
1882 #endif
1883 break;
1884
1885 case SO_LISTENQLIMIT:
1886 optval = so->so_qlimit;
1887 goto integer;
1888
1889 case SO_LISTENQLEN:
1890 optval = so->so_qlen;
1891 goto integer;
1892
1893 case SO_LISTENINCQLEN:
1894 optval = so->so_incqlen;
1895 goto integer;
1896
1897 default:
1898 error = ENOPROTOOPT;
1899 break;
1900 }
1901 return (error);
1902 }
1903 }
1904
1905 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1906 int
1907 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1908 {
1909 struct mbuf *m, *m_prev;
1910 int sopt_size = sopt->sopt_valsize;
1911
1912 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1913 if (m == NULL)
1914 return ENOBUFS;
1915 if (sopt_size > MLEN) {
1916 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
1917 if ((m->m_flags & M_EXT) == 0) {
1918 m_free(m);
1919 return ENOBUFS;
1920 }
1921 m->m_len = min(MCLBYTES, sopt_size);
1922 } else {
1923 m->m_len = min(MLEN, sopt_size);
1924 }
1925 sopt_size -= m->m_len;
1926 *mp = m;
1927 m_prev = m;
1928
1929 while (sopt_size) {
1930 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1931 if (m == NULL) {
1932 m_freem(*mp);
1933 return ENOBUFS;
1934 }
1935 if (sopt_size > MLEN) {
1936 MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
1937 M_DONTWAIT);
1938 if ((m->m_flags & M_EXT) == 0) {
1939 m_freem(m);
1940 m_freem(*mp);
1941 return ENOBUFS;
1942 }
1943 m->m_len = min(MCLBYTES, sopt_size);
1944 } else {
1945 m->m_len = min(MLEN, sopt_size);
1946 }
1947 sopt_size -= m->m_len;
1948 m_prev->m_next = m;
1949 m_prev = m;
1950 }
1951 return 0;
1952 }
1953
1954 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1955 int
1956 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1957 {
1958 struct mbuf *m0 = m;
1959
1960 if (sopt->sopt_val == NULL)
1961 return 0;
1962 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1963 if (sopt->sopt_td != NULL) {
1964 int error;
1965
1966 error = copyin(sopt->sopt_val, mtod(m, char *),
1967 m->m_len);
1968 if (error != 0) {
1969 m_freem(m0);
1970 return(error);
1971 }
1972 } else
1973 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
1974 sopt->sopt_valsize -= m->m_len;
1975 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
1976 m = m->m_next;
1977 }
1978 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1979 panic("ip6_sooptmcopyin");
1980 return 0;
1981 }
1982
1983 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1984 int
1985 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1986 {
1987 struct mbuf *m0 = m;
1988 size_t valsize = 0;
1989
1990 if (sopt->sopt_val == NULL)
1991 return 0;
1992 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1993 if (sopt->sopt_td != NULL) {
1994 int error;
1995
1996 error = copyout(mtod(m, char *), sopt->sopt_val,
1997 m->m_len);
1998 if (error != 0) {
1999 m_freem(m0);
2000 return(error);
2001 }
2002 } else
2003 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2004 sopt->sopt_valsize -= m->m_len;
2005 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2006 valsize += m->m_len;
2007 m = m->m_next;
2008 }
2009 if (m != NULL) {
2010 /* enough soopt buffer should be given from user-land */
2011 m_freem(m0);
2012 return(EINVAL);
2013 }
2014 sopt->sopt_valsize = valsize;
2015 return 0;
2016 }
2017
2018 void
2019 sohasoutofband(so)
2020 struct socket *so;
2021 {
2022 if (so->so_sigio != NULL)
2023 pgsigio(&so->so_sigio, SIGURG, 0);
2024 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2025 }
2026
2027 int
2028 sopoll(struct socket *so, int events, struct ucred *active_cred,
2029 struct thread *td)
2030 {
2031 int revents = 0;
2032
2033 SOCKBUF_LOCK(&so->so_snd);
2034 SOCKBUF_LOCK(&so->so_rcv);
2035 if (events & (POLLIN | POLLRDNORM))
2036 if (soreadable(so))
2037 revents |= events & (POLLIN | POLLRDNORM);
2038
2039 if (events & POLLINIGNEOF)
2040 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2041 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2042 revents |= POLLINIGNEOF;
2043
2044 if (events & (POLLOUT | POLLWRNORM))
2045 if (sowriteable(so))
2046 revents |= events & (POLLOUT | POLLWRNORM);
2047
2048 if (events & (POLLPRI | POLLRDBAND))
2049 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2050 revents |= events & (POLLPRI | POLLRDBAND);
2051
2052 if (revents == 0) {
2053 if (events &
2054 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2055 POLLRDBAND)) {
2056 selrecord(td, &so->so_rcv.sb_sel);
2057 so->so_rcv.sb_flags |= SB_SEL;
2058 }
2059
2060 if (events & (POLLOUT | POLLWRNORM)) {
2061 selrecord(td, &so->so_snd.sb_sel);
2062 so->so_snd.sb_flags |= SB_SEL;
2063 }
2064 }
2065
2066 SOCKBUF_UNLOCK(&so->so_rcv);
2067 SOCKBUF_UNLOCK(&so->so_snd);
2068 return (revents);
2069 }
2070
2071 int
2072 soo_kqfilter(struct file *fp, struct knote *kn)
2073 {
2074 struct socket *so = kn->kn_fp->f_data;
2075 struct sockbuf *sb;
2076
2077 switch (kn->kn_filter) {
2078 case EVFILT_READ:
2079 if (so->so_options & SO_ACCEPTCONN)
2080 kn->kn_fop = &solisten_filtops;
2081 else
2082 kn->kn_fop = &soread_filtops;
2083 sb = &so->so_rcv;
2084 break;
2085 case EVFILT_WRITE:
2086 kn->kn_fop = &sowrite_filtops;
2087 sb = &so->so_snd;
2088 break;
2089 default:
2090 return (EINVAL);
2091 }
2092
2093 SOCKBUF_LOCK(sb);
2094 knlist_add(&sb->sb_sel.si_note, kn, 1);
2095 sb->sb_flags |= SB_KNOTE;
2096 SOCKBUF_UNLOCK(sb);
2097 return (0);
2098 }
2099
2100 static void
2101 filt_sordetach(struct knote *kn)
2102 {
2103 struct socket *so = kn->kn_fp->f_data;
2104
2105 SOCKBUF_LOCK(&so->so_rcv);
2106 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2107 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2108 so->so_rcv.sb_flags &= ~SB_KNOTE;
2109 SOCKBUF_UNLOCK(&so->so_rcv);
2110 }
2111
2112 /*ARGSUSED*/
2113 static int
2114 filt_soread(struct knote *kn, long hint)
2115 {
2116 struct socket *so;
2117
2118 so = kn->kn_fp->f_data;
2119 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2120
2121 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2122 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2123 kn->kn_flags |= EV_EOF;
2124 kn->kn_fflags = so->so_error;
2125 return (1);
2126 } else if (so->so_error) /* temporary udp error */
2127 return (1);
2128 else if (kn->kn_sfflags & NOTE_LOWAT)
2129 return (kn->kn_data >= kn->kn_sdata);
2130 else
2131 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2132 }
2133
2134 static void
2135 filt_sowdetach(struct knote *kn)
2136 {
2137 struct socket *so = kn->kn_fp->f_data;
2138
2139 SOCKBUF_LOCK(&so->so_snd);
2140 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2141 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2142 so->so_snd.sb_flags &= ~SB_KNOTE;
2143 SOCKBUF_UNLOCK(&so->so_snd);
2144 }
2145
2146 /*ARGSUSED*/
2147 static int
2148 filt_sowrite(struct knote *kn, long hint)
2149 {
2150 struct socket *so;
2151
2152 so = kn->kn_fp->f_data;
2153 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2154 kn->kn_data = sbspace(&so->so_snd);
2155 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2156 kn->kn_flags |= EV_EOF;
2157 kn->kn_fflags = so->so_error;
2158 return (1);
2159 } else if (so->so_error) /* temporary udp error */
2160 return (1);
2161 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2162 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2163 return (0);
2164 else if (kn->kn_sfflags & NOTE_LOWAT)
2165 return (kn->kn_data >= kn->kn_sdata);
2166 else
2167 return (kn->kn_data >= so->so_snd.sb_lowat);
2168 }
2169
2170 /*ARGSUSED*/
2171 static int
2172 filt_solisten(struct knote *kn, long hint)
2173 {
2174 struct socket *so = kn->kn_fp->f_data;
2175
2176 kn->kn_data = so->so_qlen;
2177 return (! TAILQ_EMPTY(&so->so_comp));
2178 }
2179
2180 int
2181 socheckuid(struct socket *so, uid_t uid)
2182 {
2183
2184 if (so == NULL)
2185 return (EPERM);
2186 if (so->so_cred->cr_uid != uid)
2187 return (EPERM);
2188 return (0);
2189 }
2190
2191 static int
2192 somaxconn_sysctl(SYSCTL_HANDLER_ARGS)
2193 {
2194 int error;
2195 int val;
2196
2197 val = somaxconn;
2198 error = sysctl_handle_int(oidp, &val, sizeof(int), req);
2199 if (error || !req->newptr )
2200 return (error);
2201
2202 if (val < 1 || val > USHRT_MAX)
2203 return (EINVAL);
2204
2205 somaxconn = val;
2206 return (0);
2207 }
Cache object: ebda32d1480b4faf7859778cdcd90158
|