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/6.4/sys/kern/uipc_socket.c 171938 2007-08-23 18:17:08Z jinmei $");
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(void)
152 {
153 struct socket *so;
154
155 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
156 if (so == NULL)
157 return (NULL);
158 #ifdef MAC
159 if (mac_init_socket(so, M_NOWAIT) != 0) {
160 uma_zfree(socket_zone, so);
161 return (NULL);
162 }
163 #endif
164 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
165 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
166 TAILQ_INIT(&so->so_aiojobq);
167 mtx_lock(&so_global_mtx);
168 so->so_gencnt = ++so_gencnt;
169 ++numopensockets;
170 mtx_unlock(&so_global_mtx);
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();
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_pcb == NULL)
438 goto discard;
439 if (so->so_state & SS_ISCONNECTED) {
440 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
441 error = sodisconnect(so);
442 if (error)
443 goto drop;
444 }
445 if (so->so_options & SO_LINGER) {
446 if ((so->so_state & SS_ISDISCONNECTING) &&
447 (so->so_state & SS_NBIO))
448 goto drop;
449 while (so->so_state & SS_ISCONNECTED) {
450 error = tsleep(&so->so_timeo,
451 PSOCK | PCATCH, "soclos", so->so_linger * hz);
452 if (error)
453 break;
454 }
455 }
456 }
457 drop:
458 if (so->so_pcb != NULL) {
459 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
460 if (error == 0)
461 error = error2;
462 }
463 if (so->so_options & SO_ACCEPTCONN) {
464 struct socket *sp;
465 ACCEPT_LOCK();
466 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
467 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
468 so->so_incqlen--;
469 sp->so_qstate &= ~SQ_INCOMP;
470 sp->so_head = NULL;
471 ACCEPT_UNLOCK();
472 (void) soabort(sp);
473 ACCEPT_LOCK();
474 }
475 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
476 TAILQ_REMOVE(&so->so_comp, sp, so_list);
477 so->so_qlen--;
478 sp->so_qstate &= ~SQ_COMP;
479 sp->so_head = NULL;
480 ACCEPT_UNLOCK();
481 (void) soabort(sp);
482 ACCEPT_LOCK();
483 }
484 ACCEPT_UNLOCK();
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 if (m != NULL)
1173 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1174 else
1175 nextrecord = so->so_rcv.sb_mb;
1176 orig_resid = 0;
1177 }
1178 if (m != NULL) {
1179 if ((flags & MSG_PEEK) == 0) {
1180 KASSERT(m->m_nextpkt == nextrecord,
1181 ("soreceive: post-control, nextrecord !sync"));
1182 if (nextrecord == NULL) {
1183 KASSERT(so->so_rcv.sb_mb == m,
1184 ("soreceive: post-control, sb_mb!=m"));
1185 KASSERT(so->so_rcv.sb_lastrecord == m,
1186 ("soreceive: post-control, lastrecord!=m"));
1187 }
1188 }
1189 type = m->m_type;
1190 if (type == MT_OOBDATA)
1191 flags |= MSG_OOB;
1192 } else {
1193 if ((flags & MSG_PEEK) == 0) {
1194 KASSERT(so->so_rcv.sb_mb == nextrecord,
1195 ("soreceive: sb_mb != nextrecord"));
1196 if (so->so_rcv.sb_mb == NULL) {
1197 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1198 ("soreceive: sb_lastercord != NULL"));
1199 }
1200 }
1201 }
1202 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1203 SBLASTRECORDCHK(&so->so_rcv);
1204 SBLASTMBUFCHK(&so->so_rcv);
1205
1206 /*
1207 * Now continue to read any data mbufs off of the head of the socket
1208 * buffer until the read request is satisfied. Note that 'type' is
1209 * used to store the type of any mbuf reads that have happened so far
1210 * such that soreceive() can stop reading if the type changes, which
1211 * causes soreceive() to return only one of regular data and inline
1212 * out-of-band data in a single socket receive operation.
1213 */
1214 moff = 0;
1215 offset = 0;
1216 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1217 /*
1218 * If the type of mbuf has changed since the last mbuf
1219 * examined ('type'), end the receive operation.
1220 */
1221 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1222 if (m->m_type == MT_OOBDATA) {
1223 if (type != MT_OOBDATA)
1224 break;
1225 } else if (type == MT_OOBDATA)
1226 break;
1227 else
1228 KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1229 ("m->m_type == %d", m->m_type));
1230 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1231 len = uio->uio_resid;
1232 if (so->so_oobmark && len > so->so_oobmark - offset)
1233 len = so->so_oobmark - offset;
1234 if (len > m->m_len - moff)
1235 len = m->m_len - moff;
1236 /*
1237 * If mp is set, just pass back the mbufs.
1238 * Otherwise copy them out via the uio, then free.
1239 * Sockbuf must be consistent here (points to current mbuf,
1240 * it points to next record) when we drop priority;
1241 * we must note any additions to the sockbuf when we
1242 * block interrupts again.
1243 */
1244 if (mp == NULL) {
1245 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1246 SBLASTRECORDCHK(&so->so_rcv);
1247 SBLASTMBUFCHK(&so->so_rcv);
1248 SOCKBUF_UNLOCK(&so->so_rcv);
1249 #ifdef ZERO_COPY_SOCKETS
1250 if (so_zero_copy_receive) {
1251 int disposable;
1252
1253 if ((m->m_flags & M_EXT)
1254 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1255 disposable = 1;
1256 else
1257 disposable = 0;
1258
1259 error = uiomoveco(mtod(m, char *) + moff,
1260 (int)len, uio,
1261 disposable);
1262 } else
1263 #endif /* ZERO_COPY_SOCKETS */
1264 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1265 SOCKBUF_LOCK(&so->so_rcv);
1266 if (error) {
1267 /*
1268 * If any part of the record has been removed
1269 * (such as the MT_SONAME mbuf, which will
1270 * happen when PR_ADDR, and thus also
1271 * PR_ATOMIC, is set), then drop the entire
1272 * record to maintain the atomicity of the
1273 * receive operation.
1274 */
1275 if (m && pr->pr_flags & PR_ATOMIC &&
1276 ((flags & MSG_PEEK) == 0))
1277 (void)sbdroprecord_locked(&so->so_rcv);
1278 goto release;
1279 }
1280 } else
1281 uio->uio_resid -= len;
1282 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1283 if (len == m->m_len - moff) {
1284 if (m->m_flags & M_EOR)
1285 flags |= MSG_EOR;
1286 if (flags & MSG_PEEK) {
1287 m = m->m_next;
1288 moff = 0;
1289 } else {
1290 nextrecord = m->m_nextpkt;
1291 sbfree(&so->so_rcv, m);
1292 if (mp != NULL) {
1293 *mp = m;
1294 mp = &m->m_next;
1295 so->so_rcv.sb_mb = m = m->m_next;
1296 *mp = NULL;
1297 } else {
1298 so->so_rcv.sb_mb = m_free(m);
1299 m = so->so_rcv.sb_mb;
1300 }
1301 if (m != NULL) {
1302 m->m_nextpkt = nextrecord;
1303 if (nextrecord == NULL)
1304 so->so_rcv.sb_lastrecord = m;
1305 } else {
1306 so->so_rcv.sb_mb = nextrecord;
1307 SB_EMPTY_FIXUP(&so->so_rcv);
1308 }
1309 SBLASTRECORDCHK(&so->so_rcv);
1310 SBLASTMBUFCHK(&so->so_rcv);
1311 }
1312 } else {
1313 if (flags & MSG_PEEK)
1314 moff += len;
1315 else {
1316 if (mp != NULL) {
1317 int copy_flag;
1318
1319 if (flags & MSG_DONTWAIT)
1320 copy_flag = M_DONTWAIT;
1321 else
1322 copy_flag = M_TRYWAIT;
1323 if (copy_flag == M_TRYWAIT)
1324 SOCKBUF_UNLOCK(&so->so_rcv);
1325 *mp = m_copym(m, 0, len, copy_flag);
1326 if (copy_flag == M_TRYWAIT)
1327 SOCKBUF_LOCK(&so->so_rcv);
1328 if (*mp == NULL) {
1329 /*
1330 * m_copym() couldn't allocate an mbuf.
1331 * Adjust uio_resid back (it was adjusted
1332 * down by len bytes, which we didn't end
1333 * up "copying" over).
1334 */
1335 uio->uio_resid += len;
1336 break;
1337 }
1338 }
1339 m->m_data += len;
1340 m->m_len -= len;
1341 so->so_rcv.sb_cc -= len;
1342 }
1343 }
1344 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1345 if (so->so_oobmark) {
1346 if ((flags & MSG_PEEK) == 0) {
1347 so->so_oobmark -= len;
1348 if (so->so_oobmark == 0) {
1349 so->so_rcv.sb_state |= SBS_RCVATMARK;
1350 break;
1351 }
1352 } else {
1353 offset += len;
1354 if (offset == so->so_oobmark)
1355 break;
1356 }
1357 }
1358 if (flags & MSG_EOR)
1359 break;
1360 /*
1361 * If the MSG_WAITALL flag is set (for non-atomic socket),
1362 * we must not quit until "uio->uio_resid == 0" or an error
1363 * termination. If a signal/timeout occurs, return
1364 * with a short count but without error.
1365 * Keep sockbuf locked against other readers.
1366 */
1367 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1368 !sosendallatonce(so) && nextrecord == NULL) {
1369 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1370 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1371 break;
1372 /*
1373 * Notify the protocol that some data has been
1374 * drained before blocking.
1375 */
1376 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb != NULL) {
1377 SOCKBUF_UNLOCK(&so->so_rcv);
1378 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1379 SOCKBUF_LOCK(&so->so_rcv);
1380 }
1381 SBLASTRECORDCHK(&so->so_rcv);
1382 SBLASTMBUFCHK(&so->so_rcv);
1383 error = sbwait(&so->so_rcv);
1384 if (error)
1385 goto release;
1386 m = so->so_rcv.sb_mb;
1387 if (m != NULL)
1388 nextrecord = m->m_nextpkt;
1389 }
1390 }
1391
1392 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1393 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1394 flags |= MSG_TRUNC;
1395 if ((flags & MSG_PEEK) == 0)
1396 (void) sbdroprecord_locked(&so->so_rcv);
1397 }
1398 if ((flags & MSG_PEEK) == 0) {
1399 if (m == NULL) {
1400 /*
1401 * First part is an inline SB_EMPTY_FIXUP(). Second
1402 * part makes sure sb_lastrecord is up-to-date if
1403 * there is still data in the socket buffer.
1404 */
1405 so->so_rcv.sb_mb = nextrecord;
1406 if (so->so_rcv.sb_mb == NULL) {
1407 so->so_rcv.sb_mbtail = NULL;
1408 so->so_rcv.sb_lastrecord = NULL;
1409 } else if (nextrecord->m_nextpkt == NULL)
1410 so->so_rcv.sb_lastrecord = nextrecord;
1411 }
1412 SBLASTRECORDCHK(&so->so_rcv);
1413 SBLASTMBUFCHK(&so->so_rcv);
1414 /*
1415 * If soreceive() is being done from the socket callback, then
1416 * don't need to generate ACK to peer to update window, since
1417 * ACK will be generated on return to TCP.
1418 */
1419 if (!(flags & MSG_SOCALLBCK) &&
1420 (pr->pr_flags & PR_WANTRCVD) && so->so_pcb) {
1421 SOCKBUF_UNLOCK(&so->so_rcv);
1422 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1423 SOCKBUF_LOCK(&so->so_rcv);
1424 }
1425 }
1426 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1427 if (orig_resid == uio->uio_resid && orig_resid &&
1428 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1429 sbunlock(&so->so_rcv);
1430 goto restart;
1431 }
1432
1433 if (flagsp != NULL)
1434 *flagsp |= flags;
1435 release:
1436 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1437 sbunlock(&so->so_rcv);
1438 out:
1439 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1440 SOCKBUF_UNLOCK(&so->so_rcv);
1441 return (error);
1442 }
1443
1444 int
1445 soshutdown(so, how)
1446 struct socket *so;
1447 int how;
1448 {
1449 struct protosw *pr = so->so_proto;
1450
1451 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1452 return (EINVAL);
1453
1454 if (how != SHUT_WR)
1455 sorflush(so);
1456 if (how != SHUT_RD)
1457 return ((*pr->pr_usrreqs->pru_shutdown)(so));
1458 return (0);
1459 }
1460
1461 void
1462 sorflush(so)
1463 struct socket *so;
1464 {
1465 struct sockbuf *sb = &so->so_rcv;
1466 struct protosw *pr = so->so_proto;
1467 struct sockbuf asb;
1468
1469 /*
1470 * XXXRW: This is quite ugly. Previously, this code made a copy of
1471 * the socket buffer, then zero'd the original to clear the buffer
1472 * fields. However, with mutexes in the socket buffer, this causes
1473 * problems. We only clear the zeroable bits of the original;
1474 * however, we have to initialize and destroy the mutex in the copy
1475 * so that dom_dispose() and sbrelease() can lock t as needed.
1476 */
1477 SOCKBUF_LOCK(sb);
1478 sb->sb_flags |= SB_NOINTR;
1479 (void) sblock(sb, M_WAITOK);
1480 /*
1481 * socantrcvmore_locked() drops the socket buffer mutex so that it
1482 * can safely perform wakeups. Re-acquire the mutex before
1483 * continuing.
1484 */
1485 socantrcvmore_locked(so);
1486 SOCKBUF_LOCK(sb);
1487 sbunlock(sb);
1488 /*
1489 * Invalidate/clear most of the sockbuf structure, but leave
1490 * selinfo and mutex data unchanged.
1491 */
1492 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1493 bcopy(&sb->sb_startzero, &asb.sb_startzero,
1494 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1495 bzero(&sb->sb_startzero,
1496 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1497 SOCKBUF_UNLOCK(sb);
1498
1499 SOCKBUF_LOCK_INIT(&asb, "so_rcv");
1500 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1501 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1502 sbrelease(&asb, so);
1503 SOCKBUF_LOCK_DESTROY(&asb);
1504 }
1505
1506 /*
1507 * Perhaps this routine, and sooptcopyout(), below, ought to come in
1508 * an additional variant to handle the case where the option value needs
1509 * to be some kind of integer, but not a specific size.
1510 * In addition to their use here, these functions are also called by the
1511 * protocol-level pr_ctloutput() routines.
1512 */
1513 int
1514 sooptcopyin(sopt, buf, len, minlen)
1515 struct sockopt *sopt;
1516 void *buf;
1517 size_t len;
1518 size_t minlen;
1519 {
1520 size_t valsize;
1521
1522 /*
1523 * If the user gives us more than we wanted, we ignore it,
1524 * but if we don't get the minimum length the caller
1525 * wants, we return EINVAL. On success, sopt->sopt_valsize
1526 * is set to however much we actually retrieved.
1527 */
1528 if ((valsize = sopt->sopt_valsize) < minlen)
1529 return EINVAL;
1530 if (valsize > len)
1531 sopt->sopt_valsize = valsize = len;
1532
1533 if (sopt->sopt_td != NULL)
1534 return (copyin(sopt->sopt_val, buf, valsize));
1535
1536 bcopy(sopt->sopt_val, buf, valsize);
1537 return 0;
1538 }
1539
1540 /*
1541 * Kernel version of setsockopt(2)/
1542 * XXX: optlen is size_t, not socklen_t
1543 */
1544 int
1545 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1546 size_t optlen)
1547 {
1548 struct sockopt sopt;
1549
1550 sopt.sopt_level = level;
1551 sopt.sopt_name = optname;
1552 sopt.sopt_dir = SOPT_SET;
1553 sopt.sopt_val = optval;
1554 sopt.sopt_valsize = optlen;
1555 sopt.sopt_td = NULL;
1556 return (sosetopt(so, &sopt));
1557 }
1558
1559 int
1560 sosetopt(so, sopt)
1561 struct socket *so;
1562 struct sockopt *sopt;
1563 {
1564 int error, optval;
1565 struct linger l;
1566 struct timeval tv;
1567 u_long val;
1568 #ifdef MAC
1569 struct mac extmac;
1570 #endif
1571
1572 error = 0;
1573 if (sopt->sopt_level != SOL_SOCKET) {
1574 if (so->so_proto && so->so_proto->pr_ctloutput)
1575 return ((*so->so_proto->pr_ctloutput)
1576 (so, sopt));
1577 error = ENOPROTOOPT;
1578 } else {
1579 switch (sopt->sopt_name) {
1580 #ifdef INET
1581 case SO_ACCEPTFILTER:
1582 error = do_setopt_accept_filter(so, sopt);
1583 if (error)
1584 goto bad;
1585 break;
1586 #endif
1587 case SO_LINGER:
1588 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1589 if (error)
1590 goto bad;
1591
1592 SOCK_LOCK(so);
1593 so->so_linger = l.l_linger;
1594 if (l.l_onoff)
1595 so->so_options |= SO_LINGER;
1596 else
1597 so->so_options &= ~SO_LINGER;
1598 SOCK_UNLOCK(so);
1599 break;
1600
1601 case SO_DEBUG:
1602 case SO_KEEPALIVE:
1603 case SO_DONTROUTE:
1604 case SO_USELOOPBACK:
1605 case SO_BROADCAST:
1606 case SO_REUSEADDR:
1607 case SO_REUSEPORT:
1608 case SO_OOBINLINE:
1609 case SO_TIMESTAMP:
1610 case SO_BINTIME:
1611 case SO_NOSIGPIPE:
1612 error = sooptcopyin(sopt, &optval, sizeof optval,
1613 sizeof optval);
1614 if (error)
1615 goto bad;
1616 SOCK_LOCK(so);
1617 if (optval)
1618 so->so_options |= sopt->sopt_name;
1619 else
1620 so->so_options &= ~sopt->sopt_name;
1621 SOCK_UNLOCK(so);
1622 break;
1623
1624 case SO_SNDBUF:
1625 case SO_RCVBUF:
1626 case SO_SNDLOWAT:
1627 case SO_RCVLOWAT:
1628 error = sooptcopyin(sopt, &optval, sizeof optval,
1629 sizeof optval);
1630 if (error)
1631 goto bad;
1632
1633 /*
1634 * Values < 1 make no sense for any of these
1635 * options, so disallow them.
1636 */
1637 if (optval < 1) {
1638 error = EINVAL;
1639 goto bad;
1640 }
1641
1642 switch (sopt->sopt_name) {
1643 case SO_SNDBUF:
1644 case SO_RCVBUF:
1645 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1646 &so->so_snd : &so->so_rcv, (u_long)optval,
1647 so, curthread) == 0) {
1648 error = ENOBUFS;
1649 goto bad;
1650 }
1651 break;
1652
1653 /*
1654 * Make sure the low-water is never greater than
1655 * the high-water.
1656 */
1657 case SO_SNDLOWAT:
1658 SOCKBUF_LOCK(&so->so_snd);
1659 so->so_snd.sb_lowat =
1660 (optval > so->so_snd.sb_hiwat) ?
1661 so->so_snd.sb_hiwat : optval;
1662 SOCKBUF_UNLOCK(&so->so_snd);
1663 break;
1664 case SO_RCVLOWAT:
1665 SOCKBUF_LOCK(&so->so_rcv);
1666 so->so_rcv.sb_lowat =
1667 (optval > so->so_rcv.sb_hiwat) ?
1668 so->so_rcv.sb_hiwat : optval;
1669 SOCKBUF_UNLOCK(&so->so_rcv);
1670 break;
1671 }
1672 break;
1673
1674 case SO_SNDTIMEO:
1675 case SO_RCVTIMEO:
1676 #ifdef COMPAT_IA32
1677 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
1678 struct timeval32 tv32;
1679
1680 error = sooptcopyin(sopt, &tv32, sizeof tv32,
1681 sizeof tv32);
1682 CP(tv32, tv, tv_sec);
1683 CP(tv32, tv, tv_usec);
1684 } else
1685 #endif
1686 error = sooptcopyin(sopt, &tv, sizeof tv,
1687 sizeof tv);
1688 if (error)
1689 goto bad;
1690
1691 /* assert(hz > 0); */
1692 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1693 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1694 error = EDOM;
1695 goto bad;
1696 }
1697 /* assert(tick > 0); */
1698 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
1699 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1700 if (val > INT_MAX) {
1701 error = EDOM;
1702 goto bad;
1703 }
1704 if (val == 0 && tv.tv_usec != 0)
1705 val = 1;
1706
1707 switch (sopt->sopt_name) {
1708 case SO_SNDTIMEO:
1709 so->so_snd.sb_timeo = val;
1710 break;
1711 case SO_RCVTIMEO:
1712 so->so_rcv.sb_timeo = val;
1713 break;
1714 }
1715 break;
1716
1717 case SO_LABEL:
1718 #ifdef MAC
1719 error = sooptcopyin(sopt, &extmac, sizeof extmac,
1720 sizeof extmac);
1721 if (error)
1722 goto bad;
1723 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
1724 so, &extmac);
1725 #else
1726 error = EOPNOTSUPP;
1727 #endif
1728 break;
1729
1730 default:
1731 error = ENOPROTOOPT;
1732 break;
1733 }
1734 if (error == 0 && so->so_proto != NULL &&
1735 so->so_proto->pr_ctloutput != NULL) {
1736 (void) ((*so->so_proto->pr_ctloutput)
1737 (so, sopt));
1738 }
1739 }
1740 bad:
1741 return (error);
1742 }
1743
1744 /* Helper routine for getsockopt */
1745 int
1746 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1747 {
1748 int error;
1749 size_t valsize;
1750
1751 error = 0;
1752
1753 /*
1754 * Documented get behavior is that we always return a value,
1755 * possibly truncated to fit in the user's buffer.
1756 * Traditional behavior is that we always tell the user
1757 * precisely how much we copied, rather than something useful
1758 * like the total amount we had available for her.
1759 * Note that this interface is not idempotent; the entire answer must
1760 * generated ahead of time.
1761 */
1762 valsize = min(len, sopt->sopt_valsize);
1763 sopt->sopt_valsize = valsize;
1764 if (sopt->sopt_val != NULL) {
1765 if (sopt->sopt_td != NULL)
1766 error = copyout(buf, sopt->sopt_val, valsize);
1767 else
1768 bcopy(buf, sopt->sopt_val, valsize);
1769 }
1770 return error;
1771 }
1772
1773 int
1774 sogetopt(so, sopt)
1775 struct socket *so;
1776 struct sockopt *sopt;
1777 {
1778 int error, optval;
1779 struct linger l;
1780 struct timeval tv;
1781 #ifdef MAC
1782 struct mac extmac;
1783 #endif
1784
1785 error = 0;
1786 if (sopt->sopt_level != SOL_SOCKET) {
1787 if (so->so_proto && so->so_proto->pr_ctloutput) {
1788 return ((*so->so_proto->pr_ctloutput)
1789 (so, sopt));
1790 } else
1791 return (ENOPROTOOPT);
1792 } else {
1793 switch (sopt->sopt_name) {
1794 #ifdef INET
1795 case SO_ACCEPTFILTER:
1796 error = do_getopt_accept_filter(so, sopt);
1797 break;
1798 #endif
1799 case SO_LINGER:
1800 SOCK_LOCK(so);
1801 l.l_onoff = so->so_options & SO_LINGER;
1802 l.l_linger = so->so_linger;
1803 SOCK_UNLOCK(so);
1804 error = sooptcopyout(sopt, &l, sizeof l);
1805 break;
1806
1807 case SO_USELOOPBACK:
1808 case SO_DONTROUTE:
1809 case SO_DEBUG:
1810 case SO_KEEPALIVE:
1811 case SO_REUSEADDR:
1812 case SO_REUSEPORT:
1813 case SO_BROADCAST:
1814 case SO_OOBINLINE:
1815 case SO_ACCEPTCONN:
1816 case SO_TIMESTAMP:
1817 case SO_BINTIME:
1818 case SO_NOSIGPIPE:
1819 optval = so->so_options & sopt->sopt_name;
1820 integer:
1821 error = sooptcopyout(sopt, &optval, sizeof optval);
1822 break;
1823
1824 case SO_TYPE:
1825 optval = so->so_type;
1826 goto integer;
1827
1828 case SO_ERROR:
1829 optval = so->so_error;
1830 so->so_error = 0;
1831 goto integer;
1832
1833 case SO_SNDBUF:
1834 optval = so->so_snd.sb_hiwat;
1835 goto integer;
1836
1837 case SO_RCVBUF:
1838 optval = so->so_rcv.sb_hiwat;
1839 goto integer;
1840
1841 case SO_SNDLOWAT:
1842 optval = so->so_snd.sb_lowat;
1843 goto integer;
1844
1845 case SO_RCVLOWAT:
1846 optval = so->so_rcv.sb_lowat;
1847 goto integer;
1848
1849 case SO_SNDTIMEO:
1850 case SO_RCVTIMEO:
1851 optval = (sopt->sopt_name == SO_SNDTIMEO ?
1852 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1853
1854 tv.tv_sec = optval / hz;
1855 tv.tv_usec = (optval % hz) * tick;
1856 #ifdef COMPAT_IA32
1857 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
1858 struct timeval32 tv32;
1859
1860 CP(tv, tv32, tv_sec);
1861 CP(tv, tv32, tv_usec);
1862 error = sooptcopyout(sopt, &tv32, sizeof tv32);
1863 } else
1864 #endif
1865 error = sooptcopyout(sopt, &tv, sizeof tv);
1866 break;
1867
1868 case SO_LABEL:
1869 #ifdef MAC
1870 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
1871 sizeof(extmac));
1872 if (error)
1873 return (error);
1874 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
1875 so, &extmac);
1876 if (error)
1877 return (error);
1878 error = sooptcopyout(sopt, &extmac, sizeof extmac);
1879 #else
1880 error = EOPNOTSUPP;
1881 #endif
1882 break;
1883
1884 case SO_PEERLABEL:
1885 #ifdef MAC
1886 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
1887 sizeof(extmac));
1888 if (error)
1889 return (error);
1890 error = mac_getsockopt_peerlabel(
1891 sopt->sopt_td->td_ucred, so, &extmac);
1892 if (error)
1893 return (error);
1894 error = sooptcopyout(sopt, &extmac, sizeof extmac);
1895 #else
1896 error = EOPNOTSUPP;
1897 #endif
1898 break;
1899
1900 case SO_LISTENQLIMIT:
1901 optval = so->so_qlimit;
1902 goto integer;
1903
1904 case SO_LISTENQLEN:
1905 optval = so->so_qlen;
1906 goto integer;
1907
1908 case SO_LISTENINCQLEN:
1909 optval = so->so_incqlen;
1910 goto integer;
1911
1912 default:
1913 error = ENOPROTOOPT;
1914 break;
1915 }
1916 return (error);
1917 }
1918 }
1919
1920 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1921 int
1922 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1923 {
1924 struct mbuf *m, *m_prev;
1925 int sopt_size = sopt->sopt_valsize;
1926
1927 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1928 if (m == NULL)
1929 return ENOBUFS;
1930 if (sopt_size > MLEN) {
1931 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
1932 if ((m->m_flags & M_EXT) == 0) {
1933 m_free(m);
1934 return ENOBUFS;
1935 }
1936 m->m_len = min(MCLBYTES, sopt_size);
1937 } else {
1938 m->m_len = min(MLEN, sopt_size);
1939 }
1940 sopt_size -= m->m_len;
1941 *mp = m;
1942 m_prev = m;
1943
1944 while (sopt_size) {
1945 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1946 if (m == NULL) {
1947 m_freem(*mp);
1948 return ENOBUFS;
1949 }
1950 if (sopt_size > MLEN) {
1951 MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
1952 M_DONTWAIT);
1953 if ((m->m_flags & M_EXT) == 0) {
1954 m_freem(m);
1955 m_freem(*mp);
1956 return ENOBUFS;
1957 }
1958 m->m_len = min(MCLBYTES, sopt_size);
1959 } else {
1960 m->m_len = min(MLEN, sopt_size);
1961 }
1962 sopt_size -= m->m_len;
1963 m_prev->m_next = m;
1964 m_prev = m;
1965 }
1966 return 0;
1967 }
1968
1969 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1970 int
1971 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1972 {
1973 struct mbuf *m0 = m;
1974
1975 if (sopt->sopt_val == NULL)
1976 return 0;
1977 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1978 if (sopt->sopt_td != NULL) {
1979 int error;
1980
1981 error = copyin(sopt->sopt_val, mtod(m, char *),
1982 m->m_len);
1983 if (error != 0) {
1984 m_freem(m0);
1985 return(error);
1986 }
1987 } else
1988 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
1989 sopt->sopt_valsize -= m->m_len;
1990 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
1991 m = m->m_next;
1992 }
1993 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1994 panic("ip6_sooptmcopyin");
1995 return 0;
1996 }
1997
1998 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1999 int
2000 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2001 {
2002 struct mbuf *m0 = m;
2003 size_t valsize = 0;
2004
2005 if (sopt->sopt_val == NULL)
2006 return 0;
2007 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2008 if (sopt->sopt_td != NULL) {
2009 int error;
2010
2011 error = copyout(mtod(m, char *), sopt->sopt_val,
2012 m->m_len);
2013 if (error != 0) {
2014 m_freem(m0);
2015 return(error);
2016 }
2017 } else
2018 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2019 sopt->sopt_valsize -= m->m_len;
2020 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2021 valsize += m->m_len;
2022 m = m->m_next;
2023 }
2024 if (m != NULL) {
2025 /* enough soopt buffer should be given from user-land */
2026 m_freem(m0);
2027 return(EINVAL);
2028 }
2029 sopt->sopt_valsize = valsize;
2030 return 0;
2031 }
2032
2033 void
2034 sohasoutofband(so)
2035 struct socket *so;
2036 {
2037 if (so->so_sigio != NULL)
2038 pgsigio(&so->so_sigio, SIGURG, 0);
2039 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2040 }
2041
2042 int
2043 sopoll(struct socket *so, int events, struct ucred *active_cred,
2044 struct thread *td)
2045 {
2046 int revents = 0;
2047
2048 SOCKBUF_LOCK(&so->so_snd);
2049 SOCKBUF_LOCK(&so->so_rcv);
2050 if (events & (POLLIN | POLLRDNORM))
2051 if (soreadable(so))
2052 revents |= events & (POLLIN | POLLRDNORM);
2053
2054 if (events & POLLINIGNEOF)
2055 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2056 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2057 revents |= POLLINIGNEOF;
2058
2059 if (events & (POLLOUT | POLLWRNORM))
2060 if (sowriteable(so))
2061 revents |= events & (POLLOUT | POLLWRNORM);
2062
2063 if (events & (POLLPRI | POLLRDBAND))
2064 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2065 revents |= events & (POLLPRI | POLLRDBAND);
2066
2067 if (revents == 0) {
2068 if (events &
2069 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2070 POLLRDBAND)) {
2071 selrecord(td, &so->so_rcv.sb_sel);
2072 so->so_rcv.sb_flags |= SB_SEL;
2073 }
2074
2075 if (events & (POLLOUT | POLLWRNORM)) {
2076 selrecord(td, &so->so_snd.sb_sel);
2077 so->so_snd.sb_flags |= SB_SEL;
2078 }
2079 }
2080
2081 SOCKBUF_UNLOCK(&so->so_rcv);
2082 SOCKBUF_UNLOCK(&so->so_snd);
2083 return (revents);
2084 }
2085
2086 int
2087 soo_kqfilter(struct file *fp, struct knote *kn)
2088 {
2089 struct socket *so = kn->kn_fp->f_data;
2090 struct sockbuf *sb;
2091
2092 switch (kn->kn_filter) {
2093 case EVFILT_READ:
2094 if (so->so_options & SO_ACCEPTCONN)
2095 kn->kn_fop = &solisten_filtops;
2096 else
2097 kn->kn_fop = &soread_filtops;
2098 sb = &so->so_rcv;
2099 break;
2100 case EVFILT_WRITE:
2101 kn->kn_fop = &sowrite_filtops;
2102 sb = &so->so_snd;
2103 break;
2104 default:
2105 return (EINVAL);
2106 }
2107
2108 SOCKBUF_LOCK(sb);
2109 knlist_add(&sb->sb_sel.si_note, kn, 1);
2110 sb->sb_flags |= SB_KNOTE;
2111 SOCKBUF_UNLOCK(sb);
2112 return (0);
2113 }
2114
2115 static void
2116 filt_sordetach(struct knote *kn)
2117 {
2118 struct socket *so = kn->kn_fp->f_data;
2119
2120 SOCKBUF_LOCK(&so->so_rcv);
2121 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2122 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2123 so->so_rcv.sb_flags &= ~SB_KNOTE;
2124 SOCKBUF_UNLOCK(&so->so_rcv);
2125 }
2126
2127 /*ARGSUSED*/
2128 static int
2129 filt_soread(struct knote *kn, long hint)
2130 {
2131 struct socket *so;
2132
2133 so = kn->kn_fp->f_data;
2134 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2135
2136 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2137 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2138 kn->kn_flags |= EV_EOF;
2139 kn->kn_fflags = so->so_error;
2140 return (1);
2141 } else if (so->so_error) /* temporary udp error */
2142 return (1);
2143 else if (kn->kn_sfflags & NOTE_LOWAT)
2144 return (kn->kn_data >= kn->kn_sdata);
2145 else
2146 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2147 }
2148
2149 static void
2150 filt_sowdetach(struct knote *kn)
2151 {
2152 struct socket *so = kn->kn_fp->f_data;
2153
2154 SOCKBUF_LOCK(&so->so_snd);
2155 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2156 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2157 so->so_snd.sb_flags &= ~SB_KNOTE;
2158 SOCKBUF_UNLOCK(&so->so_snd);
2159 }
2160
2161 /*ARGSUSED*/
2162 static int
2163 filt_sowrite(struct knote *kn, long hint)
2164 {
2165 struct socket *so;
2166
2167 so = kn->kn_fp->f_data;
2168 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2169 kn->kn_data = sbspace(&so->so_snd);
2170 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2171 kn->kn_flags |= EV_EOF;
2172 kn->kn_fflags = so->so_error;
2173 return (1);
2174 } else if (so->so_error) /* temporary udp error */
2175 return (1);
2176 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2177 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2178 return (0);
2179 else if (kn->kn_sfflags & NOTE_LOWAT)
2180 return (kn->kn_data >= kn->kn_sdata);
2181 else
2182 return (kn->kn_data >= so->so_snd.sb_lowat);
2183 }
2184
2185 /*ARGSUSED*/
2186 static int
2187 filt_solisten(struct knote *kn, long hint)
2188 {
2189 struct socket *so = kn->kn_fp->f_data;
2190
2191 kn->kn_data = so->so_qlen;
2192 return (! TAILQ_EMPTY(&so->so_comp));
2193 }
2194
2195 int
2196 socheckuid(struct socket *so, uid_t uid)
2197 {
2198
2199 if (so == NULL)
2200 return (EPERM);
2201 if (so->so_cred->cr_uid != uid)
2202 return (EPERM);
2203 return (0);
2204 }
2205
2206 static int
2207 somaxconn_sysctl(SYSCTL_HANDLER_ARGS)
2208 {
2209 int error;
2210 int val;
2211
2212 val = somaxconn;
2213 error = sysctl_handle_int(oidp, &val, sizeof(int), req);
2214 if (error || !req->newptr )
2215 return (error);
2216
2217 if (val < 1 || val > USHRT_MAX)
2218 return (EINVAL);
2219
2220 somaxconn = val;
2221 return (0);
2222 }
Cache object: 5a314967918971c886190f070c0de28d
|