]> git.saurik.com Git - apple/xnu.git/blob - bsd/kern/uipc_socket.c
projects / apple / xnu.git / blob
? search:


7b259ec9f657e36a049d8383732bf7f5118088c7
[apple/xnu.git] / bsd / kern / uipc_socket.c
1 /*
2 * Copyright (c) 1998-2007 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
29 /*
30 * Copyright (c) 1982, 1986, 1988, 1990, 1993
31 * The Regents of the University of California. All rights reserved.
32 *
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
35 * are met:
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
41 * 3. All advertising materials mentioning features or use of this software
42 * must display the following acknowledgement:
43 * This product includes software developed by the University of
44 * California, Berkeley and its contributors.
45 * 4. Neither the name of the University nor the names of its contributors
46 * may be used to endorse or promote products derived from this software
47 * without specific prior written permission.
48 *
49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59 * SUCH DAMAGE.
60 *
61 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
62 * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.16 2001/06/14 20:46:06 ume Exp $
63 */
64 /*
65 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
66 * support for mandatory and extensible security protections. This notice
67 * is included in support of clause 2.2 (b) of the Apple Public License,
68 * Version 2.0.
69 */
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/filedesc.h>
74 #include <sys/proc.h>
75 #include <sys/proc_internal.h>
76 #include <sys/kauth.h>
77 #include <sys/file_internal.h>
78 #include <sys/fcntl.h>
79 #include <sys/malloc.h>
80 #include <sys/mbuf.h>
81 #include <sys/domain.h>
82 #include <sys/kernel.h>
83 #include <sys/event.h>
84 #include <sys/poll.h>
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/resourcevar.h>
89 #include <sys/signalvar.h>
90 #include <sys/sysctl.h>
91 #include <sys/uio.h>
92 #include <sys/ev.h>
93 #include <sys/kdebug.h>
94 #include <sys/un.h>
95 #include <net/route.h>
96 #include <netinet/in.h>
97 #include <netinet/in_pcb.h>
98 #include <kern/zalloc.h>
99 #include <kern/locks.h>
100 #include <machine/limits.h>
101 #include <libkern/OSAtomic.h>
102 #include <pexpert/pexpert.h>
103
104 #if CONFIG_MACF
105 #include <security/mac.h>
106 #include <security/mac_framework.h>
107 #endif /* MAC */
108
109 /* how a timeval looks to a 64-bit process */
110 struct timeval64 {
111 int64_t tv_sec;
112 int32_t tv_usec;
113 };
114
115 int so_cache_hw = 0;
116 int so_cache_timeouts = 0;
117 int so_cache_max_freed = 0;
118 int cached_sock_count = 0;
119 struct socket *socket_cache_head = 0;
120 struct socket *socket_cache_tail = 0;
121 u_long so_cache_time = 0;
122 int so_cache_init_done = 0;
123 struct zone *so_cache_zone;
124
125 static lck_grp_t *so_cache_mtx_grp;
126 static lck_attr_t *so_cache_mtx_attr;
127 static lck_grp_attr_t *so_cache_mtx_grp_attr;
128 lck_mtx_t *so_cache_mtx;
129
130 #include <machine/limits.h>
131
132 static void filt_sordetach(struct knote *kn);
133 static int filt_soread(struct knote *kn, long hint);
134 static void filt_sowdetach(struct knote *kn);
135 static int filt_sowrite(struct knote *kn, long hint);
136 static int filt_solisten(struct knote *kn, long hint);
137
138 static int
139 sooptcopyin_timeval(struct sockopt *sopt, struct timeval * tv_p);
140
141 static int
142 sooptcopyout_timeval(struct sockopt *sopt, const struct timeval * tv_p);
143
144 static struct filterops solisten_filtops =
145 { 1, NULL, filt_sordetach, filt_solisten };
146 static struct filterops soread_filtops =
147 { 1, NULL, filt_sordetach, filt_soread };
148 static struct filterops sowrite_filtops =
149 { 1, NULL, filt_sowdetach, filt_sowrite };
150
151 #define EVEN_MORE_LOCKING_DEBUG 0
152 int socket_debug = 0;
153 int socket_zone = M_SOCKET;
154 so_gen_t so_gencnt; /* generation count for sockets */
155
156 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
157 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
158
159 #define DBG_LAYER_IN_BEG NETDBG_CODE(DBG_NETSOCK, 0)
160 #define DBG_LAYER_IN_END NETDBG_CODE(DBG_NETSOCK, 2)
161 #define DBG_LAYER_OUT_BEG NETDBG_CODE(DBG_NETSOCK, 1)
162 #define DBG_LAYER_OUT_END NETDBG_CODE(DBG_NETSOCK, 3)
163 #define DBG_FNC_SOSEND NETDBG_CODE(DBG_NETSOCK, (4 << 8) | 1)
164 #define DBG_FNC_SORECEIVE NETDBG_CODE(DBG_NETSOCK, (8 << 8))
165 #define DBG_FNC_SOSHUTDOWN NETDBG_CODE(DBG_NETSOCK, (9 << 8))
166
167 #define MAX_SOOPTGETM_SIZE (128 * MCLBYTES)
168
169
170 SYSCTL_DECL(_kern_ipc);
171
172 int somaxconn = SOMAXCONN;
173 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW, &somaxconn, 0, "");
174
175 /* Should we get a maximum also ??? */
176 static int sosendmaxchain = 65536;
177 static int sosendminchain = 16384;
178 static int sorecvmincopy = 16384;
179 SYSCTL_INT(_kern_ipc, OID_AUTO, sosendminchain, CTLFLAG_RW, &sosendminchain,
180 0, "");
181 SYSCTL_INT(_kern_ipc, OID_AUTO, sorecvmincopy, CTLFLAG_RW, &sorecvmincopy,
182 0, "");
183
184 /*
185 * Set to enable jumbo clusters (if available) for large writes when
186 * the socket is marked with SOF_MULTIPAGES; see below.
187 */
188 int sosendjcl = 1;
189 SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl, CTLFLAG_RW, &sosendjcl, 0, "");
190
191 /*
192 * Set this to ignore SOF_MULTIPAGES and use jumbo clusters for large
193 * writes on the socket for all protocols on any network interfaces,
194 * depending upon sosendjcl above. Be extra careful when setting this
195 * to 1, because sending down packets that cross physical pages down to
196 * broken drivers (those that falsely assume that the physical pages
197 * are contiguous) might lead to system panics or silent data corruption.
198 * When set to 0, the system will respect SOF_MULTIPAGES, which is set
199 * only for TCP sockets whose outgoing interface is IFNET_MULTIPAGES
200 * capable. Set this to 1 only for testing/debugging purposes.
201 */
202 int sosendjcl_ignore_capab = 0;
203 SYSCTL_INT(_kern_ipc, OID_AUTO, sosendjcl_ignore_capab, CTLFLAG_RW,
204 &sosendjcl_ignore_capab, 0, "");
205
206 /*
207 * Socket operation routines.
208 * These routines are called by the routines in
209 * sys_socket.c or from a system process, and
210 * implement the semantics of socket operations by
211 * switching out to the protocol specific routines.
212 */
213
214 /* sys_generic.c */
215 extern void postevent(struct socket *, struct sockbuf *, int);
216 extern void evsofree(struct socket *);
217
218 /* TODO: these should be in header file */
219 extern int get_inpcb_str_size(void);
220 extern int get_tcp_str_size(void);
221 extern struct domain *pffinddomain(int);
222 extern struct protosw *pffindprotonotype(int, int);
223 extern int soclose_locked(struct socket *);
224 extern int soo_kqfilter(struct fileproc *, struct knote *, struct proc *);
225
226 #ifdef __APPLE__
227
228 vm_size_t so_cache_zone_element_size;
229
230 static int sodelayed_copy(struct socket *, struct uio *, struct mbuf **, int *);
231 static void cached_sock_alloc(struct socket **, int);
232 static void cached_sock_free(struct socket *);
233 static void so_cache_timer(void *);
234
235 void soclose_wait_locked(struct socket *so);
236
237
238 void
239 socketinit(void)
240 {
241 vm_size_t str_size;
242
243 if (so_cache_init_done) {
244 printf("socketinit: already called...\n");
245 return;
246 }
247
248 PE_parse_boot_arg("socket_debug", &socket_debug);
249
250 /*
251 * allocate lock group attribute and group for socket cache mutex
252 */
253 so_cache_mtx_grp_attr = lck_grp_attr_alloc_init();
254
255 so_cache_mtx_grp = lck_grp_alloc_init("so_cache",
256 so_cache_mtx_grp_attr);
257
258 /*
259 * allocate the lock attribute for socket cache mutex
260 */
261 so_cache_mtx_attr = lck_attr_alloc_init();
262
263 so_cache_init_done = 1;
264
265 /* cached sockets mutex */
266 so_cache_mtx = lck_mtx_alloc_init(so_cache_mtx_grp, so_cache_mtx_attr);
267
268 if (so_cache_mtx == NULL)
269 return; /* we're hosed... */
270
271 str_size = (vm_size_t)(sizeof (struct socket) + 4 +
272 get_inpcb_str_size() + 4 + get_tcp_str_size());
273
274 so_cache_zone = zinit(str_size, 120000*str_size, 8192, "socache zone");
275 #if TEMPDEBUG
276 printf("cached_sock_alloc -- so_cache_zone size is %x\n", str_size);
277 #endif
278 timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz));
279
280 so_cache_zone_element_size = str_size;
281
282 sflt_init();
283 }
284
285 static void
286 cached_sock_alloc(struct socket **so, int waitok)
287 {
288 caddr_t temp;
289 register u_long offset;
290
291 lck_mtx_lock(so_cache_mtx);
292
293 if (cached_sock_count) {
294 cached_sock_count--;
295 *so = socket_cache_head;
296 if (*so == 0)
297 panic("cached_sock_alloc: cached sock is null");
298
299 socket_cache_head = socket_cache_head->cache_next;
300 if (socket_cache_head)
301 socket_cache_head->cache_prev = 0;
302 else
303 socket_cache_tail = 0;
304
305 lck_mtx_unlock(so_cache_mtx);
306
307 temp = (*so)->so_saved_pcb;
308 bzero((caddr_t)*so, sizeof (struct socket));
309 #if TEMPDEBUG
310 kprintf("cached_sock_alloc - retreiving cached sock %p - "
311 "count == %d\n", *so, cached_sock_count);
312 #endif
313 (*so)->so_saved_pcb = temp;
314 (*so)->cached_in_sock_layer = 1;
315 } else {
316 #if TEMPDEBUG
317 kprintf("Allocating cached sock %p from memory\n", *so);
318 #endif
319
320 lck_mtx_unlock(so_cache_mtx);
321
322 if (waitok)
323 *so = (struct socket *)zalloc(so_cache_zone);
324 else
325 *so = (struct socket *)zalloc_noblock(so_cache_zone);
326
327 if (*so == 0)
328 return;
329
330 bzero((caddr_t)*so, sizeof (struct socket));
331
332 /*
333 * Define offsets for extra structures into our single block of
334 * memory. Align extra structures on longword boundaries.
335 */
336 offset = (u_long) *so;
337 offset += sizeof (struct socket);
338 if (offset & 0x3) {
339 offset += 4;
340 offset &= 0xfffffffc;
341 }
342 (*so)->so_saved_pcb = (caddr_t)offset;
343 offset += get_inpcb_str_size();
344 if (offset & 0x3) {
345 offset += 4;
346 offset &= 0xfffffffc;
347 }
348
349 ((struct inpcb *)(*so)->so_saved_pcb)->inp_saved_ppcb =
350 (caddr_t)offset;
351 #if TEMPDEBUG
352 kprintf("Allocating cached socket - %p, pcb=%p tcpcb=%p\n",
353 *so, (*so)->so_saved_pcb,
354 ((struct inpcb *)(*so)->so_saved_pcb)->inp_saved_ppcb);
355 #endif
356 }
357
358 (*so)->cached_in_sock_layer = 1;
359 }
360
361 static void
362 cached_sock_free(struct socket *so)
363 {
364
365 lck_mtx_lock(so_cache_mtx);
366
367 if (++cached_sock_count > MAX_CACHED_SOCKETS) {
368 --cached_sock_count;
369 lck_mtx_unlock(so_cache_mtx);
370 #if TEMPDEBUG
371 kprintf("Freeing overflowed cached socket %p\n", so);
372 #endif
373 zfree(so_cache_zone, so);
374 } else {
375 #if TEMPDEBUG
376 kprintf("Freeing socket %p into cache\n", so);
377 #endif
378 if (so_cache_hw < cached_sock_count)
379 so_cache_hw = cached_sock_count;
380
381 so->cache_next = socket_cache_head;
382 so->cache_prev = 0;
383 if (socket_cache_head)
384 socket_cache_head->cache_prev = so;
385 else
386 socket_cache_tail = so;
387
388 so->cache_timestamp = so_cache_time;
389 socket_cache_head = so;
390 lck_mtx_unlock(so_cache_mtx);
391 }
392
393 #if TEMPDEBUG
394 kprintf("Freed cached sock %p into cache - count is %d\n",
395 so, cached_sock_count);
396 #endif
397 }
398
399 static void
400 so_cache_timer(__unused void *dummy)
401 {
402 register struct socket *p;
403 register int n_freed = 0;
404
405 lck_mtx_lock(so_cache_mtx);
406
407 ++so_cache_time;
408
409 while ((p = socket_cache_tail)) {
410 if ((so_cache_time - p->cache_timestamp) < SO_CACHE_TIME_LIMIT)
411 break;
412
413 so_cache_timeouts++;
414
415 if ((socket_cache_tail = p->cache_prev))
416 p->cache_prev->cache_next = 0;
417 if (--cached_sock_count == 0)
418 socket_cache_head = 0;
419
420 zfree(so_cache_zone, p);
421
422 if (++n_freed >= SO_CACHE_MAX_FREE_BATCH) {
423 so_cache_max_freed++;
424 break;
425 }
426 }
427 lck_mtx_unlock(so_cache_mtx);
428
429 timeout(so_cache_timer, NULL, (SO_CACHE_FLUSH_INTERVAL * hz));
430 }
431 #endif /* __APPLE__ */
432
433 /*
434 * Get a socket structure from our zone, and initialize it.
435 * We don't implement `waitok' yet (see comments in uipc_domain.c).
436 * Note that it would probably be better to allocate socket
437 * and PCB at the same time, but I'm not convinced that all
438 * the protocols can be easily modified to do this.
439 */
440 struct socket *
441 soalloc(int waitok, int dom, int type)
442 {
443 struct socket *so;
444
445 if ((dom == PF_INET) && (type == SOCK_STREAM)) {
446 cached_sock_alloc(&so, waitok);
447 } else {
448 MALLOC_ZONE(so, struct socket *, sizeof (*so), socket_zone,
449 M_WAITOK);
450 if (so != NULL)
451 bzero(so, sizeof (*so));
452 }
453 /* XXX race condition for reentrant kernel */
454 //###LD Atomic add for so_gencnt
455 if (so != NULL) {
456 so->so_gencnt = ++so_gencnt;
457 so->so_zone = socket_zone;
458 #if CONFIG_MACF_SOCKET
459 /* Convert waitok to M_WAITOK/M_NOWAIT for MAC Framework. */
460 if (mac_socket_label_init(so, !waitok) != 0) {
461 sodealloc(so);
462 return (NULL);
463 }
464 #endif /* MAC_SOCKET */
465 }
466
467 return (so);
468 }
469
470 /*
471 * Returns: 0 Success
472 * EAFNOSUPPORT
473 * EPROTOTYPE
474 * EPROTONOSUPPORT
475 * ENOBUFS
476 * <pru_attach>:ENOBUFS[AF_UNIX]
477 * <pru_attach>:ENOBUFS[TCP]
478 * <pru_attach>:ENOMEM[TCP]
479 * <pru_attach>:EISCONN[TCP]
480 * <pru_attach>:??? [other protocol families, IPSEC]
481 */
482 int
483 socreate(int dom, struct socket **aso, int type, int proto)
484 {
485 struct proc *p = current_proc();
486 register struct protosw *prp;
487 register struct socket *so;
488 register int error = 0;
489 #if TCPDEBUG
490 extern int tcpconsdebug;
491 #endif
492 if (proto)
493 prp = pffindproto(dom, proto, type);
494 else
495 prp = pffindtype(dom, type);
496
497 if (prp == 0 || prp->pr_usrreqs->pru_attach == 0) {
498 if (pffinddomain(dom) == NULL) {
499 return (EAFNOSUPPORT);
500 }
501 if (proto != 0) {
502 if (pffindprotonotype(dom, proto) != NULL) {
503 return (EPROTOTYPE);
504 }
505 }
506 return (EPROTONOSUPPORT);
507 }
508 if (prp->pr_type != type)
509 return (EPROTOTYPE);
510 so = soalloc(p != 0, dom, type);
511 if (so == 0)
512 return (ENOBUFS);
513
514 TAILQ_INIT(&so->so_incomp);
515 TAILQ_INIT(&so->so_comp);
516 so->so_type = type;
517
518 if (p != 0) {
519 so->so_uid = kauth_cred_getuid(kauth_cred_get());
520 if (!suser(kauth_cred_get(), NULL))
521 so->so_state = SS_PRIV;
522 }
523 so->so_proto = prp;
524 #ifdef __APPLE__
525 so->so_rcv.sb_flags |= SB_RECV; /* XXX */
526 so->so_rcv.sb_so = so->so_snd.sb_so = so;
527 #endif
528 so->next_lock_lr = 0;
529 so->next_unlock_lr = 0;
530
531 #if CONFIG_MACF_SOCKET
532 mac_socket_label_associate(kauth_cred_get(), so);
533 #endif /* MAC_SOCKET */
534
535 //### Attachement will create the per pcb lock if necessary and increase refcount
536 /*
537 * for creation, make sure it's done before
538 * socket is inserted in lists
539 */
540 so->so_usecount++;
541
542 error = (*prp->pr_usrreqs->pru_attach)(so, proto, p);
543 if (error) {
544 /*
545 * Warning:
546 * If so_pcb is not zero, the socket will be leaked,
547 * so protocol attachment handler must be coded carefuly
548 */
549 so->so_state |= SS_NOFDREF;
550 so->so_usecount--;
551 sofreelastref(so, 1); /* will deallocate the socket */
552 return (error);
553 }
554 #ifdef __APPLE__
555 prp->pr_domain->dom_refs++;
556 TAILQ_INIT(&so->so_evlist);
557
558 /* Attach socket filters for this protocol */
559 sflt_initsock(so);
560 #if TCPDEBUG
561 if (tcpconsdebug == 2)
562 so->so_options |= SO_DEBUG;
563 #endif
564 #endif
565 *aso = so;
566 return (0);
567 }
568
569 /*
570 * Returns: 0 Success
571 * <pru_bind>:EINVAL Invalid argument [COMMON_START]
572 * <pru_bind>:EAFNOSUPPORT Address family not supported
573 * <pru_bind>:EADDRNOTAVAIL Address not available.
574 * <pru_bind>:EINVAL Invalid argument
575 * <pru_bind>:EAFNOSUPPORT Address family not supported [notdef]
576 * <pru_bind>:EACCES Permission denied
577 * <pru_bind>:EADDRINUSE Address in use
578 * <pru_bind>:EAGAIN Resource unavailable, try again
579 * <pru_bind>:EPERM Operation not permitted
580 * <pru_bind>:???
581 * <sf_bind>:???
582 *
583 * Notes: It's not possible to fully enumerate the return codes above,
584 * since socket filter authors and protocol family authors may
585 * not choose to limit their error returns to those listed, even
586 * though this may result in some software operating incorrectly.
587 *
588 * The error codes which are enumerated above are those known to
589 * be returned by the tcp_usr_bind function supplied.
590 */
591 int
592 sobind(struct socket *so, struct sockaddr *nam)
593 {
594 struct proc *p = current_proc();
595 int error = 0;
596 struct socket_filter_entry *filter;
597 int filtered = 0;
598
599 socket_lock(so, 1);
600
601 /*
602 * If this is a bind request on a previously-accepted socket
603 * that has been marked as inactive, reject it now before
604 * we go any further.
605 */
606 if (so->so_flags & SOF_DEFUNCT) {
607 error = EINVAL;
608 goto out;
609 }
610
611 /* Socket filter */
612 error = 0;
613 for (filter = so->so_filt; filter && (error == 0);
614 filter = filter->sfe_next_onsocket) {
615 if (filter->sfe_filter->sf_filter.sf_bind) {
616 if (filtered == 0) {
617 filtered = 1;
618 sflt_use(so);
619 socket_unlock(so, 0);
620 }
621 error = filter->sfe_filter->sf_filter.
622 sf_bind(filter->sfe_cookie, so, nam);
623 }
624 }
625 if (filtered != 0) {
626 socket_lock(so, 0);
627 sflt_unuse(so);
628 }
629 /* End socket filter */
630
631 if (error == 0)
632 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, p);
633 out:
634 socket_unlock(so, 1);
635
636 if (error == EJUSTRETURN)
637 error = 0;
638
639 return (error);
640 }
641
642 void
643 sodealloc(struct socket *so)
644 {
645 so->so_gencnt = ++so_gencnt;
646
647 #if CONFIG_MACF_SOCKET
648 mac_socket_label_destroy(so);
649 #endif /* MAC_SOCKET */
650 if (so->cached_in_sock_layer == 1) {
651 cached_sock_free(so);
652 } else {
653 if (so->cached_in_sock_layer == -1)
654 panic("sodealloc: double dealloc: so=%p\n", so);
655 so->cached_in_sock_layer = -1;
656 FREE_ZONE(so, sizeof (*so), so->so_zone);
657 }
658 }
659
660 /*
661 * Returns: 0 Success
662 * EINVAL
663 * EOPNOTSUPP
664 * <pru_listen>:EINVAL[AF_UNIX]
665 * <pru_listen>:EINVAL[TCP]
666 * <pru_listen>:EADDRNOTAVAIL[TCP] Address not available.
667 * <pru_listen>:EINVAL[TCP] Invalid argument
668 * <pru_listen>:EAFNOSUPPORT[TCP] Address family not supported [notdef]
669 * <pru_listen>:EACCES[TCP] Permission denied
670 * <pru_listen>:EADDRINUSE[TCP] Address in use
671 * <pru_listen>:EAGAIN[TCP] Resource unavailable, try again
672 * <pru_listen>:EPERM[TCP] Operation not permitted
673 * <sf_listen>:???
674 *
675 * Notes: Other <pru_listen> returns depend on the protocol family; all
676 * <sf_listen> returns depend on what the filter author causes
677 * their filter to return.
678 */
679 int
680 solisten(struct socket *so, int backlog)
681 {
682 struct proc *p = current_proc();
683 int error = 0;
684 struct socket_filter_entry *filter;
685 int filtered = 0;
686
687 socket_lock(so, 1);
688 if (so->so_proto == NULL) {
689 error = EINVAL;
690 goto out;
691 }
692 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0) {
693 error = EOPNOTSUPP;
694 goto out;
695 }
696
697 /*
698 * If the listen request is made on a socket that is not fully
699 * disconnected, or on a previously-accepted socket that has
700 * been marked as inactive, reject the request now.
701 */
702 if ((so->so_state &
703 (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING)) ||
704 (so->so_flags & SOF_DEFUNCT)) {
705 error = EINVAL;
706 goto out;
707 }
708
709 if ((so->so_restrictions & SO_RESTRICT_DENYIN) != 0) {
710 error = EPERM;
711 goto out;
712 }
713
714 error = 0;
715 for (filter = so->so_filt; filter && (error == 0);
716 filter = filter->sfe_next_onsocket) {
717 if (filter->sfe_filter->sf_filter.sf_listen) {
718 if (filtered == 0) {
719 filtered = 1;
720 sflt_use(so);
721 socket_unlock(so, 0);
722 }
723 error = filter->sfe_filter->sf_filter.
724 sf_listen(filter->sfe_cookie, so);
725 }
726 }
727 if (filtered != 0) {
728 socket_lock(so, 0);
729 sflt_unuse(so);
730 }
731
732 if (error == 0) {
733 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, p);
734 }
735
736 if (error) {
737 if (error == EJUSTRETURN)
738 error = 0;
739 goto out;
740 }
741
742 if (TAILQ_EMPTY(&so->so_comp))
743 so->so_options |= SO_ACCEPTCONN;
744 /*
745 * POSIX: The implementation may have an upper limit on the length of
746 * the listen queue-either global or per accepting socket. If backlog
747 * exceeds this limit, the length of the listen queue is set to the
748 * limit.
749 *
750 * If listen() is called with a backlog argument value that is less
751 * than 0, the function behaves as if it had been called with a backlog
752 * argument value of 0.
753 *
754 * A backlog argument of 0 may allow the socket to accept connections,
755 * in which case the length of the listen queue may be set to an
756 * implementation-defined minimum value.
757 */
758 if (backlog <= 0 || backlog > somaxconn)
759 backlog = somaxconn;
760
761 so->so_qlimit = backlog;
762 out:
763 socket_unlock(so, 1);
764 return (error);
765 }
766
767 void
768 sofreelastref(struct socket *so, int dealloc)
769 {
770 struct socket *head = so->so_head;
771
772 /* Assume socket is locked */
773
774 /* Remove any filters - may be called more than once */
775 sflt_termsock(so);
776
777 if ((!(so->so_flags & SOF_PCBCLEARING)) ||
778 ((so->so_state & SS_NOFDREF) == 0)) {
779 #ifdef __APPLE__
780 selthreadclear(&so->so_snd.sb_sel);
781 selthreadclear(&so->so_rcv.sb_sel);
782 so->so_rcv.sb_flags &= ~SB_UPCALL;
783 so->so_snd.sb_flags &= ~SB_UPCALL;
784 #endif
785 return;
786 }
787 if (head != NULL) {
788 socket_lock(head, 1);
789 if (so->so_state & SS_INCOMP) {
790 TAILQ_REMOVE(&head->so_incomp, so, so_list);
791 head->so_incqlen--;
792 } else if (so->so_state & SS_COMP) {
793 /*
794 * We must not decommission a socket that's
795 * on the accept(2) queue. If we do, then
796 * accept(2) may hang after select(2) indicated
797 * that the listening socket was ready.
798 */
799 #ifdef __APPLE__
800 selthreadclear(&so->so_snd.sb_sel);
801 selthreadclear(&so->so_rcv.sb_sel);
802 so->so_rcv.sb_flags &= ~SB_UPCALL;
803 so->so_snd.sb_flags &= ~SB_UPCALL;
804 #endif
805 socket_unlock(head, 1);
806 return;
807 } else {
808 panic("sofree: not queued");
809 }
810 head->so_qlen--;
811 so->so_state &= ~SS_INCOMP;
812 so->so_head = NULL;
813 socket_unlock(head, 1);
814 }
815 #ifdef __APPLE__
816 selthreadclear(&so->so_snd.sb_sel);
817 sbrelease(&so->so_snd);
818 #endif
819 sorflush(so);
820
821 /* 3932268: disable upcall */
822 so->so_rcv.sb_flags &= ~SB_UPCALL;
823 so->so_snd.sb_flags &= ~SB_UPCALL;
824
825 if (dealloc)
826 sodealloc(so);
827 }
828
829 void
830 soclose_wait_locked(struct socket *so)
831 {
832 lck_mtx_t *mutex_held;
833
834 if (so->so_proto->pr_getlock != NULL)
835 mutex_held = (*so->so_proto->pr_getlock)(so, 0);
836 else
837 mutex_held = so->so_proto->pr_domain->dom_mtx;
838 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
839
840 /*
841 * Double check here and return if there's no outstanding upcall;
842 * otherwise proceed further only if SOF_UPCALLCLOSEWAIT is set.
843 */
844 if (!(so->so_flags & SOF_UPCALLINUSE) ||
845 !(so->so_flags & SOF_UPCALLCLOSEWAIT))
846 return;
847
848 so->so_flags |= SOF_CLOSEWAIT;
849 (void) msleep((caddr_t)&so->so_upcall, mutex_held, (PZERO - 1),
850 "soclose_wait_locked", NULL);
851 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
852 so->so_flags &= ~SOF_CLOSEWAIT;
853 }
854
855 /*
856 * Close a socket on last file table reference removal.
857 * Initiate disconnect if connected.
858 * Free socket when disconnect complete.
859 */
860 int
861 soclose_locked(struct socket *so)
862 {
863 int error = 0;
864 lck_mtx_t *mutex_held;
865 struct timespec ts;
866
867 if (so->so_usecount == 0) {
868 panic("soclose: so=%p refcount=0\n", so);
869 }
870
871 sflt_notify(so, sock_evt_closing, NULL);
872
873 if ((so->so_options & SO_ACCEPTCONN)) {
874 struct socket *sp, *sonext;
875 int socklock = 0;
876
877 /*
878 * We do not want new connection to be added
879 * to the connection queues
880 */
881 so->so_options &= ~SO_ACCEPTCONN;
882
883 for (sp = TAILQ_FIRST(&so->so_incomp); sp != NULL; sp = sonext) {
884 sonext = TAILQ_NEXT(sp, so_list);
885
886 /* Radar 5350314
887 * skip sockets thrown away by tcpdropdropblreq
888 * they will get cleanup by the garbage collection.
889 * otherwise, remove the incomp socket from the queue
890 * and let soabort trigger the appropriate cleanup.
891 */
892 if (sp->so_flags & SOF_OVERFLOW)
893 continue;
894
895 if (so->so_proto->pr_getlock != NULL) {
896 /* lock ordering for consistency with the rest of the stack,
897 * we lock the socket first and then grabb the head.
898 */
899 socket_unlock(so, 0);
900 socket_lock(sp, 1);
901 socket_lock(so, 0);
902 socklock = 1;
903 }
904
905 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
906 so->so_incqlen--;
907
908 if (sp->so_state & SS_INCOMP) {
909 sp->so_state &= ~SS_INCOMP;
910 sp->so_head = NULL;
911
912 (void) soabort(sp);
913 }
914
915 if (socklock)
916 socket_unlock(sp, 1);
917 }
918
919 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
920 /* Dequeue from so_comp since sofree() won't do it */
921 TAILQ_REMOVE(&so->so_comp, sp, so_list);
922 so->so_qlen--;
923
924 if (so->so_proto->pr_getlock != NULL) {
925 socket_unlock(so, 0);
926 socket_lock(sp, 1);
927 }
928
929 if (sp->so_state & SS_COMP) {
930 sp->so_state &= ~SS_COMP;
931 sp->so_head = NULL;
932
933 (void) soabort(sp);
934 }
935
936 if (so->so_proto->pr_getlock != NULL) {
937 socket_unlock(sp, 1);
938 socket_lock(so, 0);
939 }
940 }
941 }
942 if (so->so_pcb == 0) {
943 /* 3915887: mark the socket as ready for dealloc */
944 so->so_flags |= SOF_PCBCLEARING;
945 goto discard;
946 }
947 if (so->so_state & SS_ISCONNECTED) {
948 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
949 error = sodisconnectlocked(so);
950 if (error)
951 goto drop;
952 }
953 if (so->so_options & SO_LINGER) {
954 if ((so->so_state & SS_ISDISCONNECTING) &&
955 (so->so_state & SS_NBIO))
956 goto drop;
957 if (so->so_proto->pr_getlock != NULL)
958 mutex_held = (*so->so_proto->pr_getlock)(so, 0);
959 else
960 mutex_held = so->so_proto->pr_domain->dom_mtx;
961 while (so->so_state & SS_ISCONNECTED) {
962 ts.tv_sec = (so->so_linger/100);
963 ts.tv_nsec = (so->so_linger % 100) *
964 NSEC_PER_USEC * 1000 * 10;
965 error = msleep((caddr_t)&so->so_timeo,
966 mutex_held, PSOCK | PCATCH, "soclose", &ts);
967 if (error) {
968 /*
969 * It's OK when the time fires,
970 * don't report an error
971 */
972 if (error == EWOULDBLOCK)
973 error = 0;
974 break;
975 }
976 }
977 }
978 }
979 drop:
980 if (so->so_usecount == 0)
981 panic("soclose: usecount is zero so=%p\n", so);
982 if (so->so_pcb && !(so->so_flags & SOF_PCBCLEARING)) {
983 int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
984 if (error == 0)
985 error = error2;
986 }
987 if (so->so_usecount <= 0)
988 panic("soclose: usecount is zero so=%p\n", so);
989 discard:
990 if (so->so_pcb && so->so_state & SS_NOFDREF)
991 panic("soclose: NOFDREF");
992 so->so_state |= SS_NOFDREF;
993 #ifdef __APPLE__
994 so->so_proto->pr_domain->dom_refs--;
995 evsofree(so);
996 #endif
997 so->so_usecount--;
998 sofree(so);
999 return (error);
1000 }
1001
1002 int
1003 soclose(struct socket *so)
1004 {
1005 int error = 0;
1006 socket_lock(so, 1);
1007
1008 if (so->so_flags & SOF_UPCALLINUSE)
1009 soclose_wait_locked(so);
1010
1011 if (so->so_retaincnt == 0) {
1012 error = soclose_locked(so);
1013 } else {
1014 /*
1015 * if the FD is going away, but socket is
1016 * retained in kernel remove its reference
1017 */
1018 so->so_usecount--;
1019 if (so->so_usecount < 2)
1020 panic("soclose: retaincnt non null and so=%p "
1021 "usecount=%d\n", so, so->so_usecount);
1022 }
1023 socket_unlock(so, 1);
1024 return (error);
1025 }
1026
1027 /*
1028 * Must be called at splnet...
1029 */
1030 /* Should already be locked */
1031 int
1032 soabort(struct socket *so)
1033 {
1034 int error;
1035
1036 #ifdef MORE_LOCKING_DEBUG
1037 lck_mtx_t *mutex_held;
1038
1039 if (so->so_proto->pr_getlock != NULL)
1040 mutex_held = (*so->so_proto->pr_getlock)(so, 0);
1041 else
1042 mutex_held = so->so_proto->pr_domain->dom_mtx;
1043 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
1044 #endif
1045
1046 if ((so->so_flags & SOF_ABORTED) == 0) {
1047 so->so_flags |= SOF_ABORTED;
1048 error = (*so->so_proto->pr_usrreqs->pru_abort)(so);
1049 if (error) {
1050 sofree(so);
1051 return (error);
1052 }
1053 }
1054 return (0);
1055 }
1056
1057 int
1058 soacceptlock(struct socket *so, struct sockaddr **nam, int dolock)
1059 {
1060 int error;
1061
1062 if (dolock)
1063 socket_lock(so, 1);
1064
1065 if ((so->so_state & SS_NOFDREF) == 0)
1066 panic("soaccept: !NOFDREF");
1067 so->so_state &= ~SS_NOFDREF;
1068 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1069
1070 if (dolock)
1071 socket_unlock(so, 1);
1072 return (error);
1073 }
1074
1075 int
1076 soaccept(struct socket *so, struct sockaddr **nam)
1077 {
1078 return (soacceptlock(so, nam, 1));
1079 }
1080
1081 int
1082 soacceptfilter(struct socket *so)
1083 {
1084 struct sockaddr *local = NULL, *remote = NULL;
1085 struct socket_filter_entry *filter;
1086 int error = 0, filtered = 0;
1087 struct socket *head = so->so_head;
1088
1089 /*
1090 * There's no need to hold the lock; this socket
1091 * has not been made visible to the filter(s).
1092 */
1093 if ((sock_getaddr(so, &remote, 1) != 0) ||
1094 sock_getaddr(so, &local, 0) != 0) {
1095 so->so_state &= ~(SS_NOFDREF | SS_COMP);
1096 so->so_head = NULL;
1097 soclose(so);
1098 /* Out of resources; try it again next time */
1099 error = ECONNABORTED;
1100 goto done;
1101 }
1102
1103 /*
1104 * At this point, we have a reference on the listening socket
1105 * so we know it won't be going away. Do the same for the newly
1106 * accepted socket while we invoke the accept callback routine.
1107 */
1108 socket_lock(so, 1);
1109 for (filter = so->so_filt; filter != NULL && error == 0;
1110 filter = filter->sfe_next_onsocket) {
1111 if (filter->sfe_filter->sf_filter.sf_accept != NULL) {
1112 if (!filtered) {
1113 filtered = 1;
1114 sflt_use(so);
1115 socket_unlock(so, 0);
1116 }
1117 error = filter->sfe_filter->sf_filter.
1118 sf_accept(filter->sfe_cookie,
1119 head, so, local, remote);
1120 }
1121 }
1122
1123 if (filtered) {
1124 socket_lock(so, 0);
1125 sflt_unuse(so);
1126 }
1127
1128 /*
1129 * If we get EJUSTRETURN from one of the filters, mark this socket
1130 * as inactive and return it anyway. This newly accepted socket
1131 * will be disconnected later before we hand it off to the caller.
1132 */
1133 if (error == EJUSTRETURN) {
1134 error = 0;
1135 so->so_flags |= SOF_DEFUNCT;
1136 /* Prevent data from being appended to the socket buffers */
1137 so->so_snd.sb_flags |= SB_DROP;
1138 so->so_rcv.sb_flags |= SB_DROP;
1139 }
1140
1141 if (error != 0) {
1142 /*
1143 * This may seem like a duplication to the above error
1144 * handling part when we return ECONNABORTED, except
1145 * the following is done while holding the lock since
1146 * the socket has been exposed to the filter(s) earlier.
1147 */
1148 so->so_state &= ~(SS_NOFDREF | SS_COMP);
1149 so->so_head = NULL;
1150 socket_unlock(so, 1);
1151 soclose(so);
1152 /* Propagate socket filter's error code to the caller */
1153 } else {
1154 socket_unlock(so, 1);
1155 }
1156 done:
1157 /* Callee checks for NULL pointer */
1158 sock_freeaddr(remote);
1159 sock_freeaddr(local);
1160 return (error);
1161 }
1162
1163 /*
1164 * Returns: 0 Success
1165 * EOPNOTSUPP Operation not supported on socket
1166 * EISCONN Socket is connected
1167 * <pru_connect>:EADDRNOTAVAIL Address not available.
1168 * <pru_connect>:EINVAL Invalid argument
1169 * <pru_connect>:EAFNOSUPPORT Address family not supported [notdef]
1170 * <pru_connect>:EACCES Permission denied
1171 * <pru_connect>:EADDRINUSE Address in use
1172 * <pru_connect>:EAGAIN Resource unavailable, try again
1173 * <pru_connect>:EPERM Operation not permitted
1174 * <sf_connect_out>:??? [anything a filter writer might set]
1175 */
1176 int
1177 soconnectlock(struct socket *so, struct sockaddr *nam, int dolock)
1178 {
1179 int error;
1180 struct proc *p = current_proc();
1181
1182 if (dolock)
1183 socket_lock(so, 1);
1184
1185 /*
1186 * If this is a listening socket or if this is a previously-accepted
1187 * socket that has been marked as inactive, reject the connect request.
1188 */
1189 if ((so->so_options & SO_ACCEPTCONN) || (so->so_flags & SOF_DEFUNCT)) {
1190 if (dolock)
1191 socket_unlock(so, 1);
1192 return (EOPNOTSUPP);
1193 }
1194
1195 if ((so->so_restrictions & SO_RESTRICT_DENYOUT) != 0) {
1196 if (dolock)
1197 socket_unlock(so, 1);
1198 return (EPERM);
1199 }
1200
1201 /*
1202 * If protocol is connection-based, can only connect once.
1203 * Otherwise, if connected, try to disconnect first.
1204 * This allows user to disconnect by connecting to, e.g.,
1205 * a null address.
1206 */
1207 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1208 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1209 (error = sodisconnectlocked(so)))) {
1210 error = EISCONN;
1211 } else {
1212 /*
1213 * Run connect filter before calling protocol:
1214 * - non-blocking connect returns before completion;
1215 */
1216 struct socket_filter_entry *filter;
1217 int filtered = 0;
1218
1219 error = 0;
1220 for (filter = so->so_filt; filter && (error == 0);
1221 filter = filter->sfe_next_onsocket) {
1222 if (filter->sfe_filter->sf_filter.sf_connect_out) {
1223 if (filtered == 0) {
1224 filtered = 1;
1225 sflt_use(so);
1226 socket_unlock(so, 0);
1227 }
1228 error = filter->sfe_filter->sf_filter.
1229 sf_connect_out(filter->sfe_cookie, so, nam);
1230 }
1231 }
1232 if (filtered != 0) {
1233 socket_lock(so, 0);
1234 sflt_unuse(so);
1235 }
1236
1237 if (error) {
1238 if (error == EJUSTRETURN)
1239 error = 0;
1240 if (dolock)
1241 socket_unlock(so, 1);
1242 return (error);
1243 }
1244
1245 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, p);
1246 }
1247 if (dolock)
1248 socket_unlock(so, 1);
1249 return (error);
1250 }
1251
1252 int
1253 soconnect(struct socket *so, struct sockaddr *nam)
1254 {
1255 return (soconnectlock(so, nam, 1));
1256 }
1257
1258 /*
1259 * Returns: 0 Success
1260 * <pru_connect2>:EINVAL[AF_UNIX]
1261 * <pru_connect2>:EPROTOTYPE[AF_UNIX]
1262 * <pru_connect2>:??? [other protocol families]
1263 *
1264 * Notes: <pru_connect2> is not supported by [TCP].
1265 */
1266 int
1267 soconnect2(struct socket *so1, struct socket *so2)
1268 {
1269 int error;
1270
1271 socket_lock(so1, 1);
1272 if (so2->so_proto->pr_lock)
1273 socket_lock(so2, 1);
1274
1275 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1276
1277 socket_unlock(so1, 1);
1278 if (so2->so_proto->pr_lock)
1279 socket_unlock(so2, 1);
1280 return (error);
1281 }
1282
1283 int
1284 sodisconnectlocked(struct socket *so)
1285 {
1286 int error;
1287
1288 if ((so->so_state & SS_ISCONNECTED) == 0) {
1289 error = ENOTCONN;
1290 goto bad;
1291 }
1292 if (so->so_state & SS_ISDISCONNECTING) {
1293 error = EALREADY;
1294 goto bad;
1295 }
1296
1297 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1298
1299 if (error == 0) {
1300 sflt_notify(so, sock_evt_disconnected, NULL);
1301 }
1302 bad:
1303 return (error);
1304 }
1305
1306 /* Locking version */
1307 int
1308 sodisconnect(struct socket *so)
1309 {
1310 int error;
1311
1312 socket_lock(so, 1);
1313 error = sodisconnectlocked(so);
1314 socket_unlock(so, 1);
1315 return (error);
1316 }
1317
1318 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_DONTWAIT : M_WAIT)
1319
1320 /*
1321 * sosendcheck will lock the socket buffer if it isn't locked and
1322 * verify that there is space for the data being inserted.
1323 *
1324 * Returns: 0 Success
1325 * EPIPE
1326 * sblock:EWOULDBLOCK
1327 * sblock:EINTR
1328 * sbwait:EBADF
1329 * sbwait:EINTR
1330 * [so_error]:???
1331 */
1332 static int
1333 sosendcheck(struct socket *so, struct sockaddr *addr, long resid, long clen,
1334 long atomic, int flags, int *sblocked)
1335 {
1336 int error = 0;
1337 long space;
1338 int assumelock = 0;
1339
1340 restart:
1341 if (*sblocked == 0) {
1342 if ((so->so_snd.sb_flags & SB_LOCK) != 0 &&
1343 so->so_send_filt_thread != 0 &&
1344 so->so_send_filt_thread == current_thread()) {
1345 /*
1346 * We're being called recursively from a filter,
1347 * allow this to continue. Radar 4150520.
1348 * Don't set sblocked because we don't want
1349 * to perform an unlock later.
1350 */
1351 assumelock = 1;
1352 } else {
1353 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1354 if (error) {
1355 return (error);
1356 }
1357 *sblocked = 1;
1358 }
1359 }
1360
1361 /*
1362 * If a send attempt is made on a previously-accepted socket
1363 * that has been marked as inactive (disconnected), reject
1364 * the request.
1365 */
1366 if (so->so_flags & SOF_DEFUNCT)
1367 return (ENOTCONN);
1368
1369 if (so->so_state & SS_CANTSENDMORE)
1370 return (EPIPE);
1371
1372 if (so->so_error) {
1373 error = so->so_error;
1374 so->so_error = 0;
1375 return (error);
1376 }
1377
1378 if ((so->so_state & SS_ISCONNECTED) == 0) {
1379 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1380 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1381 !(resid == 0 && clen != 0))
1382 return (ENOTCONN);
1383 } else if (addr == 0 && !(flags&MSG_HOLD)) {
1384 return ((so->so_proto->pr_flags & PR_CONNREQUIRED) ?
1385 ENOTCONN : EDESTADDRREQ);
1386 }
1387 }
1388 space = sbspace(&so->so_snd);
1389 if (flags & MSG_OOB)
1390 space += 1024;
1391 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1392 clen > so->so_snd.sb_hiwat)
1393 return (EMSGSIZE);
1394 if (space < resid + clen &&
1395 (atomic || space < (long)so->so_snd.sb_lowat || space < clen)) {
1396 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO) ||
1397 assumelock) {
1398 return (EWOULDBLOCK);
1399 }
1400 sbunlock(&so->so_snd, 1);
1401 error = sbwait(&so->so_snd);
1402 if (error) {
1403 return (error);
1404 }
1405 goto restart;
1406 }
1407
1408 return (0);
1409 }
1410
1411 /*
1412 * Send on a socket.
1413 * If send must go all at once and message is larger than
1414 * send buffering, then hard error.
1415 * Lock against other senders.
1416 * If must go all at once and not enough room now, then
1417 * inform user that this would block and do nothing.
1418 * Otherwise, if nonblocking, send as much as possible.
1419 * The data to be sent is described by "uio" if nonzero,
1420 * otherwise by the mbuf chain "top" (which must be null
1421 * if uio is not). Data provided in mbuf chain must be small
1422 * enough to send all at once.
1423 *
1424 * Returns nonzero on error, timeout or signal; callers
1425 * must check for short counts if EINTR/ERESTART are returned.
1426 * Data and control buffers are freed on return.
1427 * Experiment:
1428 * MSG_HOLD: go thru most of sosend(), but just enqueue the mbuf
1429 * MSG_SEND: go thru as for MSG_HOLD on current fragment, then
1430 * point at the mbuf chain being constructed and go from there.
1431 *
1432 * Returns: 0 Success
1433 * EOPNOTSUPP
1434 * EINVAL
1435 * ENOBUFS
1436 * uiomove:EFAULT
1437 * sosendcheck:EPIPE
1438 * sosendcheck:EWOULDBLOCK
1439 * sosendcheck:EINTR
1440 * sosendcheck:EBADF
1441 * sosendcheck:EINTR
1442 * sosendcheck:??? [value from so_error]
1443 * <pru_send>:ECONNRESET[TCP]
1444 * <pru_send>:EINVAL[TCP]
1445 * <pru_send>:ENOBUFS[TCP]
1446 * <pru_send>:EADDRINUSE[TCP]
1447 * <pru_send>:EADDRNOTAVAIL[TCP]
1448 * <pru_send>:EAFNOSUPPORT[TCP]
1449 * <pru_send>:EACCES[TCP]
1450 * <pru_send>:EAGAIN[TCP]
1451 * <pru_send>:EPERM[TCP]
1452 * <pru_send>:EMSGSIZE[TCP]
1453 * <pru_send>:EHOSTUNREACH[TCP]
1454 * <pru_send>:ENETUNREACH[TCP]
1455 * <pru_send>:ENETDOWN[TCP]
1456 * <pru_send>:ENOMEM[TCP]
1457 * <pru_send>:ENOBUFS[TCP]
1458 * <pru_send>:???[TCP] [ignorable: mostly IPSEC/firewall/DLIL]
1459 * <pru_send>:EINVAL[AF_UNIX]
1460 * <pru_send>:EOPNOTSUPP[AF_UNIX]
1461 * <pru_send>:EPIPE[AF_UNIX]
1462 * <pru_send>:ENOTCONN[AF_UNIX]
1463 * <pru_send>:EISCONN[AF_UNIX]
1464 * <pru_send>:???[AF_UNIX] [whatever a filter author chooses]
1465 * <sf_data_out>:??? [whatever a filter author chooses]
1466 *
1467 * Notes: Other <pru_send> returns depend on the protocol family; all
1468 * <sf_data_out> returns depend on what the filter author causes
1469 * their filter to return.
1470 */
1471 int
1472 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1473 struct mbuf *top, struct mbuf *control, int flags)
1474 {
1475 struct mbuf **mp;
1476 register struct mbuf *m, *freelist = NULL;
1477 register long space, len, resid;
1478 int clen = 0, error, dontroute, mlen, sendflags;
1479 int atomic = sosendallatonce(so) || top;
1480 int sblocked = 0;
1481 struct proc *p = current_proc();
1482
1483 if (uio) {
1484 // LP64todo - fix this!
1485 resid = uio_resid(uio);
1486 } else {
1487 resid = top->m_pkthdr.len;
1488 }
1489 KERNEL_DEBUG((DBG_FNC_SOSEND | DBG_FUNC_START), so, resid,
1490 so->so_snd.sb_cc, so->so_snd.sb_lowat, so->so_snd.sb_hiwat);
1491
1492 socket_lock(so, 1);
1493 if (so->so_type != SOCK_STREAM && (flags & MSG_OOB) != 0) {
1494 error = EOPNOTSUPP;
1495 socket_unlock(so, 1);
1496 goto out;
1497 }
1498
1499 /*
1500 * In theory resid should be unsigned.
1501 * However, space must be signed, as it might be less than 0
1502 * if we over-committed, and we must use a signed comparison
1503 * of space and resid. On the other hand, a negative resid
1504 * causes us to loop sending 0-length segments to the protocol.
1505 *
1506 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1507 * type sockets since that's an error.
1508 */
1509 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1510 error = EINVAL;
1511 socket_unlock(so, 1);
1512 goto out;
1513 }
1514
1515 dontroute =
1516 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1517 (so->so_proto->pr_flags & PR_ATOMIC);
1518 if (p)
1519 OSIncrementAtomic(&p->p_stats->p_ru.ru_msgsnd);
1520 if (control)
1521 clen = control->m_len;
1522
1523 do {
1524 error = sosendcheck(so, addr, resid, clen, atomic, flags,
1525 &sblocked);
1526 if (error) {
1527 goto release;
1528 }
1529 mp = &top;
1530 space = sbspace(&so->so_snd) - clen + ((flags & MSG_OOB) ?
1531 1024 : 0);
1532
1533 do {
1534 struct socket_filter_entry *filter;
1535 int filtered;
1536 boolean_t recursive;
1537
1538 if (uio == NULL) {
1539 /*
1540 * Data is prepackaged in "top".
1541 */
1542 resid = 0;
1543 if (flags & MSG_EOR)
1544 top->m_flags |= M_EOR;
1545 } else {
1546 int chainlength;
1547 int bytes_to_copy;
1548 boolean_t jumbocl;
1549
1550 bytes_to_copy = min(resid, space);
1551
1552 if (sosendminchain > 0) {
1553 chainlength = 0;
1554 } else {
1555 chainlength = sosendmaxchain;
1556 }
1557
1558 /*
1559 * Attempt to use larger than system page-size
1560 * clusters for large writes only if there is
1561 * a jumbo cluster pool and if the socket is
1562 * marked accordingly.
1563 */
1564 jumbocl = sosendjcl && njcl > 0 &&
1565 ((so->so_flags & SOF_MULTIPAGES) ||
1566 sosendjcl_ignore_capab);
1567
1568 socket_unlock(so, 0);
1569
1570 do {
1571 int num_needed;
1572 int hdrs_needed = (top == 0) ? 1 : 0;
1573
1574 /*
1575 * try to maintain a local cache of mbuf
1576 * clusters needed to complete this
1577 * write the list is further limited to
1578 * the number that are currently needed
1579 * to fill the socket this mechanism
1580 * allows a large number of mbufs/
1581 * clusters to be grabbed under a single
1582 * mbuf lock... if we can't get any
1583 * clusters, than fall back to trying
1584 * for mbufs if we fail early (or
1585 * miscalcluate the number needed) make
1586 * sure to release any clusters we
1587 * haven't yet consumed.
1588 */
1589 if (freelist == NULL &&
1590 bytes_to_copy > NBPG && jumbocl) {
1591 num_needed =
1592 bytes_to_copy / M16KCLBYTES;
1593
1594 if ((bytes_to_copy -
1595 (num_needed * M16KCLBYTES))
1596 >= MINCLSIZE)
1597 num_needed++;
1598
1599 freelist =
1600 m_getpackets_internal(
1601 (unsigned int *)&num_needed,
1602 hdrs_needed, M_WAIT, 0,
1603 M16KCLBYTES);
1604 /*
1605 * Fall back to 4K cluster size
1606 * if allocation failed
1607 */
1608 }
1609
1610 if (freelist == NULL &&
1611 bytes_to_copy > MCLBYTES) {
1612 num_needed =
1613 bytes_to_copy / NBPG;
1614
1615 if ((bytes_to_copy -
1616 (num_needed * NBPG)) >=
1617 MINCLSIZE)
1618 num_needed++;
1619
1620 freelist =
1621 m_getpackets_internal(
1622 (unsigned int *)&num_needed,
1623 hdrs_needed, M_WAIT, 0,
1624 NBPG);
1625 /*
1626 * Fall back to cluster size
1627 * if allocation failed
1628 */
1629 }
1630
1631 if (freelist == NULL &&
1632 bytes_to_copy > MINCLSIZE) {
1633 num_needed =
1634 bytes_to_copy / MCLBYTES;
1635
1636 if ((bytes_to_copy -
1637 (num_needed * MCLBYTES)) >=
1638 MINCLSIZE)
1639 num_needed++;
1640
1641 freelist =
1642 m_getpackets_internal(
1643 (unsigned int *)&num_needed,
1644 hdrs_needed, M_WAIT, 0,
1645 MCLBYTES);
1646 /*
1647 * Fall back to a single mbuf
1648 * if allocation failed
1649 */
1650 }
1651
1652 if (freelist == NULL) {
1653 if (top == 0)
1654 MGETHDR(freelist,
1655 M_WAIT, MT_DATA);
1656 else
1657 MGET(freelist,
1658 M_WAIT, MT_DATA);
1659
1660 if (freelist == NULL) {
1661 error = ENOBUFS;
1662 socket_lock(so, 0);
1663 goto release;
1664 }
1665 /*
1666 * For datagram protocols,
1667 * leave room for protocol
1668 * headers in first mbuf.
1669 */
1670 if (atomic && top == 0 &&
1671 bytes_to_copy < MHLEN) {
1672 MH_ALIGN(freelist,
1673 bytes_to_copy);
1674 }
1675 }
1676 m = freelist;
1677 freelist = m->m_next;
1678 m->m_next = NULL;
1679
1680 if ((m->m_flags & M_EXT))
1681 mlen = m->m_ext.ext_size;
1682 else if ((m->m_flags & M_PKTHDR))
1683 mlen =
1684 MHLEN - m_leadingspace(m);
1685 else
1686 mlen = MLEN;
1687 len = min(mlen, bytes_to_copy);
1688
1689 chainlength += len;
1690
1691 space -= len;
1692
1693 error = uiomove(mtod(m, caddr_t),
1694 (int)len, uio);
1695
1696 // LP64todo - fix this!
1697 resid = uio_resid(uio);
1698
1699 m->m_len = len;
1700 *mp = m;
1701 top->m_pkthdr.len += len;
1702 if (error)
1703 break;
1704 mp = &m->m_next;
1705 if (resid <= 0) {
1706 if (flags & MSG_EOR)
1707 top->m_flags |= M_EOR;
1708 break;
1709 }
1710 bytes_to_copy = min(resid, space);
1711
1712 } while (space > 0 &&
1713 (chainlength < sosendmaxchain || atomic ||
1714 resid < MINCLSIZE));
1715
1716 socket_lock(so, 0);
1717
1718 if (error)
1719 goto release;
1720 }
1721
1722 if (flags & (MSG_HOLD|MSG_SEND)) {
1723 /* Enqueue for later, go away if HOLD */
1724 register struct mbuf *mb1;
1725 if (so->so_temp && (flags & MSG_FLUSH)) {
1726 m_freem(so->so_temp);
1727 so->so_temp = NULL;
1728 }
1729 if (so->so_temp)
1730 so->so_tail->m_next = top;
1731 else
1732 so->so_temp = top;
1733 mb1 = top;
1734 while (mb1->m_next)
1735 mb1 = mb1->m_next;
1736 so->so_tail = mb1;
1737 if (flags & MSG_HOLD) {
1738 top = NULL;
1739 goto release;
1740 }
1741 top = so->so_temp;
1742 }
1743 if (dontroute)
1744 so->so_options |= SO_DONTROUTE;
1745
1746 /* Compute flags here, for pru_send and NKEs */
1747 sendflags = (flags & MSG_OOB) ? PRUS_OOB :
1748 /*
1749 * If the user set MSG_EOF, the protocol
1750 * understands this flag and nothing left to
1751 * send then use PRU_SEND_EOF instead of PRU_SEND.
1752 */
1753 ((flags & MSG_EOF) &&
1754 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1755 (resid <= 0)) ?
1756 PRUS_EOF :
1757 /* If there is more to send set PRUS_MORETOCOME */
1758 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1759
1760 /*
1761 * Socket filter processing
1762 */
1763 recursive = (so->so_send_filt_thread != NULL);
1764 filtered = 0;
1765 error = 0;
1766 for (filter = so->so_filt; filter && (error == 0);
1767 filter = filter->sfe_next_onsocket) {
1768 if (filter->sfe_filter->sf_filter.sf_data_out) {
1769 int so_flags = 0;
1770 if (filtered == 0) {
1771 filtered = 1;
1772 so->so_send_filt_thread =
1773 current_thread();
1774 sflt_use(so);
1775 socket_unlock(so, 0);
1776 so_flags =
1777 (sendflags & MSG_OOB) ?
1778 sock_data_filt_flag_oob : 0;
1779 }
1780 error = filter->sfe_filter->sf_filter.
1781 sf_data_out(filter->sfe_cookie, so,
1782 addr, &top, &control, so_flags);
1783 }
1784 }
1785
1786 if (filtered) {
1787 /*
1788 * At this point, we've run at least one
1789 * filter. The socket is unlocked as is
1790 * the socket buffer. Clear the recorded
1791 * filter thread only when we are outside
1792 * of a filter's context. This allows for
1793 * a filter to issue multiple inject calls
1794 * from its sf_data_out callback routine.
1795 */
1796 socket_lock(so, 0);
1797 sflt_unuse(so);
1798 if (!recursive)
1799 so->so_send_filt_thread = 0;
1800 if (error) {
1801 if (error == EJUSTRETURN) {
1802 error = 0;
1803 clen = 0;
1804 control = 0;
1805 top = 0;
1806 }
1807
1808 goto release;
1809 }
1810 }
1811 /*
1812 * End Socket filter processing
1813 */
1814
1815 if (error == EJUSTRETURN) {
1816 /* A socket filter handled this data */
1817 error = 0;
1818 } else {
1819 error = (*so->so_proto->pr_usrreqs->pru_send)
1820 (so, sendflags, top, addr, control, p);
1821 }
1822 #ifdef __APPLE__
1823 if (flags & MSG_SEND)
1824 so->so_temp = NULL;
1825 #endif
1826 if (dontroute)
1827 so->so_options &= ~SO_DONTROUTE;
1828
1829 clen = 0;
1830 control = 0;
1831 top = 0;
1832 mp = &top;
1833 if (error)
1834 goto release;
1835 } while (resid && space > 0);
1836 } while (resid);
1837
1838 release:
1839 if (sblocked)
1840 sbunlock(&so->so_snd, 0); /* will unlock socket */
1841 else
1842 socket_unlock(so, 1);
1843 out:
1844 if (top)
1845 m_freem(top);
1846 if (control)
1847 m_freem(control);
1848 if (freelist)
1849 m_freem_list(freelist);
1850
1851 KERNEL_DEBUG(DBG_FNC_SOSEND | DBG_FUNC_END, so, resid, so->so_snd.sb_cc,
1852 space, error);
1853
1854 return (error);
1855 }
1856
1857 /*
1858 * Implement receive operations on a socket.
1859 * We depend on the way that records are added to the sockbuf
1860 * by sbappend*. In particular, each record (mbufs linked through m_next)
1861 * must begin with an address if the protocol so specifies,
1862 * followed by an optional mbuf or mbufs containing ancillary data,
1863 * and then zero or more mbufs of data.
1864 * In order to avoid blocking network interrupts for the entire time here,
1865 * we splx() while doing the actual copy to user space.
1866 * Although the sockbuf is locked, new data may still be appended,
1867 * and thus we must maintain consistency of the sockbuf during that time.
1868 *
1869 * The caller may receive the data as a single mbuf chain by supplying
1870 * an mbuf **mp0 for use in returning the chain. The uio is then used
1871 * only for the count in uio_resid.
1872 *
1873 * Returns: 0 Success
1874 * ENOBUFS
1875 * ENOTCONN
1876 * EWOULDBLOCK
1877 * uiomove:EFAULT
1878 * sblock:EWOULDBLOCK
1879 * sblock:EINTR
1880 * sbwait:EBADF
1881 * sbwait:EINTR
1882 * sodelayed_copy:EFAULT
1883 * <pru_rcvoob>:EINVAL[TCP]
1884 * <pru_rcvoob>:EWOULDBLOCK[TCP]
1885 * <pru_rcvoob>:???
1886 * <pr_domain->dom_externalize>:EMSGSIZE[AF_UNIX]
1887 * <pr_domain->dom_externalize>:ENOBUFS[AF_UNIX]
1888 * <pr_domain->dom_externalize>:???
1889 *
1890 * Notes: Additional return values from calls through <pru_rcvoob> and
1891 * <pr_domain->dom_externalize> depend on protocols other than
1892 * TCP or AF_UNIX, which are documented above.
1893 */
1894 int
1895 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1896 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1897 {
1898 register struct mbuf *m, **mp, *ml = NULL;
1899 register int flags, len, error, offset;
1900 struct protosw *pr = so->so_proto;
1901 struct mbuf *nextrecord;
1902 int moff, type = 0;
1903 // LP64todo - fix this!
1904 int orig_resid = uio_resid(uio);
1905 struct mbuf *free_list;
1906 int delayed_copy_len;
1907 int can_delay;
1908 int need_event;
1909 struct proc *p = current_proc();
1910
1911 // LP64todo - fix this!
1912 KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_START, so, uio_resid(uio),
1913 so->so_rcv.sb_cc, so->so_rcv.sb_lowat, so->so_rcv.sb_hiwat);
1914
1915 socket_lock(so, 1);
1916
1917 #ifdef MORE_LOCKING_DEBUG
1918 if (so->so_usecount == 1)
1919 panic("soreceive: so=%x no other reference on socket\n", so);
1920 #endif
1921 mp = mp0;
1922 if (psa)
1923 *psa = 0;
1924 if (controlp)
1925 *controlp = 0;
1926 if (flagsp)
1927 flags = *flagsp &~ MSG_EOR;
1928 else
1929 flags = 0;
1930
1931 /*
1932 * If a recv attempt is made on a previously-accepted socket
1933 * that has been marked as inactive (disconnected), reject
1934 * the request.
1935 */
1936 if (so->so_flags & SOF_DEFUNCT) {
1937 struct sockbuf *sb = &so->so_rcv;
1938
1939 /*
1940 * This socket should have been disconnected and flushed
1941 * prior to being returned from accept; there should be
1942 * no data on its receive list, so panic otherwise.
1943 */
1944 sb_empty_assert(sb, __func__);
1945 socket_unlock(so, 1);
1946 return (ENOTCONN);
1947 }
1948
1949 /*
1950 * When SO_WANTOOBFLAG is set we try to get out-of-band data
1951 * regardless of the flags argument. Here is the case were
1952 * out-of-band data is not inline.
1953 */
1954 if ((flags & MSG_OOB) ||
1955 ((so->so_options & SO_WANTOOBFLAG) != 0 &&
1956 (so->so_options & SO_OOBINLINE) == 0 &&
1957 (so->so_oobmark || (so->so_state & SS_RCVATMARK)))) {
1958 m = m_get(M_WAIT, MT_DATA);
1959 if (m == NULL) {
1960 socket_unlock(so, 1);
1961 KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END,
1962 ENOBUFS, 0, 0, 0, 0);
1963 return (ENOBUFS);
1964 }
1965 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1966 if (error)
1967 goto bad;
1968 socket_unlock(so, 0);
1969 do {
1970 // LP64todo - fix this!
1971 error = uiomove(mtod(m, caddr_t),
1972 (int)min(uio_resid(uio), m->m_len), uio);
1973 m = m_free(m);
1974 } while (uio_resid(uio) && error == 0 && m);
1975 socket_lock(so, 0);
1976 bad:
1977 if (m)
1978 m_freem(m);
1979 #ifdef __APPLE__
1980 if ((so->so_options & SO_WANTOOBFLAG) != 0) {
1981 if (error == EWOULDBLOCK || error == EINVAL) {
1982 /*
1983 * Let's try to get normal data:
1984 * EWOULDBLOCK: out-of-band data not
1985 * receive yet. EINVAL: out-of-band data
1986 * already read.
1987 */
1988 error = 0;
1989 goto nooob;
1990 } else if (error == 0 && flagsp) {
1991 *flagsp |= MSG_OOB;
1992 }
1993 }
1994 socket_unlock(so, 1);
1995 KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,
1996 0, 0, 0, 0);
1997 #endif
1998 return (error);
1999 }
2000 nooob:
2001 if (mp)
2002 *mp = (struct mbuf *)0;
2003 if (so->so_state & SS_ISCONFIRMING && uio_resid(uio))
2004 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
2005
2006
2007 free_list = (struct mbuf *)0;
2008 delayed_copy_len = 0;
2009 restart:
2010 #ifdef MORE_LOCKING_DEBUG
2011 if (so->so_usecount <= 1)
2012 printf("soreceive: sblock so=%p ref=%d on socket\n",
2013 so, so->so_usecount);
2014 #endif
2015 /*
2016 * See if the socket has been closed (SS_NOFDREF|SS_CANTRCVMORE)
2017 * and if so just return to the caller. This could happen when
2018 * soreceive() is called by a socket upcall function during the
2019 * time the socket is freed. The socket buffer would have been
2020 * locked across the upcall, therefore we cannot put this thread
2021 * to sleep (else we will deadlock) or return EWOULDBLOCK (else
2022 * we may livelock), because the lock on the socket buffer will
2023 * only be released when the upcall routine returns to its caller.
2024 * Because the socket has been officially closed, there can be
2025 * no further read on it.
2026 */
2027 if ((so->so_state & (SS_NOFDREF | SS_CANTRCVMORE)) ==
2028 (SS_NOFDREF | SS_CANTRCVMORE)) {
2029 socket_unlock(so, 1);
2030 return (0);
2031 }
2032
2033 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
2034 if (error) {
2035 socket_unlock(so, 1);
2036 KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,
2037 0, 0, 0, 0);
2038 return (error);
2039 }
2040
2041 m = so->so_rcv.sb_mb;
2042 /*
2043 * If we have less data than requested, block awaiting more
2044 * (subject to any timeout) if:
2045 * 1. the current count is less than the low water mark, or
2046 * 2. MSG_WAITALL is set, and it is possible to do the entire
2047 * receive operation at once if we block (resid <= hiwat).
2048 * 3. MSG_DONTWAIT is not set
2049 * If MSG_WAITALL is set but resid is larger than the receive buffer,
2050 * we have to do the receive in sections, and thus risk returning
2051 * a short count if a timeout or signal occurs after we start.
2052 */
2053 if (m == 0 || (((flags & MSG_DONTWAIT) == 0 &&
2054 so->so_rcv.sb_cc < uio_resid(uio)) &&
2055 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
2056 ((flags & MSG_WAITALL) && uio_resid(uio) <= so->so_rcv.sb_hiwat)) &&
2057 m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
2058 /*
2059 * Panic if we notice inconsistencies in the socket's
2060 * receive list; both sb_mb and sb_cc should correctly
2061 * reflect the contents of the list, otherwise we may
2062 * end up with false positives during select() or poll()
2063 * which could put the application in a bad state.
2064 */
2065 if (m == NULL && so->so_rcv.sb_cc != 0)
2066 panic("soreceive corrupted so_rcv: m %p cc %lu",
2067 m, so->so_rcv.sb_cc);
2068
2069 if (so->so_error) {
2070 if (m)
2071 goto dontblock;
2072 error = so->so_error;
2073 if ((flags & MSG_PEEK) == 0)
2074 so->so_error = 0;
2075 goto release;
2076 }
2077 if (so->so_state & SS_CANTRCVMORE) {
2078 if (m)
2079 goto dontblock;
2080 else
2081 goto release;
2082 }
2083 for (; m; m = m->m_next)
2084 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
2085 m = so->so_rcv.sb_mb;
2086 goto dontblock;
2087 }
2088 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
2089 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
2090 error = ENOTCONN;
2091 goto release;
2092 }
2093 if (uio_resid(uio) == 0)
2094 goto release;
2095 if ((so->so_state & SS_NBIO) ||
2096 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2097 error = EWOULDBLOCK;
2098 goto release;
2099 }
2100 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 1");
2101 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 1");
2102 sbunlock(&so->so_rcv, 1);
2103 #if EVEN_MORE_LOCKING_DEBUG
2104 if (socket_debug)
2105 printf("Waiting for socket data\n");
2106 #endif
2107
2108 error = sbwait(&so->so_rcv);
2109 #if EVEN_MORE_LOCKING_DEBUG
2110 if (socket_debug)
2111 printf("SORECEIVE - sbwait returned %d\n", error);
2112 #endif
2113 if (so->so_usecount < 1)
2114 panic("soreceive: after 2nd sblock so=%p ref=%d on "
2115 "socket\n", so, so->so_usecount);
2116 if (error) {
2117 socket_unlock(so, 1);
2118 KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, error,
2119 0, 0, 0, 0);
2120 return (error);
2121 }
2122 goto restart;
2123 }
2124 dontblock:
2125 #ifndef __APPLE__
2126 if (uio->uio_procp)
2127 uio->uio_procp->p_stats->p_ru.ru_msgrcv++;
2128 #else /* __APPLE__ */
2129 /*
2130 * 2207985
2131 * This should be uio->uio-procp; however, some callers of this
2132 * function use auto variables with stack garbage, and fail to
2133 * fill out the uio structure properly.
2134 */
2135 if (p)
2136 OSIncrementAtomic(&p->p_stats->p_ru.ru_msgrcv);
2137 #endif /* __APPLE__ */
2138 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
2139 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
2140 nextrecord = m->m_nextpkt;
2141 if ((pr->pr_flags & PR_ADDR) && m->m_type == MT_SONAME) {
2142 KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
2143 #if CONFIG_MACF_SOCKET_SUBSET
2144 /*
2145 * Call the MAC framework for policy checking if we're in
2146 * the user process context and the socket isn't connected.
2147 */
2148 if (p != kernproc && !(so->so_state & SS_ISCONNECTED)) {
2149 struct mbuf *m0 = m;
2150 /*
2151 * Dequeue this record (temporarily) from the receive
2152 * list since we're about to drop the socket's lock
2153 * where a new record may arrive and be appended to
2154 * the list. Upon MAC policy failure, the record
2155 * will be freed. Otherwise, we'll add it back to
2156 * the head of the list. We cannot rely on SB_LOCK
2157 * because append operation uses the socket's lock.
2158 */
2159 do {
2160 m->m_nextpkt = NULL;
2161 sbfree(&so->so_rcv, m);
2162 m = m->m_next;
2163 } while (m != NULL);
2164 m = m0;
2165 so->so_rcv.sb_mb = nextrecord;
2166 SB_EMPTY_FIXUP(&so->so_rcv);
2167 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1a");
2168 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1a");
2169 socket_unlock(so, 0);
2170 if (mac_socket_check_received(proc_ucred(p), so,
2171 mtod(m, struct sockaddr *)) != 0) {
2172 /*
2173 * MAC policy failure; free this record and
2174 * process the next record (or block until
2175 * one is available). We have adjusted sb_cc
2176 * and sb_mbcnt above so there is no need to
2177 * call sbfree() again.
2178 */
2179 do {
2180 m = m_free(m);
2181 } while (m != NULL);
2182 /*
2183 * Clear SB_LOCK but don't unlock the socket.
2184 * Process the next record or wait for one.
2185 */
2186 socket_lock(so, 0);
2187 sbunlock(&so->so_rcv, 1);
2188 goto restart;
2189 }
2190 socket_lock(so, 0);
2191 /*
2192 * Re-adjust the socket receive list and re-enqueue
2193 * the record in front of any packets which may have
2194 * been appended while we dropped the lock.
2195 */
2196 for (m = m0; m->m_next != NULL; m = m->m_next)
2197 sballoc(&so->so_rcv, m);
2198 sballoc(&so->so_rcv, m);
2199 if (so->so_rcv.sb_mb == NULL) {
2200 so->so_rcv.sb_lastrecord = m0;
2201 so->so_rcv.sb_mbtail = m;
2202 }
2203 m = m0;
2204 nextrecord = m->m_nextpkt = so->so_rcv.sb_mb;
2205 so->so_rcv.sb_mb = m;
2206 SBLASTRECORDCHK(&so->so_rcv, "soreceive 1b");
2207 SBLASTMBUFCHK(&so->so_rcv, "soreceive 1b");
2208 }
2209 #endif /* CONFIG_MACF_SOCKET_SUBSET */
2210 orig_resid = 0;
2211 if (psa) {
2212 *psa = dup_sockaddr(mtod(m, struct sockaddr *),
2213 mp0 == 0);
2214 if ((*psa == 0) && (flags & MSG_NEEDSA)) {
2215 error = EWOULDBLOCK;
2216 goto release;
2217 }
2218 }
2219 if (flags & MSG_PEEK) {
2220 m = m->m_next;
2221 } else {
2222 sbfree(&so->so_rcv, m);
2223 if (m->m_next == 0 && so->so_rcv.sb_cc != 0)
2224 panic("soreceive: about to create invalid "
2225 "socketbuf");
2226 MFREE(m, so->so_rcv.sb_mb);
2227 m = so->so_rcv.sb_mb;
2228 if (m != NULL) {
2229 m->m_nextpkt = nextrecord;
2230 } else {
2231 so->so_rcv.sb_mb = nextrecord;
2232 SB_EMPTY_FIXUP(&so->so_rcv);
2233 }
2234 }
2235 }
2236
2237 /*
2238 * Process one or more MT_CONTROL mbufs present before any data mbufs
2239 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2240 * just copy the data; if !MSG_PEEK, we call into the protocol to
2241 * perform externalization.
2242 */
2243 if (m != NULL && m->m_type == MT_CONTROL) {
2244 struct mbuf *cm = NULL, *cmn;
2245 struct mbuf **cme = &cm;
2246 struct sockbuf *sb_rcv = &so->so_rcv;
2247
2248 /*
2249 * Externalizing the control messages would require us to
2250 * drop the socket's lock below. Once we re-acquire the
2251 * lock, the mbuf chain might change. In order to preserve
2252 * consistency, we unlink all control messages from the
2253 * first mbuf chain in one shot and link them separately
2254 * onto a different chain.
2255 */
2256 do {
2257 if (flags & MSG_PEEK) {
2258 if (controlp != NULL) {
2259 *controlp = m_copy(m, 0, m->m_len);
2260 controlp = &(*controlp)->m_next;
2261 }
2262 m = m->m_next;
2263 } else {
2264 m->m_nextpkt = NULL;
2265 sbfree(sb_rcv, m);
2266 sb_rcv->sb_mb = m->m_next;
2267 m->m_next = NULL;
2268 *cme = m;
2269 cme = &(*cme)->m_next;
2270 m = sb_rcv->sb_mb;
2271 }
2272 } while (m != NULL && m->m_type == MT_CONTROL);
2273
2274 if (!(flags & MSG_PEEK)) {
2275 if (sb_rcv->sb_mb != NULL) {
2276 sb_rcv->sb_mb->m_nextpkt = nextrecord;
2277 } else {
2278 sb_rcv->sb_mb = nextrecord;
2279 SB_EMPTY_FIXUP(sb_rcv);
2280 }
2281 if (nextrecord == NULL)
2282 sb_rcv->sb_lastrecord = m;
2283 }
2284
2285 SBLASTRECORDCHK(&so->so_rcv, "soreceive ctl");
2286 SBLASTMBUFCHK(&so->so_rcv, "soreceive ctl");
2287
2288 while (cm != NULL) {
2289 int cmsg_type;
2290
2291 cmn = cm->m_next;
2292 cm->m_next = NULL;
2293 cmsg_type = mtod(cm, struct cmsghdr *)->cmsg_type;
2294
2295 /*
2296 * Call the protocol to externalize SCM_RIGHTS message
2297 * and return the modified message to the caller upon
2298 * success. Otherwise, all other control messages are
2299 * returned unmodified to the caller. Note that we
2300 * only get into this loop if MSG_PEEK is not set.
2301 */
2302 if (pr->pr_domain->dom_externalize != NULL &&
2303 cmsg_type == SCM_RIGHTS) {
2304 /*
2305 * Release socket lock: see 3903171. This
2306 * would also allow more records to be appended
2307 * to the socket buffer. We still have SB_LOCK
2308 * set on it, so we can be sure that the head
2309 * of the mbuf chain won't change.
2310 */
2311 socket_unlock(so, 0);
2312 error = (*pr->pr_domain->dom_externalize)(cm);
2313 socket_lock(so, 0);
2314 } else {
2315 error = 0;
2316 }
2317
2318 if (controlp != NULL && error == 0) {
2319 *controlp = cm;
2320 controlp = &(*controlp)->m_next;
2321 orig_resid = 0;
2322 } else {
2323 (void) m_free(cm);
2324 }
2325 cm = cmn;
2326 }
2327 orig_resid = 0;
2328 if (sb_rcv->sb_mb != NULL)
2329 nextrecord = sb_rcv->sb_mb->m_nextpkt;
2330 else
2331 nextrecord = NULL;
2332 }
2333
2334 if (m != NULL) {
2335 if (!(flags & MSG_PEEK)) {
2336 /*
2337 * We get here because m points to an mbuf following
2338 * any MT_SONAME or MT_CONTROL mbufs which have been
2339 * processed above. In any case, m should be pointing
2340 * to the head of the mbuf chain, and the nextrecord
2341 * should be either NULL or equal to m->m_nextpkt.
2342 * See comments above about SB_LOCK.
2343 */
2344 if (m != so->so_rcv.sb_mb || m->m_nextpkt != nextrecord)
2345 panic("soreceive: post-control !sync so=%p "
2346 "m=%p nextrecord=%p\n", so, m, nextrecord);
2347
2348 if (nextrecord == NULL)
2349 so->so_rcv.sb_lastrecord = m;
2350 }
2351 type = m->m_type;
2352 if (type == MT_OOBDATA)
2353 flags |= MSG_OOB;
2354 } else {
2355 if (!(flags & MSG_PEEK)) {
2356 so->so_rcv.sb_mb = nextrecord;
2357 SB_EMPTY_FIXUP(&so->so_rcv);
2358 }
2359 }
2360 SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
2361 SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
2362
2363 moff = 0;
2364 offset = 0;
2365
2366 if (!(flags & MSG_PEEK) && uio_resid(uio) > sorecvmincopy)
2367 can_delay = 1;
2368 else
2369 can_delay = 0;
2370
2371 need_event = 0;
2372
2373 while (m && (uio_resid(uio) - delayed_copy_len) > 0 && error == 0) {
2374 if (m->m_type == MT_OOBDATA) {
2375 if (type != MT_OOBDATA)
2376 break;
2377 } else if (type == MT_OOBDATA) {
2378 break;
2379 }
2380 /*
2381 * Make sure to allways set MSG_OOB event when getting
2382 * out of band data inline.
2383 */
2384 if ((so->so_options & SO_WANTOOBFLAG) != 0 &&
2385 (so->so_options & SO_OOBINLINE) != 0 &&
2386 (so->so_state & SS_RCVATMARK) != 0) {
2387 flags |= MSG_OOB;
2388 }
2389 so->so_state &= ~SS_RCVATMARK;
2390 // LP64todo - fix this!
2391 len = uio_resid(uio) - delayed_copy_len;
2392 if (so->so_oobmark && len > so->so_oobmark - offset)
2393 len = so->so_oobmark - offset;
2394 if (len > m->m_len - moff)
2395 len = m->m_len - moff;
2396 /*
2397 * If mp is set, just pass back the mbufs.
2398 * Otherwise copy them out via the uio, then free.
2399 * Sockbuf must be consistent here (points to current mbuf,
2400 * it points to next record) when we drop priority;
2401 * we must note any additions to the sockbuf when we
2402 * block interrupts again.
2403 */
2404 if (mp == 0) {
2405 SBLASTRECORDCHK(&so->so_rcv, "soreceive uiomove");
2406 SBLASTMBUFCHK(&so->so_rcv, "soreceive uiomove");
2407 if (can_delay && len == m->m_len) {
2408 /*
2409 * only delay the copy if we're consuming the
2410 * mbuf and we're NOT in MSG_PEEK mode
2411 * and we have enough data to make it worthwile
2412 * to drop and retake the lock... can_delay
2413 * reflects the state of the 2 latter
2414 * constraints moff should always be zero
2415 * in these cases
2416 */
2417 delayed_copy_len += len;
2418 } else {
2419 if (delayed_copy_len) {
2420 error = sodelayed_copy(so, uio,
2421 &free_list, &delayed_copy_len);
2422
2423 if (error) {
2424 goto release;
2425 }
2426 /*
2427 * can only get here if MSG_PEEK is not
2428 * set therefore, m should point at the
2429 * head of the rcv queue; if it doesn't,
2430 * it means something drastically
2431 * changed while we were out from behind
2432 * the lock in sodelayed_copy. perhaps
2433 * a RST on the stream. in any event,
2434 * the stream has been interrupted. it's
2435 * probably best just to return whatever
2436 * data we've moved and let the caller
2437 * sort it out...
2438 */
2439 if (m != so->so_rcv.sb_mb) {
2440 break;
2441 }
2442 }
2443 socket_unlock(so, 0);
2444 error = uiomove(mtod(m, caddr_t) + moff,
2445 (int)len, uio);
2446 socket_lock(so, 0);
2447
2448 if (error)
2449 goto release;
2450 }
2451 } else {
2452 uio_setresid(uio, (uio_resid(uio) - len));
2453 }
2454 if (len == m->m_len - moff) {
2455 if (m->m_flags & M_EOR)
2456 flags |= MSG_EOR;
2457 if (flags & MSG_PEEK) {
2458 m = m->m_next;
2459 moff = 0;
2460 } else {
2461 nextrecord = m->m_nextpkt;
2462 sbfree(&so->so_rcv, m);
2463 m->m_nextpkt = NULL;
2464
2465 if (mp) {
2466 *mp = m;
2467 mp = &m->m_next;
2468 so->so_rcv.sb_mb = m = m->m_next;
2469 *mp = (struct mbuf *)0;
2470 } else {
2471 if (free_list == NULL)
2472 free_list = m;
2473 else
2474 ml->m_next = m;
2475 ml = m;
2476 so->so_rcv.sb_mb = m = m->m_next;
2477 ml->m_next = 0;
2478 }
2479 if (m != NULL) {
2480 m->m_nextpkt = nextrecord;
2481 if (nextrecord == NULL)
2482 so->so_rcv.sb_lastrecord = m;
2483 } else {
2484 so->so_rcv.sb_mb = nextrecord;
2485 SB_EMPTY_FIXUP(&so->so_rcv);
2486 }
2487 SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
2488 SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
2489 }
2490 } else {
2491 if (flags & MSG_PEEK) {
2492 moff += len;
2493 } else {
2494 if (mp)
2495 *mp = m_copym(m, 0, len, M_WAIT);
2496 m->m_data += len;
2497 m->m_len -= len;
2498 so->so_rcv.sb_cc -= len;
2499 }
2500 }
2501 if (so->so_oobmark) {
2502 if ((flags & MSG_PEEK) == 0) {
2503 so->so_oobmark -= len;
2504 if (so->so_oobmark == 0) {
2505 so->so_state |= SS_RCVATMARK;
2506 /*
2507 * delay posting the actual event until
2508 * after any delayed copy processing
2509 * has finished
2510 */
2511 need_event = 1;
2512 break;
2513 }
2514 } else {
2515 offset += len;
2516 if (offset == so->so_oobmark)
2517 break;
2518 }
2519 }
2520 if (flags & MSG_EOR)
2521 break;
2522 /*
2523 * If the MSG_WAITALL or MSG_WAITSTREAM flag is set
2524 * (for non-atomic socket), we must not quit until
2525 * "uio->uio_resid == 0" or an error termination.
2526 * If a signal/timeout occurs, return with a short
2527 * count but without error. Keep sockbuf locked
2528 * against other readers.
2529 */
2530 while (flags & (MSG_WAITALL|MSG_WAITSTREAM) && m == 0 &&
2531 (uio_resid(uio) - delayed_copy_len) > 0 &&
2532 !sosendallatonce(so) && !nextrecord) {
2533 if (so->so_error || so->so_state & SS_CANTRCVMORE)
2534 goto release;
2535
2536 /*
2537 * Depending on the protocol (e.g. TCP), the following
2538 * might cause the socket lock to be dropped and later
2539 * be reacquired, and more data could have arrived and
2540 * have been appended to the receive socket buffer by
2541 * the time it returns. Therefore, we only sleep in
2542 * sbwait() below if and only if the socket buffer is
2543 * empty, in order to avoid a false sleep.
2544 */
2545 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb &&
2546 (((struct inpcb *)so->so_pcb)->inp_state !=
2547 INPCB_STATE_DEAD))
2548 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2549
2550 SBLASTRECORDCHK(&so->so_rcv, "soreceive sbwait 2");
2551 SBLASTMBUFCHK(&so->so_rcv, "soreceive sbwait 2");
2552
2553 if (so->so_rcv.sb_mb == NULL && sbwait(&so->so_rcv)) {
2554 error = 0;
2555 goto release;
2556 }
2557 /*
2558 * have to wait until after we get back from the sbwait
2559 * to do the copy because we will drop the lock if we
2560 * have enough data that has been delayed... by dropping
2561 * the lock we open up a window allowing the netisr
2562 * thread to process the incoming packets and to change
2563 * the state of this socket... we're issuing the sbwait
2564 * because the socket is empty and we're expecting the
2565 * netisr thread to wake us up when more packets arrive;
2566 * if we allow that processing to happen and then sbwait
2567 * we could stall forever with packets sitting in the
2568 * socket if no further packets arrive from the remote
2569 * side.
2570 *
2571 * we want to copy before we've collected all the data
2572 * to satisfy this request to allow the copy to overlap
2573 * the incoming packet processing on an MP system
2574 */
2575 if (delayed_copy_len > sorecvmincopy &&
2576 (delayed_copy_len > (so->so_rcv.sb_hiwat / 2))) {
2577 error = sodelayed_copy(so, uio,
2578 &free_list, &delayed_copy_len);
2579
2580 if (error)
2581 goto release;
2582 }
2583 m = so->so_rcv.sb_mb;
2584 if (m) {
2585 nextrecord = m->m_nextpkt;
2586 }
2587 }
2588 }
2589 #ifdef MORE_LOCKING_DEBUG
2590 if (so->so_usecount <= 1)
2591 panic("soreceive: after big while so=%p ref=%d on socket\n",
2592 so, so->so_usecount);
2593 #endif
2594
2595 if (m && pr->pr_flags & PR_ATOMIC) {
2596 #ifdef __APPLE__
2597 if (so->so_options & SO_DONTTRUNC) {
2598 flags |= MSG_RCVMORE;
2599 } else {
2600 #endif
2601 flags |= MSG_TRUNC;
2602 if ((flags & MSG_PEEK) == 0)
2603 (void) sbdroprecord(&so->so_rcv);
2604 #ifdef __APPLE__
2605 }
2606 #endif
2607 }
2608
2609 /*
2610 * pru_rcvd below (for TCP) may cause more data to be received
2611 * if the socket lock is dropped prior to sending the ACK; some
2612 * legacy OpenTransport applications don't handle this well
2613 * (if it receives less data than requested while MSG_HAVEMORE
2614 * is set), and so we set the flag now based on what we know
2615 * prior to calling pru_rcvd.
2616 */
2617 if ((so->so_options & SO_WANTMORE) && so->so_rcv.sb_cc > 0)
2618 flags |= MSG_HAVEMORE;
2619
2620 if ((flags & MSG_PEEK) == 0) {
2621 if (m == 0) {
2622 so->so_rcv.sb_mb = nextrecord;
2623 /*
2624 * First part is an inline SB_EMPTY_FIXUP(). Second
2625 * part makes sure sb_lastrecord is up-to-date if
2626 * there is still data in the socket buffer.
2627 */
2628 if (so->so_rcv.sb_mb == NULL) {
2629 so->so_rcv.sb_mbtail = NULL;
2630 so->so_rcv.sb_lastrecord = NULL;
2631 } else if (nextrecord->m_nextpkt == NULL) {
2632 so->so_rcv.sb_lastrecord = nextrecord;
2633 }
2634 }
2635 SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
2636 SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
2637 if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
2638 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2639 }
2640 #ifdef __APPLE__
2641 if (delayed_copy_len) {
2642 error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len);
2643
2644 if (error)
2645 goto release;
2646 }
2647 if (free_list) {
2648 m_freem_list((struct mbuf *)free_list);
2649 free_list = (struct mbuf *)0;
2650 }
2651 if (need_event)
2652 postevent(so, 0, EV_OOB);
2653 #endif
2654 if (orig_resid == uio_resid(uio) && orig_resid &&
2655 (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
2656 sbunlock(&so->so_rcv, 1);
2657 goto restart;
2658 }
2659
2660 if (flagsp)
2661 *flagsp |= flags;
2662 release:
2663 #ifdef MORE_LOCKING_DEBUG
2664 if (so->so_usecount <= 1)
2665 panic("soreceive: release so=%p ref=%d on socket\n",
2666 so, so->so_usecount);
2667 #endif
2668 if (delayed_copy_len) {
2669 error = sodelayed_copy(so, uio, &free_list, &delayed_copy_len);
2670 }
2671 if (free_list) {
2672 m_freem_list((struct mbuf *)free_list);
2673 }
2674 sbunlock(&so->so_rcv, 0); /* will unlock socket */
2675
2676 // LP64todo - fix this!
2677 KERNEL_DEBUG(DBG_FNC_SORECEIVE | DBG_FUNC_END, so, uio_resid(uio),
2678 so->so_rcv.sb_cc, 0, error);
2679
2680 return (error);
2681 }
2682
2683 /*
2684 * Returns: 0 Success
2685 * uiomove:EFAULT
2686 */
2687 static int
2688 sodelayed_copy(struct socket *so, struct uio *uio, struct mbuf **free_list,
2689 int *resid)
2690 {
2691 int error = 0;
2692 struct mbuf *m;
2693
2694 m = *free_list;
2695
2696 socket_unlock(so, 0);
2697
2698 while (m && error == 0) {
2699
2700 error = uiomove(mtod(m, caddr_t), (int)m->m_len, uio);
2701
2702 m = m->m_next;
2703 }
2704 m_freem_list(*free_list);
2705
2706 *free_list = (struct mbuf *)NULL;
2707 *resid = 0;
2708
2709 socket_lock(so, 0);
2710
2711 return (error);
2712 }
2713
2714
2715 /*
2716 * Returns: 0 Success
2717 * EINVAL
2718 * ENOTCONN
2719 * <pru_shutdown>:EINVAL
2720 * <pru_shutdown>:EADDRNOTAVAIL[TCP]
2721 * <pru_shutdown>:ENOBUFS[TCP]
2722 * <pru_shutdown>:EMSGSIZE[TCP]
2723 * <pru_shutdown>:EHOSTUNREACH[TCP]
2724 * <pru_shutdown>:ENETUNREACH[TCP]
2725 * <pru_shutdown>:ENETDOWN[TCP]
2726 * <pru_shutdown>:ENOMEM[TCP]
2727 * <pru_shutdown>:EACCES[TCP]
2728 * <pru_shutdown>:EMSGSIZE[TCP]
2729 * <pru_shutdown>:ENOBUFS[TCP]
2730 * <pru_shutdown>:???[TCP] [ignorable: mostly IPSEC/firewall/DLIL]
2731 * <pru_shutdown>:??? [other protocol families]
2732 */
2733 int
2734 soshutdown(struct socket *so, int how)
2735 {
2736 int error;
2737
2738 switch (how) {
2739 case SHUT_RD:
2740 case SHUT_WR:
2741 case SHUT_RDWR:
2742 socket_lock(so, 1);
2743 if ((so->so_state &
2744 (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING)) == 0) {
2745 error = ENOTCONN;
2746 } else {
2747 error = soshutdownlock(so, how);
2748 }
2749 socket_unlock(so, 1);
2750 break;
2751 default:
2752 error = EINVAL;
2753 break;
2754 }
2755
2756 return (error);
2757 }
2758
2759 int
2760 soshutdownlock(struct socket *so, int how)
2761 {
2762 struct protosw *pr = so->so_proto;
2763 int error = 0;
2764
2765 sflt_notify(so, sock_evt_shutdown, &how);
2766
2767 if (how != SHUT_WR) {
2768 if ((so->so_state & SS_CANTRCVMORE) != 0) {
2769 /* read already shut down */
2770 error = ENOTCONN;
2771 goto done;
2772 }
2773 sorflush(so);
2774 postevent(so, 0, EV_RCLOSED);
2775 }
2776 if (how != SHUT_RD) {
2777 if ((so->so_state & SS_CANTSENDMORE) != 0) {
2778 /* write already shut down */
2779 error = ENOTCONN;
2780 goto done;
2781 }
2782 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2783 postevent(so, 0, EV_WCLOSED);
2784 }
2785 done:
2786 KERNEL_DEBUG(DBG_FNC_SOSHUTDOWN | DBG_FUNC_END, 0, 0, 0, 0, 0);
2787 return (error);
2788 }
2789
2790 void
2791 sorflush(struct socket *so)
2792 {
2793 register struct sockbuf *sb = &so->so_rcv;
2794 register struct protosw *pr = so->so_proto;
2795 struct sockbuf asb;
2796
2797 #ifdef MORE_LOCKING_DEBUG
2798 lck_mtx_t *mutex_held;
2799
2800 if (so->so_proto->pr_getlock != NULL)
2801 mutex_held = (*so->so_proto->pr_getlock)(so, 0);
2802 else
2803 mutex_held = so->so_proto->pr_domain->dom_mtx;
2804 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
2805 #endif
2806
2807 sflt_notify(so, sock_evt_flush_read, NULL);
2808
2809 sb->sb_flags |= SB_NOINTR;
2810 (void) sblock(sb, M_WAIT);
2811 socantrcvmore(so);
2812 sbunlock(sb, 1);
2813 #ifdef __APPLE__
2814 selthreadclear(&sb->sb_sel);
2815 #endif
2816 asb = *sb;
2817 bzero((caddr_t)sb, sizeof (*sb));
2818 sb->sb_so = so; /* reestablish link to socket */
2819 if (asb.sb_flags & SB_KNOTE) {
2820 sb->sb_sel.si_note = asb.sb_sel.si_note;
2821 sb->sb_flags = SB_KNOTE;
2822 }
2823 if (asb.sb_flags & SB_DROP)
2824 sb->sb_flags |= SB_DROP;
2825 if (asb.sb_flags & SB_UNIX)
2826 sb->sb_flags |= SB_UNIX;
2827 if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose) {
2828 boolean_t unp = (pr->pr_domain->dom_dispose == unp_dispose);
2829 /*
2830 * Currently AF_UNIX domain uses a global domain mutex;
2831 * unp_dispose() may end up calling soclose() on another
2832 * AF_UNIX socket and therefore the lock must not be held
2833 * across the call.
2834 */
2835 if (unp)
2836 socket_unlock(so, 0);
2837 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2838 if (unp)
2839 socket_lock(so, 0);
2840 }
2841 sbrelease(&asb);
2842 }
2843
2844 /*
2845 * Perhaps this routine, and sooptcopyout(), below, ought to come in
2846 * an additional variant to handle the case where the option value needs
2847 * to be some kind of integer, but not a specific size.
2848 * In addition to their use here, these functions are also called by the
2849 * protocol-level pr_ctloutput() routines.
2850 *
2851 * Returns: 0 Success
2852 * EINVAL
2853 * copyin:EFAULT
2854 */
2855 int
2856 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2857 {
2858 size_t valsize;
2859
2860 /*
2861 * If the user gives us more than we wanted, we ignore it,
2862 * but if we don't get the minimum length the caller
2863 * wants, we return EINVAL. On success, sopt->sopt_valsize
2864 * is set to however much we actually retrieved.
2865 */
2866 if ((valsize = sopt->sopt_valsize) < minlen)
2867 return (EINVAL);
2868 if (valsize > len)
2869 sopt->sopt_valsize = valsize = len;
2870
2871 if (sopt->sopt_p != 0)
2872 return (copyin(sopt->sopt_val, buf, valsize));
2873
2874 bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), buf, valsize);
2875 return (0);
2876 }
2877
2878 /*
2879 * sooptcopyin_timeval
2880 * Copy in a timeval value into tv_p, and take into account whether the
2881 * the calling process is 64-bit or 32-bit. Moved the sanity checking
2882 * code here so that we can verify the 64-bit tv_sec value before we lose
2883 * the top 32-bits assigning tv64.tv_sec to tv_p->tv_sec.
2884 */
2885 static int
2886 sooptcopyin_timeval(struct sockopt *sopt, struct timeval * tv_p)
2887 {
2888 int error;
2889
2890 if (proc_is64bit(sopt->sopt_p)) {
2891 struct timeval64 tv64;
2892
2893 if (sopt->sopt_valsize < sizeof(tv64)) {
2894 return (EINVAL);
2895 }
2896 sopt->sopt_valsize = sizeof(tv64);
2897 error = copyin(sopt->sopt_val, &tv64, sizeof(tv64));
2898 if (error != 0) {
2899 return (error);
2900 }
2901 if (tv64.tv_sec < 0 || tv64.tv_sec > LONG_MAX
2902 || tv64.tv_usec < 0 || tv64.tv_usec >= 1000000) {
2903 return (EDOM);
2904 }
2905 tv_p->tv_sec = tv64.tv_sec;
2906 tv_p->tv_usec = tv64.tv_usec;
2907 } else {
2908 if (sopt->sopt_valsize < sizeof(*tv_p)) {
2909 return (EINVAL);
2910 }
2911 sopt->sopt_valsize = sizeof(*tv_p);
2912 if (sopt->sopt_p != 0) {
2913 error = copyin(sopt->sopt_val, tv_p, sizeof(*tv_p));
2914 if (error != 0) {
2915 return (error);
2916 }
2917 } else {
2918 bcopy(CAST_DOWN(caddr_t, sopt->sopt_val), tv_p,
2919 sizeof(*tv_p));
2920 }
2921 if (tv_p->tv_sec < 0 || tv_p->tv_sec > LONG_MAX
2922 || tv_p->tv_usec < 0 || tv_p->tv_usec >= 1000000) {
2923 return (EDOM);
2924 }
2925 }
2926 return (0);
2927 }
2928
2929 /*
2930 * Returns: 0 Success
2931 * EINVAL
2932 * ENOPROTOOPT
2933 * ENOBUFS
2934 * EDOM
2935 * sooptcopyin:EINVAL
2936 * sooptcopyin:EFAULT
2937 * sooptcopyin_timeval:EINVAL
2938 * sooptcopyin_timeval:EFAULT
2939 * sooptcopyin_timeval:EDOM
2940 * <pr_ctloutput>:EOPNOTSUPP[AF_UNIX]
2941 * <pr_ctloutput>:???w
2942 * sflt_attach_private:??? [whatever a filter author chooses]
2943 * <sf_setoption>:??? [whatever a filter author chooses]
2944 *
2945 * Notes: Other <pru_listen> returns depend on the protocol family; all
2946 * <sf_listen> returns depend on what the filter author causes
2947 * their filter to return.
2948 */
2949 int
2950 sosetopt(struct socket *so, struct sockopt *sopt)
2951 {
2952 int error, optval;
2953 struct linger l;
2954 struct timeval tv;
2955 struct socket_filter_entry *filter;
2956 int filtered = 0;
2957 #if CONFIG_MACF_SOCKET
2958 struct mac extmac;
2959 #endif /* MAC_SOCKET */
2960
2961 socket_lock(so, 1);
2962 if ((so->so_state & (SS_CANTRCVMORE | SS_CANTSENDMORE))
2963 == (SS_CANTRCVMORE | SS_CANTSENDMORE)) {
2964 /* the socket has been shutdown, no more sockopt's */
2965 error = EINVAL;
2966 goto bad;
2967 }
2968
2969 if (sopt->sopt_dir != SOPT_SET) {
2970 sopt->sopt_dir = SOPT_SET;
2971 }
2972
2973 error = 0;
2974 for (filter = so->so_filt; filter && (error == 0);
2975 filter = filter->sfe_next_onsocket) {
2976 if (filter->sfe_filter->sf_filter.sf_setoption) {
2977 if (filtered == 0) {
2978 filtered = 1;
2979 sflt_use(so);
2980 socket_unlock(so, 0);
2981 }
2982 error = filter->sfe_filter->sf_filter.
2983 sf_setoption(filter->sfe_cookie, so, sopt);
2984 }
2985 }
2986
2987 if (filtered != 0) {
2988 socket_lock(so, 0);
2989 sflt_unuse(so);
2990
2991 if (error) {
2992 if (error == EJUSTRETURN)
2993 error = 0;
2994 goto bad;
2995 }
2996 }
2997
2998 error = 0;
2999 if (sopt->sopt_level != SOL_SOCKET) {
3000 if (so->so_proto && so->so_proto->pr_ctloutput) {
3001 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3002 socket_unlock(so, 1);
3003 return (error);
3004 }
3005 error = ENOPROTOOPT;
3006 } else {
3007 switch (sopt->sopt_name) {
3008 case SO_LINGER:
3009 case SO_LINGER_SEC:
3010 error = sooptcopyin(sopt, &l, sizeof (l), sizeof (l));
3011 if (error)
3012 goto bad;
3013
3014 so->so_linger = (sopt->sopt_name == SO_LINGER) ?
3015 l.l_linger : l.l_linger * hz;
3016 if (l.l_onoff)
3017 so->so_options |= SO_LINGER;
3018 else
3019 so->so_options &= ~SO_LINGER;
3020 break;
3021
3022 case SO_DEBUG:
3023 case SO_KEEPALIVE:
3024 case SO_DONTROUTE:
3025 case SO_USELOOPBACK:
3026 case SO_BROADCAST:
3027 case SO_REUSEADDR:
3028 case SO_REUSEPORT:
3029 case SO_OOBINLINE:
3030 case SO_TIMESTAMP:
3031 #ifdef __APPLE__
3032 case SO_DONTTRUNC:
3033 case SO_WANTMORE:
3034 case SO_WANTOOBFLAG:
3035 #endif
3036 error = sooptcopyin(sopt, &optval, sizeof (optval),
3037 sizeof (optval));
3038 if (error)
3039 goto bad;
3040 if (optval)
3041 so->so_options |= sopt->sopt_name;
3042 else
3043 so->so_options &= ~sopt->sopt_name;
3044 break;
3045
3046 case SO_SNDBUF:
3047 case SO_RCVBUF:
3048 case SO_SNDLOWAT:
3049 case SO_RCVLOWAT:
3050 error = sooptcopyin(sopt, &optval, sizeof (optval),
3051 sizeof (optval));
3052 if (error)
3053 goto bad;
3054
3055 /*
3056 * Values < 1 make no sense for any of these
3057 * options, so disallow them.
3058 */
3059 if (optval < 1) {
3060 error = EINVAL;
3061 goto bad;
3062 }
3063
3064 switch (sopt->sopt_name) {
3065 case SO_SNDBUF:
3066 case SO_RCVBUF:
3067 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
3068 &so->so_snd : &so->so_rcv,
3069 (u_long) optval) == 0) {
3070 error = ENOBUFS;
3071 goto bad;
3072 }
3073 if (sopt->sopt_name == SO_SNDBUF)
3074 so->so_snd.sb_flags |= SB_USRSIZE;
3075 else
3076 so->so_rcv.sb_flags |= SB_USRSIZE;
3077 break;
3078
3079 /*
3080 * Make sure the low-water is never greater than
3081 * the high-water.
3082 */
3083 case SO_SNDLOWAT:
3084 so->so_snd.sb_lowat =
3085 (optval > so->so_snd.sb_hiwat) ?
3086 so->so_snd.sb_hiwat : optval;
3087 break;
3088 case SO_RCVLOWAT:
3089 so->so_rcv.sb_lowat =
3090 (optval > so->so_rcv.sb_hiwat) ?
3091 so->so_rcv.sb_hiwat : optval;
3092 break;
3093 }
3094 break;
3095
3096 case SO_SNDTIMEO:
3097 case SO_RCVTIMEO:
3098 error = sooptcopyin_timeval(sopt, &tv);
3099 if (error)
3100 goto bad;
3101
3102 switch (sopt->sopt_name) {
3103 case SO_SNDTIMEO:
3104 so->so_snd.sb_timeo = tv;
3105 break;
3106 case SO_RCVTIMEO:
3107 so->so_rcv.sb_timeo = tv;
3108 break;
3109 }
3110 break;
3111
3112 case SO_NKE:
3113 {
3114 struct so_nke nke;
3115
3116 error = sooptcopyin(sopt, &nke, sizeof (nke),
3117 sizeof (nke));
3118 if (error)
3119 goto bad;
3120
3121 error = sflt_attach_private(so, NULL,
3122 nke.nke_handle, 1);
3123 break;
3124 }
3125
3126 case SO_NOSIGPIPE:
3127 error = sooptcopyin(sopt, &optval, sizeof (optval),
3128 sizeof (optval));
3129 if (error)
3130 goto bad;
3131 if (optval)
3132 so->so_flags |= SOF_NOSIGPIPE;
3133 else
3134 so->so_flags &= ~SOF_NOSIGPIPE;
3135
3136 break;
3137
3138 case SO_NOADDRERR:
3139 error = sooptcopyin(sopt, &optval, sizeof (optval),
3140 sizeof (optval));
3141 if (error)
3142 goto bad;
3143 if (optval)
3144 so->so_flags |= SOF_NOADDRAVAIL;
3145 else
3146 so->so_flags &= ~SOF_NOADDRAVAIL;
3147
3148 break;
3149
3150 case SO_REUSESHAREUID:
3151 error = sooptcopyin(sopt, &optval, sizeof (optval),
3152 sizeof (optval));
3153 if (error)
3154 goto bad;
3155 if (optval)
3156 so->so_flags |= SOF_REUSESHAREUID;
3157 else
3158 so->so_flags &= ~SOF_REUSESHAREUID;
3159 break;
3160 #ifdef __APPLE_API_PRIVATE
3161 case SO_NOTIFYCONFLICT:
3162 if (kauth_cred_issuser(kauth_cred_get()) == 0) {
3163 error = EPERM;
3164 goto bad;
3165 }
3166 error = sooptcopyin(sopt, &optval, sizeof (optval),
3167 sizeof (optval));
3168 if (error)
3169 goto bad;
3170 if (optval)
3171 so->so_flags |= SOF_NOTIFYCONFLICT;
3172 else
3173 so->so_flags &= ~SOF_NOTIFYCONFLICT;
3174 break;
3175 #endif
3176 case SO_RESTRICTIONS:
3177 if (kauth_cred_issuser(kauth_cred_get()) == 0) {
3178 error = EPERM;
3179 goto bad;
3180 }
3181 error = sooptcopyin(sopt, &optval, sizeof (optval),
3182 sizeof (optval));
3183 if (error)
3184 goto bad;
3185 so->so_restrictions = (optval & (SO_RESTRICT_DENYIN |
3186 SO_RESTRICT_DENYOUT | SO_RESTRICT_DENYSET));
3187 break;
3188
3189 case SO_LABEL:
3190 #if CONFIG_MACF_SOCKET
3191 if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac),
3192 sizeof (extmac))) != 0)
3193 goto bad;
3194
3195 error = mac_setsockopt_label(proc_ucred(sopt->sopt_p),
3196 so, &extmac);
3197 #else
3198 error = EOPNOTSUPP;
3199 #endif /* MAC_SOCKET */
3200 break;
3201
3202 #ifdef __APPLE_API_PRIVATE
3203 case SO_UPCALLCLOSEWAIT:
3204 error = sooptcopyin(sopt, &optval, sizeof (optval),
3205 sizeof (optval));
3206 if (error)
3207 goto bad;
3208 if (optval)
3209 so->so_flags |= SOF_UPCALLCLOSEWAIT;
3210 else
3211 so->so_flags &= ~SOF_UPCALLCLOSEWAIT;
3212 break;
3213 #endif
3214
3215 default:
3216 error = ENOPROTOOPT;
3217 break;
3218 }
3219 if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
3220 (void) ((*so->so_proto->pr_ctloutput)(so, sopt));
3221 }
3222 }
3223 bad:
3224 socket_unlock(so, 1);
3225 return (error);
3226 }
3227
3228 /* Helper routines for getsockopt */
3229 int
3230 sooptcopyout(struct sockopt *sopt, void *buf, size_t len)
3231 {
3232 int error;
3233 size_t valsize;
3234
3235 error = 0;
3236
3237 /*
3238 * Documented get behavior is that we always return a value,
3239 * possibly truncated to fit in the user's buffer.
3240 * Traditional behavior is that we always tell the user
3241 * precisely how much we copied, rather than something useful
3242 * like the total amount we had available for her.
3243 * Note that this interface is not idempotent; the entire answer must
3244 * generated ahead of time.
3245 */
3246 valsize = min(len, sopt->sopt_valsize);
3247 sopt->sopt_valsize = valsize;
3248 if (sopt->sopt_val != USER_ADDR_NULL) {
3249 if (sopt->sopt_p != 0)
3250 error = copyout(buf, sopt->sopt_val, valsize);
3251 else
3252 bcopy(buf, CAST_DOWN(caddr_t, sopt->sopt_val), valsize);
3253 }
3254 return (error);
3255 }
3256
3257 static int
3258 sooptcopyout_timeval(struct sockopt *sopt, const struct timeval * tv_p)
3259 {
3260 int error;
3261 size_t len;
3262 struct timeval64 tv64;
3263 const void * val;
3264 size_t valsize;
3265
3266 error = 0;
3267 if (proc_is64bit(sopt->sopt_p)) {
3268 len = sizeof(struct timeval64);
3269 tv64.tv_sec = tv_p->tv_sec;
3270 tv64.tv_usec = tv_p->tv_usec;
3271 val = &tv64;
3272 } else {
3273 len = sizeof(struct timeval);
3274 val = tv_p;
3275 }
3276 valsize = min(len, sopt->sopt_valsize);
3277 sopt->sopt_valsize = valsize;
3278 if (sopt->sopt_val != USER_ADDR_NULL) {
3279 if (sopt->sopt_p != 0)
3280 error = copyout(val, sopt->sopt_val, valsize);
3281 else
3282 bcopy(val, CAST_DOWN(caddr_t, sopt->sopt_val), valsize);
3283 }
3284 return (error);
3285 }
3286
3287 /*
3288 * Return: 0 Success
3289 * ENOPROTOOPT
3290 * <pr_ctloutput>:EOPNOTSUPP[AF_UNIX]
3291 * <pr_ctloutput>:???
3292 * <sf_getoption>:???
3293 */
3294 int
3295 sogetopt(struct socket *so, struct sockopt *sopt)
3296 {
3297 int error, optval;
3298 struct linger l;
3299 struct timeval tv;
3300 struct socket_filter_entry *filter;
3301 int filtered = 0;
3302 #if CONFIG_MACF_SOCKET
3303 struct mac extmac;
3304 #endif /* MAC_SOCKET */
3305
3306 if (sopt->sopt_dir != SOPT_GET) {
3307 sopt->sopt_dir = SOPT_GET;
3308 }
3309
3310 socket_lock(so, 1);
3311
3312 error = 0;
3313 for (filter = so->so_filt; filter && (error == 0);
3314 filter = filter->sfe_next_onsocket) {
3315 if (filter->sfe_filter->sf_filter.sf_getoption) {
3316 if (filtered == 0) {
3317 filtered = 1;
3318 sflt_use(so);
3319 socket_unlock(so, 0);
3320 }
3321 error = filter->sfe_filter->sf_filter.
3322 sf_getoption(filter->sfe_cookie, so, sopt);
3323 }
3324 }
3325 if (filtered != 0) {
3326 socket_lock(so, 0);
3327 sflt_unuse(so);
3328
3329 if (error) {
3330 if (error == EJUSTRETURN)
3331 error = 0;
3332 socket_unlock(so, 1);
3333 return (error);
3334 }
3335 }
3336
3337 error = 0;
3338 if (sopt->sopt_level != SOL_SOCKET) {
3339 if (so->so_proto && so->so_proto->pr_ctloutput) {
3340 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3341 socket_unlock(so, 1);
3342 return (error);
3343 } else {
3344 socket_unlock(so, 1);
3345 return (ENOPROTOOPT);
3346 }
3347 } else {
3348 switch (sopt->sopt_name) {
3349 case SO_LINGER:
3350 case SO_LINGER_SEC:
3351 l.l_onoff = so->so_options & SO_LINGER;
3352 l.l_linger = (sopt->sopt_name == SO_LINGER) ?
3353 so->so_linger : so->so_linger / hz;
3354 error = sooptcopyout(sopt, &l, sizeof (l));
3355 break;
3356
3357 case SO_USELOOPBACK:
3358 case SO_DONTROUTE:
3359 case SO_DEBUG:
3360 case SO_KEEPALIVE:
3361 case SO_REUSEADDR:
3362 case SO_REUSEPORT:
3363 case SO_BROADCAST:
3364 case SO_OOBINLINE:
3365 case SO_TIMESTAMP:
3366 #ifdef __APPLE__
3367 case SO_DONTTRUNC:
3368 case SO_WANTMORE:
3369 case SO_WANTOOBFLAG:
3370 #endif
3371 optval = so->so_options & sopt->sopt_name;
3372 integer:
3373 error = sooptcopyout(sopt, &optval, sizeof (optval));
3374 break;
3375
3376 case SO_TYPE:
3377 optval = so->so_type;
3378 goto integer;
3379
3380 #ifdef __APPLE__
3381 case SO_NREAD:
3382 if (so->so_proto->pr_flags & PR_ATOMIC) {
3383 int pkt_total;
3384 struct mbuf *m1;
3385
3386 pkt_total = 0;
3387 m1 = so->so_rcv.sb_mb;
3388 while (m1) {
3389 if (m1->m_type == MT_DATA || m1->m_type == MT_HEADER ||
3390 m1->m_type == MT_OOBDATA)
3391 pkt_total += m1->m_len;
3392 m1 = m1->m_next;
3393 }
3394 optval = pkt_total;
3395 } else {
3396 optval = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3397 }
3398 goto integer;
3399
3400 case SO_NWRITE:
3401 optval = so->so_snd.sb_cc;
3402 goto integer;
3403 #endif
3404 case SO_ERROR:
3405 optval = so->so_error;
3406 so->so_error = 0;
3407 goto integer;
3408
3409 case SO_SNDBUF:
3410 optval = so->so_snd.sb_hiwat;
3411 goto integer;
3412
3413 case SO_RCVBUF:
3414 optval = so->so_rcv.sb_hiwat;
3415 goto integer;
3416
3417 case SO_SNDLOWAT:
3418 optval = so->so_snd.sb_lowat;
3419 goto integer;
3420
3421 case SO_RCVLOWAT:
3422 optval = so->so_rcv.sb_lowat;
3423 goto integer;
3424
3425 case SO_SNDTIMEO:
3426 case SO_RCVTIMEO:
3427 tv = (sopt->sopt_name == SO_SNDTIMEO ?
3428 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3429
3430 error = sooptcopyout_timeval(sopt, &tv);
3431 break;
3432
3433 case SO_NOSIGPIPE:
3434 optval = (so->so_flags & SOF_NOSIGPIPE);
3435 goto integer;
3436
3437 case SO_NOADDRERR:
3438 optval = (so->so_flags & SOF_NOADDRAVAIL);
3439 goto integer;
3440
3441 case SO_REUSESHAREUID:
3442 optval = (so->so_flags & SOF_REUSESHAREUID);
3443 goto integer;
3444
3445 #ifdef __APPLE_API_PRIVATE
3446 case SO_NOTIFYCONFLICT:
3447 optval = (so->so_flags & SOF_NOTIFYCONFLICT);
3448 goto integer;
3449 #endif
3450 case SO_RESTRICTIONS:
3451 optval = so->so_restrictions & (SO_RESTRICT_DENYIN |
3452 SO_RESTRICT_DENYOUT | SO_RESTRICT_DENYSET);
3453 goto integer;
3454
3455 case SO_LABEL:
3456 #if CONFIG_MACF_SOCKET
3457 if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac),
3458 sizeof (extmac))) != 0 ||
3459 (error = mac_socket_label_get(proc_ucred(
3460 sopt->sopt_p), so, &extmac)) != 0)
3461 break;
3462
3463 error = sooptcopyout(sopt, &extmac, sizeof (extmac));
3464 #else
3465 error = EOPNOTSUPP;
3466 #endif /* MAC_SOCKET */
3467 break;
3468
3469 case SO_PEERLABEL:
3470 #if CONFIG_MACF_SOCKET
3471 if ((error = sooptcopyin(sopt, &extmac, sizeof (extmac),
3472 sizeof (extmac))) != 0 ||
3473 (error = mac_socketpeer_label_get(proc_ucred(
3474 sopt->sopt_p), so, &extmac)) != 0)
3475 break;
3476
3477 error = sooptcopyout(sopt, &extmac, sizeof (extmac));
3478 #else
3479 error = EOPNOTSUPP;
3480 #endif /* MAC_SOCKET */
3481 break;
3482
3483 #ifdef __APPLE_API_PRIVATE
3484 case SO_UPCALLCLOSEWAIT:
3485 optval = (so->so_flags & SOF_UPCALLCLOSEWAIT);
3486 goto integer;
3487 #endif
3488
3489 default:
3490 error = ENOPROTOOPT;
3491 break;
3492 }
3493 socket_unlock(so, 1);
3494 return (error);
3495 }
3496 }
3497
3498 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
3499 int
3500 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3501 {
3502 struct mbuf *m, *m_prev;
3503 int sopt_size = sopt->sopt_valsize;
3504
3505 if (sopt_size > MAX_SOOPTGETM_SIZE)
3506 return (EMSGSIZE);
3507
3508 MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_DATA);
3509 if (m == 0)
3510 return (ENOBUFS);
3511 if (sopt_size > MLEN) {
3512 MCLGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT);
3513 if ((m->m_flags & M_EXT) == 0) {
3514 m_free(m);
3515 return (ENOBUFS);
3516 }
3517 m->m_len = min(MCLBYTES, sopt_size);
3518 } else {
3519 m->m_len = min(MLEN, sopt_size);
3520 }
3521 sopt_size -= m->m_len;
3522 *mp = m;
3523 m_prev = m;
3524
3525 while (sopt_size) {
3526 MGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT, MT_DATA);
3527 if (m == 0) {
3528 m_freem(*mp);
3529 return (ENOBUFS);
3530 }
3531 if (sopt_size > MLEN) {
3532 MCLGET(m, sopt->sopt_p ? M_WAIT : M_DONTWAIT);
3533 if ((m->m_flags & M_EXT) == 0) {
3534 m_freem(*mp);
3535 return (ENOBUFS);
3536 }
3537 m->m_len = min(MCLBYTES, sopt_size);
3538 } else {
3539 m->m_len = min(MLEN, sopt_size);
3540 }
3541 sopt_size -= m->m_len;
3542 m_prev->m_next = m;
3543 m_prev = m;
3544 }
3545 return (0);
3546 }
3547
3548 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
3549 int
3550 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3551 {
3552 struct mbuf *m0 = m;
3553
3554 if (sopt->sopt_val == USER_ADDR_NULL)
3555 return (0);
3556 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3557 if (sopt->sopt_p != NULL) {
3558 int error;
3559
3560 error = copyin(sopt->sopt_val, mtod(m, char *),
3561 m->m_len);
3562 if (error != 0) {
3563 m_freem(m0);
3564 return (error);
3565 }
3566 } else {
3567 bcopy(CAST_DOWN(caddr_t, sopt->sopt_val),
3568 mtod(m, char *), m->m_len);
3569 }
3570 sopt->sopt_valsize -= m->m_len;
3571 sopt->sopt_val += m->m_len;
3572 m = m->m_next;
3573 }
3574 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3575 panic("soopt_mcopyin");
3576 return (0);
3577 }
3578
3579 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
3580 int
3581 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3582 {
3583 struct mbuf *m0 = m;
3584 size_t valsize = 0;
3585
3586 if (sopt->sopt_val == USER_ADDR_NULL)
3587 return (0);
3588 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3589 if (sopt->sopt_p != NULL) {
3590 int error;
3591
3592 error = copyout(mtod(m, char *), sopt->sopt_val,
3593 m->m_len);
3594 if (error != 0) {
3595 m_freem(m0);
3596 return (error);
3597 }
3598 } else {
3599 bcopy(mtod(m, char *),
3600 CAST_DOWN(caddr_t, sopt->sopt_val), m->m_len);
3601 }
3602 sopt->sopt_valsize -= m->m_len;
3603 sopt->sopt_val += m->m_len;
3604 valsize += m->m_len;
3605 m = m->m_next;
3606 }
3607 if (m != NULL) {
3608 /* enough soopt buffer should be given from user-land */
3609 m_freem(m0);
3610 return (EINVAL);
3611 }
3612 sopt->sopt_valsize = valsize;
3613 return (0);
3614 }
3615
3616 void
3617 sohasoutofband(struct socket *so)
3618 {
3619
3620 if (so->so_pgid < 0)
3621 gsignal(-so->so_pgid, SIGURG);
3622 else if (so->so_pgid > 0)
3623 proc_signal(so->so_pgid, SIGURG);
3624 selwakeup(&so->so_rcv.sb_sel);
3625 }
3626
3627 int
3628 sopoll(struct socket *so, int events, __unused kauth_cred_t cred, void * wql)
3629 {
3630 struct proc *p = current_proc();
3631 int revents = 0;
3632
3633 socket_lock(so, 1);
3634
3635 if (events & (POLLIN | POLLRDNORM))
3636 if (soreadable(so))
3637 revents |= events & (POLLIN | POLLRDNORM);
3638
3639 if (events & (POLLOUT | POLLWRNORM))
3640 if (sowriteable(so))
3641 revents |= events & (POLLOUT | POLLWRNORM);
3642
3643 if (events & (POLLPRI | POLLRDBAND))
3644 if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
3645 revents |= events & (POLLPRI | POLLRDBAND);
3646
3647 if (revents == 0) {
3648 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3649 /*
3650 * Darwin sets the flag first,
3651 * BSD calls selrecord first
3652 */
3653 so->so_rcv.sb_flags |= SB_SEL;
3654 selrecord(p, &so->so_rcv.sb_sel, wql);
3655 }
3656
3657 if (events & (POLLOUT | POLLWRNORM)) {
3658 /*
3659 * Darwin sets the flag first,
3660 * BSD calls selrecord first
3661 */
3662 so->so_snd.sb_flags |= SB_SEL;
3663 selrecord(p, &so->so_snd.sb_sel, wql);
3664 }
3665 }
3666
3667 socket_unlock(so, 1);
3668 return (revents);
3669 }
3670
3671 int
3672 soo_kqfilter(__unused struct fileproc *fp, struct knote *kn,
3673 __unused struct proc *p)
3674 {
3675 struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data;
3676 struct sockbuf *sb;
3677
3678 socket_lock(so, 1);
3679
3680 #if CONFIG_MACF_SOCKET
3681 if (mac_socket_check_kqfilter(proc_ucred(p), kn, so) != 0) {
3682 socket_unlock(so, 1);
3683 return (1);
3684 }
3685 #endif /* MAC_SOCKET */
3686
3687 switch (kn->kn_filter) {
3688 case EVFILT_READ:
3689 if (so->so_options & SO_ACCEPTCONN)
3690 kn->kn_fop = &solisten_filtops;
3691 else
3692 kn->kn_fop = &soread_filtops;
3693 sb = &so->so_rcv;
3694 break;
3695 case EVFILT_WRITE:
3696 kn->kn_fop = &sowrite_filtops;
3697 sb = &so->so_snd;
3698 break;
3699 default:
3700 socket_unlock(so, 1);
3701 return (1);
3702 }
3703
3704 if (KNOTE_ATTACH(&sb->sb_sel.si_note, kn))
3705 sb->sb_flags |= SB_KNOTE;
3706 socket_unlock(so, 1);
3707 return (0);
3708 }
3709
3710 static void
3711 filt_sordetach(struct knote *kn)
3712 {
3713 struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data;
3714
3715 socket_lock(so, 1);
3716 if (so->so_rcv.sb_flags & SB_KNOTE)
3717 if (KNOTE_DETACH(&so->so_rcv.sb_sel.si_note, kn))
3718 so->so_rcv.sb_flags &= ~SB_KNOTE;
3719 socket_unlock(so, 1);
3720 }
3721
3722 /*ARGSUSED*/
3723 static int
3724 filt_soread(struct knote *kn, long hint)
3725 {
3726 struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data;
3727
3728 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3729 socket_lock(so, 1);
3730
3731 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3732
3733 if (so->so_oobmark) {
3734 if (kn->kn_flags & EV_OOBAND) {
3735 kn->kn_data -= so->so_oobmark;
3736 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3737 socket_unlock(so, 1);
3738 return (1);
3739 }
3740 kn->kn_data = so->so_oobmark;
3741 kn->kn_flags |= EV_OOBAND;
3742 } else {
3743 if (so->so_state & SS_CANTRCVMORE) {
3744 kn->kn_flags |= EV_EOF;
3745 kn->kn_fflags = so->so_error;
3746 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3747 socket_unlock(so, 1);
3748 return (1);
3749 }
3750 }
3751
3752 if (so->so_state & SS_RCVATMARK) {
3753 if (kn->kn_flags & EV_OOBAND) {
3754 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3755 socket_unlock(so, 1);
3756 return (1);
3757 }
3758 kn->kn_flags |= EV_OOBAND;
3759 } else if (kn->kn_flags & EV_OOBAND) {
3760 kn->kn_data = 0;
3761 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3762 socket_unlock(so, 1);
3763 return (0);
3764 }
3765
3766 if (so->so_error) { /* temporary udp error */
3767 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3768 socket_unlock(so, 1);
3769 return (1);
3770 }
3771
3772 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3773 socket_unlock(so, 1);
3774
3775 return ((kn->kn_flags & EV_OOBAND) ||
3776 kn->kn_data >= ((kn->kn_sfflags & NOTE_LOWAT) ?
3777 kn->kn_sdata : so->so_rcv.sb_lowat));
3778 }
3779
3780 static void
3781 filt_sowdetach(struct knote *kn)
3782 {
3783 struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data;
3784 socket_lock(so, 1);
3785
3786 if (so->so_snd.sb_flags & SB_KNOTE)
3787 if (KNOTE_DETACH(&so->so_snd.sb_sel.si_note, kn))
3788 so->so_snd.sb_flags &= ~SB_KNOTE;
3789 socket_unlock(so, 1);
3790 }
3791
3792 /*ARGSUSED*/
3793 static int
3794 filt_sowrite(struct knote *kn, long hint)
3795 {
3796 struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data;
3797
3798 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3799 socket_lock(so, 1);
3800
3801 kn->kn_data = sbspace(&so->so_snd);
3802 if (so->so_state & SS_CANTSENDMORE) {
3803 kn->kn_flags |= EV_EOF;
3804 kn->kn_fflags = so->so_error;
3805 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3806 socket_unlock(so, 1);
3807 return (1);
3808 }
3809 if (so->so_error) { /* temporary udp error */
3810 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3811 socket_unlock(so, 1);
3812 return (1);
3813 }
3814 if (((so->so_state & SS_ISCONNECTED) == 0) &&
3815 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
3816 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3817 socket_unlock(so, 1);
3818 return (0);
3819 }
3820 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3821 socket_unlock(so, 1);
3822 if (kn->kn_sfflags & NOTE_LOWAT)
3823 return (kn->kn_data >= kn->kn_sdata);
3824 return (kn->kn_data >= so->so_snd.sb_lowat);
3825 }
3826
3827 /*ARGSUSED*/
3828 static int
3829 filt_solisten(struct knote *kn, long hint)
3830 {
3831 struct socket *so = (struct socket *)kn->kn_fp->f_fglob->fg_data;
3832 int isempty;
3833
3834 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3835 socket_lock(so, 1);
3836 kn->kn_data = so->so_qlen;
3837 isempty = ! TAILQ_EMPTY(&so->so_comp);
3838 if ((hint & SO_FILT_HINT_LOCKED) == 0)
3839 socket_unlock(so, 1);
3840 return (isempty);
3841 }
3842
3843
3844 int
3845 socket_lock(struct socket *so, int refcount)
3846 {
3847 int error = 0, lr_saved;
3848
3849 lr_saved = (unsigned int) __builtin_return_address(0);
3850
3851 if (so->so_proto->pr_lock) {
3852 error = (*so->so_proto->pr_lock)(so, refcount, lr_saved);
3853 } else {
3854 #ifdef MORE_LOCKING_DEBUG
3855 lck_mtx_assert(so->so_proto->pr_domain->dom_mtx,
3856 LCK_MTX_ASSERT_NOTOWNED);
3857 #endif
3858 lck_mtx_lock(so->so_proto->pr_domain->dom_mtx);
3859 if (refcount)
3860 so->so_usecount++;
3861 so->lock_lr[so->next_lock_lr] = (u_int32_t)lr_saved;
3862 so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX;
3863 }
3864
3865 return (error);
3866 }
3867
3868 int
3869 socket_unlock(struct socket *so, int refcount)
3870 {
3871 int error = 0, lr_saved;
3872 lck_mtx_t *mutex_held;
3873
3874 lr_saved = (unsigned int) __builtin_return_address(0);
3875
3876 if (so->so_proto == NULL)
3877 panic("socket_unlock null so_proto so=%p\n", so);
3878
3879 if (so && so->so_proto->pr_unlock) {
3880 error = (*so->so_proto->pr_unlock)(so, refcount, lr_saved);
3881 } else {
3882 mutex_held = so->so_proto->pr_domain->dom_mtx;
3883 #ifdef MORE_LOCKING_DEBUG
3884 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
3885 #endif
3886 so->unlock_lr[so->next_unlock_lr] = (u_int32_t)lr_saved;
3887 so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX;
3888
3889 if (refcount) {
3890 if (so->so_usecount <= 0)
3891 panic("socket_unlock: bad refcount so=%p "
3892 "value=%d\n", so, so->so_usecount);
3893 so->so_usecount--;
3894 if (so->so_usecount == 0) {
3895 sofreelastref(so, 1);
3896 }
3897 }
3898 lck_mtx_unlock(mutex_held);
3899 }
3900
3901 return (error);
3902 }
3903
3904 /* Called with socket locked, will unlock socket */
3905 void
3906 sofree(struct socket *so)
3907 {
3908
3909 lck_mtx_t *mutex_held;
3910 if (so->so_proto->pr_getlock != NULL)
3911 mutex_held = (*so->so_proto->pr_getlock)(so, 0);
3912 else
3913 mutex_held = so->so_proto->pr_domain->dom_mtx;
3914 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED);
3915
3916 sofreelastref(so, 0);
3917 }
3918
3919 void
3920 soreference(struct socket *so)
3921 {
3922 socket_lock(so, 1); /* locks & take one reference on socket */
3923 socket_unlock(so, 0); /* unlock only */
3924 }
3925
3926 void
3927 sodereference(struct socket *so)
3928 {
3929 socket_lock(so, 0);
3930 socket_unlock(so, 1);
3931 }
3932
3933 /*
3934 * Set or clear SOF_MULTIPAGES on the socket to enable or disable the
3935 * possibility of using jumbo clusters. Caller must ensure to hold
3936 * the socket lock.
3937 */
3938 void
3939 somultipages(struct socket *so, boolean_t set)
3940 {
3941 if (set)
3942 so->so_flags |= SOF_MULTIPAGES;
3943 else
3944 so->so_flags &= ~SOF_MULTIPAGES;
3945 }
mirror of XNU