1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2008 Robert N. M. Watson
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
33 */
34
35 /*
36 * Comments on the socket life cycle:
37 *
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
40 *
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
43 *
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
47 *
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
52 *
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
57 * case.
58 *
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
61 * sockets.
62 *
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
66 * listen socket.
67 *
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
71 *
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
82 *
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
86 * interface.
87 *
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
95 */
96
97 #include <sys/cdefs.h>
98 __FBSDID("$FreeBSD$");
99
100 #include "opt_inet.h"
101 #include "opt_inet6.h"
102 #include "opt_mac.h"
103 #include "opt_zero.h"
104 #include "opt_compat.h"
105
106 #include <sys/param.h>
107 #include <sys/systm.h>
108 #include <sys/fcntl.h>
109 #include <sys/limits.h>
110 #include <sys/lock.h>
111 #include <sys/mac.h>
112 #include <sys/malloc.h>
113 #include <sys/mbuf.h>
114 #include <sys/mutex.h>
115 #include <sys/domain.h>
116 #include <sys/file.h> /* for struct knote */
117 #include <sys/kernel.h>
118 #include <sys/event.h>
119 #include <sys/eventhandler.h>
120 #include <sys/poll.h>
121 #include <sys/proc.h>
122 #include <sys/protosw.h>
123 #include <sys/socket.h>
124 #include <sys/socketvar.h>
125 #include <sys/resourcevar.h>
126 #include <net/route.h>
127 #include <sys/signalvar.h>
128 #include <sys/stat.h>
129 #include <sys/sx.h>
130 #include <sys/sysctl.h>
131 #include <sys/uio.h>
132 #include <sys/jail.h>
133
134 #include <security/mac/mac_framework.h>
135
136 #include <vm/uma.h>
137
138 #ifdef COMPAT_IA32
139 #include <sys/mount.h>
140 #include <compat/freebsd32/freebsd32.h>
141
142 extern struct sysentvec ia32_freebsd_sysvec;
143 #endif
144
145 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
146 int flags);
147
148 static void filt_sordetach(struct knote *kn);
149 static int filt_soread(struct knote *kn, long hint);
150 static void filt_sowdetach(struct knote *kn);
151 static int filt_sowrite(struct knote *kn, long hint);
152 static int filt_solisten(struct knote *kn, long hint);
153
154 static struct filterops solisten_filtops =
155 { 1, NULL, filt_sordetach, filt_solisten };
156 static struct filterops soread_filtops =
157 { 1, NULL, filt_sordetach, filt_soread };
158 static struct filterops sowrite_filtops =
159 { 1, NULL, filt_sowdetach, filt_sowrite };
160
161 uma_zone_t socket_zone;
162 so_gen_t so_gencnt; /* generation count for sockets */
163
164 int maxsockets;
165
166 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
167 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
168
169 static int somaxconn = SOMAXCONN;
170 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
171 /* XXX: we dont have SYSCTL_USHORT */
172 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
173 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
174 "queue size");
175 static int numopensockets;
176 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
177 &numopensockets, 0, "Number of open sockets");
178 #ifdef ZERO_COPY_SOCKETS
179 /* These aren't static because they're used in other files. */
180 int so_zero_copy_send = 1;
181 int so_zero_copy_receive = 1;
182 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
183 "Zero copy controls");
184 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
185 &so_zero_copy_receive, 0, "Enable zero copy receive");
186 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
187 &so_zero_copy_send, 0, "Enable zero copy send");
188 #endif /* ZERO_COPY_SOCKETS */
189
190 /*
191 * accept_mtx locks down per-socket fields relating to accept queues. See
192 * socketvar.h for an annotation of the protected fields of struct socket.
193 */
194 struct mtx accept_mtx;
195 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
196
197 /*
198 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
199 * so_gencnt field.
200 */
201 static struct mtx so_global_mtx;
202 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
203
204 /*
205 * General IPC sysctl name space, used by sockets and a variety of other IPC
206 * types.
207 */
208 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
209
210 /*
211 * Sysctl to get and set the maximum global sockets limit. Notify protocols
212 * of the change so that they can update their dependent limits as required.
213 */
214 static int
215 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
216 {
217 int error, newmaxsockets;
218
219 newmaxsockets = maxsockets;
220 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
221 if (error == 0 && req->newptr) {
222 if (newmaxsockets > maxsockets) {
223 maxsockets = newmaxsockets;
224 if (maxsockets > ((maxfiles / 4) * 3)) {
225 maxfiles = (maxsockets * 5) / 4;
226 maxfilesperproc = (maxfiles * 9) / 10;
227 }
228 EVENTHANDLER_INVOKE(maxsockets_change);
229 } else
230 error = EINVAL;
231 }
232 return (error);
233 }
234
235 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
236 &maxsockets, 0, sysctl_maxsockets, "IU",
237 "Maximum number of sockets avaliable");
238
239 /*
240 * Initialise maxsockets. This SYSINIT must be run after
241 * tunable_mbinit().
242 */
243 static void
244 init_maxsockets(void *ignored)
245 {
246
247 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
248 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
249 }
250 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
251
252 /*
253 * Socket operation routines. These routines are called by the routines in
254 * sys_socket.c or from a system process, and implement the semantics of
255 * socket operations by switching out to the protocol specific routines.
256 */
257
258 /*
259 * Get a socket structure from our zone, and initialize it. Note that it
260 * would probably be better to allocate socket and PCB at the same time, but
261 * I'm not convinced that all the protocols can be easily modified to do
262 * this.
263 *
264 * soalloc() returns a socket with a ref count of 0.
265 */
266 static struct socket *
267 soalloc(void)
268 {
269 struct socket *so;
270
271 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
272 if (so == NULL)
273 return (NULL);
274 #ifdef MAC
275 if (mac_init_socket(so, M_NOWAIT) != 0) {
276 uma_zfree(socket_zone, so);
277 return (NULL);
278 }
279 #endif
280 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
281 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
282 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
283 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
284 TAILQ_INIT(&so->so_aiojobq);
285 mtx_lock(&so_global_mtx);
286 so->so_gencnt = ++so_gencnt;
287 ++numopensockets;
288 mtx_unlock(&so_global_mtx);
289 return (so);
290 }
291
292 /*
293 * Free the storage associated with a socket at the socket layer, tear down
294 * locks, labels, etc. All protocol state is assumed already to have been
295 * torn down (and possibly never set up) by the caller.
296 */
297 static void
298 sodealloc(struct socket *so)
299 {
300
301 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
302 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
303
304 mtx_lock(&so_global_mtx);
305 so->so_gencnt = ++so_gencnt;
306 --numopensockets; /* Could be below, but faster here. */
307 mtx_unlock(&so_global_mtx);
308 if (so->so_rcv.sb_hiwat)
309 (void)chgsbsize(so->so_cred->cr_uidinfo,
310 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
311 if (so->so_snd.sb_hiwat)
312 (void)chgsbsize(so->so_cred->cr_uidinfo,
313 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
314 #ifdef INET
315 /* remove acccept filter if one is present. */
316 if (so->so_accf != NULL)
317 do_setopt_accept_filter(so, NULL);
318 #endif
319 #ifdef MAC
320 mac_destroy_socket(so);
321 #endif
322 crfree(so->so_cred);
323 sx_destroy(&so->so_snd.sb_sx);
324 sx_destroy(&so->so_rcv.sb_sx);
325 SOCKBUF_LOCK_DESTROY(&so->so_snd);
326 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
327 uma_zfree(socket_zone, so);
328 }
329
330 /*
331 * socreate returns a socket with a ref count of 1. The socket should be
332 * closed with soclose().
333 */
334 int
335 socreate(int dom, struct socket **aso, int type, int proto,
336 struct ucred *cred, struct thread *td)
337 {
338 struct protosw *prp;
339 struct socket *so;
340 int error;
341
342 if (proto)
343 prp = pffindproto(dom, proto, type);
344 else
345 prp = pffindtype(dom, type);
346
347 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
348 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
349 return (EPROTONOSUPPORT);
350
351 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
352 return (EPROTONOSUPPORT);
353
354 if (prp->pr_type != type)
355 return (EPROTOTYPE);
356 so = soalloc();
357 if (so == NULL)
358 return (ENOBUFS);
359
360 TAILQ_INIT(&so->so_incomp);
361 TAILQ_INIT(&so->so_comp);
362 so->so_type = type;
363 so->so_cred = crhold(cred);
364 if ((prp->pr_domain->dom_family == PF_INET) ||
365 (prp->pr_domain->dom_family == PF_ROUTE))
366 so->so_fibnum = td->td_proc->p_fibnum;
367 else
368 so->so_fibnum = 0;
369 so->so_proto = prp;
370 #ifdef MAC
371 mac_create_socket(cred, so);
372 #endif
373 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
374 NULL, NULL, NULL);
375 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
376 NULL, NULL, NULL);
377 so->so_count = 1;
378 /*
379 * Auto-sizing of socket buffers is managed by the protocols and
380 * the appropriate flags must be set in the pru_attach function.
381 */
382 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
383 if (error) {
384 KASSERT(so->so_count == 1, ("socreate: so_count %d",
385 so->so_count));
386 so->so_count = 0;
387 sodealloc(so);
388 return (error);
389 }
390 *aso = so;
391 return (0);
392 }
393
394 #ifdef REGRESSION
395 static int regression_sonewconn_earlytest = 1;
396 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
397 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
398 #endif
399
400 /*
401 * When an attempt at a new connection is noted on a socket which accepts
402 * connections, sonewconn is called. If the connection is possible (subject
403 * to space constraints, etc.) then we allocate a new structure, propoerly
404 * linked into the data structure of the original socket, and return this.
405 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
406 *
407 * Note: the ref count on the socket is 0 on return.
408 */
409 struct socket *
410 sonewconn(struct socket *head, int connstatus)
411 {
412 struct socket *so;
413 int over;
414
415 ACCEPT_LOCK();
416 over = (head->so_qlen > 3 * head->so_qlimit / 2);
417 ACCEPT_UNLOCK();
418 #ifdef REGRESSION
419 if (regression_sonewconn_earlytest && over)
420 #else
421 if (over)
422 #endif
423 return (NULL);
424 so = soalloc();
425 if (so == NULL)
426 return (NULL);
427 if ((head->so_options & SO_ACCEPTFILTER) != 0)
428 connstatus = 0;
429 so->so_head = head;
430 so->so_type = head->so_type;
431 so->so_options = head->so_options &~ SO_ACCEPTCONN;
432 so->so_linger = head->so_linger;
433 so->so_state = head->so_state | SS_NOFDREF;
434 so->so_fibnum = head->so_fibnum;
435 so->so_proto = head->so_proto;
436 so->so_cred = crhold(head->so_cred);
437 #ifdef MAC
438 SOCK_LOCK(head);
439 mac_create_socket_from_socket(head, so);
440 SOCK_UNLOCK(head);
441 #endif
442 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
443 NULL, NULL, NULL);
444 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
445 NULL, NULL, NULL);
446 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
447 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
448 sodealloc(so);
449 return (NULL);
450 }
451 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
452 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
453 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
454 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
455 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
456 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
457 so->so_state |= connstatus;
458 ACCEPT_LOCK();
459 if (connstatus) {
460 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
461 so->so_qstate |= SQ_COMP;
462 head->so_qlen++;
463 } else {
464 /*
465 * Keep removing sockets from the head until there's room for
466 * us to insert on the tail. In pre-locking revisions, this
467 * was a simple if(), but as we could be racing with other
468 * threads and soabort() requires dropping locks, we must
469 * loop waiting for the condition to be true.
470 */
471 while (head->so_incqlen > head->so_qlimit) {
472 struct socket *sp;
473 sp = TAILQ_FIRST(&head->so_incomp);
474 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
475 head->so_incqlen--;
476 sp->so_qstate &= ~SQ_INCOMP;
477 sp->so_head = NULL;
478 ACCEPT_UNLOCK();
479 soabort(sp);
480 ACCEPT_LOCK();
481 }
482 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
483 so->so_qstate |= SQ_INCOMP;
484 head->so_incqlen++;
485 }
486 ACCEPT_UNLOCK();
487 if (connstatus) {
488 sorwakeup(head);
489 wakeup_one(&head->so_timeo);
490 }
491 return (so);
492 }
493
494 int
495 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
496 {
497
498 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
499 }
500
501 /*
502 * solisten() transitions a socket from a non-listening state to a listening
503 * state, but can also be used to update the listen queue depth on an
504 * existing listen socket. The protocol will call back into the sockets
505 * layer using solisten_proto_check() and solisten_proto() to check and set
506 * socket-layer listen state. Call backs are used so that the protocol can
507 * acquire both protocol and socket layer locks in whatever order is required
508 * by the protocol.
509 *
510 * Protocol implementors are advised to hold the socket lock across the
511 * socket-layer test and set to avoid races at the socket layer.
512 */
513 int
514 solisten(struct socket *so, int backlog, struct thread *td)
515 {
516
517 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
518 }
519
520 int
521 solisten_proto_check(struct socket *so)
522 {
523
524 SOCK_LOCK_ASSERT(so);
525
526 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
527 SS_ISDISCONNECTING))
528 return (EINVAL);
529 return (0);
530 }
531
532 void
533 solisten_proto(struct socket *so, int backlog)
534 {
535
536 SOCK_LOCK_ASSERT(so);
537
538 if (backlog < 0 || backlog > somaxconn)
539 backlog = somaxconn;
540 so->so_qlimit = backlog;
541 so->so_options |= SO_ACCEPTCONN;
542 }
543
544 /*
545 * Attempt to free a socket. This should really be sotryfree().
546 *
547 * sofree() will succeed if:
548 *
549 * - There are no outstanding file descriptor references or related consumers
550 * (so_count == 0).
551 *
552 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
553 *
554 * - The protocol does not have an outstanding strong reference on the socket
555 * (SS_PROTOREF).
556 *
557 * - The socket is not in a completed connection queue, so a process has been
558 * notified that it is present. If it is removed, the user process may
559 * block in accept() despite select() saying the socket was ready.
560 *
561 * Otherwise, it will quietly abort so that a future call to sofree(), when
562 * conditions are right, can succeed.
563 */
564 void
565 sofree(struct socket *so)
566 {
567 struct protosw *pr = so->so_proto;
568 struct socket *head;
569
570 ACCEPT_LOCK_ASSERT();
571 SOCK_LOCK_ASSERT(so);
572
573 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
574 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
575 SOCK_UNLOCK(so);
576 ACCEPT_UNLOCK();
577 return;
578 }
579
580 head = so->so_head;
581 if (head != NULL) {
582 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
583 (so->so_qstate & SQ_INCOMP) != 0,
584 ("sofree: so_head != NULL, but neither SQ_COMP nor "
585 "SQ_INCOMP"));
586 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
587 (so->so_qstate & SQ_INCOMP) == 0,
588 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
589 TAILQ_REMOVE(&head->so_incomp, so, so_list);
590 head->so_incqlen--;
591 so->so_qstate &= ~SQ_INCOMP;
592 so->so_head = NULL;
593 }
594 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
595 (so->so_qstate & SQ_INCOMP) == 0,
596 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
597 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
598 if (so->so_options & SO_ACCEPTCONN) {
599 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
600 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
601 }
602 SOCK_UNLOCK(so);
603 ACCEPT_UNLOCK();
604
605 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
606 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
607 if (pr->pr_usrreqs->pru_detach != NULL)
608 (*pr->pr_usrreqs->pru_detach)(so);
609
610 /*
611 * From this point on, we assume that no other references to this
612 * socket exist anywhere else in the stack. Therefore, no locks need
613 * to be acquired or held.
614 *
615 * We used to do a lot of socket buffer and socket locking here, as
616 * well as invoke sorflush() and perform wakeups. The direct call to
617 * dom_dispose() and sbrelease_internal() are an inlining of what was
618 * necessary from sorflush().
619 *
620 * Notice that the socket buffer and kqueue state are torn down
621 * before calling pru_detach. This means that protocols shold not
622 * assume they can perform socket wakeups, etc, in their detach code.
623 */
624 sbdestroy(&so->so_snd, so);
625 sbdestroy(&so->so_rcv, so);
626 knlist_destroy(&so->so_rcv.sb_sel.si_note);
627 knlist_destroy(&so->so_snd.sb_sel.si_note);
628 sodealloc(so);
629 }
630
631 /*
632 * Close a socket on last file table reference removal. Initiate disconnect
633 * if connected. Free socket when disconnect complete.
634 *
635 * This function will sorele() the socket. Note that soclose() may be called
636 * prior to the ref count reaching zero. The actual socket structure will
637 * not be freed until the ref count reaches zero.
638 */
639 int
640 soclose(struct socket *so)
641 {
642 int error = 0;
643
644 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
645
646 funsetown(&so->so_sigio);
647 if (so->so_state & SS_ISCONNECTED) {
648 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
649 error = sodisconnect(so);
650 if (error)
651 goto drop;
652 }
653 if (so->so_options & SO_LINGER) {
654 if ((so->so_state & SS_ISDISCONNECTING) &&
655 (so->so_state & SS_NBIO))
656 goto drop;
657 while (so->so_state & SS_ISCONNECTED) {
658 error = tsleep(&so->so_timeo,
659 PSOCK | PCATCH, "soclos", so->so_linger * hz);
660 if (error)
661 break;
662 }
663 }
664 }
665
666 drop:
667 if (so->so_proto->pr_usrreqs->pru_close != NULL)
668 (*so->so_proto->pr_usrreqs->pru_close)(so);
669 if (so->so_options & SO_ACCEPTCONN) {
670 struct socket *sp;
671 ACCEPT_LOCK();
672 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
673 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
674 so->so_incqlen--;
675 sp->so_qstate &= ~SQ_INCOMP;
676 sp->so_head = NULL;
677 ACCEPT_UNLOCK();
678 soabort(sp);
679 ACCEPT_LOCK();
680 }
681 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
682 TAILQ_REMOVE(&so->so_comp, sp, so_list);
683 so->so_qlen--;
684 sp->so_qstate &= ~SQ_COMP;
685 sp->so_head = NULL;
686 ACCEPT_UNLOCK();
687 soabort(sp);
688 ACCEPT_LOCK();
689 }
690 ACCEPT_UNLOCK();
691 }
692 ACCEPT_LOCK();
693 SOCK_LOCK(so);
694 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
695 so->so_state |= SS_NOFDREF;
696 sorele(so);
697 return (error);
698 }
699
700 /*
701 * soabort() is used to abruptly tear down a connection, such as when a
702 * resource limit is reached (listen queue depth exceeded), or if a listen
703 * socket is closed while there are sockets waiting to be accepted.
704 *
705 * This interface is tricky, because it is called on an unreferenced socket,
706 * and must be called only by a thread that has actually removed the socket
707 * from the listen queue it was on, or races with other threads are risked.
708 *
709 * This interface will call into the protocol code, so must not be called
710 * with any socket locks held. Protocols do call it while holding their own
711 * recursible protocol mutexes, but this is something that should be subject
712 * to review in the future.
713 */
714 void
715 soabort(struct socket *so)
716 {
717
718 /*
719 * In as much as is possible, assert that no references to this
720 * socket are held. This is not quite the same as asserting that the
721 * current thread is responsible for arranging for no references, but
722 * is as close as we can get for now.
723 */
724 KASSERT(so->so_count == 0, ("soabort: so_count"));
725 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
726 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
727 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
728 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
729
730 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
731 (*so->so_proto->pr_usrreqs->pru_abort)(so);
732 ACCEPT_LOCK();
733 SOCK_LOCK(so);
734 sofree(so);
735 }
736
737 int
738 soaccept(struct socket *so, struct sockaddr **nam)
739 {
740 int error;
741
742 SOCK_LOCK(so);
743 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
744 so->so_state &= ~SS_NOFDREF;
745 SOCK_UNLOCK(so);
746 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
747 return (error);
748 }
749
750 int
751 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
752 {
753 int error;
754
755 if (so->so_options & SO_ACCEPTCONN)
756 return (EOPNOTSUPP);
757 /*
758 * If protocol is connection-based, can only connect once.
759 * Otherwise, if connected, try to disconnect first. This allows
760 * user to disconnect by connecting to, e.g., a null address.
761 */
762 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
763 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
764 (error = sodisconnect(so)))) {
765 error = EISCONN;
766 } else {
767 /*
768 * Prevent accumulated error from previous connection from
769 * biting us.
770 */
771 so->so_error = 0;
772 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
773 }
774
775 return (error);
776 }
777
778 int
779 soconnect2(struct socket *so1, struct socket *so2)
780 {
781
782 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
783 }
784
785 int
786 sodisconnect(struct socket *so)
787 {
788 int error;
789
790 if ((so->so_state & SS_ISCONNECTED) == 0)
791 return (ENOTCONN);
792 if (so->so_state & SS_ISDISCONNECTING)
793 return (EALREADY);
794 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
795 return (error);
796 }
797
798 #ifdef ZERO_COPY_SOCKETS
799 struct so_zerocopy_stats{
800 int size_ok;
801 int align_ok;
802 int found_ifp;
803 };
804 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
805 #include <netinet/in.h>
806 #include <net/route.h>
807 #include <netinet/in_pcb.h>
808 #include <vm/vm.h>
809 #include <vm/vm_page.h>
810 #include <vm/vm_object.h>
811
812 /*
813 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
814 * sosend_dgram() and sosend_generic() use m_uiotombuf().
815 *
816 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
817 * all of the data referenced by the uio. If desired, it uses zero-copy.
818 * *space will be updated to reflect data copied in.
819 *
820 * NB: If atomic I/O is requested, the caller must already have checked that
821 * space can hold resid bytes.
822 *
823 * NB: In the event of an error, the caller may need to free the partial
824 * chain pointed to by *mpp. The contents of both *uio and *space may be
825 * modified even in the case of an error.
826 */
827 static int
828 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
829 int flags)
830 {
831 struct mbuf *m, **mp, *top;
832 long len, resid;
833 int error;
834 #ifdef ZERO_COPY_SOCKETS
835 int cow_send;
836 #endif
837
838 *retmp = top = NULL;
839 mp = ⊤
840 len = 0;
841 resid = uio->uio_resid;
842 error = 0;
843 do {
844 #ifdef ZERO_COPY_SOCKETS
845 cow_send = 0;
846 #endif /* ZERO_COPY_SOCKETS */
847 if (resid >= MINCLSIZE) {
848 #ifdef ZERO_COPY_SOCKETS
849 if (top == NULL) {
850 m = m_gethdr(M_WAITOK, MT_DATA);
851 m->m_pkthdr.len = 0;
852 m->m_pkthdr.rcvif = NULL;
853 } else
854 m = m_get(M_WAITOK, MT_DATA);
855 if (so_zero_copy_send &&
856 resid>=PAGE_SIZE &&
857 *space>=PAGE_SIZE &&
858 uio->uio_iov->iov_len>=PAGE_SIZE) {
859 so_zerocp_stats.size_ok++;
860 so_zerocp_stats.align_ok++;
861 cow_send = socow_setup(m, uio);
862 len = cow_send;
863 }
864 if (!cow_send) {
865 m_clget(m, M_WAITOK);
866 len = min(min(MCLBYTES, resid), *space);
867 }
868 #else /* ZERO_COPY_SOCKETS */
869 if (top == NULL) {
870 m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
871 m->m_pkthdr.len = 0;
872 m->m_pkthdr.rcvif = NULL;
873 } else
874 m = m_getcl(M_TRYWAIT, MT_DATA, 0);
875 len = min(min(MCLBYTES, resid), *space);
876 #endif /* ZERO_COPY_SOCKETS */
877 } else {
878 if (top == NULL) {
879 m = m_gethdr(M_TRYWAIT, MT_DATA);
880 m->m_pkthdr.len = 0;
881 m->m_pkthdr.rcvif = NULL;
882
883 len = min(min(MHLEN, resid), *space);
884 /*
885 * For datagram protocols, leave room
886 * for protocol headers in first mbuf.
887 */
888 if (atomic && m && len < MHLEN)
889 MH_ALIGN(m, len);
890 } else {
891 m = m_get(M_TRYWAIT, MT_DATA);
892 len = min(min(MLEN, resid), *space);
893 }
894 }
895 if (m == NULL) {
896 error = ENOBUFS;
897 goto out;
898 }
899
900 *space -= len;
901 #ifdef ZERO_COPY_SOCKETS
902 if (cow_send)
903 error = 0;
904 else
905 #endif /* ZERO_COPY_SOCKETS */
906 error = uiomove(mtod(m, void *), (int)len, uio);
907 resid = uio->uio_resid;
908 m->m_len = len;
909 *mp = m;
910 top->m_pkthdr.len += len;
911 if (error)
912 goto out;
913 mp = &m->m_next;
914 if (resid <= 0) {
915 if (flags & MSG_EOR)
916 top->m_flags |= M_EOR;
917 break;
918 }
919 } while (*space > 0 && atomic);
920 out:
921 *retmp = top;
922 return (error);
923 }
924 #endif /*ZERO_COPY_SOCKETS*/
925
926 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
927
928 int
929 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
930 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
931 {
932 long space, resid;
933 int clen = 0, error, dontroute;
934 #ifdef ZERO_COPY_SOCKETS
935 int atomic = sosendallatonce(so) || top;
936 #endif
937
938 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
939 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
940 ("sodgram_send: !PR_ATOMIC"));
941
942 if (uio != NULL)
943 resid = uio->uio_resid;
944 else
945 resid = top->m_pkthdr.len;
946 /*
947 * In theory resid should be unsigned. However, space must be
948 * signed, as it might be less than 0 if we over-committed, and we
949 * must use a signed comparison of space and resid. On the other
950 * hand, a negative resid causes us to loop sending 0-length
951 * segments to the protocol.
952 *
953 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
954 * type sockets since that's an error.
955 */
956 if (resid < 0) {
957 error = EINVAL;
958 goto out;
959 }
960
961 dontroute =
962 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
963 if (td != NULL)
964 td->td_ru.ru_msgsnd++;
965 if (control != NULL)
966 clen = control->m_len;
967
968 SOCKBUF_LOCK(&so->so_snd);
969 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
970 SOCKBUF_UNLOCK(&so->so_snd);
971 error = EPIPE;
972 goto out;
973 }
974 if (so->so_error) {
975 error = so->so_error;
976 so->so_error = 0;
977 SOCKBUF_UNLOCK(&so->so_snd);
978 goto out;
979 }
980 if ((so->so_state & SS_ISCONNECTED) == 0) {
981 /*
982 * `sendto' and `sendmsg' is allowed on a connection-based
983 * socket if it supports implied connect. Return ENOTCONN if
984 * not connected and no address is supplied.
985 */
986 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
987 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
988 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
989 !(resid == 0 && clen != 0)) {
990 SOCKBUF_UNLOCK(&so->so_snd);
991 error = ENOTCONN;
992 goto out;
993 }
994 } else if (addr == NULL) {
995 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
996 error = ENOTCONN;
997 else
998 error = EDESTADDRREQ;
999 SOCKBUF_UNLOCK(&so->so_snd);
1000 goto out;
1001 }
1002 }
1003
1004 /*
1005 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1006 * problem and need fixing.
1007 */
1008 space = sbspace(&so->so_snd);
1009 if (flags & MSG_OOB)
1010 space += 1024;
1011 space -= clen;
1012 SOCKBUF_UNLOCK(&so->so_snd);
1013 if (resid > space) {
1014 error = EMSGSIZE;
1015 goto out;
1016 }
1017 if (uio == NULL) {
1018 resid = 0;
1019 if (flags & MSG_EOR)
1020 top->m_flags |= M_EOR;
1021 } else {
1022 #ifdef ZERO_COPY_SOCKETS
1023 error = sosend_copyin(uio, &top, atomic, &space, flags);
1024 if (error)
1025 goto out;
1026 #else
1027 /*
1028 * Copy the data from userland into a mbuf chain.
1029 * If no data is to be copied in, a single empty mbuf
1030 * is returned.
1031 */
1032 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1033 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1034 if (top == NULL) {
1035 error = EFAULT; /* only possible error */
1036 goto out;
1037 }
1038 space -= resid - uio->uio_resid;
1039 #endif
1040 resid = uio->uio_resid;
1041 }
1042 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1043 /*
1044 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1045 * than with.
1046 */
1047 if (dontroute) {
1048 SOCK_LOCK(so);
1049 so->so_options |= SO_DONTROUTE;
1050 SOCK_UNLOCK(so);
1051 }
1052 /*
1053 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1054 * of date. We could have recieved a reset packet in an interrupt or
1055 * maybe we slept while doing page faults in uiomove() etc. We could
1056 * probably recheck again inside the locking protection here, but
1057 * there are probably other places that this also happens. We must
1058 * rethink this.
1059 */
1060 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1061 (flags & MSG_OOB) ? PRUS_OOB :
1062 /*
1063 * If the user set MSG_EOF, the protocol understands this flag and
1064 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1065 */
1066 ((flags & MSG_EOF) &&
1067 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1068 (resid <= 0)) ?
1069 PRUS_EOF :
1070 /* If there is more to send set PRUS_MORETOCOME */
1071 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1072 top, addr, control, td);
1073 if (dontroute) {
1074 SOCK_LOCK(so);
1075 so->so_options &= ~SO_DONTROUTE;
1076 SOCK_UNLOCK(so);
1077 }
1078 clen = 0;
1079 control = NULL;
1080 top = NULL;
1081 out:
1082 if (top != NULL)
1083 m_freem(top);
1084 if (control != NULL)
1085 m_freem(control);
1086 return (error);
1087 }
1088
1089 /*
1090 * Send on a socket. If send must go all at once and message is larger than
1091 * send buffering, then hard error. Lock against other senders. If must go
1092 * all at once and not enough room now, then inform user that this would
1093 * block and do nothing. Otherwise, if nonblocking, send as much as
1094 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1095 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1096 * in mbuf chain must be small enough to send all at once.
1097 *
1098 * Returns nonzero on error, timeout or signal; callers must check for short
1099 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1100 * on return.
1101 */
1102 int
1103 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1104 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1105 {
1106 long space, resid;
1107 int clen = 0, error, dontroute;
1108 int atomic = sosendallatonce(so) || top;
1109
1110 if (uio != NULL)
1111 resid = uio->uio_resid;
1112 else
1113 resid = top->m_pkthdr.len;
1114 /*
1115 * In theory resid should be unsigned. However, space must be
1116 * signed, as it might be less than 0 if we over-committed, and we
1117 * must use a signed comparison of space and resid. On the other
1118 * hand, a negative resid causes us to loop sending 0-length
1119 * segments to the protocol.
1120 *
1121 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1122 * type sockets since that's an error.
1123 */
1124 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1125 error = EINVAL;
1126 goto out;
1127 }
1128
1129 dontroute =
1130 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1131 (so->so_proto->pr_flags & PR_ATOMIC);
1132 if (td != NULL)
1133 td->td_ru.ru_msgsnd++;
1134 if (control != NULL)
1135 clen = control->m_len;
1136
1137 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1138 if (error)
1139 goto out;
1140
1141 restart:
1142 do {
1143 SOCKBUF_LOCK(&so->so_snd);
1144 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1145 SOCKBUF_UNLOCK(&so->so_snd);
1146 error = EPIPE;
1147 goto release;
1148 }
1149 if (so->so_error) {
1150 error = so->so_error;
1151 so->so_error = 0;
1152 SOCKBUF_UNLOCK(&so->so_snd);
1153 goto release;
1154 }
1155 if ((so->so_state & SS_ISCONNECTED) == 0) {
1156 /*
1157 * `sendto' and `sendmsg' is allowed on a connection-
1158 * based socket if it supports implied connect.
1159 * Return ENOTCONN if not connected and no address is
1160 * supplied.
1161 */
1162 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1163 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1164 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1165 !(resid == 0 && clen != 0)) {
1166 SOCKBUF_UNLOCK(&so->so_snd);
1167 error = ENOTCONN;
1168 goto release;
1169 }
1170 } else if (addr == NULL) {
1171 SOCKBUF_UNLOCK(&so->so_snd);
1172 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1173 error = ENOTCONN;
1174 else
1175 error = EDESTADDRREQ;
1176 goto release;
1177 }
1178 }
1179 space = sbspace(&so->so_snd);
1180 if (flags & MSG_OOB)
1181 space += 1024;
1182 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1183 clen > so->so_snd.sb_hiwat) {
1184 SOCKBUF_UNLOCK(&so->so_snd);
1185 error = EMSGSIZE;
1186 goto release;
1187 }
1188 if (space < resid + clen &&
1189 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1190 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1191 SOCKBUF_UNLOCK(&so->so_snd);
1192 error = EWOULDBLOCK;
1193 goto release;
1194 }
1195 error = sbwait(&so->so_snd);
1196 SOCKBUF_UNLOCK(&so->so_snd);
1197 if (error)
1198 goto release;
1199 goto restart;
1200 }
1201 SOCKBUF_UNLOCK(&so->so_snd);
1202 space -= clen;
1203 do {
1204 if (uio == NULL) {
1205 resid = 0;
1206 if (flags & MSG_EOR)
1207 top->m_flags |= M_EOR;
1208 } else {
1209 #ifdef ZERO_COPY_SOCKETS
1210 error = sosend_copyin(uio, &top, atomic,
1211 &space, flags);
1212 if (error != 0)
1213 goto release;
1214 #else
1215 /*
1216 * Copy the data from userland into a mbuf
1217 * chain. If no data is to be copied in,
1218 * a single empty mbuf is returned.
1219 */
1220 top = m_uiotombuf(uio, M_WAITOK, space,
1221 (atomic ? max_hdr : 0),
1222 (atomic ? M_PKTHDR : 0) |
1223 ((flags & MSG_EOR) ? M_EOR : 0));
1224 if (top == NULL) {
1225 error = EFAULT; /* only possible error */
1226 goto release;
1227 }
1228 space -= resid - uio->uio_resid;
1229 #endif
1230 resid = uio->uio_resid;
1231 }
1232 if (dontroute) {
1233 SOCK_LOCK(so);
1234 so->so_options |= SO_DONTROUTE;
1235 SOCK_UNLOCK(so);
1236 }
1237 /*
1238 * XXX all the SBS_CANTSENDMORE checks previously
1239 * done could be out of date. We could have recieved
1240 * a reset packet in an interrupt or maybe we slept
1241 * while doing page faults in uiomove() etc. We
1242 * could probably recheck again inside the locking
1243 * protection here, but there are probably other
1244 * places that this also happens. We must rethink
1245 * this.
1246 */
1247 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1248 (flags & MSG_OOB) ? PRUS_OOB :
1249 /*
1250 * If the user set MSG_EOF, the protocol understands
1251 * this flag and nothing left to send then use
1252 * PRU_SEND_EOF instead of PRU_SEND.
1253 */
1254 ((flags & MSG_EOF) &&
1255 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1256 (resid <= 0)) ?
1257 PRUS_EOF :
1258 /* If there is more to send set PRUS_MORETOCOME. */
1259 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1260 top, addr, control, td);
1261 if (dontroute) {
1262 SOCK_LOCK(so);
1263 so->so_options &= ~SO_DONTROUTE;
1264 SOCK_UNLOCK(so);
1265 }
1266 clen = 0;
1267 control = NULL;
1268 top = NULL;
1269 if (error)
1270 goto release;
1271 } while (resid && space > 0);
1272 } while (resid);
1273
1274 release:
1275 sbunlock(&so->so_snd);
1276 out:
1277 if (top != NULL)
1278 m_freem(top);
1279 if (control != NULL)
1280 m_freem(control);
1281 return (error);
1282 }
1283
1284 int
1285 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1286 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1287 {
1288
1289 return (so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1290 control, flags, td));
1291 }
1292
1293 /*
1294 * The part of soreceive() that implements reading non-inline out-of-band
1295 * data from a socket. For more complete comments, see soreceive(), from
1296 * which this code originated.
1297 *
1298 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1299 * unable to return an mbuf chain to the caller.
1300 */
1301 static int
1302 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1303 {
1304 struct protosw *pr = so->so_proto;
1305 struct mbuf *m;
1306 int error;
1307
1308 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1309
1310 m = m_get(M_TRYWAIT, MT_DATA);
1311 if (m == NULL)
1312 return (ENOBUFS);
1313 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1314 if (error)
1315 goto bad;
1316 do {
1317 #ifdef ZERO_COPY_SOCKETS
1318 if (so_zero_copy_receive) {
1319 int disposable;
1320
1321 if ((m->m_flags & M_EXT)
1322 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1323 disposable = 1;
1324 else
1325 disposable = 0;
1326
1327 error = uiomoveco(mtod(m, void *),
1328 min(uio->uio_resid, m->m_len),
1329 uio, disposable);
1330 } else
1331 #endif /* ZERO_COPY_SOCKETS */
1332 error = uiomove(mtod(m, void *),
1333 (int) min(uio->uio_resid, m->m_len), uio);
1334 m = m_free(m);
1335 } while (uio->uio_resid && error == 0 && m);
1336 bad:
1337 if (m != NULL)
1338 m_freem(m);
1339 return (error);
1340 }
1341
1342 /*
1343 * Following replacement or removal of the first mbuf on the first mbuf chain
1344 * of a socket buffer, push necessary state changes back into the socket
1345 * buffer so that other consumers see the values consistently. 'nextrecord'
1346 * is the callers locally stored value of the original value of
1347 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1348 * NOTE: 'nextrecord' may be NULL.
1349 */
1350 static __inline void
1351 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1352 {
1353
1354 SOCKBUF_LOCK_ASSERT(sb);
1355 /*
1356 * First, update for the new value of nextrecord. If necessary, make
1357 * it the first record.
1358 */
1359 if (sb->sb_mb != NULL)
1360 sb->sb_mb->m_nextpkt = nextrecord;
1361 else
1362 sb->sb_mb = nextrecord;
1363
1364 /*
1365 * Now update any dependent socket buffer fields to reflect the new
1366 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1367 * addition of a second clause that takes care of the case where
1368 * sb_mb has been updated, but remains the last record.
1369 */
1370 if (sb->sb_mb == NULL) {
1371 sb->sb_mbtail = NULL;
1372 sb->sb_lastrecord = NULL;
1373 } else if (sb->sb_mb->m_nextpkt == NULL)
1374 sb->sb_lastrecord = sb->sb_mb;
1375 }
1376
1377
1378 /*
1379 * Implement receive operations on a socket. We depend on the way that
1380 * records are added to the sockbuf by sbappend. In particular, each record
1381 * (mbufs linked through m_next) must begin with an address if the protocol
1382 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1383 * data, and then zero or more mbufs of data. In order to allow parallelism
1384 * between network receive and copying to user space, as well as avoid
1385 * sleeping with a mutex held, we release the socket buffer mutex during the
1386 * user space copy. Although the sockbuf is locked, new data may still be
1387 * appended, and thus we must maintain consistency of the sockbuf during that
1388 * time.
1389 *
1390 * The caller may receive the data as a single mbuf chain by supplying an
1391 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1392 * the count in uio_resid.
1393 */
1394 int
1395 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1396 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1397 {
1398 struct mbuf *m, **mp;
1399 int flags, len, error, offset;
1400 struct protosw *pr = so->so_proto;
1401 struct mbuf *nextrecord;
1402 int moff, type = 0;
1403 int orig_resid = uio->uio_resid;
1404
1405 mp = mp0;
1406 if (psa != NULL)
1407 *psa = NULL;
1408 if (controlp != NULL)
1409 *controlp = NULL;
1410 if (flagsp != NULL)
1411 flags = *flagsp &~ MSG_EOR;
1412 else
1413 flags = 0;
1414 if (flags & MSG_OOB)
1415 return (soreceive_rcvoob(so, uio, flags));
1416 if (mp != NULL)
1417 *mp = NULL;
1418 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1419 && uio->uio_resid)
1420 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1421
1422 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1423 if (error)
1424 return (error);
1425
1426 restart:
1427 SOCKBUF_LOCK(&so->so_rcv);
1428 m = so->so_rcv.sb_mb;
1429 /*
1430 * If we have less data than requested, block awaiting more (subject
1431 * to any timeout) if:
1432 * 1. the current count is less than the low water mark, or
1433 * 2. MSG_WAITALL is set, and it is possible to do the entire
1434 * receive operation at once if we block (resid <= hiwat).
1435 * 3. MSG_DONTWAIT is not set
1436 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1437 * we have to do the receive in sections, and thus risk returning a
1438 * short count if a timeout or signal occurs after we start.
1439 */
1440 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1441 so->so_rcv.sb_cc < uio->uio_resid) &&
1442 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1443 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1444 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1445 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1446 ("receive: m == %p so->so_rcv.sb_cc == %u",
1447 m, so->so_rcv.sb_cc));
1448 if (so->so_error) {
1449 if (m != NULL)
1450 goto dontblock;
1451 error = so->so_error;
1452 if ((flags & MSG_PEEK) == 0)
1453 so->so_error = 0;
1454 SOCKBUF_UNLOCK(&so->so_rcv);
1455 goto release;
1456 }
1457 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1458 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1459 if (m == NULL) {
1460 SOCKBUF_UNLOCK(&so->so_rcv);
1461 goto release;
1462 } else
1463 goto dontblock;
1464 }
1465 for (; m != NULL; m = m->m_next)
1466 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1467 m = so->so_rcv.sb_mb;
1468 goto dontblock;
1469 }
1470 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1471 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1472 SOCKBUF_UNLOCK(&so->so_rcv);
1473 error = ENOTCONN;
1474 goto release;
1475 }
1476 if (uio->uio_resid == 0) {
1477 SOCKBUF_UNLOCK(&so->so_rcv);
1478 goto release;
1479 }
1480 if ((so->so_state & SS_NBIO) ||
1481 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1482 SOCKBUF_UNLOCK(&so->so_rcv);
1483 error = EWOULDBLOCK;
1484 goto release;
1485 }
1486 SBLASTRECORDCHK(&so->so_rcv);
1487 SBLASTMBUFCHK(&so->so_rcv);
1488 error = sbwait(&so->so_rcv);
1489 SOCKBUF_UNLOCK(&so->so_rcv);
1490 if (error)
1491 goto release;
1492 goto restart;
1493 }
1494 dontblock:
1495 /*
1496 * From this point onward, we maintain 'nextrecord' as a cache of the
1497 * pointer to the next record in the socket buffer. We must keep the
1498 * various socket buffer pointers and local stack versions of the
1499 * pointers in sync, pushing out modifications before dropping the
1500 * socket buffer mutex, and re-reading them when picking it up.
1501 *
1502 * Otherwise, we will race with the network stack appending new data
1503 * or records onto the socket buffer by using inconsistent/stale
1504 * versions of the field, possibly resulting in socket buffer
1505 * corruption.
1506 *
1507 * By holding the high-level sblock(), we prevent simultaneous
1508 * readers from pulling off the front of the socket buffer.
1509 */
1510 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1511 if (uio->uio_td)
1512 uio->uio_td->td_ru.ru_msgrcv++;
1513 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1514 SBLASTRECORDCHK(&so->so_rcv);
1515 SBLASTMBUFCHK(&so->so_rcv);
1516 nextrecord = m->m_nextpkt;
1517 if (pr->pr_flags & PR_ADDR) {
1518 KASSERT(m->m_type == MT_SONAME,
1519 ("m->m_type == %d", m->m_type));
1520 orig_resid = 0;
1521 if (psa != NULL)
1522 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1523 M_NOWAIT);
1524 if (flags & MSG_PEEK) {
1525 m = m->m_next;
1526 } else {
1527 sbfree(&so->so_rcv, m);
1528 so->so_rcv.sb_mb = m_free(m);
1529 m = so->so_rcv.sb_mb;
1530 sockbuf_pushsync(&so->so_rcv, nextrecord);
1531 }
1532 }
1533
1534 /*
1535 * Process one or more MT_CONTROL mbufs present before any data mbufs
1536 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1537 * just copy the data; if !MSG_PEEK, we call into the protocol to
1538 * perform externalization (or freeing if controlp == NULL).
1539 */
1540 if (m != NULL && m->m_type == MT_CONTROL) {
1541 struct mbuf *cm = NULL, *cmn;
1542 struct mbuf **cme = &cm;
1543
1544 do {
1545 if (flags & MSG_PEEK) {
1546 if (controlp != NULL) {
1547 *controlp = m_copy(m, 0, m->m_len);
1548 controlp = &(*controlp)->m_next;
1549 }
1550 m = m->m_next;
1551 } else {
1552 sbfree(&so->so_rcv, m);
1553 so->so_rcv.sb_mb = m->m_next;
1554 m->m_next = NULL;
1555 *cme = m;
1556 cme = &(*cme)->m_next;
1557 m = so->so_rcv.sb_mb;
1558 }
1559 } while (m != NULL && m->m_type == MT_CONTROL);
1560 if ((flags & MSG_PEEK) == 0)
1561 sockbuf_pushsync(&so->so_rcv, nextrecord);
1562 while (cm != NULL) {
1563 cmn = cm->m_next;
1564 cm->m_next = NULL;
1565 if (pr->pr_domain->dom_externalize != NULL) {
1566 SOCKBUF_UNLOCK(&so->so_rcv);
1567 error = (*pr->pr_domain->dom_externalize)
1568 (cm, controlp);
1569 SOCKBUF_LOCK(&so->so_rcv);
1570 } else if (controlp != NULL)
1571 *controlp = cm;
1572 else
1573 m_freem(cm);
1574 if (controlp != NULL) {
1575 orig_resid = 0;
1576 while (*controlp != NULL)
1577 controlp = &(*controlp)->m_next;
1578 }
1579 cm = cmn;
1580 }
1581 if (m != NULL)
1582 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1583 else
1584 nextrecord = so->so_rcv.sb_mb;
1585 orig_resid = 0;
1586 }
1587 if (m != NULL) {
1588 if ((flags & MSG_PEEK) == 0) {
1589 KASSERT(m->m_nextpkt == nextrecord,
1590 ("soreceive: post-control, nextrecord !sync"));
1591 if (nextrecord == NULL) {
1592 KASSERT(so->so_rcv.sb_mb == m,
1593 ("soreceive: post-control, sb_mb!=m"));
1594 KASSERT(so->so_rcv.sb_lastrecord == m,
1595 ("soreceive: post-control, lastrecord!=m"));
1596 }
1597 }
1598 type = m->m_type;
1599 if (type == MT_OOBDATA)
1600 flags |= MSG_OOB;
1601 } else {
1602 if ((flags & MSG_PEEK) == 0) {
1603 KASSERT(so->so_rcv.sb_mb == nextrecord,
1604 ("soreceive: sb_mb != nextrecord"));
1605 if (so->so_rcv.sb_mb == NULL) {
1606 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1607 ("soreceive: sb_lastercord != NULL"));
1608 }
1609 }
1610 }
1611 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1612 SBLASTRECORDCHK(&so->so_rcv);
1613 SBLASTMBUFCHK(&so->so_rcv);
1614
1615 /*
1616 * Now continue to read any data mbufs off of the head of the socket
1617 * buffer until the read request is satisfied. Note that 'type' is
1618 * used to store the type of any mbuf reads that have happened so far
1619 * such that soreceive() can stop reading if the type changes, which
1620 * causes soreceive() to return only one of regular data and inline
1621 * out-of-band data in a single socket receive operation.
1622 */
1623 moff = 0;
1624 offset = 0;
1625 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1626 /*
1627 * If the type of mbuf has changed since the last mbuf
1628 * examined ('type'), end the receive operation.
1629 */
1630 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1631 if (m->m_type == MT_OOBDATA) {
1632 if (type != MT_OOBDATA)
1633 break;
1634 } else if (type == MT_OOBDATA)
1635 break;
1636 else
1637 KASSERT(m->m_type == MT_DATA,
1638 ("m->m_type == %d", m->m_type));
1639 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1640 len = uio->uio_resid;
1641 if (so->so_oobmark && len > so->so_oobmark - offset)
1642 len = so->so_oobmark - offset;
1643 if (len > m->m_len - moff)
1644 len = m->m_len - moff;
1645 /*
1646 * If mp is set, just pass back the mbufs. Otherwise copy
1647 * them out via the uio, then free. Sockbuf must be
1648 * consistent here (points to current mbuf, it points to next
1649 * record) when we drop priority; we must note any additions
1650 * to the sockbuf when we block interrupts again.
1651 */
1652 if (mp == NULL) {
1653 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1654 SBLASTRECORDCHK(&so->so_rcv);
1655 SBLASTMBUFCHK(&so->so_rcv);
1656 SOCKBUF_UNLOCK(&so->so_rcv);
1657 #ifdef ZERO_COPY_SOCKETS
1658 if (so_zero_copy_receive) {
1659 int disposable;
1660
1661 if ((m->m_flags & M_EXT)
1662 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1663 disposable = 1;
1664 else
1665 disposable = 0;
1666
1667 error = uiomoveco(mtod(m, char *) + moff,
1668 (int)len, uio,
1669 disposable);
1670 } else
1671 #endif /* ZERO_COPY_SOCKETS */
1672 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1673 SOCKBUF_LOCK(&so->so_rcv);
1674 if (error) {
1675 /*
1676 * The MT_SONAME mbuf has already been removed
1677 * from the record, so it is necessary to
1678 * remove the data mbufs, if any, to preserve
1679 * the invariant in the case of PR_ADDR that
1680 * requires MT_SONAME mbufs at the head of
1681 * each record.
1682 */
1683 if (m && pr->pr_flags & PR_ATOMIC &&
1684 ((flags & MSG_PEEK) == 0))
1685 (void)sbdroprecord_locked(&so->so_rcv);
1686 SOCKBUF_UNLOCK(&so->so_rcv);
1687 goto release;
1688 }
1689 } else
1690 uio->uio_resid -= len;
1691 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1692 if (len == m->m_len - moff) {
1693 if (m->m_flags & M_EOR)
1694 flags |= MSG_EOR;
1695 if (flags & MSG_PEEK) {
1696 m = m->m_next;
1697 moff = 0;
1698 } else {
1699 nextrecord = m->m_nextpkt;
1700 sbfree(&so->so_rcv, m);
1701 if (mp != NULL) {
1702 *mp = m;
1703 mp = &m->m_next;
1704 so->so_rcv.sb_mb = m = m->m_next;
1705 *mp = NULL;
1706 } else {
1707 so->so_rcv.sb_mb = m_free(m);
1708 m = so->so_rcv.sb_mb;
1709 }
1710 sockbuf_pushsync(&so->so_rcv, nextrecord);
1711 SBLASTRECORDCHK(&so->so_rcv);
1712 SBLASTMBUFCHK(&so->so_rcv);
1713 }
1714 } else {
1715 if (flags & MSG_PEEK)
1716 moff += len;
1717 else {
1718 if (mp != NULL) {
1719 int copy_flag;
1720
1721 if (flags & MSG_DONTWAIT)
1722 copy_flag = M_DONTWAIT;
1723 else
1724 copy_flag = M_TRYWAIT;
1725 if (copy_flag == M_TRYWAIT)
1726 SOCKBUF_UNLOCK(&so->so_rcv);
1727 *mp = m_copym(m, 0, len, copy_flag);
1728 if (copy_flag == M_TRYWAIT)
1729 SOCKBUF_LOCK(&so->so_rcv);
1730 if (*mp == NULL) {
1731 /*
1732 * m_copym() couldn't
1733 * allocate an mbuf. Adjust
1734 * uio_resid back (it was
1735 * adjusted down by len
1736 * bytes, which we didn't end
1737 * up "copying" over).
1738 */
1739 uio->uio_resid += len;
1740 break;
1741 }
1742 }
1743 m->m_data += len;
1744 m->m_len -= len;
1745 so->so_rcv.sb_cc -= len;
1746 }
1747 }
1748 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1749 if (so->so_oobmark) {
1750 if ((flags & MSG_PEEK) == 0) {
1751 so->so_oobmark -= len;
1752 if (so->so_oobmark == 0) {
1753 so->so_rcv.sb_state |= SBS_RCVATMARK;
1754 break;
1755 }
1756 } else {
1757 offset += len;
1758 if (offset == so->so_oobmark)
1759 break;
1760 }
1761 }
1762 if (flags & MSG_EOR)
1763 break;
1764 /*
1765 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1766 * must not quit until "uio->uio_resid == 0" or an error
1767 * termination. If a signal/timeout occurs, return with a
1768 * short count but without error. Keep sockbuf locked
1769 * against other readers.
1770 */
1771 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1772 !sosendallatonce(so) && nextrecord == NULL) {
1773 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1774 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1775 break;
1776 /*
1777 * Notify the protocol that some data has been
1778 * drained before blocking.
1779 */
1780 if (pr->pr_flags & PR_WANTRCVD) {
1781 SOCKBUF_UNLOCK(&so->so_rcv);
1782 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1783 SOCKBUF_LOCK(&so->so_rcv);
1784 }
1785 SBLASTRECORDCHK(&so->so_rcv);
1786 SBLASTMBUFCHK(&so->so_rcv);
1787 error = sbwait(&so->so_rcv);
1788 if (error) {
1789 SOCKBUF_UNLOCK(&so->so_rcv);
1790 goto release;
1791 }
1792 m = so->so_rcv.sb_mb;
1793 if (m != NULL)
1794 nextrecord = m->m_nextpkt;
1795 }
1796 }
1797
1798 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1799 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1800 flags |= MSG_TRUNC;
1801 if ((flags & MSG_PEEK) == 0)
1802 (void) sbdroprecord_locked(&so->so_rcv);
1803 }
1804 if ((flags & MSG_PEEK) == 0) {
1805 if (m == NULL) {
1806 /*
1807 * First part is an inline SB_EMPTY_FIXUP(). Second
1808 * part makes sure sb_lastrecord is up-to-date if
1809 * there is still data in the socket buffer.
1810 */
1811 so->so_rcv.sb_mb = nextrecord;
1812 if (so->so_rcv.sb_mb == NULL) {
1813 so->so_rcv.sb_mbtail = NULL;
1814 so->so_rcv.sb_lastrecord = NULL;
1815 } else if (nextrecord->m_nextpkt == NULL)
1816 so->so_rcv.sb_lastrecord = nextrecord;
1817 }
1818 SBLASTRECORDCHK(&so->so_rcv);
1819 SBLASTMBUFCHK(&so->so_rcv);
1820 /*
1821 * If soreceive() is being done from the socket callback,
1822 * then don't need to generate ACK to peer to update window,
1823 * since ACK will be generated on return to TCP.
1824 */
1825 if (!(flags & MSG_SOCALLBCK) &&
1826 (pr->pr_flags & PR_WANTRCVD)) {
1827 SOCKBUF_UNLOCK(&so->so_rcv);
1828 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1829 SOCKBUF_LOCK(&so->so_rcv);
1830 }
1831 }
1832 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1833 if (orig_resid == uio->uio_resid && orig_resid &&
1834 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1835 SOCKBUF_UNLOCK(&so->so_rcv);
1836 goto restart;
1837 }
1838 SOCKBUF_UNLOCK(&so->so_rcv);
1839
1840 if (flagsp != NULL)
1841 *flagsp |= flags;
1842 release:
1843 sbunlock(&so->so_rcv);
1844 return (error);
1845 }
1846
1847 /*
1848 * Optimized version of soreceive() for simple datagram cases from userspace.
1849 * Unlike in the stream case, we're able to drop a datagram if copyout()
1850 * fails, and because we handle datagrams atomically, we don't need to use a
1851 * sleep lock to prevent I/O interlacing.
1852 */
1853 int
1854 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
1855 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1856 {
1857 struct mbuf *m, *m2;
1858 int flags, len, error;
1859 struct protosw *pr = so->so_proto;
1860 struct mbuf *nextrecord;
1861
1862 if (psa != NULL)
1863 *psa = NULL;
1864 if (controlp != NULL)
1865 *controlp = NULL;
1866 if (flagsp != NULL)
1867 flags = *flagsp &~ MSG_EOR;
1868 else
1869 flags = 0;
1870
1871 /*
1872 * For any complicated cases, fall back to the full
1873 * soreceive_generic().
1874 */
1875 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
1876 return (soreceive_generic(so, psa, uio, mp0, controlp,
1877 flagsp));
1878
1879 /*
1880 * Enforce restrictions on use.
1881 */
1882 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
1883 ("soreceive_dgram: wantrcvd"));
1884 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
1885 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
1886 ("soreceive_dgram: SBS_RCVATMARK"));
1887 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
1888 ("soreceive_dgram: P_CONNREQUIRED"));
1889
1890 /*
1891 * Loop blocking while waiting for a datagram.
1892 */
1893 SOCKBUF_LOCK(&so->so_rcv);
1894 while ((m = so->so_rcv.sb_mb) == NULL) {
1895 KASSERT(so->so_rcv.sb_cc == 0,
1896 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
1897 so->so_rcv.sb_cc));
1898 if (so->so_error) {
1899 error = so->so_error;
1900 so->so_error = 0;
1901 SOCKBUF_UNLOCK(&so->so_rcv);
1902 return (error);
1903 }
1904 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
1905 uio->uio_resid == 0) {
1906 SOCKBUF_UNLOCK(&so->so_rcv);
1907 return (0);
1908 }
1909 if ((so->so_state & SS_NBIO) ||
1910 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1911 SOCKBUF_UNLOCK(&so->so_rcv);
1912 return (EWOULDBLOCK);
1913 }
1914 SBLASTRECORDCHK(&so->so_rcv);
1915 SBLASTMBUFCHK(&so->so_rcv);
1916 error = sbwait(&so->so_rcv);
1917 if (error) {
1918 SOCKBUF_UNLOCK(&so->so_rcv);
1919 return (error);
1920 }
1921 }
1922 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1923
1924 if (uio->uio_td)
1925 uio->uio_td->td_ru.ru_msgrcv++;
1926 SBLASTRECORDCHK(&so->so_rcv);
1927 SBLASTMBUFCHK(&so->so_rcv);
1928 nextrecord = m->m_nextpkt;
1929 if (nextrecord == NULL) {
1930 KASSERT(so->so_rcv.sb_lastrecord == m,
1931 ("soreceive_dgram: lastrecord != m"));
1932 }
1933
1934 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
1935 ("soreceive_dgram: m_nextpkt != nextrecord"));
1936
1937 /*
1938 * Pull 'm' and its chain off the front of the packet queue.
1939 */
1940 so->so_rcv.sb_mb = NULL;
1941 sockbuf_pushsync(&so->so_rcv, nextrecord);
1942
1943 /*
1944 * Walk 'm's chain and free that many bytes from the socket buffer.
1945 */
1946 for (m2 = m; m2 != NULL; m2 = m2->m_next)
1947 sbfree(&so->so_rcv, m2);
1948
1949 /*
1950 * Do a few last checks before we let go of the lock.
1951 */
1952 SBLASTRECORDCHK(&so->so_rcv);
1953 SBLASTMBUFCHK(&so->so_rcv);
1954 SOCKBUF_UNLOCK(&so->so_rcv);
1955
1956 if (pr->pr_flags & PR_ADDR) {
1957 KASSERT(m->m_type == MT_SONAME,
1958 ("m->m_type == %d", m->m_type));
1959 if (psa != NULL)
1960 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1961 M_NOWAIT);
1962 m = m_free(m);
1963 }
1964 if (m == NULL) {
1965 /* XXXRW: Can this happen? */
1966 return (0);
1967 }
1968
1969 /*
1970 * Packet to copyout() is now in 'm' and it is disconnected from the
1971 * queue.
1972 *
1973 * Process one or more MT_CONTROL mbufs present before any data mbufs
1974 * in the first mbuf chain on the socket buffer. We call into the
1975 * protocol to perform externalization (or freeing if controlp ==
1976 * NULL).
1977 */
1978 if (m->m_type == MT_CONTROL) {
1979 struct mbuf *cm = NULL, *cmn;
1980 struct mbuf **cme = &cm;
1981
1982 do {
1983 m2 = m->m_next;
1984 m->m_next = NULL;
1985 *cme = m;
1986 cme = &(*cme)->m_next;
1987 m = m2;
1988 } while (m != NULL && m->m_type == MT_CONTROL);
1989 while (cm != NULL) {
1990 cmn = cm->m_next;
1991 cm->m_next = NULL;
1992 if (pr->pr_domain->dom_externalize != NULL) {
1993 error = (*pr->pr_domain->dom_externalize)
1994 (cm, controlp);
1995 } else if (controlp != NULL)
1996 *controlp = cm;
1997 else
1998 m_freem(cm);
1999 if (controlp != NULL) {
2000 while (*controlp != NULL)
2001 controlp = &(*controlp)->m_next;
2002 }
2003 cm = cmn;
2004 }
2005 }
2006 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2007
2008 while (m != NULL && uio->uio_resid > 0) {
2009 len = uio->uio_resid;
2010 if (len > m->m_len)
2011 len = m->m_len;
2012 error = uiomove(mtod(m, char *), (int)len, uio);
2013 if (error) {
2014 m_freem(m);
2015 return (error);
2016 }
2017 m = m_free(m);
2018 }
2019 if (m != NULL)
2020 flags |= MSG_TRUNC;
2021 m_freem(m);
2022 if (flagsp != NULL)
2023 *flagsp |= flags;
2024 return (0);
2025 }
2026
2027 int
2028 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2029 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2030 {
2031
2032 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2033 controlp, flagsp));
2034 }
2035
2036 int
2037 soshutdown(struct socket *so, int how)
2038 {
2039 struct protosw *pr = so->so_proto;
2040
2041 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2042 return (EINVAL);
2043 if (pr->pr_usrreqs->pru_flush != NULL) {
2044 (*pr->pr_usrreqs->pru_flush)(so, how);
2045 }
2046 if (how != SHUT_WR)
2047 sorflush(so);
2048 if (how != SHUT_RD)
2049 return ((*pr->pr_usrreqs->pru_shutdown)(so));
2050 return (0);
2051 }
2052
2053 void
2054 sorflush(struct socket *so)
2055 {
2056 struct sockbuf *sb = &so->so_rcv;
2057 struct protosw *pr = so->so_proto;
2058 struct sockbuf asb;
2059
2060 /*
2061 * In order to avoid calling dom_dispose with the socket buffer mutex
2062 * held, and in order to generally avoid holding the lock for a long
2063 * time, we make a copy of the socket buffer and clear the original
2064 * (except locks, state). The new socket buffer copy won't have
2065 * initialized locks so we can only call routines that won't use or
2066 * assert those locks.
2067 *
2068 * Dislodge threads currently blocked in receive and wait to acquire
2069 * a lock against other simultaneous readers before clearing the
2070 * socket buffer. Don't let our acquire be interrupted by a signal
2071 * despite any existing socket disposition on interruptable waiting.
2072 */
2073 socantrcvmore(so);
2074 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2075
2076 /*
2077 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2078 * and mutex data unchanged.
2079 */
2080 SOCKBUF_LOCK(sb);
2081 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2082 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2083 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2084 bzero(&sb->sb_startzero,
2085 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2086 SOCKBUF_UNLOCK(sb);
2087 sbunlock(sb);
2088
2089 /*
2090 * Dispose of special rights and flush the socket buffer. Don't call
2091 * any unsafe routines (that rely on locks being initialized) on asb.
2092 */
2093 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2094 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2095 sbrelease_internal(&asb, so);
2096 }
2097
2098 /*
2099 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2100 * additional variant to handle the case where the option value needs to be
2101 * some kind of integer, but not a specific size. In addition to their use
2102 * here, these functions are also called by the protocol-level pr_ctloutput()
2103 * routines.
2104 */
2105 int
2106 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2107 {
2108 size_t valsize;
2109
2110 /*
2111 * If the user gives us more than we wanted, we ignore it, but if we
2112 * don't get the minimum length the caller wants, we return EINVAL.
2113 * On success, sopt->sopt_valsize is set to however much we actually
2114 * retrieved.
2115 */
2116 if ((valsize = sopt->sopt_valsize) < minlen)
2117 return EINVAL;
2118 if (valsize > len)
2119 sopt->sopt_valsize = valsize = len;
2120
2121 if (sopt->sopt_td != NULL)
2122 return (copyin(sopt->sopt_val, buf, valsize));
2123
2124 bcopy(sopt->sopt_val, buf, valsize);
2125 return (0);
2126 }
2127
2128 /*
2129 * Kernel version of setsockopt(2).
2130 *
2131 * XXX: optlen is size_t, not socklen_t
2132 */
2133 int
2134 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2135 size_t optlen)
2136 {
2137 struct sockopt sopt;
2138
2139 sopt.sopt_level = level;
2140 sopt.sopt_name = optname;
2141 sopt.sopt_dir = SOPT_SET;
2142 sopt.sopt_val = optval;
2143 sopt.sopt_valsize = optlen;
2144 sopt.sopt_td = NULL;
2145 return (sosetopt(so, &sopt));
2146 }
2147
2148 int
2149 sosetopt(struct socket *so, struct sockopt *sopt)
2150 {
2151 int error, optval;
2152 struct linger l;
2153 struct timeval tv;
2154 u_long val;
2155 #ifdef MAC
2156 struct mac extmac;
2157 #endif
2158
2159 error = 0;
2160 if (sopt->sopt_level != SOL_SOCKET) {
2161 if (so->so_proto && so->so_proto->pr_ctloutput)
2162 return ((*so->so_proto->pr_ctloutput)
2163 (so, sopt));
2164 error = ENOPROTOOPT;
2165 } else {
2166 switch (sopt->sopt_name) {
2167 #ifdef INET
2168 case SO_ACCEPTFILTER:
2169 error = do_setopt_accept_filter(so, sopt);
2170 if (error)
2171 goto bad;
2172 break;
2173 #endif
2174 case SO_LINGER:
2175 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2176 if (error)
2177 goto bad;
2178
2179 SOCK_LOCK(so);
2180 so->so_linger = l.l_linger;
2181 if (l.l_onoff)
2182 so->so_options |= SO_LINGER;
2183 else
2184 so->so_options &= ~SO_LINGER;
2185 SOCK_UNLOCK(so);
2186 break;
2187
2188 case SO_DEBUG:
2189 case SO_KEEPALIVE:
2190 case SO_DONTROUTE:
2191 case SO_USELOOPBACK:
2192 case SO_BROADCAST:
2193 case SO_REUSEADDR:
2194 case SO_REUSEPORT:
2195 case SO_OOBINLINE:
2196 case SO_TIMESTAMP:
2197 case SO_BINTIME:
2198 case SO_NOSIGPIPE:
2199 error = sooptcopyin(sopt, &optval, sizeof optval,
2200 sizeof optval);
2201 if (error)
2202 goto bad;
2203 SOCK_LOCK(so);
2204 if (optval)
2205 so->so_options |= sopt->sopt_name;
2206 else
2207 so->so_options &= ~sopt->sopt_name;
2208 SOCK_UNLOCK(so);
2209 break;
2210
2211 case SO_SETFIB:
2212 error = sooptcopyin(sopt, &optval, sizeof optval,
2213 sizeof optval);
2214 if (optval < 1 || optval > rt_numfibs) {
2215 error = EINVAL;
2216 goto bad;
2217 }
2218 if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2219 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2220 so->so_fibnum = optval;
2221 /* Note: ignore error */
2222 if (so->so_proto && so->so_proto->pr_ctloutput)
2223 (*so->so_proto->pr_ctloutput)(so, sopt);
2224 } else {
2225 so->so_fibnum = 0;
2226 }
2227 break;
2228 case SO_SNDBUF:
2229 case SO_RCVBUF:
2230 case SO_SNDLOWAT:
2231 case SO_RCVLOWAT:
2232 error = sooptcopyin(sopt, &optval, sizeof optval,
2233 sizeof optval);
2234 if (error)
2235 goto bad;
2236
2237 /*
2238 * Values < 1 make no sense for any of these options,
2239 * so disallow them.
2240 */
2241 if (optval < 1) {
2242 error = EINVAL;
2243 goto bad;
2244 }
2245
2246 switch (sopt->sopt_name) {
2247 case SO_SNDBUF:
2248 case SO_RCVBUF:
2249 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2250 &so->so_snd : &so->so_rcv, (u_long)optval,
2251 so, curthread) == 0) {
2252 error = ENOBUFS;
2253 goto bad;
2254 }
2255 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2256 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2257 break;
2258
2259 /*
2260 * Make sure the low-water is never greater than the
2261 * high-water.
2262 */
2263 case SO_SNDLOWAT:
2264 SOCKBUF_LOCK(&so->so_snd);
2265 so->so_snd.sb_lowat =
2266 (optval > so->so_snd.sb_hiwat) ?
2267 so->so_snd.sb_hiwat : optval;
2268 SOCKBUF_UNLOCK(&so->so_snd);
2269 break;
2270 case SO_RCVLOWAT:
2271 SOCKBUF_LOCK(&so->so_rcv);
2272 so->so_rcv.sb_lowat =
2273 (optval > so->so_rcv.sb_hiwat) ?
2274 so->so_rcv.sb_hiwat : optval;
2275 SOCKBUF_UNLOCK(&so->so_rcv);
2276 break;
2277 }
2278 break;
2279
2280 case SO_SNDTIMEO:
2281 case SO_RCVTIMEO:
2282 #ifdef COMPAT_IA32
2283 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2284 struct timeval32 tv32;
2285
2286 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2287 sizeof tv32);
2288 CP(tv32, tv, tv_sec);
2289 CP(tv32, tv, tv_usec);
2290 } else
2291 #endif
2292 error = sooptcopyin(sopt, &tv, sizeof tv,
2293 sizeof tv);
2294 if (error)
2295 goto bad;
2296
2297 /* assert(hz > 0); */
2298 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2299 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2300 error = EDOM;
2301 goto bad;
2302 }
2303 /* assert(tick > 0); */
2304 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2305 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2306 if (val > INT_MAX) {
2307 error = EDOM;
2308 goto bad;
2309 }
2310 if (val == 0 && tv.tv_usec != 0)
2311 val = 1;
2312
2313 switch (sopt->sopt_name) {
2314 case SO_SNDTIMEO:
2315 so->so_snd.sb_timeo = val;
2316 break;
2317 case SO_RCVTIMEO:
2318 so->so_rcv.sb_timeo = val;
2319 break;
2320 }
2321 break;
2322
2323 case SO_LABEL:
2324 #ifdef MAC
2325 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2326 sizeof extmac);
2327 if (error)
2328 goto bad;
2329 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2330 so, &extmac);
2331 #else
2332 error = EOPNOTSUPP;
2333 #endif
2334 break;
2335
2336 default:
2337 error = ENOPROTOOPT;
2338 break;
2339 }
2340 if (error == 0 && so->so_proto != NULL &&
2341 so->so_proto->pr_ctloutput != NULL) {
2342 (void) ((*so->so_proto->pr_ctloutput)
2343 (so, sopt));
2344 }
2345 }
2346 bad:
2347 return (error);
2348 }
2349
2350 /*
2351 * Helper routine for getsockopt.
2352 */
2353 int
2354 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2355 {
2356 int error;
2357 size_t valsize;
2358
2359 error = 0;
2360
2361 /*
2362 * Documented get behavior is that we always return a value, possibly
2363 * truncated to fit in the user's buffer. Traditional behavior is
2364 * that we always tell the user precisely how much we copied, rather
2365 * than something useful like the total amount we had available for
2366 * her. Note that this interface is not idempotent; the entire
2367 * answer must generated ahead of time.
2368 */
2369 valsize = min(len, sopt->sopt_valsize);
2370 sopt->sopt_valsize = valsize;
2371 if (sopt->sopt_val != NULL) {
2372 if (sopt->sopt_td != NULL)
2373 error = copyout(buf, sopt->sopt_val, valsize);
2374 else
2375 bcopy(buf, sopt->sopt_val, valsize);
2376 }
2377 return (error);
2378 }
2379
2380 int
2381 sogetopt(struct socket *so, struct sockopt *sopt)
2382 {
2383 int error, optval;
2384 struct linger l;
2385 struct timeval tv;
2386 #ifdef MAC
2387 struct mac extmac;
2388 #endif
2389
2390 error = 0;
2391 if (sopt->sopt_level != SOL_SOCKET) {
2392 if (so->so_proto && so->so_proto->pr_ctloutput) {
2393 return ((*so->so_proto->pr_ctloutput)
2394 (so, sopt));
2395 } else
2396 return (ENOPROTOOPT);
2397 } else {
2398 switch (sopt->sopt_name) {
2399 #ifdef INET
2400 case SO_ACCEPTFILTER:
2401 error = do_getopt_accept_filter(so, sopt);
2402 break;
2403 #endif
2404 case SO_LINGER:
2405 SOCK_LOCK(so);
2406 l.l_onoff = so->so_options & SO_LINGER;
2407 l.l_linger = so->so_linger;
2408 SOCK_UNLOCK(so);
2409 error = sooptcopyout(sopt, &l, sizeof l);
2410 break;
2411
2412 case SO_USELOOPBACK:
2413 case SO_DONTROUTE:
2414 case SO_DEBUG:
2415 case SO_KEEPALIVE:
2416 case SO_REUSEADDR:
2417 case SO_REUSEPORT:
2418 case SO_BROADCAST:
2419 case SO_OOBINLINE:
2420 case SO_ACCEPTCONN:
2421 case SO_TIMESTAMP:
2422 case SO_BINTIME:
2423 case SO_NOSIGPIPE:
2424 optval = so->so_options & sopt->sopt_name;
2425 integer:
2426 error = sooptcopyout(sopt, &optval, sizeof optval);
2427 break;
2428
2429 case SO_TYPE:
2430 optval = so->so_type;
2431 goto integer;
2432
2433 case SO_ERROR:
2434 SOCK_LOCK(so);
2435 optval = so->so_error;
2436 so->so_error = 0;
2437 SOCK_UNLOCK(so);
2438 goto integer;
2439
2440 case SO_SNDBUF:
2441 optval = so->so_snd.sb_hiwat;
2442 goto integer;
2443
2444 case SO_RCVBUF:
2445 optval = so->so_rcv.sb_hiwat;
2446 goto integer;
2447
2448 case SO_SNDLOWAT:
2449 optval = so->so_snd.sb_lowat;
2450 goto integer;
2451
2452 case SO_RCVLOWAT:
2453 optval = so->so_rcv.sb_lowat;
2454 goto integer;
2455
2456 case SO_SNDTIMEO:
2457 case SO_RCVTIMEO:
2458 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2459 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2460
2461 tv.tv_sec = optval / hz;
2462 tv.tv_usec = (optval % hz) * tick;
2463 #ifdef COMPAT_IA32
2464 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2465 struct timeval32 tv32;
2466
2467 CP(tv, tv32, tv_sec);
2468 CP(tv, tv32, tv_usec);
2469 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2470 } else
2471 #endif
2472 error = sooptcopyout(sopt, &tv, sizeof tv);
2473 break;
2474
2475 case SO_LABEL:
2476 #ifdef MAC
2477 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2478 sizeof(extmac));
2479 if (error)
2480 return (error);
2481 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2482 so, &extmac);
2483 if (error)
2484 return (error);
2485 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2486 #else
2487 error = EOPNOTSUPP;
2488 #endif
2489 break;
2490
2491 case SO_PEERLABEL:
2492 #ifdef MAC
2493 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2494 sizeof(extmac));
2495 if (error)
2496 return (error);
2497 error = mac_getsockopt_peerlabel(
2498 sopt->sopt_td->td_ucred, so, &extmac);
2499 if (error)
2500 return (error);
2501 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2502 #else
2503 error = EOPNOTSUPP;
2504 #endif
2505 break;
2506
2507 case SO_LISTENQLIMIT:
2508 optval = so->so_qlimit;
2509 goto integer;
2510
2511 case SO_LISTENQLEN:
2512 optval = so->so_qlen;
2513 goto integer;
2514
2515 case SO_LISTENINCQLEN:
2516 optval = so->so_incqlen;
2517 goto integer;
2518
2519 default:
2520 error = ENOPROTOOPT;
2521 break;
2522 }
2523 return (error);
2524 }
2525 }
2526
2527 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2528 int
2529 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2530 {
2531 struct mbuf *m, *m_prev;
2532 int sopt_size = sopt->sopt_valsize;
2533
2534 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2535 if (m == NULL)
2536 return ENOBUFS;
2537 if (sopt_size > MLEN) {
2538 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
2539 if ((m->m_flags & M_EXT) == 0) {
2540 m_free(m);
2541 return ENOBUFS;
2542 }
2543 m->m_len = min(MCLBYTES, sopt_size);
2544 } else {
2545 m->m_len = min(MLEN, sopt_size);
2546 }
2547 sopt_size -= m->m_len;
2548 *mp = m;
2549 m_prev = m;
2550
2551 while (sopt_size) {
2552 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2553 if (m == NULL) {
2554 m_freem(*mp);
2555 return ENOBUFS;
2556 }
2557 if (sopt_size > MLEN) {
2558 MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
2559 M_DONTWAIT);
2560 if ((m->m_flags & M_EXT) == 0) {
2561 m_freem(m);
2562 m_freem(*mp);
2563 return ENOBUFS;
2564 }
2565 m->m_len = min(MCLBYTES, sopt_size);
2566 } else {
2567 m->m_len = min(MLEN, sopt_size);
2568 }
2569 sopt_size -= m->m_len;
2570 m_prev->m_next = m;
2571 m_prev = m;
2572 }
2573 return (0);
2574 }
2575
2576 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2577 int
2578 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2579 {
2580 struct mbuf *m0 = m;
2581
2582 if (sopt->sopt_val == NULL)
2583 return (0);
2584 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2585 if (sopt->sopt_td != NULL) {
2586 int error;
2587
2588 error = copyin(sopt->sopt_val, mtod(m, char *),
2589 m->m_len);
2590 if (error != 0) {
2591 m_freem(m0);
2592 return(error);
2593 }
2594 } else
2595 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2596 sopt->sopt_valsize -= m->m_len;
2597 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2598 m = m->m_next;
2599 }
2600 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2601 panic("ip6_sooptmcopyin");
2602 return (0);
2603 }
2604
2605 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2606 int
2607 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2608 {
2609 struct mbuf *m0 = m;
2610 size_t valsize = 0;
2611
2612 if (sopt->sopt_val == NULL)
2613 return (0);
2614 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2615 if (sopt->sopt_td != NULL) {
2616 int error;
2617
2618 error = copyout(mtod(m, char *), sopt->sopt_val,
2619 m->m_len);
2620 if (error != 0) {
2621 m_freem(m0);
2622 return(error);
2623 }
2624 } else
2625 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2626 sopt->sopt_valsize -= m->m_len;
2627 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2628 valsize += m->m_len;
2629 m = m->m_next;
2630 }
2631 if (m != NULL) {
2632 /* enough soopt buffer should be given from user-land */
2633 m_freem(m0);
2634 return(EINVAL);
2635 }
2636 sopt->sopt_valsize = valsize;
2637 return (0);
2638 }
2639
2640 /*
2641 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2642 * out-of-band data, which will then notify socket consumers.
2643 */
2644 void
2645 sohasoutofband(struct socket *so)
2646 {
2647
2648 if (so->so_sigio != NULL)
2649 pgsigio(&so->so_sigio, SIGURG, 0);
2650 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2651 }
2652
2653 int
2654 sopoll(struct socket *so, int events, struct ucred *active_cred,
2655 struct thread *td)
2656 {
2657
2658 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2659 td));
2660 }
2661
2662 int
2663 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2664 struct thread *td)
2665 {
2666 int revents = 0;
2667
2668 SOCKBUF_LOCK(&so->so_snd);
2669 SOCKBUF_LOCK(&so->so_rcv);
2670 if (events & (POLLIN | POLLRDNORM))
2671 if (soreadable(so))
2672 revents |= events & (POLLIN | POLLRDNORM);
2673
2674 if (events & POLLINIGNEOF)
2675 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2676 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2677 revents |= POLLINIGNEOF;
2678
2679 if (events & (POLLOUT | POLLWRNORM))
2680 if (sowriteable(so))
2681 revents |= events & (POLLOUT | POLLWRNORM);
2682
2683 if (events & (POLLPRI | POLLRDBAND))
2684 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2685 revents |= events & (POLLPRI | POLLRDBAND);
2686
2687 if (revents == 0) {
2688 if (events &
2689 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2690 POLLRDBAND)) {
2691 selrecord(td, &so->so_rcv.sb_sel);
2692 so->so_rcv.sb_flags |= SB_SEL;
2693 }
2694
2695 if (events & (POLLOUT | POLLWRNORM)) {
2696 selrecord(td, &so->so_snd.sb_sel);
2697 so->so_snd.sb_flags |= SB_SEL;
2698 }
2699 }
2700
2701 SOCKBUF_UNLOCK(&so->so_rcv);
2702 SOCKBUF_UNLOCK(&so->so_snd);
2703 return (revents);
2704 }
2705
2706 int
2707 soo_kqfilter(struct file *fp, struct knote *kn)
2708 {
2709 struct socket *so = kn->kn_fp->f_data;
2710 struct sockbuf *sb;
2711
2712 switch (kn->kn_filter) {
2713 case EVFILT_READ:
2714 if (so->so_options & SO_ACCEPTCONN)
2715 kn->kn_fop = &solisten_filtops;
2716 else
2717 kn->kn_fop = &soread_filtops;
2718 sb = &so->so_rcv;
2719 break;
2720 case EVFILT_WRITE:
2721 kn->kn_fop = &sowrite_filtops;
2722 sb = &so->so_snd;
2723 break;
2724 default:
2725 return (EINVAL);
2726 }
2727
2728 SOCKBUF_LOCK(sb);
2729 knlist_add(&sb->sb_sel.si_note, kn, 1);
2730 sb->sb_flags |= SB_KNOTE;
2731 SOCKBUF_UNLOCK(sb);
2732 return (0);
2733 }
2734
2735 /*
2736 * Some routines that return EOPNOTSUPP for entry points that are not
2737 * supported by a protocol. Fill in as needed.
2738 */
2739 int
2740 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2741 {
2742
2743 return EOPNOTSUPP;
2744 }
2745
2746 int
2747 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2748 {
2749
2750 return EOPNOTSUPP;
2751 }
2752
2753 int
2754 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2755 {
2756
2757 return EOPNOTSUPP;
2758 }
2759
2760 int
2761 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2762 {
2763
2764 return EOPNOTSUPP;
2765 }
2766
2767 int
2768 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2769 {
2770
2771 return EOPNOTSUPP;
2772 }
2773
2774 int
2775 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2776 struct ifnet *ifp, struct thread *td)
2777 {
2778
2779 return EOPNOTSUPP;
2780 }
2781
2782 int
2783 pru_disconnect_notsupp(struct socket *so)
2784 {
2785
2786 return EOPNOTSUPP;
2787 }
2788
2789 int
2790 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
2791 {
2792
2793 return EOPNOTSUPP;
2794 }
2795
2796 int
2797 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
2798 {
2799
2800 return EOPNOTSUPP;
2801 }
2802
2803 int
2804 pru_rcvd_notsupp(struct socket *so, int flags)
2805 {
2806
2807 return EOPNOTSUPP;
2808 }
2809
2810 int
2811 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
2812 {
2813
2814 return EOPNOTSUPP;
2815 }
2816
2817 int
2818 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
2819 struct sockaddr *addr, struct mbuf *control, struct thread *td)
2820 {
2821
2822 return EOPNOTSUPP;
2823 }
2824
2825 /*
2826 * This isn't really a ``null'' operation, but it's the default one and
2827 * doesn't do anything destructive.
2828 */
2829 int
2830 pru_sense_null(struct socket *so, struct stat *sb)
2831 {
2832
2833 sb->st_blksize = so->so_snd.sb_hiwat;
2834 return 0;
2835 }
2836
2837 int
2838 pru_shutdown_notsupp(struct socket *so)
2839 {
2840
2841 return EOPNOTSUPP;
2842 }
2843
2844 int
2845 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
2846 {
2847
2848 return EOPNOTSUPP;
2849 }
2850
2851 int
2852 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
2853 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2854 {
2855
2856 return EOPNOTSUPP;
2857 }
2858
2859 int
2860 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
2861 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2862 {
2863
2864 return EOPNOTSUPP;
2865 }
2866
2867 int
2868 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
2869 struct thread *td)
2870 {
2871
2872 return EOPNOTSUPP;
2873 }
2874
2875 static void
2876 filt_sordetach(struct knote *kn)
2877 {
2878 struct socket *so = kn->kn_fp->f_data;
2879
2880 SOCKBUF_LOCK(&so->so_rcv);
2881 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2882 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2883 so->so_rcv.sb_flags &= ~SB_KNOTE;
2884 SOCKBUF_UNLOCK(&so->so_rcv);
2885 }
2886
2887 /*ARGSUSED*/
2888 static int
2889 filt_soread(struct knote *kn, long hint)
2890 {
2891 struct socket *so;
2892
2893 so = kn->kn_fp->f_data;
2894 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2895
2896 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2897 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2898 kn->kn_flags |= EV_EOF;
2899 kn->kn_fflags = so->so_error;
2900 return (1);
2901 } else if (so->so_error) /* temporary udp error */
2902 return (1);
2903 else if (kn->kn_sfflags & NOTE_LOWAT)
2904 return (kn->kn_data >= kn->kn_sdata);
2905 else
2906 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2907 }
2908
2909 static void
2910 filt_sowdetach(struct knote *kn)
2911 {
2912 struct socket *so = kn->kn_fp->f_data;
2913
2914 SOCKBUF_LOCK(&so->so_snd);
2915 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2916 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2917 so->so_snd.sb_flags &= ~SB_KNOTE;
2918 SOCKBUF_UNLOCK(&so->so_snd);
2919 }
2920
2921 /*ARGSUSED*/
2922 static int
2923 filt_sowrite(struct knote *kn, long hint)
2924 {
2925 struct socket *so;
2926
2927 so = kn->kn_fp->f_data;
2928 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2929 kn->kn_data = sbspace(&so->so_snd);
2930 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2931 kn->kn_flags |= EV_EOF;
2932 kn->kn_fflags = so->so_error;
2933 return (1);
2934 } else if (so->so_error) /* temporary udp error */
2935 return (1);
2936 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2937 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2938 return (0);
2939 else if (kn->kn_sfflags & NOTE_LOWAT)
2940 return (kn->kn_data >= kn->kn_sdata);
2941 else
2942 return (kn->kn_data >= so->so_snd.sb_lowat);
2943 }
2944
2945 /*ARGSUSED*/
2946 static int
2947 filt_solisten(struct knote *kn, long hint)
2948 {
2949 struct socket *so = kn->kn_fp->f_data;
2950
2951 kn->kn_data = so->so_qlen;
2952 return (! TAILQ_EMPTY(&so->so_comp));
2953 }
2954
2955 int
2956 socheckuid(struct socket *so, uid_t uid)
2957 {
2958
2959 if (so == NULL)
2960 return (EPERM);
2961 if (so->so_cred->cr_uid != uid)
2962 return (EPERM);
2963 return (0);
2964 }
2965
2966 static int
2967 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
2968 {
2969 int error;
2970 int val;
2971
2972 val = somaxconn;
2973 error = sysctl_handle_int(oidp, &val, 0, req);
2974 if (error || !req->newptr )
2975 return (error);
2976
2977 if (val < 1 || val > USHRT_MAX)
2978 return (EINVAL);
2979
2980 somaxconn = val;
2981 return (0);
2982 }
2983
2984 /*
2985 * These functions are used by protocols to notify the socket layer (and its
2986 * consumers) of state changes in the sockets driven by protocol-side events.
2987 */
2988
2989 /*
2990 * Procedures to manipulate state flags of socket and do appropriate wakeups.
2991 *
2992 * Normal sequence from the active (originating) side is that
2993 * soisconnecting() is called during processing of connect() call, resulting
2994 * in an eventual call to soisconnected() if/when the connection is
2995 * established. When the connection is torn down soisdisconnecting() is
2996 * called during processing of disconnect() call, and soisdisconnected() is
2997 * called when the connection to the peer is totally severed. The semantics
2998 * of these routines are such that connectionless protocols can call
2999 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3000 * calls when setting up a ``connection'' takes no time.
3001 *
3002 * From the passive side, a socket is created with two queues of sockets:
3003 * so_incomp for connections in progress and so_comp for connections already
3004 * made and awaiting user acceptance. As a protocol is preparing incoming
3005 * connections, it creates a socket structure queued on so_incomp by calling
3006 * sonewconn(). When the connection is established, soisconnected() is
3007 * called, and transfers the socket structure to so_comp, making it available
3008 * to accept().
3009 *
3010 * If a socket is closed with sockets on either so_incomp or so_comp, these
3011 * sockets are dropped.
3012 *
3013 * If higher-level protocols are implemented in the kernel, the wakeups done
3014 * here will sometimes cause software-interrupt process scheduling.
3015 */
3016 void
3017 soisconnecting(struct socket *so)
3018 {
3019
3020 SOCK_LOCK(so);
3021 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3022 so->so_state |= SS_ISCONNECTING;
3023 SOCK_UNLOCK(so);
3024 }
3025
3026 void
3027 soisconnected(struct socket *so)
3028 {
3029 struct socket *head;
3030
3031 ACCEPT_LOCK();
3032 SOCK_LOCK(so);
3033 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3034 so->so_state |= SS_ISCONNECTED;
3035 head = so->so_head;
3036 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3037 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3038 SOCK_UNLOCK(so);
3039 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3040 head->so_incqlen--;
3041 so->so_qstate &= ~SQ_INCOMP;
3042 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3043 head->so_qlen++;
3044 so->so_qstate |= SQ_COMP;
3045 ACCEPT_UNLOCK();
3046 sorwakeup(head);
3047 wakeup_one(&head->so_timeo);
3048 } else {
3049 ACCEPT_UNLOCK();
3050 so->so_upcall =
3051 head->so_accf->so_accept_filter->accf_callback;
3052 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
3053 so->so_rcv.sb_flags |= SB_UPCALL;
3054 so->so_options &= ~SO_ACCEPTFILTER;
3055 SOCK_UNLOCK(so);
3056 so->so_upcall(so, so->so_upcallarg, M_DONTWAIT);
3057 }
3058 return;
3059 }
3060 SOCK_UNLOCK(so);
3061 ACCEPT_UNLOCK();
3062 wakeup(&so->so_timeo);
3063 sorwakeup(so);
3064 sowwakeup(so);
3065 }
3066
3067 void
3068 soisdisconnecting(struct socket *so)
3069 {
3070
3071 /*
3072 * Note: This code assumes that SOCK_LOCK(so) and
3073 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3074 */
3075 SOCKBUF_LOCK(&so->so_rcv);
3076 so->so_state &= ~SS_ISCONNECTING;
3077 so->so_state |= SS_ISDISCONNECTING;
3078 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3079 sorwakeup_locked(so);
3080 SOCKBUF_LOCK(&so->so_snd);
3081 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3082 sowwakeup_locked(so);
3083 wakeup(&so->so_timeo);
3084 }
3085
3086 void
3087 soisdisconnected(struct socket *so)
3088 {
3089
3090 /*
3091 * Note: This code assumes that SOCK_LOCK(so) and
3092 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3093 */
3094 SOCKBUF_LOCK(&so->so_rcv);
3095 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3096 so->so_state |= SS_ISDISCONNECTED;
3097 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3098 sorwakeup_locked(so);
3099 SOCKBUF_LOCK(&so->so_snd);
3100 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3101 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3102 sowwakeup_locked(so);
3103 wakeup(&so->so_timeo);
3104 }
3105
3106 /*
3107 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3108 */
3109 struct sockaddr *
3110 sodupsockaddr(const struct sockaddr *sa, int mflags)
3111 {
3112 struct sockaddr *sa2;
3113
3114 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3115 if (sa2)
3116 bcopy(sa, sa2, sa->sa_len);
3117 return sa2;
3118 }
3119
3120 /*
3121 * Create an external-format (``xsocket'') structure using the information in
3122 * the kernel-format socket structure pointed to by so. This is done to
3123 * reduce the spew of irrelevant information over this interface, to isolate
3124 * user code from changes in the kernel structure, and potentially to provide
3125 * information-hiding if we decide that some of this information should be
3126 * hidden from users.
3127 */
3128 void
3129 sotoxsocket(struct socket *so, struct xsocket *xso)
3130 {
3131
3132 xso->xso_len = sizeof *xso;
3133 xso->xso_so = so;
3134 xso->so_type = so->so_type;
3135 xso->so_options = so->so_options;
3136 xso->so_linger = so->so_linger;
3137 xso->so_state = so->so_state;
3138 xso->so_pcb = so->so_pcb;
3139 xso->xso_protocol = so->so_proto->pr_protocol;
3140 xso->xso_family = so->so_proto->pr_domain->dom_family;
3141 xso->so_qlen = so->so_qlen;
3142 xso->so_incqlen = so->so_incqlen;
3143 xso->so_qlimit = so->so_qlimit;
3144 xso->so_timeo = so->so_timeo;
3145 xso->so_error = so->so_error;
3146 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3147 xso->so_oobmark = so->so_oobmark;
3148 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3149 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3150 xso->so_uid = so->so_cred->cr_uid;
3151 }
3152
3153
3154 /*
3155 * Socket accessor functions to provide external consumers with
3156 * a safe interface to socket state
3157 *
3158 */
3159
3160 void
3161 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3162 {
3163
3164 TAILQ_FOREACH(so, &so->so_comp, so_list)
3165 func(so, arg);
3166 }
3167
3168 struct sockbuf *
3169 so_sockbuf_rcv(struct socket *so)
3170 {
3171
3172 return (&so->so_rcv);
3173 }
3174
3175 struct sockbuf *
3176 so_sockbuf_snd(struct socket *so)
3177 {
3178
3179 return (&so->so_snd);
3180 }
3181
3182 int
3183 so_state_get(const struct socket *so)
3184 {
3185
3186 return (so->so_state);
3187 }
3188
3189 void
3190 so_state_set(struct socket *so, int val)
3191 {
3192
3193 so->so_state = val;
3194 }
3195
3196 int
3197 so_options_get(const struct socket *so)
3198 {
3199
3200 return (so->so_options);
3201 }
3202
3203 void
3204 so_options_set(struct socket *so, int val)
3205 {
3206
3207 so->so_options = val;
3208 }
3209
3210 int
3211 so_error_get(const struct socket *so)
3212 {
3213
3214 return (so->so_error);
3215 }
3216
3217 void
3218 so_error_set(struct socket *so, int val)
3219 {
3220
3221 so->so_error = val;
3222 }
3223
3224 int
3225 so_linger_get(const struct socket *so)
3226 {
3227
3228 return (so->so_linger);
3229 }
3230
3231 void
3232 so_linger_set(struct socket *so, int val)
3233 {
3234
3235 so->so_linger = val;
3236 }
3237
3238 struct protosw *
3239 so_protosw_get(const struct socket *so)
3240 {
3241
3242 return (so->so_proto);
3243 }
3244
3245 void
3246 so_protosw_set(struct socket *so, struct protosw *val)
3247 {
3248
3249 so->so_proto = val;
3250 }
3251
3252 void
3253 so_sorwakeup(struct socket *so)
3254 {
3255
3256 sorwakeup(so);
3257 }
3258
3259 void
3260 so_sowwakeup(struct socket *so)
3261 {
3262
3263 sowwakeup(so);
3264 }
3265
3266 void
3267 so_sorwakeup_locked(struct socket *so)
3268 {
3269
3270 sorwakeup_locked(so);
3271 }
3272
3273 void
3274 so_sowwakeup_locked(struct socket *so)
3275 {
3276
3277 sowwakeup_locked(so);
3278 }
3279
3280 void
3281 so_lock(struct socket *so)
3282 {
3283 SOCK_LOCK(so);
3284 }
3285
3286 void
3287 so_unlock(struct socket *so)
3288 {
3289 SOCK_UNLOCK(so);
3290 }
Cache object: b3513973a9256269de7f14fe527ae1fe
|