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_zero.h"
103 #include "opt_compat.h"
104
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/fcntl.h>
108 #include <sys/limits.h>
109 #include <sys/lock.h>
110 #include <sys/mac.h>
111 #include <sys/malloc.h>
112 #include <sys/mbuf.h>
113 #include <sys/mutex.h>
114 #include <sys/domain.h>
115 #include <sys/file.h> /* for struct knote */
116 #include <sys/kernel.h>
117 #include <sys/event.h>
118 #include <sys/eventhandler.h>
119 #include <sys/poll.h>
120 #include <sys/proc.h>
121 #include <sys/protosw.h>
122 #include <sys/socket.h>
123 #include <sys/socketvar.h>
124 #include <sys/resourcevar.h>
125 #include <net/route.h>
126 #include <sys/signalvar.h>
127 #include <sys/stat.h>
128 #include <sys/sx.h>
129 #include <sys/sysctl.h>
130 #include <sys/uio.h>
131 #include <sys/jail.h>
132
133 #include <net/vnet.h>
134
135 #include <security/mac/mac_framework.h>
136
137 #include <vm/uma.h>
138
139 #ifdef COMPAT_IA32
140 #include <sys/mount.h>
141 #include <sys/sysent.h>
142 #include <compat/freebsd32/freebsd32.h>
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 .f_isfd = 1,
156 .f_detach = filt_sordetach,
157 .f_event = filt_solisten,
158 };
159 static struct filterops soread_filtops = {
160 .f_isfd = 1,
161 .f_detach = filt_sordetach,
162 .f_event = filt_soread,
163 };
164 static struct filterops sowrite_filtops = {
165 .f_isfd = 1,
166 .f_detach = filt_sowdetach,
167 .f_event = filt_sowrite,
168 };
169
170 uma_zone_t socket_zone;
171 so_gen_t so_gencnt; /* generation count for sockets */
172
173 int maxsockets;
174
175 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
176 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
177
178 static int somaxconn = SOMAXCONN;
179 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
180 /* XXX: we dont have SYSCTL_USHORT */
181 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
182 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
183 "queue size");
184 static int numopensockets;
185 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
186 &numopensockets, 0, "Number of open sockets");
187 #ifdef ZERO_COPY_SOCKETS
188 /* These aren't static because they're used in other files. */
189 int so_zero_copy_send = 1;
190 int so_zero_copy_receive = 1;
191 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
192 "Zero copy controls");
193 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
194 &so_zero_copy_receive, 0, "Enable zero copy receive");
195 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
196 &so_zero_copy_send, 0, "Enable zero copy send");
197 #endif /* ZERO_COPY_SOCKETS */
198
199 /*
200 * accept_mtx locks down per-socket fields relating to accept queues. See
201 * socketvar.h for an annotation of the protected fields of struct socket.
202 */
203 struct mtx accept_mtx;
204 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
205
206 /*
207 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
208 * so_gencnt field.
209 */
210 static struct mtx so_global_mtx;
211 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
212
213 /*
214 * General IPC sysctl name space, used by sockets and a variety of other IPC
215 * types.
216 */
217 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
218
219 /*
220 * Sysctl to get and set the maximum global sockets limit. Notify protocols
221 * of the change so that they can update their dependent limits as required.
222 */
223 static int
224 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
225 {
226 int error, newmaxsockets;
227
228 newmaxsockets = maxsockets;
229 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
230 if (error == 0 && req->newptr) {
231 if (newmaxsockets > maxsockets) {
232 maxsockets = newmaxsockets;
233 if (maxsockets > ((maxfiles / 4) * 3)) {
234 maxfiles = (maxsockets * 5) / 4;
235 maxfilesperproc = (maxfiles * 9) / 10;
236 }
237 EVENTHANDLER_INVOKE(maxsockets_change);
238 } else
239 error = EINVAL;
240 }
241 return (error);
242 }
243
244 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
245 &maxsockets, 0, sysctl_maxsockets, "IU",
246 "Maximum number of sockets avaliable");
247
248 /*
249 * Initialise maxsockets. This SYSINIT must be run after
250 * tunable_mbinit().
251 */
252 static void
253 init_maxsockets(void *ignored)
254 {
255
256 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
257 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
258 }
259 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
260
261 /*
262 * Socket operation routines. These routines are called by the routines in
263 * sys_socket.c or from a system process, and implement the semantics of
264 * socket operations by switching out to the protocol specific routines.
265 */
266
267 /*
268 * Get a socket structure from our zone, and initialize it. Note that it
269 * would probably be better to allocate socket and PCB at the same time, but
270 * I'm not convinced that all the protocols can be easily modified to do
271 * this.
272 *
273 * soalloc() returns a socket with a ref count of 0.
274 */
275 static struct socket *
276 soalloc(struct vnet *vnet)
277 {
278 struct socket *so;
279
280 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
281 if (so == NULL)
282 return (NULL);
283 #ifdef MAC
284 if (mac_socket_init(so, M_NOWAIT) != 0) {
285 uma_zfree(socket_zone, so);
286 return (NULL);
287 }
288 #endif
289 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
290 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
291 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
292 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
293 TAILQ_INIT(&so->so_aiojobq);
294 mtx_lock(&so_global_mtx);
295 so->so_gencnt = ++so_gencnt;
296 ++numopensockets;
297 #ifdef VIMAGE
298 vnet->vnet_sockcnt++;
299 so->so_vnet = vnet;
300 #endif
301 mtx_unlock(&so_global_mtx);
302 return (so);
303 }
304
305 /*
306 * Free the storage associated with a socket at the socket layer, tear down
307 * locks, labels, etc. All protocol state is assumed already to have been
308 * torn down (and possibly never set up) by the caller.
309 */
310 static void
311 sodealloc(struct socket *so)
312 {
313
314 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
315 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
316
317 mtx_lock(&so_global_mtx);
318 so->so_gencnt = ++so_gencnt;
319 --numopensockets; /* Could be below, but faster here. */
320 #ifdef VIMAGE
321 so->so_vnet->vnet_sockcnt--;
322 #endif
323 mtx_unlock(&so_global_mtx);
324 if (so->so_rcv.sb_hiwat)
325 (void)chgsbsize(so->so_cred->cr_uidinfo,
326 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
327 if (so->so_snd.sb_hiwat)
328 (void)chgsbsize(so->so_cred->cr_uidinfo,
329 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
330 #ifdef INET
331 /* remove acccept filter if one is present. */
332 if (so->so_accf != NULL)
333 do_setopt_accept_filter(so, NULL);
334 #endif
335 #ifdef MAC
336 mac_socket_destroy(so);
337 #endif
338 crfree(so->so_cred);
339 sx_destroy(&so->so_snd.sb_sx);
340 sx_destroy(&so->so_rcv.sb_sx);
341 SOCKBUF_LOCK_DESTROY(&so->so_snd);
342 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
343 uma_zfree(socket_zone, so);
344 }
345
346 /*
347 * socreate returns a socket with a ref count of 1. The socket should be
348 * closed with soclose().
349 */
350 int
351 socreate(int dom, struct socket **aso, int type, int proto,
352 struct ucred *cred, struct thread *td)
353 {
354 struct protosw *prp;
355 struct socket *so;
356 int error;
357
358 if (proto)
359 prp = pffindproto(dom, proto, type);
360 else
361 prp = pffindtype(dom, type);
362
363 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
364 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
365 return (EPROTONOSUPPORT);
366
367 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
368 return (EPROTONOSUPPORT);
369
370 if (prp->pr_type != type)
371 return (EPROTOTYPE);
372 so = soalloc(CRED_TO_VNET(cred));
373 if (so == NULL)
374 return (ENOBUFS);
375
376 TAILQ_INIT(&so->so_incomp);
377 TAILQ_INIT(&so->so_comp);
378 so->so_type = type;
379 so->so_cred = crhold(cred);
380 if ((prp->pr_domain->dom_family == PF_INET) ||
381 (prp->pr_domain->dom_family == PF_ROUTE))
382 so->so_fibnum = td->td_proc->p_fibnum;
383 else
384 so->so_fibnum = 0;
385 so->so_proto = prp;
386 #ifdef MAC
387 mac_socket_create(cred, so);
388 #endif
389 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
390 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
391 so->so_count = 1;
392 /*
393 * Auto-sizing of socket buffers is managed by the protocols and
394 * the appropriate flags must be set in the pru_attach function.
395 */
396 CURVNET_SET(so->so_vnet);
397 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
398 CURVNET_RESTORE();
399 if (error) {
400 KASSERT(so->so_count == 1, ("socreate: so_count %d",
401 so->so_count));
402 so->so_count = 0;
403 sodealloc(so);
404 return (error);
405 }
406 *aso = so;
407 return (0);
408 }
409
410 #ifdef REGRESSION
411 static int regression_sonewconn_earlytest = 1;
412 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
413 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
414 #endif
415
416 /*
417 * When an attempt at a new connection is noted on a socket which accepts
418 * connections, sonewconn is called. If the connection is possible (subject
419 * to space constraints, etc.) then we allocate a new structure, propoerly
420 * linked into the data structure of the original socket, and return this.
421 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
422 *
423 * Note: the ref count on the socket is 0 on return.
424 */
425 struct socket *
426 sonewconn(struct socket *head, int connstatus)
427 {
428 struct socket *so;
429 int over;
430
431 ACCEPT_LOCK();
432 over = (head->so_qlen > 3 * head->so_qlimit / 2);
433 ACCEPT_UNLOCK();
434 #ifdef REGRESSION
435 if (regression_sonewconn_earlytest && over)
436 #else
437 if (over)
438 #endif
439 return (NULL);
440 VNET_ASSERT(head->so_vnet);
441 so = soalloc(head->so_vnet);
442 if (so == NULL)
443 return (NULL);
444 if ((head->so_options & SO_ACCEPTFILTER) != 0)
445 connstatus = 0;
446 so->so_head = head;
447 so->so_type = head->so_type;
448 so->so_options = head->so_options &~ SO_ACCEPTCONN;
449 so->so_linger = head->so_linger;
450 so->so_state = head->so_state | SS_NOFDREF;
451 so->so_fibnum = head->so_fibnum;
452 so->so_proto = head->so_proto;
453 so->so_cred = crhold(head->so_cred);
454 #ifdef MAC
455 mac_socket_newconn(head, so);
456 #endif
457 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
458 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
459 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
460 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
461 sodealloc(so);
462 return (NULL);
463 }
464 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
465 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
466 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
467 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
468 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
469 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
470 so->so_state |= connstatus;
471 ACCEPT_LOCK();
472 if (connstatus) {
473 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
474 so->so_qstate |= SQ_COMP;
475 head->so_qlen++;
476 } else {
477 /*
478 * Keep removing sockets from the head until there's room for
479 * us to insert on the tail. In pre-locking revisions, this
480 * was a simple if(), but as we could be racing with other
481 * threads and soabort() requires dropping locks, we must
482 * loop waiting for the condition to be true.
483 */
484 while (head->so_incqlen > head->so_qlimit) {
485 struct socket *sp;
486 sp = TAILQ_FIRST(&head->so_incomp);
487 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
488 head->so_incqlen--;
489 sp->so_qstate &= ~SQ_INCOMP;
490 sp->so_head = NULL;
491 ACCEPT_UNLOCK();
492 soabort(sp);
493 ACCEPT_LOCK();
494 }
495 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
496 so->so_qstate |= SQ_INCOMP;
497 head->so_incqlen++;
498 }
499 ACCEPT_UNLOCK();
500 if (connstatus) {
501 sorwakeup(head);
502 wakeup_one(&head->so_timeo);
503 }
504 return (so);
505 }
506
507 int
508 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
509 {
510 int error;
511
512 CURVNET_SET(so->so_vnet);
513 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
514 CURVNET_RESTORE();
515 return error;
516 }
517
518 /*
519 * solisten() transitions a socket from a non-listening state to a listening
520 * state, but can also be used to update the listen queue depth on an
521 * existing listen socket. The protocol will call back into the sockets
522 * layer using solisten_proto_check() and solisten_proto() to check and set
523 * socket-layer listen state. Call backs are used so that the protocol can
524 * acquire both protocol and socket layer locks in whatever order is required
525 * by the protocol.
526 *
527 * Protocol implementors are advised to hold the socket lock across the
528 * socket-layer test and set to avoid races at the socket layer.
529 */
530 int
531 solisten(struct socket *so, int backlog, struct thread *td)
532 {
533
534 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
535 }
536
537 int
538 solisten_proto_check(struct socket *so)
539 {
540
541 SOCK_LOCK_ASSERT(so);
542
543 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
544 SS_ISDISCONNECTING))
545 return (EINVAL);
546 return (0);
547 }
548
549 void
550 solisten_proto(struct socket *so, int backlog)
551 {
552
553 SOCK_LOCK_ASSERT(so);
554
555 if (backlog < 0 || backlog > somaxconn)
556 backlog = somaxconn;
557 so->so_qlimit = backlog;
558 so->so_options |= SO_ACCEPTCONN;
559 }
560
561 /*
562 * Attempt to free a socket. This should really be sotryfree().
563 *
564 * sofree() will succeed if:
565 *
566 * - There are no outstanding file descriptor references or related consumers
567 * (so_count == 0).
568 *
569 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
570 *
571 * - The protocol does not have an outstanding strong reference on the socket
572 * (SS_PROTOREF).
573 *
574 * - The socket is not in a completed connection queue, so a process has been
575 * notified that it is present. If it is removed, the user process may
576 * block in accept() despite select() saying the socket was ready.
577 *
578 * Otherwise, it will quietly abort so that a future call to sofree(), when
579 * conditions are right, can succeed.
580 */
581 void
582 sofree(struct socket *so)
583 {
584 struct protosw *pr = so->so_proto;
585 struct socket *head;
586
587 ACCEPT_LOCK_ASSERT();
588 SOCK_LOCK_ASSERT(so);
589
590 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
591 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
592 SOCK_UNLOCK(so);
593 ACCEPT_UNLOCK();
594 return;
595 }
596
597 head = so->so_head;
598 if (head != NULL) {
599 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
600 (so->so_qstate & SQ_INCOMP) != 0,
601 ("sofree: so_head != NULL, but neither SQ_COMP nor "
602 "SQ_INCOMP"));
603 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
604 (so->so_qstate & SQ_INCOMP) == 0,
605 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
606 TAILQ_REMOVE(&head->so_incomp, so, so_list);
607 head->so_incqlen--;
608 so->so_qstate &= ~SQ_INCOMP;
609 so->so_head = NULL;
610 }
611 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
612 (so->so_qstate & SQ_INCOMP) == 0,
613 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
614 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
615 if (so->so_options & SO_ACCEPTCONN) {
616 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
617 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
618 }
619 SOCK_UNLOCK(so);
620 ACCEPT_UNLOCK();
621
622 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
623 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
624 if (pr->pr_usrreqs->pru_detach != NULL)
625 (*pr->pr_usrreqs->pru_detach)(so);
626
627 /*
628 * From this point on, we assume that no other references to this
629 * socket exist anywhere else in the stack. Therefore, no locks need
630 * to be acquired or held.
631 *
632 * We used to do a lot of socket buffer and socket locking here, as
633 * well as invoke sorflush() and perform wakeups. The direct call to
634 * dom_dispose() and sbrelease_internal() are an inlining of what was
635 * necessary from sorflush().
636 *
637 * Notice that the socket buffer and kqueue state are torn down
638 * before calling pru_detach. This means that protocols shold not
639 * assume they can perform socket wakeups, etc, in their detach code.
640 */
641 sbdestroy(&so->so_snd, so);
642 sbdestroy(&so->so_rcv, so);
643 knlist_destroy(&so->so_rcv.sb_sel.si_note);
644 knlist_destroy(&so->so_snd.sb_sel.si_note);
645 sodealloc(so);
646 }
647
648 /*
649 * Close a socket on last file table reference removal. Initiate disconnect
650 * if connected. Free socket when disconnect complete.
651 *
652 * This function will sorele() the socket. Note that soclose() may be called
653 * prior to the ref count reaching zero. The actual socket structure will
654 * not be freed until the ref count reaches zero.
655 */
656 int
657 soclose(struct socket *so)
658 {
659 int error = 0;
660
661 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
662
663 CURVNET_SET(so->so_vnet);
664 funsetown(&so->so_sigio);
665 if (so->so_state & SS_ISCONNECTED) {
666 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
667 error = sodisconnect(so);
668 if (error)
669 goto drop;
670 }
671 if (so->so_options & SO_LINGER) {
672 if ((so->so_state & SS_ISDISCONNECTING) &&
673 (so->so_state & SS_NBIO))
674 goto drop;
675 while (so->so_state & SS_ISCONNECTED) {
676 error = tsleep(&so->so_timeo,
677 PSOCK | PCATCH, "soclos", so->so_linger * hz);
678 if (error)
679 break;
680 }
681 }
682 }
683
684 drop:
685 if (so->so_proto->pr_usrreqs->pru_close != NULL)
686 (*so->so_proto->pr_usrreqs->pru_close)(so);
687 if (so->so_options & SO_ACCEPTCONN) {
688 struct socket *sp;
689 ACCEPT_LOCK();
690 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
691 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
692 so->so_incqlen--;
693 sp->so_qstate &= ~SQ_INCOMP;
694 sp->so_head = NULL;
695 ACCEPT_UNLOCK();
696 soabort(sp);
697 ACCEPT_LOCK();
698 }
699 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
700 TAILQ_REMOVE(&so->so_comp, sp, so_list);
701 so->so_qlen--;
702 sp->so_qstate &= ~SQ_COMP;
703 sp->so_head = NULL;
704 ACCEPT_UNLOCK();
705 soabort(sp);
706 ACCEPT_LOCK();
707 }
708 ACCEPT_UNLOCK();
709 }
710 ACCEPT_LOCK();
711 SOCK_LOCK(so);
712 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
713 so->so_state |= SS_NOFDREF;
714 sorele(so);
715 CURVNET_RESTORE();
716 return (error);
717 }
718
719 /*
720 * soabort() is used to abruptly tear down a connection, such as when a
721 * resource limit is reached (listen queue depth exceeded), or if a listen
722 * socket is closed while there are sockets waiting to be accepted.
723 *
724 * This interface is tricky, because it is called on an unreferenced socket,
725 * and must be called only by a thread that has actually removed the socket
726 * from the listen queue it was on, or races with other threads are risked.
727 *
728 * This interface will call into the protocol code, so must not be called
729 * with any socket locks held. Protocols do call it while holding their own
730 * recursible protocol mutexes, but this is something that should be subject
731 * to review in the future.
732 */
733 void
734 soabort(struct socket *so)
735 {
736
737 /*
738 * In as much as is possible, assert that no references to this
739 * socket are held. This is not quite the same as asserting that the
740 * current thread is responsible for arranging for no references, but
741 * is as close as we can get for now.
742 */
743 KASSERT(so->so_count == 0, ("soabort: so_count"));
744 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
745 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
746 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
747 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
748
749 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
750 (*so->so_proto->pr_usrreqs->pru_abort)(so);
751 ACCEPT_LOCK();
752 SOCK_LOCK(so);
753 sofree(so);
754 }
755
756 int
757 soaccept(struct socket *so, struct sockaddr **nam)
758 {
759 int error;
760
761 SOCK_LOCK(so);
762 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
763 so->so_state &= ~SS_NOFDREF;
764 SOCK_UNLOCK(so);
765 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
766 return (error);
767 }
768
769 int
770 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
771 {
772 int error;
773
774 if (so->so_options & SO_ACCEPTCONN)
775 return (EOPNOTSUPP);
776 /*
777 * If protocol is connection-based, can only connect once.
778 * Otherwise, if connected, try to disconnect first. This allows
779 * user to disconnect by connecting to, e.g., a null address.
780 */
781 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
782 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
783 (error = sodisconnect(so)))) {
784 error = EISCONN;
785 } else {
786 /*
787 * Prevent accumulated error from previous connection from
788 * biting us.
789 */
790 so->so_error = 0;
791 CURVNET_SET(so->so_vnet);
792 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
793 CURVNET_RESTORE();
794 }
795
796 return (error);
797 }
798
799 int
800 soconnect2(struct socket *so1, struct socket *so2)
801 {
802
803 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
804 }
805
806 int
807 sodisconnect(struct socket *so)
808 {
809 int error;
810
811 if ((so->so_state & SS_ISCONNECTED) == 0)
812 return (ENOTCONN);
813 if (so->so_state & SS_ISDISCONNECTING)
814 return (EALREADY);
815 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
816 return (error);
817 }
818
819 #ifdef ZERO_COPY_SOCKETS
820 struct so_zerocopy_stats{
821 int size_ok;
822 int align_ok;
823 int found_ifp;
824 };
825 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
826 #include <netinet/in.h>
827 #include <net/route.h>
828 #include <netinet/in_pcb.h>
829 #include <vm/vm.h>
830 #include <vm/vm_page.h>
831 #include <vm/vm_object.h>
832
833 /*
834 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
835 * sosend_dgram() and sosend_generic() use m_uiotombuf().
836 *
837 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
838 * all of the data referenced by the uio. If desired, it uses zero-copy.
839 * *space will be updated to reflect data copied in.
840 *
841 * NB: If atomic I/O is requested, the caller must already have checked that
842 * space can hold resid bytes.
843 *
844 * NB: In the event of an error, the caller may need to free the partial
845 * chain pointed to by *mpp. The contents of both *uio and *space may be
846 * modified even in the case of an error.
847 */
848 static int
849 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
850 int flags)
851 {
852 struct mbuf *m, **mp, *top;
853 long len, resid;
854 int error;
855 #ifdef ZERO_COPY_SOCKETS
856 int cow_send;
857 #endif
858
859 *retmp = top = NULL;
860 mp = ⊤
861 len = 0;
862 resid = uio->uio_resid;
863 error = 0;
864 do {
865 #ifdef ZERO_COPY_SOCKETS
866 cow_send = 0;
867 #endif /* ZERO_COPY_SOCKETS */
868 if (resid >= MINCLSIZE) {
869 #ifdef ZERO_COPY_SOCKETS
870 if (top == NULL) {
871 m = m_gethdr(M_WAITOK, MT_DATA);
872 m->m_pkthdr.len = 0;
873 m->m_pkthdr.rcvif = NULL;
874 } else
875 m = m_get(M_WAITOK, MT_DATA);
876 if (so_zero_copy_send &&
877 resid>=PAGE_SIZE &&
878 *space>=PAGE_SIZE &&
879 uio->uio_iov->iov_len>=PAGE_SIZE) {
880 so_zerocp_stats.size_ok++;
881 so_zerocp_stats.align_ok++;
882 cow_send = socow_setup(m, uio);
883 len = cow_send;
884 }
885 if (!cow_send) {
886 m_clget(m, M_WAITOK);
887 len = min(min(MCLBYTES, resid), *space);
888 }
889 #else /* ZERO_COPY_SOCKETS */
890 if (top == NULL) {
891 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
892 m->m_pkthdr.len = 0;
893 m->m_pkthdr.rcvif = NULL;
894 } else
895 m = m_getcl(M_WAIT, MT_DATA, 0);
896 len = min(min(MCLBYTES, resid), *space);
897 #endif /* ZERO_COPY_SOCKETS */
898 } else {
899 if (top == NULL) {
900 m = m_gethdr(M_WAIT, MT_DATA);
901 m->m_pkthdr.len = 0;
902 m->m_pkthdr.rcvif = NULL;
903
904 len = min(min(MHLEN, resid), *space);
905 /*
906 * For datagram protocols, leave room
907 * for protocol headers in first mbuf.
908 */
909 if (atomic && m && len < MHLEN)
910 MH_ALIGN(m, len);
911 } else {
912 m = m_get(M_WAIT, MT_DATA);
913 len = min(min(MLEN, resid), *space);
914 }
915 }
916 if (m == NULL) {
917 error = ENOBUFS;
918 goto out;
919 }
920
921 *space -= len;
922 #ifdef ZERO_COPY_SOCKETS
923 if (cow_send)
924 error = 0;
925 else
926 #endif /* ZERO_COPY_SOCKETS */
927 error = uiomove(mtod(m, void *), (int)len, uio);
928 resid = uio->uio_resid;
929 m->m_len = len;
930 *mp = m;
931 top->m_pkthdr.len += len;
932 if (error)
933 goto out;
934 mp = &m->m_next;
935 if (resid <= 0) {
936 if (flags & MSG_EOR)
937 top->m_flags |= M_EOR;
938 break;
939 }
940 } while (*space > 0 && atomic);
941 out:
942 *retmp = top;
943 return (error);
944 }
945 #endif /*ZERO_COPY_SOCKETS*/
946
947 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
948
949 int
950 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
951 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
952 {
953 long space, resid;
954 int clen = 0, error, dontroute;
955 #ifdef ZERO_COPY_SOCKETS
956 int atomic = sosendallatonce(so) || top;
957 #endif
958
959 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
960 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
961 ("sodgram_send: !PR_ATOMIC"));
962
963 if (uio != NULL)
964 resid = uio->uio_resid;
965 else
966 resid = top->m_pkthdr.len;
967 /*
968 * In theory resid should be unsigned. However, space must be
969 * signed, as it might be less than 0 if we over-committed, and we
970 * must use a signed comparison of space and resid. On the other
971 * hand, a negative resid causes us to loop sending 0-length
972 * segments to the protocol.
973 */
974 if (resid < 0) {
975 error = EINVAL;
976 goto out;
977 }
978
979 dontroute =
980 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
981 if (td != NULL)
982 td->td_ru.ru_msgsnd++;
983 if (control != NULL)
984 clen = control->m_len;
985
986 SOCKBUF_LOCK(&so->so_snd);
987 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
988 SOCKBUF_UNLOCK(&so->so_snd);
989 error = EPIPE;
990 goto out;
991 }
992 if (so->so_error) {
993 error = so->so_error;
994 so->so_error = 0;
995 SOCKBUF_UNLOCK(&so->so_snd);
996 goto out;
997 }
998 if ((so->so_state & SS_ISCONNECTED) == 0) {
999 /*
1000 * `sendto' and `sendmsg' is allowed on a connection-based
1001 * socket if it supports implied connect. Return ENOTCONN if
1002 * not connected and no address is supplied.
1003 */
1004 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1005 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1006 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1007 !(resid == 0 && clen != 0)) {
1008 SOCKBUF_UNLOCK(&so->so_snd);
1009 error = ENOTCONN;
1010 goto out;
1011 }
1012 } else if (addr == NULL) {
1013 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1014 error = ENOTCONN;
1015 else
1016 error = EDESTADDRREQ;
1017 SOCKBUF_UNLOCK(&so->so_snd);
1018 goto out;
1019 }
1020 }
1021
1022 /*
1023 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1024 * problem and need fixing.
1025 */
1026 space = sbspace(&so->so_snd);
1027 if (flags & MSG_OOB)
1028 space += 1024;
1029 space -= clen;
1030 SOCKBUF_UNLOCK(&so->so_snd);
1031 if (resid > space) {
1032 error = EMSGSIZE;
1033 goto out;
1034 }
1035 if (uio == NULL) {
1036 resid = 0;
1037 if (flags & MSG_EOR)
1038 top->m_flags |= M_EOR;
1039 } else {
1040 #ifdef ZERO_COPY_SOCKETS
1041 error = sosend_copyin(uio, &top, atomic, &space, flags);
1042 if (error)
1043 goto out;
1044 #else
1045 /*
1046 * Copy the data from userland into a mbuf chain.
1047 * If no data is to be copied in, a single empty mbuf
1048 * is returned.
1049 */
1050 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1051 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1052 if (top == NULL) {
1053 error = EFAULT; /* only possible error */
1054 goto out;
1055 }
1056 space -= resid - uio->uio_resid;
1057 #endif
1058 resid = uio->uio_resid;
1059 }
1060 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1061 /*
1062 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1063 * than with.
1064 */
1065 if (dontroute) {
1066 SOCK_LOCK(so);
1067 so->so_options |= SO_DONTROUTE;
1068 SOCK_UNLOCK(so);
1069 }
1070 /*
1071 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1072 * of date. We could have recieved a reset packet in an interrupt or
1073 * maybe we slept while doing page faults in uiomove() etc. We could
1074 * probably recheck again inside the locking protection here, but
1075 * there are probably other places that this also happens. We must
1076 * rethink this.
1077 */
1078 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1079 (flags & MSG_OOB) ? PRUS_OOB :
1080 /*
1081 * If the user set MSG_EOF, the protocol understands this flag and
1082 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1083 */
1084 ((flags & MSG_EOF) &&
1085 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1086 (resid <= 0)) ?
1087 PRUS_EOF :
1088 /* If there is more to send set PRUS_MORETOCOME */
1089 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1090 top, addr, control, td);
1091 if (dontroute) {
1092 SOCK_LOCK(so);
1093 so->so_options &= ~SO_DONTROUTE;
1094 SOCK_UNLOCK(so);
1095 }
1096 clen = 0;
1097 control = NULL;
1098 top = NULL;
1099 out:
1100 if (top != NULL)
1101 m_freem(top);
1102 if (control != NULL)
1103 m_freem(control);
1104 return (error);
1105 }
1106
1107 /*
1108 * Send on a socket. If send must go all at once and message is larger than
1109 * send buffering, then hard error. Lock against other senders. If must go
1110 * all at once and not enough room now, then inform user that this would
1111 * block and do nothing. Otherwise, if nonblocking, send as much as
1112 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1113 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1114 * in mbuf chain must be small enough to send all at once.
1115 *
1116 * Returns nonzero on error, timeout or signal; callers must check for short
1117 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1118 * on return.
1119 */
1120 int
1121 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1122 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1123 {
1124 long space, resid;
1125 int clen = 0, error, dontroute;
1126 int atomic = sosendallatonce(so) || top;
1127
1128 if (uio != NULL)
1129 resid = uio->uio_resid;
1130 else
1131 resid = top->m_pkthdr.len;
1132 /*
1133 * In theory resid should be unsigned. However, space must be
1134 * signed, as it might be less than 0 if we over-committed, and we
1135 * must use a signed comparison of space and resid. On the other
1136 * hand, a negative resid causes us to loop sending 0-length
1137 * segments to the protocol.
1138 *
1139 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1140 * type sockets since that's an error.
1141 */
1142 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1143 error = EINVAL;
1144 goto out;
1145 }
1146
1147 dontroute =
1148 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1149 (so->so_proto->pr_flags & PR_ATOMIC);
1150 if (td != NULL)
1151 td->td_ru.ru_msgsnd++;
1152 if (control != NULL)
1153 clen = control->m_len;
1154
1155 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1156 if (error)
1157 goto out;
1158
1159 restart:
1160 do {
1161 SOCKBUF_LOCK(&so->so_snd);
1162 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1163 SOCKBUF_UNLOCK(&so->so_snd);
1164 error = EPIPE;
1165 goto release;
1166 }
1167 if (so->so_error) {
1168 error = so->so_error;
1169 so->so_error = 0;
1170 SOCKBUF_UNLOCK(&so->so_snd);
1171 goto release;
1172 }
1173 if ((so->so_state & SS_ISCONNECTED) == 0) {
1174 /*
1175 * `sendto' and `sendmsg' is allowed on a connection-
1176 * based socket if it supports implied connect.
1177 * Return ENOTCONN if not connected and no address is
1178 * supplied.
1179 */
1180 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1181 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1182 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1183 !(resid == 0 && clen != 0)) {
1184 SOCKBUF_UNLOCK(&so->so_snd);
1185 error = ENOTCONN;
1186 goto release;
1187 }
1188 } else if (addr == NULL) {
1189 SOCKBUF_UNLOCK(&so->so_snd);
1190 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1191 error = ENOTCONN;
1192 else
1193 error = EDESTADDRREQ;
1194 goto release;
1195 }
1196 }
1197 space = sbspace(&so->so_snd);
1198 if (flags & MSG_OOB)
1199 space += 1024;
1200 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1201 clen > so->so_snd.sb_hiwat) {
1202 SOCKBUF_UNLOCK(&so->so_snd);
1203 error = EMSGSIZE;
1204 goto release;
1205 }
1206 if (space < resid + clen &&
1207 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1208 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1209 SOCKBUF_UNLOCK(&so->so_snd);
1210 error = EWOULDBLOCK;
1211 goto release;
1212 }
1213 error = sbwait(&so->so_snd);
1214 SOCKBUF_UNLOCK(&so->so_snd);
1215 if (error)
1216 goto release;
1217 goto restart;
1218 }
1219 SOCKBUF_UNLOCK(&so->so_snd);
1220 space -= clen;
1221 do {
1222 if (uio == NULL) {
1223 resid = 0;
1224 if (flags & MSG_EOR)
1225 top->m_flags |= M_EOR;
1226 } else {
1227 #ifdef ZERO_COPY_SOCKETS
1228 error = sosend_copyin(uio, &top, atomic,
1229 &space, flags);
1230 if (error != 0)
1231 goto release;
1232 #else
1233 /*
1234 * Copy the data from userland into a mbuf
1235 * chain. If no data is to be copied in,
1236 * a single empty mbuf is returned.
1237 */
1238 top = m_uiotombuf(uio, M_WAITOK, space,
1239 (atomic ? max_hdr : 0),
1240 (atomic ? M_PKTHDR : 0) |
1241 ((flags & MSG_EOR) ? M_EOR : 0));
1242 if (top == NULL) {
1243 error = EFAULT; /* only possible error */
1244 goto release;
1245 }
1246 space -= resid - uio->uio_resid;
1247 #endif
1248 resid = uio->uio_resid;
1249 }
1250 if (dontroute) {
1251 SOCK_LOCK(so);
1252 so->so_options |= SO_DONTROUTE;
1253 SOCK_UNLOCK(so);
1254 }
1255 /*
1256 * XXX all the SBS_CANTSENDMORE checks previously
1257 * done could be out of date. We could have recieved
1258 * a reset packet in an interrupt or maybe we slept
1259 * while doing page faults in uiomove() etc. We
1260 * could probably recheck again inside the locking
1261 * protection here, but there are probably other
1262 * places that this also happens. We must rethink
1263 * this.
1264 */
1265 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1266 (flags & MSG_OOB) ? PRUS_OOB :
1267 /*
1268 * If the user set MSG_EOF, the protocol understands
1269 * this flag and nothing left to send then use
1270 * PRU_SEND_EOF instead of PRU_SEND.
1271 */
1272 ((flags & MSG_EOF) &&
1273 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1274 (resid <= 0)) ?
1275 PRUS_EOF :
1276 /* If there is more to send set PRUS_MORETOCOME. */
1277 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1278 top, addr, control, td);
1279 if (dontroute) {
1280 SOCK_LOCK(so);
1281 so->so_options &= ~SO_DONTROUTE;
1282 SOCK_UNLOCK(so);
1283 }
1284 clen = 0;
1285 control = NULL;
1286 top = NULL;
1287 if (error)
1288 goto release;
1289 } while (resid && space > 0);
1290 } while (resid);
1291
1292 release:
1293 sbunlock(&so->so_snd);
1294 out:
1295 if (top != NULL)
1296 m_freem(top);
1297 if (control != NULL)
1298 m_freem(control);
1299 return (error);
1300 }
1301
1302 int
1303 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1304 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1305 {
1306 int error;
1307
1308 CURVNET_SET(so->so_vnet);
1309 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1310 control, flags, td);
1311 CURVNET_RESTORE();
1312 return (error);
1313 }
1314
1315 /*
1316 * The part of soreceive() that implements reading non-inline out-of-band
1317 * data from a socket. For more complete comments, see soreceive(), from
1318 * which this code originated.
1319 *
1320 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1321 * unable to return an mbuf chain to the caller.
1322 */
1323 static int
1324 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1325 {
1326 struct protosw *pr = so->so_proto;
1327 struct mbuf *m;
1328 int error;
1329
1330 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1331
1332 m = m_get(M_WAIT, MT_DATA);
1333 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1334 if (error)
1335 goto bad;
1336 do {
1337 #ifdef ZERO_COPY_SOCKETS
1338 if (so_zero_copy_receive) {
1339 int disposable;
1340
1341 if ((m->m_flags & M_EXT)
1342 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1343 disposable = 1;
1344 else
1345 disposable = 0;
1346
1347 error = uiomoveco(mtod(m, void *),
1348 min(uio->uio_resid, m->m_len),
1349 uio, disposable);
1350 } else
1351 #endif /* ZERO_COPY_SOCKETS */
1352 error = uiomove(mtod(m, void *),
1353 (int) min(uio->uio_resid, m->m_len), uio);
1354 m = m_free(m);
1355 } while (uio->uio_resid && error == 0 && m);
1356 bad:
1357 if (m != NULL)
1358 m_freem(m);
1359 return (error);
1360 }
1361
1362 /*
1363 * Following replacement or removal of the first mbuf on the first mbuf chain
1364 * of a socket buffer, push necessary state changes back into the socket
1365 * buffer so that other consumers see the values consistently. 'nextrecord'
1366 * is the callers locally stored value of the original value of
1367 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1368 * NOTE: 'nextrecord' may be NULL.
1369 */
1370 static __inline void
1371 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1372 {
1373
1374 SOCKBUF_LOCK_ASSERT(sb);
1375 /*
1376 * First, update for the new value of nextrecord. If necessary, make
1377 * it the first record.
1378 */
1379 if (sb->sb_mb != NULL)
1380 sb->sb_mb->m_nextpkt = nextrecord;
1381 else
1382 sb->sb_mb = nextrecord;
1383
1384 /*
1385 * Now update any dependent socket buffer fields to reflect the new
1386 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1387 * addition of a second clause that takes care of the case where
1388 * sb_mb has been updated, but remains the last record.
1389 */
1390 if (sb->sb_mb == NULL) {
1391 sb->sb_mbtail = NULL;
1392 sb->sb_lastrecord = NULL;
1393 } else if (sb->sb_mb->m_nextpkt == NULL)
1394 sb->sb_lastrecord = sb->sb_mb;
1395 }
1396
1397
1398 /*
1399 * Implement receive operations on a socket. We depend on the way that
1400 * records are added to the sockbuf by sbappend. In particular, each record
1401 * (mbufs linked through m_next) must begin with an address if the protocol
1402 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1403 * data, and then zero or more mbufs of data. In order to allow parallelism
1404 * between network receive and copying to user space, as well as avoid
1405 * sleeping with a mutex held, we release the socket buffer mutex during the
1406 * user space copy. Although the sockbuf is locked, new data may still be
1407 * appended, and thus we must maintain consistency of the sockbuf during that
1408 * time.
1409 *
1410 * The caller may receive the data as a single mbuf chain by supplying an
1411 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1412 * the count in uio_resid.
1413 */
1414 int
1415 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1416 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1417 {
1418 struct mbuf *m, **mp;
1419 int flags, len, error, offset;
1420 struct protosw *pr = so->so_proto;
1421 struct mbuf *nextrecord;
1422 int moff, type = 0;
1423 int orig_resid = uio->uio_resid;
1424
1425 mp = mp0;
1426 if (psa != NULL)
1427 *psa = NULL;
1428 if (controlp != NULL)
1429 *controlp = NULL;
1430 if (flagsp != NULL)
1431 flags = *flagsp &~ MSG_EOR;
1432 else
1433 flags = 0;
1434 if (flags & MSG_OOB)
1435 return (soreceive_rcvoob(so, uio, flags));
1436 if (mp != NULL)
1437 *mp = NULL;
1438 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1439 && uio->uio_resid)
1440 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1441
1442 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1443 if (error)
1444 return (error);
1445
1446 restart:
1447 SOCKBUF_LOCK(&so->so_rcv);
1448 m = so->so_rcv.sb_mb;
1449 /*
1450 * If we have less data than requested, block awaiting more (subject
1451 * to any timeout) if:
1452 * 1. the current count is less than the low water mark, or
1453 * 2. MSG_WAITALL is set, and it is possible to do the entire
1454 * receive operation at once if we block (resid <= hiwat).
1455 * 3. MSG_DONTWAIT is not set
1456 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1457 * we have to do the receive in sections, and thus risk returning a
1458 * short count if a timeout or signal occurs after we start.
1459 */
1460 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1461 so->so_rcv.sb_cc < uio->uio_resid) &&
1462 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1463 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1464 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1465 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1466 ("receive: m == %p so->so_rcv.sb_cc == %u",
1467 m, so->so_rcv.sb_cc));
1468 if (so->so_error) {
1469 if (m != NULL)
1470 goto dontblock;
1471 error = so->so_error;
1472 if ((flags & MSG_PEEK) == 0)
1473 so->so_error = 0;
1474 SOCKBUF_UNLOCK(&so->so_rcv);
1475 goto release;
1476 }
1477 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1478 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1479 if (m == NULL) {
1480 SOCKBUF_UNLOCK(&so->so_rcv);
1481 goto release;
1482 } else
1483 goto dontblock;
1484 }
1485 for (; m != NULL; m = m->m_next)
1486 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1487 m = so->so_rcv.sb_mb;
1488 goto dontblock;
1489 }
1490 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1491 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1492 SOCKBUF_UNLOCK(&so->so_rcv);
1493 error = ENOTCONN;
1494 goto release;
1495 }
1496 if (uio->uio_resid == 0) {
1497 SOCKBUF_UNLOCK(&so->so_rcv);
1498 goto release;
1499 }
1500 if ((so->so_state & SS_NBIO) ||
1501 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1502 SOCKBUF_UNLOCK(&so->so_rcv);
1503 error = EWOULDBLOCK;
1504 goto release;
1505 }
1506 SBLASTRECORDCHK(&so->so_rcv);
1507 SBLASTMBUFCHK(&so->so_rcv);
1508 error = sbwait(&so->so_rcv);
1509 SOCKBUF_UNLOCK(&so->so_rcv);
1510 if (error)
1511 goto release;
1512 goto restart;
1513 }
1514 dontblock:
1515 /*
1516 * From this point onward, we maintain 'nextrecord' as a cache of the
1517 * pointer to the next record in the socket buffer. We must keep the
1518 * various socket buffer pointers and local stack versions of the
1519 * pointers in sync, pushing out modifications before dropping the
1520 * socket buffer mutex, and re-reading them when picking it up.
1521 *
1522 * Otherwise, we will race with the network stack appending new data
1523 * or records onto the socket buffer by using inconsistent/stale
1524 * versions of the field, possibly resulting in socket buffer
1525 * corruption.
1526 *
1527 * By holding the high-level sblock(), we prevent simultaneous
1528 * readers from pulling off the front of the socket buffer.
1529 */
1530 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1531 if (uio->uio_td)
1532 uio->uio_td->td_ru.ru_msgrcv++;
1533 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1534 SBLASTRECORDCHK(&so->so_rcv);
1535 SBLASTMBUFCHK(&so->so_rcv);
1536 nextrecord = m->m_nextpkt;
1537 if (pr->pr_flags & PR_ADDR) {
1538 KASSERT(m->m_type == MT_SONAME,
1539 ("m->m_type == %d", m->m_type));
1540 orig_resid = 0;
1541 if (psa != NULL)
1542 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1543 M_NOWAIT);
1544 if (flags & MSG_PEEK) {
1545 m = m->m_next;
1546 } else {
1547 sbfree(&so->so_rcv, m);
1548 so->so_rcv.sb_mb = m_free(m);
1549 m = so->so_rcv.sb_mb;
1550 sockbuf_pushsync(&so->so_rcv, nextrecord);
1551 }
1552 }
1553
1554 /*
1555 * Process one or more MT_CONTROL mbufs present before any data mbufs
1556 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1557 * just copy the data; if !MSG_PEEK, we call into the protocol to
1558 * perform externalization (or freeing if controlp == NULL).
1559 */
1560 if (m != NULL && m->m_type == MT_CONTROL) {
1561 struct mbuf *cm = NULL, *cmn;
1562 struct mbuf **cme = &cm;
1563
1564 do {
1565 if (flags & MSG_PEEK) {
1566 if (controlp != NULL) {
1567 *controlp = m_copy(m, 0, m->m_len);
1568 controlp = &(*controlp)->m_next;
1569 }
1570 m = m->m_next;
1571 } else {
1572 sbfree(&so->so_rcv, m);
1573 so->so_rcv.sb_mb = m->m_next;
1574 m->m_next = NULL;
1575 *cme = m;
1576 cme = &(*cme)->m_next;
1577 m = so->so_rcv.sb_mb;
1578 }
1579 } while (m != NULL && m->m_type == MT_CONTROL);
1580 if ((flags & MSG_PEEK) == 0)
1581 sockbuf_pushsync(&so->so_rcv, nextrecord);
1582 while (cm != NULL) {
1583 cmn = cm->m_next;
1584 cm->m_next = NULL;
1585 if (pr->pr_domain->dom_externalize != NULL) {
1586 SOCKBUF_UNLOCK(&so->so_rcv);
1587 error = (*pr->pr_domain->dom_externalize)
1588 (cm, controlp);
1589 SOCKBUF_LOCK(&so->so_rcv);
1590 } else if (controlp != NULL)
1591 *controlp = cm;
1592 else
1593 m_freem(cm);
1594 if (controlp != NULL) {
1595 orig_resid = 0;
1596 while (*controlp != NULL)
1597 controlp = &(*controlp)->m_next;
1598 }
1599 cm = cmn;
1600 }
1601 if (m != NULL)
1602 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1603 else
1604 nextrecord = so->so_rcv.sb_mb;
1605 orig_resid = 0;
1606 }
1607 if (m != NULL) {
1608 if ((flags & MSG_PEEK) == 0) {
1609 KASSERT(m->m_nextpkt == nextrecord,
1610 ("soreceive: post-control, nextrecord !sync"));
1611 if (nextrecord == NULL) {
1612 KASSERT(so->so_rcv.sb_mb == m,
1613 ("soreceive: post-control, sb_mb!=m"));
1614 KASSERT(so->so_rcv.sb_lastrecord == m,
1615 ("soreceive: post-control, lastrecord!=m"));
1616 }
1617 }
1618 type = m->m_type;
1619 if (type == MT_OOBDATA)
1620 flags |= MSG_OOB;
1621 } else {
1622 if ((flags & MSG_PEEK) == 0) {
1623 KASSERT(so->so_rcv.sb_mb == nextrecord,
1624 ("soreceive: sb_mb != nextrecord"));
1625 if (so->so_rcv.sb_mb == NULL) {
1626 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1627 ("soreceive: sb_lastercord != NULL"));
1628 }
1629 }
1630 }
1631 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1632 SBLASTRECORDCHK(&so->so_rcv);
1633 SBLASTMBUFCHK(&so->so_rcv);
1634
1635 /*
1636 * Now continue to read any data mbufs off of the head of the socket
1637 * buffer until the read request is satisfied. Note that 'type' is
1638 * used to store the type of any mbuf reads that have happened so far
1639 * such that soreceive() can stop reading if the type changes, which
1640 * causes soreceive() to return only one of regular data and inline
1641 * out-of-band data in a single socket receive operation.
1642 */
1643 moff = 0;
1644 offset = 0;
1645 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1646 /*
1647 * If the type of mbuf has changed since the last mbuf
1648 * examined ('type'), end the receive operation.
1649 */
1650 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1651 if (m->m_type == MT_OOBDATA) {
1652 if (type != MT_OOBDATA)
1653 break;
1654 } else if (type == MT_OOBDATA)
1655 break;
1656 else
1657 KASSERT(m->m_type == MT_DATA,
1658 ("m->m_type == %d", m->m_type));
1659 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1660 len = uio->uio_resid;
1661 if (so->so_oobmark && len > so->so_oobmark - offset)
1662 len = so->so_oobmark - offset;
1663 if (len > m->m_len - moff)
1664 len = m->m_len - moff;
1665 /*
1666 * If mp is set, just pass back the mbufs. Otherwise copy
1667 * them out via the uio, then free. Sockbuf must be
1668 * consistent here (points to current mbuf, it points to next
1669 * record) when we drop priority; we must note any additions
1670 * to the sockbuf when we block interrupts again.
1671 */
1672 if (mp == NULL) {
1673 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1674 SBLASTRECORDCHK(&so->so_rcv);
1675 SBLASTMBUFCHK(&so->so_rcv);
1676 SOCKBUF_UNLOCK(&so->so_rcv);
1677 #ifdef ZERO_COPY_SOCKETS
1678 if (so_zero_copy_receive) {
1679 int disposable;
1680
1681 if ((m->m_flags & M_EXT)
1682 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1683 disposable = 1;
1684 else
1685 disposable = 0;
1686
1687 error = uiomoveco(mtod(m, char *) + moff,
1688 (int)len, uio,
1689 disposable);
1690 } else
1691 #endif /* ZERO_COPY_SOCKETS */
1692 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1693 SOCKBUF_LOCK(&so->so_rcv);
1694 if (error) {
1695 /*
1696 * The MT_SONAME mbuf has already been removed
1697 * from the record, so it is necessary to
1698 * remove the data mbufs, if any, to preserve
1699 * the invariant in the case of PR_ADDR that
1700 * requires MT_SONAME mbufs at the head of
1701 * each record.
1702 */
1703 if (m && pr->pr_flags & PR_ATOMIC &&
1704 ((flags & MSG_PEEK) == 0))
1705 (void)sbdroprecord_locked(&so->so_rcv);
1706 SOCKBUF_UNLOCK(&so->so_rcv);
1707 goto release;
1708 }
1709 } else
1710 uio->uio_resid -= len;
1711 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1712 if (len == m->m_len - moff) {
1713 if (m->m_flags & M_EOR)
1714 flags |= MSG_EOR;
1715 if (flags & MSG_PEEK) {
1716 m = m->m_next;
1717 moff = 0;
1718 } else {
1719 nextrecord = m->m_nextpkt;
1720 sbfree(&so->so_rcv, m);
1721 if (mp != NULL) {
1722 *mp = m;
1723 mp = &m->m_next;
1724 so->so_rcv.sb_mb = m = m->m_next;
1725 *mp = NULL;
1726 } else {
1727 so->so_rcv.sb_mb = m_free(m);
1728 m = so->so_rcv.sb_mb;
1729 }
1730 sockbuf_pushsync(&so->so_rcv, nextrecord);
1731 SBLASTRECORDCHK(&so->so_rcv);
1732 SBLASTMBUFCHK(&so->so_rcv);
1733 }
1734 } else {
1735 if (flags & MSG_PEEK)
1736 moff += len;
1737 else {
1738 if (mp != NULL) {
1739 int copy_flag;
1740
1741 if (flags & MSG_DONTWAIT)
1742 copy_flag = M_DONTWAIT;
1743 else
1744 copy_flag = M_WAIT;
1745 if (copy_flag == M_WAIT)
1746 SOCKBUF_UNLOCK(&so->so_rcv);
1747 *mp = m_copym(m, 0, len, copy_flag);
1748 if (copy_flag == M_WAIT)
1749 SOCKBUF_LOCK(&so->so_rcv);
1750 if (*mp == NULL) {
1751 /*
1752 * m_copym() couldn't
1753 * allocate an mbuf. Adjust
1754 * uio_resid back (it was
1755 * adjusted down by len
1756 * bytes, which we didn't end
1757 * up "copying" over).
1758 */
1759 uio->uio_resid += len;
1760 break;
1761 }
1762 }
1763 m->m_data += len;
1764 m->m_len -= len;
1765 so->so_rcv.sb_cc -= len;
1766 }
1767 }
1768 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1769 if (so->so_oobmark) {
1770 if ((flags & MSG_PEEK) == 0) {
1771 so->so_oobmark -= len;
1772 if (so->so_oobmark == 0) {
1773 so->so_rcv.sb_state |= SBS_RCVATMARK;
1774 break;
1775 }
1776 } else {
1777 offset += len;
1778 if (offset == so->so_oobmark)
1779 break;
1780 }
1781 }
1782 if (flags & MSG_EOR)
1783 break;
1784 /*
1785 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1786 * must not quit until "uio->uio_resid == 0" or an error
1787 * termination. If a signal/timeout occurs, return with a
1788 * short count but without error. Keep sockbuf locked
1789 * against other readers.
1790 */
1791 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1792 !sosendallatonce(so) && nextrecord == NULL) {
1793 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1794 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1795 break;
1796 /*
1797 * Notify the protocol that some data has been
1798 * drained before blocking.
1799 */
1800 if (pr->pr_flags & PR_WANTRCVD) {
1801 SOCKBUF_UNLOCK(&so->so_rcv);
1802 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1803 SOCKBUF_LOCK(&so->so_rcv);
1804 }
1805 SBLASTRECORDCHK(&so->so_rcv);
1806 SBLASTMBUFCHK(&so->so_rcv);
1807 error = sbwait(&so->so_rcv);
1808 if (error) {
1809 SOCKBUF_UNLOCK(&so->so_rcv);
1810 goto release;
1811 }
1812 m = so->so_rcv.sb_mb;
1813 if (m != NULL)
1814 nextrecord = m->m_nextpkt;
1815 }
1816 }
1817
1818 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1819 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1820 flags |= MSG_TRUNC;
1821 if ((flags & MSG_PEEK) == 0)
1822 (void) sbdroprecord_locked(&so->so_rcv);
1823 }
1824 if ((flags & MSG_PEEK) == 0) {
1825 if (m == NULL) {
1826 /*
1827 * First part is an inline SB_EMPTY_FIXUP(). Second
1828 * part makes sure sb_lastrecord is up-to-date if
1829 * there is still data in the socket buffer.
1830 */
1831 so->so_rcv.sb_mb = nextrecord;
1832 if (so->so_rcv.sb_mb == NULL) {
1833 so->so_rcv.sb_mbtail = NULL;
1834 so->so_rcv.sb_lastrecord = NULL;
1835 } else if (nextrecord->m_nextpkt == NULL)
1836 so->so_rcv.sb_lastrecord = nextrecord;
1837 }
1838 SBLASTRECORDCHK(&so->so_rcv);
1839 SBLASTMBUFCHK(&so->so_rcv);
1840 /*
1841 * If soreceive() is being done from the socket callback,
1842 * then don't need to generate ACK to peer to update window,
1843 * since ACK will be generated on return to TCP.
1844 */
1845 if (!(flags & MSG_SOCALLBCK) &&
1846 (pr->pr_flags & PR_WANTRCVD)) {
1847 SOCKBUF_UNLOCK(&so->so_rcv);
1848 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1849 SOCKBUF_LOCK(&so->so_rcv);
1850 }
1851 }
1852 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1853 if (orig_resid == uio->uio_resid && orig_resid &&
1854 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1855 SOCKBUF_UNLOCK(&so->so_rcv);
1856 goto restart;
1857 }
1858 SOCKBUF_UNLOCK(&so->so_rcv);
1859
1860 if (flagsp != NULL)
1861 *flagsp |= flags;
1862 release:
1863 sbunlock(&so->so_rcv);
1864 return (error);
1865 }
1866
1867 /*
1868 * Optimized version of soreceive() for stream (TCP) sockets.
1869 */
1870 #ifdef TCP_SORECEIVE_STREAM
1871 int
1872 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1873 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1874 {
1875 int len = 0, error = 0, flags, oresid;
1876 struct sockbuf *sb;
1877 struct mbuf *m, *n = NULL;
1878
1879 /* We only do stream sockets. */
1880 if (so->so_type != SOCK_STREAM)
1881 return (EINVAL);
1882 if (psa != NULL)
1883 *psa = NULL;
1884 if (controlp != NULL)
1885 return (EINVAL);
1886 if (flagsp != NULL)
1887 flags = *flagsp &~ MSG_EOR;
1888 else
1889 flags = 0;
1890 if (flags & MSG_OOB)
1891 return (soreceive_rcvoob(so, uio, flags));
1892 if (mp0 != NULL)
1893 *mp0 = NULL;
1894
1895 sb = &so->so_rcv;
1896
1897 /* Prevent other readers from entering the socket. */
1898 error = sblock(sb, SBLOCKWAIT(flags));
1899 if (error)
1900 goto out;
1901 SOCKBUF_LOCK(sb);
1902
1903 /* Easy one, no space to copyout anything. */
1904 if (uio->uio_resid == 0) {
1905 error = EINVAL;
1906 goto out;
1907 }
1908 oresid = uio->uio_resid;
1909
1910 /* We will never ever get anything unless we are connected. */
1911 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1912 /* When disconnecting there may be still some data left. */
1913 if (sb->sb_cc > 0)
1914 goto deliver;
1915 if (!(so->so_state & SS_ISDISCONNECTED))
1916 error = ENOTCONN;
1917 goto out;
1918 }
1919
1920 /* Socket buffer is empty and we shall not block. */
1921 if (sb->sb_cc == 0 &&
1922 ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1923 error = EAGAIN;
1924 goto out;
1925 }
1926
1927 restart:
1928 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1929
1930 /* Abort if socket has reported problems. */
1931 if (so->so_error) {
1932 if (sb->sb_cc > 0)
1933 goto deliver;
1934 if (oresid > uio->uio_resid)
1935 goto out;
1936 error = so->so_error;
1937 if (!(flags & MSG_PEEK))
1938 so->so_error = 0;
1939 goto out;
1940 }
1941
1942 /* Door is closed. Deliver what is left, if any. */
1943 if (sb->sb_state & SBS_CANTRCVMORE) {
1944 if (sb->sb_cc > 0)
1945 goto deliver;
1946 else
1947 goto out;
1948 }
1949
1950 /* Socket buffer got some data that we shall deliver now. */
1951 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1952 ((sb->sb_flags & SS_NBIO) ||
1953 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1954 sb->sb_cc >= sb->sb_lowat ||
1955 sb->sb_cc >= uio->uio_resid ||
1956 sb->sb_cc >= sb->sb_hiwat) ) {
1957 goto deliver;
1958 }
1959
1960 /* On MSG_WAITALL we must wait until all data or error arrives. */
1961 if ((flags & MSG_WAITALL) &&
1962 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
1963 goto deliver;
1964
1965 /*
1966 * Wait and block until (more) data comes in.
1967 * NB: Drops the sockbuf lock during wait.
1968 */
1969 error = sbwait(sb);
1970 if (error)
1971 goto out;
1972 goto restart;
1973
1974 deliver:
1975 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1976 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1977 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1978
1979 /* Statistics. */
1980 if (uio->uio_td)
1981 uio->uio_td->td_ru.ru_msgrcv++;
1982
1983 /* Fill uio until full or current end of socket buffer is reached. */
1984 len = min(uio->uio_resid, sb->sb_cc);
1985 if (mp0 != NULL) {
1986 /* Dequeue as many mbufs as possible. */
1987 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1988 for (*mp0 = m = sb->sb_mb;
1989 m != NULL && m->m_len <= len;
1990 m = m->m_next) {
1991 len -= m->m_len;
1992 uio->uio_resid -= m->m_len;
1993 sbfree(sb, m);
1994 n = m;
1995 }
1996 sb->sb_mb = m;
1997 if (sb->sb_mb == NULL)
1998 SB_EMPTY_FIXUP(sb);
1999 n->m_next = NULL;
2000 }
2001 /* Copy the remainder. */
2002 if (len > 0) {
2003 KASSERT(sb->sb_mb != NULL,
2004 ("%s: len > 0 && sb->sb_mb empty", __func__));
2005
2006 m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2007 if (m == NULL)
2008 len = 0; /* Don't flush data from sockbuf. */
2009 else
2010 uio->uio_resid -= m->m_len;
2011 if (*mp0 != NULL)
2012 n->m_next = m;
2013 else
2014 *mp0 = m;
2015 if (*mp0 == NULL) {
2016 error = ENOBUFS;
2017 goto out;
2018 }
2019 }
2020 } else {
2021 /* NB: Must unlock socket buffer as uiomove may sleep. */
2022 SOCKBUF_UNLOCK(sb);
2023 error = m_mbuftouio(uio, sb->sb_mb, len);
2024 SOCKBUF_LOCK(sb);
2025 if (error)
2026 goto out;
2027 }
2028 SBLASTRECORDCHK(sb);
2029 SBLASTMBUFCHK(sb);
2030
2031 /*
2032 * Remove the delivered data from the socket buffer unless we
2033 * were only peeking.
2034 */
2035 if (!(flags & MSG_PEEK)) {
2036 if (len > 0)
2037 sbdrop_locked(sb, len);
2038
2039 /* Notify protocol that we drained some data. */
2040 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2041 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2042 !(flags & MSG_SOCALLBCK))) {
2043 SOCKBUF_UNLOCK(sb);
2044 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2045 SOCKBUF_LOCK(sb);
2046 }
2047 }
2048
2049 /*
2050 * For MSG_WAITALL we may have to loop again and wait for
2051 * more data to come in.
2052 */
2053 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2054 goto restart;
2055 out:
2056 SOCKBUF_LOCK_ASSERT(sb);
2057 SBLASTRECORDCHK(sb);
2058 SBLASTMBUFCHK(sb);
2059 SOCKBUF_UNLOCK(sb);
2060 sbunlock(sb);
2061 return (error);
2062 }
2063 #endif /* TCP_SORECEIVE_STREAM */
2064
2065 /*
2066 * Optimized version of soreceive() for simple datagram cases from userspace.
2067 * Unlike in the stream case, we're able to drop a datagram if copyout()
2068 * fails, and because we handle datagrams atomically, we don't need to use a
2069 * sleep lock to prevent I/O interlacing.
2070 */
2071 int
2072 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2073 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2074 {
2075 struct mbuf *m, *m2;
2076 int flags, len, error;
2077 struct protosw *pr = so->so_proto;
2078 struct mbuf *nextrecord;
2079
2080 if (psa != NULL)
2081 *psa = NULL;
2082 if (controlp != NULL)
2083 *controlp = NULL;
2084 if (flagsp != NULL)
2085 flags = *flagsp &~ MSG_EOR;
2086 else
2087 flags = 0;
2088
2089 /*
2090 * For any complicated cases, fall back to the full
2091 * soreceive_generic().
2092 */
2093 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2094 return (soreceive_generic(so, psa, uio, mp0, controlp,
2095 flagsp));
2096
2097 /*
2098 * Enforce restrictions on use.
2099 */
2100 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2101 ("soreceive_dgram: wantrcvd"));
2102 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2103 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2104 ("soreceive_dgram: SBS_RCVATMARK"));
2105 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2106 ("soreceive_dgram: P_CONNREQUIRED"));
2107
2108 /*
2109 * Loop blocking while waiting for a datagram.
2110 */
2111 SOCKBUF_LOCK(&so->so_rcv);
2112 while ((m = so->so_rcv.sb_mb) == NULL) {
2113 KASSERT(so->so_rcv.sb_cc == 0,
2114 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2115 so->so_rcv.sb_cc));
2116 if (so->so_error) {
2117 error = so->so_error;
2118 so->so_error = 0;
2119 SOCKBUF_UNLOCK(&so->so_rcv);
2120 return (error);
2121 }
2122 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2123 uio->uio_resid == 0) {
2124 SOCKBUF_UNLOCK(&so->so_rcv);
2125 return (0);
2126 }
2127 if ((so->so_state & SS_NBIO) ||
2128 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2129 SOCKBUF_UNLOCK(&so->so_rcv);
2130 return (EWOULDBLOCK);
2131 }
2132 SBLASTRECORDCHK(&so->so_rcv);
2133 SBLASTMBUFCHK(&so->so_rcv);
2134 error = sbwait(&so->so_rcv);
2135 if (error) {
2136 SOCKBUF_UNLOCK(&so->so_rcv);
2137 return (error);
2138 }
2139 }
2140 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2141
2142 if (uio->uio_td)
2143 uio->uio_td->td_ru.ru_msgrcv++;
2144 SBLASTRECORDCHK(&so->so_rcv);
2145 SBLASTMBUFCHK(&so->so_rcv);
2146 nextrecord = m->m_nextpkt;
2147 if (nextrecord == NULL) {
2148 KASSERT(so->so_rcv.sb_lastrecord == m,
2149 ("soreceive_dgram: lastrecord != m"));
2150 }
2151
2152 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2153 ("soreceive_dgram: m_nextpkt != nextrecord"));
2154
2155 /*
2156 * Pull 'm' and its chain off the front of the packet queue.
2157 */
2158 so->so_rcv.sb_mb = NULL;
2159 sockbuf_pushsync(&so->so_rcv, nextrecord);
2160
2161 /*
2162 * Walk 'm's chain and free that many bytes from the socket buffer.
2163 */
2164 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2165 sbfree(&so->so_rcv, m2);
2166
2167 /*
2168 * Do a few last checks before we let go of the lock.
2169 */
2170 SBLASTRECORDCHK(&so->so_rcv);
2171 SBLASTMBUFCHK(&so->so_rcv);
2172 SOCKBUF_UNLOCK(&so->so_rcv);
2173
2174 if (pr->pr_flags & PR_ADDR) {
2175 KASSERT(m->m_type == MT_SONAME,
2176 ("m->m_type == %d", m->m_type));
2177 if (psa != NULL)
2178 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2179 M_NOWAIT);
2180 m = m_free(m);
2181 }
2182 if (m == NULL) {
2183 /* XXXRW: Can this happen? */
2184 return (0);
2185 }
2186
2187 /*
2188 * Packet to copyout() is now in 'm' and it is disconnected from the
2189 * queue.
2190 *
2191 * Process one or more MT_CONTROL mbufs present before any data mbufs
2192 * in the first mbuf chain on the socket buffer. We call into the
2193 * protocol to perform externalization (or freeing if controlp ==
2194 * NULL).
2195 */
2196 if (m->m_type == MT_CONTROL) {
2197 struct mbuf *cm = NULL, *cmn;
2198 struct mbuf **cme = &cm;
2199
2200 do {
2201 m2 = m->m_next;
2202 m->m_next = NULL;
2203 *cme = m;
2204 cme = &(*cme)->m_next;
2205 m = m2;
2206 } while (m != NULL && m->m_type == MT_CONTROL);
2207 while (cm != NULL) {
2208 cmn = cm->m_next;
2209 cm->m_next = NULL;
2210 if (pr->pr_domain->dom_externalize != NULL) {
2211 error = (*pr->pr_domain->dom_externalize)
2212 (cm, controlp);
2213 } else if (controlp != NULL)
2214 *controlp = cm;
2215 else
2216 m_freem(cm);
2217 if (controlp != NULL) {
2218 while (*controlp != NULL)
2219 controlp = &(*controlp)->m_next;
2220 }
2221 cm = cmn;
2222 }
2223 }
2224 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2225
2226 while (m != NULL && uio->uio_resid > 0) {
2227 len = uio->uio_resid;
2228 if (len > m->m_len)
2229 len = m->m_len;
2230 error = uiomove(mtod(m, char *), (int)len, uio);
2231 if (error) {
2232 m_freem(m);
2233 return (error);
2234 }
2235 m = m_free(m);
2236 }
2237 if (m != NULL)
2238 flags |= MSG_TRUNC;
2239 m_freem(m);
2240 if (flagsp != NULL)
2241 *flagsp |= flags;
2242 return (0);
2243 }
2244
2245 int
2246 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2247 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2248 {
2249
2250 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2251 controlp, flagsp));
2252 }
2253
2254 int
2255 soshutdown(struct socket *so, int how)
2256 {
2257 struct protosw *pr = so->so_proto;
2258 int error;
2259
2260 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2261 return (EINVAL);
2262 if (pr->pr_usrreqs->pru_flush != NULL) {
2263 (*pr->pr_usrreqs->pru_flush)(so, how);
2264 }
2265 if (how != SHUT_WR)
2266 sorflush(so);
2267 if (how != SHUT_RD) {
2268 CURVNET_SET(so->so_vnet);
2269 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2270 CURVNET_RESTORE();
2271 return (error);
2272 }
2273 return (0);
2274 }
2275
2276 void
2277 sorflush(struct socket *so)
2278 {
2279 struct sockbuf *sb = &so->so_rcv;
2280 struct protosw *pr = so->so_proto;
2281 struct sockbuf asb;
2282
2283 /*
2284 * In order to avoid calling dom_dispose with the socket buffer mutex
2285 * held, and in order to generally avoid holding the lock for a long
2286 * time, we make a copy of the socket buffer and clear the original
2287 * (except locks, state). The new socket buffer copy won't have
2288 * initialized locks so we can only call routines that won't use or
2289 * assert those locks.
2290 *
2291 * Dislodge threads currently blocked in receive and wait to acquire
2292 * a lock against other simultaneous readers before clearing the
2293 * socket buffer. Don't let our acquire be interrupted by a signal
2294 * despite any existing socket disposition on interruptable waiting.
2295 */
2296 CURVNET_SET(so->so_vnet);
2297 socantrcvmore(so);
2298 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2299
2300 /*
2301 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2302 * and mutex data unchanged.
2303 */
2304 SOCKBUF_LOCK(sb);
2305 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2306 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2307 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2308 bzero(&sb->sb_startzero,
2309 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2310 SOCKBUF_UNLOCK(sb);
2311 sbunlock(sb);
2312
2313 /*
2314 * Dispose of special rights and flush the socket buffer. Don't call
2315 * any unsafe routines (that rely on locks being initialized) on asb.
2316 */
2317 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2318 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2319 sbrelease_internal(&asb, so);
2320 CURVNET_RESTORE();
2321 }
2322
2323 /*
2324 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2325 * additional variant to handle the case where the option value needs to be
2326 * some kind of integer, but not a specific size. In addition to their use
2327 * here, these functions are also called by the protocol-level pr_ctloutput()
2328 * routines.
2329 */
2330 int
2331 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2332 {
2333 size_t valsize;
2334
2335 /*
2336 * If the user gives us more than we wanted, we ignore it, but if we
2337 * don't get the minimum length the caller wants, we return EINVAL.
2338 * On success, sopt->sopt_valsize is set to however much we actually
2339 * retrieved.
2340 */
2341 if ((valsize = sopt->sopt_valsize) < minlen)
2342 return EINVAL;
2343 if (valsize > len)
2344 sopt->sopt_valsize = valsize = len;
2345
2346 if (sopt->sopt_td != NULL)
2347 return (copyin(sopt->sopt_val, buf, valsize));
2348
2349 bcopy(sopt->sopt_val, buf, valsize);
2350 return (0);
2351 }
2352
2353 /*
2354 * Kernel version of setsockopt(2).
2355 *
2356 * XXX: optlen is size_t, not socklen_t
2357 */
2358 int
2359 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2360 size_t optlen)
2361 {
2362 struct sockopt sopt;
2363
2364 sopt.sopt_level = level;
2365 sopt.sopt_name = optname;
2366 sopt.sopt_dir = SOPT_SET;
2367 sopt.sopt_val = optval;
2368 sopt.sopt_valsize = optlen;
2369 sopt.sopt_td = NULL;
2370 return (sosetopt(so, &sopt));
2371 }
2372
2373 int
2374 sosetopt(struct socket *so, struct sockopt *sopt)
2375 {
2376 int error, optval;
2377 struct linger l;
2378 struct timeval tv;
2379 u_long val;
2380 #ifdef MAC
2381 struct mac extmac;
2382 #endif
2383
2384 error = 0;
2385 if (sopt->sopt_level != SOL_SOCKET) {
2386 if (so->so_proto && so->so_proto->pr_ctloutput)
2387 return ((*so->so_proto->pr_ctloutput)
2388 (so, sopt));
2389 error = ENOPROTOOPT;
2390 } else {
2391 switch (sopt->sopt_name) {
2392 #ifdef INET
2393 case SO_ACCEPTFILTER:
2394 error = do_setopt_accept_filter(so, sopt);
2395 if (error)
2396 goto bad;
2397 break;
2398 #endif
2399 case SO_LINGER:
2400 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2401 if (error)
2402 goto bad;
2403
2404 SOCK_LOCK(so);
2405 so->so_linger = l.l_linger;
2406 if (l.l_onoff)
2407 so->so_options |= SO_LINGER;
2408 else
2409 so->so_options &= ~SO_LINGER;
2410 SOCK_UNLOCK(so);
2411 break;
2412
2413 case SO_DEBUG:
2414 case SO_KEEPALIVE:
2415 case SO_DONTROUTE:
2416 case SO_USELOOPBACK:
2417 case SO_BROADCAST:
2418 case SO_REUSEADDR:
2419 case SO_REUSEPORT:
2420 case SO_OOBINLINE:
2421 case SO_TIMESTAMP:
2422 case SO_BINTIME:
2423 case SO_NOSIGPIPE:
2424 case SO_NO_DDP:
2425 case SO_NO_OFFLOAD:
2426 error = sooptcopyin(sopt, &optval, sizeof optval,
2427 sizeof optval);
2428 if (error)
2429 goto bad;
2430 SOCK_LOCK(so);
2431 if (optval)
2432 so->so_options |= sopt->sopt_name;
2433 else
2434 so->so_options &= ~sopt->sopt_name;
2435 SOCK_UNLOCK(so);
2436 break;
2437
2438 case SO_SETFIB:
2439 error = sooptcopyin(sopt, &optval, sizeof optval,
2440 sizeof optval);
2441 if (optval < 1 || optval > rt_numfibs) {
2442 error = EINVAL;
2443 goto bad;
2444 }
2445 if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2446 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2447 so->so_fibnum = optval;
2448 /* Note: ignore error */
2449 if (so->so_proto && so->so_proto->pr_ctloutput)
2450 (*so->so_proto->pr_ctloutput)(so, sopt);
2451 } else {
2452 so->so_fibnum = 0;
2453 }
2454 break;
2455 case SO_SNDBUF:
2456 case SO_RCVBUF:
2457 case SO_SNDLOWAT:
2458 case SO_RCVLOWAT:
2459 error = sooptcopyin(sopt, &optval, sizeof optval,
2460 sizeof optval);
2461 if (error)
2462 goto bad;
2463
2464 /*
2465 * Values < 1 make no sense for any of these options,
2466 * so disallow them.
2467 */
2468 if (optval < 1) {
2469 error = EINVAL;
2470 goto bad;
2471 }
2472
2473 switch (sopt->sopt_name) {
2474 case SO_SNDBUF:
2475 case SO_RCVBUF:
2476 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2477 &so->so_snd : &so->so_rcv, (u_long)optval,
2478 so, curthread) == 0) {
2479 error = ENOBUFS;
2480 goto bad;
2481 }
2482 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2483 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2484 break;
2485
2486 /*
2487 * Make sure the low-water is never greater than the
2488 * high-water.
2489 */
2490 case SO_SNDLOWAT:
2491 SOCKBUF_LOCK(&so->so_snd);
2492 so->so_snd.sb_lowat =
2493 (optval > so->so_snd.sb_hiwat) ?
2494 so->so_snd.sb_hiwat : optval;
2495 SOCKBUF_UNLOCK(&so->so_snd);
2496 break;
2497 case SO_RCVLOWAT:
2498 SOCKBUF_LOCK(&so->so_rcv);
2499 so->so_rcv.sb_lowat =
2500 (optval > so->so_rcv.sb_hiwat) ?
2501 so->so_rcv.sb_hiwat : optval;
2502 SOCKBUF_UNLOCK(&so->so_rcv);
2503 break;
2504 }
2505 break;
2506
2507 case SO_SNDTIMEO:
2508 case SO_RCVTIMEO:
2509 #ifdef COMPAT_IA32
2510 if (SV_CURPROC_FLAG(SV_ILP32)) {
2511 struct timeval32 tv32;
2512
2513 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2514 sizeof tv32);
2515 CP(tv32, tv, tv_sec);
2516 CP(tv32, tv, tv_usec);
2517 } else
2518 #endif
2519 error = sooptcopyin(sopt, &tv, sizeof tv,
2520 sizeof tv);
2521 if (error)
2522 goto bad;
2523
2524 /* assert(hz > 0); */
2525 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2526 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2527 error = EDOM;
2528 goto bad;
2529 }
2530 /* assert(tick > 0); */
2531 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2532 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2533 if (val > INT_MAX) {
2534 error = EDOM;
2535 goto bad;
2536 }
2537 if (val == 0 && tv.tv_usec != 0)
2538 val = 1;
2539
2540 switch (sopt->sopt_name) {
2541 case SO_SNDTIMEO:
2542 so->so_snd.sb_timeo = val;
2543 break;
2544 case SO_RCVTIMEO:
2545 so->so_rcv.sb_timeo = val;
2546 break;
2547 }
2548 break;
2549
2550 case SO_LABEL:
2551 #ifdef MAC
2552 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2553 sizeof extmac);
2554 if (error)
2555 goto bad;
2556 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2557 so, &extmac);
2558 #else
2559 error = EOPNOTSUPP;
2560 #endif
2561 break;
2562
2563 default:
2564 error = ENOPROTOOPT;
2565 break;
2566 }
2567 if (error == 0 && so->so_proto != NULL &&
2568 so->so_proto->pr_ctloutput != NULL) {
2569 (void) ((*so->so_proto->pr_ctloutput)
2570 (so, sopt));
2571 }
2572 }
2573 bad:
2574 return (error);
2575 }
2576
2577 /*
2578 * Helper routine for getsockopt.
2579 */
2580 int
2581 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2582 {
2583 int error;
2584 size_t valsize;
2585
2586 error = 0;
2587
2588 /*
2589 * Documented get behavior is that we always return a value, possibly
2590 * truncated to fit in the user's buffer. Traditional behavior is
2591 * that we always tell the user precisely how much we copied, rather
2592 * than something useful like the total amount we had available for
2593 * her. Note that this interface is not idempotent; the entire
2594 * answer must generated ahead of time.
2595 */
2596 valsize = min(len, sopt->sopt_valsize);
2597 sopt->sopt_valsize = valsize;
2598 if (sopt->sopt_val != NULL) {
2599 if (sopt->sopt_td != NULL)
2600 error = copyout(buf, sopt->sopt_val, valsize);
2601 else
2602 bcopy(buf, sopt->sopt_val, valsize);
2603 }
2604 return (error);
2605 }
2606
2607 int
2608 sogetopt(struct socket *so, struct sockopt *sopt)
2609 {
2610 int error, optval;
2611 struct linger l;
2612 struct timeval tv;
2613 #ifdef MAC
2614 struct mac extmac;
2615 #endif
2616
2617 error = 0;
2618 if (sopt->sopt_level != SOL_SOCKET) {
2619 if (so->so_proto && so->so_proto->pr_ctloutput) {
2620 return ((*so->so_proto->pr_ctloutput)
2621 (so, sopt));
2622 } else
2623 return (ENOPROTOOPT);
2624 } else {
2625 switch (sopt->sopt_name) {
2626 #ifdef INET
2627 case SO_ACCEPTFILTER:
2628 error = do_getopt_accept_filter(so, sopt);
2629 break;
2630 #endif
2631 case SO_LINGER:
2632 SOCK_LOCK(so);
2633 l.l_onoff = so->so_options & SO_LINGER;
2634 l.l_linger = so->so_linger;
2635 SOCK_UNLOCK(so);
2636 error = sooptcopyout(sopt, &l, sizeof l);
2637 break;
2638
2639 case SO_USELOOPBACK:
2640 case SO_DONTROUTE:
2641 case SO_DEBUG:
2642 case SO_KEEPALIVE:
2643 case SO_REUSEADDR:
2644 case SO_REUSEPORT:
2645 case SO_BROADCAST:
2646 case SO_OOBINLINE:
2647 case SO_ACCEPTCONN:
2648 case SO_TIMESTAMP:
2649 case SO_BINTIME:
2650 case SO_NOSIGPIPE:
2651 optval = so->so_options & sopt->sopt_name;
2652 integer:
2653 error = sooptcopyout(sopt, &optval, sizeof optval);
2654 break;
2655
2656 case SO_TYPE:
2657 optval = so->so_type;
2658 goto integer;
2659
2660 case SO_ERROR:
2661 SOCK_LOCK(so);
2662 optval = so->so_error;
2663 so->so_error = 0;
2664 SOCK_UNLOCK(so);
2665 goto integer;
2666
2667 case SO_SNDBUF:
2668 optval = so->so_snd.sb_hiwat;
2669 goto integer;
2670
2671 case SO_RCVBUF:
2672 optval = so->so_rcv.sb_hiwat;
2673 goto integer;
2674
2675 case SO_SNDLOWAT:
2676 optval = so->so_snd.sb_lowat;
2677 goto integer;
2678
2679 case SO_RCVLOWAT:
2680 optval = so->so_rcv.sb_lowat;
2681 goto integer;
2682
2683 case SO_SNDTIMEO:
2684 case SO_RCVTIMEO:
2685 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2686 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2687
2688 tv.tv_sec = optval / hz;
2689 tv.tv_usec = (optval % hz) * tick;
2690 #ifdef COMPAT_IA32
2691 if (SV_CURPROC_FLAG(SV_ILP32)) {
2692 struct timeval32 tv32;
2693
2694 CP(tv, tv32, tv_sec);
2695 CP(tv, tv32, tv_usec);
2696 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2697 } else
2698 #endif
2699 error = sooptcopyout(sopt, &tv, sizeof tv);
2700 break;
2701
2702 case SO_LABEL:
2703 #ifdef MAC
2704 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2705 sizeof(extmac));
2706 if (error)
2707 return (error);
2708 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2709 so, &extmac);
2710 if (error)
2711 return (error);
2712 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2713 #else
2714 error = EOPNOTSUPP;
2715 #endif
2716 break;
2717
2718 case SO_PEERLABEL:
2719 #ifdef MAC
2720 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2721 sizeof(extmac));
2722 if (error)
2723 return (error);
2724 error = mac_getsockopt_peerlabel(
2725 sopt->sopt_td->td_ucred, so, &extmac);
2726 if (error)
2727 return (error);
2728 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2729 #else
2730 error = EOPNOTSUPP;
2731 #endif
2732 break;
2733
2734 case SO_LISTENQLIMIT:
2735 optval = so->so_qlimit;
2736 goto integer;
2737
2738 case SO_LISTENQLEN:
2739 optval = so->so_qlen;
2740 goto integer;
2741
2742 case SO_LISTENINCQLEN:
2743 optval = so->so_incqlen;
2744 goto integer;
2745
2746 default:
2747 error = ENOPROTOOPT;
2748 break;
2749 }
2750 return (error);
2751 }
2752 }
2753
2754 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2755 int
2756 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2757 {
2758 struct mbuf *m, *m_prev;
2759 int sopt_size = sopt->sopt_valsize;
2760
2761 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2762 if (m == NULL)
2763 return ENOBUFS;
2764 if (sopt_size > MLEN) {
2765 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2766 if ((m->m_flags & M_EXT) == 0) {
2767 m_free(m);
2768 return ENOBUFS;
2769 }
2770 m->m_len = min(MCLBYTES, sopt_size);
2771 } else {
2772 m->m_len = min(MLEN, sopt_size);
2773 }
2774 sopt_size -= m->m_len;
2775 *mp = m;
2776 m_prev = m;
2777
2778 while (sopt_size) {
2779 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2780 if (m == NULL) {
2781 m_freem(*mp);
2782 return ENOBUFS;
2783 }
2784 if (sopt_size > MLEN) {
2785 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2786 M_DONTWAIT);
2787 if ((m->m_flags & M_EXT) == 0) {
2788 m_freem(m);
2789 m_freem(*mp);
2790 return ENOBUFS;
2791 }
2792 m->m_len = min(MCLBYTES, sopt_size);
2793 } else {
2794 m->m_len = min(MLEN, sopt_size);
2795 }
2796 sopt_size -= m->m_len;
2797 m_prev->m_next = m;
2798 m_prev = m;
2799 }
2800 return (0);
2801 }
2802
2803 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2804 int
2805 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2806 {
2807 struct mbuf *m0 = m;
2808
2809 if (sopt->sopt_val == NULL)
2810 return (0);
2811 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2812 if (sopt->sopt_td != NULL) {
2813 int error;
2814
2815 error = copyin(sopt->sopt_val, mtod(m, char *),
2816 m->m_len);
2817 if (error != 0) {
2818 m_freem(m0);
2819 return(error);
2820 }
2821 } else
2822 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2823 sopt->sopt_valsize -= m->m_len;
2824 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2825 m = m->m_next;
2826 }
2827 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2828 panic("ip6_sooptmcopyin");
2829 return (0);
2830 }
2831
2832 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2833 int
2834 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2835 {
2836 struct mbuf *m0 = m;
2837 size_t valsize = 0;
2838
2839 if (sopt->sopt_val == NULL)
2840 return (0);
2841 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2842 if (sopt->sopt_td != NULL) {
2843 int error;
2844
2845 error = copyout(mtod(m, char *), sopt->sopt_val,
2846 m->m_len);
2847 if (error != 0) {
2848 m_freem(m0);
2849 return(error);
2850 }
2851 } else
2852 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2853 sopt->sopt_valsize -= m->m_len;
2854 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2855 valsize += m->m_len;
2856 m = m->m_next;
2857 }
2858 if (m != NULL) {
2859 /* enough soopt buffer should be given from user-land */
2860 m_freem(m0);
2861 return(EINVAL);
2862 }
2863 sopt->sopt_valsize = valsize;
2864 return (0);
2865 }
2866
2867 /*
2868 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2869 * out-of-band data, which will then notify socket consumers.
2870 */
2871 void
2872 sohasoutofband(struct socket *so)
2873 {
2874
2875 if (so->so_sigio != NULL)
2876 pgsigio(&so->so_sigio, SIGURG, 0);
2877 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2878 }
2879
2880 int
2881 sopoll(struct socket *so, int events, struct ucred *active_cred,
2882 struct thread *td)
2883 {
2884
2885 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2886 td));
2887 }
2888
2889 int
2890 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2891 struct thread *td)
2892 {
2893 int revents = 0;
2894
2895 SOCKBUF_LOCK(&so->so_snd);
2896 SOCKBUF_LOCK(&so->so_rcv);
2897 if (events & (POLLIN | POLLRDNORM))
2898 if (soreadabledata(so))
2899 revents |= events & (POLLIN | POLLRDNORM);
2900
2901 if (events & (POLLOUT | POLLWRNORM))
2902 if (sowriteable(so))
2903 revents |= events & (POLLOUT | POLLWRNORM);
2904
2905 if (events & (POLLPRI | POLLRDBAND))
2906 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2907 revents |= events & (POLLPRI | POLLRDBAND);
2908
2909 if ((events & POLLINIGNEOF) == 0) {
2910 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2911 revents |= events & (POLLIN | POLLRDNORM);
2912 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2913 revents |= POLLHUP;
2914 }
2915 }
2916
2917 if (revents == 0) {
2918 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2919 selrecord(td, &so->so_rcv.sb_sel);
2920 so->so_rcv.sb_flags |= SB_SEL;
2921 }
2922
2923 if (events & (POLLOUT | POLLWRNORM)) {
2924 selrecord(td, &so->so_snd.sb_sel);
2925 so->so_snd.sb_flags |= SB_SEL;
2926 }
2927 }
2928
2929 SOCKBUF_UNLOCK(&so->so_rcv);
2930 SOCKBUF_UNLOCK(&so->so_snd);
2931 return (revents);
2932 }
2933
2934 int
2935 soo_kqfilter(struct file *fp, struct knote *kn)
2936 {
2937 struct socket *so = kn->kn_fp->f_data;
2938 struct sockbuf *sb;
2939
2940 switch (kn->kn_filter) {
2941 case EVFILT_READ:
2942 if (so->so_options & SO_ACCEPTCONN)
2943 kn->kn_fop = &solisten_filtops;
2944 else
2945 kn->kn_fop = &soread_filtops;
2946 sb = &so->so_rcv;
2947 break;
2948 case EVFILT_WRITE:
2949 kn->kn_fop = &sowrite_filtops;
2950 sb = &so->so_snd;
2951 break;
2952 default:
2953 return (EINVAL);
2954 }
2955
2956 SOCKBUF_LOCK(sb);
2957 knlist_add(&sb->sb_sel.si_note, kn, 1);
2958 sb->sb_flags |= SB_KNOTE;
2959 SOCKBUF_UNLOCK(sb);
2960 return (0);
2961 }
2962
2963 /*
2964 * Some routines that return EOPNOTSUPP for entry points that are not
2965 * supported by a protocol. Fill in as needed.
2966 */
2967 int
2968 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2969 {
2970
2971 return EOPNOTSUPP;
2972 }
2973
2974 int
2975 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2976 {
2977
2978 return EOPNOTSUPP;
2979 }
2980
2981 int
2982 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2983 {
2984
2985 return EOPNOTSUPP;
2986 }
2987
2988 int
2989 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2990 {
2991
2992 return EOPNOTSUPP;
2993 }
2994
2995 int
2996 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2997 {
2998
2999 return EOPNOTSUPP;
3000 }
3001
3002 int
3003 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3004 struct ifnet *ifp, struct thread *td)
3005 {
3006
3007 return EOPNOTSUPP;
3008 }
3009
3010 int
3011 pru_disconnect_notsupp(struct socket *so)
3012 {
3013
3014 return EOPNOTSUPP;
3015 }
3016
3017 int
3018 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3019 {
3020
3021 return EOPNOTSUPP;
3022 }
3023
3024 int
3025 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3026 {
3027
3028 return EOPNOTSUPP;
3029 }
3030
3031 int
3032 pru_rcvd_notsupp(struct socket *so, int flags)
3033 {
3034
3035 return EOPNOTSUPP;
3036 }
3037
3038 int
3039 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3040 {
3041
3042 return EOPNOTSUPP;
3043 }
3044
3045 int
3046 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3047 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3048 {
3049
3050 return EOPNOTSUPP;
3051 }
3052
3053 /*
3054 * This isn't really a ``null'' operation, but it's the default one and
3055 * doesn't do anything destructive.
3056 */
3057 int
3058 pru_sense_null(struct socket *so, struct stat *sb)
3059 {
3060
3061 sb->st_blksize = so->so_snd.sb_hiwat;
3062 return 0;
3063 }
3064
3065 int
3066 pru_shutdown_notsupp(struct socket *so)
3067 {
3068
3069 return EOPNOTSUPP;
3070 }
3071
3072 int
3073 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3074 {
3075
3076 return EOPNOTSUPP;
3077 }
3078
3079 int
3080 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3081 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3082 {
3083
3084 return EOPNOTSUPP;
3085 }
3086
3087 int
3088 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3089 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3090 {
3091
3092 return EOPNOTSUPP;
3093 }
3094
3095 int
3096 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3097 struct thread *td)
3098 {
3099
3100 return EOPNOTSUPP;
3101 }
3102
3103 static void
3104 filt_sordetach(struct knote *kn)
3105 {
3106 struct socket *so = kn->kn_fp->f_data;
3107
3108 SOCKBUF_LOCK(&so->so_rcv);
3109 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3110 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3111 so->so_rcv.sb_flags &= ~SB_KNOTE;
3112 SOCKBUF_UNLOCK(&so->so_rcv);
3113 }
3114
3115 /*ARGSUSED*/
3116 static int
3117 filt_soread(struct knote *kn, long hint)
3118 {
3119 struct socket *so;
3120
3121 so = kn->kn_fp->f_data;
3122 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3123
3124 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3125 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3126 kn->kn_flags |= EV_EOF;
3127 kn->kn_fflags = so->so_error;
3128 return (1);
3129 } else if (so->so_error) /* temporary udp error */
3130 return (1);
3131 else if (kn->kn_sfflags & NOTE_LOWAT)
3132 return (kn->kn_data >= kn->kn_sdata);
3133 else
3134 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3135 }
3136
3137 static void
3138 filt_sowdetach(struct knote *kn)
3139 {
3140 struct socket *so = kn->kn_fp->f_data;
3141
3142 SOCKBUF_LOCK(&so->so_snd);
3143 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3144 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3145 so->so_snd.sb_flags &= ~SB_KNOTE;
3146 SOCKBUF_UNLOCK(&so->so_snd);
3147 }
3148
3149 /*ARGSUSED*/
3150 static int
3151 filt_sowrite(struct knote *kn, long hint)
3152 {
3153 struct socket *so;
3154
3155 so = kn->kn_fp->f_data;
3156 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3157 kn->kn_data = sbspace(&so->so_snd);
3158 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3159 kn->kn_flags |= EV_EOF;
3160 kn->kn_fflags = so->so_error;
3161 return (1);
3162 } else if (so->so_error) /* temporary udp error */
3163 return (1);
3164 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3165 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3166 return (0);
3167 else if (kn->kn_sfflags & NOTE_LOWAT)
3168 return (kn->kn_data >= kn->kn_sdata);
3169 else
3170 return (kn->kn_data >= so->so_snd.sb_lowat);
3171 }
3172
3173 /*ARGSUSED*/
3174 static int
3175 filt_solisten(struct knote *kn, long hint)
3176 {
3177 struct socket *so = kn->kn_fp->f_data;
3178
3179 kn->kn_data = so->so_qlen;
3180 return (! TAILQ_EMPTY(&so->so_comp));
3181 }
3182
3183 int
3184 socheckuid(struct socket *so, uid_t uid)
3185 {
3186
3187 if (so == NULL)
3188 return (EPERM);
3189 if (so->so_cred->cr_uid != uid)
3190 return (EPERM);
3191 return (0);
3192 }
3193
3194 static int
3195 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3196 {
3197 int error;
3198 int val;
3199
3200 val = somaxconn;
3201 error = sysctl_handle_int(oidp, &val, 0, req);
3202 if (error || !req->newptr )
3203 return (error);
3204
3205 if (val < 1 || val > USHRT_MAX)
3206 return (EINVAL);
3207
3208 somaxconn = val;
3209 return (0);
3210 }
3211
3212 /*
3213 * These functions are used by protocols to notify the socket layer (and its
3214 * consumers) of state changes in the sockets driven by protocol-side events.
3215 */
3216
3217 /*
3218 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3219 *
3220 * Normal sequence from the active (originating) side is that
3221 * soisconnecting() is called during processing of connect() call, resulting
3222 * in an eventual call to soisconnected() if/when the connection is
3223 * established. When the connection is torn down soisdisconnecting() is
3224 * called during processing of disconnect() call, and soisdisconnected() is
3225 * called when the connection to the peer is totally severed. The semantics
3226 * of these routines are such that connectionless protocols can call
3227 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3228 * calls when setting up a ``connection'' takes no time.
3229 *
3230 * From the passive side, a socket is created with two queues of sockets:
3231 * so_incomp for connections in progress and so_comp for connections already
3232 * made and awaiting user acceptance. As a protocol is preparing incoming
3233 * connections, it creates a socket structure queued on so_incomp by calling
3234 * sonewconn(). When the connection is established, soisconnected() is
3235 * called, and transfers the socket structure to so_comp, making it available
3236 * to accept().
3237 *
3238 * If a socket is closed with sockets on either so_incomp or so_comp, these
3239 * sockets are dropped.
3240 *
3241 * If higher-level protocols are implemented in the kernel, the wakeups done
3242 * here will sometimes cause software-interrupt process scheduling.
3243 */
3244 void
3245 soisconnecting(struct socket *so)
3246 {
3247
3248 SOCK_LOCK(so);
3249 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3250 so->so_state |= SS_ISCONNECTING;
3251 SOCK_UNLOCK(so);
3252 }
3253
3254 void
3255 soisconnected(struct socket *so)
3256 {
3257 struct socket *head;
3258 int ret;
3259
3260 restart:
3261 ACCEPT_LOCK();
3262 SOCK_LOCK(so);
3263 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3264 so->so_state |= SS_ISCONNECTED;
3265 head = so->so_head;
3266 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3267 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3268 SOCK_UNLOCK(so);
3269 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3270 head->so_incqlen--;
3271 so->so_qstate &= ~SQ_INCOMP;
3272 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3273 head->so_qlen++;
3274 so->so_qstate |= SQ_COMP;
3275 ACCEPT_UNLOCK();
3276 sorwakeup(head);
3277 wakeup_one(&head->so_timeo);
3278 } else {
3279 ACCEPT_UNLOCK();
3280 soupcall_set(so, SO_RCV,
3281 head->so_accf->so_accept_filter->accf_callback,
3282 head->so_accf->so_accept_filter_arg);
3283 so->so_options &= ~SO_ACCEPTFILTER;
3284 ret = head->so_accf->so_accept_filter->accf_callback(so,
3285 head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3286 if (ret == SU_ISCONNECTED)
3287 soupcall_clear(so, SO_RCV);
3288 SOCK_UNLOCK(so);
3289 if (ret == SU_ISCONNECTED)
3290 goto restart;
3291 }
3292 return;
3293 }
3294 SOCK_UNLOCK(so);
3295 ACCEPT_UNLOCK();
3296 wakeup(&so->so_timeo);
3297 sorwakeup(so);
3298 sowwakeup(so);
3299 }
3300
3301 void
3302 soisdisconnecting(struct socket *so)
3303 {
3304
3305 /*
3306 * Note: This code assumes that SOCK_LOCK(so) and
3307 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3308 */
3309 SOCKBUF_LOCK(&so->so_rcv);
3310 so->so_state &= ~SS_ISCONNECTING;
3311 so->so_state |= SS_ISDISCONNECTING;
3312 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3313 sorwakeup_locked(so);
3314 SOCKBUF_LOCK(&so->so_snd);
3315 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3316 sowwakeup_locked(so);
3317 wakeup(&so->so_timeo);
3318 }
3319
3320 void
3321 soisdisconnected(struct socket *so)
3322 {
3323
3324 /*
3325 * Note: This code assumes that SOCK_LOCK(so) and
3326 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3327 */
3328 SOCKBUF_LOCK(&so->so_rcv);
3329 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3330 so->so_state |= SS_ISDISCONNECTED;
3331 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3332 sorwakeup_locked(so);
3333 SOCKBUF_LOCK(&so->so_snd);
3334 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3335 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3336 sowwakeup_locked(so);
3337 wakeup(&so->so_timeo);
3338 }
3339
3340 /*
3341 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3342 */
3343 struct sockaddr *
3344 sodupsockaddr(const struct sockaddr *sa, int mflags)
3345 {
3346 struct sockaddr *sa2;
3347
3348 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3349 if (sa2)
3350 bcopy(sa, sa2, sa->sa_len);
3351 return sa2;
3352 }
3353
3354 /*
3355 * Register per-socket buffer upcalls.
3356 */
3357 void
3358 soupcall_set(struct socket *so, int which,
3359 int (*func)(struct socket *, void *, int), void *arg)
3360 {
3361 struct sockbuf *sb;
3362
3363 switch (which) {
3364 case SO_RCV:
3365 sb = &so->so_rcv;
3366 break;
3367 case SO_SND:
3368 sb = &so->so_snd;
3369 break;
3370 default:
3371 panic("soupcall_set: bad which");
3372 }
3373 SOCKBUF_LOCK_ASSERT(sb);
3374 #if 0
3375 /* XXX: accf_http actually wants to do this on purpose. */
3376 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3377 #endif
3378 sb->sb_upcall = func;
3379 sb->sb_upcallarg = arg;
3380 sb->sb_flags |= SB_UPCALL;
3381 }
3382
3383 void
3384 soupcall_clear(struct socket *so, int which)
3385 {
3386 struct sockbuf *sb;
3387
3388 switch (which) {
3389 case SO_RCV:
3390 sb = &so->so_rcv;
3391 break;
3392 case SO_SND:
3393 sb = &so->so_snd;
3394 break;
3395 default:
3396 panic("soupcall_clear: bad which");
3397 }
3398 SOCKBUF_LOCK_ASSERT(sb);
3399 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3400 sb->sb_upcall = NULL;
3401 sb->sb_upcallarg = NULL;
3402 sb->sb_flags &= ~SB_UPCALL;
3403 }
3404
3405 /*
3406 * Create an external-format (``xsocket'') structure using the information in
3407 * the kernel-format socket structure pointed to by so. This is done to
3408 * reduce the spew of irrelevant information over this interface, to isolate
3409 * user code from changes in the kernel structure, and potentially to provide
3410 * information-hiding if we decide that some of this information should be
3411 * hidden from users.
3412 */
3413 void
3414 sotoxsocket(struct socket *so, struct xsocket *xso)
3415 {
3416
3417 xso->xso_len = sizeof *xso;
3418 xso->xso_so = so;
3419 xso->so_type = so->so_type;
3420 xso->so_options = so->so_options;
3421 xso->so_linger = so->so_linger;
3422 xso->so_state = so->so_state;
3423 xso->so_pcb = so->so_pcb;
3424 xso->xso_protocol = so->so_proto->pr_protocol;
3425 xso->xso_family = so->so_proto->pr_domain->dom_family;
3426 xso->so_qlen = so->so_qlen;
3427 xso->so_incqlen = so->so_incqlen;
3428 xso->so_qlimit = so->so_qlimit;
3429 xso->so_timeo = so->so_timeo;
3430 xso->so_error = so->so_error;
3431 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3432 xso->so_oobmark = so->so_oobmark;
3433 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3434 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3435 xso->so_uid = so->so_cred->cr_uid;
3436 }
3437
3438
3439 /*
3440 * Socket accessor functions to provide external consumers with
3441 * a safe interface to socket state
3442 *
3443 */
3444
3445 void
3446 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3447 {
3448
3449 TAILQ_FOREACH(so, &so->so_comp, so_list)
3450 func(so, arg);
3451 }
3452
3453 struct sockbuf *
3454 so_sockbuf_rcv(struct socket *so)
3455 {
3456
3457 return (&so->so_rcv);
3458 }
3459
3460 struct sockbuf *
3461 so_sockbuf_snd(struct socket *so)
3462 {
3463
3464 return (&so->so_snd);
3465 }
3466
3467 int
3468 so_state_get(const struct socket *so)
3469 {
3470
3471 return (so->so_state);
3472 }
3473
3474 void
3475 so_state_set(struct socket *so, int val)
3476 {
3477
3478 so->so_state = val;
3479 }
3480
3481 int
3482 so_options_get(const struct socket *so)
3483 {
3484
3485 return (so->so_options);
3486 }
3487
3488 void
3489 so_options_set(struct socket *so, int val)
3490 {
3491
3492 so->so_options = val;
3493 }
3494
3495 int
3496 so_error_get(const struct socket *so)
3497 {
3498
3499 return (so->so_error);
3500 }
3501
3502 void
3503 so_error_set(struct socket *so, int val)
3504 {
3505
3506 so->so_error = val;
3507 }
3508
3509 int
3510 so_linger_get(const struct socket *so)
3511 {
3512
3513 return (so->so_linger);
3514 }
3515
3516 void
3517 so_linger_set(struct socket *so, int val)
3518 {
3519
3520 so->so_linger = val;
3521 }
3522
3523 struct protosw *
3524 so_protosw_get(const struct socket *so)
3525 {
3526
3527 return (so->so_proto);
3528 }
3529
3530 void
3531 so_protosw_set(struct socket *so, struct protosw *val)
3532 {
3533
3534 so->so_proto = val;
3535 }
3536
3537 void
3538 so_sorwakeup(struct socket *so)
3539 {
3540
3541 sorwakeup(so);
3542 }
3543
3544 void
3545 so_sowwakeup(struct socket *so)
3546 {
3547
3548 sowwakeup(so);
3549 }
3550
3551 void
3552 so_sorwakeup_locked(struct socket *so)
3553 {
3554
3555 sorwakeup_locked(so);
3556 }
3557
3558 void
3559 so_sowwakeup_locked(struct socket *so)
3560 {
3561
3562 sowwakeup_locked(so);
3563 }
3564
3565 void
3566 so_lock(struct socket *so)
3567 {
3568 SOCK_LOCK(so);
3569 }
3570
3571 void
3572 so_unlock(struct socket *so)
3573 {
3574 SOCK_UNLOCK(so);
3575 }
Cache object: 90efbbde14c328b3a25a299593303636
|