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: releng/8.1/sys/kern/uipc_socket.c 208692 2010-06-01 13:59:48Z rwatson $");
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_FREEBSD32
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 { 1, NULL, filt_sordetach, filt_solisten };
156 static struct filterops soread_filtops =
157 { 1, NULL, filt_sordetach, filt_soread };
158 static struct filterops sowrite_filtops =
159 { 1, NULL, filt_sowdetach, filt_sowrite };
160
161 uma_zone_t socket_zone;
162 so_gen_t so_gencnt; /* generation count for sockets */
163
164 int maxsockets;
165
166 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
167 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
168
169 static int somaxconn = SOMAXCONN;
170 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
171 /* XXX: we dont have SYSCTL_USHORT */
172 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
173 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
174 "queue size");
175 static int numopensockets;
176 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
177 &numopensockets, 0, "Number of open sockets");
178 #ifdef ZERO_COPY_SOCKETS
179 /* These aren't static because they're used in other files. */
180 int so_zero_copy_send = 1;
181 int so_zero_copy_receive = 1;
182 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
183 "Zero copy controls");
184 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
185 &so_zero_copy_receive, 0, "Enable zero copy receive");
186 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
187 &so_zero_copy_send, 0, "Enable zero copy send");
188 #endif /* ZERO_COPY_SOCKETS */
189
190 /*
191 * accept_mtx locks down per-socket fields relating to accept queues. See
192 * socketvar.h for an annotation of the protected fields of struct socket.
193 */
194 struct mtx accept_mtx;
195 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
196
197 /*
198 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
199 * so_gencnt field.
200 */
201 static struct mtx so_global_mtx;
202 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
203
204 /*
205 * General IPC sysctl name space, used by sockets and a variety of other IPC
206 * types.
207 */
208 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
209
210 /*
211 * Sysctl to get and set the maximum global sockets limit. Notify protocols
212 * of the change so that they can update their dependent limits as required.
213 */
214 static int
215 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
216 {
217 int error, newmaxsockets;
218
219 newmaxsockets = maxsockets;
220 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
221 if (error == 0 && req->newptr) {
222 if (newmaxsockets > maxsockets) {
223 maxsockets = newmaxsockets;
224 if (maxsockets > ((maxfiles / 4) * 3)) {
225 maxfiles = (maxsockets * 5) / 4;
226 maxfilesperproc = (maxfiles * 9) / 10;
227 }
228 EVENTHANDLER_INVOKE(maxsockets_change);
229 } else
230 error = EINVAL;
231 }
232 return (error);
233 }
234
235 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
236 &maxsockets, 0, sysctl_maxsockets, "IU",
237 "Maximum number of sockets avaliable");
238
239 /*
240 * Initialise maxsockets. This SYSINIT must be run after
241 * tunable_mbinit().
242 */
243 static void
244 init_maxsockets(void *ignored)
245 {
246
247 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
248 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
249 }
250 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
251
252 /*
253 * Socket operation routines. These routines are called by the routines in
254 * sys_socket.c or from a system process, and implement the semantics of
255 * socket operations by switching out to the protocol specific routines.
256 */
257
258 /*
259 * Get a socket structure from our zone, and initialize it. Note that it
260 * would probably be better to allocate socket and PCB at the same time, but
261 * I'm not convinced that all the protocols can be easily modified to do
262 * this.
263 *
264 * soalloc() returns a socket with a ref count of 0.
265 */
266 static struct socket *
267 soalloc(struct vnet *vnet)
268 {
269 struct socket *so;
270
271 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
272 if (so == NULL)
273 return (NULL);
274 #ifdef MAC
275 if (mac_socket_init(so, M_NOWAIT) != 0) {
276 uma_zfree(socket_zone, so);
277 return (NULL);
278 }
279 #endif
280 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
281 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
282 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
283 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
284 TAILQ_INIT(&so->so_aiojobq);
285 mtx_lock(&so_global_mtx);
286 so->so_gencnt = ++so_gencnt;
287 ++numopensockets;
288 #ifdef VIMAGE
289 vnet->vnet_sockcnt++;
290 so->so_vnet = vnet;
291 #endif
292 mtx_unlock(&so_global_mtx);
293 return (so);
294 }
295
296 /*
297 * Free the storage associated with a socket at the socket layer, tear down
298 * locks, labels, etc. All protocol state is assumed already to have been
299 * torn down (and possibly never set up) by the caller.
300 */
301 static void
302 sodealloc(struct socket *so)
303 {
304
305 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
306 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
307
308 mtx_lock(&so_global_mtx);
309 so->so_gencnt = ++so_gencnt;
310 --numopensockets; /* Could be below, but faster here. */
311 #ifdef VIMAGE
312 so->so_vnet->vnet_sockcnt--;
313 #endif
314 mtx_unlock(&so_global_mtx);
315 if (so->so_rcv.sb_hiwat)
316 (void)chgsbsize(so->so_cred->cr_uidinfo,
317 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
318 if (so->so_snd.sb_hiwat)
319 (void)chgsbsize(so->so_cred->cr_uidinfo,
320 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
321 #ifdef INET
322 /* remove acccept filter if one is present. */
323 if (so->so_accf != NULL)
324 do_setopt_accept_filter(so, NULL);
325 #endif
326 #ifdef MAC
327 mac_socket_destroy(so);
328 #endif
329 crfree(so->so_cred);
330 sx_destroy(&so->so_snd.sb_sx);
331 sx_destroy(&so->so_rcv.sb_sx);
332 SOCKBUF_LOCK_DESTROY(&so->so_snd);
333 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
334 uma_zfree(socket_zone, so);
335 }
336
337 /*
338 * socreate returns a socket with a ref count of 1. The socket should be
339 * closed with soclose().
340 */
341 int
342 socreate(int dom, struct socket **aso, int type, int proto,
343 struct ucred *cred, struct thread *td)
344 {
345 struct protosw *prp;
346 struct socket *so;
347 int error;
348
349 if (proto)
350 prp = pffindproto(dom, proto, type);
351 else
352 prp = pffindtype(dom, type);
353
354 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
355 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
356 return (EPROTONOSUPPORT);
357
358 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
359 return (EPROTONOSUPPORT);
360
361 if (prp->pr_type != type)
362 return (EPROTOTYPE);
363 so = soalloc(CRED_TO_VNET(cred));
364 if (so == NULL)
365 return (ENOBUFS);
366
367 TAILQ_INIT(&so->so_incomp);
368 TAILQ_INIT(&so->so_comp);
369 so->so_type = type;
370 so->so_cred = crhold(cred);
371 if ((prp->pr_domain->dom_family == PF_INET) ||
372 (prp->pr_domain->dom_family == PF_ROUTE))
373 so->so_fibnum = td->td_proc->p_fibnum;
374 else
375 so->so_fibnum = 0;
376 so->so_proto = prp;
377 #ifdef MAC
378 mac_socket_create(cred, so);
379 #endif
380 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
381 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
382 so->so_count = 1;
383 /*
384 * Auto-sizing of socket buffers is managed by the protocols and
385 * the appropriate flags must be set in the pru_attach function.
386 */
387 CURVNET_SET(so->so_vnet);
388 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
389 CURVNET_RESTORE();
390 if (error) {
391 KASSERT(so->so_count == 1, ("socreate: so_count %d",
392 so->so_count));
393 so->so_count = 0;
394 sodealloc(so);
395 return (error);
396 }
397 *aso = so;
398 return (0);
399 }
400
401 #ifdef REGRESSION
402 static int regression_sonewconn_earlytest = 1;
403 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
404 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
405 #endif
406
407 /*
408 * When an attempt at a new connection is noted on a socket which accepts
409 * connections, sonewconn is called. If the connection is possible (subject
410 * to space constraints, etc.) then we allocate a new structure, propoerly
411 * linked into the data structure of the original socket, and return this.
412 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
413 *
414 * Note: the ref count on the socket is 0 on return.
415 */
416 struct socket *
417 sonewconn(struct socket *head, int connstatus)
418 {
419 struct socket *so;
420 int over;
421
422 ACCEPT_LOCK();
423 over = (head->so_qlen > 3 * head->so_qlimit / 2);
424 ACCEPT_UNLOCK();
425 #ifdef REGRESSION
426 if (regression_sonewconn_earlytest && over)
427 #else
428 if (over)
429 #endif
430 return (NULL);
431 VNET_ASSERT(head->so_vnet);
432 so = soalloc(head->so_vnet);
433 if (so == NULL)
434 return (NULL);
435 if ((head->so_options & SO_ACCEPTFILTER) != 0)
436 connstatus = 0;
437 so->so_head = head;
438 so->so_type = head->so_type;
439 so->so_options = head->so_options &~ SO_ACCEPTCONN;
440 so->so_linger = head->so_linger;
441 so->so_state = head->so_state | SS_NOFDREF;
442 so->so_fibnum = head->so_fibnum;
443 so->so_proto = head->so_proto;
444 so->so_cred = crhold(head->so_cred);
445 #ifdef MAC
446 mac_socket_newconn(head, so);
447 #endif
448 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
449 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
450 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
451 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
452 sodealloc(so);
453 return (NULL);
454 }
455 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
456 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
457 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
458 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
459 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
460 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
461 so->so_state |= connstatus;
462 ACCEPT_LOCK();
463 if (connstatus) {
464 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
465 so->so_qstate |= SQ_COMP;
466 head->so_qlen++;
467 } else {
468 /*
469 * Keep removing sockets from the head until there's room for
470 * us to insert on the tail. In pre-locking revisions, this
471 * was a simple if(), but as we could be racing with other
472 * threads and soabort() requires dropping locks, we must
473 * loop waiting for the condition to be true.
474 */
475 while (head->so_incqlen > head->so_qlimit) {
476 struct socket *sp;
477 sp = TAILQ_FIRST(&head->so_incomp);
478 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
479 head->so_incqlen--;
480 sp->so_qstate &= ~SQ_INCOMP;
481 sp->so_head = NULL;
482 ACCEPT_UNLOCK();
483 soabort(sp);
484 ACCEPT_LOCK();
485 }
486 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
487 so->so_qstate |= SQ_INCOMP;
488 head->so_incqlen++;
489 }
490 ACCEPT_UNLOCK();
491 if (connstatus) {
492 sorwakeup(head);
493 wakeup_one(&head->so_timeo);
494 }
495 return (so);
496 }
497
498 int
499 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
500 {
501 int error;
502
503 CURVNET_SET(so->so_vnet);
504 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
505 CURVNET_RESTORE();
506 return error;
507 }
508
509 /*
510 * solisten() transitions a socket from a non-listening state to a listening
511 * state, but can also be used to update the listen queue depth on an
512 * existing listen socket. The protocol will call back into the sockets
513 * layer using solisten_proto_check() and solisten_proto() to check and set
514 * socket-layer listen state. Call backs are used so that the protocol can
515 * acquire both protocol and socket layer locks in whatever order is required
516 * by the protocol.
517 *
518 * Protocol implementors are advised to hold the socket lock across the
519 * socket-layer test and set to avoid races at the socket layer.
520 */
521 int
522 solisten(struct socket *so, int backlog, struct thread *td)
523 {
524
525 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
526 }
527
528 int
529 solisten_proto_check(struct socket *so)
530 {
531
532 SOCK_LOCK_ASSERT(so);
533
534 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
535 SS_ISDISCONNECTING))
536 return (EINVAL);
537 return (0);
538 }
539
540 void
541 solisten_proto(struct socket *so, int backlog)
542 {
543
544 SOCK_LOCK_ASSERT(so);
545
546 if (backlog < 0 || backlog > somaxconn)
547 backlog = somaxconn;
548 so->so_qlimit = backlog;
549 so->so_options |= SO_ACCEPTCONN;
550 }
551
552 /*
553 * Attempt to free a socket. This should really be sotryfree().
554 *
555 * sofree() will succeed if:
556 *
557 * - There are no outstanding file descriptor references or related consumers
558 * (so_count == 0).
559 *
560 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
561 *
562 * - The protocol does not have an outstanding strong reference on the socket
563 * (SS_PROTOREF).
564 *
565 * - The socket is not in a completed connection queue, so a process has been
566 * notified that it is present. If it is removed, the user process may
567 * block in accept() despite select() saying the socket was ready.
568 *
569 * Otherwise, it will quietly abort so that a future call to sofree(), when
570 * conditions are right, can succeed.
571 */
572 void
573 sofree(struct socket *so)
574 {
575 struct protosw *pr = so->so_proto;
576 struct socket *head;
577
578 ACCEPT_LOCK_ASSERT();
579 SOCK_LOCK_ASSERT(so);
580
581 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
582 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
583 SOCK_UNLOCK(so);
584 ACCEPT_UNLOCK();
585 return;
586 }
587
588 head = so->so_head;
589 if (head != NULL) {
590 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
591 (so->so_qstate & SQ_INCOMP) != 0,
592 ("sofree: so_head != NULL, but neither SQ_COMP nor "
593 "SQ_INCOMP"));
594 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
595 (so->so_qstate & SQ_INCOMP) == 0,
596 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
597 TAILQ_REMOVE(&head->so_incomp, so, so_list);
598 head->so_incqlen--;
599 so->so_qstate &= ~SQ_INCOMP;
600 so->so_head = NULL;
601 }
602 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
603 (so->so_qstate & SQ_INCOMP) == 0,
604 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
605 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
606 if (so->so_options & SO_ACCEPTCONN) {
607 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
608 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
609 }
610 SOCK_UNLOCK(so);
611 ACCEPT_UNLOCK();
612
613 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
614 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
615 if (pr->pr_usrreqs->pru_detach != NULL)
616 (*pr->pr_usrreqs->pru_detach)(so);
617
618 /*
619 * From this point on, we assume that no other references to this
620 * socket exist anywhere else in the stack. Therefore, no locks need
621 * to be acquired or held.
622 *
623 * We used to do a lot of socket buffer and socket locking here, as
624 * well as invoke sorflush() and perform wakeups. The direct call to
625 * dom_dispose() and sbrelease_internal() are an inlining of what was
626 * necessary from sorflush().
627 *
628 * Notice that the socket buffer and kqueue state are torn down
629 * before calling pru_detach. This means that protocols shold not
630 * assume they can perform socket wakeups, etc, in their detach code.
631 */
632 sbdestroy(&so->so_snd, so);
633 sbdestroy(&so->so_rcv, so);
634 knlist_destroy(&so->so_rcv.sb_sel.si_note);
635 knlist_destroy(&so->so_snd.sb_sel.si_note);
636 sodealloc(so);
637 }
638
639 /*
640 * Close a socket on last file table reference removal. Initiate disconnect
641 * if connected. Free socket when disconnect complete.
642 *
643 * This function will sorele() the socket. Note that soclose() may be called
644 * prior to the ref count reaching zero. The actual socket structure will
645 * not be freed until the ref count reaches zero.
646 */
647 int
648 soclose(struct socket *so)
649 {
650 int error = 0;
651
652 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
653
654 CURVNET_SET(so->so_vnet);
655 funsetown(&so->so_sigio);
656 if (so->so_state & SS_ISCONNECTED) {
657 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
658 error = sodisconnect(so);
659 if (error) {
660 if (error == ENOTCONN)
661 error = 0;
662 goto drop;
663 }
664 }
665 if (so->so_options & SO_LINGER) {
666 if ((so->so_state & SS_ISDISCONNECTING) &&
667 (so->so_state & SS_NBIO))
668 goto drop;
669 while (so->so_state & SS_ISCONNECTED) {
670 error = tsleep(&so->so_timeo,
671 PSOCK | PCATCH, "soclos", so->so_linger * hz);
672 if (error)
673 break;
674 }
675 }
676 }
677
678 drop:
679 if (so->so_proto->pr_usrreqs->pru_close != NULL)
680 (*so->so_proto->pr_usrreqs->pru_close)(so);
681 if (so->so_options & SO_ACCEPTCONN) {
682 struct socket *sp;
683 ACCEPT_LOCK();
684 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
685 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
686 so->so_incqlen--;
687 sp->so_qstate &= ~SQ_INCOMP;
688 sp->so_head = NULL;
689 ACCEPT_UNLOCK();
690 soabort(sp);
691 ACCEPT_LOCK();
692 }
693 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
694 TAILQ_REMOVE(&so->so_comp, sp, so_list);
695 so->so_qlen--;
696 sp->so_qstate &= ~SQ_COMP;
697 sp->so_head = NULL;
698 ACCEPT_UNLOCK();
699 soabort(sp);
700 ACCEPT_LOCK();
701 }
702 ACCEPT_UNLOCK();
703 }
704 ACCEPT_LOCK();
705 SOCK_LOCK(so);
706 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
707 so->so_state |= SS_NOFDREF;
708 sorele(so);
709 CURVNET_RESTORE();
710 return (error);
711 }
712
713 /*
714 * soabort() is used to abruptly tear down a connection, such as when a
715 * resource limit is reached (listen queue depth exceeded), or if a listen
716 * socket is closed while there are sockets waiting to be accepted.
717 *
718 * This interface is tricky, because it is called on an unreferenced socket,
719 * and must be called only by a thread that has actually removed the socket
720 * from the listen queue it was on, or races with other threads are risked.
721 *
722 * This interface will call into the protocol code, so must not be called
723 * with any socket locks held. Protocols do call it while holding their own
724 * recursible protocol mutexes, but this is something that should be subject
725 * to review in the future.
726 */
727 void
728 soabort(struct socket *so)
729 {
730
731 /*
732 * In as much as is possible, assert that no references to this
733 * socket are held. This is not quite the same as asserting that the
734 * current thread is responsible for arranging for no references, but
735 * is as close as we can get for now.
736 */
737 KASSERT(so->so_count == 0, ("soabort: so_count"));
738 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
739 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
740 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
741 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
742
743 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
744 (*so->so_proto->pr_usrreqs->pru_abort)(so);
745 ACCEPT_LOCK();
746 SOCK_LOCK(so);
747 sofree(so);
748 }
749
750 int
751 soaccept(struct socket *so, struct sockaddr **nam)
752 {
753 int error;
754
755 SOCK_LOCK(so);
756 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
757 so->so_state &= ~SS_NOFDREF;
758 SOCK_UNLOCK(so);
759 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
760 return (error);
761 }
762
763 int
764 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
765 {
766 int error;
767
768 if (so->so_options & SO_ACCEPTCONN)
769 return (EOPNOTSUPP);
770
771 CURVNET_SET(so->so_vnet);
772 /*
773 * If protocol is connection-based, can only connect once.
774 * Otherwise, if connected, try to disconnect first. This allows
775 * user to disconnect by connecting to, e.g., a null address.
776 */
777 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
778 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
779 (error = sodisconnect(so)))) {
780 error = EISCONN;
781 } else {
782 /*
783 * Prevent accumulated error from previous connection from
784 * biting us.
785 */
786 so->so_error = 0;
787 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
788 }
789 CURVNET_RESTORE();
790
791 return (error);
792 }
793
794 int
795 soconnect2(struct socket *so1, struct socket *so2)
796 {
797
798 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
799 }
800
801 int
802 sodisconnect(struct socket *so)
803 {
804 int error;
805
806 if ((so->so_state & SS_ISCONNECTED) == 0)
807 return (ENOTCONN);
808 if (so->so_state & SS_ISDISCONNECTING)
809 return (EALREADY);
810 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
811 return (error);
812 }
813
814 #ifdef ZERO_COPY_SOCKETS
815 struct so_zerocopy_stats{
816 int size_ok;
817 int align_ok;
818 int found_ifp;
819 };
820 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
821 #include <netinet/in.h>
822 #include <net/route.h>
823 #include <netinet/in_pcb.h>
824 #include <vm/vm.h>
825 #include <vm/vm_page.h>
826 #include <vm/vm_object.h>
827
828 /*
829 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
830 * sosend_dgram() and sosend_generic() use m_uiotombuf().
831 *
832 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
833 * all of the data referenced by the uio. If desired, it uses zero-copy.
834 * *space will be updated to reflect data copied in.
835 *
836 * NB: If atomic I/O is requested, the caller must already have checked that
837 * space can hold resid bytes.
838 *
839 * NB: In the event of an error, the caller may need to free the partial
840 * chain pointed to by *mpp. The contents of both *uio and *space may be
841 * modified even in the case of an error.
842 */
843 static int
844 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
845 int flags)
846 {
847 struct mbuf *m, **mp, *top;
848 long len, resid;
849 int error;
850 #ifdef ZERO_COPY_SOCKETS
851 int cow_send;
852 #endif
853
854 *retmp = top = NULL;
855 mp = ⊤
856 len = 0;
857 resid = uio->uio_resid;
858 error = 0;
859 do {
860 #ifdef ZERO_COPY_SOCKETS
861 cow_send = 0;
862 #endif /* ZERO_COPY_SOCKETS */
863 if (resid >= MINCLSIZE) {
864 #ifdef ZERO_COPY_SOCKETS
865 if (top == NULL) {
866 m = m_gethdr(M_WAITOK, MT_DATA);
867 m->m_pkthdr.len = 0;
868 m->m_pkthdr.rcvif = NULL;
869 } else
870 m = m_get(M_WAITOK, MT_DATA);
871 if (so_zero_copy_send &&
872 resid>=PAGE_SIZE &&
873 *space>=PAGE_SIZE &&
874 uio->uio_iov->iov_len>=PAGE_SIZE) {
875 so_zerocp_stats.size_ok++;
876 so_zerocp_stats.align_ok++;
877 cow_send = socow_setup(m, uio);
878 len = cow_send;
879 }
880 if (!cow_send) {
881 m_clget(m, M_WAITOK);
882 len = min(min(MCLBYTES, resid), *space);
883 }
884 #else /* ZERO_COPY_SOCKETS */
885 if (top == NULL) {
886 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
887 m->m_pkthdr.len = 0;
888 m->m_pkthdr.rcvif = NULL;
889 } else
890 m = m_getcl(M_WAIT, MT_DATA, 0);
891 len = min(min(MCLBYTES, resid), *space);
892 #endif /* ZERO_COPY_SOCKETS */
893 } else {
894 if (top == NULL) {
895 m = m_gethdr(M_WAIT, MT_DATA);
896 m->m_pkthdr.len = 0;
897 m->m_pkthdr.rcvif = NULL;
898
899 len = min(min(MHLEN, resid), *space);
900 /*
901 * For datagram protocols, leave room
902 * for protocol headers in first mbuf.
903 */
904 if (atomic && m && len < MHLEN)
905 MH_ALIGN(m, len);
906 } else {
907 m = m_get(M_WAIT, MT_DATA);
908 len = min(min(MLEN, resid), *space);
909 }
910 }
911 if (m == NULL) {
912 error = ENOBUFS;
913 goto out;
914 }
915
916 *space -= len;
917 #ifdef ZERO_COPY_SOCKETS
918 if (cow_send)
919 error = 0;
920 else
921 #endif /* ZERO_COPY_SOCKETS */
922 error = uiomove(mtod(m, void *), (int)len, uio);
923 resid = uio->uio_resid;
924 m->m_len = len;
925 *mp = m;
926 top->m_pkthdr.len += len;
927 if (error)
928 goto out;
929 mp = &m->m_next;
930 if (resid <= 0) {
931 if (flags & MSG_EOR)
932 top->m_flags |= M_EOR;
933 break;
934 }
935 } while (*space > 0 && atomic);
936 out:
937 *retmp = top;
938 return (error);
939 }
940 #endif /*ZERO_COPY_SOCKETS*/
941
942 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
943
944 int
945 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
946 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
947 {
948 long space, resid;
949 int clen = 0, error, dontroute;
950 #ifdef ZERO_COPY_SOCKETS
951 int atomic = sosendallatonce(so) || top;
952 #endif
953
954 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
955 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
956 ("sodgram_send: !PR_ATOMIC"));
957
958 if (uio != NULL)
959 resid = uio->uio_resid;
960 else
961 resid = top->m_pkthdr.len;
962 /*
963 * In theory resid should be unsigned. However, space must be
964 * signed, as it might be less than 0 if we over-committed, and we
965 * must use a signed comparison of space and resid. On the other
966 * hand, a negative resid causes us to loop sending 0-length
967 * segments to the protocol.
968 */
969 if (resid < 0) {
970 error = EINVAL;
971 goto out;
972 }
973
974 dontroute =
975 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
976 if (td != NULL)
977 td->td_ru.ru_msgsnd++;
978 if (control != NULL)
979 clen = control->m_len;
980
981 SOCKBUF_LOCK(&so->so_snd);
982 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
983 SOCKBUF_UNLOCK(&so->so_snd);
984 error = EPIPE;
985 goto out;
986 }
987 if (so->so_error) {
988 error = so->so_error;
989 so->so_error = 0;
990 SOCKBUF_UNLOCK(&so->so_snd);
991 goto out;
992 }
993 if ((so->so_state & SS_ISCONNECTED) == 0) {
994 /*
995 * `sendto' and `sendmsg' is allowed on a connection-based
996 * socket if it supports implied connect. Return ENOTCONN if
997 * not connected and no address is supplied.
998 */
999 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1000 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1001 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1002 !(resid == 0 && clen != 0)) {
1003 SOCKBUF_UNLOCK(&so->so_snd);
1004 error = ENOTCONN;
1005 goto out;
1006 }
1007 } else if (addr == NULL) {
1008 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1009 error = ENOTCONN;
1010 else
1011 error = EDESTADDRREQ;
1012 SOCKBUF_UNLOCK(&so->so_snd);
1013 goto out;
1014 }
1015 }
1016
1017 /*
1018 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1019 * problem and need fixing.
1020 */
1021 space = sbspace(&so->so_snd);
1022 if (flags & MSG_OOB)
1023 space += 1024;
1024 space -= clen;
1025 SOCKBUF_UNLOCK(&so->so_snd);
1026 if (resid > space) {
1027 error = EMSGSIZE;
1028 goto out;
1029 }
1030 if (uio == NULL) {
1031 resid = 0;
1032 if (flags & MSG_EOR)
1033 top->m_flags |= M_EOR;
1034 } else {
1035 #ifdef ZERO_COPY_SOCKETS
1036 error = sosend_copyin(uio, &top, atomic, &space, flags);
1037 if (error)
1038 goto out;
1039 #else
1040 /*
1041 * Copy the data from userland into a mbuf chain.
1042 * If no data is to be copied in, a single empty mbuf
1043 * is returned.
1044 */
1045 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1046 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1047 if (top == NULL) {
1048 error = EFAULT; /* only possible error */
1049 goto out;
1050 }
1051 space -= resid - uio->uio_resid;
1052 #endif
1053 resid = uio->uio_resid;
1054 }
1055 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1056 /*
1057 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1058 * than with.
1059 */
1060 if (dontroute) {
1061 SOCK_LOCK(so);
1062 so->so_options |= SO_DONTROUTE;
1063 SOCK_UNLOCK(so);
1064 }
1065 /*
1066 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1067 * of date. We could have recieved a reset packet in an interrupt or
1068 * maybe we slept while doing page faults in uiomove() etc. We could
1069 * probably recheck again inside the locking protection here, but
1070 * there are probably other places that this also happens. We must
1071 * rethink this.
1072 */
1073 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1074 (flags & MSG_OOB) ? PRUS_OOB :
1075 /*
1076 * If the user set MSG_EOF, the protocol understands this flag and
1077 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1078 */
1079 ((flags & MSG_EOF) &&
1080 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1081 (resid <= 0)) ?
1082 PRUS_EOF :
1083 /* If there is more to send set PRUS_MORETOCOME */
1084 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1085 top, addr, control, td);
1086 if (dontroute) {
1087 SOCK_LOCK(so);
1088 so->so_options &= ~SO_DONTROUTE;
1089 SOCK_UNLOCK(so);
1090 }
1091 clen = 0;
1092 control = NULL;
1093 top = NULL;
1094 out:
1095 if (top != NULL)
1096 m_freem(top);
1097 if (control != NULL)
1098 m_freem(control);
1099 return (error);
1100 }
1101
1102 /*
1103 * Send on a socket. If send must go all at once and message is larger than
1104 * send buffering, then hard error. Lock against other senders. If must go
1105 * all at once and not enough room now, then inform user that this would
1106 * block and do nothing. Otherwise, if nonblocking, send as much as
1107 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1108 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1109 * in mbuf chain must be small enough to send all at once.
1110 *
1111 * Returns nonzero on error, timeout or signal; callers must check for short
1112 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1113 * on return.
1114 */
1115 int
1116 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1117 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1118 {
1119 long space, resid;
1120 int clen = 0, error, dontroute;
1121 int atomic = sosendallatonce(so) || top;
1122
1123 if (uio != NULL)
1124 resid = uio->uio_resid;
1125 else
1126 resid = top->m_pkthdr.len;
1127 /*
1128 * In theory resid should be unsigned. However, space must be
1129 * signed, as it might be less than 0 if we over-committed, and we
1130 * must use a signed comparison of space and resid. On the other
1131 * hand, a negative resid causes us to loop sending 0-length
1132 * segments to the protocol.
1133 *
1134 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1135 * type sockets since that's an error.
1136 */
1137 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1138 error = EINVAL;
1139 goto out;
1140 }
1141
1142 dontroute =
1143 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1144 (so->so_proto->pr_flags & PR_ATOMIC);
1145 if (td != NULL)
1146 td->td_ru.ru_msgsnd++;
1147 if (control != NULL)
1148 clen = control->m_len;
1149
1150 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1151 if (error)
1152 goto out;
1153
1154 restart:
1155 do {
1156 SOCKBUF_LOCK(&so->so_snd);
1157 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1158 SOCKBUF_UNLOCK(&so->so_snd);
1159 error = EPIPE;
1160 goto release;
1161 }
1162 if (so->so_error) {
1163 error = so->so_error;
1164 so->so_error = 0;
1165 SOCKBUF_UNLOCK(&so->so_snd);
1166 goto release;
1167 }
1168 if ((so->so_state & SS_ISCONNECTED) == 0) {
1169 /*
1170 * `sendto' and `sendmsg' is allowed on a connection-
1171 * based socket if it supports implied connect.
1172 * Return ENOTCONN if not connected and no address is
1173 * supplied.
1174 */
1175 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1176 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1177 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1178 !(resid == 0 && clen != 0)) {
1179 SOCKBUF_UNLOCK(&so->so_snd);
1180 error = ENOTCONN;
1181 goto release;
1182 }
1183 } else if (addr == NULL) {
1184 SOCKBUF_UNLOCK(&so->so_snd);
1185 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1186 error = ENOTCONN;
1187 else
1188 error = EDESTADDRREQ;
1189 goto release;
1190 }
1191 }
1192 space = sbspace(&so->so_snd);
1193 if (flags & MSG_OOB)
1194 space += 1024;
1195 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1196 clen > so->so_snd.sb_hiwat) {
1197 SOCKBUF_UNLOCK(&so->so_snd);
1198 error = EMSGSIZE;
1199 goto release;
1200 }
1201 if (space < resid + clen &&
1202 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1203 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1204 SOCKBUF_UNLOCK(&so->so_snd);
1205 error = EWOULDBLOCK;
1206 goto release;
1207 }
1208 error = sbwait(&so->so_snd);
1209 SOCKBUF_UNLOCK(&so->so_snd);
1210 if (error)
1211 goto release;
1212 goto restart;
1213 }
1214 SOCKBUF_UNLOCK(&so->so_snd);
1215 space -= clen;
1216 do {
1217 if (uio == NULL) {
1218 resid = 0;
1219 if (flags & MSG_EOR)
1220 top->m_flags |= M_EOR;
1221 } else {
1222 #ifdef ZERO_COPY_SOCKETS
1223 error = sosend_copyin(uio, &top, atomic,
1224 &space, flags);
1225 if (error != 0)
1226 goto release;
1227 #else
1228 /*
1229 * Copy the data from userland into a mbuf
1230 * chain. If no data is to be copied in,
1231 * a single empty mbuf is returned.
1232 */
1233 top = m_uiotombuf(uio, M_WAITOK, space,
1234 (atomic ? max_hdr : 0),
1235 (atomic ? M_PKTHDR : 0) |
1236 ((flags & MSG_EOR) ? M_EOR : 0));
1237 if (top == NULL) {
1238 error = EFAULT; /* only possible error */
1239 goto release;
1240 }
1241 space -= resid - uio->uio_resid;
1242 #endif
1243 resid = uio->uio_resid;
1244 }
1245 if (dontroute) {
1246 SOCK_LOCK(so);
1247 so->so_options |= SO_DONTROUTE;
1248 SOCK_UNLOCK(so);
1249 }
1250 /*
1251 * XXX all the SBS_CANTSENDMORE checks previously
1252 * done could be out of date. We could have recieved
1253 * a reset packet in an interrupt or maybe we slept
1254 * while doing page faults in uiomove() etc. We
1255 * could probably recheck again inside the locking
1256 * protection here, but there are probably other
1257 * places that this also happens. We must rethink
1258 * this.
1259 */
1260 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1261 (flags & MSG_OOB) ? PRUS_OOB :
1262 /*
1263 * If the user set MSG_EOF, the protocol understands
1264 * this flag and nothing left to send then use
1265 * PRU_SEND_EOF instead of PRU_SEND.
1266 */
1267 ((flags & MSG_EOF) &&
1268 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1269 (resid <= 0)) ?
1270 PRUS_EOF :
1271 /* If there is more to send set PRUS_MORETOCOME. */
1272 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1273 top, addr, control, td);
1274 if (dontroute) {
1275 SOCK_LOCK(so);
1276 so->so_options &= ~SO_DONTROUTE;
1277 SOCK_UNLOCK(so);
1278 }
1279 clen = 0;
1280 control = NULL;
1281 top = NULL;
1282 if (error)
1283 goto release;
1284 } while (resid && space > 0);
1285 } while (resid);
1286
1287 release:
1288 sbunlock(&so->so_snd);
1289 out:
1290 if (top != NULL)
1291 m_freem(top);
1292 if (control != NULL)
1293 m_freem(control);
1294 return (error);
1295 }
1296
1297 int
1298 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1299 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1300 {
1301 int error;
1302
1303 CURVNET_SET(so->so_vnet);
1304 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1305 control, flags, td);
1306 CURVNET_RESTORE();
1307 return (error);
1308 }
1309
1310 /*
1311 * The part of soreceive() that implements reading non-inline out-of-band
1312 * data from a socket. For more complete comments, see soreceive(), from
1313 * which this code originated.
1314 *
1315 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1316 * unable to return an mbuf chain to the caller.
1317 */
1318 static int
1319 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1320 {
1321 struct protosw *pr = so->so_proto;
1322 struct mbuf *m;
1323 int error;
1324
1325 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1326
1327 m = m_get(M_WAIT, MT_DATA);
1328 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1329 if (error)
1330 goto bad;
1331 do {
1332 #ifdef ZERO_COPY_SOCKETS
1333 if (so_zero_copy_receive) {
1334 int disposable;
1335
1336 if ((m->m_flags & M_EXT)
1337 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1338 disposable = 1;
1339 else
1340 disposable = 0;
1341
1342 error = uiomoveco(mtod(m, void *),
1343 min(uio->uio_resid, m->m_len),
1344 uio, disposable);
1345 } else
1346 #endif /* ZERO_COPY_SOCKETS */
1347 error = uiomove(mtod(m, void *),
1348 (int) min(uio->uio_resid, m->m_len), uio);
1349 m = m_free(m);
1350 } while (uio->uio_resid && error == 0 && m);
1351 bad:
1352 if (m != NULL)
1353 m_freem(m);
1354 return (error);
1355 }
1356
1357 /*
1358 * Following replacement or removal of the first mbuf on the first mbuf chain
1359 * of a socket buffer, push necessary state changes back into the socket
1360 * buffer so that other consumers see the values consistently. 'nextrecord'
1361 * is the callers locally stored value of the original value of
1362 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1363 * NOTE: 'nextrecord' may be NULL.
1364 */
1365 static __inline void
1366 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1367 {
1368
1369 SOCKBUF_LOCK_ASSERT(sb);
1370 /*
1371 * First, update for the new value of nextrecord. If necessary, make
1372 * it the first record.
1373 */
1374 if (sb->sb_mb != NULL)
1375 sb->sb_mb->m_nextpkt = nextrecord;
1376 else
1377 sb->sb_mb = nextrecord;
1378
1379 /*
1380 * Now update any dependent socket buffer fields to reflect the new
1381 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1382 * addition of a second clause that takes care of the case where
1383 * sb_mb has been updated, but remains the last record.
1384 */
1385 if (sb->sb_mb == NULL) {
1386 sb->sb_mbtail = NULL;
1387 sb->sb_lastrecord = NULL;
1388 } else if (sb->sb_mb->m_nextpkt == NULL)
1389 sb->sb_lastrecord = sb->sb_mb;
1390 }
1391
1392
1393 /*
1394 * Implement receive operations on a socket. We depend on the way that
1395 * records are added to the sockbuf by sbappend. In particular, each record
1396 * (mbufs linked through m_next) must begin with an address if the protocol
1397 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1398 * data, and then zero or more mbufs of data. In order to allow parallelism
1399 * between network receive and copying to user space, as well as avoid
1400 * sleeping with a mutex held, we release the socket buffer mutex during the
1401 * user space copy. Although the sockbuf is locked, new data may still be
1402 * appended, and thus we must maintain consistency of the sockbuf during that
1403 * time.
1404 *
1405 * The caller may receive the data as a single mbuf chain by supplying an
1406 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1407 * the count in uio_resid.
1408 */
1409 int
1410 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1411 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1412 {
1413 struct mbuf *m, **mp;
1414 int flags, len, error, offset;
1415 struct protosw *pr = so->so_proto;
1416 struct mbuf *nextrecord;
1417 int moff, type = 0;
1418 int orig_resid = uio->uio_resid;
1419
1420 mp = mp0;
1421 if (psa != NULL)
1422 *psa = NULL;
1423 if (controlp != NULL)
1424 *controlp = NULL;
1425 if (flagsp != NULL)
1426 flags = *flagsp &~ MSG_EOR;
1427 else
1428 flags = 0;
1429 if (flags & MSG_OOB)
1430 return (soreceive_rcvoob(so, uio, flags));
1431 if (mp != NULL)
1432 *mp = NULL;
1433 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1434 && uio->uio_resid)
1435 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1436
1437 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1438 if (error)
1439 return (error);
1440
1441 restart:
1442 SOCKBUF_LOCK(&so->so_rcv);
1443 m = so->so_rcv.sb_mb;
1444 /*
1445 * If we have less data than requested, block awaiting more (subject
1446 * to any timeout) if:
1447 * 1. the current count is less than the low water mark, or
1448 * 2. MSG_WAITALL is set, and it is possible to do the entire
1449 * receive operation at once if we block (resid <= hiwat).
1450 * 3. MSG_DONTWAIT is not set
1451 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1452 * we have to do the receive in sections, and thus risk returning a
1453 * short count if a timeout or signal occurs after we start.
1454 */
1455 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1456 so->so_rcv.sb_cc < uio->uio_resid) &&
1457 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1458 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1459 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1460 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1461 ("receive: m == %p so->so_rcv.sb_cc == %u",
1462 m, so->so_rcv.sb_cc));
1463 if (so->so_error) {
1464 if (m != NULL)
1465 goto dontblock;
1466 error = so->so_error;
1467 if ((flags & MSG_PEEK) == 0)
1468 so->so_error = 0;
1469 SOCKBUF_UNLOCK(&so->so_rcv);
1470 goto release;
1471 }
1472 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1473 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1474 if (m == NULL) {
1475 SOCKBUF_UNLOCK(&so->so_rcv);
1476 goto release;
1477 } else
1478 goto dontblock;
1479 }
1480 for (; m != NULL; m = m->m_next)
1481 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1482 m = so->so_rcv.sb_mb;
1483 goto dontblock;
1484 }
1485 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1486 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1487 SOCKBUF_UNLOCK(&so->so_rcv);
1488 error = ENOTCONN;
1489 goto release;
1490 }
1491 if (uio->uio_resid == 0) {
1492 SOCKBUF_UNLOCK(&so->so_rcv);
1493 goto release;
1494 }
1495 if ((so->so_state & SS_NBIO) ||
1496 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1497 SOCKBUF_UNLOCK(&so->so_rcv);
1498 error = EWOULDBLOCK;
1499 goto release;
1500 }
1501 SBLASTRECORDCHK(&so->so_rcv);
1502 SBLASTMBUFCHK(&so->so_rcv);
1503 error = sbwait(&so->so_rcv);
1504 SOCKBUF_UNLOCK(&so->so_rcv);
1505 if (error)
1506 goto release;
1507 goto restart;
1508 }
1509 dontblock:
1510 /*
1511 * From this point onward, we maintain 'nextrecord' as a cache of the
1512 * pointer to the next record in the socket buffer. We must keep the
1513 * various socket buffer pointers and local stack versions of the
1514 * pointers in sync, pushing out modifications before dropping the
1515 * socket buffer mutex, and re-reading them when picking it up.
1516 *
1517 * Otherwise, we will race with the network stack appending new data
1518 * or records onto the socket buffer by using inconsistent/stale
1519 * versions of the field, possibly resulting in socket buffer
1520 * corruption.
1521 *
1522 * By holding the high-level sblock(), we prevent simultaneous
1523 * readers from pulling off the front of the socket buffer.
1524 */
1525 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1526 if (uio->uio_td)
1527 uio->uio_td->td_ru.ru_msgrcv++;
1528 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1529 SBLASTRECORDCHK(&so->so_rcv);
1530 SBLASTMBUFCHK(&so->so_rcv);
1531 nextrecord = m->m_nextpkt;
1532 if (pr->pr_flags & PR_ADDR) {
1533 KASSERT(m->m_type == MT_SONAME,
1534 ("m->m_type == %d", m->m_type));
1535 orig_resid = 0;
1536 if (psa != NULL)
1537 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1538 M_NOWAIT);
1539 if (flags & MSG_PEEK) {
1540 m = m->m_next;
1541 } else {
1542 sbfree(&so->so_rcv, m);
1543 so->so_rcv.sb_mb = m_free(m);
1544 m = so->so_rcv.sb_mb;
1545 sockbuf_pushsync(&so->so_rcv, nextrecord);
1546 }
1547 }
1548
1549 /*
1550 * Process one or more MT_CONTROL mbufs present before any data mbufs
1551 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1552 * just copy the data; if !MSG_PEEK, we call into the protocol to
1553 * perform externalization (or freeing if controlp == NULL).
1554 */
1555 if (m != NULL && m->m_type == MT_CONTROL) {
1556 struct mbuf *cm = NULL, *cmn;
1557 struct mbuf **cme = &cm;
1558
1559 do {
1560 if (flags & MSG_PEEK) {
1561 if (controlp != NULL) {
1562 *controlp = m_copy(m, 0, m->m_len);
1563 controlp = &(*controlp)->m_next;
1564 }
1565 m = m->m_next;
1566 } else {
1567 sbfree(&so->so_rcv, m);
1568 so->so_rcv.sb_mb = m->m_next;
1569 m->m_next = NULL;
1570 *cme = m;
1571 cme = &(*cme)->m_next;
1572 m = so->so_rcv.sb_mb;
1573 }
1574 } while (m != NULL && m->m_type == MT_CONTROL);
1575 if ((flags & MSG_PEEK) == 0)
1576 sockbuf_pushsync(&so->so_rcv, nextrecord);
1577 while (cm != NULL) {
1578 cmn = cm->m_next;
1579 cm->m_next = NULL;
1580 if (pr->pr_domain->dom_externalize != NULL) {
1581 SOCKBUF_UNLOCK(&so->so_rcv);
1582 error = (*pr->pr_domain->dom_externalize)
1583 (cm, controlp);
1584 SOCKBUF_LOCK(&so->so_rcv);
1585 } else if (controlp != NULL)
1586 *controlp = cm;
1587 else
1588 m_freem(cm);
1589 if (controlp != NULL) {
1590 orig_resid = 0;
1591 while (*controlp != NULL)
1592 controlp = &(*controlp)->m_next;
1593 }
1594 cm = cmn;
1595 }
1596 if (m != NULL)
1597 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1598 else
1599 nextrecord = so->so_rcv.sb_mb;
1600 orig_resid = 0;
1601 }
1602 if (m != NULL) {
1603 if ((flags & MSG_PEEK) == 0) {
1604 KASSERT(m->m_nextpkt == nextrecord,
1605 ("soreceive: post-control, nextrecord !sync"));
1606 if (nextrecord == NULL) {
1607 KASSERT(so->so_rcv.sb_mb == m,
1608 ("soreceive: post-control, sb_mb!=m"));
1609 KASSERT(so->so_rcv.sb_lastrecord == m,
1610 ("soreceive: post-control, lastrecord!=m"));
1611 }
1612 }
1613 type = m->m_type;
1614 if (type == MT_OOBDATA)
1615 flags |= MSG_OOB;
1616 } else {
1617 if ((flags & MSG_PEEK) == 0) {
1618 KASSERT(so->so_rcv.sb_mb == nextrecord,
1619 ("soreceive: sb_mb != nextrecord"));
1620 if (so->so_rcv.sb_mb == NULL) {
1621 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1622 ("soreceive: sb_lastercord != NULL"));
1623 }
1624 }
1625 }
1626 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1627 SBLASTRECORDCHK(&so->so_rcv);
1628 SBLASTMBUFCHK(&so->so_rcv);
1629
1630 /*
1631 * Now continue to read any data mbufs off of the head of the socket
1632 * buffer until the read request is satisfied. Note that 'type' is
1633 * used to store the type of any mbuf reads that have happened so far
1634 * such that soreceive() can stop reading if the type changes, which
1635 * causes soreceive() to return only one of regular data and inline
1636 * out-of-band data in a single socket receive operation.
1637 */
1638 moff = 0;
1639 offset = 0;
1640 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1641 /*
1642 * If the type of mbuf has changed since the last mbuf
1643 * examined ('type'), end the receive operation.
1644 */
1645 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1646 if (m->m_type == MT_OOBDATA) {
1647 if (type != MT_OOBDATA)
1648 break;
1649 } else if (type == MT_OOBDATA)
1650 break;
1651 else
1652 KASSERT(m->m_type == MT_DATA,
1653 ("m->m_type == %d", m->m_type));
1654 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1655 len = uio->uio_resid;
1656 if (so->so_oobmark && len > so->so_oobmark - offset)
1657 len = so->so_oobmark - offset;
1658 if (len > m->m_len - moff)
1659 len = m->m_len - moff;
1660 /*
1661 * If mp is set, just pass back the mbufs. Otherwise copy
1662 * them out via the uio, then free. Sockbuf must be
1663 * consistent here (points to current mbuf, it points to next
1664 * record) when we drop priority; we must note any additions
1665 * to the sockbuf when we block interrupts again.
1666 */
1667 if (mp == NULL) {
1668 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1669 SBLASTRECORDCHK(&so->so_rcv);
1670 SBLASTMBUFCHK(&so->so_rcv);
1671 SOCKBUF_UNLOCK(&so->so_rcv);
1672 #ifdef ZERO_COPY_SOCKETS
1673 if (so_zero_copy_receive) {
1674 int disposable;
1675
1676 if ((m->m_flags & M_EXT)
1677 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1678 disposable = 1;
1679 else
1680 disposable = 0;
1681
1682 error = uiomoveco(mtod(m, char *) + moff,
1683 (int)len, uio,
1684 disposable);
1685 } else
1686 #endif /* ZERO_COPY_SOCKETS */
1687 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1688 SOCKBUF_LOCK(&so->so_rcv);
1689 if (error) {
1690 /*
1691 * The MT_SONAME mbuf has already been removed
1692 * from the record, so it is necessary to
1693 * remove the data mbufs, if any, to preserve
1694 * the invariant in the case of PR_ADDR that
1695 * requires MT_SONAME mbufs at the head of
1696 * each record.
1697 */
1698 if (m && pr->pr_flags & PR_ATOMIC &&
1699 ((flags & MSG_PEEK) == 0))
1700 (void)sbdroprecord_locked(&so->so_rcv);
1701 SOCKBUF_UNLOCK(&so->so_rcv);
1702 goto release;
1703 }
1704 } else
1705 uio->uio_resid -= len;
1706 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1707 if (len == m->m_len - moff) {
1708 if (m->m_flags & M_EOR)
1709 flags |= MSG_EOR;
1710 if (flags & MSG_PEEK) {
1711 m = m->m_next;
1712 moff = 0;
1713 } else {
1714 nextrecord = m->m_nextpkt;
1715 sbfree(&so->so_rcv, m);
1716 if (mp != NULL) {
1717 *mp = m;
1718 mp = &m->m_next;
1719 so->so_rcv.sb_mb = m = m->m_next;
1720 *mp = NULL;
1721 } else {
1722 so->so_rcv.sb_mb = m_free(m);
1723 m = so->so_rcv.sb_mb;
1724 }
1725 sockbuf_pushsync(&so->so_rcv, nextrecord);
1726 SBLASTRECORDCHK(&so->so_rcv);
1727 SBLASTMBUFCHK(&so->so_rcv);
1728 }
1729 } else {
1730 if (flags & MSG_PEEK)
1731 moff += len;
1732 else {
1733 if (mp != NULL) {
1734 int copy_flag;
1735
1736 if (flags & MSG_DONTWAIT)
1737 copy_flag = M_DONTWAIT;
1738 else
1739 copy_flag = M_WAIT;
1740 if (copy_flag == M_WAIT)
1741 SOCKBUF_UNLOCK(&so->so_rcv);
1742 *mp = m_copym(m, 0, len, copy_flag);
1743 if (copy_flag == M_WAIT)
1744 SOCKBUF_LOCK(&so->so_rcv);
1745 if (*mp == NULL) {
1746 /*
1747 * m_copym() couldn't
1748 * allocate an mbuf. Adjust
1749 * uio_resid back (it was
1750 * adjusted down by len
1751 * bytes, which we didn't end
1752 * up "copying" over).
1753 */
1754 uio->uio_resid += len;
1755 break;
1756 }
1757 }
1758 m->m_data += len;
1759 m->m_len -= len;
1760 so->so_rcv.sb_cc -= len;
1761 }
1762 }
1763 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1764 if (so->so_oobmark) {
1765 if ((flags & MSG_PEEK) == 0) {
1766 so->so_oobmark -= len;
1767 if (so->so_oobmark == 0) {
1768 so->so_rcv.sb_state |= SBS_RCVATMARK;
1769 break;
1770 }
1771 } else {
1772 offset += len;
1773 if (offset == so->so_oobmark)
1774 break;
1775 }
1776 }
1777 if (flags & MSG_EOR)
1778 break;
1779 /*
1780 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1781 * must not quit until "uio->uio_resid == 0" or an error
1782 * termination. If a signal/timeout occurs, return with a
1783 * short count but without error. Keep sockbuf locked
1784 * against other readers.
1785 */
1786 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1787 !sosendallatonce(so) && nextrecord == NULL) {
1788 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1789 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1790 break;
1791 /*
1792 * Notify the protocol that some data has been
1793 * drained before blocking.
1794 */
1795 if (pr->pr_flags & PR_WANTRCVD) {
1796 SOCKBUF_UNLOCK(&so->so_rcv);
1797 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1798 SOCKBUF_LOCK(&so->so_rcv);
1799 }
1800 SBLASTRECORDCHK(&so->so_rcv);
1801 SBLASTMBUFCHK(&so->so_rcv);
1802 error = sbwait(&so->so_rcv);
1803 if (error) {
1804 SOCKBUF_UNLOCK(&so->so_rcv);
1805 goto release;
1806 }
1807 m = so->so_rcv.sb_mb;
1808 if (m != NULL)
1809 nextrecord = m->m_nextpkt;
1810 }
1811 }
1812
1813 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1814 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1815 flags |= MSG_TRUNC;
1816 if ((flags & MSG_PEEK) == 0)
1817 (void) sbdroprecord_locked(&so->so_rcv);
1818 }
1819 if ((flags & MSG_PEEK) == 0) {
1820 if (m == NULL) {
1821 /*
1822 * First part is an inline SB_EMPTY_FIXUP(). Second
1823 * part makes sure sb_lastrecord is up-to-date if
1824 * there is still data in the socket buffer.
1825 */
1826 so->so_rcv.sb_mb = nextrecord;
1827 if (so->so_rcv.sb_mb == NULL) {
1828 so->so_rcv.sb_mbtail = NULL;
1829 so->so_rcv.sb_lastrecord = NULL;
1830 } else if (nextrecord->m_nextpkt == NULL)
1831 so->so_rcv.sb_lastrecord = nextrecord;
1832 }
1833 SBLASTRECORDCHK(&so->so_rcv);
1834 SBLASTMBUFCHK(&so->so_rcv);
1835 /*
1836 * If soreceive() is being done from the socket callback,
1837 * then don't need to generate ACK to peer to update window,
1838 * since ACK will be generated on return to TCP.
1839 */
1840 if (!(flags & MSG_SOCALLBCK) &&
1841 (pr->pr_flags & PR_WANTRCVD)) {
1842 SOCKBUF_UNLOCK(&so->so_rcv);
1843 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1844 SOCKBUF_LOCK(&so->so_rcv);
1845 }
1846 }
1847 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1848 if (orig_resid == uio->uio_resid && orig_resid &&
1849 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1850 SOCKBUF_UNLOCK(&so->so_rcv);
1851 goto restart;
1852 }
1853 SOCKBUF_UNLOCK(&so->so_rcv);
1854
1855 if (flagsp != NULL)
1856 *flagsp |= flags;
1857 release:
1858 sbunlock(&so->so_rcv);
1859 return (error);
1860 }
1861
1862 /*
1863 * Optimized version of soreceive() for stream (TCP) sockets.
1864 */
1865 int
1866 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1867 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1868 {
1869 int len = 0, error = 0, flags, oresid;
1870 struct sockbuf *sb;
1871 struct mbuf *m, *n = NULL;
1872
1873 /* We only do stream sockets. */
1874 if (so->so_type != SOCK_STREAM)
1875 return (EINVAL);
1876 if (psa != NULL)
1877 *psa = NULL;
1878 if (controlp != NULL)
1879 return (EINVAL);
1880 if (flagsp != NULL)
1881 flags = *flagsp &~ MSG_EOR;
1882 else
1883 flags = 0;
1884 if (flags & MSG_OOB)
1885 return (soreceive_rcvoob(so, uio, flags));
1886 if (mp0 != NULL)
1887 *mp0 = NULL;
1888
1889 sb = &so->so_rcv;
1890
1891 /* Prevent other readers from entering the socket. */
1892 error = sblock(sb, SBLOCKWAIT(flags));
1893 if (error)
1894 goto out;
1895 SOCKBUF_LOCK(sb);
1896
1897 /* Easy one, no space to copyout anything. */
1898 if (uio->uio_resid == 0) {
1899 error = EINVAL;
1900 goto out;
1901 }
1902 oresid = uio->uio_resid;
1903
1904 /* We will never ever get anything unless we are connected. */
1905 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1906 /* When disconnecting there may be still some data left. */
1907 if (sb->sb_cc > 0)
1908 goto deliver;
1909 if (!(so->so_state & SS_ISDISCONNECTED))
1910 error = ENOTCONN;
1911 goto out;
1912 }
1913
1914 /* Socket buffer is empty and we shall not block. */
1915 if (sb->sb_cc == 0 &&
1916 ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1917 error = EAGAIN;
1918 goto out;
1919 }
1920
1921 restart:
1922 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1923
1924 /* Abort if socket has reported problems. */
1925 if (so->so_error) {
1926 if (sb->sb_cc > 0)
1927 goto deliver;
1928 if (oresid > uio->uio_resid)
1929 goto out;
1930 error = so->so_error;
1931 if (!(flags & MSG_PEEK))
1932 so->so_error = 0;
1933 goto out;
1934 }
1935
1936 /* Door is closed. Deliver what is left, if any. */
1937 if (sb->sb_state & SBS_CANTRCVMORE) {
1938 if (sb->sb_cc > 0)
1939 goto deliver;
1940 else
1941 goto out;
1942 }
1943
1944 /* Socket buffer got some data that we shall deliver now. */
1945 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1946 ((sb->sb_flags & SS_NBIO) ||
1947 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1948 sb->sb_cc >= sb->sb_lowat ||
1949 sb->sb_cc >= uio->uio_resid ||
1950 sb->sb_cc >= sb->sb_hiwat) ) {
1951 goto deliver;
1952 }
1953
1954 /* On MSG_WAITALL we must wait until all data or error arrives. */
1955 if ((flags & MSG_WAITALL) &&
1956 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
1957 goto deliver;
1958
1959 /*
1960 * Wait and block until (more) data comes in.
1961 * NB: Drops the sockbuf lock during wait.
1962 */
1963 error = sbwait(sb);
1964 if (error)
1965 goto out;
1966 goto restart;
1967
1968 deliver:
1969 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1970 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1971 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1972
1973 /* Statistics. */
1974 if (uio->uio_td)
1975 uio->uio_td->td_ru.ru_msgrcv++;
1976
1977 /* Fill uio until full or current end of socket buffer is reached. */
1978 len = min(uio->uio_resid, sb->sb_cc);
1979 if (mp0 != NULL) {
1980 /* Dequeue as many mbufs as possible. */
1981 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1982 for (*mp0 = m = sb->sb_mb;
1983 m != NULL && m->m_len <= len;
1984 m = m->m_next) {
1985 len -= m->m_len;
1986 uio->uio_resid -= m->m_len;
1987 sbfree(sb, m);
1988 n = m;
1989 }
1990 sb->sb_mb = m;
1991 if (sb->sb_mb == NULL)
1992 SB_EMPTY_FIXUP(sb);
1993 n->m_next = NULL;
1994 }
1995 /* Copy the remainder. */
1996 if (len > 0) {
1997 KASSERT(sb->sb_mb != NULL,
1998 ("%s: len > 0 && sb->sb_mb empty", __func__));
1999
2000 m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2001 if (m == NULL)
2002 len = 0; /* Don't flush data from sockbuf. */
2003 else
2004 uio->uio_resid -= m->m_len;
2005 if (*mp0 != NULL)
2006 n->m_next = m;
2007 else
2008 *mp0 = m;
2009 if (*mp0 == NULL) {
2010 error = ENOBUFS;
2011 goto out;
2012 }
2013 }
2014 } else {
2015 /* NB: Must unlock socket buffer as uiomove may sleep. */
2016 SOCKBUF_UNLOCK(sb);
2017 error = m_mbuftouio(uio, sb->sb_mb, len);
2018 SOCKBUF_LOCK(sb);
2019 if (error)
2020 goto out;
2021 }
2022 SBLASTRECORDCHK(sb);
2023 SBLASTMBUFCHK(sb);
2024
2025 /*
2026 * Remove the delivered data from the socket buffer unless we
2027 * were only peeking.
2028 */
2029 if (!(flags & MSG_PEEK)) {
2030 if (len > 0)
2031 sbdrop_locked(sb, len);
2032
2033 /* Notify protocol that we drained some data. */
2034 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2035 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2036 !(flags & MSG_SOCALLBCK))) {
2037 SOCKBUF_UNLOCK(sb);
2038 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2039 SOCKBUF_LOCK(sb);
2040 }
2041 }
2042
2043 /*
2044 * For MSG_WAITALL we may have to loop again and wait for
2045 * more data to come in.
2046 */
2047 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2048 goto restart;
2049 out:
2050 SOCKBUF_LOCK_ASSERT(sb);
2051 SBLASTRECORDCHK(sb);
2052 SBLASTMBUFCHK(sb);
2053 SOCKBUF_UNLOCK(sb);
2054 sbunlock(sb);
2055 return (error);
2056 }
2057
2058 /*
2059 * Optimized version of soreceive() for simple datagram cases from userspace.
2060 * Unlike in the stream case, we're able to drop a datagram if copyout()
2061 * fails, and because we handle datagrams atomically, we don't need to use a
2062 * sleep lock to prevent I/O interlacing.
2063 */
2064 int
2065 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2066 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2067 {
2068 struct mbuf *m, *m2;
2069 int flags, len, error;
2070 struct protosw *pr = so->so_proto;
2071 struct mbuf *nextrecord;
2072
2073 if (psa != NULL)
2074 *psa = NULL;
2075 if (controlp != NULL)
2076 *controlp = NULL;
2077 if (flagsp != NULL)
2078 flags = *flagsp &~ MSG_EOR;
2079 else
2080 flags = 0;
2081
2082 /*
2083 * For any complicated cases, fall back to the full
2084 * soreceive_generic().
2085 */
2086 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2087 return (soreceive_generic(so, psa, uio, mp0, controlp,
2088 flagsp));
2089
2090 /*
2091 * Enforce restrictions on use.
2092 */
2093 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2094 ("soreceive_dgram: wantrcvd"));
2095 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2096 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2097 ("soreceive_dgram: SBS_RCVATMARK"));
2098 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2099 ("soreceive_dgram: P_CONNREQUIRED"));
2100
2101 /*
2102 * Loop blocking while waiting for a datagram.
2103 */
2104 SOCKBUF_LOCK(&so->so_rcv);
2105 while ((m = so->so_rcv.sb_mb) == NULL) {
2106 KASSERT(so->so_rcv.sb_cc == 0,
2107 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2108 so->so_rcv.sb_cc));
2109 if (so->so_error) {
2110 error = so->so_error;
2111 so->so_error = 0;
2112 SOCKBUF_UNLOCK(&so->so_rcv);
2113 return (error);
2114 }
2115 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2116 uio->uio_resid == 0) {
2117 SOCKBUF_UNLOCK(&so->so_rcv);
2118 return (0);
2119 }
2120 if ((so->so_state & SS_NBIO) ||
2121 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2122 SOCKBUF_UNLOCK(&so->so_rcv);
2123 return (EWOULDBLOCK);
2124 }
2125 SBLASTRECORDCHK(&so->so_rcv);
2126 SBLASTMBUFCHK(&so->so_rcv);
2127 error = sbwait(&so->so_rcv);
2128 if (error) {
2129 SOCKBUF_UNLOCK(&so->so_rcv);
2130 return (error);
2131 }
2132 }
2133 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2134
2135 if (uio->uio_td)
2136 uio->uio_td->td_ru.ru_msgrcv++;
2137 SBLASTRECORDCHK(&so->so_rcv);
2138 SBLASTMBUFCHK(&so->so_rcv);
2139 nextrecord = m->m_nextpkt;
2140 if (nextrecord == NULL) {
2141 KASSERT(so->so_rcv.sb_lastrecord == m,
2142 ("soreceive_dgram: lastrecord != m"));
2143 }
2144
2145 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2146 ("soreceive_dgram: m_nextpkt != nextrecord"));
2147
2148 /*
2149 * Pull 'm' and its chain off the front of the packet queue.
2150 */
2151 so->so_rcv.sb_mb = NULL;
2152 sockbuf_pushsync(&so->so_rcv, nextrecord);
2153
2154 /*
2155 * Walk 'm's chain and free that many bytes from the socket buffer.
2156 */
2157 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2158 sbfree(&so->so_rcv, m2);
2159
2160 /*
2161 * Do a few last checks before we let go of the lock.
2162 */
2163 SBLASTRECORDCHK(&so->so_rcv);
2164 SBLASTMBUFCHK(&so->so_rcv);
2165 SOCKBUF_UNLOCK(&so->so_rcv);
2166
2167 if (pr->pr_flags & PR_ADDR) {
2168 KASSERT(m->m_type == MT_SONAME,
2169 ("m->m_type == %d", m->m_type));
2170 if (psa != NULL)
2171 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2172 M_NOWAIT);
2173 m = m_free(m);
2174 }
2175 if (m == NULL) {
2176 /* XXXRW: Can this happen? */
2177 return (0);
2178 }
2179
2180 /*
2181 * Packet to copyout() is now in 'm' and it is disconnected from the
2182 * queue.
2183 *
2184 * Process one or more MT_CONTROL mbufs present before any data mbufs
2185 * in the first mbuf chain on the socket buffer. We call into the
2186 * protocol to perform externalization (or freeing if controlp ==
2187 * NULL).
2188 */
2189 if (m->m_type == MT_CONTROL) {
2190 struct mbuf *cm = NULL, *cmn;
2191 struct mbuf **cme = &cm;
2192
2193 do {
2194 m2 = m->m_next;
2195 m->m_next = NULL;
2196 *cme = m;
2197 cme = &(*cme)->m_next;
2198 m = m2;
2199 } while (m != NULL && m->m_type == MT_CONTROL);
2200 while (cm != NULL) {
2201 cmn = cm->m_next;
2202 cm->m_next = NULL;
2203 if (pr->pr_domain->dom_externalize != NULL) {
2204 error = (*pr->pr_domain->dom_externalize)
2205 (cm, controlp);
2206 } else if (controlp != NULL)
2207 *controlp = cm;
2208 else
2209 m_freem(cm);
2210 if (controlp != NULL) {
2211 while (*controlp != NULL)
2212 controlp = &(*controlp)->m_next;
2213 }
2214 cm = cmn;
2215 }
2216 }
2217 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2218
2219 while (m != NULL && uio->uio_resid > 0) {
2220 len = uio->uio_resid;
2221 if (len > m->m_len)
2222 len = m->m_len;
2223 error = uiomove(mtod(m, char *), (int)len, uio);
2224 if (error) {
2225 m_freem(m);
2226 return (error);
2227 }
2228 m = m_free(m);
2229 }
2230 if (m != NULL)
2231 flags |= MSG_TRUNC;
2232 m_freem(m);
2233 if (flagsp != NULL)
2234 *flagsp |= flags;
2235 return (0);
2236 }
2237
2238 int
2239 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2240 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2241 {
2242
2243 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2244 controlp, flagsp));
2245 }
2246
2247 int
2248 soshutdown(struct socket *so, int how)
2249 {
2250 struct protosw *pr = so->so_proto;
2251 int error;
2252
2253 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2254 return (EINVAL);
2255 if (pr->pr_usrreqs->pru_flush != NULL) {
2256 (*pr->pr_usrreqs->pru_flush)(so, how);
2257 }
2258 if (how != SHUT_WR)
2259 sorflush(so);
2260 if (how != SHUT_RD) {
2261 CURVNET_SET(so->so_vnet);
2262 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2263 CURVNET_RESTORE();
2264 return (error);
2265 }
2266 return (0);
2267 }
2268
2269 void
2270 sorflush(struct socket *so)
2271 {
2272 struct sockbuf *sb = &so->so_rcv;
2273 struct protosw *pr = so->so_proto;
2274 struct sockbuf asb;
2275
2276 /*
2277 * In order to avoid calling dom_dispose with the socket buffer mutex
2278 * held, and in order to generally avoid holding the lock for a long
2279 * time, we make a copy of the socket buffer and clear the original
2280 * (except locks, state). The new socket buffer copy won't have
2281 * initialized locks so we can only call routines that won't use or
2282 * assert those locks.
2283 *
2284 * Dislodge threads currently blocked in receive and wait to acquire
2285 * a lock against other simultaneous readers before clearing the
2286 * socket buffer. Don't let our acquire be interrupted by a signal
2287 * despite any existing socket disposition on interruptable waiting.
2288 */
2289 CURVNET_SET(so->so_vnet);
2290 socantrcvmore(so);
2291 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2292
2293 /*
2294 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2295 * and mutex data unchanged.
2296 */
2297 SOCKBUF_LOCK(sb);
2298 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2299 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2300 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2301 bzero(&sb->sb_startzero,
2302 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2303 SOCKBUF_UNLOCK(sb);
2304 sbunlock(sb);
2305
2306 /*
2307 * Dispose of special rights and flush the socket buffer. Don't call
2308 * any unsafe routines (that rely on locks being initialized) on asb.
2309 */
2310 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2311 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2312 sbrelease_internal(&asb, so);
2313 CURVNET_RESTORE();
2314 }
2315
2316 /*
2317 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2318 * additional variant to handle the case where the option value needs to be
2319 * some kind of integer, but not a specific size. In addition to their use
2320 * here, these functions are also called by the protocol-level pr_ctloutput()
2321 * routines.
2322 */
2323 int
2324 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2325 {
2326 size_t valsize;
2327
2328 /*
2329 * If the user gives us more than we wanted, we ignore it, but if we
2330 * don't get the minimum length the caller wants, we return EINVAL.
2331 * On success, sopt->sopt_valsize is set to however much we actually
2332 * retrieved.
2333 */
2334 if ((valsize = sopt->sopt_valsize) < minlen)
2335 return EINVAL;
2336 if (valsize > len)
2337 sopt->sopt_valsize = valsize = len;
2338
2339 if (sopt->sopt_td != NULL)
2340 return (copyin(sopt->sopt_val, buf, valsize));
2341
2342 bcopy(sopt->sopt_val, buf, valsize);
2343 return (0);
2344 }
2345
2346 /*
2347 * Kernel version of setsockopt(2).
2348 *
2349 * XXX: optlen is size_t, not socklen_t
2350 */
2351 int
2352 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2353 size_t optlen)
2354 {
2355 struct sockopt sopt;
2356
2357 sopt.sopt_level = level;
2358 sopt.sopt_name = optname;
2359 sopt.sopt_dir = SOPT_SET;
2360 sopt.sopt_val = optval;
2361 sopt.sopt_valsize = optlen;
2362 sopt.sopt_td = NULL;
2363 return (sosetopt(so, &sopt));
2364 }
2365
2366 int
2367 sosetopt(struct socket *so, struct sockopt *sopt)
2368 {
2369 int error, optval;
2370 struct linger l;
2371 struct timeval tv;
2372 u_long val;
2373 #ifdef MAC
2374 struct mac extmac;
2375 #endif
2376
2377 error = 0;
2378 if (sopt->sopt_level != SOL_SOCKET) {
2379 if (so->so_proto && so->so_proto->pr_ctloutput)
2380 return ((*so->so_proto->pr_ctloutput)
2381 (so, sopt));
2382 error = ENOPROTOOPT;
2383 } else {
2384 switch (sopt->sopt_name) {
2385 #ifdef INET
2386 case SO_ACCEPTFILTER:
2387 error = do_setopt_accept_filter(so, sopt);
2388 if (error)
2389 goto bad;
2390 break;
2391 #endif
2392 case SO_LINGER:
2393 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2394 if (error)
2395 goto bad;
2396
2397 SOCK_LOCK(so);
2398 so->so_linger = l.l_linger;
2399 if (l.l_onoff)
2400 so->so_options |= SO_LINGER;
2401 else
2402 so->so_options &= ~SO_LINGER;
2403 SOCK_UNLOCK(so);
2404 break;
2405
2406 case SO_DEBUG:
2407 case SO_KEEPALIVE:
2408 case SO_DONTROUTE:
2409 case SO_USELOOPBACK:
2410 case SO_BROADCAST:
2411 case SO_REUSEADDR:
2412 case SO_REUSEPORT:
2413 case SO_OOBINLINE:
2414 case SO_TIMESTAMP:
2415 case SO_BINTIME:
2416 case SO_NOSIGPIPE:
2417 case SO_NO_DDP:
2418 case SO_NO_OFFLOAD:
2419 error = sooptcopyin(sopt, &optval, sizeof optval,
2420 sizeof optval);
2421 if (error)
2422 goto bad;
2423 SOCK_LOCK(so);
2424 if (optval)
2425 so->so_options |= sopt->sopt_name;
2426 else
2427 so->so_options &= ~sopt->sopt_name;
2428 SOCK_UNLOCK(so);
2429 break;
2430
2431 case SO_SETFIB:
2432 error = sooptcopyin(sopt, &optval, sizeof optval,
2433 sizeof optval);
2434 if (optval < 1 || optval > rt_numfibs) {
2435 error = EINVAL;
2436 goto bad;
2437 }
2438 if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2439 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2440 so->so_fibnum = optval;
2441 /* Note: ignore error */
2442 if (so->so_proto && so->so_proto->pr_ctloutput)
2443 (*so->so_proto->pr_ctloutput)(so, sopt);
2444 } else {
2445 so->so_fibnum = 0;
2446 }
2447 break;
2448 case SO_SNDBUF:
2449 case SO_RCVBUF:
2450 case SO_SNDLOWAT:
2451 case SO_RCVLOWAT:
2452 error = sooptcopyin(sopt, &optval, sizeof optval,
2453 sizeof optval);
2454 if (error)
2455 goto bad;
2456
2457 /*
2458 * Values < 1 make no sense for any of these options,
2459 * so disallow them.
2460 */
2461 if (optval < 1) {
2462 error = EINVAL;
2463 goto bad;
2464 }
2465
2466 switch (sopt->sopt_name) {
2467 case SO_SNDBUF:
2468 case SO_RCVBUF:
2469 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2470 &so->so_snd : &so->so_rcv, (u_long)optval,
2471 so, curthread) == 0) {
2472 error = ENOBUFS;
2473 goto bad;
2474 }
2475 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2476 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2477 break;
2478
2479 /*
2480 * Make sure the low-water is never greater than the
2481 * high-water.
2482 */
2483 case SO_SNDLOWAT:
2484 SOCKBUF_LOCK(&so->so_snd);
2485 so->so_snd.sb_lowat =
2486 (optval > so->so_snd.sb_hiwat) ?
2487 so->so_snd.sb_hiwat : optval;
2488 SOCKBUF_UNLOCK(&so->so_snd);
2489 break;
2490 case SO_RCVLOWAT:
2491 SOCKBUF_LOCK(&so->so_rcv);
2492 so->so_rcv.sb_lowat =
2493 (optval > so->so_rcv.sb_hiwat) ?
2494 so->so_rcv.sb_hiwat : optval;
2495 SOCKBUF_UNLOCK(&so->so_rcv);
2496 break;
2497 }
2498 break;
2499
2500 case SO_SNDTIMEO:
2501 case SO_RCVTIMEO:
2502 #ifdef COMPAT_FREEBSD32
2503 if (SV_CURPROC_FLAG(SV_ILP32)) {
2504 struct timeval32 tv32;
2505
2506 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2507 sizeof tv32);
2508 CP(tv32, tv, tv_sec);
2509 CP(tv32, tv, tv_usec);
2510 } else
2511 #endif
2512 error = sooptcopyin(sopt, &tv, sizeof tv,
2513 sizeof tv);
2514 if (error)
2515 goto bad;
2516
2517 /* assert(hz > 0); */
2518 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2519 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2520 error = EDOM;
2521 goto bad;
2522 }
2523 /* assert(tick > 0); */
2524 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2525 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2526 if (val > INT_MAX) {
2527 error = EDOM;
2528 goto bad;
2529 }
2530 if (val == 0 && tv.tv_usec != 0)
2531 val = 1;
2532
2533 switch (sopt->sopt_name) {
2534 case SO_SNDTIMEO:
2535 so->so_snd.sb_timeo = val;
2536 break;
2537 case SO_RCVTIMEO:
2538 so->so_rcv.sb_timeo = val;
2539 break;
2540 }
2541 break;
2542
2543 case SO_LABEL:
2544 #ifdef MAC
2545 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2546 sizeof extmac);
2547 if (error)
2548 goto bad;
2549 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2550 so, &extmac);
2551 #else
2552 error = EOPNOTSUPP;
2553 #endif
2554 break;
2555
2556 default:
2557 error = ENOPROTOOPT;
2558 break;
2559 }
2560 if (error == 0 && so->so_proto != NULL &&
2561 so->so_proto->pr_ctloutput != NULL) {
2562 (void) ((*so->so_proto->pr_ctloutput)
2563 (so, sopt));
2564 }
2565 }
2566 bad:
2567 return (error);
2568 }
2569
2570 /*
2571 * Helper routine for getsockopt.
2572 */
2573 int
2574 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2575 {
2576 int error;
2577 size_t valsize;
2578
2579 error = 0;
2580
2581 /*
2582 * Documented get behavior is that we always return a value, possibly
2583 * truncated to fit in the user's buffer. Traditional behavior is
2584 * that we always tell the user precisely how much we copied, rather
2585 * than something useful like the total amount we had available for
2586 * her. Note that this interface is not idempotent; the entire
2587 * answer must generated ahead of time.
2588 */
2589 valsize = min(len, sopt->sopt_valsize);
2590 sopt->sopt_valsize = valsize;
2591 if (sopt->sopt_val != NULL) {
2592 if (sopt->sopt_td != NULL)
2593 error = copyout(buf, sopt->sopt_val, valsize);
2594 else
2595 bcopy(buf, sopt->sopt_val, valsize);
2596 }
2597 return (error);
2598 }
2599
2600 int
2601 sogetopt(struct socket *so, struct sockopt *sopt)
2602 {
2603 int error, optval;
2604 struct linger l;
2605 struct timeval tv;
2606 #ifdef MAC
2607 struct mac extmac;
2608 #endif
2609
2610 error = 0;
2611 if (sopt->sopt_level != SOL_SOCKET) {
2612 if (so->so_proto && so->so_proto->pr_ctloutput) {
2613 return ((*so->so_proto->pr_ctloutput)
2614 (so, sopt));
2615 } else
2616 return (ENOPROTOOPT);
2617 } else {
2618 switch (sopt->sopt_name) {
2619 #ifdef INET
2620 case SO_ACCEPTFILTER:
2621 error = do_getopt_accept_filter(so, sopt);
2622 break;
2623 #endif
2624 case SO_LINGER:
2625 SOCK_LOCK(so);
2626 l.l_onoff = so->so_options & SO_LINGER;
2627 l.l_linger = so->so_linger;
2628 SOCK_UNLOCK(so);
2629 error = sooptcopyout(sopt, &l, sizeof l);
2630 break;
2631
2632 case SO_USELOOPBACK:
2633 case SO_DONTROUTE:
2634 case SO_DEBUG:
2635 case SO_KEEPALIVE:
2636 case SO_REUSEADDR:
2637 case SO_REUSEPORT:
2638 case SO_BROADCAST:
2639 case SO_OOBINLINE:
2640 case SO_ACCEPTCONN:
2641 case SO_TIMESTAMP:
2642 case SO_BINTIME:
2643 case SO_NOSIGPIPE:
2644 optval = so->so_options & sopt->sopt_name;
2645 integer:
2646 error = sooptcopyout(sopt, &optval, sizeof optval);
2647 break;
2648
2649 case SO_TYPE:
2650 optval = so->so_type;
2651 goto integer;
2652
2653 case SO_ERROR:
2654 SOCK_LOCK(so);
2655 optval = so->so_error;
2656 so->so_error = 0;
2657 SOCK_UNLOCK(so);
2658 goto integer;
2659
2660 case SO_SNDBUF:
2661 optval = so->so_snd.sb_hiwat;
2662 goto integer;
2663
2664 case SO_RCVBUF:
2665 optval = so->so_rcv.sb_hiwat;
2666 goto integer;
2667
2668 case SO_SNDLOWAT:
2669 optval = so->so_snd.sb_lowat;
2670 goto integer;
2671
2672 case SO_RCVLOWAT:
2673 optval = so->so_rcv.sb_lowat;
2674 goto integer;
2675
2676 case SO_SNDTIMEO:
2677 case SO_RCVTIMEO:
2678 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2679 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2680
2681 tv.tv_sec = optval / hz;
2682 tv.tv_usec = (optval % hz) * tick;
2683 #ifdef COMPAT_FREEBSD32
2684 if (SV_CURPROC_FLAG(SV_ILP32)) {
2685 struct timeval32 tv32;
2686
2687 CP(tv, tv32, tv_sec);
2688 CP(tv, tv32, tv_usec);
2689 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2690 } else
2691 #endif
2692 error = sooptcopyout(sopt, &tv, sizeof tv);
2693 break;
2694
2695 case SO_LABEL:
2696 #ifdef MAC
2697 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2698 sizeof(extmac));
2699 if (error)
2700 return (error);
2701 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2702 so, &extmac);
2703 if (error)
2704 return (error);
2705 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2706 #else
2707 error = EOPNOTSUPP;
2708 #endif
2709 break;
2710
2711 case SO_PEERLABEL:
2712 #ifdef MAC
2713 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2714 sizeof(extmac));
2715 if (error)
2716 return (error);
2717 error = mac_getsockopt_peerlabel(
2718 sopt->sopt_td->td_ucred, so, &extmac);
2719 if (error)
2720 return (error);
2721 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2722 #else
2723 error = EOPNOTSUPP;
2724 #endif
2725 break;
2726
2727 case SO_LISTENQLIMIT:
2728 optval = so->so_qlimit;
2729 goto integer;
2730
2731 case SO_LISTENQLEN:
2732 optval = so->so_qlen;
2733 goto integer;
2734
2735 case SO_LISTENINCQLEN:
2736 optval = so->so_incqlen;
2737 goto integer;
2738
2739 default:
2740 error = ENOPROTOOPT;
2741 break;
2742 }
2743 return (error);
2744 }
2745 }
2746
2747 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2748 int
2749 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2750 {
2751 struct mbuf *m, *m_prev;
2752 int sopt_size = sopt->sopt_valsize;
2753
2754 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2755 if (m == NULL)
2756 return ENOBUFS;
2757 if (sopt_size > MLEN) {
2758 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2759 if ((m->m_flags & M_EXT) == 0) {
2760 m_free(m);
2761 return ENOBUFS;
2762 }
2763 m->m_len = min(MCLBYTES, sopt_size);
2764 } else {
2765 m->m_len = min(MLEN, sopt_size);
2766 }
2767 sopt_size -= m->m_len;
2768 *mp = m;
2769 m_prev = m;
2770
2771 while (sopt_size) {
2772 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2773 if (m == NULL) {
2774 m_freem(*mp);
2775 return ENOBUFS;
2776 }
2777 if (sopt_size > MLEN) {
2778 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2779 M_DONTWAIT);
2780 if ((m->m_flags & M_EXT) == 0) {
2781 m_freem(m);
2782 m_freem(*mp);
2783 return ENOBUFS;
2784 }
2785 m->m_len = min(MCLBYTES, sopt_size);
2786 } else {
2787 m->m_len = min(MLEN, sopt_size);
2788 }
2789 sopt_size -= m->m_len;
2790 m_prev->m_next = m;
2791 m_prev = m;
2792 }
2793 return (0);
2794 }
2795
2796 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2797 int
2798 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2799 {
2800 struct mbuf *m0 = m;
2801
2802 if (sopt->sopt_val == NULL)
2803 return (0);
2804 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2805 if (sopt->sopt_td != NULL) {
2806 int error;
2807
2808 error = copyin(sopt->sopt_val, mtod(m, char *),
2809 m->m_len);
2810 if (error != 0) {
2811 m_freem(m0);
2812 return(error);
2813 }
2814 } else
2815 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2816 sopt->sopt_valsize -= m->m_len;
2817 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2818 m = m->m_next;
2819 }
2820 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2821 panic("ip6_sooptmcopyin");
2822 return (0);
2823 }
2824
2825 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2826 int
2827 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2828 {
2829 struct mbuf *m0 = m;
2830 size_t valsize = 0;
2831
2832 if (sopt->sopt_val == NULL)
2833 return (0);
2834 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2835 if (sopt->sopt_td != NULL) {
2836 int error;
2837
2838 error = copyout(mtod(m, char *), sopt->sopt_val,
2839 m->m_len);
2840 if (error != 0) {
2841 m_freem(m0);
2842 return(error);
2843 }
2844 } else
2845 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2846 sopt->sopt_valsize -= m->m_len;
2847 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2848 valsize += m->m_len;
2849 m = m->m_next;
2850 }
2851 if (m != NULL) {
2852 /* enough soopt buffer should be given from user-land */
2853 m_freem(m0);
2854 return(EINVAL);
2855 }
2856 sopt->sopt_valsize = valsize;
2857 return (0);
2858 }
2859
2860 /*
2861 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2862 * out-of-band data, which will then notify socket consumers.
2863 */
2864 void
2865 sohasoutofband(struct socket *so)
2866 {
2867
2868 if (so->so_sigio != NULL)
2869 pgsigio(&so->so_sigio, SIGURG, 0);
2870 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2871 }
2872
2873 int
2874 sopoll(struct socket *so, int events, struct ucred *active_cred,
2875 struct thread *td)
2876 {
2877
2878 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2879 td));
2880 }
2881
2882 int
2883 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2884 struct thread *td)
2885 {
2886 int revents = 0;
2887
2888 SOCKBUF_LOCK(&so->so_snd);
2889 SOCKBUF_LOCK(&so->so_rcv);
2890 if (events & (POLLIN | POLLRDNORM))
2891 if (soreadabledata(so))
2892 revents |= events & (POLLIN | POLLRDNORM);
2893
2894 if (events & (POLLOUT | POLLWRNORM))
2895 if (sowriteable(so))
2896 revents |= events & (POLLOUT | POLLWRNORM);
2897
2898 if (events & (POLLPRI | POLLRDBAND))
2899 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2900 revents |= events & (POLLPRI | POLLRDBAND);
2901
2902 if ((events & POLLINIGNEOF) == 0) {
2903 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2904 revents |= events & (POLLIN | POLLRDNORM);
2905 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2906 revents |= POLLHUP;
2907 }
2908 }
2909
2910 if (revents == 0) {
2911 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2912 selrecord(td, &so->so_rcv.sb_sel);
2913 so->so_rcv.sb_flags |= SB_SEL;
2914 }
2915
2916 if (events & (POLLOUT | POLLWRNORM)) {
2917 selrecord(td, &so->so_snd.sb_sel);
2918 so->so_snd.sb_flags |= SB_SEL;
2919 }
2920 }
2921
2922 SOCKBUF_UNLOCK(&so->so_rcv);
2923 SOCKBUF_UNLOCK(&so->so_snd);
2924 return (revents);
2925 }
2926
2927 int
2928 soo_kqfilter(struct file *fp, struct knote *kn)
2929 {
2930 struct socket *so = kn->kn_fp->f_data;
2931 struct sockbuf *sb;
2932
2933 switch (kn->kn_filter) {
2934 case EVFILT_READ:
2935 if (so->so_options & SO_ACCEPTCONN)
2936 kn->kn_fop = &solisten_filtops;
2937 else
2938 kn->kn_fop = &soread_filtops;
2939 sb = &so->so_rcv;
2940 break;
2941 case EVFILT_WRITE:
2942 kn->kn_fop = &sowrite_filtops;
2943 sb = &so->so_snd;
2944 break;
2945 default:
2946 return (EINVAL);
2947 }
2948
2949 SOCKBUF_LOCK(sb);
2950 knlist_add(&sb->sb_sel.si_note, kn, 1);
2951 sb->sb_flags |= SB_KNOTE;
2952 SOCKBUF_UNLOCK(sb);
2953 return (0);
2954 }
2955
2956 /*
2957 * Some routines that return EOPNOTSUPP for entry points that are not
2958 * supported by a protocol. Fill in as needed.
2959 */
2960 int
2961 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2962 {
2963
2964 return EOPNOTSUPP;
2965 }
2966
2967 int
2968 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2969 {
2970
2971 return EOPNOTSUPP;
2972 }
2973
2974 int
2975 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2976 {
2977
2978 return EOPNOTSUPP;
2979 }
2980
2981 int
2982 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2983 {
2984
2985 return EOPNOTSUPP;
2986 }
2987
2988 int
2989 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2990 {
2991
2992 return EOPNOTSUPP;
2993 }
2994
2995 int
2996 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2997 struct ifnet *ifp, struct thread *td)
2998 {
2999
3000 return EOPNOTSUPP;
3001 }
3002
3003 int
3004 pru_disconnect_notsupp(struct socket *so)
3005 {
3006
3007 return EOPNOTSUPP;
3008 }
3009
3010 int
3011 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3012 {
3013
3014 return EOPNOTSUPP;
3015 }
3016
3017 int
3018 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3019 {
3020
3021 return EOPNOTSUPP;
3022 }
3023
3024 int
3025 pru_rcvd_notsupp(struct socket *so, int flags)
3026 {
3027
3028 return EOPNOTSUPP;
3029 }
3030
3031 int
3032 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3033 {
3034
3035 return EOPNOTSUPP;
3036 }
3037
3038 int
3039 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3040 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3041 {
3042
3043 return EOPNOTSUPP;
3044 }
3045
3046 /*
3047 * This isn't really a ``null'' operation, but it's the default one and
3048 * doesn't do anything destructive.
3049 */
3050 int
3051 pru_sense_null(struct socket *so, struct stat *sb)
3052 {
3053
3054 sb->st_blksize = so->so_snd.sb_hiwat;
3055 return 0;
3056 }
3057
3058 int
3059 pru_shutdown_notsupp(struct socket *so)
3060 {
3061
3062 return EOPNOTSUPP;
3063 }
3064
3065 int
3066 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3067 {
3068
3069 return EOPNOTSUPP;
3070 }
3071
3072 int
3073 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3074 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3075 {
3076
3077 return EOPNOTSUPP;
3078 }
3079
3080 int
3081 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3082 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3083 {
3084
3085 return EOPNOTSUPP;
3086 }
3087
3088 int
3089 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3090 struct thread *td)
3091 {
3092
3093 return EOPNOTSUPP;
3094 }
3095
3096 static void
3097 filt_sordetach(struct knote *kn)
3098 {
3099 struct socket *so = kn->kn_fp->f_data;
3100
3101 SOCKBUF_LOCK(&so->so_rcv);
3102 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3103 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3104 so->so_rcv.sb_flags &= ~SB_KNOTE;
3105 SOCKBUF_UNLOCK(&so->so_rcv);
3106 }
3107
3108 /*ARGSUSED*/
3109 static int
3110 filt_soread(struct knote *kn, long hint)
3111 {
3112 struct socket *so;
3113
3114 so = kn->kn_fp->f_data;
3115 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3116
3117 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3118 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3119 kn->kn_flags |= EV_EOF;
3120 kn->kn_fflags = so->so_error;
3121 return (1);
3122 } else if (so->so_error) /* temporary udp error */
3123 return (1);
3124 else if (kn->kn_sfflags & NOTE_LOWAT)
3125 return (kn->kn_data >= kn->kn_sdata);
3126 else
3127 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3128 }
3129
3130 static void
3131 filt_sowdetach(struct knote *kn)
3132 {
3133 struct socket *so = kn->kn_fp->f_data;
3134
3135 SOCKBUF_LOCK(&so->so_snd);
3136 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3137 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3138 so->so_snd.sb_flags &= ~SB_KNOTE;
3139 SOCKBUF_UNLOCK(&so->so_snd);
3140 }
3141
3142 /*ARGSUSED*/
3143 static int
3144 filt_sowrite(struct knote *kn, long hint)
3145 {
3146 struct socket *so;
3147
3148 so = kn->kn_fp->f_data;
3149 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3150 kn->kn_data = sbspace(&so->so_snd);
3151 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3152 kn->kn_flags |= EV_EOF;
3153 kn->kn_fflags = so->so_error;
3154 return (1);
3155 } else if (so->so_error) /* temporary udp error */
3156 return (1);
3157 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3158 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3159 return (0);
3160 else if (kn->kn_sfflags & NOTE_LOWAT)
3161 return (kn->kn_data >= kn->kn_sdata);
3162 else
3163 return (kn->kn_data >= so->so_snd.sb_lowat);
3164 }
3165
3166 /*ARGSUSED*/
3167 static int
3168 filt_solisten(struct knote *kn, long hint)
3169 {
3170 struct socket *so = kn->kn_fp->f_data;
3171
3172 kn->kn_data = so->so_qlen;
3173 return (! TAILQ_EMPTY(&so->so_comp));
3174 }
3175
3176 int
3177 socheckuid(struct socket *so, uid_t uid)
3178 {
3179
3180 if (so == NULL)
3181 return (EPERM);
3182 if (so->so_cred->cr_uid != uid)
3183 return (EPERM);
3184 return (0);
3185 }
3186
3187 static int
3188 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3189 {
3190 int error;
3191 int val;
3192
3193 val = somaxconn;
3194 error = sysctl_handle_int(oidp, &val, 0, req);
3195 if (error || !req->newptr )
3196 return (error);
3197
3198 if (val < 1 || val > USHRT_MAX)
3199 return (EINVAL);
3200
3201 somaxconn = val;
3202 return (0);
3203 }
3204
3205 /*
3206 * These functions are used by protocols to notify the socket layer (and its
3207 * consumers) of state changes in the sockets driven by protocol-side events.
3208 */
3209
3210 /*
3211 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3212 *
3213 * Normal sequence from the active (originating) side is that
3214 * soisconnecting() is called during processing of connect() call, resulting
3215 * in an eventual call to soisconnected() if/when the connection is
3216 * established. When the connection is torn down soisdisconnecting() is
3217 * called during processing of disconnect() call, and soisdisconnected() is
3218 * called when the connection to the peer is totally severed. The semantics
3219 * of these routines are such that connectionless protocols can call
3220 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3221 * calls when setting up a ``connection'' takes no time.
3222 *
3223 * From the passive side, a socket is created with two queues of sockets:
3224 * so_incomp for connections in progress and so_comp for connections already
3225 * made and awaiting user acceptance. As a protocol is preparing incoming
3226 * connections, it creates a socket structure queued on so_incomp by calling
3227 * sonewconn(). When the connection is established, soisconnected() is
3228 * called, and transfers the socket structure to so_comp, making it available
3229 * to accept().
3230 *
3231 * If a socket is closed with sockets on either so_incomp or so_comp, these
3232 * sockets are dropped.
3233 *
3234 * If higher-level protocols are implemented in the kernel, the wakeups done
3235 * here will sometimes cause software-interrupt process scheduling.
3236 */
3237 void
3238 soisconnecting(struct socket *so)
3239 {
3240
3241 SOCK_LOCK(so);
3242 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3243 so->so_state |= SS_ISCONNECTING;
3244 SOCK_UNLOCK(so);
3245 }
3246
3247 void
3248 soisconnected(struct socket *so)
3249 {
3250 struct socket *head;
3251 int ret;
3252
3253 restart:
3254 ACCEPT_LOCK();
3255 SOCK_LOCK(so);
3256 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3257 so->so_state |= SS_ISCONNECTED;
3258 head = so->so_head;
3259 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3260 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3261 SOCK_UNLOCK(so);
3262 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3263 head->so_incqlen--;
3264 so->so_qstate &= ~SQ_INCOMP;
3265 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3266 head->so_qlen++;
3267 so->so_qstate |= SQ_COMP;
3268 ACCEPT_UNLOCK();
3269 sorwakeup(head);
3270 wakeup_one(&head->so_timeo);
3271 } else {
3272 ACCEPT_UNLOCK();
3273 soupcall_set(so, SO_RCV,
3274 head->so_accf->so_accept_filter->accf_callback,
3275 head->so_accf->so_accept_filter_arg);
3276 so->so_options &= ~SO_ACCEPTFILTER;
3277 ret = head->so_accf->so_accept_filter->accf_callback(so,
3278 head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3279 if (ret == SU_ISCONNECTED)
3280 soupcall_clear(so, SO_RCV);
3281 SOCK_UNLOCK(so);
3282 if (ret == SU_ISCONNECTED)
3283 goto restart;
3284 }
3285 return;
3286 }
3287 SOCK_UNLOCK(so);
3288 ACCEPT_UNLOCK();
3289 wakeup(&so->so_timeo);
3290 sorwakeup(so);
3291 sowwakeup(so);
3292 }
3293
3294 void
3295 soisdisconnecting(struct socket *so)
3296 {
3297
3298 /*
3299 * Note: This code assumes that SOCK_LOCK(so) and
3300 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3301 */
3302 SOCKBUF_LOCK(&so->so_rcv);
3303 so->so_state &= ~SS_ISCONNECTING;
3304 so->so_state |= SS_ISDISCONNECTING;
3305 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3306 sorwakeup_locked(so);
3307 SOCKBUF_LOCK(&so->so_snd);
3308 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3309 sowwakeup_locked(so);
3310 wakeup(&so->so_timeo);
3311 }
3312
3313 void
3314 soisdisconnected(struct socket *so)
3315 {
3316
3317 /*
3318 * Note: This code assumes that SOCK_LOCK(so) and
3319 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3320 */
3321 SOCKBUF_LOCK(&so->so_rcv);
3322 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3323 so->so_state |= SS_ISDISCONNECTED;
3324 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3325 sorwakeup_locked(so);
3326 SOCKBUF_LOCK(&so->so_snd);
3327 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3328 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3329 sowwakeup_locked(so);
3330 wakeup(&so->so_timeo);
3331 }
3332
3333 /*
3334 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3335 */
3336 struct sockaddr *
3337 sodupsockaddr(const struct sockaddr *sa, int mflags)
3338 {
3339 struct sockaddr *sa2;
3340
3341 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3342 if (sa2)
3343 bcopy(sa, sa2, sa->sa_len);
3344 return sa2;
3345 }
3346
3347 /*
3348 * Register per-socket buffer upcalls.
3349 */
3350 void
3351 soupcall_set(struct socket *so, int which,
3352 int (*func)(struct socket *, void *, int), void *arg)
3353 {
3354 struct sockbuf *sb;
3355
3356 switch (which) {
3357 case SO_RCV:
3358 sb = &so->so_rcv;
3359 break;
3360 case SO_SND:
3361 sb = &so->so_snd;
3362 break;
3363 default:
3364 panic("soupcall_set: bad which");
3365 }
3366 SOCKBUF_LOCK_ASSERT(sb);
3367 #if 0
3368 /* XXX: accf_http actually wants to do this on purpose. */
3369 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3370 #endif
3371 sb->sb_upcall = func;
3372 sb->sb_upcallarg = arg;
3373 sb->sb_flags |= SB_UPCALL;
3374 }
3375
3376 void
3377 soupcall_clear(struct socket *so, int which)
3378 {
3379 struct sockbuf *sb;
3380
3381 switch (which) {
3382 case SO_RCV:
3383 sb = &so->so_rcv;
3384 break;
3385 case SO_SND:
3386 sb = &so->so_snd;
3387 break;
3388 default:
3389 panic("soupcall_clear: bad which");
3390 }
3391 SOCKBUF_LOCK_ASSERT(sb);
3392 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3393 sb->sb_upcall = NULL;
3394 sb->sb_upcallarg = NULL;
3395 sb->sb_flags &= ~SB_UPCALL;
3396 }
3397
3398 /*
3399 * Create an external-format (``xsocket'') structure using the information in
3400 * the kernel-format socket structure pointed to by so. This is done to
3401 * reduce the spew of irrelevant information over this interface, to isolate
3402 * user code from changes in the kernel structure, and potentially to provide
3403 * information-hiding if we decide that some of this information should be
3404 * hidden from users.
3405 */
3406 void
3407 sotoxsocket(struct socket *so, struct xsocket *xso)
3408 {
3409
3410 xso->xso_len = sizeof *xso;
3411 xso->xso_so = so;
3412 xso->so_type = so->so_type;
3413 xso->so_options = so->so_options;
3414 xso->so_linger = so->so_linger;
3415 xso->so_state = so->so_state;
3416 xso->so_pcb = so->so_pcb;
3417 xso->xso_protocol = so->so_proto->pr_protocol;
3418 xso->xso_family = so->so_proto->pr_domain->dom_family;
3419 xso->so_qlen = so->so_qlen;
3420 xso->so_incqlen = so->so_incqlen;
3421 xso->so_qlimit = so->so_qlimit;
3422 xso->so_timeo = so->so_timeo;
3423 xso->so_error = so->so_error;
3424 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3425 xso->so_oobmark = so->so_oobmark;
3426 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3427 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3428 xso->so_uid = so->so_cred->cr_uid;
3429 }
3430
3431
3432 /*
3433 * Socket accessor functions to provide external consumers with
3434 * a safe interface to socket state
3435 *
3436 */
3437
3438 void
3439 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3440 {
3441
3442 TAILQ_FOREACH(so, &so->so_comp, so_list)
3443 func(so, arg);
3444 }
3445
3446 struct sockbuf *
3447 so_sockbuf_rcv(struct socket *so)
3448 {
3449
3450 return (&so->so_rcv);
3451 }
3452
3453 struct sockbuf *
3454 so_sockbuf_snd(struct socket *so)
3455 {
3456
3457 return (&so->so_snd);
3458 }
3459
3460 int
3461 so_state_get(const struct socket *so)
3462 {
3463
3464 return (so->so_state);
3465 }
3466
3467 void
3468 so_state_set(struct socket *so, int val)
3469 {
3470
3471 so->so_state = val;
3472 }
3473
3474 int
3475 so_options_get(const struct socket *so)
3476 {
3477
3478 return (so->so_options);
3479 }
3480
3481 void
3482 so_options_set(struct socket *so, int val)
3483 {
3484
3485 so->so_options = val;
3486 }
3487
3488 int
3489 so_error_get(const struct socket *so)
3490 {
3491
3492 return (so->so_error);
3493 }
3494
3495 void
3496 so_error_set(struct socket *so, int val)
3497 {
3498
3499 so->so_error = val;
3500 }
3501
3502 int
3503 so_linger_get(const struct socket *so)
3504 {
3505
3506 return (so->so_linger);
3507 }
3508
3509 void
3510 so_linger_set(struct socket *so, int val)
3511 {
3512
3513 so->so_linger = val;
3514 }
3515
3516 struct protosw *
3517 so_protosw_get(const struct socket *so)
3518 {
3519
3520 return (so->so_proto);
3521 }
3522
3523 void
3524 so_protosw_set(struct socket *so, struct protosw *val)
3525 {
3526
3527 so->so_proto = val;
3528 }
3529
3530 void
3531 so_sorwakeup(struct socket *so)
3532 {
3533
3534 sorwakeup(so);
3535 }
3536
3537 void
3538 so_sowwakeup(struct socket *so)
3539 {
3540
3541 sowwakeup(so);
3542 }
3543
3544 void
3545 so_sorwakeup_locked(struct socket *so)
3546 {
3547
3548 sorwakeup_locked(so);
3549 }
3550
3551 void
3552 so_sowwakeup_locked(struct socket *so)
3553 {
3554
3555 sowwakeup_locked(so);
3556 }
3557
3558 void
3559 so_lock(struct socket *so)
3560 {
3561 SOCK_LOCK(so);
3562 }
3563
3564 void
3565 so_unlock(struct socket *so)
3566 {
3567 SOCK_UNLOCK(so);
3568 }
Cache object: f529d8cd51337d9d2beb05bed2ecb791
|