1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2004 The FreeBSD Foundation
7 * Copyright (c) 2004-2008 Robert N. M. Watson
8 * All rights reserved.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
35 */
36
37 /*
38 * Comments on the socket life cycle:
39 *
40 * soalloc() sets of socket layer state for a socket, called only by
41 * socreate() and sonewconn(). Socket layer private.
42 *
43 * sodealloc() tears down socket layer state for a socket, called only by
44 * sofree() and sonewconn(). Socket layer private.
45 *
46 * pru_attach() associates protocol layer state with an allocated socket;
47 * called only once, may fail, aborting socket allocation. This is called
48 * from socreate() and sonewconn(). Socket layer private.
49 *
50 * pru_detach() disassociates protocol layer state from an attached socket,
51 * and will be called exactly once for sockets in which pru_attach() has
52 * been successfully called. If pru_attach() returned an error,
53 * pru_detach() will not be called. Socket layer private.
54 *
55 * pru_abort() and pru_close() notify the protocol layer that the last
56 * consumer of a socket is starting to tear down the socket, and that the
57 * protocol should terminate the connection. Historically, pru_abort() also
58 * detached protocol state from the socket state, but this is no longer the
59 * case.
60 *
61 * socreate() creates a socket and attaches protocol state. This is a public
62 * interface that may be used by socket layer consumers to create new
63 * sockets.
64 *
65 * sonewconn() creates a socket and attaches protocol state. This is a
66 * public interface that may be used by protocols to create new sockets when
67 * a new connection is received and will be available for accept() on a
68 * listen socket.
69 *
70 * soclose() destroys a socket after possibly waiting for it to disconnect.
71 * This is a public interface that socket consumers should use to close and
72 * release a socket when done with it.
73 *
74 * soabort() destroys a socket without waiting for it to disconnect (used
75 * only for incoming connections that are already partially or fully
76 * connected). This is used internally by the socket layer when clearing
77 * listen socket queues (due to overflow or close on the listen socket), but
78 * is also a public interface protocols may use to abort connections in
79 * their incomplete listen queues should they no longer be required. Sockets
80 * placed in completed connection listen queues should not be aborted for
81 * reasons described in the comment above the soclose() implementation. This
82 * is not a general purpose close routine, and except in the specific
83 * circumstances described here, should not be used.
84 *
85 * sofree() will free a socket and its protocol state if all references on
86 * the socket have been released, and is the public interface to attempt to
87 * free a socket when a reference is removed. This is a socket layer private
88 * interface.
89 *
90 * NOTE: In addition to socreate() and soclose(), which provide a single
91 * socket reference to the consumer to be managed as required, there are two
92 * calls to explicitly manage socket references, soref(), and sorele().
93 * Currently, these are generally required only when transitioning a socket
94 * from a listen queue to a file descriptor, in order to prevent garbage
95 * collection of the socket at an untimely moment. For a number of reasons,
96 * these interfaces are not preferred, and should be avoided.
97 *
98 * NOTE: With regard to VNETs the general rule is that callers do not set
99 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101 * and sorflush(), which are usually called from a pre-set VNET context.
102 * sopoll() currently does not need a VNET context to be set.
103 */
104
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
107
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_kern_tls.h"
111 #include "opt_sctp.h"
112
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/fcntl.h>
116 #include <sys/limits.h>
117 #include <sys/lock.h>
118 #include <sys/mac.h>
119 #include <sys/malloc.h>
120 #include <sys/mbuf.h>
121 #include <sys/mutex.h>
122 #include <sys/domain.h>
123 #include <sys/file.h> /* for struct knote */
124 #include <sys/hhook.h>
125 #include <sys/kernel.h>
126 #include <sys/khelp.h>
127 #include <sys/ktls.h>
128 #include <sys/event.h>
129 #include <sys/eventhandler.h>
130 #include <sys/poll.h>
131 #include <sys/proc.h>
132 #include <sys/protosw.h>
133 #include <sys/sbuf.h>
134 #include <sys/socket.h>
135 #include <sys/socketvar.h>
136 #include <sys/resourcevar.h>
137 #include <net/route.h>
138 #include <sys/signalvar.h>
139 #include <sys/stat.h>
140 #include <sys/sx.h>
141 #include <sys/sysctl.h>
142 #include <sys/taskqueue.h>
143 #include <sys/uio.h>
144 #include <sys/un.h>
145 #include <sys/unpcb.h>
146 #include <sys/jail.h>
147 #include <sys/syslog.h>
148 #include <netinet/in.h>
149 #include <netinet/in_pcb.h>
150 #include <netinet/tcp.h>
151
152 #include <net/vnet.h>
153
154 #include <security/mac/mac_framework.h>
155
156 #include <vm/uma.h>
157
158 #ifdef COMPAT_FREEBSD32
159 #include <sys/mount.h>
160 #include <sys/sysent.h>
161 #include <compat/freebsd32/freebsd32.h>
162 #endif
163
164 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
165 int flags);
166 static void so_rdknl_lock(void *);
167 static void so_rdknl_unlock(void *);
168 static void so_rdknl_assert_lock(void *, int);
169 static void so_wrknl_lock(void *);
170 static void so_wrknl_unlock(void *);
171 static void so_wrknl_assert_lock(void *, int);
172
173 static void filt_sordetach(struct knote *kn);
174 static int filt_soread(struct knote *kn, long hint);
175 static void filt_sowdetach(struct knote *kn);
176 static int filt_sowrite(struct knote *kn, long hint);
177 static int filt_soempty(struct knote *kn, long hint);
178 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
179 fo_kqfilter_t soo_kqfilter;
180
181 static struct filterops soread_filtops = {
182 .f_isfd = 1,
183 .f_detach = filt_sordetach,
184 .f_event = filt_soread,
185 };
186 static struct filterops sowrite_filtops = {
187 .f_isfd = 1,
188 .f_detach = filt_sowdetach,
189 .f_event = filt_sowrite,
190 };
191 static struct filterops soempty_filtops = {
192 .f_isfd = 1,
193 .f_detach = filt_sowdetach,
194 .f_event = filt_soempty,
195 };
196
197 so_gen_t so_gencnt; /* generation count for sockets */
198
199 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
200 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
201
202 #define VNET_SO_ASSERT(so) \
203 VNET_ASSERT(curvnet != NULL, \
204 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
205
206 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
207 #define V_socket_hhh VNET(socket_hhh)
208
209 /*
210 * Limit on the number of connections in the listen queue waiting
211 * for accept(2).
212 * NB: The original sysctl somaxconn is still available but hidden
213 * to prevent confusion about the actual purpose of this number.
214 */
215 static u_int somaxconn = SOMAXCONN;
216
217 static int
218 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
219 {
220 int error;
221 int val;
222
223 val = somaxconn;
224 error = sysctl_handle_int(oidp, &val, 0, req);
225 if (error || !req->newptr )
226 return (error);
227
228 /*
229 * The purpose of the UINT_MAX / 3 limit, is so that the formula
230 * 3 * so_qlimit / 2
231 * below, will not overflow.
232 */
233
234 if (val < 1 || val > UINT_MAX / 3)
235 return (EINVAL);
236
237 somaxconn = val;
238 return (0);
239 }
240 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
241 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
242 sysctl_somaxconn, "I",
243 "Maximum listen socket pending connection accept queue size");
244 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
245 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
246 sizeof(int), sysctl_somaxconn, "I",
247 "Maximum listen socket pending connection accept queue size (compat)");
248
249 static int numopensockets;
250 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
251 &numopensockets, 0, "Number of open sockets");
252
253 /*
254 * accept_mtx locks down per-socket fields relating to accept queues. See
255 * socketvar.h for an annotation of the protected fields of struct socket.
256 */
257 struct mtx accept_mtx;
258 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
259
260 /*
261 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
262 * so_gencnt field.
263 */
264 static struct mtx so_global_mtx;
265 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
266
267 /*
268 * General IPC sysctl name space, used by sockets and a variety of other IPC
269 * types.
270 */
271 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
272 "IPC");
273
274 /*
275 * Initialize the socket subsystem and set up the socket
276 * memory allocator.
277 */
278 static uma_zone_t socket_zone;
279 int maxsockets;
280
281 static void
282 socket_zone_change(void *tag)
283 {
284
285 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
286 }
287
288 static void
289 socket_hhook_register(int subtype)
290 {
291
292 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
293 &V_socket_hhh[subtype],
294 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
295 printf("%s: WARNING: unable to register hook\n", __func__);
296 }
297
298 static void
299 socket_hhook_deregister(int subtype)
300 {
301
302 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
303 printf("%s: WARNING: unable to deregister hook\n", __func__);
304 }
305
306 static void
307 socket_init(void *tag)
308 {
309
310 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
311 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
312 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
313 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
314 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
315 EVENTHANDLER_PRI_FIRST);
316 }
317 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
318
319 static void
320 socket_vnet_init(const void *unused __unused)
321 {
322 int i;
323
324 /* We expect a contiguous range */
325 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
326 socket_hhook_register(i);
327 }
328 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
329 socket_vnet_init, NULL);
330
331 static void
332 socket_vnet_uninit(const void *unused __unused)
333 {
334 int i;
335
336 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
337 socket_hhook_deregister(i);
338 }
339 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
340 socket_vnet_uninit, NULL);
341
342 /*
343 * Initialise maxsockets. This SYSINIT must be run after
344 * tunable_mbinit().
345 */
346 static void
347 init_maxsockets(void *ignored)
348 {
349
350 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
351 maxsockets = imax(maxsockets, maxfiles);
352 }
353 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
354
355 /*
356 * Sysctl to get and set the maximum global sockets limit. Notify protocols
357 * of the change so that they can update their dependent limits as required.
358 */
359 static int
360 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
361 {
362 int error, newmaxsockets;
363
364 newmaxsockets = maxsockets;
365 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
366 if (error == 0 && req->newptr) {
367 if (newmaxsockets > maxsockets &&
368 newmaxsockets <= maxfiles) {
369 maxsockets = newmaxsockets;
370 EVENTHANDLER_INVOKE(maxsockets_change);
371 } else
372 error = EINVAL;
373 }
374 return (error);
375 }
376 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
377 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
378 sysctl_maxsockets, "IU",
379 "Maximum number of sockets available");
380
381 /*
382 * Socket operation routines. These routines are called by the routines in
383 * sys_socket.c or from a system process, and implement the semantics of
384 * socket operations by switching out to the protocol specific routines.
385 */
386
387 /*
388 * Get a socket structure from our zone, and initialize it. Note that it
389 * would probably be better to allocate socket and PCB at the same time, but
390 * I'm not convinced that all the protocols can be easily modified to do
391 * this.
392 *
393 * soalloc() returns a socket with a ref count of 0.
394 */
395 static struct socket *
396 soalloc(struct vnet *vnet)
397 {
398 struct socket *so;
399
400 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
401 if (so == NULL)
402 return (NULL);
403 #ifdef MAC
404 if (mac_socket_init(so, M_NOWAIT) != 0) {
405 uma_zfree(socket_zone, so);
406 return (NULL);
407 }
408 #endif
409 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
410 uma_zfree(socket_zone, so);
411 return (NULL);
412 }
413
414 /*
415 * The socket locking protocol allows to lock 2 sockets at a time,
416 * however, the first one must be a listening socket. WITNESS lacks
417 * a feature to change class of an existing lock, so we use DUPOK.
418 */
419 mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
420 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
421 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
422 so->so_rcv.sb_sel = &so->so_rdsel;
423 so->so_snd.sb_sel = &so->so_wrsel;
424 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
425 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
426 TAILQ_INIT(&so->so_snd.sb_aiojobq);
427 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
428 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
429 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
430 #ifdef VIMAGE
431 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
432 __func__, __LINE__, so));
433 so->so_vnet = vnet;
434 #endif
435 /* We shouldn't need the so_global_mtx */
436 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
437 /* Do we need more comprehensive error returns? */
438 uma_zfree(socket_zone, so);
439 return (NULL);
440 }
441 mtx_lock(&so_global_mtx);
442 so->so_gencnt = ++so_gencnt;
443 ++numopensockets;
444 #ifdef VIMAGE
445 vnet->vnet_sockcnt++;
446 #endif
447 mtx_unlock(&so_global_mtx);
448
449 return (so);
450 }
451
452 /*
453 * Free the storage associated with a socket at the socket layer, tear down
454 * locks, labels, etc. All protocol state is assumed already to have been
455 * torn down (and possibly never set up) by the caller.
456 */
457 static void
458 sodealloc(struct socket *so)
459 {
460
461 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
462 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
463
464 mtx_lock(&so_global_mtx);
465 so->so_gencnt = ++so_gencnt;
466 --numopensockets; /* Could be below, but faster here. */
467 #ifdef VIMAGE
468 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
469 __func__, __LINE__, so));
470 so->so_vnet->vnet_sockcnt--;
471 #endif
472 mtx_unlock(&so_global_mtx);
473 #ifdef MAC
474 mac_socket_destroy(so);
475 #endif
476 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
477
478 crfree(so->so_cred);
479 khelp_destroy_osd(&so->osd);
480 if (SOLISTENING(so)) {
481 if (so->sol_accept_filter != NULL)
482 accept_filt_setopt(so, NULL);
483 } else {
484 if (so->so_rcv.sb_hiwat)
485 (void)chgsbsize(so->so_cred->cr_uidinfo,
486 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
487 if (so->so_snd.sb_hiwat)
488 (void)chgsbsize(so->so_cred->cr_uidinfo,
489 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
490 sx_destroy(&so->so_snd.sb_sx);
491 sx_destroy(&so->so_rcv.sb_sx);
492 SOCKBUF_LOCK_DESTROY(&so->so_snd);
493 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
494 }
495 mtx_destroy(&so->so_lock);
496 uma_zfree(socket_zone, so);
497 }
498
499 /*
500 * socreate returns a socket with a ref count of 1. The socket should be
501 * closed with soclose().
502 */
503 int
504 socreate(int dom, struct socket **aso, int type, int proto,
505 struct ucred *cred, struct thread *td)
506 {
507 struct protosw *prp;
508 struct socket *so;
509 int error;
510
511 if (proto)
512 prp = pffindproto(dom, proto, type);
513 else
514 prp = pffindtype(dom, type);
515
516 if (prp == NULL) {
517 /* No support for domain. */
518 if (pffinddomain(dom) == NULL)
519 return (EAFNOSUPPORT);
520 /* No support for socket type. */
521 if (proto == 0 && type != 0)
522 return (EPROTOTYPE);
523 return (EPROTONOSUPPORT);
524 }
525 if (prp->pr_usrreqs->pru_attach == NULL ||
526 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
527 return (EPROTONOSUPPORT);
528
529 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
530 return (EPROTONOSUPPORT);
531
532 if (prp->pr_type != type)
533 return (EPROTOTYPE);
534 so = soalloc(CRED_TO_VNET(cred));
535 if (so == NULL)
536 return (ENOBUFS);
537
538 so->so_type = type;
539 so->so_cred = crhold(cred);
540 if ((prp->pr_domain->dom_family == PF_INET) ||
541 (prp->pr_domain->dom_family == PF_INET6) ||
542 (prp->pr_domain->dom_family == PF_ROUTE))
543 so->so_fibnum = td->td_proc->p_fibnum;
544 else
545 so->so_fibnum = 0;
546 so->so_proto = prp;
547 #ifdef MAC
548 mac_socket_create(cred, so);
549 #endif
550 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
551 so_rdknl_assert_lock);
552 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
553 so_wrknl_assert_lock);
554 /*
555 * Auto-sizing of socket buffers is managed by the protocols and
556 * the appropriate flags must be set in the pru_attach function.
557 */
558 CURVNET_SET(so->so_vnet);
559 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
560 CURVNET_RESTORE();
561 if (error) {
562 sodealloc(so);
563 return (error);
564 }
565 soref(so);
566 *aso = so;
567 return (0);
568 }
569
570 #ifdef REGRESSION
571 static int regression_sonewconn_earlytest = 1;
572 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
573 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
574 #endif
575
576 static struct timeval overinterval = { 60, 0 };
577 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
578 &overinterval,
579 "Delay in seconds between warnings for listen socket overflows");
580
581 /*
582 * When an attempt at a new connection is noted on a socket which accepts
583 * connections, sonewconn is called. If the connection is possible (subject
584 * to space constraints, etc.) then we allocate a new structure, properly
585 * linked into the data structure of the original socket, and return this.
586 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
587 *
588 * Note: the ref count on the socket is 0 on return.
589 */
590 struct socket *
591 sonewconn(struct socket *head, int connstatus)
592 {
593 struct sbuf descrsb;
594 struct socket *so;
595 int len, overcount;
596 u_int qlen;
597 const char localprefix[] = "local:";
598 char descrbuf[SUNPATHLEN + sizeof(localprefix)];
599 #if defined(INET6)
600 char addrbuf[INET6_ADDRSTRLEN];
601 #elif defined(INET)
602 char addrbuf[INET_ADDRSTRLEN];
603 #endif
604 bool dolog, over;
605
606 SOLISTEN_LOCK(head);
607 over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
608 #ifdef REGRESSION
609 if (regression_sonewconn_earlytest && over) {
610 #else
611 if (over) {
612 #endif
613 head->sol_overcount++;
614 dolog = !!ratecheck(&head->sol_lastover, &overinterval);
615
616 /*
617 * If we're going to log, copy the overflow count and queue
618 * length from the listen socket before dropping the lock.
619 * Also, reset the overflow count.
620 */
621 if (dolog) {
622 overcount = head->sol_overcount;
623 head->sol_overcount = 0;
624 qlen = head->sol_qlen;
625 }
626 SOLISTEN_UNLOCK(head);
627
628 if (dolog) {
629 /*
630 * Try to print something descriptive about the
631 * socket for the error message.
632 */
633 sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
634 SBUF_FIXEDLEN);
635 switch (head->so_proto->pr_domain->dom_family) {
636 #if defined(INET) || defined(INET6)
637 #ifdef INET
638 case AF_INET:
639 #endif
640 #ifdef INET6
641 case AF_INET6:
642 if (head->so_proto->pr_domain->dom_family ==
643 AF_INET6 ||
644 (sotoinpcb(head)->inp_inc.inc_flags &
645 INC_ISIPV6)) {
646 ip6_sprintf(addrbuf,
647 &sotoinpcb(head)->inp_inc.inc6_laddr);
648 sbuf_printf(&descrsb, "[%s]", addrbuf);
649 } else
650 #endif
651 {
652 #ifdef INET
653 inet_ntoa_r(
654 sotoinpcb(head)->inp_inc.inc_laddr,
655 addrbuf);
656 sbuf_cat(&descrsb, addrbuf);
657 #endif
658 }
659 sbuf_printf(&descrsb, ":%hu (proto %u)",
660 ntohs(sotoinpcb(head)->inp_inc.inc_lport),
661 head->so_proto->pr_protocol);
662 break;
663 #endif /* INET || INET6 */
664 case AF_UNIX:
665 sbuf_cat(&descrsb, localprefix);
666 if (sotounpcb(head)->unp_addr != NULL)
667 len =
668 sotounpcb(head)->unp_addr->sun_len -
669 offsetof(struct sockaddr_un,
670 sun_path);
671 else
672 len = 0;
673 if (len > 0)
674 sbuf_bcat(&descrsb,
675 sotounpcb(head)->unp_addr->sun_path,
676 len);
677 else
678 sbuf_cat(&descrsb, "(unknown)");
679 break;
680 }
681
682 /*
683 * If we can't print something more specific, at least
684 * print the domain name.
685 */
686 if (sbuf_finish(&descrsb) != 0 ||
687 sbuf_len(&descrsb) <= 0) {
688 sbuf_clear(&descrsb);
689 sbuf_cat(&descrsb,
690 head->so_proto->pr_domain->dom_name ?:
691 "unknown");
692 sbuf_finish(&descrsb);
693 }
694 KASSERT(sbuf_len(&descrsb) > 0,
695 ("%s: sbuf creation failed", __func__));
696 log(LOG_DEBUG,
697 "%s: pcb %p (%s): Listen queue overflow: "
698 "%i already in queue awaiting acceptance "
699 "(%d occurrences)\n",
700 __func__, head->so_pcb, sbuf_data(&descrsb),
701 qlen, overcount);
702 sbuf_delete(&descrsb);
703
704 overcount = 0;
705 }
706
707 return (NULL);
708 }
709 SOLISTEN_UNLOCK(head);
710 VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
711 __func__, head));
712 so = soalloc(head->so_vnet);
713 if (so == NULL) {
714 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
715 "limit reached or out of memory\n",
716 __func__, head->so_pcb);
717 return (NULL);
718 }
719 so->so_listen = head;
720 so->so_type = head->so_type;
721 so->so_linger = head->so_linger;
722 so->so_state = head->so_state | SS_NOFDREF;
723 so->so_fibnum = head->so_fibnum;
724 so->so_proto = head->so_proto;
725 so->so_cred = crhold(head->so_cred);
726 #ifdef MAC
727 mac_socket_newconn(head, so);
728 #endif
729 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
730 so_rdknl_assert_lock);
731 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
732 so_wrknl_assert_lock);
733 VNET_SO_ASSERT(head);
734 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
735 sodealloc(so);
736 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
737 __func__, head->so_pcb);
738 return (NULL);
739 }
740 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
741 sodealloc(so);
742 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
743 __func__, head->so_pcb);
744 return (NULL);
745 }
746 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
747 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
748 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
749 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
750 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
751 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
752
753 SOLISTEN_LOCK(head);
754 if (head->sol_accept_filter != NULL)
755 connstatus = 0;
756 so->so_state |= connstatus;
757 so->so_options = head->so_options & ~SO_ACCEPTCONN;
758 soref(head); /* A socket on (in)complete queue refs head. */
759 if (connstatus) {
760 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
761 so->so_qstate = SQ_COMP;
762 head->sol_qlen++;
763 solisten_wakeup(head); /* unlocks */
764 } else {
765 /*
766 * Keep removing sockets from the head until there's room for
767 * us to insert on the tail. In pre-locking revisions, this
768 * was a simple if(), but as we could be racing with other
769 * threads and soabort() requires dropping locks, we must
770 * loop waiting for the condition to be true.
771 */
772 while (head->sol_incqlen > head->sol_qlimit) {
773 struct socket *sp;
774
775 sp = TAILQ_FIRST(&head->sol_incomp);
776 TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
777 head->sol_incqlen--;
778 SOCK_LOCK(sp);
779 sp->so_qstate = SQ_NONE;
780 sp->so_listen = NULL;
781 SOCK_UNLOCK(sp);
782 sorele(head); /* does SOLISTEN_UNLOCK, head stays */
783 soabort(sp);
784 SOLISTEN_LOCK(head);
785 }
786 TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
787 so->so_qstate = SQ_INCOMP;
788 head->sol_incqlen++;
789 SOLISTEN_UNLOCK(head);
790 }
791 return (so);
792 }
793
794 #if defined(SCTP) || defined(SCTP_SUPPORT)
795 /*
796 * Socket part of sctp_peeloff(). Detach a new socket from an
797 * association. The new socket is returned with a reference.
798 */
799 struct socket *
800 sopeeloff(struct socket *head)
801 {
802 struct socket *so;
803
804 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
805 __func__, __LINE__, head));
806 so = soalloc(head->so_vnet);
807 if (so == NULL) {
808 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
809 "limit reached or out of memory\n",
810 __func__, head->so_pcb);
811 return (NULL);
812 }
813 so->so_type = head->so_type;
814 so->so_options = head->so_options;
815 so->so_linger = head->so_linger;
816 so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
817 so->so_fibnum = head->so_fibnum;
818 so->so_proto = head->so_proto;
819 so->so_cred = crhold(head->so_cred);
820 #ifdef MAC
821 mac_socket_newconn(head, so);
822 #endif
823 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
824 so_rdknl_assert_lock);
825 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
826 so_wrknl_assert_lock);
827 VNET_SO_ASSERT(head);
828 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
829 sodealloc(so);
830 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
831 __func__, head->so_pcb);
832 return (NULL);
833 }
834 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
835 sodealloc(so);
836 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
837 __func__, head->so_pcb);
838 return (NULL);
839 }
840 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
841 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
842 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
843 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
844 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
845 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
846
847 soref(so);
848
849 return (so);
850 }
851 #endif /* SCTP */
852
853 int
854 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
855 {
856 int error;
857
858 CURVNET_SET(so->so_vnet);
859 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
860 CURVNET_RESTORE();
861 return (error);
862 }
863
864 int
865 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
866 {
867 int error;
868
869 CURVNET_SET(so->so_vnet);
870 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
871 CURVNET_RESTORE();
872 return (error);
873 }
874
875 /*
876 * solisten() transitions a socket from a non-listening state to a listening
877 * state, but can also be used to update the listen queue depth on an
878 * existing listen socket. The protocol will call back into the sockets
879 * layer using solisten_proto_check() and solisten_proto() to check and set
880 * socket-layer listen state. Call backs are used so that the protocol can
881 * acquire both protocol and socket layer locks in whatever order is required
882 * by the protocol.
883 *
884 * Protocol implementors are advised to hold the socket lock across the
885 * socket-layer test and set to avoid races at the socket layer.
886 */
887 int
888 solisten(struct socket *so, int backlog, struct thread *td)
889 {
890 int error;
891
892 CURVNET_SET(so->so_vnet);
893 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
894 CURVNET_RESTORE();
895 return (error);
896 }
897
898 int
899 solisten_proto_check(struct socket *so)
900 {
901
902 SOCK_LOCK_ASSERT(so);
903
904 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
905 SS_ISDISCONNECTING))
906 return (EINVAL);
907 return (0);
908 }
909
910 void
911 solisten_proto(struct socket *so, int backlog)
912 {
913 int sbrcv_lowat, sbsnd_lowat;
914 u_int sbrcv_hiwat, sbsnd_hiwat;
915 short sbrcv_flags, sbsnd_flags;
916 sbintime_t sbrcv_timeo, sbsnd_timeo;
917
918 SOCK_LOCK_ASSERT(so);
919
920 if (SOLISTENING(so))
921 goto listening;
922
923 /*
924 * Change this socket to listening state.
925 */
926 sbrcv_lowat = so->so_rcv.sb_lowat;
927 sbsnd_lowat = so->so_snd.sb_lowat;
928 sbrcv_hiwat = so->so_rcv.sb_hiwat;
929 sbsnd_hiwat = so->so_snd.sb_hiwat;
930 sbrcv_flags = so->so_rcv.sb_flags;
931 sbsnd_flags = so->so_snd.sb_flags;
932 sbrcv_timeo = so->so_rcv.sb_timeo;
933 sbsnd_timeo = so->so_snd.sb_timeo;
934
935 sbdestroy(&so->so_snd, so);
936 sbdestroy(&so->so_rcv, so);
937 sx_destroy(&so->so_snd.sb_sx);
938 sx_destroy(&so->so_rcv.sb_sx);
939 SOCKBUF_LOCK_DESTROY(&so->so_snd);
940 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
941
942 #ifdef INVARIANTS
943 bzero(&so->so_rcv,
944 sizeof(struct socket) - offsetof(struct socket, so_rcv));
945 #endif
946
947 so->sol_sbrcv_lowat = sbrcv_lowat;
948 so->sol_sbsnd_lowat = sbsnd_lowat;
949 so->sol_sbrcv_hiwat = sbrcv_hiwat;
950 so->sol_sbsnd_hiwat = sbsnd_hiwat;
951 so->sol_sbrcv_flags = sbrcv_flags;
952 so->sol_sbsnd_flags = sbsnd_flags;
953 so->sol_sbrcv_timeo = sbrcv_timeo;
954 so->sol_sbsnd_timeo = sbsnd_timeo;
955
956 so->sol_qlen = so->sol_incqlen = 0;
957 TAILQ_INIT(&so->sol_incomp);
958 TAILQ_INIT(&so->sol_comp);
959
960 so->sol_accept_filter = NULL;
961 so->sol_accept_filter_arg = NULL;
962 so->sol_accept_filter_str = NULL;
963
964 so->sol_upcall = NULL;
965 so->sol_upcallarg = NULL;
966
967 so->so_options |= SO_ACCEPTCONN;
968
969 listening:
970 if (backlog < 0 || backlog > somaxconn)
971 backlog = somaxconn;
972 so->sol_qlimit = backlog;
973 }
974
975 /*
976 * Wakeup listeners/subsystems once we have a complete connection.
977 * Enters with lock, returns unlocked.
978 */
979 void
980 solisten_wakeup(struct socket *sol)
981 {
982
983 if (sol->sol_upcall != NULL)
984 (void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
985 else {
986 selwakeuppri(&sol->so_rdsel, PSOCK);
987 KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
988 }
989 SOLISTEN_UNLOCK(sol);
990 wakeup_one(&sol->sol_comp);
991 if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
992 pgsigio(&sol->so_sigio, SIGIO, 0);
993 }
994
995 /*
996 * Return single connection off a listening socket queue. Main consumer of
997 * the function is kern_accept4(). Some modules, that do their own accept
998 * management also use the function.
999 *
1000 * Listening socket must be locked on entry and is returned unlocked on
1001 * return.
1002 * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1003 */
1004 int
1005 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1006 {
1007 struct socket *so;
1008 int error;
1009
1010 SOLISTEN_LOCK_ASSERT(head);
1011
1012 while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1013 head->so_error == 0) {
1014 error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
1015 "accept", 0);
1016 if (error != 0) {
1017 SOLISTEN_UNLOCK(head);
1018 return (error);
1019 }
1020 }
1021 if (head->so_error) {
1022 error = head->so_error;
1023 head->so_error = 0;
1024 } else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1025 error = EWOULDBLOCK;
1026 else
1027 error = 0;
1028 if (error) {
1029 SOLISTEN_UNLOCK(head);
1030 return (error);
1031 }
1032 so = TAILQ_FIRST(&head->sol_comp);
1033 SOCK_LOCK(so);
1034 KASSERT(so->so_qstate == SQ_COMP,
1035 ("%s: so %p not SQ_COMP", __func__, so));
1036 soref(so);
1037 head->sol_qlen--;
1038 so->so_qstate = SQ_NONE;
1039 so->so_listen = NULL;
1040 TAILQ_REMOVE(&head->sol_comp, so, so_list);
1041 if (flags & ACCEPT4_INHERIT)
1042 so->so_state |= (head->so_state & SS_NBIO);
1043 else
1044 so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1045 SOCK_UNLOCK(so);
1046 sorele(head);
1047
1048 *ret = so;
1049 return (0);
1050 }
1051
1052 /*
1053 * Evaluate the reference count and named references on a socket; if no
1054 * references remain, free it. This should be called whenever a reference is
1055 * released, such as in sorele(), but also when named reference flags are
1056 * cleared in socket or protocol code.
1057 *
1058 * sofree() will free the socket if:
1059 *
1060 * - There are no outstanding file descriptor references or related consumers
1061 * (so_count == 0).
1062 *
1063 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1064 *
1065 * - The protocol does not have an outstanding strong reference on the socket
1066 * (SS_PROTOREF).
1067 *
1068 * - The socket is not in a completed connection queue, so a process has been
1069 * notified that it is present. If it is removed, the user process may
1070 * block in accept() despite select() saying the socket was ready.
1071 */
1072 void
1073 sofree(struct socket *so)
1074 {
1075 struct protosw *pr = so->so_proto;
1076
1077 SOCK_LOCK_ASSERT(so);
1078
1079 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
1080 (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
1081 SOCK_UNLOCK(so);
1082 return;
1083 }
1084
1085 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1086 struct socket *sol;
1087
1088 sol = so->so_listen;
1089 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1090
1091 /*
1092 * To solve race between close of a listening socket and
1093 * a socket on its incomplete queue, we need to lock both.
1094 * The order is first listening socket, then regular.
1095 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1096 * function and the listening socket are the only pointers
1097 * to so. To preserve so and sol, we reference both and then
1098 * relock.
1099 * After relock the socket may not move to so_comp since it
1100 * doesn't have PCB already, but it may be removed from
1101 * so_incomp. If that happens, we share responsiblity on
1102 * freeing the socket, but soclose() has already removed
1103 * it from queue.
1104 */
1105 soref(sol);
1106 soref(so);
1107 SOCK_UNLOCK(so);
1108 SOLISTEN_LOCK(sol);
1109 SOCK_LOCK(so);
1110 if (so->so_qstate == SQ_INCOMP) {
1111 KASSERT(so->so_listen == sol,
1112 ("%s: so %p migrated out of sol %p",
1113 __func__, so, sol));
1114 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1115 sol->sol_incqlen--;
1116 /* This is guarenteed not to be the last. */
1117 refcount_release(&sol->so_count);
1118 so->so_qstate = SQ_NONE;
1119 so->so_listen = NULL;
1120 } else
1121 KASSERT(so->so_listen == NULL,
1122 ("%s: so %p not on (in)comp with so_listen",
1123 __func__, so));
1124 sorele(sol);
1125 KASSERT(so->so_count == 1,
1126 ("%s: so %p count %u", __func__, so, so->so_count));
1127 so->so_count = 0;
1128 }
1129 if (SOLISTENING(so))
1130 so->so_error = ECONNABORTED;
1131 SOCK_UNLOCK(so);
1132
1133 if (so->so_dtor != NULL)
1134 so->so_dtor(so);
1135
1136 VNET_SO_ASSERT(so);
1137 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1138 (*pr->pr_domain->dom_dispose)(so);
1139 if (pr->pr_usrreqs->pru_detach != NULL)
1140 (*pr->pr_usrreqs->pru_detach)(so);
1141
1142 /*
1143 * From this point on, we assume that no other references to this
1144 * socket exist anywhere else in the stack. Therefore, no locks need
1145 * to be acquired or held.
1146 *
1147 * We used to do a lot of socket buffer and socket locking here, as
1148 * well as invoke sorflush() and perform wakeups. The direct call to
1149 * dom_dispose() and sbdestroy() are an inlining of what was
1150 * necessary from sorflush().
1151 *
1152 * Notice that the socket buffer and kqueue state are torn down
1153 * before calling pru_detach. This means that protocols shold not
1154 * assume they can perform socket wakeups, etc, in their detach code.
1155 */
1156 if (!SOLISTENING(so)) {
1157 sbdestroy(&so->so_snd, so);
1158 sbdestroy(&so->so_rcv, so);
1159 }
1160 seldrain(&so->so_rdsel);
1161 seldrain(&so->so_wrsel);
1162 knlist_destroy(&so->so_rdsel.si_note);
1163 knlist_destroy(&so->so_wrsel.si_note);
1164 sodealloc(so);
1165 }
1166
1167 /*
1168 * Close a socket on last file table reference removal. Initiate disconnect
1169 * if connected. Free socket when disconnect complete.
1170 *
1171 * This function will sorele() the socket. Note that soclose() may be called
1172 * prior to the ref count reaching zero. The actual socket structure will
1173 * not be freed until the ref count reaches zero.
1174 */
1175 int
1176 soclose(struct socket *so)
1177 {
1178 struct accept_queue lqueue;
1179 bool listening;
1180 int error = 0;
1181
1182 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1183
1184 CURVNET_SET(so->so_vnet);
1185 funsetown(&so->so_sigio);
1186 if (so->so_state & SS_ISCONNECTED) {
1187 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1188 error = sodisconnect(so);
1189 if (error) {
1190 if (error == ENOTCONN)
1191 error = 0;
1192 goto drop;
1193 }
1194 }
1195
1196 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1197 if ((so->so_state & SS_ISDISCONNECTING) &&
1198 (so->so_state & SS_NBIO))
1199 goto drop;
1200 while (so->so_state & SS_ISCONNECTED) {
1201 error = tsleep(&so->so_timeo,
1202 PSOCK | PCATCH, "soclos",
1203 so->so_linger * hz);
1204 if (error)
1205 break;
1206 }
1207 }
1208 }
1209
1210 drop:
1211 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1212 (*so->so_proto->pr_usrreqs->pru_close)(so);
1213
1214 SOCK_LOCK(so);
1215 if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1216 struct socket *sp;
1217
1218 TAILQ_INIT(&lqueue);
1219 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1220 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1221
1222 so->sol_qlen = so->sol_incqlen = 0;
1223
1224 TAILQ_FOREACH(sp, &lqueue, so_list) {
1225 SOCK_LOCK(sp);
1226 sp->so_qstate = SQ_NONE;
1227 sp->so_listen = NULL;
1228 SOCK_UNLOCK(sp);
1229 /* Guaranteed not to be the last. */
1230 refcount_release(&so->so_count);
1231 }
1232 }
1233 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1234 so->so_state |= SS_NOFDREF;
1235 sorele(so);
1236 if (listening) {
1237 struct socket *sp, *tsp;
1238
1239 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1240 SOCK_LOCK(sp);
1241 if (sp->so_count == 0) {
1242 SOCK_UNLOCK(sp);
1243 soabort(sp);
1244 } else
1245 /* sp is now in sofree() */
1246 SOCK_UNLOCK(sp);
1247 }
1248 }
1249 CURVNET_RESTORE();
1250 return (error);
1251 }
1252
1253 /*
1254 * soabort() is used to abruptly tear down a connection, such as when a
1255 * resource limit is reached (listen queue depth exceeded), or if a listen
1256 * socket is closed while there are sockets waiting to be accepted.
1257 *
1258 * This interface is tricky, because it is called on an unreferenced socket,
1259 * and must be called only by a thread that has actually removed the socket
1260 * from the listen queue it was on, or races with other threads are risked.
1261 *
1262 * This interface will call into the protocol code, so must not be called
1263 * with any socket locks held. Protocols do call it while holding their own
1264 * recursible protocol mutexes, but this is something that should be subject
1265 * to review in the future.
1266 */
1267 void
1268 soabort(struct socket *so)
1269 {
1270
1271 /*
1272 * In as much as is possible, assert that no references to this
1273 * socket are held. This is not quite the same as asserting that the
1274 * current thread is responsible for arranging for no references, but
1275 * is as close as we can get for now.
1276 */
1277 KASSERT(so->so_count == 0, ("soabort: so_count"));
1278 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1279 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1280 VNET_SO_ASSERT(so);
1281
1282 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1283 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1284 SOCK_LOCK(so);
1285 sofree(so);
1286 }
1287
1288 int
1289 soaccept(struct socket *so, struct sockaddr **nam)
1290 {
1291 int error;
1292
1293 SOCK_LOCK(so);
1294 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1295 so->so_state &= ~SS_NOFDREF;
1296 SOCK_UNLOCK(so);
1297
1298 CURVNET_SET(so->so_vnet);
1299 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1300 CURVNET_RESTORE();
1301 return (error);
1302 }
1303
1304 int
1305 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1306 {
1307
1308 return (soconnectat(AT_FDCWD, so, nam, td));
1309 }
1310
1311 int
1312 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1313 {
1314 int error;
1315
1316 if (so->so_options & SO_ACCEPTCONN)
1317 return (EOPNOTSUPP);
1318
1319 CURVNET_SET(so->so_vnet);
1320 /*
1321 * If protocol is connection-based, can only connect once.
1322 * Otherwise, if connected, try to disconnect first. This allows
1323 * user to disconnect by connecting to, e.g., a null address.
1324 */
1325 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1326 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1327 (error = sodisconnect(so)))) {
1328 error = EISCONN;
1329 } else {
1330 /*
1331 * Prevent accumulated error from previous connection from
1332 * biting us.
1333 */
1334 so->so_error = 0;
1335 if (fd == AT_FDCWD) {
1336 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1337 nam, td);
1338 } else {
1339 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1340 so, nam, td);
1341 }
1342 }
1343 CURVNET_RESTORE();
1344
1345 return (error);
1346 }
1347
1348 int
1349 soconnect2(struct socket *so1, struct socket *so2)
1350 {
1351 int error;
1352
1353 CURVNET_SET(so1->so_vnet);
1354 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1355 CURVNET_RESTORE();
1356 return (error);
1357 }
1358
1359 int
1360 sodisconnect(struct socket *so)
1361 {
1362 int error;
1363
1364 if ((so->so_state & SS_ISCONNECTED) == 0)
1365 return (ENOTCONN);
1366 if (so->so_state & SS_ISDISCONNECTING)
1367 return (EALREADY);
1368 VNET_SO_ASSERT(so);
1369 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1370 return (error);
1371 }
1372
1373 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1374
1375 int
1376 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1377 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1378 {
1379 long space;
1380 ssize_t resid;
1381 int clen = 0, error, dontroute;
1382
1383 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1384 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1385 ("sosend_dgram: !PR_ATOMIC"));
1386
1387 if (uio != NULL)
1388 resid = uio->uio_resid;
1389 else
1390 resid = top->m_pkthdr.len;
1391 /*
1392 * In theory resid should be unsigned. However, space must be
1393 * signed, as it might be less than 0 if we over-committed, and we
1394 * must use a signed comparison of space and resid. On the other
1395 * hand, a negative resid causes us to loop sending 0-length
1396 * segments to the protocol.
1397 */
1398 if (resid < 0) {
1399 error = EINVAL;
1400 goto out;
1401 }
1402
1403 dontroute =
1404 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1405 if (td != NULL)
1406 td->td_ru.ru_msgsnd++;
1407 if (control != NULL)
1408 clen = control->m_len;
1409
1410 SOCKBUF_LOCK(&so->so_snd);
1411 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1412 SOCKBUF_UNLOCK(&so->so_snd);
1413 error = EPIPE;
1414 goto out;
1415 }
1416 if (so->so_error) {
1417 error = so->so_error;
1418 so->so_error = 0;
1419 SOCKBUF_UNLOCK(&so->so_snd);
1420 goto out;
1421 }
1422 if ((so->so_state & SS_ISCONNECTED) == 0) {
1423 /*
1424 * `sendto' and `sendmsg' is allowed on a connection-based
1425 * socket if it supports implied connect. Return ENOTCONN if
1426 * not connected and no address is supplied.
1427 */
1428 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1429 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1430 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1431 !(resid == 0 && clen != 0)) {
1432 SOCKBUF_UNLOCK(&so->so_snd);
1433 error = ENOTCONN;
1434 goto out;
1435 }
1436 } else if (addr == NULL) {
1437 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1438 error = ENOTCONN;
1439 else
1440 error = EDESTADDRREQ;
1441 SOCKBUF_UNLOCK(&so->so_snd);
1442 goto out;
1443 }
1444 }
1445
1446 /*
1447 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1448 * problem and need fixing.
1449 */
1450 space = sbspace(&so->so_snd);
1451 if (flags & MSG_OOB)
1452 space += 1024;
1453 space -= clen;
1454 SOCKBUF_UNLOCK(&so->so_snd);
1455 if (resid > space) {
1456 error = EMSGSIZE;
1457 goto out;
1458 }
1459 if (uio == NULL) {
1460 resid = 0;
1461 if (flags & MSG_EOR)
1462 top->m_flags |= M_EOR;
1463 } else {
1464 /*
1465 * Copy the data from userland into a mbuf chain.
1466 * If no data is to be copied in, a single empty mbuf
1467 * is returned.
1468 */
1469 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1470 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1471 if (top == NULL) {
1472 error = EFAULT; /* only possible error */
1473 goto out;
1474 }
1475 space -= resid - uio->uio_resid;
1476 resid = uio->uio_resid;
1477 }
1478 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1479 /*
1480 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1481 * than with.
1482 */
1483 if (dontroute) {
1484 SOCK_LOCK(so);
1485 so->so_options |= SO_DONTROUTE;
1486 SOCK_UNLOCK(so);
1487 }
1488 /*
1489 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1490 * of date. We could have received a reset packet in an interrupt or
1491 * maybe we slept while doing page faults in uiomove() etc. We could
1492 * probably recheck again inside the locking protection here, but
1493 * there are probably other places that this also happens. We must
1494 * rethink this.
1495 */
1496 VNET_SO_ASSERT(so);
1497 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1498 (flags & MSG_OOB) ? PRUS_OOB :
1499 /*
1500 * If the user set MSG_EOF, the protocol understands this flag and
1501 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1502 */
1503 ((flags & MSG_EOF) &&
1504 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1505 (resid <= 0)) ?
1506 PRUS_EOF :
1507 /* If there is more to send set PRUS_MORETOCOME */
1508 (flags & MSG_MORETOCOME) ||
1509 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1510 top, addr, control, td);
1511 if (dontroute) {
1512 SOCK_LOCK(so);
1513 so->so_options &= ~SO_DONTROUTE;
1514 SOCK_UNLOCK(so);
1515 }
1516 clen = 0;
1517 control = NULL;
1518 top = NULL;
1519 out:
1520 if (top != NULL)
1521 m_freem(top);
1522 if (control != NULL)
1523 m_freem(control);
1524 return (error);
1525 }
1526
1527 /*
1528 * Send on a socket. If send must go all at once and message is larger than
1529 * send buffering, then hard error. Lock against other senders. If must go
1530 * all at once and not enough room now, then inform user that this would
1531 * block and do nothing. Otherwise, if nonblocking, send as much as
1532 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1533 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1534 * in mbuf chain must be small enough to send all at once.
1535 *
1536 * Returns nonzero on error, timeout or signal; callers must check for short
1537 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1538 * on return.
1539 */
1540 int
1541 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1542 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1543 {
1544 long space;
1545 ssize_t resid;
1546 int clen = 0, error, dontroute;
1547 int atomic = sosendallatonce(so) || top;
1548 int pru_flag;
1549 #ifdef KERN_TLS
1550 struct ktls_session *tls;
1551 int tls_enq_cnt, tls_pruflag;
1552 uint8_t tls_rtype;
1553
1554 tls = NULL;
1555 tls_rtype = TLS_RLTYPE_APP;
1556 #endif
1557 if (uio != NULL)
1558 resid = uio->uio_resid;
1559 else if ((top->m_flags & M_PKTHDR) != 0)
1560 resid = top->m_pkthdr.len;
1561 else
1562 resid = m_length(top, NULL);
1563 /*
1564 * In theory resid should be unsigned. However, space must be
1565 * signed, as it might be less than 0 if we over-committed, and we
1566 * must use a signed comparison of space and resid. On the other
1567 * hand, a negative resid causes us to loop sending 0-length
1568 * segments to the protocol.
1569 *
1570 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1571 * type sockets since that's an error.
1572 */
1573 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1574 error = EINVAL;
1575 goto out;
1576 }
1577
1578 dontroute =
1579 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1580 (so->so_proto->pr_flags & PR_ATOMIC);
1581 if (td != NULL)
1582 td->td_ru.ru_msgsnd++;
1583 if (control != NULL)
1584 clen = control->m_len;
1585
1586 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1587 if (error)
1588 goto out;
1589
1590 #ifdef KERN_TLS
1591 tls_pruflag = 0;
1592 tls = ktls_hold(so->so_snd.sb_tls_info);
1593 if (tls != NULL) {
1594 if (tls->mode == TCP_TLS_MODE_SW)
1595 tls_pruflag = PRUS_NOTREADY;
1596
1597 if (control != NULL) {
1598 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1599
1600 if (clen >= sizeof(*cm) &&
1601 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1602 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1603 clen = 0;
1604 m_freem(control);
1605 control = NULL;
1606 atomic = 1;
1607 }
1608 }
1609 }
1610 #endif
1611
1612 restart:
1613 do {
1614 SOCKBUF_LOCK(&so->so_snd);
1615 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1616 SOCKBUF_UNLOCK(&so->so_snd);
1617 error = EPIPE;
1618 goto release;
1619 }
1620 if (so->so_error) {
1621 error = so->so_error;
1622 so->so_error = 0;
1623 SOCKBUF_UNLOCK(&so->so_snd);
1624 goto release;
1625 }
1626 if ((so->so_state & SS_ISCONNECTED) == 0) {
1627 /*
1628 * `sendto' and `sendmsg' is allowed on a connection-
1629 * based socket if it supports implied connect.
1630 * Return ENOTCONN if not connected and no address is
1631 * supplied.
1632 */
1633 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1634 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1635 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1636 !(resid == 0 && clen != 0)) {
1637 SOCKBUF_UNLOCK(&so->so_snd);
1638 error = ENOTCONN;
1639 goto release;
1640 }
1641 } else if (addr == NULL) {
1642 SOCKBUF_UNLOCK(&so->so_snd);
1643 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1644 error = ENOTCONN;
1645 else
1646 error = EDESTADDRREQ;
1647 goto release;
1648 }
1649 }
1650 space = sbspace(&so->so_snd);
1651 if (flags & MSG_OOB)
1652 space += 1024;
1653 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1654 clen > so->so_snd.sb_hiwat) {
1655 SOCKBUF_UNLOCK(&so->so_snd);
1656 error = EMSGSIZE;
1657 goto release;
1658 }
1659 if (space < resid + clen &&
1660 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1661 if ((so->so_state & SS_NBIO) ||
1662 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1663 SOCKBUF_UNLOCK(&so->so_snd);
1664 error = EWOULDBLOCK;
1665 goto release;
1666 }
1667 error = sbwait(&so->so_snd);
1668 SOCKBUF_UNLOCK(&so->so_snd);
1669 if (error)
1670 goto release;
1671 goto restart;
1672 }
1673 SOCKBUF_UNLOCK(&so->so_snd);
1674 space -= clen;
1675 do {
1676 if (uio == NULL) {
1677 resid = 0;
1678 if (flags & MSG_EOR)
1679 top->m_flags |= M_EOR;
1680 #ifdef KERN_TLS
1681 if (tls != NULL) {
1682 ktls_frame(top, tls, &tls_enq_cnt,
1683 tls_rtype);
1684 tls_rtype = TLS_RLTYPE_APP;
1685 }
1686 #endif
1687 } else {
1688 /*
1689 * Copy the data from userland into a mbuf
1690 * chain. If resid is 0, which can happen
1691 * only if we have control to send, then
1692 * a single empty mbuf is returned. This
1693 * is a workaround to prevent protocol send
1694 * methods to panic.
1695 */
1696 #ifdef KERN_TLS
1697 if (tls != NULL) {
1698 top = m_uiotombuf(uio, M_WAITOK, space,
1699 tls->params.max_frame_len,
1700 M_EXTPG |
1701 ((flags & MSG_EOR) ? M_EOR : 0));
1702 if (top != NULL) {
1703 ktls_frame(top, tls,
1704 &tls_enq_cnt, tls_rtype);
1705 }
1706 tls_rtype = TLS_RLTYPE_APP;
1707 } else
1708 #endif
1709 top = m_uiotombuf(uio, M_WAITOK, space,
1710 (atomic ? max_hdr : 0),
1711 (atomic ? M_PKTHDR : 0) |
1712 ((flags & MSG_EOR) ? M_EOR : 0));
1713 if (top == NULL) {
1714 error = EFAULT; /* only possible error */
1715 goto release;
1716 }
1717 space -= resid - uio->uio_resid;
1718 resid = uio->uio_resid;
1719 }
1720 if (dontroute) {
1721 SOCK_LOCK(so);
1722 so->so_options |= SO_DONTROUTE;
1723 SOCK_UNLOCK(so);
1724 }
1725 /*
1726 * XXX all the SBS_CANTSENDMORE checks previously
1727 * done could be out of date. We could have received
1728 * a reset packet in an interrupt or maybe we slept
1729 * while doing page faults in uiomove() etc. We
1730 * could probably recheck again inside the locking
1731 * protection here, but there are probably other
1732 * places that this also happens. We must rethink
1733 * this.
1734 */
1735 VNET_SO_ASSERT(so);
1736
1737 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1738 /*
1739 * If the user set MSG_EOF, the protocol understands
1740 * this flag and nothing left to send then use
1741 * PRU_SEND_EOF instead of PRU_SEND.
1742 */
1743 ((flags & MSG_EOF) &&
1744 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1745 (resid <= 0)) ?
1746 PRUS_EOF :
1747 /* If there is more to send set PRUS_MORETOCOME. */
1748 (flags & MSG_MORETOCOME) ||
1749 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1750
1751 #ifdef KERN_TLS
1752 pru_flag |= tls_pruflag;
1753 #endif
1754
1755 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1756 pru_flag, top, addr, control, td);
1757
1758 if (dontroute) {
1759 SOCK_LOCK(so);
1760 so->so_options &= ~SO_DONTROUTE;
1761 SOCK_UNLOCK(so);
1762 }
1763
1764 #ifdef KERN_TLS
1765 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1766 /*
1767 * Note that error is intentionally
1768 * ignored.
1769 *
1770 * Like sendfile(), we rely on the
1771 * completion routine (pru_ready())
1772 * to free the mbufs in the event that
1773 * pru_send() encountered an error and
1774 * did not append them to the sockbuf.
1775 */
1776 soref(so);
1777 ktls_enqueue(top, so, tls_enq_cnt);
1778 }
1779 #endif
1780 clen = 0;
1781 control = NULL;
1782 top = NULL;
1783 if (error)
1784 goto release;
1785 } while (resid && space > 0);
1786 } while (resid);
1787
1788 release:
1789 sbunlock(&so->so_snd);
1790 out:
1791 #ifdef KERN_TLS
1792 if (tls != NULL)
1793 ktls_free(tls);
1794 #endif
1795 if (top != NULL)
1796 m_freem(top);
1797 if (control != NULL)
1798 m_freem(control);
1799 return (error);
1800 }
1801
1802 int
1803 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1804 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1805 {
1806 int error;
1807
1808 CURVNET_SET(so->so_vnet);
1809 if (!SOLISTENING(so))
1810 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1811 top, control, flags, td);
1812 else {
1813 m_freem(top);
1814 m_freem(control);
1815 error = ENOTCONN;
1816 }
1817 CURVNET_RESTORE();
1818 return (error);
1819 }
1820
1821 /*
1822 * The part of soreceive() that implements reading non-inline out-of-band
1823 * data from a socket. For more complete comments, see soreceive(), from
1824 * which this code originated.
1825 *
1826 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1827 * unable to return an mbuf chain to the caller.
1828 */
1829 static int
1830 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1831 {
1832 struct protosw *pr = so->so_proto;
1833 struct mbuf *m;
1834 int error;
1835
1836 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1837 VNET_SO_ASSERT(so);
1838
1839 m = m_get(M_WAITOK, MT_DATA);
1840 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1841 if (error)
1842 goto bad;
1843 do {
1844 error = uiomove(mtod(m, void *),
1845 (int) min(uio->uio_resid, m->m_len), uio);
1846 m = m_free(m);
1847 } while (uio->uio_resid && error == 0 && m);
1848 bad:
1849 if (m != NULL)
1850 m_freem(m);
1851 return (error);
1852 }
1853
1854 /*
1855 * Following replacement or removal of the first mbuf on the first mbuf chain
1856 * of a socket buffer, push necessary state changes back into the socket
1857 * buffer so that other consumers see the values consistently. 'nextrecord'
1858 * is the callers locally stored value of the original value of
1859 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1860 * NOTE: 'nextrecord' may be NULL.
1861 */
1862 static __inline void
1863 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1864 {
1865
1866 SOCKBUF_LOCK_ASSERT(sb);
1867 /*
1868 * First, update for the new value of nextrecord. If necessary, make
1869 * it the first record.
1870 */
1871 if (sb->sb_mb != NULL)
1872 sb->sb_mb->m_nextpkt = nextrecord;
1873 else
1874 sb->sb_mb = nextrecord;
1875
1876 /*
1877 * Now update any dependent socket buffer fields to reflect the new
1878 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1879 * addition of a second clause that takes care of the case where
1880 * sb_mb has been updated, but remains the last record.
1881 */
1882 if (sb->sb_mb == NULL) {
1883 sb->sb_mbtail = NULL;
1884 sb->sb_lastrecord = NULL;
1885 } else if (sb->sb_mb->m_nextpkt == NULL)
1886 sb->sb_lastrecord = sb->sb_mb;
1887 }
1888
1889 /*
1890 * Implement receive operations on a socket. We depend on the way that
1891 * records are added to the sockbuf by sbappend. In particular, each record
1892 * (mbufs linked through m_next) must begin with an address if the protocol
1893 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1894 * data, and then zero or more mbufs of data. In order to allow parallelism
1895 * between network receive and copying to user space, as well as avoid
1896 * sleeping with a mutex held, we release the socket buffer mutex during the
1897 * user space copy. Although the sockbuf is locked, new data may still be
1898 * appended, and thus we must maintain consistency of the sockbuf during that
1899 * time.
1900 *
1901 * The caller may receive the data as a single mbuf chain by supplying an
1902 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1903 * the count in uio_resid.
1904 */
1905 int
1906 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1907 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1908 {
1909 struct mbuf *m, **mp;
1910 int flags, error, offset;
1911 ssize_t len;
1912 struct protosw *pr = so->so_proto;
1913 struct mbuf *nextrecord;
1914 int moff, type = 0;
1915 ssize_t orig_resid = uio->uio_resid;
1916
1917 mp = mp0;
1918 if (psa != NULL)
1919 *psa = NULL;
1920 if (controlp != NULL)
1921 *controlp = NULL;
1922 if (flagsp != NULL)
1923 flags = *flagsp &~ MSG_EOR;
1924 else
1925 flags = 0;
1926 if (flags & MSG_OOB)
1927 return (soreceive_rcvoob(so, uio, flags));
1928 if (mp != NULL)
1929 *mp = NULL;
1930 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1931 && uio->uio_resid) {
1932 VNET_SO_ASSERT(so);
1933 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1934 }
1935
1936 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1937 if (error)
1938 return (error);
1939
1940 restart:
1941 SOCKBUF_LOCK(&so->so_rcv);
1942 m = so->so_rcv.sb_mb;
1943 /*
1944 * If we have less data than requested, block awaiting more (subject
1945 * to any timeout) if:
1946 * 1. the current count is less than the low water mark, or
1947 * 2. MSG_DONTWAIT is not set
1948 */
1949 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1950 sbavail(&so->so_rcv) < uio->uio_resid) &&
1951 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1952 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1953 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1954 ("receive: m == %p sbavail == %u",
1955 m, sbavail(&so->so_rcv)));
1956 if (so->so_error) {
1957 if (m != NULL)
1958 goto dontblock;
1959 error = so->so_error;
1960 if ((flags & MSG_PEEK) == 0)
1961 so->so_error = 0;
1962 SOCKBUF_UNLOCK(&so->so_rcv);
1963 goto release;
1964 }
1965 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1966 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1967 if (m != NULL)
1968 goto dontblock;
1969 #ifdef KERN_TLS
1970 else if (so->so_rcv.sb_tlsdcc == 0 &&
1971 so->so_rcv.sb_tlscc == 0) {
1972 #else
1973 else {
1974 #endif
1975 SOCKBUF_UNLOCK(&so->so_rcv);
1976 goto release;
1977 }
1978 }
1979 for (; m != NULL; m = m->m_next)
1980 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1981 m = so->so_rcv.sb_mb;
1982 goto dontblock;
1983 }
1984 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1985 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1986 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1987 SOCKBUF_UNLOCK(&so->so_rcv);
1988 error = ENOTCONN;
1989 goto release;
1990 }
1991 if (uio->uio_resid == 0) {
1992 SOCKBUF_UNLOCK(&so->so_rcv);
1993 goto release;
1994 }
1995 if ((so->so_state & SS_NBIO) ||
1996 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1997 SOCKBUF_UNLOCK(&so->so_rcv);
1998 error = EWOULDBLOCK;
1999 goto release;
2000 }
2001 SBLASTRECORDCHK(&so->so_rcv);
2002 SBLASTMBUFCHK(&so->so_rcv);
2003 error = sbwait(&so->so_rcv);
2004 SOCKBUF_UNLOCK(&so->so_rcv);
2005 if (error)
2006 goto release;
2007 goto restart;
2008 }
2009 dontblock:
2010 /*
2011 * From this point onward, we maintain 'nextrecord' as a cache of the
2012 * pointer to the next record in the socket buffer. We must keep the
2013 * various socket buffer pointers and local stack versions of the
2014 * pointers in sync, pushing out modifications before dropping the
2015 * socket buffer mutex, and re-reading them when picking it up.
2016 *
2017 * Otherwise, we will race with the network stack appending new data
2018 * or records onto the socket buffer by using inconsistent/stale
2019 * versions of the field, possibly resulting in socket buffer
2020 * corruption.
2021 *
2022 * By holding the high-level sblock(), we prevent simultaneous
2023 * readers from pulling off the front of the socket buffer.
2024 */
2025 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2026 if (uio->uio_td)
2027 uio->uio_td->td_ru.ru_msgrcv++;
2028 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2029 SBLASTRECORDCHK(&so->so_rcv);
2030 SBLASTMBUFCHK(&so->so_rcv);
2031 nextrecord = m->m_nextpkt;
2032 if (pr->pr_flags & PR_ADDR) {
2033 KASSERT(m->m_type == MT_SONAME,
2034 ("m->m_type == %d", m->m_type));
2035 orig_resid = 0;
2036 if (psa != NULL)
2037 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2038 M_NOWAIT);
2039 if (flags & MSG_PEEK) {
2040 m = m->m_next;
2041 } else {
2042 sbfree(&so->so_rcv, m);
2043 so->so_rcv.sb_mb = m_free(m);
2044 m = so->so_rcv.sb_mb;
2045 sockbuf_pushsync(&so->so_rcv, nextrecord);
2046 }
2047 }
2048
2049 /*
2050 * Process one or more MT_CONTROL mbufs present before any data mbufs
2051 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2052 * just copy the data; if !MSG_PEEK, we call into the protocol to
2053 * perform externalization (or freeing if controlp == NULL).
2054 */
2055 if (m != NULL && m->m_type == MT_CONTROL) {
2056 struct mbuf *cm = NULL, *cmn;
2057 struct mbuf **cme = &cm;
2058 #ifdef KERN_TLS
2059 struct cmsghdr *cmsg;
2060 struct tls_get_record tgr;
2061
2062 /*
2063 * For MSG_TLSAPPDATA, check for a non-application data
2064 * record. If found, return ENXIO without removing
2065 * it from the receive queue. This allows a subsequent
2066 * call without MSG_TLSAPPDATA to receive it.
2067 * Note that, for TLS, there should only be a single
2068 * control mbuf with the TLS_GET_RECORD message in it.
2069 */
2070 if (flags & MSG_TLSAPPDATA) {
2071 cmsg = mtod(m, struct cmsghdr *);
2072 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2073 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2074 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2075 /* This will need to change for TLS 1.3. */
2076 if (tgr.tls_type != TLS_RLTYPE_APP) {
2077 SOCKBUF_UNLOCK(&so->so_rcv);
2078 error = ENXIO;
2079 goto release;
2080 }
2081 }
2082 }
2083 #endif
2084
2085 do {
2086 if (flags & MSG_PEEK) {
2087 if (controlp != NULL) {
2088 *controlp = m_copym(m, 0, m->m_len,
2089 M_NOWAIT);
2090 controlp = &(*controlp)->m_next;
2091 }
2092 m = m->m_next;
2093 } else {
2094 sbfree(&so->so_rcv, m);
2095 so->so_rcv.sb_mb = m->m_next;
2096 m->m_next = NULL;
2097 *cme = m;
2098 cme = &(*cme)->m_next;
2099 m = so->so_rcv.sb_mb;
2100 }
2101 } while (m != NULL && m->m_type == MT_CONTROL);
2102 if ((flags & MSG_PEEK) == 0)
2103 sockbuf_pushsync(&so->so_rcv, nextrecord);
2104 while (cm != NULL) {
2105 cmn = cm->m_next;
2106 cm->m_next = NULL;
2107 if (pr->pr_domain->dom_externalize != NULL) {
2108 SOCKBUF_UNLOCK(&so->so_rcv);
2109 VNET_SO_ASSERT(so);
2110 error = (*pr->pr_domain->dom_externalize)
2111 (cm, controlp, flags);
2112 SOCKBUF_LOCK(&so->so_rcv);
2113 } else if (controlp != NULL)
2114 *controlp = cm;
2115 else
2116 m_freem(cm);
2117 if (controlp != NULL) {
2118 orig_resid = 0;
2119 while (*controlp != NULL)
2120 controlp = &(*controlp)->m_next;
2121 }
2122 cm = cmn;
2123 }
2124 if (m != NULL)
2125 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2126 else
2127 nextrecord = so->so_rcv.sb_mb;
2128 orig_resid = 0;
2129 }
2130 if (m != NULL) {
2131 if ((flags & MSG_PEEK) == 0) {
2132 KASSERT(m->m_nextpkt == nextrecord,
2133 ("soreceive: post-control, nextrecord !sync"));
2134 if (nextrecord == NULL) {
2135 KASSERT(so->so_rcv.sb_mb == m,
2136 ("soreceive: post-control, sb_mb!=m"));
2137 KASSERT(so->so_rcv.sb_lastrecord == m,
2138 ("soreceive: post-control, lastrecord!=m"));
2139 }
2140 }
2141 type = m->m_type;
2142 if (type == MT_OOBDATA)
2143 flags |= MSG_OOB;
2144 } else {
2145 if ((flags & MSG_PEEK) == 0) {
2146 KASSERT(so->so_rcv.sb_mb == nextrecord,
2147 ("soreceive: sb_mb != nextrecord"));
2148 if (so->so_rcv.sb_mb == NULL) {
2149 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2150 ("soreceive: sb_lastercord != NULL"));
2151 }
2152 }
2153 }
2154 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2155 SBLASTRECORDCHK(&so->so_rcv);
2156 SBLASTMBUFCHK(&so->so_rcv);
2157
2158 /*
2159 * Now continue to read any data mbufs off of the head of the socket
2160 * buffer until the read request is satisfied. Note that 'type' is
2161 * used to store the type of any mbuf reads that have happened so far
2162 * such that soreceive() can stop reading if the type changes, which
2163 * causes soreceive() to return only one of regular data and inline
2164 * out-of-band data in a single socket receive operation.
2165 */
2166 moff = 0;
2167 offset = 0;
2168 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2169 && error == 0) {
2170 /*
2171 * If the type of mbuf has changed since the last mbuf
2172 * examined ('type'), end the receive operation.
2173 */
2174 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2175 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2176 if (type != m->m_type)
2177 break;
2178 } else if (type == MT_OOBDATA)
2179 break;
2180 else
2181 KASSERT(m->m_type == MT_DATA,
2182 ("m->m_type == %d", m->m_type));
2183 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2184 len = uio->uio_resid;
2185 if (so->so_oobmark && len > so->so_oobmark - offset)
2186 len = so->so_oobmark - offset;
2187 if (len > m->m_len - moff)
2188 len = m->m_len - moff;
2189 /*
2190 * If mp is set, just pass back the mbufs. Otherwise copy
2191 * them out via the uio, then free. Sockbuf must be
2192 * consistent here (points to current mbuf, it points to next
2193 * record) when we drop priority; we must note any additions
2194 * to the sockbuf when we block interrupts again.
2195 */
2196 if (mp == NULL) {
2197 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2198 SBLASTRECORDCHK(&so->so_rcv);
2199 SBLASTMBUFCHK(&so->so_rcv);
2200 SOCKBUF_UNLOCK(&so->so_rcv);
2201 if ((m->m_flags & M_EXTPG) != 0)
2202 error = m_unmappedtouio(m, moff, uio, (int)len);
2203 else
2204 error = uiomove(mtod(m, char *) + moff,
2205 (int)len, uio);
2206 SOCKBUF_LOCK(&so->so_rcv);
2207 if (error) {
2208 /*
2209 * The MT_SONAME mbuf has already been removed
2210 * from the record, so it is necessary to
2211 * remove the data mbufs, if any, to preserve
2212 * the invariant in the case of PR_ADDR that
2213 * requires MT_SONAME mbufs at the head of
2214 * each record.
2215 */
2216 if (pr->pr_flags & PR_ATOMIC &&
2217 ((flags & MSG_PEEK) == 0))
2218 (void)sbdroprecord_locked(&so->so_rcv);
2219 SOCKBUF_UNLOCK(&so->so_rcv);
2220 goto release;
2221 }
2222 } else
2223 uio->uio_resid -= len;
2224 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2225 if (len == m->m_len - moff) {
2226 if (m->m_flags & M_EOR)
2227 flags |= MSG_EOR;
2228 if (flags & MSG_PEEK) {
2229 m = m->m_next;
2230 moff = 0;
2231 } else {
2232 nextrecord = m->m_nextpkt;
2233 sbfree(&so->so_rcv, m);
2234 if (mp != NULL) {
2235 m->m_nextpkt = NULL;
2236 *mp = m;
2237 mp = &m->m_next;
2238 so->so_rcv.sb_mb = m = m->m_next;
2239 *mp = NULL;
2240 } else {
2241 so->so_rcv.sb_mb = m_free(m);
2242 m = so->so_rcv.sb_mb;
2243 }
2244 sockbuf_pushsync(&so->so_rcv, nextrecord);
2245 SBLASTRECORDCHK(&so->so_rcv);
2246 SBLASTMBUFCHK(&so->so_rcv);
2247 }
2248 } else {
2249 if (flags & MSG_PEEK)
2250 moff += len;
2251 else {
2252 if (mp != NULL) {
2253 if (flags & MSG_DONTWAIT) {
2254 *mp = m_copym(m, 0, len,
2255 M_NOWAIT);
2256 if (*mp == NULL) {
2257 /*
2258 * m_copym() couldn't
2259 * allocate an mbuf.
2260 * Adjust uio_resid back
2261 * (it was adjusted
2262 * down by len bytes,
2263 * which we didn't end
2264 * up "copying" over).
2265 */
2266 uio->uio_resid += len;
2267 break;
2268 }
2269 } else {
2270 SOCKBUF_UNLOCK(&so->so_rcv);
2271 *mp = m_copym(m, 0, len,
2272 M_WAITOK);
2273 SOCKBUF_LOCK(&so->so_rcv);
2274 }
2275 }
2276 sbcut_locked(&so->so_rcv, len);
2277 }
2278 }
2279 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2280 if (so->so_oobmark) {
2281 if ((flags & MSG_PEEK) == 0) {
2282 so->so_oobmark -= len;
2283 if (so->so_oobmark == 0) {
2284 so->so_rcv.sb_state |= SBS_RCVATMARK;
2285 break;
2286 }
2287 } else {
2288 offset += len;
2289 if (offset == so->so_oobmark)
2290 break;
2291 }
2292 }
2293 if (flags & MSG_EOR)
2294 break;
2295 /*
2296 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2297 * must not quit until "uio->uio_resid == 0" or an error
2298 * termination. If a signal/timeout occurs, return with a
2299 * short count but without error. Keep sockbuf locked
2300 * against other readers.
2301 */
2302 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2303 !sosendallatonce(so) && nextrecord == NULL) {
2304 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2305 if (so->so_error ||
2306 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2307 break;
2308 /*
2309 * Notify the protocol that some data has been
2310 * drained before blocking.
2311 */
2312 if (pr->pr_flags & PR_WANTRCVD) {
2313 SOCKBUF_UNLOCK(&so->so_rcv);
2314 VNET_SO_ASSERT(so);
2315 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2316 SOCKBUF_LOCK(&so->so_rcv);
2317 }
2318 SBLASTRECORDCHK(&so->so_rcv);
2319 SBLASTMBUFCHK(&so->so_rcv);
2320 /*
2321 * We could receive some data while was notifying
2322 * the protocol. Skip blocking in this case.
2323 */
2324 if (so->so_rcv.sb_mb == NULL) {
2325 error = sbwait(&so->so_rcv);
2326 if (error) {
2327 SOCKBUF_UNLOCK(&so->so_rcv);
2328 goto release;
2329 }
2330 }
2331 m = so->so_rcv.sb_mb;
2332 if (m != NULL)
2333 nextrecord = m->m_nextpkt;
2334 }
2335 }
2336
2337 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2338 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2339 flags |= MSG_TRUNC;
2340 if ((flags & MSG_PEEK) == 0)
2341 (void) sbdroprecord_locked(&so->so_rcv);
2342 }
2343 if ((flags & MSG_PEEK) == 0) {
2344 if (m == NULL) {
2345 /*
2346 * First part is an inline SB_EMPTY_FIXUP(). Second
2347 * part makes sure sb_lastrecord is up-to-date if
2348 * there is still data in the socket buffer.
2349 */
2350 so->so_rcv.sb_mb = nextrecord;
2351 if (so->so_rcv.sb_mb == NULL) {
2352 so->so_rcv.sb_mbtail = NULL;
2353 so->so_rcv.sb_lastrecord = NULL;
2354 } else if (nextrecord->m_nextpkt == NULL)
2355 so->so_rcv.sb_lastrecord = nextrecord;
2356 }
2357 SBLASTRECORDCHK(&so->so_rcv);
2358 SBLASTMBUFCHK(&so->so_rcv);
2359 /*
2360 * If soreceive() is being done from the socket callback,
2361 * then don't need to generate ACK to peer to update window,
2362 * since ACK will be generated on return to TCP.
2363 */
2364 if (!(flags & MSG_SOCALLBCK) &&
2365 (pr->pr_flags & PR_WANTRCVD)) {
2366 SOCKBUF_UNLOCK(&so->so_rcv);
2367 VNET_SO_ASSERT(so);
2368 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2369 SOCKBUF_LOCK(&so->so_rcv);
2370 }
2371 }
2372 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2373 if (orig_resid == uio->uio_resid && orig_resid &&
2374 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2375 SOCKBUF_UNLOCK(&so->so_rcv);
2376 goto restart;
2377 }
2378 SOCKBUF_UNLOCK(&so->so_rcv);
2379
2380 if (flagsp != NULL)
2381 *flagsp |= flags;
2382 release:
2383 sbunlock(&so->so_rcv);
2384 return (error);
2385 }
2386
2387 /*
2388 * Optimized version of soreceive() for stream (TCP) sockets.
2389 */
2390 int
2391 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2392 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2393 {
2394 int len = 0, error = 0, flags, oresid;
2395 struct sockbuf *sb;
2396 struct mbuf *m, *n = NULL;
2397
2398 /* We only do stream sockets. */
2399 if (so->so_type != SOCK_STREAM)
2400 return (EINVAL);
2401 if (psa != NULL)
2402 *psa = NULL;
2403 if (flagsp != NULL)
2404 flags = *flagsp &~ MSG_EOR;
2405 else
2406 flags = 0;
2407 if (controlp != NULL)
2408 *controlp = NULL;
2409 if (flags & MSG_OOB)
2410 return (soreceive_rcvoob(so, uio, flags));
2411 if (mp0 != NULL)
2412 *mp0 = NULL;
2413
2414 sb = &so->so_rcv;
2415
2416 #ifdef KERN_TLS
2417 /*
2418 * KTLS store TLS records as records with a control message to
2419 * describe the framing.
2420 *
2421 * We check once here before acquiring locks to optimize the
2422 * common case.
2423 */
2424 if (sb->sb_tls_info != NULL)
2425 return (soreceive_generic(so, psa, uio, mp0, controlp,
2426 flagsp));
2427 #endif
2428
2429 /* Prevent other readers from entering the socket. */
2430 error = sblock(sb, SBLOCKWAIT(flags));
2431 if (error)
2432 return (error);
2433 SOCKBUF_LOCK(sb);
2434
2435 #ifdef KERN_TLS
2436 if (sb->sb_tls_info != NULL) {
2437 SOCKBUF_UNLOCK(sb);
2438 sbunlock(sb);
2439 return (soreceive_generic(so, psa, uio, mp0, controlp,
2440 flagsp));
2441 }
2442 #endif
2443
2444 /* Easy one, no space to copyout anything. */
2445 if (uio->uio_resid == 0) {
2446 error = EINVAL;
2447 goto out;
2448 }
2449 oresid = uio->uio_resid;
2450
2451 /* We will never ever get anything unless we are or were connected. */
2452 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2453 error = ENOTCONN;
2454 goto out;
2455 }
2456
2457 restart:
2458 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2459
2460 /* Abort if socket has reported problems. */
2461 if (so->so_error) {
2462 if (sbavail(sb) > 0)
2463 goto deliver;
2464 if (oresid > uio->uio_resid)
2465 goto out;
2466 error = so->so_error;
2467 if (!(flags & MSG_PEEK))
2468 so->so_error = 0;
2469 goto out;
2470 }
2471
2472 /* Door is closed. Deliver what is left, if any. */
2473 if (sb->sb_state & SBS_CANTRCVMORE) {
2474 if (sbavail(sb) > 0)
2475 goto deliver;
2476 else
2477 goto out;
2478 }
2479
2480 /* Socket buffer is empty and we shall not block. */
2481 if (sbavail(sb) == 0 &&
2482 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2483 error = EAGAIN;
2484 goto out;
2485 }
2486
2487 /* Socket buffer got some data that we shall deliver now. */
2488 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2489 ((so->so_state & SS_NBIO) ||
2490 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2491 sbavail(sb) >= sb->sb_lowat ||
2492 sbavail(sb) >= uio->uio_resid ||
2493 sbavail(sb) >= sb->sb_hiwat) ) {
2494 goto deliver;
2495 }
2496
2497 /* On MSG_WAITALL we must wait until all data or error arrives. */
2498 if ((flags & MSG_WAITALL) &&
2499 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2500 goto deliver;
2501
2502 /*
2503 * Wait and block until (more) data comes in.
2504 * NB: Drops the sockbuf lock during wait.
2505 */
2506 error = sbwait(sb);
2507 if (error)
2508 goto out;
2509 goto restart;
2510
2511 deliver:
2512 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2513 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2514 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2515
2516 /* Statistics. */
2517 if (uio->uio_td)
2518 uio->uio_td->td_ru.ru_msgrcv++;
2519
2520 /* Fill uio until full or current end of socket buffer is reached. */
2521 len = min(uio->uio_resid, sbavail(sb));
2522 if (mp0 != NULL) {
2523 /* Dequeue as many mbufs as possible. */
2524 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2525 if (*mp0 == NULL)
2526 *mp0 = sb->sb_mb;
2527 else
2528 m_cat(*mp0, sb->sb_mb);
2529 for (m = sb->sb_mb;
2530 m != NULL && m->m_len <= len;
2531 m = m->m_next) {
2532 KASSERT(!(m->m_flags & M_NOTAVAIL),
2533 ("%s: m %p not available", __func__, m));
2534 len -= m->m_len;
2535 uio->uio_resid -= m->m_len;
2536 sbfree(sb, m);
2537 n = m;
2538 }
2539 n->m_next = NULL;
2540 sb->sb_mb = m;
2541 sb->sb_lastrecord = sb->sb_mb;
2542 if (sb->sb_mb == NULL)
2543 SB_EMPTY_FIXUP(sb);
2544 }
2545 /* Copy the remainder. */
2546 if (len > 0) {
2547 KASSERT(sb->sb_mb != NULL,
2548 ("%s: len > 0 && sb->sb_mb empty", __func__));
2549
2550 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2551 if (m == NULL)
2552 len = 0; /* Don't flush data from sockbuf. */
2553 else
2554 uio->uio_resid -= len;
2555 if (*mp0 != NULL)
2556 m_cat(*mp0, m);
2557 else
2558 *mp0 = m;
2559 if (*mp0 == NULL) {
2560 error = ENOBUFS;
2561 goto out;
2562 }
2563 }
2564 } else {
2565 /* NB: Must unlock socket buffer as uiomove may sleep. */
2566 SOCKBUF_UNLOCK(sb);
2567 error = m_mbuftouio(uio, sb->sb_mb, len);
2568 SOCKBUF_LOCK(sb);
2569 if (error)
2570 goto out;
2571 }
2572 SBLASTRECORDCHK(sb);
2573 SBLASTMBUFCHK(sb);
2574
2575 /*
2576 * Remove the delivered data from the socket buffer unless we
2577 * were only peeking.
2578 */
2579 if (!(flags & MSG_PEEK)) {
2580 if (len > 0)
2581 sbdrop_locked(sb, len);
2582
2583 /* Notify protocol that we drained some data. */
2584 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2585 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2586 !(flags & MSG_SOCALLBCK))) {
2587 SOCKBUF_UNLOCK(sb);
2588 VNET_SO_ASSERT(so);
2589 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2590 SOCKBUF_LOCK(sb);
2591 }
2592 }
2593
2594 /*
2595 * For MSG_WAITALL we may have to loop again and wait for
2596 * more data to come in.
2597 */
2598 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2599 goto restart;
2600 out:
2601 SOCKBUF_LOCK_ASSERT(sb);
2602 SBLASTRECORDCHK(sb);
2603 SBLASTMBUFCHK(sb);
2604 SOCKBUF_UNLOCK(sb);
2605 sbunlock(sb);
2606 return (error);
2607 }
2608
2609 /*
2610 * Optimized version of soreceive() for simple datagram cases from userspace.
2611 * Unlike in the stream case, we're able to drop a datagram if copyout()
2612 * fails, and because we handle datagrams atomically, we don't need to use a
2613 * sleep lock to prevent I/O interlacing.
2614 */
2615 int
2616 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2617 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2618 {
2619 struct mbuf *m, *m2;
2620 int flags, error;
2621 ssize_t len;
2622 struct protosw *pr = so->so_proto;
2623 struct mbuf *nextrecord;
2624
2625 if (psa != NULL)
2626 *psa = NULL;
2627 if (controlp != NULL)
2628 *controlp = NULL;
2629 if (flagsp != NULL)
2630 flags = *flagsp &~ MSG_EOR;
2631 else
2632 flags = 0;
2633
2634 /*
2635 * For any complicated cases, fall back to the full
2636 * soreceive_generic().
2637 */
2638 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2639 return (soreceive_generic(so, psa, uio, mp0, controlp,
2640 flagsp));
2641
2642 /*
2643 * Enforce restrictions on use.
2644 */
2645 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2646 ("soreceive_dgram: wantrcvd"));
2647 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2648 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2649 ("soreceive_dgram: SBS_RCVATMARK"));
2650 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2651 ("soreceive_dgram: P_CONNREQUIRED"));
2652
2653 /*
2654 * Loop blocking while waiting for a datagram.
2655 */
2656 SOCKBUF_LOCK(&so->so_rcv);
2657 while ((m = so->so_rcv.sb_mb) == NULL) {
2658 KASSERT(sbavail(&so->so_rcv) == 0,
2659 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2660 sbavail(&so->so_rcv)));
2661 if (so->so_error) {
2662 error = so->so_error;
2663 so->so_error = 0;
2664 SOCKBUF_UNLOCK(&so->so_rcv);
2665 return (error);
2666 }
2667 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2668 uio->uio_resid == 0) {
2669 SOCKBUF_UNLOCK(&so->so_rcv);
2670 return (0);
2671 }
2672 if ((so->so_state & SS_NBIO) ||
2673 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2674 SOCKBUF_UNLOCK(&so->so_rcv);
2675 return (EWOULDBLOCK);
2676 }
2677 SBLASTRECORDCHK(&so->so_rcv);
2678 SBLASTMBUFCHK(&so->so_rcv);
2679 error = sbwait(&so->so_rcv);
2680 if (error) {
2681 SOCKBUF_UNLOCK(&so->so_rcv);
2682 return (error);
2683 }
2684 }
2685 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2686
2687 if (uio->uio_td)
2688 uio->uio_td->td_ru.ru_msgrcv++;
2689 SBLASTRECORDCHK(&so->so_rcv);
2690 SBLASTMBUFCHK(&so->so_rcv);
2691 nextrecord = m->m_nextpkt;
2692 if (nextrecord == NULL) {
2693 KASSERT(so->so_rcv.sb_lastrecord == m,
2694 ("soreceive_dgram: lastrecord != m"));
2695 }
2696
2697 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2698 ("soreceive_dgram: m_nextpkt != nextrecord"));
2699
2700 /*
2701 * Pull 'm' and its chain off the front of the packet queue.
2702 */
2703 so->so_rcv.sb_mb = NULL;
2704 sockbuf_pushsync(&so->so_rcv, nextrecord);
2705
2706 /*
2707 * Walk 'm's chain and free that many bytes from the socket buffer.
2708 */
2709 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2710 sbfree(&so->so_rcv, m2);
2711
2712 /*
2713 * Do a few last checks before we let go of the lock.
2714 */
2715 SBLASTRECORDCHK(&so->so_rcv);
2716 SBLASTMBUFCHK(&so->so_rcv);
2717 SOCKBUF_UNLOCK(&so->so_rcv);
2718
2719 if (pr->pr_flags & PR_ADDR) {
2720 KASSERT(m->m_type == MT_SONAME,
2721 ("m->m_type == %d", m->m_type));
2722 if (psa != NULL)
2723 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2724 M_NOWAIT);
2725 m = m_free(m);
2726 }
2727 if (m == NULL) {
2728 /* XXXRW: Can this happen? */
2729 return (0);
2730 }
2731
2732 /*
2733 * Packet to copyout() is now in 'm' and it is disconnected from the
2734 * queue.
2735 *
2736 * Process one or more MT_CONTROL mbufs present before any data mbufs
2737 * in the first mbuf chain on the socket buffer. We call into the
2738 * protocol to perform externalization (or freeing if controlp ==
2739 * NULL). In some cases there can be only MT_CONTROL mbufs without
2740 * MT_DATA mbufs.
2741 */
2742 if (m->m_type == MT_CONTROL) {
2743 struct mbuf *cm = NULL, *cmn;
2744 struct mbuf **cme = &cm;
2745
2746 do {
2747 m2 = m->m_next;
2748 m->m_next = NULL;
2749 *cme = m;
2750 cme = &(*cme)->m_next;
2751 m = m2;
2752 } while (m != NULL && m->m_type == MT_CONTROL);
2753 while (cm != NULL) {
2754 cmn = cm->m_next;
2755 cm->m_next = NULL;
2756 if (pr->pr_domain->dom_externalize != NULL) {
2757 error = (*pr->pr_domain->dom_externalize)
2758 (cm, controlp, flags);
2759 } else if (controlp != NULL)
2760 *controlp = cm;
2761 else
2762 m_freem(cm);
2763 if (controlp != NULL) {
2764 while (*controlp != NULL)
2765 controlp = &(*controlp)->m_next;
2766 }
2767 cm = cmn;
2768 }
2769 }
2770 KASSERT(m == NULL || m->m_type == MT_DATA,
2771 ("soreceive_dgram: !data"));
2772 while (m != NULL && uio->uio_resid > 0) {
2773 len = uio->uio_resid;
2774 if (len > m->m_len)
2775 len = m->m_len;
2776 error = uiomove(mtod(m, char *), (int)len, uio);
2777 if (error) {
2778 m_freem(m);
2779 return (error);
2780 }
2781 if (len == m->m_len)
2782 m = m_free(m);
2783 else {
2784 m->m_data += len;
2785 m->m_len -= len;
2786 }
2787 }
2788 if (m != NULL) {
2789 flags |= MSG_TRUNC;
2790 m_freem(m);
2791 }
2792 if (flagsp != NULL)
2793 *flagsp |= flags;
2794 return (0);
2795 }
2796
2797 int
2798 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2799 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2800 {
2801 int error;
2802
2803 CURVNET_SET(so->so_vnet);
2804 if (!SOLISTENING(so))
2805 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2806 mp0, controlp, flagsp));
2807 else
2808 error = ENOTCONN;
2809 CURVNET_RESTORE();
2810 return (error);
2811 }
2812
2813 int
2814 soshutdown(struct socket *so, int how)
2815 {
2816 struct protosw *pr = so->so_proto;
2817 int error, soerror_enotconn;
2818
2819 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2820 return (EINVAL);
2821
2822 soerror_enotconn = 0;
2823 if ((so->so_state &
2824 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2825 /*
2826 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2827 * invoked on a datagram sockets, however historically we would
2828 * actually tear socket down. This is known to be leveraged by
2829 * some applications to unblock process waiting in recvXXX(2)
2830 * by other process that it shares that socket with. Try to meet
2831 * both backward-compatibility and POSIX requirements by forcing
2832 * ENOTCONN but still asking protocol to perform pru_shutdown().
2833 */
2834 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2835 return (ENOTCONN);
2836 soerror_enotconn = 1;
2837 }
2838
2839 if (SOLISTENING(so)) {
2840 if (how != SHUT_WR) {
2841 SOLISTEN_LOCK(so);
2842 so->so_error = ECONNABORTED;
2843 solisten_wakeup(so); /* unlocks so */
2844 }
2845 goto done;
2846 }
2847
2848 CURVNET_SET(so->so_vnet);
2849 if (pr->pr_usrreqs->pru_flush != NULL)
2850 (*pr->pr_usrreqs->pru_flush)(so, how);
2851 if (how != SHUT_WR)
2852 sorflush(so);
2853 if (how != SHUT_RD) {
2854 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2855 wakeup(&so->so_timeo);
2856 CURVNET_RESTORE();
2857 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2858 }
2859 wakeup(&so->so_timeo);
2860 CURVNET_RESTORE();
2861
2862 done:
2863 return (soerror_enotconn ? ENOTCONN : 0);
2864 }
2865
2866 void
2867 sorflush(struct socket *so)
2868 {
2869 struct sockbuf *sb = &so->so_rcv;
2870 struct protosw *pr = so->so_proto;
2871 struct socket aso;
2872
2873 VNET_SO_ASSERT(so);
2874
2875 /*
2876 * In order to avoid calling dom_dispose with the socket buffer mutex
2877 * held, and in order to generally avoid holding the lock for a long
2878 * time, we make a copy of the socket buffer and clear the original
2879 * (except locks, state). The new socket buffer copy won't have
2880 * initialized locks so we can only call routines that won't use or
2881 * assert those locks.
2882 *
2883 * Dislodge threads currently blocked in receive and wait to acquire
2884 * a lock against other simultaneous readers before clearing the
2885 * socket buffer. Don't let our acquire be interrupted by a signal
2886 * despite any existing socket disposition on interruptable waiting.
2887 */
2888 socantrcvmore(so);
2889 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2890
2891 /*
2892 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2893 * and mutex data unchanged.
2894 */
2895 SOCKBUF_LOCK(sb);
2896 bzero(&aso, sizeof(aso));
2897 aso.so_pcb = so->so_pcb;
2898 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2899 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2900 bzero(&sb->sb_startzero,
2901 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2902 SOCKBUF_UNLOCK(sb);
2903 sbunlock(sb);
2904
2905 /*
2906 * Dispose of special rights and flush the copied socket. Don't call
2907 * any unsafe routines (that rely on locks being initialized) on aso.
2908 */
2909 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2910 (*pr->pr_domain->dom_dispose)(&aso);
2911 sbrelease_internal(&aso.so_rcv, so);
2912 }
2913
2914 /*
2915 * Wrapper for Socket established helper hook.
2916 * Parameters: socket, context of the hook point, hook id.
2917 */
2918 static int inline
2919 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2920 {
2921 struct socket_hhook_data hhook_data = {
2922 .so = so,
2923 .hctx = hctx,
2924 .m = NULL,
2925 .status = 0
2926 };
2927
2928 CURVNET_SET(so->so_vnet);
2929 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2930 CURVNET_RESTORE();
2931
2932 /* Ugly but needed, since hhooks return void for now */
2933 return (hhook_data.status);
2934 }
2935
2936 /*
2937 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2938 * additional variant to handle the case where the option value needs to be
2939 * some kind of integer, but not a specific size. In addition to their use
2940 * here, these functions are also called by the protocol-level pr_ctloutput()
2941 * routines.
2942 */
2943 int
2944 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2945 {
2946 size_t valsize;
2947
2948 /*
2949 * If the user gives us more than we wanted, we ignore it, but if we
2950 * don't get the minimum length the caller wants, we return EINVAL.
2951 * On success, sopt->sopt_valsize is set to however much we actually
2952 * retrieved.
2953 */
2954 if ((valsize = sopt->sopt_valsize) < minlen)
2955 return EINVAL;
2956 if (valsize > len)
2957 sopt->sopt_valsize = valsize = len;
2958
2959 if (sopt->sopt_td != NULL)
2960 return (copyin(sopt->sopt_val, buf, valsize));
2961
2962 bcopy(sopt->sopt_val, buf, valsize);
2963 return (0);
2964 }
2965
2966 /*
2967 * Kernel version of setsockopt(2).
2968 *
2969 * XXX: optlen is size_t, not socklen_t
2970 */
2971 int
2972 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2973 size_t optlen)
2974 {
2975 struct sockopt sopt;
2976
2977 sopt.sopt_level = level;
2978 sopt.sopt_name = optname;
2979 sopt.sopt_dir = SOPT_SET;
2980 sopt.sopt_val = optval;
2981 sopt.sopt_valsize = optlen;
2982 sopt.sopt_td = NULL;
2983 return (sosetopt(so, &sopt));
2984 }
2985
2986 int
2987 sosetopt(struct socket *so, struct sockopt *sopt)
2988 {
2989 int error, optval;
2990 struct linger l;
2991 struct timeval tv;
2992 sbintime_t val;
2993 uint32_t val32;
2994 #ifdef MAC
2995 struct mac extmac;
2996 #endif
2997
2998 CURVNET_SET(so->so_vnet);
2999 error = 0;
3000 if (sopt->sopt_level != SOL_SOCKET) {
3001 if (so->so_proto->pr_ctloutput != NULL)
3002 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3003 else
3004 error = ENOPROTOOPT;
3005 } else {
3006 switch (sopt->sopt_name) {
3007 case SO_ACCEPTFILTER:
3008 error = accept_filt_setopt(so, sopt);
3009 if (error)
3010 goto bad;
3011 break;
3012
3013 case SO_LINGER:
3014 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3015 if (error)
3016 goto bad;
3017 if (l.l_linger < 0 ||
3018 l.l_linger > USHRT_MAX ||
3019 l.l_linger > (INT_MAX / hz)) {
3020 error = EDOM;
3021 goto bad;
3022 }
3023 SOCK_LOCK(so);
3024 so->so_linger = l.l_linger;
3025 if (l.l_onoff)
3026 so->so_options |= SO_LINGER;
3027 else
3028 so->so_options &= ~SO_LINGER;
3029 SOCK_UNLOCK(so);
3030 break;
3031
3032 case SO_DEBUG:
3033 case SO_KEEPALIVE:
3034 case SO_DONTROUTE:
3035 case SO_USELOOPBACK:
3036 case SO_BROADCAST:
3037 case SO_REUSEADDR:
3038 case SO_REUSEPORT:
3039 case SO_REUSEPORT_LB:
3040 case SO_OOBINLINE:
3041 case SO_TIMESTAMP:
3042 case SO_BINTIME:
3043 case SO_NOSIGPIPE:
3044 case SO_NO_DDP:
3045 case SO_NO_OFFLOAD:
3046 error = sooptcopyin(sopt, &optval, sizeof optval,
3047 sizeof optval);
3048 if (error)
3049 goto bad;
3050 SOCK_LOCK(so);
3051 if (optval)
3052 so->so_options |= sopt->sopt_name;
3053 else
3054 so->so_options &= ~sopt->sopt_name;
3055 SOCK_UNLOCK(so);
3056 break;
3057
3058 case SO_SETFIB:
3059 error = sooptcopyin(sopt, &optval, sizeof optval,
3060 sizeof optval);
3061 if (error)
3062 goto bad;
3063
3064 if (optval < 0 || optval >= rt_numfibs) {
3065 error = EINVAL;
3066 goto bad;
3067 }
3068 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3069 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3070 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3071 so->so_fibnum = optval;
3072 else
3073 so->so_fibnum = 0;
3074 break;
3075
3076 case SO_USER_COOKIE:
3077 error = sooptcopyin(sopt, &val32, sizeof val32,
3078 sizeof val32);
3079 if (error)
3080 goto bad;
3081 so->so_user_cookie = val32;
3082 break;
3083
3084 case SO_SNDBUF:
3085 case SO_RCVBUF:
3086 case SO_SNDLOWAT:
3087 case SO_RCVLOWAT:
3088 error = sooptcopyin(sopt, &optval, sizeof optval,
3089 sizeof optval);
3090 if (error)
3091 goto bad;
3092
3093 /*
3094 * Values < 1 make no sense for any of these options,
3095 * so disallow them.
3096 */
3097 if (optval < 1) {
3098 error = EINVAL;
3099 goto bad;
3100 }
3101
3102 error = sbsetopt(so, sopt->sopt_name, optval);
3103 break;
3104
3105 case SO_SNDTIMEO:
3106 case SO_RCVTIMEO:
3107 #ifdef COMPAT_FREEBSD32
3108 if (SV_CURPROC_FLAG(SV_ILP32)) {
3109 struct timeval32 tv32;
3110
3111 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3112 sizeof tv32);
3113 CP(tv32, tv, tv_sec);
3114 CP(tv32, tv, tv_usec);
3115 } else
3116 #endif
3117 error = sooptcopyin(sopt, &tv, sizeof tv,
3118 sizeof tv);
3119 if (error)
3120 goto bad;
3121 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3122 tv.tv_usec >= 1000000) {
3123 error = EDOM;
3124 goto bad;
3125 }
3126 if (tv.tv_sec > INT32_MAX)
3127 val = SBT_MAX;
3128 else
3129 val = tvtosbt(tv);
3130 switch (sopt->sopt_name) {
3131 case SO_SNDTIMEO:
3132 so->so_snd.sb_timeo = val;
3133 break;
3134 case SO_RCVTIMEO:
3135 so->so_rcv.sb_timeo = val;
3136 break;
3137 }
3138 break;
3139
3140 case SO_LABEL:
3141 #ifdef MAC
3142 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3143 sizeof extmac);
3144 if (error)
3145 goto bad;
3146 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3147 so, &extmac);
3148 #else
3149 error = EOPNOTSUPP;
3150 #endif
3151 break;
3152
3153 case SO_TS_CLOCK:
3154 error = sooptcopyin(sopt, &optval, sizeof optval,
3155 sizeof optval);
3156 if (error)
3157 goto bad;
3158 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3159 error = EINVAL;
3160 goto bad;
3161 }
3162 so->so_ts_clock = optval;
3163 break;
3164
3165 case SO_MAX_PACING_RATE:
3166 error = sooptcopyin(sopt, &val32, sizeof(val32),
3167 sizeof(val32));
3168 if (error)
3169 goto bad;
3170 so->so_max_pacing_rate = val32;
3171 break;
3172
3173 default:
3174 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3175 error = hhook_run_socket(so, sopt,
3176 HHOOK_SOCKET_OPT);
3177 else
3178 error = ENOPROTOOPT;
3179 break;
3180 }
3181 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3182 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3183 }
3184 bad:
3185 CURVNET_RESTORE();
3186 return (error);
3187 }
3188
3189 /*
3190 * Helper routine for getsockopt.
3191 */
3192 int
3193 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3194 {
3195 int error;
3196 size_t valsize;
3197
3198 error = 0;
3199
3200 /*
3201 * Documented get behavior is that we always return a value, possibly
3202 * truncated to fit in the user's buffer. Traditional behavior is
3203 * that we always tell the user precisely how much we copied, rather
3204 * than something useful like the total amount we had available for
3205 * her. Note that this interface is not idempotent; the entire
3206 * answer must be generated ahead of time.
3207 */
3208 valsize = min(len, sopt->sopt_valsize);
3209 sopt->sopt_valsize = valsize;
3210 if (sopt->sopt_val != NULL) {
3211 if (sopt->sopt_td != NULL)
3212 error = copyout(buf, sopt->sopt_val, valsize);
3213 else
3214 bcopy(buf, sopt->sopt_val, valsize);
3215 }
3216 return (error);
3217 }
3218
3219 int
3220 sogetopt(struct socket *so, struct sockopt *sopt)
3221 {
3222 int error, optval;
3223 struct linger l;
3224 struct timeval tv;
3225 #ifdef MAC
3226 struct mac extmac;
3227 #endif
3228
3229 CURVNET_SET(so->so_vnet);
3230 error = 0;
3231 if (sopt->sopt_level != SOL_SOCKET) {
3232 if (so->so_proto->pr_ctloutput != NULL)
3233 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3234 else
3235 error = ENOPROTOOPT;
3236 CURVNET_RESTORE();
3237 return (error);
3238 } else {
3239 switch (sopt->sopt_name) {
3240 case SO_ACCEPTFILTER:
3241 error = accept_filt_getopt(so, sopt);
3242 break;
3243
3244 case SO_LINGER:
3245 SOCK_LOCK(so);
3246 l.l_onoff = so->so_options & SO_LINGER;
3247 l.l_linger = so->so_linger;
3248 SOCK_UNLOCK(so);
3249 error = sooptcopyout(sopt, &l, sizeof l);
3250 break;
3251
3252 case SO_USELOOPBACK:
3253 case SO_DONTROUTE:
3254 case SO_DEBUG:
3255 case SO_KEEPALIVE:
3256 case SO_REUSEADDR:
3257 case SO_REUSEPORT:
3258 case SO_REUSEPORT_LB:
3259 case SO_BROADCAST:
3260 case SO_OOBINLINE:
3261 case SO_ACCEPTCONN:
3262 case SO_TIMESTAMP:
3263 case SO_BINTIME:
3264 case SO_NOSIGPIPE:
3265 case SO_NO_DDP:
3266 case SO_NO_OFFLOAD:
3267 optval = so->so_options & sopt->sopt_name;
3268 integer:
3269 error = sooptcopyout(sopt, &optval, sizeof optval);
3270 break;
3271
3272 case SO_DOMAIN:
3273 optval = so->so_proto->pr_domain->dom_family;
3274 goto integer;
3275
3276 case SO_TYPE:
3277 optval = so->so_type;
3278 goto integer;
3279
3280 case SO_PROTOCOL:
3281 optval = so->so_proto->pr_protocol;
3282 goto integer;
3283
3284 case SO_ERROR:
3285 SOCK_LOCK(so);
3286 optval = so->so_error;
3287 so->so_error = 0;
3288 SOCK_UNLOCK(so);
3289 goto integer;
3290
3291 case SO_SNDBUF:
3292 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3293 so->so_snd.sb_hiwat;
3294 goto integer;
3295
3296 case SO_RCVBUF:
3297 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3298 so->so_rcv.sb_hiwat;
3299 goto integer;
3300
3301 case SO_SNDLOWAT:
3302 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3303 so->so_snd.sb_lowat;
3304 goto integer;
3305
3306 case SO_RCVLOWAT:
3307 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3308 so->so_rcv.sb_lowat;
3309 goto integer;
3310
3311 case SO_SNDTIMEO:
3312 case SO_RCVTIMEO:
3313 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3314 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3315 #ifdef COMPAT_FREEBSD32
3316 if (SV_CURPROC_FLAG(SV_ILP32)) {
3317 struct timeval32 tv32;
3318
3319 CP(tv, tv32, tv_sec);
3320 CP(tv, tv32, tv_usec);
3321 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3322 } else
3323 #endif
3324 error = sooptcopyout(sopt, &tv, sizeof tv);
3325 break;
3326
3327 case SO_LABEL:
3328 #ifdef MAC
3329 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3330 sizeof(extmac));
3331 if (error)
3332 goto bad;
3333 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3334 so, &extmac);
3335 if (error)
3336 goto bad;
3337 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3338 #else
3339 error = EOPNOTSUPP;
3340 #endif
3341 break;
3342
3343 case SO_PEERLABEL:
3344 #ifdef MAC
3345 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3346 sizeof(extmac));
3347 if (error)
3348 goto bad;
3349 error = mac_getsockopt_peerlabel(
3350 sopt->sopt_td->td_ucred, so, &extmac);
3351 if (error)
3352 goto bad;
3353 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3354 #else
3355 error = EOPNOTSUPP;
3356 #endif
3357 break;
3358
3359 case SO_LISTENQLIMIT:
3360 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3361 goto integer;
3362
3363 case SO_LISTENQLEN:
3364 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3365 goto integer;
3366
3367 case SO_LISTENINCQLEN:
3368 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3369 goto integer;
3370
3371 case SO_TS_CLOCK:
3372 optval = so->so_ts_clock;
3373 goto integer;
3374
3375 case SO_MAX_PACING_RATE:
3376 optval = so->so_max_pacing_rate;
3377 goto integer;
3378
3379 default:
3380 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3381 error = hhook_run_socket(so, sopt,
3382 HHOOK_SOCKET_OPT);
3383 else
3384 error = ENOPROTOOPT;
3385 break;
3386 }
3387 }
3388 #ifdef MAC
3389 bad:
3390 #endif
3391 CURVNET_RESTORE();
3392 return (error);
3393 }
3394
3395 int
3396 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3397 {
3398 struct mbuf *m, *m_prev;
3399 int sopt_size = sopt->sopt_valsize;
3400
3401 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3402 if (m == NULL)
3403 return ENOBUFS;
3404 if (sopt_size > MLEN) {
3405 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3406 if ((m->m_flags & M_EXT) == 0) {
3407 m_free(m);
3408 return ENOBUFS;
3409 }
3410 m->m_len = min(MCLBYTES, sopt_size);
3411 } else {
3412 m->m_len = min(MLEN, sopt_size);
3413 }
3414 sopt_size -= m->m_len;
3415 *mp = m;
3416 m_prev = m;
3417
3418 while (sopt_size) {
3419 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3420 if (m == NULL) {
3421 m_freem(*mp);
3422 return ENOBUFS;
3423 }
3424 if (sopt_size > MLEN) {
3425 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3426 M_NOWAIT);
3427 if ((m->m_flags & M_EXT) == 0) {
3428 m_freem(m);
3429 m_freem(*mp);
3430 return ENOBUFS;
3431 }
3432 m->m_len = min(MCLBYTES, sopt_size);
3433 } else {
3434 m->m_len = min(MLEN, sopt_size);
3435 }
3436 sopt_size -= m->m_len;
3437 m_prev->m_next = m;
3438 m_prev = m;
3439 }
3440 return (0);
3441 }
3442
3443 int
3444 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3445 {
3446 struct mbuf *m0 = m;
3447
3448 if (sopt->sopt_val == NULL)
3449 return (0);
3450 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3451 if (sopt->sopt_td != NULL) {
3452 int error;
3453
3454 error = copyin(sopt->sopt_val, mtod(m, char *),
3455 m->m_len);
3456 if (error != 0) {
3457 m_freem(m0);
3458 return(error);
3459 }
3460 } else
3461 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3462 sopt->sopt_valsize -= m->m_len;
3463 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3464 m = m->m_next;
3465 }
3466 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3467 panic("ip6_sooptmcopyin");
3468 return (0);
3469 }
3470
3471 int
3472 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3473 {
3474 struct mbuf *m0 = m;
3475 size_t valsize = 0;
3476
3477 if (sopt->sopt_val == NULL)
3478 return (0);
3479 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3480 if (sopt->sopt_td != NULL) {
3481 int error;
3482
3483 error = copyout(mtod(m, char *), sopt->sopt_val,
3484 m->m_len);
3485 if (error != 0) {
3486 m_freem(m0);
3487 return(error);
3488 }
3489 } else
3490 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3491 sopt->sopt_valsize -= m->m_len;
3492 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3493 valsize += m->m_len;
3494 m = m->m_next;
3495 }
3496 if (m != NULL) {
3497 /* enough soopt buffer should be given from user-land */
3498 m_freem(m0);
3499 return(EINVAL);
3500 }
3501 sopt->sopt_valsize = valsize;
3502 return (0);
3503 }
3504
3505 /*
3506 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3507 * out-of-band data, which will then notify socket consumers.
3508 */
3509 void
3510 sohasoutofband(struct socket *so)
3511 {
3512
3513 if (so->so_sigio != NULL)
3514 pgsigio(&so->so_sigio, SIGURG, 0);
3515 selwakeuppri(&so->so_rdsel, PSOCK);
3516 }
3517
3518 int
3519 sopoll(struct socket *so, int events, struct ucred *active_cred,
3520 struct thread *td)
3521 {
3522
3523 /*
3524 * We do not need to set or assert curvnet as long as everyone uses
3525 * sopoll_generic().
3526 */
3527 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3528 td));
3529 }
3530
3531 int
3532 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3533 struct thread *td)
3534 {
3535 int revents;
3536
3537 SOCK_LOCK(so);
3538 if (SOLISTENING(so)) {
3539 if (!(events & (POLLIN | POLLRDNORM)))
3540 revents = 0;
3541 else if (!TAILQ_EMPTY(&so->sol_comp))
3542 revents = events & (POLLIN | POLLRDNORM);
3543 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3544 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3545 else {
3546 selrecord(td, &so->so_rdsel);
3547 revents = 0;
3548 }
3549 } else {
3550 revents = 0;
3551 SOCKBUF_LOCK(&so->so_snd);
3552 SOCKBUF_LOCK(&so->so_rcv);
3553 if (events & (POLLIN | POLLRDNORM))
3554 if (soreadabledata(so))
3555 revents |= events & (POLLIN | POLLRDNORM);
3556 if (events & (POLLOUT | POLLWRNORM))
3557 if (sowriteable(so))
3558 revents |= events & (POLLOUT | POLLWRNORM);
3559 if (events & (POLLPRI | POLLRDBAND))
3560 if (so->so_oobmark ||
3561 (so->so_rcv.sb_state & SBS_RCVATMARK))
3562 revents |= events & (POLLPRI | POLLRDBAND);
3563 if ((events & POLLINIGNEOF) == 0) {
3564 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3565 revents |= events & (POLLIN | POLLRDNORM);
3566 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3567 revents |= POLLHUP;
3568 }
3569 }
3570 if (revents == 0) {
3571 if (events &
3572 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3573 selrecord(td, &so->so_rdsel);
3574 so->so_rcv.sb_flags |= SB_SEL;
3575 }
3576 if (events & (POLLOUT | POLLWRNORM)) {
3577 selrecord(td, &so->so_wrsel);
3578 so->so_snd.sb_flags |= SB_SEL;
3579 }
3580 }
3581 SOCKBUF_UNLOCK(&so->so_rcv);
3582 SOCKBUF_UNLOCK(&so->so_snd);
3583 }
3584 SOCK_UNLOCK(so);
3585 return (revents);
3586 }
3587
3588 int
3589 soo_kqfilter(struct file *fp, struct knote *kn)
3590 {
3591 struct socket *so = kn->kn_fp->f_data;
3592 struct sockbuf *sb;
3593 struct knlist *knl;
3594
3595 switch (kn->kn_filter) {
3596 case EVFILT_READ:
3597 kn->kn_fop = &soread_filtops;
3598 knl = &so->so_rdsel.si_note;
3599 sb = &so->so_rcv;
3600 break;
3601 case EVFILT_WRITE:
3602 kn->kn_fop = &sowrite_filtops;
3603 knl = &so->so_wrsel.si_note;
3604 sb = &so->so_snd;
3605 break;
3606 case EVFILT_EMPTY:
3607 kn->kn_fop = &soempty_filtops;
3608 knl = &so->so_wrsel.si_note;
3609 sb = &so->so_snd;
3610 break;
3611 default:
3612 return (EINVAL);
3613 }
3614
3615 SOCK_LOCK(so);
3616 if (SOLISTENING(so)) {
3617 knlist_add(knl, kn, 1);
3618 } else {
3619 SOCKBUF_LOCK(sb);
3620 knlist_add(knl, kn, 1);
3621 sb->sb_flags |= SB_KNOTE;
3622 SOCKBUF_UNLOCK(sb);
3623 }
3624 SOCK_UNLOCK(so);
3625 return (0);
3626 }
3627
3628 /*
3629 * Some routines that return EOPNOTSUPP for entry points that are not
3630 * supported by a protocol. Fill in as needed.
3631 */
3632 int
3633 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3634 {
3635
3636 return EOPNOTSUPP;
3637 }
3638
3639 int
3640 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3641 {
3642
3643 return EOPNOTSUPP;
3644 }
3645
3646 int
3647 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3648 {
3649
3650 return EOPNOTSUPP;
3651 }
3652
3653 int
3654 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3655 {
3656
3657 return EOPNOTSUPP;
3658 }
3659
3660 int
3661 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3662 struct thread *td)
3663 {
3664
3665 return EOPNOTSUPP;
3666 }
3667
3668 int
3669 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3670 {
3671
3672 return EOPNOTSUPP;
3673 }
3674
3675 int
3676 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3677 struct thread *td)
3678 {
3679
3680 return EOPNOTSUPP;
3681 }
3682
3683 int
3684 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3685 {
3686
3687 return EOPNOTSUPP;
3688 }
3689
3690 int
3691 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3692 struct ifnet *ifp, struct thread *td)
3693 {
3694
3695 return EOPNOTSUPP;
3696 }
3697
3698 int
3699 pru_disconnect_notsupp(struct socket *so)
3700 {
3701
3702 return EOPNOTSUPP;
3703 }
3704
3705 int
3706 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3707 {
3708
3709 return EOPNOTSUPP;
3710 }
3711
3712 int
3713 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3714 {
3715
3716 return EOPNOTSUPP;
3717 }
3718
3719 int
3720 pru_rcvd_notsupp(struct socket *so, int flags)
3721 {
3722
3723 return EOPNOTSUPP;
3724 }
3725
3726 int
3727 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3728 {
3729
3730 return EOPNOTSUPP;
3731 }
3732
3733 int
3734 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3735 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3736 {
3737
3738 return EOPNOTSUPP;
3739 }
3740
3741 int
3742 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3743 {
3744
3745 return (EOPNOTSUPP);
3746 }
3747
3748 /*
3749 * This isn't really a ``null'' operation, but it's the default one and
3750 * doesn't do anything destructive.
3751 */
3752 int
3753 pru_sense_null(struct socket *so, struct stat *sb)
3754 {
3755
3756 sb->st_blksize = so->so_snd.sb_hiwat;
3757 return 0;
3758 }
3759
3760 int
3761 pru_shutdown_notsupp(struct socket *so)
3762 {
3763
3764 return EOPNOTSUPP;
3765 }
3766
3767 int
3768 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3769 {
3770
3771 return EOPNOTSUPP;
3772 }
3773
3774 int
3775 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3776 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3777 {
3778
3779 return EOPNOTSUPP;
3780 }
3781
3782 int
3783 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3784 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3785 {
3786
3787 return EOPNOTSUPP;
3788 }
3789
3790 int
3791 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3792 struct thread *td)
3793 {
3794
3795 return EOPNOTSUPP;
3796 }
3797
3798 static void
3799 filt_sordetach(struct knote *kn)
3800 {
3801 struct socket *so = kn->kn_fp->f_data;
3802
3803 so_rdknl_lock(so);
3804 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3805 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3806 so->so_rcv.sb_flags &= ~SB_KNOTE;
3807 so_rdknl_unlock(so);
3808 }
3809
3810 /*ARGSUSED*/
3811 static int
3812 filt_soread(struct knote *kn, long hint)
3813 {
3814 struct socket *so;
3815
3816 so = kn->kn_fp->f_data;
3817
3818 if (SOLISTENING(so)) {
3819 SOCK_LOCK_ASSERT(so);
3820 kn->kn_data = so->sol_qlen;
3821 if (so->so_error) {
3822 kn->kn_flags |= EV_EOF;
3823 kn->kn_fflags = so->so_error;
3824 return (1);
3825 }
3826 return (!TAILQ_EMPTY(&so->sol_comp));
3827 }
3828
3829 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3830
3831 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3832 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3833 kn->kn_flags |= EV_EOF;
3834 kn->kn_fflags = so->so_error;
3835 return (1);
3836 } else if (so->so_error) /* temporary udp error */
3837 return (1);
3838
3839 if (kn->kn_sfflags & NOTE_LOWAT) {
3840 if (kn->kn_data >= kn->kn_sdata)
3841 return (1);
3842 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3843 return (1);
3844
3845 /* This hook returning non-zero indicates an event, not error */
3846 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3847 }
3848
3849 static void
3850 filt_sowdetach(struct knote *kn)
3851 {
3852 struct socket *so = kn->kn_fp->f_data;
3853
3854 so_wrknl_lock(so);
3855 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3856 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3857 so->so_snd.sb_flags &= ~SB_KNOTE;
3858 so_wrknl_unlock(so);
3859 }
3860
3861 /*ARGSUSED*/
3862 static int
3863 filt_sowrite(struct knote *kn, long hint)
3864 {
3865 struct socket *so;
3866
3867 so = kn->kn_fp->f_data;
3868
3869 if (SOLISTENING(so))
3870 return (0);
3871
3872 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3873 kn->kn_data = sbspace(&so->so_snd);
3874
3875 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3876
3877 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3878 kn->kn_flags |= EV_EOF;
3879 kn->kn_fflags = so->so_error;
3880 return (1);
3881 } else if (so->so_error) /* temporary udp error */
3882 return (1);
3883 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3884 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3885 return (0);
3886 else if (kn->kn_sfflags & NOTE_LOWAT)
3887 return (kn->kn_data >= kn->kn_sdata);
3888 else
3889 return (kn->kn_data >= so->so_snd.sb_lowat);
3890 }
3891
3892 static int
3893 filt_soempty(struct knote *kn, long hint)
3894 {
3895 struct socket *so;
3896
3897 so = kn->kn_fp->f_data;
3898
3899 if (SOLISTENING(so))
3900 return (1);
3901
3902 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3903 kn->kn_data = sbused(&so->so_snd);
3904
3905 if (kn->kn_data == 0)
3906 return (1);
3907 else
3908 return (0);
3909 }
3910
3911 int
3912 socheckuid(struct socket *so, uid_t uid)
3913 {
3914
3915 if (so == NULL)
3916 return (EPERM);
3917 if (so->so_cred->cr_uid != uid)
3918 return (EPERM);
3919 return (0);
3920 }
3921
3922 /*
3923 * These functions are used by protocols to notify the socket layer (and its
3924 * consumers) of state changes in the sockets driven by protocol-side events.
3925 */
3926
3927 /*
3928 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3929 *
3930 * Normal sequence from the active (originating) side is that
3931 * soisconnecting() is called during processing of connect() call, resulting
3932 * in an eventual call to soisconnected() if/when the connection is
3933 * established. When the connection is torn down soisdisconnecting() is
3934 * called during processing of disconnect() call, and soisdisconnected() is
3935 * called when the connection to the peer is totally severed. The semantics
3936 * of these routines are such that connectionless protocols can call
3937 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3938 * calls when setting up a ``connection'' takes no time.
3939 *
3940 * From the passive side, a socket is created with two queues of sockets:
3941 * so_incomp for connections in progress and so_comp for connections already
3942 * made and awaiting user acceptance. As a protocol is preparing incoming
3943 * connections, it creates a socket structure queued on so_incomp by calling
3944 * sonewconn(). When the connection is established, soisconnected() is
3945 * called, and transfers the socket structure to so_comp, making it available
3946 * to accept().
3947 *
3948 * If a socket is closed with sockets on either so_incomp or so_comp, these
3949 * sockets are dropped.
3950 *
3951 * If higher-level protocols are implemented in the kernel, the wakeups done
3952 * here will sometimes cause software-interrupt process scheduling.
3953 */
3954 void
3955 soisconnecting(struct socket *so)
3956 {
3957
3958 SOCK_LOCK(so);
3959 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3960 so->so_state |= SS_ISCONNECTING;
3961 SOCK_UNLOCK(so);
3962 }
3963
3964 void
3965 soisconnected(struct socket *so)
3966 {
3967
3968 SOCK_LOCK(so);
3969 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3970 so->so_state |= SS_ISCONNECTED;
3971
3972 if (so->so_qstate == SQ_INCOMP) {
3973 struct socket *head = so->so_listen;
3974 int ret;
3975
3976 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3977 /*
3978 * Promoting a socket from incomplete queue to complete, we
3979 * need to go through reverse order of locking. We first do
3980 * trylock, and if that doesn't succeed, we go the hard way
3981 * leaving a reference and rechecking consistency after proper
3982 * locking.
3983 */
3984 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3985 soref(head);
3986 SOCK_UNLOCK(so);
3987 SOLISTEN_LOCK(head);
3988 SOCK_LOCK(so);
3989 if (__predict_false(head != so->so_listen)) {
3990 /*
3991 * The socket went off the listen queue,
3992 * should be lost race to close(2) of sol.
3993 * The socket is about to soabort().
3994 */
3995 SOCK_UNLOCK(so);
3996 sorele(head);
3997 return;
3998 }
3999 /* Not the last one, as so holds a ref. */
4000 refcount_release(&head->so_count);
4001 }
4002 again:
4003 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4004 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4005 head->sol_incqlen--;
4006 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4007 head->sol_qlen++;
4008 so->so_qstate = SQ_COMP;
4009 SOCK_UNLOCK(so);
4010 solisten_wakeup(head); /* unlocks */
4011 } else {
4012 SOCKBUF_LOCK(&so->so_rcv);
4013 soupcall_set(so, SO_RCV,
4014 head->sol_accept_filter->accf_callback,
4015 head->sol_accept_filter_arg);
4016 so->so_options &= ~SO_ACCEPTFILTER;
4017 ret = head->sol_accept_filter->accf_callback(so,
4018 head->sol_accept_filter_arg, M_NOWAIT);
4019 if (ret == SU_ISCONNECTED) {
4020 soupcall_clear(so, SO_RCV);
4021 SOCKBUF_UNLOCK(&so->so_rcv);
4022 goto again;
4023 }
4024 SOCKBUF_UNLOCK(&so->so_rcv);
4025 SOCK_UNLOCK(so);
4026 SOLISTEN_UNLOCK(head);
4027 }
4028 return;
4029 }
4030 SOCK_UNLOCK(so);
4031 wakeup(&so->so_timeo);
4032 sorwakeup(so);
4033 sowwakeup(so);
4034 }
4035
4036 void
4037 soisdisconnecting(struct socket *so)
4038 {
4039
4040 SOCK_LOCK(so);
4041 so->so_state &= ~SS_ISCONNECTING;
4042 so->so_state |= SS_ISDISCONNECTING;
4043
4044 if (!SOLISTENING(so)) {
4045 SOCKBUF_LOCK(&so->so_rcv);
4046 socantrcvmore_locked(so);
4047 SOCKBUF_LOCK(&so->so_snd);
4048 socantsendmore_locked(so);
4049 }
4050 SOCK_UNLOCK(so);
4051 wakeup(&so->so_timeo);
4052 }
4053
4054 void
4055 soisdisconnected(struct socket *so)
4056 {
4057
4058 SOCK_LOCK(so);
4059
4060 /*
4061 * There is at least one reader of so_state that does not
4062 * acquire socket lock, namely soreceive_generic(). Ensure
4063 * that it never sees all flags that track connection status
4064 * cleared, by ordering the update with a barrier semantic of
4065 * our release thread fence.
4066 */
4067 so->so_state |= SS_ISDISCONNECTED;
4068 atomic_thread_fence_rel();
4069 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4070
4071 if (!SOLISTENING(so)) {
4072 SOCK_UNLOCK(so);
4073 SOCKBUF_LOCK(&so->so_rcv);
4074 socantrcvmore_locked(so);
4075 SOCKBUF_LOCK(&so->so_snd);
4076 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4077 socantsendmore_locked(so);
4078 } else
4079 SOCK_UNLOCK(so);
4080 wakeup(&so->so_timeo);
4081 }
4082
4083 /*
4084 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4085 */
4086 struct sockaddr *
4087 sodupsockaddr(const struct sockaddr *sa, int mflags)
4088 {
4089 struct sockaddr *sa2;
4090
4091 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4092 if (sa2)
4093 bcopy(sa, sa2, sa->sa_len);
4094 return sa2;
4095 }
4096
4097 /*
4098 * Register per-socket destructor.
4099 */
4100 void
4101 sodtor_set(struct socket *so, so_dtor_t *func)
4102 {
4103
4104 SOCK_LOCK_ASSERT(so);
4105 so->so_dtor = func;
4106 }
4107
4108 /*
4109 * Register per-socket buffer upcalls.
4110 */
4111 void
4112 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4113 {
4114 struct sockbuf *sb;
4115
4116 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4117
4118 switch (which) {
4119 case SO_RCV:
4120 sb = &so->so_rcv;
4121 break;
4122 case SO_SND:
4123 sb = &so->so_snd;
4124 break;
4125 default:
4126 panic("soupcall_set: bad which");
4127 }
4128 SOCKBUF_LOCK_ASSERT(sb);
4129 sb->sb_upcall = func;
4130 sb->sb_upcallarg = arg;
4131 sb->sb_flags |= SB_UPCALL;
4132 }
4133
4134 void
4135 soupcall_clear(struct socket *so, int which)
4136 {
4137 struct sockbuf *sb;
4138
4139 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4140
4141 switch (which) {
4142 case SO_RCV:
4143 sb = &so->so_rcv;
4144 break;
4145 case SO_SND:
4146 sb = &so->so_snd;
4147 break;
4148 default:
4149 panic("soupcall_clear: bad which");
4150 }
4151 SOCKBUF_LOCK_ASSERT(sb);
4152 KASSERT(sb->sb_upcall != NULL,
4153 ("%s: so %p no upcall to clear", __func__, so));
4154 sb->sb_upcall = NULL;
4155 sb->sb_upcallarg = NULL;
4156 sb->sb_flags &= ~SB_UPCALL;
4157 }
4158
4159 void
4160 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4161 {
4162
4163 SOLISTEN_LOCK_ASSERT(so);
4164 so->sol_upcall = func;
4165 so->sol_upcallarg = arg;
4166 }
4167
4168 static void
4169 so_rdknl_lock(void *arg)
4170 {
4171 struct socket *so = arg;
4172
4173 if (SOLISTENING(so))
4174 SOCK_LOCK(so);
4175 else
4176 SOCKBUF_LOCK(&so->so_rcv);
4177 }
4178
4179 static void
4180 so_rdknl_unlock(void *arg)
4181 {
4182 struct socket *so = arg;
4183
4184 if (SOLISTENING(so))
4185 SOCK_UNLOCK(so);
4186 else
4187 SOCKBUF_UNLOCK(&so->so_rcv);
4188 }
4189
4190 static void
4191 so_rdknl_assert_lock(void *arg, int what)
4192 {
4193 struct socket *so = arg;
4194
4195 if (what == LA_LOCKED) {
4196 if (SOLISTENING(so))
4197 SOCK_LOCK_ASSERT(so);
4198 else
4199 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4200 } else {
4201 if (SOLISTENING(so))
4202 SOCK_UNLOCK_ASSERT(so);
4203 else
4204 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4205 }
4206 }
4207
4208 static void
4209 so_wrknl_lock(void *arg)
4210 {
4211 struct socket *so = arg;
4212
4213 if (SOLISTENING(so))
4214 SOCK_LOCK(so);
4215 else
4216 SOCKBUF_LOCK(&so->so_snd);
4217 }
4218
4219 static void
4220 so_wrknl_unlock(void *arg)
4221 {
4222 struct socket *so = arg;
4223
4224 if (SOLISTENING(so))
4225 SOCK_UNLOCK(so);
4226 else
4227 SOCKBUF_UNLOCK(&so->so_snd);
4228 }
4229
4230 static void
4231 so_wrknl_assert_lock(void *arg, int what)
4232 {
4233 struct socket *so = arg;
4234
4235 if (what == LA_LOCKED) {
4236 if (SOLISTENING(so))
4237 SOCK_LOCK_ASSERT(so);
4238 else
4239 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4240 } else {
4241 if (SOLISTENING(so))
4242 SOCK_UNLOCK_ASSERT(so);
4243 else
4244 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4245 }
4246 }
4247
4248 /*
4249 * Create an external-format (``xsocket'') structure using the information in
4250 * the kernel-format socket structure pointed to by so. This is done to
4251 * reduce the spew of irrelevant information over this interface, to isolate
4252 * user code from changes in the kernel structure, and potentially to provide
4253 * information-hiding if we decide that some of this information should be
4254 * hidden from users.
4255 */
4256 void
4257 sotoxsocket(struct socket *so, struct xsocket *xso)
4258 {
4259
4260 bzero(xso, sizeof(*xso));
4261 xso->xso_len = sizeof *xso;
4262 xso->xso_so = (uintptr_t)so;
4263 xso->so_type = so->so_type;
4264 xso->so_options = so->so_options;
4265 xso->so_linger = so->so_linger;
4266 xso->so_state = so->so_state;
4267 xso->so_pcb = (uintptr_t)so->so_pcb;
4268 xso->xso_protocol = so->so_proto->pr_protocol;
4269 xso->xso_family = so->so_proto->pr_domain->dom_family;
4270 xso->so_timeo = so->so_timeo;
4271 xso->so_error = so->so_error;
4272 xso->so_uid = so->so_cred->cr_uid;
4273 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4274 if (SOLISTENING(so)) {
4275 xso->so_qlen = so->sol_qlen;
4276 xso->so_incqlen = so->sol_incqlen;
4277 xso->so_qlimit = so->sol_qlimit;
4278 xso->so_oobmark = 0;
4279 } else {
4280 xso->so_state |= so->so_qstate;
4281 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4282 xso->so_oobmark = so->so_oobmark;
4283 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4284 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4285 }
4286 }
4287
4288 struct sockbuf *
4289 so_sockbuf_rcv(struct socket *so)
4290 {
4291
4292 return (&so->so_rcv);
4293 }
4294
4295 struct sockbuf *
4296 so_sockbuf_snd(struct socket *so)
4297 {
4298
4299 return (&so->so_snd);
4300 }
4301
4302 int
4303 so_state_get(const struct socket *so)
4304 {
4305
4306 return (so->so_state);
4307 }
4308
4309 void
4310 so_state_set(struct socket *so, int val)
4311 {
4312
4313 so->so_state = val;
4314 }
4315
4316 int
4317 so_options_get(const struct socket *so)
4318 {
4319
4320 return (so->so_options);
4321 }
4322
4323 void
4324 so_options_set(struct socket *so, int val)
4325 {
4326
4327 so->so_options = val;
4328 }
4329
4330 int
4331 so_error_get(const struct socket *so)
4332 {
4333
4334 return (so->so_error);
4335 }
4336
4337 void
4338 so_error_set(struct socket *so, int val)
4339 {
4340
4341 so->so_error = val;
4342 }
4343
4344 int
4345 so_linger_get(const struct socket *so)
4346 {
4347
4348 return (so->so_linger);
4349 }
4350
4351 void
4352 so_linger_set(struct socket *so, int val)
4353 {
4354
4355 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4356 ("%s: val %d out of range", __func__, val));
4357
4358 so->so_linger = val;
4359 }
4360
4361 struct protosw *
4362 so_protosw_get(const struct socket *so)
4363 {
4364
4365 return (so->so_proto);
4366 }
4367
4368 void
4369 so_protosw_set(struct socket *so, struct protosw *val)
4370 {
4371
4372 so->so_proto = val;
4373 }
4374
4375 void
4376 so_sorwakeup(struct socket *so)
4377 {
4378
4379 sorwakeup(so);
4380 }
4381
4382 void
4383 so_sowwakeup(struct socket *so)
4384 {
4385
4386 sowwakeup(so);
4387 }
4388
4389 void
4390 so_sorwakeup_locked(struct socket *so)
4391 {
4392
4393 sorwakeup_locked(so);
4394 }
4395
4396 void
4397 so_sowwakeup_locked(struct socket *so)
4398 {
4399
4400 sowwakeup_locked(so);
4401 }
4402
4403 void
4404 so_lock(struct socket *so)
4405 {
4406
4407 SOCK_LOCK(so);
4408 }
4409
4410 void
4411 so_unlock(struct socket *so)
4412 {
4413
4414 SOCK_UNLOCK(so);
4415 }
Cache object: 79396dcc751247a1829a0f6b43837594
|