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 khelp_destroy_osd(&so->osd);
479 if (SOLISTENING(so)) {
480 if (so->sol_accept_filter != NULL)
481 accept_filt_setopt(so, NULL);
482 } else {
483 if (so->so_rcv.sb_hiwat)
484 (void)chgsbsize(so->so_cred->cr_uidinfo,
485 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
486 if (so->so_snd.sb_hiwat)
487 (void)chgsbsize(so->so_cred->cr_uidinfo,
488 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
489 sx_destroy(&so->so_snd.sb_sx);
490 sx_destroy(&so->so_rcv.sb_sx);
491 SOCKBUF_LOCK_DESTROY(&so->so_snd);
492 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
493 }
494 crfree(so->so_cred);
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_options = head->so_options & ~SO_ACCEPTCONN;
722 so->so_linger = head->so_linger;
723 so->so_state = head->so_state | SS_NOFDREF;
724 so->so_fibnum = head->so_fibnum;
725 so->so_proto = head->so_proto;
726 so->so_cred = crhold(head->so_cred);
727 #ifdef MAC
728 mac_socket_newconn(head, so);
729 #endif
730 knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
731 so_rdknl_assert_lock);
732 knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
733 so_wrknl_assert_lock);
734 VNET_SO_ASSERT(head);
735 if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
736 sodealloc(so);
737 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
738 __func__, head->so_pcb);
739 return (NULL);
740 }
741 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
742 sodealloc(so);
743 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
744 __func__, head->so_pcb);
745 return (NULL);
746 }
747 so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
748 so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
749 so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
750 so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
751 so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
752 so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
753
754 SOLISTEN_LOCK(head);
755 if (head->sol_accept_filter != NULL)
756 connstatus = 0;
757 so->so_state |= connstatus;
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, SOCK_MTX(head), 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 bool last __diagused;
1077
1078 SOCK_LOCK_ASSERT(so);
1079
1080 if ((so->so_state & (SS_NOFDREF | SS_PROTOREF)) != SS_NOFDREF ||
1081 refcount_load(&so->so_count) != 0 || so->so_qstate == SQ_COMP) {
1082 SOCK_UNLOCK(so);
1083 return;
1084 }
1085
1086 if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1087 struct socket *sol;
1088
1089 sol = so->so_listen;
1090 KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1091
1092 /*
1093 * To solve race between close of a listening socket and
1094 * a socket on its incomplete queue, we need to lock both.
1095 * The order is first listening socket, then regular.
1096 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1097 * function and the listening socket are the only pointers
1098 * to so. To preserve so and sol, we reference both and then
1099 * relock.
1100 * After relock the socket may not move to so_comp since it
1101 * doesn't have PCB already, but it may be removed from
1102 * so_incomp. If that happens, we share responsiblity on
1103 * freeing the socket, but soclose() has already removed
1104 * it from queue.
1105 */
1106 soref(sol);
1107 soref(so);
1108 SOCK_UNLOCK(so);
1109 SOLISTEN_LOCK(sol);
1110 SOCK_LOCK(so);
1111 if (so->so_qstate == SQ_INCOMP) {
1112 KASSERT(so->so_listen == sol,
1113 ("%s: so %p migrated out of sol %p",
1114 __func__, so, sol));
1115 TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1116 sol->sol_incqlen--;
1117 last = refcount_release(&sol->so_count);
1118 KASSERT(!last, ("%s: released last reference for %p",
1119 __func__, sol));
1120 so->so_qstate = SQ_NONE;
1121 so->so_listen = NULL;
1122 } else
1123 KASSERT(so->so_listen == NULL,
1124 ("%s: so %p not on (in)comp with so_listen",
1125 __func__, so));
1126 sorele(sol);
1127 KASSERT(refcount_load(&so->so_count) == 1,
1128 ("%s: so %p count %u", __func__, so, so->so_count));
1129 so->so_count = 0;
1130 }
1131 if (SOLISTENING(so))
1132 so->so_error = ECONNABORTED;
1133 SOCK_UNLOCK(so);
1134
1135 if (so->so_dtor != NULL)
1136 so->so_dtor(so);
1137
1138 VNET_SO_ASSERT(so);
1139 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1140 (*pr->pr_domain->dom_dispose)(so);
1141 if (pr->pr_usrreqs->pru_detach != NULL)
1142 (*pr->pr_usrreqs->pru_detach)(so);
1143
1144 /*
1145 * From this point on, we assume that no other references to this
1146 * socket exist anywhere else in the stack. Therefore, no locks need
1147 * to be acquired or held.
1148 *
1149 * We used to do a lot of socket buffer and socket locking here, as
1150 * well as invoke sorflush() and perform wakeups. The direct call to
1151 * dom_dispose() and sbdestroy() are an inlining of what was
1152 * necessary from sorflush().
1153 *
1154 * Notice that the socket buffer and kqueue state are torn down
1155 * before calling pru_detach. This means that protocols shold not
1156 * assume they can perform socket wakeups, etc, in their detach code.
1157 */
1158 if (!SOLISTENING(so)) {
1159 sbdestroy(&so->so_snd, so);
1160 sbdestroy(&so->so_rcv, so);
1161 }
1162 seldrain(&so->so_rdsel);
1163 seldrain(&so->so_wrsel);
1164 knlist_destroy(&so->so_rdsel.si_note);
1165 knlist_destroy(&so->so_wrsel.si_note);
1166 sodealloc(so);
1167 }
1168
1169 /*
1170 * Close a socket on last file table reference removal. Initiate disconnect
1171 * if connected. Free socket when disconnect complete.
1172 *
1173 * This function will sorele() the socket. Note that soclose() may be called
1174 * prior to the ref count reaching zero. The actual socket structure will
1175 * not be freed until the ref count reaches zero.
1176 */
1177 int
1178 soclose(struct socket *so)
1179 {
1180 struct accept_queue lqueue;
1181 struct socket *sp, *tsp;
1182 int error = 0;
1183 bool last __diagused;
1184
1185 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1186
1187 CURVNET_SET(so->so_vnet);
1188 funsetown(&so->so_sigio);
1189 if (so->so_state & SS_ISCONNECTED) {
1190 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1191 error = sodisconnect(so);
1192 if (error) {
1193 if (error == ENOTCONN)
1194 error = 0;
1195 goto drop;
1196 }
1197 }
1198
1199 if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1200 if ((so->so_state & SS_ISDISCONNECTING) &&
1201 (so->so_state & SS_NBIO))
1202 goto drop;
1203 while (so->so_state & SS_ISCONNECTED) {
1204 error = tsleep(&so->so_timeo,
1205 PSOCK | PCATCH, "soclos",
1206 so->so_linger * hz);
1207 if (error)
1208 break;
1209 }
1210 }
1211 }
1212
1213 drop:
1214 if (so->so_proto->pr_usrreqs->pru_close != NULL)
1215 (*so->so_proto->pr_usrreqs->pru_close)(so);
1216
1217 TAILQ_INIT(&lqueue);
1218 SOCK_LOCK(so);
1219 if (SOLISTENING(so)) {
1220 TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1221 TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1222
1223 so->sol_qlen = so->sol_incqlen = 0;
1224
1225 TAILQ_FOREACH(sp, &lqueue, so_list) {
1226 SOCK_LOCK(sp);
1227 sp->so_qstate = SQ_NONE;
1228 sp->so_listen = NULL;
1229 SOCK_UNLOCK(sp);
1230 last = refcount_release(&so->so_count);
1231 KASSERT(!last, ("%s: released last reference for %p",
1232 __func__, so));
1233 }
1234 }
1235 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1236 so->so_state |= SS_NOFDREF;
1237 sorele(so);
1238 TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1239 SOCK_LOCK(sp);
1240 if (refcount_load(&sp->so_count) == 0) {
1241 SOCK_UNLOCK(sp);
1242 soabort(sp);
1243 } else {
1244 /* See the handling of queued sockets in sofree(). */
1245 SOCK_UNLOCK(sp);
1246 }
1247 }
1248 CURVNET_RESTORE();
1249 return (error);
1250 }
1251
1252 /*
1253 * soabort() is used to abruptly tear down a connection, such as when a
1254 * resource limit is reached (listen queue depth exceeded), or if a listen
1255 * socket is closed while there are sockets waiting to be accepted.
1256 *
1257 * This interface is tricky, because it is called on an unreferenced socket,
1258 * and must be called only by a thread that has actually removed the socket
1259 * from the listen queue it was on, or races with other threads are risked.
1260 *
1261 * This interface will call into the protocol code, so must not be called
1262 * with any socket locks held. Protocols do call it while holding their own
1263 * recursible protocol mutexes, but this is something that should be subject
1264 * to review in the future.
1265 */
1266 void
1267 soabort(struct socket *so)
1268 {
1269
1270 /*
1271 * In as much as is possible, assert that no references to this
1272 * socket are held. This is not quite the same as asserting that the
1273 * current thread is responsible for arranging for no references, but
1274 * is as close as we can get for now.
1275 */
1276 KASSERT(so->so_count == 0, ("soabort: so_count"));
1277 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1278 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1279 VNET_SO_ASSERT(so);
1280
1281 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1282 (*so->so_proto->pr_usrreqs->pru_abort)(so);
1283 SOCK_LOCK(so);
1284 sofree(so);
1285 }
1286
1287 int
1288 soaccept(struct socket *so, struct sockaddr **nam)
1289 {
1290 int error;
1291
1292 SOCK_LOCK(so);
1293 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1294 so->so_state &= ~SS_NOFDREF;
1295 SOCK_UNLOCK(so);
1296
1297 CURVNET_SET(so->so_vnet);
1298 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1299 CURVNET_RESTORE();
1300 return (error);
1301 }
1302
1303 int
1304 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1305 {
1306
1307 return (soconnectat(AT_FDCWD, so, nam, td));
1308 }
1309
1310 int
1311 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1312 {
1313 int error;
1314
1315 /* XXXMJ racy */
1316 if (SOLISTENING(so))
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 int
1374 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1375 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1376 {
1377 long space;
1378 ssize_t resid;
1379 int clen = 0, error, dontroute;
1380
1381 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1382 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1383 ("sosend_dgram: !PR_ATOMIC"));
1384
1385 if (uio != NULL)
1386 resid = uio->uio_resid;
1387 else
1388 resid = top->m_pkthdr.len;
1389 /*
1390 * In theory resid should be unsigned. However, space must be
1391 * signed, as it might be less than 0 if we over-committed, and we
1392 * must use a signed comparison of space and resid. On the other
1393 * hand, a negative resid causes us to loop sending 0-length
1394 * segments to the protocol.
1395 */
1396 if (resid < 0) {
1397 error = EINVAL;
1398 goto out;
1399 }
1400
1401 dontroute =
1402 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1403 if (td != NULL)
1404 td->td_ru.ru_msgsnd++;
1405 if (control != NULL)
1406 clen = control->m_len;
1407
1408 SOCKBUF_LOCK(&so->so_snd);
1409 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1410 SOCKBUF_UNLOCK(&so->so_snd);
1411 error = EPIPE;
1412 goto out;
1413 }
1414 if (so->so_error) {
1415 error = so->so_error;
1416 so->so_error = 0;
1417 SOCKBUF_UNLOCK(&so->so_snd);
1418 goto out;
1419 }
1420 if ((so->so_state & SS_ISCONNECTED) == 0) {
1421 /*
1422 * `sendto' and `sendmsg' is allowed on a connection-based
1423 * socket if it supports implied connect. Return ENOTCONN if
1424 * not connected and no address is supplied.
1425 */
1426 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1427 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1428 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1429 !(resid == 0 && clen != 0)) {
1430 SOCKBUF_UNLOCK(&so->so_snd);
1431 error = ENOTCONN;
1432 goto out;
1433 }
1434 } else if (addr == NULL) {
1435 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1436 error = ENOTCONN;
1437 else
1438 error = EDESTADDRREQ;
1439 SOCKBUF_UNLOCK(&so->so_snd);
1440 goto out;
1441 }
1442 }
1443
1444 /*
1445 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1446 * problem and need fixing.
1447 */
1448 space = sbspace(&so->so_snd);
1449 if (flags & MSG_OOB)
1450 space += 1024;
1451 space -= clen;
1452 SOCKBUF_UNLOCK(&so->so_snd);
1453 if (resid > space) {
1454 error = EMSGSIZE;
1455 goto out;
1456 }
1457 if (uio == NULL) {
1458 resid = 0;
1459 if (flags & MSG_EOR)
1460 top->m_flags |= M_EOR;
1461 } else {
1462 /*
1463 * Copy the data from userland into a mbuf chain.
1464 * If no data is to be copied in, a single empty mbuf
1465 * is returned.
1466 */
1467 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1468 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1469 if (top == NULL) {
1470 error = EFAULT; /* only possible error */
1471 goto out;
1472 }
1473 space -= resid - uio->uio_resid;
1474 resid = uio->uio_resid;
1475 }
1476 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1477 /*
1478 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1479 * than with.
1480 */
1481 if (dontroute) {
1482 SOCK_LOCK(so);
1483 so->so_options |= SO_DONTROUTE;
1484 SOCK_UNLOCK(so);
1485 }
1486 /*
1487 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1488 * of date. We could have received a reset packet in an interrupt or
1489 * maybe we slept while doing page faults in uiomove() etc. We could
1490 * probably recheck again inside the locking protection here, but
1491 * there are probably other places that this also happens. We must
1492 * rethink this.
1493 */
1494 VNET_SO_ASSERT(so);
1495 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1496 (flags & MSG_OOB) ? PRUS_OOB :
1497 /*
1498 * If the user set MSG_EOF, the protocol understands this flag and
1499 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1500 */
1501 ((flags & MSG_EOF) &&
1502 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1503 (resid <= 0)) ?
1504 PRUS_EOF :
1505 /* If there is more to send set PRUS_MORETOCOME */
1506 (flags & MSG_MORETOCOME) ||
1507 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1508 top, addr, control, td);
1509 if (dontroute) {
1510 SOCK_LOCK(so);
1511 so->so_options &= ~SO_DONTROUTE;
1512 SOCK_UNLOCK(so);
1513 }
1514 clen = 0;
1515 control = NULL;
1516 top = NULL;
1517 out:
1518 if (top != NULL)
1519 m_freem(top);
1520 if (control != NULL)
1521 m_freem(control);
1522 return (error);
1523 }
1524
1525 /*
1526 * Send on a socket. If send must go all at once and message is larger than
1527 * send buffering, then hard error. Lock against other senders. If must go
1528 * all at once and not enough room now, then inform user that this would
1529 * block and do nothing. Otherwise, if nonblocking, send as much as
1530 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1531 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1532 * in mbuf chain must be small enough to send all at once.
1533 *
1534 * Returns nonzero on error, timeout or signal; callers must check for short
1535 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1536 * on return.
1537 */
1538 int
1539 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1540 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1541 {
1542 long space;
1543 ssize_t resid;
1544 int clen = 0, error, dontroute;
1545 int atomic = sosendallatonce(so) || top;
1546 int pru_flag;
1547 #ifdef KERN_TLS
1548 struct ktls_session *tls;
1549 int tls_enq_cnt, tls_pruflag;
1550 uint8_t tls_rtype;
1551
1552 tls = NULL;
1553 tls_rtype = TLS_RLTYPE_APP;
1554 #endif
1555 if (uio != NULL)
1556 resid = uio->uio_resid;
1557 else if ((top->m_flags & M_PKTHDR) != 0)
1558 resid = top->m_pkthdr.len;
1559 else
1560 resid = m_length(top, NULL);
1561 /*
1562 * In theory resid should be unsigned. However, space must be
1563 * signed, as it might be less than 0 if we over-committed, and we
1564 * must use a signed comparison of space and resid. On the other
1565 * hand, a negative resid causes us to loop sending 0-length
1566 * segments to the protocol.
1567 *
1568 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1569 * type sockets since that's an error.
1570 */
1571 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1572 error = EINVAL;
1573 goto out;
1574 }
1575
1576 dontroute =
1577 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1578 (so->so_proto->pr_flags & PR_ATOMIC);
1579 if (td != NULL)
1580 td->td_ru.ru_msgsnd++;
1581 if (control != NULL)
1582 clen = control->m_len;
1583
1584 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1585 if (error)
1586 goto out;
1587
1588 #ifdef KERN_TLS
1589 tls_pruflag = 0;
1590 tls = ktls_hold(so->so_snd.sb_tls_info);
1591 if (tls != NULL) {
1592 if (tls->mode == TCP_TLS_MODE_SW)
1593 tls_pruflag = PRUS_NOTREADY;
1594
1595 if (control != NULL) {
1596 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1597
1598 if (clen >= sizeof(*cm) &&
1599 cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1600 tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1601 clen = 0;
1602 m_freem(control);
1603 control = NULL;
1604 atomic = 1;
1605 }
1606 }
1607 }
1608 #endif
1609
1610 restart:
1611 do {
1612 SOCKBUF_LOCK(&so->so_snd);
1613 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1614 SOCKBUF_UNLOCK(&so->so_snd);
1615 error = EPIPE;
1616 goto release;
1617 }
1618 if (so->so_error) {
1619 error = so->so_error;
1620 so->so_error = 0;
1621 SOCKBUF_UNLOCK(&so->so_snd);
1622 goto release;
1623 }
1624 if ((so->so_state & SS_ISCONNECTED) == 0) {
1625 /*
1626 * `sendto' and `sendmsg' is allowed on a connection-
1627 * based socket if it supports implied connect.
1628 * Return ENOTCONN if not connected and no address is
1629 * supplied.
1630 */
1631 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1632 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1633 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1634 !(resid == 0 && clen != 0)) {
1635 SOCKBUF_UNLOCK(&so->so_snd);
1636 error = ENOTCONN;
1637 goto release;
1638 }
1639 } else if (addr == NULL) {
1640 SOCKBUF_UNLOCK(&so->so_snd);
1641 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1642 error = ENOTCONN;
1643 else
1644 error = EDESTADDRREQ;
1645 goto release;
1646 }
1647 }
1648 space = sbspace(&so->so_snd);
1649 if (flags & MSG_OOB)
1650 space += 1024;
1651 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1652 clen > so->so_snd.sb_hiwat) {
1653 SOCKBUF_UNLOCK(&so->so_snd);
1654 error = EMSGSIZE;
1655 goto release;
1656 }
1657 if (space < resid + clen &&
1658 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1659 if ((so->so_state & SS_NBIO) ||
1660 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1661 SOCKBUF_UNLOCK(&so->so_snd);
1662 error = EWOULDBLOCK;
1663 goto release;
1664 }
1665 error = sbwait(&so->so_snd);
1666 SOCKBUF_UNLOCK(&so->so_snd);
1667 if (error)
1668 goto release;
1669 goto restart;
1670 }
1671 SOCKBUF_UNLOCK(&so->so_snd);
1672 space -= clen;
1673 do {
1674 if (uio == NULL) {
1675 resid = 0;
1676 if (flags & MSG_EOR)
1677 top->m_flags |= M_EOR;
1678 #ifdef KERN_TLS
1679 if (tls != NULL) {
1680 ktls_frame(top, tls, &tls_enq_cnt,
1681 tls_rtype);
1682 tls_rtype = TLS_RLTYPE_APP;
1683 }
1684 #endif
1685 } else {
1686 /*
1687 * Copy the data from userland into a mbuf
1688 * chain. If resid is 0, which can happen
1689 * only if we have control to send, then
1690 * a single empty mbuf is returned. This
1691 * is a workaround to prevent protocol send
1692 * methods to panic.
1693 */
1694 #ifdef KERN_TLS
1695 if (tls != NULL) {
1696 top = m_uiotombuf(uio, M_WAITOK, space,
1697 tls->params.max_frame_len,
1698 M_EXTPG |
1699 ((flags & MSG_EOR) ? M_EOR : 0));
1700 if (top != NULL) {
1701 ktls_frame(top, tls,
1702 &tls_enq_cnt, tls_rtype);
1703 }
1704 tls_rtype = TLS_RLTYPE_APP;
1705 } else
1706 #endif
1707 top = m_uiotombuf(uio, M_WAITOK, space,
1708 (atomic ? max_hdr : 0),
1709 (atomic ? M_PKTHDR : 0) |
1710 ((flags & MSG_EOR) ? M_EOR : 0));
1711 if (top == NULL) {
1712 error = EFAULT; /* only possible error */
1713 goto release;
1714 }
1715 space -= resid - uio->uio_resid;
1716 resid = uio->uio_resid;
1717 }
1718 if (dontroute) {
1719 SOCK_LOCK(so);
1720 so->so_options |= SO_DONTROUTE;
1721 SOCK_UNLOCK(so);
1722 }
1723 /*
1724 * XXX all the SBS_CANTSENDMORE checks previously
1725 * done could be out of date. We could have received
1726 * a reset packet in an interrupt or maybe we slept
1727 * while doing page faults in uiomove() etc. We
1728 * could probably recheck again inside the locking
1729 * protection here, but there are probably other
1730 * places that this also happens. We must rethink
1731 * this.
1732 */
1733 VNET_SO_ASSERT(so);
1734
1735 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1736 /*
1737 * If the user set MSG_EOF, the protocol understands
1738 * this flag and nothing left to send then use
1739 * PRU_SEND_EOF instead of PRU_SEND.
1740 */
1741 ((flags & MSG_EOF) &&
1742 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1743 (resid <= 0)) ?
1744 PRUS_EOF :
1745 /* If there is more to send set PRUS_MORETOCOME. */
1746 (flags & MSG_MORETOCOME) ||
1747 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1748
1749 #ifdef KERN_TLS
1750 pru_flag |= tls_pruflag;
1751 #endif
1752
1753 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1754 pru_flag, top, addr, control, td);
1755
1756 if (dontroute) {
1757 SOCK_LOCK(so);
1758 so->so_options &= ~SO_DONTROUTE;
1759 SOCK_UNLOCK(so);
1760 }
1761
1762 #ifdef KERN_TLS
1763 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1764 if (error != 0) {
1765 m_freem(top);
1766 top = NULL;
1767 } else {
1768 soref(so);
1769 ktls_enqueue(top, so, tls_enq_cnt);
1770 }
1771 }
1772 #endif
1773 clen = 0;
1774 control = NULL;
1775 top = NULL;
1776 if (error)
1777 goto release;
1778 } while (resid && space > 0);
1779 } while (resid);
1780
1781 release:
1782 SOCK_IO_SEND_UNLOCK(so);
1783 out:
1784 #ifdef KERN_TLS
1785 if (tls != NULL)
1786 ktls_free(tls);
1787 #endif
1788 if (top != NULL)
1789 m_freem(top);
1790 if (control != NULL)
1791 m_freem(control);
1792 return (error);
1793 }
1794
1795 int
1796 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1797 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1798 {
1799 int error;
1800
1801 CURVNET_SET(so->so_vnet);
1802 if (!SOLISTENING(so))
1803 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1804 top, control, flags, td);
1805 else {
1806 m_freem(top);
1807 m_freem(control);
1808 error = ENOTCONN;
1809 }
1810 CURVNET_RESTORE();
1811 return (error);
1812 }
1813
1814 /*
1815 * The part of soreceive() that implements reading non-inline out-of-band
1816 * data from a socket. For more complete comments, see soreceive(), from
1817 * which this code originated.
1818 *
1819 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1820 * unable to return an mbuf chain to the caller.
1821 */
1822 static int
1823 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1824 {
1825 struct protosw *pr = so->so_proto;
1826 struct mbuf *m;
1827 int error;
1828
1829 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1830 VNET_SO_ASSERT(so);
1831
1832 m = m_get(M_WAITOK, MT_DATA);
1833 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1834 if (error)
1835 goto bad;
1836 do {
1837 error = uiomove(mtod(m, void *),
1838 (int) min(uio->uio_resid, m->m_len), uio);
1839 m = m_free(m);
1840 } while (uio->uio_resid && error == 0 && m);
1841 bad:
1842 if (m != NULL)
1843 m_freem(m);
1844 return (error);
1845 }
1846
1847 /*
1848 * Following replacement or removal of the first mbuf on the first mbuf chain
1849 * of a socket buffer, push necessary state changes back into the socket
1850 * buffer so that other consumers see the values consistently. 'nextrecord'
1851 * is the callers locally stored value of the original value of
1852 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1853 * NOTE: 'nextrecord' may be NULL.
1854 */
1855 static __inline void
1856 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1857 {
1858
1859 SOCKBUF_LOCK_ASSERT(sb);
1860 /*
1861 * First, update for the new value of nextrecord. If necessary, make
1862 * it the first record.
1863 */
1864 if (sb->sb_mb != NULL)
1865 sb->sb_mb->m_nextpkt = nextrecord;
1866 else
1867 sb->sb_mb = nextrecord;
1868
1869 /*
1870 * Now update any dependent socket buffer fields to reflect the new
1871 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1872 * addition of a second clause that takes care of the case where
1873 * sb_mb has been updated, but remains the last record.
1874 */
1875 if (sb->sb_mb == NULL) {
1876 sb->sb_mbtail = NULL;
1877 sb->sb_lastrecord = NULL;
1878 } else if (sb->sb_mb->m_nextpkt == NULL)
1879 sb->sb_lastrecord = sb->sb_mb;
1880 }
1881
1882 /*
1883 * Implement receive operations on a socket. We depend on the way that
1884 * records are added to the sockbuf by sbappend. In particular, each record
1885 * (mbufs linked through m_next) must begin with an address if the protocol
1886 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1887 * data, and then zero or more mbufs of data. In order to allow parallelism
1888 * between network receive and copying to user space, as well as avoid
1889 * sleeping with a mutex held, we release the socket buffer mutex during the
1890 * user space copy. Although the sockbuf is locked, new data may still be
1891 * appended, and thus we must maintain consistency of the sockbuf during that
1892 * time.
1893 *
1894 * The caller may receive the data as a single mbuf chain by supplying an
1895 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1896 * the count in uio_resid.
1897 */
1898 int
1899 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1900 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1901 {
1902 struct mbuf *m, **mp;
1903 int flags, error, offset;
1904 ssize_t len;
1905 struct protosw *pr = so->so_proto;
1906 struct mbuf *nextrecord;
1907 int moff, type = 0;
1908 ssize_t orig_resid = uio->uio_resid;
1909
1910 mp = mp0;
1911 if (psa != NULL)
1912 *psa = NULL;
1913 if (controlp != NULL)
1914 *controlp = NULL;
1915 if (flagsp != NULL)
1916 flags = *flagsp &~ MSG_EOR;
1917 else
1918 flags = 0;
1919 if (flags & MSG_OOB)
1920 return (soreceive_rcvoob(so, uio, flags));
1921 if (mp != NULL)
1922 *mp = NULL;
1923 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1924 && uio->uio_resid) {
1925 VNET_SO_ASSERT(so);
1926 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1927 }
1928
1929 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1930 if (error)
1931 return (error);
1932
1933 restart:
1934 SOCKBUF_LOCK(&so->so_rcv);
1935 m = so->so_rcv.sb_mb;
1936 /*
1937 * If we have less data than requested, block awaiting more (subject
1938 * to any timeout) if:
1939 * 1. the current count is less than the low water mark, or
1940 * 2. MSG_DONTWAIT is not set
1941 */
1942 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1943 sbavail(&so->so_rcv) < uio->uio_resid) &&
1944 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1945 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1946 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1947 ("receive: m == %p sbavail == %u",
1948 m, sbavail(&so->so_rcv)));
1949 if (so->so_error || so->so_rerror) {
1950 if (m != NULL)
1951 goto dontblock;
1952 if (so->so_error)
1953 error = so->so_error;
1954 else
1955 error = so->so_rerror;
1956 if ((flags & MSG_PEEK) == 0) {
1957 if (so->so_error)
1958 so->so_error = 0;
1959 else
1960 so->so_rerror = 0;
1961 }
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_unmapped_uiomove(m, moff, uio,
2203 (int)len);
2204 else
2205 error = uiomove(mtod(m, char *) + moff,
2206 (int)len, uio);
2207 SOCKBUF_LOCK(&so->so_rcv);
2208 if (error) {
2209 /*
2210 * The MT_SONAME mbuf has already been removed
2211 * from the record, so it is necessary to
2212 * remove the data mbufs, if any, to preserve
2213 * the invariant in the case of PR_ADDR that
2214 * requires MT_SONAME mbufs at the head of
2215 * each record.
2216 */
2217 if (pr->pr_flags & PR_ATOMIC &&
2218 ((flags & MSG_PEEK) == 0))
2219 (void)sbdroprecord_locked(&so->so_rcv);
2220 SOCKBUF_UNLOCK(&so->so_rcv);
2221 goto release;
2222 }
2223 } else
2224 uio->uio_resid -= len;
2225 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2226 if (len == m->m_len - moff) {
2227 if (m->m_flags & M_EOR)
2228 flags |= MSG_EOR;
2229 if (flags & MSG_PEEK) {
2230 m = m->m_next;
2231 moff = 0;
2232 } else {
2233 nextrecord = m->m_nextpkt;
2234 sbfree(&so->so_rcv, m);
2235 if (mp != NULL) {
2236 m->m_nextpkt = NULL;
2237 *mp = m;
2238 mp = &m->m_next;
2239 so->so_rcv.sb_mb = m = m->m_next;
2240 *mp = NULL;
2241 } else {
2242 so->so_rcv.sb_mb = m_free(m);
2243 m = so->so_rcv.sb_mb;
2244 }
2245 sockbuf_pushsync(&so->so_rcv, nextrecord);
2246 SBLASTRECORDCHK(&so->so_rcv);
2247 SBLASTMBUFCHK(&so->so_rcv);
2248 }
2249 } else {
2250 if (flags & MSG_PEEK)
2251 moff += len;
2252 else {
2253 if (mp != NULL) {
2254 if (flags & MSG_DONTWAIT) {
2255 *mp = m_copym(m, 0, len,
2256 M_NOWAIT);
2257 if (*mp == NULL) {
2258 /*
2259 * m_copym() couldn't
2260 * allocate an mbuf.
2261 * Adjust uio_resid back
2262 * (it was adjusted
2263 * down by len bytes,
2264 * which we didn't end
2265 * up "copying" over).
2266 */
2267 uio->uio_resid += len;
2268 break;
2269 }
2270 } else {
2271 SOCKBUF_UNLOCK(&so->so_rcv);
2272 *mp = m_copym(m, 0, len,
2273 M_WAITOK);
2274 SOCKBUF_LOCK(&so->so_rcv);
2275 }
2276 }
2277 sbcut_locked(&so->so_rcv, len);
2278 }
2279 }
2280 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2281 if (so->so_oobmark) {
2282 if ((flags & MSG_PEEK) == 0) {
2283 so->so_oobmark -= len;
2284 if (so->so_oobmark == 0) {
2285 so->so_rcv.sb_state |= SBS_RCVATMARK;
2286 break;
2287 }
2288 } else {
2289 offset += len;
2290 if (offset == so->so_oobmark)
2291 break;
2292 }
2293 }
2294 if (flags & MSG_EOR)
2295 break;
2296 /*
2297 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2298 * must not quit until "uio->uio_resid == 0" or an error
2299 * termination. If a signal/timeout occurs, return with a
2300 * short count but without error. Keep sockbuf locked
2301 * against other readers.
2302 */
2303 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2304 !sosendallatonce(so) && nextrecord == NULL) {
2305 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2306 if (so->so_error || so->so_rerror ||
2307 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2308 break;
2309 /*
2310 * Notify the protocol that some data has been
2311 * drained before blocking.
2312 */
2313 if (pr->pr_flags & PR_WANTRCVD) {
2314 SOCKBUF_UNLOCK(&so->so_rcv);
2315 VNET_SO_ASSERT(so);
2316 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2317 SOCKBUF_LOCK(&so->so_rcv);
2318 }
2319 SBLASTRECORDCHK(&so->so_rcv);
2320 SBLASTMBUFCHK(&so->so_rcv);
2321 /*
2322 * We could receive some data while was notifying
2323 * the protocol. Skip blocking in this case.
2324 */
2325 if (so->so_rcv.sb_mb == NULL) {
2326 error = sbwait(&so->so_rcv);
2327 if (error) {
2328 SOCKBUF_UNLOCK(&so->so_rcv);
2329 goto release;
2330 }
2331 }
2332 m = so->so_rcv.sb_mb;
2333 if (m != NULL)
2334 nextrecord = m->m_nextpkt;
2335 }
2336 }
2337
2338 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2339 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2340 flags |= MSG_TRUNC;
2341 if ((flags & MSG_PEEK) == 0)
2342 (void) sbdroprecord_locked(&so->so_rcv);
2343 }
2344 if ((flags & MSG_PEEK) == 0) {
2345 if (m == NULL) {
2346 /*
2347 * First part is an inline SB_EMPTY_FIXUP(). Second
2348 * part makes sure sb_lastrecord is up-to-date if
2349 * there is still data in the socket buffer.
2350 */
2351 so->so_rcv.sb_mb = nextrecord;
2352 if (so->so_rcv.sb_mb == NULL) {
2353 so->so_rcv.sb_mbtail = NULL;
2354 so->so_rcv.sb_lastrecord = NULL;
2355 } else if (nextrecord->m_nextpkt == NULL)
2356 so->so_rcv.sb_lastrecord = nextrecord;
2357 }
2358 SBLASTRECORDCHK(&so->so_rcv);
2359 SBLASTMBUFCHK(&so->so_rcv);
2360 /*
2361 * If soreceive() is being done from the socket callback,
2362 * then don't need to generate ACK to peer to update window,
2363 * since ACK will be generated on return to TCP.
2364 */
2365 if (!(flags & MSG_SOCALLBCK) &&
2366 (pr->pr_flags & PR_WANTRCVD)) {
2367 SOCKBUF_UNLOCK(&so->so_rcv);
2368 VNET_SO_ASSERT(so);
2369 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2370 SOCKBUF_LOCK(&so->so_rcv);
2371 }
2372 }
2373 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2374 if (orig_resid == uio->uio_resid && orig_resid &&
2375 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2376 SOCKBUF_UNLOCK(&so->so_rcv);
2377 goto restart;
2378 }
2379 SOCKBUF_UNLOCK(&so->so_rcv);
2380
2381 if (flagsp != NULL)
2382 *flagsp |= flags;
2383 release:
2384 SOCK_IO_RECV_UNLOCK(so);
2385 return (error);
2386 }
2387
2388 /*
2389 * Optimized version of soreceive() for stream (TCP) sockets.
2390 */
2391 int
2392 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2393 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2394 {
2395 int len = 0, error = 0, flags, oresid;
2396 struct sockbuf *sb;
2397 struct mbuf *m, *n = NULL;
2398
2399 /* We only do stream sockets. */
2400 if (so->so_type != SOCK_STREAM)
2401 return (EINVAL);
2402 if (psa != NULL)
2403 *psa = NULL;
2404 if (flagsp != NULL)
2405 flags = *flagsp &~ MSG_EOR;
2406 else
2407 flags = 0;
2408 if (controlp != NULL)
2409 *controlp = NULL;
2410 if (flags & MSG_OOB)
2411 return (soreceive_rcvoob(so, uio, flags));
2412 if (mp0 != NULL)
2413 *mp0 = NULL;
2414
2415 sb = &so->so_rcv;
2416
2417 #ifdef KERN_TLS
2418 /*
2419 * KTLS store TLS records as records with a control message to
2420 * describe the framing.
2421 *
2422 * We check once here before acquiring locks to optimize the
2423 * common case.
2424 */
2425 if (sb->sb_tls_info != NULL)
2426 return (soreceive_generic(so, psa, uio, mp0, controlp,
2427 flagsp));
2428 #endif
2429
2430 /* Prevent other readers from entering the socket. */
2431 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2432 if (error)
2433 return (error);
2434 SOCKBUF_LOCK(sb);
2435
2436 #ifdef KERN_TLS
2437 if (sb->sb_tls_info != NULL) {
2438 SOCKBUF_UNLOCK(sb);
2439 SOCK_IO_RECV_UNLOCK(so);
2440 return (soreceive_generic(so, psa, uio, mp0, controlp,
2441 flagsp));
2442 }
2443 #endif
2444
2445 /* Easy one, no space to copyout anything. */
2446 if (uio->uio_resid == 0) {
2447 error = EINVAL;
2448 goto out;
2449 }
2450 oresid = uio->uio_resid;
2451
2452 /* We will never ever get anything unless we are or were connected. */
2453 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2454 error = ENOTCONN;
2455 goto out;
2456 }
2457
2458 restart:
2459 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2460
2461 /* Abort if socket has reported problems. */
2462 if (so->so_error) {
2463 if (sbavail(sb) > 0)
2464 goto deliver;
2465 if (oresid > uio->uio_resid)
2466 goto out;
2467 error = so->so_error;
2468 if (!(flags & MSG_PEEK))
2469 so->so_error = 0;
2470 goto out;
2471 }
2472
2473 /* Door is closed. Deliver what is left, if any. */
2474 if (sb->sb_state & SBS_CANTRCVMORE) {
2475 if (sbavail(sb) > 0)
2476 goto deliver;
2477 else
2478 goto out;
2479 }
2480
2481 /* Socket buffer is empty and we shall not block. */
2482 if (sbavail(sb) == 0 &&
2483 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2484 error = EAGAIN;
2485 goto out;
2486 }
2487
2488 /* Socket buffer got some data that we shall deliver now. */
2489 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2490 ((so->so_state & SS_NBIO) ||
2491 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2492 sbavail(sb) >= sb->sb_lowat ||
2493 sbavail(sb) >= uio->uio_resid ||
2494 sbavail(sb) >= sb->sb_hiwat) ) {
2495 goto deliver;
2496 }
2497
2498 /* On MSG_WAITALL we must wait until all data or error arrives. */
2499 if ((flags & MSG_WAITALL) &&
2500 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2501 goto deliver;
2502
2503 /*
2504 * Wait and block until (more) data comes in.
2505 * NB: Drops the sockbuf lock during wait.
2506 */
2507 error = sbwait(sb);
2508 if (error)
2509 goto out;
2510 goto restart;
2511
2512 deliver:
2513 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2514 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2515 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2516
2517 /* Statistics. */
2518 if (uio->uio_td)
2519 uio->uio_td->td_ru.ru_msgrcv++;
2520
2521 /* Fill uio until full or current end of socket buffer is reached. */
2522 len = min(uio->uio_resid, sbavail(sb));
2523 if (mp0 != NULL) {
2524 /* Dequeue as many mbufs as possible. */
2525 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2526 if (*mp0 == NULL)
2527 *mp0 = sb->sb_mb;
2528 else
2529 m_cat(*mp0, sb->sb_mb);
2530 for (m = sb->sb_mb;
2531 m != NULL && m->m_len <= len;
2532 m = m->m_next) {
2533 KASSERT(!(m->m_flags & M_NOTAVAIL),
2534 ("%s: m %p not available", __func__, m));
2535 len -= m->m_len;
2536 uio->uio_resid -= m->m_len;
2537 sbfree(sb, m);
2538 n = m;
2539 }
2540 n->m_next = NULL;
2541 sb->sb_mb = m;
2542 sb->sb_lastrecord = sb->sb_mb;
2543 if (sb->sb_mb == NULL)
2544 SB_EMPTY_FIXUP(sb);
2545 }
2546 /* Copy the remainder. */
2547 if (len > 0) {
2548 KASSERT(sb->sb_mb != NULL,
2549 ("%s: len > 0 && sb->sb_mb empty", __func__));
2550
2551 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2552 if (m == NULL)
2553 len = 0; /* Don't flush data from sockbuf. */
2554 else
2555 uio->uio_resid -= len;
2556 if (*mp0 != NULL)
2557 m_cat(*mp0, m);
2558 else
2559 *mp0 = m;
2560 if (*mp0 == NULL) {
2561 error = ENOBUFS;
2562 goto out;
2563 }
2564 }
2565 } else {
2566 /* NB: Must unlock socket buffer as uiomove may sleep. */
2567 SOCKBUF_UNLOCK(sb);
2568 error = m_mbuftouio(uio, sb->sb_mb, len);
2569 SOCKBUF_LOCK(sb);
2570 if (error)
2571 goto out;
2572 }
2573 SBLASTRECORDCHK(sb);
2574 SBLASTMBUFCHK(sb);
2575
2576 /*
2577 * Remove the delivered data from the socket buffer unless we
2578 * were only peeking.
2579 */
2580 if (!(flags & MSG_PEEK)) {
2581 if (len > 0)
2582 sbdrop_locked(sb, len);
2583
2584 /* Notify protocol that we drained some data. */
2585 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2586 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2587 !(flags & MSG_SOCALLBCK))) {
2588 SOCKBUF_UNLOCK(sb);
2589 VNET_SO_ASSERT(so);
2590 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2591 SOCKBUF_LOCK(sb);
2592 }
2593 }
2594
2595 /*
2596 * For MSG_WAITALL we may have to loop again and wait for
2597 * more data to come in.
2598 */
2599 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2600 goto restart;
2601 out:
2602 SBLASTRECORDCHK(sb);
2603 SBLASTMBUFCHK(sb);
2604 SOCKBUF_UNLOCK(sb);
2605 SOCK_IO_RECV_UNLOCK(so);
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 int error;
2873
2874 VNET_SO_ASSERT(so);
2875
2876 /*
2877 * In order to avoid calling dom_dispose with the socket buffer mutex
2878 * held, and in order to generally avoid holding the lock for a long
2879 * time, we make a copy of the socket buffer and clear the original
2880 * (except locks, state). The new socket buffer copy won't have
2881 * initialized locks so we can only call routines that won't use or
2882 * assert those locks.
2883 *
2884 * Dislodge threads currently blocked in receive and wait to acquire
2885 * a lock against other simultaneous readers before clearing the
2886 * socket buffer. Don't let our acquire be interrupted by a signal
2887 * despite any existing socket disposition on interruptable waiting.
2888 */
2889 socantrcvmore(so);
2890 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2891 KASSERT(error == 0, ("%s: cannot lock sock %p recv buffer",
2892 __func__, so));
2893
2894 /*
2895 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2896 * and mutex data unchanged.
2897 */
2898 SOCKBUF_LOCK(sb);
2899 bzero(&aso, sizeof(aso));
2900 aso.so_pcb = so->so_pcb;
2901 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2902 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2903 bzero(&sb->sb_startzero,
2904 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2905 SOCKBUF_UNLOCK(sb);
2906 SOCK_IO_RECV_UNLOCK(so);
2907
2908 /*
2909 * Dispose of special rights and flush the copied socket. Don't call
2910 * any unsafe routines (that rely on locks being initialized) on aso.
2911 */
2912 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2913 (*pr->pr_domain->dom_dispose)(&aso);
2914 sbrelease_internal(&aso.so_rcv, so);
2915 }
2916
2917 /*
2918 * Wrapper for Socket established helper hook.
2919 * Parameters: socket, context of the hook point, hook id.
2920 */
2921 static int inline
2922 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2923 {
2924 struct socket_hhook_data hhook_data = {
2925 .so = so,
2926 .hctx = hctx,
2927 .m = NULL,
2928 .status = 0
2929 };
2930
2931 CURVNET_SET(so->so_vnet);
2932 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2933 CURVNET_RESTORE();
2934
2935 /* Ugly but needed, since hhooks return void for now */
2936 return (hhook_data.status);
2937 }
2938
2939 /*
2940 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2941 * additional variant to handle the case where the option value needs to be
2942 * some kind of integer, but not a specific size. In addition to their use
2943 * here, these functions are also called by the protocol-level pr_ctloutput()
2944 * routines.
2945 */
2946 int
2947 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2948 {
2949 size_t valsize;
2950
2951 /*
2952 * If the user gives us more than we wanted, we ignore it, but if we
2953 * don't get the minimum length the caller wants, we return EINVAL.
2954 * On success, sopt->sopt_valsize is set to however much we actually
2955 * retrieved.
2956 */
2957 if ((valsize = sopt->sopt_valsize) < minlen)
2958 return EINVAL;
2959 if (valsize > len)
2960 sopt->sopt_valsize = valsize = len;
2961
2962 if (sopt->sopt_td != NULL)
2963 return (copyin(sopt->sopt_val, buf, valsize));
2964
2965 bcopy(sopt->sopt_val, buf, valsize);
2966 return (0);
2967 }
2968
2969 /*
2970 * Kernel version of setsockopt(2).
2971 *
2972 * XXX: optlen is size_t, not socklen_t
2973 */
2974 int
2975 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2976 size_t optlen)
2977 {
2978 struct sockopt sopt;
2979
2980 sopt.sopt_level = level;
2981 sopt.sopt_name = optname;
2982 sopt.sopt_dir = SOPT_SET;
2983 sopt.sopt_val = optval;
2984 sopt.sopt_valsize = optlen;
2985 sopt.sopt_td = NULL;
2986 return (sosetopt(so, &sopt));
2987 }
2988
2989 int
2990 sosetopt(struct socket *so, struct sockopt *sopt)
2991 {
2992 int error, optval;
2993 struct linger l;
2994 struct timeval tv;
2995 sbintime_t val;
2996 uint32_t val32;
2997 #ifdef MAC
2998 struct mac extmac;
2999 #endif
3000
3001 CURVNET_SET(so->so_vnet);
3002 error = 0;
3003 if (sopt->sopt_level != SOL_SOCKET) {
3004 if (so->so_proto->pr_ctloutput != NULL)
3005 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3006 else
3007 error = ENOPROTOOPT;
3008 } else {
3009 switch (sopt->sopt_name) {
3010 case SO_ACCEPTFILTER:
3011 error = accept_filt_setopt(so, sopt);
3012 if (error)
3013 goto bad;
3014 break;
3015
3016 case SO_LINGER:
3017 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3018 if (error)
3019 goto bad;
3020 if (l.l_linger < 0 ||
3021 l.l_linger > USHRT_MAX ||
3022 l.l_linger > (INT_MAX / hz)) {
3023 error = EDOM;
3024 goto bad;
3025 }
3026 SOCK_LOCK(so);
3027 so->so_linger = l.l_linger;
3028 if (l.l_onoff)
3029 so->so_options |= SO_LINGER;
3030 else
3031 so->so_options &= ~SO_LINGER;
3032 SOCK_UNLOCK(so);
3033 break;
3034
3035 case SO_DEBUG:
3036 case SO_KEEPALIVE:
3037 case SO_DONTROUTE:
3038 case SO_USELOOPBACK:
3039 case SO_BROADCAST:
3040 case SO_REUSEADDR:
3041 case SO_REUSEPORT:
3042 case SO_REUSEPORT_LB:
3043 case SO_OOBINLINE:
3044 case SO_TIMESTAMP:
3045 case SO_BINTIME:
3046 case SO_NOSIGPIPE:
3047 case SO_NO_DDP:
3048 case SO_NO_OFFLOAD:
3049 case SO_RERROR:
3050 error = sooptcopyin(sopt, &optval, sizeof optval,
3051 sizeof optval);
3052 if (error)
3053 goto bad;
3054 SOCK_LOCK(so);
3055 if (optval)
3056 so->so_options |= sopt->sopt_name;
3057 else
3058 so->so_options &= ~sopt->sopt_name;
3059 SOCK_UNLOCK(so);
3060 break;
3061
3062 case SO_SETFIB:
3063 error = sooptcopyin(sopt, &optval, sizeof optval,
3064 sizeof optval);
3065 if (error)
3066 goto bad;
3067
3068 if (optval < 0 || optval >= rt_numfibs) {
3069 error = EINVAL;
3070 goto bad;
3071 }
3072 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3073 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3074 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3075 so->so_fibnum = optval;
3076 else
3077 so->so_fibnum = 0;
3078 break;
3079
3080 case SO_USER_COOKIE:
3081 error = sooptcopyin(sopt, &val32, sizeof val32,
3082 sizeof val32);
3083 if (error)
3084 goto bad;
3085 so->so_user_cookie = val32;
3086 break;
3087
3088 case SO_SNDBUF:
3089 case SO_RCVBUF:
3090 case SO_SNDLOWAT:
3091 case SO_RCVLOWAT:
3092 error = sooptcopyin(sopt, &optval, sizeof optval,
3093 sizeof optval);
3094 if (error)
3095 goto bad;
3096
3097 /*
3098 * Values < 1 make no sense for any of these options,
3099 * so disallow them.
3100 */
3101 if (optval < 1) {
3102 error = EINVAL;
3103 goto bad;
3104 }
3105
3106 error = sbsetopt(so, sopt->sopt_name, optval);
3107 break;
3108
3109 case SO_SNDTIMEO:
3110 case SO_RCVTIMEO:
3111 #ifdef COMPAT_FREEBSD32
3112 if (SV_CURPROC_FLAG(SV_ILP32)) {
3113 struct timeval32 tv32;
3114
3115 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3116 sizeof tv32);
3117 CP(tv32, tv, tv_sec);
3118 CP(tv32, tv, tv_usec);
3119 } else
3120 #endif
3121 error = sooptcopyin(sopt, &tv, sizeof tv,
3122 sizeof tv);
3123 if (error)
3124 goto bad;
3125 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3126 tv.tv_usec >= 1000000) {
3127 error = EDOM;
3128 goto bad;
3129 }
3130 if (tv.tv_sec > INT32_MAX)
3131 val = SBT_MAX;
3132 else
3133 val = tvtosbt(tv);
3134 switch (sopt->sopt_name) {
3135 case SO_SNDTIMEO:
3136 so->so_snd.sb_timeo = val;
3137 break;
3138 case SO_RCVTIMEO:
3139 so->so_rcv.sb_timeo = val;
3140 break;
3141 }
3142 break;
3143
3144 case SO_LABEL:
3145 #ifdef MAC
3146 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3147 sizeof extmac);
3148 if (error)
3149 goto bad;
3150 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3151 so, &extmac);
3152 #else
3153 error = EOPNOTSUPP;
3154 #endif
3155 break;
3156
3157 case SO_TS_CLOCK:
3158 error = sooptcopyin(sopt, &optval, sizeof optval,
3159 sizeof optval);
3160 if (error)
3161 goto bad;
3162 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3163 error = EINVAL;
3164 goto bad;
3165 }
3166 so->so_ts_clock = optval;
3167 break;
3168
3169 case SO_MAX_PACING_RATE:
3170 error = sooptcopyin(sopt, &val32, sizeof(val32),
3171 sizeof(val32));
3172 if (error)
3173 goto bad;
3174 so->so_max_pacing_rate = val32;
3175 break;
3176
3177 default:
3178 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3179 error = hhook_run_socket(so, sopt,
3180 HHOOK_SOCKET_OPT);
3181 else
3182 error = ENOPROTOOPT;
3183 break;
3184 }
3185 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3186 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3187 }
3188 bad:
3189 CURVNET_RESTORE();
3190 return (error);
3191 }
3192
3193 /*
3194 * Helper routine for getsockopt.
3195 */
3196 int
3197 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3198 {
3199 int error;
3200 size_t valsize;
3201
3202 error = 0;
3203
3204 /*
3205 * Documented get behavior is that we always return a value, possibly
3206 * truncated to fit in the user's buffer. Traditional behavior is
3207 * that we always tell the user precisely how much we copied, rather
3208 * than something useful like the total amount we had available for
3209 * her. Note that this interface is not idempotent; the entire
3210 * answer must be generated ahead of time.
3211 */
3212 valsize = min(len, sopt->sopt_valsize);
3213 sopt->sopt_valsize = valsize;
3214 if (sopt->sopt_val != NULL) {
3215 if (sopt->sopt_td != NULL)
3216 error = copyout(buf, sopt->sopt_val, valsize);
3217 else
3218 bcopy(buf, sopt->sopt_val, valsize);
3219 }
3220 return (error);
3221 }
3222
3223 int
3224 sogetopt(struct socket *so, struct sockopt *sopt)
3225 {
3226 int error, optval;
3227 struct linger l;
3228 struct timeval tv;
3229 #ifdef MAC
3230 struct mac extmac;
3231 #endif
3232
3233 CURVNET_SET(so->so_vnet);
3234 error = 0;
3235 if (sopt->sopt_level != SOL_SOCKET) {
3236 if (so->so_proto->pr_ctloutput != NULL)
3237 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3238 else
3239 error = ENOPROTOOPT;
3240 CURVNET_RESTORE();
3241 return (error);
3242 } else {
3243 switch (sopt->sopt_name) {
3244 case SO_ACCEPTFILTER:
3245 error = accept_filt_getopt(so, sopt);
3246 break;
3247
3248 case SO_LINGER:
3249 SOCK_LOCK(so);
3250 l.l_onoff = so->so_options & SO_LINGER;
3251 l.l_linger = so->so_linger;
3252 SOCK_UNLOCK(so);
3253 error = sooptcopyout(sopt, &l, sizeof l);
3254 break;
3255
3256 case SO_USELOOPBACK:
3257 case SO_DONTROUTE:
3258 case SO_DEBUG:
3259 case SO_KEEPALIVE:
3260 case SO_REUSEADDR:
3261 case SO_REUSEPORT:
3262 case SO_REUSEPORT_LB:
3263 case SO_BROADCAST:
3264 case SO_OOBINLINE:
3265 case SO_ACCEPTCONN:
3266 case SO_TIMESTAMP:
3267 case SO_BINTIME:
3268 case SO_NOSIGPIPE:
3269 case SO_NO_DDP:
3270 case SO_NO_OFFLOAD:
3271 case SO_RERROR:
3272 optval = so->so_options & sopt->sopt_name;
3273 integer:
3274 error = sooptcopyout(sopt, &optval, sizeof optval);
3275 break;
3276
3277 case SO_DOMAIN:
3278 optval = so->so_proto->pr_domain->dom_family;
3279 goto integer;
3280
3281 case SO_TYPE:
3282 optval = so->so_type;
3283 goto integer;
3284
3285 case SO_PROTOCOL:
3286 optval = so->so_proto->pr_protocol;
3287 goto integer;
3288
3289 case SO_ERROR:
3290 SOCK_LOCK(so);
3291 if (so->so_error) {
3292 optval = so->so_error;
3293 so->so_error = 0;
3294 } else {
3295 optval = so->so_rerror;
3296 so->so_rerror = 0;
3297 }
3298 SOCK_UNLOCK(so);
3299 goto integer;
3300
3301 case SO_SNDBUF:
3302 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3303 so->so_snd.sb_hiwat;
3304 goto integer;
3305
3306 case SO_RCVBUF:
3307 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3308 so->so_rcv.sb_hiwat;
3309 goto integer;
3310
3311 case SO_SNDLOWAT:
3312 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3313 so->so_snd.sb_lowat;
3314 goto integer;
3315
3316 case SO_RCVLOWAT:
3317 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3318 so->so_rcv.sb_lowat;
3319 goto integer;
3320
3321 case SO_SNDTIMEO:
3322 case SO_RCVTIMEO:
3323 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3324 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3325 #ifdef COMPAT_FREEBSD32
3326 if (SV_CURPROC_FLAG(SV_ILP32)) {
3327 struct timeval32 tv32;
3328
3329 CP(tv, tv32, tv_sec);
3330 CP(tv, tv32, tv_usec);
3331 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3332 } else
3333 #endif
3334 error = sooptcopyout(sopt, &tv, sizeof tv);
3335 break;
3336
3337 case SO_LABEL:
3338 #ifdef MAC
3339 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3340 sizeof(extmac));
3341 if (error)
3342 goto bad;
3343 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3344 so, &extmac);
3345 if (error)
3346 goto bad;
3347 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3348 #else
3349 error = EOPNOTSUPP;
3350 #endif
3351 break;
3352
3353 case SO_PEERLABEL:
3354 #ifdef MAC
3355 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3356 sizeof(extmac));
3357 if (error)
3358 goto bad;
3359 error = mac_getsockopt_peerlabel(
3360 sopt->sopt_td->td_ucred, so, &extmac);
3361 if (error)
3362 goto bad;
3363 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3364 #else
3365 error = EOPNOTSUPP;
3366 #endif
3367 break;
3368
3369 case SO_LISTENQLIMIT:
3370 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3371 goto integer;
3372
3373 case SO_LISTENQLEN:
3374 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3375 goto integer;
3376
3377 case SO_LISTENINCQLEN:
3378 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3379 goto integer;
3380
3381 case SO_TS_CLOCK:
3382 optval = so->so_ts_clock;
3383 goto integer;
3384
3385 case SO_MAX_PACING_RATE:
3386 optval = so->so_max_pacing_rate;
3387 goto integer;
3388
3389 default:
3390 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3391 error = hhook_run_socket(so, sopt,
3392 HHOOK_SOCKET_OPT);
3393 else
3394 error = ENOPROTOOPT;
3395 break;
3396 }
3397 }
3398 #ifdef MAC
3399 bad:
3400 #endif
3401 CURVNET_RESTORE();
3402 return (error);
3403 }
3404
3405 int
3406 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3407 {
3408 struct mbuf *m, *m_prev;
3409 int sopt_size = sopt->sopt_valsize;
3410
3411 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3412 if (m == NULL)
3413 return ENOBUFS;
3414 if (sopt_size > MLEN) {
3415 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3416 if ((m->m_flags & M_EXT) == 0) {
3417 m_free(m);
3418 return ENOBUFS;
3419 }
3420 m->m_len = min(MCLBYTES, sopt_size);
3421 } else {
3422 m->m_len = min(MLEN, sopt_size);
3423 }
3424 sopt_size -= m->m_len;
3425 *mp = m;
3426 m_prev = m;
3427
3428 while (sopt_size) {
3429 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3430 if (m == NULL) {
3431 m_freem(*mp);
3432 return ENOBUFS;
3433 }
3434 if (sopt_size > MLEN) {
3435 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3436 M_NOWAIT);
3437 if ((m->m_flags & M_EXT) == 0) {
3438 m_freem(m);
3439 m_freem(*mp);
3440 return ENOBUFS;
3441 }
3442 m->m_len = min(MCLBYTES, sopt_size);
3443 } else {
3444 m->m_len = min(MLEN, sopt_size);
3445 }
3446 sopt_size -= m->m_len;
3447 m_prev->m_next = m;
3448 m_prev = m;
3449 }
3450 return (0);
3451 }
3452
3453 int
3454 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3455 {
3456 struct mbuf *m0 = m;
3457
3458 if (sopt->sopt_val == NULL)
3459 return (0);
3460 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3461 if (sopt->sopt_td != NULL) {
3462 int error;
3463
3464 error = copyin(sopt->sopt_val, mtod(m, char *),
3465 m->m_len);
3466 if (error != 0) {
3467 m_freem(m0);
3468 return(error);
3469 }
3470 } else
3471 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3472 sopt->sopt_valsize -= m->m_len;
3473 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3474 m = m->m_next;
3475 }
3476 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3477 panic("ip6_sooptmcopyin");
3478 return (0);
3479 }
3480
3481 int
3482 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3483 {
3484 struct mbuf *m0 = m;
3485 size_t valsize = 0;
3486
3487 if (sopt->sopt_val == NULL)
3488 return (0);
3489 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3490 if (sopt->sopt_td != NULL) {
3491 int error;
3492
3493 error = copyout(mtod(m, char *), sopt->sopt_val,
3494 m->m_len);
3495 if (error != 0) {
3496 m_freem(m0);
3497 return(error);
3498 }
3499 } else
3500 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3501 sopt->sopt_valsize -= m->m_len;
3502 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3503 valsize += m->m_len;
3504 m = m->m_next;
3505 }
3506 if (m != NULL) {
3507 /* enough soopt buffer should be given from user-land */
3508 m_freem(m0);
3509 return(EINVAL);
3510 }
3511 sopt->sopt_valsize = valsize;
3512 return (0);
3513 }
3514
3515 /*
3516 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3517 * out-of-band data, which will then notify socket consumers.
3518 */
3519 void
3520 sohasoutofband(struct socket *so)
3521 {
3522
3523 if (so->so_sigio != NULL)
3524 pgsigio(&so->so_sigio, SIGURG, 0);
3525 selwakeuppri(&so->so_rdsel, PSOCK);
3526 }
3527
3528 int
3529 sopoll(struct socket *so, int events, struct ucred *active_cred,
3530 struct thread *td)
3531 {
3532
3533 /*
3534 * We do not need to set or assert curvnet as long as everyone uses
3535 * sopoll_generic().
3536 */
3537 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3538 td));
3539 }
3540
3541 int
3542 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3543 struct thread *td)
3544 {
3545 int revents;
3546
3547 SOCK_LOCK(so);
3548 if (SOLISTENING(so)) {
3549 if (!(events & (POLLIN | POLLRDNORM)))
3550 revents = 0;
3551 else if (!TAILQ_EMPTY(&so->sol_comp))
3552 revents = events & (POLLIN | POLLRDNORM);
3553 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3554 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3555 else {
3556 selrecord(td, &so->so_rdsel);
3557 revents = 0;
3558 }
3559 } else {
3560 revents = 0;
3561 SOCKBUF_LOCK(&so->so_snd);
3562 SOCKBUF_LOCK(&so->so_rcv);
3563 if (events & (POLLIN | POLLRDNORM))
3564 if (soreadabledata(so))
3565 revents |= events & (POLLIN | POLLRDNORM);
3566 if (events & (POLLOUT | POLLWRNORM))
3567 if (sowriteable(so))
3568 revents |= events & (POLLOUT | POLLWRNORM);
3569 if (events & (POLLPRI | POLLRDBAND))
3570 if (so->so_oobmark ||
3571 (so->so_rcv.sb_state & SBS_RCVATMARK))
3572 revents |= events & (POLLPRI | POLLRDBAND);
3573 if ((events & POLLINIGNEOF) == 0) {
3574 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3575 revents |= events & (POLLIN | POLLRDNORM);
3576 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3577 revents |= POLLHUP;
3578 }
3579 }
3580 if (revents == 0) {
3581 if (events &
3582 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3583 selrecord(td, &so->so_rdsel);
3584 so->so_rcv.sb_flags |= SB_SEL;
3585 }
3586 if (events & (POLLOUT | POLLWRNORM)) {
3587 selrecord(td, &so->so_wrsel);
3588 so->so_snd.sb_flags |= SB_SEL;
3589 }
3590 }
3591 SOCKBUF_UNLOCK(&so->so_rcv);
3592 SOCKBUF_UNLOCK(&so->so_snd);
3593 }
3594 SOCK_UNLOCK(so);
3595 return (revents);
3596 }
3597
3598 int
3599 soo_kqfilter(struct file *fp, struct knote *kn)
3600 {
3601 struct socket *so = kn->kn_fp->f_data;
3602 struct sockbuf *sb;
3603 struct knlist *knl;
3604
3605 switch (kn->kn_filter) {
3606 case EVFILT_READ:
3607 kn->kn_fop = &soread_filtops;
3608 knl = &so->so_rdsel.si_note;
3609 sb = &so->so_rcv;
3610 break;
3611 case EVFILT_WRITE:
3612 kn->kn_fop = &sowrite_filtops;
3613 knl = &so->so_wrsel.si_note;
3614 sb = &so->so_snd;
3615 break;
3616 case EVFILT_EMPTY:
3617 kn->kn_fop = &soempty_filtops;
3618 knl = &so->so_wrsel.si_note;
3619 sb = &so->so_snd;
3620 break;
3621 default:
3622 return (EINVAL);
3623 }
3624
3625 SOCK_LOCK(so);
3626 if (SOLISTENING(so)) {
3627 knlist_add(knl, kn, 1);
3628 } else {
3629 SOCKBUF_LOCK(sb);
3630 knlist_add(knl, kn, 1);
3631 sb->sb_flags |= SB_KNOTE;
3632 SOCKBUF_UNLOCK(sb);
3633 }
3634 SOCK_UNLOCK(so);
3635 return (0);
3636 }
3637
3638 /*
3639 * Some routines that return EOPNOTSUPP for entry points that are not
3640 * supported by a protocol. Fill in as needed.
3641 */
3642 int
3643 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3644 {
3645
3646 return EOPNOTSUPP;
3647 }
3648
3649 int
3650 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3651 {
3652
3653 return EOPNOTSUPP;
3654 }
3655
3656 int
3657 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3658 {
3659
3660 return EOPNOTSUPP;
3661 }
3662
3663 int
3664 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3665 {
3666
3667 return EOPNOTSUPP;
3668 }
3669
3670 int
3671 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3672 struct thread *td)
3673 {
3674
3675 return EOPNOTSUPP;
3676 }
3677
3678 int
3679 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3680 {
3681
3682 return EOPNOTSUPP;
3683 }
3684
3685 int
3686 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3687 struct thread *td)
3688 {
3689
3690 return EOPNOTSUPP;
3691 }
3692
3693 int
3694 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3695 {
3696
3697 return EOPNOTSUPP;
3698 }
3699
3700 int
3701 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3702 struct ifnet *ifp, struct thread *td)
3703 {
3704
3705 return EOPNOTSUPP;
3706 }
3707
3708 int
3709 pru_disconnect_notsupp(struct socket *so)
3710 {
3711
3712 return EOPNOTSUPP;
3713 }
3714
3715 int
3716 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3717 {
3718
3719 return EOPNOTSUPP;
3720 }
3721
3722 int
3723 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3724 {
3725
3726 return EOPNOTSUPP;
3727 }
3728
3729 int
3730 pru_rcvd_notsupp(struct socket *so, int flags)
3731 {
3732
3733 return EOPNOTSUPP;
3734 }
3735
3736 int
3737 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3738 {
3739
3740 return EOPNOTSUPP;
3741 }
3742
3743 int
3744 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3745 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3746 {
3747
3748 if (control != NULL)
3749 m_freem(control);
3750 if ((flags & PRUS_NOTREADY) == 0)
3751 m_freem(m);
3752 return (EOPNOTSUPP);
3753 }
3754
3755 int
3756 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3757 {
3758
3759 return (EOPNOTSUPP);
3760 }
3761
3762 /*
3763 * This isn't really a ``null'' operation, but it's the default one and
3764 * doesn't do anything destructive.
3765 */
3766 int
3767 pru_sense_null(struct socket *so, struct stat *sb)
3768 {
3769
3770 sb->st_blksize = so->so_snd.sb_hiwat;
3771 return 0;
3772 }
3773
3774 int
3775 pru_shutdown_notsupp(struct socket *so)
3776 {
3777
3778 return EOPNOTSUPP;
3779 }
3780
3781 int
3782 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3783 {
3784
3785 return EOPNOTSUPP;
3786 }
3787
3788 int
3789 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3790 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3791 {
3792
3793 return EOPNOTSUPP;
3794 }
3795
3796 int
3797 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3798 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3799 {
3800
3801 return EOPNOTSUPP;
3802 }
3803
3804 int
3805 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3806 struct thread *td)
3807 {
3808
3809 return EOPNOTSUPP;
3810 }
3811
3812 static void
3813 filt_sordetach(struct knote *kn)
3814 {
3815 struct socket *so = kn->kn_fp->f_data;
3816
3817 so_rdknl_lock(so);
3818 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3819 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3820 so->so_rcv.sb_flags &= ~SB_KNOTE;
3821 so_rdknl_unlock(so);
3822 }
3823
3824 /*ARGSUSED*/
3825 static int
3826 filt_soread(struct knote *kn, long hint)
3827 {
3828 struct socket *so;
3829
3830 so = kn->kn_fp->f_data;
3831
3832 if (SOLISTENING(so)) {
3833 SOCK_LOCK_ASSERT(so);
3834 kn->kn_data = so->sol_qlen;
3835 if (so->so_error) {
3836 kn->kn_flags |= EV_EOF;
3837 kn->kn_fflags = so->so_error;
3838 return (1);
3839 }
3840 return (!TAILQ_EMPTY(&so->sol_comp));
3841 }
3842
3843 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3844
3845 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3846 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3847 kn->kn_flags |= EV_EOF;
3848 kn->kn_fflags = so->so_error;
3849 return (1);
3850 } else if (so->so_error || so->so_rerror)
3851 return (1);
3852
3853 if (kn->kn_sfflags & NOTE_LOWAT) {
3854 if (kn->kn_data >= kn->kn_sdata)
3855 return (1);
3856 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3857 return (1);
3858
3859 /* This hook returning non-zero indicates an event, not error */
3860 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3861 }
3862
3863 static void
3864 filt_sowdetach(struct knote *kn)
3865 {
3866 struct socket *so = kn->kn_fp->f_data;
3867
3868 so_wrknl_lock(so);
3869 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3870 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3871 so->so_snd.sb_flags &= ~SB_KNOTE;
3872 so_wrknl_unlock(so);
3873 }
3874
3875 /*ARGSUSED*/
3876 static int
3877 filt_sowrite(struct knote *kn, long hint)
3878 {
3879 struct socket *so;
3880
3881 so = kn->kn_fp->f_data;
3882
3883 if (SOLISTENING(so))
3884 return (0);
3885
3886 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3887 kn->kn_data = sbspace(&so->so_snd);
3888
3889 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3890
3891 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3892 kn->kn_flags |= EV_EOF;
3893 kn->kn_fflags = so->so_error;
3894 return (1);
3895 } else if (so->so_error) /* temporary udp error */
3896 return (1);
3897 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3898 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3899 return (0);
3900 else if (kn->kn_sfflags & NOTE_LOWAT)
3901 return (kn->kn_data >= kn->kn_sdata);
3902 else
3903 return (kn->kn_data >= so->so_snd.sb_lowat);
3904 }
3905
3906 static int
3907 filt_soempty(struct knote *kn, long hint)
3908 {
3909 struct socket *so;
3910
3911 so = kn->kn_fp->f_data;
3912
3913 if (SOLISTENING(so))
3914 return (1);
3915
3916 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3917 kn->kn_data = sbused(&so->so_snd);
3918
3919 if (kn->kn_data == 0)
3920 return (1);
3921 else
3922 return (0);
3923 }
3924
3925 int
3926 socheckuid(struct socket *so, uid_t uid)
3927 {
3928
3929 if (so == NULL)
3930 return (EPERM);
3931 if (so->so_cred->cr_uid != uid)
3932 return (EPERM);
3933 return (0);
3934 }
3935
3936 /*
3937 * These functions are used by protocols to notify the socket layer (and its
3938 * consumers) of state changes in the sockets driven by protocol-side events.
3939 */
3940
3941 /*
3942 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3943 *
3944 * Normal sequence from the active (originating) side is that
3945 * soisconnecting() is called during processing of connect() call, resulting
3946 * in an eventual call to soisconnected() if/when the connection is
3947 * established. When the connection is torn down soisdisconnecting() is
3948 * called during processing of disconnect() call, and soisdisconnected() is
3949 * called when the connection to the peer is totally severed. The semantics
3950 * of these routines are such that connectionless protocols can call
3951 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3952 * calls when setting up a ``connection'' takes no time.
3953 *
3954 * From the passive side, a socket is created with two queues of sockets:
3955 * so_incomp for connections in progress and so_comp for connections already
3956 * made and awaiting user acceptance. As a protocol is preparing incoming
3957 * connections, it creates a socket structure queued on so_incomp by calling
3958 * sonewconn(). When the connection is established, soisconnected() is
3959 * called, and transfers the socket structure to so_comp, making it available
3960 * to accept().
3961 *
3962 * If a socket is closed with sockets on either so_incomp or so_comp, these
3963 * sockets are dropped.
3964 *
3965 * If higher-level protocols are implemented in the kernel, the wakeups done
3966 * here will sometimes cause software-interrupt process scheduling.
3967 */
3968 void
3969 soisconnecting(struct socket *so)
3970 {
3971
3972 SOCK_LOCK(so);
3973 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3974 so->so_state |= SS_ISCONNECTING;
3975 SOCK_UNLOCK(so);
3976 }
3977
3978 void
3979 soisconnected(struct socket *so)
3980 {
3981 bool last __diagused;
3982
3983 SOCK_LOCK(so);
3984 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3985 so->so_state |= SS_ISCONNECTED;
3986
3987 if (so->so_qstate == SQ_INCOMP) {
3988 struct socket *head = so->so_listen;
3989 int ret;
3990
3991 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3992 /*
3993 * Promoting a socket from incomplete queue to complete, we
3994 * need to go through reverse order of locking. We first do
3995 * trylock, and if that doesn't succeed, we go the hard way
3996 * leaving a reference and rechecking consistency after proper
3997 * locking.
3998 */
3999 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4000 soref(head);
4001 SOCK_UNLOCK(so);
4002 SOLISTEN_LOCK(head);
4003 SOCK_LOCK(so);
4004 if (__predict_false(head != so->so_listen)) {
4005 /*
4006 * The socket went off the listen queue,
4007 * should be lost race to close(2) of sol.
4008 * The socket is about to soabort().
4009 */
4010 SOCK_UNLOCK(so);
4011 sorele(head);
4012 return;
4013 }
4014 last = refcount_release(&head->so_count);
4015 KASSERT(!last, ("%s: released last reference for %p",
4016 __func__, head));
4017 }
4018 again:
4019 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4020 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4021 head->sol_incqlen--;
4022 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4023 head->sol_qlen++;
4024 so->so_qstate = SQ_COMP;
4025 SOCK_UNLOCK(so);
4026 solisten_wakeup(head); /* unlocks */
4027 } else {
4028 SOCKBUF_LOCK(&so->so_rcv);
4029 soupcall_set(so, SO_RCV,
4030 head->sol_accept_filter->accf_callback,
4031 head->sol_accept_filter_arg);
4032 so->so_options &= ~SO_ACCEPTFILTER;
4033 ret = head->sol_accept_filter->accf_callback(so,
4034 head->sol_accept_filter_arg, M_NOWAIT);
4035 if (ret == SU_ISCONNECTED) {
4036 soupcall_clear(so, SO_RCV);
4037 SOCKBUF_UNLOCK(&so->so_rcv);
4038 goto again;
4039 }
4040 SOCKBUF_UNLOCK(&so->so_rcv);
4041 SOCK_UNLOCK(so);
4042 SOLISTEN_UNLOCK(head);
4043 }
4044 return;
4045 }
4046 SOCK_UNLOCK(so);
4047 wakeup(&so->so_timeo);
4048 sorwakeup(so);
4049 sowwakeup(so);
4050 }
4051
4052 void
4053 soisdisconnecting(struct socket *so)
4054 {
4055
4056 SOCK_LOCK(so);
4057 so->so_state &= ~SS_ISCONNECTING;
4058 so->so_state |= SS_ISDISCONNECTING;
4059
4060 if (!SOLISTENING(so)) {
4061 SOCKBUF_LOCK(&so->so_rcv);
4062 socantrcvmore_locked(so);
4063 SOCKBUF_LOCK(&so->so_snd);
4064 socantsendmore_locked(so);
4065 }
4066 SOCK_UNLOCK(so);
4067 wakeup(&so->so_timeo);
4068 }
4069
4070 void
4071 soisdisconnected(struct socket *so)
4072 {
4073
4074 SOCK_LOCK(so);
4075
4076 /*
4077 * There is at least one reader of so_state that does not
4078 * acquire socket lock, namely soreceive_generic(). Ensure
4079 * that it never sees all flags that track connection status
4080 * cleared, by ordering the update with a barrier semantic of
4081 * our release thread fence.
4082 */
4083 so->so_state |= SS_ISDISCONNECTED;
4084 atomic_thread_fence_rel();
4085 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4086
4087 if (!SOLISTENING(so)) {
4088 SOCK_UNLOCK(so);
4089 SOCKBUF_LOCK(&so->so_rcv);
4090 socantrcvmore_locked(so);
4091 SOCKBUF_LOCK(&so->so_snd);
4092 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4093 socantsendmore_locked(so);
4094 } else
4095 SOCK_UNLOCK(so);
4096 wakeup(&so->so_timeo);
4097 }
4098
4099 int
4100 soiolock(struct socket *so, struct sx *sx, int flags)
4101 {
4102 int error;
4103
4104 KASSERT((flags & SBL_VALID) == flags,
4105 ("soiolock: invalid flags %#x", flags));
4106
4107 if ((flags & SBL_WAIT) != 0) {
4108 if ((flags & SBL_NOINTR) != 0) {
4109 sx_xlock(sx);
4110 } else {
4111 error = sx_xlock_sig(sx);
4112 if (error != 0)
4113 return (error);
4114 }
4115 } else if (!sx_try_xlock(sx)) {
4116 return (EWOULDBLOCK);
4117 }
4118
4119 if (__predict_false(SOLISTENING(so))) {
4120 sx_xunlock(sx);
4121 return (ENOTCONN);
4122 }
4123 return (0);
4124 }
4125
4126 void
4127 soiounlock(struct sx *sx)
4128 {
4129 sx_xunlock(sx);
4130 }
4131
4132 /*
4133 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4134 */
4135 struct sockaddr *
4136 sodupsockaddr(const struct sockaddr *sa, int mflags)
4137 {
4138 struct sockaddr *sa2;
4139
4140 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4141 if (sa2)
4142 bcopy(sa, sa2, sa->sa_len);
4143 return sa2;
4144 }
4145
4146 /*
4147 * Register per-socket destructor.
4148 */
4149 void
4150 sodtor_set(struct socket *so, so_dtor_t *func)
4151 {
4152
4153 SOCK_LOCK_ASSERT(so);
4154 so->so_dtor = func;
4155 }
4156
4157 /*
4158 * Register per-socket buffer upcalls.
4159 */
4160 void
4161 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4162 {
4163 struct sockbuf *sb;
4164
4165 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4166
4167 switch (which) {
4168 case SO_RCV:
4169 sb = &so->so_rcv;
4170 break;
4171 case SO_SND:
4172 sb = &so->so_snd;
4173 break;
4174 default:
4175 panic("soupcall_set: bad which");
4176 }
4177 SOCKBUF_LOCK_ASSERT(sb);
4178 sb->sb_upcall = func;
4179 sb->sb_upcallarg = arg;
4180 sb->sb_flags |= SB_UPCALL;
4181 }
4182
4183 void
4184 soupcall_clear(struct socket *so, int which)
4185 {
4186 struct sockbuf *sb;
4187
4188 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4189
4190 switch (which) {
4191 case SO_RCV:
4192 sb = &so->so_rcv;
4193 break;
4194 case SO_SND:
4195 sb = &so->so_snd;
4196 break;
4197 default:
4198 panic("soupcall_clear: bad which");
4199 }
4200 SOCKBUF_LOCK_ASSERT(sb);
4201 KASSERT(sb->sb_upcall != NULL,
4202 ("%s: so %p no upcall to clear", __func__, so));
4203 sb->sb_upcall = NULL;
4204 sb->sb_upcallarg = NULL;
4205 sb->sb_flags &= ~SB_UPCALL;
4206 }
4207
4208 void
4209 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4210 {
4211
4212 SOLISTEN_LOCK_ASSERT(so);
4213 so->sol_upcall = func;
4214 so->sol_upcallarg = arg;
4215 }
4216
4217 static void
4218 so_rdknl_lock(void *arg)
4219 {
4220 struct socket *so = arg;
4221
4222 if (SOLISTENING(so))
4223 SOCK_LOCK(so);
4224 else
4225 SOCKBUF_LOCK(&so->so_rcv);
4226 }
4227
4228 static void
4229 so_rdknl_unlock(void *arg)
4230 {
4231 struct socket *so = arg;
4232
4233 if (SOLISTENING(so))
4234 SOCK_UNLOCK(so);
4235 else
4236 SOCKBUF_UNLOCK(&so->so_rcv);
4237 }
4238
4239 static void
4240 so_rdknl_assert_lock(void *arg, int what)
4241 {
4242 struct socket *so = arg;
4243
4244 if (what == LA_LOCKED) {
4245 if (SOLISTENING(so))
4246 SOCK_LOCK_ASSERT(so);
4247 else
4248 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4249 } else {
4250 if (SOLISTENING(so))
4251 SOCK_UNLOCK_ASSERT(so);
4252 else
4253 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4254 }
4255 }
4256
4257 static void
4258 so_wrknl_lock(void *arg)
4259 {
4260 struct socket *so = arg;
4261
4262 if (SOLISTENING(so))
4263 SOCK_LOCK(so);
4264 else
4265 SOCKBUF_LOCK(&so->so_snd);
4266 }
4267
4268 static void
4269 so_wrknl_unlock(void *arg)
4270 {
4271 struct socket *so = arg;
4272
4273 if (SOLISTENING(so))
4274 SOCK_UNLOCK(so);
4275 else
4276 SOCKBUF_UNLOCK(&so->so_snd);
4277 }
4278
4279 static void
4280 so_wrknl_assert_lock(void *arg, int what)
4281 {
4282 struct socket *so = arg;
4283
4284 if (what == LA_LOCKED) {
4285 if (SOLISTENING(so))
4286 SOCK_LOCK_ASSERT(so);
4287 else
4288 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4289 } else {
4290 if (SOLISTENING(so))
4291 SOCK_UNLOCK_ASSERT(so);
4292 else
4293 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4294 }
4295 }
4296
4297 /*
4298 * Create an external-format (``xsocket'') structure using the information in
4299 * the kernel-format socket structure pointed to by so. This is done to
4300 * reduce the spew of irrelevant information over this interface, to isolate
4301 * user code from changes in the kernel structure, and potentially to provide
4302 * information-hiding if we decide that some of this information should be
4303 * hidden from users.
4304 */
4305 void
4306 sotoxsocket(struct socket *so, struct xsocket *xso)
4307 {
4308
4309 bzero(xso, sizeof(*xso));
4310 xso->xso_len = sizeof *xso;
4311 xso->xso_so = (uintptr_t)so;
4312 xso->so_type = so->so_type;
4313 xso->so_options = so->so_options;
4314 xso->so_linger = so->so_linger;
4315 xso->so_state = so->so_state;
4316 xso->so_pcb = (uintptr_t)so->so_pcb;
4317 xso->xso_protocol = so->so_proto->pr_protocol;
4318 xso->xso_family = so->so_proto->pr_domain->dom_family;
4319 xso->so_timeo = so->so_timeo;
4320 xso->so_error = so->so_error;
4321 xso->so_uid = so->so_cred->cr_uid;
4322 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4323 if (SOLISTENING(so)) {
4324 xso->so_qlen = so->sol_qlen;
4325 xso->so_incqlen = so->sol_incqlen;
4326 xso->so_qlimit = so->sol_qlimit;
4327 xso->so_oobmark = 0;
4328 } else {
4329 xso->so_state |= so->so_qstate;
4330 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4331 xso->so_oobmark = so->so_oobmark;
4332 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4333 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4334 }
4335 }
4336
4337 struct sockbuf *
4338 so_sockbuf_rcv(struct socket *so)
4339 {
4340
4341 return (&so->so_rcv);
4342 }
4343
4344 struct sockbuf *
4345 so_sockbuf_snd(struct socket *so)
4346 {
4347
4348 return (&so->so_snd);
4349 }
4350
4351 int
4352 so_state_get(const struct socket *so)
4353 {
4354
4355 return (so->so_state);
4356 }
4357
4358 void
4359 so_state_set(struct socket *so, int val)
4360 {
4361
4362 so->so_state = val;
4363 }
4364
4365 int
4366 so_options_get(const struct socket *so)
4367 {
4368
4369 return (so->so_options);
4370 }
4371
4372 void
4373 so_options_set(struct socket *so, int val)
4374 {
4375
4376 so->so_options = val;
4377 }
4378
4379 int
4380 so_error_get(const struct socket *so)
4381 {
4382
4383 return (so->so_error);
4384 }
4385
4386 void
4387 so_error_set(struct socket *so, int val)
4388 {
4389
4390 so->so_error = val;
4391 }
4392
4393 int
4394 so_linger_get(const struct socket *so)
4395 {
4396
4397 return (so->so_linger);
4398 }
4399
4400 void
4401 so_linger_set(struct socket *so, int val)
4402 {
4403
4404 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4405 ("%s: val %d out of range", __func__, val));
4406
4407 so->so_linger = val;
4408 }
4409
4410 struct protosw *
4411 so_protosw_get(const struct socket *so)
4412 {
4413
4414 return (so->so_proto);
4415 }
4416
4417 void
4418 so_protosw_set(struct socket *so, struct protosw *val)
4419 {
4420
4421 so->so_proto = val;
4422 }
4423
4424 void
4425 so_sorwakeup(struct socket *so)
4426 {
4427
4428 sorwakeup(so);
4429 }
4430
4431 void
4432 so_sowwakeup(struct socket *so)
4433 {
4434
4435 sowwakeup(so);
4436 }
4437
4438 void
4439 so_sorwakeup_locked(struct socket *so)
4440 {
4441
4442 sorwakeup_locked(so);
4443 }
4444
4445 void
4446 so_sowwakeup_locked(struct socket *so)
4447 {
4448
4449 sowwakeup_locked(so);
4450 }
4451
4452 void
4453 so_lock(struct socket *so)
4454 {
4455
4456 SOCK_LOCK(so);
4457 }
4458
4459 void
4460 so_unlock(struct socket *so)
4461 {
4462
4463 SOCK_UNLOCK(so);
4464 }
Cache object: 0531686818d37fe1a94d6b7873b2eb07
|