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
sysctl_somaxconn(SYSCTL_HANDLER_ARGS)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_MPSAFE, 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_MPSAFE, 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
socket_zone_change(void * tag)282 socket_zone_change(void *tag)
283 {
284
285 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
286 }
287
288 static void
socket_hhook_register(int subtype)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
socket_hhook_deregister(int subtype)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
socket_init(void * tag)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
socket_vnet_init(const void * unused __unused)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
socket_vnet_uninit(const void * unused __unused)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
init_maxsockets(void * ignored)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
sysctl_maxsockets(SYSCTL_HANDLER_ARGS)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 && newmaxsockets != maxsockets) {
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_MPSAFE, &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 *
soalloc(struct vnet * vnet)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
sodealloc(struct socket * so)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
socreate(int dom,struct socket ** aso,int type,int proto,struct ucred * cred,struct thread * td)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 *
sonewconn(struct socket * head,int connstatus)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 if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1609 error = EINVAL;
1610 goto release;
1611 }
1612 }
1613 #endif
1614
1615 restart:
1616 do {
1617 SOCKBUF_LOCK(&so->so_snd);
1618 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1619 SOCKBUF_UNLOCK(&so->so_snd);
1620 error = EPIPE;
1621 goto release;
1622 }
1623 if (so->so_error) {
1624 error = so->so_error;
1625 so->so_error = 0;
1626 SOCKBUF_UNLOCK(&so->so_snd);
1627 goto release;
1628 }
1629 if ((so->so_state & SS_ISCONNECTED) == 0) {
1630 /*
1631 * `sendto' and `sendmsg' is allowed on a connection-
1632 * based socket if it supports implied connect.
1633 * Return ENOTCONN if not connected and no address is
1634 * supplied.
1635 */
1636 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1637 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1638 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1639 !(resid == 0 && clen != 0)) {
1640 SOCKBUF_UNLOCK(&so->so_snd);
1641 error = ENOTCONN;
1642 goto release;
1643 }
1644 } else if (addr == NULL) {
1645 SOCKBUF_UNLOCK(&so->so_snd);
1646 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1647 error = ENOTCONN;
1648 else
1649 error = EDESTADDRREQ;
1650 goto release;
1651 }
1652 }
1653 space = sbspace(&so->so_snd);
1654 if (flags & MSG_OOB)
1655 space += 1024;
1656 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1657 clen > so->so_snd.sb_hiwat) {
1658 SOCKBUF_UNLOCK(&so->so_snd);
1659 error = EMSGSIZE;
1660 goto release;
1661 }
1662 if (space < resid + clen &&
1663 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1664 if ((so->so_state & SS_NBIO) ||
1665 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1666 SOCKBUF_UNLOCK(&so->so_snd);
1667 error = EWOULDBLOCK;
1668 goto release;
1669 }
1670 error = sbwait(&so->so_snd);
1671 SOCKBUF_UNLOCK(&so->so_snd);
1672 if (error)
1673 goto release;
1674 goto restart;
1675 }
1676 SOCKBUF_UNLOCK(&so->so_snd);
1677 space -= clen;
1678 do {
1679 if (uio == NULL) {
1680 resid = 0;
1681 if (flags & MSG_EOR)
1682 top->m_flags |= M_EOR;
1683 #ifdef KERN_TLS
1684 if (tls != NULL) {
1685 ktls_frame(top, tls, &tls_enq_cnt,
1686 tls_rtype);
1687 tls_rtype = TLS_RLTYPE_APP;
1688 }
1689 #endif
1690 } else {
1691 /*
1692 * Copy the data from userland into a mbuf
1693 * chain. If resid is 0, which can happen
1694 * only if we have control to send, then
1695 * a single empty mbuf is returned. This
1696 * is a workaround to prevent protocol send
1697 * methods to panic.
1698 */
1699 #ifdef KERN_TLS
1700 if (tls != NULL) {
1701 top = m_uiotombuf(uio, M_WAITOK, space,
1702 tls->params.max_frame_len,
1703 M_EXTPG |
1704 ((flags & MSG_EOR) ? M_EOR : 0));
1705 if (top != NULL) {
1706 ktls_frame(top, tls,
1707 &tls_enq_cnt, tls_rtype);
1708 }
1709 tls_rtype = TLS_RLTYPE_APP;
1710 } else
1711 #endif
1712 top = m_uiotombuf(uio, M_WAITOK, space,
1713 (atomic ? max_hdr : 0),
1714 (atomic ? M_PKTHDR : 0) |
1715 ((flags & MSG_EOR) ? M_EOR : 0));
1716 if (top == NULL) {
1717 error = EFAULT; /* only possible error */
1718 goto release;
1719 }
1720 space -= resid - uio->uio_resid;
1721 resid = uio->uio_resid;
1722 }
1723 if (dontroute) {
1724 SOCK_LOCK(so);
1725 so->so_options |= SO_DONTROUTE;
1726 SOCK_UNLOCK(so);
1727 }
1728 /*
1729 * XXX all the SBS_CANTSENDMORE checks previously
1730 * done could be out of date. We could have received
1731 * a reset packet in an interrupt or maybe we slept
1732 * while doing page faults in uiomove() etc. We
1733 * could probably recheck again inside the locking
1734 * protection here, but there are probably other
1735 * places that this also happens. We must rethink
1736 * this.
1737 */
1738 VNET_SO_ASSERT(so);
1739
1740 pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1741 /*
1742 * If the user set MSG_EOF, the protocol understands
1743 * this flag and nothing left to send then use
1744 * PRU_SEND_EOF instead of PRU_SEND.
1745 */
1746 ((flags & MSG_EOF) &&
1747 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1748 (resid <= 0)) ?
1749 PRUS_EOF :
1750 /* If there is more to send set PRUS_MORETOCOME. */
1751 (flags & MSG_MORETOCOME) ||
1752 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1753
1754 #ifdef KERN_TLS
1755 pru_flag |= tls_pruflag;
1756 #endif
1757
1758 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1759 pru_flag, top, addr, control, td);
1760
1761 if (dontroute) {
1762 SOCK_LOCK(so);
1763 so->so_options &= ~SO_DONTROUTE;
1764 SOCK_UNLOCK(so);
1765 }
1766
1767 #ifdef KERN_TLS
1768 if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1769 if (error != 0) {
1770 m_freem(top);
1771 top = NULL;
1772 } else {
1773 soref(so);
1774 ktls_enqueue(top, so, tls_enq_cnt);
1775 }
1776 }
1777 #endif
1778 clen = 0;
1779 control = NULL;
1780 top = NULL;
1781 if (error)
1782 goto release;
1783 } while (resid && space > 0);
1784 } while (resid);
1785
1786 release:
1787 SOCK_IO_SEND_UNLOCK(so);
1788 out:
1789 #ifdef KERN_TLS
1790 if (tls != NULL)
1791 ktls_free(tls);
1792 #endif
1793 if (top != NULL)
1794 m_freem(top);
1795 if (control != NULL)
1796 m_freem(control);
1797 return (error);
1798 }
1799
1800 int
1801 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1802 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1803 {
1804 int error;
1805
1806 CURVNET_SET(so->so_vnet);
1807 if (!SOLISTENING(so))
1808 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1809 top, control, flags, td);
1810 else {
1811 m_freem(top);
1812 m_freem(control);
1813 error = ENOTCONN;
1814 }
1815 CURVNET_RESTORE();
1816 return (error);
1817 }
1818
1819 /*
1820 * The part of soreceive() that implements reading non-inline out-of-band
1821 * data from a socket. For more complete comments, see soreceive(), from
1822 * which this code originated.
1823 *
1824 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1825 * unable to return an mbuf chain to the caller.
1826 */
1827 static int
1828 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1829 {
1830 struct protosw *pr = so->so_proto;
1831 struct mbuf *m;
1832 int error;
1833
1834 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1835 VNET_SO_ASSERT(so);
1836
1837 m = m_get(M_WAITOK, MT_DATA);
1838 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1839 if (error)
1840 goto bad;
1841 do {
1842 error = uiomove(mtod(m, void *),
1843 (int) min(uio->uio_resid, m->m_len), uio);
1844 m = m_free(m);
1845 } while (uio->uio_resid && error == 0 && m);
1846 bad:
1847 if (m != NULL)
1848 m_freem(m);
1849 return (error);
1850 }
1851
1852 /*
1853 * Following replacement or removal of the first mbuf on the first mbuf chain
1854 * of a socket buffer, push necessary state changes back into the socket
1855 * buffer so that other consumers see the values consistently. 'nextrecord'
1856 * is the callers locally stored value of the original value of
1857 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1858 * NOTE: 'nextrecord' may be NULL.
1859 */
1860 static __inline void
1861 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1862 {
1863
1864 SOCKBUF_LOCK_ASSERT(sb);
1865 /*
1866 * First, update for the new value of nextrecord. If necessary, make
1867 * it the first record.
1868 */
1869 if (sb->sb_mb != NULL)
1870 sb->sb_mb->m_nextpkt = nextrecord;
1871 else
1872 sb->sb_mb = nextrecord;
1873
1874 /*
1875 * Now update any dependent socket buffer fields to reflect the new
1876 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1877 * addition of a second clause that takes care of the case where
1878 * sb_mb has been updated, but remains the last record.
1879 */
1880 if (sb->sb_mb == NULL) {
1881 sb->sb_mbtail = NULL;
1882 sb->sb_lastrecord = NULL;
1883 } else if (sb->sb_mb->m_nextpkt == NULL)
1884 sb->sb_lastrecord = sb->sb_mb;
1885 }
1886
1887 /*
1888 * Implement receive operations on a socket. We depend on the way that
1889 * records are added to the sockbuf by sbappend. In particular, each record
1890 * (mbufs linked through m_next) must begin with an address if the protocol
1891 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1892 * data, and then zero or more mbufs of data. In order to allow parallelism
1893 * between network receive and copying to user space, as well as avoid
1894 * sleeping with a mutex held, we release the socket buffer mutex during the
1895 * user space copy. Although the sockbuf is locked, new data may still be
1896 * appended, and thus we must maintain consistency of the sockbuf during that
1897 * time.
1898 *
1899 * The caller may receive the data as a single mbuf chain by supplying an
1900 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1901 * the count in uio_resid.
1902 */
1903 int
1904 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1905 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1906 {
1907 struct mbuf *m, **mp;
1908 int flags, error, offset;
1909 ssize_t len;
1910 struct protosw *pr = so->so_proto;
1911 struct mbuf *nextrecord;
1912 int moff, type = 0;
1913 ssize_t orig_resid = uio->uio_resid;
1914
1915 mp = mp0;
1916 if (psa != NULL)
1917 *psa = NULL;
1918 if (controlp != NULL)
1919 *controlp = NULL;
1920 if (flagsp != NULL)
1921 flags = *flagsp &~ MSG_EOR;
1922 else
1923 flags = 0;
1924 if (flags & MSG_OOB)
1925 return (soreceive_rcvoob(so, uio, flags));
1926 if (mp != NULL)
1927 *mp = NULL;
1928 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1929 && uio->uio_resid) {
1930 VNET_SO_ASSERT(so);
1931 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1932 }
1933
1934 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1935 if (error)
1936 return (error);
1937
1938 restart:
1939 SOCKBUF_LOCK(&so->so_rcv);
1940 m = so->so_rcv.sb_mb;
1941 /*
1942 * If we have less data than requested, block awaiting more (subject
1943 * to any timeout) if:
1944 * 1. the current count is less than the low water mark, or
1945 * 2. MSG_DONTWAIT is not set
1946 */
1947 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1948 sbavail(&so->so_rcv) < uio->uio_resid) &&
1949 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1950 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1951 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1952 ("receive: m == %p sbavail == %u",
1953 m, sbavail(&so->so_rcv)));
1954 if (so->so_error || so->so_rerror) {
1955 if (m != NULL)
1956 goto dontblock;
1957 if (so->so_error)
1958 error = so->so_error;
1959 else
1960 error = so->so_rerror;
1961 if ((flags & MSG_PEEK) == 0) {
1962 if (so->so_error)
1963 so->so_error = 0;
1964 else
1965 so->so_rerror = 0;
1966 }
1967 SOCKBUF_UNLOCK(&so->so_rcv);
1968 goto release;
1969 }
1970 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1971 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1972 if (m != NULL)
1973 goto dontblock;
1974 #ifdef KERN_TLS
1975 else if (so->so_rcv.sb_tlsdcc == 0 &&
1976 so->so_rcv.sb_tlscc == 0) {
1977 #else
1978 else {
1979 #endif
1980 SOCKBUF_UNLOCK(&so->so_rcv);
1981 goto release;
1982 }
1983 }
1984 for (; m != NULL; m = m->m_next)
1985 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1986 m = so->so_rcv.sb_mb;
1987 goto dontblock;
1988 }
1989 if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1990 SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1991 (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1992 SOCKBUF_UNLOCK(&so->so_rcv);
1993 error = ENOTCONN;
1994 goto release;
1995 }
1996 if (uio->uio_resid == 0) {
1997 SOCKBUF_UNLOCK(&so->so_rcv);
1998 goto release;
1999 }
2000 if ((so->so_state & SS_NBIO) ||
2001 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2002 SOCKBUF_UNLOCK(&so->so_rcv);
2003 error = EWOULDBLOCK;
2004 goto release;
2005 }
2006 SBLASTRECORDCHK(&so->so_rcv);
2007 SBLASTMBUFCHK(&so->so_rcv);
2008 error = sbwait(&so->so_rcv);
2009 SOCKBUF_UNLOCK(&so->so_rcv);
2010 if (error)
2011 goto release;
2012 goto restart;
2013 }
2014 dontblock:
2015 /*
2016 * From this point onward, we maintain 'nextrecord' as a cache of the
2017 * pointer to the next record in the socket buffer. We must keep the
2018 * various socket buffer pointers and local stack versions of the
2019 * pointers in sync, pushing out modifications before dropping the
2020 * socket buffer mutex, and re-reading them when picking it up.
2021 *
2022 * Otherwise, we will race with the network stack appending new data
2023 * or records onto the socket buffer by using inconsistent/stale
2024 * versions of the field, possibly resulting in socket buffer
2025 * corruption.
2026 *
2027 * By holding the high-level sblock(), we prevent simultaneous
2028 * readers from pulling off the front of the socket buffer.
2029 */
2030 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2031 if (uio->uio_td)
2032 uio->uio_td->td_ru.ru_msgrcv++;
2033 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2034 SBLASTRECORDCHK(&so->so_rcv);
2035 SBLASTMBUFCHK(&so->so_rcv);
2036 nextrecord = m->m_nextpkt;
2037 if (pr->pr_flags & PR_ADDR) {
2038 KASSERT(m->m_type == MT_SONAME,
2039 ("m->m_type == %d", m->m_type));
2040 orig_resid = 0;
2041 if (psa != NULL)
2042 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2043 M_NOWAIT);
2044 if (flags & MSG_PEEK) {
2045 m = m->m_next;
2046 } else {
2047 sbfree(&so->so_rcv, m);
2048 so->so_rcv.sb_mb = m_free(m);
2049 m = so->so_rcv.sb_mb;
2050 sockbuf_pushsync(&so->so_rcv, nextrecord);
2051 }
2052 }
2053
2054 /*
2055 * Process one or more MT_CONTROL mbufs present before any data mbufs
2056 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
2057 * just copy the data; if !MSG_PEEK, we call into the protocol to
2058 * perform externalization (or freeing if controlp == NULL).
2059 */
2060 if (m != NULL && m->m_type == MT_CONTROL) {
2061 struct mbuf *cm = NULL, *cmn;
2062 struct mbuf **cme = &cm;
2063 #ifdef KERN_TLS
2064 struct cmsghdr *cmsg;
2065 struct tls_get_record tgr;
2066
2067 /*
2068 * For MSG_TLSAPPDATA, check for a non-application data
2069 * record. If found, return ENXIO without removing
2070 * it from the receive queue. This allows a subsequent
2071 * call without MSG_TLSAPPDATA to receive it.
2072 * Note that, for TLS, there should only be a single
2073 * control mbuf with the TLS_GET_RECORD message in it.
2074 */
2075 if (flags & MSG_TLSAPPDATA) {
2076 cmsg = mtod(m, struct cmsghdr *);
2077 if (cmsg->cmsg_type == TLS_GET_RECORD &&
2078 cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2079 memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2080 /* This will need to change for TLS 1.3. */
2081 if (tgr.tls_type != TLS_RLTYPE_APP) {
2082 SOCKBUF_UNLOCK(&so->so_rcv);
2083 error = ENXIO;
2084 goto release;
2085 }
2086 }
2087 }
2088 #endif
2089
2090 do {
2091 if (flags & MSG_PEEK) {
2092 if (controlp != NULL) {
2093 *controlp = m_copym(m, 0, m->m_len,
2094 M_NOWAIT);
2095 controlp = &(*controlp)->m_next;
2096 }
2097 m = m->m_next;
2098 } else {
2099 sbfree(&so->so_rcv, m);
2100 so->so_rcv.sb_mb = m->m_next;
2101 m->m_next = NULL;
2102 *cme = m;
2103 cme = &(*cme)->m_next;
2104 m = so->so_rcv.sb_mb;
2105 }
2106 } while (m != NULL && m->m_type == MT_CONTROL);
2107 if ((flags & MSG_PEEK) == 0)
2108 sockbuf_pushsync(&so->so_rcv, nextrecord);
2109 while (cm != NULL) {
2110 cmn = cm->m_next;
2111 cm->m_next = NULL;
2112 if (pr->pr_domain->dom_externalize != NULL) {
2113 SOCKBUF_UNLOCK(&so->so_rcv);
2114 VNET_SO_ASSERT(so);
2115 error = (*pr->pr_domain->dom_externalize)
2116 (cm, controlp, flags);
2117 SOCKBUF_LOCK(&so->so_rcv);
2118 } else if (controlp != NULL)
2119 *controlp = cm;
2120 else
2121 m_freem(cm);
2122 if (controlp != NULL) {
2123 while (*controlp != NULL)
2124 controlp = &(*controlp)->m_next;
2125 }
2126 cm = cmn;
2127 }
2128 if (m != NULL)
2129 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2130 else
2131 nextrecord = so->so_rcv.sb_mb;
2132 orig_resid = 0;
2133 }
2134 if (m != NULL) {
2135 if ((flags & MSG_PEEK) == 0) {
2136 KASSERT(m->m_nextpkt == nextrecord,
2137 ("soreceive: post-control, nextrecord !sync"));
2138 if (nextrecord == NULL) {
2139 KASSERT(so->so_rcv.sb_mb == m,
2140 ("soreceive: post-control, sb_mb!=m"));
2141 KASSERT(so->so_rcv.sb_lastrecord == m,
2142 ("soreceive: post-control, lastrecord!=m"));
2143 }
2144 }
2145 type = m->m_type;
2146 if (type == MT_OOBDATA)
2147 flags |= MSG_OOB;
2148 } else {
2149 if ((flags & MSG_PEEK) == 0) {
2150 KASSERT(so->so_rcv.sb_mb == nextrecord,
2151 ("soreceive: sb_mb != nextrecord"));
2152 if (so->so_rcv.sb_mb == NULL) {
2153 KASSERT(so->so_rcv.sb_lastrecord == NULL,
2154 ("soreceive: sb_lastercord != NULL"));
2155 }
2156 }
2157 }
2158 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2159 SBLASTRECORDCHK(&so->so_rcv);
2160 SBLASTMBUFCHK(&so->so_rcv);
2161
2162 /*
2163 * Now continue to read any data mbufs off of the head of the socket
2164 * buffer until the read request is satisfied. Note that 'type' is
2165 * used to store the type of any mbuf reads that have happened so far
2166 * such that soreceive() can stop reading if the type changes, which
2167 * causes soreceive() to return only one of regular data and inline
2168 * out-of-band data in a single socket receive operation.
2169 */
2170 moff = 0;
2171 offset = 0;
2172 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2173 && error == 0) {
2174 /*
2175 * If the type of mbuf has changed since the last mbuf
2176 * examined ('type'), end the receive operation.
2177 */
2178 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2179 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2180 if (type != m->m_type)
2181 break;
2182 } else if (type == MT_OOBDATA)
2183 break;
2184 else
2185 KASSERT(m->m_type == MT_DATA,
2186 ("m->m_type == %d", m->m_type));
2187 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2188 len = uio->uio_resid;
2189 if (so->so_oobmark && len > so->so_oobmark - offset)
2190 len = so->so_oobmark - offset;
2191 if (len > m->m_len - moff)
2192 len = m->m_len - moff;
2193 /*
2194 * If mp is set, just pass back the mbufs. Otherwise copy
2195 * them out via the uio, then free. Sockbuf must be
2196 * consistent here (points to current mbuf, it points to next
2197 * record) when we drop priority; we must note any additions
2198 * to the sockbuf when we block interrupts again.
2199 */
2200 if (mp == NULL) {
2201 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2202 SBLASTRECORDCHK(&so->so_rcv);
2203 SBLASTMBUFCHK(&so->so_rcv);
2204 SOCKBUF_UNLOCK(&so->so_rcv);
2205 if ((m->m_flags & M_EXTPG) != 0)
2206 error = m_unmapped_uiomove(m, moff, uio,
2207 (int)len);
2208 else
2209 error = uiomove(mtod(m, char *) + moff,
2210 (int)len, uio);
2211 SOCKBUF_LOCK(&so->so_rcv);
2212 if (error) {
2213 /*
2214 * The MT_SONAME mbuf has already been removed
2215 * from the record, so it is necessary to
2216 * remove the data mbufs, if any, to preserve
2217 * the invariant in the case of PR_ADDR that
2218 * requires MT_SONAME mbufs at the head of
2219 * each record.
2220 */
2221 if (pr->pr_flags & PR_ATOMIC &&
2222 ((flags & MSG_PEEK) == 0))
2223 (void)sbdroprecord_locked(&so->so_rcv);
2224 SOCKBUF_UNLOCK(&so->so_rcv);
2225 goto release;
2226 }
2227 } else
2228 uio->uio_resid -= len;
2229 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2230 if (len == m->m_len - moff) {
2231 if (m->m_flags & M_EOR)
2232 flags |= MSG_EOR;
2233 if (flags & MSG_PEEK) {
2234 m = m->m_next;
2235 moff = 0;
2236 } else {
2237 nextrecord = m->m_nextpkt;
2238 sbfree(&so->so_rcv, m);
2239 if (mp != NULL) {
2240 m->m_nextpkt = NULL;
2241 *mp = m;
2242 mp = &m->m_next;
2243 so->so_rcv.sb_mb = m = m->m_next;
2244 *mp = NULL;
2245 } else {
2246 so->so_rcv.sb_mb = m_free(m);
2247 m = so->so_rcv.sb_mb;
2248 }
2249 sockbuf_pushsync(&so->so_rcv, nextrecord);
2250 SBLASTRECORDCHK(&so->so_rcv);
2251 SBLASTMBUFCHK(&so->so_rcv);
2252 }
2253 } else {
2254 if (flags & MSG_PEEK)
2255 moff += len;
2256 else {
2257 if (mp != NULL) {
2258 if (flags & MSG_DONTWAIT) {
2259 *mp = m_copym(m, 0, len,
2260 M_NOWAIT);
2261 if (*mp == NULL) {
2262 /*
2263 * m_copym() couldn't
2264 * allocate an mbuf.
2265 * Adjust uio_resid back
2266 * (it was adjusted
2267 * down by len bytes,
2268 * which we didn't end
2269 * up "copying" over).
2270 */
2271 uio->uio_resid += len;
2272 break;
2273 }
2274 } else {
2275 SOCKBUF_UNLOCK(&so->so_rcv);
2276 *mp = m_copym(m, 0, len,
2277 M_WAITOK);
2278 SOCKBUF_LOCK(&so->so_rcv);
2279 }
2280 }
2281 sbcut_locked(&so->so_rcv, len);
2282 }
2283 }
2284 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2285 if (so->so_oobmark) {
2286 if ((flags & MSG_PEEK) == 0) {
2287 so->so_oobmark -= len;
2288 if (so->so_oobmark == 0) {
2289 so->so_rcv.sb_state |= SBS_RCVATMARK;
2290 break;
2291 }
2292 } else {
2293 offset += len;
2294 if (offset == so->so_oobmark)
2295 break;
2296 }
2297 }
2298 if (flags & MSG_EOR)
2299 break;
2300 /*
2301 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2302 * must not quit until "uio->uio_resid == 0" or an error
2303 * termination. If a signal/timeout occurs, return with a
2304 * short count but without error. Keep sockbuf locked
2305 * against other readers.
2306 */
2307 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2308 !sosendallatonce(so) && nextrecord == NULL) {
2309 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2310 if (so->so_error || so->so_rerror ||
2311 so->so_rcv.sb_state & SBS_CANTRCVMORE)
2312 break;
2313 /*
2314 * Notify the protocol that some data has been
2315 * drained before blocking.
2316 */
2317 if (pr->pr_flags & PR_WANTRCVD) {
2318 SOCKBUF_UNLOCK(&so->so_rcv);
2319 VNET_SO_ASSERT(so);
2320 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2321 SOCKBUF_LOCK(&so->so_rcv);
2322 }
2323 SBLASTRECORDCHK(&so->so_rcv);
2324 SBLASTMBUFCHK(&so->so_rcv);
2325 /*
2326 * We could receive some data while was notifying
2327 * the protocol. Skip blocking in this case.
2328 */
2329 if (so->so_rcv.sb_mb == NULL) {
2330 error = sbwait(&so->so_rcv);
2331 if (error) {
2332 SOCKBUF_UNLOCK(&so->so_rcv);
2333 goto release;
2334 }
2335 }
2336 m = so->so_rcv.sb_mb;
2337 if (m != NULL)
2338 nextrecord = m->m_nextpkt;
2339 }
2340 }
2341
2342 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2343 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2344 flags |= MSG_TRUNC;
2345 if ((flags & MSG_PEEK) == 0)
2346 (void) sbdroprecord_locked(&so->so_rcv);
2347 }
2348 if ((flags & MSG_PEEK) == 0) {
2349 if (m == NULL) {
2350 /*
2351 * First part is an inline SB_EMPTY_FIXUP(). Second
2352 * part makes sure sb_lastrecord is up-to-date if
2353 * there is still data in the socket buffer.
2354 */
2355 so->so_rcv.sb_mb = nextrecord;
2356 if (so->so_rcv.sb_mb == NULL) {
2357 so->so_rcv.sb_mbtail = NULL;
2358 so->so_rcv.sb_lastrecord = NULL;
2359 } else if (nextrecord->m_nextpkt == NULL)
2360 so->so_rcv.sb_lastrecord = nextrecord;
2361 }
2362 SBLASTRECORDCHK(&so->so_rcv);
2363 SBLASTMBUFCHK(&so->so_rcv);
2364 /*
2365 * If soreceive() is being done from the socket callback,
2366 * then don't need to generate ACK to peer to update window,
2367 * since ACK will be generated on return to TCP.
2368 */
2369 if (!(flags & MSG_SOCALLBCK) &&
2370 (pr->pr_flags & PR_WANTRCVD)) {
2371 SOCKBUF_UNLOCK(&so->so_rcv);
2372 VNET_SO_ASSERT(so);
2373 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
2374 SOCKBUF_LOCK(&so->so_rcv);
2375 }
2376 }
2377 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2378 if (orig_resid == uio->uio_resid && orig_resid &&
2379 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2380 SOCKBUF_UNLOCK(&so->so_rcv);
2381 goto restart;
2382 }
2383 SOCKBUF_UNLOCK(&so->so_rcv);
2384
2385 if (flagsp != NULL)
2386 *flagsp |= flags;
2387 release:
2388 SOCK_IO_RECV_UNLOCK(so);
2389 return (error);
2390 }
2391
2392 /*
2393 * Optimized version of soreceive() for stream (TCP) sockets.
2394 */
2395 int
2396 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2397 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2398 {
2399 int len = 0, error = 0, flags, oresid;
2400 struct sockbuf *sb;
2401 struct mbuf *m, *n = NULL;
2402
2403 /* We only do stream sockets. */
2404 if (so->so_type != SOCK_STREAM)
2405 return (EINVAL);
2406 if (psa != NULL)
2407 *psa = NULL;
2408 if (flagsp != NULL)
2409 flags = *flagsp &~ MSG_EOR;
2410 else
2411 flags = 0;
2412 if (controlp != NULL)
2413 *controlp = NULL;
2414 if (flags & MSG_OOB)
2415 return (soreceive_rcvoob(so, uio, flags));
2416 if (mp0 != NULL)
2417 *mp0 = NULL;
2418
2419 sb = &so->so_rcv;
2420
2421 #ifdef KERN_TLS
2422 /*
2423 * KTLS store TLS records as records with a control message to
2424 * describe the framing.
2425 *
2426 * We check once here before acquiring locks to optimize the
2427 * common case.
2428 */
2429 if (sb->sb_tls_info != NULL)
2430 return (soreceive_generic(so, psa, uio, mp0, controlp,
2431 flagsp));
2432 #endif
2433
2434 /* Prevent other readers from entering the socket. */
2435 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2436 if (error)
2437 return (error);
2438 SOCKBUF_LOCK(sb);
2439
2440 #ifdef KERN_TLS
2441 if (sb->sb_tls_info != NULL) {
2442 SOCKBUF_UNLOCK(sb);
2443 SOCK_IO_RECV_UNLOCK(so);
2444 return (soreceive_generic(so, psa, uio, mp0, controlp,
2445 flagsp));
2446 }
2447 #endif
2448
2449 /* Easy one, no space to copyout anything. */
2450 if (uio->uio_resid == 0) {
2451 error = EINVAL;
2452 goto out;
2453 }
2454 oresid = uio->uio_resid;
2455
2456 /* We will never ever get anything unless we are or were connected. */
2457 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2458 error = ENOTCONN;
2459 goto out;
2460 }
2461
2462 restart:
2463 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2464
2465 /* Abort if socket has reported problems. */
2466 if (so->so_error) {
2467 if (sbavail(sb) > 0)
2468 goto deliver;
2469 if (oresid > uio->uio_resid)
2470 goto out;
2471 error = so->so_error;
2472 if (!(flags & MSG_PEEK))
2473 so->so_error = 0;
2474 goto out;
2475 }
2476
2477 /* Door is closed. Deliver what is left, if any. */
2478 if (sb->sb_state & SBS_CANTRCVMORE) {
2479 if (sbavail(sb) > 0)
2480 goto deliver;
2481 else
2482 goto out;
2483 }
2484
2485 /* Socket buffer is empty and we shall not block. */
2486 if (sbavail(sb) == 0 &&
2487 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2488 error = EAGAIN;
2489 goto out;
2490 }
2491
2492 /* Socket buffer got some data that we shall deliver now. */
2493 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2494 ((so->so_state & SS_NBIO) ||
2495 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2496 sbavail(sb) >= sb->sb_lowat ||
2497 sbavail(sb) >= uio->uio_resid ||
2498 sbavail(sb) >= sb->sb_hiwat) ) {
2499 goto deliver;
2500 }
2501
2502 /* On MSG_WAITALL we must wait until all data or error arrives. */
2503 if ((flags & MSG_WAITALL) &&
2504 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2505 goto deliver;
2506
2507 /*
2508 * Wait and block until (more) data comes in.
2509 * NB: Drops the sockbuf lock during wait.
2510 */
2511 error = sbwait(sb);
2512 if (error)
2513 goto out;
2514 goto restart;
2515
2516 deliver:
2517 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2518 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2519 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2520
2521 /* Statistics. */
2522 if (uio->uio_td)
2523 uio->uio_td->td_ru.ru_msgrcv++;
2524
2525 /* Fill uio until full or current end of socket buffer is reached. */
2526 len = min(uio->uio_resid, sbavail(sb));
2527 if (mp0 != NULL) {
2528 /* Dequeue as many mbufs as possible. */
2529 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2530 if (*mp0 == NULL)
2531 *mp0 = sb->sb_mb;
2532 else
2533 m_cat(*mp0, sb->sb_mb);
2534 for (m = sb->sb_mb;
2535 m != NULL && m->m_len <= len;
2536 m = m->m_next) {
2537 KASSERT(!(m->m_flags & M_NOTAVAIL),
2538 ("%s: m %p not available", __func__, m));
2539 len -= m->m_len;
2540 uio->uio_resid -= m->m_len;
2541 sbfree(sb, m);
2542 n = m;
2543 }
2544 n->m_next = NULL;
2545 sb->sb_mb = m;
2546 sb->sb_lastrecord = sb->sb_mb;
2547 if (sb->sb_mb == NULL)
2548 SB_EMPTY_FIXUP(sb);
2549 }
2550 /* Copy the remainder. */
2551 if (len > 0) {
2552 KASSERT(sb->sb_mb != NULL,
2553 ("%s: len > 0 && sb->sb_mb empty", __func__));
2554
2555 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2556 if (m == NULL)
2557 len = 0; /* Don't flush data from sockbuf. */
2558 else
2559 uio->uio_resid -= len;
2560 if (*mp0 != NULL)
2561 m_cat(*mp0, m);
2562 else
2563 *mp0 = m;
2564 if (*mp0 == NULL) {
2565 error = ENOBUFS;
2566 goto out;
2567 }
2568 }
2569 } else {
2570 /* NB: Must unlock socket buffer as uiomove may sleep. */
2571 SOCKBUF_UNLOCK(sb);
2572 error = m_mbuftouio(uio, sb->sb_mb, len);
2573 SOCKBUF_LOCK(sb);
2574 if (error)
2575 goto out;
2576 }
2577 SBLASTRECORDCHK(sb);
2578 SBLASTMBUFCHK(sb);
2579
2580 /*
2581 * Remove the delivered data from the socket buffer unless we
2582 * were only peeking.
2583 */
2584 if (!(flags & MSG_PEEK)) {
2585 if (len > 0)
2586 sbdrop_locked(sb, len);
2587
2588 /* Notify protocol that we drained some data. */
2589 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2590 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2591 !(flags & MSG_SOCALLBCK))) {
2592 SOCKBUF_UNLOCK(sb);
2593 VNET_SO_ASSERT(so);
2594 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2595 SOCKBUF_LOCK(sb);
2596 }
2597 }
2598
2599 /*
2600 * For MSG_WAITALL we may have to loop again and wait for
2601 * more data to come in.
2602 */
2603 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2604 goto restart;
2605 out:
2606 SBLASTRECORDCHK(sb);
2607 SBLASTMBUFCHK(sb);
2608 SOCKBUF_UNLOCK(sb);
2609 SOCK_IO_RECV_UNLOCK(so);
2610 return (error);
2611 }
2612
2613 /*
2614 * Optimized version of soreceive() for simple datagram cases from userspace.
2615 * Unlike in the stream case, we're able to drop a datagram if copyout()
2616 * fails, and because we handle datagrams atomically, we don't need to use a
2617 * sleep lock to prevent I/O interlacing.
2618 */
2619 int
2620 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2621 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2622 {
2623 struct mbuf *m, *m2;
2624 int flags, error;
2625 ssize_t len;
2626 struct protosw *pr = so->so_proto;
2627 struct mbuf *nextrecord;
2628
2629 if (psa != NULL)
2630 *psa = NULL;
2631 if (controlp != NULL)
2632 *controlp = NULL;
2633 if (flagsp != NULL)
2634 flags = *flagsp &~ MSG_EOR;
2635 else
2636 flags = 0;
2637
2638 /*
2639 * For any complicated cases, fall back to the full
2640 * soreceive_generic().
2641 */
2642 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2643 return (soreceive_generic(so, psa, uio, mp0, controlp,
2644 flagsp));
2645
2646 /*
2647 * Enforce restrictions on use.
2648 */
2649 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2650 ("soreceive_dgram: wantrcvd"));
2651 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2652 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2653 ("soreceive_dgram: SBS_RCVATMARK"));
2654 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2655 ("soreceive_dgram: P_CONNREQUIRED"));
2656
2657 /*
2658 * Loop blocking while waiting for a datagram.
2659 */
2660 SOCKBUF_LOCK(&so->so_rcv);
2661 while ((m = so->so_rcv.sb_mb) == NULL) {
2662 KASSERT(sbavail(&so->so_rcv) == 0,
2663 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2664 sbavail(&so->so_rcv)));
2665 if (so->so_error) {
2666 error = so->so_error;
2667 so->so_error = 0;
2668 SOCKBUF_UNLOCK(&so->so_rcv);
2669 return (error);
2670 }
2671 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2672 uio->uio_resid == 0) {
2673 SOCKBUF_UNLOCK(&so->so_rcv);
2674 return (0);
2675 }
2676 if ((so->so_state & SS_NBIO) ||
2677 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2678 SOCKBUF_UNLOCK(&so->so_rcv);
2679 return (EWOULDBLOCK);
2680 }
2681 SBLASTRECORDCHK(&so->so_rcv);
2682 SBLASTMBUFCHK(&so->so_rcv);
2683 error = sbwait(&so->so_rcv);
2684 if (error) {
2685 SOCKBUF_UNLOCK(&so->so_rcv);
2686 return (error);
2687 }
2688 }
2689 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2690
2691 if (uio->uio_td)
2692 uio->uio_td->td_ru.ru_msgrcv++;
2693 SBLASTRECORDCHK(&so->so_rcv);
2694 SBLASTMBUFCHK(&so->so_rcv);
2695 nextrecord = m->m_nextpkt;
2696 if (nextrecord == NULL) {
2697 KASSERT(so->so_rcv.sb_lastrecord == m,
2698 ("soreceive_dgram: lastrecord != m"));
2699 }
2700
2701 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2702 ("soreceive_dgram: m_nextpkt != nextrecord"));
2703
2704 /*
2705 * Pull 'm' and its chain off the front of the packet queue.
2706 */
2707 so->so_rcv.sb_mb = NULL;
2708 sockbuf_pushsync(&so->so_rcv, nextrecord);
2709
2710 /*
2711 * Walk 'm's chain and free that many bytes from the socket buffer.
2712 */
2713 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2714 sbfree(&so->so_rcv, m2);
2715
2716 /*
2717 * Do a few last checks before we let go of the lock.
2718 */
2719 SBLASTRECORDCHK(&so->so_rcv);
2720 SBLASTMBUFCHK(&so->so_rcv);
2721 SOCKBUF_UNLOCK(&so->so_rcv);
2722
2723 if (pr->pr_flags & PR_ADDR) {
2724 KASSERT(m->m_type == MT_SONAME,
2725 ("m->m_type == %d", m->m_type));
2726 if (psa != NULL)
2727 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2728 M_NOWAIT);
2729 m = m_free(m);
2730 }
2731 if (m == NULL) {
2732 /* XXXRW: Can this happen? */
2733 return (0);
2734 }
2735
2736 /*
2737 * Packet to copyout() is now in 'm' and it is disconnected from the
2738 * queue.
2739 *
2740 * Process one or more MT_CONTROL mbufs present before any data mbufs
2741 * in the first mbuf chain on the socket buffer. We call into the
2742 * protocol to perform externalization (or freeing if controlp ==
2743 * NULL). In some cases there can be only MT_CONTROL mbufs without
2744 * MT_DATA mbufs.
2745 */
2746 if (m->m_type == MT_CONTROL) {
2747 struct mbuf *cm = NULL, *cmn;
2748 struct mbuf **cme = &cm;
2749
2750 do {
2751 m2 = m->m_next;
2752 m->m_next = NULL;
2753 *cme = m;
2754 cme = &(*cme)->m_next;
2755 m = m2;
2756 } while (m != NULL && m->m_type == MT_CONTROL);
2757 while (cm != NULL) {
2758 cmn = cm->m_next;
2759 cm->m_next = NULL;
2760 if (pr->pr_domain->dom_externalize != NULL) {
2761 error = (*pr->pr_domain->dom_externalize)
2762 (cm, controlp, flags);
2763 } else if (controlp != NULL)
2764 *controlp = cm;
2765 else
2766 m_freem(cm);
2767 if (controlp != NULL) {
2768 while (*controlp != NULL)
2769 controlp = &(*controlp)->m_next;
2770 }
2771 cm = cmn;
2772 }
2773 }
2774 KASSERT(m == NULL || m->m_type == MT_DATA,
2775 ("soreceive_dgram: !data"));
2776 while (m != NULL && uio->uio_resid > 0) {
2777 len = uio->uio_resid;
2778 if (len > m->m_len)
2779 len = m->m_len;
2780 error = uiomove(mtod(m, char *), (int)len, uio);
2781 if (error) {
2782 m_freem(m);
2783 return (error);
2784 }
2785 if (len == m->m_len)
2786 m = m_free(m);
2787 else {
2788 m->m_data += len;
2789 m->m_len -= len;
2790 }
2791 }
2792 if (m != NULL) {
2793 flags |= MSG_TRUNC;
2794 m_freem(m);
2795 }
2796 if (flagsp != NULL)
2797 *flagsp |= flags;
2798 return (0);
2799 }
2800
2801 int
2802 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2803 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2804 {
2805 int error;
2806
2807 CURVNET_SET(so->so_vnet);
2808 if (!SOLISTENING(so))
2809 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2810 mp0, controlp, flagsp));
2811 else
2812 error = ENOTCONN;
2813 CURVNET_RESTORE();
2814 return (error);
2815 }
2816
2817 int
2818 soshutdown(struct socket *so, int how)
2819 {
2820 struct protosw *pr = so->so_proto;
2821 int error, soerror_enotconn;
2822
2823 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2824 return (EINVAL);
2825
2826 soerror_enotconn = 0;
2827 if ((so->so_state &
2828 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2829 /*
2830 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2831 * invoked on a datagram sockets, however historically we would
2832 * actually tear socket down. This is known to be leveraged by
2833 * some applications to unblock process waiting in recvXXX(2)
2834 * by other process that it shares that socket with. Try to meet
2835 * both backward-compatibility and POSIX requirements by forcing
2836 * ENOTCONN but still asking protocol to perform pru_shutdown().
2837 */
2838 if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2839 return (ENOTCONN);
2840 soerror_enotconn = 1;
2841 }
2842
2843 if (SOLISTENING(so)) {
2844 if (how != SHUT_WR) {
2845 SOLISTEN_LOCK(so);
2846 so->so_error = ECONNABORTED;
2847 solisten_wakeup(so); /* unlocks so */
2848 }
2849 goto done;
2850 }
2851
2852 CURVNET_SET(so->so_vnet);
2853 if (pr->pr_usrreqs->pru_flush != NULL)
2854 (*pr->pr_usrreqs->pru_flush)(so, how);
2855 if (how != SHUT_WR)
2856 sorflush(so);
2857 if (how != SHUT_RD) {
2858 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2859 wakeup(&so->so_timeo);
2860 CURVNET_RESTORE();
2861 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2862 }
2863 wakeup(&so->so_timeo);
2864 CURVNET_RESTORE();
2865
2866 done:
2867 return (soerror_enotconn ? ENOTCONN : 0);
2868 }
2869
2870 void
2871 sorflush(struct socket *so)
2872 {
2873 struct sockbuf *sb = &so->so_rcv;
2874 struct protosw *pr = so->so_proto;
2875 struct socket aso;
2876 int error;
2877
2878 VNET_SO_ASSERT(so);
2879
2880 /*
2881 * In order to avoid calling dom_dispose with the socket buffer mutex
2882 * held, and in order to generally avoid holding the lock for a long
2883 * time, we make a copy of the socket buffer and clear the original
2884 * (except locks, state). The new socket buffer copy won't have
2885 * initialized locks so we can only call routines that won't use or
2886 * assert those locks.
2887 *
2888 * Dislodge threads currently blocked in receive and wait to acquire
2889 * a lock against other simultaneous readers before clearing the
2890 * socket buffer. Don't let our acquire be interrupted by a signal
2891 * despite any existing socket disposition on interruptable waiting.
2892 */
2893 socantrcvmore(so);
2894 error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2895 KASSERT(error == 0, ("%s: cannot lock sock %p recv buffer",
2896 __func__, so));
2897
2898 /*
2899 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2900 * and mutex data unchanged.
2901 */
2902 SOCKBUF_LOCK(sb);
2903 bzero(&aso, sizeof(aso));
2904 aso.so_pcb = so->so_pcb;
2905 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2906 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2907 bzero(&sb->sb_startzero,
2908 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2909 SOCKBUF_UNLOCK(sb);
2910 SOCK_IO_RECV_UNLOCK(so);
2911
2912 /*
2913 * Dispose of special rights and flush the copied socket. Don't call
2914 * any unsafe routines (that rely on locks being initialized) on aso.
2915 */
2916 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2917 (*pr->pr_domain->dom_dispose)(&aso);
2918 sbrelease_internal(&aso.so_rcv, so);
2919 }
2920
2921 /*
2922 * Wrapper for Socket established helper hook.
2923 * Parameters: socket, context of the hook point, hook id.
2924 */
2925 static int inline
2926 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2927 {
2928 struct socket_hhook_data hhook_data = {
2929 .so = so,
2930 .hctx = hctx,
2931 .m = NULL,
2932 .status = 0
2933 };
2934
2935 CURVNET_SET(so->so_vnet);
2936 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2937 CURVNET_RESTORE();
2938
2939 /* Ugly but needed, since hhooks return void for now */
2940 return (hhook_data.status);
2941 }
2942
2943 /*
2944 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2945 * additional variant to handle the case where the option value needs to be
2946 * some kind of integer, but not a specific size. In addition to their use
2947 * here, these functions are also called by the protocol-level pr_ctloutput()
2948 * routines.
2949 */
2950 int
2951 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2952 {
2953 size_t valsize;
2954
2955 /*
2956 * If the user gives us more than we wanted, we ignore it, but if we
2957 * don't get the minimum length the caller wants, we return EINVAL.
2958 * On success, sopt->sopt_valsize is set to however much we actually
2959 * retrieved.
2960 */
2961 if ((valsize = sopt->sopt_valsize) < minlen)
2962 return EINVAL;
2963 if (valsize > len)
2964 sopt->sopt_valsize = valsize = len;
2965
2966 if (sopt->sopt_td != NULL)
2967 return (copyin(sopt->sopt_val, buf, valsize));
2968
2969 bcopy(sopt->sopt_val, buf, valsize);
2970 return (0);
2971 }
2972
2973 /*
2974 * Kernel version of setsockopt(2).
2975 *
2976 * XXX: optlen is size_t, not socklen_t
2977 */
2978 int
2979 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2980 size_t optlen)
2981 {
2982 struct sockopt sopt;
2983
2984 sopt.sopt_level = level;
2985 sopt.sopt_name = optname;
2986 sopt.sopt_dir = SOPT_SET;
2987 sopt.sopt_val = optval;
2988 sopt.sopt_valsize = optlen;
2989 sopt.sopt_td = NULL;
2990 return (sosetopt(so, &sopt));
2991 }
2992
2993 int
2994 sosetopt(struct socket *so, struct sockopt *sopt)
2995 {
2996 int error, optval;
2997 struct linger l;
2998 struct timeval tv;
2999 sbintime_t val;
3000 uint32_t val32;
3001 #ifdef MAC
3002 struct mac extmac;
3003 #endif
3004
3005 CURVNET_SET(so->so_vnet);
3006 error = 0;
3007 if (sopt->sopt_level != SOL_SOCKET) {
3008 if (so->so_proto->pr_ctloutput != NULL)
3009 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3010 else
3011 error = ENOPROTOOPT;
3012 } else {
3013 switch (sopt->sopt_name) {
3014 case SO_ACCEPTFILTER:
3015 error = accept_filt_setopt(so, sopt);
3016 if (error)
3017 goto bad;
3018 break;
3019
3020 case SO_LINGER:
3021 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3022 if (error)
3023 goto bad;
3024 if (l.l_linger < 0 ||
3025 l.l_linger > USHRT_MAX ||
3026 l.l_linger > (INT_MAX / hz)) {
3027 error = EDOM;
3028 goto bad;
3029 }
3030 SOCK_LOCK(so);
3031 so->so_linger = l.l_linger;
3032 if (l.l_onoff)
3033 so->so_options |= SO_LINGER;
3034 else
3035 so->so_options &= ~SO_LINGER;
3036 SOCK_UNLOCK(so);
3037 break;
3038
3039 case SO_DEBUG:
3040 case SO_KEEPALIVE:
3041 case SO_DONTROUTE:
3042 case SO_USELOOPBACK:
3043 case SO_BROADCAST:
3044 case SO_REUSEADDR:
3045 case SO_REUSEPORT:
3046 case SO_REUSEPORT_LB:
3047 case SO_OOBINLINE:
3048 case SO_TIMESTAMP:
3049 case SO_BINTIME:
3050 case SO_NOSIGPIPE:
3051 case SO_NO_DDP:
3052 case SO_NO_OFFLOAD:
3053 case SO_RERROR:
3054 error = sooptcopyin(sopt, &optval, sizeof optval,
3055 sizeof optval);
3056 if (error)
3057 goto bad;
3058 SOCK_LOCK(so);
3059 if (optval)
3060 so->so_options |= sopt->sopt_name;
3061 else
3062 so->so_options &= ~sopt->sopt_name;
3063 SOCK_UNLOCK(so);
3064 break;
3065
3066 case SO_SETFIB:
3067 error = sooptcopyin(sopt, &optval, sizeof optval,
3068 sizeof optval);
3069 if (error)
3070 goto bad;
3071
3072 if (optval < 0 || optval >= rt_numfibs) {
3073 error = EINVAL;
3074 goto bad;
3075 }
3076 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3077 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3078 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3079 so->so_fibnum = optval;
3080 else
3081 so->so_fibnum = 0;
3082 break;
3083
3084 case SO_USER_COOKIE:
3085 error = sooptcopyin(sopt, &val32, sizeof val32,
3086 sizeof val32);
3087 if (error)
3088 goto bad;
3089 so->so_user_cookie = val32;
3090 break;
3091
3092 case SO_SNDBUF:
3093 case SO_RCVBUF:
3094 case SO_SNDLOWAT:
3095 case SO_RCVLOWAT:
3096 error = sooptcopyin(sopt, &optval, sizeof optval,
3097 sizeof optval);
3098 if (error)
3099 goto bad;
3100
3101 /*
3102 * Values < 1 make no sense for any of these options,
3103 * so disallow them.
3104 */
3105 if (optval < 1) {
3106 error = EINVAL;
3107 goto bad;
3108 }
3109
3110 error = sbsetopt(so, sopt->sopt_name, optval);
3111 break;
3112
3113 case SO_SNDTIMEO:
3114 case SO_RCVTIMEO:
3115 #ifdef COMPAT_FREEBSD32
3116 if (SV_CURPROC_FLAG(SV_ILP32)) {
3117 struct timeval32 tv32;
3118
3119 error = sooptcopyin(sopt, &tv32, sizeof tv32,
3120 sizeof tv32);
3121 CP(tv32, tv, tv_sec);
3122 CP(tv32, tv, tv_usec);
3123 } else
3124 #endif
3125 error = sooptcopyin(sopt, &tv, sizeof tv,
3126 sizeof tv);
3127 if (error)
3128 goto bad;
3129 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3130 tv.tv_usec >= 1000000) {
3131 error = EDOM;
3132 goto bad;
3133 }
3134 if (tv.tv_sec > INT32_MAX)
3135 val = SBT_MAX;
3136 else
3137 val = tvtosbt(tv);
3138 switch (sopt->sopt_name) {
3139 case SO_SNDTIMEO:
3140 so->so_snd.sb_timeo = val;
3141 break;
3142 case SO_RCVTIMEO:
3143 so->so_rcv.sb_timeo = val;
3144 break;
3145 }
3146 break;
3147
3148 case SO_LABEL:
3149 #ifdef MAC
3150 error = sooptcopyin(sopt, &extmac, sizeof extmac,
3151 sizeof extmac);
3152 if (error)
3153 goto bad;
3154 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3155 so, &extmac);
3156 #else
3157 error = EOPNOTSUPP;
3158 #endif
3159 break;
3160
3161 case SO_TS_CLOCK:
3162 error = sooptcopyin(sopt, &optval, sizeof optval,
3163 sizeof optval);
3164 if (error)
3165 goto bad;
3166 if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3167 error = EINVAL;
3168 goto bad;
3169 }
3170 so->so_ts_clock = optval;
3171 break;
3172
3173 case SO_MAX_PACING_RATE:
3174 error = sooptcopyin(sopt, &val32, sizeof(val32),
3175 sizeof(val32));
3176 if (error)
3177 goto bad;
3178 so->so_max_pacing_rate = val32;
3179 break;
3180
3181 default:
3182 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3183 error = hhook_run_socket(so, sopt,
3184 HHOOK_SOCKET_OPT);
3185 else
3186 error = ENOPROTOOPT;
3187 break;
3188 }
3189 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3190 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
3191 }
3192 bad:
3193 CURVNET_RESTORE();
3194 return (error);
3195 }
3196
3197 /*
3198 * Helper routine for getsockopt.
3199 */
3200 int
3201 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3202 {
3203 int error;
3204 size_t valsize;
3205
3206 error = 0;
3207
3208 /*
3209 * Documented get behavior is that we always return a value, possibly
3210 * truncated to fit in the user's buffer. Traditional behavior is
3211 * that we always tell the user precisely how much we copied, rather
3212 * than something useful like the total amount we had available for
3213 * her. Note that this interface is not idempotent; the entire
3214 * answer must be generated ahead of time.
3215 */
3216 valsize = min(len, sopt->sopt_valsize);
3217 sopt->sopt_valsize = valsize;
3218 if (sopt->sopt_val != NULL) {
3219 if (sopt->sopt_td != NULL)
3220 error = copyout(buf, sopt->sopt_val, valsize);
3221 else
3222 bcopy(buf, sopt->sopt_val, valsize);
3223 }
3224 return (error);
3225 }
3226
3227 int
3228 sogetopt(struct socket *so, struct sockopt *sopt)
3229 {
3230 int error, optval;
3231 struct linger l;
3232 struct timeval tv;
3233 #ifdef MAC
3234 struct mac extmac;
3235 #endif
3236
3237 CURVNET_SET(so->so_vnet);
3238 error = 0;
3239 if (sopt->sopt_level != SOL_SOCKET) {
3240 if (so->so_proto->pr_ctloutput != NULL)
3241 error = (*so->so_proto->pr_ctloutput)(so, sopt);
3242 else
3243 error = ENOPROTOOPT;
3244 CURVNET_RESTORE();
3245 return (error);
3246 } else {
3247 switch (sopt->sopt_name) {
3248 case SO_ACCEPTFILTER:
3249 error = accept_filt_getopt(so, sopt);
3250 break;
3251
3252 case SO_LINGER:
3253 SOCK_LOCK(so);
3254 l.l_onoff = so->so_options & SO_LINGER;
3255 l.l_linger = so->so_linger;
3256 SOCK_UNLOCK(so);
3257 error = sooptcopyout(sopt, &l, sizeof l);
3258 break;
3259
3260 case SO_USELOOPBACK:
3261 case SO_DONTROUTE:
3262 case SO_DEBUG:
3263 case SO_KEEPALIVE:
3264 case SO_REUSEADDR:
3265 case SO_REUSEPORT:
3266 case SO_REUSEPORT_LB:
3267 case SO_BROADCAST:
3268 case SO_OOBINLINE:
3269 case SO_ACCEPTCONN:
3270 case SO_TIMESTAMP:
3271 case SO_BINTIME:
3272 case SO_NOSIGPIPE:
3273 case SO_NO_DDP:
3274 case SO_NO_OFFLOAD:
3275 case SO_RERROR:
3276 optval = so->so_options & sopt->sopt_name;
3277 integer:
3278 error = sooptcopyout(sopt, &optval, sizeof optval);
3279 break;
3280
3281 case SO_DOMAIN:
3282 optval = so->so_proto->pr_domain->dom_family;
3283 goto integer;
3284
3285 case SO_TYPE:
3286 optval = so->so_type;
3287 goto integer;
3288
3289 case SO_PROTOCOL:
3290 optval = so->so_proto->pr_protocol;
3291 goto integer;
3292
3293 case SO_ERROR:
3294 SOCK_LOCK(so);
3295 if (so->so_error) {
3296 optval = so->so_error;
3297 so->so_error = 0;
3298 } else {
3299 optval = so->so_rerror;
3300 so->so_rerror = 0;
3301 }
3302 SOCK_UNLOCK(so);
3303 goto integer;
3304
3305 case SO_SNDBUF:
3306 optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3307 so->so_snd.sb_hiwat;
3308 goto integer;
3309
3310 case SO_RCVBUF:
3311 optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3312 so->so_rcv.sb_hiwat;
3313 goto integer;
3314
3315 case SO_SNDLOWAT:
3316 optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3317 so->so_snd.sb_lowat;
3318 goto integer;
3319
3320 case SO_RCVLOWAT:
3321 optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3322 so->so_rcv.sb_lowat;
3323 goto integer;
3324
3325 case SO_SNDTIMEO:
3326 case SO_RCVTIMEO:
3327 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3328 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3329 #ifdef COMPAT_FREEBSD32
3330 if (SV_CURPROC_FLAG(SV_ILP32)) {
3331 struct timeval32 tv32;
3332
3333 CP(tv, tv32, tv_sec);
3334 CP(tv, tv32, tv_usec);
3335 error = sooptcopyout(sopt, &tv32, sizeof tv32);
3336 } else
3337 #endif
3338 error = sooptcopyout(sopt, &tv, sizeof tv);
3339 break;
3340
3341 case SO_LABEL:
3342 #ifdef MAC
3343 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3344 sizeof(extmac));
3345 if (error)
3346 goto bad;
3347 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3348 so, &extmac);
3349 if (error)
3350 goto bad;
3351 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3352 #else
3353 error = EOPNOTSUPP;
3354 #endif
3355 break;
3356
3357 case SO_PEERLABEL:
3358 #ifdef MAC
3359 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3360 sizeof(extmac));
3361 if (error)
3362 goto bad;
3363 error = mac_getsockopt_peerlabel(
3364 sopt->sopt_td->td_ucred, so, &extmac);
3365 if (error)
3366 goto bad;
3367 error = sooptcopyout(sopt, &extmac, sizeof extmac);
3368 #else
3369 error = EOPNOTSUPP;
3370 #endif
3371 break;
3372
3373 case SO_LISTENQLIMIT:
3374 optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3375 goto integer;
3376
3377 case SO_LISTENQLEN:
3378 optval = SOLISTENING(so) ? so->sol_qlen : 0;
3379 goto integer;
3380
3381 case SO_LISTENINCQLEN:
3382 optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3383 goto integer;
3384
3385 case SO_TS_CLOCK:
3386 optval = so->so_ts_clock;
3387 goto integer;
3388
3389 case SO_MAX_PACING_RATE:
3390 optval = so->so_max_pacing_rate;
3391 goto integer;
3392
3393 default:
3394 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3395 error = hhook_run_socket(so, sopt,
3396 HHOOK_SOCKET_OPT);
3397 else
3398 error = ENOPROTOOPT;
3399 break;
3400 }
3401 }
3402 #ifdef MAC
3403 bad:
3404 #endif
3405 CURVNET_RESTORE();
3406 return (error);
3407 }
3408
3409 int
3410 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3411 {
3412 struct mbuf *m, *m_prev;
3413 int sopt_size = sopt->sopt_valsize;
3414
3415 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3416 if (m == NULL)
3417 return ENOBUFS;
3418 if (sopt_size > MLEN) {
3419 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3420 if ((m->m_flags & M_EXT) == 0) {
3421 m_free(m);
3422 return ENOBUFS;
3423 }
3424 m->m_len = min(MCLBYTES, sopt_size);
3425 } else {
3426 m->m_len = min(MLEN, sopt_size);
3427 }
3428 sopt_size -= m->m_len;
3429 *mp = m;
3430 m_prev = m;
3431
3432 while (sopt_size) {
3433 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3434 if (m == NULL) {
3435 m_freem(*mp);
3436 return ENOBUFS;
3437 }
3438 if (sopt_size > MLEN) {
3439 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3440 M_NOWAIT);
3441 if ((m->m_flags & M_EXT) == 0) {
3442 m_freem(m);
3443 m_freem(*mp);
3444 return ENOBUFS;
3445 }
3446 m->m_len = min(MCLBYTES, sopt_size);
3447 } else {
3448 m->m_len = min(MLEN, sopt_size);
3449 }
3450 sopt_size -= m->m_len;
3451 m_prev->m_next = m;
3452 m_prev = m;
3453 }
3454 return (0);
3455 }
3456
3457 int
3458 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3459 {
3460 struct mbuf *m0 = m;
3461
3462 if (sopt->sopt_val == NULL)
3463 return (0);
3464 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3465 if (sopt->sopt_td != NULL) {
3466 int error;
3467
3468 error = copyin(sopt->sopt_val, mtod(m, char *),
3469 m->m_len);
3470 if (error != 0) {
3471 m_freem(m0);
3472 return(error);
3473 }
3474 } else
3475 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3476 sopt->sopt_valsize -= m->m_len;
3477 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3478 m = m->m_next;
3479 }
3480 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3481 panic("ip6_sooptmcopyin");
3482 return (0);
3483 }
3484
3485 int
3486 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3487 {
3488 struct mbuf *m0 = m;
3489 size_t valsize = 0;
3490
3491 if (sopt->sopt_val == NULL)
3492 return (0);
3493 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3494 if (sopt->sopt_td != NULL) {
3495 int error;
3496
3497 error = copyout(mtod(m, char *), sopt->sopt_val,
3498 m->m_len);
3499 if (error != 0) {
3500 m_freem(m0);
3501 return(error);
3502 }
3503 } else
3504 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3505 sopt->sopt_valsize -= m->m_len;
3506 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3507 valsize += m->m_len;
3508 m = m->m_next;
3509 }
3510 if (m != NULL) {
3511 /* enough soopt buffer should be given from user-land */
3512 m_freem(m0);
3513 return(EINVAL);
3514 }
3515 sopt->sopt_valsize = valsize;
3516 return (0);
3517 }
3518
3519 /*
3520 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3521 * out-of-band data, which will then notify socket consumers.
3522 */
3523 void
3524 sohasoutofband(struct socket *so)
3525 {
3526
3527 if (so->so_sigio != NULL)
3528 pgsigio(&so->so_sigio, SIGURG, 0);
3529 selwakeuppri(&so->so_rdsel, PSOCK);
3530 }
3531
3532 int
3533 sopoll(struct socket *so, int events, struct ucred *active_cred,
3534 struct thread *td)
3535 {
3536
3537 /*
3538 * We do not need to set or assert curvnet as long as everyone uses
3539 * sopoll_generic().
3540 */
3541 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3542 td));
3543 }
3544
3545 int
3546 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3547 struct thread *td)
3548 {
3549 int revents;
3550
3551 SOCK_LOCK(so);
3552 if (SOLISTENING(so)) {
3553 if (!(events & (POLLIN | POLLRDNORM)))
3554 revents = 0;
3555 else if (!TAILQ_EMPTY(&so->sol_comp))
3556 revents = events & (POLLIN | POLLRDNORM);
3557 else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3558 revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3559 else {
3560 selrecord(td, &so->so_rdsel);
3561 revents = 0;
3562 }
3563 } else {
3564 revents = 0;
3565 SOCKBUF_LOCK(&so->so_snd);
3566 SOCKBUF_LOCK(&so->so_rcv);
3567 if (events & (POLLIN | POLLRDNORM))
3568 if (soreadabledata(so))
3569 revents |= events & (POLLIN | POLLRDNORM);
3570 if (events & (POLLOUT | POLLWRNORM))
3571 if (sowriteable(so))
3572 revents |= events & (POLLOUT | POLLWRNORM);
3573 if (events & (POLLPRI | POLLRDBAND))
3574 if (so->so_oobmark ||
3575 (so->so_rcv.sb_state & SBS_RCVATMARK))
3576 revents |= events & (POLLPRI | POLLRDBAND);
3577 if ((events & POLLINIGNEOF) == 0) {
3578 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3579 revents |= events & (POLLIN | POLLRDNORM);
3580 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3581 revents |= POLLHUP;
3582 }
3583 }
3584 if (revents == 0) {
3585 if (events &
3586 (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3587 selrecord(td, &so->so_rdsel);
3588 so->so_rcv.sb_flags |= SB_SEL;
3589 }
3590 if (events & (POLLOUT | POLLWRNORM)) {
3591 selrecord(td, &so->so_wrsel);
3592 so->so_snd.sb_flags |= SB_SEL;
3593 }
3594 }
3595 SOCKBUF_UNLOCK(&so->so_rcv);
3596 SOCKBUF_UNLOCK(&so->so_snd);
3597 }
3598 SOCK_UNLOCK(so);
3599 return (revents);
3600 }
3601
3602 int
3603 soo_kqfilter(struct file *fp, struct knote *kn)
3604 {
3605 struct socket *so = kn->kn_fp->f_data;
3606 struct sockbuf *sb;
3607 struct knlist *knl;
3608
3609 switch (kn->kn_filter) {
3610 case EVFILT_READ:
3611 kn->kn_fop = &soread_filtops;
3612 knl = &so->so_rdsel.si_note;
3613 sb = &so->so_rcv;
3614 break;
3615 case EVFILT_WRITE:
3616 kn->kn_fop = &sowrite_filtops;
3617 knl = &so->so_wrsel.si_note;
3618 sb = &so->so_snd;
3619 break;
3620 case EVFILT_EMPTY:
3621 kn->kn_fop = &soempty_filtops;
3622 knl = &so->so_wrsel.si_note;
3623 sb = &so->so_snd;
3624 break;
3625 default:
3626 return (EINVAL);
3627 }
3628
3629 SOCK_LOCK(so);
3630 if (SOLISTENING(so)) {
3631 knlist_add(knl, kn, 1);
3632 } else {
3633 SOCKBUF_LOCK(sb);
3634 knlist_add(knl, kn, 1);
3635 sb->sb_flags |= SB_KNOTE;
3636 SOCKBUF_UNLOCK(sb);
3637 }
3638 SOCK_UNLOCK(so);
3639 return (0);
3640 }
3641
3642 /*
3643 * Some routines that return EOPNOTSUPP for entry points that are not
3644 * supported by a protocol. Fill in as needed.
3645 */
3646 int
3647 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3648 {
3649
3650 return EOPNOTSUPP;
3651 }
3652
3653 int
3654 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3655 {
3656
3657 return EOPNOTSUPP;
3658 }
3659
3660 int
3661 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3662 {
3663
3664 return EOPNOTSUPP;
3665 }
3666
3667 int
3668 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3669 {
3670
3671 return EOPNOTSUPP;
3672 }
3673
3674 int
3675 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3676 struct thread *td)
3677 {
3678
3679 return EOPNOTSUPP;
3680 }
3681
3682 int
3683 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3684 {
3685
3686 return EOPNOTSUPP;
3687 }
3688
3689 int
3690 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3691 struct thread *td)
3692 {
3693
3694 return EOPNOTSUPP;
3695 }
3696
3697 int
3698 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3699 {
3700
3701 return EOPNOTSUPP;
3702 }
3703
3704 int
3705 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3706 struct ifnet *ifp, struct thread *td)
3707 {
3708
3709 return EOPNOTSUPP;
3710 }
3711
3712 int
3713 pru_disconnect_notsupp(struct socket *so)
3714 {
3715
3716 return EOPNOTSUPP;
3717 }
3718
3719 int
3720 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3721 {
3722
3723 return EOPNOTSUPP;
3724 }
3725
3726 int
3727 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3728 {
3729
3730 return EOPNOTSUPP;
3731 }
3732
3733 int
3734 pru_rcvd_notsupp(struct socket *so, int flags)
3735 {
3736
3737 return EOPNOTSUPP;
3738 }
3739
3740 int
3741 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3742 {
3743
3744 return EOPNOTSUPP;
3745 }
3746
3747 int
3748 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3749 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3750 {
3751
3752 if (control != NULL)
3753 m_freem(control);
3754 if ((flags & PRUS_NOTREADY) == 0)
3755 m_freem(m);
3756 return (EOPNOTSUPP);
3757 }
3758
3759 int
3760 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3761 {
3762
3763 return (EOPNOTSUPP);
3764 }
3765
3766 /*
3767 * This isn't really a ``null'' operation, but it's the default one and
3768 * doesn't do anything destructive.
3769 */
3770 int
3771 pru_sense_null(struct socket *so, struct stat *sb)
3772 {
3773
3774 sb->st_blksize = so->so_snd.sb_hiwat;
3775 return 0;
3776 }
3777
3778 int
3779 pru_shutdown_notsupp(struct socket *so)
3780 {
3781
3782 return EOPNOTSUPP;
3783 }
3784
3785 int
3786 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3787 {
3788
3789 return EOPNOTSUPP;
3790 }
3791
3792 int
3793 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3794 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3795 {
3796
3797 return EOPNOTSUPP;
3798 }
3799
3800 int
3801 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3802 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3803 {
3804
3805 return EOPNOTSUPP;
3806 }
3807
3808 int
3809 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3810 struct thread *td)
3811 {
3812
3813 return EOPNOTSUPP;
3814 }
3815
3816 static void
3817 filt_sordetach(struct knote *kn)
3818 {
3819 struct socket *so = kn->kn_fp->f_data;
3820
3821 so_rdknl_lock(so);
3822 knlist_remove(&so->so_rdsel.si_note, kn, 1);
3823 if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3824 so->so_rcv.sb_flags &= ~SB_KNOTE;
3825 so_rdknl_unlock(so);
3826 }
3827
3828 /*ARGSUSED*/
3829 static int
3830 filt_soread(struct knote *kn, long hint)
3831 {
3832 struct socket *so;
3833
3834 so = kn->kn_fp->f_data;
3835
3836 if (SOLISTENING(so)) {
3837 SOCK_LOCK_ASSERT(so);
3838 kn->kn_data = so->sol_qlen;
3839 if (so->so_error) {
3840 kn->kn_flags |= EV_EOF;
3841 kn->kn_fflags = so->so_error;
3842 return (1);
3843 }
3844 return (!TAILQ_EMPTY(&so->sol_comp));
3845 }
3846
3847 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3848
3849 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3850 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3851 kn->kn_flags |= EV_EOF;
3852 kn->kn_fflags = so->so_error;
3853 return (1);
3854 } else if (so->so_error || so->so_rerror)
3855 return (1);
3856
3857 if (kn->kn_sfflags & NOTE_LOWAT) {
3858 if (kn->kn_data >= kn->kn_sdata)
3859 return (1);
3860 } else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3861 return (1);
3862
3863 /* This hook returning non-zero indicates an event, not error */
3864 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3865 }
3866
3867 static void
3868 filt_sowdetach(struct knote *kn)
3869 {
3870 struct socket *so = kn->kn_fp->f_data;
3871
3872 so_wrknl_lock(so);
3873 knlist_remove(&so->so_wrsel.si_note, kn, 1);
3874 if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3875 so->so_snd.sb_flags &= ~SB_KNOTE;
3876 so_wrknl_unlock(so);
3877 }
3878
3879 /*ARGSUSED*/
3880 static int
3881 filt_sowrite(struct knote *kn, long hint)
3882 {
3883 struct socket *so;
3884
3885 so = kn->kn_fp->f_data;
3886
3887 if (SOLISTENING(so))
3888 return (0);
3889
3890 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3891 kn->kn_data = sbspace(&so->so_snd);
3892
3893 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3894
3895 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3896 kn->kn_flags |= EV_EOF;
3897 kn->kn_fflags = so->so_error;
3898 return (1);
3899 } else if (so->so_error) /* temporary udp error */
3900 return (1);
3901 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3902 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3903 return (0);
3904 else if (kn->kn_sfflags & NOTE_LOWAT)
3905 return (kn->kn_data >= kn->kn_sdata);
3906 else
3907 return (kn->kn_data >= so->so_snd.sb_lowat);
3908 }
3909
3910 static int
3911 filt_soempty(struct knote *kn, long hint)
3912 {
3913 struct socket *so;
3914
3915 so = kn->kn_fp->f_data;
3916
3917 if (SOLISTENING(so))
3918 return (1);
3919
3920 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3921 kn->kn_data = sbused(&so->so_snd);
3922
3923 if (kn->kn_data == 0)
3924 return (1);
3925 else
3926 return (0);
3927 }
3928
3929 int
3930 socheckuid(struct socket *so, uid_t uid)
3931 {
3932
3933 if (so == NULL)
3934 return (EPERM);
3935 if (so->so_cred->cr_uid != uid)
3936 return (EPERM);
3937 return (0);
3938 }
3939
3940 /*
3941 * These functions are used by protocols to notify the socket layer (and its
3942 * consumers) of state changes in the sockets driven by protocol-side events.
3943 */
3944
3945 /*
3946 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3947 *
3948 * Normal sequence from the active (originating) side is that
3949 * soisconnecting() is called during processing of connect() call, resulting
3950 * in an eventual call to soisconnected() if/when the connection is
3951 * established. When the connection is torn down soisdisconnecting() is
3952 * called during processing of disconnect() call, and soisdisconnected() is
3953 * called when the connection to the peer is totally severed. The semantics
3954 * of these routines are such that connectionless protocols can call
3955 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3956 * calls when setting up a ``connection'' takes no time.
3957 *
3958 * From the passive side, a socket is created with two queues of sockets:
3959 * so_incomp for connections in progress and so_comp for connections already
3960 * made and awaiting user acceptance. As a protocol is preparing incoming
3961 * connections, it creates a socket structure queued on so_incomp by calling
3962 * sonewconn(). When the connection is established, soisconnected() is
3963 * called, and transfers the socket structure to so_comp, making it available
3964 * to accept().
3965 *
3966 * If a socket is closed with sockets on either so_incomp or so_comp, these
3967 * sockets are dropped.
3968 *
3969 * If higher-level protocols are implemented in the kernel, the wakeups done
3970 * here will sometimes cause software-interrupt process scheduling.
3971 */
3972 void
3973 soisconnecting(struct socket *so)
3974 {
3975
3976 SOCK_LOCK(so);
3977 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3978 so->so_state |= SS_ISCONNECTING;
3979 SOCK_UNLOCK(so);
3980 }
3981
3982 void
3983 soisconnected(struct socket *so)
3984 {
3985 bool last __diagused;
3986
3987 SOCK_LOCK(so);
3988 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3989 so->so_state |= SS_ISCONNECTED;
3990
3991 if (so->so_qstate == SQ_INCOMP) {
3992 struct socket *head = so->so_listen;
3993 int ret;
3994
3995 KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3996 /*
3997 * Promoting a socket from incomplete queue to complete, we
3998 * need to go through reverse order of locking. We first do
3999 * trylock, and if that doesn't succeed, we go the hard way
4000 * leaving a reference and rechecking consistency after proper
4001 * locking.
4002 */
4003 if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4004 soref(head);
4005 SOCK_UNLOCK(so);
4006 SOLISTEN_LOCK(head);
4007 SOCK_LOCK(so);
4008 if (__predict_false(head != so->so_listen)) {
4009 /*
4010 * The socket went off the listen queue,
4011 * should be lost race to close(2) of sol.
4012 * The socket is about to soabort().
4013 */
4014 SOCK_UNLOCK(so);
4015 sorele(head);
4016 return;
4017 }
4018 last = refcount_release(&head->so_count);
4019 KASSERT(!last, ("%s: released last reference for %p",
4020 __func__, head));
4021 }
4022 again:
4023 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4024 TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4025 head->sol_incqlen--;
4026 TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4027 head->sol_qlen++;
4028 so->so_qstate = SQ_COMP;
4029 SOCK_UNLOCK(so);
4030 solisten_wakeup(head); /* unlocks */
4031 } else {
4032 SOCKBUF_LOCK(&so->so_rcv);
4033 soupcall_set(so, SO_RCV,
4034 head->sol_accept_filter->accf_callback,
4035 head->sol_accept_filter_arg);
4036 so->so_options &= ~SO_ACCEPTFILTER;
4037 ret = head->sol_accept_filter->accf_callback(so,
4038 head->sol_accept_filter_arg, M_NOWAIT);
4039 if (ret == SU_ISCONNECTED) {
4040 soupcall_clear(so, SO_RCV);
4041 SOCKBUF_UNLOCK(&so->so_rcv);
4042 goto again;
4043 }
4044 SOCKBUF_UNLOCK(&so->so_rcv);
4045 SOCK_UNLOCK(so);
4046 SOLISTEN_UNLOCK(head);
4047 }
4048 return;
4049 }
4050 SOCK_UNLOCK(so);
4051 wakeup(&so->so_timeo);
4052 sorwakeup(so);
4053 sowwakeup(so);
4054 }
4055
4056 void
4057 soisdisconnecting(struct socket *so)
4058 {
4059
4060 SOCK_LOCK(so);
4061 so->so_state &= ~SS_ISCONNECTING;
4062 so->so_state |= SS_ISDISCONNECTING;
4063
4064 if (!SOLISTENING(so)) {
4065 SOCKBUF_LOCK(&so->so_rcv);
4066 socantrcvmore_locked(so);
4067 SOCKBUF_LOCK(&so->so_snd);
4068 socantsendmore_locked(so);
4069 }
4070 SOCK_UNLOCK(so);
4071 wakeup(&so->so_timeo);
4072 }
4073
4074 void
4075 soisdisconnected(struct socket *so)
4076 {
4077
4078 SOCK_LOCK(so);
4079
4080 /*
4081 * There is at least one reader of so_state that does not
4082 * acquire socket lock, namely soreceive_generic(). Ensure
4083 * that it never sees all flags that track connection status
4084 * cleared, by ordering the update with a barrier semantic of
4085 * our release thread fence.
4086 */
4087 so->so_state |= SS_ISDISCONNECTED;
4088 atomic_thread_fence_rel();
4089 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4090
4091 if (!SOLISTENING(so)) {
4092 SOCK_UNLOCK(so);
4093 SOCKBUF_LOCK(&so->so_rcv);
4094 socantrcvmore_locked(so);
4095 SOCKBUF_LOCK(&so->so_snd);
4096 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4097 socantsendmore_locked(so);
4098 } else
4099 SOCK_UNLOCK(so);
4100 wakeup(&so->so_timeo);
4101 }
4102
4103 int
4104 soiolock(struct socket *so, struct sx *sx, int flags)
4105 {
4106 int error;
4107
4108 KASSERT((flags & SBL_VALID) == flags,
4109 ("soiolock: invalid flags %#x", flags));
4110
4111 if ((flags & SBL_WAIT) != 0) {
4112 if ((flags & SBL_NOINTR) != 0) {
4113 sx_xlock(sx);
4114 } else {
4115 error = sx_xlock_sig(sx);
4116 if (error != 0)
4117 return (error);
4118 }
4119 } else if (!sx_try_xlock(sx)) {
4120 return (EWOULDBLOCK);
4121 }
4122
4123 if (__predict_false(SOLISTENING(so))) {
4124 sx_xunlock(sx);
4125 return (ENOTCONN);
4126 }
4127 return (0);
4128 }
4129
4130 void
4131 soiounlock(struct sx *sx)
4132 {
4133 sx_xunlock(sx);
4134 }
4135
4136 /*
4137 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4138 */
4139 struct sockaddr *
4140 sodupsockaddr(const struct sockaddr *sa, int mflags)
4141 {
4142 struct sockaddr *sa2;
4143
4144 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4145 if (sa2)
4146 bcopy(sa, sa2, sa->sa_len);
4147 return sa2;
4148 }
4149
4150 /*
4151 * Register per-socket destructor.
4152 */
4153 void
4154 sodtor_set(struct socket *so, so_dtor_t *func)
4155 {
4156
4157 SOCK_LOCK_ASSERT(so);
4158 so->so_dtor = func;
4159 }
4160
4161 /*
4162 * Register per-socket buffer upcalls.
4163 */
4164 void
4165 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4166 {
4167 struct sockbuf *sb;
4168
4169 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4170
4171 switch (which) {
4172 case SO_RCV:
4173 sb = &so->so_rcv;
4174 break;
4175 case SO_SND:
4176 sb = &so->so_snd;
4177 break;
4178 default:
4179 panic("soupcall_set: bad which");
4180 }
4181 SOCKBUF_LOCK_ASSERT(sb);
4182 sb->sb_upcall = func;
4183 sb->sb_upcallarg = arg;
4184 sb->sb_flags |= SB_UPCALL;
4185 }
4186
4187 void
4188 soupcall_clear(struct socket *so, int which)
4189 {
4190 struct sockbuf *sb;
4191
4192 KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4193
4194 switch (which) {
4195 case SO_RCV:
4196 sb = &so->so_rcv;
4197 break;
4198 case SO_SND:
4199 sb = &so->so_snd;
4200 break;
4201 default:
4202 panic("soupcall_clear: bad which");
4203 }
4204 SOCKBUF_LOCK_ASSERT(sb);
4205 KASSERT(sb->sb_upcall != NULL,
4206 ("%s: so %p no upcall to clear", __func__, so));
4207 sb->sb_upcall = NULL;
4208 sb->sb_upcallarg = NULL;
4209 sb->sb_flags &= ~SB_UPCALL;
4210 }
4211
4212 void
4213 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4214 {
4215
4216 SOLISTEN_LOCK_ASSERT(so);
4217 so->sol_upcall = func;
4218 so->sol_upcallarg = arg;
4219 }
4220
4221 static void
4222 so_rdknl_lock(void *arg)
4223 {
4224 struct socket *so = arg;
4225
4226 if (SOLISTENING(so))
4227 SOCK_LOCK(so);
4228 else
4229 SOCKBUF_LOCK(&so->so_rcv);
4230 }
4231
4232 static void
4233 so_rdknl_unlock(void *arg)
4234 {
4235 struct socket *so = arg;
4236
4237 if (SOLISTENING(so))
4238 SOCK_UNLOCK(so);
4239 else
4240 SOCKBUF_UNLOCK(&so->so_rcv);
4241 }
4242
4243 static void
4244 so_rdknl_assert_lock(void *arg, int what)
4245 {
4246 struct socket *so = arg;
4247
4248 if (what == LA_LOCKED) {
4249 if (SOLISTENING(so))
4250 SOCK_LOCK_ASSERT(so);
4251 else
4252 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4253 } else {
4254 if (SOLISTENING(so))
4255 SOCK_UNLOCK_ASSERT(so);
4256 else
4257 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4258 }
4259 }
4260
4261 static void
4262 so_wrknl_lock(void *arg)
4263 {
4264 struct socket *so = arg;
4265
4266 if (SOLISTENING(so))
4267 SOCK_LOCK(so);
4268 else
4269 SOCKBUF_LOCK(&so->so_snd);
4270 }
4271
4272 static void
4273 so_wrknl_unlock(void *arg)
4274 {
4275 struct socket *so = arg;
4276
4277 if (SOLISTENING(so))
4278 SOCK_UNLOCK(so);
4279 else
4280 SOCKBUF_UNLOCK(&so->so_snd);
4281 }
4282
4283 static void
4284 so_wrknl_assert_lock(void *arg, int what)
4285 {
4286 struct socket *so = arg;
4287
4288 if (what == LA_LOCKED) {
4289 if (SOLISTENING(so))
4290 SOCK_LOCK_ASSERT(so);
4291 else
4292 SOCKBUF_LOCK_ASSERT(&so->so_snd);
4293 } else {
4294 if (SOLISTENING(so))
4295 SOCK_UNLOCK_ASSERT(so);
4296 else
4297 SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4298 }
4299 }
4300
4301 /*
4302 * Create an external-format (``xsocket'') structure using the information in
4303 * the kernel-format socket structure pointed to by so. This is done to
4304 * reduce the spew of irrelevant information over this interface, to isolate
4305 * user code from changes in the kernel structure, and potentially to provide
4306 * information-hiding if we decide that some of this information should be
4307 * hidden from users.
4308 */
4309 void
4310 sotoxsocket(struct socket *so, struct xsocket *xso)
4311 {
4312
4313 bzero(xso, sizeof(*xso));
4314 xso->xso_len = sizeof *xso;
4315 xso->xso_so = (uintptr_t)so;
4316 xso->so_type = so->so_type;
4317 xso->so_options = so->so_options;
4318 xso->so_linger = so->so_linger;
4319 xso->so_state = so->so_state;
4320 xso->so_pcb = (uintptr_t)so->so_pcb;
4321 xso->xso_protocol = so->so_proto->pr_protocol;
4322 xso->xso_family = so->so_proto->pr_domain->dom_family;
4323 xso->so_timeo = so->so_timeo;
4324 xso->so_error = so->so_error;
4325 xso->so_uid = so->so_cred->cr_uid;
4326 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4327 if (SOLISTENING(so)) {
4328 xso->so_qlen = so->sol_qlen;
4329 xso->so_incqlen = so->sol_incqlen;
4330 xso->so_qlimit = so->sol_qlimit;
4331 xso->so_oobmark = 0;
4332 } else {
4333 xso->so_state |= so->so_qstate;
4334 xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4335 xso->so_oobmark = so->so_oobmark;
4336 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4337 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4338 }
4339 }
4340
4341 struct sockbuf *
4342 so_sockbuf_rcv(struct socket *so)
4343 {
4344
4345 return (&so->so_rcv);
4346 }
4347
4348 struct sockbuf *
4349 so_sockbuf_snd(struct socket *so)
4350 {
4351
4352 return (&so->so_snd);
4353 }
4354
4355 int
4356 so_state_get(const struct socket *so)
4357 {
4358
4359 return (so->so_state);
4360 }
4361
4362 void
4363 so_state_set(struct socket *so, int val)
4364 {
4365
4366 so->so_state = val;
4367 }
4368
4369 int
4370 so_options_get(const struct socket *so)
4371 {
4372
4373 return (so->so_options);
4374 }
4375
4376 void
4377 so_options_set(struct socket *so, int val)
4378 {
4379
4380 so->so_options = val;
4381 }
4382
4383 int
4384 so_error_get(const struct socket *so)
4385 {
4386
4387 return (so->so_error);
4388 }
4389
4390 void
4391 so_error_set(struct socket *so, int val)
4392 {
4393
4394 so->so_error = val;
4395 }
4396
4397 int
4398 so_linger_get(const struct socket *so)
4399 {
4400
4401 return (so->so_linger);
4402 }
4403
4404 void
4405 so_linger_set(struct socket *so, int val)
4406 {
4407
4408 KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4409 ("%s: val %d out of range", __func__, val));
4410
4411 so->so_linger = val;
4412 }
4413
4414 struct protosw *
4415 so_protosw_get(const struct socket *so)
4416 {
4417
4418 return (so->so_proto);
4419 }
4420
4421 void
4422 so_protosw_set(struct socket *so, struct protosw *val)
4423 {
4424
4425 so->so_proto = val;
4426 }
4427
4428 void
4429 so_sorwakeup(struct socket *so)
4430 {
4431
4432 sorwakeup(so);
4433 }
4434
4435 void
4436 so_sowwakeup(struct socket *so)
4437 {
4438
4439 sowwakeup(so);
4440 }
4441
4442 void
4443 so_sorwakeup_locked(struct socket *so)
4444 {
4445
4446 sorwakeup_locked(so);
4447 }
4448
4449 void
4450 so_sowwakeup_locked(struct socket *so)
4451 {
4452
4453 sowwakeup_locked(so);
4454 }
4455
4456 void
4457 so_lock(struct socket *so)
4458 {
4459
4460 SOCK_LOCK(so);
4461 }
4462
4463 void
4464 so_unlock(struct socket *so)
4465 {
4466
4467 SOCK_UNLOCK(so);
4468 }
4469