1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2008 Robert N. M. Watson
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
33 */
34
35 /*
36 * Comments on the socket life cycle:
37 *
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
40 *
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
43 *
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
47 *
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
52 *
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
57 * case.
58 *
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
61 * sockets.
62 *
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
66 * listen socket.
67 *
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
71 *
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
82 *
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
86 * interface.
87 *
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
95 *
96 * NOTE: With regard to VNETs the general rule is that callers do not set
97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99 * and sorflush(), which are usually called from a pre-set VNET context.
100 * sopoll() currently does not need a VNET context to be set.
101 */
102
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD: releng/11.1/sys/kern/uipc_socket.c 318268 2017-05-14 12:41:58Z sobomax $");
105
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_compat.h"
109
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
115 #include <sys/mac.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h> /* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/event.h>
125 #include <sys/eventhandler.h>
126 #include <sys/poll.h>
127 #include <sys/proc.h>
128 #include <sys/protosw.h>
129 #include <sys/socket.h>
130 #include <sys/socketvar.h>
131 #include <sys/resourcevar.h>
132 #include <net/route.h>
133 #include <sys/signalvar.h>
134 #include <sys/stat.h>
135 #include <sys/sx.h>
136 #include <sys/sysctl.h>
137 #include <sys/taskqueue.h>
138 #include <sys/uio.h>
139 #include <sys/jail.h>
140 #include <sys/syslog.h>
141 #include <netinet/in.h>
142
143 #include <net/vnet.h>
144
145 #include <security/mac/mac_framework.h>
146
147 #include <vm/uma.h>
148
149 #ifdef COMPAT_FREEBSD32
150 #include <sys/mount.h>
151 #include <sys/sysent.h>
152 #include <compat/freebsd32/freebsd32.h>
153 #endif
154
155 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
156 int flags);
157
158 static void filt_sordetach(struct knote *kn);
159 static int filt_soread(struct knote *kn, long hint);
160 static void filt_sowdetach(struct knote *kn);
161 static int filt_sowrite(struct knote *kn, long hint);
162 static int filt_solisten(struct knote *kn, long hint);
163 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
164 fo_kqfilter_t soo_kqfilter;
165
166 static struct filterops solisten_filtops = {
167 .f_isfd = 1,
168 .f_detach = filt_sordetach,
169 .f_event = filt_solisten,
170 };
171 static struct filterops soread_filtops = {
172 .f_isfd = 1,
173 .f_detach = filt_sordetach,
174 .f_event = filt_soread,
175 };
176 static struct filterops sowrite_filtops = {
177 .f_isfd = 1,
178 .f_detach = filt_sowdetach,
179 .f_event = filt_sowrite,
180 };
181
182 so_gen_t so_gencnt; /* generation count for sockets */
183
184 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
185 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
186
187 #define VNET_SO_ASSERT(so) \
188 VNET_ASSERT(curvnet != NULL, \
189 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
190
191 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
192 #define V_socket_hhh VNET(socket_hhh)
193
194 /*
195 * Limit on the number of connections in the listen queue waiting
196 * for accept(2).
197 * NB: The original sysctl somaxconn is still available but hidden
198 * to prevent confusion about the actual purpose of this number.
199 */
200 static u_int somaxconn = SOMAXCONN;
201
202 static int
203 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
204 {
205 int error;
206 int val;
207
208 val = somaxconn;
209 error = sysctl_handle_int(oidp, &val, 0, req);
210 if (error || !req->newptr )
211 return (error);
212
213 /*
214 * The purpose of the UINT_MAX / 3 limit, is so that the formula
215 * 3 * so_qlimit / 2
216 * below, will not overflow.
217 */
218
219 if (val < 1 || val > UINT_MAX / 3)
220 return (EINVAL);
221
222 somaxconn = val;
223 return (0);
224 }
225 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
226 0, sizeof(int), sysctl_somaxconn, "I",
227 "Maximum listen socket pending connection accept queue size");
228 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
229 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
230 0, sizeof(int), sysctl_somaxconn, "I",
231 "Maximum listen socket pending connection accept queue size (compat)");
232
233 static int numopensockets;
234 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
235 &numopensockets, 0, "Number of open sockets");
236
237 /*
238 * accept_mtx locks down per-socket fields relating to accept queues. See
239 * socketvar.h for an annotation of the protected fields of struct socket.
240 */
241 struct mtx accept_mtx;
242 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
243
244 /*
245 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
246 * so_gencnt field.
247 */
248 static struct mtx so_global_mtx;
249 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
250
251 /*
252 * General IPC sysctl name space, used by sockets and a variety of other IPC
253 * types.
254 */
255 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
256
257 /*
258 * Initialize the socket subsystem and set up the socket
259 * memory allocator.
260 */
261 static uma_zone_t socket_zone;
262 int maxsockets;
263
264 static void
265 socket_zone_change(void *tag)
266 {
267
268 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
269 }
270
271 static void
272 socket_hhook_register(int subtype)
273 {
274
275 if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
276 &V_socket_hhh[subtype],
277 HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
278 printf("%s: WARNING: unable to register hook\n", __func__);
279 }
280
281 static void
282 socket_hhook_deregister(int subtype)
283 {
284
285 if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
286 printf("%s: WARNING: unable to deregister hook\n", __func__);
287 }
288
289 static void
290 socket_init(void *tag)
291 {
292
293 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
294 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
296 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
297 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
298 EVENTHANDLER_PRI_FIRST);
299 }
300 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
301
302 static void
303 socket_vnet_init(const void *unused __unused)
304 {
305 int i;
306
307 /* We expect a contiguous range */
308 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
309 socket_hhook_register(i);
310 }
311 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
312 socket_vnet_init, NULL);
313
314 static void
315 socket_vnet_uninit(const void *unused __unused)
316 {
317 int i;
318
319 for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
320 socket_hhook_deregister(i);
321 }
322 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
323 socket_vnet_uninit, NULL);
324
325 /*
326 * Initialise maxsockets. This SYSINIT must be run after
327 * tunable_mbinit().
328 */
329 static void
330 init_maxsockets(void *ignored)
331 {
332
333 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
334 maxsockets = imax(maxsockets, maxfiles);
335 }
336 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
337
338 /*
339 * Sysctl to get and set the maximum global sockets limit. Notify protocols
340 * of the change so that they can update their dependent limits as required.
341 */
342 static int
343 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
344 {
345 int error, newmaxsockets;
346
347 newmaxsockets = maxsockets;
348 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
349 if (error == 0 && req->newptr) {
350 if (newmaxsockets > maxsockets &&
351 newmaxsockets <= maxfiles) {
352 maxsockets = newmaxsockets;
353 EVENTHANDLER_INVOKE(maxsockets_change);
354 } else
355 error = EINVAL;
356 }
357 return (error);
358 }
359 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
360 &maxsockets, 0, sysctl_maxsockets, "IU",
361 "Maximum number of sockets available");
362
363 /*
364 * Socket operation routines. These routines are called by the routines in
365 * sys_socket.c or from a system process, and implement the semantics of
366 * socket operations by switching out to the protocol specific routines.
367 */
368
369 /*
370 * Get a socket structure from our zone, and initialize it. Note that it
371 * would probably be better to allocate socket and PCB at the same time, but
372 * I'm not convinced that all the protocols can be easily modified to do
373 * this.
374 *
375 * soalloc() returns a socket with a ref count of 0.
376 */
377 static struct socket *
378 soalloc(struct vnet *vnet)
379 {
380 struct socket *so;
381
382 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
383 if (so == NULL)
384 return (NULL);
385 #ifdef MAC
386 if (mac_socket_init(so, M_NOWAIT) != 0) {
387 uma_zfree(socket_zone, so);
388 return (NULL);
389 }
390 #endif
391 if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
392 uma_zfree(socket_zone, so);
393 return (NULL);
394 }
395
396 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
397 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
398 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
399 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
400 TAILQ_INIT(&so->so_snd.sb_aiojobq);
401 TAILQ_INIT(&so->so_rcv.sb_aiojobq);
402 TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
403 TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
404 #ifdef VIMAGE
405 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
406 __func__, __LINE__, so));
407 so->so_vnet = vnet;
408 #endif
409 /* We shouldn't need the so_global_mtx */
410 if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
411 /* Do we need more comprehensive error returns? */
412 uma_zfree(socket_zone, so);
413 return (NULL);
414 }
415 mtx_lock(&so_global_mtx);
416 so->so_gencnt = ++so_gencnt;
417 ++numopensockets;
418 #ifdef VIMAGE
419 vnet->vnet_sockcnt++;
420 #endif
421 mtx_unlock(&so_global_mtx);
422
423 return (so);
424 }
425
426 /*
427 * Free the storage associated with a socket at the socket layer, tear down
428 * locks, labels, etc. All protocol state is assumed already to have been
429 * torn down (and possibly never set up) by the caller.
430 */
431 static void
432 sodealloc(struct socket *so)
433 {
434
435 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
436 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
437
438 mtx_lock(&so_global_mtx);
439 so->so_gencnt = ++so_gencnt;
440 --numopensockets; /* Could be below, but faster here. */
441 #ifdef VIMAGE
442 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
443 __func__, __LINE__, so));
444 so->so_vnet->vnet_sockcnt--;
445 #endif
446 mtx_unlock(&so_global_mtx);
447 if (so->so_rcv.sb_hiwat)
448 (void)chgsbsize(so->so_cred->cr_uidinfo,
449 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
450 if (so->so_snd.sb_hiwat)
451 (void)chgsbsize(so->so_cred->cr_uidinfo,
452 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
453 /* remove accept filter if one is present. */
454 if (so->so_accf != NULL)
455 do_setopt_accept_filter(so, NULL);
456 #ifdef MAC
457 mac_socket_destroy(so);
458 #endif
459 hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
460
461 crfree(so->so_cred);
462 khelp_destroy_osd(&so->osd);
463 sx_destroy(&so->so_snd.sb_sx);
464 sx_destroy(&so->so_rcv.sb_sx);
465 SOCKBUF_LOCK_DESTROY(&so->so_snd);
466 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
467 uma_zfree(socket_zone, so);
468 }
469
470 /*
471 * socreate returns a socket with a ref count of 1. The socket should be
472 * closed with soclose().
473 */
474 int
475 socreate(int dom, struct socket **aso, int type, int proto,
476 struct ucred *cred, struct thread *td)
477 {
478 struct protosw *prp;
479 struct socket *so;
480 int error;
481
482 if (proto)
483 prp = pffindproto(dom, proto, type);
484 else
485 prp = pffindtype(dom, type);
486
487 if (prp == NULL) {
488 /* No support for domain. */
489 if (pffinddomain(dom) == NULL)
490 return (EAFNOSUPPORT);
491 /* No support for socket type. */
492 if (proto == 0 && type != 0)
493 return (EPROTOTYPE);
494 return (EPROTONOSUPPORT);
495 }
496 if (prp->pr_usrreqs->pru_attach == NULL ||
497 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
498 return (EPROTONOSUPPORT);
499
500 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
501 return (EPROTONOSUPPORT);
502
503 if (prp->pr_type != type)
504 return (EPROTOTYPE);
505 so = soalloc(CRED_TO_VNET(cred));
506 if (so == NULL)
507 return (ENOBUFS);
508
509 TAILQ_INIT(&so->so_incomp);
510 TAILQ_INIT(&so->so_comp);
511 so->so_type = type;
512 so->so_cred = crhold(cred);
513 if ((prp->pr_domain->dom_family == PF_INET) ||
514 (prp->pr_domain->dom_family == PF_INET6) ||
515 (prp->pr_domain->dom_family == PF_ROUTE))
516 so->so_fibnum = td->td_proc->p_fibnum;
517 else
518 so->so_fibnum = 0;
519 so->so_proto = prp;
520 #ifdef MAC
521 mac_socket_create(cred, so);
522 #endif
523 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
524 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
525 so->so_count = 1;
526 /*
527 * Auto-sizing of socket buffers is managed by the protocols and
528 * the appropriate flags must be set in the pru_attach function.
529 */
530 CURVNET_SET(so->so_vnet);
531 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
532 CURVNET_RESTORE();
533 if (error) {
534 KASSERT(so->so_count == 1, ("socreate: so_count %d",
535 so->so_count));
536 so->so_count = 0;
537 sodealloc(so);
538 return (error);
539 }
540 *aso = so;
541 return (0);
542 }
543
544 #ifdef REGRESSION
545 static int regression_sonewconn_earlytest = 1;
546 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
547 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
548 #endif
549
550 /*
551 * When an attempt at a new connection is noted on a socket which accepts
552 * connections, sonewconn is called. If the connection is possible (subject
553 * to space constraints, etc.) then we allocate a new structure, properly
554 * linked into the data structure of the original socket, and return this.
555 * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
556 *
557 * Note: the ref count on the socket is 0 on return.
558 */
559 struct socket *
560 sonewconn(struct socket *head, int connstatus)
561 {
562 static struct timeval lastover;
563 static struct timeval overinterval = { 60, 0 };
564 static int overcount;
565
566 struct socket *so;
567 int over;
568
569 ACCEPT_LOCK();
570 over = (head->so_qlen > 3 * head->so_qlimit / 2);
571 ACCEPT_UNLOCK();
572 #ifdef REGRESSION
573 if (regression_sonewconn_earlytest && over) {
574 #else
575 if (over) {
576 #endif
577 overcount++;
578
579 if (ratecheck(&lastover, &overinterval)) {
580 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
581 "%i already in queue awaiting acceptance "
582 "(%d occurrences)\n",
583 __func__, head->so_pcb, head->so_qlen, overcount);
584
585 overcount = 0;
586 }
587
588 return (NULL);
589 }
590 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
591 __func__, __LINE__, head));
592 so = soalloc(head->so_vnet);
593 if (so == NULL) {
594 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
595 "limit reached or out of memory\n",
596 __func__, head->so_pcb);
597 return (NULL);
598 }
599 if ((head->so_options & SO_ACCEPTFILTER) != 0)
600 connstatus = 0;
601 so->so_head = head;
602 so->so_type = head->so_type;
603 so->so_options = head->so_options &~ SO_ACCEPTCONN;
604 so->so_linger = head->so_linger;
605 so->so_state = head->so_state | SS_NOFDREF;
606 so->so_fibnum = head->so_fibnum;
607 so->so_proto = head->so_proto;
608 so->so_cred = crhold(head->so_cred);
609 #ifdef MAC
610 mac_socket_newconn(head, so);
611 #endif
612 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
613 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
614 VNET_SO_ASSERT(head);
615 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
616 sodealloc(so);
617 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
618 __func__, head->so_pcb);
619 return (NULL);
620 }
621 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
622 sodealloc(so);
623 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
624 __func__, head->so_pcb);
625 return (NULL);
626 }
627 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
628 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
629 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
630 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
631 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
632 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
633 so->so_state |= connstatus;
634 ACCEPT_LOCK();
635 /*
636 * The accept socket may be tearing down but we just
637 * won a race on the ACCEPT_LOCK.
638 * However, if sctp_peeloff() is called on a 1-to-many
639 * style socket, the SO_ACCEPTCONN doesn't need to be set.
640 */
641 if (!(head->so_options & SO_ACCEPTCONN) &&
642 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
643 (head->so_type != SOCK_SEQPACKET))) {
644 SOCK_LOCK(so);
645 so->so_head = NULL;
646 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
647 return (NULL);
648 }
649 if (connstatus) {
650 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
651 so->so_qstate |= SQ_COMP;
652 head->so_qlen++;
653 } else {
654 /*
655 * Keep removing sockets from the head until there's room for
656 * us to insert on the tail. In pre-locking revisions, this
657 * was a simple if(), but as we could be racing with other
658 * threads and soabort() requires dropping locks, we must
659 * loop waiting for the condition to be true.
660 */
661 while (head->so_incqlen > head->so_qlimit) {
662 struct socket *sp;
663 sp = TAILQ_FIRST(&head->so_incomp);
664 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
665 head->so_incqlen--;
666 sp->so_qstate &= ~SQ_INCOMP;
667 sp->so_head = NULL;
668 ACCEPT_UNLOCK();
669 soabort(sp);
670 ACCEPT_LOCK();
671 }
672 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
673 so->so_qstate |= SQ_INCOMP;
674 head->so_incqlen++;
675 }
676 ACCEPT_UNLOCK();
677 if (connstatus) {
678 sorwakeup(head);
679 wakeup_one(&head->so_timeo);
680 }
681 return (so);
682 }
683
684 int
685 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
686 {
687 int error;
688
689 CURVNET_SET(so->so_vnet);
690 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
691 CURVNET_RESTORE();
692 return (error);
693 }
694
695 int
696 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
697 {
698 int error;
699
700 CURVNET_SET(so->so_vnet);
701 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
702 CURVNET_RESTORE();
703 return (error);
704 }
705
706 /*
707 * solisten() transitions a socket from a non-listening state to a listening
708 * state, but can also be used to update the listen queue depth on an
709 * existing listen socket. The protocol will call back into the sockets
710 * layer using solisten_proto_check() and solisten_proto() to check and set
711 * socket-layer listen state. Call backs are used so that the protocol can
712 * acquire both protocol and socket layer locks in whatever order is required
713 * by the protocol.
714 *
715 * Protocol implementors are advised to hold the socket lock across the
716 * socket-layer test and set to avoid races at the socket layer.
717 */
718 int
719 solisten(struct socket *so, int backlog, struct thread *td)
720 {
721 int error;
722
723 CURVNET_SET(so->so_vnet);
724 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
725 CURVNET_RESTORE();
726 return (error);
727 }
728
729 int
730 solisten_proto_check(struct socket *so)
731 {
732
733 SOCK_LOCK_ASSERT(so);
734
735 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
736 SS_ISDISCONNECTING))
737 return (EINVAL);
738 return (0);
739 }
740
741 void
742 solisten_proto(struct socket *so, int backlog)
743 {
744
745 SOCK_LOCK_ASSERT(so);
746
747 if (backlog < 0 || backlog > somaxconn)
748 backlog = somaxconn;
749 so->so_qlimit = backlog;
750 so->so_options |= SO_ACCEPTCONN;
751 }
752
753 /*
754 * Evaluate the reference count and named references on a socket; if no
755 * references remain, free it. This should be called whenever a reference is
756 * released, such as in sorele(), but also when named reference flags are
757 * cleared in socket or protocol code.
758 *
759 * sofree() will free the socket if:
760 *
761 * - There are no outstanding file descriptor references or related consumers
762 * (so_count == 0).
763 *
764 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
765 *
766 * - The protocol does not have an outstanding strong reference on the socket
767 * (SS_PROTOREF).
768 *
769 * - The socket is not in a completed connection queue, so a process has been
770 * notified that it is present. If it is removed, the user process may
771 * block in accept() despite select() saying the socket was ready.
772 */
773 void
774 sofree(struct socket *so)
775 {
776 struct protosw *pr = so->so_proto;
777 struct socket *head;
778
779 ACCEPT_LOCK_ASSERT();
780 SOCK_LOCK_ASSERT(so);
781
782 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
783 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
784 SOCK_UNLOCK(so);
785 ACCEPT_UNLOCK();
786 return;
787 }
788
789 head = so->so_head;
790 if (head != NULL) {
791 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
792 (so->so_qstate & SQ_INCOMP) != 0,
793 ("sofree: so_head != NULL, but neither SQ_COMP nor "
794 "SQ_INCOMP"));
795 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
796 (so->so_qstate & SQ_INCOMP) == 0,
797 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
798 TAILQ_REMOVE(&head->so_incomp, so, so_list);
799 head->so_incqlen--;
800 so->so_qstate &= ~SQ_INCOMP;
801 so->so_head = NULL;
802 }
803 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
804 (so->so_qstate & SQ_INCOMP) == 0,
805 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
806 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
807 if (so->so_options & SO_ACCEPTCONN) {
808 KASSERT((TAILQ_EMPTY(&so->so_comp)),
809 ("sofree: so_comp populated"));
810 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
811 ("sofree: so_incomp populated"));
812 }
813 SOCK_UNLOCK(so);
814 ACCEPT_UNLOCK();
815
816 VNET_SO_ASSERT(so);
817 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
818 (*pr->pr_domain->dom_dispose)(so);
819 if (pr->pr_usrreqs->pru_detach != NULL)
820 (*pr->pr_usrreqs->pru_detach)(so);
821
822 /*
823 * From this point on, we assume that no other references to this
824 * socket exist anywhere else in the stack. Therefore, no locks need
825 * to be acquired or held.
826 *
827 * We used to do a lot of socket buffer and socket locking here, as
828 * well as invoke sorflush() and perform wakeups. The direct call to
829 * dom_dispose() and sbrelease_internal() are an inlining of what was
830 * necessary from sorflush().
831 *
832 * Notice that the socket buffer and kqueue state are torn down
833 * before calling pru_detach. This means that protocols shold not
834 * assume they can perform socket wakeups, etc, in their detach code.
835 */
836 sbdestroy(&so->so_snd, so);
837 sbdestroy(&so->so_rcv, so);
838 seldrain(&so->so_snd.sb_sel);
839 seldrain(&so->so_rcv.sb_sel);
840 knlist_destroy(&so->so_rcv.sb_sel.si_note);
841 knlist_destroy(&so->so_snd.sb_sel.si_note);
842 sodealloc(so);
843 }
844
845 /*
846 * Close a socket on last file table reference removal. Initiate disconnect
847 * if connected. Free socket when disconnect complete.
848 *
849 * This function will sorele() the socket. Note that soclose() may be called
850 * prior to the ref count reaching zero. The actual socket structure will
851 * not be freed until the ref count reaches zero.
852 */
853 int
854 soclose(struct socket *so)
855 {
856 int error = 0;
857
858 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
859
860 CURVNET_SET(so->so_vnet);
861 funsetown(&so->so_sigio);
862 if (so->so_state & SS_ISCONNECTED) {
863 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
864 error = sodisconnect(so);
865 if (error) {
866 if (error == ENOTCONN)
867 error = 0;
868 goto drop;
869 }
870 }
871 if (so->so_options & SO_LINGER) {
872 if ((so->so_state & SS_ISDISCONNECTING) &&
873 (so->so_state & SS_NBIO))
874 goto drop;
875 while (so->so_state & SS_ISCONNECTED) {
876 error = tsleep(&so->so_timeo,
877 PSOCK | PCATCH, "soclos",
878 so->so_linger * hz);
879 if (error)
880 break;
881 }
882 }
883 }
884
885 drop:
886 if (so->so_proto->pr_usrreqs->pru_close != NULL)
887 (*so->so_proto->pr_usrreqs->pru_close)(so);
888 ACCEPT_LOCK();
889 if (so->so_options & SO_ACCEPTCONN) {
890 struct socket *sp;
891 /*
892 * Prevent new additions to the accept queues due
893 * to ACCEPT_LOCK races while we are draining them.
894 */
895 so->so_options &= ~SO_ACCEPTCONN;
896 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
897 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
898 so->so_incqlen--;
899 sp->so_qstate &= ~SQ_INCOMP;
900 sp->so_head = NULL;
901 ACCEPT_UNLOCK();
902 soabort(sp);
903 ACCEPT_LOCK();
904 }
905 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
906 TAILQ_REMOVE(&so->so_comp, sp, so_list);
907 so->so_qlen--;
908 sp->so_qstate &= ~SQ_COMP;
909 sp->so_head = NULL;
910 ACCEPT_UNLOCK();
911 soabort(sp);
912 ACCEPT_LOCK();
913 }
914 KASSERT((TAILQ_EMPTY(&so->so_comp)),
915 ("%s: so_comp populated", __func__));
916 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
917 ("%s: so_incomp populated", __func__));
918 }
919 SOCK_LOCK(so);
920 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
921 so->so_state |= SS_NOFDREF;
922 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
923 CURVNET_RESTORE();
924 return (error);
925 }
926
927 /*
928 * soabort() is used to abruptly tear down a connection, such as when a
929 * resource limit is reached (listen queue depth exceeded), or if a listen
930 * socket is closed while there are sockets waiting to be accepted.
931 *
932 * This interface is tricky, because it is called on an unreferenced socket,
933 * and must be called only by a thread that has actually removed the socket
934 * from the listen queue it was on, or races with other threads are risked.
935 *
936 * This interface will call into the protocol code, so must not be called
937 * with any socket locks held. Protocols do call it while holding their own
938 * recursible protocol mutexes, but this is something that should be subject
939 * to review in the future.
940 */
941 void
942 soabort(struct socket *so)
943 {
944
945 /*
946 * In as much as is possible, assert that no references to this
947 * socket are held. This is not quite the same as asserting that the
948 * current thread is responsible for arranging for no references, but
949 * is as close as we can get for now.
950 */
951 KASSERT(so->so_count == 0, ("soabort: so_count"));
952 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
953 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
954 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
955 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
956 VNET_SO_ASSERT(so);
957
958 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
959 (*so->so_proto->pr_usrreqs->pru_abort)(so);
960 ACCEPT_LOCK();
961 SOCK_LOCK(so);
962 sofree(so);
963 }
964
965 int
966 soaccept(struct socket *so, struct sockaddr **nam)
967 {
968 int error;
969
970 SOCK_LOCK(so);
971 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
972 so->so_state &= ~SS_NOFDREF;
973 SOCK_UNLOCK(so);
974
975 CURVNET_SET(so->so_vnet);
976 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
977 CURVNET_RESTORE();
978 return (error);
979 }
980
981 int
982 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
983 {
984
985 return (soconnectat(AT_FDCWD, so, nam, td));
986 }
987
988 int
989 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
990 {
991 int error;
992
993 if (so->so_options & SO_ACCEPTCONN)
994 return (EOPNOTSUPP);
995
996 CURVNET_SET(so->so_vnet);
997 /*
998 * If protocol is connection-based, can only connect once.
999 * Otherwise, if connected, try to disconnect first. This allows
1000 * user to disconnect by connecting to, e.g., a null address.
1001 */
1002 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1003 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1004 (error = sodisconnect(so)))) {
1005 error = EISCONN;
1006 } else {
1007 /*
1008 * Prevent accumulated error from previous connection from
1009 * biting us.
1010 */
1011 so->so_error = 0;
1012 if (fd == AT_FDCWD) {
1013 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1014 nam, td);
1015 } else {
1016 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1017 so, nam, td);
1018 }
1019 }
1020 CURVNET_RESTORE();
1021
1022 return (error);
1023 }
1024
1025 int
1026 soconnect2(struct socket *so1, struct socket *so2)
1027 {
1028 int error;
1029
1030 CURVNET_SET(so1->so_vnet);
1031 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1032 CURVNET_RESTORE();
1033 return (error);
1034 }
1035
1036 int
1037 sodisconnect(struct socket *so)
1038 {
1039 int error;
1040
1041 if ((so->so_state & SS_ISCONNECTED) == 0)
1042 return (ENOTCONN);
1043 if (so->so_state & SS_ISDISCONNECTING)
1044 return (EALREADY);
1045 VNET_SO_ASSERT(so);
1046 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1047 return (error);
1048 }
1049
1050 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1051
1052 int
1053 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1054 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1055 {
1056 long space;
1057 ssize_t resid;
1058 int clen = 0, error, dontroute;
1059
1060 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1061 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1062 ("sosend_dgram: !PR_ATOMIC"));
1063
1064 if (uio != NULL)
1065 resid = uio->uio_resid;
1066 else
1067 resid = top->m_pkthdr.len;
1068 /*
1069 * In theory resid should be unsigned. However, space must be
1070 * signed, as it might be less than 0 if we over-committed, and we
1071 * must use a signed comparison of space and resid. On the other
1072 * hand, a negative resid causes us to loop sending 0-length
1073 * segments to the protocol.
1074 */
1075 if (resid < 0) {
1076 error = EINVAL;
1077 goto out;
1078 }
1079
1080 dontroute =
1081 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1082 if (td != NULL)
1083 td->td_ru.ru_msgsnd++;
1084 if (control != NULL)
1085 clen = control->m_len;
1086
1087 SOCKBUF_LOCK(&so->so_snd);
1088 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1089 SOCKBUF_UNLOCK(&so->so_snd);
1090 error = EPIPE;
1091 goto out;
1092 }
1093 if (so->so_error) {
1094 error = so->so_error;
1095 so->so_error = 0;
1096 SOCKBUF_UNLOCK(&so->so_snd);
1097 goto out;
1098 }
1099 if ((so->so_state & SS_ISCONNECTED) == 0) {
1100 /*
1101 * `sendto' and `sendmsg' is allowed on a connection-based
1102 * socket if it supports implied connect. Return ENOTCONN if
1103 * not connected and no address is supplied.
1104 */
1105 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1106 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1107 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1108 !(resid == 0 && clen != 0)) {
1109 SOCKBUF_UNLOCK(&so->so_snd);
1110 error = ENOTCONN;
1111 goto out;
1112 }
1113 } else if (addr == NULL) {
1114 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1115 error = ENOTCONN;
1116 else
1117 error = EDESTADDRREQ;
1118 SOCKBUF_UNLOCK(&so->so_snd);
1119 goto out;
1120 }
1121 }
1122
1123 /*
1124 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1125 * problem and need fixing.
1126 */
1127 space = sbspace(&so->so_snd);
1128 if (flags & MSG_OOB)
1129 space += 1024;
1130 space -= clen;
1131 SOCKBUF_UNLOCK(&so->so_snd);
1132 if (resid > space) {
1133 error = EMSGSIZE;
1134 goto out;
1135 }
1136 if (uio == NULL) {
1137 resid = 0;
1138 if (flags & MSG_EOR)
1139 top->m_flags |= M_EOR;
1140 } else {
1141 /*
1142 * Copy the data from userland into a mbuf chain.
1143 * If no data is to be copied in, a single empty mbuf
1144 * is returned.
1145 */
1146 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1147 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1148 if (top == NULL) {
1149 error = EFAULT; /* only possible error */
1150 goto out;
1151 }
1152 space -= resid - uio->uio_resid;
1153 resid = uio->uio_resid;
1154 }
1155 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1156 /*
1157 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1158 * than with.
1159 */
1160 if (dontroute) {
1161 SOCK_LOCK(so);
1162 so->so_options |= SO_DONTROUTE;
1163 SOCK_UNLOCK(so);
1164 }
1165 /*
1166 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1167 * of date. We could have received a reset packet in an interrupt or
1168 * maybe we slept while doing page faults in uiomove() etc. We could
1169 * probably recheck again inside the locking protection here, but
1170 * there are probably other places that this also happens. We must
1171 * rethink this.
1172 */
1173 VNET_SO_ASSERT(so);
1174 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1175 (flags & MSG_OOB) ? PRUS_OOB :
1176 /*
1177 * If the user set MSG_EOF, the protocol understands this flag and
1178 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1179 */
1180 ((flags & MSG_EOF) &&
1181 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1182 (resid <= 0)) ?
1183 PRUS_EOF :
1184 /* If there is more to send set PRUS_MORETOCOME */
1185 (flags & MSG_MORETOCOME) ||
1186 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1187 top, addr, control, td);
1188 if (dontroute) {
1189 SOCK_LOCK(so);
1190 so->so_options &= ~SO_DONTROUTE;
1191 SOCK_UNLOCK(so);
1192 }
1193 clen = 0;
1194 control = NULL;
1195 top = NULL;
1196 out:
1197 if (top != NULL)
1198 m_freem(top);
1199 if (control != NULL)
1200 m_freem(control);
1201 return (error);
1202 }
1203
1204 /*
1205 * Send on a socket. If send must go all at once and message is larger than
1206 * send buffering, then hard error. Lock against other senders. If must go
1207 * all at once and not enough room now, then inform user that this would
1208 * block and do nothing. Otherwise, if nonblocking, send as much as
1209 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1210 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1211 * in mbuf chain must be small enough to send all at once.
1212 *
1213 * Returns nonzero on error, timeout or signal; callers must check for short
1214 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1215 * on return.
1216 */
1217 int
1218 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1219 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1220 {
1221 long space;
1222 ssize_t resid;
1223 int clen = 0, error, dontroute;
1224 int atomic = sosendallatonce(so) || top;
1225
1226 if (uio != NULL)
1227 resid = uio->uio_resid;
1228 else
1229 resid = top->m_pkthdr.len;
1230 /*
1231 * In theory resid should be unsigned. However, space must be
1232 * signed, as it might be less than 0 if we over-committed, and we
1233 * must use a signed comparison of space and resid. On the other
1234 * hand, a negative resid causes us to loop sending 0-length
1235 * segments to the protocol.
1236 *
1237 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1238 * type sockets since that's an error.
1239 */
1240 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1241 error = EINVAL;
1242 goto out;
1243 }
1244
1245 dontroute =
1246 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1247 (so->so_proto->pr_flags & PR_ATOMIC);
1248 if (td != NULL)
1249 td->td_ru.ru_msgsnd++;
1250 if (control != NULL)
1251 clen = control->m_len;
1252
1253 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1254 if (error)
1255 goto out;
1256
1257 restart:
1258 do {
1259 SOCKBUF_LOCK(&so->so_snd);
1260 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1261 SOCKBUF_UNLOCK(&so->so_snd);
1262 error = EPIPE;
1263 goto release;
1264 }
1265 if (so->so_error) {
1266 error = so->so_error;
1267 so->so_error = 0;
1268 SOCKBUF_UNLOCK(&so->so_snd);
1269 goto release;
1270 }
1271 if ((so->so_state & SS_ISCONNECTED) == 0) {
1272 /*
1273 * `sendto' and `sendmsg' is allowed on a connection-
1274 * based socket if it supports implied connect.
1275 * Return ENOTCONN if not connected and no address is
1276 * supplied.
1277 */
1278 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1279 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1280 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1281 !(resid == 0 && clen != 0)) {
1282 SOCKBUF_UNLOCK(&so->so_snd);
1283 error = ENOTCONN;
1284 goto release;
1285 }
1286 } else if (addr == NULL) {
1287 SOCKBUF_UNLOCK(&so->so_snd);
1288 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1289 error = ENOTCONN;
1290 else
1291 error = EDESTADDRREQ;
1292 goto release;
1293 }
1294 }
1295 space = sbspace(&so->so_snd);
1296 if (flags & MSG_OOB)
1297 space += 1024;
1298 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1299 clen > so->so_snd.sb_hiwat) {
1300 SOCKBUF_UNLOCK(&so->so_snd);
1301 error = EMSGSIZE;
1302 goto release;
1303 }
1304 if (space < resid + clen &&
1305 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1306 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1307 SOCKBUF_UNLOCK(&so->so_snd);
1308 error = EWOULDBLOCK;
1309 goto release;
1310 }
1311 error = sbwait(&so->so_snd);
1312 SOCKBUF_UNLOCK(&so->so_snd);
1313 if (error)
1314 goto release;
1315 goto restart;
1316 }
1317 SOCKBUF_UNLOCK(&so->so_snd);
1318 space -= clen;
1319 do {
1320 if (uio == NULL) {
1321 resid = 0;
1322 if (flags & MSG_EOR)
1323 top->m_flags |= M_EOR;
1324 } else {
1325 /*
1326 * Copy the data from userland into a mbuf
1327 * chain. If resid is 0, which can happen
1328 * only if we have control to send, then
1329 * a single empty mbuf is returned. This
1330 * is a workaround to prevent protocol send
1331 * methods to panic.
1332 */
1333 top = m_uiotombuf(uio, M_WAITOK, space,
1334 (atomic ? max_hdr : 0),
1335 (atomic ? M_PKTHDR : 0) |
1336 ((flags & MSG_EOR) ? M_EOR : 0));
1337 if (top == NULL) {
1338 error = EFAULT; /* only possible error */
1339 goto release;
1340 }
1341 space -= resid - uio->uio_resid;
1342 resid = uio->uio_resid;
1343 }
1344 if (dontroute) {
1345 SOCK_LOCK(so);
1346 so->so_options |= SO_DONTROUTE;
1347 SOCK_UNLOCK(so);
1348 }
1349 /*
1350 * XXX all the SBS_CANTSENDMORE checks previously
1351 * done could be out of date. We could have received
1352 * a reset packet in an interrupt or maybe we slept
1353 * while doing page faults in uiomove() etc. We
1354 * could probably recheck again inside the locking
1355 * protection here, but there are probably other
1356 * places that this also happens. We must rethink
1357 * this.
1358 */
1359 VNET_SO_ASSERT(so);
1360 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1361 (flags & MSG_OOB) ? PRUS_OOB :
1362 /*
1363 * If the user set MSG_EOF, the protocol understands
1364 * this flag and nothing left to send then use
1365 * PRU_SEND_EOF instead of PRU_SEND.
1366 */
1367 ((flags & MSG_EOF) &&
1368 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1369 (resid <= 0)) ?
1370 PRUS_EOF :
1371 /* If there is more to send set PRUS_MORETOCOME. */
1372 (flags & MSG_MORETOCOME) ||
1373 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1374 top, addr, control, td);
1375 if (dontroute) {
1376 SOCK_LOCK(so);
1377 so->so_options &= ~SO_DONTROUTE;
1378 SOCK_UNLOCK(so);
1379 }
1380 clen = 0;
1381 control = NULL;
1382 top = NULL;
1383 if (error)
1384 goto release;
1385 } while (resid && space > 0);
1386 } while (resid);
1387
1388 release:
1389 sbunlock(&so->so_snd);
1390 out:
1391 if (top != NULL)
1392 m_freem(top);
1393 if (control != NULL)
1394 m_freem(control);
1395 return (error);
1396 }
1397
1398 int
1399 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1400 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1401 {
1402 int error;
1403
1404 CURVNET_SET(so->so_vnet);
1405 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1406 control, flags, td);
1407 CURVNET_RESTORE();
1408 return (error);
1409 }
1410
1411 /*
1412 * The part of soreceive() that implements reading non-inline out-of-band
1413 * data from a socket. For more complete comments, see soreceive(), from
1414 * which this code originated.
1415 *
1416 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1417 * unable to return an mbuf chain to the caller.
1418 */
1419 static int
1420 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1421 {
1422 struct protosw *pr = so->so_proto;
1423 struct mbuf *m;
1424 int error;
1425
1426 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1427 VNET_SO_ASSERT(so);
1428
1429 m = m_get(M_WAITOK, MT_DATA);
1430 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1431 if (error)
1432 goto bad;
1433 do {
1434 error = uiomove(mtod(m, void *),
1435 (int) min(uio->uio_resid, m->m_len), uio);
1436 m = m_free(m);
1437 } while (uio->uio_resid && error == 0 && m);
1438 bad:
1439 if (m != NULL)
1440 m_freem(m);
1441 return (error);
1442 }
1443
1444 /*
1445 * Following replacement or removal of the first mbuf on the first mbuf chain
1446 * of a socket buffer, push necessary state changes back into the socket
1447 * buffer so that other consumers see the values consistently. 'nextrecord'
1448 * is the callers locally stored value of the original value of
1449 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1450 * NOTE: 'nextrecord' may be NULL.
1451 */
1452 static __inline void
1453 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1454 {
1455
1456 SOCKBUF_LOCK_ASSERT(sb);
1457 /*
1458 * First, update for the new value of nextrecord. If necessary, make
1459 * it the first record.
1460 */
1461 if (sb->sb_mb != NULL)
1462 sb->sb_mb->m_nextpkt = nextrecord;
1463 else
1464 sb->sb_mb = nextrecord;
1465
1466 /*
1467 * Now update any dependent socket buffer fields to reflect the new
1468 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1469 * addition of a second clause that takes care of the case where
1470 * sb_mb has been updated, but remains the last record.
1471 */
1472 if (sb->sb_mb == NULL) {
1473 sb->sb_mbtail = NULL;
1474 sb->sb_lastrecord = NULL;
1475 } else if (sb->sb_mb->m_nextpkt == NULL)
1476 sb->sb_lastrecord = sb->sb_mb;
1477 }
1478
1479 /*
1480 * Implement receive operations on a socket. We depend on the way that
1481 * records are added to the sockbuf by sbappend. In particular, each record
1482 * (mbufs linked through m_next) must begin with an address if the protocol
1483 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1484 * data, and then zero or more mbufs of data. In order to allow parallelism
1485 * between network receive and copying to user space, as well as avoid
1486 * sleeping with a mutex held, we release the socket buffer mutex during the
1487 * user space copy. Although the sockbuf is locked, new data may still be
1488 * appended, and thus we must maintain consistency of the sockbuf during that
1489 * time.
1490 *
1491 * The caller may receive the data as a single mbuf chain by supplying an
1492 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1493 * the count in uio_resid.
1494 */
1495 int
1496 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1497 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1498 {
1499 struct mbuf *m, **mp;
1500 int flags, error, offset;
1501 ssize_t len;
1502 struct protosw *pr = so->so_proto;
1503 struct mbuf *nextrecord;
1504 int moff, type = 0;
1505 ssize_t orig_resid = uio->uio_resid;
1506
1507 mp = mp0;
1508 if (psa != NULL)
1509 *psa = NULL;
1510 if (controlp != NULL)
1511 *controlp = NULL;
1512 if (flagsp != NULL)
1513 flags = *flagsp &~ MSG_EOR;
1514 else
1515 flags = 0;
1516 if (flags & MSG_OOB)
1517 return (soreceive_rcvoob(so, uio, flags));
1518 if (mp != NULL)
1519 *mp = NULL;
1520 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1521 && uio->uio_resid) {
1522 VNET_SO_ASSERT(so);
1523 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1524 }
1525
1526 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1527 if (error)
1528 return (error);
1529
1530 restart:
1531 SOCKBUF_LOCK(&so->so_rcv);
1532 m = so->so_rcv.sb_mb;
1533 /*
1534 * If we have less data than requested, block awaiting more (subject
1535 * to any timeout) if:
1536 * 1. the current count is less than the low water mark, or
1537 * 2. MSG_DONTWAIT is not set
1538 */
1539 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1540 sbavail(&so->so_rcv) < uio->uio_resid) &&
1541 sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1542 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1543 KASSERT(m != NULL || !sbavail(&so->so_rcv),
1544 ("receive: m == %p sbavail == %u",
1545 m, sbavail(&so->so_rcv)));
1546 if (so->so_error) {
1547 if (m != NULL)
1548 goto dontblock;
1549 error = so->so_error;
1550 if ((flags & MSG_PEEK) == 0)
1551 so->so_error = 0;
1552 SOCKBUF_UNLOCK(&so->so_rcv);
1553 goto release;
1554 }
1555 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1556 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1557 if (m == NULL) {
1558 SOCKBUF_UNLOCK(&so->so_rcv);
1559 goto release;
1560 } else
1561 goto dontblock;
1562 }
1563 for (; m != NULL; m = m->m_next)
1564 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1565 m = so->so_rcv.sb_mb;
1566 goto dontblock;
1567 }
1568 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1569 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1570 SOCKBUF_UNLOCK(&so->so_rcv);
1571 error = ENOTCONN;
1572 goto release;
1573 }
1574 if (uio->uio_resid == 0) {
1575 SOCKBUF_UNLOCK(&so->so_rcv);
1576 goto release;
1577 }
1578 if ((so->so_state & SS_NBIO) ||
1579 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1580 SOCKBUF_UNLOCK(&so->so_rcv);
1581 error = EWOULDBLOCK;
1582 goto release;
1583 }
1584 SBLASTRECORDCHK(&so->so_rcv);
1585 SBLASTMBUFCHK(&so->so_rcv);
1586 error = sbwait(&so->so_rcv);
1587 SOCKBUF_UNLOCK(&so->so_rcv);
1588 if (error)
1589 goto release;
1590 goto restart;
1591 }
1592 dontblock:
1593 /*
1594 * From this point onward, we maintain 'nextrecord' as a cache of the
1595 * pointer to the next record in the socket buffer. We must keep the
1596 * various socket buffer pointers and local stack versions of the
1597 * pointers in sync, pushing out modifications before dropping the
1598 * socket buffer mutex, and re-reading them when picking it up.
1599 *
1600 * Otherwise, we will race with the network stack appending new data
1601 * or records onto the socket buffer by using inconsistent/stale
1602 * versions of the field, possibly resulting in socket buffer
1603 * corruption.
1604 *
1605 * By holding the high-level sblock(), we prevent simultaneous
1606 * readers from pulling off the front of the socket buffer.
1607 */
1608 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1609 if (uio->uio_td)
1610 uio->uio_td->td_ru.ru_msgrcv++;
1611 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1612 SBLASTRECORDCHK(&so->so_rcv);
1613 SBLASTMBUFCHK(&so->so_rcv);
1614 nextrecord = m->m_nextpkt;
1615 if (pr->pr_flags & PR_ADDR) {
1616 KASSERT(m->m_type == MT_SONAME,
1617 ("m->m_type == %d", m->m_type));
1618 orig_resid = 0;
1619 if (psa != NULL)
1620 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1621 M_NOWAIT);
1622 if (flags & MSG_PEEK) {
1623 m = m->m_next;
1624 } else {
1625 sbfree(&so->so_rcv, m);
1626 so->so_rcv.sb_mb = m_free(m);
1627 m = so->so_rcv.sb_mb;
1628 sockbuf_pushsync(&so->so_rcv, nextrecord);
1629 }
1630 }
1631
1632 /*
1633 * Process one or more MT_CONTROL mbufs present before any data mbufs
1634 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1635 * just copy the data; if !MSG_PEEK, we call into the protocol to
1636 * perform externalization (or freeing if controlp == NULL).
1637 */
1638 if (m != NULL && m->m_type == MT_CONTROL) {
1639 struct mbuf *cm = NULL, *cmn;
1640 struct mbuf **cme = &cm;
1641
1642 do {
1643 if (flags & MSG_PEEK) {
1644 if (controlp != NULL) {
1645 *controlp = m_copy(m, 0, m->m_len);
1646 controlp = &(*controlp)->m_next;
1647 }
1648 m = m->m_next;
1649 } else {
1650 sbfree(&so->so_rcv, m);
1651 so->so_rcv.sb_mb = m->m_next;
1652 m->m_next = NULL;
1653 *cme = m;
1654 cme = &(*cme)->m_next;
1655 m = so->so_rcv.sb_mb;
1656 }
1657 } while (m != NULL && m->m_type == MT_CONTROL);
1658 if ((flags & MSG_PEEK) == 0)
1659 sockbuf_pushsync(&so->so_rcv, nextrecord);
1660 while (cm != NULL) {
1661 cmn = cm->m_next;
1662 cm->m_next = NULL;
1663 if (pr->pr_domain->dom_externalize != NULL) {
1664 SOCKBUF_UNLOCK(&so->so_rcv);
1665 VNET_SO_ASSERT(so);
1666 error = (*pr->pr_domain->dom_externalize)
1667 (cm, controlp, flags);
1668 SOCKBUF_LOCK(&so->so_rcv);
1669 } else if (controlp != NULL)
1670 *controlp = cm;
1671 else
1672 m_freem(cm);
1673 if (controlp != NULL) {
1674 orig_resid = 0;
1675 while (*controlp != NULL)
1676 controlp = &(*controlp)->m_next;
1677 }
1678 cm = cmn;
1679 }
1680 if (m != NULL)
1681 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1682 else
1683 nextrecord = so->so_rcv.sb_mb;
1684 orig_resid = 0;
1685 }
1686 if (m != NULL) {
1687 if ((flags & MSG_PEEK) == 0) {
1688 KASSERT(m->m_nextpkt == nextrecord,
1689 ("soreceive: post-control, nextrecord !sync"));
1690 if (nextrecord == NULL) {
1691 KASSERT(so->so_rcv.sb_mb == m,
1692 ("soreceive: post-control, sb_mb!=m"));
1693 KASSERT(so->so_rcv.sb_lastrecord == m,
1694 ("soreceive: post-control, lastrecord!=m"));
1695 }
1696 }
1697 type = m->m_type;
1698 if (type == MT_OOBDATA)
1699 flags |= MSG_OOB;
1700 } else {
1701 if ((flags & MSG_PEEK) == 0) {
1702 KASSERT(so->so_rcv.sb_mb == nextrecord,
1703 ("soreceive: sb_mb != nextrecord"));
1704 if (so->so_rcv.sb_mb == NULL) {
1705 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1706 ("soreceive: sb_lastercord != NULL"));
1707 }
1708 }
1709 }
1710 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1711 SBLASTRECORDCHK(&so->so_rcv);
1712 SBLASTMBUFCHK(&so->so_rcv);
1713
1714 /*
1715 * Now continue to read any data mbufs off of the head of the socket
1716 * buffer until the read request is satisfied. Note that 'type' is
1717 * used to store the type of any mbuf reads that have happened so far
1718 * such that soreceive() can stop reading if the type changes, which
1719 * causes soreceive() to return only one of regular data and inline
1720 * out-of-band data in a single socket receive operation.
1721 */
1722 moff = 0;
1723 offset = 0;
1724 while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1725 && error == 0) {
1726 /*
1727 * If the type of mbuf has changed since the last mbuf
1728 * examined ('type'), end the receive operation.
1729 */
1730 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1731 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1732 if (type != m->m_type)
1733 break;
1734 } else if (type == MT_OOBDATA)
1735 break;
1736 else
1737 KASSERT(m->m_type == MT_DATA,
1738 ("m->m_type == %d", m->m_type));
1739 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1740 len = uio->uio_resid;
1741 if (so->so_oobmark && len > so->so_oobmark - offset)
1742 len = so->so_oobmark - offset;
1743 if (len > m->m_len - moff)
1744 len = m->m_len - moff;
1745 /*
1746 * If mp is set, just pass back the mbufs. Otherwise copy
1747 * them out via the uio, then free. Sockbuf must be
1748 * consistent here (points to current mbuf, it points to next
1749 * record) when we drop priority; we must note any additions
1750 * to the sockbuf when we block interrupts again.
1751 */
1752 if (mp == NULL) {
1753 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1754 SBLASTRECORDCHK(&so->so_rcv);
1755 SBLASTMBUFCHK(&so->so_rcv);
1756 SOCKBUF_UNLOCK(&so->so_rcv);
1757 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1758 SOCKBUF_LOCK(&so->so_rcv);
1759 if (error) {
1760 /*
1761 * The MT_SONAME mbuf has already been removed
1762 * from the record, so it is necessary to
1763 * remove the data mbufs, if any, to preserve
1764 * the invariant in the case of PR_ADDR that
1765 * requires MT_SONAME mbufs at the head of
1766 * each record.
1767 */
1768 if (m && pr->pr_flags & PR_ATOMIC &&
1769 ((flags & MSG_PEEK) == 0))
1770 (void)sbdroprecord_locked(&so->so_rcv);
1771 SOCKBUF_UNLOCK(&so->so_rcv);
1772 goto release;
1773 }
1774 } else
1775 uio->uio_resid -= len;
1776 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1777 if (len == m->m_len - moff) {
1778 if (m->m_flags & M_EOR)
1779 flags |= MSG_EOR;
1780 if (flags & MSG_PEEK) {
1781 m = m->m_next;
1782 moff = 0;
1783 } else {
1784 nextrecord = m->m_nextpkt;
1785 sbfree(&so->so_rcv, m);
1786 if (mp != NULL) {
1787 m->m_nextpkt = NULL;
1788 *mp = m;
1789 mp = &m->m_next;
1790 so->so_rcv.sb_mb = m = m->m_next;
1791 *mp = NULL;
1792 } else {
1793 so->so_rcv.sb_mb = m_free(m);
1794 m = so->so_rcv.sb_mb;
1795 }
1796 sockbuf_pushsync(&so->so_rcv, nextrecord);
1797 SBLASTRECORDCHK(&so->so_rcv);
1798 SBLASTMBUFCHK(&so->so_rcv);
1799 }
1800 } else {
1801 if (flags & MSG_PEEK)
1802 moff += len;
1803 else {
1804 if (mp != NULL) {
1805 if (flags & MSG_DONTWAIT) {
1806 *mp = m_copym(m, 0, len,
1807 M_NOWAIT);
1808 if (*mp == NULL) {
1809 /*
1810 * m_copym() couldn't
1811 * allocate an mbuf.
1812 * Adjust uio_resid back
1813 * (it was adjusted
1814 * down by len bytes,
1815 * which we didn't end
1816 * up "copying" over).
1817 */
1818 uio->uio_resid += len;
1819 break;
1820 }
1821 } else {
1822 SOCKBUF_UNLOCK(&so->so_rcv);
1823 *mp = m_copym(m, 0, len,
1824 M_WAITOK);
1825 SOCKBUF_LOCK(&so->so_rcv);
1826 }
1827 }
1828 sbcut_locked(&so->so_rcv, len);
1829 }
1830 }
1831 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1832 if (so->so_oobmark) {
1833 if ((flags & MSG_PEEK) == 0) {
1834 so->so_oobmark -= len;
1835 if (so->so_oobmark == 0) {
1836 so->so_rcv.sb_state |= SBS_RCVATMARK;
1837 break;
1838 }
1839 } else {
1840 offset += len;
1841 if (offset == so->so_oobmark)
1842 break;
1843 }
1844 }
1845 if (flags & MSG_EOR)
1846 break;
1847 /*
1848 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1849 * must not quit until "uio->uio_resid == 0" or an error
1850 * termination. If a signal/timeout occurs, return with a
1851 * short count but without error. Keep sockbuf locked
1852 * against other readers.
1853 */
1854 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1855 !sosendallatonce(so) && nextrecord == NULL) {
1856 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1857 if (so->so_error ||
1858 so->so_rcv.sb_state & SBS_CANTRCVMORE)
1859 break;
1860 /*
1861 * Notify the protocol that some data has been
1862 * drained before blocking.
1863 */
1864 if (pr->pr_flags & PR_WANTRCVD) {
1865 SOCKBUF_UNLOCK(&so->so_rcv);
1866 VNET_SO_ASSERT(so);
1867 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1868 SOCKBUF_LOCK(&so->so_rcv);
1869 }
1870 SBLASTRECORDCHK(&so->so_rcv);
1871 SBLASTMBUFCHK(&so->so_rcv);
1872 /*
1873 * We could receive some data while was notifying
1874 * the protocol. Skip blocking in this case.
1875 */
1876 if (so->so_rcv.sb_mb == NULL) {
1877 error = sbwait(&so->so_rcv);
1878 if (error) {
1879 SOCKBUF_UNLOCK(&so->so_rcv);
1880 goto release;
1881 }
1882 }
1883 m = so->so_rcv.sb_mb;
1884 if (m != NULL)
1885 nextrecord = m->m_nextpkt;
1886 }
1887 }
1888
1889 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1890 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1891 flags |= MSG_TRUNC;
1892 if ((flags & MSG_PEEK) == 0)
1893 (void) sbdroprecord_locked(&so->so_rcv);
1894 }
1895 if ((flags & MSG_PEEK) == 0) {
1896 if (m == NULL) {
1897 /*
1898 * First part is an inline SB_EMPTY_FIXUP(). Second
1899 * part makes sure sb_lastrecord is up-to-date if
1900 * there is still data in the socket buffer.
1901 */
1902 so->so_rcv.sb_mb = nextrecord;
1903 if (so->so_rcv.sb_mb == NULL) {
1904 so->so_rcv.sb_mbtail = NULL;
1905 so->so_rcv.sb_lastrecord = NULL;
1906 } else if (nextrecord->m_nextpkt == NULL)
1907 so->so_rcv.sb_lastrecord = nextrecord;
1908 }
1909 SBLASTRECORDCHK(&so->so_rcv);
1910 SBLASTMBUFCHK(&so->so_rcv);
1911 /*
1912 * If soreceive() is being done from the socket callback,
1913 * then don't need to generate ACK to peer to update window,
1914 * since ACK will be generated on return to TCP.
1915 */
1916 if (!(flags & MSG_SOCALLBCK) &&
1917 (pr->pr_flags & PR_WANTRCVD)) {
1918 SOCKBUF_UNLOCK(&so->so_rcv);
1919 VNET_SO_ASSERT(so);
1920 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1921 SOCKBUF_LOCK(&so->so_rcv);
1922 }
1923 }
1924 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1925 if (orig_resid == uio->uio_resid && orig_resid &&
1926 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1927 SOCKBUF_UNLOCK(&so->so_rcv);
1928 goto restart;
1929 }
1930 SOCKBUF_UNLOCK(&so->so_rcv);
1931
1932 if (flagsp != NULL)
1933 *flagsp |= flags;
1934 release:
1935 sbunlock(&so->so_rcv);
1936 return (error);
1937 }
1938
1939 /*
1940 * Optimized version of soreceive() for stream (TCP) sockets.
1941 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1942 */
1943 int
1944 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1945 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1946 {
1947 int len = 0, error = 0, flags, oresid;
1948 struct sockbuf *sb;
1949 struct mbuf *m, *n = NULL;
1950
1951 /* We only do stream sockets. */
1952 if (so->so_type != SOCK_STREAM)
1953 return (EINVAL);
1954 if (psa != NULL)
1955 *psa = NULL;
1956 if (controlp != NULL)
1957 return (EINVAL);
1958 if (flagsp != NULL)
1959 flags = *flagsp &~ MSG_EOR;
1960 else
1961 flags = 0;
1962 if (flags & MSG_OOB)
1963 return (soreceive_rcvoob(so, uio, flags));
1964 if (mp0 != NULL)
1965 *mp0 = NULL;
1966
1967 sb = &so->so_rcv;
1968
1969 /* Prevent other readers from entering the socket. */
1970 error = sblock(sb, SBLOCKWAIT(flags));
1971 if (error)
1972 goto out;
1973 SOCKBUF_LOCK(sb);
1974
1975 /* Easy one, no space to copyout anything. */
1976 if (uio->uio_resid == 0) {
1977 error = EINVAL;
1978 goto out;
1979 }
1980 oresid = uio->uio_resid;
1981
1982 /* We will never ever get anything unless we are or were connected. */
1983 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1984 error = ENOTCONN;
1985 goto out;
1986 }
1987
1988 restart:
1989 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1990
1991 /* Abort if socket has reported problems. */
1992 if (so->so_error) {
1993 if (sbavail(sb) > 0)
1994 goto deliver;
1995 if (oresid > uio->uio_resid)
1996 goto out;
1997 error = so->so_error;
1998 if (!(flags & MSG_PEEK))
1999 so->so_error = 0;
2000 goto out;
2001 }
2002
2003 /* Door is closed. Deliver what is left, if any. */
2004 if (sb->sb_state & SBS_CANTRCVMORE) {
2005 if (sbavail(sb) > 0)
2006 goto deliver;
2007 else
2008 goto out;
2009 }
2010
2011 /* Socket buffer is empty and we shall not block. */
2012 if (sbavail(sb) == 0 &&
2013 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2014 error = EAGAIN;
2015 goto out;
2016 }
2017
2018 /* Socket buffer got some data that we shall deliver now. */
2019 if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2020 ((so->so_state & SS_NBIO) ||
2021 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2022 sbavail(sb) >= sb->sb_lowat ||
2023 sbavail(sb) >= uio->uio_resid ||
2024 sbavail(sb) >= sb->sb_hiwat) ) {
2025 goto deliver;
2026 }
2027
2028 /* On MSG_WAITALL we must wait until all data or error arrives. */
2029 if ((flags & MSG_WAITALL) &&
2030 (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2031 goto deliver;
2032
2033 /*
2034 * Wait and block until (more) data comes in.
2035 * NB: Drops the sockbuf lock during wait.
2036 */
2037 error = sbwait(sb);
2038 if (error)
2039 goto out;
2040 goto restart;
2041
2042 deliver:
2043 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2044 KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2045 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2046
2047 /* Statistics. */
2048 if (uio->uio_td)
2049 uio->uio_td->td_ru.ru_msgrcv++;
2050
2051 /* Fill uio until full or current end of socket buffer is reached. */
2052 len = min(uio->uio_resid, sbavail(sb));
2053 if (mp0 != NULL) {
2054 /* Dequeue as many mbufs as possible. */
2055 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2056 if (*mp0 == NULL)
2057 *mp0 = sb->sb_mb;
2058 else
2059 m_cat(*mp0, sb->sb_mb);
2060 for (m = sb->sb_mb;
2061 m != NULL && m->m_len <= len;
2062 m = m->m_next) {
2063 KASSERT(!(m->m_flags & M_NOTAVAIL),
2064 ("%s: m %p not available", __func__, m));
2065 len -= m->m_len;
2066 uio->uio_resid -= m->m_len;
2067 sbfree(sb, m);
2068 n = m;
2069 }
2070 n->m_next = NULL;
2071 sb->sb_mb = m;
2072 sb->sb_lastrecord = sb->sb_mb;
2073 if (sb->sb_mb == NULL)
2074 SB_EMPTY_FIXUP(sb);
2075 }
2076 /* Copy the remainder. */
2077 if (len > 0) {
2078 KASSERT(sb->sb_mb != NULL,
2079 ("%s: len > 0 && sb->sb_mb empty", __func__));
2080
2081 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2082 if (m == NULL)
2083 len = 0; /* Don't flush data from sockbuf. */
2084 else
2085 uio->uio_resid -= len;
2086 if (*mp0 != NULL)
2087 m_cat(*mp0, m);
2088 else
2089 *mp0 = m;
2090 if (*mp0 == NULL) {
2091 error = ENOBUFS;
2092 goto out;
2093 }
2094 }
2095 } else {
2096 /* NB: Must unlock socket buffer as uiomove may sleep. */
2097 SOCKBUF_UNLOCK(sb);
2098 error = m_mbuftouio(uio, sb->sb_mb, len);
2099 SOCKBUF_LOCK(sb);
2100 if (error)
2101 goto out;
2102 }
2103 SBLASTRECORDCHK(sb);
2104 SBLASTMBUFCHK(sb);
2105
2106 /*
2107 * Remove the delivered data from the socket buffer unless we
2108 * were only peeking.
2109 */
2110 if (!(flags & MSG_PEEK)) {
2111 if (len > 0)
2112 sbdrop_locked(sb, len);
2113
2114 /* Notify protocol that we drained some data. */
2115 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2116 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2117 !(flags & MSG_SOCALLBCK))) {
2118 SOCKBUF_UNLOCK(sb);
2119 VNET_SO_ASSERT(so);
2120 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2121 SOCKBUF_LOCK(sb);
2122 }
2123 }
2124
2125 /*
2126 * For MSG_WAITALL we may have to loop again and wait for
2127 * more data to come in.
2128 */
2129 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2130 goto restart;
2131 out:
2132 SOCKBUF_LOCK_ASSERT(sb);
2133 SBLASTRECORDCHK(sb);
2134 SBLASTMBUFCHK(sb);
2135 SOCKBUF_UNLOCK(sb);
2136 sbunlock(sb);
2137 return (error);
2138 }
2139
2140 /*
2141 * Optimized version of soreceive() for simple datagram cases from userspace.
2142 * Unlike in the stream case, we're able to drop a datagram if copyout()
2143 * fails, and because we handle datagrams atomically, we don't need to use a
2144 * sleep lock to prevent I/O interlacing.
2145 */
2146 int
2147 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2148 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2149 {
2150 struct mbuf *m, *m2;
2151 int flags, error;
2152 ssize_t len;
2153 struct protosw *pr = so->so_proto;
2154 struct mbuf *nextrecord;
2155
2156 if (psa != NULL)
2157 *psa = NULL;
2158 if (controlp != NULL)
2159 *controlp = NULL;
2160 if (flagsp != NULL)
2161 flags = *flagsp &~ MSG_EOR;
2162 else
2163 flags = 0;
2164
2165 /*
2166 * For any complicated cases, fall back to the full
2167 * soreceive_generic().
2168 */
2169 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2170 return (soreceive_generic(so, psa, uio, mp0, controlp,
2171 flagsp));
2172
2173 /*
2174 * Enforce restrictions on use.
2175 */
2176 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2177 ("soreceive_dgram: wantrcvd"));
2178 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2179 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2180 ("soreceive_dgram: SBS_RCVATMARK"));
2181 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2182 ("soreceive_dgram: P_CONNREQUIRED"));
2183
2184 /*
2185 * Loop blocking while waiting for a datagram.
2186 */
2187 SOCKBUF_LOCK(&so->so_rcv);
2188 while ((m = so->so_rcv.sb_mb) == NULL) {
2189 KASSERT(sbavail(&so->so_rcv) == 0,
2190 ("soreceive_dgram: sb_mb NULL but sbavail %u",
2191 sbavail(&so->so_rcv)));
2192 if (so->so_error) {
2193 error = so->so_error;
2194 so->so_error = 0;
2195 SOCKBUF_UNLOCK(&so->so_rcv);
2196 return (error);
2197 }
2198 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2199 uio->uio_resid == 0) {
2200 SOCKBUF_UNLOCK(&so->so_rcv);
2201 return (0);
2202 }
2203 if ((so->so_state & SS_NBIO) ||
2204 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2205 SOCKBUF_UNLOCK(&so->so_rcv);
2206 return (EWOULDBLOCK);
2207 }
2208 SBLASTRECORDCHK(&so->so_rcv);
2209 SBLASTMBUFCHK(&so->so_rcv);
2210 error = sbwait(&so->so_rcv);
2211 if (error) {
2212 SOCKBUF_UNLOCK(&so->so_rcv);
2213 return (error);
2214 }
2215 }
2216 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2217
2218 if (uio->uio_td)
2219 uio->uio_td->td_ru.ru_msgrcv++;
2220 SBLASTRECORDCHK(&so->so_rcv);
2221 SBLASTMBUFCHK(&so->so_rcv);
2222 nextrecord = m->m_nextpkt;
2223 if (nextrecord == NULL) {
2224 KASSERT(so->so_rcv.sb_lastrecord == m,
2225 ("soreceive_dgram: lastrecord != m"));
2226 }
2227
2228 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2229 ("soreceive_dgram: m_nextpkt != nextrecord"));
2230
2231 /*
2232 * Pull 'm' and its chain off the front of the packet queue.
2233 */
2234 so->so_rcv.sb_mb = NULL;
2235 sockbuf_pushsync(&so->so_rcv, nextrecord);
2236
2237 /*
2238 * Walk 'm's chain and free that many bytes from the socket buffer.
2239 */
2240 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2241 sbfree(&so->so_rcv, m2);
2242
2243 /*
2244 * Do a few last checks before we let go of the lock.
2245 */
2246 SBLASTRECORDCHK(&so->so_rcv);
2247 SBLASTMBUFCHK(&so->so_rcv);
2248 SOCKBUF_UNLOCK(&so->so_rcv);
2249
2250 if (pr->pr_flags & PR_ADDR) {
2251 KASSERT(m->m_type == MT_SONAME,
2252 ("m->m_type == %d", m->m_type));
2253 if (psa != NULL)
2254 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2255 M_NOWAIT);
2256 m = m_free(m);
2257 }
2258 if (m == NULL) {
2259 /* XXXRW: Can this happen? */
2260 return (0);
2261 }
2262
2263 /*
2264 * Packet to copyout() is now in 'm' and it is disconnected from the
2265 * queue.
2266 *
2267 * Process one or more MT_CONTROL mbufs present before any data mbufs
2268 * in the first mbuf chain on the socket buffer. We call into the
2269 * protocol to perform externalization (or freeing if controlp ==
2270 * NULL). In some cases there can be only MT_CONTROL mbufs without
2271 * MT_DATA mbufs.
2272 */
2273 if (m->m_type == MT_CONTROL) {
2274 struct mbuf *cm = NULL, *cmn;
2275 struct mbuf **cme = &cm;
2276
2277 do {
2278 m2 = m->m_next;
2279 m->m_next = NULL;
2280 *cme = m;
2281 cme = &(*cme)->m_next;
2282 m = m2;
2283 } while (m != NULL && m->m_type == MT_CONTROL);
2284 while (cm != NULL) {
2285 cmn = cm->m_next;
2286 cm->m_next = NULL;
2287 if (pr->pr_domain->dom_externalize != NULL) {
2288 error = (*pr->pr_domain->dom_externalize)
2289 (cm, controlp, flags);
2290 } else if (controlp != NULL)
2291 *controlp = cm;
2292 else
2293 m_freem(cm);
2294 if (controlp != NULL) {
2295 while (*controlp != NULL)
2296 controlp = &(*controlp)->m_next;
2297 }
2298 cm = cmn;
2299 }
2300 }
2301 KASSERT(m == NULL || m->m_type == MT_DATA,
2302 ("soreceive_dgram: !data"));
2303 while (m != NULL && uio->uio_resid > 0) {
2304 len = uio->uio_resid;
2305 if (len > m->m_len)
2306 len = m->m_len;
2307 error = uiomove(mtod(m, char *), (int)len, uio);
2308 if (error) {
2309 m_freem(m);
2310 return (error);
2311 }
2312 if (len == m->m_len)
2313 m = m_free(m);
2314 else {
2315 m->m_data += len;
2316 m->m_len -= len;
2317 }
2318 }
2319 if (m != NULL) {
2320 flags |= MSG_TRUNC;
2321 m_freem(m);
2322 }
2323 if (flagsp != NULL)
2324 *flagsp |= flags;
2325 return (0);
2326 }
2327
2328 int
2329 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2330 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2331 {
2332 int error;
2333
2334 CURVNET_SET(so->so_vnet);
2335 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2336 controlp, flagsp));
2337 CURVNET_RESTORE();
2338 return (error);
2339 }
2340
2341 int
2342 soshutdown(struct socket *so, int how)
2343 {
2344 struct protosw *pr = so->so_proto;
2345 int error, soerror_enotconn;
2346
2347 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2348 return (EINVAL);
2349
2350 soerror_enotconn = 0;
2351 if ((so->so_state &
2352 (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2353 /*
2354 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2355 * invoked on a datagram sockets, however historically we would
2356 * actually tear socket down. This is known to be leveraged by
2357 * some applications to unblock process waiting in recvXXX(2)
2358 * by other process that it shares that socket with. Try to meet
2359 * both backward-compatibility and POSIX requirements by forcing
2360 * ENOTCONN but still asking protocol to perform pru_shutdown().
2361 */
2362 if (so->so_type != SOCK_DGRAM)
2363 return (ENOTCONN);
2364 soerror_enotconn = 1;
2365 }
2366
2367 CURVNET_SET(so->so_vnet);
2368 if (pr->pr_usrreqs->pru_flush != NULL)
2369 (*pr->pr_usrreqs->pru_flush)(so, how);
2370 if (how != SHUT_WR)
2371 sorflush(so);
2372 if (how != SHUT_RD) {
2373 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2374 wakeup(&so->so_timeo);
2375 CURVNET_RESTORE();
2376 return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2377 }
2378 wakeup(&so->so_timeo);
2379 CURVNET_RESTORE();
2380
2381 return (soerror_enotconn ? ENOTCONN : 0);
2382 }
2383
2384 void
2385 sorflush(struct socket *so)
2386 {
2387 struct sockbuf *sb = &so->so_rcv;
2388 struct protosw *pr = so->so_proto;
2389 struct socket aso;
2390
2391 VNET_SO_ASSERT(so);
2392
2393 /*
2394 * In order to avoid calling dom_dispose with the socket buffer mutex
2395 * held, and in order to generally avoid holding the lock for a long
2396 * time, we make a copy of the socket buffer and clear the original
2397 * (except locks, state). The new socket buffer copy won't have
2398 * initialized locks so we can only call routines that won't use or
2399 * assert those locks.
2400 *
2401 * Dislodge threads currently blocked in receive and wait to acquire
2402 * a lock against other simultaneous readers before clearing the
2403 * socket buffer. Don't let our acquire be interrupted by a signal
2404 * despite any existing socket disposition on interruptable waiting.
2405 */
2406 socantrcvmore(so);
2407 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2408
2409 /*
2410 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2411 * and mutex data unchanged.
2412 */
2413 SOCKBUF_LOCK(sb);
2414 bzero(&aso, sizeof(aso));
2415 aso.so_pcb = so->so_pcb;
2416 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2417 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2418 bzero(&sb->sb_startzero,
2419 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2420 SOCKBUF_UNLOCK(sb);
2421 sbunlock(sb);
2422
2423 /*
2424 * Dispose of special rights and flush the copied socket. Don't call
2425 * any unsafe routines (that rely on locks being initialized) on aso.
2426 */
2427 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2428 (*pr->pr_domain->dom_dispose)(&aso);
2429 sbrelease_internal(&aso.so_rcv, so);
2430 }
2431
2432 /*
2433 * Wrapper for Socket established helper hook.
2434 * Parameters: socket, context of the hook point, hook id.
2435 */
2436 static int inline
2437 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2438 {
2439 struct socket_hhook_data hhook_data = {
2440 .so = so,
2441 .hctx = hctx,
2442 .m = NULL,
2443 .status = 0
2444 };
2445
2446 CURVNET_SET(so->so_vnet);
2447 HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2448 CURVNET_RESTORE();
2449
2450 /* Ugly but needed, since hhooks return void for now */
2451 return (hhook_data.status);
2452 }
2453
2454 /*
2455 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2456 * additional variant to handle the case where the option value needs to be
2457 * some kind of integer, but not a specific size. In addition to their use
2458 * here, these functions are also called by the protocol-level pr_ctloutput()
2459 * routines.
2460 */
2461 int
2462 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2463 {
2464 size_t valsize;
2465
2466 /*
2467 * If the user gives us more than we wanted, we ignore it, but if we
2468 * don't get the minimum length the caller wants, we return EINVAL.
2469 * On success, sopt->sopt_valsize is set to however much we actually
2470 * retrieved.
2471 */
2472 if ((valsize = sopt->sopt_valsize) < minlen)
2473 return EINVAL;
2474 if (valsize > len)
2475 sopt->sopt_valsize = valsize = len;
2476
2477 if (sopt->sopt_td != NULL)
2478 return (copyin(sopt->sopt_val, buf, valsize));
2479
2480 bcopy(sopt->sopt_val, buf, valsize);
2481 return (0);
2482 }
2483
2484 /*
2485 * Kernel version of setsockopt(2).
2486 *
2487 * XXX: optlen is size_t, not socklen_t
2488 */
2489 int
2490 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2491 size_t optlen)
2492 {
2493 struct sockopt sopt;
2494
2495 sopt.sopt_level = level;
2496 sopt.sopt_name = optname;
2497 sopt.sopt_dir = SOPT_SET;
2498 sopt.sopt_val = optval;
2499 sopt.sopt_valsize = optlen;
2500 sopt.sopt_td = NULL;
2501 return (sosetopt(so, &sopt));
2502 }
2503
2504 int
2505 sosetopt(struct socket *so, struct sockopt *sopt)
2506 {
2507 int error, optval;
2508 struct linger l;
2509 struct timeval tv;
2510 sbintime_t val;
2511 uint32_t val32;
2512 #ifdef MAC
2513 struct mac extmac;
2514 #endif
2515
2516 CURVNET_SET(so->so_vnet);
2517 error = 0;
2518 if (sopt->sopt_level != SOL_SOCKET) {
2519 if (so->so_proto->pr_ctloutput != NULL) {
2520 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2521 CURVNET_RESTORE();
2522 return (error);
2523 }
2524 error = ENOPROTOOPT;
2525 } else {
2526 switch (sopt->sopt_name) {
2527 case SO_ACCEPTFILTER:
2528 error = do_setopt_accept_filter(so, sopt);
2529 if (error)
2530 goto bad;
2531 break;
2532
2533 case SO_LINGER:
2534 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2535 if (error)
2536 goto bad;
2537
2538 SOCK_LOCK(so);
2539 so->so_linger = l.l_linger;
2540 if (l.l_onoff)
2541 so->so_options |= SO_LINGER;
2542 else
2543 so->so_options &= ~SO_LINGER;
2544 SOCK_UNLOCK(so);
2545 break;
2546
2547 case SO_DEBUG:
2548 case SO_KEEPALIVE:
2549 case SO_DONTROUTE:
2550 case SO_USELOOPBACK:
2551 case SO_BROADCAST:
2552 case SO_REUSEADDR:
2553 case SO_REUSEPORT:
2554 case SO_OOBINLINE:
2555 case SO_TIMESTAMP:
2556 case SO_BINTIME:
2557 case SO_NOSIGPIPE:
2558 case SO_NO_DDP:
2559 case SO_NO_OFFLOAD:
2560 error = sooptcopyin(sopt, &optval, sizeof optval,
2561 sizeof optval);
2562 if (error)
2563 goto bad;
2564 SOCK_LOCK(so);
2565 if (optval)
2566 so->so_options |= sopt->sopt_name;
2567 else
2568 so->so_options &= ~sopt->sopt_name;
2569 SOCK_UNLOCK(so);
2570 break;
2571
2572 case SO_SETFIB:
2573 error = sooptcopyin(sopt, &optval, sizeof optval,
2574 sizeof optval);
2575 if (error)
2576 goto bad;
2577
2578 if (optval < 0 || optval >= rt_numfibs) {
2579 error = EINVAL;
2580 goto bad;
2581 }
2582 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2583 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2584 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2585 so->so_fibnum = optval;
2586 else
2587 so->so_fibnum = 0;
2588 break;
2589
2590 case SO_USER_COOKIE:
2591 error = sooptcopyin(sopt, &val32, sizeof val32,
2592 sizeof val32);
2593 if (error)
2594 goto bad;
2595 so->so_user_cookie = val32;
2596 break;
2597
2598 case SO_SNDBUF:
2599 case SO_RCVBUF:
2600 case SO_SNDLOWAT:
2601 case SO_RCVLOWAT:
2602 error = sooptcopyin(sopt, &optval, sizeof optval,
2603 sizeof optval);
2604 if (error)
2605 goto bad;
2606
2607 /*
2608 * Values < 1 make no sense for any of these options,
2609 * so disallow them.
2610 */
2611 if (optval < 1) {
2612 error = EINVAL;
2613 goto bad;
2614 }
2615
2616 switch (sopt->sopt_name) {
2617 case SO_SNDBUF:
2618 case SO_RCVBUF:
2619 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2620 &so->so_snd : &so->so_rcv, (u_long)optval,
2621 so, curthread) == 0) {
2622 error = ENOBUFS;
2623 goto bad;
2624 }
2625 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2626 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2627 break;
2628
2629 /*
2630 * Make sure the low-water is never greater than the
2631 * high-water.
2632 */
2633 case SO_SNDLOWAT:
2634 SOCKBUF_LOCK(&so->so_snd);
2635 so->so_snd.sb_lowat =
2636 (optval > so->so_snd.sb_hiwat) ?
2637 so->so_snd.sb_hiwat : optval;
2638 SOCKBUF_UNLOCK(&so->so_snd);
2639 break;
2640 case SO_RCVLOWAT:
2641 SOCKBUF_LOCK(&so->so_rcv);
2642 so->so_rcv.sb_lowat =
2643 (optval > so->so_rcv.sb_hiwat) ?
2644 so->so_rcv.sb_hiwat : optval;
2645 SOCKBUF_UNLOCK(&so->so_rcv);
2646 break;
2647 }
2648 break;
2649
2650 case SO_SNDTIMEO:
2651 case SO_RCVTIMEO:
2652 #ifdef COMPAT_FREEBSD32
2653 if (SV_CURPROC_FLAG(SV_ILP32)) {
2654 struct timeval32 tv32;
2655
2656 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2657 sizeof tv32);
2658 CP(tv32, tv, tv_sec);
2659 CP(tv32, tv, tv_usec);
2660 } else
2661 #endif
2662 error = sooptcopyin(sopt, &tv, sizeof tv,
2663 sizeof tv);
2664 if (error)
2665 goto bad;
2666 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2667 tv.tv_usec >= 1000000) {
2668 error = EDOM;
2669 goto bad;
2670 }
2671 if (tv.tv_sec > INT32_MAX)
2672 val = SBT_MAX;
2673 else
2674 val = tvtosbt(tv);
2675 switch (sopt->sopt_name) {
2676 case SO_SNDTIMEO:
2677 so->so_snd.sb_timeo = val;
2678 break;
2679 case SO_RCVTIMEO:
2680 so->so_rcv.sb_timeo = val;
2681 break;
2682 }
2683 break;
2684
2685 case SO_LABEL:
2686 #ifdef MAC
2687 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2688 sizeof extmac);
2689 if (error)
2690 goto bad;
2691 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2692 so, &extmac);
2693 #else
2694 error = EOPNOTSUPP;
2695 #endif
2696 break;
2697
2698 default:
2699 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2700 error = hhook_run_socket(so, sopt,
2701 HHOOK_SOCKET_OPT);
2702 else
2703 error = ENOPROTOOPT;
2704 break;
2705 }
2706 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2707 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2708 }
2709 bad:
2710 CURVNET_RESTORE();
2711 return (error);
2712 }
2713
2714 /*
2715 * Helper routine for getsockopt.
2716 */
2717 int
2718 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2719 {
2720 int error;
2721 size_t valsize;
2722
2723 error = 0;
2724
2725 /*
2726 * Documented get behavior is that we always return a value, possibly
2727 * truncated to fit in the user's buffer. Traditional behavior is
2728 * that we always tell the user precisely how much we copied, rather
2729 * than something useful like the total amount we had available for
2730 * her. Note that this interface is not idempotent; the entire
2731 * answer must be generated ahead of time.
2732 */
2733 valsize = min(len, sopt->sopt_valsize);
2734 sopt->sopt_valsize = valsize;
2735 if (sopt->sopt_val != NULL) {
2736 if (sopt->sopt_td != NULL)
2737 error = copyout(buf, sopt->sopt_val, valsize);
2738 else
2739 bcopy(buf, sopt->sopt_val, valsize);
2740 }
2741 return (error);
2742 }
2743
2744 int
2745 sogetopt(struct socket *so, struct sockopt *sopt)
2746 {
2747 int error, optval;
2748 struct linger l;
2749 struct timeval tv;
2750 #ifdef MAC
2751 struct mac extmac;
2752 #endif
2753
2754 CURVNET_SET(so->so_vnet);
2755 error = 0;
2756 if (sopt->sopt_level != SOL_SOCKET) {
2757 if (so->so_proto->pr_ctloutput != NULL)
2758 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2759 else
2760 error = ENOPROTOOPT;
2761 CURVNET_RESTORE();
2762 return (error);
2763 } else {
2764 switch (sopt->sopt_name) {
2765 case SO_ACCEPTFILTER:
2766 error = do_getopt_accept_filter(so, sopt);
2767 break;
2768
2769 case SO_LINGER:
2770 SOCK_LOCK(so);
2771 l.l_onoff = so->so_options & SO_LINGER;
2772 l.l_linger = so->so_linger;
2773 SOCK_UNLOCK(so);
2774 error = sooptcopyout(sopt, &l, sizeof l);
2775 break;
2776
2777 case SO_USELOOPBACK:
2778 case SO_DONTROUTE:
2779 case SO_DEBUG:
2780 case SO_KEEPALIVE:
2781 case SO_REUSEADDR:
2782 case SO_REUSEPORT:
2783 case SO_BROADCAST:
2784 case SO_OOBINLINE:
2785 case SO_ACCEPTCONN:
2786 case SO_TIMESTAMP:
2787 case SO_BINTIME:
2788 case SO_NOSIGPIPE:
2789 optval = so->so_options & sopt->sopt_name;
2790 integer:
2791 error = sooptcopyout(sopt, &optval, sizeof optval);
2792 break;
2793
2794 case SO_TYPE:
2795 optval = so->so_type;
2796 goto integer;
2797
2798 case SO_PROTOCOL:
2799 optval = so->so_proto->pr_protocol;
2800 goto integer;
2801
2802 case SO_ERROR:
2803 SOCK_LOCK(so);
2804 optval = so->so_error;
2805 so->so_error = 0;
2806 SOCK_UNLOCK(so);
2807 goto integer;
2808
2809 case SO_SNDBUF:
2810 optval = so->so_snd.sb_hiwat;
2811 goto integer;
2812
2813 case SO_RCVBUF:
2814 optval = so->so_rcv.sb_hiwat;
2815 goto integer;
2816
2817 case SO_SNDLOWAT:
2818 optval = so->so_snd.sb_lowat;
2819 goto integer;
2820
2821 case SO_RCVLOWAT:
2822 optval = so->so_rcv.sb_lowat;
2823 goto integer;
2824
2825 case SO_SNDTIMEO:
2826 case SO_RCVTIMEO:
2827 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2828 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2829 #ifdef COMPAT_FREEBSD32
2830 if (SV_CURPROC_FLAG(SV_ILP32)) {
2831 struct timeval32 tv32;
2832
2833 CP(tv, tv32, tv_sec);
2834 CP(tv, tv32, tv_usec);
2835 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2836 } else
2837 #endif
2838 error = sooptcopyout(sopt, &tv, sizeof tv);
2839 break;
2840
2841 case SO_LABEL:
2842 #ifdef MAC
2843 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2844 sizeof(extmac));
2845 if (error)
2846 goto bad;
2847 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2848 so, &extmac);
2849 if (error)
2850 goto bad;
2851 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2852 #else
2853 error = EOPNOTSUPP;
2854 #endif
2855 break;
2856
2857 case SO_PEERLABEL:
2858 #ifdef MAC
2859 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2860 sizeof(extmac));
2861 if (error)
2862 goto bad;
2863 error = mac_getsockopt_peerlabel(
2864 sopt->sopt_td->td_ucred, so, &extmac);
2865 if (error)
2866 goto bad;
2867 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2868 #else
2869 error = EOPNOTSUPP;
2870 #endif
2871 break;
2872
2873 case SO_LISTENQLIMIT:
2874 optval = so->so_qlimit;
2875 goto integer;
2876
2877 case SO_LISTENQLEN:
2878 optval = so->so_qlen;
2879 goto integer;
2880
2881 case SO_LISTENINCQLEN:
2882 optval = so->so_incqlen;
2883 goto integer;
2884
2885 default:
2886 if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2887 error = hhook_run_socket(so, sopt,
2888 HHOOK_SOCKET_OPT);
2889 else
2890 error = ENOPROTOOPT;
2891 break;
2892 }
2893 }
2894 #ifdef MAC
2895 bad:
2896 #endif
2897 CURVNET_RESTORE();
2898 return (error);
2899 }
2900
2901 int
2902 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2903 {
2904 struct mbuf *m, *m_prev;
2905 int sopt_size = sopt->sopt_valsize;
2906
2907 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2908 if (m == NULL)
2909 return ENOBUFS;
2910 if (sopt_size > MLEN) {
2911 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2912 if ((m->m_flags & M_EXT) == 0) {
2913 m_free(m);
2914 return ENOBUFS;
2915 }
2916 m->m_len = min(MCLBYTES, sopt_size);
2917 } else {
2918 m->m_len = min(MLEN, sopt_size);
2919 }
2920 sopt_size -= m->m_len;
2921 *mp = m;
2922 m_prev = m;
2923
2924 while (sopt_size) {
2925 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2926 if (m == NULL) {
2927 m_freem(*mp);
2928 return ENOBUFS;
2929 }
2930 if (sopt_size > MLEN) {
2931 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2932 M_NOWAIT);
2933 if ((m->m_flags & M_EXT) == 0) {
2934 m_freem(m);
2935 m_freem(*mp);
2936 return ENOBUFS;
2937 }
2938 m->m_len = min(MCLBYTES, sopt_size);
2939 } else {
2940 m->m_len = min(MLEN, sopt_size);
2941 }
2942 sopt_size -= m->m_len;
2943 m_prev->m_next = m;
2944 m_prev = m;
2945 }
2946 return (0);
2947 }
2948
2949 int
2950 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2951 {
2952 struct mbuf *m0 = m;
2953
2954 if (sopt->sopt_val == NULL)
2955 return (0);
2956 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2957 if (sopt->sopt_td != NULL) {
2958 int error;
2959
2960 error = copyin(sopt->sopt_val, mtod(m, char *),
2961 m->m_len);
2962 if (error != 0) {
2963 m_freem(m0);
2964 return(error);
2965 }
2966 } else
2967 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2968 sopt->sopt_valsize -= m->m_len;
2969 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2970 m = m->m_next;
2971 }
2972 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2973 panic("ip6_sooptmcopyin");
2974 return (0);
2975 }
2976
2977 int
2978 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2979 {
2980 struct mbuf *m0 = m;
2981 size_t valsize = 0;
2982
2983 if (sopt->sopt_val == NULL)
2984 return (0);
2985 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2986 if (sopt->sopt_td != NULL) {
2987 int error;
2988
2989 error = copyout(mtod(m, char *), sopt->sopt_val,
2990 m->m_len);
2991 if (error != 0) {
2992 m_freem(m0);
2993 return(error);
2994 }
2995 } else
2996 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2997 sopt->sopt_valsize -= m->m_len;
2998 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2999 valsize += m->m_len;
3000 m = m->m_next;
3001 }
3002 if (m != NULL) {
3003 /* enough soopt buffer should be given from user-land */
3004 m_freem(m0);
3005 return(EINVAL);
3006 }
3007 sopt->sopt_valsize = valsize;
3008 return (0);
3009 }
3010
3011 /*
3012 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3013 * out-of-band data, which will then notify socket consumers.
3014 */
3015 void
3016 sohasoutofband(struct socket *so)
3017 {
3018
3019 if (so->so_sigio != NULL)
3020 pgsigio(&so->so_sigio, SIGURG, 0);
3021 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
3022 }
3023
3024 int
3025 sopoll(struct socket *so, int events, struct ucred *active_cred,
3026 struct thread *td)
3027 {
3028
3029 /*
3030 * We do not need to set or assert curvnet as long as everyone uses
3031 * sopoll_generic().
3032 */
3033 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3034 td));
3035 }
3036
3037 int
3038 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3039 struct thread *td)
3040 {
3041 int revents = 0;
3042
3043 SOCKBUF_LOCK(&so->so_snd);
3044 SOCKBUF_LOCK(&so->so_rcv);
3045 if (events & (POLLIN | POLLRDNORM))
3046 if (soreadabledata(so))
3047 revents |= events & (POLLIN | POLLRDNORM);
3048
3049 if (events & (POLLOUT | POLLWRNORM))
3050 if (sowriteable(so))
3051 revents |= events & (POLLOUT | POLLWRNORM);
3052
3053 if (events & (POLLPRI | POLLRDBAND))
3054 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3055 revents |= events & (POLLPRI | POLLRDBAND);
3056
3057 if ((events & POLLINIGNEOF) == 0) {
3058 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3059 revents |= events & (POLLIN | POLLRDNORM);
3060 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3061 revents |= POLLHUP;
3062 }
3063 }
3064
3065 if (revents == 0) {
3066 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3067 selrecord(td, &so->so_rcv.sb_sel);
3068 so->so_rcv.sb_flags |= SB_SEL;
3069 }
3070
3071 if (events & (POLLOUT | POLLWRNORM)) {
3072 selrecord(td, &so->so_snd.sb_sel);
3073 so->so_snd.sb_flags |= SB_SEL;
3074 }
3075 }
3076
3077 SOCKBUF_UNLOCK(&so->so_rcv);
3078 SOCKBUF_UNLOCK(&so->so_snd);
3079 return (revents);
3080 }
3081
3082 int
3083 soo_kqfilter(struct file *fp, struct knote *kn)
3084 {
3085 struct socket *so = kn->kn_fp->f_data;
3086 struct sockbuf *sb;
3087
3088 switch (kn->kn_filter) {
3089 case EVFILT_READ:
3090 if (so->so_options & SO_ACCEPTCONN)
3091 kn->kn_fop = &solisten_filtops;
3092 else
3093 kn->kn_fop = &soread_filtops;
3094 sb = &so->so_rcv;
3095 break;
3096 case EVFILT_WRITE:
3097 kn->kn_fop = &sowrite_filtops;
3098 sb = &so->so_snd;
3099 break;
3100 default:
3101 return (EINVAL);
3102 }
3103
3104 SOCKBUF_LOCK(sb);
3105 knlist_add(&sb->sb_sel.si_note, kn, 1);
3106 sb->sb_flags |= SB_KNOTE;
3107 SOCKBUF_UNLOCK(sb);
3108 return (0);
3109 }
3110
3111 /*
3112 * Some routines that return EOPNOTSUPP for entry points that are not
3113 * supported by a protocol. Fill in as needed.
3114 */
3115 int
3116 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3117 {
3118
3119 return EOPNOTSUPP;
3120 }
3121
3122 int
3123 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3124 {
3125
3126 return EOPNOTSUPP;
3127 }
3128
3129 int
3130 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3131 {
3132
3133 return EOPNOTSUPP;
3134 }
3135
3136 int
3137 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3138 {
3139
3140 return EOPNOTSUPP;
3141 }
3142
3143 int
3144 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3145 struct thread *td)
3146 {
3147
3148 return EOPNOTSUPP;
3149 }
3150
3151 int
3152 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3153 {
3154
3155 return EOPNOTSUPP;
3156 }
3157
3158 int
3159 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3160 struct thread *td)
3161 {
3162
3163 return EOPNOTSUPP;
3164 }
3165
3166 int
3167 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3168 {
3169
3170 return EOPNOTSUPP;
3171 }
3172
3173 int
3174 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3175 struct ifnet *ifp, struct thread *td)
3176 {
3177
3178 return EOPNOTSUPP;
3179 }
3180
3181 int
3182 pru_disconnect_notsupp(struct socket *so)
3183 {
3184
3185 return EOPNOTSUPP;
3186 }
3187
3188 int
3189 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3190 {
3191
3192 return EOPNOTSUPP;
3193 }
3194
3195 int
3196 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3197 {
3198
3199 return EOPNOTSUPP;
3200 }
3201
3202 int
3203 pru_rcvd_notsupp(struct socket *so, int flags)
3204 {
3205
3206 return EOPNOTSUPP;
3207 }
3208
3209 int
3210 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3211 {
3212
3213 return EOPNOTSUPP;
3214 }
3215
3216 int
3217 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3218 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3219 {
3220
3221 return EOPNOTSUPP;
3222 }
3223
3224 int
3225 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3226 {
3227
3228 return (EOPNOTSUPP);
3229 }
3230
3231 /*
3232 * This isn't really a ``null'' operation, but it's the default one and
3233 * doesn't do anything destructive.
3234 */
3235 int
3236 pru_sense_null(struct socket *so, struct stat *sb)
3237 {
3238
3239 sb->st_blksize = so->so_snd.sb_hiwat;
3240 return 0;
3241 }
3242
3243 int
3244 pru_shutdown_notsupp(struct socket *so)
3245 {
3246
3247 return EOPNOTSUPP;
3248 }
3249
3250 int
3251 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3252 {
3253
3254 return EOPNOTSUPP;
3255 }
3256
3257 int
3258 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3259 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3260 {
3261
3262 return EOPNOTSUPP;
3263 }
3264
3265 int
3266 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3267 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3268 {
3269
3270 return EOPNOTSUPP;
3271 }
3272
3273 int
3274 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3275 struct thread *td)
3276 {
3277
3278 return EOPNOTSUPP;
3279 }
3280
3281 static void
3282 filt_sordetach(struct knote *kn)
3283 {
3284 struct socket *so = kn->kn_fp->f_data;
3285
3286 SOCKBUF_LOCK(&so->so_rcv);
3287 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3288 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3289 so->so_rcv.sb_flags &= ~SB_KNOTE;
3290 SOCKBUF_UNLOCK(&so->so_rcv);
3291 }
3292
3293 /*ARGSUSED*/
3294 static int
3295 filt_soread(struct knote *kn, long hint)
3296 {
3297 struct socket *so;
3298
3299 so = kn->kn_fp->f_data;
3300 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3301
3302 kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3303 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3304 kn->kn_flags |= EV_EOF;
3305 kn->kn_fflags = so->so_error;
3306 return (1);
3307 } else if (so->so_error) /* temporary udp error */
3308 return (1);
3309
3310 if (kn->kn_sfflags & NOTE_LOWAT) {
3311 if (kn->kn_data >= kn->kn_sdata)
3312 return 1;
3313 } else {
3314 if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3315 return 1;
3316 }
3317
3318 /* This hook returning non-zero indicates an event, not error */
3319 return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3320 }
3321
3322 static void
3323 filt_sowdetach(struct knote *kn)
3324 {
3325 struct socket *so = kn->kn_fp->f_data;
3326
3327 SOCKBUF_LOCK(&so->so_snd);
3328 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3329 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3330 so->so_snd.sb_flags &= ~SB_KNOTE;
3331 SOCKBUF_UNLOCK(&so->so_snd);
3332 }
3333
3334 /*ARGSUSED*/
3335 static int
3336 filt_sowrite(struct knote *kn, long hint)
3337 {
3338 struct socket *so;
3339
3340 so = kn->kn_fp->f_data;
3341 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3342 kn->kn_data = sbspace(&so->so_snd);
3343
3344 hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3345
3346 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3347 kn->kn_flags |= EV_EOF;
3348 kn->kn_fflags = so->so_error;
3349 return (1);
3350 } else if (so->so_error) /* temporary udp error */
3351 return (1);
3352 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3353 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3354 return (0);
3355 else if (kn->kn_sfflags & NOTE_LOWAT)
3356 return (kn->kn_data >= kn->kn_sdata);
3357 else
3358 return (kn->kn_data >= so->so_snd.sb_lowat);
3359 }
3360
3361 /*ARGSUSED*/
3362 static int
3363 filt_solisten(struct knote *kn, long hint)
3364 {
3365 struct socket *so = kn->kn_fp->f_data;
3366
3367 kn->kn_data = so->so_qlen;
3368 return (!TAILQ_EMPTY(&so->so_comp));
3369 }
3370
3371 int
3372 socheckuid(struct socket *so, uid_t uid)
3373 {
3374
3375 if (so == NULL)
3376 return (EPERM);
3377 if (so->so_cred->cr_uid != uid)
3378 return (EPERM);
3379 return (0);
3380 }
3381
3382 /*
3383 * These functions are used by protocols to notify the socket layer (and its
3384 * consumers) of state changes in the sockets driven by protocol-side events.
3385 */
3386
3387 /*
3388 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3389 *
3390 * Normal sequence from the active (originating) side is that
3391 * soisconnecting() is called during processing of connect() call, resulting
3392 * in an eventual call to soisconnected() if/when the connection is
3393 * established. When the connection is torn down soisdisconnecting() is
3394 * called during processing of disconnect() call, and soisdisconnected() is
3395 * called when the connection to the peer is totally severed. The semantics
3396 * of these routines are such that connectionless protocols can call
3397 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3398 * calls when setting up a ``connection'' takes no time.
3399 *
3400 * From the passive side, a socket is created with two queues of sockets:
3401 * so_incomp for connections in progress and so_comp for connections already
3402 * made and awaiting user acceptance. As a protocol is preparing incoming
3403 * connections, it creates a socket structure queued on so_incomp by calling
3404 * sonewconn(). When the connection is established, soisconnected() is
3405 * called, and transfers the socket structure to so_comp, making it available
3406 * to accept().
3407 *
3408 * If a socket is closed with sockets on either so_incomp or so_comp, these
3409 * sockets are dropped.
3410 *
3411 * If higher-level protocols are implemented in the kernel, the wakeups done
3412 * here will sometimes cause software-interrupt process scheduling.
3413 */
3414 void
3415 soisconnecting(struct socket *so)
3416 {
3417
3418 SOCK_LOCK(so);
3419 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3420 so->so_state |= SS_ISCONNECTING;
3421 SOCK_UNLOCK(so);
3422 }
3423
3424 void
3425 soisconnected(struct socket *so)
3426 {
3427 struct socket *head;
3428 int ret;
3429
3430 restart:
3431 ACCEPT_LOCK();
3432 SOCK_LOCK(so);
3433 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3434 so->so_state |= SS_ISCONNECTED;
3435 head = so->so_head;
3436 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3437 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3438 SOCK_UNLOCK(so);
3439 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3440 head->so_incqlen--;
3441 so->so_qstate &= ~SQ_INCOMP;
3442 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3443 head->so_qlen++;
3444 so->so_qstate |= SQ_COMP;
3445 ACCEPT_UNLOCK();
3446 sorwakeup(head);
3447 wakeup_one(&head->so_timeo);
3448 } else {
3449 ACCEPT_UNLOCK();
3450 soupcall_set(so, SO_RCV,
3451 head->so_accf->so_accept_filter->accf_callback,
3452 head->so_accf->so_accept_filter_arg);
3453 so->so_options &= ~SO_ACCEPTFILTER;
3454 ret = head->so_accf->so_accept_filter->accf_callback(so,
3455 head->so_accf->so_accept_filter_arg, M_NOWAIT);
3456 if (ret == SU_ISCONNECTED)
3457 soupcall_clear(so, SO_RCV);
3458 SOCK_UNLOCK(so);
3459 if (ret == SU_ISCONNECTED)
3460 goto restart;
3461 }
3462 return;
3463 }
3464 SOCK_UNLOCK(so);
3465 ACCEPT_UNLOCK();
3466 wakeup(&so->so_timeo);
3467 sorwakeup(so);
3468 sowwakeup(so);
3469 }
3470
3471 void
3472 soisdisconnecting(struct socket *so)
3473 {
3474
3475 /*
3476 * Note: This code assumes that SOCK_LOCK(so) and
3477 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3478 */
3479 SOCKBUF_LOCK(&so->so_rcv);
3480 so->so_state &= ~SS_ISCONNECTING;
3481 so->so_state |= SS_ISDISCONNECTING;
3482 socantrcvmore_locked(so);
3483 SOCKBUF_LOCK(&so->so_snd);
3484 socantsendmore_locked(so);
3485 wakeup(&so->so_timeo);
3486 }
3487
3488 void
3489 soisdisconnected(struct socket *so)
3490 {
3491
3492 /*
3493 * Note: This code assumes that SOCK_LOCK(so) and
3494 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3495 */
3496 SOCKBUF_LOCK(&so->so_rcv);
3497 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3498 so->so_state |= SS_ISDISCONNECTED;
3499 socantrcvmore_locked(so);
3500 SOCKBUF_LOCK(&so->so_snd);
3501 sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3502 socantsendmore_locked(so);
3503 wakeup(&so->so_timeo);
3504 }
3505
3506 /*
3507 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3508 */
3509 struct sockaddr *
3510 sodupsockaddr(const struct sockaddr *sa, int mflags)
3511 {
3512 struct sockaddr *sa2;
3513
3514 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3515 if (sa2)
3516 bcopy(sa, sa2, sa->sa_len);
3517 return sa2;
3518 }
3519
3520 /*
3521 * Register per-socket buffer upcalls.
3522 */
3523 void
3524 soupcall_set(struct socket *so, int which,
3525 int (*func)(struct socket *, void *, int), void *arg)
3526 {
3527 struct sockbuf *sb;
3528
3529 switch (which) {
3530 case SO_RCV:
3531 sb = &so->so_rcv;
3532 break;
3533 case SO_SND:
3534 sb = &so->so_snd;
3535 break;
3536 default:
3537 panic("soupcall_set: bad which");
3538 }
3539 SOCKBUF_LOCK_ASSERT(sb);
3540 #if 0
3541 /* XXX: accf_http actually wants to do this on purpose. */
3542 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3543 #endif
3544 sb->sb_upcall = func;
3545 sb->sb_upcallarg = arg;
3546 sb->sb_flags |= SB_UPCALL;
3547 }
3548
3549 void
3550 soupcall_clear(struct socket *so, int which)
3551 {
3552 struct sockbuf *sb;
3553
3554 switch (which) {
3555 case SO_RCV:
3556 sb = &so->so_rcv;
3557 break;
3558 case SO_SND:
3559 sb = &so->so_snd;
3560 break;
3561 default:
3562 panic("soupcall_clear: bad which");
3563 }
3564 SOCKBUF_LOCK_ASSERT(sb);
3565 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3566 sb->sb_upcall = NULL;
3567 sb->sb_upcallarg = NULL;
3568 sb->sb_flags &= ~SB_UPCALL;
3569 }
3570
3571 /*
3572 * Create an external-format (``xsocket'') structure using the information in
3573 * the kernel-format socket structure pointed to by so. This is done to
3574 * reduce the spew of irrelevant information over this interface, to isolate
3575 * user code from changes in the kernel structure, and potentially to provide
3576 * information-hiding if we decide that some of this information should be
3577 * hidden from users.
3578 */
3579 void
3580 sotoxsocket(struct socket *so, struct xsocket *xso)
3581 {
3582
3583 xso->xso_len = sizeof *xso;
3584 xso->xso_so = so;
3585 xso->so_type = so->so_type;
3586 xso->so_options = so->so_options;
3587 xso->so_linger = so->so_linger;
3588 xso->so_state = so->so_state;
3589 xso->so_pcb = so->so_pcb;
3590 xso->xso_protocol = so->so_proto->pr_protocol;
3591 xso->xso_family = so->so_proto->pr_domain->dom_family;
3592 xso->so_qlen = so->so_qlen;
3593 xso->so_incqlen = so->so_incqlen;
3594 xso->so_qlimit = so->so_qlimit;
3595 xso->so_timeo = so->so_timeo;
3596 xso->so_error = so->so_error;
3597 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3598 xso->so_oobmark = so->so_oobmark;
3599 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3600 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3601 xso->so_uid = so->so_cred->cr_uid;
3602 }
3603
3604
3605 /*
3606 * Socket accessor functions to provide external consumers with
3607 * a safe interface to socket state
3608 *
3609 */
3610
3611 void
3612 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3613 void *arg)
3614 {
3615
3616 TAILQ_FOREACH(so, &so->so_comp, so_list)
3617 func(so, arg);
3618 }
3619
3620 struct sockbuf *
3621 so_sockbuf_rcv(struct socket *so)
3622 {
3623
3624 return (&so->so_rcv);
3625 }
3626
3627 struct sockbuf *
3628 so_sockbuf_snd(struct socket *so)
3629 {
3630
3631 return (&so->so_snd);
3632 }
3633
3634 int
3635 so_state_get(const struct socket *so)
3636 {
3637
3638 return (so->so_state);
3639 }
3640
3641 void
3642 so_state_set(struct socket *so, int val)
3643 {
3644
3645 so->so_state = val;
3646 }
3647
3648 int
3649 so_options_get(const struct socket *so)
3650 {
3651
3652 return (so->so_options);
3653 }
3654
3655 void
3656 so_options_set(struct socket *so, int val)
3657 {
3658
3659 so->so_options = val;
3660 }
3661
3662 int
3663 so_error_get(const struct socket *so)
3664 {
3665
3666 return (so->so_error);
3667 }
3668
3669 void
3670 so_error_set(struct socket *so, int val)
3671 {
3672
3673 so->so_error = val;
3674 }
3675
3676 int
3677 so_linger_get(const struct socket *so)
3678 {
3679
3680 return (so->so_linger);
3681 }
3682
3683 void
3684 so_linger_set(struct socket *so, int val)
3685 {
3686
3687 so->so_linger = val;
3688 }
3689
3690 struct protosw *
3691 so_protosw_get(const struct socket *so)
3692 {
3693
3694 return (so->so_proto);
3695 }
3696
3697 void
3698 so_protosw_set(struct socket *so, struct protosw *val)
3699 {
3700
3701 so->so_proto = val;
3702 }
3703
3704 void
3705 so_sorwakeup(struct socket *so)
3706 {
3707
3708 sorwakeup(so);
3709 }
3710
3711 void
3712 so_sowwakeup(struct socket *so)
3713 {
3714
3715 sowwakeup(so);
3716 }
3717
3718 void
3719 so_sorwakeup_locked(struct socket *so)
3720 {
3721
3722 sorwakeup_locked(so);
3723 }
3724
3725 void
3726 so_sowwakeup_locked(struct socket *so)
3727 {
3728
3729 sowwakeup_locked(so);
3730 }
3731
3732 void
3733 so_lock(struct socket *so)
3734 {
3735
3736 SOCK_LOCK(so);
3737 }
3738
3739 void
3740 so_unlock(struct socket *so)
3741 {
3742
3743 SOCK_UNLOCK(so);
3744 }
Cache object: d94c206d6d3bc2eda0f15efa72cc939c
|