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: stable/10/sys/kern/uipc_socket.c 305261 2016-09-02 00:14:28Z markj $");
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/kernel.h>
122 #include <sys/event.h>
123 #include <sys/eventhandler.h>
124 #include <sys/poll.h>
125 #include <sys/proc.h>
126 #include <sys/protosw.h>
127 #include <sys/socket.h>
128 #include <sys/socketvar.h>
129 #include <sys/resourcevar.h>
130 #include <net/route.h>
131 #include <sys/signalvar.h>
132 #include <sys/stat.h>
133 #include <sys/sx.h>
134 #include <sys/sysctl.h>
135 #include <sys/uio.h>
136 #include <sys/jail.h>
137 #include <sys/syslog.h>
138 #include <netinet/in.h>
139
140 #include <net/vnet.h>
141
142 #include <security/mac/mac_framework.h>
143
144 #include <vm/uma.h>
145
146 #ifdef COMPAT_FREEBSD32
147 #include <sys/mount.h>
148 #include <sys/sysent.h>
149 #include <compat/freebsd32/freebsd32.h>
150 #endif
151
152 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
153 int flags);
154
155 static void filt_sordetach(struct knote *kn);
156 static int filt_soread(struct knote *kn, long hint);
157 static void filt_sowdetach(struct knote *kn);
158 static int filt_sowrite(struct knote *kn, long hint);
159 static int filt_solisten(struct knote *kn, long hint);
160
161 static struct filterops solisten_filtops = {
162 .f_isfd = 1,
163 .f_detach = filt_sordetach,
164 .f_event = filt_solisten,
165 };
166 static struct filterops soread_filtops = {
167 .f_isfd = 1,
168 .f_detach = filt_sordetach,
169 .f_event = filt_soread,
170 };
171 static struct filterops sowrite_filtops = {
172 .f_isfd = 1,
173 .f_detach = filt_sowdetach,
174 .f_event = filt_sowrite,
175 };
176
177 so_gen_t so_gencnt; /* generation count for sockets */
178
179 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
180 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
181
182 #define VNET_SO_ASSERT(so) \
183 VNET_ASSERT(curvnet != NULL, \
184 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
185
186 /*
187 * Limit on the number of connections in the listen queue waiting
188 * for accept(2).
189 * NB: The original sysctl somaxconn is still available but hidden
190 * to prevent confusion about the actual purpose of this number.
191 */
192 static int somaxconn = SOMAXCONN;
193
194 static int
sysctl_somaxconn(SYSCTL_HANDLER_ARGS)195 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
196 {
197 int error;
198 int val;
199
200 val = somaxconn;
201 error = sysctl_handle_int(oidp, &val, 0, req);
202 if (error || !req->newptr )
203 return (error);
204
205 if (val < 1 || val > USHRT_MAX)
206 return (EINVAL);
207
208 somaxconn = val;
209 return (0);
210 }
211 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
212 0, sizeof(int), sysctl_somaxconn, "I",
213 "Maximum listen socket pending connection accept queue size");
214 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
215 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
216 0, sizeof(int), sysctl_somaxconn, "I",
217 "Maximum listen socket pending connection accept queue size (compat)");
218
219 static int numopensockets;
220 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
221 &numopensockets, 0, "Number of open sockets");
222
223 /*
224 * accept_mtx locks down per-socket fields relating to accept queues. See
225 * socketvar.h for an annotation of the protected fields of struct socket.
226 */
227 struct mtx accept_mtx;
228 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
229
230 /*
231 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
232 * so_gencnt field.
233 */
234 static struct mtx so_global_mtx;
235 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
236
237 /*
238 * General IPC sysctl name space, used by sockets and a variety of other IPC
239 * types.
240 */
241 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
242
243 /*
244 * Initialize the socket subsystem and set up the socket
245 * memory allocator.
246 */
247 static uma_zone_t socket_zone;
248 int maxsockets;
249
250 static void
socket_zone_change(void * tag)251 socket_zone_change(void *tag)
252 {
253
254 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
255 }
256
257 static void
socket_init(void * tag)258 socket_init(void *tag)
259 {
260
261 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
262 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
263 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
264 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
265 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
266 EVENTHANDLER_PRI_FIRST);
267 }
268 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
269
270 /*
271 * Initialise maxsockets. This SYSINIT must be run after
272 * tunable_mbinit().
273 */
274 static void
init_maxsockets(void * ignored)275 init_maxsockets(void *ignored)
276 {
277
278 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
279 maxsockets = imax(maxsockets, maxfiles);
280 }
281 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
282
283 /*
284 * Sysctl to get and set the maximum global sockets limit. Notify protocols
285 * of the change so that they can update their dependent limits as required.
286 */
287 static int
sysctl_maxsockets(SYSCTL_HANDLER_ARGS)288 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
289 {
290 int error, newmaxsockets;
291
292 newmaxsockets = maxsockets;
293 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
294 if (error == 0 && req->newptr) {
295 if (newmaxsockets > maxsockets &&
296 newmaxsockets <= maxfiles) {
297 maxsockets = newmaxsockets;
298 EVENTHANDLER_INVOKE(maxsockets_change);
299 } else
300 error = EINVAL;
301 }
302 return (error);
303 }
304 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
305 &maxsockets, 0, sysctl_maxsockets, "IU",
306 "Maximum number of sockets avaliable");
307
308 /*
309 * Socket operation routines. These routines are called by the routines in
310 * sys_socket.c or from a system process, and implement the semantics of
311 * socket operations by switching out to the protocol specific routines.
312 */
313
314 /*
315 * Get a socket structure from our zone, and initialize it. Note that it
316 * would probably be better to allocate socket and PCB at the same time, but
317 * I'm not convinced that all the protocols can be easily modified to do
318 * this.
319 *
320 * soalloc() returns a socket with a ref count of 0.
321 */
322 static struct socket *
soalloc(struct vnet * vnet)323 soalloc(struct vnet *vnet)
324 {
325 struct socket *so;
326
327 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
328 if (so == NULL)
329 return (NULL);
330 #ifdef MAC
331 if (mac_socket_init(so, M_NOWAIT) != 0) {
332 uma_zfree(socket_zone, so);
333 return (NULL);
334 }
335 #endif
336 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
337 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
338 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
339 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
340 TAILQ_INIT(&so->so_aiojobq);
341 mtx_lock(&so_global_mtx);
342 so->so_gencnt = ++so_gencnt;
343 ++numopensockets;
344 #ifdef VIMAGE
345 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
346 __func__, __LINE__, so));
347 vnet->vnet_sockcnt++;
348 so->so_vnet = vnet;
349 #endif
350 mtx_unlock(&so_global_mtx);
351 return (so);
352 }
353
354 /*
355 * Free the storage associated with a socket at the socket layer, tear down
356 * locks, labels, etc. All protocol state is assumed already to have been
357 * torn down (and possibly never set up) by the caller.
358 */
359 static void
sodealloc(struct socket * so)360 sodealloc(struct socket *so)
361 {
362
363 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
364 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
365
366 mtx_lock(&so_global_mtx);
367 so->so_gencnt = ++so_gencnt;
368 --numopensockets; /* Could be below, but faster here. */
369 #ifdef VIMAGE
370 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
371 __func__, __LINE__, so));
372 so->so_vnet->vnet_sockcnt--;
373 #endif
374 mtx_unlock(&so_global_mtx);
375 if (so->so_rcv.sb_hiwat)
376 (void)chgsbsize(so->so_cred->cr_uidinfo,
377 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
378 if (so->so_snd.sb_hiwat)
379 (void)chgsbsize(so->so_cred->cr_uidinfo,
380 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
381 #ifdef INET
382 /* remove acccept filter if one is present. */
383 if (so->so_accf != NULL)
384 do_setopt_accept_filter(so, NULL);
385 #endif
386 #ifdef MAC
387 mac_socket_destroy(so);
388 #endif
389 crfree(so->so_cred);
390 sx_destroy(&so->so_snd.sb_sx);
391 sx_destroy(&so->so_rcv.sb_sx);
392 SOCKBUF_LOCK_DESTROY(&so->so_snd);
393 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
394 uma_zfree(socket_zone, so);
395 }
396
397 /*
398 * socreate returns a socket with a ref count of 1. The socket should be
399 * closed with soclose().
400 */
401 int
socreate(int dom,struct socket ** aso,int type,int proto,struct ucred * cred,struct thread * td)402 socreate(int dom, struct socket **aso, int type, int proto,
403 struct ucred *cred, struct thread *td)
404 {
405 struct protosw *prp;
406 struct socket *so;
407 int error;
408
409 if (proto)
410 prp = pffindproto(dom, proto, type);
411 else
412 prp = pffindtype(dom, type);
413
414 if (prp == NULL) {
415 /* No support for domain. */
416 if (pffinddomain(dom) == NULL)
417 return (EAFNOSUPPORT);
418 /* No support for socket type. */
419 if (proto == 0 && type != 0)
420 return (EPROTOTYPE);
421 return (EPROTONOSUPPORT);
422 }
423 if (prp->pr_usrreqs->pru_attach == NULL ||
424 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
425 return (EPROTONOSUPPORT);
426
427 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
428 return (EPROTONOSUPPORT);
429
430 if (prp->pr_type != type)
431 return (EPROTOTYPE);
432 so = soalloc(CRED_TO_VNET(cred));
433 if (so == NULL)
434 return (ENOBUFS);
435
436 TAILQ_INIT(&so->so_incomp);
437 TAILQ_INIT(&so->so_comp);
438 so->so_type = type;
439 so->so_cred = crhold(cred);
440 if ((prp->pr_domain->dom_family == PF_INET) ||
441 (prp->pr_domain->dom_family == PF_INET6) ||
442 (prp->pr_domain->dom_family == PF_ROUTE))
443 so->so_fibnum = td->td_proc->p_fibnum;
444 else
445 so->so_fibnum = 0;
446 so->so_proto = prp;
447 #ifdef MAC
448 mac_socket_create(cred, so);
449 #endif
450 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
451 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
452 so->so_count = 1;
453 /*
454 * Auto-sizing of socket buffers is managed by the protocols and
455 * the appropriate flags must be set in the pru_attach function.
456 */
457 CURVNET_SET(so->so_vnet);
458 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
459 CURVNET_RESTORE();
460 if (error) {
461 KASSERT(so->so_count == 1, ("socreate: so_count %d",
462 so->so_count));
463 so->so_count = 0;
464 sodealloc(so);
465 return (error);
466 }
467 *aso = so;
468 return (0);
469 }
470
471 #ifdef REGRESSION
472 static int regression_sonewconn_earlytest = 1;
473 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
474 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
475 #endif
476
477 /*
478 * When an attempt at a new connection is noted on a socket which accepts
479 * connections, sonewconn is called. If the connection is possible (subject
480 * to space constraints, etc.) then we allocate a new structure, propoerly
481 * linked into the data structure of the original socket, and return this.
482 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
483 *
484 * Note: the ref count on the socket is 0 on return.
485 */
486 struct socket *
sonewconn(struct socket * head,int connstatus)487 sonewconn(struct socket *head, int connstatus)
488 {
489 static struct timeval lastover;
490 static struct timeval overinterval = { 60, 0 };
491 static int overcount;
492
493 struct socket *so;
494 int over;
495
496 ACCEPT_LOCK();
497 over = (head->so_qlen > 3 * head->so_qlimit / 2);
498 ACCEPT_UNLOCK();
499 #ifdef REGRESSION
500 if (regression_sonewconn_earlytest && over) {
501 #else
502 if (over) {
503 #endif
504 overcount++;
505
506 if (ratecheck(&lastover, &overinterval)) {
507 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
508 "%i already in queue awaiting acceptance "
509 "(%d occurrences)\n",
510 __func__, head->so_pcb, head->so_qlen, overcount);
511
512 overcount = 0;
513 }
514
515 return (NULL);
516 }
517 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
518 __func__, __LINE__, head));
519 so = soalloc(head->so_vnet);
520 if (so == NULL) {
521 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
522 "limit reached or out of memory\n",
523 __func__, head->so_pcb);
524 return (NULL);
525 }
526 if ((head->so_options & SO_ACCEPTFILTER) != 0)
527 connstatus = 0;
528 so->so_head = head;
529 so->so_type = head->so_type;
530 so->so_options = head->so_options &~ SO_ACCEPTCONN;
531 so->so_linger = head->so_linger;
532 so->so_state = head->so_state | SS_NOFDREF;
533 so->so_fibnum = head->so_fibnum;
534 so->so_proto = head->so_proto;
535 so->so_cred = crhold(head->so_cred);
536 #ifdef MAC
537 mac_socket_newconn(head, so);
538 #endif
539 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
540 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
541 VNET_SO_ASSERT(head);
542 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
543 sodealloc(so);
544 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
545 __func__, head->so_pcb);
546 return (NULL);
547 }
548 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
549 sodealloc(so);
550 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
551 __func__, head->so_pcb);
552 return (NULL);
553 }
554 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
555 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
556 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
557 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
558 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
559 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
560 so->so_state |= connstatus;
561 ACCEPT_LOCK();
562 /*
563 * The accept socket may be tearing down but we just
564 * won a race on the ACCEPT_LOCK.
565 * However, if sctp_peeloff() is called on a 1-to-many
566 * style socket, the SO_ACCEPTCONN doesn't need to be set.
567 */
568 if (!(head->so_options & SO_ACCEPTCONN) &&
569 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
570 (head->so_type != SOCK_SEQPACKET))) {
571 SOCK_LOCK(so);
572 so->so_head = NULL;
573 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
574 return (NULL);
575 }
576 if (connstatus) {
577 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
578 so->so_qstate |= SQ_COMP;
579 head->so_qlen++;
580 } else {
581 /*
582 * Keep removing sockets from the head until there's room for
583 * us to insert on the tail. In pre-locking revisions, this
584 * was a simple if(), but as we could be racing with other
585 * threads and soabort() requires dropping locks, we must
586 * loop waiting for the condition to be true.
587 */
588 while (head->so_incqlen > head->so_qlimit) {
589 struct socket *sp;
590 sp = TAILQ_FIRST(&head->so_incomp);
591 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
592 head->so_incqlen--;
593 sp->so_qstate &= ~SQ_INCOMP;
594 sp->so_head = NULL;
595 ACCEPT_UNLOCK();
596 soabort(sp);
597 ACCEPT_LOCK();
598 }
599 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
600 so->so_qstate |= SQ_INCOMP;
601 head->so_incqlen++;
602 }
603 ACCEPT_UNLOCK();
604 if (connstatus) {
605 sorwakeup(head);
606 wakeup_one(&head->so_timeo);
607 }
608 return (so);
609 }
610
611 int
612 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
613 {
614 int error;
615
616 CURVNET_SET(so->so_vnet);
617 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
618 CURVNET_RESTORE();
619 return (error);
620 }
621
622 int
623 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
624 {
625 int error;
626
627 CURVNET_SET(so->so_vnet);
628 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
629 CURVNET_RESTORE();
630 return (error);
631 }
632
633 /*
634 * solisten() transitions a socket from a non-listening state to a listening
635 * state, but can also be used to update the listen queue depth on an
636 * existing listen socket. The protocol will call back into the sockets
637 * layer using solisten_proto_check() and solisten_proto() to check and set
638 * socket-layer listen state. Call backs are used so that the protocol can
639 * acquire both protocol and socket layer locks in whatever order is required
640 * by the protocol.
641 *
642 * Protocol implementors are advised to hold the socket lock across the
643 * socket-layer test and set to avoid races at the socket layer.
644 */
645 int
646 solisten(struct socket *so, int backlog, struct thread *td)
647 {
648 int error;
649
650 CURVNET_SET(so->so_vnet);
651 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
652 CURVNET_RESTORE();
653 return (error);
654 }
655
656 int
657 solisten_proto_check(struct socket *so)
658 {
659
660 SOCK_LOCK_ASSERT(so);
661
662 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
663 SS_ISDISCONNECTING))
664 return (EINVAL);
665 return (0);
666 }
667
668 void
669 solisten_proto(struct socket *so, int backlog)
670 {
671
672 SOCK_LOCK_ASSERT(so);
673
674 if (backlog < 0 || backlog > somaxconn)
675 backlog = somaxconn;
676 so->so_qlimit = backlog;
677 so->so_options |= SO_ACCEPTCONN;
678 }
679
680 /*
681 * Evaluate the reference count and named references on a socket; if no
682 * references remain, free it. This should be called whenever a reference is
683 * released, such as in sorele(), but also when named reference flags are
684 * cleared in socket or protocol code.
685 *
686 * sofree() will free the socket if:
687 *
688 * - There are no outstanding file descriptor references or related consumers
689 * (so_count == 0).
690 *
691 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
692 *
693 * - The protocol does not have an outstanding strong reference on the socket
694 * (SS_PROTOREF).
695 *
696 * - The socket is not in a completed connection queue, so a process has been
697 * notified that it is present. If it is removed, the user process may
698 * block in accept() despite select() saying the socket was ready.
699 */
700 void
701 sofree(struct socket *so)
702 {
703 struct protosw *pr = so->so_proto;
704 struct socket *head;
705
706 ACCEPT_LOCK_ASSERT();
707 SOCK_LOCK_ASSERT(so);
708
709 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
710 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
711 SOCK_UNLOCK(so);
712 ACCEPT_UNLOCK();
713 return;
714 }
715
716 head = so->so_head;
717 if (head != NULL) {
718 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
719 (so->so_qstate & SQ_INCOMP) != 0,
720 ("sofree: so_head != NULL, but neither SQ_COMP nor "
721 "SQ_INCOMP"));
722 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
723 (so->so_qstate & SQ_INCOMP) == 0,
724 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
725 TAILQ_REMOVE(&head->so_incomp, so, so_list);
726 head->so_incqlen--;
727 so->so_qstate &= ~SQ_INCOMP;
728 so->so_head = NULL;
729 }
730 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
731 (so->so_qstate & SQ_INCOMP) == 0,
732 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
733 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
734 if (so->so_options & SO_ACCEPTCONN) {
735 KASSERT((TAILQ_EMPTY(&so->so_comp)),
736 ("sofree: so_comp populated"));
737 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
738 ("sofree: so_incomp populated"));
739 }
740 SOCK_UNLOCK(so);
741 ACCEPT_UNLOCK();
742
743 VNET_SO_ASSERT(so);
744 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) {
745 if (pr->pr_domain->dom_family == AF_LOCAL)
746 unp_dispose_so(so);
747 else
748 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
749 }
750 if (pr->pr_usrreqs->pru_detach != NULL)
751 (*pr->pr_usrreqs->pru_detach)(so);
752
753 /*
754 * From this point on, we assume that no other references to this
755 * socket exist anywhere else in the stack. Therefore, no locks need
756 * to be acquired or held.
757 *
758 * We used to do a lot of socket buffer and socket locking here, as
759 * well as invoke sorflush() and perform wakeups. The direct call to
760 * dom_dispose() and sbrelease_internal() are an inlining of what was
761 * necessary from sorflush().
762 *
763 * Notice that the socket buffer and kqueue state are torn down
764 * before calling pru_detach. This means that protocols shold not
765 * assume they can perform socket wakeups, etc, in their detach code.
766 */
767 sbdestroy(&so->so_snd, so);
768 sbdestroy(&so->so_rcv, so);
769 seldrain(&so->so_snd.sb_sel);
770 seldrain(&so->so_rcv.sb_sel);
771 knlist_destroy(&so->so_rcv.sb_sel.si_note);
772 knlist_destroy(&so->so_snd.sb_sel.si_note);
773 sodealloc(so);
774 }
775
776 /*
777 * Close a socket on last file table reference removal. Initiate disconnect
778 * if connected. Free socket when disconnect complete.
779 *
780 * This function will sorele() the socket. Note that soclose() may be called
781 * prior to the ref count reaching zero. The actual socket structure will
782 * not be freed until the ref count reaches zero.
783 */
784 int
785 soclose(struct socket *so)
786 {
787 int error = 0;
788
789 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
790
791 CURVNET_SET(so->so_vnet);
792 funsetown(&so->so_sigio);
793 if (so->so_state & SS_ISCONNECTED) {
794 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
795 error = sodisconnect(so);
796 if (error) {
797 if (error == ENOTCONN)
798 error = 0;
799 goto drop;
800 }
801 }
802 if (so->so_options & SO_LINGER) {
803 if ((so->so_state & SS_ISDISCONNECTING) &&
804 (so->so_state & SS_NBIO))
805 goto drop;
806 while (so->so_state & SS_ISCONNECTED) {
807 error = tsleep(&so->so_timeo,
808 PSOCK | PCATCH, "soclos",
809 so->so_linger * hz);
810 if (error)
811 break;
812 }
813 }
814 }
815
816 drop:
817 if (so->so_proto->pr_usrreqs->pru_close != NULL)
818 (*so->so_proto->pr_usrreqs->pru_close)(so);
819 ACCEPT_LOCK();
820 if (so->so_options & SO_ACCEPTCONN) {
821 struct socket *sp;
822 /*
823 * Prevent new additions to the accept queues due
824 * to ACCEPT_LOCK races while we are draining them.
825 */
826 so->so_options &= ~SO_ACCEPTCONN;
827 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
828 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
829 so->so_incqlen--;
830 sp->so_qstate &= ~SQ_INCOMP;
831 sp->so_head = NULL;
832 ACCEPT_UNLOCK();
833 soabort(sp);
834 ACCEPT_LOCK();
835 }
836 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
837 TAILQ_REMOVE(&so->so_comp, sp, so_list);
838 so->so_qlen--;
839 sp->so_qstate &= ~SQ_COMP;
840 sp->so_head = NULL;
841 ACCEPT_UNLOCK();
842 soabort(sp);
843 ACCEPT_LOCK();
844 }
845 KASSERT((TAILQ_EMPTY(&so->so_comp)),
846 ("%s: so_comp populated", __func__));
847 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
848 ("%s: so_incomp populated", __func__));
849 }
850 SOCK_LOCK(so);
851 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
852 so->so_state |= SS_NOFDREF;
853 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
854 CURVNET_RESTORE();
855 return (error);
856 }
857
858 /*
859 * soabort() is used to abruptly tear down a connection, such as when a
860 * resource limit is reached (listen queue depth exceeded), or if a listen
861 * socket is closed while there are sockets waiting to be accepted.
862 *
863 * This interface is tricky, because it is called on an unreferenced socket,
864 * and must be called only by a thread that has actually removed the socket
865 * from the listen queue it was on, or races with other threads are risked.
866 *
867 * This interface will call into the protocol code, so must not be called
868 * with any socket locks held. Protocols do call it while holding their own
869 * recursible protocol mutexes, but this is something that should be subject
870 * to review in the future.
871 */
872 void
873 soabort(struct socket *so)
874 {
875
876 /*
877 * In as much as is possible, assert that no references to this
878 * socket are held. This is not quite the same as asserting that the
879 * current thread is responsible for arranging for no references, but
880 * is as close as we can get for now.
881 */
882 KASSERT(so->so_count == 0, ("soabort: so_count"));
883 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
884 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
885 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
886 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
887 VNET_SO_ASSERT(so);
888
889 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
890 (*so->so_proto->pr_usrreqs->pru_abort)(so);
891 ACCEPT_LOCK();
892 SOCK_LOCK(so);
893 sofree(so);
894 }
895
896 int
897 soaccept(struct socket *so, struct sockaddr **nam)
898 {
899 int error;
900
901 SOCK_LOCK(so);
902 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
903 so->so_state &= ~SS_NOFDREF;
904 SOCK_UNLOCK(so);
905
906 CURVNET_SET(so->so_vnet);
907 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
908 CURVNET_RESTORE();
909 return (error);
910 }
911
912 int
913 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
914 {
915
916 return (soconnectat(AT_FDCWD, so, nam, td));
917 }
918
919 int
920 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
921 {
922 int error;
923
924 if (so->so_options & SO_ACCEPTCONN)
925 return (EOPNOTSUPP);
926
927 CURVNET_SET(so->so_vnet);
928 /*
929 * If protocol is connection-based, can only connect once.
930 * Otherwise, if connected, try to disconnect first. This allows
931 * user to disconnect by connecting to, e.g., a null address.
932 */
933 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
934 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
935 (error = sodisconnect(so)))) {
936 error = EISCONN;
937 } else {
938 /*
939 * Prevent accumulated error from previous connection from
940 * biting us.
941 */
942 so->so_error = 0;
943 if (fd == AT_FDCWD) {
944 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
945 nam, td);
946 } else {
947 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
948 so, nam, td);
949 }
950 }
951 CURVNET_RESTORE();
952
953 return (error);
954 }
955
956 int
957 soconnect2(struct socket *so1, struct socket *so2)
958 {
959 int error;
960
961 CURVNET_SET(so1->so_vnet);
962 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
963 CURVNET_RESTORE();
964 return (error);
965 }
966
967 int
968 sodisconnect(struct socket *so)
969 {
970 int error;
971
972 if ((so->so_state & SS_ISCONNECTED) == 0)
973 return (ENOTCONN);
974 if (so->so_state & SS_ISDISCONNECTING)
975 return (EALREADY);
976 VNET_SO_ASSERT(so);
977 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
978 return (error);
979 }
980
981 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
982
983 int
984 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
985 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
986 {
987 long space;
988 ssize_t resid;
989 int clen = 0, error, dontroute;
990
991 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
992 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
993 ("sosend_dgram: !PR_ATOMIC"));
994
995 if (uio != NULL)
996 resid = uio->uio_resid;
997 else
998 resid = top->m_pkthdr.len;
999 /*
1000 * In theory resid should be unsigned. However, space must be
1001 * signed, as it might be less than 0 if we over-committed, and we
1002 * must use a signed comparison of space and resid. On the other
1003 * hand, a negative resid causes us to loop sending 0-length
1004 * segments to the protocol.
1005 */
1006 if (resid < 0) {
1007 error = EINVAL;
1008 goto out;
1009 }
1010
1011 dontroute =
1012 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1013 if (td != NULL)
1014 td->td_ru.ru_msgsnd++;
1015 if (control != NULL)
1016 clen = control->m_len;
1017
1018 SOCKBUF_LOCK(&so->so_snd);
1019 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1020 SOCKBUF_UNLOCK(&so->so_snd);
1021 error = EPIPE;
1022 goto out;
1023 }
1024 if (so->so_error) {
1025 error = so->so_error;
1026 so->so_error = 0;
1027 SOCKBUF_UNLOCK(&so->so_snd);
1028 goto out;
1029 }
1030 if ((so->so_state & SS_ISCONNECTED) == 0) {
1031 /*
1032 * `sendto' and `sendmsg' is allowed on a connection-based
1033 * socket if it supports implied connect. Return ENOTCONN if
1034 * not connected and no address is supplied.
1035 */
1036 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1037 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1038 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1039 !(resid == 0 && clen != 0)) {
1040 SOCKBUF_UNLOCK(&so->so_snd);
1041 error = ENOTCONN;
1042 goto out;
1043 }
1044 } else if (addr == NULL) {
1045 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1046 error = ENOTCONN;
1047 else
1048 error = EDESTADDRREQ;
1049 SOCKBUF_UNLOCK(&so->so_snd);
1050 goto out;
1051 }
1052 }
1053
1054 /*
1055 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1056 * problem and need fixing.
1057 */
1058 space = sbspace(&so->so_snd);
1059 if (flags & MSG_OOB)
1060 space += 1024;
1061 space -= clen;
1062 SOCKBUF_UNLOCK(&so->so_snd);
1063 if (resid > space) {
1064 error = EMSGSIZE;
1065 goto out;
1066 }
1067 if (uio == NULL) {
1068 resid = 0;
1069 if (flags & MSG_EOR)
1070 top->m_flags |= M_EOR;
1071 } else {
1072 /*
1073 * Copy the data from userland into a mbuf chain.
1074 * If no data is to be copied in, a single empty mbuf
1075 * is returned.
1076 */
1077 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1078 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1079 if (top == NULL) {
1080 error = EFAULT; /* only possible error */
1081 goto out;
1082 }
1083 space -= resid - uio->uio_resid;
1084 resid = uio->uio_resid;
1085 }
1086 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1087 /*
1088 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1089 * than with.
1090 */
1091 if (dontroute) {
1092 SOCK_LOCK(so);
1093 so->so_options |= SO_DONTROUTE;
1094 SOCK_UNLOCK(so);
1095 }
1096 /*
1097 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1098 * of date. We could have received a reset packet in an interrupt or
1099 * maybe we slept while doing page faults in uiomove() etc. We could
1100 * probably recheck again inside the locking protection here, but
1101 * there are probably other places that this also happens. We must
1102 * rethink this.
1103 */
1104 VNET_SO_ASSERT(so);
1105 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1106 (flags & MSG_OOB) ? PRUS_OOB :
1107 /*
1108 * If the user set MSG_EOF, the protocol understands this flag and
1109 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1110 */
1111 ((flags & MSG_EOF) &&
1112 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1113 (resid <= 0)) ?
1114 PRUS_EOF :
1115 /* If there is more to send set PRUS_MORETOCOME */
1116 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1117 top, addr, control, td);
1118 if (dontroute) {
1119 SOCK_LOCK(so);
1120 so->so_options &= ~SO_DONTROUTE;
1121 SOCK_UNLOCK(so);
1122 }
1123 clen = 0;
1124 control = NULL;
1125 top = NULL;
1126 out:
1127 if (top != NULL)
1128 m_freem(top);
1129 if (control != NULL)
1130 m_freem(control);
1131 return (error);
1132 }
1133
1134 /*
1135 * Send on a socket. If send must go all at once and message is larger than
1136 * send buffering, then hard error. Lock against other senders. If must go
1137 * all at once and not enough room now, then inform user that this would
1138 * block and do nothing. Otherwise, if nonblocking, send as much as
1139 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1140 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1141 * in mbuf chain must be small enough to send all at once.
1142 *
1143 * Returns nonzero on error, timeout or signal; callers must check for short
1144 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1145 * on return.
1146 */
1147 int
1148 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1149 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1150 {
1151 long space;
1152 ssize_t resid;
1153 int clen = 0, error, dontroute;
1154 int atomic = sosendallatonce(so) || top;
1155
1156 if (uio != NULL)
1157 resid = uio->uio_resid;
1158 else
1159 resid = top->m_pkthdr.len;
1160 /*
1161 * In theory resid should be unsigned. However, space must be
1162 * signed, as it might be less than 0 if we over-committed, and we
1163 * must use a signed comparison of space and resid. On the other
1164 * hand, a negative resid causes us to loop sending 0-length
1165 * segments to the protocol.
1166 *
1167 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1168 * type sockets since that's an error.
1169 */
1170 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1171 error = EINVAL;
1172 goto out;
1173 }
1174
1175 dontroute =
1176 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1177 (so->so_proto->pr_flags & PR_ATOMIC);
1178 if (td != NULL)
1179 td->td_ru.ru_msgsnd++;
1180 if (control != NULL)
1181 clen = control->m_len;
1182
1183 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1184 if (error)
1185 goto out;
1186
1187 restart:
1188 do {
1189 SOCKBUF_LOCK(&so->so_snd);
1190 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1191 SOCKBUF_UNLOCK(&so->so_snd);
1192 error = EPIPE;
1193 goto release;
1194 }
1195 if (so->so_error) {
1196 error = so->so_error;
1197 so->so_error = 0;
1198 SOCKBUF_UNLOCK(&so->so_snd);
1199 goto release;
1200 }
1201 if ((so->so_state & SS_ISCONNECTED) == 0) {
1202 /*
1203 * `sendto' and `sendmsg' is allowed on a connection-
1204 * based socket if it supports implied connect.
1205 * Return ENOTCONN if not connected and no address is
1206 * supplied.
1207 */
1208 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1209 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1210 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1211 !(resid == 0 && clen != 0)) {
1212 SOCKBUF_UNLOCK(&so->so_snd);
1213 error = ENOTCONN;
1214 goto release;
1215 }
1216 } else if (addr == NULL) {
1217 SOCKBUF_UNLOCK(&so->so_snd);
1218 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1219 error = ENOTCONN;
1220 else
1221 error = EDESTADDRREQ;
1222 goto release;
1223 }
1224 }
1225 space = sbspace(&so->so_snd);
1226 if (flags & MSG_OOB)
1227 space += 1024;
1228 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1229 clen > so->so_snd.sb_hiwat) {
1230 SOCKBUF_UNLOCK(&so->so_snd);
1231 error = EMSGSIZE;
1232 goto release;
1233 }
1234 if (space < resid + clen &&
1235 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1236 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1237 SOCKBUF_UNLOCK(&so->so_snd);
1238 error = EWOULDBLOCK;
1239 goto release;
1240 }
1241 error = sbwait(&so->so_snd);
1242 SOCKBUF_UNLOCK(&so->so_snd);
1243 if (error)
1244 goto release;
1245 goto restart;
1246 }
1247 SOCKBUF_UNLOCK(&so->so_snd);
1248 space -= clen;
1249 do {
1250 if (uio == NULL) {
1251 resid = 0;
1252 if (flags & MSG_EOR)
1253 top->m_flags |= M_EOR;
1254 } else {
1255 /*
1256 * Copy the data from userland into a mbuf
1257 * chain. If no data is to be copied in,
1258 * a single empty mbuf is returned.
1259 */
1260 top = m_uiotombuf(uio, M_WAITOK, space,
1261 (atomic ? max_hdr : 0),
1262 (atomic ? M_PKTHDR : 0) |
1263 ((flags & MSG_EOR) ? M_EOR : 0));
1264 if (top == NULL) {
1265 error = EFAULT; /* only possible error */
1266 goto release;
1267 }
1268 space -= resid - uio->uio_resid;
1269 resid = uio->uio_resid;
1270 }
1271 if (dontroute) {
1272 SOCK_LOCK(so);
1273 so->so_options |= SO_DONTROUTE;
1274 SOCK_UNLOCK(so);
1275 }
1276 /*
1277 * XXX all the SBS_CANTSENDMORE checks previously
1278 * done could be out of date. We could have received
1279 * a reset packet in an interrupt or maybe we slept
1280 * while doing page faults in uiomove() etc. We
1281 * could probably recheck again inside the locking
1282 * protection here, but there are probably other
1283 * places that this also happens. We must rethink
1284 * this.
1285 */
1286 VNET_SO_ASSERT(so);
1287 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1288 (flags & MSG_OOB) ? PRUS_OOB :
1289 /*
1290 * If the user set MSG_EOF, the protocol understands
1291 * this flag and nothing left to send then use
1292 * PRU_SEND_EOF instead of PRU_SEND.
1293 */
1294 ((flags & MSG_EOF) &&
1295 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1296 (resid <= 0)) ?
1297 PRUS_EOF :
1298 /* If there is more to send set PRUS_MORETOCOME. */
1299 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1300 top, addr, control, td);
1301 if (dontroute) {
1302 SOCK_LOCK(so);
1303 so->so_options &= ~SO_DONTROUTE;
1304 SOCK_UNLOCK(so);
1305 }
1306 clen = 0;
1307 control = NULL;
1308 top = NULL;
1309 if (error)
1310 goto release;
1311 } while (resid && space > 0);
1312 } while (resid);
1313
1314 release:
1315 sbunlock(&so->so_snd);
1316 out:
1317 if (top != NULL)
1318 m_freem(top);
1319 if (control != NULL)
1320 m_freem(control);
1321 return (error);
1322 }
1323
1324 int
1325 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1326 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1327 {
1328 int error;
1329
1330 CURVNET_SET(so->so_vnet);
1331 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1332 control, flags, td);
1333 CURVNET_RESTORE();
1334 return (error);
1335 }
1336
1337 /*
1338 * The part of soreceive() that implements reading non-inline out-of-band
1339 * data from a socket. For more complete comments, see soreceive(), from
1340 * which this code originated.
1341 *
1342 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1343 * unable to return an mbuf chain to the caller.
1344 */
1345 static int
1346 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1347 {
1348 struct protosw *pr = so->so_proto;
1349 struct mbuf *m;
1350 int error;
1351
1352 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1353 VNET_SO_ASSERT(so);
1354
1355 m = m_get(M_WAITOK, MT_DATA);
1356 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1357 if (error)
1358 goto bad;
1359 do {
1360 error = uiomove(mtod(m, void *),
1361 (int) min(uio->uio_resid, m->m_len), uio);
1362 m = m_free(m);
1363 } while (uio->uio_resid && error == 0 && m);
1364 bad:
1365 if (m != NULL)
1366 m_freem(m);
1367 return (error);
1368 }
1369
1370 /*
1371 * Following replacement or removal of the first mbuf on the first mbuf chain
1372 * of a socket buffer, push necessary state changes back into the socket
1373 * buffer so that other consumers see the values consistently. 'nextrecord'
1374 * is the callers locally stored value of the original value of
1375 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1376 * NOTE: 'nextrecord' may be NULL.
1377 */
1378 static __inline void
1379 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1380 {
1381
1382 SOCKBUF_LOCK_ASSERT(sb);
1383 /*
1384 * First, update for the new value of nextrecord. If necessary, make
1385 * it the first record.
1386 */
1387 if (sb->sb_mb != NULL)
1388 sb->sb_mb->m_nextpkt = nextrecord;
1389 else
1390 sb->sb_mb = nextrecord;
1391
1392 /*
1393 * Now update any dependent socket buffer fields to reflect the new
1394 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1395 * addition of a second clause that takes care of the case where
1396 * sb_mb has been updated, but remains the last record.
1397 */
1398 if (sb->sb_mb == NULL) {
1399 sb->sb_mbtail = NULL;
1400 sb->sb_lastrecord = NULL;
1401 } else if (sb->sb_mb->m_nextpkt == NULL)
1402 sb->sb_lastrecord = sb->sb_mb;
1403 }
1404
1405 /*
1406 * Implement receive operations on a socket. We depend on the way that
1407 * records are added to the sockbuf by sbappend. In particular, each record
1408 * (mbufs linked through m_next) must begin with an address if the protocol
1409 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1410 * data, and then zero or more mbufs of data. In order to allow parallelism
1411 * between network receive and copying to user space, as well as avoid
1412 * sleeping with a mutex held, we release the socket buffer mutex during the
1413 * user space copy. Although the sockbuf is locked, new data may still be
1414 * appended, and thus we must maintain consistency of the sockbuf during that
1415 * time.
1416 *
1417 * The caller may receive the data as a single mbuf chain by supplying an
1418 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1419 * the count in uio_resid.
1420 */
1421 int
1422 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1423 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1424 {
1425 struct mbuf *m, **mp;
1426 int flags, error, offset;
1427 ssize_t len;
1428 struct protosw *pr = so->so_proto;
1429 struct mbuf *nextrecord;
1430 int moff, type = 0;
1431 ssize_t orig_resid = uio->uio_resid;
1432
1433 mp = mp0;
1434 if (psa != NULL)
1435 *psa = NULL;
1436 if (controlp != NULL)
1437 *controlp = NULL;
1438 if (flagsp != NULL)
1439 flags = *flagsp &~ MSG_EOR;
1440 else
1441 flags = 0;
1442 if (flags & MSG_OOB)
1443 return (soreceive_rcvoob(so, uio, flags));
1444 if (mp != NULL)
1445 *mp = NULL;
1446 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1447 && uio->uio_resid) {
1448 VNET_SO_ASSERT(so);
1449 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1450 }
1451
1452 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1453 if (error)
1454 return (error);
1455
1456 restart:
1457 SOCKBUF_LOCK(&so->so_rcv);
1458 m = so->so_rcv.sb_mb;
1459 /*
1460 * If we have less data than requested, block awaiting more (subject
1461 * to any timeout) if:
1462 * 1. the current count is less than the low water mark, or
1463 * 2. MSG_DONTWAIT is not set
1464 */
1465 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1466 so->so_rcv.sb_cc < uio->uio_resid) &&
1467 so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1468 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1469 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1470 ("receive: m == %p so->so_rcv.sb_cc == %u",
1471 m, so->so_rcv.sb_cc));
1472 if (so->so_error) {
1473 if (m != NULL)
1474 goto dontblock;
1475 error = so->so_error;
1476 if ((flags & MSG_PEEK) == 0)
1477 so->so_error = 0;
1478 SOCKBUF_UNLOCK(&so->so_rcv);
1479 goto release;
1480 }
1481 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1482 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1483 if (m == NULL) {
1484 SOCKBUF_UNLOCK(&so->so_rcv);
1485 goto release;
1486 } else
1487 goto dontblock;
1488 }
1489 for (; m != NULL; m = m->m_next)
1490 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1491 m = so->so_rcv.sb_mb;
1492 goto dontblock;
1493 }
1494 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1495 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1496 SOCKBUF_UNLOCK(&so->so_rcv);
1497 error = ENOTCONN;
1498 goto release;
1499 }
1500 if (uio->uio_resid == 0) {
1501 SOCKBUF_UNLOCK(&so->so_rcv);
1502 goto release;
1503 }
1504 if ((so->so_state & SS_NBIO) ||
1505 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1506 SOCKBUF_UNLOCK(&so->so_rcv);
1507 error = EWOULDBLOCK;
1508 goto release;
1509 }
1510 SBLASTRECORDCHK(&so->so_rcv);
1511 SBLASTMBUFCHK(&so->so_rcv);
1512 error = sbwait(&so->so_rcv);
1513 SOCKBUF_UNLOCK(&so->so_rcv);
1514 if (error)
1515 goto release;
1516 goto restart;
1517 }
1518 dontblock:
1519 /*
1520 * From this point onward, we maintain 'nextrecord' as a cache of the
1521 * pointer to the next record in the socket buffer. We must keep the
1522 * various socket buffer pointers and local stack versions of the
1523 * pointers in sync, pushing out modifications before dropping the
1524 * socket buffer mutex, and re-reading them when picking it up.
1525 *
1526 * Otherwise, we will race with the network stack appending new data
1527 * or records onto the socket buffer by using inconsistent/stale
1528 * versions of the field, possibly resulting in socket buffer
1529 * corruption.
1530 *
1531 * By holding the high-level sblock(), we prevent simultaneous
1532 * readers from pulling off the front of the socket buffer.
1533 */
1534 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1535 if (uio->uio_td)
1536 uio->uio_td->td_ru.ru_msgrcv++;
1537 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1538 SBLASTRECORDCHK(&so->so_rcv);
1539 SBLASTMBUFCHK(&so->so_rcv);
1540 nextrecord = m->m_nextpkt;
1541 if (pr->pr_flags & PR_ADDR) {
1542 KASSERT(m->m_type == MT_SONAME,
1543 ("m->m_type == %d", m->m_type));
1544 orig_resid = 0;
1545 if (psa != NULL)
1546 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1547 M_NOWAIT);
1548 if (flags & MSG_PEEK) {
1549 m = m->m_next;
1550 } else {
1551 sbfree(&so->so_rcv, m);
1552 so->so_rcv.sb_mb = m_free(m);
1553 m = so->so_rcv.sb_mb;
1554 sockbuf_pushsync(&so->so_rcv, nextrecord);
1555 }
1556 }
1557
1558 /*
1559 * Process one or more MT_CONTROL mbufs present before any data mbufs
1560 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1561 * just copy the data; if !MSG_PEEK, we call into the protocol to
1562 * perform externalization (or freeing if controlp == NULL).
1563 */
1564 if (m != NULL && m->m_type == MT_CONTROL) {
1565 struct mbuf *cm = NULL, *cmn;
1566 struct mbuf **cme = &cm;
1567
1568 do {
1569 if (flags & MSG_PEEK) {
1570 if (controlp != NULL) {
1571 *controlp = m_copy(m, 0, m->m_len);
1572 controlp = &(*controlp)->m_next;
1573 }
1574 m = m->m_next;
1575 } else {
1576 sbfree(&so->so_rcv, m);
1577 so->so_rcv.sb_mb = m->m_next;
1578 m->m_next = NULL;
1579 *cme = m;
1580 cme = &(*cme)->m_next;
1581 m = so->so_rcv.sb_mb;
1582 }
1583 } while (m != NULL && m->m_type == MT_CONTROL);
1584 if ((flags & MSG_PEEK) == 0)
1585 sockbuf_pushsync(&so->so_rcv, nextrecord);
1586 while (cm != NULL) {
1587 cmn = cm->m_next;
1588 cm->m_next = NULL;
1589 if (pr->pr_domain->dom_externalize != NULL) {
1590 SOCKBUF_UNLOCK(&so->so_rcv);
1591 VNET_SO_ASSERT(so);
1592 error = (*pr->pr_domain->dom_externalize)
1593 (cm, controlp, flags);
1594 SOCKBUF_LOCK(&so->so_rcv);
1595 } else if (controlp != NULL)
1596 *controlp = cm;
1597 else
1598 m_freem(cm);
1599 if (controlp != NULL) {
1600 orig_resid = 0;
1601 while (*controlp != NULL)
1602 controlp = &(*controlp)->m_next;
1603 }
1604 cm = cmn;
1605 }
1606 if (m != NULL)
1607 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1608 else
1609 nextrecord = so->so_rcv.sb_mb;
1610 orig_resid = 0;
1611 }
1612 if (m != NULL) {
1613 if ((flags & MSG_PEEK) == 0) {
1614 KASSERT(m->m_nextpkt == nextrecord,
1615 ("soreceive: post-control, nextrecord !sync"));
1616 if (nextrecord == NULL) {
1617 KASSERT(so->so_rcv.sb_mb == m,
1618 ("soreceive: post-control, sb_mb!=m"));
1619 KASSERT(so->so_rcv.sb_lastrecord == m,
1620 ("soreceive: post-control, lastrecord!=m"));
1621 }
1622 }
1623 type = m->m_type;
1624 if (type == MT_OOBDATA)
1625 flags |= MSG_OOB;
1626 } else {
1627 if ((flags & MSG_PEEK) == 0) {
1628 KASSERT(so->so_rcv.sb_mb == nextrecord,
1629 ("soreceive: sb_mb != nextrecord"));
1630 if (so->so_rcv.sb_mb == NULL) {
1631 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1632 ("soreceive: sb_lastercord != NULL"));
1633 }
1634 }
1635 }
1636 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1637 SBLASTRECORDCHK(&so->so_rcv);
1638 SBLASTMBUFCHK(&so->so_rcv);
1639
1640 /*
1641 * Now continue to read any data mbufs off of the head of the socket
1642 * buffer until the read request is satisfied. Note that 'type' is
1643 * used to store the type of any mbuf reads that have happened so far
1644 * such that soreceive() can stop reading if the type changes, which
1645 * causes soreceive() to return only one of regular data and inline
1646 * out-of-band data in a single socket receive operation.
1647 */
1648 moff = 0;
1649 offset = 0;
1650 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1651 /*
1652 * If the type of mbuf has changed since the last mbuf
1653 * examined ('type'), end the receive operation.
1654 */
1655 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1656 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1657 if (type != m->m_type)
1658 break;
1659 } else if (type == MT_OOBDATA)
1660 break;
1661 else
1662 KASSERT(m->m_type == MT_DATA,
1663 ("m->m_type == %d", m->m_type));
1664 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1665 len = uio->uio_resid;
1666 if (so->so_oobmark && len > so->so_oobmark - offset)
1667 len = so->so_oobmark - offset;
1668 if (len > m->m_len - moff)
1669 len = m->m_len - moff;
1670 /*
1671 * If mp is set, just pass back the mbufs. Otherwise copy
1672 * them out via the uio, then free. Sockbuf must be
1673 * consistent here (points to current mbuf, it points to next
1674 * record) when we drop priority; we must note any additions
1675 * to the sockbuf when we block interrupts again.
1676 */
1677 if (mp == NULL) {
1678 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1679 SBLASTRECORDCHK(&so->so_rcv);
1680 SBLASTMBUFCHK(&so->so_rcv);
1681 SOCKBUF_UNLOCK(&so->so_rcv);
1682 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1683 SOCKBUF_LOCK(&so->so_rcv);
1684 if (error) {
1685 /*
1686 * The MT_SONAME mbuf has already been removed
1687 * from the record, so it is necessary to
1688 * remove the data mbufs, if any, to preserve
1689 * the invariant in the case of PR_ADDR that
1690 * requires MT_SONAME mbufs at the head of
1691 * each record.
1692 */
1693 if (m && pr->pr_flags & PR_ATOMIC &&
1694 ((flags & MSG_PEEK) == 0))
1695 (void)sbdroprecord_locked(&so->so_rcv);
1696 SOCKBUF_UNLOCK(&so->so_rcv);
1697 goto release;
1698 }
1699 } else
1700 uio->uio_resid -= len;
1701 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1702 if (len == m->m_len - moff) {
1703 if (m->m_flags & M_EOR)
1704 flags |= MSG_EOR;
1705 if (flags & MSG_PEEK) {
1706 m = m->m_next;
1707 moff = 0;
1708 } else {
1709 nextrecord = m->m_nextpkt;
1710 sbfree(&so->so_rcv, m);
1711 if (mp != NULL) {
1712 m->m_nextpkt = NULL;
1713 *mp = m;
1714 mp = &m->m_next;
1715 so->so_rcv.sb_mb = m = m->m_next;
1716 *mp = NULL;
1717 } else {
1718 so->so_rcv.sb_mb = m_free(m);
1719 m = so->so_rcv.sb_mb;
1720 }
1721 sockbuf_pushsync(&so->so_rcv, nextrecord);
1722 SBLASTRECORDCHK(&so->so_rcv);
1723 SBLASTMBUFCHK(&so->so_rcv);
1724 }
1725 } else {
1726 if (flags & MSG_PEEK)
1727 moff += len;
1728 else {
1729 if (mp != NULL) {
1730 int copy_flag;
1731
1732 if (flags & MSG_DONTWAIT)
1733 copy_flag = M_NOWAIT;
1734 else
1735 copy_flag = M_WAIT;
1736 if (copy_flag == M_WAITOK)
1737 SOCKBUF_UNLOCK(&so->so_rcv);
1738 *mp = m_copym(m, 0, len, copy_flag);
1739 if (copy_flag == M_WAITOK)
1740 SOCKBUF_LOCK(&so->so_rcv);
1741 if (*mp == NULL) {
1742 /*
1743 * m_copym() couldn't
1744 * allocate an mbuf. Adjust
1745 * uio_resid back (it was
1746 * adjusted down by len
1747 * bytes, which we didn't end
1748 * up "copying" over).
1749 */
1750 uio->uio_resid += len;
1751 break;
1752 }
1753 }
1754 m->m_data += len;
1755 m->m_len -= len;
1756 so->so_rcv.sb_cc -= len;
1757 }
1758 }
1759 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1760 if (so->so_oobmark) {
1761 if ((flags & MSG_PEEK) == 0) {
1762 so->so_oobmark -= len;
1763 if (so->so_oobmark == 0) {
1764 so->so_rcv.sb_state |= SBS_RCVATMARK;
1765 break;
1766 }
1767 } else {
1768 offset += len;
1769 if (offset == so->so_oobmark)
1770 break;
1771 }
1772 }
1773 if (flags & MSG_EOR)
1774 break;
1775 /*
1776 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1777 * must not quit until "uio->uio_resid == 0" or an error
1778 * termination. If a signal/timeout occurs, return with a
1779 * short count but without error. Keep sockbuf locked
1780 * against other readers.
1781 */
1782 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1783 !sosendallatonce(so) && nextrecord == NULL) {
1784 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1785 if (so->so_error ||
1786 so->so_rcv.sb_state & SBS_CANTRCVMORE)
1787 break;
1788 /*
1789 * Notify the protocol that some data has been
1790 * drained before blocking.
1791 */
1792 if (pr->pr_flags & PR_WANTRCVD) {
1793 SOCKBUF_UNLOCK(&so->so_rcv);
1794 VNET_SO_ASSERT(so);
1795 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1796 SOCKBUF_LOCK(&so->so_rcv);
1797 }
1798 SBLASTRECORDCHK(&so->so_rcv);
1799 SBLASTMBUFCHK(&so->so_rcv);
1800 /*
1801 * We could receive some data while was notifying
1802 * the protocol. Skip blocking in this case.
1803 */
1804 if (so->so_rcv.sb_mb == NULL) {
1805 error = sbwait(&so->so_rcv);
1806 if (error) {
1807 SOCKBUF_UNLOCK(&so->so_rcv);
1808 goto release;
1809 }
1810 }
1811 m = so->so_rcv.sb_mb;
1812 if (m != NULL)
1813 nextrecord = m->m_nextpkt;
1814 }
1815 }
1816
1817 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1818 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1819 flags |= MSG_TRUNC;
1820 if ((flags & MSG_PEEK) == 0)
1821 (void) sbdroprecord_locked(&so->so_rcv);
1822 }
1823 if ((flags & MSG_PEEK) == 0) {
1824 if (m == NULL) {
1825 /*
1826 * First part is an inline SB_EMPTY_FIXUP(). Second
1827 * part makes sure sb_lastrecord is up-to-date if
1828 * there is still data in the socket buffer.
1829 */
1830 so->so_rcv.sb_mb = nextrecord;
1831 if (so->so_rcv.sb_mb == NULL) {
1832 so->so_rcv.sb_mbtail = NULL;
1833 so->so_rcv.sb_lastrecord = NULL;
1834 } else if (nextrecord->m_nextpkt == NULL)
1835 so->so_rcv.sb_lastrecord = nextrecord;
1836 }
1837 SBLASTRECORDCHK(&so->so_rcv);
1838 SBLASTMBUFCHK(&so->so_rcv);
1839 /*
1840 * If soreceive() is being done from the socket callback,
1841 * then don't need to generate ACK to peer to update window,
1842 * since ACK will be generated on return to TCP.
1843 */
1844 if (!(flags & MSG_SOCALLBCK) &&
1845 (pr->pr_flags & PR_WANTRCVD)) {
1846 SOCKBUF_UNLOCK(&so->so_rcv);
1847 VNET_SO_ASSERT(so);
1848 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1849 SOCKBUF_LOCK(&so->so_rcv);
1850 }
1851 }
1852 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1853 if (orig_resid == uio->uio_resid && orig_resid &&
1854 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1855 SOCKBUF_UNLOCK(&so->so_rcv);
1856 goto restart;
1857 }
1858 SOCKBUF_UNLOCK(&so->so_rcv);
1859
1860 if (flagsp != NULL)
1861 *flagsp |= flags;
1862 release:
1863 sbunlock(&so->so_rcv);
1864 return (error);
1865 }
1866
1867 /*
1868 * Optimized version of soreceive() for stream (TCP) sockets.
1869 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1870 */
1871 int
1872 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1873 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1874 {
1875 int len = 0, error = 0, flags, oresid;
1876 struct sockbuf *sb;
1877 struct mbuf *m, *n = NULL;
1878
1879 /* We only do stream sockets. */
1880 if (so->so_type != SOCK_STREAM)
1881 return (EINVAL);
1882 if (psa != NULL)
1883 *psa = NULL;
1884 if (controlp != NULL)
1885 return (EINVAL);
1886 if (flagsp != NULL)
1887 flags = *flagsp &~ MSG_EOR;
1888 else
1889 flags = 0;
1890 if (flags & MSG_OOB)
1891 return (soreceive_rcvoob(so, uio, flags));
1892 if (mp0 != NULL)
1893 *mp0 = NULL;
1894
1895 sb = &so->so_rcv;
1896
1897 /* Prevent other readers from entering the socket. */
1898 error = sblock(sb, SBLOCKWAIT(flags));
1899 if (error)
1900 goto out;
1901 SOCKBUF_LOCK(sb);
1902
1903 /* Easy one, no space to copyout anything. */
1904 if (uio->uio_resid == 0) {
1905 error = EINVAL;
1906 goto out;
1907 }
1908 oresid = uio->uio_resid;
1909
1910 /* We will never ever get anything unless we are or were connected. */
1911 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1912 error = ENOTCONN;
1913 goto out;
1914 }
1915
1916 restart:
1917 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1918
1919 /* Abort if socket has reported problems. */
1920 if (so->so_error) {
1921 if (sb->sb_cc > 0)
1922 goto deliver;
1923 if (oresid > uio->uio_resid)
1924 goto out;
1925 error = so->so_error;
1926 if (!(flags & MSG_PEEK))
1927 so->so_error = 0;
1928 goto out;
1929 }
1930
1931 /* Door is closed. Deliver what is left, if any. */
1932 if (sb->sb_state & SBS_CANTRCVMORE) {
1933 if (sb->sb_cc > 0)
1934 goto deliver;
1935 else
1936 goto out;
1937 }
1938
1939 /* Socket buffer is empty and we shall not block. */
1940 if (sb->sb_cc == 0 &&
1941 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1942 error = EAGAIN;
1943 goto out;
1944 }
1945
1946 /* Socket buffer got some data that we shall deliver now. */
1947 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1948 ((so->so_state & SS_NBIO) ||
1949 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1950 sb->sb_cc >= sb->sb_lowat ||
1951 sb->sb_cc >= uio->uio_resid ||
1952 sb->sb_cc >= sb->sb_hiwat) ) {
1953 goto deliver;
1954 }
1955
1956 /* On MSG_WAITALL we must wait until all data or error arrives. */
1957 if ((flags & MSG_WAITALL) &&
1958 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
1959 goto deliver;
1960
1961 /*
1962 * Wait and block until (more) data comes in.
1963 * NB: Drops the sockbuf lock during wait.
1964 */
1965 error = sbwait(sb);
1966 if (error)
1967 goto out;
1968 goto restart;
1969
1970 deliver:
1971 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1972 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1973 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1974
1975 /* Statistics. */
1976 if (uio->uio_td)
1977 uio->uio_td->td_ru.ru_msgrcv++;
1978
1979 /* Fill uio until full or current end of socket buffer is reached. */
1980 len = min(uio->uio_resid, sb->sb_cc);
1981 if (mp0 != NULL) {
1982 /* Dequeue as many mbufs as possible. */
1983 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1984 if (*mp0 == NULL)
1985 *mp0 = sb->sb_mb;
1986 else
1987 m_cat(*mp0, sb->sb_mb);
1988 for (m = sb->sb_mb;
1989 m != NULL && m->m_len <= len;
1990 m = m->m_next) {
1991 len -= m->m_len;
1992 uio->uio_resid -= m->m_len;
1993 sbfree(sb, m);
1994 n = m;
1995 }
1996 n->m_next = NULL;
1997 sb->sb_mb = m;
1998 sb->sb_lastrecord = sb->sb_mb;
1999 if (sb->sb_mb == NULL)
2000 SB_EMPTY_FIXUP(sb);
2001 }
2002 /* Copy the remainder. */
2003 if (len > 0) {
2004 KASSERT(sb->sb_mb != NULL,
2005 ("%s: len > 0 && sb->sb_mb empty", __func__));
2006
2007 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2008 if (m == NULL)
2009 len = 0; /* Don't flush data from sockbuf. */
2010 else
2011 uio->uio_resid -= len;
2012 if (*mp0 != NULL)
2013 m_cat(*mp0, m);
2014 else
2015 *mp0 = m;
2016 if (*mp0 == NULL) {
2017 error = ENOBUFS;
2018 goto out;
2019 }
2020 }
2021 } else {
2022 /* NB: Must unlock socket buffer as uiomove may sleep. */
2023 SOCKBUF_UNLOCK(sb);
2024 error = m_mbuftouio(uio, sb->sb_mb, len);
2025 SOCKBUF_LOCK(sb);
2026 if (error)
2027 goto out;
2028 }
2029 SBLASTRECORDCHK(sb);
2030 SBLASTMBUFCHK(sb);
2031
2032 /*
2033 * Remove the delivered data from the socket buffer unless we
2034 * were only peeking.
2035 */
2036 if (!(flags & MSG_PEEK)) {
2037 if (len > 0)
2038 sbdrop_locked(sb, len);
2039
2040 /* Notify protocol that we drained some data. */
2041 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2042 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2043 !(flags & MSG_SOCALLBCK))) {
2044 SOCKBUF_UNLOCK(sb);
2045 VNET_SO_ASSERT(so);
2046 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2047 SOCKBUF_LOCK(sb);
2048 }
2049 }
2050
2051 /*
2052 * For MSG_WAITALL we may have to loop again and wait for
2053 * more data to come in.
2054 */
2055 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2056 goto restart;
2057 out:
2058 SOCKBUF_LOCK_ASSERT(sb);
2059 SBLASTRECORDCHK(sb);
2060 SBLASTMBUFCHK(sb);
2061 SOCKBUF_UNLOCK(sb);
2062 sbunlock(sb);
2063 return (error);
2064 }
2065
2066 /*
2067 * Optimized version of soreceive() for simple datagram cases from userspace.
2068 * Unlike in the stream case, we're able to drop a datagram if copyout()
2069 * fails, and because we handle datagrams atomically, we don't need to use a
2070 * sleep lock to prevent I/O interlacing.
2071 */
2072 int
2073 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2074 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2075 {
2076 struct mbuf *m, *m2;
2077 int flags, error;
2078 ssize_t len;
2079 struct protosw *pr = so->so_proto;
2080 struct mbuf *nextrecord;
2081
2082 if (psa != NULL)
2083 *psa = NULL;
2084 if (controlp != NULL)
2085 *controlp = NULL;
2086 if (flagsp != NULL)
2087 flags = *flagsp &~ MSG_EOR;
2088 else
2089 flags = 0;
2090
2091 /*
2092 * For any complicated cases, fall back to the full
2093 * soreceive_generic().
2094 */
2095 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2096 return (soreceive_generic(so, psa, uio, mp0, controlp,
2097 flagsp));
2098
2099 /*
2100 * Enforce restrictions on use.
2101 */
2102 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2103 ("soreceive_dgram: wantrcvd"));
2104 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2105 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2106 ("soreceive_dgram: SBS_RCVATMARK"));
2107 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2108 ("soreceive_dgram: P_CONNREQUIRED"));
2109
2110 /*
2111 * Loop blocking while waiting for a datagram.
2112 */
2113 SOCKBUF_LOCK(&so->so_rcv);
2114 while ((m = so->so_rcv.sb_mb) == NULL) {
2115 KASSERT(so->so_rcv.sb_cc == 0,
2116 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2117 so->so_rcv.sb_cc));
2118 if (so->so_error) {
2119 error = so->so_error;
2120 so->so_error = 0;
2121 SOCKBUF_UNLOCK(&so->so_rcv);
2122 return (error);
2123 }
2124 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2125 uio->uio_resid == 0) {
2126 SOCKBUF_UNLOCK(&so->so_rcv);
2127 return (0);
2128 }
2129 if ((so->so_state & SS_NBIO) ||
2130 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2131 SOCKBUF_UNLOCK(&so->so_rcv);
2132 return (EWOULDBLOCK);
2133 }
2134 SBLASTRECORDCHK(&so->so_rcv);
2135 SBLASTMBUFCHK(&so->so_rcv);
2136 error = sbwait(&so->so_rcv);
2137 if (error) {
2138 SOCKBUF_UNLOCK(&so->so_rcv);
2139 return (error);
2140 }
2141 }
2142 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2143
2144 if (uio->uio_td)
2145 uio->uio_td->td_ru.ru_msgrcv++;
2146 SBLASTRECORDCHK(&so->so_rcv);
2147 SBLASTMBUFCHK(&so->so_rcv);
2148 nextrecord = m->m_nextpkt;
2149 if (nextrecord == NULL) {
2150 KASSERT(so->so_rcv.sb_lastrecord == m,
2151 ("soreceive_dgram: lastrecord != m"));
2152 }
2153
2154 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2155 ("soreceive_dgram: m_nextpkt != nextrecord"));
2156
2157 /*
2158 * Pull 'm' and its chain off the front of the packet queue.
2159 */
2160 so->so_rcv.sb_mb = NULL;
2161 sockbuf_pushsync(&so->so_rcv, nextrecord);
2162
2163 /*
2164 * Walk 'm's chain and free that many bytes from the socket buffer.
2165 */
2166 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2167 sbfree(&so->so_rcv, m2);
2168
2169 /*
2170 * Do a few last checks before we let go of the lock.
2171 */
2172 SBLASTRECORDCHK(&so->so_rcv);
2173 SBLASTMBUFCHK(&so->so_rcv);
2174 SOCKBUF_UNLOCK(&so->so_rcv);
2175
2176 if (pr->pr_flags & PR_ADDR) {
2177 KASSERT(m->m_type == MT_SONAME,
2178 ("m->m_type == %d", m->m_type));
2179 if (psa != NULL)
2180 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2181 M_NOWAIT);
2182 m = m_free(m);
2183 }
2184 if (m == NULL) {
2185 /* XXXRW: Can this happen? */
2186 return (0);
2187 }
2188
2189 /*
2190 * Packet to copyout() is now in 'm' and it is disconnected from the
2191 * queue.
2192 *
2193 * Process one or more MT_CONTROL mbufs present before any data mbufs
2194 * in the first mbuf chain on the socket buffer. We call into the
2195 * protocol to perform externalization (or freeing if controlp ==
2196 * NULL). In some cases there can be only MT_CONTROL mbufs without
2197 * MT_DATA mbufs.
2198 */
2199 if (m->m_type == MT_CONTROL) {
2200 struct mbuf *cm = NULL, *cmn;
2201 struct mbuf **cme = &cm;
2202
2203 do {
2204 m2 = m->m_next;
2205 m->m_next = NULL;
2206 *cme = m;
2207 cme = &(*cme)->m_next;
2208 m = m2;
2209 } while (m != NULL && m->m_type == MT_CONTROL);
2210 while (cm != NULL) {
2211 cmn = cm->m_next;
2212 cm->m_next = NULL;
2213 if (pr->pr_domain->dom_externalize != NULL) {
2214 error = (*pr->pr_domain->dom_externalize)
2215 (cm, controlp, flags);
2216 } else if (controlp != NULL)
2217 *controlp = cm;
2218 else
2219 m_freem(cm);
2220 if (controlp != NULL) {
2221 while (*controlp != NULL)
2222 controlp = &(*controlp)->m_next;
2223 }
2224 cm = cmn;
2225 }
2226 }
2227 KASSERT(m == NULL || m->m_type == MT_DATA,
2228 ("soreceive_dgram: !data"));
2229 while (m != NULL && uio->uio_resid > 0) {
2230 len = uio->uio_resid;
2231 if (len > m->m_len)
2232 len = m->m_len;
2233 error = uiomove(mtod(m, char *), (int)len, uio);
2234 if (error) {
2235 m_freem(m);
2236 return (error);
2237 }
2238 if (len == m->m_len)
2239 m = m_free(m);
2240 else {
2241 m->m_data += len;
2242 m->m_len -= len;
2243 }
2244 }
2245 if (m != NULL) {
2246 flags |= MSG_TRUNC;
2247 m_freem(m);
2248 }
2249 if (flagsp != NULL)
2250 *flagsp |= flags;
2251 return (0);
2252 }
2253
2254 int
2255 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2256 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2257 {
2258 int error;
2259
2260 CURVNET_SET(so->so_vnet);
2261 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2262 controlp, flagsp));
2263 CURVNET_RESTORE();
2264 return (error);
2265 }
2266
2267 int
2268 soshutdown(struct socket *so, int how)
2269 {
2270 struct protosw *pr = so->so_proto;
2271 int error;
2272
2273 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2274 return (EINVAL);
2275
2276 CURVNET_SET(so->so_vnet);
2277 if (pr->pr_usrreqs->pru_flush != NULL)
2278 (*pr->pr_usrreqs->pru_flush)(so, how);
2279 if (how != SHUT_WR)
2280 sorflush(so);
2281 if (how != SHUT_RD) {
2282 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2283 wakeup(&so->so_timeo);
2284 CURVNET_RESTORE();
2285 return (error);
2286 }
2287 wakeup(&so->so_timeo);
2288 CURVNET_RESTORE();
2289 return (0);
2290 }
2291
2292 void
2293 sorflush(struct socket *so)
2294 {
2295 struct sockbuf *sb = &so->so_rcv;
2296 struct protosw *pr = so->so_proto;
2297 struct socket aso;
2298
2299 VNET_SO_ASSERT(so);
2300
2301 /*
2302 * In order to avoid calling dom_dispose with the socket buffer mutex
2303 * held, and in order to generally avoid holding the lock for a long
2304 * time, we make a copy of the socket buffer and clear the original
2305 * (except locks, state). The new socket buffer copy won't have
2306 * initialized locks so we can only call routines that won't use or
2307 * assert those locks.
2308 *
2309 * Dislodge threads currently blocked in receive and wait to acquire
2310 * a lock against other simultaneous readers before clearing the
2311 * socket buffer. Don't let our acquire be interrupted by a signal
2312 * despite any existing socket disposition on interruptable waiting.
2313 */
2314 socantrcvmore(so);
2315 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2316
2317 /*
2318 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2319 * and mutex data unchanged.
2320 */
2321 SOCKBUF_LOCK(sb);
2322 bzero(&aso, sizeof(aso));
2323 aso.so_pcb = so->so_pcb;
2324 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2325 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2326 bzero(&sb->sb_startzero,
2327 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2328 SOCKBUF_UNLOCK(sb);
2329 sbunlock(sb);
2330
2331 /*
2332 * Dispose of special rights and flush the copied socket. Don't call
2333 * any unsafe routines (that rely on locks being initialized) on aso.
2334 */
2335 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) {
2336 if (pr->pr_domain->dom_family == AF_LOCAL)
2337 unp_dispose_so(&aso);
2338 else
2339 (*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
2340 }
2341 sbrelease_internal(&aso.so_rcv, so);
2342 }
2343
2344 /*
2345 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2346 * additional variant to handle the case where the option value needs to be
2347 * some kind of integer, but not a specific size. In addition to their use
2348 * here, these functions are also called by the protocol-level pr_ctloutput()
2349 * routines.
2350 */
2351 int
2352 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2353 {
2354 size_t valsize;
2355
2356 /*
2357 * If the user gives us more than we wanted, we ignore it, but if we
2358 * don't get the minimum length the caller wants, we return EINVAL.
2359 * On success, sopt->sopt_valsize is set to however much we actually
2360 * retrieved.
2361 */
2362 if ((valsize = sopt->sopt_valsize) < minlen)
2363 return EINVAL;
2364 if (valsize > len)
2365 sopt->sopt_valsize = valsize = len;
2366
2367 if (sopt->sopt_td != NULL)
2368 return (copyin(sopt->sopt_val, buf, valsize));
2369
2370 bcopy(sopt->sopt_val, buf, valsize);
2371 return (0);
2372 }
2373
2374 /*
2375 * Kernel version of setsockopt(2).
2376 *
2377 * XXX: optlen is size_t, not socklen_t
2378 */
2379 int
2380 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2381 size_t optlen)
2382 {
2383 struct sockopt sopt;
2384
2385 sopt.sopt_level = level;
2386 sopt.sopt_name = optname;
2387 sopt.sopt_dir = SOPT_SET;
2388 sopt.sopt_val = optval;
2389 sopt.sopt_valsize = optlen;
2390 sopt.sopt_td = NULL;
2391 return (sosetopt(so, &sopt));
2392 }
2393
2394 int
2395 sosetopt(struct socket *so, struct sockopt *sopt)
2396 {
2397 int error, optval;
2398 struct linger l;
2399 struct timeval tv;
2400 sbintime_t val;
2401 uint32_t val32;
2402 #ifdef MAC
2403 struct mac extmac;
2404 #endif
2405
2406 CURVNET_SET(so->so_vnet);
2407 error = 0;
2408 if (sopt->sopt_level != SOL_SOCKET) {
2409 if (so->so_proto->pr_ctloutput != NULL) {
2410 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2411 CURVNET_RESTORE();
2412 return (error);
2413 }
2414 error = ENOPROTOOPT;
2415 } else {
2416 switch (sopt->sopt_name) {
2417 #ifdef INET
2418 case SO_ACCEPTFILTER:
2419 error = do_setopt_accept_filter(so, sopt);
2420 if (error)
2421 goto bad;
2422 break;
2423 #endif
2424 case SO_LINGER:
2425 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2426 if (error)
2427 goto bad;
2428
2429 SOCK_LOCK(so);
2430 so->so_linger = l.l_linger;
2431 if (l.l_onoff)
2432 so->so_options |= SO_LINGER;
2433 else
2434 so->so_options &= ~SO_LINGER;
2435 SOCK_UNLOCK(so);
2436 break;
2437
2438 case SO_DEBUG:
2439 case SO_KEEPALIVE:
2440 case SO_DONTROUTE:
2441 case SO_USELOOPBACK:
2442 case SO_BROADCAST:
2443 case SO_REUSEADDR:
2444 case SO_REUSEPORT:
2445 case SO_OOBINLINE:
2446 case SO_TIMESTAMP:
2447 case SO_BINTIME:
2448 case SO_NOSIGPIPE:
2449 case SO_NO_DDP:
2450 case SO_NO_OFFLOAD:
2451 error = sooptcopyin(sopt, &optval, sizeof optval,
2452 sizeof optval);
2453 if (error)
2454 goto bad;
2455 SOCK_LOCK(so);
2456 if (optval)
2457 so->so_options |= sopt->sopt_name;
2458 else
2459 so->so_options &= ~sopt->sopt_name;
2460 SOCK_UNLOCK(so);
2461 break;
2462
2463 case SO_SETFIB:
2464 error = sooptcopyin(sopt, &optval, sizeof optval,
2465 sizeof optval);
2466 if (error)
2467 goto bad;
2468
2469 if (optval < 0 || optval >= rt_numfibs) {
2470 error = EINVAL;
2471 goto bad;
2472 }
2473 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2474 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2475 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2476 so->so_fibnum = optval;
2477 else
2478 so->so_fibnum = 0;
2479 break;
2480
2481 case SO_USER_COOKIE:
2482 error = sooptcopyin(sopt, &val32, sizeof val32,
2483 sizeof val32);
2484 if (error)
2485 goto bad;
2486 so->so_user_cookie = val32;
2487 break;
2488
2489 case SO_SNDBUF:
2490 case SO_RCVBUF:
2491 case SO_SNDLOWAT:
2492 case SO_RCVLOWAT:
2493 error = sooptcopyin(sopt, &optval, sizeof optval,
2494 sizeof optval);
2495 if (error)
2496 goto bad;
2497
2498 /*
2499 * Values < 1 make no sense for any of these options,
2500 * so disallow them.
2501 */
2502 if (optval < 1) {
2503 error = EINVAL;
2504 goto bad;
2505 }
2506
2507 switch (sopt->sopt_name) {
2508 case SO_SNDBUF:
2509 case SO_RCVBUF:
2510 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2511 &so->so_snd : &so->so_rcv, (u_long)optval,
2512 so, curthread) == 0) {
2513 error = ENOBUFS;
2514 goto bad;
2515 }
2516 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2517 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2518 break;
2519
2520 /*
2521 * Make sure the low-water is never greater than the
2522 * high-water.
2523 */
2524 case SO_SNDLOWAT:
2525 SOCKBUF_LOCK(&so->so_snd);
2526 so->so_snd.sb_lowat =
2527 (optval > so->so_snd.sb_hiwat) ?
2528 so->so_snd.sb_hiwat : optval;
2529 SOCKBUF_UNLOCK(&so->so_snd);
2530 break;
2531 case SO_RCVLOWAT:
2532 SOCKBUF_LOCK(&so->so_rcv);
2533 so->so_rcv.sb_lowat =
2534 (optval > so->so_rcv.sb_hiwat) ?
2535 so->so_rcv.sb_hiwat : optval;
2536 SOCKBUF_UNLOCK(&so->so_rcv);
2537 break;
2538 }
2539 break;
2540
2541 case SO_SNDTIMEO:
2542 case SO_RCVTIMEO:
2543 #ifdef COMPAT_FREEBSD32
2544 if (SV_CURPROC_FLAG(SV_ILP32)) {
2545 struct timeval32 tv32;
2546
2547 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2548 sizeof tv32);
2549 CP(tv32, tv, tv_sec);
2550 CP(tv32, tv, tv_usec);
2551 } else
2552 #endif
2553 error = sooptcopyin(sopt, &tv, sizeof tv,
2554 sizeof tv);
2555 if (error)
2556 goto bad;
2557 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2558 tv.tv_usec >= 1000000) {
2559 error = EDOM;
2560 goto bad;
2561 }
2562 if (tv.tv_sec > INT32_MAX)
2563 val = SBT_MAX;
2564 else
2565 val = tvtosbt(tv);
2566 switch (sopt->sopt_name) {
2567 case SO_SNDTIMEO:
2568 so->so_snd.sb_timeo = val;
2569 break;
2570 case SO_RCVTIMEO:
2571 so->so_rcv.sb_timeo = val;
2572 break;
2573 }
2574 break;
2575
2576 case SO_LABEL:
2577 #ifdef MAC
2578 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2579 sizeof extmac);
2580 if (error)
2581 goto bad;
2582 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2583 so, &extmac);
2584 #else
2585 error = EOPNOTSUPP;
2586 #endif
2587 break;
2588
2589 default:
2590 error = ENOPROTOOPT;
2591 break;
2592 }
2593 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2594 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2595 }
2596 bad:
2597 CURVNET_RESTORE();
2598 return (error);
2599 }
2600
2601 /*
2602 * Helper routine for getsockopt.
2603 */
2604 int
2605 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2606 {
2607 int error;
2608 size_t valsize;
2609
2610 error = 0;
2611
2612 /*
2613 * Documented get behavior is that we always return a value, possibly
2614 * truncated to fit in the user's buffer. Traditional behavior is
2615 * that we always tell the user precisely how much we copied, rather
2616 * than something useful like the total amount we had available for
2617 * her. Note that this interface is not idempotent; the entire
2618 * answer must generated ahead of time.
2619 */
2620 valsize = min(len, sopt->sopt_valsize);
2621 sopt->sopt_valsize = valsize;
2622 if (sopt->sopt_val != NULL) {
2623 if (sopt->sopt_td != NULL)
2624 error = copyout(buf, sopt->sopt_val, valsize);
2625 else
2626 bcopy(buf, sopt->sopt_val, valsize);
2627 }
2628 return (error);
2629 }
2630
2631 int
2632 sogetopt(struct socket *so, struct sockopt *sopt)
2633 {
2634 int error, optval;
2635 struct linger l;
2636 struct timeval tv;
2637 #ifdef MAC
2638 struct mac extmac;
2639 #endif
2640
2641 CURVNET_SET(so->so_vnet);
2642 error = 0;
2643 if (sopt->sopt_level != SOL_SOCKET) {
2644 if (so->so_proto->pr_ctloutput != NULL)
2645 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2646 else
2647 error = ENOPROTOOPT;
2648 CURVNET_RESTORE();
2649 return (error);
2650 } else {
2651 switch (sopt->sopt_name) {
2652 #ifdef INET
2653 case SO_ACCEPTFILTER:
2654 error = do_getopt_accept_filter(so, sopt);
2655 break;
2656 #endif
2657 case SO_LINGER:
2658 SOCK_LOCK(so);
2659 l.l_onoff = so->so_options & SO_LINGER;
2660 l.l_linger = so->so_linger;
2661 SOCK_UNLOCK(so);
2662 error = sooptcopyout(sopt, &l, sizeof l);
2663 break;
2664
2665 case SO_USELOOPBACK:
2666 case SO_DONTROUTE:
2667 case SO_DEBUG:
2668 case SO_KEEPALIVE:
2669 case SO_REUSEADDR:
2670 case SO_REUSEPORT:
2671 case SO_BROADCAST:
2672 case SO_OOBINLINE:
2673 case SO_ACCEPTCONN:
2674 case SO_TIMESTAMP:
2675 case SO_BINTIME:
2676 case SO_NOSIGPIPE:
2677 optval = so->so_options & sopt->sopt_name;
2678 integer:
2679 error = sooptcopyout(sopt, &optval, sizeof optval);
2680 break;
2681
2682 case SO_TYPE:
2683 optval = so->so_type;
2684 goto integer;
2685
2686 case SO_PROTOCOL:
2687 optval = so->so_proto->pr_protocol;
2688 goto integer;
2689
2690 case SO_ERROR:
2691 SOCK_LOCK(so);
2692 optval = so->so_error;
2693 so->so_error = 0;
2694 SOCK_UNLOCK(so);
2695 goto integer;
2696
2697 case SO_SNDBUF:
2698 optval = so->so_snd.sb_hiwat;
2699 goto integer;
2700
2701 case SO_RCVBUF:
2702 optval = so->so_rcv.sb_hiwat;
2703 goto integer;
2704
2705 case SO_SNDLOWAT:
2706 optval = so->so_snd.sb_lowat;
2707 goto integer;
2708
2709 case SO_RCVLOWAT:
2710 optval = so->so_rcv.sb_lowat;
2711 goto integer;
2712
2713 case SO_SNDTIMEO:
2714 case SO_RCVTIMEO:
2715 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2716 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2717 #ifdef COMPAT_FREEBSD32
2718 if (SV_CURPROC_FLAG(SV_ILP32)) {
2719 struct timeval32 tv32;
2720
2721 CP(tv, tv32, tv_sec);
2722 CP(tv, tv32, tv_usec);
2723 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2724 } else
2725 #endif
2726 error = sooptcopyout(sopt, &tv, sizeof tv);
2727 break;
2728
2729 case SO_LABEL:
2730 #ifdef MAC
2731 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2732 sizeof(extmac));
2733 if (error)
2734 goto bad;
2735 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2736 so, &extmac);
2737 if (error)
2738 goto bad;
2739 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2740 #else
2741 error = EOPNOTSUPP;
2742 #endif
2743 break;
2744
2745 case SO_PEERLABEL:
2746 #ifdef MAC
2747 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2748 sizeof(extmac));
2749 if (error)
2750 goto bad;
2751 error = mac_getsockopt_peerlabel(
2752 sopt->sopt_td->td_ucred, so, &extmac);
2753 if (error)
2754 goto bad;
2755 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2756 #else
2757 error = EOPNOTSUPP;
2758 #endif
2759 break;
2760
2761 case SO_LISTENQLIMIT:
2762 optval = so->so_qlimit;
2763 goto integer;
2764
2765 case SO_LISTENQLEN:
2766 optval = so->so_qlen;
2767 goto integer;
2768
2769 case SO_LISTENINCQLEN:
2770 optval = so->so_incqlen;
2771 goto integer;
2772
2773 default:
2774 error = ENOPROTOOPT;
2775 break;
2776 }
2777 }
2778 #ifdef MAC
2779 bad:
2780 #endif
2781 CURVNET_RESTORE();
2782 return (error);
2783 }
2784
2785 int
2786 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2787 {
2788 struct mbuf *m, *m_prev;
2789 int sopt_size = sopt->sopt_valsize;
2790
2791 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2792 if (m == NULL)
2793 return ENOBUFS;
2794 if (sopt_size > MLEN) {
2795 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2796 if ((m->m_flags & M_EXT) == 0) {
2797 m_free(m);
2798 return ENOBUFS;
2799 }
2800 m->m_len = min(MCLBYTES, sopt_size);
2801 } else {
2802 m->m_len = min(MLEN, sopt_size);
2803 }
2804 sopt_size -= m->m_len;
2805 *mp = m;
2806 m_prev = m;
2807
2808 while (sopt_size) {
2809 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2810 if (m == NULL) {
2811 m_freem(*mp);
2812 return ENOBUFS;
2813 }
2814 if (sopt_size > MLEN) {
2815 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2816 M_NOWAIT);
2817 if ((m->m_flags & M_EXT) == 0) {
2818 m_freem(m);
2819 m_freem(*mp);
2820 return ENOBUFS;
2821 }
2822 m->m_len = min(MCLBYTES, sopt_size);
2823 } else {
2824 m->m_len = min(MLEN, sopt_size);
2825 }
2826 sopt_size -= m->m_len;
2827 m_prev->m_next = m;
2828 m_prev = m;
2829 }
2830 return (0);
2831 }
2832
2833 int
2834 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2835 {
2836 struct mbuf *m0 = m;
2837
2838 if (sopt->sopt_val == NULL)
2839 return (0);
2840 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2841 if (sopt->sopt_td != NULL) {
2842 int error;
2843
2844 error = copyin(sopt->sopt_val, mtod(m, char *),
2845 m->m_len);
2846 if (error != 0) {
2847 m_freem(m0);
2848 return(error);
2849 }
2850 } else
2851 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2852 sopt->sopt_valsize -= m->m_len;
2853 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2854 m = m->m_next;
2855 }
2856 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2857 panic("ip6_sooptmcopyin");
2858 return (0);
2859 }
2860
2861 int
2862 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2863 {
2864 struct mbuf *m0 = m;
2865 size_t valsize = 0;
2866
2867 if (sopt->sopt_val == NULL)
2868 return (0);
2869 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2870 if (sopt->sopt_td != NULL) {
2871 int error;
2872
2873 error = copyout(mtod(m, char *), sopt->sopt_val,
2874 m->m_len);
2875 if (error != 0) {
2876 m_freem(m0);
2877 return(error);
2878 }
2879 } else
2880 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2881 sopt->sopt_valsize -= m->m_len;
2882 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2883 valsize += m->m_len;
2884 m = m->m_next;
2885 }
2886 if (m != NULL) {
2887 /* enough soopt buffer should be given from user-land */
2888 m_freem(m0);
2889 return(EINVAL);
2890 }
2891 sopt->sopt_valsize = valsize;
2892 return (0);
2893 }
2894
2895 /*
2896 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2897 * out-of-band data, which will then notify socket consumers.
2898 */
2899 void
2900 sohasoutofband(struct socket *so)
2901 {
2902
2903 if (so->so_sigio != NULL)
2904 pgsigio(&so->so_sigio, SIGURG, 0);
2905 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2906 }
2907
2908 int
2909 sopoll(struct socket *so, int events, struct ucred *active_cred,
2910 struct thread *td)
2911 {
2912
2913 /*
2914 * We do not need to set or assert curvnet as long as everyone uses
2915 * sopoll_generic().
2916 */
2917 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2918 td));
2919 }
2920
2921 int
2922 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2923 struct thread *td)
2924 {
2925 int revents = 0;
2926
2927 SOCKBUF_LOCK(&so->so_snd);
2928 SOCKBUF_LOCK(&so->so_rcv);
2929 if (events & (POLLIN | POLLRDNORM))
2930 if (soreadabledata(so))
2931 revents |= events & (POLLIN | POLLRDNORM);
2932
2933 if (events & (POLLOUT | POLLWRNORM))
2934 if (sowriteable(so))
2935 revents |= events & (POLLOUT | POLLWRNORM);
2936
2937 if (events & (POLLPRI | POLLRDBAND))
2938 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2939 revents |= events & (POLLPRI | POLLRDBAND);
2940
2941 if ((events & POLLINIGNEOF) == 0) {
2942 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2943 revents |= events & (POLLIN | POLLRDNORM);
2944 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2945 revents |= POLLHUP;
2946 }
2947 }
2948
2949 if (revents == 0) {
2950 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2951 selrecord(td, &so->so_rcv.sb_sel);
2952 so->so_rcv.sb_flags |= SB_SEL;
2953 }
2954
2955 if (events & (POLLOUT | POLLWRNORM)) {
2956 selrecord(td, &so->so_snd.sb_sel);
2957 so->so_snd.sb_flags |= SB_SEL;
2958 }
2959 }
2960
2961 SOCKBUF_UNLOCK(&so->so_rcv);
2962 SOCKBUF_UNLOCK(&so->so_snd);
2963 return (revents);
2964 }
2965
2966 int
2967 soo_kqfilter(struct file *fp, struct knote *kn)
2968 {
2969 struct socket *so = kn->kn_fp->f_data;
2970 struct sockbuf *sb;
2971
2972 switch (kn->kn_filter) {
2973 case EVFILT_READ:
2974 if (so->so_options & SO_ACCEPTCONN)
2975 kn->kn_fop = &solisten_filtops;
2976 else
2977 kn->kn_fop = &soread_filtops;
2978 sb = &so->so_rcv;
2979 break;
2980 case EVFILT_WRITE:
2981 kn->kn_fop = &sowrite_filtops;
2982 sb = &so->so_snd;
2983 break;
2984 default:
2985 return (EINVAL);
2986 }
2987
2988 SOCKBUF_LOCK(sb);
2989 knlist_add(&sb->sb_sel.si_note, kn, 1);
2990 sb->sb_flags |= SB_KNOTE;
2991 SOCKBUF_UNLOCK(sb);
2992 return (0);
2993 }
2994
2995 /*
2996 * Some routines that return EOPNOTSUPP for entry points that are not
2997 * supported by a protocol. Fill in as needed.
2998 */
2999 int
3000 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3001 {
3002
3003 return EOPNOTSUPP;
3004 }
3005
3006 int
3007 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3008 {
3009
3010 return EOPNOTSUPP;
3011 }
3012
3013 int
3014 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3015 {
3016
3017 return EOPNOTSUPP;
3018 }
3019
3020 int
3021 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3022 struct thread *td)
3023 {
3024
3025 return EOPNOTSUPP;
3026 }
3027
3028 int
3029 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3030 {
3031
3032 return EOPNOTSUPP;
3033 }
3034
3035 int
3036 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3037 struct thread *td)
3038 {
3039
3040 return EOPNOTSUPP;
3041 }
3042
3043 int
3044 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3045 {
3046
3047 return EOPNOTSUPP;
3048 }
3049
3050 int
3051 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3052 struct ifnet *ifp, struct thread *td)
3053 {
3054
3055 return EOPNOTSUPP;
3056 }
3057
3058 int
3059 pru_disconnect_notsupp(struct socket *so)
3060 {
3061
3062 return EOPNOTSUPP;
3063 }
3064
3065 int
3066 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3067 {
3068
3069 return EOPNOTSUPP;
3070 }
3071
3072 int
3073 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3074 {
3075
3076 return EOPNOTSUPP;
3077 }
3078
3079 int
3080 pru_rcvd_notsupp(struct socket *so, int flags)
3081 {
3082
3083 return EOPNOTSUPP;
3084 }
3085
3086 int
3087 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3088 {
3089
3090 return EOPNOTSUPP;
3091 }
3092
3093 int
3094 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3095 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3096 {
3097
3098 return EOPNOTSUPP;
3099 }
3100
3101 /*
3102 * This isn't really a ``null'' operation, but it's the default one and
3103 * doesn't do anything destructive.
3104 */
3105 int
3106 pru_sense_null(struct socket *so, struct stat *sb)
3107 {
3108
3109 sb->st_blksize = so->so_snd.sb_hiwat;
3110 return 0;
3111 }
3112
3113 int
3114 pru_shutdown_notsupp(struct socket *so)
3115 {
3116
3117 return EOPNOTSUPP;
3118 }
3119
3120 int
3121 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3122 {
3123
3124 return EOPNOTSUPP;
3125 }
3126
3127 int
3128 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3129 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3130 {
3131
3132 return EOPNOTSUPP;
3133 }
3134
3135 int
3136 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3137 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3138 {
3139
3140 return EOPNOTSUPP;
3141 }
3142
3143 int
3144 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3145 struct thread *td)
3146 {
3147
3148 return EOPNOTSUPP;
3149 }
3150
3151 static void
3152 filt_sordetach(struct knote *kn)
3153 {
3154 struct socket *so = kn->kn_fp->f_data;
3155
3156 SOCKBUF_LOCK(&so->so_rcv);
3157 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3158 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3159 so->so_rcv.sb_flags &= ~SB_KNOTE;
3160 SOCKBUF_UNLOCK(&so->so_rcv);
3161 }
3162
3163 /*ARGSUSED*/
3164 static int
3165 filt_soread(struct knote *kn, long hint)
3166 {
3167 struct socket *so;
3168
3169 so = kn->kn_fp->f_data;
3170 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3171
3172 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3173 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3174 kn->kn_flags |= EV_EOF;
3175 kn->kn_fflags = so->so_error;
3176 return (1);
3177 } else if (so->so_error) /* temporary udp error */
3178 return (1);
3179 else if (kn->kn_sfflags & NOTE_LOWAT)
3180 return (kn->kn_data >= kn->kn_sdata);
3181 else
3182 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3183 }
3184
3185 static void
3186 filt_sowdetach(struct knote *kn)
3187 {
3188 struct socket *so = kn->kn_fp->f_data;
3189
3190 SOCKBUF_LOCK(&so->so_snd);
3191 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3192 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3193 so->so_snd.sb_flags &= ~SB_KNOTE;
3194 SOCKBUF_UNLOCK(&so->so_snd);
3195 }
3196
3197 /*ARGSUSED*/
3198 static int
3199 filt_sowrite(struct knote *kn, long hint)
3200 {
3201 struct socket *so;
3202
3203 so = kn->kn_fp->f_data;
3204 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3205 kn->kn_data = sbspace(&so->so_snd);
3206 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3207 kn->kn_flags |= EV_EOF;
3208 kn->kn_fflags = so->so_error;
3209 return (1);
3210 } else if (so->so_error) /* temporary udp error */
3211 return (1);
3212 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3213 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3214 return (0);
3215 else if (kn->kn_sfflags & NOTE_LOWAT)
3216 return (kn->kn_data >= kn->kn_sdata);
3217 else
3218 return (kn->kn_data >= so->so_snd.sb_lowat);
3219 }
3220
3221 /*ARGSUSED*/
3222 static int
3223 filt_solisten(struct knote *kn, long hint)
3224 {
3225 struct socket *so = kn->kn_fp->f_data;
3226
3227 kn->kn_data = so->so_qlen;
3228 return (!TAILQ_EMPTY(&so->so_comp));
3229 }
3230
3231 int
3232 socheckuid(struct socket *so, uid_t uid)
3233 {
3234
3235 if (so == NULL)
3236 return (EPERM);
3237 if (so->so_cred->cr_uid != uid)
3238 return (EPERM);
3239 return (0);
3240 }
3241
3242 /*
3243 * These functions are used by protocols to notify the socket layer (and its
3244 * consumers) of state changes in the sockets driven by protocol-side events.
3245 */
3246
3247 /*
3248 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3249 *
3250 * Normal sequence from the active (originating) side is that
3251 * soisconnecting() is called during processing of connect() call, resulting
3252 * in an eventual call to soisconnected() if/when the connection is
3253 * established. When the connection is torn down soisdisconnecting() is
3254 * called during processing of disconnect() call, and soisdisconnected() is
3255 * called when the connection to the peer is totally severed. The semantics
3256 * of these routines are such that connectionless protocols can call
3257 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3258 * calls when setting up a ``connection'' takes no time.
3259 *
3260 * From the passive side, a socket is created with two queues of sockets:
3261 * so_incomp for connections in progress and so_comp for connections already
3262 * made and awaiting user acceptance. As a protocol is preparing incoming
3263 * connections, it creates a socket structure queued on so_incomp by calling
3264 * sonewconn(). When the connection is established, soisconnected() is
3265 * called, and transfers the socket structure to so_comp, making it available
3266 * to accept().
3267 *
3268 * If a socket is closed with sockets on either so_incomp or so_comp, these
3269 * sockets are dropped.
3270 *
3271 * If higher-level protocols are implemented in the kernel, the wakeups done
3272 * here will sometimes cause software-interrupt process scheduling.
3273 */
3274 void
3275 soisconnecting(struct socket *so)
3276 {
3277
3278 SOCK_LOCK(so);
3279 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3280 so->so_state |= SS_ISCONNECTING;
3281 SOCK_UNLOCK(so);
3282 }
3283
3284 void
3285 soisconnected(struct socket *so)
3286 {
3287 struct socket *head;
3288 int ret;
3289
3290 restart:
3291 ACCEPT_LOCK();
3292 SOCK_LOCK(so);
3293 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3294 so->so_state |= SS_ISCONNECTED;
3295 head = so->so_head;
3296 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3297 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3298 SOCK_UNLOCK(so);
3299 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3300 head->so_incqlen--;
3301 so->so_qstate &= ~SQ_INCOMP;
3302 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3303 head->so_qlen++;
3304 so->so_qstate |= SQ_COMP;
3305 ACCEPT_UNLOCK();
3306 sorwakeup(head);
3307 wakeup_one(&head->so_timeo);
3308 } else {
3309 ACCEPT_UNLOCK();
3310 soupcall_set(so, SO_RCV,
3311 head->so_accf->so_accept_filter->accf_callback,
3312 head->so_accf->so_accept_filter_arg);
3313 so->so_options &= ~SO_ACCEPTFILTER;
3314 ret = head->so_accf->so_accept_filter->accf_callback(so,
3315 head->so_accf->so_accept_filter_arg, M_NOWAIT);
3316 if (ret == SU_ISCONNECTED)
3317 soupcall_clear(so, SO_RCV);
3318 SOCK_UNLOCK(so);
3319 if (ret == SU_ISCONNECTED)
3320 goto restart;
3321 }
3322 return;
3323 }
3324 SOCK_UNLOCK(so);
3325 ACCEPT_UNLOCK();
3326 wakeup(&so->so_timeo);
3327 sorwakeup(so);
3328 sowwakeup(so);
3329 }
3330
3331 void
3332 soisdisconnecting(struct socket *so)
3333 {
3334
3335 /*
3336 * Note: This code assumes that SOCK_LOCK(so) and
3337 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3338 */
3339 SOCKBUF_LOCK(&so->so_rcv);
3340 so->so_state &= ~SS_ISCONNECTING;
3341 so->so_state |= SS_ISDISCONNECTING;
3342 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3343 sorwakeup_locked(so);
3344 SOCKBUF_LOCK(&so->so_snd);
3345 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3346 sowwakeup_locked(so);
3347 wakeup(&so->so_timeo);
3348 }
3349
3350 void
3351 soisdisconnected(struct socket *so)
3352 {
3353
3354 /*
3355 * Note: This code assumes that SOCK_LOCK(so) and
3356 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3357 */
3358 SOCKBUF_LOCK(&so->so_rcv);
3359 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3360 so->so_state |= SS_ISDISCONNECTED;
3361 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3362 sorwakeup_locked(so);
3363 SOCKBUF_LOCK(&so->so_snd);
3364 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3365 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3366 sowwakeup_locked(so);
3367 wakeup(&so->so_timeo);
3368 }
3369
3370 /*
3371 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3372 */
3373 struct sockaddr *
3374 sodupsockaddr(const struct sockaddr *sa, int mflags)
3375 {
3376 struct sockaddr *sa2;
3377
3378 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3379 if (sa2)
3380 bcopy(sa, sa2, sa->sa_len);
3381 return sa2;
3382 }
3383
3384 /*
3385 * Register per-socket buffer upcalls.
3386 */
3387 void
3388 soupcall_set(struct socket *so, int which,
3389 int (*func)(struct socket *, void *, int), void *arg)
3390 {
3391 struct sockbuf *sb;
3392
3393 switch (which) {
3394 case SO_RCV:
3395 sb = &so->so_rcv;
3396 break;
3397 case SO_SND:
3398 sb = &so->so_snd;
3399 break;
3400 default:
3401 panic("soupcall_set: bad which");
3402 }
3403 SOCKBUF_LOCK_ASSERT(sb);
3404 #if 0
3405 /* XXX: accf_http actually wants to do this on purpose. */
3406 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3407 #endif
3408 sb->sb_upcall = func;
3409 sb->sb_upcallarg = arg;
3410 sb->sb_flags |= SB_UPCALL;
3411 }
3412
3413 void
3414 soupcall_clear(struct socket *so, int which)
3415 {
3416 struct sockbuf *sb;
3417
3418 switch (which) {
3419 case SO_RCV:
3420 sb = &so->so_rcv;
3421 break;
3422 case SO_SND:
3423 sb = &so->so_snd;
3424 break;
3425 default:
3426 panic("soupcall_clear: bad which");
3427 }
3428 SOCKBUF_LOCK_ASSERT(sb);
3429 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3430 sb->sb_upcall = NULL;
3431 sb->sb_upcallarg = NULL;
3432 sb->sb_flags &= ~SB_UPCALL;
3433 }
3434
3435 /*
3436 * Create an external-format (``xsocket'') structure using the information in
3437 * the kernel-format socket structure pointed to by so. This is done to
3438 * reduce the spew of irrelevant information over this interface, to isolate
3439 * user code from changes in the kernel structure, and potentially to provide
3440 * information-hiding if we decide that some of this information should be
3441 * hidden from users.
3442 */
3443 void
3444 sotoxsocket(struct socket *so, struct xsocket *xso)
3445 {
3446
3447 xso->xso_len = sizeof *xso;
3448 xso->xso_so = so;
3449 xso->so_type = so->so_type;
3450 xso->so_options = so->so_options;
3451 xso->so_linger = so->so_linger;
3452 xso->so_state = so->so_state;
3453 xso->so_pcb = so->so_pcb;
3454 xso->xso_protocol = so->so_proto->pr_protocol;
3455 xso->xso_family = so->so_proto->pr_domain->dom_family;
3456 xso->so_qlen = so->so_qlen;
3457 xso->so_incqlen = so->so_incqlen;
3458 xso->so_qlimit = so->so_qlimit;
3459 xso->so_timeo = so->so_timeo;
3460 xso->so_error = so->so_error;
3461 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3462 xso->so_oobmark = so->so_oobmark;
3463 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3464 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3465 xso->so_uid = so->so_cred->cr_uid;
3466 }
3467
3468
3469 /*
3470 * Socket accessor functions to provide external consumers with
3471 * a safe interface to socket state
3472 *
3473 */
3474
3475 void
3476 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3477 void *arg)
3478 {
3479
3480 TAILQ_FOREACH(so, &so->so_comp, so_list)
3481 func(so, arg);
3482 }
3483
3484 struct sockbuf *
3485 so_sockbuf_rcv(struct socket *so)
3486 {
3487
3488 return (&so->so_rcv);
3489 }
3490
3491 struct sockbuf *
3492 so_sockbuf_snd(struct socket *so)
3493 {
3494
3495 return (&so->so_snd);
3496 }
3497
3498 int
3499 so_state_get(const struct socket *so)
3500 {
3501
3502 return (so->so_state);
3503 }
3504
3505 void
3506 so_state_set(struct socket *so, int val)
3507 {
3508
3509 so->so_state = val;
3510 }
3511
3512 int
3513 so_options_get(const struct socket *so)
3514 {
3515
3516 return (so->so_options);
3517 }
3518
3519 void
3520 so_options_set(struct socket *so, int val)
3521 {
3522
3523 so->so_options = val;
3524 }
3525
3526 int
3527 so_error_get(const struct socket *so)
3528 {
3529
3530 return (so->so_error);
3531 }
3532
3533 void
3534 so_error_set(struct socket *so, int val)
3535 {
3536
3537 so->so_error = val;
3538 }
3539
3540 int
3541 so_linger_get(const struct socket *so)
3542 {
3543
3544 return (so->so_linger);
3545 }
3546
3547 void
3548 so_linger_set(struct socket *so, int val)
3549 {
3550
3551 so->so_linger = val;
3552 }
3553
3554 struct protosw *
3555 so_protosw_get(const struct socket *so)
3556 {
3557
3558 return (so->so_proto);
3559 }
3560
3561 void
3562 so_protosw_set(struct socket *so, struct protosw *val)
3563 {
3564
3565 so->so_proto = val;
3566 }
3567
3568 void
3569 so_sorwakeup(struct socket *so)
3570 {
3571
3572 sorwakeup(so);
3573 }
3574
3575 void
3576 so_sowwakeup(struct socket *so)
3577 {
3578
3579 sowwakeup(so);
3580 }
3581
3582 void
3583 so_sorwakeup_locked(struct socket *so)
3584 {
3585
3586 sorwakeup_locked(so);
3587 }
3588
3589 void
3590 so_sowwakeup_locked(struct socket *so)
3591 {
3592
3593 sowwakeup_locked(so);
3594 }
3595
3596 void
3597 so_lock(struct socket *so)
3598 {
3599
3600 SOCK_LOCK(so);
3601 }
3602
3603 void
3604 so_unlock(struct socket *so)
3605 {
3606
3607 SOCK_UNLOCK(so);
3608 }
3609