1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2008 Robert N. M. Watson
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
33 */
34
35 /*
36 * Comments on the socket life cycle:
37 *
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
40 *
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
43 *
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
47 *
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
52 *
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
57 * case.
58 *
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
61 * sockets.
62 *
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
66 * listen socket.
67 *
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
71 *
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
82 *
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
86 * interface.
87 *
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
95 *
96 * NOTE: With regard to VNETs the general rule is that callers do not set
97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99 * and sorflush(), which are usually called from a pre-set VNET context.
100 * sopoll() currently does not need a VNET context to be set.
101 */
102
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD: releng/9.2/sys/kern/uipc_socket.c 253035 2013-07-08 13:24:18Z andre $");
105
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_zero.h"
109 #include "opt_compat.h"
110
111 #include <sys/param.h>
112 #include <sys/systm.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
116 #include <sys/mac.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h> /* for struct knote */
122 #include <sys/kernel.h>
123 #include <sys/event.h>
124 #include <sys/eventhandler.h>
125 #include <sys/poll.h>
126 #include <sys/proc.h>
127 #include <sys/protosw.h>
128 #include <sys/socket.h>
129 #include <sys/socketvar.h>
130 #include <sys/resourcevar.h>
131 #include <net/route.h>
132 #include <sys/signalvar.h>
133 #include <sys/stat.h>
134 #include <sys/sx.h>
135 #include <sys/sysctl.h>
136 #include <sys/uio.h>
137 #include <sys/jail.h>
138 #include <sys/syslog.h>
139 #include <netinet/in.h>
140
141 #include <net/vnet.h>
142
143 #include <security/mac/mac_framework.h>
144
145 #include <vm/uma.h>
146
147 #ifdef COMPAT_FREEBSD32
148 #include <sys/mount.h>
149 #include <sys/sysent.h>
150 #include <compat/freebsd32/freebsd32.h>
151 #endif
152
153 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
154 int flags);
155
156 static void filt_sordetach(struct knote *kn);
157 static int filt_soread(struct knote *kn, long hint);
158 static void filt_sowdetach(struct knote *kn);
159 static int filt_sowrite(struct knote *kn, long hint);
160 static int filt_solisten(struct knote *kn, long hint);
161
162 static struct filterops solisten_filtops = {
163 .f_isfd = 1,
164 .f_detach = filt_sordetach,
165 .f_event = filt_solisten,
166 };
167 static struct filterops soread_filtops = {
168 .f_isfd = 1,
169 .f_detach = filt_sordetach,
170 .f_event = filt_soread,
171 };
172 static struct filterops sowrite_filtops = {
173 .f_isfd = 1,
174 .f_detach = filt_sowdetach,
175 .f_event = filt_sowrite,
176 };
177
178 so_gen_t so_gencnt; /* generation count for sockets */
179
180 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
181 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
182
183 #define VNET_SO_ASSERT(so) \
184 VNET_ASSERT(curvnet != NULL, \
185 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
186
187 /*
188 * Limit on the number of connections in the listen queue waiting
189 * for accept(2).
190 */
191 static int somaxconn = SOMAXCONN;
192
193 static int
194 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
195 {
196 int error;
197 int val;
198
199 val = somaxconn;
200 error = sysctl_handle_int(oidp, &val, 0, req);
201 if (error || !req->newptr )
202 return (error);
203
204 if (val < 1 || val > USHRT_MAX)
205 return (EINVAL);
206
207 somaxconn = val;
208 return (0);
209 }
210 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
211 0, sizeof(int), sysctl_somaxconn, "I",
212 "Maximum listen socket pending connection accept queue size");
213
214 static int numopensockets;
215 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
216 &numopensockets, 0, "Number of open sockets");
217
218 #ifdef ZERO_COPY_SOCKETS
219 /* These aren't static because they're used in other files. */
220 int so_zero_copy_send = 1;
221 int so_zero_copy_receive = 1;
222 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
223 "Zero copy controls");
224 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
225 &so_zero_copy_receive, 0, "Enable zero copy receive");
226 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
227 &so_zero_copy_send, 0, "Enable zero copy send");
228 #endif /* ZERO_COPY_SOCKETS */
229
230 /*
231 * accept_mtx locks down per-socket fields relating to accept queues. See
232 * socketvar.h for an annotation of the protected fields of struct socket.
233 */
234 struct mtx accept_mtx;
235 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
236
237 /*
238 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
239 * so_gencnt field.
240 */
241 static struct mtx so_global_mtx;
242 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
243
244 /*
245 * General IPC sysctl name space, used by sockets and a variety of other IPC
246 * types.
247 */
248 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
249
250 /*
251 * Initialize the socket subsystem and set up the socket
252 * memory allocator.
253 */
254 uma_zone_t socket_zone;
255 int maxsockets;
256
257 static void
258 socket_zone_change(void *tag)
259 {
260
261 uma_zone_set_max(socket_zone, maxsockets);
262 }
263
264 static void
265 socket_init(void *tag)
266 {
267
268 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
269 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
270 uma_zone_set_max(socket_zone, maxsockets);
271 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
272 EVENTHANDLER_PRI_FIRST);
273 }
274 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
275
276 /*
277 * Initialise maxsockets. This SYSINIT must be run after
278 * tunable_mbinit().
279 */
280 static void
281 init_maxsockets(void *ignored)
282 {
283
284 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
285 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
286 }
287 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
288
289 /*
290 * Sysctl to get and set the maximum global sockets limit. Notify protocols
291 * of the change so that they can update their dependent limits as required.
292 */
293 static int
294 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
295 {
296 int error, newmaxsockets;
297
298 newmaxsockets = maxsockets;
299 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
300 if (error == 0 && req->newptr) {
301 if (newmaxsockets > maxsockets) {
302 maxsockets = newmaxsockets;
303 if (maxsockets > ((maxfiles / 4) * 3)) {
304 maxfiles = (maxsockets * 5) / 4;
305 maxfilesperproc = (maxfiles * 9) / 10;
306 }
307 EVENTHANDLER_INVOKE(maxsockets_change);
308 } else
309 error = EINVAL;
310 }
311 return (error);
312 }
313 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
314 &maxsockets, 0, sysctl_maxsockets, "IU",
315 "Maximum number of sockets avaliable");
316
317 /*
318 * Socket operation routines. These routines are called by the routines in
319 * sys_socket.c or from a system process, and implement the semantics of
320 * socket operations by switching out to the protocol specific routines.
321 */
322
323 /*
324 * Get a socket structure from our zone, and initialize it. Note that it
325 * would probably be better to allocate socket and PCB at the same time, but
326 * I'm not convinced that all the protocols can be easily modified to do
327 * this.
328 *
329 * soalloc() returns a socket with a ref count of 0.
330 */
331 static struct socket *
332 soalloc(struct vnet *vnet)
333 {
334 struct socket *so;
335
336 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
337 if (so == NULL)
338 return (NULL);
339 #ifdef MAC
340 if (mac_socket_init(so, M_NOWAIT) != 0) {
341 uma_zfree(socket_zone, so);
342 return (NULL);
343 }
344 #endif
345 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
346 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
347 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
348 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
349 TAILQ_INIT(&so->so_aiojobq);
350 mtx_lock(&so_global_mtx);
351 so->so_gencnt = ++so_gencnt;
352 ++numopensockets;
353 #ifdef VIMAGE
354 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
355 __func__, __LINE__, so));
356 vnet->vnet_sockcnt++;
357 so->so_vnet = vnet;
358 #endif
359 mtx_unlock(&so_global_mtx);
360 return (so);
361 }
362
363 /*
364 * Free the storage associated with a socket at the socket layer, tear down
365 * locks, labels, etc. All protocol state is assumed already to have been
366 * torn down (and possibly never set up) by the caller.
367 */
368 static void
369 sodealloc(struct socket *so)
370 {
371
372 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
373 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
374
375 mtx_lock(&so_global_mtx);
376 so->so_gencnt = ++so_gencnt;
377 --numopensockets; /* Could be below, but faster here. */
378 #ifdef VIMAGE
379 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
380 __func__, __LINE__, so));
381 so->so_vnet->vnet_sockcnt--;
382 #endif
383 mtx_unlock(&so_global_mtx);
384 if (so->so_rcv.sb_hiwat)
385 (void)chgsbsize(so->so_cred->cr_uidinfo,
386 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
387 if (so->so_snd.sb_hiwat)
388 (void)chgsbsize(so->so_cred->cr_uidinfo,
389 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
390 #ifdef INET
391 /* remove acccept filter if one is present. */
392 if (so->so_accf != NULL)
393 do_setopt_accept_filter(so, NULL);
394 #endif
395 #ifdef MAC
396 mac_socket_destroy(so);
397 #endif
398 crfree(so->so_cred);
399 sx_destroy(&so->so_snd.sb_sx);
400 sx_destroy(&so->so_rcv.sb_sx);
401 SOCKBUF_LOCK_DESTROY(&so->so_snd);
402 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
403 uma_zfree(socket_zone, so);
404 }
405
406 /*
407 * socreate returns a socket with a ref count of 1. The socket should be
408 * closed with soclose().
409 */
410 int
411 socreate(int dom, struct socket **aso, int type, int proto,
412 struct ucred *cred, struct thread *td)
413 {
414 struct protosw *prp;
415 struct socket *so;
416 int error;
417
418 if (proto)
419 prp = pffindproto(dom, proto, type);
420 else
421 prp = pffindtype(dom, type);
422
423 if (prp == NULL || 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 *
487 sonewconn(struct socket *head, int connstatus)
488 {
489 struct socket *so;
490 int over;
491
492 ACCEPT_LOCK();
493 over = (head->so_qlen > 3 * head->so_qlimit / 2);
494 ACCEPT_UNLOCK();
495 #ifdef REGRESSION
496 if (regression_sonewconn_earlytest && over) {
497 #else
498 if (over) {
499 #endif
500 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
501 "%i already in queue awaiting acceptance\n",
502 __func__, head->so_pcb, head->so_qlen);
503 return (NULL);
504 }
505 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
506 __func__, __LINE__, head));
507 so = soalloc(head->so_vnet);
508 if (so == NULL) {
509 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
510 "limit reached or out of memory\n",
511 __func__, head->so_pcb);
512 return (NULL);
513 }
514 if ((head->so_options & SO_ACCEPTFILTER) != 0)
515 connstatus = 0;
516 so->so_head = head;
517 so->so_type = head->so_type;
518 so->so_options = head->so_options &~ SO_ACCEPTCONN;
519 so->so_linger = head->so_linger;
520 so->so_state = head->so_state | SS_NOFDREF;
521 so->so_fibnum = head->so_fibnum;
522 so->so_proto = head->so_proto;
523 so->so_cred = crhold(head->so_cred);
524 #ifdef MAC
525 mac_socket_newconn(head, so);
526 #endif
527 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
528 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
529 VNET_SO_ASSERT(head);
530 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
531 sodealloc(so);
532 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
533 __func__, head->so_pcb);
534 return (NULL);
535 }
536 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
537 sodealloc(so);
538 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
539 __func__, head->so_pcb);
540 return (NULL);
541 }
542 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
543 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
544 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
545 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
546 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
547 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
548 so->so_state |= connstatus;
549 ACCEPT_LOCK();
550 /*
551 * The accept socket may be tearing down but we just
552 * won a race on the ACCEPT_LOCK.
553 * However, if sctp_peeloff() is called on a 1-to-many
554 * style socket, the SO_ACCEPTCONN doesn't need to be set.
555 */
556 if (!(head->so_options & SO_ACCEPTCONN) &&
557 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
558 (head->so_type != SOCK_SEQPACKET))) {
559 SOCK_LOCK(so);
560 so->so_head = NULL;
561 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
562 return (NULL);
563 }
564 if (connstatus) {
565 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
566 so->so_qstate |= SQ_COMP;
567 head->so_qlen++;
568 } else {
569 /*
570 * Keep removing sockets from the head until there's room for
571 * us to insert on the tail. In pre-locking revisions, this
572 * was a simple if(), but as we could be racing with other
573 * threads and soabort() requires dropping locks, we must
574 * loop waiting for the condition to be true.
575 */
576 while (head->so_incqlen > head->so_qlimit) {
577 struct socket *sp;
578 sp = TAILQ_FIRST(&head->so_incomp);
579 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
580 head->so_incqlen--;
581 sp->so_qstate &= ~SQ_INCOMP;
582 sp->so_head = NULL;
583 ACCEPT_UNLOCK();
584 soabort(sp);
585 ACCEPT_LOCK();
586 }
587 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
588 so->so_qstate |= SQ_INCOMP;
589 head->so_incqlen++;
590 }
591 ACCEPT_UNLOCK();
592 if (connstatus) {
593 sorwakeup(head);
594 wakeup_one(&head->so_timeo);
595 }
596 return (so);
597 }
598
599 int
600 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
601 {
602 int error;
603
604 CURVNET_SET(so->so_vnet);
605 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
606 CURVNET_RESTORE();
607 return error;
608 }
609
610 /*
611 * solisten() transitions a socket from a non-listening state to a listening
612 * state, but can also be used to update the listen queue depth on an
613 * existing listen socket. The protocol will call back into the sockets
614 * layer using solisten_proto_check() and solisten_proto() to check and set
615 * socket-layer listen state. Call backs are used so that the protocol can
616 * acquire both protocol and socket layer locks in whatever order is required
617 * by the protocol.
618 *
619 * Protocol implementors are advised to hold the socket lock across the
620 * socket-layer test and set to avoid races at the socket layer.
621 */
622 int
623 solisten(struct socket *so, int backlog, struct thread *td)
624 {
625 int error;
626
627 CURVNET_SET(so->so_vnet);
628 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
629 CURVNET_RESTORE();
630 return error;
631 }
632
633 int
634 solisten_proto_check(struct socket *so)
635 {
636
637 SOCK_LOCK_ASSERT(so);
638
639 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
640 SS_ISDISCONNECTING))
641 return (EINVAL);
642 return (0);
643 }
644
645 void
646 solisten_proto(struct socket *so, int backlog)
647 {
648
649 SOCK_LOCK_ASSERT(so);
650
651 if (backlog < 0 || backlog > somaxconn)
652 backlog = somaxconn;
653 so->so_qlimit = backlog;
654 so->so_options |= SO_ACCEPTCONN;
655 }
656
657 /*
658 * Evaluate the reference count and named references on a socket; if no
659 * references remain, free it. This should be called whenever a reference is
660 * released, such as in sorele(), but also when named reference flags are
661 * cleared in socket or protocol code.
662 *
663 * sofree() will free the socket if:
664 *
665 * - There are no outstanding file descriptor references or related consumers
666 * (so_count == 0).
667 *
668 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
669 *
670 * - The protocol does not have an outstanding strong reference on the socket
671 * (SS_PROTOREF).
672 *
673 * - The socket is not in a completed connection queue, so a process has been
674 * notified that it is present. If it is removed, the user process may
675 * block in accept() despite select() saying the socket was ready.
676 */
677 void
678 sofree(struct socket *so)
679 {
680 struct protosw *pr = so->so_proto;
681 struct socket *head;
682
683 ACCEPT_LOCK_ASSERT();
684 SOCK_LOCK_ASSERT(so);
685
686 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
687 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
688 SOCK_UNLOCK(so);
689 ACCEPT_UNLOCK();
690 return;
691 }
692
693 head = so->so_head;
694 if (head != NULL) {
695 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
696 (so->so_qstate & SQ_INCOMP) != 0,
697 ("sofree: so_head != NULL, but neither SQ_COMP nor "
698 "SQ_INCOMP"));
699 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
700 (so->so_qstate & SQ_INCOMP) == 0,
701 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
702 TAILQ_REMOVE(&head->so_incomp, so, so_list);
703 head->so_incqlen--;
704 so->so_qstate &= ~SQ_INCOMP;
705 so->so_head = NULL;
706 }
707 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
708 (so->so_qstate & SQ_INCOMP) == 0,
709 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
710 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
711 if (so->so_options & SO_ACCEPTCONN) {
712 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
713 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_incomp populated"));
714 }
715 SOCK_UNLOCK(so);
716 ACCEPT_UNLOCK();
717
718 VNET_SO_ASSERT(so);
719 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
720 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
721 if (pr->pr_usrreqs->pru_detach != NULL)
722 (*pr->pr_usrreqs->pru_detach)(so);
723
724 /*
725 * From this point on, we assume that no other references to this
726 * socket exist anywhere else in the stack. Therefore, no locks need
727 * to be acquired or held.
728 *
729 * We used to do a lot of socket buffer and socket locking here, as
730 * well as invoke sorflush() and perform wakeups. The direct call to
731 * dom_dispose() and sbrelease_internal() are an inlining of what was
732 * necessary from sorflush().
733 *
734 * Notice that the socket buffer and kqueue state are torn down
735 * before calling pru_detach. This means that protocols shold not
736 * assume they can perform socket wakeups, etc, in their detach code.
737 */
738 sbdestroy(&so->so_snd, so);
739 sbdestroy(&so->so_rcv, so);
740 seldrain(&so->so_snd.sb_sel);
741 seldrain(&so->so_rcv.sb_sel);
742 knlist_destroy(&so->so_rcv.sb_sel.si_note);
743 knlist_destroy(&so->so_snd.sb_sel.si_note);
744 sodealloc(so);
745 }
746
747 /*
748 * Close a socket on last file table reference removal. Initiate disconnect
749 * if connected. Free socket when disconnect complete.
750 *
751 * This function will sorele() the socket. Note that soclose() may be called
752 * prior to the ref count reaching zero. The actual socket structure will
753 * not be freed until the ref count reaches zero.
754 */
755 int
756 soclose(struct socket *so)
757 {
758 int error = 0;
759
760 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
761
762 CURVNET_SET(so->so_vnet);
763 funsetown(&so->so_sigio);
764 if (so->so_state & SS_ISCONNECTED) {
765 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
766 error = sodisconnect(so);
767 if (error) {
768 if (error == ENOTCONN)
769 error = 0;
770 goto drop;
771 }
772 }
773 if (so->so_options & SO_LINGER) {
774 if ((so->so_state & SS_ISDISCONNECTING) &&
775 (so->so_state & SS_NBIO))
776 goto drop;
777 while (so->so_state & SS_ISCONNECTED) {
778 error = tsleep(&so->so_timeo,
779 PSOCK | PCATCH, "soclos", so->so_linger * hz);
780 if (error)
781 break;
782 }
783 }
784 }
785
786 drop:
787 if (so->so_proto->pr_usrreqs->pru_close != NULL)
788 (*so->so_proto->pr_usrreqs->pru_close)(so);
789 ACCEPT_LOCK();
790 if (so->so_options & SO_ACCEPTCONN) {
791 struct socket *sp;
792 /*
793 * Prevent new additions to the accept queues due
794 * to ACCEPT_LOCK races while we are draining them.
795 */
796 so->so_options &= ~SO_ACCEPTCONN;
797 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
798 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
799 so->so_incqlen--;
800 sp->so_qstate &= ~SQ_INCOMP;
801 sp->so_head = NULL;
802 ACCEPT_UNLOCK();
803 soabort(sp);
804 ACCEPT_LOCK();
805 }
806 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
807 TAILQ_REMOVE(&so->so_comp, sp, so_list);
808 so->so_qlen--;
809 sp->so_qstate &= ~SQ_COMP;
810 sp->so_head = NULL;
811 ACCEPT_UNLOCK();
812 soabort(sp);
813 ACCEPT_LOCK();
814 }
815 KASSERT((TAILQ_EMPTY(&so->so_comp)),
816 ("%s: so_comp populated", __func__));
817 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
818 ("%s: so_incomp populated", __func__));
819 }
820 SOCK_LOCK(so);
821 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
822 so->so_state |= SS_NOFDREF;
823 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
824 CURVNET_RESTORE();
825 return (error);
826 }
827
828 /*
829 * soabort() is used to abruptly tear down a connection, such as when a
830 * resource limit is reached (listen queue depth exceeded), or if a listen
831 * socket is closed while there are sockets waiting to be accepted.
832 *
833 * This interface is tricky, because it is called on an unreferenced socket,
834 * and must be called only by a thread that has actually removed the socket
835 * from the listen queue it was on, or races with other threads are risked.
836 *
837 * This interface will call into the protocol code, so must not be called
838 * with any socket locks held. Protocols do call it while holding their own
839 * recursible protocol mutexes, but this is something that should be subject
840 * to review in the future.
841 */
842 void
843 soabort(struct socket *so)
844 {
845
846 /*
847 * In as much as is possible, assert that no references to this
848 * socket are held. This is not quite the same as asserting that the
849 * current thread is responsible for arranging for no references, but
850 * is as close as we can get for now.
851 */
852 KASSERT(so->so_count == 0, ("soabort: so_count"));
853 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
854 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
855 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
856 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
857 VNET_SO_ASSERT(so);
858
859 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
860 (*so->so_proto->pr_usrreqs->pru_abort)(so);
861 ACCEPT_LOCK();
862 SOCK_LOCK(so);
863 sofree(so);
864 }
865
866 int
867 soaccept(struct socket *so, struct sockaddr **nam)
868 {
869 int error;
870
871 SOCK_LOCK(so);
872 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
873 so->so_state &= ~SS_NOFDREF;
874 SOCK_UNLOCK(so);
875
876 CURVNET_SET(so->so_vnet);
877 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
878 CURVNET_RESTORE();
879 return (error);
880 }
881
882 int
883 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
884 {
885 int error;
886
887 if (so->so_options & SO_ACCEPTCONN)
888 return (EOPNOTSUPP);
889
890 CURVNET_SET(so->so_vnet);
891 /*
892 * If protocol is connection-based, can only connect once.
893 * Otherwise, if connected, try to disconnect first. This allows
894 * user to disconnect by connecting to, e.g., a null address.
895 */
896 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
897 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
898 (error = sodisconnect(so)))) {
899 error = EISCONN;
900 } else {
901 /*
902 * Prevent accumulated error from previous connection from
903 * biting us.
904 */
905 so->so_error = 0;
906 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
907 }
908 CURVNET_RESTORE();
909
910 return (error);
911 }
912
913 int
914 soconnect2(struct socket *so1, struct socket *so2)
915 {
916 int error;
917
918 CURVNET_SET(so1->so_vnet);
919 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
920 CURVNET_RESTORE();
921 return (error);
922 }
923
924 int
925 sodisconnect(struct socket *so)
926 {
927 int error;
928
929 if ((so->so_state & SS_ISCONNECTED) == 0)
930 return (ENOTCONN);
931 if (so->so_state & SS_ISDISCONNECTING)
932 return (EALREADY);
933 VNET_SO_ASSERT(so);
934 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
935 return (error);
936 }
937
938 #ifdef ZERO_COPY_SOCKETS
939 struct so_zerocopy_stats{
940 int size_ok;
941 int align_ok;
942 int found_ifp;
943 };
944 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
945
946 /*
947 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
948 * sosend_dgram() and sosend_generic() use m_uiotombuf().
949 *
950 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
951 * all of the data referenced by the uio. If desired, it uses zero-copy.
952 * *space will be updated to reflect data copied in.
953 *
954 * NB: If atomic I/O is requested, the caller must already have checked that
955 * space can hold resid bytes.
956 *
957 * NB: In the event of an error, the caller may need to free the partial
958 * chain pointed to by *mpp. The contents of both *uio and *space may be
959 * modified even in the case of an error.
960 */
961 static int
962 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
963 int flags)
964 {
965 struct mbuf *m, **mp, *top;
966 long len;
967 ssize_t resid;
968 int error;
969 int cow_send;
970
971 *retmp = top = NULL;
972 mp = ⊤
973 len = 0;
974 resid = uio->uio_resid;
975 error = 0;
976 do {
977 cow_send = 0;
978 if (resid >= MINCLSIZE) {
979 if (top == NULL) {
980 m = m_gethdr(M_WAITOK, MT_DATA);
981 m->m_pkthdr.len = 0;
982 m->m_pkthdr.rcvif = NULL;
983 } else
984 m = m_get(M_WAITOK, MT_DATA);
985 if (so_zero_copy_send &&
986 resid >= PAGE_SIZE &&
987 *space >= PAGE_SIZE &&
988 uio->uio_iov->iov_len >= PAGE_SIZE) {
989 so_zerocp_stats.size_ok++;
990 so_zerocp_stats.align_ok++;
991 cow_send = socow_setup(m, uio);
992 len = cow_send;
993 }
994 if (!cow_send) {
995 m_clget(m, M_WAITOK);
996 len = min(min(MCLBYTES, resid), *space);
997 }
998 } else {
999 if (top == NULL) {
1000 m = m_gethdr(M_WAIT, MT_DATA);
1001 m->m_pkthdr.len = 0;
1002 m->m_pkthdr.rcvif = NULL;
1003
1004 len = min(min(MHLEN, resid), *space);
1005 /*
1006 * For datagram protocols, leave room
1007 * for protocol headers in first mbuf.
1008 */
1009 if (atomic && m && len < MHLEN)
1010 MH_ALIGN(m, len);
1011 } else {
1012 m = m_get(M_WAIT, MT_DATA);
1013 len = min(min(MLEN, resid), *space);
1014 }
1015 }
1016 if (m == NULL) {
1017 error = ENOBUFS;
1018 goto out;
1019 }
1020
1021 *space -= len;
1022 if (cow_send)
1023 error = 0;
1024 else
1025 error = uiomove(mtod(m, void *), (int)len, uio);
1026 resid = uio->uio_resid;
1027 m->m_len = len;
1028 *mp = m;
1029 top->m_pkthdr.len += len;
1030 if (error)
1031 goto out;
1032 mp = &m->m_next;
1033 if (resid <= 0) {
1034 if (flags & MSG_EOR)
1035 top->m_flags |= M_EOR;
1036 break;
1037 }
1038 } while (*space > 0 && atomic);
1039 out:
1040 *retmp = top;
1041 return (error);
1042 }
1043 #endif /* ZERO_COPY_SOCKETS */
1044
1045 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1046
1047 int
1048 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1049 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1050 {
1051 long space;
1052 ssize_t resid;
1053 int clen = 0, error, dontroute;
1054 #ifdef ZERO_COPY_SOCKETS
1055 int atomic = sosendallatonce(so) || top;
1056 #endif
1057
1058 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
1059 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1060 ("sodgram_send: !PR_ATOMIC"));
1061
1062 if (uio != NULL)
1063 resid = uio->uio_resid;
1064 else
1065 resid = top->m_pkthdr.len;
1066 /*
1067 * In theory resid should be unsigned. However, space must be
1068 * signed, as it might be less than 0 if we over-committed, and we
1069 * must use a signed comparison of space and resid. On the other
1070 * hand, a negative resid causes us to loop sending 0-length
1071 * segments to the protocol.
1072 */
1073 if (resid < 0) {
1074 error = EINVAL;
1075 goto out;
1076 }
1077
1078 dontroute =
1079 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1080 if (td != NULL)
1081 td->td_ru.ru_msgsnd++;
1082 if (control != NULL)
1083 clen = control->m_len;
1084
1085 SOCKBUF_LOCK(&so->so_snd);
1086 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1087 SOCKBUF_UNLOCK(&so->so_snd);
1088 error = EPIPE;
1089 goto out;
1090 }
1091 if (so->so_error) {
1092 error = so->so_error;
1093 so->so_error = 0;
1094 SOCKBUF_UNLOCK(&so->so_snd);
1095 goto out;
1096 }
1097 if ((so->so_state & SS_ISCONNECTED) == 0) {
1098 /*
1099 * `sendto' and `sendmsg' is allowed on a connection-based
1100 * socket if it supports implied connect. Return ENOTCONN if
1101 * not connected and no address is supplied.
1102 */
1103 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1104 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1105 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1106 !(resid == 0 && clen != 0)) {
1107 SOCKBUF_UNLOCK(&so->so_snd);
1108 error = ENOTCONN;
1109 goto out;
1110 }
1111 } else if (addr == NULL) {
1112 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1113 error = ENOTCONN;
1114 else
1115 error = EDESTADDRREQ;
1116 SOCKBUF_UNLOCK(&so->so_snd);
1117 goto out;
1118 }
1119 }
1120
1121 /*
1122 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1123 * problem and need fixing.
1124 */
1125 space = sbspace(&so->so_snd);
1126 if (flags & MSG_OOB)
1127 space += 1024;
1128 space -= clen;
1129 SOCKBUF_UNLOCK(&so->so_snd);
1130 if (resid > space) {
1131 error = EMSGSIZE;
1132 goto out;
1133 }
1134 if (uio == NULL) {
1135 resid = 0;
1136 if (flags & MSG_EOR)
1137 top->m_flags |= M_EOR;
1138 } else {
1139 #ifdef ZERO_COPY_SOCKETS
1140 error = sosend_copyin(uio, &top, atomic, &space, flags);
1141 if (error)
1142 goto out;
1143 #else
1144 /*
1145 * Copy the data from userland into a mbuf chain.
1146 * If no data is to be copied in, a single empty mbuf
1147 * is returned.
1148 */
1149 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1150 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1151 if (top == NULL) {
1152 error = EFAULT; /* only possible error */
1153 goto out;
1154 }
1155 space -= resid - uio->uio_resid;
1156 #endif
1157 resid = uio->uio_resid;
1158 }
1159 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1160 /*
1161 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1162 * than with.
1163 */
1164 if (dontroute) {
1165 SOCK_LOCK(so);
1166 so->so_options |= SO_DONTROUTE;
1167 SOCK_UNLOCK(so);
1168 }
1169 /*
1170 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1171 * of date. We could have recieved a reset packet in an interrupt or
1172 * maybe we slept while doing page faults in uiomove() etc. We could
1173 * probably recheck again inside the locking protection here, but
1174 * there are probably other places that this also happens. We must
1175 * rethink this.
1176 */
1177 VNET_SO_ASSERT(so);
1178 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1179 (flags & MSG_OOB) ? PRUS_OOB :
1180 /*
1181 * If the user set MSG_EOF, the protocol understands this flag and
1182 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1183 */
1184 ((flags & MSG_EOF) &&
1185 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1186 (resid <= 0)) ?
1187 PRUS_EOF :
1188 /* If there is more to send set PRUS_MORETOCOME */
1189 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1190 top, addr, control, td);
1191 if (dontroute) {
1192 SOCK_LOCK(so);
1193 so->so_options &= ~SO_DONTROUTE;
1194 SOCK_UNLOCK(so);
1195 }
1196 clen = 0;
1197 control = NULL;
1198 top = NULL;
1199 out:
1200 if (top != NULL)
1201 m_freem(top);
1202 if (control != NULL)
1203 m_freem(control);
1204 return (error);
1205 }
1206
1207 /*
1208 * Send on a socket. If send must go all at once and message is larger than
1209 * send buffering, then hard error. Lock against other senders. If must go
1210 * all at once and not enough room now, then inform user that this would
1211 * block and do nothing. Otherwise, if nonblocking, send as much as
1212 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1213 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1214 * in mbuf chain must be small enough to send all at once.
1215 *
1216 * Returns nonzero on error, timeout or signal; callers must check for short
1217 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1218 * on return.
1219 */
1220 int
1221 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1222 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1223 {
1224 long space;
1225 ssize_t resid;
1226 int clen = 0, error, dontroute;
1227 int atomic = sosendallatonce(so) || top;
1228
1229 if (uio != NULL)
1230 resid = uio->uio_resid;
1231 else
1232 resid = top->m_pkthdr.len;
1233 /*
1234 * In theory resid should be unsigned. However, space must be
1235 * signed, as it might be less than 0 if we over-committed, and we
1236 * must use a signed comparison of space and resid. On the other
1237 * hand, a negative resid causes us to loop sending 0-length
1238 * segments to the protocol.
1239 *
1240 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1241 * type sockets since that's an error.
1242 */
1243 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1244 error = EINVAL;
1245 goto out;
1246 }
1247
1248 dontroute =
1249 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1250 (so->so_proto->pr_flags & PR_ATOMIC);
1251 if (td != NULL)
1252 td->td_ru.ru_msgsnd++;
1253 if (control != NULL)
1254 clen = control->m_len;
1255
1256 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1257 if (error)
1258 goto out;
1259
1260 restart:
1261 do {
1262 SOCKBUF_LOCK(&so->so_snd);
1263 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1264 SOCKBUF_UNLOCK(&so->so_snd);
1265 error = EPIPE;
1266 goto release;
1267 }
1268 if (so->so_error) {
1269 error = so->so_error;
1270 so->so_error = 0;
1271 SOCKBUF_UNLOCK(&so->so_snd);
1272 goto release;
1273 }
1274 if ((so->so_state & SS_ISCONNECTED) == 0) {
1275 /*
1276 * `sendto' and `sendmsg' is allowed on a connection-
1277 * based socket if it supports implied connect.
1278 * Return ENOTCONN if not connected and no address is
1279 * supplied.
1280 */
1281 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1282 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1283 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1284 !(resid == 0 && clen != 0)) {
1285 SOCKBUF_UNLOCK(&so->so_snd);
1286 error = ENOTCONN;
1287 goto release;
1288 }
1289 } else if (addr == NULL) {
1290 SOCKBUF_UNLOCK(&so->so_snd);
1291 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1292 error = ENOTCONN;
1293 else
1294 error = EDESTADDRREQ;
1295 goto release;
1296 }
1297 }
1298 space = sbspace(&so->so_snd);
1299 if (flags & MSG_OOB)
1300 space += 1024;
1301 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1302 clen > so->so_snd.sb_hiwat) {
1303 SOCKBUF_UNLOCK(&so->so_snd);
1304 error = EMSGSIZE;
1305 goto release;
1306 }
1307 if (space < resid + clen &&
1308 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1309 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1310 SOCKBUF_UNLOCK(&so->so_snd);
1311 error = EWOULDBLOCK;
1312 goto release;
1313 }
1314 error = sbwait(&so->so_snd);
1315 SOCKBUF_UNLOCK(&so->so_snd);
1316 if (error)
1317 goto release;
1318 goto restart;
1319 }
1320 SOCKBUF_UNLOCK(&so->so_snd);
1321 space -= clen;
1322 do {
1323 if (uio == NULL) {
1324 resid = 0;
1325 if (flags & MSG_EOR)
1326 top->m_flags |= M_EOR;
1327 } else {
1328 #ifdef ZERO_COPY_SOCKETS
1329 error = sosend_copyin(uio, &top, atomic,
1330 &space, flags);
1331 if (error != 0)
1332 goto release;
1333 #else
1334 /*
1335 * Copy the data from userland into a mbuf
1336 * chain. If no data is to be copied in,
1337 * a single empty mbuf is returned.
1338 */
1339 top = m_uiotombuf(uio, M_WAITOK, space,
1340 (atomic ? max_hdr : 0),
1341 (atomic ? M_PKTHDR : 0) |
1342 ((flags & MSG_EOR) ? M_EOR : 0));
1343 if (top == NULL) {
1344 error = EFAULT; /* only possible error */
1345 goto release;
1346 }
1347 space -= resid - uio->uio_resid;
1348 #endif
1349 resid = uio->uio_resid;
1350 }
1351 if (dontroute) {
1352 SOCK_LOCK(so);
1353 so->so_options |= SO_DONTROUTE;
1354 SOCK_UNLOCK(so);
1355 }
1356 /*
1357 * XXX all the SBS_CANTSENDMORE checks previously
1358 * done could be out of date. We could have recieved
1359 * a reset packet in an interrupt or maybe we slept
1360 * while doing page faults in uiomove() etc. We
1361 * could probably recheck again inside the locking
1362 * protection here, but there are probably other
1363 * places that this also happens. We must rethink
1364 * this.
1365 */
1366 VNET_SO_ASSERT(so);
1367 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1368 (flags & MSG_OOB) ? PRUS_OOB :
1369 /*
1370 * If the user set MSG_EOF, the protocol understands
1371 * this flag and nothing left to send then use
1372 * PRU_SEND_EOF instead of PRU_SEND.
1373 */
1374 ((flags & MSG_EOF) &&
1375 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1376 (resid <= 0)) ?
1377 PRUS_EOF :
1378 /* If there is more to send set PRUS_MORETOCOME. */
1379 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1380 top, addr, control, td);
1381 if (dontroute) {
1382 SOCK_LOCK(so);
1383 so->so_options &= ~SO_DONTROUTE;
1384 SOCK_UNLOCK(so);
1385 }
1386 clen = 0;
1387 control = NULL;
1388 top = NULL;
1389 if (error)
1390 goto release;
1391 } while (resid && space > 0);
1392 } while (resid);
1393
1394 release:
1395 sbunlock(&so->so_snd);
1396 out:
1397 if (top != NULL)
1398 m_freem(top);
1399 if (control != NULL)
1400 m_freem(control);
1401 return (error);
1402 }
1403
1404 int
1405 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1406 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1407 {
1408 int error;
1409
1410 CURVNET_SET(so->so_vnet);
1411 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1412 control, flags, td);
1413 CURVNET_RESTORE();
1414 return (error);
1415 }
1416
1417 /*
1418 * The part of soreceive() that implements reading non-inline out-of-band
1419 * data from a socket. For more complete comments, see soreceive(), from
1420 * which this code originated.
1421 *
1422 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1423 * unable to return an mbuf chain to the caller.
1424 */
1425 static int
1426 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1427 {
1428 struct protosw *pr = so->so_proto;
1429 struct mbuf *m;
1430 int error;
1431
1432 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1433 VNET_SO_ASSERT(so);
1434
1435 m = m_get(M_WAIT, MT_DATA);
1436 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1437 if (error)
1438 goto bad;
1439 do {
1440 #ifdef ZERO_COPY_SOCKETS
1441 if (so_zero_copy_receive) {
1442 int disposable;
1443
1444 if ((m->m_flags & M_EXT)
1445 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1446 disposable = 1;
1447 else
1448 disposable = 0;
1449
1450 error = uiomoveco(mtod(m, void *),
1451 min(uio->uio_resid, m->m_len),
1452 uio, disposable);
1453 } else
1454 #endif /* ZERO_COPY_SOCKETS */
1455 error = uiomove(mtod(m, void *),
1456 (int) min(uio->uio_resid, m->m_len), uio);
1457 m = m_free(m);
1458 } while (uio->uio_resid && error == 0 && m);
1459 bad:
1460 if (m != NULL)
1461 m_freem(m);
1462 return (error);
1463 }
1464
1465 /*
1466 * Following replacement or removal of the first mbuf on the first mbuf chain
1467 * of a socket buffer, push necessary state changes back into the socket
1468 * buffer so that other consumers see the values consistently. 'nextrecord'
1469 * is the callers locally stored value of the original value of
1470 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1471 * NOTE: 'nextrecord' may be NULL.
1472 */
1473 static __inline void
1474 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1475 {
1476
1477 SOCKBUF_LOCK_ASSERT(sb);
1478 /*
1479 * First, update for the new value of nextrecord. If necessary, make
1480 * it the first record.
1481 */
1482 if (sb->sb_mb != NULL)
1483 sb->sb_mb->m_nextpkt = nextrecord;
1484 else
1485 sb->sb_mb = nextrecord;
1486
1487 /*
1488 * Now update any dependent socket buffer fields to reflect the new
1489 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1490 * addition of a second clause that takes care of the case where
1491 * sb_mb has been updated, but remains the last record.
1492 */
1493 if (sb->sb_mb == NULL) {
1494 sb->sb_mbtail = NULL;
1495 sb->sb_lastrecord = NULL;
1496 } else if (sb->sb_mb->m_nextpkt == NULL)
1497 sb->sb_lastrecord = sb->sb_mb;
1498 }
1499
1500
1501 /*
1502 * Implement receive operations on a socket. We depend on the way that
1503 * records are added to the sockbuf by sbappend. In particular, each record
1504 * (mbufs linked through m_next) must begin with an address if the protocol
1505 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1506 * data, and then zero or more mbufs of data. In order to allow parallelism
1507 * between network receive and copying to user space, as well as avoid
1508 * sleeping with a mutex held, we release the socket buffer mutex during the
1509 * user space copy. Although the sockbuf is locked, new data may still be
1510 * appended, and thus we must maintain consistency of the sockbuf during that
1511 * time.
1512 *
1513 * The caller may receive the data as a single mbuf chain by supplying an
1514 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1515 * the count in uio_resid.
1516 */
1517 int
1518 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1519 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1520 {
1521 struct mbuf *m, **mp;
1522 int flags, error, offset;
1523 ssize_t len;
1524 struct protosw *pr = so->so_proto;
1525 struct mbuf *nextrecord;
1526 int moff, type = 0;
1527 ssize_t orig_resid = uio->uio_resid;
1528
1529 mp = mp0;
1530 if (psa != NULL)
1531 *psa = NULL;
1532 if (controlp != NULL)
1533 *controlp = NULL;
1534 if (flagsp != NULL)
1535 flags = *flagsp &~ MSG_EOR;
1536 else
1537 flags = 0;
1538 if (flags & MSG_OOB)
1539 return (soreceive_rcvoob(so, uio, flags));
1540 if (mp != NULL)
1541 *mp = NULL;
1542 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1543 && uio->uio_resid) {
1544 VNET_SO_ASSERT(so);
1545 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1546 }
1547
1548 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1549 if (error)
1550 return (error);
1551
1552 restart:
1553 SOCKBUF_LOCK(&so->so_rcv);
1554 m = so->so_rcv.sb_mb;
1555 /*
1556 * If we have less data than requested, block awaiting more (subject
1557 * to any timeout) if:
1558 * 1. the current count is less than the low water mark, or
1559 * 2. MSG_DONTWAIT is not set
1560 */
1561 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1562 so->so_rcv.sb_cc < uio->uio_resid) &&
1563 so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1564 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1565 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1566 ("receive: m == %p so->so_rcv.sb_cc == %u",
1567 m, so->so_rcv.sb_cc));
1568 if (so->so_error) {
1569 if (m != NULL)
1570 goto dontblock;
1571 error = so->so_error;
1572 if ((flags & MSG_PEEK) == 0)
1573 so->so_error = 0;
1574 SOCKBUF_UNLOCK(&so->so_rcv);
1575 goto release;
1576 }
1577 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1578 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1579 if (m == NULL) {
1580 SOCKBUF_UNLOCK(&so->so_rcv);
1581 goto release;
1582 } else
1583 goto dontblock;
1584 }
1585 for (; m != NULL; m = m->m_next)
1586 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1587 m = so->so_rcv.sb_mb;
1588 goto dontblock;
1589 }
1590 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1591 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1592 SOCKBUF_UNLOCK(&so->so_rcv);
1593 error = ENOTCONN;
1594 goto release;
1595 }
1596 if (uio->uio_resid == 0) {
1597 SOCKBUF_UNLOCK(&so->so_rcv);
1598 goto release;
1599 }
1600 if ((so->so_state & SS_NBIO) ||
1601 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1602 SOCKBUF_UNLOCK(&so->so_rcv);
1603 error = EWOULDBLOCK;
1604 goto release;
1605 }
1606 SBLASTRECORDCHK(&so->so_rcv);
1607 SBLASTMBUFCHK(&so->so_rcv);
1608 error = sbwait(&so->so_rcv);
1609 SOCKBUF_UNLOCK(&so->so_rcv);
1610 if (error)
1611 goto release;
1612 goto restart;
1613 }
1614 dontblock:
1615 /*
1616 * From this point onward, we maintain 'nextrecord' as a cache of the
1617 * pointer to the next record in the socket buffer. We must keep the
1618 * various socket buffer pointers and local stack versions of the
1619 * pointers in sync, pushing out modifications before dropping the
1620 * socket buffer mutex, and re-reading them when picking it up.
1621 *
1622 * Otherwise, we will race with the network stack appending new data
1623 * or records onto the socket buffer by using inconsistent/stale
1624 * versions of the field, possibly resulting in socket buffer
1625 * corruption.
1626 *
1627 * By holding the high-level sblock(), we prevent simultaneous
1628 * readers from pulling off the front of the socket buffer.
1629 */
1630 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1631 if (uio->uio_td)
1632 uio->uio_td->td_ru.ru_msgrcv++;
1633 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1634 SBLASTRECORDCHK(&so->so_rcv);
1635 SBLASTMBUFCHK(&so->so_rcv);
1636 nextrecord = m->m_nextpkt;
1637 if (pr->pr_flags & PR_ADDR) {
1638 KASSERT(m->m_type == MT_SONAME,
1639 ("m->m_type == %d", m->m_type));
1640 orig_resid = 0;
1641 if (psa != NULL)
1642 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1643 M_NOWAIT);
1644 if (flags & MSG_PEEK) {
1645 m = m->m_next;
1646 } else {
1647 sbfree(&so->so_rcv, m);
1648 so->so_rcv.sb_mb = m_free(m);
1649 m = so->so_rcv.sb_mb;
1650 sockbuf_pushsync(&so->so_rcv, nextrecord);
1651 }
1652 }
1653
1654 /*
1655 * Process one or more MT_CONTROL mbufs present before any data mbufs
1656 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1657 * just copy the data; if !MSG_PEEK, we call into the protocol to
1658 * perform externalization (or freeing if controlp == NULL).
1659 */
1660 if (m != NULL && m->m_type == MT_CONTROL) {
1661 struct mbuf *cm = NULL, *cmn;
1662 struct mbuf **cme = &cm;
1663
1664 do {
1665 if (flags & MSG_PEEK) {
1666 if (controlp != NULL) {
1667 *controlp = m_copy(m, 0, m->m_len);
1668 controlp = &(*controlp)->m_next;
1669 }
1670 m = m->m_next;
1671 } else {
1672 sbfree(&so->so_rcv, m);
1673 so->so_rcv.sb_mb = m->m_next;
1674 m->m_next = NULL;
1675 *cme = m;
1676 cme = &(*cme)->m_next;
1677 m = so->so_rcv.sb_mb;
1678 }
1679 } while (m != NULL && m->m_type == MT_CONTROL);
1680 if ((flags & MSG_PEEK) == 0)
1681 sockbuf_pushsync(&so->so_rcv, nextrecord);
1682 while (cm != NULL) {
1683 cmn = cm->m_next;
1684 cm->m_next = NULL;
1685 if (pr->pr_domain->dom_externalize != NULL) {
1686 SOCKBUF_UNLOCK(&so->so_rcv);
1687 VNET_SO_ASSERT(so);
1688 error = (*pr->pr_domain->dom_externalize)
1689 (cm, controlp);
1690 SOCKBUF_LOCK(&so->so_rcv);
1691 } else if (controlp != NULL)
1692 *controlp = cm;
1693 else
1694 m_freem(cm);
1695 if (controlp != NULL) {
1696 orig_resid = 0;
1697 while (*controlp != NULL)
1698 controlp = &(*controlp)->m_next;
1699 }
1700 cm = cmn;
1701 }
1702 if (m != NULL)
1703 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1704 else
1705 nextrecord = so->so_rcv.sb_mb;
1706 orig_resid = 0;
1707 }
1708 if (m != NULL) {
1709 if ((flags & MSG_PEEK) == 0) {
1710 KASSERT(m->m_nextpkt == nextrecord,
1711 ("soreceive: post-control, nextrecord !sync"));
1712 if (nextrecord == NULL) {
1713 KASSERT(so->so_rcv.sb_mb == m,
1714 ("soreceive: post-control, sb_mb!=m"));
1715 KASSERT(so->so_rcv.sb_lastrecord == m,
1716 ("soreceive: post-control, lastrecord!=m"));
1717 }
1718 }
1719 type = m->m_type;
1720 if (type == MT_OOBDATA)
1721 flags |= MSG_OOB;
1722 } else {
1723 if ((flags & MSG_PEEK) == 0) {
1724 KASSERT(so->so_rcv.sb_mb == nextrecord,
1725 ("soreceive: sb_mb != nextrecord"));
1726 if (so->so_rcv.sb_mb == NULL) {
1727 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1728 ("soreceive: sb_lastercord != NULL"));
1729 }
1730 }
1731 }
1732 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1733 SBLASTRECORDCHK(&so->so_rcv);
1734 SBLASTMBUFCHK(&so->so_rcv);
1735
1736 /*
1737 * Now continue to read any data mbufs off of the head of the socket
1738 * buffer until the read request is satisfied. Note that 'type' is
1739 * used to store the type of any mbuf reads that have happened so far
1740 * such that soreceive() can stop reading if the type changes, which
1741 * causes soreceive() to return only one of regular data and inline
1742 * out-of-band data in a single socket receive operation.
1743 */
1744 moff = 0;
1745 offset = 0;
1746 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1747 /*
1748 * If the type of mbuf has changed since the last mbuf
1749 * examined ('type'), end the receive operation.
1750 */
1751 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1752 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1753 if (type != m->m_type)
1754 break;
1755 } else if (type == MT_OOBDATA)
1756 break;
1757 else
1758 KASSERT(m->m_type == MT_DATA,
1759 ("m->m_type == %d", m->m_type));
1760 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1761 len = uio->uio_resid;
1762 if (so->so_oobmark && len > so->so_oobmark - offset)
1763 len = so->so_oobmark - offset;
1764 if (len > m->m_len - moff)
1765 len = m->m_len - moff;
1766 /*
1767 * If mp is set, just pass back the mbufs. Otherwise copy
1768 * them out via the uio, then free. Sockbuf must be
1769 * consistent here (points to current mbuf, it points to next
1770 * record) when we drop priority; we must note any additions
1771 * to the sockbuf when we block interrupts again.
1772 */
1773 if (mp == NULL) {
1774 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1775 SBLASTRECORDCHK(&so->so_rcv);
1776 SBLASTMBUFCHK(&so->so_rcv);
1777 SOCKBUF_UNLOCK(&so->so_rcv);
1778 #ifdef ZERO_COPY_SOCKETS
1779 if (so_zero_copy_receive) {
1780 int disposable;
1781
1782 if ((m->m_flags & M_EXT)
1783 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1784 disposable = 1;
1785 else
1786 disposable = 0;
1787
1788 error = uiomoveco(mtod(m, char *) + moff,
1789 (int)len, uio,
1790 disposable);
1791 } else
1792 #endif /* ZERO_COPY_SOCKETS */
1793 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1794 SOCKBUF_LOCK(&so->so_rcv);
1795 if (error) {
1796 /*
1797 * The MT_SONAME mbuf has already been removed
1798 * from the record, so it is necessary to
1799 * remove the data mbufs, if any, to preserve
1800 * the invariant in the case of PR_ADDR that
1801 * requires MT_SONAME mbufs at the head of
1802 * each record.
1803 */
1804 if (m && pr->pr_flags & PR_ATOMIC &&
1805 ((flags & MSG_PEEK) == 0))
1806 (void)sbdroprecord_locked(&so->so_rcv);
1807 SOCKBUF_UNLOCK(&so->so_rcv);
1808 goto release;
1809 }
1810 } else
1811 uio->uio_resid -= len;
1812 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1813 if (len == m->m_len - moff) {
1814 if (m->m_flags & M_EOR)
1815 flags |= MSG_EOR;
1816 if (flags & MSG_PEEK) {
1817 m = m->m_next;
1818 moff = 0;
1819 } else {
1820 nextrecord = m->m_nextpkt;
1821 sbfree(&so->so_rcv, m);
1822 if (mp != NULL) {
1823 *mp = m;
1824 mp = &m->m_next;
1825 so->so_rcv.sb_mb = m = m->m_next;
1826 *mp = NULL;
1827 } else {
1828 so->so_rcv.sb_mb = m_free(m);
1829 m = so->so_rcv.sb_mb;
1830 }
1831 sockbuf_pushsync(&so->so_rcv, nextrecord);
1832 SBLASTRECORDCHK(&so->so_rcv);
1833 SBLASTMBUFCHK(&so->so_rcv);
1834 }
1835 } else {
1836 if (flags & MSG_PEEK)
1837 moff += len;
1838 else {
1839 if (mp != NULL) {
1840 int copy_flag;
1841
1842 if (flags & MSG_DONTWAIT)
1843 copy_flag = M_DONTWAIT;
1844 else
1845 copy_flag = M_WAIT;
1846 if (copy_flag == M_WAIT)
1847 SOCKBUF_UNLOCK(&so->so_rcv);
1848 *mp = m_copym(m, 0, len, copy_flag);
1849 if (copy_flag == M_WAIT)
1850 SOCKBUF_LOCK(&so->so_rcv);
1851 if (*mp == NULL) {
1852 /*
1853 * m_copym() couldn't
1854 * allocate an mbuf. Adjust
1855 * uio_resid back (it was
1856 * adjusted down by len
1857 * bytes, which we didn't end
1858 * up "copying" over).
1859 */
1860 uio->uio_resid += len;
1861 break;
1862 }
1863 }
1864 m->m_data += len;
1865 m->m_len -= len;
1866 so->so_rcv.sb_cc -= len;
1867 }
1868 }
1869 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1870 if (so->so_oobmark) {
1871 if ((flags & MSG_PEEK) == 0) {
1872 so->so_oobmark -= len;
1873 if (so->so_oobmark == 0) {
1874 so->so_rcv.sb_state |= SBS_RCVATMARK;
1875 break;
1876 }
1877 } else {
1878 offset += len;
1879 if (offset == so->so_oobmark)
1880 break;
1881 }
1882 }
1883 if (flags & MSG_EOR)
1884 break;
1885 /*
1886 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1887 * must not quit until "uio->uio_resid == 0" or an error
1888 * termination. If a signal/timeout occurs, return with a
1889 * short count but without error. Keep sockbuf locked
1890 * against other readers.
1891 */
1892 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1893 !sosendallatonce(so) && nextrecord == NULL) {
1894 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1895 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1896 break;
1897 /*
1898 * Notify the protocol that some data has been
1899 * drained before blocking.
1900 */
1901 if (pr->pr_flags & PR_WANTRCVD) {
1902 SOCKBUF_UNLOCK(&so->so_rcv);
1903 VNET_SO_ASSERT(so);
1904 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1905 SOCKBUF_LOCK(&so->so_rcv);
1906 }
1907 SBLASTRECORDCHK(&so->so_rcv);
1908 SBLASTMBUFCHK(&so->so_rcv);
1909 /*
1910 * We could receive some data while was notifying
1911 * the protocol. Skip blocking in this case.
1912 */
1913 if (so->so_rcv.sb_mb == NULL) {
1914 error = sbwait(&so->so_rcv);
1915 if (error) {
1916 SOCKBUF_UNLOCK(&so->so_rcv);
1917 goto release;
1918 }
1919 }
1920 m = so->so_rcv.sb_mb;
1921 if (m != NULL)
1922 nextrecord = m->m_nextpkt;
1923 }
1924 }
1925
1926 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1927 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1928 flags |= MSG_TRUNC;
1929 if ((flags & MSG_PEEK) == 0)
1930 (void) sbdroprecord_locked(&so->so_rcv);
1931 }
1932 if ((flags & MSG_PEEK) == 0) {
1933 if (m == NULL) {
1934 /*
1935 * First part is an inline SB_EMPTY_FIXUP(). Second
1936 * part makes sure sb_lastrecord is up-to-date if
1937 * there is still data in the socket buffer.
1938 */
1939 so->so_rcv.sb_mb = nextrecord;
1940 if (so->so_rcv.sb_mb == NULL) {
1941 so->so_rcv.sb_mbtail = NULL;
1942 so->so_rcv.sb_lastrecord = NULL;
1943 } else if (nextrecord->m_nextpkt == NULL)
1944 so->so_rcv.sb_lastrecord = nextrecord;
1945 }
1946 SBLASTRECORDCHK(&so->so_rcv);
1947 SBLASTMBUFCHK(&so->so_rcv);
1948 /*
1949 * If soreceive() is being done from the socket callback,
1950 * then don't need to generate ACK to peer to update window,
1951 * since ACK will be generated on return to TCP.
1952 */
1953 if (!(flags & MSG_SOCALLBCK) &&
1954 (pr->pr_flags & PR_WANTRCVD)) {
1955 SOCKBUF_UNLOCK(&so->so_rcv);
1956 VNET_SO_ASSERT(so);
1957 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1958 SOCKBUF_LOCK(&so->so_rcv);
1959 }
1960 }
1961 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1962 if (orig_resid == uio->uio_resid && orig_resid &&
1963 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1964 SOCKBUF_UNLOCK(&so->so_rcv);
1965 goto restart;
1966 }
1967 SOCKBUF_UNLOCK(&so->so_rcv);
1968
1969 if (flagsp != NULL)
1970 *flagsp |= flags;
1971 release:
1972 sbunlock(&so->so_rcv);
1973 return (error);
1974 }
1975
1976 /*
1977 * Optimized version of soreceive() for stream (TCP) sockets.
1978 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1979 */
1980 int
1981 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1982 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1983 {
1984 int len = 0, error = 0, flags, oresid;
1985 struct sockbuf *sb;
1986 struct mbuf *m, *n = NULL;
1987
1988 /* We only do stream sockets. */
1989 if (so->so_type != SOCK_STREAM)
1990 return (EINVAL);
1991 if (psa != NULL)
1992 *psa = NULL;
1993 if (controlp != NULL)
1994 return (EINVAL);
1995 if (flagsp != NULL)
1996 flags = *flagsp &~ MSG_EOR;
1997 else
1998 flags = 0;
1999 if (flags & MSG_OOB)
2000 return (soreceive_rcvoob(so, uio, flags));
2001 if (mp0 != NULL)
2002 *mp0 = NULL;
2003
2004 sb = &so->so_rcv;
2005
2006 /* Prevent other readers from entering the socket. */
2007 error = sblock(sb, SBLOCKWAIT(flags));
2008 if (error)
2009 goto out;
2010 SOCKBUF_LOCK(sb);
2011
2012 /* Easy one, no space to copyout anything. */
2013 if (uio->uio_resid == 0) {
2014 error = EINVAL;
2015 goto out;
2016 }
2017 oresid = uio->uio_resid;
2018
2019 /* We will never ever get anything unless we are or were connected. */
2020 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2021 error = ENOTCONN;
2022 goto out;
2023 }
2024
2025 restart:
2026 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2027
2028 /* Abort if socket has reported problems. */
2029 if (so->so_error) {
2030 if (sb->sb_cc > 0)
2031 goto deliver;
2032 if (oresid > uio->uio_resid)
2033 goto out;
2034 error = so->so_error;
2035 if (!(flags & MSG_PEEK))
2036 so->so_error = 0;
2037 goto out;
2038 }
2039
2040 /* Door is closed. Deliver what is left, if any. */
2041 if (sb->sb_state & SBS_CANTRCVMORE) {
2042 if (sb->sb_cc > 0)
2043 goto deliver;
2044 else
2045 goto out;
2046 }
2047
2048 /* Socket buffer is empty and we shall not block. */
2049 if (sb->sb_cc == 0 &&
2050 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2051 error = EAGAIN;
2052 goto out;
2053 }
2054
2055 /* Socket buffer got some data that we shall deliver now. */
2056 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
2057 ((sb->sb_flags & SS_NBIO) ||
2058 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2059 sb->sb_cc >= sb->sb_lowat ||
2060 sb->sb_cc >= uio->uio_resid ||
2061 sb->sb_cc >= sb->sb_hiwat) ) {
2062 goto deliver;
2063 }
2064
2065 /* On MSG_WAITALL we must wait until all data or error arrives. */
2066 if ((flags & MSG_WAITALL) &&
2067 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
2068 goto deliver;
2069
2070 /*
2071 * Wait and block until (more) data comes in.
2072 * NB: Drops the sockbuf lock during wait.
2073 */
2074 error = sbwait(sb);
2075 if (error)
2076 goto out;
2077 goto restart;
2078
2079 deliver:
2080 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2081 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2082 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2083
2084 /* Statistics. */
2085 if (uio->uio_td)
2086 uio->uio_td->td_ru.ru_msgrcv++;
2087
2088 /* Fill uio until full or current end of socket buffer is reached. */
2089 len = min(uio->uio_resid, sb->sb_cc);
2090 if (mp0 != NULL) {
2091 /* Dequeue as many mbufs as possible. */
2092 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2093 if (*mp0 == NULL)
2094 *mp0 = sb->sb_mb;
2095 else
2096 m_cat(*mp0, sb->sb_mb);
2097 for (m = sb->sb_mb;
2098 m != NULL && m->m_len <= len;
2099 m = m->m_next) {
2100 len -= m->m_len;
2101 uio->uio_resid -= m->m_len;
2102 sbfree(sb, m);
2103 n = m;
2104 }
2105 n->m_next = NULL;
2106 sb->sb_mb = m;
2107 sb->sb_lastrecord = sb->sb_mb;
2108 if (sb->sb_mb == NULL)
2109 SB_EMPTY_FIXUP(sb);
2110 }
2111 /* Copy the remainder. */
2112 if (len > 0) {
2113 KASSERT(sb->sb_mb != NULL,
2114 ("%s: len > 0 && sb->sb_mb empty", __func__));
2115
2116 m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2117 if (m == NULL)
2118 len = 0; /* Don't flush data from sockbuf. */
2119 else
2120 uio->uio_resid -= len;
2121 if (*mp0 != NULL)
2122 m_cat(*mp0, m);
2123 else
2124 *mp0 = m;
2125 if (*mp0 == NULL) {
2126 error = ENOBUFS;
2127 goto out;
2128 }
2129 }
2130 } else {
2131 /* NB: Must unlock socket buffer as uiomove may sleep. */
2132 SOCKBUF_UNLOCK(sb);
2133 error = m_mbuftouio(uio, sb->sb_mb, len);
2134 SOCKBUF_LOCK(sb);
2135 if (error)
2136 goto out;
2137 }
2138 SBLASTRECORDCHK(sb);
2139 SBLASTMBUFCHK(sb);
2140
2141 /*
2142 * Remove the delivered data from the socket buffer unless we
2143 * were only peeking.
2144 */
2145 if (!(flags & MSG_PEEK)) {
2146 if (len > 0)
2147 sbdrop_locked(sb, len);
2148
2149 /* Notify protocol that we drained some data. */
2150 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2151 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2152 !(flags & MSG_SOCALLBCK))) {
2153 SOCKBUF_UNLOCK(sb);
2154 VNET_SO_ASSERT(so);
2155 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2156 SOCKBUF_LOCK(sb);
2157 }
2158 }
2159
2160 /*
2161 * For MSG_WAITALL we may have to loop again and wait for
2162 * more data to come in.
2163 */
2164 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2165 goto restart;
2166 out:
2167 SOCKBUF_LOCK_ASSERT(sb);
2168 SBLASTRECORDCHK(sb);
2169 SBLASTMBUFCHK(sb);
2170 SOCKBUF_UNLOCK(sb);
2171 sbunlock(sb);
2172 return (error);
2173 }
2174
2175 /*
2176 * Optimized version of soreceive() for simple datagram cases from userspace.
2177 * Unlike in the stream case, we're able to drop a datagram if copyout()
2178 * fails, and because we handle datagrams atomically, we don't need to use a
2179 * sleep lock to prevent I/O interlacing.
2180 */
2181 int
2182 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2183 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2184 {
2185 struct mbuf *m, *m2;
2186 int flags, error;
2187 ssize_t len;
2188 struct protosw *pr = so->so_proto;
2189 struct mbuf *nextrecord;
2190
2191 if (psa != NULL)
2192 *psa = NULL;
2193 if (controlp != NULL)
2194 *controlp = NULL;
2195 if (flagsp != NULL)
2196 flags = *flagsp &~ MSG_EOR;
2197 else
2198 flags = 0;
2199
2200 /*
2201 * For any complicated cases, fall back to the full
2202 * soreceive_generic().
2203 */
2204 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2205 return (soreceive_generic(so, psa, uio, mp0, controlp,
2206 flagsp));
2207
2208 /*
2209 * Enforce restrictions on use.
2210 */
2211 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2212 ("soreceive_dgram: wantrcvd"));
2213 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2214 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2215 ("soreceive_dgram: SBS_RCVATMARK"));
2216 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2217 ("soreceive_dgram: P_CONNREQUIRED"));
2218
2219 /*
2220 * Loop blocking while waiting for a datagram.
2221 */
2222 SOCKBUF_LOCK(&so->so_rcv);
2223 while ((m = so->so_rcv.sb_mb) == NULL) {
2224 KASSERT(so->so_rcv.sb_cc == 0,
2225 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2226 so->so_rcv.sb_cc));
2227 if (so->so_error) {
2228 error = so->so_error;
2229 so->so_error = 0;
2230 SOCKBUF_UNLOCK(&so->so_rcv);
2231 return (error);
2232 }
2233 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2234 uio->uio_resid == 0) {
2235 SOCKBUF_UNLOCK(&so->so_rcv);
2236 return (0);
2237 }
2238 if ((so->so_state & SS_NBIO) ||
2239 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2240 SOCKBUF_UNLOCK(&so->so_rcv);
2241 return (EWOULDBLOCK);
2242 }
2243 SBLASTRECORDCHK(&so->so_rcv);
2244 SBLASTMBUFCHK(&so->so_rcv);
2245 error = sbwait(&so->so_rcv);
2246 if (error) {
2247 SOCKBUF_UNLOCK(&so->so_rcv);
2248 return (error);
2249 }
2250 }
2251 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2252
2253 if (uio->uio_td)
2254 uio->uio_td->td_ru.ru_msgrcv++;
2255 SBLASTRECORDCHK(&so->so_rcv);
2256 SBLASTMBUFCHK(&so->so_rcv);
2257 nextrecord = m->m_nextpkt;
2258 if (nextrecord == NULL) {
2259 KASSERT(so->so_rcv.sb_lastrecord == m,
2260 ("soreceive_dgram: lastrecord != m"));
2261 }
2262
2263 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2264 ("soreceive_dgram: m_nextpkt != nextrecord"));
2265
2266 /*
2267 * Pull 'm' and its chain off the front of the packet queue.
2268 */
2269 so->so_rcv.sb_mb = NULL;
2270 sockbuf_pushsync(&so->so_rcv, nextrecord);
2271
2272 /*
2273 * Walk 'm's chain and free that many bytes from the socket buffer.
2274 */
2275 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2276 sbfree(&so->so_rcv, m2);
2277
2278 /*
2279 * Do a few last checks before we let go of the lock.
2280 */
2281 SBLASTRECORDCHK(&so->so_rcv);
2282 SBLASTMBUFCHK(&so->so_rcv);
2283 SOCKBUF_UNLOCK(&so->so_rcv);
2284
2285 if (pr->pr_flags & PR_ADDR) {
2286 KASSERT(m->m_type == MT_SONAME,
2287 ("m->m_type == %d", m->m_type));
2288 if (psa != NULL)
2289 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2290 M_NOWAIT);
2291 m = m_free(m);
2292 }
2293 if (m == NULL) {
2294 /* XXXRW: Can this happen? */
2295 return (0);
2296 }
2297
2298 /*
2299 * Packet to copyout() is now in 'm' and it is disconnected from the
2300 * queue.
2301 *
2302 * Process one or more MT_CONTROL mbufs present before any data mbufs
2303 * in the first mbuf chain on the socket buffer. We call into the
2304 * protocol to perform externalization (or freeing if controlp ==
2305 * NULL).
2306 */
2307 if (m->m_type == MT_CONTROL) {
2308 struct mbuf *cm = NULL, *cmn;
2309 struct mbuf **cme = &cm;
2310
2311 do {
2312 m2 = m->m_next;
2313 m->m_next = NULL;
2314 *cme = m;
2315 cme = &(*cme)->m_next;
2316 m = m2;
2317 } while (m != NULL && m->m_type == MT_CONTROL);
2318 while (cm != NULL) {
2319 cmn = cm->m_next;
2320 cm->m_next = NULL;
2321 if (pr->pr_domain->dom_externalize != NULL) {
2322 error = (*pr->pr_domain->dom_externalize)
2323 (cm, controlp);
2324 } else if (controlp != NULL)
2325 *controlp = cm;
2326 else
2327 m_freem(cm);
2328 if (controlp != NULL) {
2329 while (*controlp != NULL)
2330 controlp = &(*controlp)->m_next;
2331 }
2332 cm = cmn;
2333 }
2334 }
2335 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2336
2337 while (m != NULL && uio->uio_resid > 0) {
2338 len = uio->uio_resid;
2339 if (len > m->m_len)
2340 len = m->m_len;
2341 error = uiomove(mtod(m, char *), (int)len, uio);
2342 if (error) {
2343 m_freem(m);
2344 return (error);
2345 }
2346 if (len == m->m_len)
2347 m = m_free(m);
2348 else {
2349 m->m_data += len;
2350 m->m_len -= len;
2351 }
2352 }
2353 if (m != NULL)
2354 flags |= MSG_TRUNC;
2355 m_freem(m);
2356 if (flagsp != NULL)
2357 *flagsp |= flags;
2358 return (0);
2359 }
2360
2361 int
2362 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2363 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2364 {
2365 int error;
2366
2367 CURVNET_SET(so->so_vnet);
2368 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2369 controlp, flagsp));
2370 CURVNET_RESTORE();
2371 return (error);
2372 }
2373
2374 int
2375 soshutdown(struct socket *so, int how)
2376 {
2377 struct protosw *pr = so->so_proto;
2378 int error;
2379
2380 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2381 return (EINVAL);
2382
2383 CURVNET_SET(so->so_vnet);
2384 if (pr->pr_usrreqs->pru_flush != NULL) {
2385 (*pr->pr_usrreqs->pru_flush)(so, how);
2386 }
2387 if (how != SHUT_WR)
2388 sorflush(so);
2389 if (how != SHUT_RD) {
2390 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2391 wakeup(&so->so_timeo);
2392 CURVNET_RESTORE();
2393 return (error);
2394 }
2395 wakeup(&so->so_timeo);
2396 CURVNET_RESTORE();
2397 return (0);
2398 }
2399
2400 void
2401 sorflush(struct socket *so)
2402 {
2403 struct sockbuf *sb = &so->so_rcv;
2404 struct protosw *pr = so->so_proto;
2405 struct sockbuf asb;
2406
2407 VNET_SO_ASSERT(so);
2408
2409 /*
2410 * In order to avoid calling dom_dispose with the socket buffer mutex
2411 * held, and in order to generally avoid holding the lock for a long
2412 * time, we make a copy of the socket buffer and clear the original
2413 * (except locks, state). The new socket buffer copy won't have
2414 * initialized locks so we can only call routines that won't use or
2415 * assert those locks.
2416 *
2417 * Dislodge threads currently blocked in receive and wait to acquire
2418 * a lock against other simultaneous readers before clearing the
2419 * socket buffer. Don't let our acquire be interrupted by a signal
2420 * despite any existing socket disposition on interruptable waiting.
2421 */
2422 socantrcvmore(so);
2423 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2424
2425 /*
2426 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2427 * and mutex data unchanged.
2428 */
2429 SOCKBUF_LOCK(sb);
2430 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2431 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2432 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2433 bzero(&sb->sb_startzero,
2434 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2435 SOCKBUF_UNLOCK(sb);
2436 sbunlock(sb);
2437
2438 /*
2439 * Dispose of special rights and flush the socket buffer. Don't call
2440 * any unsafe routines (that rely on locks being initialized) on asb.
2441 */
2442 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2443 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2444 sbrelease_internal(&asb, so);
2445 }
2446
2447 /*
2448 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2449 * additional variant to handle the case where the option value needs to be
2450 * some kind of integer, but not a specific size. In addition to their use
2451 * here, these functions are also called by the protocol-level pr_ctloutput()
2452 * routines.
2453 */
2454 int
2455 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2456 {
2457 size_t valsize;
2458
2459 /*
2460 * If the user gives us more than we wanted, we ignore it, but if we
2461 * don't get the minimum length the caller wants, we return EINVAL.
2462 * On success, sopt->sopt_valsize is set to however much we actually
2463 * retrieved.
2464 */
2465 if ((valsize = sopt->sopt_valsize) < minlen)
2466 return EINVAL;
2467 if (valsize > len)
2468 sopt->sopt_valsize = valsize = len;
2469
2470 if (sopt->sopt_td != NULL)
2471 return (copyin(sopt->sopt_val, buf, valsize));
2472
2473 bcopy(sopt->sopt_val, buf, valsize);
2474 return (0);
2475 }
2476
2477 /*
2478 * Kernel version of setsockopt(2).
2479 *
2480 * XXX: optlen is size_t, not socklen_t
2481 */
2482 int
2483 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2484 size_t optlen)
2485 {
2486 struct sockopt sopt;
2487
2488 sopt.sopt_level = level;
2489 sopt.sopt_name = optname;
2490 sopt.sopt_dir = SOPT_SET;
2491 sopt.sopt_val = optval;
2492 sopt.sopt_valsize = optlen;
2493 sopt.sopt_td = NULL;
2494 return (sosetopt(so, &sopt));
2495 }
2496
2497 int
2498 sosetopt(struct socket *so, struct sockopt *sopt)
2499 {
2500 int error, optval;
2501 struct linger l;
2502 struct timeval tv;
2503 u_long val;
2504 uint32_t val32;
2505 #ifdef MAC
2506 struct mac extmac;
2507 #endif
2508
2509 CURVNET_SET(so->so_vnet);
2510 error = 0;
2511 if (sopt->sopt_level != SOL_SOCKET) {
2512 if (so->so_proto->pr_ctloutput != NULL) {
2513 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2514 CURVNET_RESTORE();
2515 return (error);
2516 }
2517 error = ENOPROTOOPT;
2518 } else {
2519 switch (sopt->sopt_name) {
2520 #ifdef INET
2521 case SO_ACCEPTFILTER:
2522 error = do_setopt_accept_filter(so, sopt);
2523 if (error)
2524 goto bad;
2525 break;
2526 #endif
2527 case SO_LINGER:
2528 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2529 if (error)
2530 goto bad;
2531
2532 SOCK_LOCK(so);
2533 so->so_linger = l.l_linger;
2534 if (l.l_onoff)
2535 so->so_options |= SO_LINGER;
2536 else
2537 so->so_options &= ~SO_LINGER;
2538 SOCK_UNLOCK(so);
2539 break;
2540
2541 case SO_DEBUG:
2542 case SO_KEEPALIVE:
2543 case SO_DONTROUTE:
2544 case SO_USELOOPBACK:
2545 case SO_BROADCAST:
2546 case SO_REUSEADDR:
2547 case SO_REUSEPORT:
2548 case SO_OOBINLINE:
2549 case SO_TIMESTAMP:
2550 case SO_BINTIME:
2551 case SO_NOSIGPIPE:
2552 case SO_NO_DDP:
2553 case SO_NO_OFFLOAD:
2554 error = sooptcopyin(sopt, &optval, sizeof optval,
2555 sizeof optval);
2556 if (error)
2557 goto bad;
2558 SOCK_LOCK(so);
2559 if (optval)
2560 so->so_options |= sopt->sopt_name;
2561 else
2562 so->so_options &= ~sopt->sopt_name;
2563 SOCK_UNLOCK(so);
2564 break;
2565
2566 case SO_SETFIB:
2567 error = sooptcopyin(sopt, &optval, sizeof optval,
2568 sizeof optval);
2569 if (error)
2570 goto bad;
2571
2572 if (optval < 0 || optval >= rt_numfibs) {
2573 error = EINVAL;
2574 goto bad;
2575 }
2576 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2577 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2578 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2579 so->so_fibnum = optval;
2580 else
2581 so->so_fibnum = 0;
2582 break;
2583
2584 case SO_USER_COOKIE:
2585 error = sooptcopyin(sopt, &val32, sizeof val32,
2586 sizeof val32);
2587 if (error)
2588 goto bad;
2589 so->so_user_cookie = val32;
2590 break;
2591
2592 case SO_SNDBUF:
2593 case SO_RCVBUF:
2594 case SO_SNDLOWAT:
2595 case SO_RCVLOWAT:
2596 error = sooptcopyin(sopt, &optval, sizeof optval,
2597 sizeof optval);
2598 if (error)
2599 goto bad;
2600
2601 /*
2602 * Values < 1 make no sense for any of these options,
2603 * so disallow them.
2604 */
2605 if (optval < 1) {
2606 error = EINVAL;
2607 goto bad;
2608 }
2609
2610 switch (sopt->sopt_name) {
2611 case SO_SNDBUF:
2612 case SO_RCVBUF:
2613 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2614 &so->so_snd : &so->so_rcv, (u_long)optval,
2615 so, curthread) == 0) {
2616 error = ENOBUFS;
2617 goto bad;
2618 }
2619 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2620 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2621 break;
2622
2623 /*
2624 * Make sure the low-water is never greater than the
2625 * high-water.
2626 */
2627 case SO_SNDLOWAT:
2628 SOCKBUF_LOCK(&so->so_snd);
2629 so->so_snd.sb_lowat =
2630 (optval > so->so_snd.sb_hiwat) ?
2631 so->so_snd.sb_hiwat : optval;
2632 SOCKBUF_UNLOCK(&so->so_snd);
2633 break;
2634 case SO_RCVLOWAT:
2635 SOCKBUF_LOCK(&so->so_rcv);
2636 so->so_rcv.sb_lowat =
2637 (optval > so->so_rcv.sb_hiwat) ?
2638 so->so_rcv.sb_hiwat : optval;
2639 SOCKBUF_UNLOCK(&so->so_rcv);
2640 break;
2641 }
2642 break;
2643
2644 case SO_SNDTIMEO:
2645 case SO_RCVTIMEO:
2646 #ifdef COMPAT_FREEBSD32
2647 if (SV_CURPROC_FLAG(SV_ILP32)) {
2648 struct timeval32 tv32;
2649
2650 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2651 sizeof tv32);
2652 CP(tv32, tv, tv_sec);
2653 CP(tv32, tv, tv_usec);
2654 } else
2655 #endif
2656 error = sooptcopyin(sopt, &tv, sizeof tv,
2657 sizeof tv);
2658 if (error)
2659 goto bad;
2660
2661 /* assert(hz > 0); */
2662 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2663 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2664 error = EDOM;
2665 goto bad;
2666 }
2667 /* assert(tick > 0); */
2668 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2669 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2670 if (val > INT_MAX) {
2671 error = EDOM;
2672 goto bad;
2673 }
2674 if (val == 0 && tv.tv_usec != 0)
2675 val = 1;
2676
2677 switch (sopt->sopt_name) {
2678 case SO_SNDTIMEO:
2679 so->so_snd.sb_timeo = val;
2680 break;
2681 case SO_RCVTIMEO:
2682 so->so_rcv.sb_timeo = val;
2683 break;
2684 }
2685 break;
2686
2687 case SO_LABEL:
2688 #ifdef MAC
2689 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2690 sizeof extmac);
2691 if (error)
2692 goto bad;
2693 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2694 so, &extmac);
2695 #else
2696 error = EOPNOTSUPP;
2697 #endif
2698 break;
2699
2700 default:
2701 error = ENOPROTOOPT;
2702 break;
2703 }
2704 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2705 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2706 }
2707 bad:
2708 CURVNET_RESTORE();
2709 return (error);
2710 }
2711
2712 /*
2713 * Helper routine for getsockopt.
2714 */
2715 int
2716 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2717 {
2718 int error;
2719 size_t valsize;
2720
2721 error = 0;
2722
2723 /*
2724 * Documented get behavior is that we always return a value, possibly
2725 * truncated to fit in the user's buffer. Traditional behavior is
2726 * that we always tell the user precisely how much we copied, rather
2727 * than something useful like the total amount we had available for
2728 * her. Note that this interface is not idempotent; the entire
2729 * answer must generated ahead of time.
2730 */
2731 valsize = min(len, sopt->sopt_valsize);
2732 sopt->sopt_valsize = valsize;
2733 if (sopt->sopt_val != NULL) {
2734 if (sopt->sopt_td != NULL)
2735 error = copyout(buf, sopt->sopt_val, valsize);
2736 else
2737 bcopy(buf, sopt->sopt_val, valsize);
2738 }
2739 return (error);
2740 }
2741
2742 int
2743 sogetopt(struct socket *so, struct sockopt *sopt)
2744 {
2745 int error, optval;
2746 struct linger l;
2747 struct timeval tv;
2748 #ifdef MAC
2749 struct mac extmac;
2750 #endif
2751
2752 CURVNET_SET(so->so_vnet);
2753 error = 0;
2754 if (sopt->sopt_level != SOL_SOCKET) {
2755 if (so->so_proto->pr_ctloutput != NULL)
2756 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2757 else
2758 error = ENOPROTOOPT;
2759 CURVNET_RESTORE();
2760 return (error);
2761 } else {
2762 switch (sopt->sopt_name) {
2763 #ifdef INET
2764 case SO_ACCEPTFILTER:
2765 error = do_getopt_accept_filter(so, sopt);
2766 break;
2767 #endif
2768 case SO_LINGER:
2769 SOCK_LOCK(so);
2770 l.l_onoff = so->so_options & SO_LINGER;
2771 l.l_linger = so->so_linger;
2772 SOCK_UNLOCK(so);
2773 error = sooptcopyout(sopt, &l, sizeof l);
2774 break;
2775
2776 case SO_USELOOPBACK:
2777 case SO_DONTROUTE:
2778 case SO_DEBUG:
2779 case SO_KEEPALIVE:
2780 case SO_REUSEADDR:
2781 case SO_REUSEPORT:
2782 case SO_BROADCAST:
2783 case SO_OOBINLINE:
2784 case SO_ACCEPTCONN:
2785 case SO_TIMESTAMP:
2786 case SO_BINTIME:
2787 case SO_NOSIGPIPE:
2788 optval = so->so_options & sopt->sopt_name;
2789 integer:
2790 error = sooptcopyout(sopt, &optval, sizeof optval);
2791 break;
2792
2793 case SO_TYPE:
2794 optval = so->so_type;
2795 goto integer;
2796
2797 case SO_PROTOCOL:
2798 optval = so->so_proto->pr_protocol;
2799 goto integer;
2800
2801 case SO_ERROR:
2802 SOCK_LOCK(so);
2803 optval = so->so_error;
2804 so->so_error = 0;
2805 SOCK_UNLOCK(so);
2806 goto integer;
2807
2808 case SO_SNDBUF:
2809 optval = so->so_snd.sb_hiwat;
2810 goto integer;
2811
2812 case SO_RCVBUF:
2813 optval = so->so_rcv.sb_hiwat;
2814 goto integer;
2815
2816 case SO_SNDLOWAT:
2817 optval = so->so_snd.sb_lowat;
2818 goto integer;
2819
2820 case SO_RCVLOWAT:
2821 optval = so->so_rcv.sb_lowat;
2822 goto integer;
2823
2824 case SO_SNDTIMEO:
2825 case SO_RCVTIMEO:
2826 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2827 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2828
2829 tv.tv_sec = optval / hz;
2830 tv.tv_usec = (optval % hz) * tick;
2831 #ifdef COMPAT_FREEBSD32
2832 if (SV_CURPROC_FLAG(SV_ILP32)) {
2833 struct timeval32 tv32;
2834
2835 CP(tv, tv32, tv_sec);
2836 CP(tv, tv32, tv_usec);
2837 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2838 } else
2839 #endif
2840 error = sooptcopyout(sopt, &tv, sizeof tv);
2841 break;
2842
2843 case SO_LABEL:
2844 #ifdef MAC
2845 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2846 sizeof(extmac));
2847 if (error)
2848 goto bad;
2849 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2850 so, &extmac);
2851 if (error)
2852 goto bad;
2853 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2854 #else
2855 error = EOPNOTSUPP;
2856 #endif
2857 break;
2858
2859 case SO_PEERLABEL:
2860 #ifdef MAC
2861 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2862 sizeof(extmac));
2863 if (error)
2864 goto bad;
2865 error = mac_getsockopt_peerlabel(
2866 sopt->sopt_td->td_ucred, so, &extmac);
2867 if (error)
2868 goto bad;
2869 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2870 #else
2871 error = EOPNOTSUPP;
2872 #endif
2873 break;
2874
2875 case SO_LISTENQLIMIT:
2876 optval = so->so_qlimit;
2877 goto integer;
2878
2879 case SO_LISTENQLEN:
2880 optval = so->so_qlen;
2881 goto integer;
2882
2883 case SO_LISTENINCQLEN:
2884 optval = so->so_incqlen;
2885 goto integer;
2886
2887 default:
2888 error = ENOPROTOOPT;
2889 break;
2890 }
2891 }
2892 #ifdef MAC
2893 bad:
2894 #endif
2895 CURVNET_RESTORE();
2896 return (error);
2897 }
2898
2899 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2900 int
2901 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2902 {
2903 struct mbuf *m, *m_prev;
2904 int sopt_size = sopt->sopt_valsize;
2905
2906 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2907 if (m == NULL)
2908 return ENOBUFS;
2909 if (sopt_size > MLEN) {
2910 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2911 if ((m->m_flags & M_EXT) == 0) {
2912 m_free(m);
2913 return ENOBUFS;
2914 }
2915 m->m_len = min(MCLBYTES, sopt_size);
2916 } else {
2917 m->m_len = min(MLEN, sopt_size);
2918 }
2919 sopt_size -= m->m_len;
2920 *mp = m;
2921 m_prev = m;
2922
2923 while (sopt_size) {
2924 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2925 if (m == NULL) {
2926 m_freem(*mp);
2927 return ENOBUFS;
2928 }
2929 if (sopt_size > MLEN) {
2930 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2931 M_DONTWAIT);
2932 if ((m->m_flags & M_EXT) == 0) {
2933 m_freem(m);
2934 m_freem(*mp);
2935 return ENOBUFS;
2936 }
2937 m->m_len = min(MCLBYTES, sopt_size);
2938 } else {
2939 m->m_len = min(MLEN, sopt_size);
2940 }
2941 sopt_size -= m->m_len;
2942 m_prev->m_next = m;
2943 m_prev = m;
2944 }
2945 return (0);
2946 }
2947
2948 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2949 int
2950 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2951 {
2952 struct mbuf *m0 = m;
2953
2954 if (sopt->sopt_val == NULL)
2955 return (0);
2956 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2957 if (sopt->sopt_td != NULL) {
2958 int error;
2959
2960 error = copyin(sopt->sopt_val, mtod(m, char *),
2961 m->m_len);
2962 if (error != 0) {
2963 m_freem(m0);
2964 return(error);
2965 }
2966 } else
2967 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2968 sopt->sopt_valsize -= m->m_len;
2969 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2970 m = m->m_next;
2971 }
2972 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2973 panic("ip6_sooptmcopyin");
2974 return (0);
2975 }
2976
2977 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2978 int
2979 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2980 {
2981 struct mbuf *m0 = m;
2982 size_t valsize = 0;
2983
2984 if (sopt->sopt_val == NULL)
2985 return (0);
2986 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2987 if (sopt->sopt_td != NULL) {
2988 int error;
2989
2990 error = copyout(mtod(m, char *), sopt->sopt_val,
2991 m->m_len);
2992 if (error != 0) {
2993 m_freem(m0);
2994 return(error);
2995 }
2996 } else
2997 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2998 sopt->sopt_valsize -= m->m_len;
2999 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3000 valsize += m->m_len;
3001 m = m->m_next;
3002 }
3003 if (m != NULL) {
3004 /* enough soopt buffer should be given from user-land */
3005 m_freem(m0);
3006 return(EINVAL);
3007 }
3008 sopt->sopt_valsize = valsize;
3009 return (0);
3010 }
3011
3012 /*
3013 * sohasoutofband(): protocol notifies socket layer of the arrival of new
3014 * out-of-band data, which will then notify socket consumers.
3015 */
3016 void
3017 sohasoutofband(struct socket *so)
3018 {
3019
3020 if (so->so_sigio != NULL)
3021 pgsigio(&so->so_sigio, SIGURG, 0);
3022 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
3023 }
3024
3025 int
3026 sopoll(struct socket *so, int events, struct ucred *active_cred,
3027 struct thread *td)
3028 {
3029
3030 /*
3031 * We do not need to set or assert curvnet as long as everyone uses
3032 * sopoll_generic().
3033 */
3034 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3035 td));
3036 }
3037
3038 int
3039 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3040 struct thread *td)
3041 {
3042 int revents = 0;
3043
3044 SOCKBUF_LOCK(&so->so_snd);
3045 SOCKBUF_LOCK(&so->so_rcv);
3046 if (events & (POLLIN | POLLRDNORM))
3047 if (soreadabledata(so))
3048 revents |= events & (POLLIN | POLLRDNORM);
3049
3050 if (events & (POLLOUT | POLLWRNORM))
3051 if (sowriteable(so))
3052 revents |= events & (POLLOUT | POLLWRNORM);
3053
3054 if (events & (POLLPRI | POLLRDBAND))
3055 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3056 revents |= events & (POLLPRI | POLLRDBAND);
3057
3058 if ((events & POLLINIGNEOF) == 0) {
3059 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3060 revents |= events & (POLLIN | POLLRDNORM);
3061 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3062 revents |= POLLHUP;
3063 }
3064 }
3065
3066 if (revents == 0) {
3067 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3068 selrecord(td, &so->so_rcv.sb_sel);
3069 so->so_rcv.sb_flags |= SB_SEL;
3070 }
3071
3072 if (events & (POLLOUT | POLLWRNORM)) {
3073 selrecord(td, &so->so_snd.sb_sel);
3074 so->so_snd.sb_flags |= SB_SEL;
3075 }
3076 }
3077
3078 SOCKBUF_UNLOCK(&so->so_rcv);
3079 SOCKBUF_UNLOCK(&so->so_snd);
3080 return (revents);
3081 }
3082
3083 int
3084 soo_kqfilter(struct file *fp, struct knote *kn)
3085 {
3086 struct socket *so = kn->kn_fp->f_data;
3087 struct sockbuf *sb;
3088
3089 switch (kn->kn_filter) {
3090 case EVFILT_READ:
3091 if (so->so_options & SO_ACCEPTCONN)
3092 kn->kn_fop = &solisten_filtops;
3093 else
3094 kn->kn_fop = &soread_filtops;
3095 sb = &so->so_rcv;
3096 break;
3097 case EVFILT_WRITE:
3098 kn->kn_fop = &sowrite_filtops;
3099 sb = &so->so_snd;
3100 break;
3101 default:
3102 return (EINVAL);
3103 }
3104
3105 SOCKBUF_LOCK(sb);
3106 knlist_add(&sb->sb_sel.si_note, kn, 1);
3107 sb->sb_flags |= SB_KNOTE;
3108 SOCKBUF_UNLOCK(sb);
3109 return (0);
3110 }
3111
3112 /*
3113 * Some routines that return EOPNOTSUPP for entry points that are not
3114 * supported by a protocol. Fill in as needed.
3115 */
3116 int
3117 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3118 {
3119
3120 return EOPNOTSUPP;
3121 }
3122
3123 int
3124 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3125 {
3126
3127 return EOPNOTSUPP;
3128 }
3129
3130 int
3131 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3132 {
3133
3134 return EOPNOTSUPP;
3135 }
3136
3137 int
3138 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3139 {
3140
3141 return EOPNOTSUPP;
3142 }
3143
3144 int
3145 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3146 {
3147
3148 return EOPNOTSUPP;
3149 }
3150
3151 int
3152 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3153 struct ifnet *ifp, struct thread *td)
3154 {
3155
3156 return EOPNOTSUPP;
3157 }
3158
3159 int
3160 pru_disconnect_notsupp(struct socket *so)
3161 {
3162
3163 return EOPNOTSUPP;
3164 }
3165
3166 int
3167 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3168 {
3169
3170 return EOPNOTSUPP;
3171 }
3172
3173 int
3174 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3175 {
3176
3177 return EOPNOTSUPP;
3178 }
3179
3180 int
3181 pru_rcvd_notsupp(struct socket *so, int flags)
3182 {
3183
3184 return EOPNOTSUPP;
3185 }
3186
3187 int
3188 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3189 {
3190
3191 return EOPNOTSUPP;
3192 }
3193
3194 int
3195 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3196 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3197 {
3198
3199 return EOPNOTSUPP;
3200 }
3201
3202 /*
3203 * This isn't really a ``null'' operation, but it's the default one and
3204 * doesn't do anything destructive.
3205 */
3206 int
3207 pru_sense_null(struct socket *so, struct stat *sb)
3208 {
3209
3210 sb->st_blksize = so->so_snd.sb_hiwat;
3211 return 0;
3212 }
3213
3214 int
3215 pru_shutdown_notsupp(struct socket *so)
3216 {
3217
3218 return EOPNOTSUPP;
3219 }
3220
3221 int
3222 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3223 {
3224
3225 return EOPNOTSUPP;
3226 }
3227
3228 int
3229 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3230 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3231 {
3232
3233 return EOPNOTSUPP;
3234 }
3235
3236 int
3237 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3238 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3239 {
3240
3241 return EOPNOTSUPP;
3242 }
3243
3244 int
3245 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3246 struct thread *td)
3247 {
3248
3249 return EOPNOTSUPP;
3250 }
3251
3252 static void
3253 filt_sordetach(struct knote *kn)
3254 {
3255 struct socket *so = kn->kn_fp->f_data;
3256
3257 SOCKBUF_LOCK(&so->so_rcv);
3258 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3259 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3260 so->so_rcv.sb_flags &= ~SB_KNOTE;
3261 SOCKBUF_UNLOCK(&so->so_rcv);
3262 }
3263
3264 /*ARGSUSED*/
3265 static int
3266 filt_soread(struct knote *kn, long hint)
3267 {
3268 struct socket *so;
3269
3270 so = kn->kn_fp->f_data;
3271 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3272
3273 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3274 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3275 kn->kn_flags |= EV_EOF;
3276 kn->kn_fflags = so->so_error;
3277 return (1);
3278 } else if (so->so_error) /* temporary udp error */
3279 return (1);
3280 else if (kn->kn_sfflags & NOTE_LOWAT)
3281 return (kn->kn_data >= kn->kn_sdata);
3282 else
3283 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3284 }
3285
3286 static void
3287 filt_sowdetach(struct knote *kn)
3288 {
3289 struct socket *so = kn->kn_fp->f_data;
3290
3291 SOCKBUF_LOCK(&so->so_snd);
3292 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3293 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3294 so->so_snd.sb_flags &= ~SB_KNOTE;
3295 SOCKBUF_UNLOCK(&so->so_snd);
3296 }
3297
3298 /*ARGSUSED*/
3299 static int
3300 filt_sowrite(struct knote *kn, long hint)
3301 {
3302 struct socket *so;
3303
3304 so = kn->kn_fp->f_data;
3305 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3306 kn->kn_data = sbspace(&so->so_snd);
3307 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3308 kn->kn_flags |= EV_EOF;
3309 kn->kn_fflags = so->so_error;
3310 return (1);
3311 } else if (so->so_error) /* temporary udp error */
3312 return (1);
3313 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3314 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3315 return (0);
3316 else if (kn->kn_sfflags & NOTE_LOWAT)
3317 return (kn->kn_data >= kn->kn_sdata);
3318 else
3319 return (kn->kn_data >= so->so_snd.sb_lowat);
3320 }
3321
3322 /*ARGSUSED*/
3323 static int
3324 filt_solisten(struct knote *kn, long hint)
3325 {
3326 struct socket *so = kn->kn_fp->f_data;
3327
3328 kn->kn_data = so->so_qlen;
3329 return (! TAILQ_EMPTY(&so->so_comp));
3330 }
3331
3332 int
3333 socheckuid(struct socket *so, uid_t uid)
3334 {
3335
3336 if (so == NULL)
3337 return (EPERM);
3338 if (so->so_cred->cr_uid != uid)
3339 return (EPERM);
3340 return (0);
3341 }
3342
3343 /*
3344 * These functions are used by protocols to notify the socket layer (and its
3345 * consumers) of state changes in the sockets driven by protocol-side events.
3346 */
3347
3348 /*
3349 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3350 *
3351 * Normal sequence from the active (originating) side is that
3352 * soisconnecting() is called during processing of connect() call, resulting
3353 * in an eventual call to soisconnected() if/when the connection is
3354 * established. When the connection is torn down soisdisconnecting() is
3355 * called during processing of disconnect() call, and soisdisconnected() is
3356 * called when the connection to the peer is totally severed. The semantics
3357 * of these routines are such that connectionless protocols can call
3358 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3359 * calls when setting up a ``connection'' takes no time.
3360 *
3361 * From the passive side, a socket is created with two queues of sockets:
3362 * so_incomp for connections in progress and so_comp for connections already
3363 * made and awaiting user acceptance. As a protocol is preparing incoming
3364 * connections, it creates a socket structure queued on so_incomp by calling
3365 * sonewconn(). When the connection is established, soisconnected() is
3366 * called, and transfers the socket structure to so_comp, making it available
3367 * to accept().
3368 *
3369 * If a socket is closed with sockets on either so_incomp or so_comp, these
3370 * sockets are dropped.
3371 *
3372 * If higher-level protocols are implemented in the kernel, the wakeups done
3373 * here will sometimes cause software-interrupt process scheduling.
3374 */
3375 void
3376 soisconnecting(struct socket *so)
3377 {
3378
3379 SOCK_LOCK(so);
3380 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3381 so->so_state |= SS_ISCONNECTING;
3382 SOCK_UNLOCK(so);
3383 }
3384
3385 void
3386 soisconnected(struct socket *so)
3387 {
3388 struct socket *head;
3389 int ret;
3390
3391 restart:
3392 ACCEPT_LOCK();
3393 SOCK_LOCK(so);
3394 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3395 so->so_state |= SS_ISCONNECTED;
3396 head = so->so_head;
3397 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3398 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3399 SOCK_UNLOCK(so);
3400 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3401 head->so_incqlen--;
3402 so->so_qstate &= ~SQ_INCOMP;
3403 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3404 head->so_qlen++;
3405 so->so_qstate |= SQ_COMP;
3406 ACCEPT_UNLOCK();
3407 sorwakeup(head);
3408 wakeup_one(&head->so_timeo);
3409 } else {
3410 ACCEPT_UNLOCK();
3411 soupcall_set(so, SO_RCV,
3412 head->so_accf->so_accept_filter->accf_callback,
3413 head->so_accf->so_accept_filter_arg);
3414 so->so_options &= ~SO_ACCEPTFILTER;
3415 ret = head->so_accf->so_accept_filter->accf_callback(so,
3416 head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3417 if (ret == SU_ISCONNECTED)
3418 soupcall_clear(so, SO_RCV);
3419 SOCK_UNLOCK(so);
3420 if (ret == SU_ISCONNECTED)
3421 goto restart;
3422 }
3423 return;
3424 }
3425 SOCK_UNLOCK(so);
3426 ACCEPT_UNLOCK();
3427 wakeup(&so->so_timeo);
3428 sorwakeup(so);
3429 sowwakeup(so);
3430 }
3431
3432 void
3433 soisdisconnecting(struct socket *so)
3434 {
3435
3436 /*
3437 * Note: This code assumes that SOCK_LOCK(so) and
3438 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3439 */
3440 SOCKBUF_LOCK(&so->so_rcv);
3441 so->so_state &= ~SS_ISCONNECTING;
3442 so->so_state |= SS_ISDISCONNECTING;
3443 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3444 sorwakeup_locked(so);
3445 SOCKBUF_LOCK(&so->so_snd);
3446 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3447 sowwakeup_locked(so);
3448 wakeup(&so->so_timeo);
3449 }
3450
3451 void
3452 soisdisconnected(struct socket *so)
3453 {
3454
3455 /*
3456 * Note: This code assumes that SOCK_LOCK(so) and
3457 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3458 */
3459 SOCKBUF_LOCK(&so->so_rcv);
3460 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3461 so->so_state |= SS_ISDISCONNECTED;
3462 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3463 sorwakeup_locked(so);
3464 SOCKBUF_LOCK(&so->so_snd);
3465 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3466 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3467 sowwakeup_locked(so);
3468 wakeup(&so->so_timeo);
3469 }
3470
3471 /*
3472 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3473 */
3474 struct sockaddr *
3475 sodupsockaddr(const struct sockaddr *sa, int mflags)
3476 {
3477 struct sockaddr *sa2;
3478
3479 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3480 if (sa2)
3481 bcopy(sa, sa2, sa->sa_len);
3482 return sa2;
3483 }
3484
3485 /*
3486 * Register per-socket buffer upcalls.
3487 */
3488 void
3489 soupcall_set(struct socket *so, int which,
3490 int (*func)(struct socket *, void *, int), void *arg)
3491 {
3492 struct sockbuf *sb;
3493
3494 switch (which) {
3495 case SO_RCV:
3496 sb = &so->so_rcv;
3497 break;
3498 case SO_SND:
3499 sb = &so->so_snd;
3500 break;
3501 default:
3502 panic("soupcall_set: bad which");
3503 }
3504 SOCKBUF_LOCK_ASSERT(sb);
3505 #if 0
3506 /* XXX: accf_http actually wants to do this on purpose. */
3507 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3508 #endif
3509 sb->sb_upcall = func;
3510 sb->sb_upcallarg = arg;
3511 sb->sb_flags |= SB_UPCALL;
3512 }
3513
3514 void
3515 soupcall_clear(struct socket *so, int which)
3516 {
3517 struct sockbuf *sb;
3518
3519 switch (which) {
3520 case SO_RCV:
3521 sb = &so->so_rcv;
3522 break;
3523 case SO_SND:
3524 sb = &so->so_snd;
3525 break;
3526 default:
3527 panic("soupcall_clear: bad which");
3528 }
3529 SOCKBUF_LOCK_ASSERT(sb);
3530 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3531 sb->sb_upcall = NULL;
3532 sb->sb_upcallarg = NULL;
3533 sb->sb_flags &= ~SB_UPCALL;
3534 }
3535
3536 /*
3537 * Create an external-format (``xsocket'') structure using the information in
3538 * the kernel-format socket structure pointed to by so. This is done to
3539 * reduce the spew of irrelevant information over this interface, to isolate
3540 * user code from changes in the kernel structure, and potentially to provide
3541 * information-hiding if we decide that some of this information should be
3542 * hidden from users.
3543 */
3544 void
3545 sotoxsocket(struct socket *so, struct xsocket *xso)
3546 {
3547
3548 xso->xso_len = sizeof *xso;
3549 xso->xso_so = so;
3550 xso->so_type = so->so_type;
3551 xso->so_options = so->so_options;
3552 xso->so_linger = so->so_linger;
3553 xso->so_state = so->so_state;
3554 xso->so_pcb = so->so_pcb;
3555 xso->xso_protocol = so->so_proto->pr_protocol;
3556 xso->xso_family = so->so_proto->pr_domain->dom_family;
3557 xso->so_qlen = so->so_qlen;
3558 xso->so_incqlen = so->so_incqlen;
3559 xso->so_qlimit = so->so_qlimit;
3560 xso->so_timeo = so->so_timeo;
3561 xso->so_error = so->so_error;
3562 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3563 xso->so_oobmark = so->so_oobmark;
3564 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3565 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3566 xso->so_uid = so->so_cred->cr_uid;
3567 }
3568
3569
3570 /*
3571 * Socket accessor functions to provide external consumers with
3572 * a safe interface to socket state
3573 *
3574 */
3575
3576 void
3577 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3578 {
3579
3580 TAILQ_FOREACH(so, &so->so_comp, so_list)
3581 func(so, arg);
3582 }
3583
3584 struct sockbuf *
3585 so_sockbuf_rcv(struct socket *so)
3586 {
3587
3588 return (&so->so_rcv);
3589 }
3590
3591 struct sockbuf *
3592 so_sockbuf_snd(struct socket *so)
3593 {
3594
3595 return (&so->so_snd);
3596 }
3597
3598 int
3599 so_state_get(const struct socket *so)
3600 {
3601
3602 return (so->so_state);
3603 }
3604
3605 void
3606 so_state_set(struct socket *so, int val)
3607 {
3608
3609 so->so_state = val;
3610 }
3611
3612 int
3613 so_options_get(const struct socket *so)
3614 {
3615
3616 return (so->so_options);
3617 }
3618
3619 void
3620 so_options_set(struct socket *so, int val)
3621 {
3622
3623 so->so_options = val;
3624 }
3625
3626 int
3627 so_error_get(const struct socket *so)
3628 {
3629
3630 return (so->so_error);
3631 }
3632
3633 void
3634 so_error_set(struct socket *so, int val)
3635 {
3636
3637 so->so_error = val;
3638 }
3639
3640 int
3641 so_linger_get(const struct socket *so)
3642 {
3643
3644 return (so->so_linger);
3645 }
3646
3647 void
3648 so_linger_set(struct socket *so, int val)
3649 {
3650
3651 so->so_linger = val;
3652 }
3653
3654 struct protosw *
3655 so_protosw_get(const struct socket *so)
3656 {
3657
3658 return (so->so_proto);
3659 }
3660
3661 void
3662 so_protosw_set(struct socket *so, struct protosw *val)
3663 {
3664
3665 so->so_proto = val;
3666 }
3667
3668 void
3669 so_sorwakeup(struct socket *so)
3670 {
3671
3672 sorwakeup(so);
3673 }
3674
3675 void
3676 so_sowwakeup(struct socket *so)
3677 {
3678
3679 sowwakeup(so);
3680 }
3681
3682 void
3683 so_sorwakeup_locked(struct socket *so)
3684 {
3685
3686 sorwakeup_locked(so);
3687 }
3688
3689 void
3690 so_sowwakeup_locked(struct socket *so)
3691 {
3692
3693 sowwakeup_locked(so);
3694 }
3695
3696 void
3697 so_lock(struct socket *so)
3698 {
3699 SOCK_LOCK(so);
3700 }
3701
3702 void
3703 so_unlock(struct socket *so)
3704 {
3705 SOCK_UNLOCK(so);
3706 }
Cache object: b465798b7f2c9f7b2d9e915c4e41b64f
|