1 /* $NetBSD: uipc_socket2.c,v 1.142 2022/10/26 23:38:09 riastradh Exp $ */
2
3 /*-
4 * Copyright (c) 2008 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
26 * POSSIBILITY OF SUCH DAMAGE.
27 */
28
29 /*
30 * Copyright (c) 1982, 1986, 1988, 1990, 1993
31 * The Regents of the University of California. All rights reserved.
32 *
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
35 * are met:
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
41 * 3. Neither the name of the University nor the names of its contributors
42 * may be used to endorse or promote products derived from this software
43 * without specific prior written permission.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * SUCH DAMAGE.
56 *
57 * @(#)uipc_socket2.c 8.2 (Berkeley) 2/14/95
58 */
59
60 #include <sys/cdefs.h>
61 __KERNEL_RCSID(0, "$NetBSD: uipc_socket2.c,v 1.142 2022/10/26 23:38:09 riastradh Exp $");
62
63 #ifdef _KERNEL_OPT
64 #include "opt_ddb.h"
65 #include "opt_inet.h"
66 #include "opt_mbuftrace.h"
67 #include "opt_sb_max.h"
68 #endif
69
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/proc.h>
73 #include <sys/file.h>
74 #include <sys/buf.h>
75 #include <sys/mbuf.h>
76 #include <sys/protosw.h>
77 #include <sys/domain.h>
78 #include <sys/poll.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/signalvar.h>
82 #include <sys/kauth.h>
83 #include <sys/pool.h>
84 #include <sys/uidinfo.h>
85
86 #ifdef DDB
87 #include <sys/filedesc.h>
88 #include <ddb/db_active.h>
89 #endif
90
91 /*
92 * Primitive routines for operating on sockets and socket buffers.
93 *
94 * Connection life-cycle:
95 *
96 * Normal sequence from the active (originating) side:
97 *
98 * - soisconnecting() is called during processing of connect() call,
99 * - resulting in an eventual call to soisconnected() if/when the
100 * connection is established.
101 *
102 * When the connection is torn down during processing of disconnect():
103 *
104 * - soisdisconnecting() is called and,
105 * - soisdisconnected() is called when the connection to the peer
106 * is totally severed.
107 *
108 * The semantics of these routines are such that connectionless protocols
109 * can call soisconnected() and soisdisconnected() only, bypassing the
110 * in-progress calls when setting up a ``connection'' takes no time.
111 *
112 * From the passive side, a socket is created with two queues of sockets:
113 *
114 * - so_q0 (0) for partial connections (i.e. connections in progress)
115 * - so_q (1) for connections already made and awaiting user acceptance.
116 *
117 * As a protocol is preparing incoming connections, it creates a socket
118 * structure queued on so_q0 by calling sonewconn(). When the connection
119 * is established, soisconnected() is called, and transfers the
120 * socket structure to so_q, making it available to accept().
121 *
122 * If a socket is closed with sockets on either so_q0 or so_q, these
123 * sockets are dropped.
124 *
125 * Locking rules and assumptions:
126 *
127 * o socket::so_lock can change on the fly. The low level routines used
128 * to lock sockets are aware of this. When so_lock is acquired, the
129 * routine locking must check to see if so_lock still points to the
130 * lock that was acquired. If so_lock has changed in the meantime, the
131 * now irrelevant lock that was acquired must be dropped and the lock
132 * operation retried. Although not proven here, this is completely safe
133 * on a multiprocessor system, even with relaxed memory ordering, given
134 * the next two rules:
135 *
136 * o In order to mutate so_lock, the lock pointed to by the current value
137 * of so_lock must be held: i.e., the socket must be held locked by the
138 * changing thread. The thread must issue membar_release() to prevent
139 * memory accesses being reordered, and can set so_lock to the desired
140 * value. If the lock pointed to by the new value of so_lock is not
141 * held by the changing thread, the socket must then be considered
142 * unlocked.
143 *
144 * o If so_lock is mutated, and the previous lock referred to by so_lock
145 * could still be visible to other threads in the system (e.g. via file
146 * descriptor or protocol-internal reference), then the old lock must
147 * remain valid until the socket and/or protocol control block has been
148 * torn down.
149 *
150 * o If a socket has a non-NULL so_head value (i.e. is in the process of
151 * connecting), then locking the socket must also lock the socket pointed
152 * to by so_head: their lock pointers must match.
153 *
154 * o If a socket has connections in progress (so_q, so_q0 not empty) then
155 * locking the socket must also lock the sockets attached to both queues.
156 * Again, their lock pointers must match.
157 *
158 * o Beyond the initial lock assignment in socreate(), assigning locks to
159 * sockets is the responsibility of the individual protocols / protocol
160 * domains.
161 */
162
163 static pool_cache_t socket_cache;
164 u_long sb_max = SB_MAX;/* maximum socket buffer size */
165 static u_long sb_max_adj; /* adjusted sb_max */
166
167 void
soisconnecting(struct socket * so)168 soisconnecting(struct socket *so)
169 {
170
171 KASSERT(solocked(so));
172
173 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
174 so->so_state |= SS_ISCONNECTING;
175 }
176
177 void
soisconnected(struct socket * so)178 soisconnected(struct socket *so)
179 {
180 struct socket *head;
181
182 head = so->so_head;
183
184 KASSERT(solocked(so));
185 KASSERT(head == NULL || solocked2(so, head));
186
187 so->so_state &= ~(SS_ISCONNECTING | SS_ISDISCONNECTING);
188 so->so_state |= SS_ISCONNECTED;
189 if (head && so->so_onq == &head->so_q0) {
190 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
191 /*
192 * Re-enqueue and wake up any waiters, e.g.
193 * processes blocking on accept().
194 */
195 soqremque(so, 0);
196 soqinsque(head, so, 1);
197 sorwakeup(head);
198 cv_broadcast(&head->so_cv);
199 } else {
200 so->so_upcall =
201 head->so_accf->so_accept_filter->accf_callback;
202 so->so_upcallarg = head->so_accf->so_accept_filter_arg;
203 so->so_rcv.sb_flags |= SB_UPCALL;
204 so->so_options &= ~SO_ACCEPTFILTER;
205 (*so->so_upcall)(so, so->so_upcallarg,
206 POLLIN|POLLRDNORM, M_DONTWAIT);
207 }
208 } else {
209 cv_broadcast(&so->so_cv);
210 sorwakeup(so);
211 sowwakeup(so);
212 }
213 }
214
215 void
soisdisconnecting(struct socket * so)216 soisdisconnecting(struct socket *so)
217 {
218
219 KASSERT(solocked(so));
220
221 so->so_state &= ~SS_ISCONNECTING;
222 so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
223 cv_broadcast(&so->so_cv);
224 sowwakeup(so);
225 sorwakeup(so);
226 }
227
228 void
soisdisconnected(struct socket * so)229 soisdisconnected(struct socket *so)
230 {
231
232 KASSERT(solocked(so));
233
234 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
235 so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE|SS_ISDISCONNECTED);
236 cv_broadcast(&so->so_cv);
237 sowwakeup(so);
238 sorwakeup(so);
239 }
240
241 void
soinit2(void)242 soinit2(void)
243 {
244
245 socket_cache = pool_cache_init(sizeof(struct socket), 0, 0, 0,
246 "socket", NULL, IPL_SOFTNET, NULL, NULL, NULL);
247 }
248
249 /*
250 * sonewconn: accept a new connection.
251 *
252 * When an attempt at a new connection is noted on a socket which accepts
253 * connections, sonewconn(9) is called. If the connection is possible
254 * (subject to space constraints, etc) then we allocate a new structure,
255 * properly linked into the data structure of the original socket.
256 *
257 * => If 'soready' is true, then socket will become ready for accept() i.e.
258 * inserted into the so_q queue, SS_ISCONNECTED set and waiters awoken.
259 * => May be called from soft-interrupt context.
260 * => Listening socket should be locked.
261 * => Returns the new socket locked.
262 */
263 struct socket *
sonewconn(struct socket * head,bool soready)264 sonewconn(struct socket *head, bool soready)
265 {
266 struct socket *so;
267 int soqueue, error;
268
269 KASSERT(solocked(head));
270
271 if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2) {
272 /*
273 * Listen queue overflow. If there is an accept filter
274 * active, pass through the oldest cxn it's handling.
275 */
276 if (head->so_accf == NULL) {
277 return NULL;
278 } else {
279 struct socket *so2, *next;
280
281 /* Pass the oldest connection waiting in the
282 accept filter */
283 for (so2 = TAILQ_FIRST(&head->so_q0);
284 so2 != NULL; so2 = next) {
285 next = TAILQ_NEXT(so2, so_qe);
286 if (so2->so_upcall == NULL) {
287 continue;
288 }
289 so2->so_upcall = NULL;
290 so2->so_upcallarg = NULL;
291 so2->so_options &= ~SO_ACCEPTFILTER;
292 so2->so_rcv.sb_flags &= ~SB_UPCALL;
293 soisconnected(so2);
294 break;
295 }
296
297 /* If nothing was nudged out of the acept filter, bail
298 * out; otherwise proceed allocating the socket. */
299 if (so2 == NULL) {
300 return NULL;
301 }
302 }
303 }
304 if ((head->so_options & SO_ACCEPTFILTER) != 0) {
305 soready = false;
306 }
307 soqueue = soready ? 1 : 0;
308
309 if ((so = soget(false)) == NULL) {
310 return NULL;
311 }
312 so->so_type = head->so_type;
313 so->so_options = head->so_options & ~SO_ACCEPTCONN;
314 so->so_linger = head->so_linger;
315 so->so_state = head->so_state | SS_NOFDREF;
316 so->so_proto = head->so_proto;
317 so->so_timeo = head->so_timeo;
318 so->so_pgid = head->so_pgid;
319 so->so_send = head->so_send;
320 so->so_receive = head->so_receive;
321 so->so_uidinfo = head->so_uidinfo;
322 so->so_egid = head->so_egid;
323 so->so_cpid = head->so_cpid;
324
325 /*
326 * Share the lock with the listening-socket, it may get unshared
327 * once the connection is complete.
328 *
329 * so_lock is stable while we hold the socket locked, so no
330 * need for atomic_load_* here.
331 */
332 mutex_obj_hold(head->so_lock);
333 so->so_lock = head->so_lock;
334
335 /*
336 * Reserve the space for socket buffers.
337 */
338 #ifdef MBUFTRACE
339 so->so_mowner = head->so_mowner;
340 so->so_rcv.sb_mowner = head->so_rcv.sb_mowner;
341 so->so_snd.sb_mowner = head->so_snd.sb_mowner;
342 #endif
343 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
344 goto out;
345 }
346 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
347 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
348 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
349 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
350 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & (SB_AUTOSIZE | SB_ASYNC);
351 so->so_snd.sb_flags |= head->so_snd.sb_flags & (SB_AUTOSIZE | SB_ASYNC);
352
353 /*
354 * Finally, perform the protocol attach. Note: a new socket
355 * lock may be assigned at this point (if so, it will be held).
356 */
357 error = (*so->so_proto->pr_usrreqs->pr_attach)(so, 0);
358 if (error) {
359 out:
360 KASSERT(solocked(so));
361 KASSERT(so->so_accf == NULL);
362 soput(so);
363
364 /* Note: the listening socket shall stay locked. */
365 KASSERT(solocked(head));
366 return NULL;
367 }
368 KASSERT(solocked2(head, so));
369
370 /*
371 * Insert into the queue. If ready, update the connection status
372 * and wake up any waiters, e.g. processes blocking on accept().
373 */
374 soqinsque(head, so, soqueue);
375 if (soready) {
376 so->so_state |= SS_ISCONNECTED;
377 sorwakeup(head);
378 cv_broadcast(&head->so_cv);
379 }
380 return so;
381 }
382
383 struct socket *
soget(bool waitok)384 soget(bool waitok)
385 {
386 struct socket *so;
387
388 so = pool_cache_get(socket_cache, (waitok ? PR_WAITOK : PR_NOWAIT));
389 if (__predict_false(so == NULL))
390 return (NULL);
391 memset(so, 0, sizeof(*so));
392 TAILQ_INIT(&so->so_q0);
393 TAILQ_INIT(&so->so_q);
394 cv_init(&so->so_cv, "socket");
395 cv_init(&so->so_rcv.sb_cv, "netio");
396 cv_init(&so->so_snd.sb_cv, "netio");
397 selinit(&so->so_rcv.sb_sel);
398 selinit(&so->so_snd.sb_sel);
399 so->so_rcv.sb_so = so;
400 so->so_snd.sb_so = so;
401 return so;
402 }
403
404 void
soput(struct socket * so)405 soput(struct socket *so)
406 {
407
408 KASSERT(!cv_has_waiters(&so->so_cv));
409 KASSERT(!cv_has_waiters(&so->so_rcv.sb_cv));
410 KASSERT(!cv_has_waiters(&so->so_snd.sb_cv));
411 seldestroy(&so->so_rcv.sb_sel);
412 seldestroy(&so->so_snd.sb_sel);
413 mutex_obj_free(so->so_lock);
414 cv_destroy(&so->so_cv);
415 cv_destroy(&so->so_rcv.sb_cv);
416 cv_destroy(&so->so_snd.sb_cv);
417 pool_cache_put(socket_cache, so);
418 }
419
420 /*
421 * soqinsque: insert socket of a new connection into the specified
422 * accept queue of the listening socket (head).
423 *
424 * q = 0: queue of partial connections
425 * q = 1: queue of incoming connections
426 */
427 void
soqinsque(struct socket * head,struct socket * so,int q)428 soqinsque(struct socket *head, struct socket *so, int q)
429 {
430 KASSERT(q == 0 || q == 1);
431 KASSERT(solocked2(head, so));
432 KASSERT(so->so_onq == NULL);
433 KASSERT(so->so_head == NULL);
434
435 so->so_head = head;
436 if (q == 0) {
437 head->so_q0len++;
438 so->so_onq = &head->so_q0;
439 } else {
440 head->so_qlen++;
441 so->so_onq = &head->so_q;
442 }
443 TAILQ_INSERT_TAIL(so->so_onq, so, so_qe);
444 }
445
446 /*
447 * soqremque: remove socket from the specified queue.
448 *
449 * => Returns true if socket was removed from the specified queue.
450 * => False if socket was not removed (because it was in other queue).
451 */
452 bool
soqremque(struct socket * so,int q)453 soqremque(struct socket *so, int q)
454 {
455 struct socket *head = so->so_head;
456
457 KASSERT(q == 0 || q == 1);
458 KASSERT(solocked(so));
459 KASSERT(so->so_onq != NULL);
460 KASSERT(head != NULL);
461
462 if (q == 0) {
463 if (so->so_onq != &head->so_q0)
464 return false;
465 head->so_q0len--;
466 } else {
467 if (so->so_onq != &head->so_q)
468 return false;
469 head->so_qlen--;
470 }
471 KASSERT(solocked2(so, head));
472 TAILQ_REMOVE(so->so_onq, so, so_qe);
473 so->so_onq = NULL;
474 so->so_head = NULL;
475 return true;
476 }
477
478 /*
479 * socantsendmore: indicates that no more data will be sent on the
480 * socket; it would normally be applied to a socket when the user
481 * informs the system that no more data is to be sent, by the protocol
482 * code (in case pr_shutdown()).
483 */
484 void
socantsendmore(struct socket * so)485 socantsendmore(struct socket *so)
486 {
487 KASSERT(solocked(so));
488
489 so->so_state |= SS_CANTSENDMORE;
490 sowwakeup(so);
491 }
492
493 /*
494 * socantrcvmore(): indicates that no more data will be received and
495 * will normally be applied to the socket by a protocol when it detects
496 * that the peer will send no more data. Data queued for reading in
497 * the socket may yet be read.
498 */
499 void
socantrcvmore(struct socket * so)500 socantrcvmore(struct socket *so)
501 {
502 KASSERT(solocked(so));
503
504 so->so_state |= SS_CANTRCVMORE;
505 sorwakeup(so);
506 }
507
508 /*
509 * soroverflow(): indicates that data was attempted to be sent
510 * but the receiving buffer overflowed.
511 */
512 void
soroverflow(struct socket * so)513 soroverflow(struct socket *so)
514 {
515 KASSERT(solocked(so));
516
517 so->so_rcv.sb_overflowed++;
518 if (so->so_options & SO_RERROR) {
519 so->so_rerror = ENOBUFS;
520 sorwakeup(so);
521 }
522 }
523
524 /*
525 * Wait for data to arrive at/drain from a socket buffer.
526 */
527 int
sbwait(struct sockbuf * sb)528 sbwait(struct sockbuf *sb)
529 {
530 struct socket *so;
531 kmutex_t *lock;
532 int error;
533
534 so = sb->sb_so;
535
536 KASSERT(solocked(so));
537
538 sb->sb_flags |= SB_NOTIFY;
539 lock = so->so_lock;
540 if ((sb->sb_flags & SB_NOINTR) != 0)
541 error = cv_timedwait(&sb->sb_cv, lock, sb->sb_timeo);
542 else
543 error = cv_timedwait_sig(&sb->sb_cv, lock, sb->sb_timeo);
544 if (__predict_false(lock != atomic_load_relaxed(&so->so_lock)))
545 solockretry(so, lock);
546 return error;
547 }
548
549 /*
550 * Wakeup processes waiting on a socket buffer.
551 * Do asynchronous notification via SIGIO
552 * if the socket buffer has the SB_ASYNC flag set.
553 */
554 void
sowakeup(struct socket * so,struct sockbuf * sb,int code)555 sowakeup(struct socket *so, struct sockbuf *sb, int code)
556 {
557 int band;
558
559 KASSERT(solocked(so));
560 KASSERT(sb->sb_so == so);
561
562 switch (code) {
563 case POLL_IN:
564 band = POLLIN|POLLRDNORM;
565 break;
566
567 case POLL_OUT:
568 band = POLLOUT|POLLWRNORM;
569 break;
570
571 case POLL_HUP:
572 band = POLLHUP;
573 break;
574
575 default:
576 band = 0;
577 #ifdef DIAGNOSTIC
578 printf("bad siginfo code %d in socket notification.\n", code);
579 #endif
580 break;
581 }
582
583 sb->sb_flags &= ~SB_NOTIFY;
584 selnotify(&sb->sb_sel, band, NOTE_SUBMIT);
585 cv_broadcast(&sb->sb_cv);
586 if (sb->sb_flags & SB_ASYNC)
587 fownsignal(so->so_pgid, SIGIO, code, band, so);
588 if (sb->sb_flags & SB_UPCALL)
589 (*so->so_upcall)(so, so->so_upcallarg, band, M_DONTWAIT);
590 }
591
592 /*
593 * Reset a socket's lock pointer. Wake all threads waiting on the
594 * socket's condition variables so that they can restart their waits
595 * using the new lock. The existing lock must be held.
596 *
597 * Caller must have issued membar_release before this.
598 */
599 void
solockreset(struct socket * so,kmutex_t * lock)600 solockreset(struct socket *so, kmutex_t *lock)
601 {
602
603 KASSERT(solocked(so));
604
605 so->so_lock = lock;
606 cv_broadcast(&so->so_snd.sb_cv);
607 cv_broadcast(&so->so_rcv.sb_cv);
608 cv_broadcast(&so->so_cv);
609 }
610
611 /*
612 * Socket buffer (struct sockbuf) utility routines.
613 *
614 * Each socket contains two socket buffers: one for sending data and
615 * one for receiving data. Each buffer contains a queue of mbufs,
616 * information about the number of mbufs and amount of data in the
617 * queue, and other fields allowing poll() statements and notification
618 * on data availability to be implemented.
619 *
620 * Data stored in a socket buffer is maintained as a list of records.
621 * Each record is a list of mbufs chained together with the m_next
622 * field. Records are chained together with the m_nextpkt field. The upper
623 * level routine soreceive() expects the following conventions to be
624 * observed when placing information in the receive buffer:
625 *
626 * 1. If the protocol requires each message be preceded by the sender's
627 * name, then a record containing that name must be present before
628 * any associated data (mbuf's must be of type MT_SONAME).
629 * 2. If the protocol supports the exchange of ``access rights'' (really
630 * just additional data associated with the message), and there are
631 * ``rights'' to be received, then a record containing this data
632 * should be present (mbuf's must be of type MT_CONTROL).
633 * 3. If a name or rights record exists, then it must be followed by
634 * a data record, perhaps of zero length.
635 *
636 * Before using a new socket structure it is first necessary to reserve
637 * buffer space to the socket, by calling sbreserve(). This should commit
638 * some of the available buffer space in the system buffer pool for the
639 * socket (currently, it does nothing but enforce limits). The space
640 * should be released by calling sbrelease() when the socket is destroyed.
641 */
642
643 int
sb_max_set(u_long new_sbmax)644 sb_max_set(u_long new_sbmax)
645 {
646 int s;
647
648 if (new_sbmax < (16 * 1024))
649 return (EINVAL);
650
651 s = splsoftnet();
652 sb_max = new_sbmax;
653 sb_max_adj = (u_quad_t)new_sbmax * MCLBYTES / (MSIZE + MCLBYTES);
654 splx(s);
655
656 return (0);
657 }
658
659 int
soreserve(struct socket * so,u_long sndcc,u_long rcvcc)660 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
661 {
662 KASSERT(so->so_pcb == NULL || solocked(so));
663
664 /*
665 * there's at least one application (a configure script of screen)
666 * which expects a fifo is writable even if it has "some" bytes
667 * in its buffer.
668 * so we want to make sure (hiwat - lowat) >= (some bytes).
669 *
670 * PIPE_BUF here is an arbitrary value chosen as (some bytes) above.
671 * we expect it's large enough for such applications.
672 */
673 u_long lowat = MAX(sock_loan_thresh, MCLBYTES);
674 u_long hiwat = lowat + PIPE_BUF;
675
676 if (sndcc < hiwat)
677 sndcc = hiwat;
678 if (sbreserve(&so->so_snd, sndcc, so) == 0)
679 goto bad;
680 if (sbreserve(&so->so_rcv, rcvcc, so) == 0)
681 goto bad2;
682 if (so->so_rcv.sb_lowat == 0)
683 so->so_rcv.sb_lowat = 1;
684 if (so->so_snd.sb_lowat == 0)
685 so->so_snd.sb_lowat = lowat;
686 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
687 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
688 return (0);
689 bad2:
690 sbrelease(&so->so_snd, so);
691 bad:
692 return (ENOBUFS);
693 }
694
695 /*
696 * Allot mbufs to a sockbuf.
697 * Attempt to scale mbmax so that mbcnt doesn't become limiting
698 * if buffering efficiency is near the normal case.
699 */
700 int
sbreserve(struct sockbuf * sb,u_long cc,struct socket * so)701 sbreserve(struct sockbuf *sb, u_long cc, struct socket *so)
702 {
703 struct lwp *l = curlwp; /* XXX */
704 rlim_t maxcc;
705 struct uidinfo *uidinfo;
706
707 KASSERT(so->so_pcb == NULL || solocked(so));
708 KASSERT(sb->sb_so == so);
709 KASSERT(sb_max_adj != 0);
710
711 if (cc == 0 || cc > sb_max_adj)
712 return (0);
713
714 maxcc = l->l_proc->p_rlimit[RLIMIT_SBSIZE].rlim_cur;
715
716 uidinfo = so->so_uidinfo;
717 if (!chgsbsize(uidinfo, &sb->sb_hiwat, cc, maxcc))
718 return 0;
719 sb->sb_mbmax = uimin(cc * 2, sb_max);
720 if (sb->sb_lowat > sb->sb_hiwat)
721 sb->sb_lowat = sb->sb_hiwat;
722
723 return (1);
724 }
725
726 /*
727 * Free mbufs held by a socket, and reserved mbuf space. We do not assert
728 * that the socket is held locked here: see sorflush().
729 */
730 void
sbrelease(struct sockbuf * sb,struct socket * so)731 sbrelease(struct sockbuf *sb, struct socket *so)
732 {
733
734 KASSERT(sb->sb_so == so);
735
736 sbflush(sb);
737 (void)chgsbsize(so->so_uidinfo, &sb->sb_hiwat, 0, RLIM_INFINITY);
738 sb->sb_mbmax = 0;
739 }
740
741 /*
742 * Routines to add and remove
743 * data from an mbuf queue.
744 *
745 * The routines sbappend() or sbappendrecord() are normally called to
746 * append new mbufs to a socket buffer, after checking that adequate
747 * space is available, comparing the function sbspace() with the amount
748 * of data to be added. sbappendrecord() differs from sbappend() in
749 * that data supplied is treated as the beginning of a new record.
750 * To place a sender's address, optional access rights, and data in a
751 * socket receive buffer, sbappendaddr() should be used. To place
752 * access rights and data in a socket receive buffer, sbappendrights()
753 * should be used. In either case, the new data begins a new record.
754 * Note that unlike sbappend() and sbappendrecord(), these routines check
755 * for the caller that there will be enough space to store the data.
756 * Each fails if there is not enough space, or if it cannot find mbufs
757 * to store additional information in.
758 *
759 * Reliable protocols may use the socket send buffer to hold data
760 * awaiting acknowledgement. Data is normally copied from a socket
761 * send buffer in a protocol with m_copym for output to a peer,
762 * and then removing the data from the socket buffer with sbdrop()
763 * or sbdroprecord() when the data is acknowledged by the peer.
764 */
765
766 #ifdef SOCKBUF_DEBUG
767 void
sblastrecordchk(struct sockbuf * sb,const char * where)768 sblastrecordchk(struct sockbuf *sb, const char *where)
769 {
770 struct mbuf *m = sb->sb_mb;
771
772 KASSERT(solocked(sb->sb_so));
773
774 while (m && m->m_nextpkt)
775 m = m->m_nextpkt;
776
777 if (m != sb->sb_lastrecord) {
778 printf("sblastrecordchk: sb_mb %p sb_lastrecord %p last %p\n",
779 sb->sb_mb, sb->sb_lastrecord, m);
780 printf("packet chain:\n");
781 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
782 printf("\t%p\n", m);
783 panic("sblastrecordchk from %s", where);
784 }
785 }
786
787 void
sblastmbufchk(struct sockbuf * sb,const char * where)788 sblastmbufchk(struct sockbuf *sb, const char *where)
789 {
790 struct mbuf *m = sb->sb_mb;
791 struct mbuf *n;
792
793 KASSERT(solocked(sb->sb_so));
794
795 while (m && m->m_nextpkt)
796 m = m->m_nextpkt;
797
798 while (m && m->m_next)
799 m = m->m_next;
800
801 if (m != sb->sb_mbtail) {
802 printf("sblastmbufchk: sb_mb %p sb_mbtail %p last %p\n",
803 sb->sb_mb, sb->sb_mbtail, m);
804 printf("packet tree:\n");
805 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
806 printf("\t");
807 for (n = m; n != NULL; n = n->m_next)
808 printf("%p ", n);
809 printf("\n");
810 }
811 panic("sblastmbufchk from %s", where);
812 }
813 }
814 #endif /* SOCKBUF_DEBUG */
815
816 /*
817 * Link a chain of records onto a socket buffer
818 */
819 #define SBLINKRECORDCHAIN(sb, m0, mlast) \
820 do { \
821 if ((sb)->sb_lastrecord != NULL) \
822 (sb)->sb_lastrecord->m_nextpkt = (m0); \
823 else \
824 (sb)->sb_mb = (m0); \
825 (sb)->sb_lastrecord = (mlast); \
826 } while (/*CONSTCOND*/0)
827
828
829 #define SBLINKRECORD(sb, m0) \
830 SBLINKRECORDCHAIN(sb, m0, m0)
831
832 /*
833 * Append mbuf chain m to the last record in the
834 * socket buffer sb. The additional space associated
835 * the mbuf chain is recorded in sb. Empty mbufs are
836 * discarded and mbufs are compacted where possible.
837 */
838 void
sbappend(struct sockbuf * sb,struct mbuf * m)839 sbappend(struct sockbuf *sb, struct mbuf *m)
840 {
841 struct mbuf *n;
842
843 KASSERT(solocked(sb->sb_so));
844
845 if (m == NULL)
846 return;
847
848 #ifdef MBUFTRACE
849 m_claimm(m, sb->sb_mowner);
850 #endif
851
852 SBLASTRECORDCHK(sb, "sbappend 1");
853
854 if ((n = sb->sb_lastrecord) != NULL) {
855 /*
856 * XXX Would like to simply use sb_mbtail here, but
857 * XXX I need to verify that I won't miss an EOR that
858 * XXX way.
859 */
860 do {
861 if (n->m_flags & M_EOR) {
862 sbappendrecord(sb, m); /* XXXXXX!!!! */
863 return;
864 }
865 } while (n->m_next && (n = n->m_next));
866 } else {
867 /*
868 * If this is the first record in the socket buffer, it's
869 * also the last record.
870 */
871 sb->sb_lastrecord = m;
872 }
873 sbcompress(sb, m, n);
874 SBLASTRECORDCHK(sb, "sbappend 2");
875 }
876
877 /*
878 * This version of sbappend() should only be used when the caller
879 * absolutely knows that there will never be more than one record
880 * in the socket buffer, that is, a stream protocol (such as TCP).
881 */
882 void
sbappendstream(struct sockbuf * sb,struct mbuf * m)883 sbappendstream(struct sockbuf *sb, struct mbuf *m)
884 {
885
886 KASSERT(solocked(sb->sb_so));
887 KDASSERT(m->m_nextpkt == NULL);
888 KASSERT(sb->sb_mb == sb->sb_lastrecord);
889
890 SBLASTMBUFCHK(sb, __func__);
891
892 #ifdef MBUFTRACE
893 m_claimm(m, sb->sb_mowner);
894 #endif
895
896 sbcompress(sb, m, sb->sb_mbtail);
897
898 sb->sb_lastrecord = sb->sb_mb;
899 SBLASTRECORDCHK(sb, __func__);
900 }
901
902 #ifdef SOCKBUF_DEBUG
903 void
sbcheck(struct sockbuf * sb)904 sbcheck(struct sockbuf *sb)
905 {
906 struct mbuf *m, *m2;
907 u_long len, mbcnt;
908
909 KASSERT(solocked(sb->sb_so));
910
911 len = 0;
912 mbcnt = 0;
913 for (m = sb->sb_mb; m; m = m->m_nextpkt) {
914 for (m2 = m; m2 != NULL; m2 = m2->m_next) {
915 len += m2->m_len;
916 mbcnt += MSIZE;
917 if (m2->m_flags & M_EXT)
918 mbcnt += m2->m_ext.ext_size;
919 if (m2->m_nextpkt != NULL)
920 panic("sbcheck nextpkt");
921 }
922 }
923 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
924 printf("cc %lu != %lu || mbcnt %lu != %lu\n", len, sb->sb_cc,
925 mbcnt, sb->sb_mbcnt);
926 panic("sbcheck");
927 }
928 }
929 #endif
930
931 /*
932 * As above, except the mbuf chain
933 * begins a new record.
934 */
935 void
sbappendrecord(struct sockbuf * sb,struct mbuf * m0)936 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
937 {
938 struct mbuf *m;
939
940 KASSERT(solocked(sb->sb_so));
941
942 if (m0 == NULL)
943 return;
944
945 #ifdef MBUFTRACE
946 m_claimm(m0, sb->sb_mowner);
947 #endif
948 /*
949 * Put the first mbuf on the queue.
950 * Note this permits zero length records.
951 */
952 sballoc(sb, m0);
953 SBLASTRECORDCHK(sb, "sbappendrecord 1");
954 SBLINKRECORD(sb, m0);
955 m = m0->m_next;
956 m0->m_next = 0;
957 if (m && (m0->m_flags & M_EOR)) {
958 m0->m_flags &= ~M_EOR;
959 m->m_flags |= M_EOR;
960 }
961 sbcompress(sb, m, m0);
962 SBLASTRECORDCHK(sb, "sbappendrecord 2");
963 }
964
965 /*
966 * As above except that OOB data
967 * is inserted at the beginning of the sockbuf,
968 * but after any other OOB data.
969 */
970 void
sbinsertoob(struct sockbuf * sb,struct mbuf * m0)971 sbinsertoob(struct sockbuf *sb, struct mbuf *m0)
972 {
973 struct mbuf *m, **mp;
974
975 KASSERT(solocked(sb->sb_so));
976
977 if (m0 == NULL)
978 return;
979
980 SBLASTRECORDCHK(sb, "sbinsertoob 1");
981
982 for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) {
983 again:
984 switch (m->m_type) {
985
986 case MT_OOBDATA:
987 continue; /* WANT next train */
988
989 case MT_CONTROL:
990 if ((m = m->m_next) != NULL)
991 goto again; /* inspect THIS train further */
992 }
993 break;
994 }
995 /*
996 * Put the first mbuf on the queue.
997 * Note this permits zero length records.
998 */
999 sballoc(sb, m0);
1000 m0->m_nextpkt = *mp;
1001 if (*mp == NULL) {
1002 /* m0 is actually the new tail */
1003 sb->sb_lastrecord = m0;
1004 }
1005 *mp = m0;
1006 m = m0->m_next;
1007 m0->m_next = 0;
1008 if (m && (m0->m_flags & M_EOR)) {
1009 m0->m_flags &= ~M_EOR;
1010 m->m_flags |= M_EOR;
1011 }
1012 sbcompress(sb, m, m0);
1013 SBLASTRECORDCHK(sb, "sbinsertoob 2");
1014 }
1015
1016 /*
1017 * Append address and data, and optionally, control (ancillary) data
1018 * to the receive queue of a socket. If present,
1019 * m0 must include a packet header with total length.
1020 * Returns 0 if no space in sockbuf or insufficient mbufs.
1021 */
1022 int
sbappendaddr(struct sockbuf * sb,const struct sockaddr * asa,struct mbuf * m0,struct mbuf * control)1023 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa, struct mbuf *m0,
1024 struct mbuf *control)
1025 {
1026 struct mbuf *m, *n, *nlast;
1027 int space, len;
1028
1029 KASSERT(solocked(sb->sb_so));
1030
1031 space = asa->sa_len;
1032
1033 if (m0 != NULL) {
1034 if ((m0->m_flags & M_PKTHDR) == 0)
1035 panic("sbappendaddr");
1036 space += m0->m_pkthdr.len;
1037 #ifdef MBUFTRACE
1038 m_claimm(m0, sb->sb_mowner);
1039 #endif
1040 }
1041 for (n = control; n; n = n->m_next) {
1042 space += n->m_len;
1043 MCLAIM(n, sb->sb_mowner);
1044 if (n->m_next == NULL) /* keep pointer to last control buf */
1045 break;
1046 }
1047 if (space > sbspace(sb))
1048 return (0);
1049 m = m_get(M_DONTWAIT, MT_SONAME);
1050 if (m == NULL)
1051 return (0);
1052 MCLAIM(m, sb->sb_mowner);
1053 /*
1054 * XXX avoid 'comparison always true' warning which isn't easily
1055 * avoided.
1056 */
1057 len = asa->sa_len;
1058 if (len > MLEN) {
1059 MEXTMALLOC(m, asa->sa_len, M_NOWAIT);
1060 if ((m->m_flags & M_EXT) == 0) {
1061 m_free(m);
1062 return (0);
1063 }
1064 }
1065 m->m_len = asa->sa_len;
1066 memcpy(mtod(m, void *), asa, asa->sa_len);
1067 if (n)
1068 n->m_next = m0; /* concatenate data to control */
1069 else
1070 control = m0;
1071 m->m_next = control;
1072
1073 SBLASTRECORDCHK(sb, "sbappendaddr 1");
1074
1075 for (n = m; n->m_next != NULL; n = n->m_next)
1076 sballoc(sb, n);
1077 sballoc(sb, n);
1078 nlast = n;
1079 SBLINKRECORD(sb, m);
1080
1081 sb->sb_mbtail = nlast;
1082 SBLASTMBUFCHK(sb, "sbappendaddr");
1083 SBLASTRECORDCHK(sb, "sbappendaddr 2");
1084
1085 return (1);
1086 }
1087
1088 /*
1089 * Helper for sbappendchainaddr: prepend a struct sockaddr* to
1090 * an mbuf chain.
1091 */
1092 static inline struct mbuf *
m_prepend_sockaddr(struct sockbuf * sb,struct mbuf * m0,const struct sockaddr * asa)1093 m_prepend_sockaddr(struct sockbuf *sb, struct mbuf *m0,
1094 const struct sockaddr *asa)
1095 {
1096 struct mbuf *m;
1097 const int salen = asa->sa_len;
1098
1099 KASSERT(solocked(sb->sb_so));
1100
1101 /* only the first in each chain need be a pkthdr */
1102 m = m_gethdr(M_DONTWAIT, MT_SONAME);
1103 if (m == NULL)
1104 return NULL;
1105 MCLAIM(m, sb->sb_mowner);
1106 #ifdef notyet
1107 if (salen > MHLEN) {
1108 MEXTMALLOC(m, salen, M_NOWAIT);
1109 if ((m->m_flags & M_EXT) == 0) {
1110 m_free(m);
1111 return NULL;
1112 }
1113 }
1114 #else
1115 KASSERT(salen <= MHLEN);
1116 #endif
1117 m->m_len = salen;
1118 memcpy(mtod(m, void *), asa, salen);
1119 m->m_next = m0;
1120 m->m_pkthdr.len = salen + m0->m_pkthdr.len;
1121
1122 return m;
1123 }
1124
1125 int
sbappendaddrchain(struct sockbuf * sb,const struct sockaddr * asa,struct mbuf * m0,int sbprio)1126 sbappendaddrchain(struct sockbuf *sb, const struct sockaddr *asa,
1127 struct mbuf *m0, int sbprio)
1128 {
1129 struct mbuf *m, *n, *n0, *nlast;
1130 int error;
1131
1132 KASSERT(solocked(sb->sb_so));
1133
1134 /*
1135 * XXX sbprio reserved for encoding priority of this* request:
1136 * SB_PRIO_NONE --> honour normal sb limits
1137 * SB_PRIO_ONESHOT_OVERFLOW --> if socket has any space,
1138 * take whole chain. Intended for large requests
1139 * that should be delivered atomically (all, or none).
1140 * SB_PRIO_OVERDRAFT -- allow a small (2*MLEN) overflow
1141 * over normal socket limits, for messages indicating
1142 * buffer overflow in earlier normal/lower-priority messages
1143 * SB_PRIO_BESTEFFORT --> ignore limits entirely.
1144 * Intended for kernel-generated messages only.
1145 * Up to generator to avoid total mbuf resource exhaustion.
1146 */
1147 (void)sbprio;
1148
1149 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
1150 panic("sbappendaddrchain");
1151
1152 #ifdef notyet
1153 space = sbspace(sb);
1154
1155 /*
1156 * Enforce SB_PRIO_* limits as described above.
1157 */
1158 #endif
1159
1160 n0 = NULL;
1161 nlast = NULL;
1162 for (m = m0; m; m = m->m_nextpkt) {
1163 struct mbuf *np;
1164
1165 #ifdef MBUFTRACE
1166 m_claimm(m, sb->sb_mowner);
1167 #endif
1168
1169 /* Prepend sockaddr to this record (m) of input chain m0 */
1170 n = m_prepend_sockaddr(sb, m, asa);
1171 if (n == NULL) {
1172 error = ENOBUFS;
1173 goto bad;
1174 }
1175
1176 /* Append record (asa+m) to end of new chain n0 */
1177 if (n0 == NULL) {
1178 n0 = n;
1179 } else {
1180 nlast->m_nextpkt = n;
1181 }
1182 /* Keep track of last record on new chain */
1183 nlast = n;
1184
1185 for (np = n; np; np = np->m_next)
1186 sballoc(sb, np);
1187 }
1188
1189 SBLASTRECORDCHK(sb, "sbappendaddrchain 1");
1190
1191 /* Drop the entire chain of (asa+m) records onto the socket */
1192 SBLINKRECORDCHAIN(sb, n0, nlast);
1193
1194 SBLASTRECORDCHK(sb, "sbappendaddrchain 2");
1195
1196 for (m = nlast; m->m_next; m = m->m_next)
1197 ;
1198 sb->sb_mbtail = m;
1199 SBLASTMBUFCHK(sb, "sbappendaddrchain");
1200
1201 return (1);
1202
1203 bad:
1204 /*
1205 * On error, free the prepended addreseses. For consistency
1206 * with sbappendaddr(), leave it to our caller to free
1207 * the input record chain passed to us as m0.
1208 */
1209 while ((n = n0) != NULL) {
1210 struct mbuf *np;
1211
1212 /* Undo the sballoc() of this record */
1213 for (np = n; np; np = np->m_next)
1214 sbfree(sb, np);
1215
1216 n0 = n->m_nextpkt; /* iterate at next prepended address */
1217 np = m_free(n); /* free prepended address (not data) */
1218 }
1219 return error;
1220 }
1221
1222
1223 int
sbappendcontrol(struct sockbuf * sb,struct mbuf * m0,struct mbuf * control)1224 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
1225 {
1226 struct mbuf *m, *mlast, *n;
1227 int space;
1228
1229 KASSERT(solocked(sb->sb_so));
1230
1231 space = 0;
1232 if (control == NULL)
1233 panic("sbappendcontrol");
1234 for (m = control; ; m = m->m_next) {
1235 space += m->m_len;
1236 MCLAIM(m, sb->sb_mowner);
1237 if (m->m_next == NULL)
1238 break;
1239 }
1240 n = m; /* save pointer to last control buffer */
1241 for (m = m0; m; m = m->m_next) {
1242 MCLAIM(m, sb->sb_mowner);
1243 space += m->m_len;
1244 }
1245 if (space > sbspace(sb))
1246 return (0);
1247 n->m_next = m0; /* concatenate data to control */
1248
1249 SBLASTRECORDCHK(sb, "sbappendcontrol 1");
1250
1251 for (m = control; m->m_next != NULL; m = m->m_next)
1252 sballoc(sb, m);
1253 sballoc(sb, m);
1254 mlast = m;
1255 SBLINKRECORD(sb, control);
1256
1257 sb->sb_mbtail = mlast;
1258 SBLASTMBUFCHK(sb, "sbappendcontrol");
1259 SBLASTRECORDCHK(sb, "sbappendcontrol 2");
1260
1261 return (1);
1262 }
1263
1264 /*
1265 * Compress mbuf chain m into the socket
1266 * buffer sb following mbuf n. If n
1267 * is null, the buffer is presumed empty.
1268 */
1269 void
sbcompress(struct sockbuf * sb,struct mbuf * m,struct mbuf * n)1270 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1271 {
1272 int eor;
1273 struct mbuf *o;
1274
1275 KASSERT(solocked(sb->sb_so));
1276
1277 eor = 0;
1278 while (m) {
1279 eor |= m->m_flags & M_EOR;
1280 if (m->m_len == 0 &&
1281 (eor == 0 ||
1282 (((o = m->m_next) || (o = n)) &&
1283 o->m_type == m->m_type))) {
1284 if (sb->sb_lastrecord == m)
1285 sb->sb_lastrecord = m->m_next;
1286 m = m_free(m);
1287 continue;
1288 }
1289 if (n && (n->m_flags & M_EOR) == 0 &&
1290 /* M_TRAILINGSPACE() checks buffer writeability */
1291 m->m_len <= MCLBYTES / 4 && /* XXX Don't copy too much */
1292 m->m_len <= M_TRAILINGSPACE(n) &&
1293 n->m_type == m->m_type) {
1294 memcpy(mtod(n, char *) + n->m_len, mtod(m, void *),
1295 (unsigned)m->m_len);
1296 n->m_len += m->m_len;
1297 sb->sb_cc += m->m_len;
1298 m = m_free(m);
1299 continue;
1300 }
1301 if (n)
1302 n->m_next = m;
1303 else
1304 sb->sb_mb = m;
1305 sb->sb_mbtail = m;
1306 sballoc(sb, m);
1307 n = m;
1308 m->m_flags &= ~M_EOR;
1309 m = m->m_next;
1310 n->m_next = 0;
1311 }
1312 if (eor) {
1313 if (n)
1314 n->m_flags |= eor;
1315 else
1316 printf("semi-panic: sbcompress\n");
1317 }
1318 SBLASTMBUFCHK(sb, __func__);
1319 }
1320
1321 /*
1322 * Free all mbufs in a sockbuf.
1323 * Check that all resources are reclaimed.
1324 */
1325 void
sbflush(struct sockbuf * sb)1326 sbflush(struct sockbuf *sb)
1327 {
1328
1329 KASSERT(solocked(sb->sb_so));
1330 KASSERT((sb->sb_flags & SB_LOCK) == 0);
1331
1332 while (sb->sb_mbcnt)
1333 sbdrop(sb, (int)sb->sb_cc);
1334
1335 KASSERT(sb->sb_cc == 0);
1336 KASSERT(sb->sb_mb == NULL);
1337 KASSERT(sb->sb_mbtail == NULL);
1338 KASSERT(sb->sb_lastrecord == NULL);
1339 }
1340
1341 /*
1342 * Drop data from (the front of) a sockbuf.
1343 */
1344 void
sbdrop(struct sockbuf * sb,int len)1345 sbdrop(struct sockbuf *sb, int len)
1346 {
1347 struct mbuf *m, *next;
1348
1349 KASSERT(solocked(sb->sb_so));
1350
1351 next = (m = sb->sb_mb) ? m->m_nextpkt : NULL;
1352 while (len > 0) {
1353 if (m == NULL) {
1354 if (next == NULL)
1355 panic("sbdrop(%p,%d): cc=%lu",
1356 sb, len, sb->sb_cc);
1357 m = next;
1358 next = m->m_nextpkt;
1359 continue;
1360 }
1361 if (m->m_len > len) {
1362 m->m_len -= len;
1363 m->m_data += len;
1364 sb->sb_cc -= len;
1365 break;
1366 }
1367 len -= m->m_len;
1368 sbfree(sb, m);
1369 m = m_free(m);
1370 }
1371 while (m && m->m_len == 0) {
1372 sbfree(sb, m);
1373 m = m_free(m);
1374 }
1375 if (m) {
1376 sb->sb_mb = m;
1377 m->m_nextpkt = next;
1378 } else
1379 sb->sb_mb = next;
1380 /*
1381 * First part is an inline SB_EMPTY_FIXUP(). Second part
1382 * makes sure sb_lastrecord is up-to-date if we dropped
1383 * part of the last record.
1384 */
1385 m = sb->sb_mb;
1386 if (m == NULL) {
1387 sb->sb_mbtail = NULL;
1388 sb->sb_lastrecord = NULL;
1389 } else if (m->m_nextpkt == NULL)
1390 sb->sb_lastrecord = m;
1391 }
1392
1393 /*
1394 * Drop a record off the front of a sockbuf
1395 * and move the next record to the front.
1396 */
1397 void
sbdroprecord(struct sockbuf * sb)1398 sbdroprecord(struct sockbuf *sb)
1399 {
1400 struct mbuf *m, *mn;
1401
1402 KASSERT(solocked(sb->sb_so));
1403
1404 m = sb->sb_mb;
1405 if (m) {
1406 sb->sb_mb = m->m_nextpkt;
1407 do {
1408 sbfree(sb, m);
1409 mn = m_free(m);
1410 } while ((m = mn) != NULL);
1411 }
1412 SB_EMPTY_FIXUP(sb);
1413 }
1414
1415 /*
1416 * Create a "control" mbuf containing the specified data
1417 * with the specified type for presentation on a socket buffer.
1418 */
1419 struct mbuf *
sbcreatecontrol1(void ** p,int size,int type,int level,int flags)1420 sbcreatecontrol1(void **p, int size, int type, int level, int flags)
1421 {
1422 struct cmsghdr *cp;
1423 struct mbuf *m;
1424 int space = CMSG_SPACE(size);
1425
1426 if ((flags & M_DONTWAIT) && space > MCLBYTES) {
1427 printf("%s: message too large %d\n", __func__, space);
1428 return NULL;
1429 }
1430
1431 if ((m = m_get(flags, MT_CONTROL)) == NULL)
1432 return NULL;
1433 if (space > MLEN) {
1434 if (space > MCLBYTES)
1435 MEXTMALLOC(m, space, M_WAITOK);
1436 else
1437 MCLGET(m, flags);
1438 if ((m->m_flags & M_EXT) == 0) {
1439 m_free(m);
1440 return NULL;
1441 }
1442 }
1443 cp = mtod(m, struct cmsghdr *);
1444 *p = CMSG_DATA(cp);
1445 m->m_len = space;
1446 cp->cmsg_len = CMSG_LEN(size);
1447 cp->cmsg_level = level;
1448 cp->cmsg_type = type;
1449
1450 memset(cp + 1, 0, CMSG_LEN(0) - sizeof(*cp));
1451 memset((uint8_t *)*p + size, 0, CMSG_ALIGN(size) - size);
1452
1453 return m;
1454 }
1455
1456 struct mbuf *
sbcreatecontrol(void * p,int size,int type,int level)1457 sbcreatecontrol(void *p, int size, int type, int level)
1458 {
1459 struct mbuf *m;
1460 void *v;
1461
1462 m = sbcreatecontrol1(&v, size, type, level, M_DONTWAIT);
1463 if (m == NULL)
1464 return NULL;
1465 memcpy(v, p, size);
1466 return m;
1467 }
1468
1469 void
solockretry(struct socket * so,kmutex_t * lock)1470 solockretry(struct socket *so, kmutex_t *lock)
1471 {
1472
1473 while (lock != atomic_load_relaxed(&so->so_lock)) {
1474 mutex_exit(lock);
1475 lock = atomic_load_consume(&so->so_lock);
1476 mutex_enter(lock);
1477 }
1478 }
1479
1480 bool
solocked(const struct socket * so)1481 solocked(const struct socket *so)
1482 {
1483
1484 /*
1485 * Used only for diagnostic assertions, so so_lock should be
1486 * stable at this point, hence on need for atomic_load_*.
1487 */
1488 return mutex_owned(so->so_lock);
1489 }
1490
1491 bool
solocked2(const struct socket * so1,const struct socket * so2)1492 solocked2(const struct socket *so1, const struct socket *so2)
1493 {
1494 const kmutex_t *lock;
1495
1496 /*
1497 * Used only for diagnostic assertions, so so_lock should be
1498 * stable at this point, hence on need for atomic_load_*.
1499 */
1500 lock = so1->so_lock;
1501 if (lock != so2->so_lock)
1502 return false;
1503 return mutex_owned(lock);
1504 }
1505
1506 /*
1507 * sosetlock: assign a default lock to a new socket.
1508 */
1509 void
sosetlock(struct socket * so)1510 sosetlock(struct socket *so)
1511 {
1512 if (so->so_lock == NULL) {
1513 kmutex_t *lock = softnet_lock;
1514
1515 so->so_lock = lock;
1516 mutex_obj_hold(lock);
1517 mutex_enter(lock);
1518 }
1519 KASSERT(solocked(so));
1520 }
1521
1522 /*
1523 * Set lock on sockbuf sb; sleep if lock is already held.
1524 * Unless SB_NOINTR is set on sockbuf, sleep is interruptible.
1525 * Returns error without lock if sleep is interrupted.
1526 */
1527 int
sblock(struct sockbuf * sb,int wf)1528 sblock(struct sockbuf *sb, int wf)
1529 {
1530 struct socket *so;
1531 kmutex_t *lock;
1532 int error;
1533
1534 KASSERT(solocked(sb->sb_so));
1535
1536 for (;;) {
1537 if (__predict_true((sb->sb_flags & SB_LOCK) == 0)) {
1538 sb->sb_flags |= SB_LOCK;
1539 return 0;
1540 }
1541 if (wf != M_WAITOK)
1542 return EWOULDBLOCK;
1543 so = sb->sb_so;
1544 lock = so->so_lock;
1545 if ((sb->sb_flags & SB_NOINTR) != 0) {
1546 cv_wait(&so->so_cv, lock);
1547 error = 0;
1548 } else
1549 error = cv_wait_sig(&so->so_cv, lock);
1550 if (__predict_false(lock != atomic_load_relaxed(&so->so_lock)))
1551 solockretry(so, lock);
1552 if (error != 0)
1553 return error;
1554 }
1555 }
1556
1557 void
sbunlock(struct sockbuf * sb)1558 sbunlock(struct sockbuf *sb)
1559 {
1560 struct socket *so;
1561
1562 so = sb->sb_so;
1563
1564 KASSERT(solocked(so));
1565 KASSERT((sb->sb_flags & SB_LOCK) != 0);
1566
1567 sb->sb_flags &= ~SB_LOCK;
1568 cv_broadcast(&so->so_cv);
1569 }
1570
1571 int
sowait(struct socket * so,bool catch_p,int timo)1572 sowait(struct socket *so, bool catch_p, int timo)
1573 {
1574 kmutex_t *lock;
1575 int error;
1576
1577 KASSERT(solocked(so));
1578 KASSERT(catch_p || timo != 0);
1579
1580 lock = so->so_lock;
1581 if (catch_p)
1582 error = cv_timedwait_sig(&so->so_cv, lock, timo);
1583 else
1584 error = cv_timedwait(&so->so_cv, lock, timo);
1585 if (__predict_false(lock != atomic_load_relaxed(&so->so_lock)))
1586 solockretry(so, lock);
1587 return error;
1588 }
1589
1590 #ifdef DDB
1591
1592 /*
1593 * Currently, sofindproc() is used only from DDB. It could be used from others
1594 * by using db_mutex_enter()
1595 */
1596
1597 static inline int
db_mutex_enter(kmutex_t * mtx)1598 db_mutex_enter(kmutex_t *mtx)
1599 {
1600 int rv;
1601
1602 if (!db_active) {
1603 mutex_enter(mtx);
1604 rv = 1;
1605 } else
1606 rv = mutex_tryenter(mtx);
1607
1608 return rv;
1609 }
1610
1611 int
sofindproc(struct socket * so,int all,void (* pr)(const char *,...))1612 sofindproc(struct socket *so, int all, void (*pr)(const char *, ...))
1613 {
1614 proc_t *p;
1615 filedesc_t *fdp;
1616 fdtab_t *dt;
1617 fdfile_t *ff;
1618 file_t *fp = NULL;
1619 int found = 0;
1620 int i, t;
1621
1622 if (so == NULL)
1623 return 0;
1624
1625 t = db_mutex_enter(&proc_lock);
1626 if (!t) {
1627 pr("could not acquire proc_lock mutex\n");
1628 return 0;
1629 }
1630 PROCLIST_FOREACH(p, &allproc) {
1631 if (p->p_stat == SIDL)
1632 continue;
1633 fdp = p->p_fd;
1634 t = db_mutex_enter(&fdp->fd_lock);
1635 if (!t) {
1636 pr("could not acquire fd_lock mutex\n");
1637 continue;
1638 }
1639 dt = atomic_load_consume(&fdp->fd_dt);
1640 for (i = 0; i < dt->dt_nfiles; i++) {
1641 ff = dt->dt_ff[i];
1642 if (ff == NULL)
1643 continue;
1644
1645 fp = atomic_load_consume(&ff->ff_file);
1646 if (fp == NULL)
1647 continue;
1648
1649 t = db_mutex_enter(&fp->f_lock);
1650 if (!t) {
1651 pr("could not acquire f_lock mutex\n");
1652 continue;
1653 }
1654 if ((struct socket *)fp->f_data != so) {
1655 mutex_exit(&fp->f_lock);
1656 continue;
1657 }
1658 found++;
1659 if (pr)
1660 pr("socket %p: owner %s(pid=%d)\n",
1661 so, p->p_comm, p->p_pid);
1662 mutex_exit(&fp->f_lock);
1663 if (all == 0)
1664 break;
1665 }
1666 mutex_exit(&fdp->fd_lock);
1667 if (all == 0 && found != 0)
1668 break;
1669 }
1670 mutex_exit(&proc_lock);
1671
1672 return found;
1673 }
1674
1675 void
socket_print(const char * modif,void (* pr)(const char *,...))1676 socket_print(const char *modif, void (*pr)(const char *, ...))
1677 {
1678 file_t *fp;
1679 struct socket *so;
1680 struct sockbuf *sb_snd, *sb_rcv;
1681 struct mbuf *m_rec, *m;
1682 bool opt_v = false;
1683 bool opt_m = false;
1684 bool opt_a = false;
1685 bool opt_p = false;
1686 int nrecs, nmbufs;
1687 char ch;
1688 const char *family;
1689
1690 while ( (ch = *(modif++)) != '\0') {
1691 switch (ch) {
1692 case 'v':
1693 opt_v = true;
1694 break;
1695 case 'm':
1696 opt_m = true;
1697 break;
1698 case 'a':
1699 opt_a = true;
1700 break;
1701 case 'p':
1702 opt_p = true;
1703 break;
1704 }
1705 }
1706 if (opt_v == false && pr)
1707 (pr)("Ignore empty sockets. use /v to print all.\n");
1708 if (opt_p == true && pr)
1709 (pr)("Don't search owner process.\n");
1710
1711 LIST_FOREACH(fp, &filehead, f_list) {
1712 if (fp->f_type != DTYPE_SOCKET)
1713 continue;
1714 so = (struct socket *)fp->f_data;
1715 if (so == NULL)
1716 continue;
1717
1718 if (so->so_proto->pr_domain->dom_family == AF_INET)
1719 family = "INET";
1720 #ifdef INET6
1721 else if (so->so_proto->pr_domain->dom_family == AF_INET6)
1722 family = "INET6";
1723 #endif
1724 else if (so->so_proto->pr_domain->dom_family == pseudo_AF_KEY)
1725 family = "KEY";
1726 else if (so->so_proto->pr_domain->dom_family == AF_ROUTE)
1727 family = "ROUTE";
1728 else
1729 continue;
1730
1731 sb_snd = &so->so_snd;
1732 sb_rcv = &so->so_rcv;
1733
1734 if (opt_v != true &&
1735 sb_snd->sb_cc == 0 && sb_rcv->sb_cc == 0)
1736 continue;
1737
1738 pr("---SOCKET %p: type %s\n", so, family);
1739 if (opt_p != true)
1740 sofindproc(so, opt_a == true ? 1 : 0, pr);
1741 pr("Send Buffer Bytes: %d [bytes]\n", sb_snd->sb_cc);
1742 pr("Send Buffer mbufs:\n");
1743 m_rec = m = sb_snd->sb_mb;
1744 nrecs = 0;
1745 nmbufs = 0;
1746 while (m_rec) {
1747 nrecs++;
1748 if (opt_m == true)
1749 pr(" mbuf chain %p\n", m_rec);
1750 while (m) {
1751 nmbufs++;
1752 m = m->m_next;
1753 }
1754 m_rec = m = m_rec->m_nextpkt;
1755 }
1756 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs);
1757
1758 pr("Recv Buffer Usage: %d [bytes]\n", sb_rcv->sb_cc);
1759 pr("Recv Buffer mbufs:\n");
1760 m_rec = m = sb_rcv->sb_mb;
1761 nrecs = 0;
1762 nmbufs = 0;
1763 while (m_rec) {
1764 nrecs++;
1765 if (opt_m == true)
1766 pr(" mbuf chain %p\n", m_rec);
1767 while (m) {
1768 nmbufs++;
1769 m = m->m_next;
1770 }
1771 m_rec = m = m_rec->m_nextpkt;
1772 }
1773 pr(" Total %d records, %d mbufs.\n", nrecs, nmbufs);
1774 }
1775 }
1776 #endif /* DDB */
1777