uipc_socket2.c revision 1.20 1 1.20 thorpej /* $NetBSD: uipc_socket2.c,v 1.20 1997/06/26 05:56:38 thorpej Exp $ */
2 1.9 cgd
3 1.1 cgd /*
4 1.7 mycroft * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 1.7 mycroft * The Regents of the University of California. All rights reserved.
6 1.1 cgd *
7 1.1 cgd * Redistribution and use in source and binary forms, with or without
8 1.1 cgd * modification, are permitted provided that the following conditions
9 1.1 cgd * are met:
10 1.1 cgd * 1. Redistributions of source code must retain the above copyright
11 1.1 cgd * notice, this list of conditions and the following disclaimer.
12 1.1 cgd * 2. Redistributions in binary form must reproduce the above copyright
13 1.1 cgd * notice, this list of conditions and the following disclaimer in the
14 1.1 cgd * documentation and/or other materials provided with the distribution.
15 1.1 cgd * 3. All advertising materials mentioning features or use of this software
16 1.1 cgd * must display the following acknowledgement:
17 1.1 cgd * This product includes software developed by the University of
18 1.1 cgd * California, Berkeley and its contributors.
19 1.1 cgd * 4. Neither the name of the University nor the names of its contributors
20 1.1 cgd * may be used to endorse or promote products derived from this software
21 1.1 cgd * without specific prior written permission.
22 1.1 cgd *
23 1.1 cgd * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 1.1 cgd * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 1.1 cgd * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 1.1 cgd * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 1.1 cgd * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 1.1 cgd * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 1.1 cgd * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 1.1 cgd * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 1.1 cgd * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 1.1 cgd * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 1.1 cgd * SUCH DAMAGE.
34 1.1 cgd *
35 1.9 cgd * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
36 1.1 cgd */
37 1.1 cgd
38 1.5 mycroft #include <sys/param.h>
39 1.5 mycroft #include <sys/systm.h>
40 1.5 mycroft #include <sys/proc.h>
41 1.5 mycroft #include <sys/file.h>
42 1.5 mycroft #include <sys/buf.h>
43 1.5 mycroft #include <sys/malloc.h>
44 1.5 mycroft #include <sys/mbuf.h>
45 1.5 mycroft #include <sys/protosw.h>
46 1.5 mycroft #include <sys/socket.h>
47 1.5 mycroft #include <sys/socketvar.h>
48 1.11 christos #include <sys/signalvar.h>
49 1.1 cgd
50 1.1 cgd /*
51 1.1 cgd * Primitive routines for operating on sockets and socket buffers
52 1.1 cgd */
53 1.1 cgd
54 1.1 cgd /* strings for sleep message: */
55 1.1 cgd char netio[] = "netio";
56 1.1 cgd char netcon[] = "netcon";
57 1.1 cgd char netcls[] = "netcls";
58 1.1 cgd
59 1.1 cgd u_long sb_max = SB_MAX; /* patchable */
60 1.1 cgd
61 1.1 cgd /*
62 1.1 cgd * Procedures to manipulate state flags of socket
63 1.1 cgd * and do appropriate wakeups. Normal sequence from the
64 1.1 cgd * active (originating) side is that soisconnecting() is
65 1.1 cgd * called during processing of connect() call,
66 1.1 cgd * resulting in an eventual call to soisconnected() if/when the
67 1.1 cgd * connection is established. When the connection is torn down
68 1.1 cgd * soisdisconnecting() is called during processing of disconnect() call,
69 1.1 cgd * and soisdisconnected() is called when the connection to the peer
70 1.1 cgd * is totally severed. The semantics of these routines are such that
71 1.1 cgd * connectionless protocols can call soisconnected() and soisdisconnected()
72 1.1 cgd * only, bypassing the in-progress calls when setting up a ``connection''
73 1.1 cgd * takes no time.
74 1.1 cgd *
75 1.1 cgd * From the passive side, a socket is created with
76 1.1 cgd * two queues of sockets: so_q0 for connections in progress
77 1.1 cgd * and so_q for connections already made and awaiting user acceptance.
78 1.1 cgd * As a protocol is preparing incoming connections, it creates a socket
79 1.1 cgd * structure queued on so_q0 by calling sonewconn(). When the connection
80 1.1 cgd * is established, soisconnected() is called, and transfers the
81 1.1 cgd * socket structure to so_q, making it available to accept().
82 1.1 cgd *
83 1.1 cgd * If a socket is closed with sockets on either
84 1.1 cgd * so_q0 or so_q, these sockets are dropped.
85 1.1 cgd *
86 1.1 cgd * If higher level protocols are implemented in
87 1.1 cgd * the kernel, the wakeups done here will sometimes
88 1.1 cgd * cause software-interrupt process scheduling.
89 1.1 cgd */
90 1.1 cgd
91 1.7 mycroft void
92 1.1 cgd soisconnecting(so)
93 1.1 cgd register struct socket *so;
94 1.1 cgd {
95 1.1 cgd
96 1.1 cgd so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
97 1.1 cgd so->so_state |= SS_ISCONNECTING;
98 1.1 cgd }
99 1.1 cgd
100 1.7 mycroft void
101 1.1 cgd soisconnected(so)
102 1.1 cgd register struct socket *so;
103 1.1 cgd {
104 1.1 cgd register struct socket *head = so->so_head;
105 1.1 cgd
106 1.1 cgd so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
107 1.1 cgd so->so_state |= SS_ISCONNECTED;
108 1.1 cgd if (head && soqremque(so, 0)) {
109 1.1 cgd soqinsque(head, so, 1);
110 1.1 cgd sorwakeup(head);
111 1.1 cgd wakeup((caddr_t)&head->so_timeo);
112 1.1 cgd } else {
113 1.1 cgd wakeup((caddr_t)&so->so_timeo);
114 1.1 cgd sorwakeup(so);
115 1.1 cgd sowwakeup(so);
116 1.1 cgd }
117 1.1 cgd }
118 1.1 cgd
119 1.7 mycroft void
120 1.1 cgd soisdisconnecting(so)
121 1.1 cgd register struct socket *so;
122 1.1 cgd {
123 1.1 cgd
124 1.1 cgd so->so_state &= ~SS_ISCONNECTING;
125 1.1 cgd so->so_state |= (SS_ISDISCONNECTING|SS_CANTRCVMORE|SS_CANTSENDMORE);
126 1.1 cgd wakeup((caddr_t)&so->so_timeo);
127 1.1 cgd sowwakeup(so);
128 1.1 cgd sorwakeup(so);
129 1.1 cgd }
130 1.1 cgd
131 1.7 mycroft void
132 1.1 cgd soisdisconnected(so)
133 1.1 cgd register struct socket *so;
134 1.1 cgd {
135 1.1 cgd
136 1.1 cgd so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
137 1.1 cgd so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
138 1.1 cgd wakeup((caddr_t)&so->so_timeo);
139 1.1 cgd sowwakeup(so);
140 1.1 cgd sorwakeup(so);
141 1.1 cgd }
142 1.1 cgd
143 1.1 cgd /*
144 1.1 cgd * When an attempt at a new connection is noted on a socket
145 1.1 cgd * which accepts connections, sonewconn is called. If the
146 1.1 cgd * connection is possible (subject to space constraints, etc.)
147 1.1 cgd * then we allocate a new structure, propoerly linked into the
148 1.1 cgd * data structure of the original socket, and return this.
149 1.1 cgd * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
150 1.1 cgd *
151 1.1 cgd * Currently, sonewconn() is defined as sonewconn1() in socketvar.h
152 1.1 cgd * to catch calls that are missing the (new) second parameter.
153 1.1 cgd */
154 1.1 cgd struct socket *
155 1.1 cgd sonewconn1(head, connstatus)
156 1.1 cgd register struct socket *head;
157 1.1 cgd int connstatus;
158 1.1 cgd {
159 1.1 cgd register struct socket *so;
160 1.1 cgd int soqueue = connstatus ? 1 : 0;
161 1.1 cgd
162 1.1 cgd if (head->so_qlen + head->so_q0len > 3 * head->so_qlimit / 2)
163 1.1 cgd return ((struct socket *)0);
164 1.1 cgd MALLOC(so, struct socket *, sizeof(*so), M_SOCKET, M_DONTWAIT);
165 1.1 cgd if (so == NULL)
166 1.1 cgd return ((struct socket *)0);
167 1.1 cgd bzero((caddr_t)so, sizeof(*so));
168 1.1 cgd so->so_type = head->so_type;
169 1.1 cgd so->so_options = head->so_options &~ SO_ACCEPTCONN;
170 1.1 cgd so->so_linger = head->so_linger;
171 1.1 cgd so->so_state = head->so_state | SS_NOFDREF;
172 1.1 cgd so->so_proto = head->so_proto;
173 1.1 cgd so->so_timeo = head->so_timeo;
174 1.1 cgd so->so_pgid = head->so_pgid;
175 1.1 cgd (void) soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat);
176 1.1 cgd soqinsque(head, so, soqueue);
177 1.1 cgd if ((*so->so_proto->pr_usrreq)(so, PRU_ATTACH,
178 1.12 mycroft (struct mbuf *)0, (struct mbuf *)0, (struct mbuf *)0,
179 1.12 mycroft (struct proc *)0)) {
180 1.1 cgd (void) soqremque(so, soqueue);
181 1.1 cgd (void) free((caddr_t)so, M_SOCKET);
182 1.1 cgd return ((struct socket *)0);
183 1.1 cgd }
184 1.1 cgd if (connstatus) {
185 1.1 cgd sorwakeup(head);
186 1.1 cgd wakeup((caddr_t)&head->so_timeo);
187 1.1 cgd so->so_state |= connstatus;
188 1.1 cgd }
189 1.1 cgd return (so);
190 1.1 cgd }
191 1.1 cgd
192 1.7 mycroft void
193 1.1 cgd soqinsque(head, so, q)
194 1.1 cgd register struct socket *head, *so;
195 1.1 cgd int q;
196 1.1 cgd {
197 1.1 cgd
198 1.17 thorpej register struct socket **prev;
199 1.1 cgd so->so_head = head;
200 1.1 cgd if (q == 0) {
201 1.1 cgd head->so_q0len++;
202 1.17 thorpej so->so_q0 = 0;
203 1.17 thorpej for (prev = &(head->so_q0); *prev; )
204 1.17 thorpej prev = &((*prev)->so_q0);
205 1.1 cgd } else {
206 1.1 cgd head->so_qlen++;
207 1.17 thorpej so->so_q = 0;
208 1.17 thorpej for (prev = &(head->so_q); *prev; )
209 1.17 thorpej prev = &((*prev)->so_q);
210 1.1 cgd }
211 1.17 thorpej *prev = so;
212 1.1 cgd }
213 1.1 cgd
214 1.7 mycroft int
215 1.1 cgd soqremque(so, q)
216 1.1 cgd register struct socket *so;
217 1.1 cgd int q;
218 1.1 cgd {
219 1.17 thorpej register struct socket *head, *prev, *next;
220 1.1 cgd
221 1.17 thorpej head = so->so_head;
222 1.17 thorpej prev = head;
223 1.17 thorpej for (;;) {
224 1.17 thorpej next = q ? prev->so_q : prev->so_q0;
225 1.17 thorpej if (next == so)
226 1.17 thorpej break;
227 1.17 thorpej if (next == 0)
228 1.17 thorpej return (0);
229 1.17 thorpej prev = next;
230 1.1 cgd }
231 1.1 cgd if (q == 0) {
232 1.17 thorpej prev->so_q0 = next->so_q0;
233 1.1 cgd head->so_q0len--;
234 1.1 cgd } else {
235 1.17 thorpej prev->so_q = next->so_q;
236 1.1 cgd head->so_qlen--;
237 1.1 cgd }
238 1.17 thorpej next->so_q0 = next->so_q = 0;
239 1.17 thorpej next->so_head = 0;
240 1.1 cgd return (1);
241 1.1 cgd }
242 1.1 cgd
243 1.1 cgd /*
244 1.1 cgd * Socantsendmore indicates that no more data will be sent on the
245 1.1 cgd * socket; it would normally be applied to a socket when the user
246 1.1 cgd * informs the system that no more data is to be sent, by the protocol
247 1.1 cgd * code (in case PRU_SHUTDOWN). Socantrcvmore indicates that no more data
248 1.1 cgd * will be received, and will normally be applied to the socket by a
249 1.1 cgd * protocol when it detects that the peer will send no more data.
250 1.1 cgd * Data queued for reading in the socket may yet be read.
251 1.1 cgd */
252 1.1 cgd
253 1.4 andrew void
254 1.1 cgd socantsendmore(so)
255 1.1 cgd struct socket *so;
256 1.1 cgd {
257 1.1 cgd
258 1.1 cgd so->so_state |= SS_CANTSENDMORE;
259 1.1 cgd sowwakeup(so);
260 1.1 cgd }
261 1.1 cgd
262 1.4 andrew void
263 1.1 cgd socantrcvmore(so)
264 1.1 cgd struct socket *so;
265 1.1 cgd {
266 1.1 cgd
267 1.1 cgd so->so_state |= SS_CANTRCVMORE;
268 1.1 cgd sorwakeup(so);
269 1.1 cgd }
270 1.1 cgd
271 1.1 cgd /*
272 1.1 cgd * Wait for data to arrive at/drain from a socket buffer.
273 1.1 cgd */
274 1.7 mycroft int
275 1.1 cgd sbwait(sb)
276 1.1 cgd struct sockbuf *sb;
277 1.1 cgd {
278 1.1 cgd
279 1.1 cgd sb->sb_flags |= SB_WAIT;
280 1.1 cgd return (tsleep((caddr_t)&sb->sb_cc,
281 1.1 cgd (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, netio,
282 1.1 cgd sb->sb_timeo));
283 1.1 cgd }
284 1.1 cgd
285 1.1 cgd /*
286 1.1 cgd * Lock a sockbuf already known to be locked;
287 1.1 cgd * return any error returned from sleep (EINTR).
288 1.1 cgd */
289 1.7 mycroft int
290 1.1 cgd sb_lock(sb)
291 1.1 cgd register struct sockbuf *sb;
292 1.1 cgd {
293 1.1 cgd int error;
294 1.1 cgd
295 1.1 cgd while (sb->sb_flags & SB_LOCK) {
296 1.1 cgd sb->sb_flags |= SB_WANT;
297 1.11 christos error = tsleep((caddr_t)&sb->sb_flags,
298 1.11 christos (sb->sb_flags & SB_NOINTR) ?
299 1.11 christos PSOCK : PSOCK|PCATCH, netio, 0);
300 1.11 christos if (error)
301 1.1 cgd return (error);
302 1.1 cgd }
303 1.1 cgd sb->sb_flags |= SB_LOCK;
304 1.1 cgd return (0);
305 1.1 cgd }
306 1.1 cgd
307 1.1 cgd /*
308 1.1 cgd * Wakeup processes waiting on a socket buffer.
309 1.1 cgd * Do asynchronous notification via SIGIO
310 1.1 cgd * if the socket has the SS_ASYNC flag set.
311 1.1 cgd */
312 1.7 mycroft void
313 1.1 cgd sowakeup(so, sb)
314 1.1 cgd register struct socket *so;
315 1.1 cgd register struct sockbuf *sb;
316 1.1 cgd {
317 1.1 cgd struct proc *p;
318 1.1 cgd
319 1.3 cgd selwakeup(&sb->sb_sel);
320 1.7 mycroft sb->sb_flags &= ~SB_SEL;
321 1.1 cgd if (sb->sb_flags & SB_WAIT) {
322 1.1 cgd sb->sb_flags &= ~SB_WAIT;
323 1.1 cgd wakeup((caddr_t)&sb->sb_cc);
324 1.1 cgd }
325 1.1 cgd if (so->so_state & SS_ASYNC) {
326 1.1 cgd if (so->so_pgid < 0)
327 1.1 cgd gsignal(-so->so_pgid, SIGIO);
328 1.1 cgd else if (so->so_pgid > 0 && (p = pfind(so->so_pgid)) != 0)
329 1.1 cgd psignal(p, SIGIO);
330 1.1 cgd }
331 1.1 cgd }
332 1.1 cgd
333 1.1 cgd /*
334 1.1 cgd * Socket buffer (struct sockbuf) utility routines.
335 1.1 cgd *
336 1.1 cgd * Each socket contains two socket buffers: one for sending data and
337 1.1 cgd * one for receiving data. Each buffer contains a queue of mbufs,
338 1.1 cgd * information about the number of mbufs and amount of data in the
339 1.13 mycroft * queue, and other fields allowing poll() statements and notification
340 1.1 cgd * on data availability to be implemented.
341 1.1 cgd *
342 1.1 cgd * Data stored in a socket buffer is maintained as a list of records.
343 1.1 cgd * Each record is a list of mbufs chained together with the m_next
344 1.1 cgd * field. Records are chained together with the m_nextpkt field. The upper
345 1.1 cgd * level routine soreceive() expects the following conventions to be
346 1.1 cgd * observed when placing information in the receive buffer:
347 1.1 cgd *
348 1.1 cgd * 1. If the protocol requires each message be preceded by the sender's
349 1.1 cgd * name, then a record containing that name must be present before
350 1.1 cgd * any associated data (mbuf's must be of type MT_SONAME).
351 1.1 cgd * 2. If the protocol supports the exchange of ``access rights'' (really
352 1.1 cgd * just additional data associated with the message), and there are
353 1.1 cgd * ``rights'' to be received, then a record containing this data
354 1.10 mycroft * should be present (mbuf's must be of type MT_CONTROL).
355 1.1 cgd * 3. If a name or rights record exists, then it must be followed by
356 1.1 cgd * a data record, perhaps of zero length.
357 1.1 cgd *
358 1.1 cgd * Before using a new socket structure it is first necessary to reserve
359 1.1 cgd * buffer space to the socket, by calling sbreserve(). This should commit
360 1.1 cgd * some of the available buffer space in the system buffer pool for the
361 1.1 cgd * socket (currently, it does nothing but enforce limits). The space
362 1.1 cgd * should be released by calling sbrelease() when the socket is destroyed.
363 1.1 cgd */
364 1.1 cgd
365 1.7 mycroft int
366 1.1 cgd soreserve(so, sndcc, rcvcc)
367 1.1 cgd register struct socket *so;
368 1.1 cgd u_long sndcc, rcvcc;
369 1.1 cgd {
370 1.1 cgd
371 1.1 cgd if (sbreserve(&so->so_snd, sndcc) == 0)
372 1.1 cgd goto bad;
373 1.1 cgd if (sbreserve(&so->so_rcv, rcvcc) == 0)
374 1.1 cgd goto bad2;
375 1.1 cgd if (so->so_rcv.sb_lowat == 0)
376 1.1 cgd so->so_rcv.sb_lowat = 1;
377 1.1 cgd if (so->so_snd.sb_lowat == 0)
378 1.1 cgd so->so_snd.sb_lowat = MCLBYTES;
379 1.1 cgd if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
380 1.1 cgd so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
381 1.1 cgd return (0);
382 1.1 cgd bad2:
383 1.1 cgd sbrelease(&so->so_snd);
384 1.1 cgd bad:
385 1.1 cgd return (ENOBUFS);
386 1.1 cgd }
387 1.1 cgd
388 1.1 cgd /*
389 1.1 cgd * Allot mbufs to a sockbuf.
390 1.1 cgd * Attempt to scale mbmax so that mbcnt doesn't become limiting
391 1.1 cgd * if buffering efficiency is near the normal case.
392 1.1 cgd */
393 1.7 mycroft int
394 1.1 cgd sbreserve(sb, cc)
395 1.1 cgd struct sockbuf *sb;
396 1.1 cgd u_long cc;
397 1.1 cgd {
398 1.1 cgd
399 1.18 thorpej if (cc == 0 || cc > sb_max * MCLBYTES / (MSIZE + MCLBYTES))
400 1.1 cgd return (0);
401 1.1 cgd sb->sb_hiwat = cc;
402 1.1 cgd sb->sb_mbmax = min(cc * 2, sb_max);
403 1.1 cgd if (sb->sb_lowat > sb->sb_hiwat)
404 1.1 cgd sb->sb_lowat = sb->sb_hiwat;
405 1.1 cgd return (1);
406 1.1 cgd }
407 1.1 cgd
408 1.1 cgd /*
409 1.1 cgd * Free mbufs held by a socket, and reserved mbuf space.
410 1.1 cgd */
411 1.7 mycroft void
412 1.1 cgd sbrelease(sb)
413 1.1 cgd struct sockbuf *sb;
414 1.1 cgd {
415 1.1 cgd
416 1.1 cgd sbflush(sb);
417 1.1 cgd sb->sb_hiwat = sb->sb_mbmax = 0;
418 1.1 cgd }
419 1.1 cgd
420 1.1 cgd /*
421 1.1 cgd * Routines to add and remove
422 1.1 cgd * data from an mbuf queue.
423 1.1 cgd *
424 1.1 cgd * The routines sbappend() or sbappendrecord() are normally called to
425 1.1 cgd * append new mbufs to a socket buffer, after checking that adequate
426 1.1 cgd * space is available, comparing the function sbspace() with the amount
427 1.1 cgd * of data to be added. sbappendrecord() differs from sbappend() in
428 1.1 cgd * that data supplied is treated as the beginning of a new record.
429 1.1 cgd * To place a sender's address, optional access rights, and data in a
430 1.1 cgd * socket receive buffer, sbappendaddr() should be used. To place
431 1.1 cgd * access rights and data in a socket receive buffer, sbappendrights()
432 1.1 cgd * should be used. In either case, the new data begins a new record.
433 1.1 cgd * Note that unlike sbappend() and sbappendrecord(), these routines check
434 1.1 cgd * for the caller that there will be enough space to store the data.
435 1.1 cgd * Each fails if there is not enough space, or if it cannot find mbufs
436 1.1 cgd * to store additional information in.
437 1.1 cgd *
438 1.1 cgd * Reliable protocols may use the socket send buffer to hold data
439 1.1 cgd * awaiting acknowledgement. Data is normally copied from a socket
440 1.1 cgd * send buffer in a protocol with m_copy for output to a peer,
441 1.1 cgd * and then removing the data from the socket buffer with sbdrop()
442 1.1 cgd * or sbdroprecord() when the data is acknowledged by the peer.
443 1.1 cgd */
444 1.1 cgd
445 1.1 cgd /*
446 1.1 cgd * Append mbuf chain m to the last record in the
447 1.1 cgd * socket buffer sb. The additional space associated
448 1.1 cgd * the mbuf chain is recorded in sb. Empty mbufs are
449 1.1 cgd * discarded and mbufs are compacted where possible.
450 1.1 cgd */
451 1.7 mycroft void
452 1.1 cgd sbappend(sb, m)
453 1.1 cgd struct sockbuf *sb;
454 1.1 cgd struct mbuf *m;
455 1.1 cgd {
456 1.1 cgd register struct mbuf *n;
457 1.1 cgd
458 1.1 cgd if (m == 0)
459 1.1 cgd return;
460 1.11 christos if ((n = sb->sb_mb) != NULL) {
461 1.1 cgd while (n->m_nextpkt)
462 1.1 cgd n = n->m_nextpkt;
463 1.1 cgd do {
464 1.1 cgd if (n->m_flags & M_EOR) {
465 1.1 cgd sbappendrecord(sb, m); /* XXXXXX!!!! */
466 1.1 cgd return;
467 1.1 cgd }
468 1.1 cgd } while (n->m_next && (n = n->m_next));
469 1.1 cgd }
470 1.1 cgd sbcompress(sb, m, n);
471 1.1 cgd }
472 1.1 cgd
473 1.1 cgd #ifdef SOCKBUF_DEBUG
474 1.7 mycroft void
475 1.1 cgd sbcheck(sb)
476 1.1 cgd register struct sockbuf *sb;
477 1.1 cgd {
478 1.1 cgd register struct mbuf *m;
479 1.1 cgd register int len = 0, mbcnt = 0;
480 1.1 cgd
481 1.1 cgd for (m = sb->sb_mb; m; m = m->m_next) {
482 1.1 cgd len += m->m_len;
483 1.1 cgd mbcnt += MSIZE;
484 1.1 cgd if (m->m_flags & M_EXT)
485 1.1 cgd mbcnt += m->m_ext.ext_size;
486 1.1 cgd if (m->m_nextpkt)
487 1.1 cgd panic("sbcheck nextpkt");
488 1.1 cgd }
489 1.1 cgd if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
490 1.15 christos printf("cc %d != %d || mbcnt %d != %d\n", len, sb->sb_cc,
491 1.1 cgd mbcnt, sb->sb_mbcnt);
492 1.1 cgd panic("sbcheck");
493 1.1 cgd }
494 1.1 cgd }
495 1.1 cgd #endif
496 1.1 cgd
497 1.1 cgd /*
498 1.1 cgd * As above, except the mbuf chain
499 1.1 cgd * begins a new record.
500 1.1 cgd */
501 1.7 mycroft void
502 1.1 cgd sbappendrecord(sb, m0)
503 1.1 cgd register struct sockbuf *sb;
504 1.1 cgd register struct mbuf *m0;
505 1.1 cgd {
506 1.1 cgd register struct mbuf *m;
507 1.1 cgd
508 1.1 cgd if (m0 == 0)
509 1.1 cgd return;
510 1.11 christos if ((m = sb->sb_mb) != NULL)
511 1.1 cgd while (m->m_nextpkt)
512 1.1 cgd m = m->m_nextpkt;
513 1.1 cgd /*
514 1.1 cgd * Put the first mbuf on the queue.
515 1.1 cgd * Note this permits zero length records.
516 1.1 cgd */
517 1.1 cgd sballoc(sb, m0);
518 1.1 cgd if (m)
519 1.1 cgd m->m_nextpkt = m0;
520 1.1 cgd else
521 1.1 cgd sb->sb_mb = m0;
522 1.1 cgd m = m0->m_next;
523 1.1 cgd m0->m_next = 0;
524 1.1 cgd if (m && (m0->m_flags & M_EOR)) {
525 1.1 cgd m0->m_flags &= ~M_EOR;
526 1.1 cgd m->m_flags |= M_EOR;
527 1.1 cgd }
528 1.1 cgd sbcompress(sb, m, m0);
529 1.1 cgd }
530 1.1 cgd
531 1.1 cgd /*
532 1.1 cgd * As above except that OOB data
533 1.1 cgd * is inserted at the beginning of the sockbuf,
534 1.1 cgd * but after any other OOB data.
535 1.1 cgd */
536 1.7 mycroft void
537 1.1 cgd sbinsertoob(sb, m0)
538 1.1 cgd register struct sockbuf *sb;
539 1.1 cgd register struct mbuf *m0;
540 1.1 cgd {
541 1.1 cgd register struct mbuf *m;
542 1.1 cgd register struct mbuf **mp;
543 1.1 cgd
544 1.1 cgd if (m0 == 0)
545 1.1 cgd return;
546 1.11 christos for (mp = &sb->sb_mb; (m = *mp) != NULL; mp = &((*mp)->m_nextpkt)) {
547 1.1 cgd again:
548 1.1 cgd switch (m->m_type) {
549 1.1 cgd
550 1.1 cgd case MT_OOBDATA:
551 1.1 cgd continue; /* WANT next train */
552 1.1 cgd
553 1.1 cgd case MT_CONTROL:
554 1.11 christos if ((m = m->m_next) != NULL)
555 1.1 cgd goto again; /* inspect THIS train further */
556 1.1 cgd }
557 1.1 cgd break;
558 1.1 cgd }
559 1.1 cgd /*
560 1.1 cgd * Put the first mbuf on the queue.
561 1.1 cgd * Note this permits zero length records.
562 1.1 cgd */
563 1.1 cgd sballoc(sb, m0);
564 1.1 cgd m0->m_nextpkt = *mp;
565 1.1 cgd *mp = m0;
566 1.1 cgd m = m0->m_next;
567 1.1 cgd m0->m_next = 0;
568 1.1 cgd if (m && (m0->m_flags & M_EOR)) {
569 1.1 cgd m0->m_flags &= ~M_EOR;
570 1.1 cgd m->m_flags |= M_EOR;
571 1.1 cgd }
572 1.1 cgd sbcompress(sb, m, m0);
573 1.1 cgd }
574 1.1 cgd
575 1.1 cgd /*
576 1.1 cgd * Append address and data, and optionally, control (ancillary) data
577 1.1 cgd * to the receive queue of a socket. If present,
578 1.1 cgd * m0 must include a packet header with total length.
579 1.1 cgd * Returns 0 if no space in sockbuf or insufficient mbufs.
580 1.1 cgd */
581 1.7 mycroft int
582 1.1 cgd sbappendaddr(sb, asa, m0, control)
583 1.1 cgd register struct sockbuf *sb;
584 1.1 cgd struct sockaddr *asa;
585 1.1 cgd struct mbuf *m0, *control;
586 1.1 cgd {
587 1.1 cgd register struct mbuf *m, *n;
588 1.1 cgd int space = asa->sa_len;
589 1.1 cgd
590 1.1 cgd if (m0 && (m0->m_flags & M_PKTHDR) == 0)
591 1.1 cgd panic("sbappendaddr");
592 1.1 cgd if (m0)
593 1.1 cgd space += m0->m_pkthdr.len;
594 1.1 cgd for (n = control; n; n = n->m_next) {
595 1.1 cgd space += n->m_len;
596 1.1 cgd if (n->m_next == 0) /* keep pointer to last control buf */
597 1.1 cgd break;
598 1.1 cgd }
599 1.1 cgd if (space > sbspace(sb))
600 1.1 cgd return (0);
601 1.1 cgd MGET(m, M_DONTWAIT, MT_SONAME);
602 1.1 cgd if (m == 0)
603 1.1 cgd return (0);
604 1.20 thorpej if (asa->sa_len > MLEN) {
605 1.20 thorpej MEXTMALLOC(m, asa->sa_len, M_NOWAIT);
606 1.20 thorpej if ((m->m_flags & M_EXT) == 0) {
607 1.20 thorpej m_free(m);
608 1.20 thorpej return (0);
609 1.20 thorpej }
610 1.20 thorpej }
611 1.1 cgd m->m_len = asa->sa_len;
612 1.1 cgd bcopy((caddr_t)asa, mtod(m, caddr_t), asa->sa_len);
613 1.1 cgd if (n)
614 1.1 cgd n->m_next = m0; /* concatenate data to control */
615 1.1 cgd else
616 1.1 cgd control = m0;
617 1.1 cgd m->m_next = control;
618 1.1 cgd for (n = m; n; n = n->m_next)
619 1.1 cgd sballoc(sb, n);
620 1.11 christos if ((n = sb->sb_mb) != NULL) {
621 1.1 cgd while (n->m_nextpkt)
622 1.1 cgd n = n->m_nextpkt;
623 1.1 cgd n->m_nextpkt = m;
624 1.1 cgd } else
625 1.1 cgd sb->sb_mb = m;
626 1.1 cgd return (1);
627 1.1 cgd }
628 1.1 cgd
629 1.7 mycroft int
630 1.1 cgd sbappendcontrol(sb, m0, control)
631 1.1 cgd struct sockbuf *sb;
632 1.7 mycroft struct mbuf *m0, *control;
633 1.1 cgd {
634 1.1 cgd register struct mbuf *m, *n;
635 1.1 cgd int space = 0;
636 1.1 cgd
637 1.1 cgd if (control == 0)
638 1.1 cgd panic("sbappendcontrol");
639 1.1 cgd for (m = control; ; m = m->m_next) {
640 1.1 cgd space += m->m_len;
641 1.1 cgd if (m->m_next == 0)
642 1.1 cgd break;
643 1.1 cgd }
644 1.1 cgd n = m; /* save pointer to last control buffer */
645 1.1 cgd for (m = m0; m; m = m->m_next)
646 1.1 cgd space += m->m_len;
647 1.1 cgd if (space > sbspace(sb))
648 1.1 cgd return (0);
649 1.1 cgd n->m_next = m0; /* concatenate data to control */
650 1.1 cgd for (m = control; m; m = m->m_next)
651 1.1 cgd sballoc(sb, m);
652 1.11 christos if ((n = sb->sb_mb) != NULL) {
653 1.1 cgd while (n->m_nextpkt)
654 1.1 cgd n = n->m_nextpkt;
655 1.1 cgd n->m_nextpkt = control;
656 1.1 cgd } else
657 1.1 cgd sb->sb_mb = control;
658 1.1 cgd return (1);
659 1.1 cgd }
660 1.1 cgd
661 1.1 cgd /*
662 1.1 cgd * Compress mbuf chain m into the socket
663 1.1 cgd * buffer sb following mbuf n. If n
664 1.1 cgd * is null, the buffer is presumed empty.
665 1.1 cgd */
666 1.7 mycroft void
667 1.1 cgd sbcompress(sb, m, n)
668 1.1 cgd register struct sockbuf *sb;
669 1.1 cgd register struct mbuf *m, *n;
670 1.1 cgd {
671 1.1 cgd register int eor = 0;
672 1.1 cgd register struct mbuf *o;
673 1.1 cgd
674 1.1 cgd while (m) {
675 1.1 cgd eor |= m->m_flags & M_EOR;
676 1.1 cgd if (m->m_len == 0 &&
677 1.1 cgd (eor == 0 ||
678 1.1 cgd (((o = m->m_next) || (o = n)) &&
679 1.1 cgd o->m_type == m->m_type))) {
680 1.1 cgd m = m_free(m);
681 1.1 cgd continue;
682 1.1 cgd }
683 1.1 cgd if (n && (n->m_flags & (M_EXT | M_EOR)) == 0 &&
684 1.1 cgd (n->m_data + n->m_len + m->m_len) < &n->m_dat[MLEN] &&
685 1.1 cgd n->m_type == m->m_type) {
686 1.1 cgd bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
687 1.1 cgd (unsigned)m->m_len);
688 1.1 cgd n->m_len += m->m_len;
689 1.1 cgd sb->sb_cc += m->m_len;
690 1.1 cgd m = m_free(m);
691 1.1 cgd continue;
692 1.1 cgd }
693 1.1 cgd if (n)
694 1.1 cgd n->m_next = m;
695 1.1 cgd else
696 1.1 cgd sb->sb_mb = m;
697 1.1 cgd sballoc(sb, m);
698 1.1 cgd n = m;
699 1.1 cgd m->m_flags &= ~M_EOR;
700 1.1 cgd m = m->m_next;
701 1.1 cgd n->m_next = 0;
702 1.1 cgd }
703 1.1 cgd if (eor) {
704 1.1 cgd if (n)
705 1.1 cgd n->m_flags |= eor;
706 1.1 cgd else
707 1.15 christos printf("semi-panic: sbcompress\n");
708 1.1 cgd }
709 1.1 cgd }
710 1.1 cgd
711 1.1 cgd /*
712 1.1 cgd * Free all mbufs in a sockbuf.
713 1.1 cgd * Check that all resources are reclaimed.
714 1.1 cgd */
715 1.7 mycroft void
716 1.1 cgd sbflush(sb)
717 1.1 cgd register struct sockbuf *sb;
718 1.1 cgd {
719 1.1 cgd
720 1.1 cgd if (sb->sb_flags & SB_LOCK)
721 1.1 cgd panic("sbflush");
722 1.1 cgd while (sb->sb_mbcnt)
723 1.1 cgd sbdrop(sb, (int)sb->sb_cc);
724 1.1 cgd if (sb->sb_cc || sb->sb_mb)
725 1.1 cgd panic("sbflush 2");
726 1.1 cgd }
727 1.1 cgd
728 1.1 cgd /*
729 1.1 cgd * Drop data from (the front of) a sockbuf.
730 1.1 cgd */
731 1.7 mycroft void
732 1.1 cgd sbdrop(sb, len)
733 1.1 cgd register struct sockbuf *sb;
734 1.1 cgd register int len;
735 1.1 cgd {
736 1.1 cgd register struct mbuf *m, *mn;
737 1.1 cgd struct mbuf *next;
738 1.1 cgd
739 1.1 cgd next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
740 1.1 cgd while (len > 0) {
741 1.1 cgd if (m == 0) {
742 1.1 cgd if (next == 0)
743 1.1 cgd panic("sbdrop");
744 1.1 cgd m = next;
745 1.1 cgd next = m->m_nextpkt;
746 1.1 cgd continue;
747 1.1 cgd }
748 1.1 cgd if (m->m_len > len) {
749 1.1 cgd m->m_len -= len;
750 1.1 cgd m->m_data += len;
751 1.1 cgd sb->sb_cc -= len;
752 1.1 cgd break;
753 1.1 cgd }
754 1.1 cgd len -= m->m_len;
755 1.1 cgd sbfree(sb, m);
756 1.1 cgd MFREE(m, mn);
757 1.1 cgd m = mn;
758 1.1 cgd }
759 1.1 cgd while (m && m->m_len == 0) {
760 1.1 cgd sbfree(sb, m);
761 1.1 cgd MFREE(m, mn);
762 1.1 cgd m = mn;
763 1.1 cgd }
764 1.1 cgd if (m) {
765 1.1 cgd sb->sb_mb = m;
766 1.1 cgd m->m_nextpkt = next;
767 1.1 cgd } else
768 1.1 cgd sb->sb_mb = next;
769 1.1 cgd }
770 1.1 cgd
771 1.1 cgd /*
772 1.1 cgd * Drop a record off the front of a sockbuf
773 1.1 cgd * and move the next record to the front.
774 1.1 cgd */
775 1.7 mycroft void
776 1.1 cgd sbdroprecord(sb)
777 1.1 cgd register struct sockbuf *sb;
778 1.1 cgd {
779 1.1 cgd register struct mbuf *m, *mn;
780 1.1 cgd
781 1.1 cgd m = sb->sb_mb;
782 1.1 cgd if (m) {
783 1.1 cgd sb->sb_mb = m->m_nextpkt;
784 1.1 cgd do {
785 1.1 cgd sbfree(sb, m);
786 1.1 cgd MFREE(m, mn);
787 1.11 christos } while ((m = mn) != NULL);
788 1.1 cgd }
789 1.19 thorpej }
790 1.19 thorpej
791 1.19 thorpej /*
792 1.19 thorpej * Create a "control" mbuf containing the specified data
793 1.19 thorpej * with the specified type for presentation on a socket buffer.
794 1.19 thorpej */
795 1.19 thorpej struct mbuf *
796 1.19 thorpej sbcreatecontrol(p, size, type, level)
797 1.19 thorpej caddr_t p;
798 1.19 thorpej register int size;
799 1.19 thorpej int type, level;
800 1.19 thorpej {
801 1.19 thorpej register struct cmsghdr *cp;
802 1.19 thorpej struct mbuf *m;
803 1.19 thorpej
804 1.19 thorpej if ((m = m_get(M_DONTWAIT, MT_CONTROL)) == NULL)
805 1.19 thorpej return ((struct mbuf *) NULL);
806 1.19 thorpej cp = mtod(m, struct cmsghdr *);
807 1.19 thorpej bcopy(p, CMSG_DATA(cp), size);
808 1.19 thorpej size += sizeof(*cp);
809 1.19 thorpej m->m_len = size;
810 1.19 thorpej cp->cmsg_len = size;
811 1.19 thorpej cp->cmsg_level = level;
812 1.19 thorpej cp->cmsg_type = type;
813 1.19 thorpej return (m);
814 1.1 cgd }
815