1/*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
30 */
31
32#include <sys/cdefs.h>
| 1/*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
30 */
31
32#include <sys/cdefs.h>
|
33__FBSDID("$FreeBSD: head/sys/kern/uipc_sockbuf.c 169236 2007-05-03 14:42:42Z rwatson $");
| 33__FBSDID("$FreeBSD: head/sys/kern/uipc_sockbuf.c 169624 2007-05-16 20:41:08Z rwatson $");
|
34
35#include "opt_param.h"
36
37#include <sys/param.h>
38#include <sys/aio.h> /* for aio_swake proto */
39#include <sys/kernel.h>
40#include <sys/lock.h>
41#include <sys/mbuf.h>
42#include <sys/mutex.h>
43#include <sys/proc.h>
44#include <sys/protosw.h>
45#include <sys/resourcevar.h>
46#include <sys/signalvar.h>
47#include <sys/socket.h>
48#include <sys/socketvar.h>
49#include <sys/sx.h>
50#include <sys/sysctl.h>
51
52/*
53 * Function pointer set by the AIO routines so that the socket buffer code
54 * can call back into the AIO module if it is loaded.
55 */
56void (*aio_swake)(struct socket *, struct sockbuf *);
57
58/*
59 * Primitive routines for operating on socket buffers
60 */
61
62u_long sb_max = SB_MAX;
63static u_long sb_max_adj =
64 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
65
66static u_long sb_efficiency = 8; /* parameter for sbreserve() */
67
68static void sbdrop_internal(struct sockbuf *sb, int len);
69static void sbflush_internal(struct sockbuf *sb);
70static void sbrelease_internal(struct sockbuf *sb, struct socket *so);
71
72/*
73 * Socantsendmore indicates that no more data will be sent on the socket; it
74 * would normally be applied to a socket when the user informs the system
75 * that no more data is to be sent, by the protocol code (in case
76 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
77 * received, and will normally be applied to the socket by a protocol when it
78 * detects that the peer will send no more data. Data queued for reading in
79 * the socket may yet be read.
80 */
81void
82socantsendmore_locked(struct socket *so)
83{
84
85 SOCKBUF_LOCK_ASSERT(&so->so_snd);
86
87 so->so_snd.sb_state |= SBS_CANTSENDMORE;
88 sowwakeup_locked(so);
89 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
90}
91
92void
93socantsendmore(struct socket *so)
94{
95
96 SOCKBUF_LOCK(&so->so_snd);
97 socantsendmore_locked(so);
98 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
99}
100
101void
102socantrcvmore_locked(struct socket *so)
103{
104
105 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
106
107 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
108 sorwakeup_locked(so);
109 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
110}
111
112void
113socantrcvmore(struct socket *so)
114{
115
116 SOCKBUF_LOCK(&so->so_rcv);
117 socantrcvmore_locked(so);
118 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
119}
120
121/*
122 * Wait for data to arrive at/drain from a socket buffer.
123 */
124int
125sbwait(struct sockbuf *sb)
126{
127
128 SOCKBUF_LOCK_ASSERT(sb);
129
130 sb->sb_flags |= SB_WAIT;
131 return (msleep(&sb->sb_cc, &sb->sb_mtx,
132 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
133 sb->sb_timeo));
134}
135
136int
137sblock(struct sockbuf *sb, int flags)
138{
139
140 if (flags == M_WAITOK) {
141 sx_xlock(&sb->sb_sx);
142 return (0);
143 } else {
144 if (sx_try_xlock(&sb->sb_sx) == 0)
145 return (EWOULDBLOCK);
146 return (0);
147 }
148}
149
150void
151sbunlock(struct sockbuf *sb)
152{
153
154 sx_xunlock(&sb->sb_sx);
155}
156
157/*
158 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
159 * via SIGIO if the socket has the SS_ASYNC flag set.
160 *
161 * Called with the socket buffer lock held; will release the lock by the end
162 * of the function. This allows the caller to acquire the socket buffer lock
163 * while testing for the need for various sorts of wakeup and hold it through
164 * to the point where it's no longer required. We currently hold the lock
165 * through calls out to other subsystems (with the exception of kqueue), and
166 * then release it to avoid lock order issues. It's not clear that's
167 * correct.
168 */
169void
170sowakeup(struct socket *so, struct sockbuf *sb)
171{
172
173 SOCKBUF_LOCK_ASSERT(sb);
174
175 selwakeuppri(&sb->sb_sel, PSOCK);
176 sb->sb_flags &= ~SB_SEL;
177 if (sb->sb_flags & SB_WAIT) {
178 sb->sb_flags &= ~SB_WAIT;
179 wakeup(&sb->sb_cc);
180 }
181 KNOTE_LOCKED(&sb->sb_sel.si_note, 0);
182 SOCKBUF_UNLOCK(sb);
183 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
184 pgsigio(&so->so_sigio, SIGIO, 0);
185 if (sb->sb_flags & SB_UPCALL)
186 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
187 if (sb->sb_flags & SB_AIO)
188 aio_swake(so, sb);
189 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
190}
191
192/*
193 * Socket buffer (struct sockbuf) utility routines.
194 *
195 * Each socket contains two socket buffers: one for sending data and one for
196 * receiving data. Each buffer contains a queue of mbufs, information about
197 * the number of mbufs and amount of data in the queue, and other fields
198 * allowing select() statements and notification on data availability to be
199 * implemented.
200 *
201 * Data stored in a socket buffer is maintained as a list of records. Each
202 * record is a list of mbufs chained together with the m_next field. Records
203 * are chained together with the m_nextpkt field. The upper level routine
204 * soreceive() expects the following conventions to be observed when placing
205 * information in the receive buffer:
206 *
207 * 1. If the protocol requires each message be preceded by the sender's name,
208 * then a record containing that name must be present before any
209 * associated data (mbuf's must be of type MT_SONAME).
210 * 2. If the protocol supports the exchange of ``access rights'' (really just
211 * additional data associated with the message), and there are ``rights''
212 * to be received, then a record containing this data should be present
213 * (mbuf's must be of type MT_RIGHTS).
214 * 3. If a name or rights record exists, then it must be followed by a data
215 * record, perhaps of zero length.
216 *
217 * Before using a new socket structure it is first necessary to reserve
218 * buffer space to the socket, by calling sbreserve(). This should commit
219 * some of the available buffer space in the system buffer pool for the
220 * socket (currently, it does nothing but enforce limits). The space should
221 * be released by calling sbrelease() when the socket is destroyed.
222 */
223int
224soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
225{
226 struct thread *td = curthread;
227
228 SOCKBUF_LOCK(&so->so_snd);
229 SOCKBUF_LOCK(&so->so_rcv);
230 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
231 goto bad;
232 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
233 goto bad2;
234 if (so->so_rcv.sb_lowat == 0)
235 so->so_rcv.sb_lowat = 1;
236 if (so->so_snd.sb_lowat == 0)
237 so->so_snd.sb_lowat = MCLBYTES;
238 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
239 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
240 SOCKBUF_UNLOCK(&so->so_rcv);
241 SOCKBUF_UNLOCK(&so->so_snd);
242 return (0);
243bad2:
244 sbrelease_locked(&so->so_snd, so);
245bad:
246 SOCKBUF_UNLOCK(&so->so_rcv);
247 SOCKBUF_UNLOCK(&so->so_snd);
248 return (ENOBUFS);
249}
250
251static int
252sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
253{
254 int error = 0;
255 u_long tmp_sb_max = sb_max;
256
257 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
258 if (error || !req->newptr)
259 return (error);
260 if (tmp_sb_max < MSIZE + MCLBYTES)
261 return (EINVAL);
262 sb_max = tmp_sb_max;
263 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
264 return (0);
265}
266
267/*
268 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
269 * become limiting if buffering efficiency is near the normal case.
270 */
271int
272sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
273 struct thread *td)
274{
275 rlim_t sbsize_limit;
276
277 SOCKBUF_LOCK_ASSERT(sb);
278
279 /*
280 * td will only be NULL when we're in an interrupt (e.g. in
281 * tcp_input()).
282 *
283 * XXXRW: This comment needs updating, as might the code.
284 */
285 if (cc > sb_max_adj)
286 return (0);
287 if (td != NULL) {
288 PROC_LOCK(td->td_proc);
289 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
290 PROC_UNLOCK(td->td_proc);
291 } else
292 sbsize_limit = RLIM_INFINITY;
293 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
294 sbsize_limit))
295 return (0);
296 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
297 if (sb->sb_lowat > sb->sb_hiwat)
298 sb->sb_lowat = sb->sb_hiwat;
299 return (1);
300}
301
302int
303sbreserve(struct sockbuf *sb, u_long cc, struct socket *so,
304 struct thread *td)
305{
306 int error;
307
308 SOCKBUF_LOCK(sb);
309 error = sbreserve_locked(sb, cc, so, td);
310 SOCKBUF_UNLOCK(sb);
311 return (error);
312}
313
314/*
315 * Free mbufs held by a socket, and reserved mbuf space.
316 */
317static void
318sbrelease_internal(struct sockbuf *sb, struct socket *so)
319{
320
321 sbflush_internal(sb);
322 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
323 RLIM_INFINITY);
324 sb->sb_mbmax = 0;
325}
326
327void
328sbrelease_locked(struct sockbuf *sb, struct socket *so)
329{
330
331 SOCKBUF_LOCK_ASSERT(sb);
332
333 sbrelease_internal(sb, so);
334}
335
336void
337sbrelease(struct sockbuf *sb, struct socket *so)
338{
339
340 SOCKBUF_LOCK(sb);
341 sbrelease_locked(sb, so);
342 SOCKBUF_UNLOCK(sb);
343}
344
345void
346sbdestroy(struct sockbuf *sb, struct socket *so)
347{
348
349 sbrelease_internal(sb, so);
350}
351
| 34
35#include "opt_param.h"
36
37#include <sys/param.h>
38#include <sys/aio.h> /* for aio_swake proto */
39#include <sys/kernel.h>
40#include <sys/lock.h>
41#include <sys/mbuf.h>
42#include <sys/mutex.h>
43#include <sys/proc.h>
44#include <sys/protosw.h>
45#include <sys/resourcevar.h>
46#include <sys/signalvar.h>
47#include <sys/socket.h>
48#include <sys/socketvar.h>
49#include <sys/sx.h>
50#include <sys/sysctl.h>
51
52/*
53 * Function pointer set by the AIO routines so that the socket buffer code
54 * can call back into the AIO module if it is loaded.
55 */
56void (*aio_swake)(struct socket *, struct sockbuf *);
57
58/*
59 * Primitive routines for operating on socket buffers
60 */
61
62u_long sb_max = SB_MAX;
63static u_long sb_max_adj =
64 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
65
66static u_long sb_efficiency = 8; /* parameter for sbreserve() */
67
68static void sbdrop_internal(struct sockbuf *sb, int len);
69static void sbflush_internal(struct sockbuf *sb);
70static void sbrelease_internal(struct sockbuf *sb, struct socket *so);
71
72/*
73 * Socantsendmore indicates that no more data will be sent on the socket; it
74 * would normally be applied to a socket when the user informs the system
75 * that no more data is to be sent, by the protocol code (in case
76 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
77 * received, and will normally be applied to the socket by a protocol when it
78 * detects that the peer will send no more data. Data queued for reading in
79 * the socket may yet be read.
80 */
81void
82socantsendmore_locked(struct socket *so)
83{
84
85 SOCKBUF_LOCK_ASSERT(&so->so_snd);
86
87 so->so_snd.sb_state |= SBS_CANTSENDMORE;
88 sowwakeup_locked(so);
89 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
90}
91
92void
93socantsendmore(struct socket *so)
94{
95
96 SOCKBUF_LOCK(&so->so_snd);
97 socantsendmore_locked(so);
98 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
99}
100
101void
102socantrcvmore_locked(struct socket *so)
103{
104
105 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
106
107 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
108 sorwakeup_locked(so);
109 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
110}
111
112void
113socantrcvmore(struct socket *so)
114{
115
116 SOCKBUF_LOCK(&so->so_rcv);
117 socantrcvmore_locked(so);
118 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
119}
120
121/*
122 * Wait for data to arrive at/drain from a socket buffer.
123 */
124int
125sbwait(struct sockbuf *sb)
126{
127
128 SOCKBUF_LOCK_ASSERT(sb);
129
130 sb->sb_flags |= SB_WAIT;
131 return (msleep(&sb->sb_cc, &sb->sb_mtx,
132 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
133 sb->sb_timeo));
134}
135
136int
137sblock(struct sockbuf *sb, int flags)
138{
139
140 if (flags == M_WAITOK) {
141 sx_xlock(&sb->sb_sx);
142 return (0);
143 } else {
144 if (sx_try_xlock(&sb->sb_sx) == 0)
145 return (EWOULDBLOCK);
146 return (0);
147 }
148}
149
150void
151sbunlock(struct sockbuf *sb)
152{
153
154 sx_xunlock(&sb->sb_sx);
155}
156
157/*
158 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
159 * via SIGIO if the socket has the SS_ASYNC flag set.
160 *
161 * Called with the socket buffer lock held; will release the lock by the end
162 * of the function. This allows the caller to acquire the socket buffer lock
163 * while testing for the need for various sorts of wakeup and hold it through
164 * to the point where it's no longer required. We currently hold the lock
165 * through calls out to other subsystems (with the exception of kqueue), and
166 * then release it to avoid lock order issues. It's not clear that's
167 * correct.
168 */
169void
170sowakeup(struct socket *so, struct sockbuf *sb)
171{
172
173 SOCKBUF_LOCK_ASSERT(sb);
174
175 selwakeuppri(&sb->sb_sel, PSOCK);
176 sb->sb_flags &= ~SB_SEL;
177 if (sb->sb_flags & SB_WAIT) {
178 sb->sb_flags &= ~SB_WAIT;
179 wakeup(&sb->sb_cc);
180 }
181 KNOTE_LOCKED(&sb->sb_sel.si_note, 0);
182 SOCKBUF_UNLOCK(sb);
183 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
184 pgsigio(&so->so_sigio, SIGIO, 0);
185 if (sb->sb_flags & SB_UPCALL)
186 (*so->so_upcall)(so, so->so_upcallarg, M_DONTWAIT);
187 if (sb->sb_flags & SB_AIO)
188 aio_swake(so, sb);
189 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
190}
191
192/*
193 * Socket buffer (struct sockbuf) utility routines.
194 *
195 * Each socket contains two socket buffers: one for sending data and one for
196 * receiving data. Each buffer contains a queue of mbufs, information about
197 * the number of mbufs and amount of data in the queue, and other fields
198 * allowing select() statements and notification on data availability to be
199 * implemented.
200 *
201 * Data stored in a socket buffer is maintained as a list of records. Each
202 * record is a list of mbufs chained together with the m_next field. Records
203 * are chained together with the m_nextpkt field. The upper level routine
204 * soreceive() expects the following conventions to be observed when placing
205 * information in the receive buffer:
206 *
207 * 1. If the protocol requires each message be preceded by the sender's name,
208 * then a record containing that name must be present before any
209 * associated data (mbuf's must be of type MT_SONAME).
210 * 2. If the protocol supports the exchange of ``access rights'' (really just
211 * additional data associated with the message), and there are ``rights''
212 * to be received, then a record containing this data should be present
213 * (mbuf's must be of type MT_RIGHTS).
214 * 3. If a name or rights record exists, then it must be followed by a data
215 * record, perhaps of zero length.
216 *
217 * Before using a new socket structure it is first necessary to reserve
218 * buffer space to the socket, by calling sbreserve(). This should commit
219 * some of the available buffer space in the system buffer pool for the
220 * socket (currently, it does nothing but enforce limits). The space should
221 * be released by calling sbrelease() when the socket is destroyed.
222 */
223int
224soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
225{
226 struct thread *td = curthread;
227
228 SOCKBUF_LOCK(&so->so_snd);
229 SOCKBUF_LOCK(&so->so_rcv);
230 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
231 goto bad;
232 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
233 goto bad2;
234 if (so->so_rcv.sb_lowat == 0)
235 so->so_rcv.sb_lowat = 1;
236 if (so->so_snd.sb_lowat == 0)
237 so->so_snd.sb_lowat = MCLBYTES;
238 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
239 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
240 SOCKBUF_UNLOCK(&so->so_rcv);
241 SOCKBUF_UNLOCK(&so->so_snd);
242 return (0);
243bad2:
244 sbrelease_locked(&so->so_snd, so);
245bad:
246 SOCKBUF_UNLOCK(&so->so_rcv);
247 SOCKBUF_UNLOCK(&so->so_snd);
248 return (ENOBUFS);
249}
250
251static int
252sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
253{
254 int error = 0;
255 u_long tmp_sb_max = sb_max;
256
257 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
258 if (error || !req->newptr)
259 return (error);
260 if (tmp_sb_max < MSIZE + MCLBYTES)
261 return (EINVAL);
262 sb_max = tmp_sb_max;
263 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
264 return (0);
265}
266
267/*
268 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
269 * become limiting if buffering efficiency is near the normal case.
270 */
271int
272sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
273 struct thread *td)
274{
275 rlim_t sbsize_limit;
276
277 SOCKBUF_LOCK_ASSERT(sb);
278
279 /*
280 * td will only be NULL when we're in an interrupt (e.g. in
281 * tcp_input()).
282 *
283 * XXXRW: This comment needs updating, as might the code.
284 */
285 if (cc > sb_max_adj)
286 return (0);
287 if (td != NULL) {
288 PROC_LOCK(td->td_proc);
289 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
290 PROC_UNLOCK(td->td_proc);
291 } else
292 sbsize_limit = RLIM_INFINITY;
293 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
294 sbsize_limit))
295 return (0);
296 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
297 if (sb->sb_lowat > sb->sb_hiwat)
298 sb->sb_lowat = sb->sb_hiwat;
299 return (1);
300}
301
302int
303sbreserve(struct sockbuf *sb, u_long cc, struct socket *so,
304 struct thread *td)
305{
306 int error;
307
308 SOCKBUF_LOCK(sb);
309 error = sbreserve_locked(sb, cc, so, td);
310 SOCKBUF_UNLOCK(sb);
311 return (error);
312}
313
314/*
315 * Free mbufs held by a socket, and reserved mbuf space.
316 */
317static void
318sbrelease_internal(struct sockbuf *sb, struct socket *so)
319{
320
321 sbflush_internal(sb);
322 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
323 RLIM_INFINITY);
324 sb->sb_mbmax = 0;
325}
326
327void
328sbrelease_locked(struct sockbuf *sb, struct socket *so)
329{
330
331 SOCKBUF_LOCK_ASSERT(sb);
332
333 sbrelease_internal(sb, so);
334}
335
336void
337sbrelease(struct sockbuf *sb, struct socket *so)
338{
339
340 SOCKBUF_LOCK(sb);
341 sbrelease_locked(sb, so);
342 SOCKBUF_UNLOCK(sb);
343}
344
345void
346sbdestroy(struct sockbuf *sb, struct socket *so)
347{
348
349 sbrelease_internal(sb, so);
350}
351
|
352
| |
353/*
354 * Routines to add and remove data from an mbuf queue.
355 *
356 * The routines sbappend() or sbappendrecord() are normally called to append
357 * new mbufs to a socket buffer, after checking that adequate space is
358 * available, comparing the function sbspace() with the amount of data to be
359 * added. sbappendrecord() differs from sbappend() in that data supplied is
360 * treated as the beginning of a new record. To place a sender's address,
361 * optional access rights, and data in a socket receive buffer,
362 * sbappendaddr() should be used. To place access rights and data in a
363 * socket receive buffer, sbappendrights() should be used. In either case,
364 * the new data begins a new record. Note that unlike sbappend() and
365 * sbappendrecord(), these routines check for the caller that there will be
366 * enough space to store the data. Each fails if there is not enough space,
367 * or if it cannot find mbufs to store additional information in.
368 *
369 * Reliable protocols may use the socket send buffer to hold data awaiting
370 * acknowledgement. Data is normally copied from a socket send buffer in a
371 * protocol with m_copy for output to a peer, and then removing the data from
372 * the socket buffer with sbdrop() or sbdroprecord() when the data is
373 * acknowledged by the peer.
374 */
375#ifdef SOCKBUF_DEBUG
376void
377sblastrecordchk(struct sockbuf *sb, const char *file, int line)
378{
379 struct mbuf *m = sb->sb_mb;
380
381 SOCKBUF_LOCK_ASSERT(sb);
382
383 while (m && m->m_nextpkt)
384 m = m->m_nextpkt;
385
386 if (m != sb->sb_lastrecord) {
387 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
388 __func__, sb->sb_mb, sb->sb_lastrecord, m);
389 printf("packet chain:\n");
390 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
391 printf("\t%p\n", m);
392 panic("%s from %s:%u", __func__, file, line);
393 }
394}
395
396void
397sblastmbufchk(struct sockbuf *sb, const char *file, int line)
398{
399 struct mbuf *m = sb->sb_mb;
400 struct mbuf *n;
401
402 SOCKBUF_LOCK_ASSERT(sb);
403
404 while (m && m->m_nextpkt)
405 m = m->m_nextpkt;
406
407 while (m && m->m_next)
408 m = m->m_next;
409
410 if (m != sb->sb_mbtail) {
411 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
412 __func__, sb->sb_mb, sb->sb_mbtail, m);
413 printf("packet tree:\n");
414 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
415 printf("\t");
416 for (n = m; n != NULL; n = n->m_next)
417 printf("%p ", n);
418 printf("\n");
419 }
420 panic("%s from %s:%u", __func__, file, line);
421 }
422}
423#endif /* SOCKBUF_DEBUG */
424
425#define SBLINKRECORD(sb, m0) do { \
426 SOCKBUF_LOCK_ASSERT(sb); \
427 if ((sb)->sb_lastrecord != NULL) \
428 (sb)->sb_lastrecord->m_nextpkt = (m0); \
429 else \
430 (sb)->sb_mb = (m0); \
431 (sb)->sb_lastrecord = (m0); \
432} while (/*CONSTCOND*/0)
433
434/*
435 * Append mbuf chain m to the last record in the socket buffer sb. The
436 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
437 * are discarded and mbufs are compacted where possible.
438 */
439void
440sbappend_locked(struct sockbuf *sb, struct mbuf *m)
441{
442 struct mbuf *n;
443
444 SOCKBUF_LOCK_ASSERT(sb);
445
446 if (m == 0)
447 return;
448
449 SBLASTRECORDCHK(sb);
450 n = sb->sb_mb;
451 if (n) {
452 while (n->m_nextpkt)
453 n = n->m_nextpkt;
454 do {
455 if (n->m_flags & M_EOR) {
456 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
457 return;
458 }
459 } while (n->m_next && (n = n->m_next));
460 } else {
461 /*
462 * XXX Would like to simply use sb_mbtail here, but
463 * XXX I need to verify that I won't miss an EOR that
464 * XXX way.
465 */
466 if ((n = sb->sb_lastrecord) != NULL) {
467 do {
468 if (n->m_flags & M_EOR) {
469 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
470 return;
471 }
472 } while (n->m_next && (n = n->m_next));
473 } else {
474 /*
475 * If this is the first record in the socket buffer,
476 * it's also the last record.
477 */
478 sb->sb_lastrecord = m;
479 }
480 }
481 sbcompress(sb, m, n);
482 SBLASTRECORDCHK(sb);
483}
484
485/*
486 * Append mbuf chain m to the last record in the socket buffer sb. The
487 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
488 * are discarded and mbufs are compacted where possible.
489 */
490void
491sbappend(struct sockbuf *sb, struct mbuf *m)
492{
493
494 SOCKBUF_LOCK(sb);
495 sbappend_locked(sb, m);
496 SOCKBUF_UNLOCK(sb);
497}
498
499/*
500 * This version of sbappend() should only be used when the caller absolutely
501 * knows that there will never be more than one record in the socket buffer,
502 * that is, a stream protocol (such as TCP).
503 */
504void
505sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
506{
507 SOCKBUF_LOCK_ASSERT(sb);
508
509 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
510 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
511
512 SBLASTMBUFCHK(sb);
513
514 sbcompress(sb, m, sb->sb_mbtail);
515
516 sb->sb_lastrecord = sb->sb_mb;
517 SBLASTRECORDCHK(sb);
518}
519
520/*
521 * This version of sbappend() should only be used when the caller absolutely
522 * knows that there will never be more than one record in the socket buffer,
523 * that is, a stream protocol (such as TCP).
524 */
525void
526sbappendstream(struct sockbuf *sb, struct mbuf *m)
527{
528
529 SOCKBUF_LOCK(sb);
530 sbappendstream_locked(sb, m);
531 SOCKBUF_UNLOCK(sb);
532}
533
534#ifdef SOCKBUF_DEBUG
535void
536sbcheck(struct sockbuf *sb)
537{
538 struct mbuf *m;
539 struct mbuf *n = 0;
540 u_long len = 0, mbcnt = 0;
541
542 SOCKBUF_LOCK_ASSERT(sb);
543
544 for (m = sb->sb_mb; m; m = n) {
545 n = m->m_nextpkt;
546 for (; m; m = m->m_next) {
547 len += m->m_len;
548 mbcnt += MSIZE;
549 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
550 mbcnt += m->m_ext.ext_size;
551 }
552 }
553 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
554 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
555 mbcnt, sb->sb_mbcnt);
556 panic("sbcheck");
557 }
558}
559#endif
560
561/*
562 * As above, except the mbuf chain begins a new record.
563 */
564void
565sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
566{
567 struct mbuf *m;
568
569 SOCKBUF_LOCK_ASSERT(sb);
570
571 if (m0 == 0)
572 return;
573 m = sb->sb_mb;
574 if (m)
575 while (m->m_nextpkt)
576 m = m->m_nextpkt;
577 /*
578 * Put the first mbuf on the queue. Note this permits zero length
579 * records.
580 */
581 sballoc(sb, m0);
582 SBLASTRECORDCHK(sb);
583 SBLINKRECORD(sb, m0);
584 if (m)
585 m->m_nextpkt = m0;
586 else
587 sb->sb_mb = m0;
588 m = m0->m_next;
589 m0->m_next = 0;
590 if (m && (m0->m_flags & M_EOR)) {
591 m0->m_flags &= ~M_EOR;
592 m->m_flags |= M_EOR;
593 }
594 sbcompress(sb, m, m0);
595}
596
597/*
598 * As above, except the mbuf chain begins a new record.
599 */
600void
601sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
602{
603
604 SOCKBUF_LOCK(sb);
605 sbappendrecord_locked(sb, m0);
606 SOCKBUF_UNLOCK(sb);
607}
608
609/*
610 * Append address and data, and optionally, control (ancillary) data to the
611 * receive queue of a socket. If present, m0 must include a packet header
612 * with total length. Returns 0 if no space in sockbuf or insufficient
613 * mbufs.
614 */
615int
616sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
617 struct mbuf *m0, struct mbuf *control)
618{
619 struct mbuf *m, *n, *nlast;
620 int space = asa->sa_len;
621
622 SOCKBUF_LOCK_ASSERT(sb);
623
624 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
625 panic("sbappendaddr_locked");
626 if (m0)
627 space += m0->m_pkthdr.len;
628 space += m_length(control, &n);
629
630 if (space > sbspace(sb))
631 return (0);
632#if MSIZE <= 256
633 if (asa->sa_len > MLEN)
634 return (0);
635#endif
636 MGET(m, M_DONTWAIT, MT_SONAME);
637 if (m == 0)
638 return (0);
639 m->m_len = asa->sa_len;
640 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
641 if (n)
642 n->m_next = m0; /* concatenate data to control */
643 else
644 control = m0;
645 m->m_next = control;
646 for (n = m; n->m_next != NULL; n = n->m_next)
647 sballoc(sb, n);
648 sballoc(sb, n);
649 nlast = n;
650 SBLINKRECORD(sb, m);
651
652 sb->sb_mbtail = nlast;
653 SBLASTMBUFCHK(sb);
654
655 SBLASTRECORDCHK(sb);
656 return (1);
657}
658
659/*
660 * Append address and data, and optionally, control (ancillary) data to the
661 * receive queue of a socket. If present, m0 must include a packet header
662 * with total length. Returns 0 if no space in sockbuf or insufficient
663 * mbufs.
664 */
665int
666sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
667 struct mbuf *m0, struct mbuf *control)
668{
669 int retval;
670
671 SOCKBUF_LOCK(sb);
672 retval = sbappendaddr_locked(sb, asa, m0, control);
673 SOCKBUF_UNLOCK(sb);
674 return (retval);
675}
676
677int
678sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
679 struct mbuf *control)
680{
681 struct mbuf *m, *n, *mlast;
682 int space;
683
684 SOCKBUF_LOCK_ASSERT(sb);
685
686 if (control == 0)
687 panic("sbappendcontrol_locked");
688 space = m_length(control, &n) + m_length(m0, NULL);
689
690 if (space > sbspace(sb))
691 return (0);
692 n->m_next = m0; /* concatenate data to control */
693
694 SBLASTRECORDCHK(sb);
695
696 for (m = control; m->m_next; m = m->m_next)
697 sballoc(sb, m);
698 sballoc(sb, m);
699 mlast = m;
700 SBLINKRECORD(sb, control);
701
702 sb->sb_mbtail = mlast;
703 SBLASTMBUFCHK(sb);
704
705 SBLASTRECORDCHK(sb);
706 return (1);
707}
708
709int
710sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
711{
712 int retval;
713
714 SOCKBUF_LOCK(sb);
715 retval = sbappendcontrol_locked(sb, m0, control);
716 SOCKBUF_UNLOCK(sb);
717 return (retval);
718}
719
720/*
721 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
722 * (n). If (n) is NULL, the buffer is presumed empty.
723 *
724 * When the data is compressed, mbufs in the chain may be handled in one of
725 * three ways:
726 *
727 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
728 * record boundary, and no change in data type).
729 *
730 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
731 * an mbuf already in the socket buffer. This can occur if an
732 * appropriate mbuf exists, there is room, and no merging of data types
733 * will occur.
734 *
735 * (3) The mbuf may be appended to the end of the existing mbuf chain.
736 *
737 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
738 * end-of-record.
739 */
740void
741sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
742{
743 int eor = 0;
744 struct mbuf *o;
745
746 SOCKBUF_LOCK_ASSERT(sb);
747
748 while (m) {
749 eor |= m->m_flags & M_EOR;
750 if (m->m_len == 0 &&
751 (eor == 0 ||
752 (((o = m->m_next) || (o = n)) &&
753 o->m_type == m->m_type))) {
754 if (sb->sb_lastrecord == m)
755 sb->sb_lastrecord = m->m_next;
756 m = m_free(m);
757 continue;
758 }
759 if (n && (n->m_flags & M_EOR) == 0 &&
760 M_WRITABLE(n) &&
761 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
762 m->m_len <= M_TRAILINGSPACE(n) &&
763 n->m_type == m->m_type) {
764 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
765 (unsigned)m->m_len);
766 n->m_len += m->m_len;
767 sb->sb_cc += m->m_len;
768 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
769 /* XXX: Probably don't need.*/
770 sb->sb_ctl += m->m_len;
771 m = m_free(m);
772 continue;
773 }
774 if (n)
775 n->m_next = m;
776 else
777 sb->sb_mb = m;
778 sb->sb_mbtail = m;
779 sballoc(sb, m);
780 n = m;
781 m->m_flags &= ~M_EOR;
782 m = m->m_next;
783 n->m_next = 0;
784 }
785 if (eor) {
786 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
787 n->m_flags |= eor;
788 }
789 SBLASTMBUFCHK(sb);
790}
791
792/*
793 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
794 */
795static void
796sbflush_internal(struct sockbuf *sb)
797{
798
799 while (sb->sb_mbcnt) {
800 /*
801 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
802 * we would loop forever. Panic instead.
803 */
804 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
805 break;
806 sbdrop_internal(sb, (int)sb->sb_cc);
807 }
808 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
809 panic("sbflush_internal: cc %u || mb %p || mbcnt %u",
810 sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
811}
812
813void
814sbflush_locked(struct sockbuf *sb)
815{
816
817 SOCKBUF_LOCK_ASSERT(sb);
818 sbflush_internal(sb);
819}
820
821void
822sbflush(struct sockbuf *sb)
823{
824
825 SOCKBUF_LOCK(sb);
826 sbflush_locked(sb);
827 SOCKBUF_UNLOCK(sb);
828}
829
830/*
831 * Drop data from (the front of) a sockbuf.
832 */
833static void
834sbdrop_internal(struct sockbuf *sb, int len)
835{
836 struct mbuf *m;
837 struct mbuf *next;
838
839 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
840 while (len > 0) {
841 if (m == 0) {
842 if (next == 0)
843 panic("sbdrop");
844 m = next;
845 next = m->m_nextpkt;
846 continue;
847 }
848 if (m->m_len > len) {
849 m->m_len -= len;
850 m->m_data += len;
851 sb->sb_cc -= len;
852 if (sb->sb_sndptroff != 0)
853 sb->sb_sndptroff -= len;
854 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
855 sb->sb_ctl -= len;
856 break;
857 }
858 len -= m->m_len;
859 sbfree(sb, m);
860 m = m_free(m);
861 }
862 while (m && m->m_len == 0) {
863 sbfree(sb, m);
864 m = m_free(m);
865 }
866 if (m) {
867 sb->sb_mb = m;
868 m->m_nextpkt = next;
869 } else
870 sb->sb_mb = next;
871 /*
872 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
873 * sb_lastrecord is up-to-date if we dropped part of the last record.
874 */
875 m = sb->sb_mb;
876 if (m == NULL) {
877 sb->sb_mbtail = NULL;
878 sb->sb_lastrecord = NULL;
879 } else if (m->m_nextpkt == NULL) {
880 sb->sb_lastrecord = m;
881 }
882}
883
884/*
885 * Drop data from (the front of) a sockbuf.
886 */
887void
888sbdrop_locked(struct sockbuf *sb, int len)
889{
890
891 SOCKBUF_LOCK_ASSERT(sb);
892
893 sbdrop_internal(sb, len);
894}
895
896void
897sbdrop(struct sockbuf *sb, int len)
898{
899
900 SOCKBUF_LOCK(sb);
901 sbdrop_locked(sb, len);
902 SOCKBUF_UNLOCK(sb);
903}
904
| 352/*
353 * Routines to add and remove data from an mbuf queue.
354 *
355 * The routines sbappend() or sbappendrecord() are normally called to append
356 * new mbufs to a socket buffer, after checking that adequate space is
357 * available, comparing the function sbspace() with the amount of data to be
358 * added. sbappendrecord() differs from sbappend() in that data supplied is
359 * treated as the beginning of a new record. To place a sender's address,
360 * optional access rights, and data in a socket receive buffer,
361 * sbappendaddr() should be used. To place access rights and data in a
362 * socket receive buffer, sbappendrights() should be used. In either case,
363 * the new data begins a new record. Note that unlike sbappend() and
364 * sbappendrecord(), these routines check for the caller that there will be
365 * enough space to store the data. Each fails if there is not enough space,
366 * or if it cannot find mbufs to store additional information in.
367 *
368 * Reliable protocols may use the socket send buffer to hold data awaiting
369 * acknowledgement. Data is normally copied from a socket send buffer in a
370 * protocol with m_copy for output to a peer, and then removing the data from
371 * the socket buffer with sbdrop() or sbdroprecord() when the data is
372 * acknowledged by the peer.
373 */
374#ifdef SOCKBUF_DEBUG
375void
376sblastrecordchk(struct sockbuf *sb, const char *file, int line)
377{
378 struct mbuf *m = sb->sb_mb;
379
380 SOCKBUF_LOCK_ASSERT(sb);
381
382 while (m && m->m_nextpkt)
383 m = m->m_nextpkt;
384
385 if (m != sb->sb_lastrecord) {
386 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
387 __func__, sb->sb_mb, sb->sb_lastrecord, m);
388 printf("packet chain:\n");
389 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
390 printf("\t%p\n", m);
391 panic("%s from %s:%u", __func__, file, line);
392 }
393}
394
395void
396sblastmbufchk(struct sockbuf *sb, const char *file, int line)
397{
398 struct mbuf *m = sb->sb_mb;
399 struct mbuf *n;
400
401 SOCKBUF_LOCK_ASSERT(sb);
402
403 while (m && m->m_nextpkt)
404 m = m->m_nextpkt;
405
406 while (m && m->m_next)
407 m = m->m_next;
408
409 if (m != sb->sb_mbtail) {
410 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
411 __func__, sb->sb_mb, sb->sb_mbtail, m);
412 printf("packet tree:\n");
413 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
414 printf("\t");
415 for (n = m; n != NULL; n = n->m_next)
416 printf("%p ", n);
417 printf("\n");
418 }
419 panic("%s from %s:%u", __func__, file, line);
420 }
421}
422#endif /* SOCKBUF_DEBUG */
423
424#define SBLINKRECORD(sb, m0) do { \
425 SOCKBUF_LOCK_ASSERT(sb); \
426 if ((sb)->sb_lastrecord != NULL) \
427 (sb)->sb_lastrecord->m_nextpkt = (m0); \
428 else \
429 (sb)->sb_mb = (m0); \
430 (sb)->sb_lastrecord = (m0); \
431} while (/*CONSTCOND*/0)
432
433/*
434 * Append mbuf chain m to the last record in the socket buffer sb. The
435 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
436 * are discarded and mbufs are compacted where possible.
437 */
438void
439sbappend_locked(struct sockbuf *sb, struct mbuf *m)
440{
441 struct mbuf *n;
442
443 SOCKBUF_LOCK_ASSERT(sb);
444
445 if (m == 0)
446 return;
447
448 SBLASTRECORDCHK(sb);
449 n = sb->sb_mb;
450 if (n) {
451 while (n->m_nextpkt)
452 n = n->m_nextpkt;
453 do {
454 if (n->m_flags & M_EOR) {
455 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
456 return;
457 }
458 } while (n->m_next && (n = n->m_next));
459 } else {
460 /*
461 * XXX Would like to simply use sb_mbtail here, but
462 * XXX I need to verify that I won't miss an EOR that
463 * XXX way.
464 */
465 if ((n = sb->sb_lastrecord) != NULL) {
466 do {
467 if (n->m_flags & M_EOR) {
468 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
469 return;
470 }
471 } while (n->m_next && (n = n->m_next));
472 } else {
473 /*
474 * If this is the first record in the socket buffer,
475 * it's also the last record.
476 */
477 sb->sb_lastrecord = m;
478 }
479 }
480 sbcompress(sb, m, n);
481 SBLASTRECORDCHK(sb);
482}
483
484/*
485 * Append mbuf chain m to the last record in the socket buffer sb. The
486 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
487 * are discarded and mbufs are compacted where possible.
488 */
489void
490sbappend(struct sockbuf *sb, struct mbuf *m)
491{
492
493 SOCKBUF_LOCK(sb);
494 sbappend_locked(sb, m);
495 SOCKBUF_UNLOCK(sb);
496}
497
498/*
499 * This version of sbappend() should only be used when the caller absolutely
500 * knows that there will never be more than one record in the socket buffer,
501 * that is, a stream protocol (such as TCP).
502 */
503void
504sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
505{
506 SOCKBUF_LOCK_ASSERT(sb);
507
508 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
509 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
510
511 SBLASTMBUFCHK(sb);
512
513 sbcompress(sb, m, sb->sb_mbtail);
514
515 sb->sb_lastrecord = sb->sb_mb;
516 SBLASTRECORDCHK(sb);
517}
518
519/*
520 * This version of sbappend() should only be used when the caller absolutely
521 * knows that there will never be more than one record in the socket buffer,
522 * that is, a stream protocol (such as TCP).
523 */
524void
525sbappendstream(struct sockbuf *sb, struct mbuf *m)
526{
527
528 SOCKBUF_LOCK(sb);
529 sbappendstream_locked(sb, m);
530 SOCKBUF_UNLOCK(sb);
531}
532
533#ifdef SOCKBUF_DEBUG
534void
535sbcheck(struct sockbuf *sb)
536{
537 struct mbuf *m;
538 struct mbuf *n = 0;
539 u_long len = 0, mbcnt = 0;
540
541 SOCKBUF_LOCK_ASSERT(sb);
542
543 for (m = sb->sb_mb; m; m = n) {
544 n = m->m_nextpkt;
545 for (; m; m = m->m_next) {
546 len += m->m_len;
547 mbcnt += MSIZE;
548 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
549 mbcnt += m->m_ext.ext_size;
550 }
551 }
552 if (len != sb->sb_cc || mbcnt != sb->sb_mbcnt) {
553 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
554 mbcnt, sb->sb_mbcnt);
555 panic("sbcheck");
556 }
557}
558#endif
559
560/*
561 * As above, except the mbuf chain begins a new record.
562 */
563void
564sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
565{
566 struct mbuf *m;
567
568 SOCKBUF_LOCK_ASSERT(sb);
569
570 if (m0 == 0)
571 return;
572 m = sb->sb_mb;
573 if (m)
574 while (m->m_nextpkt)
575 m = m->m_nextpkt;
576 /*
577 * Put the first mbuf on the queue. Note this permits zero length
578 * records.
579 */
580 sballoc(sb, m0);
581 SBLASTRECORDCHK(sb);
582 SBLINKRECORD(sb, m0);
583 if (m)
584 m->m_nextpkt = m0;
585 else
586 sb->sb_mb = m0;
587 m = m0->m_next;
588 m0->m_next = 0;
589 if (m && (m0->m_flags & M_EOR)) {
590 m0->m_flags &= ~M_EOR;
591 m->m_flags |= M_EOR;
592 }
593 sbcompress(sb, m, m0);
594}
595
596/*
597 * As above, except the mbuf chain begins a new record.
598 */
599void
600sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
601{
602
603 SOCKBUF_LOCK(sb);
604 sbappendrecord_locked(sb, m0);
605 SOCKBUF_UNLOCK(sb);
606}
607
608/*
609 * Append address and data, and optionally, control (ancillary) data to the
610 * receive queue of a socket. If present, m0 must include a packet header
611 * with total length. Returns 0 if no space in sockbuf or insufficient
612 * mbufs.
613 */
614int
615sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
616 struct mbuf *m0, struct mbuf *control)
617{
618 struct mbuf *m, *n, *nlast;
619 int space = asa->sa_len;
620
621 SOCKBUF_LOCK_ASSERT(sb);
622
623 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
624 panic("sbappendaddr_locked");
625 if (m0)
626 space += m0->m_pkthdr.len;
627 space += m_length(control, &n);
628
629 if (space > sbspace(sb))
630 return (0);
631#if MSIZE <= 256
632 if (asa->sa_len > MLEN)
633 return (0);
634#endif
635 MGET(m, M_DONTWAIT, MT_SONAME);
636 if (m == 0)
637 return (0);
638 m->m_len = asa->sa_len;
639 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
640 if (n)
641 n->m_next = m0; /* concatenate data to control */
642 else
643 control = m0;
644 m->m_next = control;
645 for (n = m; n->m_next != NULL; n = n->m_next)
646 sballoc(sb, n);
647 sballoc(sb, n);
648 nlast = n;
649 SBLINKRECORD(sb, m);
650
651 sb->sb_mbtail = nlast;
652 SBLASTMBUFCHK(sb);
653
654 SBLASTRECORDCHK(sb);
655 return (1);
656}
657
658/*
659 * Append address and data, and optionally, control (ancillary) data to the
660 * receive queue of a socket. If present, m0 must include a packet header
661 * with total length. Returns 0 if no space in sockbuf or insufficient
662 * mbufs.
663 */
664int
665sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
666 struct mbuf *m0, struct mbuf *control)
667{
668 int retval;
669
670 SOCKBUF_LOCK(sb);
671 retval = sbappendaddr_locked(sb, asa, m0, control);
672 SOCKBUF_UNLOCK(sb);
673 return (retval);
674}
675
676int
677sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
678 struct mbuf *control)
679{
680 struct mbuf *m, *n, *mlast;
681 int space;
682
683 SOCKBUF_LOCK_ASSERT(sb);
684
685 if (control == 0)
686 panic("sbappendcontrol_locked");
687 space = m_length(control, &n) + m_length(m0, NULL);
688
689 if (space > sbspace(sb))
690 return (0);
691 n->m_next = m0; /* concatenate data to control */
692
693 SBLASTRECORDCHK(sb);
694
695 for (m = control; m->m_next; m = m->m_next)
696 sballoc(sb, m);
697 sballoc(sb, m);
698 mlast = m;
699 SBLINKRECORD(sb, control);
700
701 sb->sb_mbtail = mlast;
702 SBLASTMBUFCHK(sb);
703
704 SBLASTRECORDCHK(sb);
705 return (1);
706}
707
708int
709sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
710{
711 int retval;
712
713 SOCKBUF_LOCK(sb);
714 retval = sbappendcontrol_locked(sb, m0, control);
715 SOCKBUF_UNLOCK(sb);
716 return (retval);
717}
718
719/*
720 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
721 * (n). If (n) is NULL, the buffer is presumed empty.
722 *
723 * When the data is compressed, mbufs in the chain may be handled in one of
724 * three ways:
725 *
726 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
727 * record boundary, and no change in data type).
728 *
729 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
730 * an mbuf already in the socket buffer. This can occur if an
731 * appropriate mbuf exists, there is room, and no merging of data types
732 * will occur.
733 *
734 * (3) The mbuf may be appended to the end of the existing mbuf chain.
735 *
736 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
737 * end-of-record.
738 */
739void
740sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
741{
742 int eor = 0;
743 struct mbuf *o;
744
745 SOCKBUF_LOCK_ASSERT(sb);
746
747 while (m) {
748 eor |= m->m_flags & M_EOR;
749 if (m->m_len == 0 &&
750 (eor == 0 ||
751 (((o = m->m_next) || (o = n)) &&
752 o->m_type == m->m_type))) {
753 if (sb->sb_lastrecord == m)
754 sb->sb_lastrecord = m->m_next;
755 m = m_free(m);
756 continue;
757 }
758 if (n && (n->m_flags & M_EOR) == 0 &&
759 M_WRITABLE(n) &&
760 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
761 m->m_len <= M_TRAILINGSPACE(n) &&
762 n->m_type == m->m_type) {
763 bcopy(mtod(m, caddr_t), mtod(n, caddr_t) + n->m_len,
764 (unsigned)m->m_len);
765 n->m_len += m->m_len;
766 sb->sb_cc += m->m_len;
767 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
768 /* XXX: Probably don't need.*/
769 sb->sb_ctl += m->m_len;
770 m = m_free(m);
771 continue;
772 }
773 if (n)
774 n->m_next = m;
775 else
776 sb->sb_mb = m;
777 sb->sb_mbtail = m;
778 sballoc(sb, m);
779 n = m;
780 m->m_flags &= ~M_EOR;
781 m = m->m_next;
782 n->m_next = 0;
783 }
784 if (eor) {
785 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
786 n->m_flags |= eor;
787 }
788 SBLASTMBUFCHK(sb);
789}
790
791/*
792 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
793 */
794static void
795sbflush_internal(struct sockbuf *sb)
796{
797
798 while (sb->sb_mbcnt) {
799 /*
800 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
801 * we would loop forever. Panic instead.
802 */
803 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
804 break;
805 sbdrop_internal(sb, (int)sb->sb_cc);
806 }
807 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
808 panic("sbflush_internal: cc %u || mb %p || mbcnt %u",
809 sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
810}
811
812void
813sbflush_locked(struct sockbuf *sb)
814{
815
816 SOCKBUF_LOCK_ASSERT(sb);
817 sbflush_internal(sb);
818}
819
820void
821sbflush(struct sockbuf *sb)
822{
823
824 SOCKBUF_LOCK(sb);
825 sbflush_locked(sb);
826 SOCKBUF_UNLOCK(sb);
827}
828
829/*
830 * Drop data from (the front of) a sockbuf.
831 */
832static void
833sbdrop_internal(struct sockbuf *sb, int len)
834{
835 struct mbuf *m;
836 struct mbuf *next;
837
838 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
839 while (len > 0) {
840 if (m == 0) {
841 if (next == 0)
842 panic("sbdrop");
843 m = next;
844 next = m->m_nextpkt;
845 continue;
846 }
847 if (m->m_len > len) {
848 m->m_len -= len;
849 m->m_data += len;
850 sb->sb_cc -= len;
851 if (sb->sb_sndptroff != 0)
852 sb->sb_sndptroff -= len;
853 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
854 sb->sb_ctl -= len;
855 break;
856 }
857 len -= m->m_len;
858 sbfree(sb, m);
859 m = m_free(m);
860 }
861 while (m && m->m_len == 0) {
862 sbfree(sb, m);
863 m = m_free(m);
864 }
865 if (m) {
866 sb->sb_mb = m;
867 m->m_nextpkt = next;
868 } else
869 sb->sb_mb = next;
870 /*
871 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
872 * sb_lastrecord is up-to-date if we dropped part of the last record.
873 */
874 m = sb->sb_mb;
875 if (m == NULL) {
876 sb->sb_mbtail = NULL;
877 sb->sb_lastrecord = NULL;
878 } else if (m->m_nextpkt == NULL) {
879 sb->sb_lastrecord = m;
880 }
881}
882
883/*
884 * Drop data from (the front of) a sockbuf.
885 */
886void
887sbdrop_locked(struct sockbuf *sb, int len)
888{
889
890 SOCKBUF_LOCK_ASSERT(sb);
891
892 sbdrop_internal(sb, len);
893}
894
895void
896sbdrop(struct sockbuf *sb, int len)
897{
898
899 SOCKBUF_LOCK(sb);
900 sbdrop_locked(sb, len);
901 SOCKBUF_UNLOCK(sb);
902}
903
|
905
| |
906/*
907 * Maintain a pointer and offset pair into the socket buffer mbuf chain to
908 * avoid traversal of the entire socket buffer for larger offsets.
909 */
910struct mbuf *
911sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff)
912{
913 struct mbuf *m, *ret;
914
915 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
916 KASSERT(off + len <= sb->sb_cc, ("%s: beyond sb", __func__));
917 KASSERT(sb->sb_sndptroff <= sb->sb_cc, ("%s: sndptroff broken", __func__));
918
919 /*
920 * Is off below stored offset? Happens on retransmits.
921 * Just return, we can't help here.
922 */
923 if (sb->sb_sndptroff > off) {
924 *moff = off;
925 return (sb->sb_mb);
926 }
927
928 /* Return closest mbuf in chain for current offset. */
929 *moff = off - sb->sb_sndptroff;
930 m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb;
931
932 /* Advance by len to be as close as possible for the next transmit. */
933 for (off = off - sb->sb_sndptroff + len - 1;
934 off > 0 && off >= m->m_len;
935 m = m->m_next) {
936 sb->sb_sndptroff += m->m_len;
937 off -= m->m_len;
938 }
939 sb->sb_sndptr = m;
940
941 return (ret);
942}
943
944/*
945 * Drop a record off the front of a sockbuf and move the next record to the
946 * front.
947 */
948void
949sbdroprecord_locked(struct sockbuf *sb)
950{
951 struct mbuf *m;
952
953 SOCKBUF_LOCK_ASSERT(sb);
954
955 m = sb->sb_mb;
956 if (m) {
957 sb->sb_mb = m->m_nextpkt;
958 do {
959 sbfree(sb, m);
960 m = m_free(m);
961 } while (m);
962 }
963 SB_EMPTY_FIXUP(sb);
964}
965
966/*
967 * Drop a record off the front of a sockbuf and move the next record to the
968 * front.
969 */
970void
971sbdroprecord(struct sockbuf *sb)
972{
973
974 SOCKBUF_LOCK(sb);
975 sbdroprecord_locked(sb);
976 SOCKBUF_UNLOCK(sb);
977}
978
979/*
980 * Create a "control" mbuf containing the specified data with the specified
981 * type for presentation on a socket buffer.
982 */
983struct mbuf *
| 904/*
905 * Maintain a pointer and offset pair into the socket buffer mbuf chain to
906 * avoid traversal of the entire socket buffer for larger offsets.
907 */
908struct mbuf *
909sbsndptr(struct sockbuf *sb, u_int off, u_int len, u_int *moff)
910{
911 struct mbuf *m, *ret;
912
913 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
914 KASSERT(off + len <= sb->sb_cc, ("%s: beyond sb", __func__));
915 KASSERT(sb->sb_sndptroff <= sb->sb_cc, ("%s: sndptroff broken", __func__));
916
917 /*
918 * Is off below stored offset? Happens on retransmits.
919 * Just return, we can't help here.
920 */
921 if (sb->sb_sndptroff > off) {
922 *moff = off;
923 return (sb->sb_mb);
924 }
925
926 /* Return closest mbuf in chain for current offset. */
927 *moff = off - sb->sb_sndptroff;
928 m = ret = sb->sb_sndptr ? sb->sb_sndptr : sb->sb_mb;
929
930 /* Advance by len to be as close as possible for the next transmit. */
931 for (off = off - sb->sb_sndptroff + len - 1;
932 off > 0 && off >= m->m_len;
933 m = m->m_next) {
934 sb->sb_sndptroff += m->m_len;
935 off -= m->m_len;
936 }
937 sb->sb_sndptr = m;
938
939 return (ret);
940}
941
942/*
943 * Drop a record off the front of a sockbuf and move the next record to the
944 * front.
945 */
946void
947sbdroprecord_locked(struct sockbuf *sb)
948{
949 struct mbuf *m;
950
951 SOCKBUF_LOCK_ASSERT(sb);
952
953 m = sb->sb_mb;
954 if (m) {
955 sb->sb_mb = m->m_nextpkt;
956 do {
957 sbfree(sb, m);
958 m = m_free(m);
959 } while (m);
960 }
961 SB_EMPTY_FIXUP(sb);
962}
963
964/*
965 * Drop a record off the front of a sockbuf and move the next record to the
966 * front.
967 */
968void
969sbdroprecord(struct sockbuf *sb)
970{
971
972 SOCKBUF_LOCK(sb);
973 sbdroprecord_locked(sb);
974 SOCKBUF_UNLOCK(sb);
975}
976
977/*
978 * Create a "control" mbuf containing the specified data with the specified
979 * type for presentation on a socket buffer.
980 */
981struct mbuf *
|
984sbcreatecontrol(p, size, type, level)
985 caddr_t p;
986 register int size;
987 int type, level;
| 982sbcreatecontrol(caddr_t p, int size, int type, int level)
|
988{
| 983{
|
989 register struct cmsghdr *cp;
| 984 struct cmsghdr *cp;
|
990 struct mbuf *m;
991
992 if (CMSG_SPACE((u_int)size) > MCLBYTES)
993 return ((struct mbuf *) NULL);
994 if (CMSG_SPACE((u_int)size) > MLEN)
995 m = m_getcl(M_DONTWAIT, MT_CONTROL, 0);
996 else
997 m = m_get(M_DONTWAIT, MT_CONTROL);
998 if (m == NULL)
999 return ((struct mbuf *) NULL);
1000 cp = mtod(m, struct cmsghdr *);
1001 m->m_len = 0;
1002 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
1003 ("sbcreatecontrol: short mbuf"));
1004 if (p != NULL)
1005 (void)memcpy(CMSG_DATA(cp), p, size);
1006 m->m_len = CMSG_SPACE(size);
1007 cp->cmsg_len = CMSG_LEN(size);
1008 cp->cmsg_level = level;
1009 cp->cmsg_type = type;
1010 return (m);
1011}
1012
1013/*
1014 * This does the same for socket buffers that sotoxsocket does for sockets:
1015 * generate an user-format data structure describing the socket buffer. Note
1016 * that the xsockbuf structure, since it is always embedded in a socket, does
1017 * not include a self pointer nor a length. We make this entry point public
1018 * in case some other mechanism needs it.
1019 */
1020void
1021sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1022{
| 985 struct mbuf *m;
986
987 if (CMSG_SPACE((u_int)size) > MCLBYTES)
988 return ((struct mbuf *) NULL);
989 if (CMSG_SPACE((u_int)size) > MLEN)
990 m = m_getcl(M_DONTWAIT, MT_CONTROL, 0);
991 else
992 m = m_get(M_DONTWAIT, MT_CONTROL);
993 if (m == NULL)
994 return ((struct mbuf *) NULL);
995 cp = mtod(m, struct cmsghdr *);
996 m->m_len = 0;
997 KASSERT(CMSG_SPACE((u_int)size) <= M_TRAILINGSPACE(m),
998 ("sbcreatecontrol: short mbuf"));
999 if (p != NULL)
1000 (void)memcpy(CMSG_DATA(cp), p, size);
1001 m->m_len = CMSG_SPACE(size);
1002 cp->cmsg_len = CMSG_LEN(size);
1003 cp->cmsg_level = level;
1004 cp->cmsg_type = type;
1005 return (m);
1006}
1007
1008/*
1009 * This does the same for socket buffers that sotoxsocket does for sockets:
1010 * generate an user-format data structure describing the socket buffer. Note
1011 * that the xsockbuf structure, since it is always embedded in a socket, does
1012 * not include a self pointer nor a length. We make this entry point public
1013 * in case some other mechanism needs it.
1014 */
1015void
1016sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1017{
|
| 1018
|
1023 xsb->sb_cc = sb->sb_cc;
1024 xsb->sb_hiwat = sb->sb_hiwat;
1025 xsb->sb_mbcnt = sb->sb_mbcnt;
1026 xsb->sb_mbmax = sb->sb_mbmax;
1027 xsb->sb_lowat = sb->sb_lowat;
1028 xsb->sb_flags = sb->sb_flags;
1029 xsb->sb_timeo = sb->sb_timeo;
1030}
1031
1032/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1033static int dummy;
1034SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1035SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1036 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1037SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1038 &sb_efficiency, 0, "");
| 1019 xsb->sb_cc = sb->sb_cc;
1020 xsb->sb_hiwat = sb->sb_hiwat;
1021 xsb->sb_mbcnt = sb->sb_mbcnt;
1022 xsb->sb_mbmax = sb->sb_mbmax;
1023 xsb->sb_lowat = sb->sb_lowat;
1024 xsb->sb_flags = sb->sb_flags;
1025 xsb->sb_timeo = sb->sb_timeo;
1026}
1027
1028/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1029static int dummy;
1030SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
1031SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
1032 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
1033SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1034 &sb_efficiency, 0, "");
|