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
--- 16 unchanged lines hidden (view full) ---
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 160915 2006-08-02 13:01:58Z 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>
--- 17 unchanged lines hidden (view full) ---
59 */
60
61u_long sb_max = SB_MAX;
62static u_long sb_max_adj =
63 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
64
65static u_long sb_efficiency = 8; /* parameter for sbreserve() */
66
67static void sbdrop_internal(struct sockbuf *sb, int len);
68static void sbflush_internal(struct sockbuf *sb);
69static void sbrelease_internal(struct sockbuf *sb, struct socket *so);
70
71/*
72 * Socantsendmore indicates that no more data will be sent on the socket; it
73 * would normally be applied to a socket when the user informs the system
74 * that no more data is to be sent, by the protocol code (in case
75 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
76 * received, and will normally be applied to the socket by a protocol when it
77 * detects that the peer will send no more data. Data queued for reading in
78 * the socket may yet be read.
79 */
80void
81socantsendmore_locked(struct socket *so)
82{
83
84 SOCKBUF_LOCK_ASSERT(&so->so_snd);
85
86 so->so_snd.sb_state |= SBS_CANTSENDMORE;
87 sowwakeup_locked(so);
88 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
89}
90
91void
92socantsendmore(struct socket *so)
93{
94
95 SOCKBUF_LOCK(&so->so_snd);
96 socantsendmore_locked(so);
97 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
98}
99
100void
101socantrcvmore_locked(struct socket *so)
102{
103
104 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
105
106 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
107 sorwakeup_locked(so);
108 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
109}
110
111void
112socantrcvmore(struct socket *so)
113{
114
115 SOCKBUF_LOCK(&so->so_rcv);
116 socantrcvmore_locked(so);
117 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
118}
119
120/*
121 * Wait for data to arrive at/drain from a socket buffer.
122 */
123int
124sbwait(struct sockbuf *sb)
125{
126
127 SOCKBUF_LOCK_ASSERT(sb);
128
129 sb->sb_flags |= SB_WAIT;
130 return (msleep(&sb->sb_cc, &sb->sb_mtx,
131 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
132 sb->sb_timeo));
133}
134
135/*
136 * Lock a sockbuf already known to be locked; return any error returned from
137 * sleep (EINTR).
138 */
139int
140sb_lock(struct sockbuf *sb)
141{
142 int error;
143
144 SOCKBUF_LOCK_ASSERT(sb);
145
146 while (sb->sb_flags & SB_LOCK) {
147 sb->sb_flags |= SB_WANT;
148 error = msleep(&sb->sb_flags, &sb->sb_mtx,
149 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
150 "sblock", 0);
151 if (error)
152 return (error);
153 }
154 sb->sb_flags |= SB_LOCK;
155 return (0);
156}
157
158/*
159 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
160 * via SIGIO if the socket has the SS_ASYNC flag set.
161 *
162 * Called with the socket buffer lock held; will release the lock by the end
163 * of the function. This allows the caller to acquire the socket buffer lock
164 * while testing for the need for various sorts of wakeup and hold it through
165 * to the point where it's no longer required. We currently hold the lock
166 * through calls out to other subsystems (with the exception of kqueue), and
167 * then release it to avoid lock order issues. It's not clear that's
168 * correct.
169 */
170void
171sowakeup(struct socket *so, struct sockbuf *sb)
172{
173
174 SOCKBUF_LOCK_ASSERT(sb);
175
176 selwakeuppri(&sb->sb_sel, PSOCK);
177 sb->sb_flags &= ~SB_SEL;
178 if (sb->sb_flags & SB_WAIT) {
179 sb->sb_flags &= ~SB_WAIT;
--- 8 unchanged lines hidden (view full) ---
188 if (sb->sb_flags & SB_AIO)
189 aio_swake(so, sb);
190 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
191}
192
193/*
194 * Socket buffer (struct sockbuf) utility routines.
195 *
196 * Each socket contains two socket buffers: one for sending data and one for
197 * receiving data. Each buffer contains a queue of mbufs, information about
198 * the number of mbufs and amount of data in the queue, and other fields
199 * allowing select() statements and notification on data availability to be
200 * implemented.
201 *
202 * Data stored in a socket buffer is maintained as a list of records. Each
203 * record is a list of mbufs chained together with the m_next field. Records
204 * are chained together with the m_nextpkt field. The upper level routine
205 * soreceive() expects the following conventions to be observed when placing
206 * information in the receive buffer:
207 *
208 * 1. If the protocol requires each message be preceded by the sender's name,
209 * then a record containing that name must be present before any
210 * associated data (mbuf's must be of type MT_SONAME).
211 * 2. If the protocol supports the exchange of ``access rights'' (really just
212 * additional data associated with the message), and there are ``rights''
213 * to be received, then a record containing this data should be present
214 * (mbuf's must be of type MT_RIGHTS).
215 * 3. If a name or rights record exists, then it must be followed by a data
216 * record, perhaps of zero length.
217 *
218 * Before using a new socket structure it is first necessary to reserve
219 * buffer space to the socket, by calling sbreserve(). This should commit
220 * some of the available buffer space in the system buffer pool for the
221 * socket (currently, it does nothing but enforce limits). The space should
222 * be released by calling sbrelease() when the socket is destroyed.
223 */
224int
225soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
226{
227 struct thread *td = curthread;
228
229 SOCKBUF_LOCK(&so->so_snd);
230 SOCKBUF_LOCK(&so->so_rcv);
231 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
232 goto bad;
233 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
--- 31 unchanged lines hidden (view full) ---
265 sb_max = old_sb_max;
266 return (EINVAL);
267 }
268 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
269 return (0);
270}
271
272/*
273 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
274 * become limiting if buffering efficiency is near the normal case.
275 */
276int
277sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
278 struct thread *td)
279{
280 rlim_t sbsize_limit;
281
282 SOCKBUF_LOCK_ASSERT(sb);
283
284 /*
285 * td will only be NULL when we're in an interrupt (e.g. in
286 * tcp_input()).
287 *
288 * XXXRW: This comment needs updating, as might the code.
289 */
290 if (cc > sb_max_adj)
291 return (0);
292 if (td != NULL) {
293 PROC_LOCK(td->td_proc);
294 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
295 PROC_UNLOCK(td->td_proc);
296 } else
297 sbsize_limit = RLIM_INFINITY;
298 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
299 sbsize_limit))
300 return (0);
301 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
302 if (sb->sb_lowat > sb->sb_hiwat)
303 sb->sb_lowat = sb->sb_hiwat;
304 return (1);
305}
306
307int
308sbreserve(struct sockbuf *sb, u_long cc, struct socket *so,
309 struct thread *td)
310{
311 int error;
312
313 SOCKBUF_LOCK(sb);
314 error = sbreserve_locked(sb, cc, so, td);
315 SOCKBUF_UNLOCK(sb);
316 return (error);
317}
318
319/*
320 * Free mbufs held by a socket, and reserved mbuf space.
321 */
322static void
323sbrelease_internal(struct sockbuf *sb, struct socket *so)
324{
325
326 sbflush_internal(sb);
327 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
328 RLIM_INFINITY);
329 sb->sb_mbmax = 0;
330}
331
332void
333sbrelease_locked(struct sockbuf *sb, struct socket *so)
334{
335
336 SOCKBUF_LOCK_ASSERT(sb);
337
338 sbrelease_internal(sb, so);
339}
340
341void
342sbrelease(struct sockbuf *sb, struct socket *so)
343{
344
345 SOCKBUF_LOCK(sb);
346 sbrelease_locked(sb, so);
347 SOCKBUF_UNLOCK(sb);
348}
349
350void
351sbdestroy(struct sockbuf *sb, struct socket *so)
352{
353
354 sbrelease_internal(sb, so);
355}
356
357
358/*
359 * Routines to add and remove data from an mbuf queue.
360 *
361 * The routines sbappend() or sbappendrecord() are normally called to append
362 * new mbufs to a socket buffer, after checking that adequate space is
363 * available, comparing the function sbspace() with the amount of data to be
364 * added. sbappendrecord() differs from sbappend() in that data supplied is
365 * treated as the beginning of a new record. To place a sender's address,
366 * optional access rights, and data in a socket receive buffer,
367 * sbappendaddr() should be used. To place access rights and data in a
368 * socket receive buffer, sbappendrights() should be used. In either case,
369 * the new data begins a new record. Note that unlike sbappend() and
370 * sbappendrecord(), these routines check for the caller that there will be
371 * enough space to store the data. Each fails if there is not enough space,
372 * or if it cannot find mbufs to store additional information in.
373 *
374 * Reliable protocols may use the socket send buffer to hold data awaiting
375 * acknowledgement. Data is normally copied from a socket send buffer in a
376 * protocol with m_copy for output to a peer, and then removing the data from
377 * the socket buffer with sbdrop() or sbdroprecord() when the data is
378 * acknowledged by the peer.
379 */
380#ifdef SOCKBUF_DEBUG
381void
382sblastrecordchk(struct sockbuf *sb, const char *file, int line)
383{
384 struct mbuf *m = sb->sb_mb;
385
386 SOCKBUF_LOCK_ASSERT(sb);
387
--- 44 unchanged lines hidden (view full) ---
432 if ((sb)->sb_lastrecord != NULL) \
433 (sb)->sb_lastrecord->m_nextpkt = (m0); \
434 else \
435 (sb)->sb_mb = (m0); \
436 (sb)->sb_lastrecord = (m0); \
437} while (/*CONSTCOND*/0)
438
439/*
440 * Append mbuf chain m to the last record in the socket buffer sb. The
441 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
442 * are discarded and mbufs are compacted where possible.
443 */
444void
445sbappend_locked(struct sockbuf *sb, struct mbuf *m)
446{
447 struct mbuf *n;
448
449 SOCKBUF_LOCK_ASSERT(sb);
450
451 if (m == 0)
452 return;
453
454 SBLASTRECORDCHK(sb);
455 n = sb->sb_mb;
--- 27 unchanged lines hidden (view full) ---
483 sb->sb_lastrecord = m;
484 }
485 }
486 sbcompress(sb, m, n);
487 SBLASTRECORDCHK(sb);
488}
489
490/*
491 * Append mbuf chain m to the last record in the socket buffer sb. The
492 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
493 * are discarded and mbufs are compacted where possible.
494 */
495void
496sbappend(struct sockbuf *sb, struct mbuf *m)
497{
498
499 SOCKBUF_LOCK(sb);
500 sbappend_locked(sb, m);
501 SOCKBUF_UNLOCK(sb);
502}
503
504/*
505 * This version of sbappend() should only be used when the caller absolutely
506 * knows that there will never be more than one record in the socket buffer,
507 * that is, a stream protocol (such as TCP).
508 */
509void
510sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
511{
512 SOCKBUF_LOCK_ASSERT(sb);
513
514 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
515 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
516
517 SBLASTMBUFCHK(sb);
518
519 sbcompress(sb, m, sb->sb_mbtail);
520
521 sb->sb_lastrecord = sb->sb_mb;
522 SBLASTRECORDCHK(sb);
523}
524
525/*
526 * This version of sbappend() should only be used when the caller absolutely
527 * knows that there will never be more than one record in the socket buffer,
528 * that is, a stream protocol (such as TCP).
529 */
530void
531sbappendstream(struct sockbuf *sb, struct mbuf *m)
532{
533
534 SOCKBUF_LOCK(sb);
535 sbappendstream_locked(sb, m);
536 SOCKBUF_UNLOCK(sb);
537}
538
539#ifdef SOCKBUF_DEBUG
540void
541sbcheck(struct sockbuf *sb)
542{
543 struct mbuf *m;
544 struct mbuf *n = 0;
545 u_long len = 0, mbcnt = 0;
546
547 SOCKBUF_LOCK_ASSERT(sb);
548
549 for (m = sb->sb_mb; m; m = n) {
--- 9 unchanged lines hidden (view full) ---
559 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
560 mbcnt, sb->sb_mbcnt);
561 panic("sbcheck");
562 }
563}
564#endif
565
566/*
567 * As above, except the mbuf chain begins a new record.
568 */
569void
570sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
571{
572 struct mbuf *m;
573
574 SOCKBUF_LOCK_ASSERT(sb);
575
576 if (m0 == 0)
577 return;
578 m = sb->sb_mb;
579 if (m)
580 while (m->m_nextpkt)
581 m = m->m_nextpkt;
582 /*
583 * Put the first mbuf on the queue. Note this permits zero length
584 * records.
585 */
586 sballoc(sb, m0);
587 SBLASTRECORDCHK(sb);
588 SBLINKRECORD(sb, m0);
589 if (m)
590 m->m_nextpkt = m0;
591 else
592 sb->sb_mb = m0;
593 m = m0->m_next;
594 m0->m_next = 0;
595 if (m && (m0->m_flags & M_EOR)) {
596 m0->m_flags &= ~M_EOR;
597 m->m_flags |= M_EOR;
598 }
599 sbcompress(sb, m, m0);
600}
601
602/*
603 * As above, except the mbuf chain begins a new record.
604 */
605void
606sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
607{
608
609 SOCKBUF_LOCK(sb);
610 sbappendrecord_locked(sb, m0);
611 SOCKBUF_UNLOCK(sb);
612}
613
614/*
615 * Append address and data, and optionally, control (ancillary) data to the
616 * receive queue of a socket. If present, m0 must include a packet header
617 * with total length. Returns 0 if no space in sockbuf or insufficient
618 * mbufs.
619 */
620int
621sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
622 struct mbuf *m0, struct mbuf *control)
623{
624 struct mbuf *m, *n, *nlast;
625 int space = asa->sa_len;
626
627 SOCKBUF_LOCK_ASSERT(sb);
628
629 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
630 panic("sbappendaddr_locked");
--- 26 unchanged lines hidden (view full) ---
657 sb->sb_mbtail = nlast;
658 SBLASTMBUFCHK(sb);
659
660 SBLASTRECORDCHK(sb);
661 return (1);
662}
663
664/*
665 * Append address and data, and optionally, control (ancillary) data to the
666 * receive queue of a socket. If present, m0 must include a packet header
667 * with total length. Returns 0 if no space in sockbuf or insufficient
668 * mbufs.
669 */
670int
671sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
672 struct mbuf *m0, struct mbuf *control)
673{
674 int retval;
675
676 SOCKBUF_LOCK(sb);
677 retval = sbappendaddr_locked(sb, asa, m0, control);
678 SOCKBUF_UNLOCK(sb);
679 return (retval);
680}
681
682int
683sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
684 struct mbuf *control)
685{
686 struct mbuf *m, *n, *mlast;
687 int space;
688
689 SOCKBUF_LOCK_ASSERT(sb);
690
691 if (control == 0)
692 panic("sbappendcontrol_locked");
--- 14 unchanged lines hidden (view full) ---
707 sb->sb_mbtail = mlast;
708 SBLASTMBUFCHK(sb);
709
710 SBLASTRECORDCHK(sb);
711 return (1);
712}
713
714int
715sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
716{
717 int retval;
718
719 SOCKBUF_LOCK(sb);
720 retval = sbappendcontrol_locked(sb, m0, control);
721 SOCKBUF_UNLOCK(sb);
722 return (retval);
723}
--- 14 unchanged lines hidden (view full) ---
738 * will occur.
739 *
740 * (3) The mbuf may be appended to the end of the existing mbuf chain.
741 *
742 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
743 * end-of-record.
744 */
745void
746sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
747{
748 int eor = 0;
749 struct mbuf *o;
750
751 SOCKBUF_LOCK_ASSERT(sb);
752
753 while (m) {
754 eor |= m->m_flags & M_EOR;
755 if (m->m_len == 0 &&
756 (eor == 0 ||
757 (((o = m->m_next) || (o = n)) &&
--- 32 unchanged lines hidden (view full) ---
790 if (eor) {
791 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
792 n->m_flags |= eor;
793 }
794 SBLASTMBUFCHK(sb);
795}
796
797/*
798 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
799 */
800static void
801sbflush_internal(struct sockbuf *sb)
802{
803
804 if (sb->sb_flags & SB_LOCK)
805 panic("sbflush_internal: locked");
806 while (sb->sb_mbcnt) {
807 /*
808 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
809 * we would loop forever. Panic instead.
810 */
811 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
812 break;
813 sbdrop_internal(sb, (int)sb->sb_cc);
814 }
815 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
816 panic("sbflush_internal: cc %u || mb %p || mbcnt %u",
817 sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
818}
819
820void
821sbflush_locked(struct sockbuf *sb)
822{
823
824 SOCKBUF_LOCK_ASSERT(sb);
825 sbflush_internal(sb);
826}
827
828void
829sbflush(struct sockbuf *sb)
830{
831
832 SOCKBUF_LOCK(sb);
833 sbflush_locked(sb);
834 SOCKBUF_UNLOCK(sb);
835}
836
837/*
838 * Drop data from (the front of) a sockbuf.
839 */
840static void
841sbdrop_internal(struct sockbuf *sb, int len)
842{
843 struct mbuf *m;
844 struct mbuf *next;
845
846 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
847 while (len > 0) {
848 if (m == 0) {
849 if (next == 0)
850 panic("sbdrop");
851 m = next;
--- 17 unchanged lines hidden (view full) ---
869 m = m_free(m);
870 }
871 if (m) {
872 sb->sb_mb = m;
873 m->m_nextpkt = next;
874 } else
875 sb->sb_mb = next;
876 /*
877 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
878 * sb_lastrecord is up-to-date if we dropped part of the last record.
879 */
880 m = sb->sb_mb;
881 if (m == NULL) {
882 sb->sb_mbtail = NULL;
883 sb->sb_lastrecord = NULL;
884 } else if (m->m_nextpkt == NULL) {
885 sb->sb_lastrecord = m;
886 }
887}
888
889/*
890 * Drop data from (the front of) a sockbuf.
891 */
892void
893sbdrop_locked(struct sockbuf *sb, int len)
894{
895
896 SOCKBUF_LOCK_ASSERT(sb);
897
898 sbdrop_internal(sb, len);
899}
900
901void
902sbdrop(struct sockbuf *sb, int len)
903{
904
905 SOCKBUF_LOCK(sb);
906 sbdrop_locked(sb, len);
907 SOCKBUF_UNLOCK(sb);
908}
909
910/*
911 * Drop a record off the front of a sockbuf and move the next record to the
912 * front.
913 */
914void
915sbdroprecord_locked(struct sockbuf *sb)
916{
917 struct mbuf *m;
918
919 SOCKBUF_LOCK_ASSERT(sb);
920
921 m = sb->sb_mb;
922 if (m) {
923 sb->sb_mb = m->m_nextpkt;
924 do {
925 sbfree(sb, m);
926 m = m_free(m);
927 } while (m);
928 }
929 SB_EMPTY_FIXUP(sb);
930}
931
932/*
933 * Drop a record off the front of a sockbuf and move the next record to the
934 * front.
935 */
936void
937sbdroprecord(struct sockbuf *sb)
938{
939
940 SOCKBUF_LOCK(sb);
941 sbdroprecord_locked(sb);
942 SOCKBUF_UNLOCK(sb);
943}
944
945/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
946static int dummy;
947SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
948SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
949 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
950SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
951 &sb_efficiency, 0, "");
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
--- 16 unchanged lines hidden (view full) ---
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 160915 2006-08-02 13:01:58Z 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>
--- 17 unchanged lines hidden (view full) ---
59 */
60
61u_long sb_max = SB_MAX;
62static u_long sb_max_adj =
63 SB_MAX * MCLBYTES / (MSIZE + MCLBYTES); /* adjusted sb_max */
64
65static u_long sb_efficiency = 8; /* parameter for sbreserve() */
66
67static void sbdrop_internal(struct sockbuf *sb, int len);
68static void sbflush_internal(struct sockbuf *sb);
69static void sbrelease_internal(struct sockbuf *sb, struct socket *so);
70
71/*
72 * Socantsendmore indicates that no more data will be sent on the socket; it
73 * would normally be applied to a socket when the user informs the system
74 * that no more data is to be sent, by the protocol code (in case
75 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
76 * received, and will normally be applied to the socket by a protocol when it
77 * detects that the peer will send no more data. Data queued for reading in
78 * the socket may yet be read.
79 */
80void
81socantsendmore_locked(struct socket *so)
82{
83
84 SOCKBUF_LOCK_ASSERT(&so->so_snd);
85
86 so->so_snd.sb_state |= SBS_CANTSENDMORE;
87 sowwakeup_locked(so);
88 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
89}
90
91void
92socantsendmore(struct socket *so)
93{
94
95 SOCKBUF_LOCK(&so->so_snd);
96 socantsendmore_locked(so);
97 mtx_assert(SOCKBUF_MTX(&so->so_snd), MA_NOTOWNED);
98}
99
100void
101socantrcvmore_locked(struct socket *so)
102{
103
104 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
105
106 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
107 sorwakeup_locked(so);
108 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
109}
110
111void
112socantrcvmore(struct socket *so)
113{
114
115 SOCKBUF_LOCK(&so->so_rcv);
116 socantrcvmore_locked(so);
117 mtx_assert(SOCKBUF_MTX(&so->so_rcv), MA_NOTOWNED);
118}
119
120/*
121 * Wait for data to arrive at/drain from a socket buffer.
122 */
123int
124sbwait(struct sockbuf *sb)
125{
126
127 SOCKBUF_LOCK_ASSERT(sb);
128
129 sb->sb_flags |= SB_WAIT;
130 return (msleep(&sb->sb_cc, &sb->sb_mtx,
131 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
132 sb->sb_timeo));
133}
134
135/*
136 * Lock a sockbuf already known to be locked; return any error returned from
137 * sleep (EINTR).
138 */
139int
140sb_lock(struct sockbuf *sb)
141{
142 int error;
143
144 SOCKBUF_LOCK_ASSERT(sb);
145
146 while (sb->sb_flags & SB_LOCK) {
147 sb->sb_flags |= SB_WANT;
148 error = msleep(&sb->sb_flags, &sb->sb_mtx,
149 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK|PCATCH,
150 "sblock", 0);
151 if (error)
152 return (error);
153 }
154 sb->sb_flags |= SB_LOCK;
155 return (0);
156}
157
158/*
159 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
160 * via SIGIO if the socket has the SS_ASYNC flag set.
161 *
162 * Called with the socket buffer lock held; will release the lock by the end
163 * of the function. This allows the caller to acquire the socket buffer lock
164 * while testing for the need for various sorts of wakeup and hold it through
165 * to the point where it's no longer required. We currently hold the lock
166 * through calls out to other subsystems (with the exception of kqueue), and
167 * then release it to avoid lock order issues. It's not clear that's
168 * correct.
169 */
170void
171sowakeup(struct socket *so, struct sockbuf *sb)
172{
173
174 SOCKBUF_LOCK_ASSERT(sb);
175
176 selwakeuppri(&sb->sb_sel, PSOCK);
177 sb->sb_flags &= ~SB_SEL;
178 if (sb->sb_flags & SB_WAIT) {
179 sb->sb_flags &= ~SB_WAIT;
--- 8 unchanged lines hidden (view full) ---
188 if (sb->sb_flags & SB_AIO)
189 aio_swake(so, sb);
190 mtx_assert(SOCKBUF_MTX(sb), MA_NOTOWNED);
191}
192
193/*
194 * Socket buffer (struct sockbuf) utility routines.
195 *
196 * Each socket contains two socket buffers: one for sending data and one for
197 * receiving data. Each buffer contains a queue of mbufs, information about
198 * the number of mbufs and amount of data in the queue, and other fields
199 * allowing select() statements and notification on data availability to be
200 * implemented.
201 *
202 * Data stored in a socket buffer is maintained as a list of records. Each
203 * record is a list of mbufs chained together with the m_next field. Records
204 * are chained together with the m_nextpkt field. The upper level routine
205 * soreceive() expects the following conventions to be observed when placing
206 * information in the receive buffer:
207 *
208 * 1. If the protocol requires each message be preceded by the sender's name,
209 * then a record containing that name must be present before any
210 * associated data (mbuf's must be of type MT_SONAME).
211 * 2. If the protocol supports the exchange of ``access rights'' (really just
212 * additional data associated with the message), and there are ``rights''
213 * to be received, then a record containing this data should be present
214 * (mbuf's must be of type MT_RIGHTS).
215 * 3. If a name or rights record exists, then it must be followed by a data
216 * record, perhaps of zero length.
217 *
218 * Before using a new socket structure it is first necessary to reserve
219 * buffer space to the socket, by calling sbreserve(). This should commit
220 * some of the available buffer space in the system buffer pool for the
221 * socket (currently, it does nothing but enforce limits). The space should
222 * be released by calling sbrelease() when the socket is destroyed.
223 */
224int
225soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
226{
227 struct thread *td = curthread;
228
229 SOCKBUF_LOCK(&so->so_snd);
230 SOCKBUF_LOCK(&so->so_rcv);
231 if (sbreserve_locked(&so->so_snd, sndcc, so, td) == 0)
232 goto bad;
233 if (sbreserve_locked(&so->so_rcv, rcvcc, so, td) == 0)
--- 31 unchanged lines hidden (view full) ---
265 sb_max = old_sb_max;
266 return (EINVAL);
267 }
268 sb_max_adj = (u_quad_t)sb_max * MCLBYTES / (MSIZE + MCLBYTES);
269 return (0);
270}
271
272/*
273 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
274 * become limiting if buffering efficiency is near the normal case.
275 */
276int
277sbreserve_locked(struct sockbuf *sb, u_long cc, struct socket *so,
278 struct thread *td)
279{
280 rlim_t sbsize_limit;
281
282 SOCKBUF_LOCK_ASSERT(sb);
283
284 /*
285 * td will only be NULL when we're in an interrupt (e.g. in
286 * tcp_input()).
287 *
288 * XXXRW: This comment needs updating, as might the code.
289 */
290 if (cc > sb_max_adj)
291 return (0);
292 if (td != NULL) {
293 PROC_LOCK(td->td_proc);
294 sbsize_limit = lim_cur(td->td_proc, RLIMIT_SBSIZE);
295 PROC_UNLOCK(td->td_proc);
296 } else
297 sbsize_limit = RLIM_INFINITY;
298 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
299 sbsize_limit))
300 return (0);
301 sb->sb_mbmax = min(cc * sb_efficiency, sb_max);
302 if (sb->sb_lowat > sb->sb_hiwat)
303 sb->sb_lowat = sb->sb_hiwat;
304 return (1);
305}
306
307int
308sbreserve(struct sockbuf *sb, u_long cc, struct socket *so,
309 struct thread *td)
310{
311 int error;
312
313 SOCKBUF_LOCK(sb);
314 error = sbreserve_locked(sb, cc, so, td);
315 SOCKBUF_UNLOCK(sb);
316 return (error);
317}
318
319/*
320 * Free mbufs held by a socket, and reserved mbuf space.
321 */
322static void
323sbrelease_internal(struct sockbuf *sb, struct socket *so)
324{
325
326 sbflush_internal(sb);
327 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
328 RLIM_INFINITY);
329 sb->sb_mbmax = 0;
330}
331
332void
333sbrelease_locked(struct sockbuf *sb, struct socket *so)
334{
335
336 SOCKBUF_LOCK_ASSERT(sb);
337
338 sbrelease_internal(sb, so);
339}
340
341void
342sbrelease(struct sockbuf *sb, struct socket *so)
343{
344
345 SOCKBUF_LOCK(sb);
346 sbrelease_locked(sb, so);
347 SOCKBUF_UNLOCK(sb);
348}
349
350void
351sbdestroy(struct sockbuf *sb, struct socket *so)
352{
353
354 sbrelease_internal(sb, so);
355}
356
357
358/*
359 * Routines to add and remove data from an mbuf queue.
360 *
361 * The routines sbappend() or sbappendrecord() are normally called to append
362 * new mbufs to a socket buffer, after checking that adequate space is
363 * available, comparing the function sbspace() with the amount of data to be
364 * added. sbappendrecord() differs from sbappend() in that data supplied is
365 * treated as the beginning of a new record. To place a sender's address,
366 * optional access rights, and data in a socket receive buffer,
367 * sbappendaddr() should be used. To place access rights and data in a
368 * socket receive buffer, sbappendrights() should be used. In either case,
369 * the new data begins a new record. Note that unlike sbappend() and
370 * sbappendrecord(), these routines check for the caller that there will be
371 * enough space to store the data. Each fails if there is not enough space,
372 * or if it cannot find mbufs to store additional information in.
373 *
374 * Reliable protocols may use the socket send buffer to hold data awaiting
375 * acknowledgement. Data is normally copied from a socket send buffer in a
376 * protocol with m_copy for output to a peer, and then removing the data from
377 * the socket buffer with sbdrop() or sbdroprecord() when the data is
378 * acknowledged by the peer.
379 */
380#ifdef SOCKBUF_DEBUG
381void
382sblastrecordchk(struct sockbuf *sb, const char *file, int line)
383{
384 struct mbuf *m = sb->sb_mb;
385
386 SOCKBUF_LOCK_ASSERT(sb);
387
--- 44 unchanged lines hidden (view full) ---
432 if ((sb)->sb_lastrecord != NULL) \
433 (sb)->sb_lastrecord->m_nextpkt = (m0); \
434 else \
435 (sb)->sb_mb = (m0); \
436 (sb)->sb_lastrecord = (m0); \
437} while (/*CONSTCOND*/0)
438
439/*
440 * Append mbuf chain m to the last record in the socket buffer sb. The
441 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
442 * are discarded and mbufs are compacted where possible.
443 */
444void
445sbappend_locked(struct sockbuf *sb, struct mbuf *m)
446{
447 struct mbuf *n;
448
449 SOCKBUF_LOCK_ASSERT(sb);
450
451 if (m == 0)
452 return;
453
454 SBLASTRECORDCHK(sb);
455 n = sb->sb_mb;
--- 27 unchanged lines hidden (view full) ---
483 sb->sb_lastrecord = m;
484 }
485 }
486 sbcompress(sb, m, n);
487 SBLASTRECORDCHK(sb);
488}
489
490/*
491 * Append mbuf chain m to the last record in the socket buffer sb. The
492 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
493 * are discarded and mbufs are compacted where possible.
494 */
495void
496sbappend(struct sockbuf *sb, struct mbuf *m)
497{
498
499 SOCKBUF_LOCK(sb);
500 sbappend_locked(sb, m);
501 SOCKBUF_UNLOCK(sb);
502}
503
504/*
505 * This version of sbappend() should only be used when the caller absolutely
506 * knows that there will never be more than one record in the socket buffer,
507 * that is, a stream protocol (such as TCP).
508 */
509void
510sbappendstream_locked(struct sockbuf *sb, struct mbuf *m)
511{
512 SOCKBUF_LOCK_ASSERT(sb);
513
514 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
515 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
516
517 SBLASTMBUFCHK(sb);
518
519 sbcompress(sb, m, sb->sb_mbtail);
520
521 sb->sb_lastrecord = sb->sb_mb;
522 SBLASTRECORDCHK(sb);
523}
524
525/*
526 * This version of sbappend() should only be used when the caller absolutely
527 * knows that there will never be more than one record in the socket buffer,
528 * that is, a stream protocol (such as TCP).
529 */
530void
531sbappendstream(struct sockbuf *sb, struct mbuf *m)
532{
533
534 SOCKBUF_LOCK(sb);
535 sbappendstream_locked(sb, m);
536 SOCKBUF_UNLOCK(sb);
537}
538
539#ifdef SOCKBUF_DEBUG
540void
541sbcheck(struct sockbuf *sb)
542{
543 struct mbuf *m;
544 struct mbuf *n = 0;
545 u_long len = 0, mbcnt = 0;
546
547 SOCKBUF_LOCK_ASSERT(sb);
548
549 for (m = sb->sb_mb; m; m = n) {
--- 9 unchanged lines hidden (view full) ---
559 printf("cc %ld != %u || mbcnt %ld != %u\n", len, sb->sb_cc,
560 mbcnt, sb->sb_mbcnt);
561 panic("sbcheck");
562 }
563}
564#endif
565
566/*
567 * As above, except the mbuf chain begins a new record.
568 */
569void
570sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
571{
572 struct mbuf *m;
573
574 SOCKBUF_LOCK_ASSERT(sb);
575
576 if (m0 == 0)
577 return;
578 m = sb->sb_mb;
579 if (m)
580 while (m->m_nextpkt)
581 m = m->m_nextpkt;
582 /*
583 * Put the first mbuf on the queue. Note this permits zero length
584 * records.
585 */
586 sballoc(sb, m0);
587 SBLASTRECORDCHK(sb);
588 SBLINKRECORD(sb, m0);
589 if (m)
590 m->m_nextpkt = m0;
591 else
592 sb->sb_mb = m0;
593 m = m0->m_next;
594 m0->m_next = 0;
595 if (m && (m0->m_flags & M_EOR)) {
596 m0->m_flags &= ~M_EOR;
597 m->m_flags |= M_EOR;
598 }
599 sbcompress(sb, m, m0);
600}
601
602/*
603 * As above, except the mbuf chain begins a new record.
604 */
605void
606sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
607{
608
609 SOCKBUF_LOCK(sb);
610 sbappendrecord_locked(sb, m0);
611 SOCKBUF_UNLOCK(sb);
612}
613
614/*
615 * Append address and data, and optionally, control (ancillary) data to the
616 * receive queue of a socket. If present, m0 must include a packet header
617 * with total length. Returns 0 if no space in sockbuf or insufficient
618 * mbufs.
619 */
620int
621sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
622 struct mbuf *m0, struct mbuf *control)
623{
624 struct mbuf *m, *n, *nlast;
625 int space = asa->sa_len;
626
627 SOCKBUF_LOCK_ASSERT(sb);
628
629 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
630 panic("sbappendaddr_locked");
--- 26 unchanged lines hidden (view full) ---
657 sb->sb_mbtail = nlast;
658 SBLASTMBUFCHK(sb);
659
660 SBLASTRECORDCHK(sb);
661 return (1);
662}
663
664/*
665 * Append address and data, and optionally, control (ancillary) data to the
666 * receive queue of a socket. If present, m0 must include a packet header
667 * with total length. Returns 0 if no space in sockbuf or insufficient
668 * mbufs.
669 */
670int
671sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
672 struct mbuf *m0, struct mbuf *control)
673{
674 int retval;
675
676 SOCKBUF_LOCK(sb);
677 retval = sbappendaddr_locked(sb, asa, m0, control);
678 SOCKBUF_UNLOCK(sb);
679 return (retval);
680}
681
682int
683sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
684 struct mbuf *control)
685{
686 struct mbuf *m, *n, *mlast;
687 int space;
688
689 SOCKBUF_LOCK_ASSERT(sb);
690
691 if (control == 0)
692 panic("sbappendcontrol_locked");
--- 14 unchanged lines hidden (view full) ---
707 sb->sb_mbtail = mlast;
708 SBLASTMBUFCHK(sb);
709
710 SBLASTRECORDCHK(sb);
711 return (1);
712}
713
714int
715sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control)
716{
717 int retval;
718
719 SOCKBUF_LOCK(sb);
720 retval = sbappendcontrol_locked(sb, m0, control);
721 SOCKBUF_UNLOCK(sb);
722 return (retval);
723}
--- 14 unchanged lines hidden (view full) ---
738 * will occur.
739 *
740 * (3) The mbuf may be appended to the end of the existing mbuf chain.
741 *
742 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
743 * end-of-record.
744 */
745void
746sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
747{
748 int eor = 0;
749 struct mbuf *o;
750
751 SOCKBUF_LOCK_ASSERT(sb);
752
753 while (m) {
754 eor |= m->m_flags & M_EOR;
755 if (m->m_len == 0 &&
756 (eor == 0 ||
757 (((o = m->m_next) || (o = n)) &&
--- 32 unchanged lines hidden (view full) ---
790 if (eor) {
791 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
792 n->m_flags |= eor;
793 }
794 SBLASTMBUFCHK(sb);
795}
796
797/*
798 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
799 */
800static void
801sbflush_internal(struct sockbuf *sb)
802{
803
804 if (sb->sb_flags & SB_LOCK)
805 panic("sbflush_internal: locked");
806 while (sb->sb_mbcnt) {
807 /*
808 * Don't call sbdrop(sb, 0) if the leading mbuf is non-empty:
809 * we would loop forever. Panic instead.
810 */
811 if (!sb->sb_cc && (sb->sb_mb == NULL || sb->sb_mb->m_len))
812 break;
813 sbdrop_internal(sb, (int)sb->sb_cc);
814 }
815 if (sb->sb_cc || sb->sb_mb || sb->sb_mbcnt)
816 panic("sbflush_internal: cc %u || mb %p || mbcnt %u",
817 sb->sb_cc, (void *)sb->sb_mb, sb->sb_mbcnt);
818}
819
820void
821sbflush_locked(struct sockbuf *sb)
822{
823
824 SOCKBUF_LOCK_ASSERT(sb);
825 sbflush_internal(sb);
826}
827
828void
829sbflush(struct sockbuf *sb)
830{
831
832 SOCKBUF_LOCK(sb);
833 sbflush_locked(sb);
834 SOCKBUF_UNLOCK(sb);
835}
836
837/*
838 * Drop data from (the front of) a sockbuf.
839 */
840static void
841sbdrop_internal(struct sockbuf *sb, int len)
842{
843 struct mbuf *m;
844 struct mbuf *next;
845
846 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
847 while (len > 0) {
848 if (m == 0) {
849 if (next == 0)
850 panic("sbdrop");
851 m = next;
--- 17 unchanged lines hidden (view full) ---
869 m = m_free(m);
870 }
871 if (m) {
872 sb->sb_mb = m;
873 m->m_nextpkt = next;
874 } else
875 sb->sb_mb = next;
876 /*
877 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
878 * sb_lastrecord is up-to-date if we dropped part of the last record.
879 */
880 m = sb->sb_mb;
881 if (m == NULL) {
882 sb->sb_mbtail = NULL;
883 sb->sb_lastrecord = NULL;
884 } else if (m->m_nextpkt == NULL) {
885 sb->sb_lastrecord = m;
886 }
887}
888
889/*
890 * Drop data from (the front of) a sockbuf.
891 */
892void
893sbdrop_locked(struct sockbuf *sb, int len)
894{
895
896 SOCKBUF_LOCK_ASSERT(sb);
897
898 sbdrop_internal(sb, len);
899}
900
901void
902sbdrop(struct sockbuf *sb, int len)
903{
904
905 SOCKBUF_LOCK(sb);
906 sbdrop_locked(sb, len);
907 SOCKBUF_UNLOCK(sb);
908}
909
910/*
911 * Drop a record off the front of a sockbuf and move the next record to the
912 * front.
913 */
914void
915sbdroprecord_locked(struct sockbuf *sb)
916{
917 struct mbuf *m;
918
919 SOCKBUF_LOCK_ASSERT(sb);
920
921 m = sb->sb_mb;
922 if (m) {
923 sb->sb_mb = m->m_nextpkt;
924 do {
925 sbfree(sb, m);
926 m = m_free(m);
927 } while (m);
928 }
929 SB_EMPTY_FIXUP(sb);
930}
931
932/*
933 * Drop a record off the front of a sockbuf and move the next record to the
934 * front.
935 */
936void
937sbdroprecord(struct sockbuf *sb)
938{
939
940 SOCKBUF_LOCK(sb);
941 sbdroprecord_locked(sb);
942 SOCKBUF_UNLOCK(sb);
943}
944
945/* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
946static int dummy;
947SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW, &dummy, 0, "");
948SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf, CTLTYPE_ULONG|CTLFLAG_RW,
949 &sb_max, 0, sysctl_handle_sb_max, "LU", "Maximum socket buffer size");
950SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
951 &sb_efficiency, 0, "");