1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * @(#)uipc_socket2.c 8.1 (Berkeley) 6/10/93
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_kern_tls.h"
38 #include "opt_param.h"
39
40 #include <sys/param.h>
41 #include <sys/aio.h> /* for aio_swake proto */
42 #include <sys/kernel.h>
43 #include <sys/ktls.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/mutex.h>
48 #include <sys/proc.h>
49 #include <sys/protosw.h>
50 #include <sys/resourcevar.h>
51 #include <sys/signalvar.h>
52 #include <sys/socket.h>
53 #include <sys/socketvar.h>
54 #include <sys/sx.h>
55 #include <sys/sysctl.h>
56
57 #include <netinet/in.h>
58
59 /*
60 * Function pointer set by the AIO routines so that the socket buffer code
61 * can call back into the AIO module if it is loaded.
62 */
63 void (*aio_swake)(struct socket *, struct sockbuf *);
64
65 /*
66 * Primitive routines for operating on socket buffers
67 */
68
69 #define BUF_MAX_ADJ(_sz) (((u_quad_t)(_sz)) * MCLBYTES / (MSIZE + MCLBYTES))
70
71 u_long sb_max = SB_MAX;
72 u_long sb_max_adj = BUF_MAX_ADJ(SB_MAX);
73
74 static u_long sb_efficiency = 8; /* parameter for sbreserve() */
75
76 #ifdef KERN_TLS
77 static void sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m,
78 struct mbuf *n);
79 #endif
80 static struct mbuf *sbcut_internal(struct sockbuf *sb, int len);
81 static void sbflush_internal(struct sockbuf *sb);
82
83 /*
84 * Our own version of m_clrprotoflags(), that can preserve M_NOTREADY.
85 */
86 static void
87 sbm_clrprotoflags(struct mbuf *m, int flags)
88 {
89 int mask;
90
91 mask = ~M_PROTOFLAGS;
92 if (flags & PRUS_NOTREADY)
93 mask |= M_NOTREADY;
94 while (m) {
95 m->m_flags &= mask;
96 m = m->m_next;
97 }
98 }
99
100 /*
101 * Compress M_NOTREADY mbufs after they have been readied by sbready().
102 *
103 * sbcompress() skips M_NOTREADY mbufs since the data is not available to
104 * be copied at the time of sbcompress(). This function combines small
105 * mbufs similar to sbcompress() once mbufs are ready. 'm0' is the first
106 * mbuf sbready() marked ready, and 'end' is the first mbuf still not
107 * ready.
108 */
109 static void
110 sbready_compress(struct sockbuf *sb, struct mbuf *m0, struct mbuf *end)
111 {
112 struct mbuf *m, *n;
113 int ext_size;
114
115 SOCKBUF_LOCK_ASSERT(sb);
116
117 if ((sb->sb_flags & SB_NOCOALESCE) != 0)
118 return;
119
120 for (m = m0; m != end; m = m->m_next) {
121 MPASS((m->m_flags & M_NOTREADY) == 0);
122 /*
123 * NB: In sbcompress(), 'n' is the last mbuf in the
124 * socket buffer and 'm' is the new mbuf being copied
125 * into the trailing space of 'n'. Here, the roles
126 * are reversed and 'n' is the next mbuf after 'm'
127 * that is being copied into the trailing space of
128 * 'm'.
129 */
130 n = m->m_next;
131 #ifdef KERN_TLS
132 /* Try to coalesce adjacent ktls mbuf hdr/trailers. */
133 if ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 &&
134 (m->m_flags & M_EXTPG) &&
135 (n->m_flags & M_EXTPG) &&
136 !mbuf_has_tls_session(m) &&
137 !mbuf_has_tls_session(n)) {
138 int hdr_len, trail_len;
139
140 hdr_len = n->m_epg_hdrlen;
141 trail_len = m->m_epg_trllen;
142 if (trail_len != 0 && hdr_len != 0 &&
143 trail_len + hdr_len <= MBUF_PEXT_TRAIL_LEN) {
144 /* copy n's header to m's trailer */
145 memcpy(&m->m_epg_trail[trail_len],
146 n->m_epg_hdr, hdr_len);
147 m->m_epg_trllen += hdr_len;
148 m->m_len += hdr_len;
149 n->m_epg_hdrlen = 0;
150 n->m_len -= hdr_len;
151 }
152 }
153 #endif
154
155 /* Compress small unmapped mbufs into plain mbufs. */
156 if ((m->m_flags & M_EXTPG) && m->m_len <= MLEN &&
157 !mbuf_has_tls_session(m)) {
158 ext_size = m->m_ext.ext_size;
159 if (mb_unmapped_compress(m) == 0)
160 sb->sb_mbcnt -= ext_size;
161 }
162
163 while ((n != NULL) && (n != end) && (m->m_flags & M_EOR) == 0 &&
164 M_WRITABLE(m) &&
165 (m->m_flags & M_EXTPG) == 0 &&
166 !mbuf_has_tls_session(n) &&
167 !mbuf_has_tls_session(m) &&
168 n->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
169 n->m_len <= M_TRAILINGSPACE(m) &&
170 m->m_type == n->m_type) {
171 KASSERT(sb->sb_lastrecord != n,
172 ("%s: merging start of record (%p) into previous mbuf (%p)",
173 __func__, n, m));
174 m_copydata(n, 0, n->m_len, mtodo(m, m->m_len));
175 m->m_len += n->m_len;
176 m->m_next = n->m_next;
177 m->m_flags |= n->m_flags & M_EOR;
178 if (sb->sb_mbtail == n)
179 sb->sb_mbtail = m;
180
181 sb->sb_mbcnt -= MSIZE;
182 if (n->m_flags & M_EXT)
183 sb->sb_mbcnt -= n->m_ext.ext_size;
184 m_free(n);
185 n = m->m_next;
186 }
187 }
188 SBLASTRECORDCHK(sb);
189 SBLASTMBUFCHK(sb);
190 }
191
192 /*
193 * Mark ready "count" units of I/O starting with "m". Most mbufs
194 * count as a single unit of I/O except for M_EXTPG mbufs which
195 * are backed by multiple pages.
196 */
197 int
198 sbready(struct sockbuf *sb, struct mbuf *m0, int count)
199 {
200 struct mbuf *m;
201 u_int blocker;
202
203 SOCKBUF_LOCK_ASSERT(sb);
204 KASSERT(sb->sb_fnrdy != NULL, ("%s: sb %p NULL fnrdy", __func__, sb));
205 KASSERT(count > 0, ("%s: invalid count %d", __func__, count));
206
207 m = m0;
208 blocker = (sb->sb_fnrdy == m) ? M_BLOCKED : 0;
209
210 while (count > 0) {
211 KASSERT(m->m_flags & M_NOTREADY,
212 ("%s: m %p !M_NOTREADY", __func__, m));
213 if ((m->m_flags & M_EXTPG) != 0 && m->m_epg_npgs != 0) {
214 if (count < m->m_epg_nrdy) {
215 m->m_epg_nrdy -= count;
216 count = 0;
217 break;
218 }
219 count -= m->m_epg_nrdy;
220 m->m_epg_nrdy = 0;
221 } else
222 count--;
223
224 m->m_flags &= ~(M_NOTREADY | blocker);
225 if (blocker)
226 sb->sb_acc += m->m_len;
227 m = m->m_next;
228 }
229
230 /*
231 * If the first mbuf is still not fully ready because only
232 * some of its backing pages were readied, no further progress
233 * can be made.
234 */
235 if (m0 == m) {
236 MPASS(m->m_flags & M_NOTREADY);
237 return (EINPROGRESS);
238 }
239
240 if (!blocker) {
241 sbready_compress(sb, m0, m);
242 return (EINPROGRESS);
243 }
244
245 /* This one was blocking all the queue. */
246 for (; m && (m->m_flags & M_NOTREADY) == 0; m = m->m_next) {
247 KASSERT(m->m_flags & M_BLOCKED,
248 ("%s: m %p !M_BLOCKED", __func__, m));
249 m->m_flags &= ~M_BLOCKED;
250 sb->sb_acc += m->m_len;
251 }
252
253 sb->sb_fnrdy = m;
254 sbready_compress(sb, m0, m);
255
256 return (0);
257 }
258
259 /*
260 * Adjust sockbuf state reflecting allocation of m.
261 */
262 void
263 sballoc(struct sockbuf *sb, struct mbuf *m)
264 {
265
266 SOCKBUF_LOCK_ASSERT(sb);
267
268 sb->sb_ccc += m->m_len;
269
270 if (sb->sb_fnrdy == NULL) {
271 if (m->m_flags & M_NOTREADY)
272 sb->sb_fnrdy = m;
273 else
274 sb->sb_acc += m->m_len;
275 } else
276 m->m_flags |= M_BLOCKED;
277
278 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
279 sb->sb_ctl += m->m_len;
280
281 sb->sb_mbcnt += MSIZE;
282
283 if (m->m_flags & M_EXT)
284 sb->sb_mbcnt += m->m_ext.ext_size;
285 }
286
287 /*
288 * Adjust sockbuf state reflecting freeing of m.
289 */
290 void
291 sbfree(struct sockbuf *sb, struct mbuf *m)
292 {
293
294 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
295 SOCKBUF_LOCK_ASSERT(sb);
296 #endif
297
298 sb->sb_ccc -= m->m_len;
299
300 if (!(m->m_flags & M_NOTAVAIL))
301 sb->sb_acc -= m->m_len;
302
303 if (m == sb->sb_fnrdy) {
304 struct mbuf *n;
305
306 KASSERT(m->m_flags & M_NOTREADY,
307 ("%s: m %p !M_NOTREADY", __func__, m));
308
309 n = m->m_next;
310 while (n != NULL && !(n->m_flags & M_NOTREADY)) {
311 n->m_flags &= ~M_BLOCKED;
312 sb->sb_acc += n->m_len;
313 n = n->m_next;
314 }
315 sb->sb_fnrdy = n;
316 }
317
318 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
319 sb->sb_ctl -= m->m_len;
320
321 sb->sb_mbcnt -= MSIZE;
322 if (m->m_flags & M_EXT)
323 sb->sb_mbcnt -= m->m_ext.ext_size;
324
325 if (sb->sb_sndptr == m) {
326 sb->sb_sndptr = NULL;
327 sb->sb_sndptroff = 0;
328 }
329 if (sb->sb_sndptroff != 0)
330 sb->sb_sndptroff -= m->m_len;
331 }
332
333 #ifdef KERN_TLS
334 /*
335 * Similar to sballoc/sbfree but does not adjust state associated with
336 * the sb_mb chain such as sb_fnrdy or sb_sndptr*. Also assumes mbufs
337 * are not ready.
338 */
339 void
340 sballoc_ktls_rx(struct sockbuf *sb, struct mbuf *m)
341 {
342
343 SOCKBUF_LOCK_ASSERT(sb);
344
345 sb->sb_ccc += m->m_len;
346 sb->sb_tlscc += m->m_len;
347
348 sb->sb_mbcnt += MSIZE;
349
350 if (m->m_flags & M_EXT)
351 sb->sb_mbcnt += m->m_ext.ext_size;
352 }
353
354 void
355 sbfree_ktls_rx(struct sockbuf *sb, struct mbuf *m)
356 {
357
358 #if 0 /* XXX: not yet: soclose() call path comes here w/o lock. */
359 SOCKBUF_LOCK_ASSERT(sb);
360 #endif
361
362 sb->sb_ccc -= m->m_len;
363 sb->sb_tlscc -= m->m_len;
364
365 sb->sb_mbcnt -= MSIZE;
366
367 if (m->m_flags & M_EXT)
368 sb->sb_mbcnt -= m->m_ext.ext_size;
369 }
370 #endif
371
372 /*
373 * Socantsendmore indicates that no more data will be sent on the socket; it
374 * would normally be applied to a socket when the user informs the system
375 * that no more data is to be sent, by the protocol code (in case
376 * PRU_SHUTDOWN). Socantrcvmore indicates that no more data will be
377 * received, and will normally be applied to the socket by a protocol when it
378 * detects that the peer will send no more data. Data queued for reading in
379 * the socket may yet be read.
380 */
381 void
382 socantsendmore_locked(struct socket *so)
383 {
384
385 SOCK_SENDBUF_LOCK_ASSERT(so);
386
387 so->so_snd.sb_state |= SBS_CANTSENDMORE;
388 sowwakeup_locked(so);
389 SOCK_SENDBUF_UNLOCK_ASSERT(so);
390 }
391
392 void
393 socantsendmore(struct socket *so)
394 {
395
396 SOCK_SENDBUF_LOCK(so);
397 socantsendmore_locked(so);
398 SOCK_SENDBUF_UNLOCK_ASSERT(so);
399 }
400
401 void
402 socantrcvmore_locked(struct socket *so)
403 {
404
405 SOCK_RECVBUF_LOCK_ASSERT(so);
406
407 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
408 #ifdef KERN_TLS
409 if (so->so_rcv.sb_flags & SB_TLS_RX)
410 ktls_check_rx(&so->so_rcv);
411 #endif
412 sorwakeup_locked(so);
413 SOCK_RECVBUF_UNLOCK_ASSERT(so);
414 }
415
416 void
417 socantrcvmore(struct socket *so)
418 {
419
420 SOCK_RECVBUF_LOCK(so);
421 socantrcvmore_locked(so);
422 SOCK_RECVBUF_UNLOCK_ASSERT(so);
423 }
424
425 void
426 soroverflow_locked(struct socket *so)
427 {
428
429 SOCK_RECVBUF_LOCK_ASSERT(so);
430
431 if (so->so_options & SO_RERROR) {
432 so->so_rerror = ENOBUFS;
433 sorwakeup_locked(so);
434 } else
435 SOCK_RECVBUF_UNLOCK(so);
436
437 SOCK_RECVBUF_UNLOCK_ASSERT(so);
438 }
439
440 void
441 soroverflow(struct socket *so)
442 {
443
444 SOCK_RECVBUF_LOCK(so);
445 soroverflow_locked(so);
446 SOCK_RECVBUF_UNLOCK_ASSERT(so);
447 }
448
449 /*
450 * Wait for data to arrive at/drain from a socket buffer.
451 */
452 int
453 sbwait(struct socket *so, sb_which which)
454 {
455 struct sockbuf *sb;
456
457 SOCK_BUF_LOCK_ASSERT(so, which);
458
459 sb = sobuf(so, which);
460 sb->sb_flags |= SB_WAIT;
461 return (msleep_sbt(&sb->sb_acc, soeventmtx(so, which),
462 (sb->sb_flags & SB_NOINTR) ? PSOCK : PSOCK | PCATCH, "sbwait",
463 sb->sb_timeo, 0, 0));
464 }
465
466 /*
467 * Wakeup processes waiting on a socket buffer. Do asynchronous notification
468 * via SIGIO if the socket has the SS_ASYNC flag set.
469 *
470 * Called with the socket buffer lock held; will release the lock by the end
471 * of the function. This allows the caller to acquire the socket buffer lock
472 * while testing for the need for various sorts of wakeup and hold it through
473 * to the point where it's no longer required. We currently hold the lock
474 * through calls out to other subsystems (with the exception of kqueue), and
475 * then release it to avoid lock order issues. It's not clear that's
476 * correct.
477 */
478 static __always_inline void
479 sowakeup(struct socket *so, const sb_which which)
480 {
481 struct sockbuf *sb;
482 int ret;
483
484 SOCK_BUF_LOCK_ASSERT(so, which);
485
486 sb = sobuf(so, which);
487 selwakeuppri(sb->sb_sel, PSOCK);
488 if (!SEL_WAITING(sb->sb_sel))
489 sb->sb_flags &= ~SB_SEL;
490 if (sb->sb_flags & SB_WAIT) {
491 sb->sb_flags &= ~SB_WAIT;
492 wakeup(&sb->sb_acc);
493 }
494 KNOTE_LOCKED(&sb->sb_sel->si_note, 0);
495 if (sb->sb_upcall != NULL) {
496 ret = sb->sb_upcall(so, sb->sb_upcallarg, M_NOWAIT);
497 if (ret == SU_ISCONNECTED) {
498 KASSERT(sb == &so->so_rcv,
499 ("SO_SND upcall returned SU_ISCONNECTED"));
500 soupcall_clear(so, SO_RCV);
501 }
502 } else
503 ret = SU_OK;
504 if (sb->sb_flags & SB_AIO)
505 sowakeup_aio(so, which);
506 SOCK_BUF_UNLOCK(so, which);
507 if (ret == SU_ISCONNECTED)
508 soisconnected(so);
509 if ((so->so_state & SS_ASYNC) && so->so_sigio != NULL)
510 pgsigio(&so->so_sigio, SIGIO, 0);
511 SOCK_BUF_UNLOCK_ASSERT(so, which);
512 }
513
514 /*
515 * Do we need to notify the other side when I/O is possible?
516 */
517 static __always_inline bool
518 sb_notify(const struct sockbuf *sb)
519 {
520 return ((sb->sb_flags & (SB_WAIT | SB_SEL | SB_ASYNC |
521 SB_UPCALL | SB_AIO | SB_KNOTE)) != 0);
522 }
523
524 void
525 sorwakeup_locked(struct socket *so)
526 {
527 SOCK_RECVBUF_LOCK_ASSERT(so);
528 if (sb_notify(&so->so_rcv))
529 sowakeup(so, SO_RCV);
530 else
531 SOCK_RECVBUF_UNLOCK(so);
532 }
533
534 void
535 sowwakeup_locked(struct socket *so)
536 {
537 SOCK_SENDBUF_LOCK_ASSERT(so);
538 if (sb_notify(&so->so_snd))
539 sowakeup(so, SO_SND);
540 else
541 SOCK_SENDBUF_UNLOCK(so);
542 }
543
544 /*
545 * Socket buffer (struct sockbuf) utility routines.
546 *
547 * Each socket contains two socket buffers: one for sending data and one for
548 * receiving data. Each buffer contains a queue of mbufs, information about
549 * the number of mbufs and amount of data in the queue, and other fields
550 * allowing select() statements and notification on data availability to be
551 * implemented.
552 *
553 * Data stored in a socket buffer is maintained as a list of records. Each
554 * record is a list of mbufs chained together with the m_next field. Records
555 * are chained together with the m_nextpkt field. The upper level routine
556 * soreceive() expects the following conventions to be observed when placing
557 * information in the receive buffer:
558 *
559 * 1. If the protocol requires each message be preceded by the sender's name,
560 * then a record containing that name must be present before any
561 * associated data (mbuf's must be of type MT_SONAME).
562 * 2. If the protocol supports the exchange of ``access rights'' (really just
563 * additional data associated with the message), and there are ``rights''
564 * to be received, then a record containing this data should be present
565 * (mbuf's must be of type MT_RIGHTS).
566 * 3. If a name or rights record exists, then it must be followed by a data
567 * record, perhaps of zero length.
568 *
569 * Before using a new socket structure it is first necessary to reserve
570 * buffer space to the socket, by calling sbreserve(). This should commit
571 * some of the available buffer space in the system buffer pool for the
572 * socket (currently, it does nothing but enforce limits). The space should
573 * be released by calling sbrelease() when the socket is destroyed.
574 */
575 int
576 soreserve(struct socket *so, u_long sndcc, u_long rcvcc)
577 {
578 struct thread *td = curthread;
579
580 SOCK_SENDBUF_LOCK(so);
581 SOCK_RECVBUF_LOCK(so);
582 if (sbreserve_locked(so, SO_SND, sndcc, td) == 0)
583 goto bad;
584 if (sbreserve_locked(so, SO_RCV, rcvcc, td) == 0)
585 goto bad2;
586 if (so->so_rcv.sb_lowat == 0)
587 so->so_rcv.sb_lowat = 1;
588 if (so->so_snd.sb_lowat == 0)
589 so->so_snd.sb_lowat = MCLBYTES;
590 if (so->so_snd.sb_lowat > so->so_snd.sb_hiwat)
591 so->so_snd.sb_lowat = so->so_snd.sb_hiwat;
592 SOCK_RECVBUF_UNLOCK(so);
593 SOCK_SENDBUF_UNLOCK(so);
594 return (0);
595 bad2:
596 sbrelease_locked(so, SO_SND);
597 bad:
598 SOCK_RECVBUF_UNLOCK(so);
599 SOCK_SENDBUF_UNLOCK(so);
600 return (ENOBUFS);
601 }
602
603 static int
604 sysctl_handle_sb_max(SYSCTL_HANDLER_ARGS)
605 {
606 int error = 0;
607 u_long tmp_sb_max = sb_max;
608
609 error = sysctl_handle_long(oidp, &tmp_sb_max, arg2, req);
610 if (error || !req->newptr)
611 return (error);
612 if (tmp_sb_max < MSIZE + MCLBYTES)
613 return (EINVAL);
614 sb_max = tmp_sb_max;
615 sb_max_adj = BUF_MAX_ADJ(sb_max);
616 return (0);
617 }
618
619 /*
620 * Allot mbufs to a sockbuf. Attempt to scale mbmax so that mbcnt doesn't
621 * become limiting if buffering efficiency is near the normal case.
622 */
623 bool
624 sbreserve_locked_limit(struct socket *so, sb_which which, u_long cc,
625 u_long buf_max, struct thread *td)
626 {
627 struct sockbuf *sb = sobuf(so, which);
628 rlim_t sbsize_limit;
629
630 SOCK_BUF_LOCK_ASSERT(so, which);
631
632 /*
633 * When a thread is passed, we take into account the thread's socket
634 * buffer size limit. The caller will generally pass curthread, but
635 * in the TCP input path, NULL will be passed to indicate that no
636 * appropriate thread resource limits are available. In that case,
637 * we don't apply a process limit.
638 */
639 if (cc > BUF_MAX_ADJ(buf_max))
640 return (false);
641 if (td != NULL) {
642 sbsize_limit = lim_cur(td, RLIMIT_SBSIZE);
643 } else
644 sbsize_limit = RLIM_INFINITY;
645 if (!chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, cc,
646 sbsize_limit))
647 return (false);
648 sb->sb_mbmax = min(cc * sb_efficiency, buf_max);
649 if (sb->sb_lowat > sb->sb_hiwat)
650 sb->sb_lowat = sb->sb_hiwat;
651 return (true);
652 }
653
654 bool
655 sbreserve_locked(struct socket *so, sb_which which, u_long cc,
656 struct thread *td)
657 {
658 return (sbreserve_locked_limit(so, which, cc, sb_max, td));
659 }
660
661 int
662 sbsetopt(struct socket *so, struct sockopt *sopt)
663 {
664 struct sockbuf *sb;
665 sb_which wh;
666 short *flags;
667 u_int cc, *hiwat, *lowat;
668 int error, optval;
669
670 error = sooptcopyin(sopt, &optval, sizeof optval, sizeof optval);
671 if (error != 0)
672 return (error);
673
674 /*
675 * Values < 1 make no sense for any of these options,
676 * so disallow them.
677 */
678 if (optval < 1)
679 return (EINVAL);
680 cc = optval;
681
682 sb = NULL;
683 SOCK_LOCK(so);
684 if (SOLISTENING(so)) {
685 switch (sopt->sopt_name) {
686 case SO_SNDLOWAT:
687 case SO_SNDBUF:
688 lowat = &so->sol_sbsnd_lowat;
689 hiwat = &so->sol_sbsnd_hiwat;
690 flags = &so->sol_sbsnd_flags;
691 break;
692 case SO_RCVLOWAT:
693 case SO_RCVBUF:
694 lowat = &so->sol_sbrcv_lowat;
695 hiwat = &so->sol_sbrcv_hiwat;
696 flags = &so->sol_sbrcv_flags;
697 break;
698 }
699 } else {
700 switch (sopt->sopt_name) {
701 case SO_SNDLOWAT:
702 case SO_SNDBUF:
703 sb = &so->so_snd;
704 wh = SO_SND;
705 break;
706 case SO_RCVLOWAT:
707 case SO_RCVBUF:
708 sb = &so->so_rcv;
709 wh = SO_RCV;
710 break;
711 }
712 flags = &sb->sb_flags;
713 hiwat = &sb->sb_hiwat;
714 lowat = &sb->sb_lowat;
715 SOCK_BUF_LOCK(so, wh);
716 }
717
718 error = 0;
719 switch (sopt->sopt_name) {
720 case SO_SNDBUF:
721 case SO_RCVBUF:
722 if (SOLISTENING(so)) {
723 if (cc > sb_max_adj) {
724 error = ENOBUFS;
725 break;
726 }
727 *hiwat = cc;
728 if (*lowat > *hiwat)
729 *lowat = *hiwat;
730 } else {
731 if (!sbreserve_locked(so, wh, cc, curthread))
732 error = ENOBUFS;
733 }
734 if (error == 0)
735 *flags &= ~SB_AUTOSIZE;
736 break;
737 case SO_SNDLOWAT:
738 case SO_RCVLOWAT:
739 /*
740 * Make sure the low-water is never greater than the
741 * high-water.
742 */
743 *lowat = (cc > *hiwat) ? *hiwat : cc;
744 break;
745 }
746
747 if (!SOLISTENING(so))
748 SOCK_BUF_UNLOCK(so, wh);
749 SOCK_UNLOCK(so);
750 return (error);
751 }
752
753 /*
754 * Free mbufs held by a socket, and reserved mbuf space.
755 */
756 static void
757 sbrelease_internal(struct socket *so, sb_which which)
758 {
759 struct sockbuf *sb = sobuf(so, which);
760
761 sbflush_internal(sb);
762 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
763 RLIM_INFINITY);
764 sb->sb_mbmax = 0;
765 }
766
767 void
768 sbrelease_locked(struct socket *so, sb_which which)
769 {
770
771 SOCK_BUF_LOCK_ASSERT(so, which);
772
773 sbrelease_internal(so, which);
774 }
775
776 void
777 sbrelease(struct socket *so, sb_which which)
778 {
779
780 SOCK_BUF_LOCK(so, which);
781 sbrelease_locked(so, which);
782 SOCK_BUF_UNLOCK(so, which);
783 }
784
785 void
786 sbdestroy(struct socket *so, sb_which which)
787 {
788 #ifdef KERN_TLS
789 struct sockbuf *sb = sobuf(so, which);
790
791 if (sb->sb_tls_info != NULL)
792 ktls_free(sb->sb_tls_info);
793 sb->sb_tls_info = NULL;
794 #endif
795 sbrelease_internal(so, which);
796 }
797
798 /*
799 * Routines to add and remove data from an mbuf queue.
800 *
801 * The routines sbappend() or sbappendrecord() are normally called to append
802 * new mbufs to a socket buffer, after checking that adequate space is
803 * available, comparing the function sbspace() with the amount of data to be
804 * added. sbappendrecord() differs from sbappend() in that data supplied is
805 * treated as the beginning of a new record. To place a sender's address,
806 * optional access rights, and data in a socket receive buffer,
807 * sbappendaddr() should be used. To place access rights and data in a
808 * socket receive buffer, sbappendrights() should be used. In either case,
809 * the new data begins a new record. Note that unlike sbappend() and
810 * sbappendrecord(), these routines check for the caller that there will be
811 * enough space to store the data. Each fails if there is not enough space,
812 * or if it cannot find mbufs to store additional information in.
813 *
814 * Reliable protocols may use the socket send buffer to hold data awaiting
815 * acknowledgement. Data is normally copied from a socket send buffer in a
816 * protocol with m_copy for output to a peer, and then removing the data from
817 * the socket buffer with sbdrop() or sbdroprecord() when the data is
818 * acknowledged by the peer.
819 */
820 #ifdef SOCKBUF_DEBUG
821 void
822 sblastrecordchk(struct sockbuf *sb, const char *file, int line)
823 {
824 struct mbuf *m = sb->sb_mb;
825
826 SOCKBUF_LOCK_ASSERT(sb);
827
828 while (m && m->m_nextpkt)
829 m = m->m_nextpkt;
830
831 if (m != sb->sb_lastrecord) {
832 printf("%s: sb_mb %p sb_lastrecord %p last %p\n",
833 __func__, sb->sb_mb, sb->sb_lastrecord, m);
834 printf("packet chain:\n");
835 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt)
836 printf("\t%p\n", m);
837 panic("%s from %s:%u", __func__, file, line);
838 }
839 }
840
841 void
842 sblastmbufchk(struct sockbuf *sb, const char *file, int line)
843 {
844 struct mbuf *m = sb->sb_mb;
845 struct mbuf *n;
846
847 SOCKBUF_LOCK_ASSERT(sb);
848
849 while (m && m->m_nextpkt)
850 m = m->m_nextpkt;
851
852 while (m && m->m_next)
853 m = m->m_next;
854
855 if (m != sb->sb_mbtail) {
856 printf("%s: sb_mb %p sb_mbtail %p last %p\n",
857 __func__, sb->sb_mb, sb->sb_mbtail, m);
858 printf("packet tree:\n");
859 for (m = sb->sb_mb; m != NULL; m = m->m_nextpkt) {
860 printf("\t");
861 for (n = m; n != NULL; n = n->m_next)
862 printf("%p ", n);
863 printf("\n");
864 }
865 panic("%s from %s:%u", __func__, file, line);
866 }
867
868 #ifdef KERN_TLS
869 m = sb->sb_mtls;
870 while (m && m->m_next)
871 m = m->m_next;
872
873 if (m != sb->sb_mtlstail) {
874 printf("%s: sb_mtls %p sb_mtlstail %p last %p\n",
875 __func__, sb->sb_mtls, sb->sb_mtlstail, m);
876 printf("TLS packet tree:\n");
877 printf("\t");
878 for (m = sb->sb_mtls; m != NULL; m = m->m_next) {
879 printf("%p ", m);
880 }
881 printf("\n");
882 panic("%s from %s:%u", __func__, file, line);
883 }
884 #endif
885 }
886 #endif /* SOCKBUF_DEBUG */
887
888 #define SBLINKRECORD(sb, m0) do { \
889 SOCKBUF_LOCK_ASSERT(sb); \
890 if ((sb)->sb_lastrecord != NULL) \
891 (sb)->sb_lastrecord->m_nextpkt = (m0); \
892 else \
893 (sb)->sb_mb = (m0); \
894 (sb)->sb_lastrecord = (m0); \
895 } while (/*CONSTCOND*/0)
896
897 /*
898 * Append mbuf chain m to the last record in the socket buffer sb. The
899 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
900 * are discarded and mbufs are compacted where possible.
901 */
902 void
903 sbappend_locked(struct sockbuf *sb, struct mbuf *m, int flags)
904 {
905 struct mbuf *n;
906
907 SOCKBUF_LOCK_ASSERT(sb);
908
909 if (m == NULL)
910 return;
911 sbm_clrprotoflags(m, flags);
912 SBLASTRECORDCHK(sb);
913 n = sb->sb_mb;
914 if (n) {
915 while (n->m_nextpkt)
916 n = n->m_nextpkt;
917 do {
918 if (n->m_flags & M_EOR) {
919 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
920 return;
921 }
922 } while (n->m_next && (n = n->m_next));
923 } else {
924 /*
925 * XXX Would like to simply use sb_mbtail here, but
926 * XXX I need to verify that I won't miss an EOR that
927 * XXX way.
928 */
929 if ((n = sb->sb_lastrecord) != NULL) {
930 do {
931 if (n->m_flags & M_EOR) {
932 sbappendrecord_locked(sb, m); /* XXXXXX!!!! */
933 return;
934 }
935 } while (n->m_next && (n = n->m_next));
936 } else {
937 /*
938 * If this is the first record in the socket buffer,
939 * it's also the last record.
940 */
941 sb->sb_lastrecord = m;
942 }
943 }
944 sbcompress(sb, m, n);
945 SBLASTRECORDCHK(sb);
946 }
947
948 /*
949 * Append mbuf chain m to the last record in the socket buffer sb. The
950 * additional space associated the mbuf chain is recorded in sb. Empty mbufs
951 * are discarded and mbufs are compacted where possible.
952 */
953 void
954 sbappend(struct sockbuf *sb, struct mbuf *m, int flags)
955 {
956
957 SOCKBUF_LOCK(sb);
958 sbappend_locked(sb, m, flags);
959 SOCKBUF_UNLOCK(sb);
960 }
961
962 #ifdef KERN_TLS
963 /*
964 * Append an mbuf containing encrypted TLS data. The data
965 * is marked M_NOTREADY until it has been decrypted and
966 * stored as a TLS record.
967 */
968 static void
969 sbappend_ktls_rx(struct sockbuf *sb, struct mbuf *m)
970 {
971 struct ifnet *ifp;
972 struct mbuf *n;
973 int flags;
974
975 ifp = NULL;
976 flags = M_NOTREADY;
977
978 SBLASTMBUFCHK(sb);
979
980 /* Mbuf chain must start with a packet header. */
981 MPASS((m->m_flags & M_PKTHDR) != 0);
982
983 /* Remove all packet headers and mbuf tags to get a pure data chain. */
984 for (n = m; n != NULL; n = n->m_next) {
985 if (n->m_flags & M_PKTHDR) {
986 ifp = m->m_pkthdr.leaf_rcvif;
987 if ((n->m_pkthdr.csum_flags & CSUM_TLS_MASK) ==
988 CSUM_TLS_DECRYPTED) {
989 /* Mark all mbufs in this packet decrypted. */
990 flags = M_NOTREADY | M_DECRYPTED;
991 } else {
992 flags = M_NOTREADY;
993 }
994 m_demote_pkthdr(n);
995 }
996
997 n->m_flags &= M_DEMOTEFLAGS;
998 n->m_flags |= flags;
999
1000 MPASS((n->m_flags & M_NOTREADY) != 0);
1001 }
1002
1003 sbcompress_ktls_rx(sb, m, sb->sb_mtlstail);
1004 ktls_check_rx(sb);
1005
1006 /* Check for incoming packet route changes: */
1007 if (ifp != NULL && sb->sb_tls_info->rx_ifp != NULL &&
1008 sb->sb_tls_info->rx_ifp != ifp)
1009 ktls_input_ifp_mismatch(sb, ifp);
1010 }
1011 #endif
1012
1013 /*
1014 * This version of sbappend() should only be used when the caller absolutely
1015 * knows that there will never be more than one record in the socket buffer,
1016 * that is, a stream protocol (such as TCP).
1017 */
1018 void
1019 sbappendstream_locked(struct sockbuf *sb, struct mbuf *m, int flags)
1020 {
1021 SOCKBUF_LOCK_ASSERT(sb);
1022
1023 KASSERT(m->m_nextpkt == NULL,("sbappendstream 0"));
1024
1025 #ifdef KERN_TLS
1026 /*
1027 * Decrypted TLS records are appended as records via
1028 * sbappendrecord(). TCP passes encrypted TLS records to this
1029 * function which must be scheduled for decryption.
1030 */
1031 if (sb->sb_flags & SB_TLS_RX) {
1032 sbappend_ktls_rx(sb, m);
1033 return;
1034 }
1035 #endif
1036
1037 KASSERT(sb->sb_mb == sb->sb_lastrecord,("sbappendstream 1"));
1038
1039 SBLASTMBUFCHK(sb);
1040
1041 #ifdef KERN_TLS
1042 if (sb->sb_tls_info != NULL)
1043 ktls_seq(sb, m);
1044 #endif
1045
1046 /* Remove all packet headers and mbuf tags to get a pure data chain. */
1047 m_demote(m, 1, flags & PRUS_NOTREADY ? M_NOTREADY : 0);
1048
1049 sbcompress(sb, m, sb->sb_mbtail);
1050
1051 sb->sb_lastrecord = sb->sb_mb;
1052 SBLASTRECORDCHK(sb);
1053 }
1054
1055 /*
1056 * This version of sbappend() should only be used when the caller absolutely
1057 * knows that there will never be more than one record in the socket buffer,
1058 * that is, a stream protocol (such as TCP).
1059 */
1060 void
1061 sbappendstream(struct sockbuf *sb, struct mbuf *m, int flags)
1062 {
1063
1064 SOCKBUF_LOCK(sb);
1065 sbappendstream_locked(sb, m, flags);
1066 SOCKBUF_UNLOCK(sb);
1067 }
1068
1069 #ifdef SOCKBUF_DEBUG
1070 void
1071 sbcheck(struct sockbuf *sb, const char *file, int line)
1072 {
1073 struct mbuf *m, *n, *fnrdy;
1074 u_long acc, ccc, mbcnt;
1075 #ifdef KERN_TLS
1076 u_long tlscc;
1077 #endif
1078
1079 SOCKBUF_LOCK_ASSERT(sb);
1080
1081 acc = ccc = mbcnt = 0;
1082 fnrdy = NULL;
1083
1084 for (m = sb->sb_mb; m; m = n) {
1085 n = m->m_nextpkt;
1086 for (; m; m = m->m_next) {
1087 if (m->m_len == 0) {
1088 printf("sb %p empty mbuf %p\n", sb, m);
1089 goto fail;
1090 }
1091 if ((m->m_flags & M_NOTREADY) && fnrdy == NULL) {
1092 if (m != sb->sb_fnrdy) {
1093 printf("sb %p: fnrdy %p != m %p\n",
1094 sb, sb->sb_fnrdy, m);
1095 goto fail;
1096 }
1097 fnrdy = m;
1098 }
1099 if (fnrdy) {
1100 if (!(m->m_flags & M_NOTAVAIL)) {
1101 printf("sb %p: fnrdy %p, m %p is avail\n",
1102 sb, sb->sb_fnrdy, m);
1103 goto fail;
1104 }
1105 } else
1106 acc += m->m_len;
1107 ccc += m->m_len;
1108 mbcnt += MSIZE;
1109 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
1110 mbcnt += m->m_ext.ext_size;
1111 }
1112 }
1113 #ifdef KERN_TLS
1114 /*
1115 * Account for mbufs "detached" by ktls_detach_record() while
1116 * they are decrypted by ktls_decrypt(). tlsdcc gives a count
1117 * of the detached bytes that are included in ccc. The mbufs
1118 * and clusters are not included in the socket buffer
1119 * accounting.
1120 */
1121 ccc += sb->sb_tlsdcc;
1122
1123 tlscc = 0;
1124 for (m = sb->sb_mtls; m; m = m->m_next) {
1125 if (m->m_nextpkt != NULL) {
1126 printf("sb %p TLS mbuf %p with nextpkt\n", sb, m);
1127 goto fail;
1128 }
1129 if ((m->m_flags & M_NOTREADY) == 0) {
1130 printf("sb %p TLS mbuf %p ready\n", sb, m);
1131 goto fail;
1132 }
1133 tlscc += m->m_len;
1134 ccc += m->m_len;
1135 mbcnt += MSIZE;
1136 if (m->m_flags & M_EXT) /*XXX*/ /* pretty sure this is bogus */
1137 mbcnt += m->m_ext.ext_size;
1138 }
1139
1140 if (sb->sb_tlscc != tlscc) {
1141 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc,
1142 sb->sb_tlsdcc);
1143 goto fail;
1144 }
1145 #endif
1146 if (acc != sb->sb_acc || ccc != sb->sb_ccc || mbcnt != sb->sb_mbcnt) {
1147 printf("acc %ld/%u ccc %ld/%u mbcnt %ld/%u\n",
1148 acc, sb->sb_acc, ccc, sb->sb_ccc, mbcnt, sb->sb_mbcnt);
1149 #ifdef KERN_TLS
1150 printf("tlscc %ld/%u dcc %u\n", tlscc, sb->sb_tlscc,
1151 sb->sb_tlsdcc);
1152 #endif
1153 goto fail;
1154 }
1155 return;
1156 fail:
1157 panic("%s from %s:%u", __func__, file, line);
1158 }
1159 #endif
1160
1161 /*
1162 * As above, except the mbuf chain begins a new record.
1163 */
1164 void
1165 sbappendrecord_locked(struct sockbuf *sb, struct mbuf *m0)
1166 {
1167 struct mbuf *m;
1168
1169 SOCKBUF_LOCK_ASSERT(sb);
1170
1171 if (m0 == NULL)
1172 return;
1173 m_clrprotoflags(m0);
1174 /*
1175 * Put the first mbuf on the queue. Note this permits zero length
1176 * records.
1177 */
1178 sballoc(sb, m0);
1179 SBLASTRECORDCHK(sb);
1180 SBLINKRECORD(sb, m0);
1181 sb->sb_mbtail = m0;
1182 m = m0->m_next;
1183 m0->m_next = 0;
1184 if (m && (m0->m_flags & M_EOR)) {
1185 m0->m_flags &= ~M_EOR;
1186 m->m_flags |= M_EOR;
1187 }
1188 /* always call sbcompress() so it can do SBLASTMBUFCHK() */
1189 sbcompress(sb, m, m0);
1190 }
1191
1192 /*
1193 * As above, except the mbuf chain begins a new record.
1194 */
1195 void
1196 sbappendrecord(struct sockbuf *sb, struct mbuf *m0)
1197 {
1198
1199 SOCKBUF_LOCK(sb);
1200 sbappendrecord_locked(sb, m0);
1201 SOCKBUF_UNLOCK(sb);
1202 }
1203
1204 /* Helper routine that appends data, control, and address to a sockbuf. */
1205 static int
1206 sbappendaddr_locked_internal(struct sockbuf *sb, const struct sockaddr *asa,
1207 struct mbuf *m0, struct mbuf *control, struct mbuf *ctrl_last)
1208 {
1209 struct mbuf *m, *n, *nlast;
1210 #if MSIZE <= 256
1211 if (asa->sa_len > MLEN)
1212 return (0);
1213 #endif
1214 m = m_get(M_NOWAIT, MT_SONAME);
1215 if (m == NULL)
1216 return (0);
1217 m->m_len = asa->sa_len;
1218 bcopy(asa, mtod(m, caddr_t), asa->sa_len);
1219 if (m0) {
1220 M_ASSERT_NO_SND_TAG(m0);
1221 m_clrprotoflags(m0);
1222 m_tag_delete_chain(m0, NULL);
1223 /*
1224 * Clear some persistent info from pkthdr.
1225 * We don't use m_demote(), because some netgraph consumers
1226 * expect M_PKTHDR presence.
1227 */
1228 m0->m_pkthdr.rcvif = NULL;
1229 m0->m_pkthdr.flowid = 0;
1230 m0->m_pkthdr.csum_flags = 0;
1231 m0->m_pkthdr.fibnum = 0;
1232 m0->m_pkthdr.rsstype = 0;
1233 }
1234 if (ctrl_last)
1235 ctrl_last->m_next = m0; /* concatenate data to control */
1236 else
1237 control = m0;
1238 m->m_next = control;
1239 for (n = m; n->m_next != NULL; n = n->m_next)
1240 sballoc(sb, n);
1241 sballoc(sb, n);
1242 nlast = n;
1243 SBLINKRECORD(sb, m);
1244
1245 sb->sb_mbtail = nlast;
1246 SBLASTMBUFCHK(sb);
1247
1248 SBLASTRECORDCHK(sb);
1249 return (1);
1250 }
1251
1252 /*
1253 * Append address and data, and optionally, control (ancillary) data to the
1254 * receive queue of a socket. If present, m0 must include a packet header
1255 * with total length. Returns 0 if no space in sockbuf or insufficient
1256 * mbufs.
1257 */
1258 int
1259 sbappendaddr_locked(struct sockbuf *sb, const struct sockaddr *asa,
1260 struct mbuf *m0, struct mbuf *control)
1261 {
1262 struct mbuf *ctrl_last;
1263 int space = asa->sa_len;
1264
1265 SOCKBUF_LOCK_ASSERT(sb);
1266
1267 if (m0 && (m0->m_flags & M_PKTHDR) == 0)
1268 panic("sbappendaddr_locked");
1269 if (m0)
1270 space += m0->m_pkthdr.len;
1271 space += m_length(control, &ctrl_last);
1272
1273 if (space > sbspace(sb))
1274 return (0);
1275 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
1276 }
1277
1278 /*
1279 * Append address and data, and optionally, control (ancillary) data to the
1280 * receive queue of a socket. If present, m0 must include a packet header
1281 * with total length. Returns 0 if insufficient mbufs. Does not validate space
1282 * on the receiving sockbuf.
1283 */
1284 int
1285 sbappendaddr_nospacecheck_locked(struct sockbuf *sb, const struct sockaddr *asa,
1286 struct mbuf *m0, struct mbuf *control)
1287 {
1288 struct mbuf *ctrl_last;
1289
1290 SOCKBUF_LOCK_ASSERT(sb);
1291
1292 ctrl_last = (control == NULL) ? NULL : m_last(control);
1293 return (sbappendaddr_locked_internal(sb, asa, m0, control, ctrl_last));
1294 }
1295
1296 /*
1297 * Append address and data, and optionally, control (ancillary) data to the
1298 * receive queue of a socket. If present, m0 must include a packet header
1299 * with total length. Returns 0 if no space in sockbuf or insufficient
1300 * mbufs.
1301 */
1302 int
1303 sbappendaddr(struct sockbuf *sb, const struct sockaddr *asa,
1304 struct mbuf *m0, struct mbuf *control)
1305 {
1306 int retval;
1307
1308 SOCKBUF_LOCK(sb);
1309 retval = sbappendaddr_locked(sb, asa, m0, control);
1310 SOCKBUF_UNLOCK(sb);
1311 return (retval);
1312 }
1313
1314 void
1315 sbappendcontrol_locked(struct sockbuf *sb, struct mbuf *m0,
1316 struct mbuf *control, int flags)
1317 {
1318 struct mbuf *m, *mlast;
1319
1320 sbm_clrprotoflags(m0, flags);
1321 m_last(control)->m_next = m0;
1322
1323 SBLASTRECORDCHK(sb);
1324
1325 for (m = control; m->m_next; m = m->m_next)
1326 sballoc(sb, m);
1327 sballoc(sb, m);
1328 mlast = m;
1329 SBLINKRECORD(sb, control);
1330
1331 sb->sb_mbtail = mlast;
1332 SBLASTMBUFCHK(sb);
1333
1334 SBLASTRECORDCHK(sb);
1335 }
1336
1337 void
1338 sbappendcontrol(struct sockbuf *sb, struct mbuf *m0, struct mbuf *control,
1339 int flags)
1340 {
1341
1342 SOCKBUF_LOCK(sb);
1343 sbappendcontrol_locked(sb, m0, control, flags);
1344 SOCKBUF_UNLOCK(sb);
1345 }
1346
1347 /*
1348 * Append the data in mbuf chain (m) into the socket buffer sb following mbuf
1349 * (n). If (n) is NULL, the buffer is presumed empty.
1350 *
1351 * When the data is compressed, mbufs in the chain may be handled in one of
1352 * three ways:
1353 *
1354 * (1) The mbuf may simply be dropped, if it contributes nothing (no data, no
1355 * record boundary, and no change in data type).
1356 *
1357 * (2) The mbuf may be coalesced -- i.e., data in the mbuf may be copied into
1358 * an mbuf already in the socket buffer. This can occur if an
1359 * appropriate mbuf exists, there is room, both mbufs are not marked as
1360 * not ready, and no merging of data types will occur.
1361 *
1362 * (3) The mbuf may be appended to the end of the existing mbuf chain.
1363 *
1364 * If any of the new mbufs is marked as M_EOR, mark the last mbuf appended as
1365 * end-of-record.
1366 */
1367 void
1368 sbcompress(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1369 {
1370 int eor = 0;
1371 struct mbuf *o;
1372
1373 SOCKBUF_LOCK_ASSERT(sb);
1374
1375 while (m) {
1376 eor |= m->m_flags & M_EOR;
1377 if (m->m_len == 0 &&
1378 (eor == 0 ||
1379 (((o = m->m_next) || (o = n)) &&
1380 o->m_type == m->m_type))) {
1381 if (sb->sb_lastrecord == m)
1382 sb->sb_lastrecord = m->m_next;
1383 m = m_free(m);
1384 continue;
1385 }
1386 if (n && (n->m_flags & M_EOR) == 0 &&
1387 M_WRITABLE(n) &&
1388 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1389 !(m->m_flags & M_NOTREADY) &&
1390 !(n->m_flags & (M_NOTREADY | M_EXTPG)) &&
1391 !mbuf_has_tls_session(m) &&
1392 !mbuf_has_tls_session(n) &&
1393 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1394 m->m_len <= M_TRAILINGSPACE(n) &&
1395 n->m_type == m->m_type) {
1396 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len));
1397 n->m_len += m->m_len;
1398 sb->sb_ccc += m->m_len;
1399 if (sb->sb_fnrdy == NULL)
1400 sb->sb_acc += m->m_len;
1401 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1402 /* XXX: Probably don't need.*/
1403 sb->sb_ctl += m->m_len;
1404 m = m_free(m);
1405 continue;
1406 }
1407 if (m->m_len <= MLEN && (m->m_flags & M_EXTPG) &&
1408 (m->m_flags & M_NOTREADY) == 0 &&
1409 !mbuf_has_tls_session(m))
1410 (void)mb_unmapped_compress(m);
1411 if (n)
1412 n->m_next = m;
1413 else
1414 sb->sb_mb = m;
1415 sb->sb_mbtail = m;
1416 sballoc(sb, m);
1417 n = m;
1418 m->m_flags &= ~M_EOR;
1419 m = m->m_next;
1420 n->m_next = 0;
1421 }
1422 if (eor) {
1423 KASSERT(n != NULL, ("sbcompress: eor && n == NULL"));
1424 n->m_flags |= eor;
1425 }
1426 SBLASTMBUFCHK(sb);
1427 }
1428
1429 #ifdef KERN_TLS
1430 /*
1431 * A version of sbcompress() for encrypted TLS RX mbufs. These mbufs
1432 * are appended to the 'sb_mtls' chain instead of 'sb_mb' and are also
1433 * a bit simpler (no EOR markers, always MT_DATA, etc.).
1434 */
1435 static void
1436 sbcompress_ktls_rx(struct sockbuf *sb, struct mbuf *m, struct mbuf *n)
1437 {
1438
1439 SOCKBUF_LOCK_ASSERT(sb);
1440
1441 while (m) {
1442 KASSERT((m->m_flags & M_EOR) == 0,
1443 ("TLS RX mbuf %p with EOR", m));
1444 KASSERT(m->m_type == MT_DATA,
1445 ("TLS RX mbuf %p is not MT_DATA", m));
1446 KASSERT((m->m_flags & M_NOTREADY) != 0,
1447 ("TLS RX mbuf %p ready", m));
1448 KASSERT((m->m_flags & M_EXTPG) == 0,
1449 ("TLS RX mbuf %p unmapped", m));
1450
1451 if (m->m_len == 0) {
1452 m = m_free(m);
1453 continue;
1454 }
1455
1456 /*
1457 * Even though both 'n' and 'm' are NOTREADY, it's ok
1458 * to coalesce the data.
1459 */
1460 if (n &&
1461 M_WRITABLE(n) &&
1462 ((sb->sb_flags & SB_NOCOALESCE) == 0) &&
1463 !((m->m_flags ^ n->m_flags) & M_DECRYPTED) &&
1464 !(n->m_flags & M_EXTPG) &&
1465 m->m_len <= MCLBYTES / 4 && /* XXX: Don't copy too much */
1466 m->m_len <= M_TRAILINGSPACE(n)) {
1467 m_copydata(m, 0, m->m_len, mtodo(n, n->m_len));
1468 n->m_len += m->m_len;
1469 sb->sb_ccc += m->m_len;
1470 sb->sb_tlscc += m->m_len;
1471 m = m_free(m);
1472 continue;
1473 }
1474 if (n)
1475 n->m_next = m;
1476 else
1477 sb->sb_mtls = m;
1478 sb->sb_mtlstail = m;
1479 sballoc_ktls_rx(sb, m);
1480 n = m;
1481 m = m->m_next;
1482 n->m_next = NULL;
1483 }
1484 SBLASTMBUFCHK(sb);
1485 }
1486 #endif
1487
1488 /*
1489 * Free all mbufs in a sockbuf. Check that all resources are reclaimed.
1490 */
1491 static void
1492 sbflush_internal(struct sockbuf *sb)
1493 {
1494
1495 while (sb->sb_mbcnt || sb->sb_tlsdcc) {
1496 /*
1497 * Don't call sbcut(sb, 0) if the leading mbuf is non-empty:
1498 * we would loop forever. Panic instead.
1499 */
1500 if (sb->sb_ccc == 0 && (sb->sb_mb == NULL || sb->sb_mb->m_len))
1501 break;
1502 m_freem(sbcut_internal(sb, (int)sb->sb_ccc));
1503 }
1504 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
1505 ("%s: ccc %u mb %p mbcnt %u", __func__,
1506 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
1507 }
1508
1509 void
1510 sbflush_locked(struct sockbuf *sb)
1511 {
1512
1513 SOCKBUF_LOCK_ASSERT(sb);
1514 sbflush_internal(sb);
1515 }
1516
1517 void
1518 sbflush(struct sockbuf *sb)
1519 {
1520
1521 SOCKBUF_LOCK(sb);
1522 sbflush_locked(sb);
1523 SOCKBUF_UNLOCK(sb);
1524 }
1525
1526 /*
1527 * Cut data from (the front of) a sockbuf.
1528 */
1529 static struct mbuf *
1530 sbcut_internal(struct sockbuf *sb, int len)
1531 {
1532 struct mbuf *m, *next, *mfree;
1533 bool is_tls;
1534
1535 KASSERT(len >= 0, ("%s: len is %d but it is supposed to be >= 0",
1536 __func__, len));
1537 KASSERT(len <= sb->sb_ccc, ("%s: len: %d is > ccc: %u",
1538 __func__, len, sb->sb_ccc));
1539
1540 next = (m = sb->sb_mb) ? m->m_nextpkt : 0;
1541 is_tls = false;
1542 mfree = NULL;
1543
1544 while (len > 0) {
1545 if (m == NULL) {
1546 #ifdef KERN_TLS
1547 if (next == NULL && !is_tls) {
1548 if (sb->sb_tlsdcc != 0) {
1549 MPASS(len >= sb->sb_tlsdcc);
1550 len -= sb->sb_tlsdcc;
1551 sb->sb_ccc -= sb->sb_tlsdcc;
1552 sb->sb_tlsdcc = 0;
1553 if (len == 0)
1554 break;
1555 }
1556 next = sb->sb_mtls;
1557 is_tls = true;
1558 }
1559 #endif
1560 KASSERT(next, ("%s: no next, len %d", __func__, len));
1561 m = next;
1562 next = m->m_nextpkt;
1563 }
1564 if (m->m_len > len) {
1565 KASSERT(!(m->m_flags & M_NOTAVAIL),
1566 ("%s: m %p M_NOTAVAIL", __func__, m));
1567 m->m_len -= len;
1568 m->m_data += len;
1569 sb->sb_ccc -= len;
1570 sb->sb_acc -= len;
1571 if (sb->sb_sndptroff != 0)
1572 sb->sb_sndptroff -= len;
1573 if (m->m_type != MT_DATA && m->m_type != MT_OOBDATA)
1574 sb->sb_ctl -= len;
1575 break;
1576 }
1577 len -= m->m_len;
1578 #ifdef KERN_TLS
1579 if (is_tls)
1580 sbfree_ktls_rx(sb, m);
1581 else
1582 #endif
1583 sbfree(sb, m);
1584 /*
1585 * Do not put M_NOTREADY buffers to the free list, they
1586 * are referenced from outside.
1587 */
1588 if (m->m_flags & M_NOTREADY && !is_tls)
1589 m = m->m_next;
1590 else {
1591 struct mbuf *n;
1592
1593 n = m->m_next;
1594 m->m_next = mfree;
1595 mfree = m;
1596 m = n;
1597 }
1598 }
1599 /*
1600 * Free any zero-length mbufs from the buffer.
1601 * For SOCK_DGRAM sockets such mbufs represent empty records.
1602 * XXX: For SOCK_STREAM sockets such mbufs can appear in the buffer,
1603 * when sosend_generic() needs to send only control data.
1604 */
1605 while (m && m->m_len == 0) {
1606 struct mbuf *n;
1607
1608 sbfree(sb, m);
1609 n = m->m_next;
1610 m->m_next = mfree;
1611 mfree = m;
1612 m = n;
1613 }
1614 #ifdef KERN_TLS
1615 if (is_tls) {
1616 sb->sb_mb = NULL;
1617 sb->sb_mtls = m;
1618 if (m == NULL)
1619 sb->sb_mtlstail = NULL;
1620 } else
1621 #endif
1622 if (m) {
1623 sb->sb_mb = m;
1624 m->m_nextpkt = next;
1625 } else
1626 sb->sb_mb = next;
1627 /*
1628 * First part is an inline SB_EMPTY_FIXUP(). Second part makes sure
1629 * sb_lastrecord is up-to-date if we dropped part of the last record.
1630 */
1631 m = sb->sb_mb;
1632 if (m == NULL) {
1633 sb->sb_mbtail = NULL;
1634 sb->sb_lastrecord = NULL;
1635 } else if (m->m_nextpkt == NULL) {
1636 sb->sb_lastrecord = m;
1637 }
1638
1639 return (mfree);
1640 }
1641
1642 /*
1643 * Drop data from (the front of) a sockbuf.
1644 */
1645 void
1646 sbdrop_locked(struct sockbuf *sb, int len)
1647 {
1648
1649 SOCKBUF_LOCK_ASSERT(sb);
1650 m_freem(sbcut_internal(sb, len));
1651 }
1652
1653 /*
1654 * Drop data from (the front of) a sockbuf,
1655 * and return it to caller.
1656 */
1657 struct mbuf *
1658 sbcut_locked(struct sockbuf *sb, int len)
1659 {
1660
1661 SOCKBUF_LOCK_ASSERT(sb);
1662 return (sbcut_internal(sb, len));
1663 }
1664
1665 void
1666 sbdrop(struct sockbuf *sb, int len)
1667 {
1668 struct mbuf *mfree;
1669
1670 SOCKBUF_LOCK(sb);
1671 mfree = sbcut_internal(sb, len);
1672 SOCKBUF_UNLOCK(sb);
1673
1674 m_freem(mfree);
1675 }
1676
1677 struct mbuf *
1678 sbsndptr_noadv(struct sockbuf *sb, uint32_t off, uint32_t *moff)
1679 {
1680 struct mbuf *m;
1681
1682 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1683 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1684 *moff = off;
1685 if (sb->sb_sndptr == NULL) {
1686 sb->sb_sndptr = sb->sb_mb;
1687 sb->sb_sndptroff = 0;
1688 }
1689 return (sb->sb_mb);
1690 } else {
1691 m = sb->sb_sndptr;
1692 off -= sb->sb_sndptroff;
1693 }
1694 *moff = off;
1695 return (m);
1696 }
1697
1698 void
1699 sbsndptr_adv(struct sockbuf *sb, struct mbuf *mb, uint32_t len)
1700 {
1701 /*
1702 * A small copy was done, advance forward the sb_sbsndptr to cover
1703 * it.
1704 */
1705 struct mbuf *m;
1706
1707 if (mb != sb->sb_sndptr) {
1708 /* Did not copyout at the same mbuf */
1709 return;
1710 }
1711 m = mb;
1712 while (m && (len > 0)) {
1713 if (len >= m->m_len) {
1714 len -= m->m_len;
1715 if (m->m_next) {
1716 sb->sb_sndptroff += m->m_len;
1717 sb->sb_sndptr = m->m_next;
1718 }
1719 m = m->m_next;
1720 } else {
1721 len = 0;
1722 }
1723 }
1724 }
1725
1726 /*
1727 * Return the first mbuf and the mbuf data offset for the provided
1728 * send offset without changing the "sb_sndptroff" field.
1729 */
1730 struct mbuf *
1731 sbsndmbuf(struct sockbuf *sb, u_int off, u_int *moff)
1732 {
1733 struct mbuf *m;
1734
1735 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb is NULL", __func__));
1736
1737 /*
1738 * If the "off" is below the stored offset, which happens on
1739 * retransmits, just use "sb_mb":
1740 */
1741 if (sb->sb_sndptr == NULL || sb->sb_sndptroff > off) {
1742 m = sb->sb_mb;
1743 } else {
1744 m = sb->sb_sndptr;
1745 off -= sb->sb_sndptroff;
1746 }
1747 while (off > 0 && m != NULL) {
1748 if (off < m->m_len)
1749 break;
1750 off -= m->m_len;
1751 m = m->m_next;
1752 }
1753 *moff = off;
1754 return (m);
1755 }
1756
1757 /*
1758 * Drop a record off the front of a sockbuf and move the next record to the
1759 * front.
1760 */
1761 void
1762 sbdroprecord_locked(struct sockbuf *sb)
1763 {
1764 struct mbuf *m;
1765
1766 SOCKBUF_LOCK_ASSERT(sb);
1767
1768 m = sb->sb_mb;
1769 if (m) {
1770 sb->sb_mb = m->m_nextpkt;
1771 do {
1772 sbfree(sb, m);
1773 m = m_free(m);
1774 } while (m);
1775 }
1776 SB_EMPTY_FIXUP(sb);
1777 }
1778
1779 /*
1780 * Drop a record off the front of a sockbuf and move the next record to the
1781 * front.
1782 */
1783 void
1784 sbdroprecord(struct sockbuf *sb)
1785 {
1786
1787 SOCKBUF_LOCK(sb);
1788 sbdroprecord_locked(sb);
1789 SOCKBUF_UNLOCK(sb);
1790 }
1791
1792 /*
1793 * Create a "control" mbuf containing the specified data with the specified
1794 * type for presentation on a socket buffer.
1795 */
1796 struct mbuf *
1797 sbcreatecontrol(const void *p, u_int size, int type, int level, int wait)
1798 {
1799 struct cmsghdr *cp;
1800 struct mbuf *m;
1801
1802 MBUF_CHECKSLEEP(wait);
1803
1804 if (wait == M_NOWAIT) {
1805 if (CMSG_SPACE(size) > MCLBYTES)
1806 return (NULL);
1807 } else
1808 KASSERT(CMSG_SPACE(size) <= MCLBYTES,
1809 ("%s: passed CMSG_SPACE(%u) > MCLBYTES", __func__, size));
1810
1811 if (CMSG_SPACE(size) > MLEN)
1812 m = m_getcl(wait, MT_CONTROL, 0);
1813 else
1814 m = m_get(wait, MT_CONTROL);
1815 if (m == NULL)
1816 return (NULL);
1817
1818 KASSERT(CMSG_SPACE(size) <= M_TRAILINGSPACE(m),
1819 ("sbcreatecontrol: short mbuf"));
1820 /*
1821 * Don't leave the padding between the msg header and the
1822 * cmsg data and the padding after the cmsg data un-initialized.
1823 */
1824 cp = mtod(m, struct cmsghdr *);
1825 bzero(cp, CMSG_SPACE(size));
1826 if (p != NULL)
1827 (void)memcpy(CMSG_DATA(cp), p, size);
1828 m->m_len = CMSG_SPACE(size);
1829 cp->cmsg_len = CMSG_LEN(size);
1830 cp->cmsg_level = level;
1831 cp->cmsg_type = type;
1832 return (m);
1833 }
1834
1835 /*
1836 * This does the same for socket buffers that sotoxsocket does for sockets:
1837 * generate an user-format data structure describing the socket buffer. Note
1838 * that the xsockbuf structure, since it is always embedded in a socket, does
1839 * not include a self pointer nor a length. We make this entry point public
1840 * in case some other mechanism needs it.
1841 */
1842 void
1843 sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb)
1844 {
1845
1846 xsb->sb_cc = sb->sb_ccc;
1847 xsb->sb_hiwat = sb->sb_hiwat;
1848 xsb->sb_mbcnt = sb->sb_mbcnt;
1849 xsb->sb_mbmax = sb->sb_mbmax;
1850 xsb->sb_lowat = sb->sb_lowat;
1851 xsb->sb_flags = sb->sb_flags;
1852 xsb->sb_timeo = sb->sb_timeo;
1853 }
1854
1855 /* This takes the place of kern.maxsockbuf, which moved to kern.ipc. */
1856 static int dummy;
1857 SYSCTL_INT(_kern, KERN_DUMMY, dummy, CTLFLAG_RW | CTLFLAG_SKIP, &dummy, 0, "");
1858 SYSCTL_OID(_kern_ipc, KIPC_MAXSOCKBUF, maxsockbuf,
1859 CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, &sb_max, 0,
1860 sysctl_handle_sb_max, "LU",
1861 "Maximum socket buffer size");
1862 SYSCTL_ULONG(_kern_ipc, KIPC_SOCKBUF_WASTE, sockbuf_waste_factor, CTLFLAG_RW,
1863 &sb_efficiency, 0, "Socket buffer size waste factor");
Cache object: 3650f998b29d92ec5ca66192e22497d1
|