xref: /original-bsd/sys/kern/uipc_mbuf.c (revision e59fb703)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1991 Regents of the University of California.
3  * All rights reserved.
4  *
5  * %sccs.include.redist.c%
6  *
7  *	@(#)uipc_mbuf.c	7.21 (Berkeley) 12/19/91
8  */
9 
10 #include "param.h"
11 #include "proc.h"
12 #include "malloc.h"
13 #include "map.h"
14 #define MBTYPES
15 #include "mbuf.h"
16 #include "kernel.h"
17 #include "syslog.h"
18 #include "domain.h"
19 #include "protosw.h"
20 #include "vm/vm.h"
21 
22 extern	vm_map_t mb_map;
23 struct	mbuf *mbutl;
24 char	*mclrefcnt;
25 
26 mbinit()
27 {
28 	int s;
29 
30 #if CLBYTES < 4096
31 #define NCL_INIT	(4096/CLBYTES)
32 #else
33 #define NCL_INIT	1
34 #endif
35 	s = splimp();
36 	if (m_clalloc(NCL_INIT, M_DONTWAIT) == 0)
37 		goto bad;
38 	splx(s);
39 	return;
40 bad:
41 	panic("mbinit");
42 }
43 
44 /*
45  * Allocate some number of mbuf clusters
46  * and place on cluster free list.
47  * Must be called at splimp.
48  */
49 /* ARGSUSED */
50 m_clalloc(ncl, nowait)
51 	register int ncl;
52 	int nowait;
53 {
54 	int npg, mbx;
55 	register caddr_t p;
56 	register int i;
57 	static int logged;
58 
59 	npg = ncl * CLSIZE;
60 	p = (caddr_t)kmem_malloc(mb_map, ctob(npg), !nowait);
61 	if (p == NULL) {
62 		if (logged == 0) {
63 			logged++;
64 			log(LOG_ERR, "mb_map full\n");
65 		}
66 		return (0);
67 	}
68 	ncl = ncl * CLBYTES / MCLBYTES;
69 	for (i = 0; i < ncl; i++) {
70 		((union mcluster *)p)->mcl_next = mclfree;
71 		mclfree = (union mcluster *)p;
72 		p += MCLBYTES;
73 		mbstat.m_clfree++;
74 	}
75 	mbstat.m_clusters += ncl;
76 	return (1);
77 }
78 
79 /*
80  * When MGET failes, ask protocols to free space when short of memory,
81  * then re-attempt to allocate an mbuf.
82  */
83 struct mbuf *
84 m_retry(i, t)
85 	int i, t;
86 {
87 	register struct mbuf *m;
88 
89 	m_reclaim();
90 #define m_retry(i, t)	(struct mbuf *)0
91 	MGET(m, i, t);
92 #undef m_retry
93 	return (m);
94 }
95 
96 /*
97  * As above; retry an MGETHDR.
98  */
99 struct mbuf *
100 m_retryhdr(i, t)
101 	int i, t;
102 {
103 	register struct mbuf *m;
104 
105 	m_reclaim();
106 #define m_retryhdr(i, t) (struct mbuf *)0
107 	MGETHDR(m, i, t);
108 #undef m_retryhdr
109 	return (m);
110 }
111 
112 m_reclaim()
113 {
114 	register struct domain *dp;
115 	register struct protosw *pr;
116 	int s = splimp();
117 
118 	for (dp = domains; dp; dp = dp->dom_next)
119 		for (pr = dp->dom_protosw; pr < dp->dom_protoswNPROTOSW; pr++)
120 			if (pr->pr_drain)
121 				(*pr->pr_drain)();
122 	splx(s);
123 	mbstat.m_drain++;
124 }
125 
126 /*
127  * Space allocation routines.
128  * These are also available as macros
129  * for critical paths.
130  */
131 struct mbuf *
132 m_get(nowait, type)
133 	int nowait, type;
134 {
135 	register struct mbuf *m;
136 
137 	MGET(m, nowait, type);
138 	return (m);
139 }
140 
141 struct mbuf *
142 m_gethdr(nowait, type)
143 	int nowait, type;
144 {
145 	register struct mbuf *m;
146 
147 	MGETHDR(m, nowait, type);
148 	return (m);
149 }
150 
151 struct mbuf *
152 m_getclr(nowait, type)
153 	int nowait, type;
154 {
155 	register struct mbuf *m;
156 
157 	MGET(m, nowait, type);
158 	if (m == 0)
159 		return (0);
160 	bzero(mtod(m, caddr_t), MLEN);
161 	return (m);
162 }
163 
164 struct mbuf *
165 m_free(m)
166 	struct mbuf *m;
167 {
168 	register struct mbuf *n;
169 
170 	MFREE(m, n);
171 	return (n);
172 }
173 
174 m_freem(m)
175 	register struct mbuf *m;
176 {
177 	register struct mbuf *n;
178 
179 	if (m == NULL)
180 		return;
181 	do {
182 		MFREE(m, n);
183 	} while (m = n);
184 }
185 
186 /*
187  * Mbuffer utility routines.
188  */
189 
190 /*
191  * Lesser-used path for M_PREPEND:
192  * allocate new mbuf to prepend to chain,
193  * copy junk along.
194  */
195 struct mbuf *
196 m_prepend(m, len, how)
197 	register struct mbuf *m;
198 	int len, how;
199 {
200 	struct mbuf *mn;
201 
202 	MGET(mn, how, m->m_type);
203 	if (mn == (struct mbuf *)NULL) {
204 		m_freem(m);
205 		return ((struct mbuf *)NULL);
206 	}
207 	if (m->m_flags & M_PKTHDR) {
208 		M_COPY_PKTHDR(mn, m);
209 		m->m_flags &= ~M_PKTHDR;
210 	}
211 	mn->m_next = m;
212 	m = mn;
213 	if (len < MHLEN)
214 		MH_ALIGN(m, len);
215 	m->m_len = len;
216 	return (m);
217 }
218 
219 /*
220  * Make a copy of an mbuf chain starting "off0" bytes from the beginning,
221  * continuing for "len" bytes.  If len is M_COPYALL, copy to end of mbuf.
222  * The wait parameter is a choice of M_WAIT/M_DONTWAIT from caller.
223  */
224 int MCFail;
225 
226 struct mbuf *
227 m_copym(m, off0, len, wait)
228 	register struct mbuf *m;
229 	int off0, wait;
230 	register int len;
231 {
232 	register struct mbuf *n, **np;
233 	register int off = off0;
234 	struct mbuf *top;
235 	int copyhdr = 0;
236 
237 	if (off < 0 || len < 0)
238 		panic("m_copym");
239 	if (off == 0 && m->m_flags & M_PKTHDR)
240 		copyhdr = 1;
241 	while (off > 0) {
242 		if (m == 0)
243 			panic("m_copym");
244 		if (off < m->m_len)
245 			break;
246 		off -= m->m_len;
247 		m = m->m_next;
248 	}
249 	np = &top;
250 	top = 0;
251 	while (len > 0) {
252 		if (m == 0) {
253 			if (len != M_COPYALL)
254 				panic("m_copym");
255 			break;
256 		}
257 		MGET(n, wait, m->m_type);
258 		*np = n;
259 		if (n == 0)
260 			goto nospace;
261 		if (copyhdr) {
262 			M_COPY_PKTHDR(n, m);
263 			if (len == M_COPYALL)
264 				n->m_pkthdr.len -= off0;
265 			else
266 				n->m_pkthdr.len = len;
267 			copyhdr = 0;
268 		}
269 		n->m_len = MIN(len, m->m_len - off);
270 		if (m->m_flags & M_EXT) {
271 			n->m_data = m->m_data + off;
272 			mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
273 			n->m_ext = m->m_ext;
274 			n->m_flags |= M_EXT;
275 		} else
276 			bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t),
277 			    (unsigned)n->m_len);
278 		if (len != M_COPYALL)
279 			len -= n->m_len;
280 		off = 0;
281 		m = m->m_next;
282 		np = &n->m_next;
283 	}
284 	if (top == 0)
285 		MCFail++;
286 	return (top);
287 nospace:
288 	m_freem(top);
289 	MCFail++;
290 	return (0);
291 }
292 
293 /*
294  * Copy data from an mbuf chain starting "off" bytes from the beginning,
295  * continuing for "len" bytes, into the indicated buffer.
296  */
297 m_copydata(m, off, len, cp)
298 	register struct mbuf *m;
299 	register int off;
300 	register int len;
301 	caddr_t cp;
302 {
303 	register unsigned count;
304 
305 	if (off < 0 || len < 0)
306 		panic("m_copydata");
307 	while (off > 0) {
308 		if (m == 0)
309 			panic("m_copydata");
310 		if (off < m->m_len)
311 			break;
312 		off -= m->m_len;
313 		m = m->m_next;
314 	}
315 	while (len > 0) {
316 		if (m == 0)
317 			panic("m_copydata");
318 		count = MIN(m->m_len - off, len);
319 		bcopy(mtod(m, caddr_t) + off, cp, count);
320 		len -= count;
321 		cp += count;
322 		off = 0;
323 		m = m->m_next;
324 	}
325 }
326 
327 /*
328  * Concatenate mbuf chain n to m.
329  * Both chains must be of the same type (e.g. MT_DATA).
330  * Any m_pkthdr is not updated.
331  */
332 m_cat(m, n)
333 	register struct mbuf *m, *n;
334 {
335 	while (m->m_next)
336 		m = m->m_next;
337 	while (n) {
338 		if (m->m_flags & M_EXT ||
339 		    m->m_data + m->m_len + n->m_len >= &m->m_dat[MLEN]) {
340 			/* just join the two chains */
341 			m->m_next = n;
342 			return;
343 		}
344 		/* splat the data from one into the other */
345 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
346 		    (u_int)n->m_len);
347 		m->m_len += n->m_len;
348 		n = m_free(n);
349 	}
350 }
351 
352 m_adj(mp, req_len)
353 	struct mbuf *mp;
354 {
355 	register int len = req_len;
356 	register struct mbuf *m;
357 	register count;
358 
359 	if ((m = mp) == NULL)
360 		return;
361 	if (len >= 0) {
362 		/*
363 		 * Trim from head.
364 		 */
365 		while (m != NULL && len > 0) {
366 			if (m->m_len <= len) {
367 				len -= m->m_len;
368 				m->m_len = 0;
369 				m = m->m_next;
370 			} else {
371 				m->m_len -= len;
372 				m->m_data += len;
373 				len = 0;
374 			}
375 		}
376 		m = mp;
377 		if (mp->m_flags & M_PKTHDR)
378 			m->m_pkthdr.len -= (req_len - len);
379 	} else {
380 		/*
381 		 * Trim from tail.  Scan the mbuf chain,
382 		 * calculating its length and finding the last mbuf.
383 		 * If the adjustment only affects this mbuf, then just
384 		 * adjust and return.  Otherwise, rescan and truncate
385 		 * after the remaining size.
386 		 */
387 		len = -len;
388 		count = 0;
389 		for (;;) {
390 			count += m->m_len;
391 			if (m->m_next == (struct mbuf *)0)
392 				break;
393 			m = m->m_next;
394 		}
395 		if (m->m_len >= len) {
396 			m->m_len -= len;
397 			if ((mp = m)->m_flags & M_PKTHDR)
398 				m->m_pkthdr.len -= len;
399 			return;
400 		}
401 		count -= len;
402 		if (count < 0)
403 			count = 0;
404 		/*
405 		 * Correct length for chain is "count".
406 		 * Find the mbuf with last data, adjust its length,
407 		 * and toss data from remaining mbufs on chain.
408 		 */
409 		m = mp;
410 		if (m->m_flags & M_PKTHDR)
411 			m->m_pkthdr.len = count;
412 		for (; m; m = m->m_next) {
413 			if (m->m_len >= count) {
414 				m->m_len = count;
415 				break;
416 			}
417 			count -= m->m_len;
418 		}
419 		while (m = m->m_next)
420 			m->m_len = 0;
421 	}
422 }
423 
424 /*
425  * Rearange an mbuf chain so that len bytes are contiguous
426  * and in the data area of an mbuf (so that mtod and dtom
427  * will work for a structure of size len).  Returns the resulting
428  * mbuf chain on success, frees it and returns null on failure.
429  * If there is room, it will add up to max_protohdr-len extra bytes to the
430  * contiguous region in an attempt to avoid being called next time.
431  */
432 int MPFail;
433 
434 struct mbuf *
435 m_pullup(n, len)
436 	register struct mbuf *n;
437 	int len;
438 {
439 	register struct mbuf *m;
440 	register int count;
441 	int space;
442 
443 	/*
444 	 * If first mbuf has no cluster, and has room for len bytes
445 	 * without shifting current data, pullup into it,
446 	 * otherwise allocate a new mbuf to prepend to the chain.
447 	 */
448 	if ((n->m_flags & M_EXT) == 0 &&
449 	    n->m_data + len < &n->m_dat[MLEN] && n->m_next) {
450 		if (n->m_len >= len)
451 			return (n);
452 		m = n;
453 		n = n->m_next;
454 		len -= m->m_len;
455 	} else {
456 		if (len > MHLEN)
457 			goto bad;
458 		MGET(m, M_DONTWAIT, n->m_type);
459 		if (m == 0)
460 			goto bad;
461 		m->m_len = 0;
462 		if (n->m_flags & M_PKTHDR) {
463 			M_COPY_PKTHDR(m, n);
464 			n->m_flags &= ~M_PKTHDR;
465 		}
466 	}
467 	space = &m->m_dat[MLEN] - (m->m_data + m->m_len);
468 	do {
469 		count = min(min(max(len, max_protohdr), space), n->m_len);
470 		bcopy(mtod(n, caddr_t), mtod(m, caddr_t) + m->m_len,
471 		  (unsigned)count);
472 		len -= count;
473 		m->m_len += count;
474 		n->m_len -= count;
475 		space -= count;
476 		if (n->m_len)
477 			n->m_data += count;
478 		else
479 			n = m_free(n);
480 	} while (len > 0 && n);
481 	if (len > 0) {
482 		(void) m_free(m);
483 		goto bad;
484 	}
485 	m->m_next = n;
486 	return (m);
487 bad:
488 	m_freem(n);
489 	MPFail++;
490 	return (0);
491 }
492 
493 /*
494  * Partition an mbuf chain in two pieces, returning the tail --
495  * all but the first len0 bytes.  In case of failure, it returns NULL and
496  * attempts to restore the chain to its original state.
497  */
498 struct mbuf *
499 m_split(m0, len0, wait)
500 register struct mbuf *m0;
501 int len0;
502 {
503 	register struct mbuf *m, *n;
504 	unsigned len = len0, remain;
505 
506 	for (m = m0; m && len > m->m_len; m = m->m_next)
507 		len -= m->m_len;
508 	if (m == 0)
509 		return (0);
510 	remain = m->m_len - len;
511 	if (m0->m_flags & M_PKTHDR) {
512 		MGETHDR(n, wait, m0->m_type);
513 		if (n == 0)
514 			return (0);
515 		n->m_pkthdr.rcvif = m0->m_pkthdr.rcvif;
516 		n->m_pkthdr.len = m0->m_pkthdr.len - len0;
517 		m0->m_pkthdr.len = len0;
518 		if (m->m_flags & M_EXT)
519 			goto extpacket;
520 		if (remain > MHLEN) {
521 			/* m can't be the lead packet */
522 			MH_ALIGN(n, 0);
523 			n->m_next = m_split(m, len, wait);
524 			if (n->m_next == 0) {
525 				(void) m_free(n);
526 				return (0);
527 			} else
528 				return (n);
529 		} else
530 			MH_ALIGN(n, remain);
531 	} else if (remain == 0) {
532 		n = m->m_next;
533 		m->m_next = 0;
534 		return (n);
535 	} else {
536 		MGET(n, wait, m->m_type);
537 		if (n == 0)
538 			return (0);
539 		M_ALIGN(n, remain);
540 	}
541 extpacket:
542 	if (m->m_flags & M_EXT) {
543 		n->m_flags |= M_EXT;
544 		n->m_ext = m->m_ext;
545 		mclrefcnt[mtocl(m->m_ext.ext_buf)]++;
546 		m->m_ext.ext_size = 0; /* For Accounting XXXXXX danger */
547 		n->m_data = m->m_data + len;
548 	} else {
549 		bcopy(mtod(m, caddr_t) + len, mtod(n, caddr_t), remain);
550 	}
551 	n->m_len = remain;
552 	m->m_len = len;
553 	n->m_next = m->m_next;
554 	m->m_next = 0;
555 	return (n);
556 }
557 /*
558  * Routine to copy from device local memory into mbufs.
559  */
560 struct mbuf *
561 m_devget(buf, totlen, off0, ifp, copy)
562 	char *buf;
563 	int totlen, off0;
564 	struct ifnet *ifp;
565 	void (*copy)();
566 {
567 	register struct mbuf *m;
568 	struct mbuf *top = 0, **mp = &top;
569 	register int off = off0, len;
570 	register char *cp;
571 	char *epkt;
572 
573 	cp = buf;
574 	epkt = cp + totlen;
575 	if (off) {
576 		cp += off + 2 * sizeof(u_short);
577 		totlen -= 2 * sizeof(u_short);
578 	}
579 	MGETHDR(m, M_DONTWAIT, MT_DATA);
580 	if (m == 0)
581 		return (0);
582 	m->m_pkthdr.rcvif = ifp;
583 	m->m_pkthdr.len = totlen;
584 	m->m_len = MHLEN;
585 
586 	while (totlen > 0) {
587 		if (top) {
588 			MGET(m, M_DONTWAIT, MT_DATA);
589 			if (m == 0) {
590 				m_freem(top);
591 				return (0);
592 			}
593 			m->m_len = MLEN;
594 		}
595 		len = min(totlen, epkt - cp);
596 		if (len >= MINCLSIZE) {
597 			MCLGET(m, M_DONTWAIT);
598 			if (m->m_flags & M_EXT)
599 				m->m_len = len = min(len, MCLBYTES);
600 			else
601 				len = m->m_len;
602 		} else {
603 			/*
604 			 * Place initial small packet/header at end of mbuf.
605 			 */
606 			if (len < m->m_len) {
607 				if (top == 0 && len + max_linkhdr <= m->m_len)
608 					m->m_data += max_linkhdr;
609 				m->m_len = len;
610 			} else
611 				len = m->m_len;
612 		}
613 		if (copy)
614 			copy(cp, mtod(m, caddr_t), (unsigned)len);
615 		else
616 			bcopy(cp, mtod(m, caddr_t), (unsigned)len);
617 		cp += len;
618 		*mp = m;
619 		mp = &m->m_next;
620 		totlen -= len;
621 		if (cp == epkt)
622 			cp = buf;
623 	}
624 	return (top);
625 }
626