1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
22 /*	  All Rights Reserved	*/
23 
24 /*
25  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
26  * Use is subject to license terms.
27  *
28  * Copyright 2017 Nexenta Systems, Inc.  All rights reserved.
29  * Copyright 2022 Garrett D'Amore
30  */
31 
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/thread.h>
35 #include <sys/sysmacros.h>
36 #include <sys/stropts.h>
37 #include <sys/stream.h>
38 #include <sys/strsubr.h>
39 #include <sys/strsun.h>
40 #include <sys/conf.h>
41 #include <sys/debug.h>
42 #include <sys/cmn_err.h>
43 #include <sys/kmem.h>
44 #include <sys/atomic.h>
45 #include <sys/errno.h>
46 #include <sys/vtrace.h>
47 #include <sys/ftrace.h>
48 #include <sys/ontrap.h>
49 #include <sys/sdt.h>
50 #include <sys/strft.h>
51 
52 /*
53  * This file contains selected functions from io/stream.c
54  * needed by this library, mostly unmodified.
55  */
56 
57 /*
58  * STREAMS message allocator: principles of operation
59  * (See usr/src/uts/common/io/stream.c)
60  */
61 #define	DBLK_MAX_CACHE		73728
62 #define	DBLK_CACHE_ALIGN	64
63 #define	DBLK_MIN_SIZE		8
64 #define	DBLK_SIZE_SHIFT		3
65 
66 #ifdef _BIG_ENDIAN
67 #define	DBLK_RTFU_SHIFT(field)	\
68 	(8 * (&((dblk_t *)0)->db_struioflag - &((dblk_t *)0)->field))
69 #else
70 #define	DBLK_RTFU_SHIFT(field)	\
71 	(8 * (&((dblk_t *)0)->field - &((dblk_t *)0)->db_ref))
72 #endif
73 
74 #define	DBLK_RTFU(ref, type, flags, uioflag)	\
75 	(((ref) << DBLK_RTFU_SHIFT(db_ref)) | \
76 	((type) << DBLK_RTFU_SHIFT(db_type)) | \
77 	(((flags) | (ref - 1)) << DBLK_RTFU_SHIFT(db_flags)) | \
78 	((uioflag) << DBLK_RTFU_SHIFT(db_struioflag)))
79 #define	DBLK_RTFU_REF_MASK	(DBLK_REFMAX << DBLK_RTFU_SHIFT(db_ref))
80 #define	DBLK_RTFU_WORD(dbp)	(*((uint32_t *)&(dbp)->db_ref))
81 #define	MBLK_BAND_FLAG_WORD(mp)	(*((uint32_t *)&(mp)->b_band))
82 
83 static size_t dblk_sizes[] = {
84 #ifdef _LP64
85 	16, 80, 144, 208, 272, 336, 528, 1040, 1488, 1936, 2576, 3856,
86 	8192, 12048, 16384, 20240, 24576, 28432, 32768, 36624,
87 	40960, 44816, 49152, 53008, 57344, 61200, 65536, 69392,
88 #else
89 	64, 128, 320, 576, 1088, 1536, 1984, 2624, 3904,
90 	8192, 12096, 16384, 20288, 24576, 28480, 32768, 36672,
91 	40960, 44864, 49152, 53056, 57344, 61248, 65536, 69440,
92 #endif
93 	DBLK_MAX_CACHE, 0
94 };
95 
96 static struct kmem_cache *dblk_cache[DBLK_MAX_CACHE / DBLK_MIN_SIZE];
97 static struct kmem_cache *mblk_cache;
98 static struct kmem_cache *dblk_esb_cache;
99 
100 static void dblk_lastfree(mblk_t *mp, dblk_t *dbp);
101 static mblk_t *allocb_oversize(size_t size, int flags);
102 static int allocb_tryhard_fails;
103 static void frnop_func(void *arg);
104 frtn_t frnop = { frnop_func };
105 static void bcache_dblk_lastfree(mblk_t *mp, dblk_t *dbp);
106 
107 /*
108  * Patchable mblk/dblk kmem_cache flags.
109  */
110 int dblk_kmem_flags = 0;
111 int mblk_kmem_flags = 0;
112 
113 static int
dblk_constructor(void * buf,void * cdrarg,int kmflags)114 dblk_constructor(void *buf, void *cdrarg, int kmflags)
115 {
116 	dblk_t *dbp = buf;
117 	ssize_t msg_size = (ssize_t)cdrarg;
118 	size_t index;
119 
120 	ASSERT(msg_size != 0);
121 
122 	index = (msg_size - 1) >> DBLK_SIZE_SHIFT;
123 
124 	ASSERT(index < (DBLK_MAX_CACHE >> DBLK_SIZE_SHIFT));
125 
126 	if ((dbp->db_mblk = kmem_cache_alloc(mblk_cache, kmflags)) == NULL)
127 		return (-1);
128 	if ((msg_size & PAGEOFFSET) == 0) {
129 		dbp->db_base = kmem_alloc(msg_size, kmflags);
130 		if (dbp->db_base == NULL) {
131 			kmem_cache_free(mblk_cache, dbp->db_mblk);
132 			return (-1);
133 		}
134 	} else {
135 		dbp->db_base = (unsigned char *)&dbp[1];
136 	}
137 
138 	dbp->db_mblk->b_datap = dbp;
139 	dbp->db_cache = dblk_cache[index];
140 	dbp->db_lim = dbp->db_base + msg_size;
141 	dbp->db_free = dbp->db_lastfree = dblk_lastfree;
142 	dbp->db_frtnp = NULL;
143 	dbp->db_fthdr = NULL;
144 	dbp->db_credp = NULL;
145 	dbp->db_cpid = -1;
146 	dbp->db_struioflag = 0;
147 	dbp->db_struioun.cksum.flags = 0;
148 	return (0);
149 }
150 
151 /*ARGSUSED*/
152 static int
dblk_esb_constructor(void * buf,void * cdrarg,int kmflags)153 dblk_esb_constructor(void *buf, void *cdrarg, int kmflags)
154 {
155 	dblk_t *dbp = buf;
156 
157 	if ((dbp->db_mblk = kmem_cache_alloc(mblk_cache, kmflags)) == NULL)
158 		return (-1);
159 	dbp->db_mblk->b_datap = dbp;
160 	dbp->db_cache = dblk_esb_cache;
161 	dbp->db_fthdr = NULL;
162 	dbp->db_credp = NULL;
163 	dbp->db_cpid = -1;
164 	dbp->db_struioflag = 0;
165 	dbp->db_struioun.cksum.flags = 0;
166 	return (0);
167 }
168 
169 static int
bcache_dblk_constructor(void * buf,void * cdrarg,int kmflags)170 bcache_dblk_constructor(void *buf, void *cdrarg, int kmflags)
171 {
172 	dblk_t *dbp = buf;
173 	bcache_t *bcp = cdrarg;
174 
175 	if ((dbp->db_mblk = kmem_cache_alloc(mblk_cache, kmflags)) == NULL)
176 		return (-1);
177 
178 	dbp->db_base = kmem_cache_alloc(bcp->buffer_cache, kmflags);
179 	if (dbp->db_base == NULL) {
180 		kmem_cache_free(mblk_cache, dbp->db_mblk);
181 		return (-1);
182 	}
183 
184 	dbp->db_mblk->b_datap = dbp;
185 	dbp->db_cache = (void *)bcp;
186 	dbp->db_lim = dbp->db_base + bcp->size;
187 	dbp->db_free = dbp->db_lastfree = bcache_dblk_lastfree;
188 	dbp->db_frtnp = NULL;
189 	dbp->db_fthdr = NULL;
190 	dbp->db_credp = NULL;
191 	dbp->db_cpid = -1;
192 	dbp->db_struioflag = 0;
193 	dbp->db_struioun.cksum.flags = 0;
194 	return (0);
195 }
196 
197 /*ARGSUSED*/
198 static void
dblk_destructor(void * buf,void * cdrarg)199 dblk_destructor(void *buf, void *cdrarg)
200 {
201 	dblk_t *dbp = buf;
202 	ssize_t msg_size = (ssize_t)cdrarg;
203 
204 	ASSERT(dbp->db_mblk->b_datap == dbp);
205 	ASSERT(msg_size != 0);
206 	ASSERT(dbp->db_struioflag == 0);
207 	ASSERT(dbp->db_struioun.cksum.flags == 0);
208 
209 	if ((msg_size & PAGEOFFSET) == 0) {
210 		kmem_free(dbp->db_base, msg_size);
211 	}
212 
213 	kmem_cache_free(mblk_cache, dbp->db_mblk);
214 }
215 
216 static void
bcache_dblk_destructor(void * buf,void * cdrarg)217 bcache_dblk_destructor(void *buf, void *cdrarg)
218 {
219 	dblk_t *dbp = buf;
220 	bcache_t *bcp = cdrarg;
221 
222 	kmem_cache_free(bcp->buffer_cache, dbp->db_base);
223 
224 	ASSERT(dbp->db_mblk->b_datap == dbp);
225 	ASSERT(dbp->db_struioflag == 0);
226 	ASSERT(dbp->db_struioun.cksum.flags == 0);
227 
228 	kmem_cache_free(mblk_cache, dbp->db_mblk);
229 }
230 
231 /* Needed in the ASSERT below */
232 #ifdef	DEBUG
233 #ifdef	_KERNEL
234 #define	KMEM_SLAB_T_SZ	sizeof (kmem_slab_t)
235 #else	/* _KERNEL */
236 #define	KMEM_SLAB_T_SZ	64	/* fakekernel */
237 #endif	/* _KERNEL */
238 #endif	/* DEBUG */
239 
240 void
streams_msg_init(void)241 streams_msg_init(void)
242 {
243 	char name[40];
244 	size_t size;
245 	size_t lastsize = DBLK_MIN_SIZE;
246 	size_t *sizep;
247 	struct kmem_cache *cp;
248 	size_t tot_size;
249 	int offset;
250 
251 	mblk_cache = kmem_cache_create("streams_mblk", sizeof (mblk_t), 32,
252 	    NULL, NULL, NULL, NULL, NULL, mblk_kmem_flags);
253 
254 	for (sizep = dblk_sizes; (size = *sizep) != 0; sizep++) {
255 
256 		if ((offset = (size & PAGEOFFSET)) != 0) {
257 			/*
258 			 * We are in the middle of a page, dblk should
259 			 * be allocated on the same page
260 			 */
261 			tot_size = size + sizeof (dblk_t);
262 			ASSERT((offset + sizeof (dblk_t) + KMEM_SLAB_T_SZ)
263 			    < PAGESIZE);
264 			ASSERT((tot_size & (DBLK_CACHE_ALIGN - 1)) == 0);
265 
266 		} else {
267 
268 			/*
269 			 * buf size is multiple of page size, dblk and
270 			 * buffer are allocated separately.
271 			 */
272 
273 			ASSERT((size & (DBLK_CACHE_ALIGN - 1)) == 0);
274 			tot_size = sizeof (dblk_t);
275 		}
276 
277 		(void) sprintf(name, "streams_dblk_%ld", (long)size);
278 		cp = kmem_cache_create(name, tot_size, DBLK_CACHE_ALIGN,
279 		    dblk_constructor, dblk_destructor, NULL, (void *)(size),
280 		    NULL, dblk_kmem_flags);
281 
282 		while (lastsize <= size) {
283 			dblk_cache[(lastsize - 1) >> DBLK_SIZE_SHIFT] = cp;
284 			lastsize += DBLK_MIN_SIZE;
285 		}
286 	}
287 
288 	dblk_esb_cache = kmem_cache_create("streams_dblk_esb", sizeof (dblk_t),
289 	    DBLK_CACHE_ALIGN, dblk_esb_constructor, dblk_destructor, NULL,
290 	    (void *)sizeof (dblk_t), NULL, dblk_kmem_flags);
291 
292 	/* fthdr_cache, ftblk_cache, ... */
293 }
294 
295 /*ARGSUSED*/
296 mblk_t *
allocb(size_t size,uint_t pri)297 allocb(size_t size, uint_t pri)
298 {
299 	dblk_t *dbp;
300 	mblk_t *mp;
301 	size_t index;
302 
303 	index =  (size - 1)  >> DBLK_SIZE_SHIFT;
304 
305 	if (index >= (DBLK_MAX_CACHE >> DBLK_SIZE_SHIFT)) {
306 		if (size != 0) {
307 			mp = allocb_oversize(size, KM_NOSLEEP);
308 			goto out;
309 		}
310 		index = 0;
311 	}
312 
313 	if ((dbp = kmem_cache_alloc(dblk_cache[index], KM_NOSLEEP)) == NULL) {
314 		mp = NULL;
315 		goto out;
316 	}
317 
318 	mp = dbp->db_mblk;
319 	DBLK_RTFU_WORD(dbp) = DBLK_RTFU(1, M_DATA, 0, 0);
320 	mp->b_next = mp->b_prev = mp->b_cont = NULL;
321 	mp->b_rptr = mp->b_wptr = dbp->db_base;
322 	mp->b_queue = NULL;
323 	MBLK_BAND_FLAG_WORD(mp) = 0;
324 	STR_FTALLOC(&dbp->db_fthdr, FTEV_ALLOCB, size);
325 out:
326 	FTRACE_1("allocb(): mp=0x%p", (uintptr_t)mp);
327 
328 	return (mp);
329 }
330 
331 /*
332  * Allocate an mblk taking db_credp and db_cpid from the template.
333  * Allow the cred to be NULL.
334  */
335 mblk_t *
allocb_tmpl(size_t size,const mblk_t * tmpl)336 allocb_tmpl(size_t size, const mblk_t *tmpl)
337 {
338 	mblk_t *mp = allocb(size, 0);
339 
340 	if (mp != NULL) {
341 		dblk_t *src = tmpl->b_datap;
342 		dblk_t *dst = mp->b_datap;
343 		cred_t *cr;
344 		pid_t cpid;
345 
346 		cr = msg_getcred(tmpl, &cpid);
347 		if (cr != NULL)
348 			crhold(dst->db_credp = cr);
349 		dst->db_cpid = cpid;
350 		dst->db_type = src->db_type;
351 	}
352 	return (mp);
353 }
354 
355 mblk_t *
allocb_cred(size_t size,cred_t * cr,pid_t cpid)356 allocb_cred(size_t size, cred_t *cr, pid_t cpid)
357 {
358 	mblk_t *mp = allocb(size, 0);
359 
360 	ASSERT(cr != NULL);
361 	if (mp != NULL) {
362 		dblk_t *dbp = mp->b_datap;
363 
364 		crhold(dbp->db_credp = cr);
365 		dbp->db_cpid = cpid;
366 	}
367 	return (mp);
368 }
369 
370 mblk_t *
allocb_cred_wait(size_t size,uint_t flags,int * error,cred_t * cr,pid_t cpid)371 allocb_cred_wait(size_t size, uint_t flags, int *error, cred_t *cr, pid_t cpid)
372 {
373 	mblk_t *mp = allocb_wait(size, 0, flags, error);
374 
375 	ASSERT(cr != NULL);
376 	if (mp != NULL) {
377 		dblk_t *dbp = mp->b_datap;
378 
379 		crhold(dbp->db_credp = cr);
380 		dbp->db_cpid = cpid;
381 	}
382 
383 	return (mp);
384 }
385 
386 /*
387  * Extract the db_cred (and optionally db_cpid) from a message.
388  * We find the first mblk which has a non-NULL db_cred and use that.
389  * If none found we return NULL.
390  * Does NOT get a hold on the cred.
391  */
392 cred_t *
msg_getcred(const mblk_t * mp,pid_t * cpidp)393 msg_getcred(const mblk_t *mp, pid_t *cpidp)
394 {
395 	cred_t *cr = NULL;
396 
397 	while (mp != NULL) {
398 		dblk_t *dbp = mp->b_datap;
399 
400 		cr = dbp->db_credp;
401 		if (cr == NULL) {
402 			mp = mp->b_cont;
403 			continue;
404 		}
405 		if (cpidp != NULL)
406 			*cpidp = dbp->db_cpid;
407 
408 		/* DEBUG check for only one db_credp */
409 		return (cr);
410 	}
411 	if (cpidp != NULL)
412 		*cpidp = NOPID;
413 	return (NULL);
414 }
415 
416 /*
417  * Variant of msg_getcred which, when a cred is found
418  * 1. Returns with a hold on the cred
419  * 2. Clears the first cred in the mblk.
420  * This is more efficient to use than a msg_getcred() + crhold() when
421  * the message is freed after the cred has been extracted.
422  *
423  * The caller is responsible for ensuring that there is no other reference
424  * on the message since db_credp can not be cleared when there are other
425  * references.
426  */
427 cred_t *
msg_extractcred(mblk_t * mp,pid_t * cpidp)428 msg_extractcred(mblk_t *mp, pid_t *cpidp)
429 {
430 	cred_t *cr = NULL;
431 
432 	while (mp != NULL) {
433 		dblk_t *dbp = mp->b_datap;
434 
435 		cr = dbp->db_credp;
436 		if (cr == NULL) {
437 			mp = mp->b_cont;
438 			continue;
439 		}
440 		ASSERT(dbp->db_ref == 1);
441 		dbp->db_credp = NULL;
442 		if (cpidp != NULL)
443 			*cpidp = dbp->db_cpid;
444 
445 		/* DEBUG check for only one db_credp */
446 		return (cr);
447 	}
448 	return (NULL);
449 }
450 
451 /* _KERNEL msg_getlabel() */
452 
453 void
freeb(mblk_t * mp)454 freeb(mblk_t *mp)
455 {
456 	dblk_t *dbp = mp->b_datap;
457 
458 	ASSERT(dbp->db_ref > 0);
459 	ASSERT(mp->b_next == NULL && mp->b_prev == NULL);
460 	FTRACE_1("freeb(): mp=0x%lx", (uintptr_t)mp);
461 
462 	STR_FTEVENT_MBLK(mp, caller(), FTEV_FREEB, dbp->db_ref);
463 
464 	dbp->db_free(mp, dbp);
465 }
466 
467 void
freemsg(mblk_t * mp)468 freemsg(mblk_t *mp)
469 {
470 	FTRACE_1("freemsg(): mp=0x%lx", (uintptr_t)mp);
471 	while (mp) {
472 		dblk_t *dbp = mp->b_datap;
473 		mblk_t *mp_cont = mp->b_cont;
474 
475 		ASSERT(dbp->db_ref > 0);
476 		ASSERT(mp->b_next == NULL && mp->b_prev == NULL);
477 
478 		STR_FTEVENT_MBLK(mp, caller(), FTEV_FREEB, dbp->db_ref);
479 
480 		dbp->db_free(mp, dbp);
481 		mp = mp_cont;
482 	}
483 }
484 
485 /*
486  * Reallocate a block for another use.  Try hard to use the old block.
487  * If the old data is wanted (copy), leave b_wptr at the end of the data,
488  * otherwise return b_wptr = b_rptr.
489  *
490  * This routine is private and unstable.
491  */
492 mblk_t	*
reallocb(mblk_t * mp,size_t size,uint_t copy)493 reallocb(mblk_t *mp, size_t size, uint_t copy)
494 {
495 	mblk_t		*mp1;
496 	unsigned char	*old_rptr;
497 	ptrdiff_t	cur_size;
498 
499 	if (mp == NULL)
500 		return (allocb(size, BPRI_HI));
501 
502 	cur_size = mp->b_wptr - mp->b_rptr;
503 	old_rptr = mp->b_rptr;
504 
505 	ASSERT(mp->b_datap->db_ref != 0);
506 
507 	if (mp->b_datap->db_ref == 1 && MBLKSIZE(mp) >= size) {
508 		/*
509 		 * If the data is wanted and it will fit where it is, no
510 		 * work is required.
511 		 */
512 		if (copy && mp->b_datap->db_lim - mp->b_rptr >= size)
513 			return (mp);
514 
515 		mp->b_wptr = mp->b_rptr = mp->b_datap->db_base;
516 		mp1 = mp;
517 	} else if ((mp1 = allocb_tmpl(size, mp)) != NULL) {
518 		/* XXX other mp state could be copied too, db_flags ... ? */
519 		mp1->b_cont = mp->b_cont;
520 	} else {
521 		return (NULL);
522 	}
523 
524 	if (copy) {
525 		bcopy(old_rptr, mp1->b_rptr, cur_size);
526 		mp1->b_wptr = mp1->b_rptr + cur_size;
527 	}
528 
529 	if (mp != mp1)
530 		freeb(mp);
531 
532 	return (mp1);
533 }
534 
535 static void
dblk_lastfree(mblk_t * mp,dblk_t * dbp)536 dblk_lastfree(mblk_t *mp, dblk_t *dbp)
537 {
538 	ASSERT(dbp->db_mblk == mp);
539 	if (dbp->db_fthdr != NULL)
540 		str_ftfree(dbp);
541 
542 	/* set credp and projid to be 'unspecified' before returning to cache */
543 	if (dbp->db_credp != NULL) {
544 		crfree(dbp->db_credp);
545 		dbp->db_credp = NULL;
546 	}
547 	dbp->db_cpid = -1;
548 
549 	/* Reset the struioflag and the checksum flag fields */
550 	dbp->db_struioflag = 0;
551 	dbp->db_struioun.cksum.flags = 0;
552 
553 	/* and the COOKED and/or UIOA flag(s) */
554 	dbp->db_flags &= ~(DBLK_COOKED | DBLK_UIOA);
555 
556 	kmem_cache_free(dbp->db_cache, dbp);
557 }
558 
559 static void
dblk_decref(mblk_t * mp,dblk_t * dbp)560 dblk_decref(mblk_t *mp, dblk_t *dbp)
561 {
562 	if (dbp->db_ref != 1) {
563 		uint32_t rtfu = atomic_add_32_nv(&DBLK_RTFU_WORD(dbp),
564 		    -(1 << DBLK_RTFU_SHIFT(db_ref)));
565 		/*
566 		 * atomic_add_32_nv() just decremented db_ref, so we no longer
567 		 * have a reference to the dblk, which means another thread
568 		 * could free it.  Therefore we cannot examine the dblk to
569 		 * determine whether ours was the last reference.  Instead,
570 		 * we extract the new and minimum reference counts from rtfu.
571 		 * Note that all we're really saying is "if (ref != refmin)".
572 		 */
573 		if (((rtfu >> DBLK_RTFU_SHIFT(db_ref)) & DBLK_REFMAX) !=
574 		    ((rtfu >> DBLK_RTFU_SHIFT(db_flags)) & DBLK_REFMIN)) {
575 			kmem_cache_free(mblk_cache, mp);
576 			return;
577 		}
578 	}
579 	dbp->db_mblk = mp;
580 	dbp->db_free = dbp->db_lastfree;
581 	dbp->db_lastfree(mp, dbp);
582 }
583 
584 mblk_t *
dupb(mblk_t * mp)585 dupb(mblk_t *mp)
586 {
587 	dblk_t *dbp = mp->b_datap;
588 	mblk_t *new_mp;
589 	uint32_t oldrtfu, newrtfu;
590 
591 	if ((new_mp = kmem_cache_alloc(mblk_cache, KM_NOSLEEP)) == NULL)
592 		goto out;
593 
594 	new_mp->b_next = new_mp->b_prev = new_mp->b_cont = NULL;
595 	new_mp->b_rptr = mp->b_rptr;
596 	new_mp->b_wptr = mp->b_wptr;
597 	new_mp->b_datap = dbp;
598 	new_mp->b_queue = NULL;
599 	MBLK_BAND_FLAG_WORD(new_mp) = MBLK_BAND_FLAG_WORD(mp);
600 
601 	STR_FTEVENT_MBLK(mp, caller(), FTEV_DUPB, dbp->db_ref);
602 
603 	dbp->db_free = dblk_decref;
604 	do {
605 		ASSERT(dbp->db_ref > 0);
606 		oldrtfu = DBLK_RTFU_WORD(dbp);
607 		newrtfu = oldrtfu + (1 << DBLK_RTFU_SHIFT(db_ref));
608 		/*
609 		 * If db_ref is maxed out we can't dup this message anymore.
610 		 */
611 		if ((oldrtfu & DBLK_RTFU_REF_MASK) == DBLK_RTFU_REF_MASK) {
612 			kmem_cache_free(mblk_cache, new_mp);
613 			new_mp = NULL;
614 			goto out;
615 		}
616 	} while (atomic_cas_32(&DBLK_RTFU_WORD(dbp), oldrtfu, newrtfu) !=
617 	    oldrtfu);
618 
619 out:
620 	FTRACE_1("dupb(): new_mp=0x%lx", (uintptr_t)new_mp);
621 	return (new_mp);
622 }
623 
624 /*ARGSUSED*/
625 static void
frnop_func(void * arg)626 frnop_func(void *arg)
627 {
628 }
629 
630 /*
631  * Generic esballoc used to implement the four flavors: [d]esballoc[a].
632  * and allocb_oversize
633  */
634 static mblk_t *
gesballoc(unsigned char * base,size_t size,uint32_t db_rtfu,frtn_t * frp,void (* lastfree)(mblk_t *,dblk_t *),int kmflags)635 gesballoc(unsigned char *base, size_t size, uint32_t db_rtfu, frtn_t *frp,
636     void (*lastfree)(mblk_t *, dblk_t *), int kmflags)
637 {
638 	dblk_t *dbp;
639 	mblk_t *mp;
640 
641 	ASSERT(base != NULL && frp != NULL);
642 
643 	if ((dbp = kmem_cache_alloc(dblk_esb_cache, kmflags)) == NULL) {
644 		mp = NULL;
645 		goto out;
646 	}
647 
648 	mp = dbp->db_mblk;
649 	dbp->db_base = base;
650 	dbp->db_lim = base + size;
651 	dbp->db_free = dbp->db_lastfree = lastfree;
652 	dbp->db_frtnp = frp;
653 	DBLK_RTFU_WORD(dbp) = db_rtfu;
654 	mp->b_next = mp->b_prev = mp->b_cont = NULL;
655 	mp->b_rptr = mp->b_wptr = base;
656 	mp->b_queue = NULL;
657 	MBLK_BAND_FLAG_WORD(mp) = 0;
658 
659 out:
660 	FTRACE_1("gesballoc(): mp=0x%lx", (uintptr_t)mp);
661 	return (mp);
662 }
663 
664 static void
bcache_dblk_lastfree(mblk_t * mp,dblk_t * dbp)665 bcache_dblk_lastfree(mblk_t *mp, dblk_t *dbp)
666 {
667 	bcache_t *bcp = dbp->db_cache;
668 
669 	ASSERT(dbp->db_mblk == mp);
670 	if (dbp->db_fthdr != NULL)
671 		str_ftfree(dbp);
672 
673 	/* set credp and projid to be 'unspecified' before returning to cache */
674 	if (dbp->db_credp != NULL) {
675 		crfree(dbp->db_credp);
676 		dbp->db_credp = NULL;
677 	}
678 	dbp->db_cpid = -1;
679 	dbp->db_struioflag = 0;
680 	dbp->db_struioun.cksum.flags = 0;
681 
682 	mutex_enter(&bcp->mutex);
683 	kmem_cache_free(bcp->dblk_cache, dbp);
684 	bcp->alloc--;
685 
686 	if (bcp->alloc == 0 && bcp->destroy != 0) {
687 		kmem_cache_destroy(bcp->dblk_cache);
688 		kmem_cache_destroy(bcp->buffer_cache);
689 		mutex_exit(&bcp->mutex);
690 		mutex_destroy(&bcp->mutex);
691 		kmem_free(bcp, sizeof (bcache_t));
692 	} else {
693 		mutex_exit(&bcp->mutex);
694 	}
695 }
696 
697 bcache_t *
bcache_create(char * name,size_t size,uint_t align)698 bcache_create(char *name, size_t size, uint_t align)
699 {
700 	bcache_t *bcp;
701 	char buffer[255];
702 
703 	ASSERT((align & (align - 1)) == 0);
704 
705 	if ((bcp = kmem_alloc(sizeof (bcache_t), KM_NOSLEEP)) == NULL)
706 		return (NULL);
707 
708 	bcp->size = size;
709 	bcp->align = align;
710 	bcp->alloc = 0;
711 	bcp->destroy = 0;
712 
713 	mutex_init(&bcp->mutex, NULL, MUTEX_DRIVER, NULL);
714 
715 	(void) sprintf(buffer, "%s_buffer_cache", name);
716 	bcp->buffer_cache = kmem_cache_create(buffer, size, align, NULL, NULL,
717 	    NULL, NULL, NULL, 0);
718 	(void) sprintf(buffer, "%s_dblk_cache", name);
719 	bcp->dblk_cache = kmem_cache_create(buffer, sizeof (dblk_t),
720 	    DBLK_CACHE_ALIGN, bcache_dblk_constructor, bcache_dblk_destructor,
721 	    NULL, (void *)bcp, NULL, 0);
722 
723 	return (bcp);
724 }
725 
726 void
bcache_destroy(bcache_t * bcp)727 bcache_destroy(bcache_t *bcp)
728 {
729 	ASSERT(bcp != NULL);
730 
731 	mutex_enter(&bcp->mutex);
732 	if (bcp->alloc == 0) {
733 		kmem_cache_destroy(bcp->dblk_cache);
734 		kmem_cache_destroy(bcp->buffer_cache);
735 		mutex_exit(&bcp->mutex);
736 		mutex_destroy(&bcp->mutex);
737 		kmem_free(bcp, sizeof (bcache_t));
738 	} else {
739 		bcp->destroy++;
740 		mutex_exit(&bcp->mutex);
741 	}
742 }
743 
744 /*ARGSUSED*/
745 mblk_t *
bcache_allocb(bcache_t * bcp,uint_t pri)746 bcache_allocb(bcache_t *bcp, uint_t pri)
747 {
748 	dblk_t *dbp;
749 	mblk_t *mp = NULL;
750 
751 	ASSERT(bcp != NULL);
752 
753 	mutex_enter(&bcp->mutex);
754 	if (bcp->destroy != 0) {
755 		mutex_exit(&bcp->mutex);
756 		goto out;
757 	}
758 
759 	if ((dbp = kmem_cache_alloc(bcp->dblk_cache, KM_NOSLEEP)) == NULL) {
760 		mutex_exit(&bcp->mutex);
761 		goto out;
762 	}
763 	bcp->alloc++;
764 	mutex_exit(&bcp->mutex);
765 
766 	ASSERT(((uintptr_t)(dbp->db_base) & (bcp->align - 1)) == 0);
767 
768 	mp = dbp->db_mblk;
769 	DBLK_RTFU_WORD(dbp) = DBLK_RTFU(1, M_DATA, 0, 0);
770 	mp->b_next = mp->b_prev = mp->b_cont = NULL;
771 	mp->b_rptr = mp->b_wptr = dbp->db_base;
772 	mp->b_queue = NULL;
773 	MBLK_BAND_FLAG_WORD(mp) = 0;
774 	STR_FTALLOC(&dbp->db_fthdr, FTEV_BCALLOCB, bcp->size);
775 out:
776 	FTRACE_1("bcache_allocb(): mp=0x%p", (uintptr_t)mp);
777 
778 	return (mp);
779 }
780 
781 static void
dblk_lastfree_oversize(mblk_t * mp,dblk_t * dbp)782 dblk_lastfree_oversize(mblk_t *mp, dblk_t *dbp)
783 {
784 	ASSERT(dbp->db_mblk == mp);
785 	if (dbp->db_fthdr != NULL)
786 		str_ftfree(dbp);
787 
788 	/* set credp and projid to be 'unspecified' before returning to cache */
789 	if (dbp->db_credp != NULL) {
790 		crfree(dbp->db_credp);
791 		dbp->db_credp = NULL;
792 	}
793 	dbp->db_cpid = -1;
794 	dbp->db_struioflag = 0;
795 	dbp->db_struioun.cksum.flags = 0;
796 
797 	kmem_free(dbp->db_base, dbp->db_lim - dbp->db_base);
798 	kmem_cache_free(dbp->db_cache, dbp);
799 }
800 
801 static mblk_t *
allocb_oversize(size_t size,int kmflags)802 allocb_oversize(size_t size, int kmflags)
803 {
804 	mblk_t *mp;
805 	void *buf;
806 
807 	size = P2ROUNDUP(size, DBLK_CACHE_ALIGN);
808 	if ((buf = kmem_alloc(size, kmflags)) == NULL)
809 		return (NULL);
810 	if ((mp = gesballoc(buf, size, DBLK_RTFU(1, M_DATA, 0, 0),
811 	    &frnop, dblk_lastfree_oversize, kmflags)) == NULL)
812 		kmem_free(buf, size);
813 
814 	if (mp != NULL)
815 		STR_FTALLOC(&DB_FTHDR(mp), FTEV_ALLOCBIG, size);
816 
817 	return (mp);
818 }
819 
820 mblk_t *
allocb_tryhard(size_t target_size)821 allocb_tryhard(size_t target_size)
822 {
823 	size_t size;
824 	mblk_t *bp;
825 
826 	for (size = target_size; size < target_size + 512;
827 	    size += DBLK_CACHE_ALIGN)
828 		if ((bp = allocb(size, BPRI_HI)) != NULL)
829 			return (bp);
830 	allocb_tryhard_fails++;
831 	return (NULL);
832 }
833 
834 /*
835  * This routine is consolidation private for STREAMS internal use
836  * This routine may only be called from sync routines (i.e., not
837  * from put or service procedures).  It is located here (rather
838  * than strsubr.c) so that we don't have to expose all of the
839  * allocb() implementation details in header files.
840  */
841 mblk_t *
allocb_wait(size_t size,uint_t pri,uint_t flags,int * error)842 allocb_wait(size_t size, uint_t pri, uint_t flags, int *error)
843 {
844 	dblk_t *dbp;
845 	mblk_t *mp;
846 	size_t index;
847 
848 	index = (size -1) >> DBLK_SIZE_SHIFT;
849 
850 	if (flags & STR_NOSIG) {
851 		if (index >= (DBLK_MAX_CACHE >> DBLK_SIZE_SHIFT)) {
852 			if (size != 0) {
853 				mp = allocb_oversize(size, KM_SLEEP);
854 				FTRACE_1("allocb_wait (NOSIG): mp=0x%lx",
855 				    (uintptr_t)mp);
856 				return (mp);
857 			}
858 			index = 0;
859 		}
860 
861 		dbp = kmem_cache_alloc(dblk_cache[index], KM_SLEEP);
862 		mp = dbp->db_mblk;
863 		DBLK_RTFU_WORD(dbp) = DBLK_RTFU(1, M_DATA, 0, 0);
864 		mp->b_next = mp->b_prev = mp->b_cont = NULL;
865 		mp->b_rptr = mp->b_wptr = dbp->db_base;
866 		mp->b_queue = NULL;
867 		MBLK_BAND_FLAG_WORD(mp) = 0;
868 		STR_FTALLOC(&DB_FTHDR(mp), FTEV_ALLOCBW, size);
869 
870 		FTRACE_1("allocb_wait (NOSIG): mp=0x%lx", (uintptr_t)mp);
871 
872 	} else {
873 		while ((mp = allocb(size, pri)) == NULL) {
874 			if ((*error = strwaitbuf(size, BPRI_HI)) != 0)
875 				return (NULL);
876 		}
877 	}
878 
879 	return (mp);
880 }
881 
882 /*
883  * Call function 'func' with 'arg' when a class zero block can
884  * be allocated with priority 'pri'.
885  */
886 bufcall_id_t
esbbcall(uint_t pri,void (* func)(void *),void * arg)887 esbbcall(uint_t pri, void (*func)(void *), void *arg)
888 {
889 	return (bufcall(1, pri, func, arg));
890 }
891 
892 /*
893  * Allocates an iocblk (M_IOCTL) block. Properly sets the credentials
894  * ioc_id, rval and error of the struct ioctl to set up an ioctl call.
895  * This provides consistency for all internal allocators of ioctl.
896  */
897 mblk_t *
mkiocb(uint_t cmd)898 mkiocb(uint_t cmd)
899 {
900 	struct iocblk	*ioc;
901 	mblk_t		*mp;
902 
903 	/*
904 	 * Allocate enough space for any of the ioctl related messages.
905 	 */
906 	if ((mp = allocb(sizeof (union ioctypes), BPRI_MED)) == NULL)
907 		return (NULL);
908 
909 	bzero(mp->b_rptr, sizeof (union ioctypes));
910 
911 	/*
912 	 * Set the mblk_t information and ptrs correctly.
913 	 */
914 	mp->b_wptr += sizeof (struct iocblk);
915 	mp->b_datap->db_type = M_IOCTL;
916 
917 	/*
918 	 * Fill in the fields.
919 	 */
920 	ioc		= (struct iocblk *)mp->b_rptr;
921 	ioc->ioc_cmd	= cmd;
922 	ioc->ioc_cr	= kcred;
923 	ioc->ioc_id	= getiocseqno();
924 	ioc->ioc_flag	= IOC_NATIVE;
925 	return (mp);
926 }
927 
928 /*
929  * test if block of given size can be allocated with a request of
930  * the given priority.
931  * 'pri' is no longer used, but is retained for compatibility.
932  */
933 /* ARGSUSED */
934 int
testb(size_t size,uint_t pri)935 testb(size_t size, uint_t pri)
936 {
937 	return ((size + sizeof (dblk_t)) <= kmem_avail());
938 }
939 
940 /* _KERNEL: bufcall, unbufcall */
941 
942 /*
943  * Duplicate a message block by block (uses dupb), returning
944  * a pointer to the duplicate message.
945  * Returns a non-NULL value only if the entire message
946  * was dup'd.
947  */
948 mblk_t *
dupmsg(mblk_t * bp)949 dupmsg(mblk_t *bp)
950 {
951 	mblk_t *head, *nbp;
952 
953 	if (!bp || !(nbp = head = dupb(bp)))
954 		return (NULL);
955 
956 	while (bp->b_cont) {
957 		if (!(nbp->b_cont = dupb(bp->b_cont))) {
958 			freemsg(head);
959 			return (NULL);
960 		}
961 		nbp = nbp->b_cont;
962 		bp = bp->b_cont;
963 	}
964 	return (head);
965 }
966 
967 #define	DUPB_NOLOAN(bp) \
968 	((((bp)->b_datap->db_struioflag & STRUIO_ZC) != 0) ? \
969 	copyb((bp)) : dupb((bp)))
970 
971 mblk_t *
dupmsg_noloan(mblk_t * bp)972 dupmsg_noloan(mblk_t *bp)
973 {
974 	mblk_t *head, *nbp;
975 
976 	if (bp == NULL || DB_TYPE(bp) != M_DATA ||
977 	    ((nbp = head = DUPB_NOLOAN(bp)) == NULL))
978 		return (NULL);
979 
980 	while (bp->b_cont) {
981 		if ((nbp->b_cont = DUPB_NOLOAN(bp->b_cont)) == NULL) {
982 			freemsg(head);
983 			return (NULL);
984 		}
985 		nbp = nbp->b_cont;
986 		bp = bp->b_cont;
987 	}
988 	return (head);
989 }
990 
991 /*
992  * Copy data from message and data block to newly allocated message and
993  * data block. Returns new message block pointer, or NULL if error.
994  * The alignment of rptr (w.r.t. word alignment) will be the same in the copy
995  * as in the original even when db_base is not word aligned. (bug 1052877)
996  */
997 mblk_t *
copyb(mblk_t * bp)998 copyb(mblk_t *bp)
999 {
1000 	mblk_t	*nbp;
1001 	dblk_t	*dp, *ndp;
1002 	uchar_t *base;
1003 	size_t	size;
1004 	size_t	unaligned;
1005 
1006 	ASSERT(bp->b_wptr >= bp->b_rptr);
1007 
1008 	dp = bp->b_datap;
1009 	if (dp->db_fthdr != NULL)
1010 		STR_FTEVENT_MBLK(bp, caller(), FTEV_COPYB, 0);
1011 
1012 	size = dp->db_lim - dp->db_base;
1013 	unaligned = P2PHASE((uintptr_t)dp->db_base, sizeof (uint_t));
1014 	if ((nbp = allocb_tmpl(size + unaligned, bp)) == NULL)
1015 		return (NULL);
1016 	nbp->b_flag = bp->b_flag;
1017 	nbp->b_band = bp->b_band;
1018 	ndp = nbp->b_datap;
1019 
1020 	/*
1021 	 * Well, here is a potential issue.  If we are trying to
1022 	 * trace a flow, and we copy the message, we might lose
1023 	 * information about where this message might have been.
1024 	 * So we should inherit the FT data.  On the other hand,
1025 	 * a user might be interested only in alloc to free data.
1026 	 * So I guess the real answer is to provide a tunable.
1027 	 */
1028 	STR_FTEVENT_MBLK(nbp, caller(), FTEV_COPYB, 1);
1029 
1030 	base = ndp->db_base + unaligned;
1031 	bcopy(dp->db_base, ndp->db_base + unaligned, size);
1032 
1033 	nbp->b_rptr = base + (bp->b_rptr - dp->db_base);
1034 	nbp->b_wptr = nbp->b_rptr + MBLKL(bp);
1035 
1036 	return (nbp);
1037 }
1038 
1039 /*
1040  * Copy data from message to newly allocated message using new
1041  * data blocks.  Returns a pointer to the new message, or NULL if error.
1042  */
1043 mblk_t *
copymsg(mblk_t * bp)1044 copymsg(mblk_t *bp)
1045 {
1046 	mblk_t *head, *nbp;
1047 
1048 	if (!bp || !(nbp = head = copyb(bp)))
1049 		return (NULL);
1050 
1051 	while (bp->b_cont) {
1052 		if (!(nbp->b_cont = copyb(bp->b_cont))) {
1053 			freemsg(head);
1054 			return (NULL);
1055 		}
1056 		nbp = nbp->b_cont;
1057 		bp = bp->b_cont;
1058 	}
1059 	return (head);
1060 }
1061 
1062 /*
1063  * link a message block to tail of message
1064  */
1065 void
linkb(mblk_t * mp,mblk_t * bp)1066 linkb(mblk_t *mp, mblk_t *bp)
1067 {
1068 	ASSERT(mp && bp);
1069 
1070 	for (; mp->b_cont; mp = mp->b_cont)
1071 		;
1072 	mp->b_cont = bp;
1073 }
1074 
1075 /*
1076  * unlink a message block from head of message
1077  * return pointer to new message.
1078  * NULL if message becomes empty.
1079  */
1080 mblk_t *
unlinkb(mblk_t * bp)1081 unlinkb(mblk_t *bp)
1082 {
1083 	mblk_t *bp1;
1084 
1085 	bp1 = bp->b_cont;
1086 	bp->b_cont = NULL;
1087 	return (bp1);
1088 }
1089 
1090 /*
1091  * remove a message block "bp" from message "mp"
1092  *
1093  * Return pointer to new message or NULL if no message remains.
1094  * Return -1 if bp is not found in message.
1095  */
1096 mblk_t *
rmvb(mblk_t * mp,mblk_t * bp)1097 rmvb(mblk_t *mp, mblk_t *bp)
1098 {
1099 	mblk_t *tmp;
1100 	mblk_t *lastp = NULL;
1101 
1102 	ASSERT(mp && bp);
1103 	for (tmp = mp; tmp; tmp = tmp->b_cont) {
1104 		if (tmp == bp) {
1105 			if (lastp)
1106 				lastp->b_cont = tmp->b_cont;
1107 			else
1108 				mp = tmp->b_cont;
1109 			tmp->b_cont = NULL;
1110 			return (mp);
1111 		}
1112 		lastp = tmp;
1113 	}
1114 	return ((mblk_t *)-1);
1115 }
1116 
1117 /*
1118  * Concatenate and align first len bytes of common
1119  * message type.  Len == -1, means concat everything.
1120  * Returns 1 on success, 0 on failure
1121  * After the pullup, mp points to the pulled up data.
1122  */
1123 int
pullupmsg(mblk_t * mp,ssize_t len)1124 pullupmsg(mblk_t *mp, ssize_t len)
1125 {
1126 	mblk_t *bp, *b_cont;
1127 	dblk_t *dbp;
1128 	ssize_t n;
1129 
1130 	ASSERT(mp->b_datap->db_ref > 0);
1131 	ASSERT(mp->b_next == NULL && mp->b_prev == NULL);
1132 
1133 	/*
1134 	 * We won't handle Multidata message, since it contains
1135 	 * metadata which this function has no knowledge of; we
1136 	 * assert on DEBUG, and return failure otherwise.
1137 	 */
1138 	ASSERT(mp->b_datap->db_type != M_MULTIDATA);
1139 	if (mp->b_datap->db_type == M_MULTIDATA)
1140 		return (0);
1141 
1142 	if (len == -1) {
1143 		if (mp->b_cont == NULL && str_aligned(mp->b_rptr))
1144 			return (1);
1145 		len = xmsgsize(mp);
1146 	} else {
1147 		ssize_t first_mblk_len = mp->b_wptr - mp->b_rptr;
1148 		ASSERT(first_mblk_len >= 0);
1149 		/*
1150 		 * If the length is less than that of the first mblk,
1151 		 * we want to pull up the message into an aligned mblk.
1152 		 * Though not part of the spec, some callers assume it.
1153 		 */
1154 		if (len <= first_mblk_len) {
1155 			if (str_aligned(mp->b_rptr))
1156 				return (1);
1157 			len = first_mblk_len;
1158 		} else if (xmsgsize(mp) < len)
1159 			return (0);
1160 	}
1161 
1162 	if ((bp = allocb_tmpl(len, mp)) == NULL)
1163 		return (0);
1164 
1165 	dbp = bp->b_datap;
1166 	*bp = *mp;		/* swap mblks so bp heads the old msg... */
1167 	mp->b_datap = dbp;	/* ... and mp heads the new message */
1168 	mp->b_datap->db_mblk = mp;
1169 	bp->b_datap->db_mblk = bp;
1170 	mp->b_rptr = mp->b_wptr = dbp->db_base;
1171 
1172 	do {
1173 		ASSERT(bp->b_datap->db_ref > 0);
1174 		ASSERT(bp->b_wptr >= bp->b_rptr);
1175 		n = MIN(bp->b_wptr - bp->b_rptr, len);
1176 		ASSERT(n >= 0);		/* allow zero-length mblk_t's */
1177 		if (n > 0)
1178 			bcopy(bp->b_rptr, mp->b_wptr, (size_t)n);
1179 		mp->b_wptr += n;
1180 		bp->b_rptr += n;
1181 		len -= n;
1182 		if (bp->b_rptr != bp->b_wptr)
1183 			break;
1184 		b_cont = bp->b_cont;
1185 		freeb(bp);
1186 		bp = b_cont;
1187 	} while (len && bp);
1188 
1189 	mp->b_cont = bp;	/* tack on whatever wasn't pulled up */
1190 
1191 	return (1);
1192 }
1193 
1194 /*
1195  * Concatenate and align at least the first len bytes of common message
1196  * type.  Len == -1 means concatenate everything.  The original message is
1197  * unaltered.  Returns a pointer to a new message on success, otherwise
1198  * returns NULL.
1199  */
1200 mblk_t *
msgpullup(mblk_t * mp,ssize_t len)1201 msgpullup(mblk_t *mp, ssize_t len)
1202 {
1203 	mblk_t	*newmp;
1204 	ssize_t	totlen;
1205 	ssize_t	n;
1206 
1207 	/*
1208 	 * We won't handle Multidata message, since it contains
1209 	 * metadata which this function has no knowledge of; we
1210 	 * assert on DEBUG, and return failure otherwise.
1211 	 */
1212 	ASSERT(mp->b_datap->db_type != M_MULTIDATA);
1213 	if (mp->b_datap->db_type == M_MULTIDATA)
1214 		return (NULL);
1215 
1216 	totlen = xmsgsize(mp);
1217 
1218 	if ((len > 0) && (len > totlen))
1219 		return (NULL);
1220 
1221 	/*
1222 	 * Copy all of the first msg type into one new mblk, then dupmsg
1223 	 * and link the rest onto this.
1224 	 */
1225 
1226 	len = totlen;
1227 
1228 	if ((newmp = allocb_tmpl(len, mp)) == NULL)
1229 		return (NULL);
1230 
1231 	newmp->b_flag = mp->b_flag;
1232 	newmp->b_band = mp->b_band;
1233 
1234 	while (len > 0) {
1235 		n = mp->b_wptr - mp->b_rptr;
1236 		ASSERT(n >= 0);		/* allow zero-length mblk_t's */
1237 		if (n > 0)
1238 			bcopy(mp->b_rptr, newmp->b_wptr, n);
1239 		newmp->b_wptr += n;
1240 		len -= n;
1241 		mp = mp->b_cont;
1242 	}
1243 
1244 	if (mp != NULL) {
1245 		newmp->b_cont = dupmsg(mp);
1246 		if (newmp->b_cont == NULL) {
1247 			freemsg(newmp);
1248 			return (NULL);
1249 		}
1250 	}
1251 
1252 	return (newmp);
1253 }
1254 
1255 /*
1256  * Trim bytes from message
1257  *  len > 0, trim from head
1258  *  len < 0, trim from tail
1259  * Returns 1 on success, 0 on failure.
1260  */
1261 int
adjmsg(mblk_t * mp,ssize_t len)1262 adjmsg(mblk_t *mp, ssize_t len)
1263 {
1264 	mblk_t *bp;
1265 	mblk_t *save_bp = NULL;
1266 	mblk_t *prev_bp;
1267 	mblk_t *bcont;
1268 	unsigned char type;
1269 	ssize_t n;
1270 	int fromhead;
1271 	int first;
1272 
1273 	ASSERT(mp != NULL);
1274 	/*
1275 	 * We won't handle Multidata message, since it contains
1276 	 * metadata which this function has no knowledge of; we
1277 	 * assert on DEBUG, and return failure otherwise.
1278 	 */
1279 	ASSERT(mp->b_datap->db_type != M_MULTIDATA);
1280 	if (mp->b_datap->db_type == M_MULTIDATA)
1281 		return (0);
1282 
1283 	if (len < 0) {
1284 		fromhead = 0;
1285 		len = -len;
1286 	} else {
1287 		fromhead = 1;
1288 	}
1289 
1290 	if (xmsgsize(mp) < len)
1291 		return (0);
1292 
1293 	if (fromhead) {
1294 		first = 1;
1295 		while (len) {
1296 			ASSERT(mp->b_wptr >= mp->b_rptr);
1297 			n = MIN(mp->b_wptr - mp->b_rptr, len);
1298 			mp->b_rptr += n;
1299 			len -= n;
1300 
1301 			/*
1302 			 * If this is not the first zero length
1303 			 * message remove it
1304 			 */
1305 			if (!first && (mp->b_wptr == mp->b_rptr)) {
1306 				bcont = mp->b_cont;
1307 				freeb(mp);
1308 				mp = save_bp->b_cont = bcont;
1309 			} else {
1310 				save_bp = mp;
1311 				mp = mp->b_cont;
1312 			}
1313 			first = 0;
1314 		}
1315 	} else {
1316 		type = mp->b_datap->db_type;
1317 		while (len) {
1318 			bp = mp;
1319 			save_bp = NULL;
1320 
1321 			/*
1322 			 * Find the last message of same type
1323 			 */
1324 			while (bp && bp->b_datap->db_type == type) {
1325 				ASSERT(bp->b_wptr >= bp->b_rptr);
1326 				prev_bp = save_bp;
1327 				save_bp = bp;
1328 				bp = bp->b_cont;
1329 			}
1330 			if (save_bp == NULL)
1331 				break;
1332 			n = MIN(save_bp->b_wptr - save_bp->b_rptr, len);
1333 			save_bp->b_wptr -= n;
1334 			len -= n;
1335 
1336 			/*
1337 			 * If this is not the first message
1338 			 * and we have taken away everything
1339 			 * from this message, remove it
1340 			 */
1341 
1342 			if ((save_bp != mp) &&
1343 			    (save_bp->b_wptr == save_bp->b_rptr)) {
1344 				bcont = save_bp->b_cont;
1345 				freeb(save_bp);
1346 				prev_bp->b_cont = bcont;
1347 			}
1348 		}
1349 	}
1350 	return (1);
1351 }
1352 
1353 /*
1354  * get number of data bytes in message
1355  */
1356 size_t
msgdsize(mblk_t * bp)1357 msgdsize(mblk_t *bp)
1358 {
1359 	size_t count = 0;
1360 
1361 	for (; bp; bp = bp->b_cont)
1362 		if (bp->b_datap->db_type == M_DATA) {
1363 			ASSERT(bp->b_wptr >= bp->b_rptr);
1364 			count += bp->b_wptr - bp->b_rptr;
1365 		}
1366 	return (count);
1367 }
1368 
1369 /* getq() etc to EOF removed */
1370