xref: /freebsd/sys/kern/subr_memdesc.c (revision aa1a8ff2d6dbc51ef058f46f3db5a8bb77967145)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2023 Chelsio Communications, Inc.
5  * Written by: John Baldwin <jhb@FreeBSD.org>
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  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/mbuf.h>
31 #include <sys/memdesc.h>
32 #include <sys/systm.h>
33 #include <sys/uio.h>
34 #include <vm/vm.h>
35 #include <vm/pmap.h>
36 #include <vm/vm_page.h>
37 #include <vm/vm_param.h>
38 #include <machine/bus.h>
39 
40 /*
41  * memdesc_copyback copies data from a source buffer into a buffer
42  * described by a memory descriptor.
43  */
44 static void
45 phys_copyback(vm_paddr_t pa, int off, int size, const void *src)
46 {
47 	const char *cp;
48 	u_int page_off;
49 	int todo;
50 	void *p;
51 
52 	KASSERT(PMAP_HAS_DMAP, ("direct-map required"));
53 
54 	cp = src;
55 	pa += off;
56 	page_off = pa & PAGE_MASK;
57 	while (size > 0) {
58 		todo = min(PAGE_SIZE - page_off, size);
59 		p = (void *)PHYS_TO_DMAP(pa);
60 		memcpy(p, cp, todo);
61 		size -= todo;
62 		cp += todo;
63 		pa += todo;
64 		page_off = 0;
65 	}
66 }
67 
68 static void
69 vlist_copyback(struct bus_dma_segment *vlist, int sglist_cnt, int off,
70     int size, const void *src)
71 {
72 	const char *p;
73 	int todo;
74 
75 	while (vlist->ds_len <= off) {
76 		KASSERT(sglist_cnt > 1, ("out of sglist entries"));
77 
78 		off -= vlist->ds_len;
79 		vlist++;
80 		sglist_cnt--;
81 	}
82 
83 	p = src;
84 	while (size > 0) {
85 		KASSERT(sglist_cnt >= 1, ("out of sglist entries"));
86 
87 		todo = size;
88 		if (todo > vlist->ds_len - off)
89 			todo = vlist->ds_len - off;
90 
91 		memcpy((char *)(uintptr_t)vlist->ds_addr + off, p, todo);
92 		off = 0;
93 		vlist++;
94 		sglist_cnt--;
95 		size -= todo;
96 		p += todo;
97 	}
98 }
99 
100 static void
101 plist_copyback(struct bus_dma_segment *plist, int sglist_cnt, int off,
102     int size, const void *src)
103 {
104 	const char *p;
105 	int todo;
106 
107 	while (plist->ds_len <= off) {
108 		KASSERT(sglist_cnt > 1, ("out of sglist entries"));
109 
110 		off -= plist->ds_len;
111 		plist++;
112 		sglist_cnt--;
113 	}
114 
115 	p = src;
116 	while (size > 0) {
117 		KASSERT(sglist_cnt >= 1, ("out of sglist entries"));
118 
119 		todo = size;
120 		if (todo > plist->ds_len - off)
121 			todo = plist->ds_len - off;
122 
123 		phys_copyback(plist->ds_addr, off, todo, p);
124 		off = 0;
125 		plist++;
126 		sglist_cnt--;
127 		size -= todo;
128 		p += todo;
129 	}
130 }
131 
132 static void
133 vmpages_copyback(vm_page_t *m, int off, int size, const void *src)
134 {
135 	struct iovec iov[1];
136 	struct uio uio;
137 	int error __diagused;
138 
139 	iov[0].iov_base = __DECONST(void *, src);
140 	iov[0].iov_len = size;
141 	uio.uio_iov = iov;
142 	uio.uio_iovcnt = 1;
143 	uio.uio_offset = 0;
144 	uio.uio_resid = size;
145 	uio.uio_segflg = UIO_SYSSPACE;
146 	uio.uio_rw = UIO_WRITE;
147 	error = uiomove_fromphys(m, off, size, &uio);
148 	KASSERT(error == 0 && uio.uio_resid == 0, ("copy failed"));
149 }
150 
151 void
152 memdesc_copyback(struct memdesc *mem, int off, int size, const void *src)
153 {
154 	KASSERT(off >= 0, ("%s: invalid offset %d", __func__, off));
155 	KASSERT(size >= 0, ("%s: invalid size %d", __func__, off));
156 
157 	switch (mem->md_type) {
158 	case MEMDESC_VADDR:
159 		KASSERT(off + size <= mem->md_len, ("copy out of bounds"));
160 		memcpy((char *)mem->u.md_vaddr + off, src, size);
161 		break;
162 	case MEMDESC_PADDR:
163 		KASSERT(off + size <= mem->md_len, ("copy out of bounds"));
164 		phys_copyback(mem->u.md_paddr, off, size, src);
165 		break;
166 	case MEMDESC_VLIST:
167 		vlist_copyback(mem->u.md_list, mem->md_nseg, off, size, src);
168 		break;
169 	case MEMDESC_PLIST:
170 		plist_copyback(mem->u.md_list, mem->md_nseg, off, size, src);
171 		break;
172 	case MEMDESC_UIO:
173 		panic("Use uiomove instead");
174 		break;
175 	case MEMDESC_MBUF:
176 		m_copyback(mem->u.md_mbuf, off, size, src);
177 		break;
178 	case MEMDESC_VMPAGES:
179 		KASSERT(off + size <= mem->md_len, ("copy out of bounds"));
180 		vmpages_copyback(mem->u.md_ma, mem->md_offset + off, size,
181 		    src);
182 		break;
183 	default:
184 		__assert_unreachable();
185 	}
186 }
187 
188 /*
189  * memdesc_copydata copies data from a buffer described by a memory
190  * descriptor into a destination buffer.
191  */
192 static void
193 phys_copydata(vm_paddr_t pa, int off, int size, void *dst)
194 {
195 	char *cp;
196 	u_int page_off;
197 	int todo;
198 	const void *p;
199 
200 	KASSERT(PMAP_HAS_DMAP, ("direct-map required"));
201 
202 	cp = dst;
203 	pa += off;
204 	page_off = pa & PAGE_MASK;
205 	while (size > 0) {
206 		todo = min(PAGE_SIZE - page_off, size);
207 		p = (const void *)PHYS_TO_DMAP(pa);
208 		memcpy(cp, p, todo);
209 		size -= todo;
210 		cp += todo;
211 		pa += todo;
212 		page_off = 0;
213 	}
214 }
215 
216 static void
217 vlist_copydata(struct bus_dma_segment *vlist, int sglist_cnt, int off,
218     int size, void *dst)
219 {
220 	char *p;
221 	int todo;
222 
223 	while (vlist->ds_len <= off) {
224 		KASSERT(sglist_cnt > 1, ("out of sglist entries"));
225 
226 		off -= vlist->ds_len;
227 		vlist++;
228 		sglist_cnt--;
229 	}
230 
231 	p = dst;
232 	while (size > 0) {
233 		KASSERT(sglist_cnt >= 1, ("out of sglist entries"));
234 
235 		todo = size;
236 		if (todo > vlist->ds_len - off)
237 			todo = vlist->ds_len - off;
238 
239 		memcpy(p, (char *)(uintptr_t)vlist->ds_addr + off, todo);
240 		off = 0;
241 		vlist++;
242 		sglist_cnt--;
243 		size -= todo;
244 		p += todo;
245 	}
246 }
247 
248 static void
249 plist_copydata(struct bus_dma_segment *plist, int sglist_cnt, int off,
250     int size, void *dst)
251 {
252 	char *p;
253 	int todo;
254 
255 	while (plist->ds_len <= off) {
256 		KASSERT(sglist_cnt > 1, ("out of sglist entries"));
257 
258 		off -= plist->ds_len;
259 		plist++;
260 		sglist_cnt--;
261 	}
262 
263 	p = dst;
264 	while (size > 0) {
265 		KASSERT(sglist_cnt >= 1, ("out of sglist entries"));
266 
267 		todo = size;
268 		if (todo > plist->ds_len - off)
269 			todo = plist->ds_len - off;
270 
271 		phys_copydata(plist->ds_addr, off, todo, p);
272 		off = 0;
273 		plist++;
274 		sglist_cnt--;
275 		size -= todo;
276 		p += todo;
277 	}
278 }
279 
280 static void
281 vmpages_copydata(vm_page_t *m, int off, int size, void *dst)
282 {
283 	struct iovec iov[1];
284 	struct uio uio;
285 	int error __diagused;
286 
287 	iov[0].iov_base = dst;
288 	iov[0].iov_len = size;
289 	uio.uio_iov = iov;
290 	uio.uio_iovcnt = 1;
291 	uio.uio_offset = 0;
292 	uio.uio_resid = size;
293 	uio.uio_segflg = UIO_SYSSPACE;
294 	uio.uio_rw = UIO_READ;
295 	error = uiomove_fromphys(m, off, size, &uio);
296 	KASSERT(error == 0 && uio.uio_resid == 0, ("copy failed"));
297 }
298 
299 void
300 memdesc_copydata(struct memdesc *mem, int off, int size, void *dst)
301 {
302 	KASSERT(off >= 0, ("%s: invalid offset %d", __func__, off));
303 	KASSERT(size >= 0, ("%s: invalid size %d", __func__, off));
304 
305 	switch (mem->md_type) {
306 	case MEMDESC_VADDR:
307 		KASSERT(off + size <= mem->md_len, ("copy out of bounds"));
308 		memcpy(dst, (const char *)mem->u.md_vaddr + off, size);
309 		break;
310 	case MEMDESC_PADDR:
311 		KASSERT(off + size <= mem->md_len, ("copy out of bounds"));
312 		phys_copydata(mem->u.md_paddr, off, size, dst);
313 		break;
314 	case MEMDESC_VLIST:
315 		vlist_copydata(mem->u.md_list, mem->md_nseg, off, size, dst);
316 		break;
317 	case MEMDESC_PLIST:
318 		plist_copydata(mem->u.md_list, mem->md_nseg, off, size, dst);
319 		break;
320 	case MEMDESC_UIO:
321 		panic("Use uiomove instead");
322 		break;
323 	case MEMDESC_MBUF:
324 		m_copydata(mem->u.md_mbuf, off, size, dst);
325 		break;
326 	case MEMDESC_VMPAGES:
327 		KASSERT(off + size <= mem->md_len, ("copy out of bounds"));
328 		vmpages_copydata(mem->u.md_ma, mem->md_offset + off, size,
329 		    dst);
330 		break;
331 	default:
332 		__assert_unreachable();
333 	}
334 }
335 
336 /*
337  * memdesc_alloc_ext_mbufs allocates a chain of external mbufs backed
338  * by the storage of a memory descriptor's data buffer.
339  */
340 static struct mbuf *
341 vaddr_ext_mbuf(memdesc_alloc_ext_mbuf_t *ext_alloc, void *cb_arg, int how,
342     void *buf, size_t len, size_t *actual_len)
343 {
344 	*actual_len = len;
345 	return (ext_alloc(cb_arg, how, buf, len));
346 }
347 
348 static bool
349 can_append_paddr(struct mbuf *m, vm_paddr_t pa)
350 {
351 	u_int last_len;
352 
353 	/* Can always append to an empty mbuf. */
354 	if (m->m_epg_npgs == 0)
355 		return (true);
356 
357 	/* Can't append to a full mbuf. */
358 	if (m->m_epg_npgs == MBUF_PEXT_MAX_PGS)
359 		return (false);
360 
361 	/* Can't append a non-page-aligned address to a non-empty mbuf. */
362 	if ((pa & PAGE_MASK) != 0)
363 		return (false);
364 
365 	/* Can't append if the last page is not a full page. */
366 	last_len = m->m_epg_last_len;
367 	if (m->m_epg_npgs == 1)
368 		last_len += m->m_epg_1st_off;
369 	return (last_len == PAGE_SIZE);
370 }
371 
372 /*
373  * Returns amount of data added to an M_EXTPG mbuf.
374  */
375 static size_t
376 append_paddr_range(struct mbuf *m, vm_paddr_t pa, size_t len)
377 {
378 	size_t appended;
379 
380 	appended = 0;
381 
382 	/* Append the first page. */
383 	if (m->m_epg_npgs == 0) {
384 		m->m_epg_pa[0] = trunc_page(pa);
385 		m->m_epg_npgs = 1;
386 		m->m_epg_1st_off = pa & PAGE_MASK;
387 		m->m_epg_last_len = PAGE_SIZE - m->m_epg_1st_off;
388 		if (m->m_epg_last_len > len)
389 			m->m_epg_last_len = len;
390 		m->m_len = m->m_epg_last_len;
391 		len -= m->m_epg_last_len;
392 		pa += m->m_epg_last_len;
393 		appended += m->m_epg_last_len;
394 	}
395 	KASSERT(len == 0 || (pa & PAGE_MASK) == 0,
396 	    ("PA not aligned before full pages"));
397 
398 	/* Full pages. */
399 	while (len >= PAGE_SIZE && m->m_epg_npgs < MBUF_PEXT_MAX_PGS) {
400 		m->m_epg_pa[m->m_epg_npgs] = pa;
401 		m->m_epg_npgs++;
402 		m->m_epg_last_len = PAGE_SIZE;
403 		m->m_len += PAGE_SIZE;
404 		pa += PAGE_SIZE;
405 		len -= PAGE_SIZE;
406 		appended += PAGE_SIZE;
407 	}
408 
409 	/* Final partial page. */
410 	if (len > 0 && m->m_epg_npgs < MBUF_PEXT_MAX_PGS) {
411 		KASSERT(len < PAGE_SIZE, ("final page is full page"));
412 		m->m_epg_pa[m->m_epg_npgs] = pa;
413 		m->m_epg_npgs++;
414 		m->m_epg_last_len = len;
415 		m->m_len += len;
416 		appended += len;
417 	}
418 
419 	return (appended);
420 }
421 
422 static struct mbuf *
423 paddr_ext_mbuf(memdesc_alloc_extpg_mbuf_t *extpg_alloc, void *cb_arg, int how,
424     vm_paddr_t pa, size_t len, size_t *actual_len, bool can_truncate)
425 {
426 	struct mbuf *m, *tail;
427 	size_t appended;
428 
429 	if (can_truncate) {
430 		vm_paddr_t end;
431 
432 		/*
433 		 * Trim any partial page at the end, but not if it's
434 		 * the only page.
435 		 */
436 		end = trunc_page(pa + len);
437 		if (end > pa)
438 			len = end - pa;
439 	}
440 	*actual_len = len;
441 
442 	m = tail = extpg_alloc(cb_arg, how);
443 	if (m == NULL)
444 		return (NULL);
445 	while (len > 0) {
446 		if (!can_append_paddr(tail, pa)) {
447 			MBUF_EXT_PGS_ASSERT_SANITY(tail);
448 			tail->m_next = extpg_alloc(cb_arg, how);
449 			if (tail->m_next == NULL)
450 				goto error;
451 			tail = tail->m_next;
452 		}
453 
454 		appended = append_paddr_range(tail, pa, len);
455 		KASSERT(appended > 0, ("did not append anything"));
456 		KASSERT(appended <= len, ("appended too much"));
457 
458 		pa += appended;
459 		len -= appended;
460 	}
461 
462 	MBUF_EXT_PGS_ASSERT_SANITY(tail);
463 	return (m);
464 error:
465 	m_freem(m);
466 	return (NULL);
467 }
468 
469 static struct mbuf *
470 vlist_ext_mbuf(memdesc_alloc_ext_mbuf_t *ext_alloc, void *cb_arg, int how,
471     struct bus_dma_segment *vlist, u_int sglist_cnt, size_t offset,
472     size_t len, size_t *actual_len)
473 {
474 	struct mbuf *m, *n, *tail;
475 	size_t todo;
476 
477 	*actual_len = len;
478 
479 	while (vlist->ds_len <= offset) {
480 		KASSERT(sglist_cnt > 1, ("out of sglist entries"));
481 
482 		offset -= vlist->ds_len;
483 		vlist++;
484 		sglist_cnt--;
485 	}
486 
487 	m = tail = NULL;
488 	while (len > 0) {
489 		KASSERT(sglist_cnt >= 1, ("out of sglist entries"));
490 
491 		todo = len;
492 		if (todo > vlist->ds_len - offset)
493 			todo = vlist->ds_len - offset;
494 
495 		n = ext_alloc(cb_arg, how, (char *)(uintptr_t)vlist->ds_addr +
496 		    offset, todo);
497 		if (n == NULL)
498 			goto error;
499 
500 		if (m == NULL) {
501 			m = n;
502 			tail = m;
503 		} else {
504 			tail->m_next = n;
505 			tail = n;
506 		}
507 
508 		offset = 0;
509 		vlist++;
510 		sglist_cnt--;
511 		len -= todo;
512 	}
513 
514 	return (m);
515 error:
516 	m_freem(m);
517 	return (NULL);
518 }
519 
520 static struct mbuf *
521 plist_ext_mbuf(memdesc_alloc_extpg_mbuf_t *extpg_alloc, void *cb_arg, int how,
522     struct bus_dma_segment *plist, u_int sglist_cnt, size_t offset, size_t len,
523     size_t *actual_len, bool can_truncate)
524 {
525 	vm_paddr_t pa;
526 	struct mbuf *m, *tail;
527 	size_t appended, totlen, todo;
528 
529 	while (plist->ds_len <= offset) {
530 		KASSERT(sglist_cnt > 1, ("out of sglist entries"));
531 
532 		offset -= plist->ds_len;
533 		plist++;
534 		sglist_cnt--;
535 	}
536 
537 	totlen = 0;
538 	m = tail = extpg_alloc(cb_arg, how);
539 	if (m == NULL)
540 		return (NULL);
541 	while (len > 0) {
542 		KASSERT(sglist_cnt >= 1, ("out of sglist entries"));
543 
544 		pa = plist->ds_addr + offset;
545 		todo = len;
546 		if (todo > plist->ds_len - offset)
547 			todo = plist->ds_len - offset;
548 
549 		/*
550 		 * If truncation is enabled, avoid sending a final
551 		 * partial page, but only if there is more data
552 		 * available in the current segment.  Also, at least
553 		 * some data must be sent, so only drop the final page
554 		 * for this segment if the segment spans multiple
555 		 * pages or some other data is already queued.
556 		 */
557 		else if (can_truncate) {
558 			vm_paddr_t end;
559 
560 			end = trunc_page(pa + len);
561 			if (end <= pa && totlen != 0) {
562 				/*
563 				 * This last segment is only a partial
564 				 * page.
565 				 */
566 				len = 0;
567 				break;
568 			}
569 			todo = end - pa;
570 		}
571 
572 		offset = 0;
573 		len -= todo;
574 		totlen += todo;
575 
576 		while (todo > 0) {
577 			if (!can_append_paddr(tail, pa)) {
578 				MBUF_EXT_PGS_ASSERT_SANITY(tail);
579 				tail->m_next = extpg_alloc(cb_arg, how);
580 				if (tail->m_next == NULL)
581 					goto error;
582 				tail = tail->m_next;
583 			}
584 
585 			appended = append_paddr_range(tail, pa, todo);
586 			KASSERT(appended > 0, ("did not append anything"));
587 
588 			pa += appended;
589 			todo -= appended;
590 		}
591 	}
592 
593 	MBUF_EXT_PGS_ASSERT_SANITY(tail);
594 	*actual_len = totlen;
595 	return (m);
596 error:
597 	m_freem(m);
598 	return (NULL);
599 }
600 
601 static struct mbuf *
602 vmpages_ext_mbuf(memdesc_alloc_extpg_mbuf_t *extpg_alloc, void *cb_arg, int how,
603     vm_page_t *ma, size_t offset, size_t len, size_t *actual_len,
604     bool can_truncate)
605 {
606 	struct mbuf *m, *tail;
607 
608 	while (offset >= PAGE_SIZE) {
609 		ma++;
610 		offset -= PAGE_SIZE;
611 	}
612 
613 	if (can_truncate) {
614 		size_t end;
615 
616 		/*
617 		 * Trim any partial page at the end, but not if it's
618 		 * the only page.
619 		 */
620 		end = trunc_page(offset + len);
621 		if (end > offset)
622 			len = end - offset;
623 	}
624 	*actual_len = len;
625 
626 	m = tail = extpg_alloc(cb_arg, how);
627 	if (m == NULL)
628 		return (NULL);
629 
630 	/* First page. */
631 	m->m_epg_pa[0] = VM_PAGE_TO_PHYS(*ma);
632 	ma++;
633 	m->m_epg_npgs = 1;
634 	m->m_epg_1st_off = offset;
635 	m->m_epg_last_len = PAGE_SIZE - offset;
636 	if (m->m_epg_last_len > len)
637 		m->m_epg_last_len = len;
638 	m->m_len = m->m_epg_last_len;
639 	len -= m->m_epg_last_len;
640 
641 	/* Full pages. */
642 	while (len >= PAGE_SIZE) {
643 		if (tail->m_epg_npgs == MBUF_PEXT_MAX_PGS) {
644 			MBUF_EXT_PGS_ASSERT_SANITY(tail);
645 			tail->m_next = extpg_alloc(cb_arg, how);
646 			if (tail->m_next == NULL)
647 				goto error;
648 			tail = tail->m_next;
649 		}
650 
651 		tail->m_epg_pa[tail->m_epg_npgs] = VM_PAGE_TO_PHYS(*ma);
652 		ma++;
653 		tail->m_epg_npgs++;
654 		tail->m_epg_last_len = PAGE_SIZE;
655 		tail->m_len += PAGE_SIZE;
656 		len -= PAGE_SIZE;
657 	}
658 
659 	/* Last partial page. */
660 	if (len > 0) {
661 		if (tail->m_epg_npgs == MBUF_PEXT_MAX_PGS) {
662 			MBUF_EXT_PGS_ASSERT_SANITY(tail);
663 			tail->m_next = extpg_alloc(cb_arg, how);
664 			if (tail->m_next == NULL)
665 				goto error;
666 			tail = tail->m_next;
667 		}
668 
669 		tail->m_epg_pa[tail->m_epg_npgs] = VM_PAGE_TO_PHYS(*ma);
670 		ma++;
671 		tail->m_epg_npgs++;
672 		tail->m_epg_last_len = len;
673 		tail->m_len += len;
674 	}
675 
676 	MBUF_EXT_PGS_ASSERT_SANITY(tail);
677 	return (m);
678 error:
679 	m_freem(m);
680 	return (NULL);
681 }
682 
683 /*
684  * Somewhat similar to m_copym but optionally avoids a partial mbuf at
685  * the end.
686  */
687 static struct mbuf *
688 mbuf_subchain(struct mbuf *m0, size_t offset, size_t len,
689     size_t *actual_len, bool can_truncate, int how)
690 {
691 	struct mbuf *m, *tail;
692 	size_t totlen;
693 
694 	while (offset >= m0->m_len) {
695 		offset -= m0->m_len;
696 		m0 = m0->m_next;
697 	}
698 
699 	/* Always return at least one mbuf. */
700 	totlen = m0->m_len - offset;
701 	if (totlen > len)
702 		totlen = len;
703 
704 	m = m_get(how, MT_DATA);
705 	if (m == NULL)
706 		return (NULL);
707 	m->m_len = totlen;
708 	if (m0->m_flags & (M_EXT | M_EXTPG)) {
709 		m->m_data = m0->m_data + offset;
710 		mb_dupcl(m, m0);
711 	} else
712 		memcpy(mtod(m, void *), mtodo(m0, offset), m->m_len);
713 
714 	tail = m;
715 	m0 = m0->m_next;
716 	len -= totlen;
717 	while (len > 0) {
718 		/*
719 		 * If truncation is enabled, don't send any partial
720 		 * mbufs besides the first one.
721 		 */
722 		if (can_truncate && m0->m_len > len)
723 			break;
724 
725 		tail->m_next = m_get(how, MT_DATA);
726 		if (tail->m_next == NULL)
727 			goto error;
728 		tail = tail->m_next;
729 		tail->m_len = m0->m_len;
730 		if (m0->m_flags & (M_EXT | M_EXTPG)) {
731 			tail->m_data = m0->m_data;
732 			mb_dupcl(tail, m0);
733 		} else
734 			memcpy(mtod(tail, void *), mtod(m0, void *),
735 			    tail->m_len);
736 
737 		totlen += tail->m_len;
738 		m0 = m0->m_next;
739 		len -= tail->m_len;
740 	}
741 	*actual_len = totlen;
742 	return (m);
743 error:
744 	m_freem(m);
745 	return (NULL);
746 }
747 
748 struct mbuf *
749 memdesc_alloc_ext_mbufs(struct memdesc *mem,
750     memdesc_alloc_ext_mbuf_t *ext_alloc,
751     memdesc_alloc_extpg_mbuf_t *extpg_alloc, void *cb_arg, int how,
752     size_t offset, size_t len, size_t *actual_len, bool can_truncate)
753 {
754 	struct mbuf *m;
755 	size_t done;
756 
757 	switch (mem->md_type) {
758 	case MEMDESC_VADDR:
759 		m = vaddr_ext_mbuf(ext_alloc, cb_arg, how,
760 		    (char *)mem->u.md_vaddr + offset, len, &done);
761 		break;
762 	case MEMDESC_PADDR:
763 		m = paddr_ext_mbuf(extpg_alloc, cb_arg, how, mem->u.md_paddr +
764 		    offset, len, &done, can_truncate);
765 		break;
766 	case MEMDESC_VLIST:
767 		m = vlist_ext_mbuf(ext_alloc, cb_arg, how, mem->u.md_list,
768 		    mem->md_nseg, offset, len, &done);
769 		break;
770 	case MEMDESC_PLIST:
771 		m = plist_ext_mbuf(extpg_alloc, cb_arg, how, mem->u.md_list,
772 		    mem->md_nseg, offset, len, &done, can_truncate);
773 		break;
774 	case MEMDESC_UIO:
775 		panic("uio not supported");
776 	case MEMDESC_MBUF:
777 		m = mbuf_subchain(mem->u.md_mbuf, offset, len, &done,
778 		    can_truncate, how);
779 		break;
780 	case MEMDESC_VMPAGES:
781 		m = vmpages_ext_mbuf(extpg_alloc, cb_arg, how, mem->u.md_ma,
782 		    mem->md_offset + offset, len, &done, can_truncate);
783 		break;
784 	default:
785 		__assert_unreachable();
786 	}
787 	if (m == NULL)
788 		return (NULL);
789 
790 	if (can_truncate) {
791 		KASSERT(done <= len, ("chain too long"));
792 	} else {
793 		KASSERT(done == len, ("short chain with no limit"));
794 	}
795 	KASSERT(m_length(m, NULL) == done, ("length mismatch"));
796 	if (actual_len != NULL)
797 		*actual_len = done;
798 	return (m);
799 }
800