xref: /freebsd/sys/kern/subr_sglist.c (revision 2f513db7)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2008 Yahoo!, Inc.
5  * All rights reserved.
6  * Written by: John Baldwin <jhb@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the author nor the names of any co-contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/bio.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/proc.h>
42 #include <sys/sglist.h>
43 #include <sys/uio.h>
44 
45 #include <vm/vm.h>
46 #include <vm/vm_page.h>
47 #include <vm/pmap.h>
48 #include <vm/vm_map.h>
49 
50 #include <sys/ktr.h>
51 
52 static MALLOC_DEFINE(M_SGLIST, "sglist", "scatter/gather lists");
53 
54 /*
55  * Convenience macros to save the state of an sglist so it can be restored
56  * if an append attempt fails.  Since sglist's only grow we only need to
57  * save the current count of segments and the length of the ending segment.
58  * Earlier segments will not be changed by an append, and the only change
59  * that can occur to the ending segment is that it can be extended.
60  */
61 struct sgsave {
62 	u_short sg_nseg;
63 	size_t ss_len;
64 };
65 
66 #define	SGLIST_SAVE(sg, sgsave) do {					\
67 	(sgsave).sg_nseg = (sg)->sg_nseg;				\
68 	if ((sgsave).sg_nseg > 0)					\
69 		(sgsave).ss_len = (sg)->sg_segs[(sgsave).sg_nseg - 1].ss_len; \
70 	else								\
71 		(sgsave).ss_len = 0;					\
72 } while (0)
73 
74 #define	SGLIST_RESTORE(sg, sgsave) do {					\
75 	(sg)->sg_nseg = (sgsave).sg_nseg;				\
76 	if ((sgsave).sg_nseg > 0)					\
77 		(sg)->sg_segs[(sgsave).sg_nseg - 1].ss_len = (sgsave).ss_len; \
78 } while (0)
79 
80 /*
81  * Append a single (paddr, len) to a sglist.  sg is the list and ss is
82  * the current segment in the list.  If we run out of segments then
83  * EFBIG will be returned.
84  */
85 static __inline int
86 _sglist_append_range(struct sglist *sg, struct sglist_seg **ssp,
87     vm_paddr_t paddr, size_t len)
88 {
89 	struct sglist_seg *ss;
90 
91 	ss = *ssp;
92 	if (ss->ss_paddr + ss->ss_len == paddr)
93 		ss->ss_len += len;
94 	else {
95 		if (sg->sg_nseg == sg->sg_maxseg)
96 			return (EFBIG);
97 		ss++;
98 		ss->ss_paddr = paddr;
99 		ss->ss_len = len;
100 		sg->sg_nseg++;
101 		*ssp = ss;
102 	}
103 	return (0);
104 }
105 
106 /*
107  * Worker routine to append a virtual address range (either kernel or
108  * user) to a scatter/gather list.
109  */
110 static __inline int
111 _sglist_append_buf(struct sglist *sg, void *buf, size_t len, pmap_t pmap,
112     size_t *donep)
113 {
114 	struct sglist_seg *ss;
115 	vm_offset_t vaddr, offset;
116 	vm_paddr_t paddr;
117 	size_t seglen;
118 	int error;
119 
120 	if (donep)
121 		*donep = 0;
122 	if (len == 0)
123 		return (0);
124 
125 	/* Do the first page.  It may have an offset. */
126 	vaddr = (vm_offset_t)buf;
127 	offset = vaddr & PAGE_MASK;
128 	if (pmap != NULL)
129 		paddr = pmap_extract(pmap, vaddr);
130 	else
131 		paddr = pmap_kextract(vaddr);
132 	seglen = MIN(len, PAGE_SIZE - offset);
133 	if (sg->sg_nseg == 0) {
134 		ss = sg->sg_segs;
135 		ss->ss_paddr = paddr;
136 		ss->ss_len = seglen;
137 		sg->sg_nseg = 1;
138 	} else {
139 		ss = &sg->sg_segs[sg->sg_nseg - 1];
140 		error = _sglist_append_range(sg, &ss, paddr, seglen);
141 		if (error)
142 			return (error);
143 	}
144 	vaddr += seglen;
145 	len -= seglen;
146 	if (donep)
147 		*donep += seglen;
148 
149 	while (len > 0) {
150 		seglen = MIN(len, PAGE_SIZE);
151 		if (pmap != NULL)
152 			paddr = pmap_extract(pmap, vaddr);
153 		else
154 			paddr = pmap_kextract(vaddr);
155 		error = _sglist_append_range(sg, &ss, paddr, seglen);
156 		if (error)
157 			return (error);
158 		vaddr += seglen;
159 		len -= seglen;
160 		if (donep)
161 			*donep += seglen;
162 	}
163 
164 	return (0);
165 }
166 
167 /*
168  * Determine the number of scatter/gather list elements needed to
169  * describe a kernel virtual address range.
170  */
171 int
172 sglist_count(void *buf, size_t len)
173 {
174 	vm_offset_t vaddr, vendaddr;
175 	vm_paddr_t lastaddr, paddr;
176 	int nsegs;
177 
178 	if (len == 0)
179 		return (0);
180 
181 	vaddr = trunc_page((vm_offset_t)buf);
182 	vendaddr = (vm_offset_t)buf + len;
183 	nsegs = 1;
184 	lastaddr = pmap_kextract(vaddr);
185 	vaddr += PAGE_SIZE;
186 	while (vaddr < vendaddr) {
187 		paddr = pmap_kextract(vaddr);
188 		if (lastaddr + PAGE_SIZE != paddr)
189 			nsegs++;
190 		lastaddr = paddr;
191 		vaddr += PAGE_SIZE;
192 	}
193 	return (nsegs);
194 }
195 
196 /*
197  * Determine the number of scatter/gather list elements needed to
198  * describe a buffer backed by an array of VM pages.
199  */
200 int
201 sglist_count_vmpages(vm_page_t *m, size_t pgoff, size_t len)
202 {
203 	vm_paddr_t lastaddr, paddr;
204 	int i, nsegs;
205 
206 	if (len == 0)
207 		return (0);
208 
209 	len += pgoff;
210 	nsegs = 1;
211 	lastaddr = VM_PAGE_TO_PHYS(m[0]);
212 	for (i = 1; len > PAGE_SIZE; len -= PAGE_SIZE, i++) {
213 		paddr = VM_PAGE_TO_PHYS(m[i]);
214 		if (lastaddr + PAGE_SIZE != paddr)
215 			nsegs++;
216 		lastaddr = paddr;
217 	}
218 	return (nsegs);
219 }
220 
221 /*
222  * Determine the number of scatter/gather list elements needed to
223  * describe an EXT_PGS buffer.
224  */
225 int
226 sglist_count_ext_pgs(struct mbuf_ext_pgs *ext_pgs, size_t off, size_t len)
227 {
228 	vm_paddr_t nextaddr, paddr;
229 	size_t seglen, segoff;
230 	int i, nsegs, pglen, pgoff;
231 
232 	if (len == 0)
233 		return (0);
234 
235 	nsegs = 0;
236 	if (ext_pgs->hdr_len != 0) {
237 		if (off >= ext_pgs->hdr_len) {
238 			off -= ext_pgs->hdr_len;
239 		} else {
240 			seglen = ext_pgs->hdr_len - off;
241 			segoff = off;
242 			seglen = MIN(seglen, len);
243 			off = 0;
244 			len -= seglen;
245 			nsegs += sglist_count(&ext_pgs->hdr[segoff], seglen);
246 		}
247 	}
248 	nextaddr = 0;
249 	pgoff = ext_pgs->first_pg_off;
250 	for (i = 0; i < ext_pgs->npgs && len > 0; i++) {
251 		pglen = mbuf_ext_pg_len(ext_pgs, i, pgoff);
252 		if (off >= pglen) {
253 			off -= pglen;
254 			pgoff = 0;
255 			continue;
256 		}
257 		seglen = pglen - off;
258 		segoff = pgoff + off;
259 		off = 0;
260 		seglen = MIN(seglen, len);
261 		len -= seglen;
262 		paddr = ext_pgs->pa[i] + segoff;
263 		if (paddr != nextaddr)
264 			nsegs++;
265 		nextaddr = paddr + seglen;
266 		pgoff = 0;
267 	};
268 	if (len != 0) {
269 		seglen = MIN(len, ext_pgs->trail_len - off);
270 		len -= seglen;
271 		nsegs += sglist_count(&ext_pgs->trail[off], seglen);
272 	}
273 	KASSERT(len == 0, ("len != 0"));
274 	return (nsegs);
275 }
276 
277 /*
278  * Determine the number of scatter/gather list elements needed to
279  * describe an EXT_PGS mbuf.
280  */
281 int
282 sglist_count_mb_ext_pgs(struct mbuf *m)
283 {
284 
285 	MBUF_EXT_PGS_ASSERT(m);
286 	return (sglist_count_ext_pgs(m->m_ext.ext_pgs, mtod(m, vm_offset_t),
287 	    m->m_len));
288 }
289 
290 /*
291  * Allocate a scatter/gather list along with 'nsegs' segments.  The
292  * 'mflags' parameters are the same as passed to malloc(9).  The caller
293  * should use sglist_free() to free this list.
294  */
295 struct sglist *
296 sglist_alloc(int nsegs, int mflags)
297 {
298 	struct sglist *sg;
299 
300 	sg = malloc(sizeof(struct sglist) + nsegs * sizeof(struct sglist_seg),
301 	    M_SGLIST, mflags);
302 	if (sg == NULL)
303 		return (NULL);
304 	sglist_init(sg, nsegs, (struct sglist_seg *)(sg + 1));
305 	return (sg);
306 }
307 
308 /*
309  * Free a scatter/gather list allocated via sglist_allc().
310  */
311 void
312 sglist_free(struct sglist *sg)
313 {
314 
315 	if (sg == NULL)
316 		return;
317 
318 	if (refcount_release(&sg->sg_refs))
319 		free(sg, M_SGLIST);
320 }
321 
322 /*
323  * Append the segments to describe a single kernel virtual address
324  * range to a scatter/gather list.  If there are insufficient
325  * segments, then this fails with EFBIG.
326  */
327 int
328 sglist_append(struct sglist *sg, void *buf, size_t len)
329 {
330 	struct sgsave save;
331 	int error;
332 
333 	if (sg->sg_maxseg == 0)
334 		return (EINVAL);
335 	SGLIST_SAVE(sg, save);
336 	error = _sglist_append_buf(sg, buf, len, NULL, NULL);
337 	if (error)
338 		SGLIST_RESTORE(sg, save);
339 	return (error);
340 }
341 
342 /*
343  * Append the segments to describe a bio's data to a scatter/gather list.
344  * If there are insufficient segments, then this fails with EFBIG.
345  *
346  * NOTE: This function expects bio_bcount to be initialized.
347  */
348 int
349 sglist_append_bio(struct sglist *sg, struct bio *bp)
350 {
351 	int error;
352 
353 	if ((bp->bio_flags & BIO_UNMAPPED) == 0)
354 		error = sglist_append(sg, bp->bio_data, bp->bio_bcount);
355 	else
356 		error = sglist_append_vmpages(sg, bp->bio_ma,
357 		    bp->bio_ma_offset, bp->bio_bcount);
358 	return (error);
359 }
360 
361 /*
362  * Append a single physical address range to a scatter/gather list.
363  * If there are insufficient segments, then this fails with EFBIG.
364  */
365 int
366 sglist_append_phys(struct sglist *sg, vm_paddr_t paddr, size_t len)
367 {
368 	struct sglist_seg *ss;
369 	struct sgsave save;
370 	int error;
371 
372 	if (sg->sg_maxseg == 0)
373 		return (EINVAL);
374 	if (len == 0)
375 		return (0);
376 
377 	if (sg->sg_nseg == 0) {
378 		sg->sg_segs[0].ss_paddr = paddr;
379 		sg->sg_segs[0].ss_len = len;
380 		sg->sg_nseg = 1;
381 		return (0);
382 	}
383 	ss = &sg->sg_segs[sg->sg_nseg - 1];
384 	SGLIST_SAVE(sg, save);
385 	error = _sglist_append_range(sg, &ss, paddr, len);
386 	if (error)
387 		SGLIST_RESTORE(sg, save);
388 	return (error);
389 }
390 
391 /*
392  * Append the segments to describe an EXT_PGS buffer to a
393  * scatter/gather list.  If there are insufficient segments, then this
394  * fails with EFBIG.
395  */
396 int
397 sglist_append_ext_pgs(struct sglist *sg, struct mbuf_ext_pgs *ext_pgs,
398     size_t off, size_t len)
399 {
400 	size_t seglen, segoff;
401 	vm_paddr_t paddr;
402 	int error, i, pglen, pgoff;
403 
404 	error = 0;
405 	if (ext_pgs->hdr_len != 0) {
406 		if (off >= ext_pgs->hdr_len) {
407 			off -= ext_pgs->hdr_len;
408 		} else {
409 			seglen = ext_pgs->hdr_len - off;
410 			segoff = off;
411 			seglen = MIN(seglen, len);
412 			off = 0;
413 			len -= seglen;
414 			error = sglist_append(sg,
415 			    &ext_pgs->hdr[segoff], seglen);
416 		}
417 	}
418 	pgoff = ext_pgs->first_pg_off;
419 	for (i = 0; i < ext_pgs->npgs && error == 0 && len > 0; i++) {
420 		pglen = mbuf_ext_pg_len(ext_pgs, i, pgoff);
421 		if (off >= pglen) {
422 			off -= pglen;
423 			pgoff = 0;
424 			continue;
425 		}
426 		seglen = pglen - off;
427 		segoff = pgoff + off;
428 		off = 0;
429 		seglen = MIN(seglen, len);
430 		len -= seglen;
431 		paddr = ext_pgs->pa[i] + segoff;
432 		error = sglist_append_phys(sg, paddr, seglen);
433 		pgoff = 0;
434 	};
435 	if (error == 0 && len > 0) {
436 		seglen = MIN(len, ext_pgs->trail_len - off);
437 		len -= seglen;
438 		error = sglist_append(sg,
439 		    &ext_pgs->trail[off], seglen);
440 	}
441 	if (error == 0)
442 		KASSERT(len == 0, ("len != 0"));
443 	return (error);
444 }
445 
446 /*
447  * Append the segments to describe an EXT_PGS mbuf to a scatter/gather
448  * list.  If there are insufficient segments, then this fails with
449  * EFBIG.
450  */
451 int
452 sglist_append_mb_ext_pgs(struct sglist *sg, struct mbuf *m)
453 {
454 
455 	/* for now, all unmapped mbufs are assumed to be EXT_PGS */
456 	MBUF_EXT_PGS_ASSERT(m);
457 	return (sglist_append_ext_pgs(sg, m->m_ext.ext_pgs,
458 	    mtod(m, vm_offset_t), m->m_len));
459 }
460 
461 /*
462  * Append the segments that describe a single mbuf chain to a
463  * scatter/gather list.  If there are insufficient segments, then this
464  * fails with EFBIG.
465  */
466 int
467 sglist_append_mbuf(struct sglist *sg, struct mbuf *m0)
468 {
469 	struct sgsave save;
470 	struct mbuf *m;
471 	int error;
472 
473 	if (sg->sg_maxseg == 0)
474 		return (EINVAL);
475 
476 	error = 0;
477 	SGLIST_SAVE(sg, save);
478 	for (m = m0; m != NULL; m = m->m_next) {
479 		if (m->m_len > 0) {
480 			if ((m->m_flags & M_NOMAP) != 0)
481 				error = sglist_append_mb_ext_pgs(sg, m);
482 			else
483 				error = sglist_append(sg, m->m_data,
484 				    m->m_len);
485 			if (error) {
486 				SGLIST_RESTORE(sg, save);
487 				return (error);
488 			}
489 		}
490 	}
491 	return (0);
492 }
493 
494 /*
495  * Append the segments that describe a buffer spanning an array of VM
496  * pages.  The buffer begins at an offset of 'pgoff' in the first
497  * page.
498  */
499 int
500 sglist_append_vmpages(struct sglist *sg, vm_page_t *m, size_t pgoff,
501     size_t len)
502 {
503 	struct sgsave save;
504 	struct sglist_seg *ss;
505 	vm_paddr_t paddr;
506 	size_t seglen;
507 	int error, i;
508 
509 	if (sg->sg_maxseg == 0)
510 		return (EINVAL);
511 	if (len == 0)
512 		return (0);
513 
514 	SGLIST_SAVE(sg, save);
515 	i = 0;
516 	if (sg->sg_nseg == 0) {
517 		seglen = min(PAGE_SIZE - pgoff, len);
518 		sg->sg_segs[0].ss_paddr = VM_PAGE_TO_PHYS(m[0]) + pgoff;
519 		sg->sg_segs[0].ss_len = seglen;
520 		sg->sg_nseg = 1;
521 		pgoff = 0;
522 		len -= seglen;
523 		i++;
524 	}
525 	ss = &sg->sg_segs[sg->sg_nseg - 1];
526 	for (; len > 0; i++, len -= seglen) {
527 		seglen = min(PAGE_SIZE - pgoff, len);
528 		paddr = VM_PAGE_TO_PHYS(m[i]) + pgoff;
529 		error = _sglist_append_range(sg, &ss, paddr, seglen);
530 		if (error) {
531 			SGLIST_RESTORE(sg, save);
532 			return (error);
533 		}
534 		pgoff = 0;
535 	}
536 	return (0);
537 }
538 
539 /*
540  * Append the segments that describe a single user address range to a
541  * scatter/gather list.  If there are insufficient segments, then this
542  * fails with EFBIG.
543  */
544 int
545 sglist_append_user(struct sglist *sg, void *buf, size_t len, struct thread *td)
546 {
547 	struct sgsave save;
548 	int error;
549 
550 	if (sg->sg_maxseg == 0)
551 		return (EINVAL);
552 	SGLIST_SAVE(sg, save);
553 	error = _sglist_append_buf(sg, buf, len,
554 	    vmspace_pmap(td->td_proc->p_vmspace), NULL);
555 	if (error)
556 		SGLIST_RESTORE(sg, save);
557 	return (error);
558 }
559 
560 /*
561  * Append a subset of an existing scatter/gather list 'source' to a
562  * the scatter/gather list 'sg'.  If there are insufficient segments,
563  * then this fails with EFBIG.
564  */
565 int
566 sglist_append_sglist(struct sglist *sg, struct sglist *source, size_t offset,
567     size_t length)
568 {
569 	struct sgsave save;
570 	struct sglist_seg *ss;
571 	size_t seglen;
572 	int error, i;
573 
574 	if (sg->sg_maxseg == 0 || length == 0)
575 		return (EINVAL);
576 	SGLIST_SAVE(sg, save);
577 	error = EINVAL;
578 	ss = &sg->sg_segs[sg->sg_nseg - 1];
579 	for (i = 0; i < source->sg_nseg; i++) {
580 		if (offset >= source->sg_segs[i].ss_len) {
581 			offset -= source->sg_segs[i].ss_len;
582 			continue;
583 		}
584 		seglen = source->sg_segs[i].ss_len - offset;
585 		if (seglen > length)
586 			seglen = length;
587 		error = _sglist_append_range(sg, &ss,
588 		    source->sg_segs[i].ss_paddr + offset, seglen);
589 		if (error)
590 			break;
591 		offset = 0;
592 		length -= seglen;
593 		if (length == 0)
594 			break;
595 	}
596 	if (length != 0)
597 		error = EINVAL;
598 	if (error)
599 		SGLIST_RESTORE(sg, save);
600 	return (error);
601 }
602 
603 /*
604  * Append the segments that describe a single uio to a scatter/gather
605  * list.  If there are insufficient segments, then this fails with
606  * EFBIG.
607  */
608 int
609 sglist_append_uio(struct sglist *sg, struct uio *uio)
610 {
611 	struct iovec *iov;
612 	struct sgsave save;
613 	size_t resid, minlen;
614 	pmap_t pmap;
615 	int error, i;
616 
617 	if (sg->sg_maxseg == 0)
618 		return (EINVAL);
619 
620 	resid = uio->uio_resid;
621 	iov = uio->uio_iov;
622 
623 	if (uio->uio_segflg == UIO_USERSPACE) {
624 		KASSERT(uio->uio_td != NULL,
625 		    ("sglist_append_uio: USERSPACE but no thread"));
626 		pmap = vmspace_pmap(uio->uio_td->td_proc->p_vmspace);
627 	} else
628 		pmap = NULL;
629 
630 	error = 0;
631 	SGLIST_SAVE(sg, save);
632 	for (i = 0; i < uio->uio_iovcnt && resid != 0; i++) {
633 		/*
634 		 * Now at the first iovec to load.  Load each iovec
635 		 * until we have exhausted the residual count.
636 		 */
637 		minlen = MIN(resid, iov[i].iov_len);
638 		if (minlen > 0) {
639 			error = _sglist_append_buf(sg, iov[i].iov_base, minlen,
640 			    pmap, NULL);
641 			if (error) {
642 				SGLIST_RESTORE(sg, save);
643 				return (error);
644 			}
645 			resid -= minlen;
646 		}
647 	}
648 	return (0);
649 }
650 
651 /*
652  * Append the segments that describe at most 'resid' bytes from a
653  * single uio to a scatter/gather list.  If there are insufficient
654  * segments, then only the amount that fits is appended.
655  */
656 int
657 sglist_consume_uio(struct sglist *sg, struct uio *uio, size_t resid)
658 {
659 	struct iovec *iov;
660 	size_t done;
661 	pmap_t pmap;
662 	int error, len;
663 
664 	if (sg->sg_maxseg == 0)
665 		return (EINVAL);
666 
667 	if (uio->uio_segflg == UIO_USERSPACE) {
668 		KASSERT(uio->uio_td != NULL,
669 		    ("sglist_consume_uio: USERSPACE but no thread"));
670 		pmap = vmspace_pmap(uio->uio_td->td_proc->p_vmspace);
671 	} else
672 		pmap = NULL;
673 
674 	error = 0;
675 	while (resid > 0 && uio->uio_resid) {
676 		iov = uio->uio_iov;
677 		len = iov->iov_len;
678 		if (len == 0) {
679 			uio->uio_iov++;
680 			uio->uio_iovcnt--;
681 			continue;
682 		}
683 		if (len > resid)
684 			len = resid;
685 
686 		/*
687 		 * Try to append this iovec.  If we run out of room,
688 		 * then break out of the loop.
689 		 */
690 		error = _sglist_append_buf(sg, iov->iov_base, len, pmap, &done);
691 		iov->iov_base = (char *)iov->iov_base + done;
692 		iov->iov_len -= done;
693 		uio->uio_resid -= done;
694 		uio->uio_offset += done;
695 		resid -= done;
696 		if (error)
697 			break;
698 	}
699 	return (0);
700 }
701 
702 /*
703  * Allocate and populate a scatter/gather list to describe a single
704  * kernel virtual address range.
705  */
706 struct sglist *
707 sglist_build(void *buf, size_t len, int mflags)
708 {
709 	struct sglist *sg;
710 	int nsegs;
711 
712 	if (len == 0)
713 		return (NULL);
714 
715 	nsegs = sglist_count(buf, len);
716 	sg = sglist_alloc(nsegs, mflags);
717 	if (sg == NULL)
718 		return (NULL);
719 	if (sglist_append(sg, buf, len) != 0) {
720 		sglist_free(sg);
721 		return (NULL);
722 	}
723 	return (sg);
724 }
725 
726 /*
727  * Clone a new copy of a scatter/gather list.
728  */
729 struct sglist *
730 sglist_clone(struct sglist *sg, int mflags)
731 {
732 	struct sglist *new;
733 
734 	if (sg == NULL)
735 		return (NULL);
736 	new = sglist_alloc(sg->sg_maxseg, mflags);
737 	if (new == NULL)
738 		return (NULL);
739 	new->sg_nseg = sg->sg_nseg;
740 	bcopy(sg->sg_segs, new->sg_segs, sizeof(struct sglist_seg) *
741 	    sg->sg_nseg);
742 	return (new);
743 }
744 
745 /*
746  * Calculate the total length of the segments described in a
747  * scatter/gather list.
748  */
749 size_t
750 sglist_length(struct sglist *sg)
751 {
752 	size_t space;
753 	int i;
754 
755 	space = 0;
756 	for (i = 0; i < sg->sg_nseg; i++)
757 		space += sg->sg_segs[i].ss_len;
758 	return (space);
759 }
760 
761 /*
762  * Split a scatter/gather list into two lists.  The scatter/gather
763  * entries for the first 'length' bytes of the 'original' list are
764  * stored in the '*head' list and are removed from 'original'.
765  *
766  * If '*head' is NULL, then a new list will be allocated using
767  * 'mflags'.  If M_NOWAIT is specified and the allocation fails,
768  * ENOMEM will be returned.
769  *
770  * If '*head' is not NULL, it should point to an empty sglist.  If it
771  * does not have enough room for the remaining space, then EFBIG will
772  * be returned.  If '*head' is not empty, then EINVAL will be
773  * returned.
774  *
775  * If 'original' is shared (refcount > 1), then EDOOFUS will be
776  * returned.
777  */
778 int
779 sglist_split(struct sglist *original, struct sglist **head, size_t length,
780     int mflags)
781 {
782 	struct sglist *sg;
783 	size_t space, split;
784 	int count, i;
785 
786 	if (original->sg_refs > 1)
787 		return (EDOOFUS);
788 
789 	/* Figure out how big of a sglist '*head' has to hold. */
790 	count = 0;
791 	space = 0;
792 	split = 0;
793 	for (i = 0; i < original->sg_nseg; i++) {
794 		space += original->sg_segs[i].ss_len;
795 		count++;
796 		if (space >= length) {
797 			/*
798 			 * If 'length' falls in the middle of a
799 			 * scatter/gather list entry, then 'split'
800 			 * holds how much of that entry will remain in
801 			 * 'original'.
802 			 */
803 			split = space - length;
804 			break;
805 		}
806 	}
807 
808 	/* Nothing to do, so leave head empty. */
809 	if (count == 0)
810 		return (0);
811 
812 	if (*head == NULL) {
813 		sg = sglist_alloc(count, mflags);
814 		if (sg == NULL)
815 			return (ENOMEM);
816 		*head = sg;
817 	} else {
818 		sg = *head;
819 		if (sg->sg_maxseg < count)
820 			return (EFBIG);
821 		if (sg->sg_nseg != 0)
822 			return (EINVAL);
823 	}
824 
825 	/* Copy 'count' entries to 'sg' from 'original'. */
826 	bcopy(original->sg_segs, sg->sg_segs, count *
827 	    sizeof(struct sglist_seg));
828 	sg->sg_nseg = count;
829 
830 	/*
831 	 * If we had to split a list entry, fixup the last entry in
832 	 * 'sg' and the new first entry in 'original'.  We also
833 	 * decrement 'count' by 1 since we will only be removing
834 	 * 'count - 1' segments from 'original' now.
835 	 */
836 	if (split != 0) {
837 		count--;
838 		sg->sg_segs[count].ss_len -= split;
839 		original->sg_segs[count].ss_paddr =
840 		    sg->sg_segs[count].ss_paddr + split;
841 		original->sg_segs[count].ss_len = split;
842 	}
843 
844 	/* Trim 'count' entries from the front of 'original'. */
845 	original->sg_nseg -= count;
846 	bcopy(original->sg_segs + count, original->sg_segs, count *
847 	    sizeof(struct sglist_seg));
848 	return (0);
849 }
850 
851 /*
852  * Append the scatter/gather list elements in 'second' to the
853  * scatter/gather list 'first'.  If there is not enough space in
854  * 'first', EFBIG is returned.
855  */
856 int
857 sglist_join(struct sglist *first, struct sglist *second)
858 {
859 	struct sglist_seg *flast, *sfirst;
860 	int append;
861 
862 	/* If 'second' is empty, there is nothing to do. */
863 	if (second->sg_nseg == 0)
864 		return (0);
865 
866 	/*
867 	 * If the first entry in 'second' can be appended to the last entry
868 	 * in 'first' then set append to '1'.
869 	 */
870 	append = 0;
871 	flast = &first->sg_segs[first->sg_nseg - 1];
872 	sfirst = &second->sg_segs[0];
873 	if (first->sg_nseg != 0 &&
874 	    flast->ss_paddr + flast->ss_len == sfirst->ss_paddr)
875 		append = 1;
876 
877 	/* Make sure 'first' has enough room. */
878 	if (first->sg_nseg + second->sg_nseg - append > first->sg_maxseg)
879 		return (EFBIG);
880 
881 	/* Merge last in 'first' and first in 'second' if needed. */
882 	if (append)
883 		flast->ss_len += sfirst->ss_len;
884 
885 	/* Append new segments from 'second' to 'first'. */
886 	bcopy(first->sg_segs + first->sg_nseg, second->sg_segs + append,
887 	    (second->sg_nseg - append) * sizeof(struct sglist_seg));
888 	first->sg_nseg += second->sg_nseg - append;
889 	sglist_reset(second);
890 	return (0);
891 }
892 
893 /*
894  * Generate a new scatter/gather list from a range of an existing
895  * scatter/gather list.  The 'offset' and 'length' parameters specify
896  * the logical range of the 'original' list to extract.  If that range
897  * is not a subset of the length of 'original', then EINVAL is
898  * returned.  The new scatter/gather list is stored in '*slice'.
899  *
900  * If '*slice' is NULL, then a new list will be allocated using
901  * 'mflags'.  If M_NOWAIT is specified and the allocation fails,
902  * ENOMEM will be returned.
903  *
904  * If '*slice' is not NULL, it should point to an empty sglist.  If it
905  * does not have enough room for the remaining space, then EFBIG will
906  * be returned.  If '*slice' is not empty, then EINVAL will be
907  * returned.
908  */
909 int
910 sglist_slice(struct sglist *original, struct sglist **slice, size_t offset,
911     size_t length, int mflags)
912 {
913 	struct sglist *sg;
914 	size_t space, end, foffs, loffs;
915 	int count, i, fseg;
916 
917 	/* Nothing to do. */
918 	if (length == 0)
919 		return (0);
920 
921 	/* Figure out how many segments '*slice' needs to have. */
922 	end = offset + length;
923 	space = 0;
924 	count = 0;
925 	fseg = 0;
926 	foffs = loffs = 0;
927 	for (i = 0; i < original->sg_nseg; i++) {
928 		space += original->sg_segs[i].ss_len;
929 		if (space > offset) {
930 			/*
931 			 * When we hit the first segment, store its index
932 			 * in 'fseg' and the offset into the first segment
933 			 * of 'offset' in 'foffs'.
934 			 */
935 			if (count == 0) {
936 				fseg = i;
937 				foffs = offset - (space -
938 				    original->sg_segs[i].ss_len);
939 				CTR1(KTR_DEV, "sglist_slice: foffs = %08lx",
940 				    foffs);
941 			}
942 			count++;
943 
944 			/*
945 			 * When we hit the last segment, break out of
946 			 * the loop.  Store the amount of extra space
947 			 * at the end of this segment in 'loffs'.
948 			 */
949 			if (space >= end) {
950 				loffs = space - end;
951 				CTR1(KTR_DEV, "sglist_slice: loffs = %08lx",
952 				    loffs);
953 				break;
954 			}
955 		}
956 	}
957 
958 	/* If we never hit 'end', then 'length' ran off the end, so fail. */
959 	if (space < end)
960 		return (EINVAL);
961 
962 	if (*slice == NULL) {
963 		sg = sglist_alloc(count, mflags);
964 		if (sg == NULL)
965 			return (ENOMEM);
966 		*slice = sg;
967 	} else {
968 		sg = *slice;
969 		if (sg->sg_maxseg < count)
970 			return (EFBIG);
971 		if (sg->sg_nseg != 0)
972 			return (EINVAL);
973 	}
974 
975 	/*
976 	 * Copy over 'count' segments from 'original' starting at
977 	 * 'fseg' to 'sg'.
978 	 */
979 	bcopy(original->sg_segs + fseg, sg->sg_segs,
980 	    count * sizeof(struct sglist_seg));
981 	sg->sg_nseg = count;
982 
983 	/* Fixup first and last segments if needed. */
984 	if (foffs != 0) {
985 		sg->sg_segs[0].ss_paddr += foffs;
986 		sg->sg_segs[0].ss_len -= foffs;
987 		CTR2(KTR_DEV, "sglist_slice seg[0]: %08lx:%08lx",
988 		    (long)sg->sg_segs[0].ss_paddr, sg->sg_segs[0].ss_len);
989 	}
990 	if (loffs != 0) {
991 		sg->sg_segs[count - 1].ss_len -= loffs;
992 		CTR2(KTR_DEV, "sglist_slice seg[%d]: len %08x", count - 1,
993 		    sg->sg_segs[count - 1].ss_len);
994 	}
995 	return (0);
996 }
997