xref: /freebsd/sys/dev/md/md.c (revision b3e7694832e81d7a904a10f525f8797b753bf0d3)
1 /*-
2  * SPDX-License-Identifier: (Beerware AND BSD-3-Clause)
3  *
4  * ----------------------------------------------------------------------------
5  * "THE BEER-WARE LICENSE" (Revision 42):
6  * <phk@FreeBSD.ORG> wrote this file.  As long as you retain this notice you
7  * can do whatever you want with this stuff. If we meet some day, and you think
8  * this stuff is worth it, you can buy me a beer in return.   Poul-Henning Kamp
9  * ----------------------------------------------------------------------------
10  *
11  */
12 
13 /*-
14  * The following functions are based on the vn(4) driver: mdstart_swap(),
15  * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
16  * and as such under the following copyright:
17  *
18  * Copyright (c) 1988 University of Utah.
19  * Copyright (c) 1990, 1993
20  *	The Regents of the University of California.  All rights reserved.
21  * Copyright (c) 2013 The FreeBSD Foundation
22  * All rights reserved.
23  *
24  * This code is derived from software contributed to Berkeley by
25  * the Systems Programming Group of the University of Utah Computer
26  * Science Department.
27  *
28  * Portions of this software were developed by Konstantin Belousov
29  * under sponsorship from the FreeBSD Foundation.
30  *
31  * Redistribution and use in source and binary forms, with or without
32  * modification, are permitted provided that the following conditions
33  * are met:
34  * 1. Redistributions of source code must retain the above copyright
35  *    notice, this list of conditions and the following disclaimer.
36  * 2. Redistributions in binary form must reproduce the above copyright
37  *    notice, this list of conditions and the following disclaimer in the
38  *    documentation and/or other materials provided with the distribution.
39  * 3. Neither the name of the University nor the names of its contributors
40  *    may be used to endorse or promote products derived from this software
41  *    without specific prior written permission.
42  *
43  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
44  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
47  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53  * SUCH DAMAGE.
54  *
55  * from: Utah Hdr: vn.c 1.13 94/04/02
56  *
57  *	from: @(#)vn.c	8.6 (Berkeley) 4/1/94
58  * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
59  */
60 
61 #include "opt_rootdevname.h"
62 #include "opt_geom.h"
63 #include "opt_md.h"
64 
65 #include <sys/param.h>
66 #include <sys/systm.h>
67 #include <sys/bio.h>
68 #include <sys/buf.h>
69 #include <sys/conf.h>
70 #include <sys/devicestat.h>
71 #include <sys/fcntl.h>
72 #include <sys/kernel.h>
73 #include <sys/kthread.h>
74 #include <sys/limits.h>
75 #include <sys/linker.h>
76 #include <sys/lock.h>
77 #include <sys/malloc.h>
78 #include <sys/mdioctl.h>
79 #include <sys/mount.h>
80 #include <sys/mutex.h>
81 #include <sys/sx.h>
82 #include <sys/namei.h>
83 #include <sys/proc.h>
84 #include <sys/queue.h>
85 #include <sys/rwlock.h>
86 #include <sys/sbuf.h>
87 #include <sys/sched.h>
88 #include <sys/sf_buf.h>
89 #include <sys/sysctl.h>
90 #include <sys/uio.h>
91 #include <sys/unistd.h>
92 #include <sys/vnode.h>
93 #include <sys/disk.h>
94 
95 #include <geom/geom.h>
96 #include <geom/geom_int.h>
97 
98 #include <vm/vm.h>
99 #include <vm/vm_extern.h>
100 #include <vm/vm_param.h>
101 #include <vm/vm_object.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_pager.h>
104 #include <vm/swap_pager.h>
105 #include <vm/uma.h>
106 
107 #include <machine/bus.h>
108 
109 #define MD_MODVER 1
110 
111 #define MD_SHUTDOWN	0x10000		/* Tell worker thread to terminate. */
112 #define	MD_EXITING	0x20000		/* Worker thread is exiting. */
113 #define MD_PROVIDERGONE	0x40000		/* Safe to free the softc */
114 
115 #ifndef MD_NSECT
116 #define MD_NSECT (10000 * 2)
117 #endif
118 
119 struct md_req {
120 	unsigned	md_unit;	/* unit number */
121 	enum md_types	md_type;	/* type of disk */
122 	off_t		md_mediasize;	/* size of disk in bytes */
123 	unsigned	md_sectorsize;	/* sectorsize */
124 	unsigned	md_options;	/* options */
125 	int		md_fwheads;	/* firmware heads */
126 	int		md_fwsectors;	/* firmware sectors */
127 	char		*md_file;	/* pathname of file to mount */
128 	enum uio_seg	md_file_seg;	/* location of md_file */
129 	char		*md_label;	/* label of the device (userspace) */
130 	int		*md_units;	/* pointer to units array (kernel) */
131 	size_t		md_units_nitems; /* items in md_units array */
132 };
133 
134 #ifdef COMPAT_FREEBSD32
135 struct md_ioctl32 {
136 	unsigned	md_version;
137 	unsigned	md_unit;
138 	enum md_types	md_type;
139 	uint32_t	md_file;
140 	off_t		md_mediasize;
141 	unsigned	md_sectorsize;
142 	unsigned	md_options;
143 	uint64_t	md_base;
144 	int		md_fwheads;
145 	int		md_fwsectors;
146 	uint32_t	md_label;
147 	int		md_pad[MDNPAD];
148 }
149 #ifdef __amd64__
150 __attribute__((__packed__))
151 #endif
152 ;
153 #ifndef __amd64__
154 CTASSERT((sizeof(struct md_ioctl32)) == 440);
155 #else
156 CTASSERT((sizeof(struct md_ioctl32)) == 436);
157 #endif
158 
159 #define	MDIOCATTACH_32	_IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
160 #define	MDIOCDETACH_32	_IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
161 #define	MDIOCQUERY_32	_IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
162 #define	MDIOCRESIZE_32	_IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
163 #endif /* COMPAT_FREEBSD32 */
164 
165 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
166 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
167 
168 static int md_debug;
169 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
170     "Enable md(4) debug messages");
171 static int md_malloc_wait;
172 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
173     "Allow malloc to wait for memory allocations");
174 
175 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
176 #define	MD_ROOT_FSTYPE	"ufs"
177 #endif
178 
179 #if defined(MD_ROOT)
180 /*
181  * Preloaded image gets put here.
182  */
183 #if defined(MD_ROOT_SIZE)
184 /*
185  * We put the mfs_root symbol into the oldmfs section of the kernel object file.
186  * Applications that patch the object with the image can determine
187  * the size looking at the oldmfs section size within the kernel.
188  */
189 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
190 const int mfs_root_size = sizeof(mfs_root);
191 #elif defined(MD_ROOT_MEM)
192 /* MD region already mapped in the memory */
193 u_char *mfs_root;
194 int mfs_root_size;
195 #else
196 extern volatile u_char __weak_symbol mfs_root;
197 extern volatile u_char __weak_symbol mfs_root_end;
198 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
199 #endif
200 #endif
201 
202 static g_init_t g_md_init;
203 static g_fini_t g_md_fini;
204 static g_start_t g_md_start;
205 static g_access_t g_md_access;
206 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
207     struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
208 static g_provgone_t g_md_providergone;
209 
210 static struct cdev *status_dev = NULL;
211 static struct sx md_sx;
212 static struct unrhdr *md_uh;
213 
214 static d_ioctl_t mdctlioctl;
215 
216 static struct cdevsw mdctl_cdevsw = {
217 	.d_version =	D_VERSION,
218 	.d_ioctl =	mdctlioctl,
219 	.d_name =	MD_NAME,
220 };
221 
222 struct g_class g_md_class = {
223 	.name = "MD",
224 	.version = G_VERSION,
225 	.init = g_md_init,
226 	.fini = g_md_fini,
227 	.start = g_md_start,
228 	.access = g_md_access,
229 	.dumpconf = g_md_dumpconf,
230 	.providergone = g_md_providergone,
231 };
232 
233 DECLARE_GEOM_CLASS(g_md_class, g_md);
234 MODULE_VERSION(geom_md, 0);
235 
236 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
237 
238 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
239 #define NMASK	(NINDIR-1)
240 static int nshift;
241 
242 struct indir {
243 	uintptr_t	*array;
244 	u_int		total;
245 	u_int		used;
246 	u_int		shift;
247 };
248 
249 struct md_s {
250 	int unit;
251 	LIST_ENTRY(md_s) list;
252 	struct bio_queue_head bio_queue;
253 	struct mtx queue_mtx;
254 	struct cdev *dev;
255 	enum md_types type;
256 	off_t mediasize;
257 	unsigned sectorsize;
258 	unsigned opencount;
259 	unsigned fwheads;
260 	unsigned fwsectors;
261 	char ident[32];
262 	unsigned flags;
263 	char name[20];
264 	struct proc *procp;
265 	struct g_geom *gp;
266 	struct g_provider *pp;
267 	int (*start)(struct md_s *sc, struct bio *bp);
268 	struct devstat *devstat;
269 	bool candelete;
270 
271 	/* MD_MALLOC related fields */
272 	struct indir *indir;
273 	uma_zone_t uma;
274 
275 	/* MD_PRELOAD related fields */
276 	u_char *pl_ptr;
277 	size_t pl_len;
278 
279 	/* MD_VNODE related fields */
280 	struct vnode *vnode;
281 	char file[PATH_MAX];
282 	char label[PATH_MAX];
283 	struct ucred *cred;
284 	vm_offset_t kva;
285 
286 	/* MD_SWAP related fields */
287 	vm_object_t object;
288 };
289 
290 static struct indir *
291 new_indir(u_int shift)
292 {
293 	struct indir *ip;
294 
295 	ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
296 	    | M_ZERO);
297 	if (ip == NULL)
298 		return (NULL);
299 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
300 	    M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
301 	if (ip->array == NULL) {
302 		free(ip, M_MD);
303 		return (NULL);
304 	}
305 	ip->total = NINDIR;
306 	ip->shift = shift;
307 	return (ip);
308 }
309 
310 static void
311 del_indir(struct indir *ip)
312 {
313 
314 	free(ip->array, M_MDSECT);
315 	free(ip, M_MD);
316 }
317 
318 static void
319 destroy_indir(struct md_s *sc, struct indir *ip)
320 {
321 	int i;
322 
323 	for (i = 0; i < NINDIR; i++) {
324 		if (!ip->array[i])
325 			continue;
326 		if (ip->shift)
327 			destroy_indir(sc, (struct indir*)(ip->array[i]));
328 		else if (ip->array[i] > 255)
329 			uma_zfree(sc->uma, (void *)(ip->array[i]));
330 	}
331 	del_indir(ip);
332 }
333 
334 /*
335  * This function does the math and allocates the top level "indir" structure
336  * for a device of "size" sectors.
337  */
338 
339 static struct indir *
340 dimension(off_t size)
341 {
342 	off_t rcnt;
343 	struct indir *ip;
344 	int layer;
345 
346 	rcnt = size;
347 	layer = 0;
348 	while (rcnt > NINDIR) {
349 		rcnt /= NINDIR;
350 		layer++;
351 	}
352 
353 	/*
354 	 * XXX: the top layer is probably not fully populated, so we allocate
355 	 * too much space for ip->array in here.
356 	 */
357 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
358 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
359 	    M_MDSECT, M_WAITOK | M_ZERO);
360 	ip->total = NINDIR;
361 	ip->shift = layer * nshift;
362 	return (ip);
363 }
364 
365 /*
366  * Read a given sector
367  */
368 
369 static uintptr_t
370 s_read(struct indir *ip, off_t offset)
371 {
372 	struct indir *cip;
373 	int idx;
374 	uintptr_t up;
375 
376 	if (md_debug > 1)
377 		printf("s_read(%jd)\n", (intmax_t)offset);
378 	up = 0;
379 	for (cip = ip; cip != NULL;) {
380 		if (cip->shift) {
381 			idx = (offset >> cip->shift) & NMASK;
382 			up = cip->array[idx];
383 			cip = (struct indir *)up;
384 			continue;
385 		}
386 		idx = offset & NMASK;
387 		return (cip->array[idx]);
388 	}
389 	return (0);
390 }
391 
392 /*
393  * Write a given sector, prune the tree if the value is 0
394  */
395 
396 static int
397 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
398 {
399 	struct indir *cip, *lip[10];
400 	int idx, li;
401 	uintptr_t up;
402 
403 	if (md_debug > 1)
404 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
405 	up = 0;
406 	li = 0;
407 	cip = ip;
408 	for (;;) {
409 		lip[li++] = cip;
410 		if (cip->shift) {
411 			idx = (offset >> cip->shift) & NMASK;
412 			up = cip->array[idx];
413 			if (up != 0) {
414 				cip = (struct indir *)up;
415 				continue;
416 			}
417 			/* Allocate branch */
418 			cip->array[idx] =
419 			    (uintptr_t)new_indir(cip->shift - nshift);
420 			if (cip->array[idx] == 0)
421 				return (ENOSPC);
422 			cip->used++;
423 			up = cip->array[idx];
424 			cip = (struct indir *)up;
425 			continue;
426 		}
427 		/* leafnode */
428 		idx = offset & NMASK;
429 		up = cip->array[idx];
430 		if (up != 0)
431 			cip->used--;
432 		cip->array[idx] = ptr;
433 		if (ptr != 0)
434 			cip->used++;
435 		break;
436 	}
437 	if (cip->used != 0 || li == 1)
438 		return (0);
439 	li--;
440 	while (cip->used == 0 && cip != ip) {
441 		li--;
442 		idx = (offset >> lip[li]->shift) & NMASK;
443 		up = lip[li]->array[idx];
444 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
445 		del_indir(cip);
446 		lip[li]->array[idx] = 0;
447 		lip[li]->used--;
448 		cip = lip[li];
449 	}
450 	return (0);
451 }
452 
453 static int
454 g_md_access(struct g_provider *pp, int r, int w, int e)
455 {
456 	struct md_s *sc;
457 
458 	sc = pp->geom->softc;
459 	if (sc == NULL) {
460 		if (r <= 0 && w <= 0 && e <= 0)
461 			return (0);
462 		return (ENXIO);
463 	}
464 	r += pp->acr;
465 	w += pp->acw;
466 	e += pp->ace;
467 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
468 		return (EROFS);
469 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
470 		sc->opencount = 1;
471 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
472 		sc->opencount = 0;
473 	}
474 	return (0);
475 }
476 
477 static void
478 g_md_start(struct bio *bp)
479 {
480 	struct md_s *sc;
481 
482 	sc = bp->bio_to->geom->softc;
483 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
484 		devstat_start_transaction_bio(sc->devstat, bp);
485 	}
486 	mtx_lock(&sc->queue_mtx);
487 	bioq_disksort(&sc->bio_queue, bp);
488 	wakeup(sc);
489 	mtx_unlock(&sc->queue_mtx);
490 }
491 
492 #define	MD_MALLOC_MOVE_ZERO	1
493 #define	MD_MALLOC_MOVE_FILL	2
494 #define	MD_MALLOC_MOVE_READ	3
495 #define	MD_MALLOC_MOVE_WRITE	4
496 #define	MD_MALLOC_MOVE_CMP	5
497 
498 static int
499 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
500     void *ptr, u_char fill, int op)
501 {
502 	struct sf_buf *sf;
503 	vm_page_t m, *mp1;
504 	char *p, first;
505 	off_t *uc;
506 	unsigned n;
507 	int error, i, ma_offs1, sz, first_read;
508 
509 	m = NULL;
510 	error = 0;
511 	sf = NULL;
512 	/* if (op == MD_MALLOC_MOVE_CMP) { gcc */
513 		first = 0;
514 		first_read = 0;
515 		uc = ptr;
516 		mp1 = *mp;
517 		ma_offs1 = *ma_offs;
518 	/* } */
519 	sched_pin();
520 	for (n = sectorsize; n != 0; n -= sz) {
521 		sz = imin(PAGE_SIZE - *ma_offs, n);
522 		if (m != **mp) {
523 			if (sf != NULL)
524 				sf_buf_free(sf);
525 			m = **mp;
526 			sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
527 			    (md_malloc_wait ? 0 : SFB_NOWAIT));
528 			if (sf == NULL) {
529 				error = ENOMEM;
530 				break;
531 			}
532 		}
533 		p = (char *)sf_buf_kva(sf) + *ma_offs;
534 		switch (op) {
535 		case MD_MALLOC_MOVE_ZERO:
536 			bzero(p, sz);
537 			break;
538 		case MD_MALLOC_MOVE_FILL:
539 			memset(p, fill, sz);
540 			break;
541 		case MD_MALLOC_MOVE_READ:
542 			bcopy(ptr, p, sz);
543 			cpu_flush_dcache(p, sz);
544 			break;
545 		case MD_MALLOC_MOVE_WRITE:
546 			bcopy(p, ptr, sz);
547 			break;
548 		case MD_MALLOC_MOVE_CMP:
549 			for (i = 0; i < sz; i++, p++) {
550 				if (!first_read) {
551 					*uc = (u_char)*p;
552 					first = *p;
553 					first_read = 1;
554 				} else if (*p != first) {
555 					error = EDOOFUS;
556 					break;
557 				}
558 			}
559 			break;
560 		default:
561 			KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
562 			break;
563 		}
564 		if (error != 0)
565 			break;
566 		*ma_offs += sz;
567 		*ma_offs %= PAGE_SIZE;
568 		if (*ma_offs == 0)
569 			(*mp)++;
570 		ptr = (char *)ptr + sz;
571 	}
572 
573 	if (sf != NULL)
574 		sf_buf_free(sf);
575 	sched_unpin();
576 	if (op == MD_MALLOC_MOVE_CMP && error != 0) {
577 		*mp = mp1;
578 		*ma_offs = ma_offs1;
579 	}
580 	return (error);
581 }
582 
583 static int
584 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
585     unsigned len, void *ptr, u_char fill, int op)
586 {
587 	bus_dma_segment_t *vlist;
588 	uint8_t *p, *end, first;
589 	off_t *uc;
590 	int ma_offs, seg_len;
591 
592 	vlist = *pvlist;
593 	ma_offs = *pma_offs;
594 	uc = ptr;
595 
596 	for (; len != 0; len -= seg_len) {
597 		seg_len = imin(vlist->ds_len - ma_offs, len);
598 		p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
599 		switch (op) {
600 		case MD_MALLOC_MOVE_ZERO:
601 			bzero(p, seg_len);
602 			break;
603 		case MD_MALLOC_MOVE_FILL:
604 			memset(p, fill, seg_len);
605 			break;
606 		case MD_MALLOC_MOVE_READ:
607 			bcopy(ptr, p, seg_len);
608 			cpu_flush_dcache(p, seg_len);
609 			break;
610 		case MD_MALLOC_MOVE_WRITE:
611 			bcopy(p, ptr, seg_len);
612 			break;
613 		case MD_MALLOC_MOVE_CMP:
614 			end = p + seg_len;
615 			first = *uc = *p;
616 			/* Confirm all following bytes match the first */
617 			while (++p < end) {
618 				if (*p != first)
619 					return (EDOOFUS);
620 			}
621 			break;
622 		default:
623 			KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
624 			break;
625 		}
626 
627 		ma_offs += seg_len;
628 		if (ma_offs == vlist->ds_len) {
629 			ma_offs = 0;
630 			vlist++;
631 		}
632 		ptr = (uint8_t *)ptr + seg_len;
633 	}
634 	*pvlist = vlist;
635 	*pma_offs = ma_offs;
636 
637 	return (0);
638 }
639 
640 static int
641 mdstart_malloc(struct md_s *sc, struct bio *bp)
642 {
643 	u_char *dst;
644 	vm_page_t *m;
645 	bus_dma_segment_t *vlist;
646 	int i, error, error1, ma_offs, notmapped;
647 	off_t secno, nsec, uc;
648 	uintptr_t sp, osp;
649 
650 	switch (bp->bio_cmd) {
651 	case BIO_READ:
652 	case BIO_WRITE:
653 	case BIO_DELETE:
654 		break;
655 	case BIO_FLUSH:
656 		return (0);
657 	default:
658 		return (EOPNOTSUPP);
659 	}
660 
661 	notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
662 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
663 	    (bus_dma_segment_t *)bp->bio_data : NULL;
664 	if (notmapped) {
665 		m = bp->bio_ma;
666 		ma_offs = bp->bio_ma_offset;
667 		dst = NULL;
668 		KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
669 	} else if (vlist != NULL) {
670 		ma_offs = bp->bio_ma_offset;
671 		dst = NULL;
672 	} else {
673 		dst = bp->bio_data;
674 	}
675 
676 	nsec = bp->bio_length / sc->sectorsize;
677 	secno = bp->bio_offset / sc->sectorsize;
678 	error = 0;
679 	while (nsec--) {
680 		osp = s_read(sc->indir, secno);
681 		if (bp->bio_cmd == BIO_DELETE) {
682 			if (osp != 0)
683 				error = s_write(sc->indir, secno, 0);
684 		} else if (bp->bio_cmd == BIO_READ) {
685 			if (osp == 0) {
686 				if (notmapped) {
687 					error = md_malloc_move_ma(&m, &ma_offs,
688 					    sc->sectorsize, NULL, 0,
689 					    MD_MALLOC_MOVE_ZERO);
690 				} else if (vlist != NULL) {
691 					error = md_malloc_move_vlist(&vlist,
692 					    &ma_offs, sc->sectorsize, NULL, 0,
693 					    MD_MALLOC_MOVE_ZERO);
694 				} else
695 					bzero(dst, sc->sectorsize);
696 			} else if (osp <= 255) {
697 				if (notmapped) {
698 					error = md_malloc_move_ma(&m, &ma_offs,
699 					    sc->sectorsize, NULL, osp,
700 					    MD_MALLOC_MOVE_FILL);
701 				} else if (vlist != NULL) {
702 					error = md_malloc_move_vlist(&vlist,
703 					    &ma_offs, sc->sectorsize, NULL, osp,
704 					    MD_MALLOC_MOVE_FILL);
705 				} else
706 					memset(dst, osp, sc->sectorsize);
707 			} else {
708 				if (notmapped) {
709 					error = md_malloc_move_ma(&m, &ma_offs,
710 					    sc->sectorsize, (void *)osp, 0,
711 					    MD_MALLOC_MOVE_READ);
712 				} else if (vlist != NULL) {
713 					error = md_malloc_move_vlist(&vlist,
714 					    &ma_offs, sc->sectorsize,
715 					    (void *)osp, 0,
716 					    MD_MALLOC_MOVE_READ);
717 				} else {
718 					bcopy((void *)osp, dst, sc->sectorsize);
719 					cpu_flush_dcache(dst, sc->sectorsize);
720 				}
721 			}
722 			osp = 0;
723 		} else if (bp->bio_cmd == BIO_WRITE) {
724 			if (sc->flags & MD_COMPRESS) {
725 				if (notmapped) {
726 					error1 = md_malloc_move_ma(&m, &ma_offs,
727 					    sc->sectorsize, &uc, 0,
728 					    MD_MALLOC_MOVE_CMP);
729 					i = error1 == 0 ? sc->sectorsize : 0;
730 				} else if (vlist != NULL) {
731 					error1 = md_malloc_move_vlist(&vlist,
732 					    &ma_offs, sc->sectorsize, &uc, 0,
733 					    MD_MALLOC_MOVE_CMP);
734 					i = error1 == 0 ? sc->sectorsize : 0;
735 				} else {
736 					uc = dst[0];
737 					for (i = 1; i < sc->sectorsize; i++) {
738 						if (dst[i] != uc)
739 							break;
740 					}
741 				}
742 			} else {
743 				i = 0;
744 				uc = 0;
745 			}
746 			if (i == sc->sectorsize) {
747 				if (osp != uc)
748 					error = s_write(sc->indir, secno, uc);
749 			} else {
750 				if (osp <= 255) {
751 					sp = (uintptr_t)uma_zalloc(sc->uma,
752 					    md_malloc_wait ? M_WAITOK :
753 					    M_NOWAIT);
754 					if (sp == 0) {
755 						error = ENOSPC;
756 						break;
757 					}
758 					if (notmapped) {
759 						error = md_malloc_move_ma(&m,
760 						    &ma_offs, sc->sectorsize,
761 						    (void *)sp, 0,
762 						    MD_MALLOC_MOVE_WRITE);
763 					} else if (vlist != NULL) {
764 						error = md_malloc_move_vlist(
765 						    &vlist, &ma_offs,
766 						    sc->sectorsize, (void *)sp,
767 						    0, MD_MALLOC_MOVE_WRITE);
768 					} else {
769 						bcopy(dst, (void *)sp,
770 						    sc->sectorsize);
771 					}
772 					error = s_write(sc->indir, secno, sp);
773 				} else {
774 					if (notmapped) {
775 						error = md_malloc_move_ma(&m,
776 						    &ma_offs, sc->sectorsize,
777 						    (void *)osp, 0,
778 						    MD_MALLOC_MOVE_WRITE);
779 					} else if (vlist != NULL) {
780 						error = md_malloc_move_vlist(
781 						    &vlist, &ma_offs,
782 						    sc->sectorsize, (void *)osp,
783 						    0, MD_MALLOC_MOVE_WRITE);
784 					} else {
785 						bcopy(dst, (void *)osp,
786 						    sc->sectorsize);
787 					}
788 					osp = 0;
789 				}
790 			}
791 		} else {
792 			error = EOPNOTSUPP;
793 		}
794 		if (osp > 255)
795 			uma_zfree(sc->uma, (void*)osp);
796 		if (error != 0)
797 			break;
798 		secno++;
799 		if (!notmapped && vlist == NULL)
800 			dst += sc->sectorsize;
801 	}
802 	bp->bio_resid = 0;
803 	return (error);
804 }
805 
806 static void
807 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
808 {
809 	off_t seg_len;
810 
811 	while (offset >= vlist->ds_len) {
812 		offset -= vlist->ds_len;
813 		vlist++;
814 	}
815 
816 	while (len != 0) {
817 		seg_len = omin(len, vlist->ds_len - offset);
818 		bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
819 		    seg_len);
820 		offset = 0;
821 		src = (uint8_t *)src + seg_len;
822 		len -= seg_len;
823 		vlist++;
824 	}
825 }
826 
827 static void
828 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
829 {
830 	off_t seg_len;
831 
832 	while (offset >= vlist->ds_len) {
833 		offset -= vlist->ds_len;
834 		vlist++;
835 	}
836 
837 	while (len != 0) {
838 		seg_len = omin(len, vlist->ds_len - offset);
839 		bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
840 		    seg_len);
841 		offset = 0;
842 		dst = (uint8_t *)dst + seg_len;
843 		len -= seg_len;
844 		vlist++;
845 	}
846 }
847 
848 static int
849 mdstart_preload(struct md_s *sc, struct bio *bp)
850 {
851 	uint8_t *p;
852 
853 	p = sc->pl_ptr + bp->bio_offset;
854 	switch (bp->bio_cmd) {
855 	case BIO_READ:
856 		if ((bp->bio_flags & BIO_VLIST) != 0) {
857 			mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
858 			    bp->bio_ma_offset, bp->bio_length);
859 		} else {
860 			bcopy(p, bp->bio_data, bp->bio_length);
861 		}
862 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
863 		break;
864 	case BIO_WRITE:
865 		if ((bp->bio_flags & BIO_VLIST) != 0) {
866 			mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
867 			    bp->bio_ma_offset, p, bp->bio_length);
868 		} else {
869 			bcopy(bp->bio_data, p, bp->bio_length);
870 		}
871 		break;
872 	}
873 	bp->bio_resid = 0;
874 	return (0);
875 }
876 
877 static int
878 mdstart_vnode(struct md_s *sc, struct bio *bp)
879 {
880 	int error;
881 	struct uio auio;
882 	struct iovec aiov;
883 	struct iovec *piov;
884 	struct mount *mp;
885 	struct vnode *vp;
886 	bus_dma_segment_t *vlist;
887 	struct thread *td;
888 	off_t iolen, iostart, off, len;
889 	int ma_offs, npages;
890 	bool mapped;
891 
892 	switch (bp->bio_cmd) {
893 	case BIO_READ:
894 		auio.uio_rw = UIO_READ;
895 		break;
896 	case BIO_WRITE:
897 		auio.uio_rw = UIO_WRITE;
898 		break;
899 	case BIO_FLUSH:
900 		break;
901 	case BIO_DELETE:
902 		if (sc->candelete)
903 			break;
904 		/* FALLTHROUGH */
905 	default:
906 		return (EOPNOTSUPP);
907 	}
908 
909 	td = curthread;
910 	vp = sc->vnode;
911 	piov = NULL;
912 	ma_offs = bp->bio_ma_offset;
913 	off = bp->bio_offset;
914 	len = bp->bio_length;
915 	mapped = false;
916 
917 	/*
918 	 * VNODE I/O
919 	 *
920 	 * If an error occurs, we set BIO_ERROR but we do not set
921 	 * B_INVAL because (for a write anyway), the buffer is
922 	 * still valid.
923 	 */
924 
925 	if (bp->bio_cmd == BIO_FLUSH) {
926 		do {
927 			(void)vn_start_write(vp, &mp, V_WAIT);
928 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
929 			error = VOP_FSYNC(vp, MNT_WAIT, td);
930 			VOP_UNLOCK(vp);
931 			vn_finished_write(mp);
932 		} while (error == ERELOOKUP);
933 		return (error);
934 	} else if (bp->bio_cmd == BIO_DELETE) {
935 		error = vn_deallocate(vp, &off, &len, 0,
936 		    sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred, NOCRED);
937 		bp->bio_resid = len;
938 		return (error);
939 	}
940 
941 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
942 	auio.uio_resid = bp->bio_length;
943 	auio.uio_segflg = UIO_SYSSPACE;
944 	auio.uio_td = td;
945 
946 	if ((bp->bio_flags & BIO_VLIST) != 0) {
947 		piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
948 		auio.uio_iov = piov;
949 		vlist = (bus_dma_segment_t *)bp->bio_data;
950 		while (len > 0) {
951 			piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
952 			    ma_offs);
953 			piov->iov_len = vlist->ds_len - ma_offs;
954 			if (piov->iov_len > len)
955 				piov->iov_len = len;
956 			len -= piov->iov_len;
957 			ma_offs = 0;
958 			vlist++;
959 			piov++;
960 		}
961 		auio.uio_iovcnt = piov - auio.uio_iov;
962 		piov = auio.uio_iov;
963 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
964 		bp->bio_resid = len;
965 unmapped_step:
966 		npages = atop(min(maxphys, round_page(len + (ma_offs &
967 		    PAGE_MASK))));
968 		iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
969 		KASSERT(iolen > 0, ("zero iolen"));
970 		KASSERT(npages <= atop(MAXPHYS + PAGE_SIZE),
971 		    ("npages %d too large", npages));
972 		pmap_qenter(sc->kva, &bp->bio_ma[atop(ma_offs)], npages);
973 		aiov.iov_base = (void *)(sc->kva + (ma_offs & PAGE_MASK));
974 		aiov.iov_len = iolen;
975 		auio.uio_iov = &aiov;
976 		auio.uio_iovcnt = 1;
977 		auio.uio_resid = iolen;
978 		mapped = true;
979 	} else {
980 		aiov.iov_base = bp->bio_data;
981 		aiov.iov_len = bp->bio_length;
982 		auio.uio_iov = &aiov;
983 		auio.uio_iovcnt = 1;
984 	}
985 	iostart = auio.uio_offset;
986 	if (auio.uio_rw == UIO_READ) {
987 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
988 		error = VOP_READ(vp, &auio, 0, sc->cred);
989 		VOP_UNLOCK(vp);
990 	} else {
991 		(void) vn_start_write(vp, &mp, V_WAIT);
992 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
993 		error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
994 		    sc->cred);
995 		VOP_UNLOCK(vp);
996 		vn_finished_write(mp);
997 		if (error == 0)
998 			sc->flags &= ~MD_VERIFY;
999 	}
1000 
1001 	/* When MD_CACHE is set, try to avoid double-caching the data. */
1002 	if (error == 0 && (sc->flags & MD_CACHE) == 0)
1003 		VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1004 		    POSIX_FADV_DONTNEED);
1005 
1006 	if (mapped) {
1007 		pmap_qremove(sc->kva, npages);
1008 		if (error == 0) {
1009 			len -= iolen;
1010 			bp->bio_resid -= iolen;
1011 			ma_offs += iolen;
1012 			if (len > 0)
1013 				goto unmapped_step;
1014 		}
1015 	} else {
1016 		bp->bio_resid = auio.uio_resid;
1017 	}
1018 
1019 	free(piov, M_MD);
1020 	return (error);
1021 }
1022 
1023 static int
1024 mdstart_swap(struct md_s *sc, struct bio *bp)
1025 {
1026 	vm_page_t m;
1027 	u_char *p;
1028 	vm_pindex_t i, lastp;
1029 	bus_dma_segment_t *vlist;
1030 	int rv, ma_offs, offs, len, lastend;
1031 
1032 	switch (bp->bio_cmd) {
1033 	case BIO_READ:
1034 	case BIO_WRITE:
1035 	case BIO_DELETE:
1036 		break;
1037 	case BIO_FLUSH:
1038 		return (0);
1039 	default:
1040 		return (EOPNOTSUPP);
1041 	}
1042 
1043 	p = bp->bio_data;
1044 	ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1045 	    bp->bio_ma_offset : 0;
1046 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1047 	    (bus_dma_segment_t *)bp->bio_data : NULL;
1048 
1049 	/*
1050 	 * offs is the offset at which to start operating on the
1051 	 * next (ie, first) page.  lastp is the last page on
1052 	 * which we're going to operate.  lastend is the ending
1053 	 * position within that last page (ie, PAGE_SIZE if
1054 	 * we're operating on complete aligned pages).
1055 	 */
1056 	offs = bp->bio_offset % PAGE_SIZE;
1057 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1058 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1059 
1060 	rv = VM_PAGER_OK;
1061 	vm_object_pip_add(sc->object, 1);
1062 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1063 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1064 		m = vm_page_grab_unlocked(sc->object, i, VM_ALLOC_SYSTEM);
1065 		if (bp->bio_cmd == BIO_READ) {
1066 			if (vm_page_all_valid(m))
1067 				rv = VM_PAGER_OK;
1068 			else
1069 				rv = vm_pager_get_pages(sc->object, &m, 1,
1070 				    NULL, NULL);
1071 			if (rv == VM_PAGER_ERROR) {
1072 				VM_OBJECT_WLOCK(sc->object);
1073 				vm_page_free(m);
1074 				VM_OBJECT_WUNLOCK(sc->object);
1075 				break;
1076 			} else if (rv == VM_PAGER_FAIL) {
1077 				/*
1078 				 * Pager does not have the page.  Zero
1079 				 * the allocated page, and mark it as
1080 				 * valid. Do not set dirty, the page
1081 				 * can be recreated if thrown out.
1082 				 */
1083 				pmap_zero_page(m);
1084 				vm_page_valid(m);
1085 			}
1086 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1087 				pmap_copy_pages(&m, offs, bp->bio_ma,
1088 				    ma_offs, len);
1089 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1090 				physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1091 				    vlist, ma_offs, len);
1092 				cpu_flush_dcache(p, len);
1093 			} else {
1094 				physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1095 				cpu_flush_dcache(p, len);
1096 			}
1097 		} else if (bp->bio_cmd == BIO_WRITE) {
1098 			if (len == PAGE_SIZE || vm_page_all_valid(m))
1099 				rv = VM_PAGER_OK;
1100 			else
1101 				rv = vm_pager_get_pages(sc->object, &m, 1,
1102 				    NULL, NULL);
1103 			if (rv == VM_PAGER_ERROR) {
1104 				VM_OBJECT_WLOCK(sc->object);
1105 				vm_page_free(m);
1106 				VM_OBJECT_WUNLOCK(sc->object);
1107 				break;
1108 			} else if (rv == VM_PAGER_FAIL)
1109 				pmap_zero_page(m);
1110 
1111 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1112 				pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1113 				    offs, len);
1114 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1115 				physcopyin_vlist(vlist, ma_offs,
1116 				    VM_PAGE_TO_PHYS(m) + offs, len);
1117 			} else {
1118 				physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1119 			}
1120 
1121 			vm_page_valid(m);
1122 			vm_page_set_dirty(m);
1123 		} else if (bp->bio_cmd == BIO_DELETE) {
1124 			if (len == PAGE_SIZE || vm_page_all_valid(m))
1125 				rv = VM_PAGER_OK;
1126 			else
1127 				rv = vm_pager_get_pages(sc->object, &m, 1,
1128 				    NULL, NULL);
1129 			VM_OBJECT_WLOCK(sc->object);
1130 			if (rv == VM_PAGER_ERROR) {
1131 				vm_page_free(m);
1132 				VM_OBJECT_WUNLOCK(sc->object);
1133 				break;
1134 			} else if (rv == VM_PAGER_FAIL) {
1135 				vm_page_free(m);
1136 				m = NULL;
1137 			} else {
1138 				/* Page is valid. */
1139 				if (len != PAGE_SIZE) {
1140 					pmap_zero_page_area(m, offs, len);
1141 					vm_page_set_dirty(m);
1142 				} else {
1143 					vm_pager_page_unswapped(m);
1144 					vm_page_free(m);
1145 					m = NULL;
1146 				}
1147 			}
1148 			VM_OBJECT_WUNLOCK(sc->object);
1149 		}
1150 		if (m != NULL) {
1151 			/*
1152 			 * The page may be deactivated prior to setting
1153 			 * PGA_REFERENCED, but in this case it will be
1154 			 * reactivated by the page daemon.
1155 			 */
1156 			if (vm_page_active(m))
1157 				vm_page_reference(m);
1158 			else
1159 				vm_page_activate(m);
1160 			vm_page_xunbusy(m);
1161 		}
1162 
1163 		/* Actions on further pages start at offset 0 */
1164 		p += PAGE_SIZE - offs;
1165 		offs = 0;
1166 		ma_offs += len;
1167 	}
1168 	vm_object_pip_wakeup(sc->object);
1169 	return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1170 }
1171 
1172 static int
1173 mdstart_null(struct md_s *sc, struct bio *bp)
1174 {
1175 
1176 	switch (bp->bio_cmd) {
1177 	case BIO_READ:
1178 		bzero(bp->bio_data, bp->bio_length);
1179 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
1180 		break;
1181 	case BIO_WRITE:
1182 		break;
1183 	}
1184 	bp->bio_resid = 0;
1185 	return (0);
1186 }
1187 
1188 static void
1189 md_handleattr(struct md_s *sc, struct bio *bp)
1190 {
1191 	if (sc->fwsectors && sc->fwheads &&
1192 	    (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 ||
1193 	    g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0))
1194 		return;
1195 	if (g_handleattr_int(bp, "GEOM::candelete", sc->candelete) != 0)
1196 		return;
1197 	if (sc->ident[0] != '\0' &&
1198 	    g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1199 		return;
1200 	if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1201 		return;
1202 	g_io_deliver(bp, EOPNOTSUPP);
1203 }
1204 
1205 static void
1206 md_kthread(void *arg)
1207 {
1208 	struct md_s *sc;
1209 	struct bio *bp;
1210 	int error;
1211 
1212 	sc = arg;
1213 	thread_lock(curthread);
1214 	sched_prio(curthread, PRIBIO);
1215 	thread_unlock(curthread);
1216 	if (sc->type == MD_VNODE)
1217 		curthread->td_pflags |= TDP_NORUNNINGBUF;
1218 
1219 	for (;;) {
1220 		mtx_lock(&sc->queue_mtx);
1221 		if (sc->flags & MD_SHUTDOWN) {
1222 			sc->flags |= MD_EXITING;
1223 			mtx_unlock(&sc->queue_mtx);
1224 			kproc_exit(0);
1225 		}
1226 		bp = bioq_takefirst(&sc->bio_queue);
1227 		if (!bp) {
1228 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1229 			continue;
1230 		}
1231 		mtx_unlock(&sc->queue_mtx);
1232 		if (bp->bio_cmd == BIO_GETATTR) {
1233 			md_handleattr(sc, bp);
1234 		} else {
1235 			error = sc->start(sc, bp);
1236 			if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1237 				/*
1238 				 * Devstat uses (bio_bcount, bio_resid) for
1239 				 * determining the length of the completed part
1240 				 * of the i/o.  g_io_deliver() will translate
1241 				 * from bio_completed to that, but it also
1242 				 * destroys the bio so we must do our own
1243 				 * translation.
1244 				 */
1245 				bp->bio_bcount = bp->bio_length;
1246 				devstat_end_transaction_bio(sc->devstat, bp);
1247 			}
1248 			bp->bio_completed = bp->bio_length - bp->bio_resid;
1249 			g_io_deliver(bp, error);
1250 		}
1251 	}
1252 }
1253 
1254 static struct md_s *
1255 mdfind(int unit)
1256 {
1257 	struct md_s *sc;
1258 
1259 	LIST_FOREACH(sc, &md_softc_list, list) {
1260 		if (sc->unit == unit)
1261 			break;
1262 	}
1263 	return (sc);
1264 }
1265 
1266 static struct md_s *
1267 mdnew(int unit, int *errp, enum md_types type)
1268 {
1269 	struct md_s *sc;
1270 	int error;
1271 
1272 	*errp = 0;
1273 	if (unit == -1)
1274 		unit = alloc_unr(md_uh);
1275 	else
1276 		unit = alloc_unr_specific(md_uh, unit);
1277 
1278 	if (unit == -1) {
1279 		*errp = EBUSY;
1280 		return (NULL);
1281 	}
1282 
1283 	sc = malloc(sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
1284 	sc->type = type;
1285 	bioq_init(&sc->bio_queue);
1286 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1287 	sc->unit = unit;
1288 	sprintf(sc->name, "md%d", unit);
1289 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
1290 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1291 	if (error == 0)
1292 		return (sc);
1293 	LIST_REMOVE(sc, list);
1294 	mtx_destroy(&sc->queue_mtx);
1295 	free_unr(md_uh, sc->unit);
1296 	free(sc, M_MD);
1297 	*errp = error;
1298 	return (NULL);
1299 }
1300 
1301 static void
1302 mdinit(struct md_s *sc)
1303 {
1304 	struct g_geom *gp;
1305 	struct g_provider *pp;
1306 
1307 	g_topology_lock();
1308 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1309 	gp->softc = sc;
1310 	pp = g_new_providerf(gp, "md%d", sc->unit);
1311 	devstat_remove_entry(pp->stat);
1312 	pp->stat = NULL;
1313 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1314 	pp->mediasize = sc->mediasize;
1315 	pp->sectorsize = sc->sectorsize;
1316 	switch (sc->type) {
1317 	case MD_MALLOC:
1318 	case MD_VNODE:
1319 	case MD_SWAP:
1320 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1321 		break;
1322 	case MD_PRELOAD:
1323 	case MD_NULL:
1324 		break;
1325 	}
1326 	sc->gp = gp;
1327 	sc->pp = pp;
1328 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1329 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1330 	sc->devstat->id = pp;
1331 	g_error_provider(pp, 0);
1332 	g_topology_unlock();
1333 }
1334 
1335 static int
1336 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1337 {
1338 	uintptr_t sp;
1339 	int error;
1340 	off_t u;
1341 
1342 	error = 0;
1343 	if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1344 		return (EINVAL);
1345 	if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1346 		return (EINVAL);
1347 	/* Compression doesn't make sense if we have reserved space */
1348 	if (mdr->md_options & MD_RESERVE)
1349 		mdr->md_options &= ~MD_COMPRESS;
1350 	if (mdr->md_fwsectors != 0)
1351 		sc->fwsectors = mdr->md_fwsectors;
1352 	if (mdr->md_fwheads != 0)
1353 		sc->fwheads = mdr->md_fwheads;
1354 	sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1355 	sc->indir = dimension(sc->mediasize / sc->sectorsize);
1356 	sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1357 	    0x1ff, 0);
1358 	if (mdr->md_options & MD_RESERVE) {
1359 		off_t nsectors;
1360 
1361 		nsectors = sc->mediasize / sc->sectorsize;
1362 		for (u = 0; u < nsectors; u++) {
1363 			sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1364 			    M_WAITOK : M_NOWAIT) | M_ZERO);
1365 			if (sp != 0)
1366 				error = s_write(sc->indir, u, sp);
1367 			else
1368 				error = ENOMEM;
1369 			if (error != 0)
1370 				break;
1371 		}
1372 	}
1373 	return (error);
1374 }
1375 
1376 static int
1377 mdsetcred(struct md_s *sc, struct ucred *cred)
1378 {
1379 	char *tmpbuf;
1380 	int error = 0;
1381 
1382 	/*
1383 	 * Set credits in our softc
1384 	 */
1385 
1386 	if (sc->cred)
1387 		crfree(sc->cred);
1388 	sc->cred = crhold(cred);
1389 
1390 	/*
1391 	 * Horrible kludge to establish credentials for NFS  XXX.
1392 	 */
1393 
1394 	if (sc->vnode) {
1395 		struct uio auio;
1396 		struct iovec aiov;
1397 
1398 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1399 		bzero(&auio, sizeof(auio));
1400 
1401 		aiov.iov_base = tmpbuf;
1402 		aiov.iov_len = sc->sectorsize;
1403 		auio.uio_iov = &aiov;
1404 		auio.uio_iovcnt = 1;
1405 		auio.uio_offset = 0;
1406 		auio.uio_rw = UIO_READ;
1407 		auio.uio_segflg = UIO_SYSSPACE;
1408 		auio.uio_resid = aiov.iov_len;
1409 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1410 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1411 		VOP_UNLOCK(sc->vnode);
1412 		free(tmpbuf, M_TEMP);
1413 	}
1414 	return (error);
1415 }
1416 
1417 static int
1418 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1419 {
1420 	struct vattr vattr;
1421 	struct nameidata nd;
1422 	char *fname;
1423 	int error, flags;
1424 	long v;
1425 
1426 	fname = mdr->md_file;
1427 	if (mdr->md_file_seg == UIO_USERSPACE) {
1428 		error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1429 		if (error != 0)
1430 			return (error);
1431 	} else if (mdr->md_file_seg == UIO_SYSSPACE)
1432 		strlcpy(sc->file, fname, sizeof(sc->file));
1433 	else
1434 		return (EDOOFUS);
1435 
1436 	/*
1437 	 * If the user specified that this is a read only device, don't
1438 	 * set the FWRITE mask before trying to open the backing store.
1439 	 */
1440 	flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1441 	    | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1442 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file);
1443 	error = vn_open(&nd, &flags, 0, NULL);
1444 	if (error != 0)
1445 		return (error);
1446 	NDFREE_PNBUF(&nd);
1447 	if (nd.ni_vp->v_type != VREG) {
1448 		error = EINVAL;
1449 		goto bad;
1450 	}
1451 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1452 	if (error != 0)
1453 		goto bad;
1454 	if ((mdr->md_options & MD_MUSTDEALLOC) != 0) {
1455 		error = VOP_PATHCONF(nd.ni_vp, _PC_DEALLOC_PRESENT, &v);
1456 		if (error != 0)
1457 			goto bad;
1458 		if (v == 0)
1459 			sc->candelete = false;
1460 	}
1461 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1462 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1463 		if (VN_IS_DOOMED(nd.ni_vp)) {
1464 			/* Forced unmount. */
1465 			error = EBADF;
1466 			goto bad;
1467 		}
1468 	}
1469 	nd.ni_vp->v_vflag |= VV_MD;
1470 	VOP_UNLOCK(nd.ni_vp);
1471 
1472 	if (mdr->md_fwsectors != 0)
1473 		sc->fwsectors = mdr->md_fwsectors;
1474 	if (mdr->md_fwheads != 0)
1475 		sc->fwheads = mdr->md_fwheads;
1476 	snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1477 	    (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1478 	sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1479 	    MD_VERIFY);
1480 	if (!(flags & FWRITE))
1481 		sc->flags |= MD_READONLY;
1482 	sc->vnode = nd.ni_vp;
1483 
1484 	error = mdsetcred(sc, td->td_ucred);
1485 	if (error != 0) {
1486 		sc->vnode = NULL;
1487 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1488 		nd.ni_vp->v_vflag &= ~VV_MD;
1489 		goto bad;
1490 	}
1491 
1492 	sc->kva = kva_alloc(MAXPHYS + PAGE_SIZE);
1493 	return (0);
1494 bad:
1495 	VOP_UNLOCK(nd.ni_vp);
1496 	(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1497 	return (error);
1498 }
1499 
1500 static void
1501 g_md_providergone(struct g_provider *pp)
1502 {
1503 	struct md_s *sc = pp->geom->softc;
1504 
1505 	mtx_lock(&sc->queue_mtx);
1506 	sc->flags |= MD_PROVIDERGONE;
1507 	wakeup(&sc->flags);
1508 	mtx_unlock(&sc->queue_mtx);
1509 }
1510 
1511 static int
1512 mddestroy(struct md_s *sc, struct thread *td)
1513 {
1514 
1515 	if (sc->gp) {
1516 		g_topology_lock();
1517 		g_wither_geom(sc->gp, ENXIO);
1518 		g_topology_unlock();
1519 
1520 		mtx_lock(&sc->queue_mtx);
1521 		while (!(sc->flags & MD_PROVIDERGONE))
1522 			msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1523 		mtx_unlock(&sc->queue_mtx);
1524 	}
1525 	if (sc->devstat) {
1526 		devstat_remove_entry(sc->devstat);
1527 		sc->devstat = NULL;
1528 	}
1529 	mtx_lock(&sc->queue_mtx);
1530 	sc->flags |= MD_SHUTDOWN;
1531 	wakeup(sc);
1532 	while (!(sc->flags & MD_EXITING))
1533 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1534 	mtx_unlock(&sc->queue_mtx);
1535 	mtx_destroy(&sc->queue_mtx);
1536 	if (sc->vnode != NULL) {
1537 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1538 		sc->vnode->v_vflag &= ~VV_MD;
1539 		VOP_UNLOCK(sc->vnode);
1540 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1541 		    FREAD : (FREAD|FWRITE), sc->cred, td);
1542 	}
1543 	if (sc->cred != NULL)
1544 		crfree(sc->cred);
1545 	if (sc->object != NULL)
1546 		vm_object_deallocate(sc->object);
1547 	if (sc->indir)
1548 		destroy_indir(sc, sc->indir);
1549 	if (sc->uma)
1550 		uma_zdestroy(sc->uma);
1551 	if (sc->kva)
1552 		kva_free(sc->kva, MAXPHYS + PAGE_SIZE);
1553 
1554 	LIST_REMOVE(sc, list);
1555 	free_unr(md_uh, sc->unit);
1556 	free(sc, M_MD);
1557 	return (0);
1558 }
1559 
1560 static int
1561 mdresize(struct md_s *sc, struct md_req *mdr)
1562 {
1563 	int error, res;
1564 	vm_pindex_t oldpages, newpages;
1565 
1566 	switch (sc->type) {
1567 	case MD_VNODE:
1568 	case MD_NULL:
1569 		break;
1570 	case MD_SWAP:
1571 		if (mdr->md_mediasize <= 0 ||
1572 		    (mdr->md_mediasize % PAGE_SIZE) != 0)
1573 			return (EDOM);
1574 		oldpages = OFF_TO_IDX(sc->mediasize);
1575 		newpages = OFF_TO_IDX(mdr->md_mediasize);
1576 		if (newpages < oldpages) {
1577 			VM_OBJECT_WLOCK(sc->object);
1578 			vm_object_page_remove(sc->object, newpages, 0, 0);
1579 			swap_release_by_cred(IDX_TO_OFF(oldpages -
1580 			    newpages), sc->cred);
1581 			sc->object->charge = IDX_TO_OFF(newpages);
1582 			sc->object->size = newpages;
1583 			VM_OBJECT_WUNLOCK(sc->object);
1584 		} else if (newpages > oldpages) {
1585 			res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1586 			    oldpages), sc->cred);
1587 			if (!res)
1588 				return (ENOMEM);
1589 			if ((mdr->md_options & MD_RESERVE) ||
1590 			    (sc->flags & MD_RESERVE)) {
1591 				error = swap_pager_reserve(sc->object,
1592 				    oldpages, newpages - oldpages);
1593 				if (error < 0) {
1594 					swap_release_by_cred(
1595 					    IDX_TO_OFF(newpages - oldpages),
1596 					    sc->cred);
1597 					return (EDOM);
1598 				}
1599 			}
1600 			VM_OBJECT_WLOCK(sc->object);
1601 			sc->object->charge = IDX_TO_OFF(newpages);
1602 			sc->object->size = newpages;
1603 			VM_OBJECT_WUNLOCK(sc->object);
1604 		}
1605 		break;
1606 	default:
1607 		return (EOPNOTSUPP);
1608 	}
1609 
1610 	sc->mediasize = mdr->md_mediasize;
1611 
1612 	g_topology_lock();
1613 	g_resize_provider(sc->pp, sc->mediasize);
1614 	g_topology_unlock();
1615 	return (0);
1616 }
1617 
1618 static int
1619 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1620 {
1621 	vm_ooffset_t npage;
1622 	int error;
1623 
1624 	/*
1625 	 * Range check.  Disallow negative sizes and sizes not being
1626 	 * multiple of page size.
1627 	 */
1628 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1629 		return (EDOM);
1630 
1631 	/*
1632 	 * Allocate an OBJT_SWAP object.
1633 	 *
1634 	 * Note the truncation.
1635 	 */
1636 
1637 	if ((mdr->md_options & MD_VERIFY) != 0)
1638 		return (EINVAL);
1639 	npage = mdr->md_mediasize / PAGE_SIZE;
1640 	if (mdr->md_fwsectors != 0)
1641 		sc->fwsectors = mdr->md_fwsectors;
1642 	if (mdr->md_fwheads != 0)
1643 		sc->fwheads = mdr->md_fwheads;
1644 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1645 	    VM_PROT_DEFAULT, 0, td->td_ucred);
1646 	if (sc->object == NULL)
1647 		return (ENOMEM);
1648 	sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1649 	if (mdr->md_options & MD_RESERVE) {
1650 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1651 			error = EDOM;
1652 			goto finish;
1653 		}
1654 	}
1655 	error = mdsetcred(sc, td->td_ucred);
1656  finish:
1657 	if (error != 0) {
1658 		vm_object_deallocate(sc->object);
1659 		sc->object = NULL;
1660 	}
1661 	return (error);
1662 }
1663 
1664 static int
1665 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1666 {
1667 
1668 	/*
1669 	 * Range check.  Disallow negative sizes and sizes not being
1670 	 * multiple of page size.
1671 	 */
1672 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1673 		return (EDOM);
1674 
1675 	return (0);
1676 }
1677 
1678 static int
1679 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1680 {
1681 	struct md_s *sc;
1682 	unsigned sectsize;
1683 	int error, i;
1684 
1685 	sx_assert(&md_sx, SA_XLOCKED);
1686 
1687 	switch (mdr->md_type) {
1688 	case MD_MALLOC:
1689 	case MD_PRELOAD:
1690 	case MD_VNODE:
1691 	case MD_SWAP:
1692 	case MD_NULL:
1693 		break;
1694 	default:
1695 		return (EINVAL);
1696 	}
1697 	if (mdr->md_sectorsize == 0)
1698 		sectsize = DEV_BSIZE;
1699 	else
1700 		sectsize = mdr->md_sectorsize;
1701 	if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1702 		return (EINVAL);
1703 	if (mdr->md_options & MD_AUTOUNIT)
1704 		sc = mdnew(-1, &error, mdr->md_type);
1705 	else {
1706 		if (mdr->md_unit > INT_MAX)
1707 			return (EINVAL);
1708 		sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1709 	}
1710 	if (sc == NULL)
1711 		return (error);
1712 	if (mdr->md_label != NULL)
1713 		error = copyinstr(mdr->md_label, sc->label,
1714 		    sizeof(sc->label), NULL);
1715 	if (error != 0)
1716 		goto err_after_new;
1717 	if (mdr->md_options & MD_AUTOUNIT)
1718 		mdr->md_unit = sc->unit;
1719 	sc->mediasize = mdr->md_mediasize;
1720 	sc->sectorsize = sectsize;
1721 	sc->candelete = true;
1722 	error = EDOOFUS;
1723 	switch (sc->type) {
1724 	case MD_MALLOC:
1725 		sc->start = mdstart_malloc;
1726 		error = mdcreate_malloc(sc, mdr);
1727 		break;
1728 	case MD_PRELOAD:
1729 		/*
1730 		 * We disallow attaching preloaded memory disks via
1731 		 * ioctl. Preloaded memory disks are automatically
1732 		 * attached in g_md_init().
1733 		 */
1734 		error = EOPNOTSUPP;
1735 		break;
1736 	case MD_VNODE:
1737 		sc->start = mdstart_vnode;
1738 		error = mdcreate_vnode(sc, mdr, td);
1739 		break;
1740 	case MD_SWAP:
1741 		sc->start = mdstart_swap;
1742 		error = mdcreate_swap(sc, mdr, td);
1743 		break;
1744 	case MD_NULL:
1745 		sc->start = mdstart_null;
1746 		error = mdcreate_null(sc, mdr, td);
1747 		break;
1748 	}
1749 err_after_new:
1750 	if (error != 0) {
1751 		mddestroy(sc, td);
1752 		return (error);
1753 	}
1754 
1755 	/* Prune off any residual fractional sector */
1756 	i = sc->mediasize % sc->sectorsize;
1757 	sc->mediasize -= i;
1758 
1759 	mdinit(sc);
1760 	return (0);
1761 }
1762 
1763 static int
1764 kern_mdattach(struct thread *td, struct md_req *mdr)
1765 {
1766 	int error;
1767 
1768 	sx_xlock(&md_sx);
1769 	error = kern_mdattach_locked(td, mdr);
1770 	sx_xunlock(&md_sx);
1771 	return (error);
1772 }
1773 
1774 static int
1775 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1776 {
1777 	struct md_s *sc;
1778 
1779 	sx_assert(&md_sx, SA_XLOCKED);
1780 
1781 	if (mdr->md_mediasize != 0 ||
1782 	    (mdr->md_options & ~MD_FORCE) != 0)
1783 		return (EINVAL);
1784 
1785 	sc = mdfind(mdr->md_unit);
1786 	if (sc == NULL)
1787 		return (ENOENT);
1788 	if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1789 	    !(mdr->md_options & MD_FORCE))
1790 		return (EBUSY);
1791 	return (mddestroy(sc, td));
1792 }
1793 
1794 static int
1795 kern_mddetach(struct thread *td, struct md_req *mdr)
1796 {
1797 	int error;
1798 
1799 	sx_xlock(&md_sx);
1800 	error = kern_mddetach_locked(td, mdr);
1801 	sx_xunlock(&md_sx);
1802 	return (error);
1803 }
1804 
1805 static int
1806 kern_mdresize_locked(struct md_req *mdr)
1807 {
1808 	struct md_s *sc;
1809 
1810 	sx_assert(&md_sx, SA_XLOCKED);
1811 
1812 	if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1813 		return (EINVAL);
1814 
1815 	sc = mdfind(mdr->md_unit);
1816 	if (sc == NULL)
1817 		return (ENOENT);
1818 	if (mdr->md_mediasize < sc->sectorsize)
1819 		return (EINVAL);
1820 	mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize;
1821 	if (mdr->md_mediasize < sc->mediasize &&
1822 	    !(sc->flags & MD_FORCE) &&
1823 	    !(mdr->md_options & MD_FORCE))
1824 		return (EBUSY);
1825 	return (mdresize(sc, mdr));
1826 }
1827 
1828 static int
1829 kern_mdresize(struct md_req *mdr)
1830 {
1831 	int error;
1832 
1833 	sx_xlock(&md_sx);
1834 	error = kern_mdresize_locked(mdr);
1835 	sx_xunlock(&md_sx);
1836 	return (error);
1837 }
1838 
1839 static int
1840 kern_mdquery_locked(struct md_req *mdr)
1841 {
1842 	struct md_s *sc;
1843 	int error;
1844 
1845 	sx_assert(&md_sx, SA_XLOCKED);
1846 
1847 	sc = mdfind(mdr->md_unit);
1848 	if (sc == NULL)
1849 		return (ENOENT);
1850 	mdr->md_type = sc->type;
1851 	mdr->md_options = sc->flags;
1852 	mdr->md_mediasize = sc->mediasize;
1853 	mdr->md_sectorsize = sc->sectorsize;
1854 	error = 0;
1855 	if (mdr->md_label != NULL) {
1856 		error = copyout(sc->label, mdr->md_label,
1857 		    strlen(sc->label) + 1);
1858 		if (error != 0)
1859 			return (error);
1860 	}
1861 	if (sc->type == MD_VNODE ||
1862 	    (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1863 		error = copyout(sc->file, mdr->md_file,
1864 		    strlen(sc->file) + 1);
1865 	return (error);
1866 }
1867 
1868 static int
1869 kern_mdquery(struct md_req *mdr)
1870 {
1871 	int error;
1872 
1873 	sx_xlock(&md_sx);
1874 	error = kern_mdquery_locked(mdr);
1875 	sx_xunlock(&md_sx);
1876 	return (error);
1877 }
1878 
1879 /* Copy members that are not userspace pointers. */
1880 #define	MD_IOCTL2REQ(mdio, mdr) do {					\
1881 	(mdr)->md_unit = (mdio)->md_unit;				\
1882 	(mdr)->md_type = (mdio)->md_type;				\
1883 	(mdr)->md_mediasize = (mdio)->md_mediasize;			\
1884 	(mdr)->md_sectorsize = (mdio)->md_sectorsize;			\
1885 	(mdr)->md_options = (mdio)->md_options;				\
1886 	(mdr)->md_fwheads = (mdio)->md_fwheads;				\
1887 	(mdr)->md_fwsectors = (mdio)->md_fwsectors;			\
1888 	(mdr)->md_units = &(mdio)->md_pad[0];				\
1889 	(mdr)->md_units_nitems = nitems((mdio)->md_pad);		\
1890 } while(0)
1891 
1892 /* Copy members that might have been updated */
1893 #define MD_REQ2IOCTL(mdr, mdio) do {					\
1894 	(mdio)->md_unit = (mdr)->md_unit;				\
1895 	(mdio)->md_type = (mdr)->md_type;				\
1896 	(mdio)->md_mediasize = (mdr)->md_mediasize;			\
1897 	(mdio)->md_sectorsize = (mdr)->md_sectorsize;			\
1898 	(mdio)->md_options = (mdr)->md_options;				\
1899 	(mdio)->md_fwheads = (mdr)->md_fwheads;				\
1900 	(mdio)->md_fwsectors = (mdr)->md_fwsectors;			\
1901 } while(0)
1902 
1903 static int
1904 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1905     struct thread *td)
1906 {
1907 	struct md_req mdr;
1908 	int error;
1909 
1910 	if (md_debug)
1911 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1912 			devtoname(dev), cmd, addr, flags, td);
1913 
1914 	bzero(&mdr, sizeof(mdr));
1915 	switch (cmd) {
1916 	case MDIOCATTACH:
1917 	case MDIOCDETACH:
1918 	case MDIOCRESIZE:
1919 	case MDIOCQUERY: {
1920 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1921 		if (mdio->md_version != MDIOVERSION)
1922 			return (EINVAL);
1923 		MD_IOCTL2REQ(mdio, &mdr);
1924 		mdr.md_file = mdio->md_file;
1925 		mdr.md_file_seg = UIO_USERSPACE;
1926 		/* If the file is adjacent to the md_ioctl it's in kernel. */
1927 		if ((void *)mdio->md_file == (void *)(mdio + 1))
1928 			mdr.md_file_seg = UIO_SYSSPACE;
1929 		mdr.md_label = mdio->md_label;
1930 		break;
1931 	}
1932 #ifdef COMPAT_FREEBSD32
1933 	case MDIOCATTACH_32:
1934 	case MDIOCDETACH_32:
1935 	case MDIOCRESIZE_32:
1936 	case MDIOCQUERY_32: {
1937 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1938 		if (mdio->md_version != MDIOVERSION)
1939 			return (EINVAL);
1940 		MD_IOCTL2REQ(mdio, &mdr);
1941 		mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1942 		mdr.md_file_seg = UIO_USERSPACE;
1943 		mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1944 		break;
1945 	}
1946 #endif
1947 	default:
1948 		/* Fall through to handler switch. */
1949 		break;
1950 	}
1951 
1952 	error = 0;
1953 	switch (cmd) {
1954 	case MDIOCATTACH:
1955 #ifdef COMPAT_FREEBSD32
1956 	case MDIOCATTACH_32:
1957 #endif
1958 		error = kern_mdattach(td, &mdr);
1959 		break;
1960 	case MDIOCDETACH:
1961 #ifdef COMPAT_FREEBSD32
1962 	case MDIOCDETACH_32:
1963 #endif
1964 		error = kern_mddetach(td, &mdr);
1965 		break;
1966 	case MDIOCRESIZE:
1967 #ifdef COMPAT_FREEBSD32
1968 	case MDIOCRESIZE_32:
1969 #endif
1970 		error = kern_mdresize(&mdr);
1971 		break;
1972 	case MDIOCQUERY:
1973 #ifdef COMPAT_FREEBSD32
1974 	case MDIOCQUERY_32:
1975 #endif
1976 		error = kern_mdquery(&mdr);
1977 		break;
1978 	default:
1979 		error = ENOIOCTL;
1980 	}
1981 
1982 	switch (cmd) {
1983 	case MDIOCATTACH:
1984 	case MDIOCQUERY: {
1985 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1986 		MD_REQ2IOCTL(&mdr, mdio);
1987 		break;
1988 	}
1989 #ifdef COMPAT_FREEBSD32
1990 	case MDIOCATTACH_32:
1991 	case MDIOCQUERY_32: {
1992 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1993 		MD_REQ2IOCTL(&mdr, mdio);
1994 		break;
1995 	}
1996 #endif
1997 	default:
1998 		/* Other commands to not alter mdr. */
1999 		break;
2000 	}
2001 
2002 	return (error);
2003 }
2004 
2005 static void
2006 md_preloaded(u_char *image, size_t length, const char *name)
2007 {
2008 	struct md_s *sc;
2009 	int error;
2010 
2011 	sc = mdnew(-1, &error, MD_PRELOAD);
2012 	if (sc == NULL)
2013 		return;
2014 	sc->mediasize = length;
2015 	sc->sectorsize = DEV_BSIZE;
2016 	sc->pl_ptr = image;
2017 	sc->pl_len = length;
2018 	sc->start = mdstart_preload;
2019 	if (name != NULL)
2020 		strlcpy(sc->file, name, sizeof(sc->file));
2021 #ifdef MD_ROOT
2022 	if (sc->unit == 0) {
2023 #ifndef ROOTDEVNAME
2024 		rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2025 #endif
2026 #ifdef MD_ROOT_READONLY
2027 		sc->flags |= MD_READONLY;
2028 #endif
2029 	}
2030 #endif
2031 	mdinit(sc);
2032 	if (name != NULL) {
2033 		printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2034 		    MD_NAME, sc->unit, name, length, image);
2035 	} else {
2036 		printf("%s%d: Embedded image %zd bytes at %p\n",
2037 		    MD_NAME, sc->unit, length, image);
2038 	}
2039 }
2040 
2041 static void
2042 g_md_init(struct g_class *mp __unused)
2043 {
2044 	caddr_t mod;
2045 	u_char *ptr, *name, *type;
2046 	unsigned len;
2047 	int i;
2048 
2049 	/* figure out log2(NINDIR) */
2050 	for (i = NINDIR, nshift = -1; i; nshift++)
2051 		i >>= 1;
2052 
2053 	mod = NULL;
2054 	sx_init(&md_sx, "MD config lock");
2055 	g_topology_unlock();
2056 	md_uh = new_unrhdr(0, INT_MAX, NULL);
2057 #ifdef MD_ROOT
2058 	if (mfs_root_size != 0) {
2059 		sx_xlock(&md_sx);
2060 #ifdef MD_ROOT_MEM
2061 		md_preloaded(mfs_root, mfs_root_size, NULL);
2062 #else
2063 		md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2064 		    NULL);
2065 #endif
2066 		sx_xunlock(&md_sx);
2067 	}
2068 #endif
2069 	/* XXX: are preload_* static or do they need Giant ? */
2070 	while ((mod = preload_search_next_name(mod)) != NULL) {
2071 		name = (char *)preload_search_info(mod, MODINFO_NAME);
2072 		if (name == NULL)
2073 			continue;
2074 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
2075 		if (type == NULL)
2076 			continue;
2077 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2078 			continue;
2079 		ptr = preload_fetch_addr(mod);
2080 		len = preload_fetch_size(mod);
2081 		if (ptr != NULL && len != 0) {
2082 			sx_xlock(&md_sx);
2083 			md_preloaded(ptr, len, name);
2084 			sx_xunlock(&md_sx);
2085 		}
2086 	}
2087 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2088 	    0600, MDCTL_NAME);
2089 	g_topology_lock();
2090 }
2091 
2092 static void
2093 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2094     struct g_consumer *cp __unused, struct g_provider *pp)
2095 {
2096 	struct md_s *mp;
2097 	char *type;
2098 
2099 	mp = gp->softc;
2100 	if (mp == NULL)
2101 		return;
2102 
2103 	switch (mp->type) {
2104 	case MD_MALLOC:
2105 		type = "malloc";
2106 		break;
2107 	case MD_PRELOAD:
2108 		type = "preload";
2109 		break;
2110 	case MD_VNODE:
2111 		type = "vnode";
2112 		break;
2113 	case MD_SWAP:
2114 		type = "swap";
2115 		break;
2116 	case MD_NULL:
2117 		type = "null";
2118 		break;
2119 	default:
2120 		type = "unknown";
2121 		break;
2122 	}
2123 
2124 	if (pp != NULL) {
2125 		if (indent == NULL) {
2126 			sbuf_printf(sb, " u %d", mp->unit);
2127 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2128 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2129 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2130 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2131 			sbuf_printf(sb, " t %s", type);
2132 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2133 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2134 				sbuf_printf(sb, " file %s", mp->file);
2135 			sbuf_printf(sb, " label %s", mp->label);
2136 		} else {
2137 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2138 			    mp->unit);
2139 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2140 			    indent, (uintmax_t) mp->sectorsize);
2141 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2142 			    indent, (uintmax_t) mp->fwheads);
2143 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2144 			    indent, (uintmax_t) mp->fwsectors);
2145 			if (mp->ident[0] != '\0') {
2146 				sbuf_printf(sb, "%s<ident>", indent);
2147 				g_conf_printf_escaped(sb, "%s", mp->ident);
2148 				sbuf_printf(sb, "</ident>\n");
2149 			}
2150 			sbuf_printf(sb, "%s<length>%ju</length>\n",
2151 			    indent, (uintmax_t) mp->mediasize);
2152 			sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2153 			    (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2154 			sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2155 			    (mp->flags & MD_READONLY) == 0 ? "read-write":
2156 			    "read-only");
2157 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2158 			    type);
2159 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2160 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2161 				sbuf_printf(sb, "%s<file>", indent);
2162 				g_conf_printf_escaped(sb, "%s", mp->file);
2163 				sbuf_printf(sb, "</file>\n");
2164 			}
2165 			if (mp->type == MD_VNODE)
2166 				sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2167 				    (mp->flags & MD_CACHE) == 0 ? "off": "on");
2168 			sbuf_printf(sb, "%s<label>", indent);
2169 			g_conf_printf_escaped(sb, "%s", mp->label);
2170 			sbuf_printf(sb, "</label>\n");
2171 		}
2172 	}
2173 }
2174 
2175 static void
2176 g_md_fini(struct g_class *mp __unused)
2177 {
2178 
2179 	sx_destroy(&md_sx);
2180 	if (status_dev != NULL)
2181 		destroy_dev(status_dev);
2182 	delete_unrhdr(md_uh);
2183 }
2184