xref: /freebsd/sys/dev/md/md.c (revision 3a3deb00a5e449c9478156b162dfa10ec82a2a3f)
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  * $FreeBSD$
12  *
13  */
14 
15 /*-
16  * The following functions are based in the vn(4) driver: mdstart_swap(),
17  * mdstart_vnode(), mdcreate_swap(), mdcreate_vnode() and mddestroy(),
18  * and as such under the following copyright:
19  *
20  * Copyright (c) 1988 University of Utah.
21  * Copyright (c) 1990, 1993
22  *	The Regents of the University of California.  All rights reserved.
23  * Copyright (c) 2013 The FreeBSD Foundation
24  * All rights reserved.
25  *
26  * This code is derived from software contributed to Berkeley by
27  * the Systems Programming Group of the University of Utah Computer
28  * Science Department.
29  *
30  * Portions of this software were developed by Konstantin Belousov
31  * under sponsorship from the FreeBSD Foundation.
32  *
33  * Redistribution and use in source and binary forms, with or without
34  * modification, are permitted provided that the following conditions
35  * are met:
36  * 1. Redistributions of source code must retain the above copyright
37  *    notice, this list of conditions and the following disclaimer.
38  * 2. Redistributions in binary form must reproduce the above copyright
39  *    notice, this list of conditions and the following disclaimer in the
40  *    documentation and/or other materials provided with the distribution.
41  * 3. Neither the name of the University nor the names of its contributors
42  *    may be used to endorse or promote products derived from this software
43  *    without specific prior written permission.
44  *
45  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
46  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
49  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55  * SUCH DAMAGE.
56  *
57  * from: Utah Hdr: vn.c 1.13 94/04/02
58  *
59  *	from: @(#)vn.c	8.6 (Berkeley) 4/1/94
60  * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
61  */
62 
63 #include "opt_rootdevname.h"
64 #include "opt_geom.h"
65 #include "opt_md.h"
66 
67 #include <sys/param.h>
68 #include <sys/systm.h>
69 #include <sys/bio.h>
70 #include <sys/buf.h>
71 #include <sys/conf.h>
72 #include <sys/devicestat.h>
73 #include <sys/fcntl.h>
74 #include <sys/kernel.h>
75 #include <sys/kthread.h>
76 #include <sys/limits.h>
77 #include <sys/linker.h>
78 #include <sys/lock.h>
79 #include <sys/malloc.h>
80 #include <sys/mdioctl.h>
81 #include <sys/mount.h>
82 #include <sys/mutex.h>
83 #include <sys/sx.h>
84 #include <sys/namei.h>
85 #include <sys/proc.h>
86 #include <sys/queue.h>
87 #include <sys/rwlock.h>
88 #include <sys/sbuf.h>
89 #include <sys/sched.h>
90 #include <sys/sf_buf.h>
91 #include <sys/sysctl.h>
92 #include <sys/uio.h>
93 #include <sys/vnode.h>
94 #include <sys/disk.h>
95 
96 #include <geom/geom.h>
97 #include <geom/geom_int.h>
98 
99 #include <vm/vm.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 } __attribute__((__packed__));
149 CTASSERT((sizeof(struct md_ioctl32)) == 436);
150 
151 #define	MDIOCATTACH_32	_IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
152 #define	MDIOCDETACH_32	_IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
153 #define	MDIOCQUERY_32	_IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
154 #define	MDIOCRESIZE_32	_IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
155 #endif /* COMPAT_FREEBSD32 */
156 
157 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
158 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
159 
160 static int md_debug;
161 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
162     "Enable md(4) debug messages");
163 static int md_malloc_wait;
164 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
165     "Allow malloc to wait for memory allocations");
166 
167 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
168 #define	MD_ROOT_FSTYPE	"ufs"
169 #endif
170 
171 #if defined(MD_ROOT)
172 /*
173  * Preloaded image gets put here.
174  */
175 #if defined(MD_ROOT_SIZE)
176 /*
177  * We put the mfs_root symbol into the oldmfs section of the kernel object file.
178  * Applications that patch the object with the image can determine
179  * the size looking at the oldmfs section size within the kernel.
180  */
181 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
182 const int mfs_root_size = sizeof(mfs_root);
183 #elif defined(MD_ROOT_MEM)
184 /* MD region already mapped in the memory */
185 u_char *mfs_root;
186 int mfs_root_size;
187 #else
188 extern volatile u_char __weak_symbol mfs_root;
189 extern volatile u_char __weak_symbol mfs_root_end;
190 __GLOBL(mfs_root);
191 __GLOBL(mfs_root_end);
192 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
193 #endif
194 #endif
195 
196 static g_init_t g_md_init;
197 static g_fini_t g_md_fini;
198 static g_start_t g_md_start;
199 static g_access_t g_md_access;
200 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
201     struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
202 static g_provgone_t g_md_providergone;
203 
204 static struct cdev *status_dev = NULL;
205 static struct sx md_sx;
206 static struct unrhdr *md_uh;
207 
208 static d_ioctl_t mdctlioctl;
209 
210 static struct cdevsw mdctl_cdevsw = {
211 	.d_version =	D_VERSION,
212 	.d_ioctl =	mdctlioctl,
213 	.d_name =	MD_NAME,
214 };
215 
216 struct g_class g_md_class = {
217 	.name = "MD",
218 	.version = G_VERSION,
219 	.init = g_md_init,
220 	.fini = g_md_fini,
221 	.start = g_md_start,
222 	.access = g_md_access,
223 	.dumpconf = g_md_dumpconf,
224 	.providergone = g_md_providergone,
225 };
226 
227 DECLARE_GEOM_CLASS(g_md_class, g_md);
228 
229 
230 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
231 
232 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
233 #define NMASK	(NINDIR-1)
234 static int nshift;
235 
236 static uma_zone_t md_pbuf_zone;
237 
238 struct indir {
239 	uintptr_t	*array;
240 	u_int		total;
241 	u_int		used;
242 	u_int		shift;
243 };
244 
245 struct md_s {
246 	int unit;
247 	LIST_ENTRY(md_s) list;
248 	struct bio_queue_head bio_queue;
249 	struct mtx queue_mtx;
250 	struct mtx stat_mtx;
251 	struct cdev *dev;
252 	enum md_types type;
253 	off_t mediasize;
254 	unsigned sectorsize;
255 	unsigned opencount;
256 	unsigned fwheads;
257 	unsigned fwsectors;
258 	char ident[32];
259 	unsigned flags;
260 	char name[20];
261 	struct proc *procp;
262 	struct g_geom *gp;
263 	struct g_provider *pp;
264 	int (*start)(struct md_s *sc, struct bio *bp);
265 	struct devstat *devstat;
266 
267 	/* MD_MALLOC related fields */
268 	struct indir *indir;
269 	uma_zone_t uma;
270 
271 	/* MD_PRELOAD related fields */
272 	u_char *pl_ptr;
273 	size_t pl_len;
274 
275 	/* MD_VNODE related fields */
276 	struct vnode *vnode;
277 	char file[PATH_MAX];
278 	char label[PATH_MAX];
279 	struct ucred *cred;
280 
281 	/* MD_SWAP related fields */
282 	vm_object_t object;
283 };
284 
285 static struct indir *
286 new_indir(u_int shift)
287 {
288 	struct indir *ip;
289 
290 	ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
291 	    | M_ZERO);
292 	if (ip == NULL)
293 		return (NULL);
294 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
295 	    M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
296 	if (ip->array == NULL) {
297 		free(ip, M_MD);
298 		return (NULL);
299 	}
300 	ip->total = NINDIR;
301 	ip->shift = shift;
302 	return (ip);
303 }
304 
305 static void
306 del_indir(struct indir *ip)
307 {
308 
309 	free(ip->array, M_MDSECT);
310 	free(ip, M_MD);
311 }
312 
313 static void
314 destroy_indir(struct md_s *sc, struct indir *ip)
315 {
316 	int i;
317 
318 	for (i = 0; i < NINDIR; i++) {
319 		if (!ip->array[i])
320 			continue;
321 		if (ip->shift)
322 			destroy_indir(sc, (struct indir*)(ip->array[i]));
323 		else if (ip->array[i] > 255)
324 			uma_zfree(sc->uma, (void *)(ip->array[i]));
325 	}
326 	del_indir(ip);
327 }
328 
329 /*
330  * This function does the math and allocates the top level "indir" structure
331  * for a device of "size" sectors.
332  */
333 
334 static struct indir *
335 dimension(off_t size)
336 {
337 	off_t rcnt;
338 	struct indir *ip;
339 	int layer;
340 
341 	rcnt = size;
342 	layer = 0;
343 	while (rcnt > NINDIR) {
344 		rcnt /= NINDIR;
345 		layer++;
346 	}
347 
348 	/*
349 	 * XXX: the top layer is probably not fully populated, so we allocate
350 	 * too much space for ip->array in here.
351 	 */
352 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
353 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
354 	    M_MDSECT, M_WAITOK | M_ZERO);
355 	ip->total = NINDIR;
356 	ip->shift = layer * nshift;
357 	return (ip);
358 }
359 
360 /*
361  * Read a given sector
362  */
363 
364 static uintptr_t
365 s_read(struct indir *ip, off_t offset)
366 {
367 	struct indir *cip;
368 	int idx;
369 	uintptr_t up;
370 
371 	if (md_debug > 1)
372 		printf("s_read(%jd)\n", (intmax_t)offset);
373 	up = 0;
374 	for (cip = ip; cip != NULL;) {
375 		if (cip->shift) {
376 			idx = (offset >> cip->shift) & NMASK;
377 			up = cip->array[idx];
378 			cip = (struct indir *)up;
379 			continue;
380 		}
381 		idx = offset & NMASK;
382 		return (cip->array[idx]);
383 	}
384 	return (0);
385 }
386 
387 /*
388  * Write a given sector, prune the tree if the value is 0
389  */
390 
391 static int
392 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
393 {
394 	struct indir *cip, *lip[10];
395 	int idx, li;
396 	uintptr_t up;
397 
398 	if (md_debug > 1)
399 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
400 	up = 0;
401 	li = 0;
402 	cip = ip;
403 	for (;;) {
404 		lip[li++] = cip;
405 		if (cip->shift) {
406 			idx = (offset >> cip->shift) & NMASK;
407 			up = cip->array[idx];
408 			if (up != 0) {
409 				cip = (struct indir *)up;
410 				continue;
411 			}
412 			/* Allocate branch */
413 			cip->array[idx] =
414 			    (uintptr_t)new_indir(cip->shift - nshift);
415 			if (cip->array[idx] == 0)
416 				return (ENOSPC);
417 			cip->used++;
418 			up = cip->array[idx];
419 			cip = (struct indir *)up;
420 			continue;
421 		}
422 		/* leafnode */
423 		idx = offset & NMASK;
424 		up = cip->array[idx];
425 		if (up != 0)
426 			cip->used--;
427 		cip->array[idx] = ptr;
428 		if (ptr != 0)
429 			cip->used++;
430 		break;
431 	}
432 	if (cip->used != 0 || li == 1)
433 		return (0);
434 	li--;
435 	while (cip->used == 0 && cip != ip) {
436 		li--;
437 		idx = (offset >> lip[li]->shift) & NMASK;
438 		up = lip[li]->array[idx];
439 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
440 		del_indir(cip);
441 		lip[li]->array[idx] = 0;
442 		lip[li]->used--;
443 		cip = lip[li];
444 	}
445 	return (0);
446 }
447 
448 
449 static int
450 g_md_access(struct g_provider *pp, int r, int w, int e)
451 {
452 	struct md_s *sc;
453 
454 	sc = pp->geom->softc;
455 	if (sc == NULL) {
456 		if (r <= 0 && w <= 0 && e <= 0)
457 			return (0);
458 		return (ENXIO);
459 	}
460 	r += pp->acr;
461 	w += pp->acw;
462 	e += pp->ace;
463 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
464 		return (EROFS);
465 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
466 		sc->opencount = 1;
467 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
468 		sc->opencount = 0;
469 	}
470 	return (0);
471 }
472 
473 static void
474 g_md_start(struct bio *bp)
475 {
476 	struct md_s *sc;
477 
478 	sc = bp->bio_to->geom->softc;
479 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
480 		mtx_lock(&sc->stat_mtx);
481 		devstat_start_transaction_bio(sc->devstat, bp);
482 		mtx_unlock(&sc->stat_mtx);
483 	}
484 	mtx_lock(&sc->queue_mtx);
485 	bioq_disksort(&sc->bio_queue, bp);
486 	wakeup(sc);
487 	mtx_unlock(&sc->queue_mtx);
488 }
489 
490 #define	MD_MALLOC_MOVE_ZERO	1
491 #define	MD_MALLOC_MOVE_FILL	2
492 #define	MD_MALLOC_MOVE_READ	3
493 #define	MD_MALLOC_MOVE_WRITE	4
494 #define	MD_MALLOC_MOVE_CMP	5
495 
496 static int
497 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
498     void *ptr, u_char fill, int op)
499 {
500 	struct sf_buf *sf;
501 	vm_page_t m, *mp1;
502 	char *p, first;
503 	off_t *uc;
504 	unsigned n;
505 	int error, i, ma_offs1, sz, first_read;
506 
507 	m = NULL;
508 	error = 0;
509 	sf = NULL;
510 	/* if (op == MD_MALLOC_MOVE_CMP) { gcc */
511 		first = 0;
512 		first_read = 0;
513 		uc = ptr;
514 		mp1 = *mp;
515 		ma_offs1 = *ma_offs;
516 	/* } */
517 	sched_pin();
518 	for (n = sectorsize; n != 0; n -= sz) {
519 		sz = imin(PAGE_SIZE - *ma_offs, n);
520 		if (m != **mp) {
521 			if (sf != NULL)
522 				sf_buf_free(sf);
523 			m = **mp;
524 			sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
525 			    (md_malloc_wait ? 0 : SFB_NOWAIT));
526 			if (sf == NULL) {
527 				error = ENOMEM;
528 				break;
529 			}
530 		}
531 		p = (char *)sf_buf_kva(sf) + *ma_offs;
532 		switch (op) {
533 		case MD_MALLOC_MOVE_ZERO:
534 			bzero(p, sz);
535 			break;
536 		case MD_MALLOC_MOVE_FILL:
537 			memset(p, fill, sz);
538 			break;
539 		case MD_MALLOC_MOVE_READ:
540 			bcopy(ptr, p, sz);
541 			cpu_flush_dcache(p, sz);
542 			break;
543 		case MD_MALLOC_MOVE_WRITE:
544 			bcopy(p, ptr, sz);
545 			break;
546 		case MD_MALLOC_MOVE_CMP:
547 			for (i = 0; i < sz; i++, p++) {
548 				if (!first_read) {
549 					*uc = (u_char)*p;
550 					first = *p;
551 					first_read = 1;
552 				} else if (*p != first) {
553 					error = EDOOFUS;
554 					break;
555 				}
556 			}
557 			break;
558 		default:
559 			KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
560 			break;
561 		}
562 		if (error != 0)
563 			break;
564 		*ma_offs += sz;
565 		*ma_offs %= PAGE_SIZE;
566 		if (*ma_offs == 0)
567 			(*mp)++;
568 		ptr = (char *)ptr + sz;
569 	}
570 
571 	if (sf != NULL)
572 		sf_buf_free(sf);
573 	sched_unpin();
574 	if (op == MD_MALLOC_MOVE_CMP && error != 0) {
575 		*mp = mp1;
576 		*ma_offs = ma_offs1;
577 	}
578 	return (error);
579 }
580 
581 static int
582 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
583     unsigned len, void *ptr, u_char fill, int op)
584 {
585 	bus_dma_segment_t *vlist;
586 	uint8_t *p, *end, first;
587 	off_t *uc;
588 	int ma_offs, seg_len;
589 
590 	vlist = *pvlist;
591 	ma_offs = *pma_offs;
592 	uc = ptr;
593 
594 	for (; len != 0; len -= seg_len) {
595 		seg_len = imin(vlist->ds_len - ma_offs, len);
596 		p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
597 		switch (op) {
598 		case MD_MALLOC_MOVE_ZERO:
599 			bzero(p, seg_len);
600 			break;
601 		case MD_MALLOC_MOVE_FILL:
602 			memset(p, fill, seg_len);
603 			break;
604 		case MD_MALLOC_MOVE_READ:
605 			bcopy(ptr, p, seg_len);
606 			cpu_flush_dcache(p, seg_len);
607 			break;
608 		case MD_MALLOC_MOVE_WRITE:
609 			bcopy(p, ptr, seg_len);
610 			break;
611 		case MD_MALLOC_MOVE_CMP:
612 			end = p + seg_len;
613 			first = *uc = *p;
614 			/* Confirm all following bytes match the first */
615 			while (++p < end) {
616 				if (*p != first)
617 					return (EDOOFUS);
618 			}
619 			break;
620 		default:
621 			KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
622 			break;
623 		}
624 
625 		ma_offs += seg_len;
626 		if (ma_offs == vlist->ds_len) {
627 			ma_offs = 0;
628 			vlist++;
629 		}
630 		ptr = (uint8_t *)ptr + seg_len;
631 	}
632 	*pvlist = vlist;
633 	*pma_offs = ma_offs;
634 
635 	return (0);
636 }
637 
638 static int
639 mdstart_malloc(struct md_s *sc, struct bio *bp)
640 {
641 	u_char *dst;
642 	vm_page_t *m;
643 	bus_dma_segment_t *vlist;
644 	int i, error, error1, ma_offs, notmapped;
645 	off_t secno, nsec, uc;
646 	uintptr_t sp, osp;
647 
648 	switch (bp->bio_cmd) {
649 	case BIO_READ:
650 	case BIO_WRITE:
651 	case BIO_DELETE:
652 		break;
653 	default:
654 		return (EOPNOTSUPP);
655 	}
656 
657 	notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
658 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
659 	    (bus_dma_segment_t *)bp->bio_data : NULL;
660 	if (notmapped) {
661 		m = bp->bio_ma;
662 		ma_offs = bp->bio_ma_offset;
663 		dst = NULL;
664 		KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
665 	} else if (vlist != NULL) {
666 		ma_offs = bp->bio_ma_offset;
667 		dst = NULL;
668 	} else {
669 		dst = bp->bio_data;
670 	}
671 
672 	nsec = bp->bio_length / sc->sectorsize;
673 	secno = bp->bio_offset / sc->sectorsize;
674 	error = 0;
675 	while (nsec--) {
676 		osp = s_read(sc->indir, secno);
677 		if (bp->bio_cmd == BIO_DELETE) {
678 			if (osp != 0)
679 				error = s_write(sc->indir, secno, 0);
680 		} else if (bp->bio_cmd == BIO_READ) {
681 			if (osp == 0) {
682 				if (notmapped) {
683 					error = md_malloc_move_ma(&m, &ma_offs,
684 					    sc->sectorsize, NULL, 0,
685 					    MD_MALLOC_MOVE_ZERO);
686 				} else if (vlist != NULL) {
687 					error = md_malloc_move_vlist(&vlist,
688 					    &ma_offs, sc->sectorsize, NULL, 0,
689 					    MD_MALLOC_MOVE_ZERO);
690 				} else
691 					bzero(dst, sc->sectorsize);
692 			} else if (osp <= 255) {
693 				if (notmapped) {
694 					error = md_malloc_move_ma(&m, &ma_offs,
695 					    sc->sectorsize, NULL, osp,
696 					    MD_MALLOC_MOVE_FILL);
697 				} else if (vlist != NULL) {
698 					error = md_malloc_move_vlist(&vlist,
699 					    &ma_offs, sc->sectorsize, NULL, osp,
700 					    MD_MALLOC_MOVE_FILL);
701 				} else
702 					memset(dst, osp, sc->sectorsize);
703 			} else {
704 				if (notmapped) {
705 					error = md_malloc_move_ma(&m, &ma_offs,
706 					    sc->sectorsize, (void *)osp, 0,
707 					    MD_MALLOC_MOVE_READ);
708 				} else if (vlist != NULL) {
709 					error = md_malloc_move_vlist(&vlist,
710 					    &ma_offs, sc->sectorsize,
711 					    (void *)osp, 0,
712 					    MD_MALLOC_MOVE_READ);
713 				} else {
714 					bcopy((void *)osp, dst, sc->sectorsize);
715 					cpu_flush_dcache(dst, sc->sectorsize);
716 				}
717 			}
718 			osp = 0;
719 		} else if (bp->bio_cmd == BIO_WRITE) {
720 			if (sc->flags & MD_COMPRESS) {
721 				if (notmapped) {
722 					error1 = md_malloc_move_ma(&m, &ma_offs,
723 					    sc->sectorsize, &uc, 0,
724 					    MD_MALLOC_MOVE_CMP);
725 					i = error1 == 0 ? sc->sectorsize : 0;
726 				} else if (vlist != NULL) {
727 					error1 = md_malloc_move_vlist(&vlist,
728 					    &ma_offs, sc->sectorsize, &uc, 0,
729 					    MD_MALLOC_MOVE_CMP);
730 					i = error1 == 0 ? sc->sectorsize : 0;
731 				} else {
732 					uc = dst[0];
733 					for (i = 1; i < sc->sectorsize; i++) {
734 						if (dst[i] != uc)
735 							break;
736 					}
737 				}
738 			} else {
739 				i = 0;
740 				uc = 0;
741 			}
742 			if (i == sc->sectorsize) {
743 				if (osp != uc)
744 					error = s_write(sc->indir, secno, uc);
745 			} else {
746 				if (osp <= 255) {
747 					sp = (uintptr_t)uma_zalloc(sc->uma,
748 					    md_malloc_wait ? M_WAITOK :
749 					    M_NOWAIT);
750 					if (sp == 0) {
751 						error = ENOSPC;
752 						break;
753 					}
754 					if (notmapped) {
755 						error = md_malloc_move_ma(&m,
756 						    &ma_offs, sc->sectorsize,
757 						    (void *)sp, 0,
758 						    MD_MALLOC_MOVE_WRITE);
759 					} else if (vlist != NULL) {
760 						error = md_malloc_move_vlist(
761 						    &vlist, &ma_offs,
762 						    sc->sectorsize, (void *)sp,
763 						    0, MD_MALLOC_MOVE_WRITE);
764 					} else {
765 						bcopy(dst, (void *)sp,
766 						    sc->sectorsize);
767 					}
768 					error = s_write(sc->indir, secno, sp);
769 				} else {
770 					if (notmapped) {
771 						error = md_malloc_move_ma(&m,
772 						    &ma_offs, sc->sectorsize,
773 						    (void *)osp, 0,
774 						    MD_MALLOC_MOVE_WRITE);
775 					} else if (vlist != NULL) {
776 						error = md_malloc_move_vlist(
777 						    &vlist, &ma_offs,
778 						    sc->sectorsize, (void *)osp,
779 						    0, MD_MALLOC_MOVE_WRITE);
780 					} else {
781 						bcopy(dst, (void *)osp,
782 						    sc->sectorsize);
783 					}
784 					osp = 0;
785 				}
786 			}
787 		} else {
788 			error = EOPNOTSUPP;
789 		}
790 		if (osp > 255)
791 			uma_zfree(sc->uma, (void*)osp);
792 		if (error != 0)
793 			break;
794 		secno++;
795 		if (!notmapped && vlist == NULL)
796 			dst += sc->sectorsize;
797 	}
798 	bp->bio_resid = 0;
799 	return (error);
800 }
801 
802 static void
803 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
804 {
805 	off_t seg_len;
806 
807 	while (offset >= vlist->ds_len) {
808 		offset -= vlist->ds_len;
809 		vlist++;
810 	}
811 
812 	while (len != 0) {
813 		seg_len = omin(len, vlist->ds_len - offset);
814 		bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
815 		    seg_len);
816 		offset = 0;
817 		src = (uint8_t *)src + seg_len;
818 		len -= seg_len;
819 		vlist++;
820 	}
821 }
822 
823 static void
824 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
825 {
826 	off_t seg_len;
827 
828 	while (offset >= vlist->ds_len) {
829 		offset -= vlist->ds_len;
830 		vlist++;
831 	}
832 
833 	while (len != 0) {
834 		seg_len = omin(len, vlist->ds_len - offset);
835 		bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
836 		    seg_len);
837 		offset = 0;
838 		dst = (uint8_t *)dst + seg_len;
839 		len -= seg_len;
840 		vlist++;
841 	}
842 }
843 
844 static int
845 mdstart_preload(struct md_s *sc, struct bio *bp)
846 {
847 	uint8_t *p;
848 
849 	p = sc->pl_ptr + bp->bio_offset;
850 	switch (bp->bio_cmd) {
851 	case BIO_READ:
852 		if ((bp->bio_flags & BIO_VLIST) != 0) {
853 			mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
854 			    bp->bio_ma_offset, bp->bio_length);
855 		} else {
856 			bcopy(p, bp->bio_data, bp->bio_length);
857 		}
858 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
859 		break;
860 	case BIO_WRITE:
861 		if ((bp->bio_flags & BIO_VLIST) != 0) {
862 			mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
863 			    bp->bio_ma_offset, p, bp->bio_length);
864 		} else {
865 			bcopy(bp->bio_data, p, bp->bio_length);
866 		}
867 		break;
868 	}
869 	bp->bio_resid = 0;
870 	return (0);
871 }
872 
873 static int
874 mdstart_vnode(struct md_s *sc, struct bio *bp)
875 {
876 	int error;
877 	struct uio auio;
878 	struct iovec aiov;
879 	struct iovec *piov;
880 	struct mount *mp;
881 	struct vnode *vp;
882 	struct buf *pb;
883 	bus_dma_segment_t *vlist;
884 	struct thread *td;
885 	off_t iolen, iostart, len, zerosize;
886 	int ma_offs, npages;
887 
888 	switch (bp->bio_cmd) {
889 	case BIO_READ:
890 		auio.uio_rw = UIO_READ;
891 		break;
892 	case BIO_WRITE:
893 	case BIO_DELETE:
894 		auio.uio_rw = UIO_WRITE;
895 		break;
896 	case BIO_FLUSH:
897 		break;
898 	default:
899 		return (EOPNOTSUPP);
900 	}
901 
902 	td = curthread;
903 	vp = sc->vnode;
904 	pb = NULL;
905 	piov = NULL;
906 	ma_offs = bp->bio_ma_offset;
907 	len = bp->bio_length;
908 
909 	/*
910 	 * VNODE I/O
911 	 *
912 	 * If an error occurs, we set BIO_ERROR but we do not set
913 	 * B_INVAL because (for a write anyway), the buffer is
914 	 * still valid.
915 	 */
916 
917 	if (bp->bio_cmd == BIO_FLUSH) {
918 		(void) vn_start_write(vp, &mp, V_WAIT);
919 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
920 		error = VOP_FSYNC(vp, MNT_WAIT, td);
921 		VOP_UNLOCK(vp, 0);
922 		vn_finished_write(mp);
923 		return (error);
924 	}
925 
926 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
927 	auio.uio_resid = bp->bio_length;
928 	auio.uio_segflg = UIO_SYSSPACE;
929 	auio.uio_td = td;
930 
931 	if (bp->bio_cmd == BIO_DELETE) {
932 		/*
933 		 * Emulate BIO_DELETE by writing zeros.
934 		 */
935 		zerosize = ZERO_REGION_SIZE -
936 		    (ZERO_REGION_SIZE % sc->sectorsize);
937 		auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
938 		piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
939 		auio.uio_iov = piov;
940 		while (len > 0) {
941 			piov->iov_base = __DECONST(void *, zero_region);
942 			piov->iov_len = len;
943 			if (len > zerosize)
944 				piov->iov_len = zerosize;
945 			len -= piov->iov_len;
946 			piov++;
947 		}
948 		piov = auio.uio_iov;
949 	} else if ((bp->bio_flags & BIO_VLIST) != 0) {
950 		piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
951 		auio.uio_iov = piov;
952 		vlist = (bus_dma_segment_t *)bp->bio_data;
953 		while (len > 0) {
954 			piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
955 			    ma_offs);
956 			piov->iov_len = vlist->ds_len - ma_offs;
957 			if (piov->iov_len > len)
958 				piov->iov_len = len;
959 			len -= piov->iov_len;
960 			ma_offs = 0;
961 			vlist++;
962 			piov++;
963 		}
964 		auio.uio_iovcnt = piov - auio.uio_iov;
965 		piov = auio.uio_iov;
966 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
967 		pb = uma_zalloc(md_pbuf_zone, M_WAITOK);
968 		bp->bio_resid = len;
969 unmapped_step:
970 		npages = atop(min(MAXPHYS, round_page(len + (ma_offs &
971 		    PAGE_MASK))));
972 		iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
973 		KASSERT(iolen > 0, ("zero iolen"));
974 		pmap_qenter((vm_offset_t)pb->b_data,
975 		    &bp->bio_ma[atop(ma_offs)], npages);
976 		aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
977 		    (ma_offs & PAGE_MASK));
978 		aiov.iov_len = iolen;
979 		auio.uio_iov = &aiov;
980 		auio.uio_iovcnt = 1;
981 		auio.uio_resid = iolen;
982 	} else {
983 		aiov.iov_base = bp->bio_data;
984 		aiov.iov_len = bp->bio_length;
985 		auio.uio_iov = &aiov;
986 		auio.uio_iovcnt = 1;
987 	}
988 	iostart = auio.uio_offset;
989 	if (auio.uio_rw == UIO_READ) {
990 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
991 		error = VOP_READ(vp, &auio, 0, sc->cred);
992 		VOP_UNLOCK(vp, 0);
993 	} else {
994 		(void) vn_start_write(vp, &mp, V_WAIT);
995 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
996 		error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
997 		    sc->cred);
998 		VOP_UNLOCK(vp, 0);
999 		vn_finished_write(mp);
1000 		if (error == 0)
1001 			sc->flags &= ~MD_VERIFY;
1002 	}
1003 
1004 	/* When MD_CACHE is set, try to avoid double-caching the data. */
1005 	if (error == 0 && (sc->flags & MD_CACHE) == 0)
1006 		VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
1007 		    POSIX_FADV_DONTNEED);
1008 
1009 	if (pb != NULL) {
1010 		pmap_qremove((vm_offset_t)pb->b_data, npages);
1011 		if (error == 0) {
1012 			len -= iolen;
1013 			bp->bio_resid -= iolen;
1014 			ma_offs += iolen;
1015 			if (len > 0)
1016 				goto unmapped_step;
1017 		}
1018 		uma_zfree(md_pbuf_zone, pb);
1019 	}
1020 
1021 	free(piov, M_MD);
1022 	if (pb == NULL)
1023 		bp->bio_resid = auio.uio_resid;
1024 	return (error);
1025 }
1026 
1027 static int
1028 mdstart_swap(struct md_s *sc, struct bio *bp)
1029 {
1030 	vm_page_t m;
1031 	u_char *p;
1032 	vm_pindex_t i, lastp;
1033 	bus_dma_segment_t *vlist;
1034 	int rv, ma_offs, offs, len, lastend;
1035 
1036 	switch (bp->bio_cmd) {
1037 	case BIO_READ:
1038 	case BIO_WRITE:
1039 	case BIO_DELETE:
1040 		break;
1041 	default:
1042 		return (EOPNOTSUPP);
1043 	}
1044 
1045 	p = bp->bio_data;
1046 	ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1047 	    bp->bio_ma_offset : 0;
1048 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1049 	    (bus_dma_segment_t *)bp->bio_data : NULL;
1050 
1051 	/*
1052 	 * offs is the offset at which to start operating on the
1053 	 * next (ie, first) page.  lastp is the last page on
1054 	 * which we're going to operate.  lastend is the ending
1055 	 * position within that last page (ie, PAGE_SIZE if
1056 	 * we're operating on complete aligned pages).
1057 	 */
1058 	offs = bp->bio_offset % PAGE_SIZE;
1059 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1060 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1061 
1062 	rv = VM_PAGER_OK;
1063 	VM_OBJECT_WLOCK(sc->object);
1064 	vm_object_pip_add(sc->object, 1);
1065 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1066 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1067 		m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
1068 		if (bp->bio_cmd == BIO_READ) {
1069 			if (vm_page_all_valid(m))
1070 				rv = VM_PAGER_OK;
1071 			else
1072 				rv = vm_pager_get_pages(sc->object, &m, 1,
1073 				    NULL, NULL);
1074 			if (rv == VM_PAGER_ERROR) {
1075 				vm_page_free(m);
1076 				break;
1077 			} else if (rv == VM_PAGER_FAIL) {
1078 				/*
1079 				 * Pager does not have the page.  Zero
1080 				 * the allocated page, and mark it as
1081 				 * valid. Do not set dirty, the page
1082 				 * can be recreated if thrown out.
1083 				 */
1084 				pmap_zero_page(m);
1085 				vm_page_valid(m);
1086 			}
1087 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1088 				pmap_copy_pages(&m, offs, bp->bio_ma,
1089 				    ma_offs, len);
1090 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1091 				physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1092 				    vlist, ma_offs, len);
1093 				cpu_flush_dcache(p, len);
1094 			} else {
1095 				physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1096 				cpu_flush_dcache(p, len);
1097 			}
1098 		} else if (bp->bio_cmd == BIO_WRITE) {
1099 			if (len == PAGE_SIZE || vm_page_all_valid(m))
1100 				rv = VM_PAGER_OK;
1101 			else
1102 				rv = vm_pager_get_pages(sc->object, &m, 1,
1103 				    NULL, NULL);
1104 			if (rv == VM_PAGER_ERROR) {
1105 				vm_page_free(m);
1106 				break;
1107 			} else if (rv == VM_PAGER_FAIL)
1108 				pmap_zero_page(m);
1109 
1110 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1111 				pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1112 				    offs, len);
1113 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1114 				physcopyin_vlist(vlist, ma_offs,
1115 				    VM_PAGE_TO_PHYS(m) + offs, len);
1116 			} else {
1117 				physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1118 			}
1119 
1120 			vm_page_valid(m);
1121 			if (m->dirty != VM_PAGE_BITS_ALL) {
1122 				vm_page_dirty(m);
1123 				vm_pager_page_unswapped(m);
1124 			}
1125 		} else if (bp->bio_cmd == BIO_DELETE) {
1126 			if (len == PAGE_SIZE || vm_page_all_valid(m))
1127 				rv = VM_PAGER_OK;
1128 			else
1129 				rv = vm_pager_get_pages(sc->object, &m, 1,
1130 				    NULL, NULL);
1131 			if (rv == VM_PAGER_ERROR) {
1132 				vm_page_free(m);
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 					if (m->dirty != VM_PAGE_BITS_ALL) {
1142 						vm_page_dirty(m);
1143 						vm_pager_page_unswapped(m);
1144 					}
1145 				} else {
1146 					vm_pager_page_unswapped(m);
1147 					vm_page_free(m);
1148 					m = NULL;
1149 				}
1150 			}
1151 		}
1152 		if (m != NULL) {
1153 			vm_page_xunbusy(m);
1154 			vm_page_lock(m);
1155 			if (vm_page_active(m))
1156 				vm_page_reference(m);
1157 			else
1158 				vm_page_activate(m);
1159 			vm_page_unlock(m);
1160 		}
1161 
1162 		/* Actions on further pages start at offset 0 */
1163 		p += PAGE_SIZE - offs;
1164 		offs = 0;
1165 		ma_offs += len;
1166 	}
1167 	vm_object_pip_wakeup(sc->object);
1168 	VM_OBJECT_WUNLOCK(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_kthread(void *arg)
1190 {
1191 	struct md_s *sc;
1192 	struct bio *bp;
1193 	int error;
1194 
1195 	sc = arg;
1196 	thread_lock(curthread);
1197 	sched_prio(curthread, PRIBIO);
1198 	thread_unlock(curthread);
1199 	if (sc->type == MD_VNODE)
1200 		curthread->td_pflags |= TDP_NORUNNINGBUF;
1201 
1202 	for (;;) {
1203 		mtx_lock(&sc->queue_mtx);
1204 		if (sc->flags & MD_SHUTDOWN) {
1205 			sc->flags |= MD_EXITING;
1206 			mtx_unlock(&sc->queue_mtx);
1207 			kproc_exit(0);
1208 		}
1209 		bp = bioq_takefirst(&sc->bio_queue);
1210 		if (!bp) {
1211 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1212 			continue;
1213 		}
1214 		mtx_unlock(&sc->queue_mtx);
1215 		if (bp->bio_cmd == BIO_GETATTR) {
1216 			int isv = ((sc->flags & MD_VERIFY) != 0);
1217 
1218 			if ((sc->fwsectors && sc->fwheads &&
1219 			    (g_handleattr_int(bp, "GEOM::fwsectors",
1220 			    sc->fwsectors) ||
1221 			    g_handleattr_int(bp, "GEOM::fwheads",
1222 			    sc->fwheads))) ||
1223 			    g_handleattr_int(bp, "GEOM::candelete", 1))
1224 				error = -1;
1225 			else if (sc->ident[0] != '\0' &&
1226 			    g_handleattr_str(bp, "GEOM::ident", sc->ident))
1227 				error = -1;
1228 			else if (g_handleattr_int(bp, "MNT::verified", isv))
1229 				error = -1;
1230 			else
1231 				error = EOPNOTSUPP;
1232 		} else {
1233 			error = sc->start(sc, bp);
1234 		}
1235 
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 of
1240 			 * the i/o.  g_io_deliver() will translate from
1241 			 * bio_completed to that, but it also destroys the
1242 			 * bio so we must do our own translation.
1243 			 */
1244 			bp->bio_bcount = bp->bio_length;
1245 			bp->bio_resid = (error == -1 ? bp->bio_bcount : 0);
1246 			devstat_end_transaction_bio(sc->devstat, bp);
1247 		}
1248 		if (error != -1) {
1249 			bp->bio_completed = bp->bio_length;
1250 			g_io_deliver(bp, error);
1251 		}
1252 	}
1253 }
1254 
1255 static struct md_s *
1256 mdfind(int unit)
1257 {
1258 	struct md_s *sc;
1259 
1260 	LIST_FOREACH(sc, &md_softc_list, list) {
1261 		if (sc->unit == unit)
1262 			break;
1263 	}
1264 	return (sc);
1265 }
1266 
1267 static struct md_s *
1268 mdnew(int unit, int *errp, enum md_types type)
1269 {
1270 	struct md_s *sc;
1271 	int error;
1272 
1273 	*errp = 0;
1274 	if (unit == -1)
1275 		unit = alloc_unr(md_uh);
1276 	else
1277 		unit = alloc_unr_specific(md_uh, unit);
1278 
1279 	if (unit == -1) {
1280 		*errp = EBUSY;
1281 		return (NULL);
1282 	}
1283 
1284 	sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1285 	sc->type = type;
1286 	bioq_init(&sc->bio_queue);
1287 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1288 	mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1289 	sc->unit = unit;
1290 	sprintf(sc->name, "md%d", unit);
1291 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
1292 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1293 	if (error == 0)
1294 		return (sc);
1295 	LIST_REMOVE(sc, list);
1296 	mtx_destroy(&sc->stat_mtx);
1297 	mtx_destroy(&sc->queue_mtx);
1298 	free_unr(md_uh, sc->unit);
1299 	free(sc, M_MD);
1300 	*errp = error;
1301 	return (NULL);
1302 }
1303 
1304 static void
1305 mdinit(struct md_s *sc)
1306 {
1307 	struct g_geom *gp;
1308 	struct g_provider *pp;
1309 
1310 	g_topology_lock();
1311 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1312 	gp->softc = sc;
1313 	pp = g_new_providerf(gp, "md%d", sc->unit);
1314 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1315 	pp->mediasize = sc->mediasize;
1316 	pp->sectorsize = sc->sectorsize;
1317 	switch (sc->type) {
1318 	case MD_MALLOC:
1319 	case MD_VNODE:
1320 	case MD_SWAP:
1321 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1322 		break;
1323 	case MD_PRELOAD:
1324 	case MD_NULL:
1325 		break;
1326 	}
1327 	sc->gp = gp;
1328 	sc->pp = pp;
1329 	g_error_provider(pp, 0);
1330 	g_topology_unlock();
1331 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1332 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
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 
1377 static int
1378 mdsetcred(struct md_s *sc, struct ucred *cred)
1379 {
1380 	char *tmpbuf;
1381 	int error = 0;
1382 
1383 	/*
1384 	 * Set credits in our softc
1385 	 */
1386 
1387 	if (sc->cred)
1388 		crfree(sc->cred);
1389 	sc->cred = crhold(cred);
1390 
1391 	/*
1392 	 * Horrible kludge to establish credentials for NFS  XXX.
1393 	 */
1394 
1395 	if (sc->vnode) {
1396 		struct uio auio;
1397 		struct iovec aiov;
1398 
1399 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1400 		bzero(&auio, sizeof(auio));
1401 
1402 		aiov.iov_base = tmpbuf;
1403 		aiov.iov_len = sc->sectorsize;
1404 		auio.uio_iov = &aiov;
1405 		auio.uio_iovcnt = 1;
1406 		auio.uio_offset = 0;
1407 		auio.uio_rw = UIO_READ;
1408 		auio.uio_segflg = UIO_SYSSPACE;
1409 		auio.uio_resid = aiov.iov_len;
1410 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1411 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1412 		VOP_UNLOCK(sc->vnode, 0);
1413 		free(tmpbuf, M_TEMP);
1414 	}
1415 	return (error);
1416 }
1417 
1418 static int
1419 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1420 {
1421 	struct vattr vattr;
1422 	struct nameidata nd;
1423 	char *fname;
1424 	int error, flags;
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, td);
1443 	error = vn_open(&nd, &flags, 0, NULL);
1444 	if (error != 0)
1445 		return (error);
1446 	NDFREE(&nd, NDF_ONLY_PNBUF);
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 (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1455 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1456 		if (VN_IS_DOOMED(nd.ni_vp)) {
1457 			/* Forced unmount. */
1458 			error = EBADF;
1459 			goto bad;
1460 		}
1461 	}
1462 	nd.ni_vp->v_vflag |= VV_MD;
1463 	VOP_UNLOCK(nd.ni_vp, 0);
1464 
1465 	if (mdr->md_fwsectors != 0)
1466 		sc->fwsectors = mdr->md_fwsectors;
1467 	if (mdr->md_fwheads != 0)
1468 		sc->fwheads = mdr->md_fwheads;
1469 	snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1470 	    (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1471 	sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1472 	    MD_VERIFY);
1473 	if (!(flags & FWRITE))
1474 		sc->flags |= MD_READONLY;
1475 	sc->vnode = nd.ni_vp;
1476 
1477 	error = mdsetcred(sc, td->td_ucred);
1478 	if (error != 0) {
1479 		sc->vnode = NULL;
1480 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1481 		nd.ni_vp->v_vflag &= ~VV_MD;
1482 		goto bad;
1483 	}
1484 	return (0);
1485 bad:
1486 	VOP_UNLOCK(nd.ni_vp, 0);
1487 	(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1488 	return (error);
1489 }
1490 
1491 static void
1492 g_md_providergone(struct g_provider *pp)
1493 {
1494 	struct md_s *sc = pp->geom->softc;
1495 
1496 	mtx_lock(&sc->queue_mtx);
1497 	sc->flags |= MD_PROVIDERGONE;
1498 	wakeup(&sc->flags);
1499 	mtx_unlock(&sc->queue_mtx);
1500 }
1501 
1502 static int
1503 mddestroy(struct md_s *sc, struct thread *td)
1504 {
1505 
1506 	if (sc->gp) {
1507 		g_topology_lock();
1508 		g_wither_geom(sc->gp, ENXIO);
1509 		g_topology_unlock();
1510 
1511 		mtx_lock(&sc->queue_mtx);
1512 		while (!(sc->flags & MD_PROVIDERGONE))
1513 			msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1514 		mtx_unlock(&sc->queue_mtx);
1515 	}
1516 	if (sc->devstat) {
1517 		devstat_remove_entry(sc->devstat);
1518 		sc->devstat = NULL;
1519 	}
1520 	mtx_lock(&sc->queue_mtx);
1521 	sc->flags |= MD_SHUTDOWN;
1522 	wakeup(sc);
1523 	while (!(sc->flags & MD_EXITING))
1524 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1525 	mtx_unlock(&sc->queue_mtx);
1526 	mtx_destroy(&sc->stat_mtx);
1527 	mtx_destroy(&sc->queue_mtx);
1528 	if (sc->vnode != NULL) {
1529 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1530 		sc->vnode->v_vflag &= ~VV_MD;
1531 		VOP_UNLOCK(sc->vnode, 0);
1532 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1533 		    FREAD : (FREAD|FWRITE), sc->cred, td);
1534 	}
1535 	if (sc->cred != NULL)
1536 		crfree(sc->cred);
1537 	if (sc->object != NULL)
1538 		vm_object_deallocate(sc->object);
1539 	if (sc->indir)
1540 		destroy_indir(sc, sc->indir);
1541 	if (sc->uma)
1542 		uma_zdestroy(sc->uma);
1543 
1544 	LIST_REMOVE(sc, list);
1545 	free_unr(md_uh, sc->unit);
1546 	free(sc, M_MD);
1547 	return (0);
1548 }
1549 
1550 static int
1551 mdresize(struct md_s *sc, struct md_req *mdr)
1552 {
1553 	int error, res;
1554 	vm_pindex_t oldpages, newpages;
1555 
1556 	switch (sc->type) {
1557 	case MD_VNODE:
1558 	case MD_NULL:
1559 		break;
1560 	case MD_SWAP:
1561 		if (mdr->md_mediasize <= 0 ||
1562 		    (mdr->md_mediasize % PAGE_SIZE) != 0)
1563 			return (EDOM);
1564 		oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1565 		newpages = OFF_TO_IDX(round_page(mdr->md_mediasize));
1566 		if (newpages < oldpages) {
1567 			VM_OBJECT_WLOCK(sc->object);
1568 			vm_object_page_remove(sc->object, newpages, 0, 0);
1569 			swap_pager_freespace(sc->object, newpages,
1570 			    oldpages - newpages);
1571 			swap_release_by_cred(IDX_TO_OFF(oldpages -
1572 			    newpages), sc->cred);
1573 			sc->object->charge = IDX_TO_OFF(newpages);
1574 			sc->object->size = newpages;
1575 			VM_OBJECT_WUNLOCK(sc->object);
1576 		} else if (newpages > oldpages) {
1577 			res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1578 			    oldpages), sc->cred);
1579 			if (!res)
1580 				return (ENOMEM);
1581 			if ((mdr->md_options & MD_RESERVE) ||
1582 			    (sc->flags & MD_RESERVE)) {
1583 				error = swap_pager_reserve(sc->object,
1584 				    oldpages, newpages - oldpages);
1585 				if (error < 0) {
1586 					swap_release_by_cred(
1587 					    IDX_TO_OFF(newpages - oldpages),
1588 					    sc->cred);
1589 					return (EDOM);
1590 				}
1591 			}
1592 			VM_OBJECT_WLOCK(sc->object);
1593 			sc->object->charge = IDX_TO_OFF(newpages);
1594 			sc->object->size = newpages;
1595 			VM_OBJECT_WUNLOCK(sc->object);
1596 		}
1597 		break;
1598 	default:
1599 		return (EOPNOTSUPP);
1600 	}
1601 
1602 	sc->mediasize = mdr->md_mediasize;
1603 	g_topology_lock();
1604 	g_resize_provider(sc->pp, sc->mediasize);
1605 	g_topology_unlock();
1606 	return (0);
1607 }
1608 
1609 static int
1610 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1611 {
1612 	vm_ooffset_t npage;
1613 	int error;
1614 
1615 	/*
1616 	 * Range check.  Disallow negative sizes and sizes not being
1617 	 * multiple of page size.
1618 	 */
1619 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1620 		return (EDOM);
1621 
1622 	/*
1623 	 * Allocate an OBJT_SWAP object.
1624 	 *
1625 	 * Note the truncation.
1626 	 */
1627 
1628 	if ((mdr->md_options & MD_VERIFY) != 0)
1629 		return (EINVAL);
1630 	npage = mdr->md_mediasize / PAGE_SIZE;
1631 	if (mdr->md_fwsectors != 0)
1632 		sc->fwsectors = mdr->md_fwsectors;
1633 	if (mdr->md_fwheads != 0)
1634 		sc->fwheads = mdr->md_fwheads;
1635 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1636 	    VM_PROT_DEFAULT, 0, td->td_ucred);
1637 	if (sc->object == NULL)
1638 		return (ENOMEM);
1639 	sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1640 	if (mdr->md_options & MD_RESERVE) {
1641 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1642 			error = EDOM;
1643 			goto finish;
1644 		}
1645 	}
1646 	error = mdsetcred(sc, td->td_ucred);
1647  finish:
1648 	if (error != 0) {
1649 		vm_object_deallocate(sc->object);
1650 		sc->object = NULL;
1651 	}
1652 	return (error);
1653 }
1654 
1655 static int
1656 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1657 {
1658 
1659 	/*
1660 	 * Range check.  Disallow negative sizes and sizes not being
1661 	 * multiple of page size.
1662 	 */
1663 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1664 		return (EDOM);
1665 
1666 	return (0);
1667 }
1668 
1669 static int
1670 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1671 {
1672 	struct md_s *sc;
1673 	unsigned sectsize;
1674 	int error, i;
1675 
1676 	sx_assert(&md_sx, SA_XLOCKED);
1677 
1678 	switch (mdr->md_type) {
1679 	case MD_MALLOC:
1680 	case MD_PRELOAD:
1681 	case MD_VNODE:
1682 	case MD_SWAP:
1683 	case MD_NULL:
1684 		break;
1685 	default:
1686 		return (EINVAL);
1687 	}
1688 	if (mdr->md_sectorsize == 0)
1689 		sectsize = DEV_BSIZE;
1690 	else
1691 		sectsize = mdr->md_sectorsize;
1692 	if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize)
1693 		return (EINVAL);
1694 	if (mdr->md_options & MD_AUTOUNIT)
1695 		sc = mdnew(-1, &error, mdr->md_type);
1696 	else {
1697 		if (mdr->md_unit > INT_MAX)
1698 			return (EINVAL);
1699 		sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1700 	}
1701 	if (sc == NULL)
1702 		return (error);
1703 	if (mdr->md_label != NULL)
1704 		error = copyinstr(mdr->md_label, sc->label,
1705 		    sizeof(sc->label), NULL);
1706 	if (error != 0)
1707 		goto err_after_new;
1708 	if (mdr->md_options & MD_AUTOUNIT)
1709 		mdr->md_unit = sc->unit;
1710 	sc->mediasize = mdr->md_mediasize;
1711 	sc->sectorsize = sectsize;
1712 	error = EDOOFUS;
1713 	switch (sc->type) {
1714 	case MD_MALLOC:
1715 		sc->start = mdstart_malloc;
1716 		error = mdcreate_malloc(sc, mdr);
1717 		break;
1718 	case MD_PRELOAD:
1719 		/*
1720 		 * We disallow attaching preloaded memory disks via
1721 		 * ioctl. Preloaded memory disks are automatically
1722 		 * attached in g_md_init().
1723 		 */
1724 		error = EOPNOTSUPP;
1725 		break;
1726 	case MD_VNODE:
1727 		sc->start = mdstart_vnode;
1728 		error = mdcreate_vnode(sc, mdr, td);
1729 		break;
1730 	case MD_SWAP:
1731 		sc->start = mdstart_swap;
1732 		error = mdcreate_swap(sc, mdr, td);
1733 		break;
1734 	case MD_NULL:
1735 		sc->start = mdstart_null;
1736 		error = mdcreate_null(sc, mdr, td);
1737 		break;
1738 	}
1739 err_after_new:
1740 	if (error != 0) {
1741 		mddestroy(sc, td);
1742 		return (error);
1743 	}
1744 
1745 	/* Prune off any residual fractional sector */
1746 	i = sc->mediasize % sc->sectorsize;
1747 	sc->mediasize -= i;
1748 
1749 	mdinit(sc);
1750 	return (0);
1751 }
1752 
1753 static int
1754 kern_mdattach(struct thread *td, struct md_req *mdr)
1755 {
1756 	int error;
1757 
1758 	sx_xlock(&md_sx);
1759 	error = kern_mdattach_locked(td, mdr);
1760 	sx_xunlock(&md_sx);
1761 	return (error);
1762 }
1763 
1764 static int
1765 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1766 {
1767 	struct md_s *sc;
1768 
1769 	sx_assert(&md_sx, SA_XLOCKED);
1770 
1771 	if (mdr->md_mediasize != 0 ||
1772 	    (mdr->md_options & ~MD_FORCE) != 0)
1773 		return (EINVAL);
1774 
1775 	sc = mdfind(mdr->md_unit);
1776 	if (sc == NULL)
1777 		return (ENOENT);
1778 	if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1779 	    !(mdr->md_options & MD_FORCE))
1780 		return (EBUSY);
1781 	return (mddestroy(sc, td));
1782 }
1783 
1784 static int
1785 kern_mddetach(struct thread *td, struct md_req *mdr)
1786 {
1787 	int error;
1788 
1789 	sx_xlock(&md_sx);
1790 	error = kern_mddetach_locked(td, mdr);
1791 	sx_xunlock(&md_sx);
1792 	return (error);
1793 }
1794 
1795 static int
1796 kern_mdresize_locked(struct md_req *mdr)
1797 {
1798 	struct md_s *sc;
1799 
1800 	sx_assert(&md_sx, SA_XLOCKED);
1801 
1802 	if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1803 		return (EINVAL);
1804 
1805 	sc = mdfind(mdr->md_unit);
1806 	if (sc == NULL)
1807 		return (ENOENT);
1808 	if (mdr->md_mediasize < sc->sectorsize)
1809 		return (EINVAL);
1810 	if (mdr->md_mediasize < sc->mediasize &&
1811 	    !(sc->flags & MD_FORCE) &&
1812 	    !(mdr->md_options & MD_FORCE))
1813 		return (EBUSY);
1814 	return (mdresize(sc, mdr));
1815 }
1816 
1817 static int
1818 kern_mdresize(struct md_req *mdr)
1819 {
1820 	int error;
1821 
1822 	sx_xlock(&md_sx);
1823 	error = kern_mdresize_locked(mdr);
1824 	sx_xunlock(&md_sx);
1825 	return (error);
1826 }
1827 
1828 static int
1829 kern_mdquery_locked(struct md_req *mdr)
1830 {
1831 	struct md_s *sc;
1832 	int error;
1833 
1834 	sx_assert(&md_sx, SA_XLOCKED);
1835 
1836 	sc = mdfind(mdr->md_unit);
1837 	if (sc == NULL)
1838 		return (ENOENT);
1839 	mdr->md_type = sc->type;
1840 	mdr->md_options = sc->flags;
1841 	mdr->md_mediasize = sc->mediasize;
1842 	mdr->md_sectorsize = sc->sectorsize;
1843 	error = 0;
1844 	if (mdr->md_label != NULL) {
1845 		error = copyout(sc->label, mdr->md_label,
1846 		    strlen(sc->label) + 1);
1847 		if (error != 0)
1848 			return (error);
1849 	}
1850 	if (sc->type == MD_VNODE ||
1851 	    (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1852 		error = copyout(sc->file, mdr->md_file,
1853 		    strlen(sc->file) + 1);
1854 	return (error);
1855 }
1856 
1857 static int
1858 kern_mdquery(struct md_req *mdr)
1859 {
1860 	int error;
1861 
1862 	sx_xlock(&md_sx);
1863 	error = kern_mdquery_locked(mdr);
1864 	sx_xunlock(&md_sx);
1865 	return (error);
1866 }
1867 
1868 /* Copy members that are not userspace pointers. */
1869 #define	MD_IOCTL2REQ(mdio, mdr) do {					\
1870 	(mdr)->md_unit = (mdio)->md_unit;				\
1871 	(mdr)->md_type = (mdio)->md_type;				\
1872 	(mdr)->md_mediasize = (mdio)->md_mediasize;			\
1873 	(mdr)->md_sectorsize = (mdio)->md_sectorsize;			\
1874 	(mdr)->md_options = (mdio)->md_options;				\
1875 	(mdr)->md_fwheads = (mdio)->md_fwheads;				\
1876 	(mdr)->md_fwsectors = (mdio)->md_fwsectors;			\
1877 	(mdr)->md_units = &(mdio)->md_pad[0];				\
1878 	(mdr)->md_units_nitems = nitems((mdio)->md_pad);		\
1879 } while(0)
1880 
1881 /* Copy members that might have been updated */
1882 #define MD_REQ2IOCTL(mdr, mdio) do {					\
1883 	(mdio)->md_unit = (mdr)->md_unit;				\
1884 	(mdio)->md_type = (mdr)->md_type;				\
1885 	(mdio)->md_mediasize = (mdr)->md_mediasize;			\
1886 	(mdio)->md_sectorsize = (mdr)->md_sectorsize;			\
1887 	(mdio)->md_options = (mdr)->md_options;				\
1888 	(mdio)->md_fwheads = (mdr)->md_fwheads;				\
1889 	(mdio)->md_fwsectors = (mdr)->md_fwsectors;			\
1890 } while(0)
1891 
1892 static int
1893 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1894     struct thread *td)
1895 {
1896 	struct md_req mdr;
1897 	int error;
1898 
1899 	if (md_debug)
1900 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1901 			devtoname(dev), cmd, addr, flags, td);
1902 
1903 	bzero(&mdr, sizeof(mdr));
1904 	switch (cmd) {
1905 	case MDIOCATTACH:
1906 	case MDIOCDETACH:
1907 	case MDIOCRESIZE:
1908 	case MDIOCQUERY: {
1909 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1910 		if (mdio->md_version != MDIOVERSION)
1911 			return (EINVAL);
1912 		MD_IOCTL2REQ(mdio, &mdr);
1913 		mdr.md_file = mdio->md_file;
1914 		mdr.md_file_seg = UIO_USERSPACE;
1915 		/* If the file is adjacent to the md_ioctl it's in kernel. */
1916 		if ((void *)mdio->md_file == (void *)(mdio + 1))
1917 			mdr.md_file_seg = UIO_SYSSPACE;
1918 		mdr.md_label = mdio->md_label;
1919 		break;
1920 	}
1921 #ifdef COMPAT_FREEBSD32
1922 	case MDIOCATTACH_32:
1923 	case MDIOCDETACH_32:
1924 	case MDIOCRESIZE_32:
1925 	case MDIOCQUERY_32: {
1926 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1927 		if (mdio->md_version != MDIOVERSION)
1928 			return (EINVAL);
1929 		MD_IOCTL2REQ(mdio, &mdr);
1930 		mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1931 		mdr.md_file_seg = UIO_USERSPACE;
1932 		mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1933 		break;
1934 	}
1935 #endif
1936 	default:
1937 		/* Fall through to handler switch. */
1938 		break;
1939 	}
1940 
1941 	error = 0;
1942 	switch (cmd) {
1943 	case MDIOCATTACH:
1944 #ifdef COMPAT_FREEBSD32
1945 	case MDIOCATTACH_32:
1946 #endif
1947 		error = kern_mdattach(td, &mdr);
1948 		break;
1949 	case MDIOCDETACH:
1950 #ifdef COMPAT_FREEBSD32
1951 	case MDIOCDETACH_32:
1952 #endif
1953 		error = kern_mddetach(td, &mdr);
1954 		break;
1955 	case MDIOCRESIZE:
1956 #ifdef COMPAT_FREEBSD32
1957 	case MDIOCRESIZE_32:
1958 #endif
1959 		error = kern_mdresize(&mdr);
1960 		break;
1961 	case MDIOCQUERY:
1962 #ifdef COMPAT_FREEBSD32
1963 	case MDIOCQUERY_32:
1964 #endif
1965 		error = kern_mdquery(&mdr);
1966 		break;
1967 	default:
1968 		error = ENOIOCTL;
1969 	}
1970 
1971 	switch (cmd) {
1972 	case MDIOCATTACH:
1973 	case MDIOCQUERY: {
1974 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1975 		MD_REQ2IOCTL(&mdr, mdio);
1976 		break;
1977 	}
1978 #ifdef COMPAT_FREEBSD32
1979 	case MDIOCATTACH_32:
1980 	case MDIOCQUERY_32: {
1981 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1982 		MD_REQ2IOCTL(&mdr, mdio);
1983 		break;
1984 	}
1985 #endif
1986 	default:
1987 		/* Other commands to not alter mdr. */
1988 		break;
1989 	}
1990 
1991 	return (error);
1992 }
1993 
1994 static void
1995 md_preloaded(u_char *image, size_t length, const char *name)
1996 {
1997 	struct md_s *sc;
1998 	int error;
1999 
2000 	sc = mdnew(-1, &error, MD_PRELOAD);
2001 	if (sc == NULL)
2002 		return;
2003 	sc->mediasize = length;
2004 	sc->sectorsize = DEV_BSIZE;
2005 	sc->pl_ptr = image;
2006 	sc->pl_len = length;
2007 	sc->start = mdstart_preload;
2008 	if (name != NULL)
2009 		strlcpy(sc->file, name, sizeof(sc->file));
2010 #ifdef MD_ROOT
2011 	if (sc->unit == 0) {
2012 #ifndef ROOTDEVNAME
2013 		rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2014 #endif
2015 #ifdef MD_ROOT_READONLY
2016 		sc->flags |= MD_READONLY;
2017 #endif
2018 	}
2019 #endif
2020 	mdinit(sc);
2021 	if (name != NULL) {
2022 		printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2023 		    MD_NAME, sc->unit, name, length, image);
2024 	} else {
2025 		printf("%s%d: Embedded image %zd bytes at %p\n",
2026 		    MD_NAME, sc->unit, length, image);
2027 	}
2028 }
2029 
2030 static void
2031 g_md_init(struct g_class *mp __unused)
2032 {
2033 	caddr_t mod;
2034 	u_char *ptr, *name, *type;
2035 	unsigned len;
2036 	int i;
2037 
2038 	/* figure out log2(NINDIR) */
2039 	for (i = NINDIR, nshift = -1; i; nshift++)
2040 		i >>= 1;
2041 
2042 	mod = NULL;
2043 	sx_init(&md_sx, "MD config lock");
2044 	g_topology_unlock();
2045 	md_uh = new_unrhdr(0, INT_MAX, NULL);
2046 #ifdef MD_ROOT
2047 	if (mfs_root_size != 0) {
2048 		sx_xlock(&md_sx);
2049 #ifdef MD_ROOT_MEM
2050 		md_preloaded(mfs_root, mfs_root_size, NULL);
2051 #else
2052 		md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2053 		    NULL);
2054 #endif
2055 		sx_xunlock(&md_sx);
2056 	}
2057 #endif
2058 	/* XXX: are preload_* static or do they need Giant ? */
2059 	while ((mod = preload_search_next_name(mod)) != NULL) {
2060 		name = (char *)preload_search_info(mod, MODINFO_NAME);
2061 		if (name == NULL)
2062 			continue;
2063 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
2064 		if (type == NULL)
2065 			continue;
2066 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2067 			continue;
2068 		ptr = preload_fetch_addr(mod);
2069 		len = preload_fetch_size(mod);
2070 		if (ptr != NULL && len != 0) {
2071 			sx_xlock(&md_sx);
2072 			md_preloaded(ptr, len, name);
2073 			sx_xunlock(&md_sx);
2074 		}
2075 	}
2076 	md_pbuf_zone = pbuf_zsecond_create("mdpbuf", nswbuf / 10);
2077 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2078 	    0600, MDCTL_NAME);
2079 	g_topology_lock();
2080 }
2081 
2082 static void
2083 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2084     struct g_consumer *cp __unused, struct g_provider *pp)
2085 {
2086 	struct md_s *mp;
2087 	char *type;
2088 
2089 	mp = gp->softc;
2090 	if (mp == NULL)
2091 		return;
2092 
2093 	switch (mp->type) {
2094 	case MD_MALLOC:
2095 		type = "malloc";
2096 		break;
2097 	case MD_PRELOAD:
2098 		type = "preload";
2099 		break;
2100 	case MD_VNODE:
2101 		type = "vnode";
2102 		break;
2103 	case MD_SWAP:
2104 		type = "swap";
2105 		break;
2106 	case MD_NULL:
2107 		type = "null";
2108 		break;
2109 	default:
2110 		type = "unknown";
2111 		break;
2112 	}
2113 
2114 	if (pp != NULL) {
2115 		if (indent == NULL) {
2116 			sbuf_printf(sb, " u %d", mp->unit);
2117 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2118 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2119 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2120 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2121 			sbuf_printf(sb, " t %s", type);
2122 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2123 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2124 				sbuf_printf(sb, " file %s", mp->file);
2125 			sbuf_printf(sb, " label %s", mp->label);
2126 		} else {
2127 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2128 			    mp->unit);
2129 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2130 			    indent, (uintmax_t) mp->sectorsize);
2131 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2132 			    indent, (uintmax_t) mp->fwheads);
2133 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2134 			    indent, (uintmax_t) mp->fwsectors);
2135 			if (mp->ident[0] != '\0') {
2136 				sbuf_printf(sb, "%s<ident>", indent);
2137 				g_conf_printf_escaped(sb, "%s", mp->ident);
2138 				sbuf_printf(sb, "</ident>\n");
2139 			}
2140 			sbuf_printf(sb, "%s<length>%ju</length>\n",
2141 			    indent, (uintmax_t) mp->mediasize);
2142 			sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2143 			    (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2144 			sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2145 			    (mp->flags & MD_READONLY) == 0 ? "read-write":
2146 			    "read-only");
2147 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2148 			    type);
2149 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2150 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2151 				sbuf_printf(sb, "%s<file>", indent);
2152 				g_conf_printf_escaped(sb, "%s", mp->file);
2153 				sbuf_printf(sb, "</file>\n");
2154 			}
2155 			if (mp->type == MD_VNODE)
2156 				sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2157 				    (mp->flags & MD_CACHE) == 0 ? "off": "on");
2158 			sbuf_printf(sb, "%s<label>", indent);
2159 			g_conf_printf_escaped(sb, "%s", mp->label);
2160 			sbuf_printf(sb, "</label>\n");
2161 		}
2162 	}
2163 }
2164 
2165 static void
2166 g_md_fini(struct g_class *mp __unused)
2167 {
2168 
2169 	sx_destroy(&md_sx);
2170 	if (status_dev != NULL)
2171 		destroy_dev(status_dev);
2172 	uma_zdestroy(md_pbuf_zone);
2173 	delete_unrhdr(md_uh);
2174 }
2175