xref: /freebsd/sys/dev/md/md.c (revision 7937bfbc0ca53fe7cdd0d54414f9296e273a518e)
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  * From: src/sys/dev/vn/vn.c,v 1.122 2000/12/16 16:06:03
57  */
58 
59 #include "opt_rootdevname.h"
60 #include "opt_geom.h"
61 #include "opt_md.h"
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/bio.h>
66 #include <sys/buf.h>
67 #include <sys/conf.h>
68 #include <sys/devicestat.h>
69 #include <sys/fcntl.h>
70 #include <sys/kernel.h>
71 #include <sys/kthread.h>
72 #include <sys/limits.h>
73 #include <sys/linker.h>
74 #include <sys/lock.h>
75 #include <sys/malloc.h>
76 #include <sys/mdioctl.h>
77 #include <sys/mount.h>
78 #include <sys/mutex.h>
79 #include <sys/sx.h>
80 #include <sys/namei.h>
81 #include <sys/proc.h>
82 #include <sys/queue.h>
83 #include <sys/rwlock.h>
84 #include <sys/sbuf.h>
85 #include <sys/sched.h>
86 #include <sys/sf_buf.h>
87 #include <sys/sysctl.h>
88 #include <sys/uio.h>
89 #include <sys/unistd.h>
90 #include <sys/vnode.h>
91 #include <sys/disk.h>
92 
93 #include <geom/geom.h>
94 #include <geom/geom_int.h>
95 
96 #include <vm/vm.h>
97 #include <vm/vm_extern.h>
98 #include <vm/vm_param.h>
99 #include <vm/vm_object.h>
100 #include <vm/vm_page.h>
101 #include <vm/vm_pager.h>
102 #include <vm/swap_pager.h>
103 #include <vm/uma.h>
104 
105 #include <machine/bus.h>
106 
107 #define MD_MODVER 1
108 
109 #define MD_SHUTDOWN	0x10000		/* Tell worker thread to terminate. */
110 #define	MD_EXITING	0x20000		/* Worker thread is exiting. */
111 #define MD_PROVIDERGONE	0x40000		/* Safe to free the softc */
112 
113 #ifndef MD_NSECT
114 #define MD_NSECT (10000 * 2)
115 #endif
116 
117 struct md_req {
118 	unsigned	md_unit;	/* unit number */
119 	enum md_types	md_type;	/* type of disk */
120 	off_t		md_mediasize;	/* size of disk in bytes */
121 	unsigned	md_sectorsize;	/* sectorsize */
122 	unsigned	md_options;	/* options */
123 	int		md_fwheads;	/* firmware heads */
124 	int		md_fwsectors;	/* firmware sectors */
125 	char		*md_file;	/* pathname of file to mount */
126 	enum uio_seg	md_file_seg;	/* location of md_file */
127 	char		*md_label;	/* label of the device (userspace) */
128 	int		*md_units;	/* pointer to units array (kernel) */
129 	size_t		md_units_nitems; /* items in md_units array */
130 };
131 
132 #ifdef COMPAT_FREEBSD32
133 struct md_ioctl32 {
134 	unsigned	md_version;
135 	unsigned	md_unit;
136 	enum md_types	md_type;
137 	uint32_t	md_file;
138 	off_t		md_mediasize;
139 	unsigned	md_sectorsize;
140 	unsigned	md_options;
141 	uint64_t	md_base;
142 	int		md_fwheads;
143 	int		md_fwsectors;
144 	uint32_t	md_label;
145 	int		md_pad[MDNPAD];
146 }
147 #ifdef __amd64__
148 __attribute__((__packed__))
149 #endif
150 ;
151 #ifndef __amd64__
152 CTASSERT((sizeof(struct md_ioctl32)) == 440);
153 #else
154 CTASSERT((sizeof(struct md_ioctl32)) == 436);
155 #endif
156 
157 #define	MDIOCATTACH_32	_IOC_NEWTYPE(MDIOCATTACH, struct md_ioctl32)
158 #define	MDIOCDETACH_32	_IOC_NEWTYPE(MDIOCDETACH, struct md_ioctl32)
159 #define	MDIOCQUERY_32	_IOC_NEWTYPE(MDIOCQUERY, struct md_ioctl32)
160 #define	MDIOCRESIZE_32	_IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
161 #endif /* COMPAT_FREEBSD32 */
162 
163 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
164 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
165 
166 static int md_debug;
167 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
168     "Enable md(4) debug messages");
169 static int md_malloc_wait;
170 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
171     "Allow malloc to wait for memory allocations");
172 
173 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
174 #define	MD_ROOT_FSTYPE	"ufs"
175 #endif
176 
177 #if defined(MD_ROOT)
178 /*
179  * Preloaded image gets put here.
180  */
181 #if defined(MD_ROOT_SIZE)
182 /*
183  * We put the mfs_root symbol into the oldmfs section of the kernel object file.
184  * Applications that patch the object with the image can determine
185  * the size looking at the oldmfs section size within the kernel.
186  */
187 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
188 const int mfs_root_size = sizeof(mfs_root);
189 #elif defined(MD_ROOT_MEM)
190 /* MD region already mapped in the memory */
191 u_char *mfs_root;
192 int mfs_root_size;
193 #else
194 extern volatile u_char __weak_symbol mfs_root;
195 extern volatile u_char __weak_symbol mfs_root_end;
196 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
197 #endif
198 #endif
199 
200 static g_init_t g_md_init;
201 static g_fini_t g_md_fini;
202 static g_start_t g_md_start;
203 static g_access_t g_md_access;
204 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
205     struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
206 static g_provgone_t g_md_providergone;
207 
208 static struct cdev *status_dev = NULL;
209 static struct sx md_sx;
210 static struct unrhdr *md_uh;
211 
212 static d_ioctl_t mdctlioctl;
213 
214 static struct cdevsw mdctl_cdevsw = {
215 	.d_version =	D_VERSION,
216 	.d_ioctl =	mdctlioctl,
217 	.d_name =	MD_NAME,
218 };
219 
220 struct g_class g_md_class = {
221 	.name = "MD",
222 	.version = G_VERSION,
223 	.init = g_md_init,
224 	.fini = g_md_fini,
225 	.start = g_md_start,
226 	.access = g_md_access,
227 	.dumpconf = g_md_dumpconf,
228 	.providergone = g_md_providergone,
229 };
230 
231 DECLARE_GEOM_CLASS(g_md_class, g_md);
232 MODULE_VERSION(geom_md, 0);
233 
234 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
235 
236 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
237 #define NMASK	(NINDIR-1)
238 static int nshift;
239 
240 struct indir {
241 	uintptr_t	*array;
242 	u_int		total;
243 	u_int		used;
244 	u_int		shift;
245 };
246 
247 struct md_s {
248 	int unit;
249 	LIST_ENTRY(md_s) list;
250 	struct bio_queue_head bio_queue;
251 	struct mtx queue_mtx;
252 	struct cdev *dev;
253 	enum md_types type;
254 	off_t mediasize;
255 	unsigned sectorsize;
256 	unsigned opencount;
257 	unsigned fwheads;
258 	unsigned fwsectors;
259 	char ident[32];
260 	unsigned flags;
261 	char name[20];
262 	struct proc *procp;
263 	struct g_geom *gp;
264 	struct g_provider *pp;
265 	int (*start)(struct md_s *sc, struct bio *bp);
266 	struct devstat *devstat;
267 	bool candelete;
268 
269 	/* MD_MALLOC related fields */
270 	struct indir *indir;
271 	uma_zone_t uma;
272 
273 	/* MD_PRELOAD related fields */
274 	u_char *pl_ptr;
275 	size_t pl_len;
276 
277 	/* MD_VNODE related fields */
278 	struct vnode *vnode;
279 	char file[PATH_MAX];
280 	char label[PATH_MAX];
281 	struct ucred *cred;
282 	vm_offset_t kva;
283 
284 	/* MD_SWAP related fields */
285 	vm_object_t object;
286 };
287 
288 static struct indir *
289 new_indir(u_int shift)
290 {
291 	struct indir *ip;
292 
293 	ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
294 	    | M_ZERO);
295 	if (ip == NULL)
296 		return (NULL);
297 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
298 	    M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
299 	if (ip->array == NULL) {
300 		free(ip, M_MD);
301 		return (NULL);
302 	}
303 	ip->total = NINDIR;
304 	ip->shift = shift;
305 	return (ip);
306 }
307 
308 static void
309 del_indir(struct indir *ip)
310 {
311 
312 	free(ip->array, M_MDSECT);
313 	free(ip, M_MD);
314 }
315 
316 static void
317 destroy_indir(struct md_s *sc, struct indir *ip)
318 {
319 	int i;
320 
321 	for (i = 0; i < NINDIR; i++) {
322 		if (!ip->array[i])
323 			continue;
324 		if (ip->shift)
325 			destroy_indir(sc, (struct indir*)(ip->array[i]));
326 		else if (ip->array[i] > 255)
327 			uma_zfree(sc->uma, (void *)(ip->array[i]));
328 	}
329 	del_indir(ip);
330 }
331 
332 /*
333  * This function does the math and allocates the top level "indir" structure
334  * for a device of "size" sectors.
335  */
336 
337 static struct indir *
338 dimension(off_t size)
339 {
340 	off_t rcnt;
341 	struct indir *ip;
342 	int layer;
343 
344 	rcnt = size;
345 	layer = 0;
346 	while (rcnt > NINDIR) {
347 		rcnt /= NINDIR;
348 		layer++;
349 	}
350 
351 	/*
352 	 * XXX: the top layer is probably not fully populated, so we allocate
353 	 * too much space for ip->array in here.
354 	 */
355 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
356 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
357 	    M_MDSECT, M_WAITOK | M_ZERO);
358 	ip->total = NINDIR;
359 	ip->shift = layer * nshift;
360 	return (ip);
361 }
362 
363 /*
364  * Read a given sector
365  */
366 
367 static uintptr_t
368 s_read(struct indir *ip, off_t offset)
369 {
370 	struct indir *cip;
371 	int idx;
372 	uintptr_t up;
373 
374 	if (md_debug > 1)
375 		printf("s_read(%jd)\n", (intmax_t)offset);
376 	up = 0;
377 	for (cip = ip; cip != NULL;) {
378 		if (cip->shift) {
379 			idx = (offset >> cip->shift) & NMASK;
380 			up = cip->array[idx];
381 			cip = (struct indir *)up;
382 			continue;
383 		}
384 		idx = offset & NMASK;
385 		return (cip->array[idx]);
386 	}
387 	return (0);
388 }
389 
390 /*
391  * Write a given sector, prune the tree if the value is 0
392  */
393 
394 static int
395 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
396 {
397 	struct indir *cip, *lip[10];
398 	int idx, li;
399 	uintptr_t up;
400 
401 	if (md_debug > 1)
402 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
403 	up = 0;
404 	li = 0;
405 	cip = ip;
406 	for (;;) {
407 		lip[li++] = cip;
408 		if (cip->shift) {
409 			idx = (offset >> cip->shift) & NMASK;
410 			up = cip->array[idx];
411 			if (up != 0) {
412 				cip = (struct indir *)up;
413 				continue;
414 			}
415 			/* Allocate branch */
416 			cip->array[idx] =
417 			    (uintptr_t)new_indir(cip->shift - nshift);
418 			if (cip->array[idx] == 0)
419 				return (ENOSPC);
420 			cip->used++;
421 			up = cip->array[idx];
422 			cip = (struct indir *)up;
423 			continue;
424 		}
425 		/* leafnode */
426 		idx = offset & NMASK;
427 		up = cip->array[idx];
428 		if (up != 0)
429 			cip->used--;
430 		cip->array[idx] = ptr;
431 		if (ptr != 0)
432 			cip->used++;
433 		break;
434 	}
435 	if (cip->used != 0 || li == 1)
436 		return (0);
437 	li--;
438 	while (cip->used == 0 && cip != ip) {
439 		li--;
440 		idx = (offset >> lip[li]->shift) & NMASK;
441 		up = lip[li]->array[idx];
442 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
443 		del_indir(cip);
444 		lip[li]->array[idx] = 0;
445 		lip[li]->used--;
446 		cip = lip[li];
447 	}
448 	return (0);
449 }
450 
451 static int
452 g_md_access(struct g_provider *pp, int r, int w, int e)
453 {
454 	struct md_s *sc;
455 
456 	sc = pp->geom->softc;
457 	if (sc == NULL) {
458 		if (r <= 0 && w <= 0 && e <= 0)
459 			return (0);
460 		return (ENXIO);
461 	}
462 	r += pp->acr;
463 	w += pp->acw;
464 	e += pp->ace;
465 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
466 		return (EROFS);
467 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
468 		sc->opencount = 1;
469 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
470 		sc->opencount = 0;
471 	}
472 	return (0);
473 }
474 
475 static void
476 g_md_start(struct bio *bp)
477 {
478 	struct md_s *sc;
479 
480 	sc = bp->bio_to->geom->softc;
481 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
482 		devstat_start_transaction_bio(sc->devstat, bp);
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 	case BIO_FLUSH:
654 		return (0);
655 	default:
656 		return (EOPNOTSUPP);
657 	}
658 
659 	notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
660 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
661 	    (bus_dma_segment_t *)bp->bio_data : NULL;
662 	if (notmapped) {
663 		m = bp->bio_ma;
664 		ma_offs = bp->bio_ma_offset;
665 		dst = NULL;
666 		KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
667 	} else if (vlist != NULL) {
668 		ma_offs = bp->bio_ma_offset;
669 		dst = NULL;
670 	} else {
671 		dst = bp->bio_data;
672 	}
673 
674 	nsec = bp->bio_length / sc->sectorsize;
675 	secno = bp->bio_offset / sc->sectorsize;
676 	error = 0;
677 	while (nsec--) {
678 		osp = s_read(sc->indir, secno);
679 		if (bp->bio_cmd == BIO_DELETE) {
680 			if (osp != 0)
681 				error = s_write(sc->indir, secno, 0);
682 		} else if (bp->bio_cmd == BIO_READ) {
683 			if (osp == 0) {
684 				if (notmapped) {
685 					error = md_malloc_move_ma(&m, &ma_offs,
686 					    sc->sectorsize, NULL, 0,
687 					    MD_MALLOC_MOVE_ZERO);
688 				} else if (vlist != NULL) {
689 					error = md_malloc_move_vlist(&vlist,
690 					    &ma_offs, sc->sectorsize, NULL, 0,
691 					    MD_MALLOC_MOVE_ZERO);
692 				} else
693 					bzero(dst, sc->sectorsize);
694 			} else if (osp <= 255) {
695 				if (notmapped) {
696 					error = md_malloc_move_ma(&m, &ma_offs,
697 					    sc->sectorsize, NULL, osp,
698 					    MD_MALLOC_MOVE_FILL);
699 				} else if (vlist != NULL) {
700 					error = md_malloc_move_vlist(&vlist,
701 					    &ma_offs, sc->sectorsize, NULL, osp,
702 					    MD_MALLOC_MOVE_FILL);
703 				} else
704 					memset(dst, osp, sc->sectorsize);
705 			} else {
706 				if (notmapped) {
707 					error = md_malloc_move_ma(&m, &ma_offs,
708 					    sc->sectorsize, (void *)osp, 0,
709 					    MD_MALLOC_MOVE_READ);
710 				} else if (vlist != NULL) {
711 					error = md_malloc_move_vlist(&vlist,
712 					    &ma_offs, sc->sectorsize,
713 					    (void *)osp, 0,
714 					    MD_MALLOC_MOVE_READ);
715 				} else {
716 					bcopy((void *)osp, dst, sc->sectorsize);
717 					cpu_flush_dcache(dst, sc->sectorsize);
718 				}
719 			}
720 			osp = 0;
721 		} else if (bp->bio_cmd == BIO_WRITE) {
722 			if (sc->flags & MD_COMPRESS) {
723 				if (notmapped) {
724 					error1 = md_malloc_move_ma(&m, &ma_offs,
725 					    sc->sectorsize, &uc, 0,
726 					    MD_MALLOC_MOVE_CMP);
727 					i = error1 == 0 ? sc->sectorsize : 0;
728 				} else if (vlist != NULL) {
729 					error1 = md_malloc_move_vlist(&vlist,
730 					    &ma_offs, sc->sectorsize, &uc, 0,
731 					    MD_MALLOC_MOVE_CMP);
732 					i = error1 == 0 ? sc->sectorsize : 0;
733 				} else {
734 					uc = dst[0];
735 					for (i = 1; i < sc->sectorsize; i++) {
736 						if (dst[i] != uc)
737 							break;
738 					}
739 				}
740 			} else {
741 				i = 0;
742 				uc = 0;
743 			}
744 			if (i == sc->sectorsize) {
745 				if (osp != uc)
746 					error = s_write(sc->indir, secno, uc);
747 			} else {
748 				if (osp <= 255) {
749 					sp = (uintptr_t)uma_zalloc(sc->uma,
750 					    md_malloc_wait ? M_WAITOK :
751 					    M_NOWAIT);
752 					if (sp == 0) {
753 						error = ENOSPC;
754 						break;
755 					}
756 					if (notmapped) {
757 						error = md_malloc_move_ma(&m,
758 						    &ma_offs, sc->sectorsize,
759 						    (void *)sp, 0,
760 						    MD_MALLOC_MOVE_WRITE);
761 					} else if (vlist != NULL) {
762 						error = md_malloc_move_vlist(
763 						    &vlist, &ma_offs,
764 						    sc->sectorsize, (void *)sp,
765 						    0, MD_MALLOC_MOVE_WRITE);
766 					} else {
767 						bcopy(dst, (void *)sp,
768 						    sc->sectorsize);
769 					}
770 					error = s_write(sc->indir, secno, sp);
771 				} else {
772 					if (notmapped) {
773 						error = md_malloc_move_ma(&m,
774 						    &ma_offs, sc->sectorsize,
775 						    (void *)osp, 0,
776 						    MD_MALLOC_MOVE_WRITE);
777 					} else if (vlist != NULL) {
778 						error = md_malloc_move_vlist(
779 						    &vlist, &ma_offs,
780 						    sc->sectorsize, (void *)osp,
781 						    0, MD_MALLOC_MOVE_WRITE);
782 					} else {
783 						bcopy(dst, (void *)osp,
784 						    sc->sectorsize);
785 					}
786 					osp = 0;
787 				}
788 			}
789 		} else {
790 			error = EOPNOTSUPP;
791 		}
792 		if (osp > 255)
793 			uma_zfree(sc->uma, (void*)osp);
794 		if (error != 0)
795 			break;
796 		secno++;
797 		if (!notmapped && vlist == NULL)
798 			dst += sc->sectorsize;
799 	}
800 	bp->bio_resid = 0;
801 	return (error);
802 }
803 
804 static void
805 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
806 {
807 	off_t seg_len;
808 
809 	while (offset >= vlist->ds_len) {
810 		offset -= vlist->ds_len;
811 		vlist++;
812 	}
813 
814 	while (len != 0) {
815 		seg_len = omin(len, vlist->ds_len - offset);
816 		bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
817 		    seg_len);
818 		offset = 0;
819 		src = (uint8_t *)src + seg_len;
820 		len -= seg_len;
821 		vlist++;
822 	}
823 }
824 
825 static void
826 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
827 {
828 	off_t seg_len;
829 
830 	while (offset >= vlist->ds_len) {
831 		offset -= vlist->ds_len;
832 		vlist++;
833 	}
834 
835 	while (len != 0) {
836 		seg_len = omin(len, vlist->ds_len - offset);
837 		bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
838 		    seg_len);
839 		offset = 0;
840 		dst = (uint8_t *)dst + seg_len;
841 		len -= seg_len;
842 		vlist++;
843 	}
844 }
845 
846 static int
847 mdstart_preload(struct md_s *sc, struct bio *bp)
848 {
849 	uint8_t *p;
850 
851 	p = sc->pl_ptr + bp->bio_offset;
852 	switch (bp->bio_cmd) {
853 	case BIO_READ:
854 		if ((bp->bio_flags & BIO_VLIST) != 0) {
855 			mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
856 			    bp->bio_ma_offset, bp->bio_length);
857 		} else {
858 			bcopy(p, bp->bio_data, bp->bio_length);
859 		}
860 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
861 		break;
862 	case BIO_WRITE:
863 		if ((bp->bio_flags & BIO_VLIST) != 0) {
864 			mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
865 			    bp->bio_ma_offset, p, bp->bio_length);
866 		} else {
867 			bcopy(bp->bio_data, p, bp->bio_length);
868 		}
869 		break;
870 	}
871 	bp->bio_resid = 0;
872 	return (0);
873 }
874 
875 static int
876 mdstart_vnode(struct md_s *sc, struct bio *bp)
877 {
878 	int error;
879 	struct uio auio;
880 	struct iovec aiov;
881 	struct iovec *piov;
882 	struct mount *mp;
883 	struct vnode *vp;
884 	bus_dma_segment_t *vlist;
885 	struct thread *td;
886 	off_t iolen, iostart, off, len;
887 	int ma_offs, npages;
888 	bool mapped;
889 
890 	td = curthread;
891 	vp = sc->vnode;
892 	piov = NULL;
893 	ma_offs = bp->bio_ma_offset;
894 	off = bp->bio_offset;
895 	len = bp->bio_length;
896 	mapped = false;
897 
898 	/*
899 	 * VNODE I/O
900 	 *
901 	 * If an error occurs, we set BIO_ERROR but we do not set
902 	 * B_INVAL because (for a write anyway), the buffer is
903 	 * still valid.
904 	 */
905 
906 	switch (bp->bio_cmd) {
907 	case BIO_READ:
908 		auio.uio_rw = UIO_READ;
909 		break;
910 	case BIO_WRITE:
911 		auio.uio_rw = UIO_WRITE;
912 		break;
913 	case BIO_FLUSH:
914 		do {
915 			(void)vn_start_write(vp, &mp, V_WAIT);
916 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
917 			error = VOP_FSYNC(vp, MNT_WAIT, td);
918 			VOP_UNLOCK(vp);
919 			vn_finished_write(mp);
920 		} while (error == ERELOOKUP);
921 		return (error);
922 	case BIO_DELETE:
923 		if (sc->candelete) {
924 			error = vn_deallocate(vp, &off, &len, 0,
925 			    sc->flags & MD_ASYNC ? 0 : IO_SYNC, sc->cred,
926 			    NOCRED);
927 			bp->bio_resid = len;
928 			return (error);
929 		}
930 		/* FALLTHROUGH */
931 	default:
932 		return (EOPNOTSUPP);
933 	}
934 
935 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
936 	auio.uio_resid = bp->bio_length;
937 	auio.uio_segflg = UIO_SYSSPACE;
938 	auio.uio_td = td;
939 
940 	if ((bp->bio_flags & BIO_VLIST) != 0) {
941 		piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
942 		auio.uio_iov = piov;
943 		vlist = (bus_dma_segment_t *)bp->bio_data;
944 		while (len > 0) {
945 			piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
946 			    ma_offs);
947 			piov->iov_len = vlist->ds_len - ma_offs;
948 			if (piov->iov_len > len)
949 				piov->iov_len = len;
950 			len -= piov->iov_len;
951 			ma_offs = 0;
952 			vlist++;
953 			piov++;
954 		}
955 		auio.uio_iovcnt = piov - auio.uio_iov;
956 		piov = auio.uio_iov;
957 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
958 		bp->bio_resid = len;
959 unmapped_step:
960 		npages = atop(min(maxphys, round_page(len + (ma_offs &
961 		    PAGE_MASK))));
962 		iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
963 		KASSERT(iolen > 0, ("zero iolen"));
964 		KASSERT(npages <= atop(maxphys + PAGE_SIZE),
965 		    ("npages %d too large", npages));
966 		pmap_qenter(sc->kva, &bp->bio_ma[atop(ma_offs)], npages);
967 		aiov.iov_base = (void *)(sc->kva + (ma_offs & PAGE_MASK));
968 		aiov.iov_len = iolen;
969 		auio.uio_iov = &aiov;
970 		auio.uio_iovcnt = 1;
971 		auio.uio_resid = iolen;
972 		mapped = true;
973 	} else {
974 		aiov.iov_base = bp->bio_data;
975 		aiov.iov_len = bp->bio_length;
976 		auio.uio_iov = &aiov;
977 		auio.uio_iovcnt = 1;
978 	}
979 	iostart = auio.uio_offset;
980 	if (bp->bio_cmd == BIO_READ) {
981 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
982 		error = VOP_READ(vp, &auio, 0, sc->cred);
983 		VOP_UNLOCK(vp);
984 	} else {
985 		(void) vn_start_write(vp, &mp, V_WAIT);
986 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
987 		error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
988 		    sc->cred);
989 		VOP_UNLOCK(vp);
990 		vn_finished_write(mp);
991 		if (error == 0)
992 			sc->flags &= ~MD_VERIFY;
993 	}
994 
995 	/* When MD_CACHE is set, try to avoid double-caching the data. */
996 	if (error == 0 && (sc->flags & MD_CACHE) == 0)
997 		VOP_ADVISE(vp, iostart, auio.uio_offset - 1,
998 		    POSIX_FADV_DONTNEED);
999 
1000 	if (mapped) {
1001 		pmap_qremove(sc->kva, npages);
1002 		if (error == 0) {
1003 			len -= iolen;
1004 			bp->bio_resid -= iolen;
1005 			ma_offs += iolen;
1006 			if (len > 0)
1007 				goto unmapped_step;
1008 		}
1009 	} else {
1010 		bp->bio_resid = auio.uio_resid;
1011 	}
1012 
1013 	free(piov, M_MD);
1014 	return (error);
1015 }
1016 
1017 static int
1018 mdstart_swap(struct md_s *sc, struct bio *bp)
1019 {
1020 	vm_page_t m;
1021 	u_char *p;
1022 	vm_pindex_t i, lastp;
1023 	bus_dma_segment_t *vlist;
1024 	int rv, ma_offs, offs, len, lastend;
1025 
1026 	switch (bp->bio_cmd) {
1027 	case BIO_READ:
1028 	case BIO_WRITE:
1029 	case BIO_DELETE:
1030 		break;
1031 	case BIO_FLUSH:
1032 		return (0);
1033 	default:
1034 		return (EOPNOTSUPP);
1035 	}
1036 
1037 	p = bp->bio_data;
1038 	ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1039 	    bp->bio_ma_offset : 0;
1040 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1041 	    (bus_dma_segment_t *)bp->bio_data : NULL;
1042 
1043 	/*
1044 	 * offs is the offset at which to start operating on the
1045 	 * next (ie, first) page.  lastp is the last page on
1046 	 * which we're going to operate.  lastend is the ending
1047 	 * position within that last page (ie, PAGE_SIZE if
1048 	 * we're operating on complete aligned pages).
1049 	 */
1050 	offs = bp->bio_offset % PAGE_SIZE;
1051 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1052 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1053 
1054 	rv = VM_PAGER_OK;
1055 	vm_object_pip_add(sc->object, 1);
1056 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1057 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1058 		m = vm_page_grab_unlocked(sc->object, i, VM_ALLOC_SYSTEM);
1059 		if (bp->bio_cmd == BIO_READ) {
1060 			if (vm_page_all_valid(m))
1061 				rv = VM_PAGER_OK;
1062 			else
1063 				rv = vm_pager_get_pages(sc->object, &m, 1,
1064 				    NULL, NULL);
1065 			if (rv == VM_PAGER_ERROR) {
1066 				VM_OBJECT_WLOCK(sc->object);
1067 				vm_page_free(m);
1068 				VM_OBJECT_WUNLOCK(sc->object);
1069 				break;
1070 			} else if (rv == VM_PAGER_FAIL) {
1071 				/*
1072 				 * Pager does not have the page.  Zero
1073 				 * the allocated page, and mark it as
1074 				 * valid. Do not set dirty, the page
1075 				 * can be recreated if thrown out.
1076 				 */
1077 				pmap_zero_page(m);
1078 				vm_page_valid(m);
1079 			}
1080 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1081 				pmap_copy_pages(&m, offs, bp->bio_ma,
1082 				    ma_offs, len);
1083 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1084 				physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1085 				    vlist, ma_offs, len);
1086 				cpu_flush_dcache(p, len);
1087 			} else {
1088 				physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1089 				cpu_flush_dcache(p, len);
1090 			}
1091 		} else if (bp->bio_cmd == BIO_WRITE) {
1092 			if (len == PAGE_SIZE || vm_page_all_valid(m))
1093 				rv = VM_PAGER_OK;
1094 			else
1095 				rv = vm_pager_get_pages(sc->object, &m, 1,
1096 				    NULL, NULL);
1097 			if (rv == VM_PAGER_ERROR) {
1098 				VM_OBJECT_WLOCK(sc->object);
1099 				vm_page_free(m);
1100 				VM_OBJECT_WUNLOCK(sc->object);
1101 				break;
1102 			} else if (rv == VM_PAGER_FAIL)
1103 				pmap_zero_page(m);
1104 
1105 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1106 				pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1107 				    offs, len);
1108 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1109 				physcopyin_vlist(vlist, ma_offs,
1110 				    VM_PAGE_TO_PHYS(m) + offs, len);
1111 			} else {
1112 				physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1113 			}
1114 
1115 			vm_page_valid(m);
1116 			vm_page_set_dirty(m);
1117 		} else if (bp->bio_cmd == BIO_DELETE) {
1118 			if (len == PAGE_SIZE || vm_page_all_valid(m))
1119 				rv = VM_PAGER_OK;
1120 			else
1121 				rv = vm_pager_get_pages(sc->object, &m, 1,
1122 				    NULL, NULL);
1123 			VM_OBJECT_WLOCK(sc->object);
1124 			if (rv == VM_PAGER_ERROR) {
1125 				vm_page_free(m);
1126 				VM_OBJECT_WUNLOCK(sc->object);
1127 				break;
1128 			} else if (rv == VM_PAGER_FAIL) {
1129 				vm_page_free(m);
1130 				m = NULL;
1131 			} else {
1132 				/* Page is valid. */
1133 				if (len != PAGE_SIZE) {
1134 					pmap_zero_page_area(m, offs, len);
1135 					vm_page_set_dirty(m);
1136 				} else {
1137 					vm_pager_page_unswapped(m);
1138 					vm_page_free(m);
1139 					m = NULL;
1140 				}
1141 			}
1142 			VM_OBJECT_WUNLOCK(sc->object);
1143 		}
1144 		if (m != NULL) {
1145 			/*
1146 			 * The page may be deactivated prior to setting
1147 			 * PGA_REFERENCED, but in this case it will be
1148 			 * reactivated by the page daemon.
1149 			 */
1150 			if (vm_page_active(m))
1151 				vm_page_reference(m);
1152 			else
1153 				vm_page_activate(m);
1154 			vm_page_xunbusy(m);
1155 		}
1156 
1157 		/* Actions on further pages start at offset 0 */
1158 		p += PAGE_SIZE - offs;
1159 		offs = 0;
1160 		ma_offs += len;
1161 	}
1162 	vm_object_pip_wakeup(sc->object);
1163 	return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1164 }
1165 
1166 static int
1167 mdstart_null(struct md_s *sc, struct bio *bp)
1168 {
1169 
1170 	switch (bp->bio_cmd) {
1171 	case BIO_READ:
1172 		bzero(bp->bio_data, bp->bio_length);
1173 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
1174 		break;
1175 	case BIO_WRITE:
1176 		break;
1177 	}
1178 	bp->bio_resid = 0;
1179 	return (0);
1180 }
1181 
1182 static void
1183 md_handleattr(struct md_s *sc, struct bio *bp)
1184 {
1185 	if (sc->fwsectors && sc->fwheads &&
1186 	    (g_handleattr_int(bp, "GEOM::fwsectors", sc->fwsectors) != 0 ||
1187 	    g_handleattr_int(bp, "GEOM::fwheads", sc->fwheads) != 0))
1188 		return;
1189 	if (g_handleattr_int(bp, "GEOM::candelete", sc->candelete) != 0)
1190 		return;
1191 	if (sc->ident[0] != '\0' &&
1192 	    g_handleattr_str(bp, "GEOM::ident", sc->ident) != 0)
1193 		return;
1194 	if (g_handleattr_int(bp, "MNT::verified", (sc->flags & MD_VERIFY) != 0))
1195 		return;
1196 	g_io_deliver(bp, EOPNOTSUPP);
1197 }
1198 
1199 static void
1200 md_kthread(void *arg)
1201 {
1202 	struct md_s *sc;
1203 	struct bio *bp;
1204 	int error;
1205 
1206 	sc = arg;
1207 	thread_lock(curthread);
1208 	sched_prio(curthread, PRIBIO);
1209 	thread_unlock(curthread);
1210 	if (sc->type == MD_VNODE)
1211 		curthread->td_pflags |= TDP_NORUNNINGBUF;
1212 
1213 	for (;;) {
1214 		mtx_lock(&sc->queue_mtx);
1215 		if (sc->flags & MD_SHUTDOWN) {
1216 			sc->flags |= MD_EXITING;
1217 			mtx_unlock(&sc->queue_mtx);
1218 			kproc_exit(0);
1219 		}
1220 		bp = bioq_takefirst(&sc->bio_queue);
1221 		if (!bp) {
1222 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1223 			continue;
1224 		}
1225 		mtx_unlock(&sc->queue_mtx);
1226 		if (bp->bio_cmd == BIO_GETATTR) {
1227 			md_handleattr(sc, bp);
1228 		} else {
1229 			error = sc->start(sc, bp);
1230 			if (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE) {
1231 				/*
1232 				 * Devstat uses (bio_bcount, bio_resid) for
1233 				 * determining the length of the completed part
1234 				 * of the i/o.  g_io_deliver() will translate
1235 				 * from bio_completed to that, but it also
1236 				 * destroys the bio so we must do our own
1237 				 * translation.
1238 				 */
1239 				bp->bio_bcount = bp->bio_length;
1240 				devstat_end_transaction_bio(sc->devstat, bp);
1241 			}
1242 			bp->bio_completed = bp->bio_length - bp->bio_resid;
1243 			g_io_deliver(bp, error);
1244 		}
1245 	}
1246 }
1247 
1248 static struct md_s *
1249 mdfind(int unit)
1250 {
1251 	struct md_s *sc;
1252 
1253 	LIST_FOREACH(sc, &md_softc_list, list) {
1254 		if (sc->unit == unit)
1255 			break;
1256 	}
1257 	return (sc);
1258 }
1259 
1260 static struct md_s *
1261 mdnew(int unit, int *errp, enum md_types type)
1262 {
1263 	struct md_s *sc;
1264 	int error;
1265 
1266 	*errp = 0;
1267 	if (unit == -1)
1268 		unit = alloc_unr(md_uh);
1269 	else
1270 		unit = alloc_unr_specific(md_uh, unit);
1271 
1272 	if (unit == -1) {
1273 		*errp = EBUSY;
1274 		return (NULL);
1275 	}
1276 
1277 	sc = malloc(sizeof(*sc), M_MD, M_WAITOK | M_ZERO);
1278 	sc->type = type;
1279 	bioq_init(&sc->bio_queue);
1280 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1281 	sc->unit = unit;
1282 	sprintf(sc->name, "md%d", unit);
1283 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
1284 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1285 	if (error == 0)
1286 		return (sc);
1287 	LIST_REMOVE(sc, list);
1288 	mtx_destroy(&sc->queue_mtx);
1289 	free_unr(md_uh, sc->unit);
1290 	free(sc, M_MD);
1291 	*errp = error;
1292 	return (NULL);
1293 }
1294 
1295 static void
1296 mdinit(struct md_s *sc)
1297 {
1298 	struct g_geom *gp;
1299 	struct g_provider *pp;
1300 	unsigned remn;
1301 
1302 	g_topology_lock();
1303 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1304 	gp->softc = sc;
1305 	pp = g_new_providerf(gp, "md%d", sc->unit);
1306 	devstat_remove_entry(pp->stat);
1307 	pp->stat = NULL;
1308 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1309 	/* Prune off any residual fractional sector. */
1310 	remn = sc->mediasize % sc->sectorsize;
1311 	if (remn != 0) {
1312 		printf("md%d: truncating fractional last sector by %u bytes\n",
1313 		    sc->unit, remn);
1314 		sc->mediasize -= remn;
1315 	}
1316 	pp->mediasize = sc->mediasize;
1317 	pp->sectorsize = sc->sectorsize;
1318 	switch (sc->type) {
1319 	case MD_MALLOC:
1320 	case MD_VNODE:
1321 	case MD_SWAP:
1322 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1323 		break;
1324 	case MD_PRELOAD:
1325 	case MD_NULL:
1326 		break;
1327 	}
1328 	sc->gp = gp;
1329 	sc->pp = pp;
1330 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1331 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1332 	sc->devstat->id = pp;
1333 	g_error_provider(pp, 0);
1334 	g_topology_unlock();
1335 }
1336 
1337 static int
1338 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1339 {
1340 	uintptr_t sp;
1341 	int error;
1342 	off_t u;
1343 
1344 	error = 0;
1345 	if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1346 		return (EINVAL);
1347 	if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1348 		return (EINVAL);
1349 	/* Compression doesn't make sense if we have reserved space */
1350 	if (mdr->md_options & MD_RESERVE)
1351 		mdr->md_options &= ~MD_COMPRESS;
1352 	if (mdr->md_fwsectors != 0)
1353 		sc->fwsectors = mdr->md_fwsectors;
1354 	if (mdr->md_fwheads != 0)
1355 		sc->fwheads = mdr->md_fwheads;
1356 	sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE | MD_RESERVE);
1357 	sc->indir = dimension(sc->mediasize / sc->sectorsize);
1358 	sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1359 	    0x1ff, 0);
1360 	if (mdr->md_options & MD_RESERVE) {
1361 		off_t nsectors;
1362 
1363 		nsectors = sc->mediasize / sc->sectorsize;
1364 		for (u = 0; u < nsectors; u++) {
1365 			sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1366 			    M_WAITOK : M_NOWAIT) | M_ZERO);
1367 			if (sp != 0)
1368 				error = s_write(sc->indir, u, sp);
1369 			else
1370 				error = ENOMEM;
1371 			if (error != 0)
1372 				break;
1373 		}
1374 	}
1375 	return (error);
1376 }
1377 
1378 static int
1379 mdsetcred(struct md_s *sc, struct ucred *cred)
1380 {
1381 	char *tmpbuf;
1382 	int error = 0;
1383 
1384 	/*
1385 	 * Set credits in our softc
1386 	 */
1387 
1388 	if (sc->cred)
1389 		crfree(sc->cred);
1390 	sc->cred = crhold(cred);
1391 
1392 	/*
1393 	 * Horrible kludge to establish credentials for NFS  XXX.
1394 	 */
1395 
1396 	if (sc->vnode) {
1397 		struct uio auio;
1398 		struct iovec aiov;
1399 
1400 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1401 		bzero(&auio, sizeof(auio));
1402 
1403 		aiov.iov_base = tmpbuf;
1404 		aiov.iov_len = sc->sectorsize;
1405 		auio.uio_iov = &aiov;
1406 		auio.uio_iovcnt = 1;
1407 		auio.uio_offset = 0;
1408 		auio.uio_rw = UIO_READ;
1409 		auio.uio_segflg = UIO_SYSSPACE;
1410 		auio.uio_resid = aiov.iov_len;
1411 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1412 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1413 		VOP_UNLOCK(sc->vnode);
1414 		free(tmpbuf, M_TEMP);
1415 	}
1416 	return (error);
1417 }
1418 
1419 static int
1420 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1421 {
1422 	struct vattr vattr;
1423 	struct nameidata nd;
1424 	char *fname;
1425 	int error, flags;
1426 	long v;
1427 
1428 	fname = mdr->md_file;
1429 	if (mdr->md_file_seg == UIO_USERSPACE) {
1430 		error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1431 		if (error != 0)
1432 			return (error);
1433 	} else if (mdr->md_file_seg == UIO_SYSSPACE)
1434 		strlcpy(sc->file, fname, sizeof(sc->file));
1435 	else
1436 		return (EDOOFUS);
1437 
1438 	/*
1439 	 * If the user specified that this is a read only device, don't
1440 	 * set the FWRITE mask before trying to open the backing store.
1441 	 */
1442 	flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1443 	    | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1444 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file);
1445 	error = vn_open(&nd, &flags, 0, NULL);
1446 	if (error != 0)
1447 		return (error);
1448 	NDFREE_PNBUF(&nd);
1449 	if (nd.ni_vp->v_type != VREG) {
1450 		error = EINVAL;
1451 		goto bad;
1452 	}
1453 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1454 	if (error != 0)
1455 		goto bad;
1456 	if ((mdr->md_options & MD_MUSTDEALLOC) != 0) {
1457 		error = VOP_PATHCONF(nd.ni_vp, _PC_DEALLOC_PRESENT, &v);
1458 		if (error != 0)
1459 			goto bad;
1460 		if (v == 0)
1461 			sc->candelete = false;
1462 	}
1463 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1464 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1465 		if (VN_IS_DOOMED(nd.ni_vp)) {
1466 			/* Forced unmount. */
1467 			error = EBADF;
1468 			goto bad;
1469 		}
1470 	}
1471 	nd.ni_vp->v_vflag |= VV_MD;
1472 	VOP_UNLOCK(nd.ni_vp);
1473 
1474 	if (mdr->md_fwsectors != 0)
1475 		sc->fwsectors = mdr->md_fwsectors;
1476 	if (mdr->md_fwheads != 0)
1477 		sc->fwheads = mdr->md_fwheads;
1478 	snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1479 	    (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1480 	sc->flags = mdr->md_options & (MD_ASYNC | MD_CACHE | MD_FORCE |
1481 	    MD_VERIFY | MD_MUSTDEALLOC);
1482 	if (!(flags & FWRITE))
1483 		sc->flags |= MD_READONLY;
1484 	sc->vnode = nd.ni_vp;
1485 
1486 	error = mdsetcred(sc, td->td_ucred);
1487 	if (error != 0) {
1488 		sc->vnode = NULL;
1489 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1490 		nd.ni_vp->v_vflag &= ~VV_MD;
1491 		goto bad;
1492 	}
1493 
1494 	sc->kva = kva_alloc(maxphys + PAGE_SIZE);
1495 	return (0);
1496 bad:
1497 	VOP_UNLOCK(nd.ni_vp);
1498 	(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1499 	return (error);
1500 }
1501 
1502 static void
1503 g_md_providergone(struct g_provider *pp)
1504 {
1505 	struct md_s *sc = pp->geom->softc;
1506 
1507 	mtx_lock(&sc->queue_mtx);
1508 	sc->flags |= MD_PROVIDERGONE;
1509 	wakeup(&sc->flags);
1510 	mtx_unlock(&sc->queue_mtx);
1511 }
1512 
1513 static int
1514 mddestroy(struct md_s *sc, struct thread *td)
1515 {
1516 
1517 	if (sc->gp) {
1518 		g_topology_lock();
1519 		g_wither_geom(sc->gp, ENXIO);
1520 		g_topology_unlock();
1521 
1522 		mtx_lock(&sc->queue_mtx);
1523 		while (!(sc->flags & MD_PROVIDERGONE))
1524 			msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1525 		mtx_unlock(&sc->queue_mtx);
1526 	}
1527 	if (sc->devstat) {
1528 		devstat_remove_entry(sc->devstat);
1529 		sc->devstat = NULL;
1530 	}
1531 	mtx_lock(&sc->queue_mtx);
1532 	sc->flags |= MD_SHUTDOWN;
1533 	wakeup(sc);
1534 	while (!(sc->flags & MD_EXITING))
1535 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1536 	mtx_unlock(&sc->queue_mtx);
1537 	mtx_destroy(&sc->queue_mtx);
1538 	if (sc->vnode != NULL) {
1539 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1540 		sc->vnode->v_vflag &= ~VV_MD;
1541 		VOP_UNLOCK(sc->vnode);
1542 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1543 		    FREAD : (FREAD|FWRITE), sc->cred, td);
1544 	}
1545 	if (sc->cred != NULL)
1546 		crfree(sc->cred);
1547 	if (sc->object != NULL)
1548 		vm_object_deallocate(sc->object);
1549 	if (sc->indir)
1550 		destroy_indir(sc, sc->indir);
1551 	if (sc->uma)
1552 		uma_zdestroy(sc->uma);
1553 	if (sc->kva)
1554 		kva_free(sc->kva, maxphys + PAGE_SIZE);
1555 
1556 	LIST_REMOVE(sc, list);
1557 	free_unr(md_uh, sc->unit);
1558 	free(sc, M_MD);
1559 	return (0);
1560 }
1561 
1562 static int
1563 mdresize(struct md_s *sc, struct md_req *mdr)
1564 {
1565 	int error, res;
1566 	vm_pindex_t oldpages, newpages;
1567 
1568 	switch (sc->type) {
1569 	case MD_VNODE:
1570 	case MD_NULL:
1571 		break;
1572 	case MD_SWAP:
1573 		if (mdr->md_mediasize <= 0 ||
1574 		    (mdr->md_mediasize % PAGE_SIZE) != 0)
1575 			return (EDOM);
1576 		oldpages = OFF_TO_IDX(sc->mediasize);
1577 		newpages = OFF_TO_IDX(mdr->md_mediasize);
1578 		if (newpages < oldpages) {
1579 			VM_OBJECT_WLOCK(sc->object);
1580 			vm_object_page_remove(sc->object, newpages, 0, 0);
1581 			swap_release_by_cred(IDX_TO_OFF(oldpages -
1582 			    newpages), sc->cred);
1583 			sc->object->charge = IDX_TO_OFF(newpages);
1584 			sc->object->size = newpages;
1585 			VM_OBJECT_WUNLOCK(sc->object);
1586 		} else if (newpages > oldpages) {
1587 			res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1588 			    oldpages), sc->cred);
1589 			if (!res)
1590 				return (ENOMEM);
1591 			if ((mdr->md_options & MD_RESERVE) ||
1592 			    (sc->flags & MD_RESERVE)) {
1593 				error = swap_pager_reserve(sc->object,
1594 				    oldpages, newpages - oldpages);
1595 				if (error < 0) {
1596 					swap_release_by_cred(
1597 					    IDX_TO_OFF(newpages - oldpages),
1598 					    sc->cred);
1599 					return (EDOM);
1600 				}
1601 			}
1602 			VM_OBJECT_WLOCK(sc->object);
1603 			sc->object->charge = IDX_TO_OFF(newpages);
1604 			sc->object->size = newpages;
1605 			VM_OBJECT_WUNLOCK(sc->object);
1606 		}
1607 		break;
1608 	default:
1609 		return (EOPNOTSUPP);
1610 	}
1611 
1612 	sc->mediasize = mdr->md_mediasize;
1613 
1614 	g_topology_lock();
1615 	g_resize_provider(sc->pp, sc->mediasize);
1616 	g_topology_unlock();
1617 	return (0);
1618 }
1619 
1620 static int
1621 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1622 {
1623 	vm_ooffset_t npage;
1624 	int error;
1625 
1626 	/*
1627 	 * Range check.  Disallow negative sizes and sizes not being
1628 	 * multiple of page size.
1629 	 */
1630 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1631 		return (EDOM);
1632 
1633 	/*
1634 	 * Allocate an OBJT_SWAP object.
1635 	 *
1636 	 * Note the truncation.
1637 	 */
1638 
1639 	if ((mdr->md_options & MD_VERIFY) != 0)
1640 		return (EINVAL);
1641 	npage = mdr->md_mediasize / PAGE_SIZE;
1642 	if (mdr->md_fwsectors != 0)
1643 		sc->fwsectors = mdr->md_fwsectors;
1644 	if (mdr->md_fwheads != 0)
1645 		sc->fwheads = mdr->md_fwheads;
1646 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1647 	    VM_PROT_DEFAULT, 0, td->td_ucred);
1648 	if (sc->object == NULL)
1649 		return (ENOMEM);
1650 	sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1651 	if (mdr->md_options & MD_RESERVE) {
1652 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1653 			error = EDOM;
1654 			goto finish;
1655 		}
1656 	}
1657 	error = mdsetcred(sc, td->td_ucred);
1658  finish:
1659 	if (error != 0) {
1660 		vm_object_deallocate(sc->object);
1661 		sc->object = NULL;
1662 	}
1663 	return (error);
1664 }
1665 
1666 static int
1667 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1668 {
1669 
1670 	/*
1671 	 * Range check.  Disallow negative sizes and sizes not being
1672 	 * multiple of page size.
1673 	 */
1674 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1675 		return (EDOM);
1676 
1677 	return (0);
1678 }
1679 
1680 static int
1681 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1682 {
1683 	struct md_s *sc;
1684 	unsigned sectsize;
1685 	int error;
1686 
1687 	sx_assert(&md_sx, SA_XLOCKED);
1688 
1689 	switch (mdr->md_type) {
1690 	case MD_MALLOC:
1691 	case MD_PRELOAD:
1692 	case MD_VNODE:
1693 	case MD_SWAP:
1694 	case MD_NULL:
1695 		break;
1696 	default:
1697 		return (EINVAL);
1698 	}
1699 	if (mdr->md_sectorsize == 0)
1700 		sectsize = DEV_BSIZE;
1701 	else
1702 		sectsize = mdr->md_sectorsize;
1703 	if (sectsize > maxphys || mdr->md_mediasize < sectsize)
1704 		return (EINVAL);
1705 	if (mdr->md_options & MD_AUTOUNIT)
1706 		sc = mdnew(-1, &error, mdr->md_type);
1707 	else {
1708 		if (mdr->md_unit > INT_MAX)
1709 			return (EINVAL);
1710 		sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1711 	}
1712 	if (sc == NULL)
1713 		return (error);
1714 	if (mdr->md_label != NULL)
1715 		error = copyinstr(mdr->md_label, sc->label,
1716 		    sizeof(sc->label), NULL);
1717 	if (error != 0)
1718 		goto err_after_new;
1719 	if (mdr->md_options & MD_AUTOUNIT)
1720 		mdr->md_unit = sc->unit;
1721 	sc->mediasize = mdr->md_mediasize;
1722 	sc->sectorsize = sectsize;
1723 	sc->candelete = true;
1724 	error = EDOOFUS;
1725 	switch (sc->type) {
1726 	case MD_MALLOC:
1727 		sc->start = mdstart_malloc;
1728 		error = mdcreate_malloc(sc, mdr);
1729 		break;
1730 	case MD_PRELOAD:
1731 		/*
1732 		 * We disallow attaching preloaded memory disks via
1733 		 * ioctl. Preloaded memory disks are automatically
1734 		 * attached in g_md_init().
1735 		 */
1736 		error = EOPNOTSUPP;
1737 		break;
1738 	case MD_VNODE:
1739 		sc->start = mdstart_vnode;
1740 		error = mdcreate_vnode(sc, mdr, td);
1741 		break;
1742 	case MD_SWAP:
1743 		sc->start = mdstart_swap;
1744 		error = mdcreate_swap(sc, mdr, td);
1745 		break;
1746 	case MD_NULL:
1747 		sc->start = mdstart_null;
1748 		error = mdcreate_null(sc, mdr, td);
1749 		break;
1750 	}
1751 err_after_new:
1752 	if (error != 0) {
1753 		mddestroy(sc, td);
1754 		return (error);
1755 	}
1756 
1757 	mdinit(sc);
1758 	return (0);
1759 }
1760 
1761 static int
1762 kern_mdattach(struct thread *td, struct md_req *mdr)
1763 {
1764 	int error;
1765 
1766 	sx_xlock(&md_sx);
1767 	error = kern_mdattach_locked(td, mdr);
1768 	sx_xunlock(&md_sx);
1769 	return (error);
1770 }
1771 
1772 static int
1773 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1774 {
1775 	struct md_s *sc;
1776 
1777 	sx_assert(&md_sx, SA_XLOCKED);
1778 
1779 	if (mdr->md_mediasize != 0 ||
1780 	    (mdr->md_options & ~MD_FORCE) != 0)
1781 		return (EINVAL);
1782 
1783 	sc = mdfind(mdr->md_unit);
1784 	if (sc == NULL)
1785 		return (ENOENT);
1786 	if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1787 	    !(mdr->md_options & MD_FORCE))
1788 		return (EBUSY);
1789 	return (mddestroy(sc, td));
1790 }
1791 
1792 static int
1793 kern_mddetach(struct thread *td, struct md_req *mdr)
1794 {
1795 	int error;
1796 
1797 	sx_xlock(&md_sx);
1798 	error = kern_mddetach_locked(td, mdr);
1799 	sx_xunlock(&md_sx);
1800 	return (error);
1801 }
1802 
1803 static int
1804 kern_mdresize_locked(struct md_req *mdr)
1805 {
1806 	struct md_s *sc;
1807 
1808 	sx_assert(&md_sx, SA_XLOCKED);
1809 
1810 	if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1811 		return (EINVAL);
1812 
1813 	sc = mdfind(mdr->md_unit);
1814 	if (sc == NULL)
1815 		return (ENOENT);
1816 	if (mdr->md_mediasize < sc->sectorsize)
1817 		return (EINVAL);
1818 	mdr->md_mediasize -= mdr->md_mediasize % sc->sectorsize;
1819 	if (mdr->md_mediasize < sc->mediasize &&
1820 	    !(sc->flags & MD_FORCE) &&
1821 	    !(mdr->md_options & MD_FORCE))
1822 		return (EBUSY);
1823 	return (mdresize(sc, mdr));
1824 }
1825 
1826 static int
1827 kern_mdresize(struct md_req *mdr)
1828 {
1829 	int error;
1830 
1831 	sx_xlock(&md_sx);
1832 	error = kern_mdresize_locked(mdr);
1833 	sx_xunlock(&md_sx);
1834 	return (error);
1835 }
1836 
1837 static int
1838 kern_mdquery_locked(struct md_req *mdr)
1839 {
1840 	struct md_s *sc;
1841 	int error;
1842 
1843 	sx_assert(&md_sx, SA_XLOCKED);
1844 
1845 	sc = mdfind(mdr->md_unit);
1846 	if (sc == NULL)
1847 		return (ENOENT);
1848 	mdr->md_type = sc->type;
1849 	mdr->md_options = sc->flags;
1850 	mdr->md_mediasize = sc->mediasize;
1851 	mdr->md_sectorsize = sc->sectorsize;
1852 	error = 0;
1853 	if (mdr->md_label != NULL) {
1854 		error = copyout(sc->label, mdr->md_label,
1855 		    strlen(sc->label) + 1);
1856 		if (error != 0)
1857 			return (error);
1858 	}
1859 	if (sc->type == MD_VNODE ||
1860 	    (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1861 		error = copyout(sc->file, mdr->md_file,
1862 		    strlen(sc->file) + 1);
1863 	return (error);
1864 }
1865 
1866 static int
1867 kern_mdquery(struct md_req *mdr)
1868 {
1869 	int error;
1870 
1871 	sx_xlock(&md_sx);
1872 	error = kern_mdquery_locked(mdr);
1873 	sx_xunlock(&md_sx);
1874 	return (error);
1875 }
1876 
1877 /* Copy members that are not userspace pointers. */
1878 #define	MD_IOCTL2REQ(mdio, mdr) do {					\
1879 	(mdr)->md_unit = (mdio)->md_unit;				\
1880 	(mdr)->md_type = (mdio)->md_type;				\
1881 	(mdr)->md_mediasize = (mdio)->md_mediasize;			\
1882 	(mdr)->md_sectorsize = (mdio)->md_sectorsize;			\
1883 	(mdr)->md_options = (mdio)->md_options;				\
1884 	(mdr)->md_fwheads = (mdio)->md_fwheads;				\
1885 	(mdr)->md_fwsectors = (mdio)->md_fwsectors;			\
1886 	(mdr)->md_units = &(mdio)->md_pad[0];				\
1887 	(mdr)->md_units_nitems = nitems((mdio)->md_pad);		\
1888 } while(0)
1889 
1890 /* Copy members that might have been updated */
1891 #define MD_REQ2IOCTL(mdr, mdio) do {					\
1892 	(mdio)->md_unit = (mdr)->md_unit;				\
1893 	(mdio)->md_type = (mdr)->md_type;				\
1894 	(mdio)->md_mediasize = (mdr)->md_mediasize;			\
1895 	(mdio)->md_sectorsize = (mdr)->md_sectorsize;			\
1896 	(mdio)->md_options = (mdr)->md_options;				\
1897 	(mdio)->md_fwheads = (mdr)->md_fwheads;				\
1898 	(mdio)->md_fwsectors = (mdr)->md_fwsectors;			\
1899 } while(0)
1900 
1901 static int
1902 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1903     struct thread *td)
1904 {
1905 	struct md_req mdr;
1906 	int error;
1907 
1908 	if (md_debug)
1909 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1910 			devtoname(dev), cmd, addr, flags, td);
1911 
1912 	bzero(&mdr, sizeof(mdr));
1913 	switch (cmd) {
1914 	case MDIOCATTACH:
1915 	case MDIOCDETACH:
1916 	case MDIOCRESIZE:
1917 	case MDIOCQUERY: {
1918 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1919 		if (mdio->md_version != MDIOVERSION)
1920 			return (EINVAL);
1921 		MD_IOCTL2REQ(mdio, &mdr);
1922 		mdr.md_file = mdio->md_file;
1923 		mdr.md_file_seg = UIO_USERSPACE;
1924 		/* If the file is adjacent to the md_ioctl it's in kernel. */
1925 		if ((void *)mdio->md_file == (void *)(mdio + 1))
1926 			mdr.md_file_seg = UIO_SYSSPACE;
1927 		mdr.md_label = mdio->md_label;
1928 		break;
1929 	}
1930 #ifdef COMPAT_FREEBSD32
1931 	case MDIOCATTACH_32:
1932 	case MDIOCDETACH_32:
1933 	case MDIOCRESIZE_32:
1934 	case MDIOCQUERY_32: {
1935 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1936 		if (mdio->md_version != MDIOVERSION)
1937 			return (EINVAL);
1938 		MD_IOCTL2REQ(mdio, &mdr);
1939 		mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1940 		mdr.md_file_seg = UIO_USERSPACE;
1941 		mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1942 		break;
1943 	}
1944 #endif
1945 	default:
1946 		/* Fall through to handler switch. */
1947 		break;
1948 	}
1949 
1950 	error = 0;
1951 	switch (cmd) {
1952 	case MDIOCATTACH:
1953 #ifdef COMPAT_FREEBSD32
1954 	case MDIOCATTACH_32:
1955 #endif
1956 		error = kern_mdattach(td, &mdr);
1957 		break;
1958 	case MDIOCDETACH:
1959 #ifdef COMPAT_FREEBSD32
1960 	case MDIOCDETACH_32:
1961 #endif
1962 		error = kern_mddetach(td, &mdr);
1963 		break;
1964 	case MDIOCRESIZE:
1965 #ifdef COMPAT_FREEBSD32
1966 	case MDIOCRESIZE_32:
1967 #endif
1968 		error = kern_mdresize(&mdr);
1969 		break;
1970 	case MDIOCQUERY:
1971 #ifdef COMPAT_FREEBSD32
1972 	case MDIOCQUERY_32:
1973 #endif
1974 		error = kern_mdquery(&mdr);
1975 		break;
1976 	default:
1977 		error = ENOIOCTL;
1978 	}
1979 
1980 	switch (cmd) {
1981 	case MDIOCATTACH:
1982 	case MDIOCQUERY: {
1983 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1984 		MD_REQ2IOCTL(&mdr, mdio);
1985 		break;
1986 	}
1987 #ifdef COMPAT_FREEBSD32
1988 	case MDIOCATTACH_32:
1989 	case MDIOCQUERY_32: {
1990 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1991 		MD_REQ2IOCTL(&mdr, mdio);
1992 		break;
1993 	}
1994 #endif
1995 	default:
1996 		/* Other commands to not alter mdr. */
1997 		break;
1998 	}
1999 
2000 	return (error);
2001 }
2002 
2003 static void
2004 md_preloaded(u_char *image, size_t length, const char *name)
2005 {
2006 	struct md_s *sc;
2007 	int error;
2008 
2009 	sc = mdnew(-1, &error, MD_PRELOAD);
2010 	if (sc == NULL)
2011 		return;
2012 	sc->mediasize = length;
2013 	sc->sectorsize = DEV_BSIZE;
2014 	sc->pl_ptr = image;
2015 	sc->pl_len = length;
2016 	sc->start = mdstart_preload;
2017 	if (name != NULL)
2018 		strlcpy(sc->file, name, sizeof(sc->file));
2019 #ifdef MD_ROOT
2020 	if (sc->unit == 0) {
2021 #ifndef ROOTDEVNAME
2022 		rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2023 #endif
2024 #ifdef MD_ROOT_READONLY
2025 		sc->flags |= MD_READONLY;
2026 #endif
2027 	}
2028 #endif
2029 	mdinit(sc);
2030 	if (name != NULL) {
2031 		printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2032 		    MD_NAME, sc->unit, name, length, image);
2033 	} else {
2034 		printf("%s%d: Embedded image %zd bytes at %p\n",
2035 		    MD_NAME, sc->unit, length, image);
2036 	}
2037 }
2038 
2039 static void
2040 g_md_init(struct g_class *mp __unused)
2041 {
2042 	caddr_t mod;
2043 	u_char *ptr, *name, *type;
2044 	unsigned len;
2045 	int i;
2046 
2047 	/* figure out log2(NINDIR) */
2048 	for (i = NINDIR, nshift = -1; i; nshift++)
2049 		i >>= 1;
2050 
2051 	mod = NULL;
2052 	sx_init(&md_sx, "MD config lock");
2053 	g_topology_unlock();
2054 	md_uh = new_unrhdr(0, INT_MAX, NULL);
2055 #ifdef MD_ROOT
2056 	if (mfs_root_size != 0) {
2057 		sx_xlock(&md_sx);
2058 #ifdef MD_ROOT_MEM
2059 		md_preloaded(mfs_root, mfs_root_size, NULL);
2060 #else
2061 		md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2062 		    NULL);
2063 #endif
2064 		sx_xunlock(&md_sx);
2065 	}
2066 #endif
2067 	/* XXX: are preload_* static or do they need Giant ? */
2068 	while ((mod = preload_search_next_name(mod)) != NULL) {
2069 		name = (char *)preload_search_info(mod, MODINFO_NAME);
2070 		if (name == NULL)
2071 			continue;
2072 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
2073 		if (type == NULL)
2074 			continue;
2075 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2076 			continue;
2077 		ptr = preload_fetch_addr(mod);
2078 		len = preload_fetch_size(mod);
2079 		if (ptr != NULL && len != 0) {
2080 			sx_xlock(&md_sx);
2081 			md_preloaded(ptr, len, name);
2082 			sx_xunlock(&md_sx);
2083 		}
2084 	}
2085 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2086 	    0600, MDCTL_NAME);
2087 	g_topology_lock();
2088 }
2089 
2090 static void
2091 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2092     struct g_consumer *cp __unused, struct g_provider *pp)
2093 {
2094 	struct md_s *mp;
2095 	char *type;
2096 
2097 	mp = gp->softc;
2098 	if (mp == NULL)
2099 		return;
2100 
2101 	switch (mp->type) {
2102 	case MD_MALLOC:
2103 		type = "malloc";
2104 		break;
2105 	case MD_PRELOAD:
2106 		type = "preload";
2107 		break;
2108 	case MD_VNODE:
2109 		type = "vnode";
2110 		break;
2111 	case MD_SWAP:
2112 		type = "swap";
2113 		break;
2114 	case MD_NULL:
2115 		type = "null";
2116 		break;
2117 	default:
2118 		type = "unknown";
2119 		break;
2120 	}
2121 
2122 	if (pp != NULL) {
2123 		if (indent == NULL) {
2124 			sbuf_printf(sb, " u %d", mp->unit);
2125 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2126 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2127 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2128 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2129 			sbuf_printf(sb, " t %s", type);
2130 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2131 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2132 				sbuf_printf(sb, " file %s", mp->file);
2133 			sbuf_printf(sb, " label %s", mp->label);
2134 		} else {
2135 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2136 			    mp->unit);
2137 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2138 			    indent, (uintmax_t) mp->sectorsize);
2139 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2140 			    indent, (uintmax_t) mp->fwheads);
2141 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2142 			    indent, (uintmax_t) mp->fwsectors);
2143 			if (mp->ident[0] != '\0') {
2144 				sbuf_printf(sb, "%s<ident>", indent);
2145 				g_conf_printf_escaped(sb, "%s", mp->ident);
2146 				sbuf_printf(sb, "</ident>\n");
2147 			}
2148 			sbuf_printf(sb, "%s<length>%ju</length>\n",
2149 			    indent, (uintmax_t) mp->mediasize);
2150 			sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2151 			    (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2152 			sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2153 			    (mp->flags & MD_READONLY) == 0 ? "read-write":
2154 			    "read-only");
2155 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2156 			    type);
2157 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2158 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2159 				sbuf_printf(sb, "%s<file>", indent);
2160 				g_conf_printf_escaped(sb, "%s", mp->file);
2161 				sbuf_printf(sb, "</file>\n");
2162 			}
2163 			if (mp->type == MD_VNODE)
2164 				sbuf_printf(sb, "%s<cache>%s</cache>\n", indent,
2165 				    (mp->flags & MD_CACHE) == 0 ? "off": "on");
2166 			sbuf_printf(sb, "%s<label>", indent);
2167 			g_conf_printf_escaped(sb, "%s", mp->label);
2168 			sbuf_printf(sb, "</label>\n");
2169 		}
2170 	}
2171 }
2172 
2173 static void
2174 g_md_fini(struct g_class *mp __unused)
2175 {
2176 
2177 	sx_destroy(&md_sx);
2178 	if (status_dev != NULL)
2179 		destroy_dev(status_dev);
2180 	delete_unrhdr(md_uh);
2181 }
2182