xref: /freebsd-12.1/sys/dev/md/md.c (revision b8a4423f)
1 /*-
2  * SPDX-License-Identifier: (Beerware AND BSD-3-Clause)
3  *
4  * ----------------------------------------------------------------------------
5  * "THE BEER-WARE LICENSE" (Revision 42):
6  * <[email protected]> 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	MDIOCLIST_32	_IOC_NEWTYPE(MDIOCLIST, struct md_ioctl32)
155 #define	MDIOCRESIZE_32	_IOC_NEWTYPE(MDIOCRESIZE, struct md_ioctl32)
156 #endif /* COMPAT_FREEBSD32 */
157 
158 static MALLOC_DEFINE(M_MD, "md_disk", "Memory Disk");
159 static MALLOC_DEFINE(M_MDSECT, "md_sectors", "Memory Disk Sectors");
160 
161 static int md_debug;
162 SYSCTL_INT(_debug, OID_AUTO, mddebug, CTLFLAG_RW, &md_debug, 0,
163     "Enable md(4) debug messages");
164 static int md_malloc_wait;
165 SYSCTL_INT(_vm, OID_AUTO, md_malloc_wait, CTLFLAG_RW, &md_malloc_wait, 0,
166     "Allow malloc to wait for memory allocations");
167 
168 #if defined(MD_ROOT) && !defined(MD_ROOT_FSTYPE)
169 #define	MD_ROOT_FSTYPE	"ufs"
170 #endif
171 
172 #if defined(MD_ROOT)
173 /*
174  * Preloaded image gets put here.
175  */
176 #if defined(MD_ROOT_SIZE)
177 /*
178  * We put the mfs_root symbol into the oldmfs section of the kernel object file.
179  * Applications that patch the object with the image can determine
180  * the size looking at the oldmfs section size within the kernel.
181  */
182 u_char mfs_root[MD_ROOT_SIZE*1024] __attribute__ ((section ("oldmfs")));
183 const int mfs_root_size = sizeof(mfs_root);
184 #elif defined(MD_ROOT_MEM)
185 /* MD region already mapped in the memory */
186 u_char *mfs_root;
187 int mfs_root_size;
188 #else
189 extern volatile u_char __weak_symbol mfs_root;
190 extern volatile u_char __weak_symbol mfs_root_end;
191 __GLOBL(mfs_root);
192 __GLOBL(mfs_root_end);
193 #define mfs_root_size ((uintptr_t)(&mfs_root_end - &mfs_root))
194 #endif
195 #endif
196 
197 static g_init_t g_md_init;
198 static g_fini_t g_md_fini;
199 static g_start_t g_md_start;
200 static g_access_t g_md_access;
201 static void g_md_dumpconf(struct sbuf *sb, const char *indent,
202     struct g_geom *gp, struct g_consumer *cp __unused, struct g_provider *pp);
203 static g_provgone_t g_md_providergone;
204 
205 static struct cdev *status_dev = NULL;
206 static struct sx md_sx;
207 static struct unrhdr *md_uh;
208 
209 static d_ioctl_t mdctlioctl;
210 
211 static struct cdevsw mdctl_cdevsw = {
212 	.d_version =	D_VERSION,
213 	.d_ioctl =	mdctlioctl,
214 	.d_name =	MD_NAME,
215 };
216 
217 struct g_class g_md_class = {
218 	.name = "MD",
219 	.version = G_VERSION,
220 	.init = g_md_init,
221 	.fini = g_md_fini,
222 	.start = g_md_start,
223 	.access = g_md_access,
224 	.dumpconf = g_md_dumpconf,
225 	.providergone = g_md_providergone,
226 };
227 
228 DECLARE_GEOM_CLASS(g_md_class, g_md);
229 
230 
231 static LIST_HEAD(, md_s) md_softc_list = LIST_HEAD_INITIALIZER(md_softc_list);
232 
233 #define NINDIR	(PAGE_SIZE / sizeof(uintptr_t))
234 #define NMASK	(NINDIR-1)
235 static int nshift;
236 
237 static int md_vnode_pbuf_freecnt;
238 
239 struct indir {
240 	uintptr_t	*array;
241 	u_int		total;
242 	u_int		used;
243 	u_int		shift;
244 };
245 
246 struct md_s {
247 	int unit;
248 	LIST_ENTRY(md_s) list;
249 	struct bio_queue_head bio_queue;
250 	struct mtx queue_mtx;
251 	struct mtx stat_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 
268 	/* MD_MALLOC related fields */
269 	struct indir *indir;
270 	uma_zone_t uma;
271 
272 	/* MD_PRELOAD related fields */
273 	u_char *pl_ptr;
274 	size_t pl_len;
275 
276 	/* MD_VNODE related fields */
277 	struct vnode *vnode;
278 	char file[PATH_MAX];
279 	char label[PATH_MAX];
280 	struct ucred *cred;
281 
282 	/* MD_SWAP related fields */
283 	vm_object_t object;
284 };
285 
286 static struct indir *
new_indir(u_int shift)287 new_indir(u_int shift)
288 {
289 	struct indir *ip;
290 
291 	ip = malloc(sizeof *ip, M_MD, (md_malloc_wait ? M_WAITOK : M_NOWAIT)
292 	    | M_ZERO);
293 	if (ip == NULL)
294 		return (NULL);
295 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
296 	    M_MDSECT, (md_malloc_wait ? M_WAITOK : M_NOWAIT) | M_ZERO);
297 	if (ip->array == NULL) {
298 		free(ip, M_MD);
299 		return (NULL);
300 	}
301 	ip->total = NINDIR;
302 	ip->shift = shift;
303 	return (ip);
304 }
305 
306 static void
del_indir(struct indir * ip)307 del_indir(struct indir *ip)
308 {
309 
310 	free(ip->array, M_MDSECT);
311 	free(ip, M_MD);
312 }
313 
314 static void
destroy_indir(struct md_s * sc,struct indir * ip)315 destroy_indir(struct md_s *sc, struct indir *ip)
316 {
317 	int i;
318 
319 	for (i = 0; i < NINDIR; i++) {
320 		if (!ip->array[i])
321 			continue;
322 		if (ip->shift)
323 			destroy_indir(sc, (struct indir*)(ip->array[i]));
324 		else if (ip->array[i] > 255)
325 			uma_zfree(sc->uma, (void *)(ip->array[i]));
326 	}
327 	del_indir(ip);
328 }
329 
330 /*
331  * This function does the math and allocates the top level "indir" structure
332  * for a device of "size" sectors.
333  */
334 
335 static struct indir *
dimension(off_t size)336 dimension(off_t size)
337 {
338 	off_t rcnt;
339 	struct indir *ip;
340 	int layer;
341 
342 	rcnt = size;
343 	layer = 0;
344 	while (rcnt > NINDIR) {
345 		rcnt /= NINDIR;
346 		layer++;
347 	}
348 
349 	/*
350 	 * XXX: the top layer is probably not fully populated, so we allocate
351 	 * too much space for ip->array in here.
352 	 */
353 	ip = malloc(sizeof *ip, M_MD, M_WAITOK | M_ZERO);
354 	ip->array = malloc(sizeof(uintptr_t) * NINDIR,
355 	    M_MDSECT, M_WAITOK | M_ZERO);
356 	ip->total = NINDIR;
357 	ip->shift = layer * nshift;
358 	return (ip);
359 }
360 
361 /*
362  * Read a given sector
363  */
364 
365 static uintptr_t
s_read(struct indir * ip,off_t offset)366 s_read(struct indir *ip, off_t offset)
367 {
368 	struct indir *cip;
369 	int idx;
370 	uintptr_t up;
371 
372 	if (md_debug > 1)
373 		printf("s_read(%jd)\n", (intmax_t)offset);
374 	up = 0;
375 	for (cip = ip; cip != NULL;) {
376 		if (cip->shift) {
377 			idx = (offset >> cip->shift) & NMASK;
378 			up = cip->array[idx];
379 			cip = (struct indir *)up;
380 			continue;
381 		}
382 		idx = offset & NMASK;
383 		return (cip->array[idx]);
384 	}
385 	return (0);
386 }
387 
388 /*
389  * Write a given sector, prune the tree if the value is 0
390  */
391 
392 static int
s_write(struct indir * ip,off_t offset,uintptr_t ptr)393 s_write(struct indir *ip, off_t offset, uintptr_t ptr)
394 {
395 	struct indir *cip, *lip[10];
396 	int idx, li;
397 	uintptr_t up;
398 
399 	if (md_debug > 1)
400 		printf("s_write(%jd, %p)\n", (intmax_t)offset, (void *)ptr);
401 	up = 0;
402 	li = 0;
403 	cip = ip;
404 	for (;;) {
405 		lip[li++] = cip;
406 		if (cip->shift) {
407 			idx = (offset >> cip->shift) & NMASK;
408 			up = cip->array[idx];
409 			if (up != 0) {
410 				cip = (struct indir *)up;
411 				continue;
412 			}
413 			/* Allocate branch */
414 			cip->array[idx] =
415 			    (uintptr_t)new_indir(cip->shift - nshift);
416 			if (cip->array[idx] == 0)
417 				return (ENOSPC);
418 			cip->used++;
419 			up = cip->array[idx];
420 			cip = (struct indir *)up;
421 			continue;
422 		}
423 		/* leafnode */
424 		idx = offset & NMASK;
425 		up = cip->array[idx];
426 		if (up != 0)
427 			cip->used--;
428 		cip->array[idx] = ptr;
429 		if (ptr != 0)
430 			cip->used++;
431 		break;
432 	}
433 	if (cip->used != 0 || li == 1)
434 		return (0);
435 	li--;
436 	while (cip->used == 0 && cip != ip) {
437 		li--;
438 		idx = (offset >> lip[li]->shift) & NMASK;
439 		up = lip[li]->array[idx];
440 		KASSERT(up == (uintptr_t)cip, ("md screwed up"));
441 		del_indir(cip);
442 		lip[li]->array[idx] = 0;
443 		lip[li]->used--;
444 		cip = lip[li];
445 	}
446 	return (0);
447 }
448 
449 
450 static int
g_md_access(struct g_provider * pp,int r,int w,int e)451 g_md_access(struct g_provider *pp, int r, int w, int e)
452 {
453 	struct md_s *sc;
454 
455 	sc = pp->geom->softc;
456 	if (sc == NULL) {
457 		if (r <= 0 && w <= 0 && e <= 0)
458 			return (0);
459 		return (ENXIO);
460 	}
461 	r += pp->acr;
462 	w += pp->acw;
463 	e += pp->ace;
464 	if ((sc->flags & MD_READONLY) != 0 && w > 0)
465 		return (EROFS);
466 	if ((pp->acr + pp->acw + pp->ace) == 0 && (r + w + e) > 0) {
467 		sc->opencount = 1;
468 	} else if ((pp->acr + pp->acw + pp->ace) > 0 && (r + w + e) == 0) {
469 		sc->opencount = 0;
470 	}
471 	return (0);
472 }
473 
474 static void
g_md_start(struct bio * bp)475 g_md_start(struct bio *bp)
476 {
477 	struct md_s *sc;
478 
479 	sc = bp->bio_to->geom->softc;
480 	if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE)) {
481 		mtx_lock(&sc->stat_mtx);
482 		devstat_start_transaction_bio(sc->devstat, bp);
483 		mtx_unlock(&sc->stat_mtx);
484 	}
485 	mtx_lock(&sc->queue_mtx);
486 	bioq_disksort(&sc->bio_queue, bp);
487 	wakeup(sc);
488 	mtx_unlock(&sc->queue_mtx);
489 }
490 
491 #define	MD_MALLOC_MOVE_ZERO	1
492 #define	MD_MALLOC_MOVE_FILL	2
493 #define	MD_MALLOC_MOVE_READ	3
494 #define	MD_MALLOC_MOVE_WRITE	4
495 #define	MD_MALLOC_MOVE_CMP	5
496 
497 static int
md_malloc_move_ma(vm_page_t ** mp,int * ma_offs,unsigned sectorsize,void * ptr,u_char fill,int op)498 md_malloc_move_ma(vm_page_t **mp, int *ma_offs, unsigned sectorsize,
499     void *ptr, u_char fill, int op)
500 {
501 	struct sf_buf *sf;
502 	vm_page_t m, *mp1;
503 	char *p, first;
504 	off_t *uc;
505 	unsigned n;
506 	int error, i, ma_offs1, sz, first_read;
507 
508 	m = NULL;
509 	error = 0;
510 	sf = NULL;
511 	/* if (op == MD_MALLOC_MOVE_CMP) { gcc */
512 		first = 0;
513 		first_read = 0;
514 		uc = ptr;
515 		mp1 = *mp;
516 		ma_offs1 = *ma_offs;
517 	/* } */
518 	sched_pin();
519 	for (n = sectorsize; n != 0; n -= sz) {
520 		sz = imin(PAGE_SIZE - *ma_offs, n);
521 		if (m != **mp) {
522 			if (sf != NULL)
523 				sf_buf_free(sf);
524 			m = **mp;
525 			sf = sf_buf_alloc(m, SFB_CPUPRIVATE |
526 			    (md_malloc_wait ? 0 : SFB_NOWAIT));
527 			if (sf == NULL) {
528 				error = ENOMEM;
529 				break;
530 			}
531 		}
532 		p = (char *)sf_buf_kva(sf) + *ma_offs;
533 		switch (op) {
534 		case MD_MALLOC_MOVE_ZERO:
535 			bzero(p, sz);
536 			break;
537 		case MD_MALLOC_MOVE_FILL:
538 			memset(p, fill, sz);
539 			break;
540 		case MD_MALLOC_MOVE_READ:
541 			bcopy(ptr, p, sz);
542 			cpu_flush_dcache(p, sz);
543 			break;
544 		case MD_MALLOC_MOVE_WRITE:
545 			bcopy(p, ptr, sz);
546 			break;
547 		case MD_MALLOC_MOVE_CMP:
548 			for (i = 0; i < sz; i++, p++) {
549 				if (!first_read) {
550 					*uc = (u_char)*p;
551 					first = *p;
552 					first_read = 1;
553 				} else if (*p != first) {
554 					error = EDOOFUS;
555 					break;
556 				}
557 			}
558 			break;
559 		default:
560 			KASSERT(0, ("md_malloc_move_ma unknown op %d\n", op));
561 			break;
562 		}
563 		if (error != 0)
564 			break;
565 		*ma_offs += sz;
566 		*ma_offs %= PAGE_SIZE;
567 		if (*ma_offs == 0)
568 			(*mp)++;
569 		ptr = (char *)ptr + sz;
570 	}
571 
572 	if (sf != NULL)
573 		sf_buf_free(sf);
574 	sched_unpin();
575 	if (op == MD_MALLOC_MOVE_CMP && error != 0) {
576 		*mp = mp1;
577 		*ma_offs = ma_offs1;
578 	}
579 	return (error);
580 }
581 
582 static int
md_malloc_move_vlist(bus_dma_segment_t ** pvlist,int * pma_offs,unsigned len,void * ptr,u_char fill,int op)583 md_malloc_move_vlist(bus_dma_segment_t **pvlist, int *pma_offs,
584     unsigned len, void *ptr, u_char fill, int op)
585 {
586 	bus_dma_segment_t *vlist;
587 	uint8_t *p, *end, first;
588 	off_t *uc;
589 	int ma_offs, seg_len;
590 
591 	vlist = *pvlist;
592 	ma_offs = *pma_offs;
593 	uc = ptr;
594 
595 	for (; len != 0; len -= seg_len) {
596 		seg_len = imin(vlist->ds_len - ma_offs, len);
597 		p = (uint8_t *)(uintptr_t)vlist->ds_addr + ma_offs;
598 		switch (op) {
599 		case MD_MALLOC_MOVE_ZERO:
600 			bzero(p, seg_len);
601 			break;
602 		case MD_MALLOC_MOVE_FILL:
603 			memset(p, fill, seg_len);
604 			break;
605 		case MD_MALLOC_MOVE_READ:
606 			bcopy(ptr, p, seg_len);
607 			cpu_flush_dcache(p, seg_len);
608 			break;
609 		case MD_MALLOC_MOVE_WRITE:
610 			bcopy(p, ptr, seg_len);
611 			break;
612 		case MD_MALLOC_MOVE_CMP:
613 			end = p + seg_len;
614 			first = *uc = *p;
615 			/* Confirm all following bytes match the first */
616 			while (++p < end) {
617 				if (*p != first)
618 					return (EDOOFUS);
619 			}
620 			break;
621 		default:
622 			KASSERT(0, ("md_malloc_move_vlist unknown op %d\n", op));
623 			break;
624 		}
625 
626 		ma_offs += seg_len;
627 		if (ma_offs == vlist->ds_len) {
628 			ma_offs = 0;
629 			vlist++;
630 		}
631 		ptr = (uint8_t *)ptr + seg_len;
632 	}
633 	*pvlist = vlist;
634 	*pma_offs = ma_offs;
635 
636 	return (0);
637 }
638 
639 static int
mdstart_malloc(struct md_s * sc,struct bio * bp)640 mdstart_malloc(struct md_s *sc, struct bio *bp)
641 {
642 	u_char *dst;
643 	vm_page_t *m;
644 	bus_dma_segment_t *vlist;
645 	int i, error, error1, ma_offs, notmapped;
646 	off_t secno, nsec, uc;
647 	uintptr_t sp, osp;
648 
649 	switch (bp->bio_cmd) {
650 	case BIO_READ:
651 	case BIO_WRITE:
652 	case BIO_DELETE:
653 		break;
654 	default:
655 		return (EOPNOTSUPP);
656 	}
657 
658 	notmapped = (bp->bio_flags & BIO_UNMAPPED) != 0;
659 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
660 	    (bus_dma_segment_t *)bp->bio_data : NULL;
661 	if (notmapped) {
662 		m = bp->bio_ma;
663 		ma_offs = bp->bio_ma_offset;
664 		dst = NULL;
665 		KASSERT(vlist == NULL, ("vlists cannot be unmapped"));
666 	} else if (vlist != NULL) {
667 		ma_offs = bp->bio_ma_offset;
668 		dst = NULL;
669 	} else {
670 		dst = bp->bio_data;
671 	}
672 
673 	nsec = bp->bio_length / sc->sectorsize;
674 	secno = bp->bio_offset / sc->sectorsize;
675 	error = 0;
676 	while (nsec--) {
677 		osp = s_read(sc->indir, secno);
678 		if (bp->bio_cmd == BIO_DELETE) {
679 			if (osp != 0)
680 				error = s_write(sc->indir, secno, 0);
681 		} else if (bp->bio_cmd == BIO_READ) {
682 			if (osp == 0) {
683 				if (notmapped) {
684 					error = md_malloc_move_ma(&m, &ma_offs,
685 					    sc->sectorsize, NULL, 0,
686 					    MD_MALLOC_MOVE_ZERO);
687 				} else if (vlist != NULL) {
688 					error = md_malloc_move_vlist(&vlist,
689 					    &ma_offs, sc->sectorsize, NULL, 0,
690 					    MD_MALLOC_MOVE_ZERO);
691 				} else
692 					bzero(dst, sc->sectorsize);
693 			} else if (osp <= 255) {
694 				if (notmapped) {
695 					error = md_malloc_move_ma(&m, &ma_offs,
696 					    sc->sectorsize, NULL, osp,
697 					    MD_MALLOC_MOVE_FILL);
698 				} else if (vlist != NULL) {
699 					error = md_malloc_move_vlist(&vlist,
700 					    &ma_offs, sc->sectorsize, NULL, osp,
701 					    MD_MALLOC_MOVE_FILL);
702 				} else
703 					memset(dst, osp, sc->sectorsize);
704 			} else {
705 				if (notmapped) {
706 					error = md_malloc_move_ma(&m, &ma_offs,
707 					    sc->sectorsize, (void *)osp, 0,
708 					    MD_MALLOC_MOVE_READ);
709 				} else if (vlist != NULL) {
710 					error = md_malloc_move_vlist(&vlist,
711 					    &ma_offs, sc->sectorsize,
712 					    (void *)osp, 0,
713 					    MD_MALLOC_MOVE_READ);
714 				} else {
715 					bcopy((void *)osp, dst, sc->sectorsize);
716 					cpu_flush_dcache(dst, sc->sectorsize);
717 				}
718 			}
719 			osp = 0;
720 		} else if (bp->bio_cmd == BIO_WRITE) {
721 			if (sc->flags & MD_COMPRESS) {
722 				if (notmapped) {
723 					error1 = md_malloc_move_ma(&m, &ma_offs,
724 					    sc->sectorsize, &uc, 0,
725 					    MD_MALLOC_MOVE_CMP);
726 					i = error1 == 0 ? sc->sectorsize : 0;
727 				} else if (vlist != NULL) {
728 					error1 = md_malloc_move_vlist(&vlist,
729 					    &ma_offs, sc->sectorsize, &uc, 0,
730 					    MD_MALLOC_MOVE_CMP);
731 					i = error1 == 0 ? sc->sectorsize : 0;
732 				} else {
733 					uc = dst[0];
734 					for (i = 1; i < sc->sectorsize; i++) {
735 						if (dst[i] != uc)
736 							break;
737 					}
738 				}
739 			} else {
740 				i = 0;
741 				uc = 0;
742 			}
743 			if (i == sc->sectorsize) {
744 				if (osp != uc)
745 					error = s_write(sc->indir, secno, uc);
746 			} else {
747 				if (osp <= 255) {
748 					sp = (uintptr_t)uma_zalloc(sc->uma,
749 					    md_malloc_wait ? M_WAITOK :
750 					    M_NOWAIT);
751 					if (sp == 0) {
752 						error = ENOSPC;
753 						break;
754 					}
755 					if (notmapped) {
756 						error = md_malloc_move_ma(&m,
757 						    &ma_offs, sc->sectorsize,
758 						    (void *)sp, 0,
759 						    MD_MALLOC_MOVE_WRITE);
760 					} else if (vlist != NULL) {
761 						error = md_malloc_move_vlist(
762 						    &vlist, &ma_offs,
763 						    sc->sectorsize, (void *)sp,
764 						    0, MD_MALLOC_MOVE_WRITE);
765 					} else {
766 						bcopy(dst, (void *)sp,
767 						    sc->sectorsize);
768 					}
769 					error = s_write(sc->indir, secno, sp);
770 				} else {
771 					if (notmapped) {
772 						error = md_malloc_move_ma(&m,
773 						    &ma_offs, sc->sectorsize,
774 						    (void *)osp, 0,
775 						    MD_MALLOC_MOVE_WRITE);
776 					} else if (vlist != NULL) {
777 						error = md_malloc_move_vlist(
778 						    &vlist, &ma_offs,
779 						    sc->sectorsize, (void *)osp,
780 						    0, MD_MALLOC_MOVE_WRITE);
781 					} else {
782 						bcopy(dst, (void *)osp,
783 						    sc->sectorsize);
784 					}
785 					osp = 0;
786 				}
787 			}
788 		} else {
789 			error = EOPNOTSUPP;
790 		}
791 		if (osp > 255)
792 			uma_zfree(sc->uma, (void*)osp);
793 		if (error != 0)
794 			break;
795 		secno++;
796 		if (!notmapped && vlist == NULL)
797 			dst += sc->sectorsize;
798 	}
799 	bp->bio_resid = 0;
800 	return (error);
801 }
802 
803 static void
mdcopyto_vlist(void * src,bus_dma_segment_t * vlist,off_t offset,off_t len)804 mdcopyto_vlist(void *src, bus_dma_segment_t *vlist, off_t offset, off_t len)
805 {
806 	off_t seg_len;
807 
808 	while (offset >= vlist->ds_len) {
809 		offset -= vlist->ds_len;
810 		vlist++;
811 	}
812 
813 	while (len != 0) {
814 		seg_len = omin(len, vlist->ds_len - offset);
815 		bcopy(src, (void *)(uintptr_t)(vlist->ds_addr + offset),
816 		    seg_len);
817 		offset = 0;
818 		src = (uint8_t *)src + seg_len;
819 		len -= seg_len;
820 		vlist++;
821 	}
822 }
823 
824 static void
mdcopyfrom_vlist(bus_dma_segment_t * vlist,off_t offset,void * dst,off_t len)825 mdcopyfrom_vlist(bus_dma_segment_t *vlist, off_t offset, void *dst, off_t len)
826 {
827 	off_t seg_len;
828 
829 	while (offset >= vlist->ds_len) {
830 		offset -= vlist->ds_len;
831 		vlist++;
832 	}
833 
834 	while (len != 0) {
835 		seg_len = omin(len, vlist->ds_len - offset);
836 		bcopy((void *)(uintptr_t)(vlist->ds_addr + offset), dst,
837 		    seg_len);
838 		offset = 0;
839 		dst = (uint8_t *)dst + seg_len;
840 		len -= seg_len;
841 		vlist++;
842 	}
843 }
844 
845 static int
mdstart_preload(struct md_s * sc,struct bio * bp)846 mdstart_preload(struct md_s *sc, struct bio *bp)
847 {
848 	uint8_t *p;
849 
850 	p = sc->pl_ptr + bp->bio_offset;
851 	switch (bp->bio_cmd) {
852 	case BIO_READ:
853 		if ((bp->bio_flags & BIO_VLIST) != 0) {
854 			mdcopyto_vlist(p, (bus_dma_segment_t *)bp->bio_data,
855 			    bp->bio_ma_offset, bp->bio_length);
856 		} else {
857 			bcopy(p, bp->bio_data, bp->bio_length);
858 		}
859 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
860 		break;
861 	case BIO_WRITE:
862 		if ((bp->bio_flags & BIO_VLIST) != 0) {
863 			mdcopyfrom_vlist((bus_dma_segment_t *)bp->bio_data,
864 			    bp->bio_ma_offset, p, bp->bio_length);
865 		} else {
866 			bcopy(bp->bio_data, p, bp->bio_length);
867 		}
868 		break;
869 	}
870 	bp->bio_resid = 0;
871 	return (0);
872 }
873 
874 static int
mdstart_vnode(struct md_s * sc,struct bio * bp)875 mdstart_vnode(struct md_s *sc, struct bio *bp)
876 {
877 	int error;
878 	struct uio auio;
879 	struct iovec aiov;
880 	struct iovec *piov;
881 	struct mount *mp;
882 	struct vnode *vp;
883 	struct buf *pb;
884 	bus_dma_segment_t *vlist;
885 	struct thread *td;
886 	off_t iolen, len, zerosize;
887 	int ma_offs, npages;
888 
889 	switch (bp->bio_cmd) {
890 	case BIO_READ:
891 		auio.uio_rw = UIO_READ;
892 		break;
893 	case BIO_WRITE:
894 	case BIO_DELETE:
895 		auio.uio_rw = UIO_WRITE;
896 		break;
897 	case BIO_FLUSH:
898 		break;
899 	default:
900 		return (EOPNOTSUPP);
901 	}
902 
903 	td = curthread;
904 	vp = sc->vnode;
905 	pb = NULL;
906 	piov = NULL;
907 	ma_offs = bp->bio_ma_offset;
908 	len = bp->bio_length;
909 
910 	/*
911 	 * VNODE I/O
912 	 *
913 	 * If an error occurs, we set BIO_ERROR but we do not set
914 	 * B_INVAL because (for a write anyway), the buffer is
915 	 * still valid.
916 	 */
917 
918 	if (bp->bio_cmd == BIO_FLUSH) {
919 		(void) vn_start_write(vp, &mp, V_WAIT);
920 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
921 		error = VOP_FSYNC(vp, MNT_WAIT, td);
922 		VOP_UNLOCK(vp, 0);
923 		vn_finished_write(mp);
924 		return (error);
925 	}
926 
927 	auio.uio_offset = (vm_ooffset_t)bp->bio_offset;
928 	auio.uio_resid = bp->bio_length;
929 	auio.uio_segflg = UIO_SYSSPACE;
930 	auio.uio_td = td;
931 
932 	if (bp->bio_cmd == BIO_DELETE) {
933 		/*
934 		 * Emulate BIO_DELETE by writing zeros.
935 		 */
936 		zerosize = ZERO_REGION_SIZE -
937 		    (ZERO_REGION_SIZE % sc->sectorsize);
938 		auio.uio_iovcnt = howmany(bp->bio_length, zerosize);
939 		piov = malloc(sizeof(*piov) * auio.uio_iovcnt, M_MD, M_WAITOK);
940 		auio.uio_iov = piov;
941 		while (len > 0) {
942 			piov->iov_base = __DECONST(void *, zero_region);
943 			piov->iov_len = len;
944 			if (len > zerosize)
945 				piov->iov_len = zerosize;
946 			len -= piov->iov_len;
947 			piov++;
948 		}
949 		piov = auio.uio_iov;
950 	} else if ((bp->bio_flags & BIO_VLIST) != 0) {
951 		piov = malloc(sizeof(*piov) * bp->bio_ma_n, M_MD, M_WAITOK);
952 		auio.uio_iov = piov;
953 		vlist = (bus_dma_segment_t *)bp->bio_data;
954 		while (len > 0) {
955 			piov->iov_base = (void *)(uintptr_t)(vlist->ds_addr +
956 			    ma_offs);
957 			piov->iov_len = vlist->ds_len - ma_offs;
958 			if (piov->iov_len > len)
959 				piov->iov_len = len;
960 			len -= piov->iov_len;
961 			ma_offs = 0;
962 			vlist++;
963 			piov++;
964 		}
965 		auio.uio_iovcnt = piov - auio.uio_iov;
966 		piov = auio.uio_iov;
967 	} else if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
968 		pb = getpbuf(&md_vnode_pbuf_freecnt);
969 		bp->bio_resid = len;
970 unmapped_step:
971 		npages = atop(min(MAXPHYS, round_page(len + (ma_offs &
972 		    PAGE_MASK))));
973 		iolen = min(ptoa(npages) - (ma_offs & PAGE_MASK), len);
974 		KASSERT(iolen > 0, ("zero iolen"));
975 		pmap_qenter((vm_offset_t)pb->b_data,
976 		    &bp->bio_ma[atop(ma_offs)], npages);
977 		aiov.iov_base = (void *)((vm_offset_t)pb->b_data +
978 		    (ma_offs & PAGE_MASK));
979 		aiov.iov_len = iolen;
980 		auio.uio_iov = &aiov;
981 		auio.uio_iovcnt = 1;
982 		auio.uio_resid = iolen;
983 	} else {
984 		aiov.iov_base = bp->bio_data;
985 		aiov.iov_len = bp->bio_length;
986 		auio.uio_iov = &aiov;
987 		auio.uio_iovcnt = 1;
988 	}
989 	/*
990 	 * When reading set IO_DIRECT to try to avoid double-caching
991 	 * the data.  When writing IO_DIRECT is not optimal.
992 	 */
993 	if (auio.uio_rw == UIO_READ) {
994 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
995 		error = VOP_READ(vp, &auio, IO_DIRECT, sc->cred);
996 		VOP_UNLOCK(vp, 0);
997 	} else {
998 		(void) vn_start_write(vp, &mp, V_WAIT);
999 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1000 		error = VOP_WRITE(vp, &auio, sc->flags & MD_ASYNC ? 0 : IO_SYNC,
1001 		    sc->cred);
1002 		VOP_UNLOCK(vp, 0);
1003 		vn_finished_write(mp);
1004 		if (error == 0)
1005 			sc->flags &= ~MD_VERIFY;
1006 	}
1007 
1008 	if (pb != NULL) {
1009 		pmap_qremove((vm_offset_t)pb->b_data, npages);
1010 		if (error == 0) {
1011 			len -= iolen;
1012 			bp->bio_resid -= iolen;
1013 			ma_offs += iolen;
1014 			if (len > 0)
1015 				goto unmapped_step;
1016 		}
1017 		relpbuf(pb, &md_vnode_pbuf_freecnt);
1018 	}
1019 
1020 	free(piov, M_MD);
1021 	if (pb == NULL)
1022 		bp->bio_resid = auio.uio_resid;
1023 	return (error);
1024 }
1025 
1026 static void
md_swap_page_free(vm_page_t m)1027 md_swap_page_free(vm_page_t m)
1028 {
1029 
1030 	vm_page_xunbusy(m);
1031 	vm_page_lock(m);
1032 	vm_page_free(m);
1033 	vm_page_unlock(m);
1034 }
1035 
1036 static int
mdstart_swap(struct md_s * sc,struct bio * bp)1037 mdstart_swap(struct md_s *sc, struct bio *bp)
1038 {
1039 	vm_page_t m;
1040 	u_char *p;
1041 	vm_pindex_t i, lastp;
1042 	bus_dma_segment_t *vlist;
1043 	int rv, ma_offs, offs, len, lastend;
1044 
1045 	switch (bp->bio_cmd) {
1046 	case BIO_READ:
1047 	case BIO_WRITE:
1048 	case BIO_DELETE:
1049 		break;
1050 	default:
1051 		return (EOPNOTSUPP);
1052 	}
1053 
1054 	p = bp->bio_data;
1055 	ma_offs = (bp->bio_flags & (BIO_UNMAPPED|BIO_VLIST)) != 0 ?
1056 	    bp->bio_ma_offset : 0;
1057 	vlist = (bp->bio_flags & BIO_VLIST) != 0 ?
1058 	    (bus_dma_segment_t *)bp->bio_data : NULL;
1059 
1060 	/*
1061 	 * offs is the offset at which to start operating on the
1062 	 * next (ie, first) page.  lastp is the last page on
1063 	 * which we're going to operate.  lastend is the ending
1064 	 * position within that last page (ie, PAGE_SIZE if
1065 	 * we're operating on complete aligned pages).
1066 	 */
1067 	offs = bp->bio_offset % PAGE_SIZE;
1068 	lastp = (bp->bio_offset + bp->bio_length - 1) / PAGE_SIZE;
1069 	lastend = (bp->bio_offset + bp->bio_length - 1) % PAGE_SIZE + 1;
1070 
1071 	rv = VM_PAGER_OK;
1072 	VM_OBJECT_WLOCK(sc->object);
1073 	vm_object_pip_add(sc->object, 1);
1074 	for (i = bp->bio_offset / PAGE_SIZE; i <= lastp; i++) {
1075 		len = ((i == lastp) ? lastend : PAGE_SIZE) - offs;
1076 		m = vm_page_grab(sc->object, i, VM_ALLOC_SYSTEM);
1077 		if (bp->bio_cmd == BIO_READ) {
1078 			if (m->valid == VM_PAGE_BITS_ALL)
1079 				rv = VM_PAGER_OK;
1080 			else
1081 				rv = vm_pager_get_pages(sc->object, &m, 1,
1082 				    NULL, NULL);
1083 			if (rv == VM_PAGER_ERROR) {
1084 				md_swap_page_free(m);
1085 				break;
1086 			} else if (rv == VM_PAGER_FAIL) {
1087 				/*
1088 				 * Pager does not have the page.  Zero
1089 				 * the allocated page, and mark it as
1090 				 * valid. Do not set dirty, the page
1091 				 * can be recreated if thrown out.
1092 				 */
1093 				pmap_zero_page(m);
1094 				m->valid = VM_PAGE_BITS_ALL;
1095 			}
1096 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1097 				pmap_copy_pages(&m, offs, bp->bio_ma,
1098 				    ma_offs, len);
1099 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1100 				physcopyout_vlist(VM_PAGE_TO_PHYS(m) + offs,
1101 				    vlist, ma_offs, len);
1102 				cpu_flush_dcache(p, len);
1103 			} else {
1104 				physcopyout(VM_PAGE_TO_PHYS(m) + offs, p, len);
1105 				cpu_flush_dcache(p, len);
1106 			}
1107 		} else if (bp->bio_cmd == BIO_WRITE) {
1108 			if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1109 				rv = VM_PAGER_OK;
1110 			else
1111 				rv = vm_pager_get_pages(sc->object, &m, 1,
1112 				    NULL, NULL);
1113 			if (rv == VM_PAGER_ERROR) {
1114 				md_swap_page_free(m);
1115 				break;
1116 			} else if (rv == VM_PAGER_FAIL)
1117 				pmap_zero_page(m);
1118 
1119 			if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
1120 				pmap_copy_pages(bp->bio_ma, ma_offs, &m,
1121 				    offs, len);
1122 			} else if ((bp->bio_flags & BIO_VLIST) != 0) {
1123 				physcopyin_vlist(vlist, ma_offs,
1124 				    VM_PAGE_TO_PHYS(m) + offs, len);
1125 			} else {
1126 				physcopyin(p, VM_PAGE_TO_PHYS(m) + offs, len);
1127 			}
1128 
1129 			m->valid = VM_PAGE_BITS_ALL;
1130 			if (m->dirty != VM_PAGE_BITS_ALL) {
1131 				vm_page_dirty(m);
1132 				vm_pager_page_unswapped(m);
1133 			}
1134 		} else if (bp->bio_cmd == BIO_DELETE) {
1135 			if (len == PAGE_SIZE || m->valid == VM_PAGE_BITS_ALL)
1136 				rv = VM_PAGER_OK;
1137 			else
1138 				rv = vm_pager_get_pages(sc->object, &m, 1,
1139 				    NULL, NULL);
1140 			if (rv == VM_PAGER_ERROR) {
1141 				md_swap_page_free(m);
1142 				break;
1143 			} else if (rv == VM_PAGER_FAIL) {
1144 				md_swap_page_free(m);
1145 				m = NULL;
1146 			} else {
1147 				/* Page is valid. */
1148 				if (len != PAGE_SIZE) {
1149 					pmap_zero_page_area(m, offs, len);
1150 					if (m->dirty != VM_PAGE_BITS_ALL) {
1151 						vm_page_dirty(m);
1152 						vm_pager_page_unswapped(m);
1153 					}
1154 				} else {
1155 					vm_pager_page_unswapped(m);
1156 					md_swap_page_free(m);
1157 					m = NULL;
1158 				}
1159 			}
1160 		}
1161 		if (m != NULL) {
1162 			vm_page_xunbusy(m);
1163 			vm_page_lock(m);
1164 			if (vm_page_active(m))
1165 				vm_page_reference(m);
1166 			else
1167 				vm_page_activate(m);
1168 			vm_page_unlock(m);
1169 		}
1170 
1171 		/* Actions on further pages start at offset 0 */
1172 		p += PAGE_SIZE - offs;
1173 		offs = 0;
1174 		ma_offs += len;
1175 	}
1176 	vm_object_pip_wakeup(sc->object);
1177 	VM_OBJECT_WUNLOCK(sc->object);
1178 	return (rv != VM_PAGER_ERROR ? 0 : ENOSPC);
1179 }
1180 
1181 static int
mdstart_null(struct md_s * sc,struct bio * bp)1182 mdstart_null(struct md_s *sc, struct bio *bp)
1183 {
1184 
1185 	switch (bp->bio_cmd) {
1186 	case BIO_READ:
1187 		bzero(bp->bio_data, bp->bio_length);
1188 		cpu_flush_dcache(bp->bio_data, bp->bio_length);
1189 		break;
1190 	case BIO_WRITE:
1191 		break;
1192 	}
1193 	bp->bio_resid = 0;
1194 	return (0);
1195 }
1196 
1197 static void
md_kthread(void * arg)1198 md_kthread(void *arg)
1199 {
1200 	struct md_s *sc;
1201 	struct bio *bp;
1202 	int error;
1203 
1204 	sc = arg;
1205 	thread_lock(curthread);
1206 	sched_prio(curthread, PRIBIO);
1207 	thread_unlock(curthread);
1208 	if (sc->type == MD_VNODE)
1209 		curthread->td_pflags |= TDP_NORUNNINGBUF;
1210 
1211 	for (;;) {
1212 		mtx_lock(&sc->queue_mtx);
1213 		if (sc->flags & MD_SHUTDOWN) {
1214 			sc->flags |= MD_EXITING;
1215 			mtx_unlock(&sc->queue_mtx);
1216 			kproc_exit(0);
1217 		}
1218 		bp = bioq_takefirst(&sc->bio_queue);
1219 		if (!bp) {
1220 			msleep(sc, &sc->queue_mtx, PRIBIO | PDROP, "mdwait", 0);
1221 			continue;
1222 		}
1223 		mtx_unlock(&sc->queue_mtx);
1224 		if (bp->bio_cmd == BIO_GETATTR) {
1225 			int isv = ((sc->flags & MD_VERIFY) != 0);
1226 
1227 			if ((sc->fwsectors && sc->fwheads &&
1228 			    (g_handleattr_int(bp, "GEOM::fwsectors",
1229 			    sc->fwsectors) ||
1230 			    g_handleattr_int(bp, "GEOM::fwheads",
1231 			    sc->fwheads))) ||
1232 			    g_handleattr_int(bp, "GEOM::candelete", 1))
1233 				error = -1;
1234 			else if (sc->ident[0] != '\0' &&
1235 			    g_handleattr_str(bp, "GEOM::ident", sc->ident))
1236 				error = -1;
1237 			else if (g_handleattr_int(bp, "MNT::verified", isv))
1238 				error = -1;
1239 			else
1240 				error = EOPNOTSUPP;
1241 		} else {
1242 			error = sc->start(sc, bp);
1243 		}
1244 
1245 		if (error != -1) {
1246 			bp->bio_completed = bp->bio_length;
1247 			if ((bp->bio_cmd == BIO_READ) || (bp->bio_cmd == BIO_WRITE))
1248 				devstat_end_transaction_bio(sc->devstat, bp);
1249 			g_io_deliver(bp, error);
1250 		}
1251 	}
1252 }
1253 
1254 static struct md_s *
mdfind(int unit)1255 mdfind(int unit)
1256 {
1257 	struct md_s *sc;
1258 
1259 	LIST_FOREACH(sc, &md_softc_list, list) {
1260 		if (sc->unit == unit)
1261 			break;
1262 	}
1263 	return (sc);
1264 }
1265 
1266 static struct md_s *
mdnew(int unit,int * errp,enum md_types type)1267 mdnew(int unit, int *errp, enum md_types type)
1268 {
1269 	struct md_s *sc;
1270 	int error;
1271 
1272 	*errp = 0;
1273 	if (unit == -1)
1274 		unit = alloc_unr(md_uh);
1275 	else
1276 		unit = alloc_unr_specific(md_uh, unit);
1277 
1278 	if (unit == -1) {
1279 		*errp = EBUSY;
1280 		return (NULL);
1281 	}
1282 
1283 	sc = (struct md_s *)malloc(sizeof *sc, M_MD, M_WAITOK | M_ZERO);
1284 	sc->type = type;
1285 	bioq_init(&sc->bio_queue);
1286 	mtx_init(&sc->queue_mtx, "md bio queue", NULL, MTX_DEF);
1287 	mtx_init(&sc->stat_mtx, "md stat", NULL, MTX_DEF);
1288 	sc->unit = unit;
1289 	sprintf(sc->name, "md%d", unit);
1290 	LIST_INSERT_HEAD(&md_softc_list, sc, list);
1291 	error = kproc_create(md_kthread, sc, &sc->procp, 0, 0,"%s", sc->name);
1292 	if (error == 0)
1293 		return (sc);
1294 	LIST_REMOVE(sc, list);
1295 	mtx_destroy(&sc->stat_mtx);
1296 	mtx_destroy(&sc->queue_mtx);
1297 	free_unr(md_uh, sc->unit);
1298 	free(sc, M_MD);
1299 	*errp = error;
1300 	return (NULL);
1301 }
1302 
1303 static void
mdinit(struct md_s * sc)1304 mdinit(struct md_s *sc)
1305 {
1306 	struct g_geom *gp;
1307 	struct g_provider *pp;
1308 
1309 	g_topology_lock();
1310 	gp = g_new_geomf(&g_md_class, "md%d", sc->unit);
1311 	gp->softc = sc;
1312 	pp = g_new_providerf(gp, "md%d", sc->unit);
1313 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
1314 	pp->mediasize = sc->mediasize;
1315 	pp->sectorsize = sc->sectorsize;
1316 	switch (sc->type) {
1317 	case MD_MALLOC:
1318 	case MD_VNODE:
1319 	case MD_SWAP:
1320 		pp->flags |= G_PF_ACCEPT_UNMAPPED;
1321 		break;
1322 	case MD_PRELOAD:
1323 	case MD_NULL:
1324 		break;
1325 	}
1326 	sc->gp = gp;
1327 	sc->pp = pp;
1328 	g_error_provider(pp, 0);
1329 	g_topology_unlock();
1330 	sc->devstat = devstat_new_entry("md", sc->unit, sc->sectorsize,
1331 	    DEVSTAT_ALL_SUPPORTED, DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
1332 }
1333 
1334 static int
mdcreate_malloc(struct md_s * sc,struct md_req * mdr)1335 mdcreate_malloc(struct md_s *sc, struct md_req *mdr)
1336 {
1337 	uintptr_t sp;
1338 	int error;
1339 	off_t u;
1340 
1341 	error = 0;
1342 	if (mdr->md_options & ~(MD_AUTOUNIT | MD_COMPRESS | MD_RESERVE))
1343 		return (EINVAL);
1344 	if (mdr->md_sectorsize != 0 && !powerof2(mdr->md_sectorsize))
1345 		return (EINVAL);
1346 	/* Compression doesn't make sense if we have reserved space */
1347 	if (mdr->md_options & MD_RESERVE)
1348 		mdr->md_options &= ~MD_COMPRESS;
1349 	if (mdr->md_fwsectors != 0)
1350 		sc->fwsectors = mdr->md_fwsectors;
1351 	if (mdr->md_fwheads != 0)
1352 		sc->fwheads = mdr->md_fwheads;
1353 	sc->flags = mdr->md_options & (MD_COMPRESS | MD_FORCE);
1354 	sc->indir = dimension(sc->mediasize / sc->sectorsize);
1355 	sc->uma = uma_zcreate(sc->name, sc->sectorsize, NULL, NULL, NULL, NULL,
1356 	    0x1ff, 0);
1357 	if (mdr->md_options & MD_RESERVE) {
1358 		off_t nsectors;
1359 
1360 		nsectors = sc->mediasize / sc->sectorsize;
1361 		for (u = 0; u < nsectors; u++) {
1362 			sp = (uintptr_t)uma_zalloc(sc->uma, (md_malloc_wait ?
1363 			    M_WAITOK : M_NOWAIT) | M_ZERO);
1364 			if (sp != 0)
1365 				error = s_write(sc->indir, u, sp);
1366 			else
1367 				error = ENOMEM;
1368 			if (error != 0)
1369 				break;
1370 		}
1371 	}
1372 	return (error);
1373 }
1374 
1375 
1376 static int
mdsetcred(struct md_s * sc,struct ucred * cred)1377 mdsetcred(struct md_s *sc, struct ucred *cred)
1378 {
1379 	char *tmpbuf;
1380 	int error = 0;
1381 
1382 	/*
1383 	 * Set credits in our softc
1384 	 */
1385 
1386 	if (sc->cred)
1387 		crfree(sc->cred);
1388 	sc->cred = crhold(cred);
1389 
1390 	/*
1391 	 * Horrible kludge to establish credentials for NFS  XXX.
1392 	 */
1393 
1394 	if (sc->vnode) {
1395 		struct uio auio;
1396 		struct iovec aiov;
1397 
1398 		tmpbuf = malloc(sc->sectorsize, M_TEMP, M_WAITOK);
1399 		bzero(&auio, sizeof(auio));
1400 
1401 		aiov.iov_base = tmpbuf;
1402 		aiov.iov_len = sc->sectorsize;
1403 		auio.uio_iov = &aiov;
1404 		auio.uio_iovcnt = 1;
1405 		auio.uio_offset = 0;
1406 		auio.uio_rw = UIO_READ;
1407 		auio.uio_segflg = UIO_SYSSPACE;
1408 		auio.uio_resid = aiov.iov_len;
1409 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1410 		error = VOP_READ(sc->vnode, &auio, 0, sc->cred);
1411 		VOP_UNLOCK(sc->vnode, 0);
1412 		free(tmpbuf, M_TEMP);
1413 	}
1414 	return (error);
1415 }
1416 
1417 static int
mdcreate_vnode(struct md_s * sc,struct md_req * mdr,struct thread * td)1418 mdcreate_vnode(struct md_s *sc, struct md_req *mdr, struct thread *td)
1419 {
1420 	struct vattr vattr;
1421 	struct nameidata nd;
1422 	char *fname;
1423 	int error, flags;
1424 
1425 	fname = mdr->md_file;
1426 	if (mdr->md_file_seg == UIO_USERSPACE) {
1427 		error = copyinstr(fname, sc->file, sizeof(sc->file), NULL);
1428 		if (error != 0)
1429 			return (error);
1430 	} else if (mdr->md_file_seg == UIO_SYSSPACE)
1431 		strlcpy(sc->file, fname, sizeof(sc->file));
1432 	else
1433 		return (EDOOFUS);
1434 
1435 	/*
1436 	 * If the user specified that this is a read only device, don't
1437 	 * set the FWRITE mask before trying to open the backing store.
1438 	 */
1439 	flags = FREAD | ((mdr->md_options & MD_READONLY) ? 0 : FWRITE) \
1440 	    | ((mdr->md_options & MD_VERIFY) ? O_VERIFY : 0);
1441 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_SYSSPACE, sc->file, td);
1442 	error = vn_open(&nd, &flags, 0, NULL);
1443 	if (error != 0)
1444 		return (error);
1445 	NDFREE(&nd, NDF_ONLY_PNBUF);
1446 	if (nd.ni_vp->v_type != VREG) {
1447 		error = EINVAL;
1448 		goto bad;
1449 	}
1450 	error = VOP_GETATTR(nd.ni_vp, &vattr, td->td_ucred);
1451 	if (error != 0)
1452 		goto bad;
1453 	if (VOP_ISLOCKED(nd.ni_vp) != LK_EXCLUSIVE) {
1454 		vn_lock(nd.ni_vp, LK_UPGRADE | LK_RETRY);
1455 		if (nd.ni_vp->v_iflag & VI_DOOMED) {
1456 			/* Forced unmount. */
1457 			error = EBADF;
1458 			goto bad;
1459 		}
1460 	}
1461 	nd.ni_vp->v_vflag |= VV_MD;
1462 	VOP_UNLOCK(nd.ni_vp, 0);
1463 
1464 	if (mdr->md_fwsectors != 0)
1465 		sc->fwsectors = mdr->md_fwsectors;
1466 	if (mdr->md_fwheads != 0)
1467 		sc->fwheads = mdr->md_fwheads;
1468 	snprintf(sc->ident, sizeof(sc->ident), "MD-DEV%ju-INO%ju",
1469 	    (uintmax_t)vattr.va_fsid, (uintmax_t)vattr.va_fileid);
1470 	sc->flags = mdr->md_options & (MD_FORCE | MD_ASYNC | MD_VERIFY);
1471 	if (!(flags & FWRITE))
1472 		sc->flags |= MD_READONLY;
1473 	sc->vnode = nd.ni_vp;
1474 
1475 	error = mdsetcred(sc, td->td_ucred);
1476 	if (error != 0) {
1477 		sc->vnode = NULL;
1478 		vn_lock(nd.ni_vp, LK_EXCLUSIVE | LK_RETRY);
1479 		nd.ni_vp->v_vflag &= ~VV_MD;
1480 		goto bad;
1481 	}
1482 	return (0);
1483 bad:
1484 	VOP_UNLOCK(nd.ni_vp, 0);
1485 	(void)vn_close(nd.ni_vp, flags, td->td_ucred, td);
1486 	return (error);
1487 }
1488 
1489 static void
g_md_providergone(struct g_provider * pp)1490 g_md_providergone(struct g_provider *pp)
1491 {
1492 	struct md_s *sc = pp->geom->softc;
1493 
1494 	mtx_lock(&sc->queue_mtx);
1495 	sc->flags |= MD_PROVIDERGONE;
1496 	wakeup(&sc->flags);
1497 	mtx_unlock(&sc->queue_mtx);
1498 }
1499 
1500 static int
mddestroy(struct md_s * sc,struct thread * td)1501 mddestroy(struct md_s *sc, struct thread *td)
1502 {
1503 
1504 	if (sc->gp) {
1505 		g_topology_lock();
1506 		g_wither_geom(sc->gp, ENXIO);
1507 		g_topology_unlock();
1508 
1509 		mtx_lock(&sc->queue_mtx);
1510 		while (!(sc->flags & MD_PROVIDERGONE))
1511 			msleep(&sc->flags, &sc->queue_mtx, PRIBIO, "mddestroy", 0);
1512 		mtx_unlock(&sc->queue_mtx);
1513 	}
1514 	if (sc->devstat) {
1515 		devstat_remove_entry(sc->devstat);
1516 		sc->devstat = NULL;
1517 	}
1518 	mtx_lock(&sc->queue_mtx);
1519 	sc->flags |= MD_SHUTDOWN;
1520 	wakeup(sc);
1521 	while (!(sc->flags & MD_EXITING))
1522 		msleep(sc->procp, &sc->queue_mtx, PRIBIO, "mddestroy", hz / 10);
1523 	mtx_unlock(&sc->queue_mtx);
1524 	mtx_destroy(&sc->stat_mtx);
1525 	mtx_destroy(&sc->queue_mtx);
1526 	if (sc->vnode != NULL) {
1527 		vn_lock(sc->vnode, LK_EXCLUSIVE | LK_RETRY);
1528 		sc->vnode->v_vflag &= ~VV_MD;
1529 		VOP_UNLOCK(sc->vnode, 0);
1530 		(void)vn_close(sc->vnode, sc->flags & MD_READONLY ?
1531 		    FREAD : (FREAD|FWRITE), sc->cred, td);
1532 	}
1533 	if (sc->cred != NULL)
1534 		crfree(sc->cred);
1535 	if (sc->object != NULL)
1536 		vm_object_deallocate(sc->object);
1537 	if (sc->indir)
1538 		destroy_indir(sc, sc->indir);
1539 	if (sc->uma)
1540 		uma_zdestroy(sc->uma);
1541 
1542 	LIST_REMOVE(sc, list);
1543 	free_unr(md_uh, sc->unit);
1544 	free(sc, M_MD);
1545 	return (0);
1546 }
1547 
1548 static int
mdresize(struct md_s * sc,struct md_req * mdr)1549 mdresize(struct md_s *sc, struct md_req *mdr)
1550 {
1551 	int error, res;
1552 	vm_pindex_t oldpages, newpages;
1553 
1554 	switch (sc->type) {
1555 	case MD_VNODE:
1556 	case MD_NULL:
1557 		break;
1558 	case MD_SWAP:
1559 		if (mdr->md_mediasize <= 0 ||
1560 		    (mdr->md_mediasize % PAGE_SIZE) != 0)
1561 			return (EDOM);
1562 		oldpages = OFF_TO_IDX(round_page(sc->mediasize));
1563 		newpages = OFF_TO_IDX(round_page(mdr->md_mediasize));
1564 		if (newpages < oldpages) {
1565 			VM_OBJECT_WLOCK(sc->object);
1566 			vm_object_page_remove(sc->object, newpages, 0, 0);
1567 			swap_pager_freespace(sc->object, newpages,
1568 			    oldpages - newpages);
1569 			swap_release_by_cred(IDX_TO_OFF(oldpages -
1570 			    newpages), sc->cred);
1571 			sc->object->charge = IDX_TO_OFF(newpages);
1572 			sc->object->size = newpages;
1573 			VM_OBJECT_WUNLOCK(sc->object);
1574 		} else if (newpages > oldpages) {
1575 			res = swap_reserve_by_cred(IDX_TO_OFF(newpages -
1576 			    oldpages), sc->cred);
1577 			if (!res)
1578 				return (ENOMEM);
1579 			if ((mdr->md_options & MD_RESERVE) ||
1580 			    (sc->flags & MD_RESERVE)) {
1581 				error = swap_pager_reserve(sc->object,
1582 				    oldpages, newpages - oldpages);
1583 				if (error < 0) {
1584 					swap_release_by_cred(
1585 					    IDX_TO_OFF(newpages - oldpages),
1586 					    sc->cred);
1587 					return (EDOM);
1588 				}
1589 			}
1590 			VM_OBJECT_WLOCK(sc->object);
1591 			sc->object->charge = IDX_TO_OFF(newpages);
1592 			sc->object->size = newpages;
1593 			VM_OBJECT_WUNLOCK(sc->object);
1594 		}
1595 		break;
1596 	default:
1597 		return (EOPNOTSUPP);
1598 	}
1599 
1600 	sc->mediasize = mdr->md_mediasize;
1601 	g_topology_lock();
1602 	g_resize_provider(sc->pp, sc->mediasize);
1603 	g_topology_unlock();
1604 	return (0);
1605 }
1606 
1607 static int
mdcreate_swap(struct md_s * sc,struct md_req * mdr,struct thread * td)1608 mdcreate_swap(struct md_s *sc, struct md_req *mdr, struct thread *td)
1609 {
1610 	vm_ooffset_t npage;
1611 	int error;
1612 
1613 	/*
1614 	 * Range check.  Disallow negative sizes and sizes not being
1615 	 * multiple of page size.
1616 	 */
1617 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1618 		return (EDOM);
1619 
1620 	/*
1621 	 * Allocate an OBJT_SWAP object.
1622 	 *
1623 	 * Note the truncation.
1624 	 */
1625 
1626 	if ((mdr->md_options & MD_VERIFY) != 0)
1627 		return (EINVAL);
1628 	npage = mdr->md_mediasize / PAGE_SIZE;
1629 	if (mdr->md_fwsectors != 0)
1630 		sc->fwsectors = mdr->md_fwsectors;
1631 	if (mdr->md_fwheads != 0)
1632 		sc->fwheads = mdr->md_fwheads;
1633 	sc->object = vm_pager_allocate(OBJT_SWAP, NULL, PAGE_SIZE * npage,
1634 	    VM_PROT_DEFAULT, 0, td->td_ucred);
1635 	if (sc->object == NULL)
1636 		return (ENOMEM);
1637 	sc->flags = mdr->md_options & (MD_FORCE | MD_RESERVE);
1638 	if (mdr->md_options & MD_RESERVE) {
1639 		if (swap_pager_reserve(sc->object, 0, npage) < 0) {
1640 			error = EDOM;
1641 			goto finish;
1642 		}
1643 	}
1644 	error = mdsetcred(sc, td->td_ucred);
1645  finish:
1646 	if (error != 0) {
1647 		vm_object_deallocate(sc->object);
1648 		sc->object = NULL;
1649 	}
1650 	return (error);
1651 }
1652 
1653 static int
mdcreate_null(struct md_s * sc,struct md_req * mdr,struct thread * td)1654 mdcreate_null(struct md_s *sc, struct md_req *mdr, struct thread *td)
1655 {
1656 
1657 	/*
1658 	 * Range check.  Disallow negative sizes and sizes not being
1659 	 * multiple of page size.
1660 	 */
1661 	if (sc->mediasize <= 0 || (sc->mediasize % PAGE_SIZE) != 0)
1662 		return (EDOM);
1663 
1664 	return (0);
1665 }
1666 
1667 static int
kern_mdattach_locked(struct thread * td,struct md_req * mdr)1668 kern_mdattach_locked(struct thread *td, struct md_req *mdr)
1669 {
1670 	struct md_s *sc;
1671 	unsigned sectsize;
1672 	int error, i;
1673 
1674 	sx_assert(&md_sx, SA_XLOCKED);
1675 
1676 	switch (mdr->md_type) {
1677 	case MD_MALLOC:
1678 	case MD_PRELOAD:
1679 	case MD_VNODE:
1680 	case MD_SWAP:
1681 	case MD_NULL:
1682 		break;
1683 	default:
1684 		return (EINVAL);
1685 	}
1686 	if (mdr->md_sectorsize == 0)
1687 		sectsize = DEV_BSIZE;
1688 	else
1689 		sectsize = mdr->md_sectorsize;
1690 	if (sectsize > MAXPHYS || mdr->md_mediasize < sectsize)
1691 		return (EINVAL);
1692 	if (mdr->md_options & MD_AUTOUNIT)
1693 		sc = mdnew(-1, &error, mdr->md_type);
1694 	else {
1695 		if (mdr->md_unit > INT_MAX)
1696 			return (EINVAL);
1697 		sc = mdnew(mdr->md_unit, &error, mdr->md_type);
1698 	}
1699 	if (sc == NULL)
1700 		return (error);
1701 	if (mdr->md_label != NULL)
1702 		error = copyinstr(mdr->md_label, sc->label,
1703 		    sizeof(sc->label), NULL);
1704 	if (error != 0)
1705 		goto err_after_new;
1706 	if (mdr->md_options & MD_AUTOUNIT)
1707 		mdr->md_unit = sc->unit;
1708 	sc->mediasize = mdr->md_mediasize;
1709 	sc->sectorsize = sectsize;
1710 	error = EDOOFUS;
1711 	switch (sc->type) {
1712 	case MD_MALLOC:
1713 		sc->start = mdstart_malloc;
1714 		error = mdcreate_malloc(sc, mdr);
1715 		break;
1716 	case MD_PRELOAD:
1717 		/*
1718 		 * We disallow attaching preloaded memory disks via
1719 		 * ioctl. Preloaded memory disks are automatically
1720 		 * attached in g_md_init().
1721 		 */
1722 		error = EOPNOTSUPP;
1723 		break;
1724 	case MD_VNODE:
1725 		sc->start = mdstart_vnode;
1726 		error = mdcreate_vnode(sc, mdr, td);
1727 		break;
1728 	case MD_SWAP:
1729 		sc->start = mdstart_swap;
1730 		error = mdcreate_swap(sc, mdr, td);
1731 		break;
1732 	case MD_NULL:
1733 		sc->start = mdstart_null;
1734 		error = mdcreate_null(sc, mdr, td);
1735 		break;
1736 	}
1737 err_after_new:
1738 	if (error != 0) {
1739 		mddestroy(sc, td);
1740 		return (error);
1741 	}
1742 
1743 	/* Prune off any residual fractional sector */
1744 	i = sc->mediasize % sc->sectorsize;
1745 	sc->mediasize -= i;
1746 
1747 	mdinit(sc);
1748 	return (0);
1749 }
1750 
1751 static int
kern_mdattach(struct thread * td,struct md_req * mdr)1752 kern_mdattach(struct thread *td, struct md_req *mdr)
1753 {
1754 	int error;
1755 
1756 	sx_xlock(&md_sx);
1757 	error = kern_mdattach_locked(td, mdr);
1758 	sx_xunlock(&md_sx);
1759 	return (error);
1760 }
1761 
1762 static int
kern_mddetach_locked(struct thread * td,struct md_req * mdr)1763 kern_mddetach_locked(struct thread *td, struct md_req *mdr)
1764 {
1765 	struct md_s *sc;
1766 
1767 	sx_assert(&md_sx, SA_XLOCKED);
1768 
1769 	if (mdr->md_mediasize != 0 ||
1770 	    (mdr->md_options & ~MD_FORCE) != 0)
1771 		return (EINVAL);
1772 
1773 	sc = mdfind(mdr->md_unit);
1774 	if (sc == NULL)
1775 		return (ENOENT);
1776 	if (sc->opencount != 0 && !(sc->flags & MD_FORCE) &&
1777 	    !(mdr->md_options & MD_FORCE))
1778 		return (EBUSY);
1779 	return (mddestroy(sc, td));
1780 }
1781 
1782 static int
kern_mddetach(struct thread * td,struct md_req * mdr)1783 kern_mddetach(struct thread *td, struct md_req *mdr)
1784 {
1785 	int error;
1786 
1787 	sx_xlock(&md_sx);
1788 	error = kern_mddetach_locked(td, mdr);
1789 	sx_xunlock(&md_sx);
1790 	return (error);
1791 }
1792 
1793 static int
kern_mdresize_locked(struct md_req * mdr)1794 kern_mdresize_locked(struct md_req *mdr)
1795 {
1796 	struct md_s *sc;
1797 
1798 	sx_assert(&md_sx, SA_XLOCKED);
1799 
1800 	if ((mdr->md_options & ~(MD_FORCE | MD_RESERVE)) != 0)
1801 		return (EINVAL);
1802 
1803 	sc = mdfind(mdr->md_unit);
1804 	if (sc == NULL)
1805 		return (ENOENT);
1806 	if (mdr->md_mediasize < sc->sectorsize)
1807 		return (EINVAL);
1808 	if (mdr->md_mediasize < sc->mediasize &&
1809 	    !(sc->flags & MD_FORCE) &&
1810 	    !(mdr->md_options & MD_FORCE))
1811 		return (EBUSY);
1812 	return (mdresize(sc, mdr));
1813 }
1814 
1815 static int
kern_mdresize(struct md_req * mdr)1816 kern_mdresize(struct md_req *mdr)
1817 {
1818 	int error;
1819 
1820 	sx_xlock(&md_sx);
1821 	error = kern_mdresize_locked(mdr);
1822 	sx_xunlock(&md_sx);
1823 	return (error);
1824 }
1825 
1826 static int
kern_mdquery_locked(struct md_req * mdr)1827 kern_mdquery_locked(struct md_req *mdr)
1828 {
1829 	struct md_s *sc;
1830 	int error;
1831 
1832 	sx_assert(&md_sx, SA_XLOCKED);
1833 
1834 	sc = mdfind(mdr->md_unit);
1835 	if (sc == NULL)
1836 		return (ENOENT);
1837 	mdr->md_type = sc->type;
1838 	mdr->md_options = sc->flags;
1839 	mdr->md_mediasize = sc->mediasize;
1840 	mdr->md_sectorsize = sc->sectorsize;
1841 	error = 0;
1842 	if (mdr->md_label != NULL) {
1843 		error = copyout(sc->label, mdr->md_label,
1844 		    strlen(sc->label) + 1);
1845 		if (error != 0)
1846 			return (error);
1847 	}
1848 	if (sc->type == MD_VNODE ||
1849 	    (sc->type == MD_PRELOAD && mdr->md_file != NULL))
1850 		error = copyout(sc->file, mdr->md_file,
1851 		    strlen(sc->file) + 1);
1852 	return (error);
1853 }
1854 
1855 static int
kern_mdquery(struct md_req * mdr)1856 kern_mdquery(struct md_req *mdr)
1857 {
1858 	int error;
1859 
1860 	sx_xlock(&md_sx);
1861 	error = kern_mdquery_locked(mdr);
1862 	sx_xunlock(&md_sx);
1863 	return (error);
1864 }
1865 
1866 static int
kern_mdlist_locked(struct md_req * mdr)1867 kern_mdlist_locked(struct md_req *mdr)
1868 {
1869 	struct md_s *sc;
1870 	int i;
1871 
1872 	sx_assert(&md_sx, SA_XLOCKED);
1873 
1874 	/*
1875 	 * Write the number of md devices to mdr->md_units[0].
1876 	 * Write the unit number of the first (mdr->md_units_nitems - 2)
1877 	 * units to mdr->md_units[1::(mdr->md_units - 2)] and terminate the
1878 	 * list with -1.
1879 	 *
1880 	 * XXX: There is currently no mechanism to retrieve unit
1881 	 * numbers for more than (MDNPAD - 2) units.
1882 	 *
1883 	 * XXX: Due to the use of LIST_INSERT_HEAD in mdnew(), the
1884 	 * list of visible unit numbers not stable.
1885 	 */
1886 	i = 1;
1887 	LIST_FOREACH(sc, &md_softc_list, list) {
1888 		if (i < mdr->md_units_nitems - 1)
1889 			mdr->md_units[i] = sc->unit;
1890 		i++;
1891 	}
1892 	mdr->md_units[MIN(i, mdr->md_units_nitems - 1)] = -1;
1893 	mdr->md_units[0] = i - 1;
1894 	return (0);
1895 }
1896 
1897 static int
kern_mdlist(struct md_req * mdr)1898 kern_mdlist(struct md_req *mdr)
1899 {
1900 	int error;
1901 
1902 	sx_xlock(&md_sx);
1903 	error = kern_mdlist_locked(mdr);
1904 	sx_xunlock(&md_sx);
1905 	return (error);
1906 }
1907 
1908 /* Copy members that are not userspace pointers. */
1909 #define	MD_IOCTL2REQ(mdio, mdr) do {					\
1910 	(mdr)->md_unit = (mdio)->md_unit;				\
1911 	(mdr)->md_type = (mdio)->md_type;				\
1912 	(mdr)->md_mediasize = (mdio)->md_mediasize;			\
1913 	(mdr)->md_sectorsize = (mdio)->md_sectorsize;			\
1914 	(mdr)->md_options = (mdio)->md_options;				\
1915 	(mdr)->md_fwheads = (mdio)->md_fwheads;				\
1916 	(mdr)->md_fwsectors = (mdio)->md_fwsectors;			\
1917 	(mdr)->md_units = &(mdio)->md_pad[0];				\
1918 	(mdr)->md_units_nitems = nitems((mdio)->md_pad);		\
1919 } while(0)
1920 
1921 /* Copy members that might have been updated */
1922 #define MD_REQ2IOCTL(mdr, mdio) do {					\
1923 	(mdio)->md_unit = (mdr)->md_unit;				\
1924 	(mdio)->md_type = (mdr)->md_type;				\
1925 	(mdio)->md_mediasize = (mdr)->md_mediasize;			\
1926 	(mdio)->md_sectorsize = (mdr)->md_sectorsize;			\
1927 	(mdio)->md_options = (mdr)->md_options;				\
1928 	(mdio)->md_fwheads = (mdr)->md_fwheads;				\
1929 	(mdio)->md_fwsectors = (mdr)->md_fwsectors;			\
1930 } while(0)
1931 
1932 static int
mdctlioctl(struct cdev * dev,u_long cmd,caddr_t addr,int flags,struct thread * td)1933 mdctlioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1934     struct thread *td)
1935 {
1936 	struct md_req mdr;
1937 	int error;
1938 
1939 	if (md_debug)
1940 		printf("mdctlioctl(%s %lx %p %x %p)\n",
1941 			devtoname(dev), cmd, addr, flags, td);
1942 
1943 	bzero(&mdr, sizeof(mdr));
1944 	switch (cmd) {
1945 	case MDIOCATTACH:
1946 	case MDIOCDETACH:
1947 	case MDIOCRESIZE:
1948 	case MDIOCQUERY:
1949 	case MDIOCLIST: {
1950 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
1951 		if (mdio->md_version != MDIOVERSION)
1952 			return (EINVAL);
1953 		MD_IOCTL2REQ(mdio, &mdr);
1954 		mdr.md_file = mdio->md_file;
1955 		mdr.md_file_seg = UIO_USERSPACE;
1956 		/* If the file is adjacent to the md_ioctl it's in kernel. */
1957 		if ((void *)mdio->md_file == (void *)(mdio + 1))
1958 			mdr.md_file_seg = UIO_SYSSPACE;
1959 		mdr.md_label = mdio->md_label;
1960 		break;
1961 	}
1962 #ifdef COMPAT_FREEBSD32
1963 	case MDIOCATTACH_32:
1964 	case MDIOCDETACH_32:
1965 	case MDIOCRESIZE_32:
1966 	case MDIOCQUERY_32:
1967 	case MDIOCLIST_32: {
1968 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
1969 		if (mdio->md_version != MDIOVERSION)
1970 			return (EINVAL);
1971 		MD_IOCTL2REQ(mdio, &mdr);
1972 		mdr.md_file = (void *)(uintptr_t)mdio->md_file;
1973 		mdr.md_file_seg = UIO_USERSPACE;
1974 		mdr.md_label = (void *)(uintptr_t)mdio->md_label;
1975 		break;
1976 	}
1977 #endif
1978 	default:
1979 		/* Fall through to handler switch. */
1980 		break;
1981 	}
1982 
1983 	error = 0;
1984 	switch (cmd) {
1985 	case MDIOCATTACH:
1986 #ifdef COMPAT_FREEBSD32
1987 	case MDIOCATTACH_32:
1988 #endif
1989 		error = kern_mdattach(td, &mdr);
1990 		break;
1991 	case MDIOCDETACH:
1992 #ifdef COMPAT_FREEBSD32
1993 	case MDIOCDETACH_32:
1994 #endif
1995 		error = kern_mddetach(td, &mdr);
1996 		break;
1997 	case MDIOCRESIZE:
1998 #ifdef COMPAT_FREEBSD32
1999 	case MDIOCRESIZE_32:
2000 #endif
2001 		error = kern_mdresize(&mdr);
2002 		break;
2003 	case MDIOCQUERY:
2004 #ifdef COMPAT_FREEBSD32
2005 	case MDIOCQUERY_32:
2006 #endif
2007 		error = kern_mdquery(&mdr);
2008 		break;
2009 	case MDIOCLIST:
2010 #ifdef COMPAT_FREEBSD32
2011 	case MDIOCLIST_32:
2012 #endif
2013 		error = kern_mdlist(&mdr);
2014 		break;
2015 	default:
2016 		error = ENOIOCTL;
2017 	}
2018 
2019 	switch (cmd) {
2020 	case MDIOCATTACH:
2021 	case MDIOCQUERY: {
2022 		struct md_ioctl *mdio = (struct md_ioctl *)addr;
2023 		MD_REQ2IOCTL(&mdr, mdio);
2024 		break;
2025 	}
2026 #ifdef COMPAT_FREEBSD32
2027 	case MDIOCATTACH_32:
2028 	case MDIOCQUERY_32: {
2029 		struct md_ioctl32 *mdio = (struct md_ioctl32 *)addr;
2030 		MD_REQ2IOCTL(&mdr, mdio);
2031 		break;
2032 	}
2033 #endif
2034 	default:
2035 		/* Other commands to not alter mdr. */
2036 		break;
2037 	}
2038 
2039 	return (error);
2040 }
2041 
2042 static void
md_preloaded(u_char * image,size_t length,const char * name)2043 md_preloaded(u_char *image, size_t length, const char *name)
2044 {
2045 	struct md_s *sc;
2046 	int error;
2047 
2048 	sc = mdnew(-1, &error, MD_PRELOAD);
2049 	if (sc == NULL)
2050 		return;
2051 	sc->mediasize = length;
2052 	sc->sectorsize = DEV_BSIZE;
2053 	sc->pl_ptr = image;
2054 	sc->pl_len = length;
2055 	sc->start = mdstart_preload;
2056 	if (name != NULL)
2057 		strlcpy(sc->file, name, sizeof(sc->file));
2058 #ifdef MD_ROOT
2059 	if (sc->unit == 0) {
2060 #ifndef ROOTDEVNAME
2061 		rootdevnames[0] = MD_ROOT_FSTYPE ":/dev/md0";
2062 #endif
2063 #ifdef MD_ROOT_READONLY
2064 		sc->flags |= MD_READONLY;
2065 #endif
2066 	}
2067 #endif
2068 	mdinit(sc);
2069 	if (name != NULL) {
2070 		printf("%s%d: Preloaded image <%s> %zd bytes at %p\n",
2071 		    MD_NAME, sc->unit, name, length, image);
2072 	} else {
2073 		printf("%s%d: Embedded image %zd bytes at %p\n",
2074 		    MD_NAME, sc->unit, length, image);
2075 	}
2076 }
2077 
2078 static void
g_md_init(struct g_class * mp __unused)2079 g_md_init(struct g_class *mp __unused)
2080 {
2081 	caddr_t mod;
2082 	u_char *ptr, *name, *type;
2083 	unsigned len;
2084 	int i;
2085 
2086 	/* figure out log2(NINDIR) */
2087 	for (i = NINDIR, nshift = -1; i; nshift++)
2088 		i >>= 1;
2089 
2090 	mod = NULL;
2091 	sx_init(&md_sx, "MD config lock");
2092 	g_topology_unlock();
2093 	md_uh = new_unrhdr(0, INT_MAX, NULL);
2094 #ifdef MD_ROOT
2095 	if (mfs_root_size != 0) {
2096 		sx_xlock(&md_sx);
2097 #ifdef MD_ROOT_MEM
2098 		md_preloaded(mfs_root, mfs_root_size, NULL);
2099 #else
2100 		md_preloaded(__DEVOLATILE(u_char *, &mfs_root), mfs_root_size,
2101 		    NULL);
2102 #endif
2103 		sx_xunlock(&md_sx);
2104 	}
2105 #endif
2106 	/* XXX: are preload_* static or do they need Giant ? */
2107 	while ((mod = preload_search_next_name(mod)) != NULL) {
2108 		name = (char *)preload_search_info(mod, MODINFO_NAME);
2109 		if (name == NULL)
2110 			continue;
2111 		type = (char *)preload_search_info(mod, MODINFO_TYPE);
2112 		if (type == NULL)
2113 			continue;
2114 		if (strcmp(type, "md_image") && strcmp(type, "mfs_root"))
2115 			continue;
2116 		ptr = preload_fetch_addr(mod);
2117 		len = preload_fetch_size(mod);
2118 		if (ptr != NULL && len != 0) {
2119 			sx_xlock(&md_sx);
2120 			md_preloaded(ptr, len, name);
2121 			sx_xunlock(&md_sx);
2122 		}
2123 	}
2124 	md_vnode_pbuf_freecnt = nswbuf / 10;
2125 	status_dev = make_dev(&mdctl_cdevsw, INT_MAX, UID_ROOT, GID_WHEEL,
2126 	    0600, MDCTL_NAME);
2127 	g_topology_lock();
2128 }
2129 
2130 static void
g_md_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp __unused,struct g_provider * pp)2131 g_md_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2132     struct g_consumer *cp __unused, struct g_provider *pp)
2133 {
2134 	struct md_s *mp;
2135 	char *type;
2136 
2137 	mp = gp->softc;
2138 	if (mp == NULL)
2139 		return;
2140 
2141 	switch (mp->type) {
2142 	case MD_MALLOC:
2143 		type = "malloc";
2144 		break;
2145 	case MD_PRELOAD:
2146 		type = "preload";
2147 		break;
2148 	case MD_VNODE:
2149 		type = "vnode";
2150 		break;
2151 	case MD_SWAP:
2152 		type = "swap";
2153 		break;
2154 	case MD_NULL:
2155 		type = "null";
2156 		break;
2157 	default:
2158 		type = "unknown";
2159 		break;
2160 	}
2161 
2162 	if (pp != NULL) {
2163 		if (indent == NULL) {
2164 			sbuf_printf(sb, " u %d", mp->unit);
2165 			sbuf_printf(sb, " s %ju", (uintmax_t) mp->sectorsize);
2166 			sbuf_printf(sb, " f %ju", (uintmax_t) mp->fwheads);
2167 			sbuf_printf(sb, " fs %ju", (uintmax_t) mp->fwsectors);
2168 			sbuf_printf(sb, " l %ju", (uintmax_t) mp->mediasize);
2169 			sbuf_printf(sb, " t %s", type);
2170 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2171 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0'))
2172 				sbuf_printf(sb, " file %s", mp->file);
2173 			sbuf_printf(sb, " label %s", mp->label);
2174 		} else {
2175 			sbuf_printf(sb, "%s<unit>%d</unit>\n", indent,
2176 			    mp->unit);
2177 			sbuf_printf(sb, "%s<sectorsize>%ju</sectorsize>\n",
2178 			    indent, (uintmax_t) mp->sectorsize);
2179 			sbuf_printf(sb, "%s<fwheads>%ju</fwheads>\n",
2180 			    indent, (uintmax_t) mp->fwheads);
2181 			sbuf_printf(sb, "%s<fwsectors>%ju</fwsectors>\n",
2182 			    indent, (uintmax_t) mp->fwsectors);
2183 			if (mp->ident[0] != '\0') {
2184 				sbuf_printf(sb, "%s<ident>", indent);
2185 				g_conf_printf_escaped(sb, "%s", mp->ident);
2186 				sbuf_printf(sb, "</ident>\n");
2187 			}
2188 			sbuf_printf(sb, "%s<length>%ju</length>\n",
2189 			    indent, (uintmax_t) mp->mediasize);
2190 			sbuf_printf(sb, "%s<compression>%s</compression>\n", indent,
2191 			    (mp->flags & MD_COMPRESS) == 0 ? "off": "on");
2192 			sbuf_printf(sb, "%s<access>%s</access>\n", indent,
2193 			    (mp->flags & MD_READONLY) == 0 ? "read-write":
2194 			    "read-only");
2195 			sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2196 			    type);
2197 			if ((mp->type == MD_VNODE && mp->vnode != NULL) ||
2198 			    (mp->type == MD_PRELOAD && mp->file[0] != '\0')) {
2199 				sbuf_printf(sb, "%s<file>", indent);
2200 				g_conf_printf_escaped(sb, "%s", mp->file);
2201 				sbuf_printf(sb, "</file>\n");
2202 			}
2203 			sbuf_printf(sb, "%s<label>", indent);
2204 			g_conf_printf_escaped(sb, "%s", mp->label);
2205 			sbuf_printf(sb, "</label>\n");
2206 		}
2207 	}
2208 }
2209 
2210 static void
g_md_fini(struct g_class * mp __unused)2211 g_md_fini(struct g_class *mp __unused)
2212 {
2213 
2214 	sx_destroy(&md_sx);
2215 	if (status_dev != NULL)
2216 		destroy_dev(status_dev);
2217 	delete_unrhdr(md_uh);
2218 }
2219