g_part.c revision 281303
1/*-
2 * Copyright (c) 2002, 2005-2009 Marcel Moolenaar
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: stable/10/sys/geom/part/g_part.c 281303 2015-04-09 10:10:05Z mav $");
29
30#include <sys/param.h>
31#include <sys/bio.h>
32#include <sys/endian.h>
33#include <sys/kernel.h>
34#include <sys/kobj.h>
35#include <sys/limits.h>
36#include <sys/lock.h>
37#include <sys/malloc.h>
38#include <sys/mutex.h>
39#include <sys/queue.h>
40#include <sys/sbuf.h>
41#include <sys/sysctl.h>
42#include <sys/systm.h>
43#include <sys/uuid.h>
44#include <geom/geom.h>
45#include <geom/geom_ctl.h>
46#include <geom/geom_int.h>
47#include <geom/part/g_part.h>
48
49#include "g_part_if.h"
50
51#ifndef _PATH_DEV
52#define _PATH_DEV "/dev/"
53#endif
54
55static kobj_method_t g_part_null_methods[] = {
56	{ 0, 0 }
57};
58
59static struct g_part_scheme g_part_null_scheme = {
60	"(none)",
61	g_part_null_methods,
62	sizeof(struct g_part_table),
63};
64
65TAILQ_HEAD(, g_part_scheme) g_part_schemes =
66    TAILQ_HEAD_INITIALIZER(g_part_schemes);
67
68struct g_part_alias_list {
69	const char *lexeme;
70	enum g_part_alias alias;
71} g_part_alias_list[G_PART_ALIAS_COUNT] = {
72	{ "apple-boot", G_PART_ALIAS_APPLE_BOOT },
73	{ "apple-core-storage", G_PART_ALIAS_APPLE_CORE_STORAGE },
74	{ "apple-hfs", G_PART_ALIAS_APPLE_HFS },
75	{ "apple-label", G_PART_ALIAS_APPLE_LABEL },
76	{ "apple-raid", G_PART_ALIAS_APPLE_RAID },
77	{ "apple-raid-offline", G_PART_ALIAS_APPLE_RAID_OFFLINE },
78	{ "apple-tv-recovery", G_PART_ALIAS_APPLE_TV_RECOVERY },
79	{ "apple-ufs", G_PART_ALIAS_APPLE_UFS },
80	{ "bios-boot", G_PART_ALIAS_BIOS_BOOT },
81	{ "ebr", G_PART_ALIAS_EBR },
82	{ "efi", G_PART_ALIAS_EFI },
83	{ "fat16", G_PART_ALIAS_MS_FAT16 },
84	{ "fat32", G_PART_ALIAS_MS_FAT32 },
85	{ "freebsd", G_PART_ALIAS_FREEBSD },
86	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
87	{ "freebsd-nandfs", G_PART_ALIAS_FREEBSD_NANDFS },
88	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
89	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
90	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
91	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
92	{ "linux-data", G_PART_ALIAS_LINUX_DATA },
93	{ "linux-lvm", G_PART_ALIAS_LINUX_LVM },
94	{ "linux-raid", G_PART_ALIAS_LINUX_RAID },
95	{ "linux-swap", G_PART_ALIAS_LINUX_SWAP },
96	{ "mbr", G_PART_ALIAS_MBR },
97	{ "ms-basic-data", G_PART_ALIAS_MS_BASIC_DATA },
98	{ "ms-ldm-data", G_PART_ALIAS_MS_LDM_DATA },
99	{ "ms-ldm-metadata", G_PART_ALIAS_MS_LDM_METADATA },
100	{ "ms-reserved", G_PART_ALIAS_MS_RESERVED },
101	{ "ntfs", G_PART_ALIAS_MS_NTFS },
102	{ "netbsd-ccd", G_PART_ALIAS_NETBSD_CCD },
103	{ "netbsd-cgd", G_PART_ALIAS_NETBSD_CGD },
104	{ "netbsd-ffs", G_PART_ALIAS_NETBSD_FFS },
105	{ "netbsd-lfs", G_PART_ALIAS_NETBSD_LFS },
106	{ "netbsd-raid", G_PART_ALIAS_NETBSD_RAID },
107	{ "netbsd-swap", G_PART_ALIAS_NETBSD_SWAP },
108	{ "vmware-vmfs", G_PART_ALIAS_VMFS },
109	{ "vmware-vmkdiag", G_PART_ALIAS_VMKDIAG },
110	{ "vmware-reserved", G_PART_ALIAS_VMRESERVED },
111	{ "vmware-vsanhdr", G_PART_ALIAS_VMVSANHDR },
112	{ "dragonfly-label32", G_PART_ALIAS_DFBSD },
113	{ "dragonfly-label64", G_PART_ALIAS_DFBSD64 },
114	{ "dragonfly-swap", G_PART_ALIAS_DFBSD_SWAP },
115	{ "dragonfly-ufs", G_PART_ALIAS_DFBSD_UFS },
116	{ "dragonfly-vinum", G_PART_ALIAS_DFBSD_VINUM },
117	{ "dragonfly-ccd", G_PART_ALIAS_DFBSD_CCD },
118	{ "dragonfly-legacy", G_PART_ALIAS_DFBSD_LEGACY },
119	{ "dragonfly-hammer", G_PART_ALIAS_DFBSD_HAMMER },
120	{ "dragonfly-hammer2", G_PART_ALIAS_DFBSD_HAMMER2 },
121	{ "prep-boot", G_PART_ALIAS_PREP_BOOT },
122};
123
124SYSCTL_DECL(_kern_geom);
125SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
126    "GEOM_PART stuff");
127static u_int check_integrity = 1;
128TUNABLE_INT("kern.geom.part.check_integrity", &check_integrity);
129SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity,
130    CTLFLAG_RW | CTLFLAG_TUN, &check_integrity, 1,
131    "Enable integrity checking");
132
133/*
134 * The GEOM partitioning class.
135 */
136static g_ctl_req_t g_part_ctlreq;
137static g_ctl_destroy_geom_t g_part_destroy_geom;
138static g_fini_t g_part_fini;
139static g_init_t g_part_init;
140static g_taste_t g_part_taste;
141
142static g_access_t g_part_access;
143static g_dumpconf_t g_part_dumpconf;
144static g_orphan_t g_part_orphan;
145static g_spoiled_t g_part_spoiled;
146static g_start_t g_part_start;
147static g_resize_t g_part_resize;
148
149static struct g_class g_part_class = {
150	.name = "PART",
151	.version = G_VERSION,
152	/* Class methods. */
153	.ctlreq = g_part_ctlreq,
154	.destroy_geom = g_part_destroy_geom,
155	.fini = g_part_fini,
156	.init = g_part_init,
157	.taste = g_part_taste,
158	/* Geom methods. */
159	.access = g_part_access,
160	.dumpconf = g_part_dumpconf,
161	.orphan = g_part_orphan,
162	.spoiled = g_part_spoiled,
163	.start = g_part_start,
164	.resize = g_part_resize
165};
166
167DECLARE_GEOM_CLASS(g_part_class, g_part);
168MODULE_VERSION(g_part, 0);
169
170/*
171 * Support functions.
172 */
173
174static void g_part_wither(struct g_geom *, int);
175
176const char *
177g_part_alias_name(enum g_part_alias alias)
178{
179	int i;
180
181	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
182		if (g_part_alias_list[i].alias != alias)
183			continue;
184		return (g_part_alias_list[i].lexeme);
185	}
186
187	return (NULL);
188}
189
190void
191g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
192    u_int *bestheads)
193{
194	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
195	off_t chs, cylinders;
196	u_int heads;
197	int idx;
198
199	*bestchs = 0;
200	*bestheads = 0;
201	for (idx = 0; candidate_heads[idx] != 0; idx++) {
202		heads = candidate_heads[idx];
203		cylinders = blocks / heads / sectors;
204		if (cylinders < heads || cylinders < sectors)
205			break;
206		if (cylinders > 1023)
207			continue;
208		chs = cylinders * heads * sectors;
209		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
210			*bestchs = chs;
211			*bestheads = heads;
212		}
213	}
214}
215
216static void
217g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
218    off_t blocks)
219{
220	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
221	off_t chs, bestchs;
222	u_int heads, sectors;
223	int idx;
224
225	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
226	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
227		table->gpt_fixgeom = 0;
228		table->gpt_heads = 0;
229		table->gpt_sectors = 0;
230		bestchs = 0;
231		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
232			sectors = candidate_sectors[idx];
233			g_part_geometry_heads(blocks, sectors, &chs, &heads);
234			if (chs == 0)
235				continue;
236			/*
237			 * Prefer a geometry with sectors > 1, but only if
238			 * it doesn't bump down the number of heads to 1.
239			 */
240			if (chs > bestchs || (chs == bestchs && heads > 1 &&
241			    table->gpt_sectors == 1)) {
242				bestchs = chs;
243				table->gpt_heads = heads;
244				table->gpt_sectors = sectors;
245			}
246		}
247		/*
248		 * If we didn't find a geometry at all, then the disk is
249		 * too big. This means we can use the maximum number of
250		 * heads and sectors.
251		 */
252		if (bestchs == 0) {
253			table->gpt_heads = 255;
254			table->gpt_sectors = 63;
255		}
256	} else {
257		table->gpt_fixgeom = 1;
258		table->gpt_heads = heads;
259		table->gpt_sectors = sectors;
260	}
261}
262
263#define	DPRINTF(...)	if (bootverbose) {	\
264	printf("GEOM_PART: " __VA_ARGS__);	\
265}
266
267static int
268g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
269{
270	struct g_part_entry *e1, *e2;
271	struct g_provider *pp;
272	off_t offset;
273	int failed;
274
275	failed = 0;
276	pp = cp->provider;
277	if (table->gpt_last < table->gpt_first) {
278		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
279		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
280		failed++;
281	}
282	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
283		DPRINTF("last LBA extends beyond mediasize: "
284		    "%jd > %jd\n", (intmax_t)table->gpt_last,
285		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
286		failed++;
287	}
288	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
289		if (e1->gpe_deleted || e1->gpe_internal)
290			continue;
291		if (e1->gpe_start < table->gpt_first) {
292			DPRINTF("partition %d has start offset below first "
293			    "LBA: %jd < %jd\n", e1->gpe_index,
294			    (intmax_t)e1->gpe_start,
295			    (intmax_t)table->gpt_first);
296			failed++;
297		}
298		if (e1->gpe_start > table->gpt_last) {
299			DPRINTF("partition %d has start offset beyond last "
300			    "LBA: %jd > %jd\n", e1->gpe_index,
301			    (intmax_t)e1->gpe_start,
302			    (intmax_t)table->gpt_last);
303			failed++;
304		}
305		if (e1->gpe_end < e1->gpe_start) {
306			DPRINTF("partition %d has end offset below start "
307			    "offset: %jd < %jd\n", e1->gpe_index,
308			    (intmax_t)e1->gpe_end,
309			    (intmax_t)e1->gpe_start);
310			failed++;
311		}
312		if (e1->gpe_end > table->gpt_last) {
313			DPRINTF("partition %d has end offset beyond last "
314			    "LBA: %jd > %jd\n", e1->gpe_index,
315			    (intmax_t)e1->gpe_end,
316			    (intmax_t)table->gpt_last);
317			failed++;
318		}
319		if (pp->stripesize > 0) {
320			offset = e1->gpe_start * pp->sectorsize;
321			if (e1->gpe_offset > offset)
322				offset = e1->gpe_offset;
323			if ((offset + pp->stripeoffset) % pp->stripesize) {
324				DPRINTF("partition %d is not aligned on %u "
325				    "bytes\n", e1->gpe_index, pp->stripesize);
326				/* Don't treat this as a critical failure */
327			}
328		}
329		e2 = e1;
330		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
331			if (e2->gpe_deleted || e2->gpe_internal)
332				continue;
333			if (e1->gpe_start >= e2->gpe_start &&
334			    e1->gpe_start <= e2->gpe_end) {
335				DPRINTF("partition %d has start offset inside "
336				    "partition %d: start[%d] %jd >= start[%d] "
337				    "%jd <= end[%d] %jd\n",
338				    e1->gpe_index, e2->gpe_index,
339				    e2->gpe_index, (intmax_t)e2->gpe_start,
340				    e1->gpe_index, (intmax_t)e1->gpe_start,
341				    e2->gpe_index, (intmax_t)e2->gpe_end);
342				failed++;
343			}
344			if (e1->gpe_end >= e2->gpe_start &&
345			    e1->gpe_end <= e2->gpe_end) {
346				DPRINTF("partition %d has end offset inside "
347				    "partition %d: start[%d] %jd >= end[%d] "
348				    "%jd <= end[%d] %jd\n",
349				    e1->gpe_index, e2->gpe_index,
350				    e2->gpe_index, (intmax_t)e2->gpe_start,
351				    e1->gpe_index, (intmax_t)e1->gpe_end,
352				    e2->gpe_index, (intmax_t)e2->gpe_end);
353				failed++;
354			}
355			if (e1->gpe_start < e2->gpe_start &&
356			    e1->gpe_end > e2->gpe_end) {
357				DPRINTF("partition %d contains partition %d: "
358				    "start[%d] %jd > start[%d] %jd, end[%d] "
359				    "%jd < end[%d] %jd\n",
360				    e1->gpe_index, e2->gpe_index,
361				    e1->gpe_index, (intmax_t)e1->gpe_start,
362				    e2->gpe_index, (intmax_t)e2->gpe_start,
363				    e2->gpe_index, (intmax_t)e2->gpe_end,
364				    e1->gpe_index, (intmax_t)e1->gpe_end);
365				failed++;
366			}
367		}
368	}
369	if (failed != 0) {
370		printf("GEOM_PART: integrity check failed (%s, %s)\n",
371		    pp->name, table->gpt_scheme->name);
372		if (check_integrity != 0)
373			return (EINVAL);
374		table->gpt_corrupt = 1;
375	}
376	return (0);
377}
378#undef	DPRINTF
379
380struct g_part_entry *
381g_part_new_entry(struct g_part_table *table, int index, quad_t start,
382    quad_t end)
383{
384	struct g_part_entry *entry, *last;
385
386	last = NULL;
387	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
388		if (entry->gpe_index == index)
389			break;
390		if (entry->gpe_index > index) {
391			entry = NULL;
392			break;
393		}
394		last = entry;
395	}
396	if (entry == NULL) {
397		entry = g_malloc(table->gpt_scheme->gps_entrysz,
398		    M_WAITOK | M_ZERO);
399		entry->gpe_index = index;
400		if (last == NULL)
401			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
402		else
403			LIST_INSERT_AFTER(last, entry, gpe_entry);
404	} else
405		entry->gpe_offset = 0;
406	entry->gpe_start = start;
407	entry->gpe_end = end;
408	return (entry);
409}
410
411static void
412g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
413    struct g_part_entry *entry)
414{
415	struct g_consumer *cp;
416	struct g_provider *pp;
417	struct sbuf *sb;
418	off_t offset;
419
420	cp = LIST_FIRST(&gp->consumer);
421	pp = cp->provider;
422
423	offset = entry->gpe_start * pp->sectorsize;
424	if (entry->gpe_offset < offset)
425		entry->gpe_offset = offset;
426
427	if (entry->gpe_pp == NULL) {
428		sb = sbuf_new_auto();
429		G_PART_FULLNAME(table, entry, sb, gp->name);
430		sbuf_finish(sb);
431		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
432		sbuf_delete(sb);
433		entry->gpe_pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
434		entry->gpe_pp->private = entry;		/* Close the circle. */
435	}
436	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
437	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
438	    pp->sectorsize;
439	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
440	entry->gpe_pp->sectorsize = pp->sectorsize;
441	entry->gpe_pp->stripesize = pp->stripesize;
442	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
443	if (pp->stripesize > 0)
444		entry->gpe_pp->stripeoffset %= pp->stripesize;
445	entry->gpe_pp->flags |= pp->flags & G_PF_ACCEPT_UNMAPPED;
446	g_error_provider(entry->gpe_pp, 0);
447}
448
449static struct g_geom*
450g_part_find_geom(const char *name)
451{
452	struct g_geom *gp;
453	LIST_FOREACH(gp, &g_part_class.geom, geom) {
454		if ((gp->flags & G_GEOM_WITHER) == 0 &&
455		    strcmp(name, gp->name) == 0)
456			break;
457	}
458	return (gp);
459}
460
461static int
462g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
463{
464	struct g_geom *gp;
465	const char *gname;
466
467	gname = gctl_get_asciiparam(req, name);
468	if (gname == NULL)
469		return (ENOATTR);
470	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
471		gname += sizeof(_PATH_DEV) - 1;
472	gp = g_part_find_geom(gname);
473	if (gp == NULL) {
474		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
475		return (EINVAL);
476	}
477	*v = gp;
478	return (0);
479}
480
481static int
482g_part_parm_provider(struct gctl_req *req, const char *name,
483    struct g_provider **v)
484{
485	struct g_provider *pp;
486	const char *pname;
487
488	pname = gctl_get_asciiparam(req, name);
489	if (pname == NULL)
490		return (ENOATTR);
491	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
492		pname += sizeof(_PATH_DEV) - 1;
493	pp = g_provider_by_name(pname);
494	if (pp == NULL) {
495		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
496		return (EINVAL);
497	}
498	*v = pp;
499	return (0);
500}
501
502static int
503g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
504{
505	const char *p;
506	char *x;
507	quad_t q;
508
509	p = gctl_get_asciiparam(req, name);
510	if (p == NULL)
511		return (ENOATTR);
512	q = strtoq(p, &x, 0);
513	if (*x != '\0' || q < 0) {
514		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
515		return (EINVAL);
516	}
517	*v = q;
518	return (0);
519}
520
521static int
522g_part_parm_scheme(struct gctl_req *req, const char *name,
523    struct g_part_scheme **v)
524{
525	struct g_part_scheme *s;
526	const char *p;
527
528	p = gctl_get_asciiparam(req, name);
529	if (p == NULL)
530		return (ENOATTR);
531	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
532		if (s == &g_part_null_scheme)
533			continue;
534		if (!strcasecmp(s->name, p))
535			break;
536	}
537	if (s == NULL) {
538		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
539		return (EINVAL);
540	}
541	*v = s;
542	return (0);
543}
544
545static int
546g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
547{
548	const char *p;
549
550	p = gctl_get_asciiparam(req, name);
551	if (p == NULL)
552		return (ENOATTR);
553	/* An empty label is always valid. */
554	if (strcmp(name, "label") != 0 && p[0] == '\0') {
555		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
556		return (EINVAL);
557	}
558	*v = p;
559	return (0);
560}
561
562static int
563g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
564{
565	const intmax_t *p;
566	int size;
567
568	p = gctl_get_param(req, name, &size);
569	if (p == NULL)
570		return (ENOATTR);
571	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
572		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
573		return (EINVAL);
574	}
575	*v = (u_int)*p;
576	return (0);
577}
578
579static int
580g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
581{
582	const uint32_t *p;
583	int size;
584
585	p = gctl_get_param(req, name, &size);
586	if (p == NULL)
587		return (ENOATTR);
588	if (size != sizeof(*p) || *p > INT_MAX) {
589		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
590		return (EINVAL);
591	}
592	*v = (u_int)*p;
593	return (0);
594}
595
596static int
597g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
598    unsigned int *s)
599{
600	const void *p;
601	int size;
602
603	p = gctl_get_param(req, name, &size);
604	if (p == NULL)
605		return (ENOATTR);
606	*v = p;
607	*s = size;
608	return (0);
609}
610
611static int
612g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
613{
614	struct g_part_scheme *iter, *scheme;
615	struct g_part_table *table;
616	int pri, probe;
617
618	table = gp->softc;
619	scheme = (table != NULL) ? table->gpt_scheme : NULL;
620	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
621	if (pri == 0)
622		goto done;
623	if (pri > 0) {	/* error */
624		scheme = NULL;
625		pri = INT_MIN;
626	}
627
628	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
629		if (iter == &g_part_null_scheme)
630			continue;
631		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
632		    M_WAITOK);
633		table->gpt_gp = gp;
634		table->gpt_scheme = iter;
635		table->gpt_depth = depth;
636		probe = G_PART_PROBE(table, cp);
637		if (probe <= 0 && probe > pri) {
638			pri = probe;
639			scheme = iter;
640			if (gp->softc != NULL)
641				kobj_delete((kobj_t)gp->softc, M_GEOM);
642			gp->softc = table;
643			if (pri == 0)
644				goto done;
645		} else
646			kobj_delete((kobj_t)table, M_GEOM);
647	}
648
649done:
650	return ((scheme == NULL) ? ENXIO : 0);
651}
652
653/*
654 * Control request functions.
655 */
656
657static int
658g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
659{
660	struct g_geom *gp;
661	struct g_provider *pp;
662	struct g_part_entry *delent, *last, *entry;
663	struct g_part_table *table;
664	struct sbuf *sb;
665	quad_t end;
666	unsigned int index;
667	int error;
668
669	gp = gpp->gpp_geom;
670	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
671	g_topology_assert();
672
673	pp = LIST_FIRST(&gp->consumer)->provider;
674	table = gp->softc;
675	end = gpp->gpp_start + gpp->gpp_size - 1;
676
677	if (gpp->gpp_start < table->gpt_first ||
678	    gpp->gpp_start > table->gpt_last) {
679		gctl_error(req, "%d start '%jd'", EINVAL,
680		    (intmax_t)gpp->gpp_start);
681		return (EINVAL);
682	}
683	if (end < gpp->gpp_start || end > table->gpt_last) {
684		gctl_error(req, "%d size '%jd'", EINVAL,
685		    (intmax_t)gpp->gpp_size);
686		return (EINVAL);
687	}
688	if (gpp->gpp_index > table->gpt_entries) {
689		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
690		return (EINVAL);
691	}
692
693	delent = last = NULL;
694	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
695	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
696		if (entry->gpe_deleted) {
697			if (entry->gpe_index == index)
698				delent = entry;
699			continue;
700		}
701		if (entry->gpe_index == index)
702			index = entry->gpe_index + 1;
703		if (entry->gpe_index < index)
704			last = entry;
705		if (entry->gpe_internal)
706			continue;
707		if (gpp->gpp_start >= entry->gpe_start &&
708		    gpp->gpp_start <= entry->gpe_end) {
709			gctl_error(req, "%d start '%jd'", ENOSPC,
710			    (intmax_t)gpp->gpp_start);
711			return (ENOSPC);
712		}
713		if (end >= entry->gpe_start && end <= entry->gpe_end) {
714			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
715			return (ENOSPC);
716		}
717		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
718			gctl_error(req, "%d size '%jd'", ENOSPC,
719			    (intmax_t)gpp->gpp_size);
720			return (ENOSPC);
721		}
722	}
723	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
724		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
725		return (EEXIST);
726	}
727	if (index > table->gpt_entries) {
728		gctl_error(req, "%d index '%d'", ENOSPC, index);
729		return (ENOSPC);
730	}
731
732	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
733	    M_WAITOK | M_ZERO) : delent;
734	entry->gpe_index = index;
735	entry->gpe_start = gpp->gpp_start;
736	entry->gpe_end = end;
737	error = G_PART_ADD(table, entry, gpp);
738	if (error) {
739		gctl_error(req, "%d", error);
740		if (delent == NULL)
741			g_free(entry);
742		return (error);
743	}
744	if (delent == NULL) {
745		if (last == NULL)
746			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
747		else
748			LIST_INSERT_AFTER(last, entry, gpe_entry);
749		entry->gpe_created = 1;
750	} else {
751		entry->gpe_deleted = 0;
752		entry->gpe_modified = 1;
753	}
754	g_part_new_provider(gp, table, entry);
755
756	/* Provide feedback if so requested. */
757	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
758		sb = sbuf_new_auto();
759		G_PART_FULLNAME(table, entry, sb, gp->name);
760		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
761			sbuf_printf(sb, " added, but partition is not "
762			    "aligned on %u bytes\n", pp->stripesize);
763		else
764			sbuf_cat(sb, " added\n");
765		sbuf_finish(sb);
766		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
767		sbuf_delete(sb);
768	}
769	return (0);
770}
771
772static int
773g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
774{
775	struct g_geom *gp;
776	struct g_part_table *table;
777	struct sbuf *sb;
778	int error, sz;
779
780	gp = gpp->gpp_geom;
781	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
782	g_topology_assert();
783
784	table = gp->softc;
785	sz = table->gpt_scheme->gps_bootcodesz;
786	if (sz == 0) {
787		error = ENODEV;
788		goto fail;
789	}
790	if (gpp->gpp_codesize > sz) {
791		error = EFBIG;
792		goto fail;
793	}
794
795	error = G_PART_BOOTCODE(table, gpp);
796	if (error)
797		goto fail;
798
799	/* Provide feedback if so requested. */
800	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
801		sb = sbuf_new_auto();
802		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
803		sbuf_finish(sb);
804		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
805		sbuf_delete(sb);
806	}
807	return (0);
808
809 fail:
810	gctl_error(req, "%d", error);
811	return (error);
812}
813
814static int
815g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
816{
817	struct g_consumer *cp;
818	struct g_geom *gp;
819	struct g_provider *pp;
820	struct g_part_entry *entry, *tmp;
821	struct g_part_table *table;
822	char *buf;
823	int error, i;
824
825	gp = gpp->gpp_geom;
826	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
827	g_topology_assert();
828
829	table = gp->softc;
830	if (!table->gpt_opened) {
831		gctl_error(req, "%d", EPERM);
832		return (EPERM);
833	}
834
835	g_topology_unlock();
836
837	cp = LIST_FIRST(&gp->consumer);
838	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
839		pp = cp->provider;
840		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
841		while (table->gpt_smhead != 0) {
842			i = ffs(table->gpt_smhead) - 1;
843			error = g_write_data(cp, i * pp->sectorsize, buf,
844			    pp->sectorsize);
845			if (error) {
846				g_free(buf);
847				goto fail;
848			}
849			table->gpt_smhead &= ~(1 << i);
850		}
851		while (table->gpt_smtail != 0) {
852			i = ffs(table->gpt_smtail) - 1;
853			error = g_write_data(cp, pp->mediasize - (i + 1) *
854			    pp->sectorsize, buf, pp->sectorsize);
855			if (error) {
856				g_free(buf);
857				goto fail;
858			}
859			table->gpt_smtail &= ~(1 << i);
860		}
861		g_free(buf);
862	}
863
864	if (table->gpt_scheme == &g_part_null_scheme) {
865		g_topology_lock();
866		g_access(cp, -1, -1, -1);
867		g_part_wither(gp, ENXIO);
868		return (0);
869	}
870
871	error = G_PART_WRITE(table, cp);
872	if (error)
873		goto fail;
874
875	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
876		if (!entry->gpe_deleted) {
877			entry->gpe_created = 0;
878			entry->gpe_modified = 0;
879			continue;
880		}
881		LIST_REMOVE(entry, gpe_entry);
882		g_free(entry);
883	}
884	table->gpt_created = 0;
885	table->gpt_opened = 0;
886
887	g_topology_lock();
888	g_access(cp, -1, -1, -1);
889	return (0);
890
891fail:
892	g_topology_lock();
893	gctl_error(req, "%d", error);
894	return (error);
895}
896
897static int
898g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
899{
900	struct g_consumer *cp;
901	struct g_geom *gp;
902	struct g_provider *pp;
903	struct g_part_scheme *scheme;
904	struct g_part_table *null, *table;
905	struct sbuf *sb;
906	int attr, error;
907
908	pp = gpp->gpp_provider;
909	scheme = gpp->gpp_scheme;
910	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
911	g_topology_assert();
912
913	/* Check that there isn't already a g_part geom on the provider. */
914	gp = g_part_find_geom(pp->name);
915	if (gp != NULL) {
916		null = gp->softc;
917		if (null->gpt_scheme != &g_part_null_scheme) {
918			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
919			return (EEXIST);
920		}
921	} else
922		null = NULL;
923
924	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
925	    (gpp->gpp_entries < scheme->gps_minent ||
926	     gpp->gpp_entries > scheme->gps_maxent)) {
927		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
928		return (EINVAL);
929	}
930
931	if (null == NULL)
932		gp = g_new_geomf(&g_part_class, "%s", pp->name);
933	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
934	    M_WAITOK);
935	table = gp->softc;
936	table->gpt_gp = gp;
937	table->gpt_scheme = gpp->gpp_scheme;
938	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
939	    gpp->gpp_entries : scheme->gps_minent;
940	LIST_INIT(&table->gpt_entry);
941	if (null == NULL) {
942		cp = g_new_consumer(gp);
943		cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
944		error = g_attach(cp, pp);
945		if (error == 0)
946			error = g_access(cp, 1, 1, 1);
947		if (error != 0) {
948			g_part_wither(gp, error);
949			gctl_error(req, "%d geom '%s'", error, pp->name);
950			return (error);
951		}
952		table->gpt_opened = 1;
953	} else {
954		cp = LIST_FIRST(&gp->consumer);
955		table->gpt_opened = null->gpt_opened;
956		table->gpt_smhead = null->gpt_smhead;
957		table->gpt_smtail = null->gpt_smtail;
958	}
959
960	g_topology_unlock();
961
962	/* Make sure the provider has media. */
963	if (pp->mediasize == 0 || pp->sectorsize == 0) {
964		error = ENODEV;
965		goto fail;
966	}
967
968	/* Make sure we can nest and if so, determine our depth. */
969	error = g_getattr("PART::isleaf", cp, &attr);
970	if (!error && attr) {
971		error = ENODEV;
972		goto fail;
973	}
974	error = g_getattr("PART::depth", cp, &attr);
975	table->gpt_depth = (!error) ? attr + 1 : 0;
976
977	/*
978	 * Synthesize a disk geometry. Some partitioning schemes
979	 * depend on it and since some file systems need it even
980	 * when the partitition scheme doesn't, we do it here in
981	 * scheme-independent code.
982	 */
983	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
984
985	error = G_PART_CREATE(table, gpp);
986	if (error)
987		goto fail;
988
989	g_topology_lock();
990
991	table->gpt_created = 1;
992	if (null != NULL)
993		kobj_delete((kobj_t)null, M_GEOM);
994
995	/*
996	 * Support automatic commit by filling in the gpp_geom
997	 * parameter.
998	 */
999	gpp->gpp_parms |= G_PART_PARM_GEOM;
1000	gpp->gpp_geom = gp;
1001
1002	/* Provide feedback if so requested. */
1003	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1004		sb = sbuf_new_auto();
1005		sbuf_printf(sb, "%s created\n", gp->name);
1006		sbuf_finish(sb);
1007		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1008		sbuf_delete(sb);
1009	}
1010	return (0);
1011
1012fail:
1013	g_topology_lock();
1014	if (null == NULL) {
1015		g_access(cp, -1, -1, -1);
1016		g_part_wither(gp, error);
1017	} else {
1018		kobj_delete((kobj_t)gp->softc, M_GEOM);
1019		gp->softc = null;
1020	}
1021	gctl_error(req, "%d provider", error);
1022	return (error);
1023}
1024
1025static int
1026g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1027{
1028	struct g_geom *gp;
1029	struct g_provider *pp;
1030	struct g_part_entry *entry;
1031	struct g_part_table *table;
1032	struct sbuf *sb;
1033
1034	gp = gpp->gpp_geom;
1035	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1036	g_topology_assert();
1037
1038	table = gp->softc;
1039
1040	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1041		if (entry->gpe_deleted || entry->gpe_internal)
1042			continue;
1043		if (entry->gpe_index == gpp->gpp_index)
1044			break;
1045	}
1046	if (entry == NULL) {
1047		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1048		return (ENOENT);
1049	}
1050
1051	pp = entry->gpe_pp;
1052	if (pp != NULL) {
1053		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1054			gctl_error(req, "%d", EBUSY);
1055			return (EBUSY);
1056		}
1057
1058		pp->private = NULL;
1059		entry->gpe_pp = NULL;
1060	}
1061
1062	if (pp != NULL)
1063		g_wither_provider(pp, ENXIO);
1064
1065	/* Provide feedback if so requested. */
1066	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1067		sb = sbuf_new_auto();
1068		G_PART_FULLNAME(table, entry, sb, gp->name);
1069		sbuf_cat(sb, " deleted\n");
1070		sbuf_finish(sb);
1071		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1072		sbuf_delete(sb);
1073	}
1074
1075	if (entry->gpe_created) {
1076		LIST_REMOVE(entry, gpe_entry);
1077		g_free(entry);
1078	} else {
1079		entry->gpe_modified = 0;
1080		entry->gpe_deleted = 1;
1081	}
1082	return (0);
1083}
1084
1085static int
1086g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1087{
1088	struct g_consumer *cp;
1089	struct g_geom *gp;
1090	struct g_provider *pp;
1091	struct g_part_entry *entry, *tmp;
1092	struct g_part_table *null, *table;
1093	struct sbuf *sb;
1094	int error;
1095
1096	gp = gpp->gpp_geom;
1097	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1098	g_topology_assert();
1099
1100	table = gp->softc;
1101	/* Check for busy providers. */
1102	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1103		if (entry->gpe_deleted || entry->gpe_internal)
1104			continue;
1105		if (gpp->gpp_force) {
1106			pp = entry->gpe_pp;
1107			if (pp == NULL)
1108				continue;
1109			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1110				continue;
1111		}
1112		gctl_error(req, "%d", EBUSY);
1113		return (EBUSY);
1114	}
1115
1116	if (gpp->gpp_force) {
1117		/* Destroy all providers. */
1118		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1119			pp = entry->gpe_pp;
1120			if (pp != NULL) {
1121				pp->private = NULL;
1122				g_wither_provider(pp, ENXIO);
1123			}
1124			LIST_REMOVE(entry, gpe_entry);
1125			g_free(entry);
1126		}
1127	}
1128
1129	error = G_PART_DESTROY(table, gpp);
1130	if (error) {
1131		gctl_error(req, "%d", error);
1132		return (error);
1133	}
1134
1135	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1136	    M_WAITOK);
1137	null = gp->softc;
1138	null->gpt_gp = gp;
1139	null->gpt_scheme = &g_part_null_scheme;
1140	LIST_INIT(&null->gpt_entry);
1141
1142	cp = LIST_FIRST(&gp->consumer);
1143	pp = cp->provider;
1144	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1145
1146	null->gpt_depth = table->gpt_depth;
1147	null->gpt_opened = table->gpt_opened;
1148	null->gpt_smhead = table->gpt_smhead;
1149	null->gpt_smtail = table->gpt_smtail;
1150
1151	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1152		LIST_REMOVE(entry, gpe_entry);
1153		g_free(entry);
1154	}
1155	kobj_delete((kobj_t)table, M_GEOM);
1156
1157	/* Provide feedback if so requested. */
1158	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1159		sb = sbuf_new_auto();
1160		sbuf_printf(sb, "%s destroyed\n", gp->name);
1161		sbuf_finish(sb);
1162		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1163		sbuf_delete(sb);
1164	}
1165	return (0);
1166}
1167
1168static int
1169g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1170{
1171	struct g_geom *gp;
1172	struct g_part_entry *entry;
1173	struct g_part_table *table;
1174	struct sbuf *sb;
1175	int error;
1176
1177	gp = gpp->gpp_geom;
1178	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1179	g_topology_assert();
1180
1181	table = gp->softc;
1182
1183	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1184		if (entry->gpe_deleted || entry->gpe_internal)
1185			continue;
1186		if (entry->gpe_index == gpp->gpp_index)
1187			break;
1188	}
1189	if (entry == NULL) {
1190		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1191		return (ENOENT);
1192	}
1193
1194	error = G_PART_MODIFY(table, entry, gpp);
1195	if (error) {
1196		gctl_error(req, "%d", error);
1197		return (error);
1198	}
1199
1200	if (!entry->gpe_created)
1201		entry->gpe_modified = 1;
1202
1203	/* Provide feedback if so requested. */
1204	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1205		sb = sbuf_new_auto();
1206		G_PART_FULLNAME(table, entry, sb, gp->name);
1207		sbuf_cat(sb, " modified\n");
1208		sbuf_finish(sb);
1209		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1210		sbuf_delete(sb);
1211	}
1212	return (0);
1213}
1214
1215static int
1216g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1217{
1218	gctl_error(req, "%d verb 'move'", ENOSYS);
1219	return (ENOSYS);
1220}
1221
1222static int
1223g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1224{
1225	struct g_part_table *table;
1226	struct g_geom *gp;
1227	struct sbuf *sb;
1228	int error, recovered;
1229
1230	gp = gpp->gpp_geom;
1231	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1232	g_topology_assert();
1233	table = gp->softc;
1234	error = recovered = 0;
1235
1236	if (table->gpt_corrupt) {
1237		error = G_PART_RECOVER(table);
1238		if (error == 0)
1239			error = g_part_check_integrity(table,
1240			    LIST_FIRST(&gp->consumer));
1241		if (error) {
1242			gctl_error(req, "%d recovering '%s' failed",
1243			    error, gp->name);
1244			return (error);
1245		}
1246		recovered = 1;
1247	}
1248	/* Provide feedback if so requested. */
1249	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1250		sb = sbuf_new_auto();
1251		if (recovered)
1252			sbuf_printf(sb, "%s recovered\n", gp->name);
1253		else
1254			sbuf_printf(sb, "%s recovering is not needed\n",
1255			    gp->name);
1256		sbuf_finish(sb);
1257		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1258		sbuf_delete(sb);
1259	}
1260	return (0);
1261}
1262
1263static int
1264g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1265{
1266	struct g_geom *gp;
1267	struct g_provider *pp;
1268	struct g_part_entry *pe, *entry;
1269	struct g_part_table *table;
1270	struct sbuf *sb;
1271	quad_t end;
1272	int error;
1273	off_t mediasize;
1274
1275	gp = gpp->gpp_geom;
1276	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1277	g_topology_assert();
1278	table = gp->softc;
1279
1280	/* check gpp_index */
1281	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1282		if (entry->gpe_deleted || entry->gpe_internal)
1283			continue;
1284		if (entry->gpe_index == gpp->gpp_index)
1285			break;
1286	}
1287	if (entry == NULL) {
1288		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1289		return (ENOENT);
1290	}
1291
1292	/* check gpp_size */
1293	end = entry->gpe_start + gpp->gpp_size - 1;
1294	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1295		gctl_error(req, "%d size '%jd'", EINVAL,
1296		    (intmax_t)gpp->gpp_size);
1297		return (EINVAL);
1298	}
1299
1300	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1301		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1302			continue;
1303		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1304			gctl_error(req, "%d end '%jd'", ENOSPC,
1305			    (intmax_t)end);
1306			return (ENOSPC);
1307		}
1308		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1309			gctl_error(req, "%d size '%jd'", ENOSPC,
1310			    (intmax_t)gpp->gpp_size);
1311			return (ENOSPC);
1312		}
1313	}
1314
1315	pp = entry->gpe_pp;
1316	if ((g_debugflags & 16) == 0 &&
1317	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1318		if (entry->gpe_end - entry->gpe_start + 1 > gpp->gpp_size) {
1319			/* Deny shrinking of an opened partition. */
1320			gctl_error(req, "%d", EBUSY);
1321			return (EBUSY);
1322		}
1323	}
1324
1325	error = G_PART_RESIZE(table, entry, gpp);
1326	if (error) {
1327		gctl_error(req, "%d%s", error, error != EBUSY ? "":
1328		    " resizing will lead to unexpected shrinking"
1329		    " due to alignment");
1330		return (error);
1331	}
1332
1333	if (!entry->gpe_created)
1334		entry->gpe_modified = 1;
1335
1336	/* update mediasize of changed provider */
1337	mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1338		pp->sectorsize;
1339	g_resize_provider(pp, mediasize);
1340
1341	/* Provide feedback if so requested. */
1342	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1343		sb = sbuf_new_auto();
1344		G_PART_FULLNAME(table, entry, sb, gp->name);
1345		sbuf_cat(sb, " resized\n");
1346		sbuf_finish(sb);
1347		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1348		sbuf_delete(sb);
1349	}
1350	return (0);
1351}
1352
1353static int
1354g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1355    unsigned int set)
1356{
1357	struct g_geom *gp;
1358	struct g_part_entry *entry;
1359	struct g_part_table *table;
1360	struct sbuf *sb;
1361	int error;
1362
1363	gp = gpp->gpp_geom;
1364	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1365	g_topology_assert();
1366
1367	table = gp->softc;
1368
1369	if (gpp->gpp_parms & G_PART_PARM_INDEX) {
1370		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1371			if (entry->gpe_deleted || entry->gpe_internal)
1372				continue;
1373			if (entry->gpe_index == gpp->gpp_index)
1374				break;
1375		}
1376		if (entry == NULL) {
1377			gctl_error(req, "%d index '%d'", ENOENT,
1378			    gpp->gpp_index);
1379			return (ENOENT);
1380		}
1381	} else
1382		entry = NULL;
1383
1384	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1385	if (error) {
1386		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1387		return (error);
1388	}
1389
1390	/* Provide feedback if so requested. */
1391	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1392		sb = sbuf_new_auto();
1393		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1394		    (set) ? "" : "un");
1395		if (entry)
1396			G_PART_FULLNAME(table, entry, sb, gp->name);
1397		else
1398			sbuf_cat(sb, gp->name);
1399		sbuf_cat(sb, "\n");
1400		sbuf_finish(sb);
1401		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1402		sbuf_delete(sb);
1403	}
1404	return (0);
1405}
1406
1407static int
1408g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1409{
1410	struct g_consumer *cp;
1411	struct g_provider *pp;
1412	struct g_geom *gp;
1413	struct g_part_entry *entry, *tmp;
1414	struct g_part_table *table;
1415	int error, reprobe;
1416
1417	gp = gpp->gpp_geom;
1418	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1419	g_topology_assert();
1420
1421	table = gp->softc;
1422	if (!table->gpt_opened) {
1423		gctl_error(req, "%d", EPERM);
1424		return (EPERM);
1425	}
1426
1427	cp = LIST_FIRST(&gp->consumer);
1428	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1429		entry->gpe_modified = 0;
1430		if (entry->gpe_created) {
1431			pp = entry->gpe_pp;
1432			if (pp != NULL) {
1433				pp->private = NULL;
1434				entry->gpe_pp = NULL;
1435				g_wither_provider(pp, ENXIO);
1436			}
1437			entry->gpe_deleted = 1;
1438		}
1439		if (entry->gpe_deleted) {
1440			LIST_REMOVE(entry, gpe_entry);
1441			g_free(entry);
1442		}
1443	}
1444
1445	g_topology_unlock();
1446
1447	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1448	    table->gpt_created) ? 1 : 0;
1449
1450	if (reprobe) {
1451		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1452			if (entry->gpe_internal)
1453				continue;
1454			error = EBUSY;
1455			goto fail;
1456		}
1457		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1458			LIST_REMOVE(entry, gpe_entry);
1459			g_free(entry);
1460		}
1461		error = g_part_probe(gp, cp, table->gpt_depth);
1462		if (error) {
1463			g_topology_lock();
1464			g_access(cp, -1, -1, -1);
1465			g_part_wither(gp, error);
1466			return (0);
1467		}
1468		table = gp->softc;
1469
1470		/*
1471		 * Synthesize a disk geometry. Some partitioning schemes
1472		 * depend on it and since some file systems need it even
1473		 * when the partitition scheme doesn't, we do it here in
1474		 * scheme-independent code.
1475		 */
1476		pp = cp->provider;
1477		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1478	}
1479
1480	error = G_PART_READ(table, cp);
1481	if (error)
1482		goto fail;
1483	error = g_part_check_integrity(table, cp);
1484	if (error)
1485		goto fail;
1486
1487	g_topology_lock();
1488	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1489		if (!entry->gpe_internal)
1490			g_part_new_provider(gp, table, entry);
1491	}
1492
1493	table->gpt_opened = 0;
1494	g_access(cp, -1, -1, -1);
1495	return (0);
1496
1497fail:
1498	g_topology_lock();
1499	gctl_error(req, "%d", error);
1500	return (error);
1501}
1502
1503static void
1504g_part_wither(struct g_geom *gp, int error)
1505{
1506	struct g_part_entry *entry;
1507	struct g_part_table *table;
1508
1509	table = gp->softc;
1510	if (table != NULL) {
1511		G_PART_DESTROY(table, NULL);
1512		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1513			LIST_REMOVE(entry, gpe_entry);
1514			g_free(entry);
1515		}
1516		if (gp->softc != NULL) {
1517			kobj_delete((kobj_t)gp->softc, M_GEOM);
1518			gp->softc = NULL;
1519		}
1520	}
1521	g_wither_geom(gp, error);
1522}
1523
1524/*
1525 * Class methods.
1526 */
1527
1528static void
1529g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1530{
1531	struct g_part_parms gpp;
1532	struct g_part_table *table;
1533	struct gctl_req_arg *ap;
1534	enum g_part_ctl ctlreq;
1535	unsigned int i, mparms, oparms, parm;
1536	int auto_commit, close_on_error;
1537	int error, modifies;
1538
1539	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1540	g_topology_assert();
1541
1542	ctlreq = G_PART_CTL_NONE;
1543	modifies = 1;
1544	mparms = 0;
1545	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1546	switch (*verb) {
1547	case 'a':
1548		if (!strcmp(verb, "add")) {
1549			ctlreq = G_PART_CTL_ADD;
1550			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1551			    G_PART_PARM_START | G_PART_PARM_TYPE;
1552			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1553		}
1554		break;
1555	case 'b':
1556		if (!strcmp(verb, "bootcode")) {
1557			ctlreq = G_PART_CTL_BOOTCODE;
1558			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1559		}
1560		break;
1561	case 'c':
1562		if (!strcmp(verb, "commit")) {
1563			ctlreq = G_PART_CTL_COMMIT;
1564			mparms |= G_PART_PARM_GEOM;
1565			modifies = 0;
1566		} else if (!strcmp(verb, "create")) {
1567			ctlreq = G_PART_CTL_CREATE;
1568			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1569			oparms |= G_PART_PARM_ENTRIES;
1570		}
1571		break;
1572	case 'd':
1573		if (!strcmp(verb, "delete")) {
1574			ctlreq = G_PART_CTL_DELETE;
1575			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1576		} else if (!strcmp(verb, "destroy")) {
1577			ctlreq = G_PART_CTL_DESTROY;
1578			mparms |= G_PART_PARM_GEOM;
1579			oparms |= G_PART_PARM_FORCE;
1580		}
1581		break;
1582	case 'm':
1583		if (!strcmp(verb, "modify")) {
1584			ctlreq = G_PART_CTL_MODIFY;
1585			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1586			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1587		} else if (!strcmp(verb, "move")) {
1588			ctlreq = G_PART_CTL_MOVE;
1589			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1590		}
1591		break;
1592	case 'r':
1593		if (!strcmp(verb, "recover")) {
1594			ctlreq = G_PART_CTL_RECOVER;
1595			mparms |= G_PART_PARM_GEOM;
1596		} else if (!strcmp(verb, "resize")) {
1597			ctlreq = G_PART_CTL_RESIZE;
1598			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1599			    G_PART_PARM_SIZE;
1600		}
1601		break;
1602	case 's':
1603		if (!strcmp(verb, "set")) {
1604			ctlreq = G_PART_CTL_SET;
1605			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM;
1606			oparms |= G_PART_PARM_INDEX;
1607		}
1608		break;
1609	case 'u':
1610		if (!strcmp(verb, "undo")) {
1611			ctlreq = G_PART_CTL_UNDO;
1612			mparms |= G_PART_PARM_GEOM;
1613			modifies = 0;
1614		} else if (!strcmp(verb, "unset")) {
1615			ctlreq = G_PART_CTL_UNSET;
1616			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM;
1617			oparms |= G_PART_PARM_INDEX;
1618		}
1619		break;
1620	}
1621	if (ctlreq == G_PART_CTL_NONE) {
1622		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1623		return;
1624	}
1625
1626	bzero(&gpp, sizeof(gpp));
1627	for (i = 0; i < req->narg; i++) {
1628		ap = &req->arg[i];
1629		parm = 0;
1630		switch (ap->name[0]) {
1631		case 'a':
1632			if (!strcmp(ap->name, "arg0")) {
1633				parm = mparms &
1634				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1635			}
1636			if (!strcmp(ap->name, "attrib"))
1637				parm = G_PART_PARM_ATTRIB;
1638			break;
1639		case 'b':
1640			if (!strcmp(ap->name, "bootcode"))
1641				parm = G_PART_PARM_BOOTCODE;
1642			break;
1643		case 'c':
1644			if (!strcmp(ap->name, "class"))
1645				continue;
1646			break;
1647		case 'e':
1648			if (!strcmp(ap->name, "entries"))
1649				parm = G_PART_PARM_ENTRIES;
1650			break;
1651		case 'f':
1652			if (!strcmp(ap->name, "flags"))
1653				parm = G_PART_PARM_FLAGS;
1654			else if (!strcmp(ap->name, "force"))
1655				parm = G_PART_PARM_FORCE;
1656			break;
1657		case 'i':
1658			if (!strcmp(ap->name, "index"))
1659				parm = G_PART_PARM_INDEX;
1660			break;
1661		case 'l':
1662			if (!strcmp(ap->name, "label"))
1663				parm = G_PART_PARM_LABEL;
1664			break;
1665		case 'o':
1666			if (!strcmp(ap->name, "output"))
1667				parm = G_PART_PARM_OUTPUT;
1668			break;
1669		case 's':
1670			if (!strcmp(ap->name, "scheme"))
1671				parm = G_PART_PARM_SCHEME;
1672			else if (!strcmp(ap->name, "size"))
1673				parm = G_PART_PARM_SIZE;
1674			else if (!strcmp(ap->name, "start"))
1675				parm = G_PART_PARM_START;
1676			break;
1677		case 't':
1678			if (!strcmp(ap->name, "type"))
1679				parm = G_PART_PARM_TYPE;
1680			break;
1681		case 'v':
1682			if (!strcmp(ap->name, "verb"))
1683				continue;
1684			else if (!strcmp(ap->name, "version"))
1685				parm = G_PART_PARM_VERSION;
1686			break;
1687		}
1688		if ((parm & (mparms | oparms)) == 0) {
1689			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1690			return;
1691		}
1692		switch (parm) {
1693		case G_PART_PARM_ATTRIB:
1694			error = g_part_parm_str(req, ap->name,
1695			    &gpp.gpp_attrib);
1696			break;
1697		case G_PART_PARM_BOOTCODE:
1698			error = g_part_parm_bootcode(req, ap->name,
1699			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1700			break;
1701		case G_PART_PARM_ENTRIES:
1702			error = g_part_parm_intmax(req, ap->name,
1703			    &gpp.gpp_entries);
1704			break;
1705		case G_PART_PARM_FLAGS:
1706			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1707			break;
1708		case G_PART_PARM_FORCE:
1709			error = g_part_parm_uint32(req, ap->name,
1710			    &gpp.gpp_force);
1711			break;
1712		case G_PART_PARM_GEOM:
1713			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1714			break;
1715		case G_PART_PARM_INDEX:
1716			error = g_part_parm_intmax(req, ap->name,
1717			    &gpp.gpp_index);
1718			break;
1719		case G_PART_PARM_LABEL:
1720			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1721			break;
1722		case G_PART_PARM_OUTPUT:
1723			error = 0;	/* Write-only parameter */
1724			break;
1725		case G_PART_PARM_PROVIDER:
1726			error = g_part_parm_provider(req, ap->name,
1727			    &gpp.gpp_provider);
1728			break;
1729		case G_PART_PARM_SCHEME:
1730			error = g_part_parm_scheme(req, ap->name,
1731			    &gpp.gpp_scheme);
1732			break;
1733		case G_PART_PARM_SIZE:
1734			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1735			break;
1736		case G_PART_PARM_START:
1737			error = g_part_parm_quad(req, ap->name,
1738			    &gpp.gpp_start);
1739			break;
1740		case G_PART_PARM_TYPE:
1741			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1742			break;
1743		case G_PART_PARM_VERSION:
1744			error = g_part_parm_uint32(req, ap->name,
1745			    &gpp.gpp_version);
1746			break;
1747		default:
1748			error = EDOOFUS;
1749			gctl_error(req, "%d %s", error, ap->name);
1750			break;
1751		}
1752		if (error != 0) {
1753			if (error == ENOATTR) {
1754				gctl_error(req, "%d param '%s'", error,
1755				    ap->name);
1756			}
1757			return;
1758		}
1759		gpp.gpp_parms |= parm;
1760	}
1761	if ((gpp.gpp_parms & mparms) != mparms) {
1762		parm = mparms - (gpp.gpp_parms & mparms);
1763		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1764		return;
1765	}
1766
1767	/* Obtain permissions if possible/necessary. */
1768	close_on_error = 0;
1769	table = NULL;
1770	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1771		table = gpp.gpp_geom->softc;
1772		if (table != NULL && table->gpt_corrupt &&
1773		    ctlreq != G_PART_CTL_DESTROY &&
1774		    ctlreq != G_PART_CTL_RECOVER) {
1775			gctl_error(req, "%d table '%s' is corrupt",
1776			    EPERM, gpp.gpp_geom->name);
1777			return;
1778		}
1779		if (table != NULL && !table->gpt_opened) {
1780			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1781			    1, 1, 1);
1782			if (error) {
1783				gctl_error(req, "%d geom '%s'", error,
1784				    gpp.gpp_geom->name);
1785				return;
1786			}
1787			table->gpt_opened = 1;
1788			close_on_error = 1;
1789		}
1790	}
1791
1792	/* Allow the scheme to check or modify the parameters. */
1793	if (table != NULL) {
1794		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1795		if (error) {
1796			gctl_error(req, "%d pre-check failed", error);
1797			goto out;
1798		}
1799	} else
1800		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1801
1802	switch (ctlreq) {
1803	case G_PART_CTL_NONE:
1804		panic("%s", __func__);
1805	case G_PART_CTL_ADD:
1806		error = g_part_ctl_add(req, &gpp);
1807		break;
1808	case G_PART_CTL_BOOTCODE:
1809		error = g_part_ctl_bootcode(req, &gpp);
1810		break;
1811	case G_PART_CTL_COMMIT:
1812		error = g_part_ctl_commit(req, &gpp);
1813		break;
1814	case G_PART_CTL_CREATE:
1815		error = g_part_ctl_create(req, &gpp);
1816		break;
1817	case G_PART_CTL_DELETE:
1818		error = g_part_ctl_delete(req, &gpp);
1819		break;
1820	case G_PART_CTL_DESTROY:
1821		error = g_part_ctl_destroy(req, &gpp);
1822		break;
1823	case G_PART_CTL_MODIFY:
1824		error = g_part_ctl_modify(req, &gpp);
1825		break;
1826	case G_PART_CTL_MOVE:
1827		error = g_part_ctl_move(req, &gpp);
1828		break;
1829	case G_PART_CTL_RECOVER:
1830		error = g_part_ctl_recover(req, &gpp);
1831		break;
1832	case G_PART_CTL_RESIZE:
1833		error = g_part_ctl_resize(req, &gpp);
1834		break;
1835	case G_PART_CTL_SET:
1836		error = g_part_ctl_setunset(req, &gpp, 1);
1837		break;
1838	case G_PART_CTL_UNDO:
1839		error = g_part_ctl_undo(req, &gpp);
1840		break;
1841	case G_PART_CTL_UNSET:
1842		error = g_part_ctl_setunset(req, &gpp, 0);
1843		break;
1844	}
1845
1846	/* Implement automatic commit. */
1847	if (!error) {
1848		auto_commit = (modifies &&
1849		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1850		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1851		if (auto_commit) {
1852			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1853			    __func__));
1854			error = g_part_ctl_commit(req, &gpp);
1855		}
1856	}
1857
1858 out:
1859	if (error && close_on_error) {
1860		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1861		table->gpt_opened = 0;
1862	}
1863}
1864
1865static int
1866g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1867    struct g_geom *gp)
1868{
1869
1870	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1871	g_topology_assert();
1872
1873	g_part_wither(gp, EINVAL);
1874	return (0);
1875}
1876
1877static struct g_geom *
1878g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1879{
1880	struct g_consumer *cp;
1881	struct g_geom *gp;
1882	struct g_part_entry *entry;
1883	struct g_part_table *table;
1884	struct root_hold_token *rht;
1885	int attr, depth;
1886	int error;
1887
1888	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1889	g_topology_assert();
1890
1891	/* Skip providers that are already open for writing. */
1892	if (pp->acw > 0)
1893		return (NULL);
1894
1895	/*
1896	 * Create a GEOM with consumer and hook it up to the provider.
1897	 * With that we become part of the topology. Optain read access
1898	 * to the provider.
1899	 */
1900	gp = g_new_geomf(mp, "%s", pp->name);
1901	cp = g_new_consumer(gp);
1902	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
1903	error = g_attach(cp, pp);
1904	if (error == 0)
1905		error = g_access(cp, 1, 0, 0);
1906	if (error != 0) {
1907		if (cp->provider)
1908			g_detach(cp);
1909		g_destroy_consumer(cp);
1910		g_destroy_geom(gp);
1911		return (NULL);
1912	}
1913
1914	rht = root_mount_hold(mp->name);
1915	g_topology_unlock();
1916
1917	/*
1918	 * Short-circuit the whole probing galore when there's no
1919	 * media present.
1920	 */
1921	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1922		error = ENODEV;
1923		goto fail;
1924	}
1925
1926	/* Make sure we can nest and if so, determine our depth. */
1927	error = g_getattr("PART::isleaf", cp, &attr);
1928	if (!error && attr) {
1929		error = ENODEV;
1930		goto fail;
1931	}
1932	error = g_getattr("PART::depth", cp, &attr);
1933	depth = (!error) ? attr + 1 : 0;
1934
1935	error = g_part_probe(gp, cp, depth);
1936	if (error)
1937		goto fail;
1938
1939	table = gp->softc;
1940
1941	/*
1942	 * Synthesize a disk geometry. Some partitioning schemes
1943	 * depend on it and since some file systems need it even
1944	 * when the partitition scheme doesn't, we do it here in
1945	 * scheme-independent code.
1946	 */
1947	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1948
1949	error = G_PART_READ(table, cp);
1950	if (error)
1951		goto fail;
1952	error = g_part_check_integrity(table, cp);
1953	if (error)
1954		goto fail;
1955
1956	g_topology_lock();
1957	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1958		if (!entry->gpe_internal)
1959			g_part_new_provider(gp, table, entry);
1960	}
1961
1962	root_mount_rel(rht);
1963	g_access(cp, -1, 0, 0);
1964	return (gp);
1965
1966 fail:
1967	g_topology_lock();
1968	root_mount_rel(rht);
1969	g_access(cp, -1, 0, 0);
1970	g_detach(cp);
1971	g_destroy_consumer(cp);
1972	g_destroy_geom(gp);
1973	return (NULL);
1974}
1975
1976/*
1977 * Geom methods.
1978 */
1979
1980static int
1981g_part_access(struct g_provider *pp, int dr, int dw, int de)
1982{
1983	struct g_consumer *cp;
1984
1985	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1986	    dw, de));
1987
1988	cp = LIST_FIRST(&pp->geom->consumer);
1989
1990	/* We always gain write-exclusive access. */
1991	return (g_access(cp, dr, dw, dw + de));
1992}
1993
1994static void
1995g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1996    struct g_consumer *cp, struct g_provider *pp)
1997{
1998	char buf[64];
1999	struct g_part_entry *entry;
2000	struct g_part_table *table;
2001
2002	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
2003	table = gp->softc;
2004
2005	if (indent == NULL) {
2006		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
2007		entry = pp->private;
2008		if (entry == NULL)
2009			return;
2010		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
2011		    (uintmax_t)entry->gpe_offset,
2012		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2013		/*
2014		 * libdisk compatibility quirk - the scheme dumps the
2015		 * slicer name and partition type in a way that is
2016		 * compatible with libdisk. When libdisk is not used
2017		 * anymore, this should go away.
2018		 */
2019		G_PART_DUMPCONF(table, entry, sb, indent);
2020	} else if (cp != NULL) {	/* Consumer configuration. */
2021		KASSERT(pp == NULL, ("%s", __func__));
2022		/* none */
2023	} else if (pp != NULL) {	/* Provider configuration. */
2024		entry = pp->private;
2025		if (entry == NULL)
2026			return;
2027		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
2028		    (uintmax_t)entry->gpe_start);
2029		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
2030		    (uintmax_t)entry->gpe_end);
2031		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
2032		    entry->gpe_index);
2033		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2034		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2035		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
2036		    (uintmax_t)entry->gpe_offset);
2037		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
2038		    (uintmax_t)pp->mediasize);
2039		G_PART_DUMPCONF(table, entry, sb, indent);
2040	} else {			/* Geom configuration. */
2041		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2042		    table->gpt_scheme->name);
2043		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2044		    table->gpt_entries);
2045		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2046		    (uintmax_t)table->gpt_first);
2047		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2048		    (uintmax_t)table->gpt_last);
2049		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2050		    table->gpt_sectors);
2051		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2052		    table->gpt_heads);
2053		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2054		    table->gpt_corrupt ? "CORRUPT": "OK");
2055		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2056		    table->gpt_opened ? "true": "false");
2057		G_PART_DUMPCONF(table, NULL, sb, indent);
2058	}
2059}
2060
2061static void
2062g_part_resize(struct g_consumer *cp)
2063{
2064	struct g_part_table *table;
2065
2066	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2067	g_topology_assert();
2068
2069	table = cp->geom->softc;
2070	if (table->gpt_opened == 0) {
2071		if (g_access(cp, 1, 1, 1) != 0)
2072			return;
2073		table->gpt_opened = 1;
2074	}
2075	if (G_PART_RESIZE(table, NULL, NULL) == 0)
2076		printf("GEOM_PART: %s was automatically resized.\n"
2077		    "  Use `gpart commit %s` to save changes or "
2078		    "`gpart undo %s` to revert them.\n", cp->geom->name,
2079		    cp->geom->name, cp->geom->name);
2080	if (g_part_check_integrity(table, cp) != 0) {
2081		g_access(cp, -1, -1, -1);
2082		table->gpt_opened = 0;
2083		g_part_wither(table->gpt_gp, ENXIO);
2084	}
2085}
2086
2087static void
2088g_part_orphan(struct g_consumer *cp)
2089{
2090	struct g_provider *pp;
2091	struct g_part_table *table;
2092
2093	pp = cp->provider;
2094	KASSERT(pp != NULL, ("%s", __func__));
2095	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2096	g_topology_assert();
2097
2098	KASSERT(pp->error != 0, ("%s", __func__));
2099	table = cp->geom->softc;
2100	if (table != NULL && table->gpt_opened)
2101		g_access(cp, -1, -1, -1);
2102	g_part_wither(cp->geom, pp->error);
2103}
2104
2105static void
2106g_part_spoiled(struct g_consumer *cp)
2107{
2108
2109	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2110	g_topology_assert();
2111
2112	cp->flags |= G_CF_ORPHAN;
2113	g_part_wither(cp->geom, ENXIO);
2114}
2115
2116static void
2117g_part_start(struct bio *bp)
2118{
2119	struct bio *bp2;
2120	struct g_consumer *cp;
2121	struct g_geom *gp;
2122	struct g_part_entry *entry;
2123	struct g_part_table *table;
2124	struct g_kerneldump *gkd;
2125	struct g_provider *pp;
2126	char buf[64];
2127
2128	pp = bp->bio_to;
2129	gp = pp->geom;
2130	table = gp->softc;
2131	cp = LIST_FIRST(&gp->consumer);
2132
2133	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2134	    pp->name));
2135
2136	entry = pp->private;
2137	if (entry == NULL) {
2138		g_io_deliver(bp, ENXIO);
2139		return;
2140	}
2141
2142	switch(bp->bio_cmd) {
2143	case BIO_DELETE:
2144	case BIO_READ:
2145	case BIO_WRITE:
2146		if (bp->bio_offset >= pp->mediasize) {
2147			g_io_deliver(bp, EIO);
2148			return;
2149		}
2150		bp2 = g_clone_bio(bp);
2151		if (bp2 == NULL) {
2152			g_io_deliver(bp, ENOMEM);
2153			return;
2154		}
2155		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2156			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2157		bp2->bio_done = g_std_done;
2158		bp2->bio_offset += entry->gpe_offset;
2159		g_io_request(bp2, cp);
2160		return;
2161	case BIO_FLUSH:
2162		break;
2163	case BIO_GETATTR:
2164		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2165			return;
2166		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2167			return;
2168		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2169			return;
2170		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2171			return;
2172		if (g_handleattr_str(bp, "PART::scheme",
2173		    table->gpt_scheme->name))
2174			return;
2175		if (g_handleattr_str(bp, "PART::type",
2176		    G_PART_TYPE(table, entry, buf, sizeof(buf))))
2177			return;
2178		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2179			/*
2180			 * Check that the partition is suitable for kernel
2181			 * dumps. Typically only swap partitions should be
2182			 * used. If the request comes from the nested scheme
2183			 * we allow dumping there as well.
2184			 */
2185			if ((bp->bio_from == NULL ||
2186			    bp->bio_from->geom->class != &g_part_class) &&
2187			    G_PART_DUMPTO(table, entry) == 0) {
2188				g_io_deliver(bp, ENODEV);
2189				printf("GEOM_PART: Partition '%s' not suitable"
2190				    " for kernel dumps (wrong type?)\n",
2191				    pp->name);
2192				return;
2193			}
2194			gkd = (struct g_kerneldump *)bp->bio_data;
2195			if (gkd->offset >= pp->mediasize) {
2196				g_io_deliver(bp, EIO);
2197				return;
2198			}
2199			if (gkd->offset + gkd->length > pp->mediasize)
2200				gkd->length = pp->mediasize - gkd->offset;
2201			gkd->offset += entry->gpe_offset;
2202		}
2203		break;
2204	default:
2205		g_io_deliver(bp, EOPNOTSUPP);
2206		return;
2207	}
2208
2209	bp2 = g_clone_bio(bp);
2210	if (bp2 == NULL) {
2211		g_io_deliver(bp, ENOMEM);
2212		return;
2213	}
2214	bp2->bio_done = g_std_done;
2215	g_io_request(bp2, cp);
2216}
2217
2218static void
2219g_part_init(struct g_class *mp)
2220{
2221
2222	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2223}
2224
2225static void
2226g_part_fini(struct g_class *mp)
2227{
2228
2229	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2230}
2231
2232static void
2233g_part_unload_event(void *arg, int flag)
2234{
2235	struct g_consumer *cp;
2236	struct g_geom *gp;
2237	struct g_provider *pp;
2238	struct g_part_scheme *scheme;
2239	struct g_part_table *table;
2240	uintptr_t *xchg;
2241	int acc, error;
2242
2243	if (flag == EV_CANCEL)
2244		return;
2245
2246	xchg = arg;
2247	error = 0;
2248	scheme = (void *)(*xchg);
2249
2250	g_topology_assert();
2251
2252	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2253		table = gp->softc;
2254		if (table->gpt_scheme != scheme)
2255			continue;
2256
2257		acc = 0;
2258		LIST_FOREACH(pp, &gp->provider, provider)
2259			acc += pp->acr + pp->acw + pp->ace;
2260		LIST_FOREACH(cp, &gp->consumer, consumer)
2261			acc += cp->acr + cp->acw + cp->ace;
2262
2263		if (!acc)
2264			g_part_wither(gp, ENOSYS);
2265		else
2266			error = EBUSY;
2267	}
2268
2269	if (!error)
2270		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2271
2272	*xchg = error;
2273}
2274
2275int
2276g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2277{
2278	struct g_part_scheme *iter;
2279	uintptr_t arg;
2280	int error;
2281
2282	error = 0;
2283	switch (type) {
2284	case MOD_LOAD:
2285		TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
2286			if (scheme == iter) {
2287				printf("GEOM_PART: scheme %s is already "
2288				    "registered!\n", scheme->name);
2289				break;
2290			}
2291		}
2292		if (iter == NULL) {
2293			TAILQ_INSERT_TAIL(&g_part_schemes, scheme,
2294			    scheme_list);
2295			g_retaste(&g_part_class);
2296		}
2297		break;
2298	case MOD_UNLOAD:
2299		arg = (uintptr_t)scheme;
2300		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2301		    NULL);
2302		if (error == 0)
2303			error = arg;
2304		break;
2305	default:
2306		error = EOPNOTSUPP;
2307		break;
2308	}
2309
2310	return (error);
2311}
2312