g_part.c revision 236023
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: head/sys/geom/part/g_part.c 236023 2012-05-25 20:33:34Z marcel $");
29
30#include <sys/param.h>
31#include <sys/bio.h>
32#include <sys/diskmbr.h>
33#include <sys/endian.h>
34#include <sys/kernel.h>
35#include <sys/kobj.h>
36#include <sys/limits.h>
37#include <sys/lock.h>
38#include <sys/malloc.h>
39#include <sys/mutex.h>
40#include <sys/queue.h>
41#include <sys/sbuf.h>
42#include <sys/sysctl.h>
43#include <sys/systm.h>
44#include <sys/uuid.h>
45#include <geom/geom.h>
46#include <geom/geom_ctl.h>
47#include <geom/geom_int.h>
48#include <geom/part/g_part.h>
49
50#include "g_part_if.h"
51
52#ifndef _PATH_DEV
53#define _PATH_DEV "/dev/"
54#endif
55
56static kobj_method_t g_part_null_methods[] = {
57	{ 0, 0 }
58};
59
60static struct g_part_scheme g_part_null_scheme = {
61	"(none)",
62	g_part_null_methods,
63	sizeof(struct g_part_table),
64};
65
66TAILQ_HEAD(, g_part_scheme) g_part_schemes =
67    TAILQ_HEAD_INITIALIZER(g_part_schemes);
68
69struct g_part_alias_list {
70	const char *lexeme;
71	enum g_part_alias alias;
72} g_part_alias_list[G_PART_ALIAS_COUNT] = {
73	{ "apple-boot", G_PART_ALIAS_APPLE_BOOT },
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	{ "fat32", G_PART_ALIAS_MS_FAT32 },
84	{ "freebsd", G_PART_ALIAS_FREEBSD },
85	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
86	{ "freebsd-nandfs", G_PART_ALIAS_FREEBSD_NANDFS },
87	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
88	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
89	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
90	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
91	{ "linux-data", G_PART_ALIAS_LINUX_DATA },
92	{ "linux-lvm", G_PART_ALIAS_LINUX_LVM },
93	{ "linux-raid", G_PART_ALIAS_LINUX_RAID },
94	{ "linux-swap", G_PART_ALIAS_LINUX_SWAP },
95	{ "mbr", G_PART_ALIAS_MBR },
96	{ "ms-basic-data", G_PART_ALIAS_MS_BASIC_DATA },
97	{ "ms-ldm-data", G_PART_ALIAS_MS_LDM_DATA },
98	{ "ms-ldm-metadata", G_PART_ALIAS_MS_LDM_METADATA },
99	{ "ms-reserved", G_PART_ALIAS_MS_RESERVED },
100	{ "ntfs", G_PART_ALIAS_MS_NTFS },
101	{ "netbsd-ccd", G_PART_ALIAS_NETBSD_CCD },
102	{ "netbsd-cgd", G_PART_ALIAS_NETBSD_CGD },
103	{ "netbsd-ffs", G_PART_ALIAS_NETBSD_FFS },
104	{ "netbsd-lfs", G_PART_ALIAS_NETBSD_LFS },
105	{ "netbsd-raid", G_PART_ALIAS_NETBSD_RAID },
106	{ "netbsd-swap", G_PART_ALIAS_NETBSD_SWAP },
107	{ "vmware-vmfs", G_PART_ALIAS_VMFS },
108	{ "vmware-vmkdiag", G_PART_ALIAS_VMKDIAG },
109	{ "vmware-reserved", G_PART_ALIAS_VMRESERVED },
110};
111
112SYSCTL_DECL(_kern_geom);
113SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
114    "GEOM_PART stuff");
115static u_int check_integrity = 1;
116TUNABLE_INT("kern.geom.part.check_integrity", &check_integrity);
117SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity,
118    CTLFLAG_RW | CTLFLAG_TUN, &check_integrity, 1,
119    "Enable integrity checking");
120
121/*
122 * The GEOM partitioning class.
123 */
124static g_ctl_req_t g_part_ctlreq;
125static g_ctl_destroy_geom_t g_part_destroy_geom;
126static g_fini_t g_part_fini;
127static g_init_t g_part_init;
128static g_taste_t g_part_taste;
129
130static g_access_t g_part_access;
131static g_dumpconf_t g_part_dumpconf;
132static g_orphan_t g_part_orphan;
133static g_spoiled_t g_part_spoiled;
134static g_start_t g_part_start;
135
136static struct g_class g_part_class = {
137	.name = "PART",
138	.version = G_VERSION,
139	/* Class methods. */
140	.ctlreq = g_part_ctlreq,
141	.destroy_geom = g_part_destroy_geom,
142	.fini = g_part_fini,
143	.init = g_part_init,
144	.taste = g_part_taste,
145	/* Geom methods. */
146	.access = g_part_access,
147	.dumpconf = g_part_dumpconf,
148	.orphan = g_part_orphan,
149	.spoiled = g_part_spoiled,
150	.start = g_part_start,
151};
152
153DECLARE_GEOM_CLASS(g_part_class, g_part);
154MODULE_VERSION(g_part, 0);
155
156/*
157 * Support functions.
158 */
159
160static void g_part_wither(struct g_geom *, int);
161
162const char *
163g_part_alias_name(enum g_part_alias alias)
164{
165	int i;
166
167	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
168		if (g_part_alias_list[i].alias != alias)
169			continue;
170		return (g_part_alias_list[i].lexeme);
171	}
172
173	return (NULL);
174}
175
176void
177g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
178    u_int *bestheads)
179{
180	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
181	off_t chs, cylinders;
182	u_int heads;
183	int idx;
184
185	*bestchs = 0;
186	*bestheads = 0;
187	for (idx = 0; candidate_heads[idx] != 0; idx++) {
188		heads = candidate_heads[idx];
189		cylinders = blocks / heads / sectors;
190		if (cylinders < heads || cylinders < sectors)
191			break;
192		if (cylinders > 1023)
193			continue;
194		chs = cylinders * heads * sectors;
195		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
196			*bestchs = chs;
197			*bestheads = heads;
198		}
199	}
200}
201
202static void
203g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
204    off_t blocks)
205{
206	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
207	off_t chs, bestchs;
208	u_int heads, sectors;
209	int idx;
210
211	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
212	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
213		table->gpt_fixgeom = 0;
214		table->gpt_heads = 0;
215		table->gpt_sectors = 0;
216		bestchs = 0;
217		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
218			sectors = candidate_sectors[idx];
219			g_part_geometry_heads(blocks, sectors, &chs, &heads);
220			if (chs == 0)
221				continue;
222			/*
223			 * Prefer a geometry with sectors > 1, but only if
224			 * it doesn't bump down the number of heads to 1.
225			 */
226			if (chs > bestchs || (chs == bestchs && heads > 1 &&
227			    table->gpt_sectors == 1)) {
228				bestchs = chs;
229				table->gpt_heads = heads;
230				table->gpt_sectors = sectors;
231			}
232		}
233		/*
234		 * If we didn't find a geometry at all, then the disk is
235		 * too big. This means we can use the maximum number of
236		 * heads and sectors.
237		 */
238		if (bestchs == 0) {
239			table->gpt_heads = 255;
240			table->gpt_sectors = 63;
241		}
242	} else {
243		table->gpt_fixgeom = 1;
244		table->gpt_heads = heads;
245		table->gpt_sectors = sectors;
246	}
247}
248
249#define	DPRINTF(...)	if (bootverbose) {	\
250	printf("GEOM_PART: " __VA_ARGS__);	\
251}
252
253static int
254g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
255{
256	struct g_part_entry *e1, *e2;
257	struct g_provider *pp;
258	off_t offset;
259	int failed;
260
261	failed = 0;
262	pp = cp->provider;
263	if (table->gpt_last < table->gpt_first) {
264		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
265		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
266		failed++;
267	}
268	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
269		DPRINTF("last LBA extends beyond mediasize: "
270		    "%jd > %jd\n", (intmax_t)table->gpt_last,
271		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
272		failed++;
273	}
274	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
275		if (e1->gpe_deleted || e1->gpe_internal)
276			continue;
277		if (e1->gpe_start < table->gpt_first) {
278			DPRINTF("partition %d has start offset below first "
279			    "LBA: %jd < %jd\n", e1->gpe_index,
280			    (intmax_t)e1->gpe_start,
281			    (intmax_t)table->gpt_first);
282			failed++;
283		}
284		if (e1->gpe_start > table->gpt_last) {
285			DPRINTF("partition %d has start offset beyond last "
286			    "LBA: %jd > %jd\n", e1->gpe_index,
287			    (intmax_t)e1->gpe_start,
288			    (intmax_t)table->gpt_last);
289			failed++;
290		}
291		if (e1->gpe_end < e1->gpe_start) {
292			DPRINTF("partition %d has end offset below start "
293			    "offset: %jd < %jd\n", e1->gpe_index,
294			    (intmax_t)e1->gpe_end,
295			    (intmax_t)e1->gpe_start);
296			failed++;
297		}
298		if (e1->gpe_end > table->gpt_last) {
299			DPRINTF("partition %d has end offset beyond last "
300			    "LBA: %jd > %jd\n", e1->gpe_index,
301			    (intmax_t)e1->gpe_end,
302			    (intmax_t)table->gpt_last);
303			failed++;
304		}
305		if (pp->stripesize > 0) {
306			offset = e1->gpe_start * pp->sectorsize;
307			if (e1->gpe_offset > offset)
308				offset = e1->gpe_offset;
309			if ((offset + pp->stripeoffset) % pp->stripesize) {
310				DPRINTF("partition %d is not aligned on %u "
311				    "bytes\n", e1->gpe_index, pp->stripesize);
312				/* Don't treat this as a critical failure */
313			}
314		}
315		e2 = e1;
316		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
317			if (e2->gpe_deleted || e2->gpe_internal)
318				continue;
319			if (e1->gpe_start >= e2->gpe_start &&
320			    e1->gpe_start <= e2->gpe_end) {
321				DPRINTF("partition %d has start offset inside "
322				    "partition %d: start[%d] %jd >= start[%d] "
323				    "%jd <= end[%d] %jd\n",
324				    e1->gpe_index, e2->gpe_index,
325				    e2->gpe_index, (intmax_t)e2->gpe_start,
326				    e1->gpe_index, (intmax_t)e1->gpe_start,
327				    e2->gpe_index, (intmax_t)e2->gpe_end);
328				failed++;
329			}
330			if (e1->gpe_end >= e2->gpe_start &&
331			    e1->gpe_end <= e2->gpe_end) {
332				DPRINTF("partition %d has end offset inside "
333				    "partition %d: start[%d] %jd >= end[%d] "
334				    "%jd <= end[%d] %jd\n",
335				    e1->gpe_index, e2->gpe_index,
336				    e2->gpe_index, (intmax_t)e2->gpe_start,
337				    e1->gpe_index, (intmax_t)e1->gpe_end,
338				    e2->gpe_index, (intmax_t)e2->gpe_end);
339				failed++;
340			}
341			if (e1->gpe_start < e2->gpe_start &&
342			    e1->gpe_end > e2->gpe_end) {
343				DPRINTF("partition %d contains partition %d: "
344				    "start[%d] %jd > start[%d] %jd, end[%d] "
345				    "%jd < end[%d] %jd\n",
346				    e1->gpe_index, e2->gpe_index,
347				    e1->gpe_index, (intmax_t)e1->gpe_start,
348				    e2->gpe_index, (intmax_t)e2->gpe_start,
349				    e2->gpe_index, (intmax_t)e2->gpe_end,
350				    e1->gpe_index, (intmax_t)e1->gpe_end);
351				failed++;
352			}
353		}
354	}
355	if (failed != 0) {
356		printf("GEOM_PART: integrity check failed (%s, %s)\n",
357		    pp->name, table->gpt_scheme->name);
358		if (check_integrity != 0)
359			return (EINVAL);
360		table->gpt_corrupt = 1;
361	}
362	return (0);
363}
364#undef	DPRINTF
365
366struct g_part_entry *
367g_part_new_entry(struct g_part_table *table, int index, quad_t start,
368    quad_t end)
369{
370	struct g_part_entry *entry, *last;
371
372	last = NULL;
373	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
374		if (entry->gpe_index == index)
375			break;
376		if (entry->gpe_index > index) {
377			entry = NULL;
378			break;
379		}
380		last = entry;
381	}
382	if (entry == NULL) {
383		entry = g_malloc(table->gpt_scheme->gps_entrysz,
384		    M_WAITOK | M_ZERO);
385		entry->gpe_index = index;
386		if (last == NULL)
387			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
388		else
389			LIST_INSERT_AFTER(last, entry, gpe_entry);
390	} else
391		entry->gpe_offset = 0;
392	entry->gpe_start = start;
393	entry->gpe_end = end;
394	return (entry);
395}
396
397static void
398g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
399    struct g_part_entry *entry)
400{
401	struct g_consumer *cp;
402	struct g_provider *pp;
403	struct sbuf *sb;
404	off_t offset;
405
406	cp = LIST_FIRST(&gp->consumer);
407	pp = cp->provider;
408
409	offset = entry->gpe_start * pp->sectorsize;
410	if (entry->gpe_offset < offset)
411		entry->gpe_offset = offset;
412
413	if (entry->gpe_pp == NULL) {
414		sb = sbuf_new_auto();
415		G_PART_FULLNAME(table, entry, sb, gp->name);
416		sbuf_finish(sb);
417		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
418		sbuf_delete(sb);
419		entry->gpe_pp->private = entry;		/* Close the circle. */
420	}
421	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
422	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
423	    pp->sectorsize;
424	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
425	entry->gpe_pp->sectorsize = pp->sectorsize;
426	entry->gpe_pp->flags = pp->flags & G_PF_CANDELETE;
427	entry->gpe_pp->stripesize = pp->stripesize;
428	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
429	if (pp->stripesize > 0)
430		entry->gpe_pp->stripeoffset %= pp->stripesize;
431	g_error_provider(entry->gpe_pp, 0);
432}
433
434static struct g_geom*
435g_part_find_geom(const char *name)
436{
437	struct g_geom *gp;
438	LIST_FOREACH(gp, &g_part_class.geom, geom) {
439		if (!strcmp(name, gp->name))
440			break;
441	}
442	return (gp);
443}
444
445static int
446g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
447{
448	struct g_geom *gp;
449	const char *gname;
450
451	gname = gctl_get_asciiparam(req, name);
452	if (gname == NULL)
453		return (ENOATTR);
454	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
455		gname += sizeof(_PATH_DEV) - 1;
456	gp = g_part_find_geom(gname);
457	if (gp == NULL) {
458		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
459		return (EINVAL);
460	}
461	if ((gp->flags & G_GEOM_WITHER) != 0) {
462		gctl_error(req, "%d %s", ENXIO, gname);
463		return (ENXIO);
464	}
465	*v = gp;
466	return (0);
467}
468
469static int
470g_part_parm_provider(struct gctl_req *req, const char *name,
471    struct g_provider **v)
472{
473	struct g_provider *pp;
474	const char *pname;
475
476	pname = gctl_get_asciiparam(req, name);
477	if (pname == NULL)
478		return (ENOATTR);
479	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
480		pname += sizeof(_PATH_DEV) - 1;
481	pp = g_provider_by_name(pname);
482	if (pp == NULL) {
483		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
484		return (EINVAL);
485	}
486	*v = pp;
487	return (0);
488}
489
490static int
491g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
492{
493	const char *p;
494	char *x;
495	quad_t q;
496
497	p = gctl_get_asciiparam(req, name);
498	if (p == NULL)
499		return (ENOATTR);
500	q = strtoq(p, &x, 0);
501	if (*x != '\0' || q < 0) {
502		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
503		return (EINVAL);
504	}
505	*v = q;
506	return (0);
507}
508
509static int
510g_part_parm_scheme(struct gctl_req *req, const char *name,
511    struct g_part_scheme **v)
512{
513	struct g_part_scheme *s;
514	const char *p;
515
516	p = gctl_get_asciiparam(req, name);
517	if (p == NULL)
518		return (ENOATTR);
519	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
520		if (s == &g_part_null_scheme)
521			continue;
522		if (!strcasecmp(s->name, p))
523			break;
524	}
525	if (s == NULL) {
526		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
527		return (EINVAL);
528	}
529	*v = s;
530	return (0);
531}
532
533static int
534g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
535{
536	const char *p;
537
538	p = gctl_get_asciiparam(req, name);
539	if (p == NULL)
540		return (ENOATTR);
541	/* An empty label is always valid. */
542	if (strcmp(name, "label") != 0 && p[0] == '\0') {
543		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
544		return (EINVAL);
545	}
546	*v = p;
547	return (0);
548}
549
550static int
551g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
552{
553	const intmax_t *p;
554	int size;
555
556	p = gctl_get_param(req, name, &size);
557	if (p == NULL)
558		return (ENOATTR);
559	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
560		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
561		return (EINVAL);
562	}
563	*v = (u_int)*p;
564	return (0);
565}
566
567static int
568g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
569{
570	const uint32_t *p;
571	int size;
572
573	p = gctl_get_param(req, name, &size);
574	if (p == NULL)
575		return (ENOATTR);
576	if (size != sizeof(*p) || *p > INT_MAX) {
577		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
578		return (EINVAL);
579	}
580	*v = (u_int)*p;
581	return (0);
582}
583
584static int
585g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
586    unsigned int *s)
587{
588	const void *p;
589	int size;
590
591	p = gctl_get_param(req, name, &size);
592	if (p == NULL)
593		return (ENOATTR);
594	*v = p;
595	*s = size;
596	return (0);
597}
598
599static int
600g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
601{
602	struct g_part_scheme *iter, *scheme;
603	struct g_part_table *table;
604	int pri, probe;
605
606	table = gp->softc;
607	scheme = (table != NULL) ? table->gpt_scheme : NULL;
608	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
609	if (pri == 0)
610		goto done;
611	if (pri > 0) {	/* error */
612		scheme = NULL;
613		pri = INT_MIN;
614	}
615
616	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
617		if (iter == &g_part_null_scheme)
618			continue;
619		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
620		    M_WAITOK);
621		table->gpt_gp = gp;
622		table->gpt_scheme = iter;
623		table->gpt_depth = depth;
624		probe = G_PART_PROBE(table, cp);
625		if (probe <= 0 && probe > pri) {
626			pri = probe;
627			scheme = iter;
628			if (gp->softc != NULL)
629				kobj_delete((kobj_t)gp->softc, M_GEOM);
630			gp->softc = table;
631			if (pri == 0)
632				goto done;
633		} else
634			kobj_delete((kobj_t)table, M_GEOM);
635	}
636
637done:
638	return ((scheme == NULL) ? ENXIO : 0);
639}
640
641/*
642 * Control request functions.
643 */
644
645static int
646g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
647{
648	struct g_geom *gp;
649	struct g_provider *pp;
650	struct g_part_entry *delent, *last, *entry;
651	struct g_part_table *table;
652	struct sbuf *sb;
653	quad_t end;
654	unsigned int index;
655	int error;
656
657	gp = gpp->gpp_geom;
658	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
659	g_topology_assert();
660
661	pp = LIST_FIRST(&gp->consumer)->provider;
662	table = gp->softc;
663	end = gpp->gpp_start + gpp->gpp_size - 1;
664
665	if (gpp->gpp_start < table->gpt_first ||
666	    gpp->gpp_start > table->gpt_last) {
667		gctl_error(req, "%d start '%jd'", EINVAL,
668		    (intmax_t)gpp->gpp_start);
669		return (EINVAL);
670	}
671	if (end < gpp->gpp_start || end > table->gpt_last) {
672		gctl_error(req, "%d size '%jd'", EINVAL,
673		    (intmax_t)gpp->gpp_size);
674		return (EINVAL);
675	}
676	if (gpp->gpp_index > table->gpt_entries) {
677		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
678		return (EINVAL);
679	}
680
681	delent = last = NULL;
682	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
683	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
684		if (entry->gpe_deleted) {
685			if (entry->gpe_index == index)
686				delent = entry;
687			continue;
688		}
689		if (entry->gpe_index == index)
690			index = entry->gpe_index + 1;
691		if (entry->gpe_index < index)
692			last = entry;
693		if (entry->gpe_internal)
694			continue;
695		if (gpp->gpp_start >= entry->gpe_start &&
696		    gpp->gpp_start <= entry->gpe_end) {
697			gctl_error(req, "%d start '%jd'", ENOSPC,
698			    (intmax_t)gpp->gpp_start);
699			return (ENOSPC);
700		}
701		if (end >= entry->gpe_start && end <= entry->gpe_end) {
702			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
703			return (ENOSPC);
704		}
705		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
706			gctl_error(req, "%d size '%jd'", ENOSPC,
707			    (intmax_t)gpp->gpp_size);
708			return (ENOSPC);
709		}
710	}
711	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
712		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
713		return (EEXIST);
714	}
715	if (index > table->gpt_entries) {
716		gctl_error(req, "%d index '%d'", ENOSPC, index);
717		return (ENOSPC);
718	}
719
720	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
721	    M_WAITOK | M_ZERO) : delent;
722	entry->gpe_index = index;
723	entry->gpe_start = gpp->gpp_start;
724	entry->gpe_end = end;
725	error = G_PART_ADD(table, entry, gpp);
726	if (error) {
727		gctl_error(req, "%d", error);
728		if (delent == NULL)
729			g_free(entry);
730		return (error);
731	}
732	if (delent == NULL) {
733		if (last == NULL)
734			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
735		else
736			LIST_INSERT_AFTER(last, entry, gpe_entry);
737		entry->gpe_created = 1;
738	} else {
739		entry->gpe_deleted = 0;
740		entry->gpe_modified = 1;
741	}
742	g_part_new_provider(gp, table, entry);
743
744	/* Provide feedback if so requested. */
745	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
746		sb = sbuf_new_auto();
747		G_PART_FULLNAME(table, entry, sb, gp->name);
748		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
749			sbuf_printf(sb, " added, but partition is not "
750			    "aligned on %u bytes\n", pp->stripesize);
751		else
752			sbuf_cat(sb, " added\n");
753		sbuf_finish(sb);
754		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
755		sbuf_delete(sb);
756	}
757	return (0);
758}
759
760static int
761g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
762{
763	struct g_geom *gp;
764	struct g_part_table *table;
765	struct sbuf *sb;
766	int error, sz;
767
768	gp = gpp->gpp_geom;
769	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
770	g_topology_assert();
771
772	table = gp->softc;
773	sz = table->gpt_scheme->gps_bootcodesz;
774	if (sz == 0) {
775		error = ENODEV;
776		goto fail;
777	}
778	if (gpp->gpp_codesize > sz) {
779		error = EFBIG;
780		goto fail;
781	}
782
783	error = G_PART_BOOTCODE(table, gpp);
784	if (error)
785		goto fail;
786
787	/* Provide feedback if so requested. */
788	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
789		sb = sbuf_new_auto();
790		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
791		sbuf_finish(sb);
792		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
793		sbuf_delete(sb);
794	}
795	return (0);
796
797 fail:
798	gctl_error(req, "%d", error);
799	return (error);
800}
801
802static int
803g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
804{
805	struct g_consumer *cp;
806	struct g_geom *gp;
807	struct g_provider *pp;
808	struct g_part_entry *entry, *tmp;
809	struct g_part_table *table;
810	char *buf;
811	int error, i;
812
813	gp = gpp->gpp_geom;
814	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
815	g_topology_assert();
816
817	table = gp->softc;
818	if (!table->gpt_opened) {
819		gctl_error(req, "%d", EPERM);
820		return (EPERM);
821	}
822
823	g_topology_unlock();
824
825	cp = LIST_FIRST(&gp->consumer);
826	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
827		pp = cp->provider;
828		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
829		while (table->gpt_smhead != 0) {
830			i = ffs(table->gpt_smhead) - 1;
831			error = g_write_data(cp, i * pp->sectorsize, buf,
832			    pp->sectorsize);
833			if (error) {
834				g_free(buf);
835				goto fail;
836			}
837			table->gpt_smhead &= ~(1 << i);
838		}
839		while (table->gpt_smtail != 0) {
840			i = ffs(table->gpt_smtail) - 1;
841			error = g_write_data(cp, pp->mediasize - (i + 1) *
842			    pp->sectorsize, buf, pp->sectorsize);
843			if (error) {
844				g_free(buf);
845				goto fail;
846			}
847			table->gpt_smtail &= ~(1 << i);
848		}
849		g_free(buf);
850	}
851
852	if (table->gpt_scheme == &g_part_null_scheme) {
853		g_topology_lock();
854		g_access(cp, -1, -1, -1);
855		g_part_wither(gp, ENXIO);
856		return (0);
857	}
858
859	error = G_PART_WRITE(table, cp);
860	if (error)
861		goto fail;
862
863	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
864		if (!entry->gpe_deleted) {
865			entry->gpe_created = 0;
866			entry->gpe_modified = 0;
867			continue;
868		}
869		LIST_REMOVE(entry, gpe_entry);
870		g_free(entry);
871	}
872	table->gpt_created = 0;
873	table->gpt_opened = 0;
874
875	g_topology_lock();
876	g_access(cp, -1, -1, -1);
877	return (0);
878
879fail:
880	g_topology_lock();
881	gctl_error(req, "%d", error);
882	return (error);
883}
884
885static int
886g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
887{
888	struct g_consumer *cp;
889	struct g_geom *gp;
890	struct g_provider *pp;
891	struct g_part_scheme *scheme;
892	struct g_part_table *null, *table;
893	struct sbuf *sb;
894	int attr, error;
895
896	pp = gpp->gpp_provider;
897	scheme = gpp->gpp_scheme;
898	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
899	g_topology_assert();
900
901	/* Check that there isn't already a g_part geom on the provider. */
902	gp = g_part_find_geom(pp->name);
903	if (gp != NULL) {
904		null = gp->softc;
905		if (null->gpt_scheme != &g_part_null_scheme) {
906			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
907			return (EEXIST);
908		}
909	} else
910		null = NULL;
911
912	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
913	    (gpp->gpp_entries < scheme->gps_minent ||
914	     gpp->gpp_entries > scheme->gps_maxent)) {
915		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
916		return (EINVAL);
917	}
918
919	if (null == NULL)
920		gp = g_new_geomf(&g_part_class, "%s", pp->name);
921	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
922	    M_WAITOK);
923	table = gp->softc;
924	table->gpt_gp = gp;
925	table->gpt_scheme = gpp->gpp_scheme;
926	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
927	    gpp->gpp_entries : scheme->gps_minent;
928	LIST_INIT(&table->gpt_entry);
929	if (null == NULL) {
930		cp = g_new_consumer(gp);
931		error = g_attach(cp, pp);
932		if (error == 0)
933			error = g_access(cp, 1, 1, 1);
934		if (error != 0) {
935			g_part_wither(gp, error);
936			gctl_error(req, "%d geom '%s'", error, pp->name);
937			return (error);
938		}
939		table->gpt_opened = 1;
940	} else {
941		cp = LIST_FIRST(&gp->consumer);
942		table->gpt_opened = null->gpt_opened;
943		table->gpt_smhead = null->gpt_smhead;
944		table->gpt_smtail = null->gpt_smtail;
945	}
946
947	g_topology_unlock();
948
949	/* Make sure the provider has media. */
950	if (pp->mediasize == 0 || pp->sectorsize == 0) {
951		error = ENODEV;
952		goto fail;
953	}
954
955	/* Make sure we can nest and if so, determine our depth. */
956	error = g_getattr("PART::isleaf", cp, &attr);
957	if (!error && attr) {
958		error = ENODEV;
959		goto fail;
960	}
961	error = g_getattr("PART::depth", cp, &attr);
962	table->gpt_depth = (!error) ? attr + 1 : 0;
963
964	/*
965	 * Synthesize a disk geometry. Some partitioning schemes
966	 * depend on it and since some file systems need it even
967	 * when the partitition scheme doesn't, we do it here in
968	 * scheme-independent code.
969	 */
970	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
971
972	error = G_PART_CREATE(table, gpp);
973	if (error)
974		goto fail;
975
976	g_topology_lock();
977
978	table->gpt_created = 1;
979	if (null != NULL)
980		kobj_delete((kobj_t)null, M_GEOM);
981
982	/*
983	 * Support automatic commit by filling in the gpp_geom
984	 * parameter.
985	 */
986	gpp->gpp_parms |= G_PART_PARM_GEOM;
987	gpp->gpp_geom = gp;
988
989	/* Provide feedback if so requested. */
990	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
991		sb = sbuf_new_auto();
992		sbuf_printf(sb, "%s created\n", gp->name);
993		sbuf_finish(sb);
994		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
995		sbuf_delete(sb);
996	}
997	return (0);
998
999fail:
1000	g_topology_lock();
1001	if (null == NULL) {
1002		g_access(cp, -1, -1, -1);
1003		g_part_wither(gp, error);
1004	} else {
1005		kobj_delete((kobj_t)gp->softc, M_GEOM);
1006		gp->softc = null;
1007	}
1008	gctl_error(req, "%d provider", error);
1009	return (error);
1010}
1011
1012static int
1013g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1014{
1015	struct g_geom *gp;
1016	struct g_provider *pp;
1017	struct g_part_entry *entry;
1018	struct g_part_table *table;
1019	struct sbuf *sb;
1020
1021	gp = gpp->gpp_geom;
1022	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1023	g_topology_assert();
1024
1025	table = gp->softc;
1026
1027	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1028		if (entry->gpe_deleted || entry->gpe_internal)
1029			continue;
1030		if (entry->gpe_index == gpp->gpp_index)
1031			break;
1032	}
1033	if (entry == NULL) {
1034		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1035		return (ENOENT);
1036	}
1037
1038	pp = entry->gpe_pp;
1039	if (pp != NULL) {
1040		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1041			gctl_error(req, "%d", EBUSY);
1042			return (EBUSY);
1043		}
1044
1045		pp->private = NULL;
1046		entry->gpe_pp = NULL;
1047	}
1048
1049	if (pp != NULL)
1050		g_wither_provider(pp, ENXIO);
1051
1052	/* Provide feedback if so requested. */
1053	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1054		sb = sbuf_new_auto();
1055		G_PART_FULLNAME(table, entry, sb, gp->name);
1056		sbuf_cat(sb, " deleted\n");
1057		sbuf_finish(sb);
1058		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1059		sbuf_delete(sb);
1060	}
1061
1062	if (entry->gpe_created) {
1063		LIST_REMOVE(entry, gpe_entry);
1064		g_free(entry);
1065	} else {
1066		entry->gpe_modified = 0;
1067		entry->gpe_deleted = 1;
1068	}
1069	return (0);
1070}
1071
1072static int
1073g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1074{
1075	struct g_consumer *cp;
1076	struct g_geom *gp;
1077	struct g_provider *pp;
1078	struct g_part_entry *entry, *tmp;
1079	struct g_part_table *null, *table;
1080	struct sbuf *sb;
1081	int error;
1082
1083	gp = gpp->gpp_geom;
1084	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1085	g_topology_assert();
1086
1087	table = gp->softc;
1088	/* Check for busy providers. */
1089	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1090		if (entry->gpe_deleted || entry->gpe_internal)
1091			continue;
1092		if (gpp->gpp_force) {
1093			pp = entry->gpe_pp;
1094			if (pp == NULL)
1095				continue;
1096			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1097				continue;
1098		}
1099		gctl_error(req, "%d", EBUSY);
1100		return (EBUSY);
1101	}
1102
1103	if (gpp->gpp_force) {
1104		/* Destroy all providers. */
1105		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1106			pp = entry->gpe_pp;
1107			if (pp != NULL) {
1108				pp->private = NULL;
1109				g_wither_provider(pp, ENXIO);
1110			}
1111			LIST_REMOVE(entry, gpe_entry);
1112			g_free(entry);
1113		}
1114	}
1115
1116	error = G_PART_DESTROY(table, gpp);
1117	if (error) {
1118		gctl_error(req, "%d", error);
1119		return (error);
1120	}
1121
1122	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1123	    M_WAITOK);
1124	null = gp->softc;
1125	null->gpt_gp = gp;
1126	null->gpt_scheme = &g_part_null_scheme;
1127	LIST_INIT(&null->gpt_entry);
1128
1129	cp = LIST_FIRST(&gp->consumer);
1130	pp = cp->provider;
1131	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1132
1133	null->gpt_depth = table->gpt_depth;
1134	null->gpt_opened = table->gpt_opened;
1135	null->gpt_smhead = table->gpt_smhead;
1136	null->gpt_smtail = table->gpt_smtail;
1137
1138	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1139		LIST_REMOVE(entry, gpe_entry);
1140		g_free(entry);
1141	}
1142	kobj_delete((kobj_t)table, M_GEOM);
1143
1144	/* Provide feedback if so requested. */
1145	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1146		sb = sbuf_new_auto();
1147		sbuf_printf(sb, "%s destroyed\n", gp->name);
1148		sbuf_finish(sb);
1149		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1150		sbuf_delete(sb);
1151	}
1152	return (0);
1153}
1154
1155static int
1156g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1157{
1158	struct g_geom *gp;
1159	struct g_part_entry *entry;
1160	struct g_part_table *table;
1161	struct sbuf *sb;
1162	int error;
1163
1164	gp = gpp->gpp_geom;
1165	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1166	g_topology_assert();
1167
1168	table = gp->softc;
1169
1170	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1171		if (entry->gpe_deleted || entry->gpe_internal)
1172			continue;
1173		if (entry->gpe_index == gpp->gpp_index)
1174			break;
1175	}
1176	if (entry == NULL) {
1177		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1178		return (ENOENT);
1179	}
1180
1181	error = G_PART_MODIFY(table, entry, gpp);
1182	if (error) {
1183		gctl_error(req, "%d", error);
1184		return (error);
1185	}
1186
1187	if (!entry->gpe_created)
1188		entry->gpe_modified = 1;
1189
1190	/* Provide feedback if so requested. */
1191	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1192		sb = sbuf_new_auto();
1193		G_PART_FULLNAME(table, entry, sb, gp->name);
1194		sbuf_cat(sb, " modified\n");
1195		sbuf_finish(sb);
1196		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1197		sbuf_delete(sb);
1198	}
1199	return (0);
1200}
1201
1202static int
1203g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1204{
1205	gctl_error(req, "%d verb 'move'", ENOSYS);
1206	return (ENOSYS);
1207}
1208
1209static int
1210g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1211{
1212	struct g_part_table *table;
1213	struct g_geom *gp;
1214	struct sbuf *sb;
1215	int error, recovered;
1216
1217	gp = gpp->gpp_geom;
1218	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1219	g_topology_assert();
1220	table = gp->softc;
1221	error = recovered = 0;
1222
1223	if (table->gpt_corrupt) {
1224		error = G_PART_RECOVER(table);
1225		if (error == 0)
1226			error = g_part_check_integrity(table,
1227			    LIST_FIRST(&gp->consumer));
1228		if (error) {
1229			gctl_error(req, "%d recovering '%s' failed",
1230			    error, gp->name);
1231			return (error);
1232		}
1233		recovered = 1;
1234	}
1235	/* Provide feedback if so requested. */
1236	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1237		sb = sbuf_new_auto();
1238		if (recovered)
1239			sbuf_printf(sb, "%s recovered\n", gp->name);
1240		else
1241			sbuf_printf(sb, "%s recovering is not needed\n",
1242			    gp->name);
1243		sbuf_finish(sb);
1244		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1245		sbuf_delete(sb);
1246	}
1247	return (0);
1248}
1249
1250static int
1251g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1252{
1253	struct g_geom *gp;
1254	struct g_provider *pp;
1255	struct g_part_entry *pe, *entry;
1256	struct g_part_table *table;
1257	struct sbuf *sb;
1258	quad_t end;
1259	int error;
1260
1261	gp = gpp->gpp_geom;
1262	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1263	g_topology_assert();
1264	table = gp->softc;
1265
1266	/* check gpp_index */
1267	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1268		if (entry->gpe_deleted || entry->gpe_internal)
1269			continue;
1270		if (entry->gpe_index == gpp->gpp_index)
1271			break;
1272	}
1273	if (entry == NULL) {
1274		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1275		return (ENOENT);
1276	}
1277
1278	/* check gpp_size */
1279	end = entry->gpe_start + gpp->gpp_size - 1;
1280	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1281		gctl_error(req, "%d size '%jd'", EINVAL,
1282		    (intmax_t)gpp->gpp_size);
1283		return (EINVAL);
1284	}
1285
1286	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1287		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1288			continue;
1289		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1290			gctl_error(req, "%d end '%jd'", ENOSPC,
1291			    (intmax_t)end);
1292			return (ENOSPC);
1293		}
1294		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1295			gctl_error(req, "%d size '%jd'", ENOSPC,
1296			    (intmax_t)gpp->gpp_size);
1297			return (ENOSPC);
1298		}
1299	}
1300
1301	pp = entry->gpe_pp;
1302	if ((g_debugflags & 16) == 0 &&
1303	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1304		gctl_error(req, "%d", EBUSY);
1305		return (EBUSY);
1306	}
1307
1308	error = G_PART_RESIZE(table, entry, gpp);
1309	if (error) {
1310		gctl_error(req, "%d", error);
1311		return (error);
1312	}
1313
1314	if (!entry->gpe_created)
1315		entry->gpe_modified = 1;
1316
1317	/* update mediasize of changed provider */
1318	pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1319		pp->sectorsize;
1320
1321	/* Provide feedback if so requested. */
1322	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1323		sb = sbuf_new_auto();
1324		G_PART_FULLNAME(table, entry, sb, gp->name);
1325		sbuf_cat(sb, " resized\n");
1326		sbuf_finish(sb);
1327		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1328		sbuf_delete(sb);
1329	}
1330	return (0);
1331}
1332
1333static int
1334g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1335    unsigned int set)
1336{
1337	struct g_geom *gp;
1338	struct g_part_entry *entry;
1339	struct g_part_table *table;
1340	struct sbuf *sb;
1341	int error;
1342
1343	gp = gpp->gpp_geom;
1344	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1345	g_topology_assert();
1346
1347	table = gp->softc;
1348
1349	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1350		if (entry->gpe_deleted || entry->gpe_internal)
1351			continue;
1352		if (entry->gpe_index == gpp->gpp_index)
1353			break;
1354	}
1355	if (entry == NULL) {
1356		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1357		return (ENOENT);
1358	}
1359
1360	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1361	if (error) {
1362		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1363		return (error);
1364	}
1365
1366	/* Provide feedback if so requested. */
1367	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1368		sb = sbuf_new_auto();
1369		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1370		    (set) ? "" : "un");
1371		G_PART_FULLNAME(table, entry, sb, gp->name);
1372		sbuf_printf(sb, "\n");
1373		sbuf_finish(sb);
1374		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1375		sbuf_delete(sb);
1376	}
1377	return (0);
1378}
1379
1380static int
1381g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1382{
1383	struct g_consumer *cp;
1384	struct g_provider *pp;
1385	struct g_geom *gp;
1386	struct g_part_entry *entry, *tmp;
1387	struct g_part_table *table;
1388	int error, reprobe;
1389
1390	gp = gpp->gpp_geom;
1391	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1392	g_topology_assert();
1393
1394	table = gp->softc;
1395	if (!table->gpt_opened) {
1396		gctl_error(req, "%d", EPERM);
1397		return (EPERM);
1398	}
1399
1400	cp = LIST_FIRST(&gp->consumer);
1401	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1402		entry->gpe_modified = 0;
1403		if (entry->gpe_created) {
1404			pp = entry->gpe_pp;
1405			if (pp != NULL) {
1406				pp->private = NULL;
1407				entry->gpe_pp = NULL;
1408				g_wither_provider(pp, ENXIO);
1409			}
1410			entry->gpe_deleted = 1;
1411		}
1412		if (entry->gpe_deleted) {
1413			LIST_REMOVE(entry, gpe_entry);
1414			g_free(entry);
1415		}
1416	}
1417
1418	g_topology_unlock();
1419
1420	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1421	    table->gpt_created) ? 1 : 0;
1422
1423	if (reprobe) {
1424		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1425			if (entry->gpe_internal)
1426				continue;
1427			error = EBUSY;
1428			goto fail;
1429		}
1430		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1431			LIST_REMOVE(entry, gpe_entry);
1432			g_free(entry);
1433		}
1434		error = g_part_probe(gp, cp, table->gpt_depth);
1435		if (error) {
1436			g_topology_lock();
1437			g_access(cp, -1, -1, -1);
1438			g_part_wither(gp, error);
1439			return (0);
1440		}
1441		table = gp->softc;
1442
1443		/*
1444		 * Synthesize a disk geometry. Some partitioning schemes
1445		 * depend on it and since some file systems need it even
1446		 * when the partitition scheme doesn't, we do it here in
1447		 * scheme-independent code.
1448		 */
1449		pp = cp->provider;
1450		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1451	}
1452
1453	error = G_PART_READ(table, cp);
1454	if (error)
1455		goto fail;
1456	error = g_part_check_integrity(table, cp);
1457	if (error)
1458		goto fail;
1459
1460	g_topology_lock();
1461	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1462		if (!entry->gpe_internal)
1463			g_part_new_provider(gp, table, entry);
1464	}
1465
1466	table->gpt_opened = 0;
1467	g_access(cp, -1, -1, -1);
1468	return (0);
1469
1470fail:
1471	g_topology_lock();
1472	gctl_error(req, "%d", error);
1473	return (error);
1474}
1475
1476static void
1477g_part_wither(struct g_geom *gp, int error)
1478{
1479	struct g_part_entry *entry;
1480	struct g_part_table *table;
1481
1482	table = gp->softc;
1483	if (table != NULL) {
1484		G_PART_DESTROY(table, NULL);
1485		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1486			LIST_REMOVE(entry, gpe_entry);
1487			g_free(entry);
1488		}
1489		if (gp->softc != NULL) {
1490			kobj_delete((kobj_t)gp->softc, M_GEOM);
1491			gp->softc = NULL;
1492		}
1493	}
1494	g_wither_geom(gp, error);
1495}
1496
1497/*
1498 * Class methods.
1499 */
1500
1501static void
1502g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1503{
1504	struct g_part_parms gpp;
1505	struct g_part_table *table;
1506	struct gctl_req_arg *ap;
1507	enum g_part_ctl ctlreq;
1508	unsigned int i, mparms, oparms, parm;
1509	int auto_commit, close_on_error;
1510	int error, modifies;
1511
1512	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1513	g_topology_assert();
1514
1515	ctlreq = G_PART_CTL_NONE;
1516	modifies = 1;
1517	mparms = 0;
1518	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1519	switch (*verb) {
1520	case 'a':
1521		if (!strcmp(verb, "add")) {
1522			ctlreq = G_PART_CTL_ADD;
1523			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1524			    G_PART_PARM_START | G_PART_PARM_TYPE;
1525			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1526		}
1527		break;
1528	case 'b':
1529		if (!strcmp(verb, "bootcode")) {
1530			ctlreq = G_PART_CTL_BOOTCODE;
1531			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1532		}
1533		break;
1534	case 'c':
1535		if (!strcmp(verb, "commit")) {
1536			ctlreq = G_PART_CTL_COMMIT;
1537			mparms |= G_PART_PARM_GEOM;
1538			modifies = 0;
1539		} else if (!strcmp(verb, "create")) {
1540			ctlreq = G_PART_CTL_CREATE;
1541			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1542			oparms |= G_PART_PARM_ENTRIES;
1543		}
1544		break;
1545	case 'd':
1546		if (!strcmp(verb, "delete")) {
1547			ctlreq = G_PART_CTL_DELETE;
1548			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1549		} else if (!strcmp(verb, "destroy")) {
1550			ctlreq = G_PART_CTL_DESTROY;
1551			mparms |= G_PART_PARM_GEOM;
1552			oparms |= G_PART_PARM_FORCE;
1553		}
1554		break;
1555	case 'm':
1556		if (!strcmp(verb, "modify")) {
1557			ctlreq = G_PART_CTL_MODIFY;
1558			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1559			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1560		} else if (!strcmp(verb, "move")) {
1561			ctlreq = G_PART_CTL_MOVE;
1562			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1563		}
1564		break;
1565	case 'r':
1566		if (!strcmp(verb, "recover")) {
1567			ctlreq = G_PART_CTL_RECOVER;
1568			mparms |= G_PART_PARM_GEOM;
1569		} else if (!strcmp(verb, "resize")) {
1570			ctlreq = G_PART_CTL_RESIZE;
1571			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1572			    G_PART_PARM_SIZE;
1573		}
1574		break;
1575	case 's':
1576		if (!strcmp(verb, "set")) {
1577			ctlreq = G_PART_CTL_SET;
1578			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1579			    G_PART_PARM_INDEX;
1580		}
1581		break;
1582	case 'u':
1583		if (!strcmp(verb, "undo")) {
1584			ctlreq = G_PART_CTL_UNDO;
1585			mparms |= G_PART_PARM_GEOM;
1586			modifies = 0;
1587		} else if (!strcmp(verb, "unset")) {
1588			ctlreq = G_PART_CTL_UNSET;
1589			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1590			    G_PART_PARM_INDEX;
1591		}
1592		break;
1593	}
1594	if (ctlreq == G_PART_CTL_NONE) {
1595		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1596		return;
1597	}
1598
1599	bzero(&gpp, sizeof(gpp));
1600	for (i = 0; i < req->narg; i++) {
1601		ap = &req->arg[i];
1602		parm = 0;
1603		switch (ap->name[0]) {
1604		case 'a':
1605			if (!strcmp(ap->name, "arg0")) {
1606				parm = mparms &
1607				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1608			}
1609			if (!strcmp(ap->name, "attrib"))
1610				parm = G_PART_PARM_ATTRIB;
1611			break;
1612		case 'b':
1613			if (!strcmp(ap->name, "bootcode"))
1614				parm = G_PART_PARM_BOOTCODE;
1615			break;
1616		case 'c':
1617			if (!strcmp(ap->name, "class"))
1618				continue;
1619			break;
1620		case 'e':
1621			if (!strcmp(ap->name, "entries"))
1622				parm = G_PART_PARM_ENTRIES;
1623			break;
1624		case 'f':
1625			if (!strcmp(ap->name, "flags"))
1626				parm = G_PART_PARM_FLAGS;
1627			else if (!strcmp(ap->name, "force"))
1628				parm = G_PART_PARM_FORCE;
1629			break;
1630		case 'i':
1631			if (!strcmp(ap->name, "index"))
1632				parm = G_PART_PARM_INDEX;
1633			break;
1634		case 'l':
1635			if (!strcmp(ap->name, "label"))
1636				parm = G_PART_PARM_LABEL;
1637			break;
1638		case 'o':
1639			if (!strcmp(ap->name, "output"))
1640				parm = G_PART_PARM_OUTPUT;
1641			break;
1642		case 's':
1643			if (!strcmp(ap->name, "scheme"))
1644				parm = G_PART_PARM_SCHEME;
1645			else if (!strcmp(ap->name, "size"))
1646				parm = G_PART_PARM_SIZE;
1647			else if (!strcmp(ap->name, "start"))
1648				parm = G_PART_PARM_START;
1649			break;
1650		case 't':
1651			if (!strcmp(ap->name, "type"))
1652				parm = G_PART_PARM_TYPE;
1653			break;
1654		case 'v':
1655			if (!strcmp(ap->name, "verb"))
1656				continue;
1657			else if (!strcmp(ap->name, "version"))
1658				parm = G_PART_PARM_VERSION;
1659			break;
1660		}
1661		if ((parm & (mparms | oparms)) == 0) {
1662			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1663			return;
1664		}
1665		switch (parm) {
1666		case G_PART_PARM_ATTRIB:
1667			error = g_part_parm_str(req, ap->name,
1668			    &gpp.gpp_attrib);
1669			break;
1670		case G_PART_PARM_BOOTCODE:
1671			error = g_part_parm_bootcode(req, ap->name,
1672			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1673			break;
1674		case G_PART_PARM_ENTRIES:
1675			error = g_part_parm_intmax(req, ap->name,
1676			    &gpp.gpp_entries);
1677			break;
1678		case G_PART_PARM_FLAGS:
1679			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1680			break;
1681		case G_PART_PARM_FORCE:
1682			error = g_part_parm_uint32(req, ap->name,
1683			    &gpp.gpp_force);
1684			break;
1685		case G_PART_PARM_GEOM:
1686			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1687			break;
1688		case G_PART_PARM_INDEX:
1689			error = g_part_parm_intmax(req, ap->name,
1690			    &gpp.gpp_index);
1691			break;
1692		case G_PART_PARM_LABEL:
1693			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1694			break;
1695		case G_PART_PARM_OUTPUT:
1696			error = 0;	/* Write-only parameter */
1697			break;
1698		case G_PART_PARM_PROVIDER:
1699			error = g_part_parm_provider(req, ap->name,
1700			    &gpp.gpp_provider);
1701			break;
1702		case G_PART_PARM_SCHEME:
1703			error = g_part_parm_scheme(req, ap->name,
1704			    &gpp.gpp_scheme);
1705			break;
1706		case G_PART_PARM_SIZE:
1707			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1708			break;
1709		case G_PART_PARM_START:
1710			error = g_part_parm_quad(req, ap->name,
1711			    &gpp.gpp_start);
1712			break;
1713		case G_PART_PARM_TYPE:
1714			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1715			break;
1716		case G_PART_PARM_VERSION:
1717			error = g_part_parm_uint32(req, ap->name,
1718			    &gpp.gpp_version);
1719			break;
1720		default:
1721			error = EDOOFUS;
1722			gctl_error(req, "%d %s", error, ap->name);
1723			break;
1724		}
1725		if (error != 0) {
1726			if (error == ENOATTR) {
1727				gctl_error(req, "%d param '%s'", error,
1728				    ap->name);
1729			}
1730			return;
1731		}
1732		gpp.gpp_parms |= parm;
1733	}
1734	if ((gpp.gpp_parms & mparms) != mparms) {
1735		parm = mparms - (gpp.gpp_parms & mparms);
1736		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1737		return;
1738	}
1739
1740	/* Obtain permissions if possible/necessary. */
1741	close_on_error = 0;
1742	table = NULL;
1743	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1744		table = gpp.gpp_geom->softc;
1745		if (table != NULL && table->gpt_corrupt &&
1746		    ctlreq != G_PART_CTL_DESTROY &&
1747		    ctlreq != G_PART_CTL_RECOVER) {
1748			gctl_error(req, "%d table '%s' is corrupt",
1749			    EPERM, gpp.gpp_geom->name);
1750			return;
1751		}
1752		if (table != NULL && !table->gpt_opened) {
1753			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1754			    1, 1, 1);
1755			if (error) {
1756				gctl_error(req, "%d geom '%s'", error,
1757				    gpp.gpp_geom->name);
1758				return;
1759			}
1760			table->gpt_opened = 1;
1761			close_on_error = 1;
1762		}
1763	}
1764
1765	/* Allow the scheme to check or modify the parameters. */
1766	if (table != NULL) {
1767		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1768		if (error) {
1769			gctl_error(req, "%d pre-check failed", error);
1770			goto out;
1771		}
1772	} else
1773		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1774
1775	switch (ctlreq) {
1776	case G_PART_CTL_NONE:
1777		panic("%s", __func__);
1778	case G_PART_CTL_ADD:
1779		error = g_part_ctl_add(req, &gpp);
1780		break;
1781	case G_PART_CTL_BOOTCODE:
1782		error = g_part_ctl_bootcode(req, &gpp);
1783		break;
1784	case G_PART_CTL_COMMIT:
1785		error = g_part_ctl_commit(req, &gpp);
1786		break;
1787	case G_PART_CTL_CREATE:
1788		error = g_part_ctl_create(req, &gpp);
1789		break;
1790	case G_PART_CTL_DELETE:
1791		error = g_part_ctl_delete(req, &gpp);
1792		break;
1793	case G_PART_CTL_DESTROY:
1794		error = g_part_ctl_destroy(req, &gpp);
1795		break;
1796	case G_PART_CTL_MODIFY:
1797		error = g_part_ctl_modify(req, &gpp);
1798		break;
1799	case G_PART_CTL_MOVE:
1800		error = g_part_ctl_move(req, &gpp);
1801		break;
1802	case G_PART_CTL_RECOVER:
1803		error = g_part_ctl_recover(req, &gpp);
1804		break;
1805	case G_PART_CTL_RESIZE:
1806		error = g_part_ctl_resize(req, &gpp);
1807		break;
1808	case G_PART_CTL_SET:
1809		error = g_part_ctl_setunset(req, &gpp, 1);
1810		break;
1811	case G_PART_CTL_UNDO:
1812		error = g_part_ctl_undo(req, &gpp);
1813		break;
1814	case G_PART_CTL_UNSET:
1815		error = g_part_ctl_setunset(req, &gpp, 0);
1816		break;
1817	}
1818
1819	/* Implement automatic commit. */
1820	if (!error) {
1821		auto_commit = (modifies &&
1822		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1823		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1824		if (auto_commit) {
1825			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1826			    __func__));
1827			error = g_part_ctl_commit(req, &gpp);
1828		}
1829	}
1830
1831 out:
1832	if (error && close_on_error) {
1833		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1834		table->gpt_opened = 0;
1835	}
1836}
1837
1838static int
1839g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1840    struct g_geom *gp)
1841{
1842
1843	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1844	g_topology_assert();
1845
1846	g_part_wither(gp, EINVAL);
1847	return (0);
1848}
1849
1850static struct g_geom *
1851g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1852{
1853	struct g_consumer *cp;
1854	struct g_geom *gp;
1855	struct g_part_entry *entry;
1856	struct g_part_table *table;
1857	struct root_hold_token *rht;
1858	int attr, depth;
1859	int error;
1860
1861	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1862	g_topology_assert();
1863
1864	/* Skip providers that are already open for writing. */
1865	if (pp->acw > 0)
1866		return (NULL);
1867
1868	/*
1869	 * Create a GEOM with consumer and hook it up to the provider.
1870	 * With that we become part of the topology. Optain read access
1871	 * to the provider.
1872	 */
1873	gp = g_new_geomf(mp, "%s", pp->name);
1874	cp = g_new_consumer(gp);
1875	error = g_attach(cp, pp);
1876	if (error == 0)
1877		error = g_access(cp, 1, 0, 0);
1878	if (error != 0) {
1879		g_part_wither(gp, error);
1880		return (NULL);
1881	}
1882
1883	rht = root_mount_hold(mp->name);
1884	g_topology_unlock();
1885
1886	/*
1887	 * Short-circuit the whole probing galore when there's no
1888	 * media present.
1889	 */
1890	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1891		error = ENODEV;
1892		goto fail;
1893	}
1894
1895	/* Make sure we can nest and if so, determine our depth. */
1896	error = g_getattr("PART::isleaf", cp, &attr);
1897	if (!error && attr) {
1898		error = ENODEV;
1899		goto fail;
1900	}
1901	error = g_getattr("PART::depth", cp, &attr);
1902	depth = (!error) ? attr + 1 : 0;
1903
1904	error = g_part_probe(gp, cp, depth);
1905	if (error)
1906		goto fail;
1907
1908	table = gp->softc;
1909
1910	/*
1911	 * Synthesize a disk geometry. Some partitioning schemes
1912	 * depend on it and since some file systems need it even
1913	 * when the partitition scheme doesn't, we do it here in
1914	 * scheme-independent code.
1915	 */
1916	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1917
1918	error = G_PART_READ(table, cp);
1919	if (error)
1920		goto fail;
1921	error = g_part_check_integrity(table, cp);
1922	if (error)
1923		goto fail;
1924
1925	g_topology_lock();
1926	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1927		if (!entry->gpe_internal)
1928			g_part_new_provider(gp, table, entry);
1929	}
1930
1931	root_mount_rel(rht);
1932	g_access(cp, -1, 0, 0);
1933	return (gp);
1934
1935 fail:
1936	g_topology_lock();
1937	root_mount_rel(rht);
1938	g_access(cp, -1, 0, 0);
1939	g_part_wither(gp, error);
1940	return (NULL);
1941}
1942
1943/*
1944 * Geom methods.
1945 */
1946
1947static int
1948g_part_access(struct g_provider *pp, int dr, int dw, int de)
1949{
1950	struct g_consumer *cp;
1951
1952	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1953	    dw, de));
1954
1955	cp = LIST_FIRST(&pp->geom->consumer);
1956
1957	/* We always gain write-exclusive access. */
1958	return (g_access(cp, dr, dw, dw + de));
1959}
1960
1961static void
1962g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1963    struct g_consumer *cp, struct g_provider *pp)
1964{
1965	char buf[64];
1966	struct g_part_entry *entry;
1967	struct g_part_table *table;
1968
1969	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
1970	table = gp->softc;
1971
1972	if (indent == NULL) {
1973		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
1974		entry = pp->private;
1975		if (entry == NULL)
1976			return;
1977		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
1978		    (uintmax_t)entry->gpe_offset,
1979		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1980		/*
1981		 * libdisk compatibility quirk - the scheme dumps the
1982		 * slicer name and partition type in a way that is
1983		 * compatible with libdisk. When libdisk is not used
1984		 * anymore, this should go away.
1985		 */
1986		G_PART_DUMPCONF(table, entry, sb, indent);
1987	} else if (cp != NULL) {	/* Consumer configuration. */
1988		KASSERT(pp == NULL, ("%s", __func__));
1989		/* none */
1990	} else if (pp != NULL) {	/* Provider configuration. */
1991		entry = pp->private;
1992		if (entry == NULL)
1993			return;
1994		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
1995		    (uintmax_t)entry->gpe_start);
1996		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
1997		    (uintmax_t)entry->gpe_end);
1998		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
1999		    entry->gpe_index);
2000		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2001		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2002		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
2003		    (uintmax_t)entry->gpe_offset);
2004		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
2005		    (uintmax_t)pp->mediasize);
2006		G_PART_DUMPCONF(table, entry, sb, indent);
2007	} else {			/* Geom configuration. */
2008		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2009		    table->gpt_scheme->name);
2010		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2011		    table->gpt_entries);
2012		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2013		    (uintmax_t)table->gpt_first);
2014		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2015		    (uintmax_t)table->gpt_last);
2016		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2017		    table->gpt_sectors);
2018		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2019		    table->gpt_heads);
2020		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2021		    table->gpt_corrupt ? "CORRUPT": "OK");
2022		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2023		    table->gpt_opened ? "true": "false");
2024		G_PART_DUMPCONF(table, NULL, sb, indent);
2025	}
2026}
2027
2028static void
2029g_part_orphan(struct g_consumer *cp)
2030{
2031	struct g_provider *pp;
2032	struct g_part_table *table;
2033
2034	pp = cp->provider;
2035	KASSERT(pp != NULL, ("%s", __func__));
2036	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2037	g_topology_assert();
2038
2039	KASSERT(pp->error != 0, ("%s", __func__));
2040	table = cp->geom->softc;
2041	if (table != NULL && table->gpt_opened)
2042		g_access(cp, -1, -1, -1);
2043	g_part_wither(cp->geom, pp->error);
2044}
2045
2046static void
2047g_part_spoiled(struct g_consumer *cp)
2048{
2049
2050	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2051	g_topology_assert();
2052
2053	g_part_wither(cp->geom, ENXIO);
2054}
2055
2056static void
2057g_part_start(struct bio *bp)
2058{
2059	struct bio *bp2;
2060	struct g_consumer *cp;
2061	struct g_geom *gp;
2062	struct g_part_entry *entry;
2063	struct g_part_table *table;
2064	struct g_kerneldump *gkd;
2065	struct g_provider *pp;
2066	char buf[64];
2067
2068	pp = bp->bio_to;
2069	gp = pp->geom;
2070	table = gp->softc;
2071	cp = LIST_FIRST(&gp->consumer);
2072
2073	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2074	    pp->name));
2075
2076	entry = pp->private;
2077	if (entry == NULL) {
2078		g_io_deliver(bp, ENXIO);
2079		return;
2080	}
2081
2082	switch(bp->bio_cmd) {
2083	case BIO_DELETE:
2084	case BIO_READ:
2085	case BIO_WRITE:
2086		if (bp->bio_offset >= pp->mediasize) {
2087			g_io_deliver(bp, EIO);
2088			return;
2089		}
2090		bp2 = g_clone_bio(bp);
2091		if (bp2 == NULL) {
2092			g_io_deliver(bp, ENOMEM);
2093			return;
2094		}
2095		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2096			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2097		bp2->bio_done = g_std_done;
2098		bp2->bio_offset += entry->gpe_offset;
2099		g_io_request(bp2, cp);
2100		return;
2101	case BIO_FLUSH:
2102		break;
2103	case BIO_GETATTR:
2104		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2105			return;
2106		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2107			return;
2108		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2109			return;
2110		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2111			return;
2112		if (g_handleattr_str(bp, "PART::scheme",
2113		    table->gpt_scheme->name))
2114			return;
2115		if (g_handleattr_str(bp, "PART::type",
2116		    G_PART_TYPE(table, entry, buf, sizeof(buf))))
2117			return;
2118		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2119			/*
2120			 * Check that the partition is suitable for kernel
2121			 * dumps. Typically only swap partitions should be
2122			 * used. If the request comes from the nested scheme
2123			 * we allow dumping there as well.
2124			 */
2125			if ((bp->bio_from == NULL ||
2126			    bp->bio_from->geom->class != &g_part_class) &&
2127			    G_PART_DUMPTO(table, entry) == 0) {
2128				g_io_deliver(bp, ENODEV);
2129				printf("GEOM_PART: Partition '%s' not suitable"
2130				    " for kernel dumps (wrong type?)\n",
2131				    pp->name);
2132				return;
2133			}
2134			gkd = (struct g_kerneldump *)bp->bio_data;
2135			if (gkd->offset >= pp->mediasize) {
2136				g_io_deliver(bp, EIO);
2137				return;
2138			}
2139			if (gkd->offset + gkd->length > pp->mediasize)
2140				gkd->length = pp->mediasize - gkd->offset;
2141			gkd->offset += entry->gpe_offset;
2142		}
2143		break;
2144	default:
2145		g_io_deliver(bp, EOPNOTSUPP);
2146		return;
2147	}
2148
2149	bp2 = g_clone_bio(bp);
2150	if (bp2 == NULL) {
2151		g_io_deliver(bp, ENOMEM);
2152		return;
2153	}
2154	bp2->bio_done = g_std_done;
2155	g_io_request(bp2, cp);
2156}
2157
2158static void
2159g_part_init(struct g_class *mp)
2160{
2161
2162	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2163}
2164
2165static void
2166g_part_fini(struct g_class *mp)
2167{
2168
2169	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2170}
2171
2172static void
2173g_part_unload_event(void *arg, int flag)
2174{
2175	struct g_consumer *cp;
2176	struct g_geom *gp;
2177	struct g_provider *pp;
2178	struct g_part_scheme *scheme;
2179	struct g_part_table *table;
2180	uintptr_t *xchg;
2181	int acc, error;
2182
2183	if (flag == EV_CANCEL)
2184		return;
2185
2186	xchg = arg;
2187	error = 0;
2188	scheme = (void *)(*xchg);
2189
2190	g_topology_assert();
2191
2192	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2193		table = gp->softc;
2194		if (table->gpt_scheme != scheme)
2195			continue;
2196
2197		acc = 0;
2198		LIST_FOREACH(pp, &gp->provider, provider)
2199			acc += pp->acr + pp->acw + pp->ace;
2200		LIST_FOREACH(cp, &gp->consumer, consumer)
2201			acc += cp->acr + cp->acw + cp->ace;
2202
2203		if (!acc)
2204			g_part_wither(gp, ENOSYS);
2205		else
2206			error = EBUSY;
2207	}
2208
2209	if (!error)
2210		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2211
2212	*xchg = error;
2213}
2214
2215int
2216g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2217{
2218	struct g_part_scheme *iter;
2219	uintptr_t arg;
2220	int error;
2221
2222	error = 0;
2223	switch (type) {
2224	case MOD_LOAD:
2225		TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
2226			if (scheme == iter) {
2227				printf("GEOM_PART: scheme %s is already "
2228				    "registered!\n", scheme->name);
2229				break;
2230			}
2231		}
2232		if (iter == NULL) {
2233			TAILQ_INSERT_TAIL(&g_part_schemes, scheme,
2234			    scheme_list);
2235			g_retaste(&g_part_class);
2236		}
2237		break;
2238	case MOD_UNLOAD:
2239		arg = (uintptr_t)scheme;
2240		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2241		    NULL);
2242		if (error == 0)
2243			error = arg;
2244		break;
2245	default:
2246		error = EOPNOTSUPP;
2247		break;
2248	}
2249
2250	return (error);
2251}
2252