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/11/sys/geom/part/g_part.c 365923 2020-09-20 05:57:23Z eugen $");
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	{ "chromeos-firmware", G_PART_ALIAS_CHROMEOS_FIRMWARE },
82	{ "chromeos-kernel", G_PART_ALIAS_CHROMEOS_KERNEL },
83	{ "chromeos-reserved", G_PART_ALIAS_CHROMEOS_RESERVED },
84	{ "chromeos-root", G_PART_ALIAS_CHROMEOS_ROOT },
85	{ "dragonfly-ccd", G_PART_ALIAS_DFBSD_CCD },
86	{ "dragonfly-hammer", G_PART_ALIAS_DFBSD_HAMMER },
87	{ "dragonfly-hammer2", G_PART_ALIAS_DFBSD_HAMMER2 },
88	{ "dragonfly-label32", G_PART_ALIAS_DFBSD },
89	{ "dragonfly-label64", G_PART_ALIAS_DFBSD64 },
90	{ "dragonfly-legacy", G_PART_ALIAS_DFBSD_LEGACY },
91	{ "dragonfly-swap", G_PART_ALIAS_DFBSD_SWAP },
92	{ "dragonfly-ufs", G_PART_ALIAS_DFBSD_UFS },
93	{ "dragonfly-vinum", G_PART_ALIAS_DFBSD_VINUM },
94	{ "ebr", G_PART_ALIAS_EBR },
95	{ "efi", G_PART_ALIAS_EFI },
96	{ "fat16", G_PART_ALIAS_MS_FAT16 },
97	{ "fat32", G_PART_ALIAS_MS_FAT32 },
98	{ "fat32lba", G_PART_ALIAS_MS_FAT32LBA },
99	{ "freebsd", G_PART_ALIAS_FREEBSD },
100	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
101	{ "freebsd-nandfs", G_PART_ALIAS_FREEBSD_NANDFS },
102	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
103	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
104	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
105	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
106	{ "linux-data", G_PART_ALIAS_LINUX_DATA },
107	{ "linux-lvm", G_PART_ALIAS_LINUX_LVM },
108	{ "linux-raid", G_PART_ALIAS_LINUX_RAID },
109	{ "linux-swap", G_PART_ALIAS_LINUX_SWAP },
110	{ "mbr", G_PART_ALIAS_MBR },
111	{ "ms-basic-data", G_PART_ALIAS_MS_BASIC_DATA },
112	{ "ms-ldm-data", G_PART_ALIAS_MS_LDM_DATA },
113	{ "ms-ldm-metadata", G_PART_ALIAS_MS_LDM_METADATA },
114	{ "ms-recovery", G_PART_ALIAS_MS_RECOVERY },
115	{ "ms-reserved", G_PART_ALIAS_MS_RESERVED },
116	{ "ms-spaces", G_PART_ALIAS_MS_SPACES },
117	{ "netbsd-ccd", G_PART_ALIAS_NETBSD_CCD },
118	{ "netbsd-cgd", G_PART_ALIAS_NETBSD_CGD },
119	{ "netbsd-ffs", G_PART_ALIAS_NETBSD_FFS },
120	{ "netbsd-lfs", G_PART_ALIAS_NETBSD_LFS },
121	{ "netbsd-raid", G_PART_ALIAS_NETBSD_RAID },
122	{ "netbsd-swap", G_PART_ALIAS_NETBSD_SWAP },
123	{ "ntfs", G_PART_ALIAS_MS_NTFS },
124	{ "openbsd-data", G_PART_ALIAS_OPENBSD_DATA },
125	{ "prep-boot", G_PART_ALIAS_PREP_BOOT },
126	{ "vmware-reserved", G_PART_ALIAS_VMRESERVED },
127	{ "vmware-vmfs", G_PART_ALIAS_VMFS },
128	{ "vmware-vmkdiag", G_PART_ALIAS_VMKDIAG },
129	{ "vmware-vsanhdr", G_PART_ALIAS_VMVSANHDR },
130};
131
132SYSCTL_DECL(_kern_geom);
133SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
134    "GEOM_PART stuff");
135u_int geom_part_check_integrity = 1;
136SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity,
137    CTLFLAG_RWTUN, &geom_part_check_integrity, 1,
138    "Enable integrity checking");
139
140/*
141 * The GEOM partitioning class.
142 */
143static g_ctl_req_t g_part_ctlreq;
144static g_ctl_destroy_geom_t g_part_destroy_geom;
145static g_fini_t g_part_fini;
146static g_init_t g_part_init;
147static g_taste_t g_part_taste;
148
149static g_access_t g_part_access;
150static g_dumpconf_t g_part_dumpconf;
151static g_orphan_t g_part_orphan;
152static g_spoiled_t g_part_spoiled;
153static g_start_t g_part_start;
154static g_resize_t g_part_resize;
155static g_ioctl_t g_part_ioctl;
156
157static struct g_class g_part_class = {
158	.name = "PART",
159	.version = G_VERSION,
160	/* Class methods. */
161	.ctlreq = g_part_ctlreq,
162	.destroy_geom = g_part_destroy_geom,
163	.fini = g_part_fini,
164	.init = g_part_init,
165	.taste = g_part_taste,
166	/* Geom methods. */
167	.access = g_part_access,
168	.dumpconf = g_part_dumpconf,
169	.orphan = g_part_orphan,
170	.spoiled = g_part_spoiled,
171	.start = g_part_start,
172	.resize = g_part_resize,
173	.ioctl = g_part_ioctl,
174};
175
176DECLARE_GEOM_CLASS(g_part_class, g_part);
177MODULE_VERSION(g_part, 0);
178
179/*
180 * Support functions.
181 */
182
183static void g_part_wither(struct g_geom *, int);
184
185const char *
186g_part_alias_name(enum g_part_alias alias)
187{
188	int i;
189
190	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
191		if (g_part_alias_list[i].alias != alias)
192			continue;
193		return (g_part_alias_list[i].lexeme);
194	}
195
196	return (NULL);
197}
198
199void
200g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
201    u_int *bestheads)
202{
203	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
204	off_t chs, cylinders;
205	u_int heads;
206	int idx;
207
208	*bestchs = 0;
209	*bestheads = 0;
210	for (idx = 0; candidate_heads[idx] != 0; idx++) {
211		heads = candidate_heads[idx];
212		cylinders = blocks / heads / sectors;
213		if (cylinders < heads || cylinders < sectors)
214			break;
215		if (cylinders > 1023)
216			continue;
217		chs = cylinders * heads * sectors;
218		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
219			*bestchs = chs;
220			*bestheads = heads;
221		}
222	}
223}
224
225static void
226g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
227    off_t blocks)
228{
229	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
230	off_t chs, bestchs;
231	u_int heads, sectors;
232	int idx;
233
234	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
235	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
236		table->gpt_fixgeom = 0;
237		table->gpt_heads = 0;
238		table->gpt_sectors = 0;
239		bestchs = 0;
240		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
241			sectors = candidate_sectors[idx];
242			g_part_geometry_heads(blocks, sectors, &chs, &heads);
243			if (chs == 0)
244				continue;
245			/*
246			 * Prefer a geometry with sectors > 1, but only if
247			 * it doesn't bump down the number of heads to 1.
248			 */
249			if (chs > bestchs || (chs == bestchs && heads > 1 &&
250			    table->gpt_sectors == 1)) {
251				bestchs = chs;
252				table->gpt_heads = heads;
253				table->gpt_sectors = sectors;
254			}
255		}
256		/*
257		 * If we didn't find a geometry at all, then the disk is
258		 * too big. This means we can use the maximum number of
259		 * heads and sectors.
260		 */
261		if (bestchs == 0) {
262			table->gpt_heads = 255;
263			table->gpt_sectors = 63;
264		}
265	} else {
266		table->gpt_fixgeom = 1;
267		table->gpt_heads = heads;
268		table->gpt_sectors = sectors;
269	}
270}
271
272static void
273g_part_get_physpath_done(struct bio *bp)
274{
275	struct g_geom *gp;
276	struct g_part_entry *entry;
277	struct g_part_table *table;
278	struct g_provider *pp;
279	struct bio *pbp;
280
281	pbp = bp->bio_parent;
282	pp = pbp->bio_to;
283	gp = pp->geom;
284	table = gp->softc;
285	entry = pp->private;
286
287	if (bp->bio_error == 0) {
288		char *end;
289		size_t len, remainder;
290		len = strlcat(bp->bio_data, "/", bp->bio_length);
291		if (len < bp->bio_length) {
292			end = bp->bio_data + len;
293			remainder = bp->bio_length - len;
294			G_PART_NAME(table, entry, end, remainder);
295		}
296	}
297	g_std_done(bp);
298}
299
300
301#define	DPRINTF(...)	if (bootverbose) {	\
302	printf("GEOM_PART: " __VA_ARGS__);	\
303}
304
305static int
306g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
307{
308	struct g_part_entry *e1, *e2;
309	struct g_provider *pp;
310	off_t offset;
311	int failed;
312
313	failed = 0;
314	pp = cp->provider;
315	if (table->gpt_last < table->gpt_first) {
316		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
317		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
318		failed++;
319	}
320	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
321		DPRINTF("last LBA extends beyond mediasize: "
322		    "%jd > %jd\n", (intmax_t)table->gpt_last,
323		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
324		failed++;
325	}
326	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
327		if (e1->gpe_deleted || e1->gpe_internal)
328			continue;
329		if (e1->gpe_start < table->gpt_first) {
330			DPRINTF("partition %d has start offset below first "
331			    "LBA: %jd < %jd\n", e1->gpe_index,
332			    (intmax_t)e1->gpe_start,
333			    (intmax_t)table->gpt_first);
334			failed++;
335		}
336		if (e1->gpe_start > table->gpt_last) {
337			DPRINTF("partition %d has start offset beyond last "
338			    "LBA: %jd > %jd\n", e1->gpe_index,
339			    (intmax_t)e1->gpe_start,
340			    (intmax_t)table->gpt_last);
341			failed++;
342		}
343		if (e1->gpe_end < e1->gpe_start) {
344			DPRINTF("partition %d has end offset below start "
345			    "offset: %jd < %jd\n", e1->gpe_index,
346			    (intmax_t)e1->gpe_end,
347			    (intmax_t)e1->gpe_start);
348			failed++;
349		}
350		if (e1->gpe_end > table->gpt_last) {
351			DPRINTF("partition %d has end offset beyond last "
352			    "LBA: %jd > %jd\n", e1->gpe_index,
353			    (intmax_t)e1->gpe_end,
354			    (intmax_t)table->gpt_last);
355			failed++;
356		}
357		if (pp->stripesize > 0) {
358			offset = e1->gpe_start * pp->sectorsize;
359			if (e1->gpe_offset > offset)
360				offset = e1->gpe_offset;
361			if ((offset + pp->stripeoffset) % pp->stripesize) {
362				DPRINTF("partition %d on (%s, %s) is not "
363				    "aligned on %u bytes\n", e1->gpe_index,
364				    pp->name, table->gpt_scheme->name,
365				    pp->stripesize);
366				/* Don't treat this as a critical failure */
367			}
368		}
369		e2 = e1;
370		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
371			if (e2->gpe_deleted || e2->gpe_internal)
372				continue;
373			if (e1->gpe_start >= e2->gpe_start &&
374			    e1->gpe_start <= e2->gpe_end) {
375				DPRINTF("partition %d has start offset inside "
376				    "partition %d: start[%d] %jd >= start[%d] "
377				    "%jd <= end[%d] %jd\n",
378				    e1->gpe_index, e2->gpe_index,
379				    e2->gpe_index, (intmax_t)e2->gpe_start,
380				    e1->gpe_index, (intmax_t)e1->gpe_start,
381				    e2->gpe_index, (intmax_t)e2->gpe_end);
382				failed++;
383			}
384			if (e1->gpe_end >= e2->gpe_start &&
385			    e1->gpe_end <= e2->gpe_end) {
386				DPRINTF("partition %d has end offset inside "
387				    "partition %d: start[%d] %jd >= end[%d] "
388				    "%jd <= end[%d] %jd\n",
389				    e1->gpe_index, e2->gpe_index,
390				    e2->gpe_index, (intmax_t)e2->gpe_start,
391				    e1->gpe_index, (intmax_t)e1->gpe_end,
392				    e2->gpe_index, (intmax_t)e2->gpe_end);
393				failed++;
394			}
395			if (e1->gpe_start < e2->gpe_start &&
396			    e1->gpe_end > e2->gpe_end) {
397				DPRINTF("partition %d contains partition %d: "
398				    "start[%d] %jd > start[%d] %jd, end[%d] "
399				    "%jd < end[%d] %jd\n",
400				    e1->gpe_index, e2->gpe_index,
401				    e1->gpe_index, (intmax_t)e1->gpe_start,
402				    e2->gpe_index, (intmax_t)e2->gpe_start,
403				    e2->gpe_index, (intmax_t)e2->gpe_end,
404				    e1->gpe_index, (intmax_t)e1->gpe_end);
405				failed++;
406			}
407		}
408	}
409	if (failed != 0) {
410		printf("GEOM_PART: integrity check failed (%s, %s)\n",
411		    pp->name, table->gpt_scheme->name);
412		if (geom_part_check_integrity != 0)
413			return (EINVAL);
414		table->gpt_corrupt = 1;
415	}
416	return (0);
417}
418#undef	DPRINTF
419
420struct g_part_entry *
421g_part_new_entry(struct g_part_table *table, int index, quad_t start,
422    quad_t end)
423{
424	struct g_part_entry *entry, *last;
425
426	last = NULL;
427	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
428		if (entry->gpe_index == index)
429			break;
430		if (entry->gpe_index > index) {
431			entry = NULL;
432			break;
433		}
434		last = entry;
435	}
436	if (entry == NULL) {
437		entry = g_malloc(table->gpt_scheme->gps_entrysz,
438		    M_WAITOK | M_ZERO);
439		entry->gpe_index = index;
440		if (last == NULL)
441			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
442		else
443			LIST_INSERT_AFTER(last, entry, gpe_entry);
444	} else
445		entry->gpe_offset = 0;
446	entry->gpe_start = start;
447	entry->gpe_end = end;
448	return (entry);
449}
450
451static void
452g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
453    struct g_part_entry *entry)
454{
455	struct g_consumer *cp;
456	struct g_provider *pp;
457	struct sbuf *sb;
458	off_t offset;
459
460	cp = LIST_FIRST(&gp->consumer);
461	pp = cp->provider;
462
463	offset = entry->gpe_start * pp->sectorsize;
464	if (entry->gpe_offset < offset)
465		entry->gpe_offset = offset;
466
467	if (entry->gpe_pp == NULL) {
468		sb = sbuf_new_auto();
469		G_PART_FULLNAME(table, entry, sb, gp->name);
470		sbuf_finish(sb);
471		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
472		sbuf_delete(sb);
473		entry->gpe_pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
474		entry->gpe_pp->private = entry;		/* Close the circle. */
475	}
476	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
477	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
478	    pp->sectorsize;
479	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
480	entry->gpe_pp->sectorsize = pp->sectorsize;
481	entry->gpe_pp->stripesize = pp->stripesize;
482	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
483	if (pp->stripesize > 0)
484		entry->gpe_pp->stripeoffset %= pp->stripesize;
485	entry->gpe_pp->flags |= pp->flags & G_PF_ACCEPT_UNMAPPED;
486	g_error_provider(entry->gpe_pp, 0);
487}
488
489static struct g_geom*
490g_part_find_geom(const char *name)
491{
492	struct g_geom *gp;
493	LIST_FOREACH(gp, &g_part_class.geom, geom) {
494		if ((gp->flags & G_GEOM_WITHER) == 0 &&
495		    strcmp(name, gp->name) == 0)
496			break;
497	}
498	return (gp);
499}
500
501static int
502g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
503{
504	struct g_geom *gp;
505	const char *gname;
506
507	gname = gctl_get_asciiparam(req, name);
508	if (gname == NULL)
509		return (ENOATTR);
510	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
511		gname += sizeof(_PATH_DEV) - 1;
512	gp = g_part_find_geom(gname);
513	if (gp == NULL) {
514		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
515		return (EINVAL);
516	}
517	*v = gp;
518	return (0);
519}
520
521static int
522g_part_parm_provider(struct gctl_req *req, const char *name,
523    struct g_provider **v)
524{
525	struct g_provider *pp;
526	const char *pname;
527
528	pname = gctl_get_asciiparam(req, name);
529	if (pname == NULL)
530		return (ENOATTR);
531	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
532		pname += sizeof(_PATH_DEV) - 1;
533	pp = g_provider_by_name(pname);
534	if (pp == NULL) {
535		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
536		return (EINVAL);
537	}
538	*v = pp;
539	return (0);
540}
541
542static int
543g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
544{
545	const char *p;
546	char *x;
547	quad_t q;
548
549	p = gctl_get_asciiparam(req, name);
550	if (p == NULL)
551		return (ENOATTR);
552	q = strtoq(p, &x, 0);
553	if (*x != '\0' || q < 0) {
554		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
555		return (EINVAL);
556	}
557	*v = q;
558	return (0);
559}
560
561static int
562g_part_parm_scheme(struct gctl_req *req, const char *name,
563    struct g_part_scheme **v)
564{
565	struct g_part_scheme *s;
566	const char *p;
567
568	p = gctl_get_asciiparam(req, name);
569	if (p == NULL)
570		return (ENOATTR);
571	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
572		if (s == &g_part_null_scheme)
573			continue;
574		if (!strcasecmp(s->name, p))
575			break;
576	}
577	if (s == NULL) {
578		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
579		return (EINVAL);
580	}
581	*v = s;
582	return (0);
583}
584
585static int
586g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
587{
588	const char *p;
589
590	p = gctl_get_asciiparam(req, name);
591	if (p == NULL)
592		return (ENOATTR);
593	/* An empty label is always valid. */
594	if (strcmp(name, "label") != 0 && p[0] == '\0') {
595		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
596		return (EINVAL);
597	}
598	*v = p;
599	return (0);
600}
601
602static int
603g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
604{
605	const intmax_t *p;
606	int size;
607
608	p = gctl_get_param(req, name, &size);
609	if (p == NULL)
610		return (ENOATTR);
611	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
612		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
613		return (EINVAL);
614	}
615	*v = (u_int)*p;
616	return (0);
617}
618
619static int
620g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
621{
622	const uint32_t *p;
623	int size;
624
625	p = gctl_get_param(req, name, &size);
626	if (p == NULL)
627		return (ENOATTR);
628	if (size != sizeof(*p) || *p > INT_MAX) {
629		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
630		return (EINVAL);
631	}
632	*v = (u_int)*p;
633	return (0);
634}
635
636static int
637g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
638    unsigned int *s)
639{
640	const void *p;
641	int size;
642
643	p = gctl_get_param(req, name, &size);
644	if (p == NULL)
645		return (ENOATTR);
646	*v = p;
647	*s = size;
648	return (0);
649}
650
651static int
652g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
653{
654	struct g_part_scheme *iter, *scheme;
655	struct g_part_table *table;
656	int pri, probe;
657
658	table = gp->softc;
659	scheme = (table != NULL) ? table->gpt_scheme : NULL;
660	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
661	if (pri == 0)
662		goto done;
663	if (pri > 0) {	/* error */
664		scheme = NULL;
665		pri = INT_MIN;
666	}
667
668	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
669		if (iter == &g_part_null_scheme)
670			continue;
671		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
672		    M_WAITOK);
673		table->gpt_gp = gp;
674		table->gpt_scheme = iter;
675		table->gpt_depth = depth;
676		probe = G_PART_PROBE(table, cp);
677		if (probe <= 0 && probe > pri) {
678			pri = probe;
679			scheme = iter;
680			if (gp->softc != NULL)
681				kobj_delete((kobj_t)gp->softc, M_GEOM);
682			gp->softc = table;
683			if (pri == 0)
684				goto done;
685		} else
686			kobj_delete((kobj_t)table, M_GEOM);
687	}
688
689done:
690	return ((scheme == NULL) ? ENXIO : 0);
691}
692
693/*
694 * Control request functions.
695 */
696
697static int
698g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
699{
700	struct g_geom *gp;
701	struct g_provider *pp;
702	struct g_part_entry *delent, *last, *entry;
703	struct g_part_table *table;
704	struct sbuf *sb;
705	quad_t end;
706	unsigned int index;
707	int error;
708
709	gp = gpp->gpp_geom;
710	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
711	g_topology_assert();
712
713	pp = LIST_FIRST(&gp->consumer)->provider;
714	table = gp->softc;
715	end = gpp->gpp_start + gpp->gpp_size - 1;
716
717	if (gpp->gpp_start < table->gpt_first ||
718	    gpp->gpp_start > table->gpt_last) {
719		gctl_error(req, "%d start '%jd'", EINVAL,
720		    (intmax_t)gpp->gpp_start);
721		return (EINVAL);
722	}
723	if (end < gpp->gpp_start || end > table->gpt_last) {
724		gctl_error(req, "%d size '%jd'", EINVAL,
725		    (intmax_t)gpp->gpp_size);
726		return (EINVAL);
727	}
728	if (gpp->gpp_index > table->gpt_entries) {
729		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
730		return (EINVAL);
731	}
732
733	delent = last = NULL;
734	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
735	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
736		if (entry->gpe_deleted) {
737			if (entry->gpe_index == index)
738				delent = entry;
739			continue;
740		}
741		if (entry->gpe_index == index)
742			index = entry->gpe_index + 1;
743		if (entry->gpe_index < index)
744			last = entry;
745		if (entry->gpe_internal)
746			continue;
747		if (gpp->gpp_start >= entry->gpe_start &&
748		    gpp->gpp_start <= entry->gpe_end) {
749			gctl_error(req, "%d start '%jd'", ENOSPC,
750			    (intmax_t)gpp->gpp_start);
751			return (ENOSPC);
752		}
753		if (end >= entry->gpe_start && end <= entry->gpe_end) {
754			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
755			return (ENOSPC);
756		}
757		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
758			gctl_error(req, "%d size '%jd'", ENOSPC,
759			    (intmax_t)gpp->gpp_size);
760			return (ENOSPC);
761		}
762	}
763	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
764		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
765		return (EEXIST);
766	}
767	if (index > table->gpt_entries) {
768		gctl_error(req, "%d index '%d'", ENOSPC, index);
769		return (ENOSPC);
770	}
771
772	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
773	    M_WAITOK | M_ZERO) : delent;
774	entry->gpe_index = index;
775	entry->gpe_start = gpp->gpp_start;
776	entry->gpe_end = end;
777	error = G_PART_ADD(table, entry, gpp);
778	if (error) {
779		gctl_error(req, "%d", error);
780		if (delent == NULL)
781			g_free(entry);
782		return (error);
783	}
784	if (delent == NULL) {
785		if (last == NULL)
786			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
787		else
788			LIST_INSERT_AFTER(last, entry, gpe_entry);
789		entry->gpe_created = 1;
790	} else {
791		entry->gpe_deleted = 0;
792		entry->gpe_modified = 1;
793	}
794	g_part_new_provider(gp, table, entry);
795
796	/* Provide feedback if so requested. */
797	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
798		sb = sbuf_new_auto();
799		G_PART_FULLNAME(table, entry, sb, gp->name);
800		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
801			sbuf_printf(sb, " added, but partition is not "
802			    "aligned on %u bytes\n", pp->stripesize);
803		else
804			sbuf_cat(sb, " added\n");
805		sbuf_finish(sb);
806		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
807		sbuf_delete(sb);
808	}
809	return (0);
810}
811
812static int
813g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
814{
815	struct g_geom *gp;
816	struct g_part_table *table;
817	struct sbuf *sb;
818	int error, sz;
819
820	gp = gpp->gpp_geom;
821	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
822	g_topology_assert();
823
824	table = gp->softc;
825	sz = table->gpt_scheme->gps_bootcodesz;
826	if (sz == 0) {
827		error = ENODEV;
828		goto fail;
829	}
830	if (gpp->gpp_codesize > sz) {
831		error = EFBIG;
832		goto fail;
833	}
834
835	error = G_PART_BOOTCODE(table, gpp);
836	if (error)
837		goto fail;
838
839	/* Provide feedback if so requested. */
840	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
841		sb = sbuf_new_auto();
842		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
843		sbuf_finish(sb);
844		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
845		sbuf_delete(sb);
846	}
847	return (0);
848
849 fail:
850	gctl_error(req, "%d", error);
851	return (error);
852}
853
854static int
855g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
856{
857	struct g_consumer *cp;
858	struct g_geom *gp;
859	struct g_provider *pp;
860	struct g_part_entry *entry, *tmp;
861	struct g_part_table *table;
862	char *buf;
863	int error, i;
864
865	gp = gpp->gpp_geom;
866	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
867	g_topology_assert();
868
869	table = gp->softc;
870	if (!table->gpt_opened) {
871		gctl_error(req, "%d", EPERM);
872		return (EPERM);
873	}
874
875	g_topology_unlock();
876
877	cp = LIST_FIRST(&gp->consumer);
878	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
879		pp = cp->provider;
880		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
881		while (table->gpt_smhead != 0) {
882			i = ffs(table->gpt_smhead) - 1;
883			error = g_write_data(cp, i * pp->sectorsize, buf,
884			    pp->sectorsize);
885			if (error) {
886				g_free(buf);
887				goto fail;
888			}
889			table->gpt_smhead &= ~(1 << i);
890		}
891		while (table->gpt_smtail != 0) {
892			i = ffs(table->gpt_smtail) - 1;
893			error = g_write_data(cp, pp->mediasize - (i + 1) *
894			    pp->sectorsize, buf, pp->sectorsize);
895			if (error) {
896				g_free(buf);
897				goto fail;
898			}
899			table->gpt_smtail &= ~(1 << i);
900		}
901		g_free(buf);
902	}
903
904	if (table->gpt_scheme == &g_part_null_scheme) {
905		g_topology_lock();
906		g_access(cp, -1, -1, -1);
907		g_part_wither(gp, ENXIO);
908		return (0);
909	}
910
911	error = G_PART_WRITE(table, cp);
912	if (error)
913		goto fail;
914
915	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
916		if (!entry->gpe_deleted) {
917			/* Notify consumers that provider might be changed. */
918			if (entry->gpe_modified && (
919			    entry->gpe_pp->acw + entry->gpe_pp->ace +
920			    entry->gpe_pp->acr) == 0)
921				g_media_changed(entry->gpe_pp, M_NOWAIT);
922			entry->gpe_created = 0;
923			entry->gpe_modified = 0;
924			continue;
925		}
926		LIST_REMOVE(entry, gpe_entry);
927		g_free(entry);
928	}
929	table->gpt_created = 0;
930	table->gpt_opened = 0;
931
932	g_topology_lock();
933	g_access(cp, -1, -1, -1);
934	return (0);
935
936fail:
937	g_topology_lock();
938	gctl_error(req, "%d", error);
939	return (error);
940}
941
942static int
943g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
944{
945	struct g_consumer *cp;
946	struct g_geom *gp;
947	struct g_provider *pp;
948	struct g_part_scheme *scheme;
949	struct g_part_table *null, *table;
950	struct sbuf *sb;
951	int attr, error;
952
953	pp = gpp->gpp_provider;
954	scheme = gpp->gpp_scheme;
955	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
956	g_topology_assert();
957
958	/* Check that there isn't already a g_part geom on the provider. */
959	gp = g_part_find_geom(pp->name);
960	if (gp != NULL) {
961		null = gp->softc;
962		if (null->gpt_scheme != &g_part_null_scheme) {
963			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
964			return (EEXIST);
965		}
966	} else
967		null = NULL;
968
969	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
970	    (gpp->gpp_entries < scheme->gps_minent ||
971	     gpp->gpp_entries > scheme->gps_maxent)) {
972		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
973		return (EINVAL);
974	}
975
976	if (null == NULL)
977		gp = g_new_geomf(&g_part_class, "%s", pp->name);
978	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
979	    M_WAITOK);
980	table = gp->softc;
981	table->gpt_gp = gp;
982	table->gpt_scheme = gpp->gpp_scheme;
983	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
984	    gpp->gpp_entries : scheme->gps_minent;
985	LIST_INIT(&table->gpt_entry);
986	if (null == NULL) {
987		cp = g_new_consumer(gp);
988		cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
989		error = g_attach(cp, pp);
990		if (error == 0)
991			error = g_access(cp, 1, 1, 1);
992		if (error != 0) {
993			g_part_wither(gp, error);
994			gctl_error(req, "%d geom '%s'", error, pp->name);
995			return (error);
996		}
997		table->gpt_opened = 1;
998	} else {
999		cp = LIST_FIRST(&gp->consumer);
1000		table->gpt_opened = null->gpt_opened;
1001		table->gpt_smhead = null->gpt_smhead;
1002		table->gpt_smtail = null->gpt_smtail;
1003	}
1004
1005	g_topology_unlock();
1006
1007	/* Make sure the provider has media. */
1008	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1009		error = ENODEV;
1010		goto fail;
1011	}
1012
1013	/* Make sure we can nest and if so, determine our depth. */
1014	error = g_getattr("PART::isleaf", cp, &attr);
1015	if (!error && attr) {
1016		error = ENODEV;
1017		goto fail;
1018	}
1019	error = g_getattr("PART::depth", cp, &attr);
1020	table->gpt_depth = (!error) ? attr + 1 : 0;
1021
1022	/*
1023	 * Synthesize a disk geometry. Some partitioning schemes
1024	 * depend on it and since some file systems need it even
1025	 * when the partitition scheme doesn't, we do it here in
1026	 * scheme-independent code.
1027	 */
1028	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1029
1030	error = G_PART_CREATE(table, gpp);
1031	if (error)
1032		goto fail;
1033
1034	g_topology_lock();
1035
1036	table->gpt_created = 1;
1037	if (null != NULL)
1038		kobj_delete((kobj_t)null, M_GEOM);
1039
1040	/*
1041	 * Support automatic commit by filling in the gpp_geom
1042	 * parameter.
1043	 */
1044	gpp->gpp_parms |= G_PART_PARM_GEOM;
1045	gpp->gpp_geom = gp;
1046
1047	/* Provide feedback if so requested. */
1048	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1049		sb = sbuf_new_auto();
1050		sbuf_printf(sb, "%s created\n", gp->name);
1051		sbuf_finish(sb);
1052		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1053		sbuf_delete(sb);
1054	}
1055	return (0);
1056
1057fail:
1058	g_topology_lock();
1059	if (null == NULL) {
1060		g_access(cp, -1, -1, -1);
1061		g_part_wither(gp, error);
1062	} else {
1063		kobj_delete((kobj_t)gp->softc, M_GEOM);
1064		gp->softc = null;
1065	}
1066	gctl_error(req, "%d provider", error);
1067	return (error);
1068}
1069
1070static int
1071g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1072{
1073	struct g_geom *gp;
1074	struct g_provider *pp;
1075	struct g_part_entry *entry;
1076	struct g_part_table *table;
1077	struct sbuf *sb;
1078
1079	gp = gpp->gpp_geom;
1080	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1081	g_topology_assert();
1082
1083	table = gp->softc;
1084
1085	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1086		if (entry->gpe_deleted || entry->gpe_internal)
1087			continue;
1088		if (entry->gpe_index == gpp->gpp_index)
1089			break;
1090	}
1091	if (entry == NULL) {
1092		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1093		return (ENOENT);
1094	}
1095
1096	pp = entry->gpe_pp;
1097	if (pp != NULL) {
1098		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1099			gctl_error(req, "%d", EBUSY);
1100			return (EBUSY);
1101		}
1102
1103		pp->private = NULL;
1104		entry->gpe_pp = NULL;
1105	}
1106
1107	if (pp != NULL)
1108		g_wither_provider(pp, ENXIO);
1109
1110	/* Provide feedback if so requested. */
1111	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1112		sb = sbuf_new_auto();
1113		G_PART_FULLNAME(table, entry, sb, gp->name);
1114		sbuf_cat(sb, " deleted\n");
1115		sbuf_finish(sb);
1116		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1117		sbuf_delete(sb);
1118	}
1119
1120	if (entry->gpe_created) {
1121		LIST_REMOVE(entry, gpe_entry);
1122		g_free(entry);
1123	} else {
1124		entry->gpe_modified = 0;
1125		entry->gpe_deleted = 1;
1126	}
1127	return (0);
1128}
1129
1130static int
1131g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1132{
1133	struct g_consumer *cp;
1134	struct g_geom *gp;
1135	struct g_provider *pp;
1136	struct g_part_entry *entry, *tmp;
1137	struct g_part_table *null, *table;
1138	struct sbuf *sb;
1139	int error;
1140
1141	gp = gpp->gpp_geom;
1142	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1143	g_topology_assert();
1144
1145	table = gp->softc;
1146	/* Check for busy providers. */
1147	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1148		if (entry->gpe_deleted || entry->gpe_internal)
1149			continue;
1150		if (gpp->gpp_force) {
1151			pp = entry->gpe_pp;
1152			if (pp == NULL)
1153				continue;
1154			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1155				continue;
1156		}
1157		gctl_error(req, "%d", EBUSY);
1158		return (EBUSY);
1159	}
1160
1161	if (gpp->gpp_force) {
1162		/* Destroy all providers. */
1163		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1164			pp = entry->gpe_pp;
1165			if (pp != NULL) {
1166				pp->private = NULL;
1167				g_wither_provider(pp, ENXIO);
1168			}
1169			LIST_REMOVE(entry, gpe_entry);
1170			g_free(entry);
1171		}
1172	}
1173
1174	error = G_PART_DESTROY(table, gpp);
1175	if (error) {
1176		gctl_error(req, "%d", error);
1177		return (error);
1178	}
1179
1180	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1181	    M_WAITOK);
1182	null = gp->softc;
1183	null->gpt_gp = gp;
1184	null->gpt_scheme = &g_part_null_scheme;
1185	LIST_INIT(&null->gpt_entry);
1186
1187	cp = LIST_FIRST(&gp->consumer);
1188	pp = cp->provider;
1189	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1190
1191	null->gpt_depth = table->gpt_depth;
1192	null->gpt_opened = table->gpt_opened;
1193	null->gpt_smhead = table->gpt_smhead;
1194	null->gpt_smtail = table->gpt_smtail;
1195
1196	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1197		LIST_REMOVE(entry, gpe_entry);
1198		g_free(entry);
1199	}
1200	kobj_delete((kobj_t)table, M_GEOM);
1201
1202	/* Provide feedback if so requested. */
1203	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1204		sb = sbuf_new_auto();
1205		sbuf_printf(sb, "%s destroyed\n", gp->name);
1206		sbuf_finish(sb);
1207		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1208		sbuf_delete(sb);
1209	}
1210	return (0);
1211}
1212
1213static int
1214g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1215{
1216	struct g_geom *gp;
1217	struct g_part_entry *entry;
1218	struct g_part_table *table;
1219	struct sbuf *sb;
1220	int error;
1221
1222	gp = gpp->gpp_geom;
1223	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1224	g_topology_assert();
1225
1226	table = gp->softc;
1227
1228	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1229		if (entry->gpe_deleted || entry->gpe_internal)
1230			continue;
1231		if (entry->gpe_index == gpp->gpp_index)
1232			break;
1233	}
1234	if (entry == NULL) {
1235		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1236		return (ENOENT);
1237	}
1238
1239	error = G_PART_MODIFY(table, entry, gpp);
1240	if (error) {
1241		gctl_error(req, "%d", error);
1242		return (error);
1243	}
1244
1245	if (!entry->gpe_created)
1246		entry->gpe_modified = 1;
1247
1248	/* Provide feedback if so requested. */
1249	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1250		sb = sbuf_new_auto();
1251		G_PART_FULLNAME(table, entry, sb, gp->name);
1252		sbuf_cat(sb, " modified\n");
1253		sbuf_finish(sb);
1254		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1255		sbuf_delete(sb);
1256	}
1257	return (0);
1258}
1259
1260static int
1261g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1262{
1263	gctl_error(req, "%d verb 'move'", ENOSYS);
1264	return (ENOSYS);
1265}
1266
1267static int
1268g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1269{
1270	struct g_part_table *table;
1271	struct g_geom *gp;
1272	struct sbuf *sb;
1273	int error, recovered;
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	error = recovered = 0;
1280
1281	if (table->gpt_corrupt) {
1282		error = G_PART_RECOVER(table);
1283		if (error == 0)
1284			error = g_part_check_integrity(table,
1285			    LIST_FIRST(&gp->consumer));
1286		if (error) {
1287			gctl_error(req, "%d recovering '%s' failed",
1288			    error, gp->name);
1289			return (error);
1290		}
1291		recovered = 1;
1292	}
1293	/* Provide feedback if so requested. */
1294	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1295		sb = sbuf_new_auto();
1296		if (recovered)
1297			sbuf_printf(sb, "%s recovered\n", gp->name);
1298		else
1299			sbuf_printf(sb, "%s recovering is not needed\n",
1300			    gp->name);
1301		sbuf_finish(sb);
1302		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1303		sbuf_delete(sb);
1304	}
1305	return (0);
1306}
1307
1308static int
1309g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1310{
1311	struct g_geom *gp;
1312	struct g_provider *pp;
1313	struct g_part_entry *pe, *entry;
1314	struct g_part_table *table;
1315	struct sbuf *sb;
1316	quad_t end;
1317	int error;
1318	off_t mediasize;
1319
1320	gp = gpp->gpp_geom;
1321	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1322	g_topology_assert();
1323	table = gp->softc;
1324
1325	/* check gpp_index */
1326	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1327		if (entry->gpe_deleted || entry->gpe_internal)
1328			continue;
1329		if (entry->gpe_index == gpp->gpp_index)
1330			break;
1331	}
1332	if (entry == NULL) {
1333		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1334		return (ENOENT);
1335	}
1336
1337	/* check gpp_size */
1338	end = entry->gpe_start + gpp->gpp_size - 1;
1339	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1340		gctl_error(req, "%d size '%jd'", EINVAL,
1341		    (intmax_t)gpp->gpp_size);
1342		return (EINVAL);
1343	}
1344
1345	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1346		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1347			continue;
1348		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1349			gctl_error(req, "%d end '%jd'", ENOSPC,
1350			    (intmax_t)end);
1351			return (ENOSPC);
1352		}
1353		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1354			gctl_error(req, "%d size '%jd'", ENOSPC,
1355			    (intmax_t)gpp->gpp_size);
1356			return (ENOSPC);
1357		}
1358	}
1359
1360	pp = entry->gpe_pp;
1361	if ((g_debugflags & 16) == 0 &&
1362	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1363		if (entry->gpe_end - entry->gpe_start + 1 > gpp->gpp_size) {
1364			/* Deny shrinking of an opened partition. */
1365			gctl_error(req, "%d", EBUSY);
1366			return (EBUSY);
1367		}
1368	}
1369
1370	error = G_PART_RESIZE(table, entry, gpp);
1371	if (error) {
1372		gctl_error(req, "%d%s", error, error != EBUSY ? "":
1373		    " resizing will lead to unexpected shrinking"
1374		    " due to alignment");
1375		return (error);
1376	}
1377
1378	if (!entry->gpe_created)
1379		entry->gpe_modified = 1;
1380
1381	/* update mediasize of changed provider */
1382	mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1383		pp->sectorsize;
1384	g_resize_provider(pp, mediasize);
1385
1386	/* Provide feedback if so requested. */
1387	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1388		sb = sbuf_new_auto();
1389		G_PART_FULLNAME(table, entry, sb, gp->name);
1390		sbuf_cat(sb, " resized\n");
1391		sbuf_finish(sb);
1392		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1393		sbuf_delete(sb);
1394	}
1395	return (0);
1396}
1397
1398static int
1399g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1400    unsigned int set)
1401{
1402	struct g_geom *gp;
1403	struct g_part_entry *entry;
1404	struct g_part_table *table;
1405	struct sbuf *sb;
1406	int error;
1407
1408	gp = gpp->gpp_geom;
1409	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1410	g_topology_assert();
1411
1412	table = gp->softc;
1413
1414	if (gpp->gpp_parms & G_PART_PARM_INDEX) {
1415		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1416			if (entry->gpe_deleted || entry->gpe_internal)
1417				continue;
1418			if (entry->gpe_index == gpp->gpp_index)
1419				break;
1420		}
1421		if (entry == NULL) {
1422			gctl_error(req, "%d index '%d'", ENOENT,
1423			    gpp->gpp_index);
1424			return (ENOENT);
1425		}
1426	} else
1427		entry = NULL;
1428
1429	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1430	if (error) {
1431		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1432		return (error);
1433	}
1434
1435	/* Provide feedback if so requested. */
1436	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1437		sb = sbuf_new_auto();
1438		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1439		    (set) ? "" : "un");
1440		if (entry)
1441			G_PART_FULLNAME(table, entry, sb, gp->name);
1442		else
1443			sbuf_cat(sb, gp->name);
1444		sbuf_cat(sb, "\n");
1445		sbuf_finish(sb);
1446		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1447		sbuf_delete(sb);
1448	}
1449	return (0);
1450}
1451
1452static int
1453g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1454{
1455	struct g_consumer *cp;
1456	struct g_provider *pp;
1457	struct g_geom *gp;
1458	struct g_part_entry *entry, *tmp;
1459	struct g_part_table *table;
1460	int error, reprobe;
1461
1462	gp = gpp->gpp_geom;
1463	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1464	g_topology_assert();
1465
1466	table = gp->softc;
1467	if (!table->gpt_opened) {
1468		gctl_error(req, "%d", EPERM);
1469		return (EPERM);
1470	}
1471
1472	cp = LIST_FIRST(&gp->consumer);
1473	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1474		entry->gpe_modified = 0;
1475		if (entry->gpe_created) {
1476			pp = entry->gpe_pp;
1477			if (pp != NULL) {
1478				pp->private = NULL;
1479				entry->gpe_pp = NULL;
1480				g_wither_provider(pp, ENXIO);
1481			}
1482			entry->gpe_deleted = 1;
1483		}
1484		if (entry->gpe_deleted) {
1485			LIST_REMOVE(entry, gpe_entry);
1486			g_free(entry);
1487		}
1488	}
1489
1490	g_topology_unlock();
1491
1492	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1493	    table->gpt_created) ? 1 : 0;
1494
1495	if (reprobe) {
1496		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1497			if (entry->gpe_internal)
1498				continue;
1499			error = EBUSY;
1500			goto fail;
1501		}
1502		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1503			LIST_REMOVE(entry, gpe_entry);
1504			g_free(entry);
1505		}
1506		error = g_part_probe(gp, cp, table->gpt_depth);
1507		if (error) {
1508			g_topology_lock();
1509			g_access(cp, -1, -1, -1);
1510			g_part_wither(gp, error);
1511			return (0);
1512		}
1513		table = gp->softc;
1514
1515		/*
1516		 * Synthesize a disk geometry. Some partitioning schemes
1517		 * depend on it and since some file systems need it even
1518		 * when the partitition scheme doesn't, we do it here in
1519		 * scheme-independent code.
1520		 */
1521		pp = cp->provider;
1522		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1523	}
1524
1525	error = G_PART_READ(table, cp);
1526	if (error)
1527		goto fail;
1528	error = g_part_check_integrity(table, cp);
1529	if (error)
1530		goto fail;
1531
1532	g_topology_lock();
1533	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1534		if (!entry->gpe_internal)
1535			g_part_new_provider(gp, table, entry);
1536	}
1537
1538	table->gpt_opened = 0;
1539	g_access(cp, -1, -1, -1);
1540	return (0);
1541
1542fail:
1543	g_topology_lock();
1544	gctl_error(req, "%d", error);
1545	return (error);
1546}
1547
1548static void
1549g_part_wither(struct g_geom *gp, int error)
1550{
1551	struct g_part_entry *entry;
1552	struct g_part_table *table;
1553
1554	table = gp->softc;
1555	if (table != NULL) {
1556		G_PART_DESTROY(table, NULL);
1557		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1558			LIST_REMOVE(entry, gpe_entry);
1559			g_free(entry);
1560		}
1561		if (gp->softc != NULL) {
1562			kobj_delete((kobj_t)gp->softc, M_GEOM);
1563			gp->softc = NULL;
1564		}
1565	}
1566	g_wither_geom(gp, error);
1567}
1568
1569/*
1570 * Class methods.
1571 */
1572
1573static void
1574g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1575{
1576	struct g_part_parms gpp;
1577	struct g_part_table *table;
1578	struct gctl_req_arg *ap;
1579	enum g_part_ctl ctlreq;
1580	unsigned int i, mparms, oparms, parm;
1581	int auto_commit, close_on_error;
1582	int error, modifies;
1583
1584	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1585	g_topology_assert();
1586
1587	ctlreq = G_PART_CTL_NONE;
1588	modifies = 1;
1589	mparms = 0;
1590	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1591	switch (*verb) {
1592	case 'a':
1593		if (!strcmp(verb, "add")) {
1594			ctlreq = G_PART_CTL_ADD;
1595			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1596			    G_PART_PARM_START | G_PART_PARM_TYPE;
1597			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1598		}
1599		break;
1600	case 'b':
1601		if (!strcmp(verb, "bootcode")) {
1602			ctlreq = G_PART_CTL_BOOTCODE;
1603			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1604		}
1605		break;
1606	case 'c':
1607		if (!strcmp(verb, "commit")) {
1608			ctlreq = G_PART_CTL_COMMIT;
1609			mparms |= G_PART_PARM_GEOM;
1610			modifies = 0;
1611		} else if (!strcmp(verb, "create")) {
1612			ctlreq = G_PART_CTL_CREATE;
1613			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1614			oparms |= G_PART_PARM_ENTRIES;
1615		}
1616		break;
1617	case 'd':
1618		if (!strcmp(verb, "delete")) {
1619			ctlreq = G_PART_CTL_DELETE;
1620			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1621		} else if (!strcmp(verb, "destroy")) {
1622			ctlreq = G_PART_CTL_DESTROY;
1623			mparms |= G_PART_PARM_GEOM;
1624			oparms |= G_PART_PARM_FORCE;
1625		}
1626		break;
1627	case 'm':
1628		if (!strcmp(verb, "modify")) {
1629			ctlreq = G_PART_CTL_MODIFY;
1630			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1631			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1632		} else if (!strcmp(verb, "move")) {
1633			ctlreq = G_PART_CTL_MOVE;
1634			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1635		}
1636		break;
1637	case 'r':
1638		if (!strcmp(verb, "recover")) {
1639			ctlreq = G_PART_CTL_RECOVER;
1640			mparms |= G_PART_PARM_GEOM;
1641		} else if (!strcmp(verb, "resize")) {
1642			ctlreq = G_PART_CTL_RESIZE;
1643			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1644			    G_PART_PARM_SIZE;
1645		}
1646		break;
1647	case 's':
1648		if (!strcmp(verb, "set")) {
1649			ctlreq = G_PART_CTL_SET;
1650			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM;
1651			oparms |= G_PART_PARM_INDEX;
1652		}
1653		break;
1654	case 'u':
1655		if (!strcmp(verb, "undo")) {
1656			ctlreq = G_PART_CTL_UNDO;
1657			mparms |= G_PART_PARM_GEOM;
1658			modifies = 0;
1659		} else if (!strcmp(verb, "unset")) {
1660			ctlreq = G_PART_CTL_UNSET;
1661			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM;
1662			oparms |= G_PART_PARM_INDEX;
1663		}
1664		break;
1665	}
1666	if (ctlreq == G_PART_CTL_NONE) {
1667		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1668		return;
1669	}
1670
1671	bzero(&gpp, sizeof(gpp));
1672	for (i = 0; i < req->narg; i++) {
1673		ap = &req->arg[i];
1674		parm = 0;
1675		switch (ap->name[0]) {
1676		case 'a':
1677			if (!strcmp(ap->name, "arg0")) {
1678				parm = mparms &
1679				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1680			}
1681			if (!strcmp(ap->name, "attrib"))
1682				parm = G_PART_PARM_ATTRIB;
1683			break;
1684		case 'b':
1685			if (!strcmp(ap->name, "bootcode"))
1686				parm = G_PART_PARM_BOOTCODE;
1687			break;
1688		case 'c':
1689			if (!strcmp(ap->name, "class"))
1690				continue;
1691			break;
1692		case 'e':
1693			if (!strcmp(ap->name, "entries"))
1694				parm = G_PART_PARM_ENTRIES;
1695			break;
1696		case 'f':
1697			if (!strcmp(ap->name, "flags"))
1698				parm = G_PART_PARM_FLAGS;
1699			else if (!strcmp(ap->name, "force"))
1700				parm = G_PART_PARM_FORCE;
1701			break;
1702		case 'i':
1703			if (!strcmp(ap->name, "index"))
1704				parm = G_PART_PARM_INDEX;
1705			break;
1706		case 'l':
1707			if (!strcmp(ap->name, "label"))
1708				parm = G_PART_PARM_LABEL;
1709			break;
1710		case 'o':
1711			if (!strcmp(ap->name, "output"))
1712				parm = G_PART_PARM_OUTPUT;
1713			break;
1714		case 's':
1715			if (!strcmp(ap->name, "scheme"))
1716				parm = G_PART_PARM_SCHEME;
1717			else if (!strcmp(ap->name, "size"))
1718				parm = G_PART_PARM_SIZE;
1719			else if (!strcmp(ap->name, "start"))
1720				parm = G_PART_PARM_START;
1721			break;
1722		case 't':
1723			if (!strcmp(ap->name, "type"))
1724				parm = G_PART_PARM_TYPE;
1725			break;
1726		case 'v':
1727			if (!strcmp(ap->name, "verb"))
1728				continue;
1729			else if (!strcmp(ap->name, "version"))
1730				parm = G_PART_PARM_VERSION;
1731			break;
1732		}
1733		if ((parm & (mparms | oparms)) == 0) {
1734			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1735			return;
1736		}
1737		switch (parm) {
1738		case G_PART_PARM_ATTRIB:
1739			error = g_part_parm_str(req, ap->name,
1740			    &gpp.gpp_attrib);
1741			break;
1742		case G_PART_PARM_BOOTCODE:
1743			error = g_part_parm_bootcode(req, ap->name,
1744			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1745			break;
1746		case G_PART_PARM_ENTRIES:
1747			error = g_part_parm_intmax(req, ap->name,
1748			    &gpp.gpp_entries);
1749			break;
1750		case G_PART_PARM_FLAGS:
1751			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1752			break;
1753		case G_PART_PARM_FORCE:
1754			error = g_part_parm_uint32(req, ap->name,
1755			    &gpp.gpp_force);
1756			break;
1757		case G_PART_PARM_GEOM:
1758			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1759			break;
1760		case G_PART_PARM_INDEX:
1761			error = g_part_parm_intmax(req, ap->name,
1762			    &gpp.gpp_index);
1763			break;
1764		case G_PART_PARM_LABEL:
1765			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1766			break;
1767		case G_PART_PARM_OUTPUT:
1768			error = 0;	/* Write-only parameter */
1769			break;
1770		case G_PART_PARM_PROVIDER:
1771			error = g_part_parm_provider(req, ap->name,
1772			    &gpp.gpp_provider);
1773			break;
1774		case G_PART_PARM_SCHEME:
1775			error = g_part_parm_scheme(req, ap->name,
1776			    &gpp.gpp_scheme);
1777			break;
1778		case G_PART_PARM_SIZE:
1779			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1780			break;
1781		case G_PART_PARM_START:
1782			error = g_part_parm_quad(req, ap->name,
1783			    &gpp.gpp_start);
1784			break;
1785		case G_PART_PARM_TYPE:
1786			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1787			break;
1788		case G_PART_PARM_VERSION:
1789			error = g_part_parm_uint32(req, ap->name,
1790			    &gpp.gpp_version);
1791			break;
1792		default:
1793			error = EDOOFUS;
1794			gctl_error(req, "%d %s", error, ap->name);
1795			break;
1796		}
1797		if (error != 0) {
1798			if (error == ENOATTR) {
1799				gctl_error(req, "%d param '%s'", error,
1800				    ap->name);
1801			}
1802			return;
1803		}
1804		gpp.gpp_parms |= parm;
1805	}
1806	if ((gpp.gpp_parms & mparms) != mparms) {
1807		parm = mparms - (gpp.gpp_parms & mparms);
1808		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1809		return;
1810	}
1811
1812	/* Obtain permissions if possible/necessary. */
1813	close_on_error = 0;
1814	table = NULL;
1815	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1816		table = gpp.gpp_geom->softc;
1817		if (table != NULL && table->gpt_corrupt &&
1818		    ctlreq != G_PART_CTL_DESTROY &&
1819		    ctlreq != G_PART_CTL_RECOVER &&
1820		    geom_part_check_integrity) {
1821			gctl_error(req, "%d table '%s' is corrupt",
1822			    EPERM, gpp.gpp_geom->name);
1823			return;
1824		}
1825		if (table != NULL && !table->gpt_opened) {
1826			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1827			    1, 1, 1);
1828			if (error) {
1829				gctl_error(req, "%d geom '%s'", error,
1830				    gpp.gpp_geom->name);
1831				return;
1832			}
1833			table->gpt_opened = 1;
1834			close_on_error = 1;
1835		}
1836	}
1837
1838	/* Allow the scheme to check or modify the parameters. */
1839	if (table != NULL) {
1840		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1841		if (error) {
1842			gctl_error(req, "%d pre-check failed", error);
1843			goto out;
1844		}
1845	} else
1846		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1847
1848	switch (ctlreq) {
1849	case G_PART_CTL_NONE:
1850		panic("%s", __func__);
1851	case G_PART_CTL_ADD:
1852		error = g_part_ctl_add(req, &gpp);
1853		break;
1854	case G_PART_CTL_BOOTCODE:
1855		error = g_part_ctl_bootcode(req, &gpp);
1856		break;
1857	case G_PART_CTL_COMMIT:
1858		error = g_part_ctl_commit(req, &gpp);
1859		break;
1860	case G_PART_CTL_CREATE:
1861		error = g_part_ctl_create(req, &gpp);
1862		break;
1863	case G_PART_CTL_DELETE:
1864		error = g_part_ctl_delete(req, &gpp);
1865		break;
1866	case G_PART_CTL_DESTROY:
1867		error = g_part_ctl_destroy(req, &gpp);
1868		break;
1869	case G_PART_CTL_MODIFY:
1870		error = g_part_ctl_modify(req, &gpp);
1871		break;
1872	case G_PART_CTL_MOVE:
1873		error = g_part_ctl_move(req, &gpp);
1874		break;
1875	case G_PART_CTL_RECOVER:
1876		error = g_part_ctl_recover(req, &gpp);
1877		break;
1878	case G_PART_CTL_RESIZE:
1879		error = g_part_ctl_resize(req, &gpp);
1880		break;
1881	case G_PART_CTL_SET:
1882		error = g_part_ctl_setunset(req, &gpp, 1);
1883		break;
1884	case G_PART_CTL_UNDO:
1885		error = g_part_ctl_undo(req, &gpp);
1886		break;
1887	case G_PART_CTL_UNSET:
1888		error = g_part_ctl_setunset(req, &gpp, 0);
1889		break;
1890	}
1891
1892	/* Implement automatic commit. */
1893	if (!error) {
1894		auto_commit = (modifies &&
1895		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1896		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1897		if (auto_commit) {
1898			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1899			    __func__));
1900			error = g_part_ctl_commit(req, &gpp);
1901		}
1902	}
1903
1904 out:
1905	if (error && close_on_error) {
1906		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1907		table->gpt_opened = 0;
1908	}
1909}
1910
1911static int
1912g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1913    struct g_geom *gp)
1914{
1915
1916	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1917	g_topology_assert();
1918
1919	g_part_wither(gp, EINVAL);
1920	return (0);
1921}
1922
1923static struct g_geom *
1924g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1925{
1926	struct g_consumer *cp;
1927	struct g_geom *gp;
1928	struct g_part_entry *entry;
1929	struct g_part_table *table;
1930	struct root_hold_token *rht;
1931	int attr, depth;
1932	int error;
1933
1934	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1935	g_topology_assert();
1936
1937	/* Skip providers that are already open for writing. */
1938	if (pp->acw > 0)
1939		return (NULL);
1940
1941	/*
1942	 * Create a GEOM with consumer and hook it up to the provider.
1943	 * With that we become part of the topology. Optain read access
1944	 * to the provider.
1945	 */
1946	gp = g_new_geomf(mp, "%s", pp->name);
1947	cp = g_new_consumer(gp);
1948	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
1949	error = g_attach(cp, pp);
1950	if (error == 0)
1951		error = g_access(cp, 1, 0, 0);
1952	if (error != 0) {
1953		if (cp->provider)
1954			g_detach(cp);
1955		g_destroy_consumer(cp);
1956		g_destroy_geom(gp);
1957		return (NULL);
1958	}
1959
1960	rht = root_mount_hold(mp->name);
1961	g_topology_unlock();
1962
1963	/*
1964	 * Short-circuit the whole probing galore when there's no
1965	 * media present.
1966	 */
1967	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1968		error = ENODEV;
1969		goto fail;
1970	}
1971
1972	/* Make sure we can nest and if so, determine our depth. */
1973	error = g_getattr("PART::isleaf", cp, &attr);
1974	if (!error && attr) {
1975		error = ENODEV;
1976		goto fail;
1977	}
1978	error = g_getattr("PART::depth", cp, &attr);
1979	depth = (!error) ? attr + 1 : 0;
1980
1981	error = g_part_probe(gp, cp, depth);
1982	if (error)
1983		goto fail;
1984
1985	table = gp->softc;
1986
1987	/*
1988	 * Synthesize a disk geometry. Some partitioning schemes
1989	 * depend on it and since some file systems need it even
1990	 * when the partitition scheme doesn't, we do it here in
1991	 * scheme-independent code.
1992	 */
1993	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1994
1995	error = G_PART_READ(table, cp);
1996	if (error)
1997		goto fail;
1998	error = g_part_check_integrity(table, cp);
1999	if (error)
2000		goto fail;
2001
2002	g_topology_lock();
2003	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
2004		if (!entry->gpe_internal)
2005			g_part_new_provider(gp, table, entry);
2006	}
2007
2008	root_mount_rel(rht);
2009	g_access(cp, -1, 0, 0);
2010	return (gp);
2011
2012 fail:
2013	g_topology_lock();
2014	root_mount_rel(rht);
2015	g_access(cp, -1, 0, 0);
2016	g_detach(cp);
2017	g_destroy_consumer(cp);
2018	g_destroy_geom(gp);
2019	return (NULL);
2020}
2021
2022/*
2023 * Geom methods.
2024 */
2025
2026static int
2027g_part_access(struct g_provider *pp, int dr, int dw, int de)
2028{
2029	struct g_consumer *cp;
2030
2031	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
2032	    dw, de));
2033
2034	cp = LIST_FIRST(&pp->geom->consumer);
2035
2036	/* We always gain write-exclusive access. */
2037	return (g_access(cp, dr, dw, dw + de));
2038}
2039
2040static void
2041g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2042    struct g_consumer *cp, struct g_provider *pp)
2043{
2044	char buf[64];
2045	struct g_part_entry *entry;
2046	struct g_part_table *table;
2047
2048	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
2049	table = gp->softc;
2050
2051	if (indent == NULL) {
2052		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
2053		entry = pp->private;
2054		if (entry == NULL)
2055			return;
2056		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
2057		    (uintmax_t)entry->gpe_offset,
2058		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2059		/*
2060		 * libdisk compatibility quirk - the scheme dumps the
2061		 * slicer name and partition type in a way that is
2062		 * compatible with libdisk. When libdisk is not used
2063		 * anymore, this should go away.
2064		 */
2065		G_PART_DUMPCONF(table, entry, sb, indent);
2066	} else if (cp != NULL) {	/* Consumer configuration. */
2067		KASSERT(pp == NULL, ("%s", __func__));
2068		/* none */
2069	} else if (pp != NULL) {	/* Provider configuration. */
2070		entry = pp->private;
2071		if (entry == NULL)
2072			return;
2073		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
2074		    (uintmax_t)entry->gpe_start);
2075		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
2076		    (uintmax_t)entry->gpe_end);
2077		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
2078		    entry->gpe_index);
2079		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2080		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2081		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
2082		    (uintmax_t)entry->gpe_offset);
2083		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
2084		    (uintmax_t)pp->mediasize);
2085		G_PART_DUMPCONF(table, entry, sb, indent);
2086	} else {			/* Geom configuration. */
2087		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2088		    table->gpt_scheme->name);
2089		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2090		    table->gpt_entries);
2091		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2092		    (uintmax_t)table->gpt_first);
2093		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2094		    (uintmax_t)table->gpt_last);
2095		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2096		    table->gpt_sectors);
2097		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2098		    table->gpt_heads);
2099		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2100		    table->gpt_corrupt ? "CORRUPT": "OK");
2101		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2102		    table->gpt_opened ? "true": "false");
2103		G_PART_DUMPCONF(table, NULL, sb, indent);
2104	}
2105}
2106
2107/*-
2108 * This start routine is only called for non-trivial requests, all the
2109 * trivial ones are handled autonomously by the slice code.
2110 * For requests we handle here, we must call the g_io_deliver() on the
2111 * bio, and return non-zero to indicate to the slice code that we did so.
2112 * This code executes in the "DOWN" I/O path, this means:
2113 *    * No sleeping.
2114 *    * Don't grab the topology lock.
2115 *    * Don't call biowait, g_getattr(), g_setattr() or g_read_data()
2116 */
2117static int
2118g_part_ioctl(struct g_provider *pp, u_long cmd, void *data, int fflag, struct thread *td)
2119{
2120	struct g_part_table *table;
2121
2122	table = pp->geom->softc;
2123	return G_PART_IOCTL(table, pp, cmd, data, fflag, td);
2124}
2125
2126static void
2127g_part_resize(struct g_consumer *cp)
2128{
2129	struct g_part_table *table;
2130
2131	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2132	g_topology_assert();
2133
2134	table = cp->geom->softc;
2135	if (table->gpt_opened == 0) {
2136		if (g_access(cp, 1, 1, 1) != 0)
2137			return;
2138		table->gpt_opened = 1;
2139	}
2140	if (G_PART_RESIZE(table, NULL, NULL) == 0)
2141		printf("GEOM_PART: %s was automatically resized.\n"
2142		    "  Use `gpart commit %s` to save changes or "
2143		    "`gpart undo %s` to revert them.\n", cp->geom->name,
2144		    cp->geom->name, cp->geom->name);
2145	if (g_part_check_integrity(table, cp) != 0) {
2146		g_access(cp, -1, -1, -1);
2147		table->gpt_opened = 0;
2148		g_part_wither(table->gpt_gp, ENXIO);
2149	}
2150}
2151
2152static void
2153g_part_orphan(struct g_consumer *cp)
2154{
2155	struct g_provider *pp;
2156	struct g_part_table *table;
2157
2158	pp = cp->provider;
2159	KASSERT(pp != NULL, ("%s", __func__));
2160	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2161	g_topology_assert();
2162
2163	KASSERT(pp->error != 0, ("%s", __func__));
2164	table = cp->geom->softc;
2165	if (table != NULL && table->gpt_opened)
2166		g_access(cp, -1, -1, -1);
2167	g_part_wither(cp->geom, pp->error);
2168}
2169
2170static void
2171g_part_spoiled(struct g_consumer *cp)
2172{
2173
2174	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2175	g_topology_assert();
2176
2177	cp->flags |= G_CF_ORPHAN;
2178	g_part_wither(cp->geom, ENXIO);
2179}
2180
2181static void
2182g_part_start(struct bio *bp)
2183{
2184	struct bio *bp2;
2185	struct g_consumer *cp;
2186	struct g_geom *gp;
2187	struct g_part_entry *entry;
2188	struct g_part_table *table;
2189	struct g_kerneldump *gkd;
2190	struct g_provider *pp;
2191	void (*done_func)(struct bio *) = g_std_done;
2192	char buf[64];
2193
2194	pp = bp->bio_to;
2195	gp = pp->geom;
2196	table = gp->softc;
2197	cp = LIST_FIRST(&gp->consumer);
2198
2199	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2200	    pp->name));
2201
2202	entry = pp->private;
2203	if (entry == NULL) {
2204		g_io_deliver(bp, ENXIO);
2205		return;
2206	}
2207
2208	switch(bp->bio_cmd) {
2209	case BIO_DELETE:
2210	case BIO_READ:
2211	case BIO_WRITE:
2212		if (bp->bio_offset >= pp->mediasize) {
2213			g_io_deliver(bp, EIO);
2214			return;
2215		}
2216		bp2 = g_clone_bio(bp);
2217		if (bp2 == NULL) {
2218			g_io_deliver(bp, ENOMEM);
2219			return;
2220		}
2221		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2222			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2223		bp2->bio_done = g_std_done;
2224		bp2->bio_offset += entry->gpe_offset;
2225		g_io_request(bp2, cp);
2226		return;
2227	case BIO_FLUSH:
2228		break;
2229	case BIO_GETATTR:
2230		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2231			return;
2232		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2233			return;
2234		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2235			return;
2236		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2237			return;
2238		if (g_handleattr_str(bp, "PART::scheme",
2239		    table->gpt_scheme->name))
2240			return;
2241		if (g_handleattr_str(bp, "PART::type",
2242		    G_PART_TYPE(table, entry, buf, sizeof(buf))))
2243			return;
2244		if (!strcmp("GEOM::physpath", bp->bio_attribute)) {
2245			done_func = g_part_get_physpath_done;
2246			break;
2247		}
2248		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2249			/*
2250			 * Check that the partition is suitable for kernel
2251			 * dumps. Typically only swap partitions should be
2252			 * used. If the request comes from the nested scheme
2253			 * we allow dumping there as well.
2254			 */
2255			if ((bp->bio_from == NULL ||
2256			    bp->bio_from->geom->class != &g_part_class) &&
2257			    G_PART_DUMPTO(table, entry) == 0) {
2258				g_io_deliver(bp, ENODEV);
2259				printf("GEOM_PART: Partition '%s' not suitable"
2260				    " for kernel dumps (wrong type?)\n",
2261				    pp->name);
2262				return;
2263			}
2264			gkd = (struct g_kerneldump *)bp->bio_data;
2265			if (gkd->offset >= pp->mediasize) {
2266				g_io_deliver(bp, EIO);
2267				return;
2268			}
2269			if (gkd->offset + gkd->length > pp->mediasize)
2270				gkd->length = pp->mediasize - gkd->offset;
2271			gkd->offset += entry->gpe_offset;
2272		}
2273		break;
2274	default:
2275		g_io_deliver(bp, EOPNOTSUPP);
2276		return;
2277	}
2278
2279	bp2 = g_clone_bio(bp);
2280	if (bp2 == NULL) {
2281		g_io_deliver(bp, ENOMEM);
2282		return;
2283	}
2284	bp2->bio_done = done_func;
2285	g_io_request(bp2, cp);
2286}
2287
2288static void
2289g_part_init(struct g_class *mp)
2290{
2291
2292	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2293}
2294
2295static void
2296g_part_fini(struct g_class *mp)
2297{
2298
2299	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2300}
2301
2302static void
2303g_part_unload_event(void *arg, int flag)
2304{
2305	struct g_consumer *cp;
2306	struct g_geom *gp;
2307	struct g_provider *pp;
2308	struct g_part_scheme *scheme;
2309	struct g_part_table *table;
2310	uintptr_t *xchg;
2311	int acc, error;
2312
2313	if (flag == EV_CANCEL)
2314		return;
2315
2316	xchg = arg;
2317	error = 0;
2318	scheme = (void *)(*xchg);
2319
2320	g_topology_assert();
2321
2322	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2323		table = gp->softc;
2324		if (table->gpt_scheme != scheme)
2325			continue;
2326
2327		acc = 0;
2328		LIST_FOREACH(pp, &gp->provider, provider)
2329			acc += pp->acr + pp->acw + pp->ace;
2330		LIST_FOREACH(cp, &gp->consumer, consumer)
2331			acc += cp->acr + cp->acw + cp->ace;
2332
2333		if (!acc)
2334			g_part_wither(gp, ENOSYS);
2335		else
2336			error = EBUSY;
2337	}
2338
2339	if (!error)
2340		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2341
2342	*xchg = error;
2343}
2344
2345int
2346g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2347{
2348	struct g_part_scheme *iter;
2349	uintptr_t arg;
2350	int error;
2351
2352	error = 0;
2353	switch (type) {
2354	case MOD_LOAD:
2355		TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
2356			if (scheme == iter) {
2357				printf("GEOM_PART: scheme %s is already "
2358				    "registered!\n", scheme->name);
2359				break;
2360			}
2361		}
2362		if (iter == NULL) {
2363			TAILQ_INSERT_TAIL(&g_part_schemes, scheme,
2364			    scheme_list);
2365			g_retaste(&g_part_class);
2366		}
2367		break;
2368	case MOD_UNLOAD:
2369		arg = (uintptr_t)scheme;
2370		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2371		    NULL);
2372		if (error == 0)
2373			error = arg;
2374		break;
2375	default:
2376		error = EOPNOTSUPP;
2377		break;
2378	}
2379
2380	return (error);
2381}
2382