xref: /freebsd/sys/geom/part/g_part.c (revision a812392203d7c4c3f0db9d8a0f3391374c49c71f)
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("$FreeBSD$");
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 
55 static kobj_method_t g_part_null_methods[] = {
56 	{ 0, 0 }
57 };
58 
59 static struct g_part_scheme g_part_null_scheme = {
60 	"(none)",
61 	g_part_null_methods,
62 	sizeof(struct g_part_table),
63 };
64 
65 TAILQ_HEAD(, g_part_scheme) g_part_schemes =
66     TAILQ_HEAD_INITIALIZER(g_part_schemes);
67 
68 struct 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-hfs", G_PART_ALIAS_APPLE_HFS },
74 	{ "apple-label", G_PART_ALIAS_APPLE_LABEL },
75 	{ "apple-raid", G_PART_ALIAS_APPLE_RAID },
76 	{ "apple-raid-offline", G_PART_ALIAS_APPLE_RAID_OFFLINE },
77 	{ "apple-tv-recovery", G_PART_ALIAS_APPLE_TV_RECOVERY },
78 	{ "apple-ufs", G_PART_ALIAS_APPLE_UFS },
79 	{ "bios-boot", G_PART_ALIAS_BIOS_BOOT },
80 	{ "ebr", G_PART_ALIAS_EBR },
81 	{ "efi", G_PART_ALIAS_EFI },
82 	{ "fat16", G_PART_ALIAS_MS_FAT16 },
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 	{ "vmware-vsanhdr", G_PART_ALIAS_VMVSANHDR },
111 	{ "dragonfly-label32", G_PART_ALIAS_DFBSD },
112 	{ "dragonfly-label64", G_PART_ALIAS_DFBSD64 },
113 	{ "dragonfly-swap", G_PART_ALIAS_DFBSD_SWAP },
114 	{ "dragonfly-ufs", G_PART_ALIAS_DFBSD_UFS },
115 	{ "dragonfly-vinum", G_PART_ALIAS_DFBSD_VINUM },
116 	{ "dragonfly-ccd", G_PART_ALIAS_DFBSD_CCD },
117 	{ "dragonfly-legacy", G_PART_ALIAS_DFBSD_LEGACY },
118 	{ "dragonfly-hammer", G_PART_ALIAS_DFBSD_HAMMER },
119 	{ "dragonfly-hammer2", G_PART_ALIAS_DFBSD_HAMMER2 },
120 	{ "prep-boot", G_PART_ALIAS_PREP_BOOT },
121 };
122 
123 SYSCTL_DECL(_kern_geom);
124 SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
125     "GEOM_PART stuff");
126 static u_int check_integrity = 1;
127 SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity,
128     CTLFLAG_RWTUN, &check_integrity, 1,
129     "Enable integrity checking");
130 
131 /*
132  * The GEOM partitioning class.
133  */
134 static g_ctl_req_t g_part_ctlreq;
135 static g_ctl_destroy_geom_t g_part_destroy_geom;
136 static g_fini_t g_part_fini;
137 static g_init_t g_part_init;
138 static g_taste_t g_part_taste;
139 
140 static g_access_t g_part_access;
141 static g_dumpconf_t g_part_dumpconf;
142 static g_orphan_t g_part_orphan;
143 static g_spoiled_t g_part_spoiled;
144 static g_start_t g_part_start;
145 static g_resize_t g_part_resize;
146 static g_ioctl_t g_part_ioctl;
147 
148 static struct g_class g_part_class = {
149 	.name = "PART",
150 	.version = G_VERSION,
151 	/* Class methods. */
152 	.ctlreq = g_part_ctlreq,
153 	.destroy_geom = g_part_destroy_geom,
154 	.fini = g_part_fini,
155 	.init = g_part_init,
156 	.taste = g_part_taste,
157 	/* Geom methods. */
158 	.access = g_part_access,
159 	.dumpconf = g_part_dumpconf,
160 	.orphan = g_part_orphan,
161 	.spoiled = g_part_spoiled,
162 	.start = g_part_start,
163 	.resize = g_part_resize,
164 	.ioctl = g_part_ioctl,
165 };
166 
167 DECLARE_GEOM_CLASS(g_part_class, g_part);
168 MODULE_VERSION(g_part, 0);
169 
170 /*
171  * Support functions.
172  */
173 
174 static void g_part_wither(struct g_geom *, int);
175 
176 const char *
177 g_part_alias_name(enum g_part_alias alias)
178 {
179 	int i;
180 
181 	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
182 		if (g_part_alias_list[i].alias != alias)
183 			continue;
184 		return (g_part_alias_list[i].lexeme);
185 	}
186 
187 	return (NULL);
188 }
189 
190 void
191 g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
192     u_int *bestheads)
193 {
194 	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
195 	off_t chs, cylinders;
196 	u_int heads;
197 	int idx;
198 
199 	*bestchs = 0;
200 	*bestheads = 0;
201 	for (idx = 0; candidate_heads[idx] != 0; idx++) {
202 		heads = candidate_heads[idx];
203 		cylinders = blocks / heads / sectors;
204 		if (cylinders < heads || cylinders < sectors)
205 			break;
206 		if (cylinders > 1023)
207 			continue;
208 		chs = cylinders * heads * sectors;
209 		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
210 			*bestchs = chs;
211 			*bestheads = heads;
212 		}
213 	}
214 }
215 
216 static void
217 g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
218     off_t blocks)
219 {
220 	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
221 	off_t chs, bestchs;
222 	u_int heads, sectors;
223 	int idx;
224 
225 	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
226 	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
227 		table->gpt_fixgeom = 0;
228 		table->gpt_heads = 0;
229 		table->gpt_sectors = 0;
230 		bestchs = 0;
231 		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
232 			sectors = candidate_sectors[idx];
233 			g_part_geometry_heads(blocks, sectors, &chs, &heads);
234 			if (chs == 0)
235 				continue;
236 			/*
237 			 * Prefer a geometry with sectors > 1, but only if
238 			 * it doesn't bump down the number of heads to 1.
239 			 */
240 			if (chs > bestchs || (chs == bestchs && heads > 1 &&
241 			    table->gpt_sectors == 1)) {
242 				bestchs = chs;
243 				table->gpt_heads = heads;
244 				table->gpt_sectors = sectors;
245 			}
246 		}
247 		/*
248 		 * If we didn't find a geometry at all, then the disk is
249 		 * too big. This means we can use the maximum number of
250 		 * heads and sectors.
251 		 */
252 		if (bestchs == 0) {
253 			table->gpt_heads = 255;
254 			table->gpt_sectors = 63;
255 		}
256 	} else {
257 		table->gpt_fixgeom = 1;
258 		table->gpt_heads = heads;
259 		table->gpt_sectors = sectors;
260 	}
261 }
262 
263 #define	DPRINTF(...)	if (bootverbose) {	\
264 	printf("GEOM_PART: " __VA_ARGS__);	\
265 }
266 
267 static int
268 g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
269 {
270 	struct g_part_entry *e1, *e2;
271 	struct g_provider *pp;
272 	off_t offset;
273 	int failed;
274 
275 	failed = 0;
276 	pp = cp->provider;
277 	if (table->gpt_last < table->gpt_first) {
278 		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
279 		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
280 		failed++;
281 	}
282 	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
283 		DPRINTF("last LBA extends beyond mediasize: "
284 		    "%jd > %jd\n", (intmax_t)table->gpt_last,
285 		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
286 		failed++;
287 	}
288 	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
289 		if (e1->gpe_deleted || e1->gpe_internal)
290 			continue;
291 		if (e1->gpe_start < table->gpt_first) {
292 			DPRINTF("partition %d has start offset below first "
293 			    "LBA: %jd < %jd\n", e1->gpe_index,
294 			    (intmax_t)e1->gpe_start,
295 			    (intmax_t)table->gpt_first);
296 			failed++;
297 		}
298 		if (e1->gpe_start > table->gpt_last) {
299 			DPRINTF("partition %d has start offset beyond last "
300 			    "LBA: %jd > %jd\n", e1->gpe_index,
301 			    (intmax_t)e1->gpe_start,
302 			    (intmax_t)table->gpt_last);
303 			failed++;
304 		}
305 		if (e1->gpe_end < e1->gpe_start) {
306 			DPRINTF("partition %d has end offset below start "
307 			    "offset: %jd < %jd\n", e1->gpe_index,
308 			    (intmax_t)e1->gpe_end,
309 			    (intmax_t)e1->gpe_start);
310 			failed++;
311 		}
312 		if (e1->gpe_end > table->gpt_last) {
313 			DPRINTF("partition %d has end offset beyond last "
314 			    "LBA: %jd > %jd\n", e1->gpe_index,
315 			    (intmax_t)e1->gpe_end,
316 			    (intmax_t)table->gpt_last);
317 			failed++;
318 		}
319 		if (pp->stripesize > 0) {
320 			offset = e1->gpe_start * pp->sectorsize;
321 			if (e1->gpe_offset > offset)
322 				offset = e1->gpe_offset;
323 			if ((offset + pp->stripeoffset) % pp->stripesize) {
324 				DPRINTF("partition %d is not aligned on %u "
325 				    "bytes\n", e1->gpe_index, pp->stripesize);
326 				/* Don't treat this as a critical failure */
327 			}
328 		}
329 		e2 = e1;
330 		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
331 			if (e2->gpe_deleted || e2->gpe_internal)
332 				continue;
333 			if (e1->gpe_start >= e2->gpe_start &&
334 			    e1->gpe_start <= e2->gpe_end) {
335 				DPRINTF("partition %d has start offset inside "
336 				    "partition %d: start[%d] %jd >= start[%d] "
337 				    "%jd <= end[%d] %jd\n",
338 				    e1->gpe_index, e2->gpe_index,
339 				    e2->gpe_index, (intmax_t)e2->gpe_start,
340 				    e1->gpe_index, (intmax_t)e1->gpe_start,
341 				    e2->gpe_index, (intmax_t)e2->gpe_end);
342 				failed++;
343 			}
344 			if (e1->gpe_end >= e2->gpe_start &&
345 			    e1->gpe_end <= e2->gpe_end) {
346 				DPRINTF("partition %d has end offset inside "
347 				    "partition %d: start[%d] %jd >= end[%d] "
348 				    "%jd <= end[%d] %jd\n",
349 				    e1->gpe_index, e2->gpe_index,
350 				    e2->gpe_index, (intmax_t)e2->gpe_start,
351 				    e1->gpe_index, (intmax_t)e1->gpe_end,
352 				    e2->gpe_index, (intmax_t)e2->gpe_end);
353 				failed++;
354 			}
355 			if (e1->gpe_start < e2->gpe_start &&
356 			    e1->gpe_end > e2->gpe_end) {
357 				DPRINTF("partition %d contains partition %d: "
358 				    "start[%d] %jd > start[%d] %jd, end[%d] "
359 				    "%jd < end[%d] %jd\n",
360 				    e1->gpe_index, e2->gpe_index,
361 				    e1->gpe_index, (intmax_t)e1->gpe_start,
362 				    e2->gpe_index, (intmax_t)e2->gpe_start,
363 				    e2->gpe_index, (intmax_t)e2->gpe_end,
364 				    e1->gpe_index, (intmax_t)e1->gpe_end);
365 				failed++;
366 			}
367 		}
368 	}
369 	if (failed != 0) {
370 		printf("GEOM_PART: integrity check failed (%s, %s)\n",
371 		    pp->name, table->gpt_scheme->name);
372 		if (check_integrity != 0)
373 			return (EINVAL);
374 		table->gpt_corrupt = 1;
375 	}
376 	return (0);
377 }
378 #undef	DPRINTF
379 
380 struct g_part_entry *
381 g_part_new_entry(struct g_part_table *table, int index, quad_t start,
382     quad_t end)
383 {
384 	struct g_part_entry *entry, *last;
385 
386 	last = NULL;
387 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
388 		if (entry->gpe_index == index)
389 			break;
390 		if (entry->gpe_index > index) {
391 			entry = NULL;
392 			break;
393 		}
394 		last = entry;
395 	}
396 	if (entry == NULL) {
397 		entry = g_malloc(table->gpt_scheme->gps_entrysz,
398 		    M_WAITOK | M_ZERO);
399 		entry->gpe_index = index;
400 		if (last == NULL)
401 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
402 		else
403 			LIST_INSERT_AFTER(last, entry, gpe_entry);
404 	} else
405 		entry->gpe_offset = 0;
406 	entry->gpe_start = start;
407 	entry->gpe_end = end;
408 	return (entry);
409 }
410 
411 static void
412 g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
413     struct g_part_entry *entry)
414 {
415 	struct g_consumer *cp;
416 	struct g_provider *pp;
417 	struct sbuf *sb;
418 	off_t offset;
419 
420 	cp = LIST_FIRST(&gp->consumer);
421 	pp = cp->provider;
422 
423 	offset = entry->gpe_start * pp->sectorsize;
424 	if (entry->gpe_offset < offset)
425 		entry->gpe_offset = offset;
426 
427 	if (entry->gpe_pp == NULL) {
428 		sb = sbuf_new_auto();
429 		G_PART_FULLNAME(table, entry, sb, gp->name);
430 		sbuf_finish(sb);
431 		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
432 		sbuf_delete(sb);
433 		entry->gpe_pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE;
434 		entry->gpe_pp->private = entry;		/* Close the circle. */
435 	}
436 	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
437 	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
438 	    pp->sectorsize;
439 	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
440 	entry->gpe_pp->sectorsize = pp->sectorsize;
441 	entry->gpe_pp->stripesize = pp->stripesize;
442 	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
443 	if (pp->stripesize > 0)
444 		entry->gpe_pp->stripeoffset %= pp->stripesize;
445 	entry->gpe_pp->flags |= pp->flags & G_PF_ACCEPT_UNMAPPED;
446 	g_error_provider(entry->gpe_pp, 0);
447 }
448 
449 static struct g_geom*
450 g_part_find_geom(const char *name)
451 {
452 	struct g_geom *gp;
453 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
454 		if (!strcmp(name, gp->name))
455 			break;
456 	}
457 	return (gp);
458 }
459 
460 static int
461 g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
462 {
463 	struct g_geom *gp;
464 	const char *gname;
465 
466 	gname = gctl_get_asciiparam(req, name);
467 	if (gname == NULL)
468 		return (ENOATTR);
469 	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
470 		gname += sizeof(_PATH_DEV) - 1;
471 	gp = g_part_find_geom(gname);
472 	if (gp == NULL) {
473 		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
474 		return (EINVAL);
475 	}
476 	if ((gp->flags & G_GEOM_WITHER) != 0) {
477 		gctl_error(req, "%d %s", ENXIO, gname);
478 		return (ENXIO);
479 	}
480 	*v = gp;
481 	return (0);
482 }
483 
484 static int
485 g_part_parm_provider(struct gctl_req *req, const char *name,
486     struct g_provider **v)
487 {
488 	struct g_provider *pp;
489 	const char *pname;
490 
491 	pname = gctl_get_asciiparam(req, name);
492 	if (pname == NULL)
493 		return (ENOATTR);
494 	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
495 		pname += sizeof(_PATH_DEV) - 1;
496 	pp = g_provider_by_name(pname);
497 	if (pp == NULL) {
498 		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
499 		return (EINVAL);
500 	}
501 	*v = pp;
502 	return (0);
503 }
504 
505 static int
506 g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
507 {
508 	const char *p;
509 	char *x;
510 	quad_t q;
511 
512 	p = gctl_get_asciiparam(req, name);
513 	if (p == NULL)
514 		return (ENOATTR);
515 	q = strtoq(p, &x, 0);
516 	if (*x != '\0' || q < 0) {
517 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
518 		return (EINVAL);
519 	}
520 	*v = q;
521 	return (0);
522 }
523 
524 static int
525 g_part_parm_scheme(struct gctl_req *req, const char *name,
526     struct g_part_scheme **v)
527 {
528 	struct g_part_scheme *s;
529 	const char *p;
530 
531 	p = gctl_get_asciiparam(req, name);
532 	if (p == NULL)
533 		return (ENOATTR);
534 	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
535 		if (s == &g_part_null_scheme)
536 			continue;
537 		if (!strcasecmp(s->name, p))
538 			break;
539 	}
540 	if (s == NULL) {
541 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
542 		return (EINVAL);
543 	}
544 	*v = s;
545 	return (0);
546 }
547 
548 static int
549 g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
550 {
551 	const char *p;
552 
553 	p = gctl_get_asciiparam(req, name);
554 	if (p == NULL)
555 		return (ENOATTR);
556 	/* An empty label is always valid. */
557 	if (strcmp(name, "label") != 0 && p[0] == '\0') {
558 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
559 		return (EINVAL);
560 	}
561 	*v = p;
562 	return (0);
563 }
564 
565 static int
566 g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
567 {
568 	const intmax_t *p;
569 	int size;
570 
571 	p = gctl_get_param(req, name, &size);
572 	if (p == NULL)
573 		return (ENOATTR);
574 	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
575 		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
576 		return (EINVAL);
577 	}
578 	*v = (u_int)*p;
579 	return (0);
580 }
581 
582 static int
583 g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
584 {
585 	const uint32_t *p;
586 	int size;
587 
588 	p = gctl_get_param(req, name, &size);
589 	if (p == NULL)
590 		return (ENOATTR);
591 	if (size != sizeof(*p) || *p > INT_MAX) {
592 		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
593 		return (EINVAL);
594 	}
595 	*v = (u_int)*p;
596 	return (0);
597 }
598 
599 static int
600 g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
601     unsigned int *s)
602 {
603 	const void *p;
604 	int size;
605 
606 	p = gctl_get_param(req, name, &size);
607 	if (p == NULL)
608 		return (ENOATTR);
609 	*v = p;
610 	*s = size;
611 	return (0);
612 }
613 
614 static int
615 g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
616 {
617 	struct g_part_scheme *iter, *scheme;
618 	struct g_part_table *table;
619 	int pri, probe;
620 
621 	table = gp->softc;
622 	scheme = (table != NULL) ? table->gpt_scheme : NULL;
623 	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
624 	if (pri == 0)
625 		goto done;
626 	if (pri > 0) {	/* error */
627 		scheme = NULL;
628 		pri = INT_MIN;
629 	}
630 
631 	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
632 		if (iter == &g_part_null_scheme)
633 			continue;
634 		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
635 		    M_WAITOK);
636 		table->gpt_gp = gp;
637 		table->gpt_scheme = iter;
638 		table->gpt_depth = depth;
639 		probe = G_PART_PROBE(table, cp);
640 		if (probe <= 0 && probe > pri) {
641 			pri = probe;
642 			scheme = iter;
643 			if (gp->softc != NULL)
644 				kobj_delete((kobj_t)gp->softc, M_GEOM);
645 			gp->softc = table;
646 			if (pri == 0)
647 				goto done;
648 		} else
649 			kobj_delete((kobj_t)table, M_GEOM);
650 	}
651 
652 done:
653 	return ((scheme == NULL) ? ENXIO : 0);
654 }
655 
656 /*
657  * Control request functions.
658  */
659 
660 static int
661 g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
662 {
663 	struct g_geom *gp;
664 	struct g_provider *pp;
665 	struct g_part_entry *delent, *last, *entry;
666 	struct g_part_table *table;
667 	struct sbuf *sb;
668 	quad_t end;
669 	unsigned int index;
670 	int error;
671 
672 	gp = gpp->gpp_geom;
673 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
674 	g_topology_assert();
675 
676 	pp = LIST_FIRST(&gp->consumer)->provider;
677 	table = gp->softc;
678 	end = gpp->gpp_start + gpp->gpp_size - 1;
679 
680 	if (gpp->gpp_start < table->gpt_first ||
681 	    gpp->gpp_start > table->gpt_last) {
682 		gctl_error(req, "%d start '%jd'", EINVAL,
683 		    (intmax_t)gpp->gpp_start);
684 		return (EINVAL);
685 	}
686 	if (end < gpp->gpp_start || end > table->gpt_last) {
687 		gctl_error(req, "%d size '%jd'", EINVAL,
688 		    (intmax_t)gpp->gpp_size);
689 		return (EINVAL);
690 	}
691 	if (gpp->gpp_index > table->gpt_entries) {
692 		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
693 		return (EINVAL);
694 	}
695 
696 	delent = last = NULL;
697 	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
698 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
699 		if (entry->gpe_deleted) {
700 			if (entry->gpe_index == index)
701 				delent = entry;
702 			continue;
703 		}
704 		if (entry->gpe_index == index)
705 			index = entry->gpe_index + 1;
706 		if (entry->gpe_index < index)
707 			last = entry;
708 		if (entry->gpe_internal)
709 			continue;
710 		if (gpp->gpp_start >= entry->gpe_start &&
711 		    gpp->gpp_start <= entry->gpe_end) {
712 			gctl_error(req, "%d start '%jd'", ENOSPC,
713 			    (intmax_t)gpp->gpp_start);
714 			return (ENOSPC);
715 		}
716 		if (end >= entry->gpe_start && end <= entry->gpe_end) {
717 			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
718 			return (ENOSPC);
719 		}
720 		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
721 			gctl_error(req, "%d size '%jd'", ENOSPC,
722 			    (intmax_t)gpp->gpp_size);
723 			return (ENOSPC);
724 		}
725 	}
726 	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
727 		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
728 		return (EEXIST);
729 	}
730 	if (index > table->gpt_entries) {
731 		gctl_error(req, "%d index '%d'", ENOSPC, index);
732 		return (ENOSPC);
733 	}
734 
735 	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
736 	    M_WAITOK | M_ZERO) : delent;
737 	entry->gpe_index = index;
738 	entry->gpe_start = gpp->gpp_start;
739 	entry->gpe_end = end;
740 	error = G_PART_ADD(table, entry, gpp);
741 	if (error) {
742 		gctl_error(req, "%d", error);
743 		if (delent == NULL)
744 			g_free(entry);
745 		return (error);
746 	}
747 	if (delent == NULL) {
748 		if (last == NULL)
749 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
750 		else
751 			LIST_INSERT_AFTER(last, entry, gpe_entry);
752 		entry->gpe_created = 1;
753 	} else {
754 		entry->gpe_deleted = 0;
755 		entry->gpe_modified = 1;
756 	}
757 	g_part_new_provider(gp, table, entry);
758 
759 	/* Provide feedback if so requested. */
760 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
761 		sb = sbuf_new_auto();
762 		G_PART_FULLNAME(table, entry, sb, gp->name);
763 		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
764 			sbuf_printf(sb, " added, but partition is not "
765 			    "aligned on %u bytes\n", pp->stripesize);
766 		else
767 			sbuf_cat(sb, " added\n");
768 		sbuf_finish(sb);
769 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
770 		sbuf_delete(sb);
771 	}
772 	return (0);
773 }
774 
775 static int
776 g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
777 {
778 	struct g_geom *gp;
779 	struct g_part_table *table;
780 	struct sbuf *sb;
781 	int error, sz;
782 
783 	gp = gpp->gpp_geom;
784 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
785 	g_topology_assert();
786 
787 	table = gp->softc;
788 	sz = table->gpt_scheme->gps_bootcodesz;
789 	if (sz == 0) {
790 		error = ENODEV;
791 		goto fail;
792 	}
793 	if (gpp->gpp_codesize > sz) {
794 		error = EFBIG;
795 		goto fail;
796 	}
797 
798 	error = G_PART_BOOTCODE(table, gpp);
799 	if (error)
800 		goto fail;
801 
802 	/* Provide feedback if so requested. */
803 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
804 		sb = sbuf_new_auto();
805 		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
806 		sbuf_finish(sb);
807 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
808 		sbuf_delete(sb);
809 	}
810 	return (0);
811 
812  fail:
813 	gctl_error(req, "%d", error);
814 	return (error);
815 }
816 
817 static int
818 g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
819 {
820 	struct g_consumer *cp;
821 	struct g_geom *gp;
822 	struct g_provider *pp;
823 	struct g_part_entry *entry, *tmp;
824 	struct g_part_table *table;
825 	char *buf;
826 	int error, i;
827 
828 	gp = gpp->gpp_geom;
829 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
830 	g_topology_assert();
831 
832 	table = gp->softc;
833 	if (!table->gpt_opened) {
834 		gctl_error(req, "%d", EPERM);
835 		return (EPERM);
836 	}
837 
838 	g_topology_unlock();
839 
840 	cp = LIST_FIRST(&gp->consumer);
841 	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
842 		pp = cp->provider;
843 		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
844 		while (table->gpt_smhead != 0) {
845 			i = ffs(table->gpt_smhead) - 1;
846 			error = g_write_data(cp, i * pp->sectorsize, buf,
847 			    pp->sectorsize);
848 			if (error) {
849 				g_free(buf);
850 				goto fail;
851 			}
852 			table->gpt_smhead &= ~(1 << i);
853 		}
854 		while (table->gpt_smtail != 0) {
855 			i = ffs(table->gpt_smtail) - 1;
856 			error = g_write_data(cp, pp->mediasize - (i + 1) *
857 			    pp->sectorsize, buf, pp->sectorsize);
858 			if (error) {
859 				g_free(buf);
860 				goto fail;
861 			}
862 			table->gpt_smtail &= ~(1 << i);
863 		}
864 		g_free(buf);
865 	}
866 
867 	if (table->gpt_scheme == &g_part_null_scheme) {
868 		g_topology_lock();
869 		g_access(cp, -1, -1, -1);
870 		g_part_wither(gp, ENXIO);
871 		return (0);
872 	}
873 
874 	error = G_PART_WRITE(table, cp);
875 	if (error)
876 		goto fail;
877 
878 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
879 		if (!entry->gpe_deleted) {
880 			entry->gpe_created = 0;
881 			entry->gpe_modified = 0;
882 			continue;
883 		}
884 		LIST_REMOVE(entry, gpe_entry);
885 		g_free(entry);
886 	}
887 	table->gpt_created = 0;
888 	table->gpt_opened = 0;
889 
890 	g_topology_lock();
891 	g_access(cp, -1, -1, -1);
892 	return (0);
893 
894 fail:
895 	g_topology_lock();
896 	gctl_error(req, "%d", error);
897 	return (error);
898 }
899 
900 static int
901 g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
902 {
903 	struct g_consumer *cp;
904 	struct g_geom *gp;
905 	struct g_provider *pp;
906 	struct g_part_scheme *scheme;
907 	struct g_part_table *null, *table;
908 	struct sbuf *sb;
909 	int attr, error;
910 
911 	pp = gpp->gpp_provider;
912 	scheme = gpp->gpp_scheme;
913 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
914 	g_topology_assert();
915 
916 	/* Check that there isn't already a g_part geom on the provider. */
917 	gp = g_part_find_geom(pp->name);
918 	if (gp != NULL) {
919 		null = gp->softc;
920 		if (null->gpt_scheme != &g_part_null_scheme) {
921 			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
922 			return (EEXIST);
923 		}
924 	} else
925 		null = NULL;
926 
927 	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
928 	    (gpp->gpp_entries < scheme->gps_minent ||
929 	     gpp->gpp_entries > scheme->gps_maxent)) {
930 		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
931 		return (EINVAL);
932 	}
933 
934 	if (null == NULL)
935 		gp = g_new_geomf(&g_part_class, "%s", pp->name);
936 	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
937 	    M_WAITOK);
938 	table = gp->softc;
939 	table->gpt_gp = gp;
940 	table->gpt_scheme = gpp->gpp_scheme;
941 	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
942 	    gpp->gpp_entries : scheme->gps_minent;
943 	LIST_INIT(&table->gpt_entry);
944 	if (null == NULL) {
945 		cp = g_new_consumer(gp);
946 		cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
947 		error = g_attach(cp, pp);
948 		if (error == 0)
949 			error = g_access(cp, 1, 1, 1);
950 		if (error != 0) {
951 			g_part_wither(gp, error);
952 			gctl_error(req, "%d geom '%s'", error, pp->name);
953 			return (error);
954 		}
955 		table->gpt_opened = 1;
956 	} else {
957 		cp = LIST_FIRST(&gp->consumer);
958 		table->gpt_opened = null->gpt_opened;
959 		table->gpt_smhead = null->gpt_smhead;
960 		table->gpt_smtail = null->gpt_smtail;
961 	}
962 
963 	g_topology_unlock();
964 
965 	/* Make sure the provider has media. */
966 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
967 		error = ENODEV;
968 		goto fail;
969 	}
970 
971 	/* Make sure we can nest and if so, determine our depth. */
972 	error = g_getattr("PART::isleaf", cp, &attr);
973 	if (!error && attr) {
974 		error = ENODEV;
975 		goto fail;
976 	}
977 	error = g_getattr("PART::depth", cp, &attr);
978 	table->gpt_depth = (!error) ? attr + 1 : 0;
979 
980 	/*
981 	 * Synthesize a disk geometry. Some partitioning schemes
982 	 * depend on it and since some file systems need it even
983 	 * when the partitition scheme doesn't, we do it here in
984 	 * scheme-independent code.
985 	 */
986 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
987 
988 	error = G_PART_CREATE(table, gpp);
989 	if (error)
990 		goto fail;
991 
992 	g_topology_lock();
993 
994 	table->gpt_created = 1;
995 	if (null != NULL)
996 		kobj_delete((kobj_t)null, M_GEOM);
997 
998 	/*
999 	 * Support automatic commit by filling in the gpp_geom
1000 	 * parameter.
1001 	 */
1002 	gpp->gpp_parms |= G_PART_PARM_GEOM;
1003 	gpp->gpp_geom = gp;
1004 
1005 	/* Provide feedback if so requested. */
1006 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1007 		sb = sbuf_new_auto();
1008 		sbuf_printf(sb, "%s created\n", gp->name);
1009 		sbuf_finish(sb);
1010 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1011 		sbuf_delete(sb);
1012 	}
1013 	return (0);
1014 
1015 fail:
1016 	g_topology_lock();
1017 	if (null == NULL) {
1018 		g_access(cp, -1, -1, -1);
1019 		g_part_wither(gp, error);
1020 	} else {
1021 		kobj_delete((kobj_t)gp->softc, M_GEOM);
1022 		gp->softc = null;
1023 	}
1024 	gctl_error(req, "%d provider", error);
1025 	return (error);
1026 }
1027 
1028 static int
1029 g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1030 {
1031 	struct g_geom *gp;
1032 	struct g_provider *pp;
1033 	struct g_part_entry *entry;
1034 	struct g_part_table *table;
1035 	struct sbuf *sb;
1036 
1037 	gp = gpp->gpp_geom;
1038 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1039 	g_topology_assert();
1040 
1041 	table = gp->softc;
1042 
1043 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1044 		if (entry->gpe_deleted || entry->gpe_internal)
1045 			continue;
1046 		if (entry->gpe_index == gpp->gpp_index)
1047 			break;
1048 	}
1049 	if (entry == NULL) {
1050 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1051 		return (ENOENT);
1052 	}
1053 
1054 	pp = entry->gpe_pp;
1055 	if (pp != NULL) {
1056 		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1057 			gctl_error(req, "%d", EBUSY);
1058 			return (EBUSY);
1059 		}
1060 
1061 		pp->private = NULL;
1062 		entry->gpe_pp = NULL;
1063 	}
1064 
1065 	if (pp != NULL)
1066 		g_wither_provider(pp, ENXIO);
1067 
1068 	/* Provide feedback if so requested. */
1069 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1070 		sb = sbuf_new_auto();
1071 		G_PART_FULLNAME(table, entry, sb, gp->name);
1072 		sbuf_cat(sb, " deleted\n");
1073 		sbuf_finish(sb);
1074 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1075 		sbuf_delete(sb);
1076 	}
1077 
1078 	if (entry->gpe_created) {
1079 		LIST_REMOVE(entry, gpe_entry);
1080 		g_free(entry);
1081 	} else {
1082 		entry->gpe_modified = 0;
1083 		entry->gpe_deleted = 1;
1084 	}
1085 	return (0);
1086 }
1087 
1088 static int
1089 g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1090 {
1091 	struct g_consumer *cp;
1092 	struct g_geom *gp;
1093 	struct g_provider *pp;
1094 	struct g_part_entry *entry, *tmp;
1095 	struct g_part_table *null, *table;
1096 	struct sbuf *sb;
1097 	int error;
1098 
1099 	gp = gpp->gpp_geom;
1100 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1101 	g_topology_assert();
1102 
1103 	table = gp->softc;
1104 	/* Check for busy providers. */
1105 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1106 		if (entry->gpe_deleted || entry->gpe_internal)
1107 			continue;
1108 		if (gpp->gpp_force) {
1109 			pp = entry->gpe_pp;
1110 			if (pp == NULL)
1111 				continue;
1112 			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1113 				continue;
1114 		}
1115 		gctl_error(req, "%d", EBUSY);
1116 		return (EBUSY);
1117 	}
1118 
1119 	if (gpp->gpp_force) {
1120 		/* Destroy all providers. */
1121 		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1122 			pp = entry->gpe_pp;
1123 			if (pp != NULL) {
1124 				pp->private = NULL;
1125 				g_wither_provider(pp, ENXIO);
1126 			}
1127 			LIST_REMOVE(entry, gpe_entry);
1128 			g_free(entry);
1129 		}
1130 	}
1131 
1132 	error = G_PART_DESTROY(table, gpp);
1133 	if (error) {
1134 		gctl_error(req, "%d", error);
1135 		return (error);
1136 	}
1137 
1138 	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1139 	    M_WAITOK);
1140 	null = gp->softc;
1141 	null->gpt_gp = gp;
1142 	null->gpt_scheme = &g_part_null_scheme;
1143 	LIST_INIT(&null->gpt_entry);
1144 
1145 	cp = LIST_FIRST(&gp->consumer);
1146 	pp = cp->provider;
1147 	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1148 
1149 	null->gpt_depth = table->gpt_depth;
1150 	null->gpt_opened = table->gpt_opened;
1151 	null->gpt_smhead = table->gpt_smhead;
1152 	null->gpt_smtail = table->gpt_smtail;
1153 
1154 	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1155 		LIST_REMOVE(entry, gpe_entry);
1156 		g_free(entry);
1157 	}
1158 	kobj_delete((kobj_t)table, M_GEOM);
1159 
1160 	/* Provide feedback if so requested. */
1161 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1162 		sb = sbuf_new_auto();
1163 		sbuf_printf(sb, "%s destroyed\n", gp->name);
1164 		sbuf_finish(sb);
1165 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1166 		sbuf_delete(sb);
1167 	}
1168 	return (0);
1169 }
1170 
1171 static int
1172 g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1173 {
1174 	struct g_geom *gp;
1175 	struct g_part_entry *entry;
1176 	struct g_part_table *table;
1177 	struct sbuf *sb;
1178 	int error;
1179 
1180 	gp = gpp->gpp_geom;
1181 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1182 	g_topology_assert();
1183 
1184 	table = gp->softc;
1185 
1186 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1187 		if (entry->gpe_deleted || entry->gpe_internal)
1188 			continue;
1189 		if (entry->gpe_index == gpp->gpp_index)
1190 			break;
1191 	}
1192 	if (entry == NULL) {
1193 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1194 		return (ENOENT);
1195 	}
1196 
1197 	error = G_PART_MODIFY(table, entry, gpp);
1198 	if (error) {
1199 		gctl_error(req, "%d", error);
1200 		return (error);
1201 	}
1202 
1203 	if (!entry->gpe_created)
1204 		entry->gpe_modified = 1;
1205 
1206 	/* Provide feedback if so requested. */
1207 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1208 		sb = sbuf_new_auto();
1209 		G_PART_FULLNAME(table, entry, sb, gp->name);
1210 		sbuf_cat(sb, " modified\n");
1211 		sbuf_finish(sb);
1212 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1213 		sbuf_delete(sb);
1214 	}
1215 	return (0);
1216 }
1217 
1218 static int
1219 g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1220 {
1221 	gctl_error(req, "%d verb 'move'", ENOSYS);
1222 	return (ENOSYS);
1223 }
1224 
1225 static int
1226 g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1227 {
1228 	struct g_part_table *table;
1229 	struct g_geom *gp;
1230 	struct sbuf *sb;
1231 	int error, recovered;
1232 
1233 	gp = gpp->gpp_geom;
1234 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1235 	g_topology_assert();
1236 	table = gp->softc;
1237 	error = recovered = 0;
1238 
1239 	if (table->gpt_corrupt) {
1240 		error = G_PART_RECOVER(table);
1241 		if (error == 0)
1242 			error = g_part_check_integrity(table,
1243 			    LIST_FIRST(&gp->consumer));
1244 		if (error) {
1245 			gctl_error(req, "%d recovering '%s' failed",
1246 			    error, gp->name);
1247 			return (error);
1248 		}
1249 		recovered = 1;
1250 	}
1251 	/* Provide feedback if so requested. */
1252 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1253 		sb = sbuf_new_auto();
1254 		if (recovered)
1255 			sbuf_printf(sb, "%s recovered\n", gp->name);
1256 		else
1257 			sbuf_printf(sb, "%s recovering is not needed\n",
1258 			    gp->name);
1259 		sbuf_finish(sb);
1260 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1261 		sbuf_delete(sb);
1262 	}
1263 	return (0);
1264 }
1265 
1266 static int
1267 g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1268 {
1269 	struct g_geom *gp;
1270 	struct g_provider *pp;
1271 	struct g_part_entry *pe, *entry;
1272 	struct g_part_table *table;
1273 	struct sbuf *sb;
1274 	quad_t end;
1275 	int error;
1276 	off_t mediasize;
1277 
1278 	gp = gpp->gpp_geom;
1279 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1280 	g_topology_assert();
1281 	table = gp->softc;
1282 
1283 	/* check gpp_index */
1284 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1285 		if (entry->gpe_deleted || entry->gpe_internal)
1286 			continue;
1287 		if (entry->gpe_index == gpp->gpp_index)
1288 			break;
1289 	}
1290 	if (entry == NULL) {
1291 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1292 		return (ENOENT);
1293 	}
1294 
1295 	/* check gpp_size */
1296 	end = entry->gpe_start + gpp->gpp_size - 1;
1297 	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1298 		gctl_error(req, "%d size '%jd'", EINVAL,
1299 		    (intmax_t)gpp->gpp_size);
1300 		return (EINVAL);
1301 	}
1302 
1303 	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1304 		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1305 			continue;
1306 		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1307 			gctl_error(req, "%d end '%jd'", ENOSPC,
1308 			    (intmax_t)end);
1309 			return (ENOSPC);
1310 		}
1311 		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1312 			gctl_error(req, "%d size '%jd'", ENOSPC,
1313 			    (intmax_t)gpp->gpp_size);
1314 			return (ENOSPC);
1315 		}
1316 	}
1317 
1318 	pp = entry->gpe_pp;
1319 	if ((g_debugflags & 16) == 0 &&
1320 	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1321 		if (entry->gpe_end - entry->gpe_start + 1 > gpp->gpp_size) {
1322 			/* Deny shrinking of an opened partition. */
1323 			gctl_error(req, "%d", EBUSY);
1324 			return (EBUSY);
1325 		}
1326 	}
1327 
1328 	error = G_PART_RESIZE(table, entry, gpp);
1329 	if (error) {
1330 		gctl_error(req, "%d%s", error, error != EBUSY ? "":
1331 		    " resizing will lead to unexpected shrinking"
1332 		    " due to alignment");
1333 		return (error);
1334 	}
1335 
1336 	if (!entry->gpe_created)
1337 		entry->gpe_modified = 1;
1338 
1339 	/* update mediasize of changed provider */
1340 	mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1341 		pp->sectorsize;
1342 	g_resize_provider(pp, mediasize);
1343 
1344 	/* Provide feedback if so requested. */
1345 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1346 		sb = sbuf_new_auto();
1347 		G_PART_FULLNAME(table, entry, sb, gp->name);
1348 		sbuf_cat(sb, " resized\n");
1349 		sbuf_finish(sb);
1350 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1351 		sbuf_delete(sb);
1352 	}
1353 	return (0);
1354 }
1355 
1356 static int
1357 g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1358     unsigned int set)
1359 {
1360 	struct g_geom *gp;
1361 	struct g_part_entry *entry;
1362 	struct g_part_table *table;
1363 	struct sbuf *sb;
1364 	int error;
1365 
1366 	gp = gpp->gpp_geom;
1367 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1368 	g_topology_assert();
1369 
1370 	table = gp->softc;
1371 
1372 	if (gpp->gpp_parms & G_PART_PARM_INDEX) {
1373 		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1374 			if (entry->gpe_deleted || entry->gpe_internal)
1375 				continue;
1376 			if (entry->gpe_index == gpp->gpp_index)
1377 				break;
1378 		}
1379 		if (entry == NULL) {
1380 			gctl_error(req, "%d index '%d'", ENOENT,
1381 			    gpp->gpp_index);
1382 			return (ENOENT);
1383 		}
1384 	} else
1385 		entry = NULL;
1386 
1387 	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1388 	if (error) {
1389 		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1390 		return (error);
1391 	}
1392 
1393 	/* Provide feedback if so requested. */
1394 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1395 		sb = sbuf_new_auto();
1396 		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1397 		    (set) ? "" : "un");
1398 		if (entry)
1399 			G_PART_FULLNAME(table, entry, sb, gp->name);
1400 		else
1401 			sbuf_cat(sb, gp->name);
1402 		sbuf_cat(sb, "\n");
1403 		sbuf_finish(sb);
1404 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1405 		sbuf_delete(sb);
1406 	}
1407 	return (0);
1408 }
1409 
1410 static int
1411 g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1412 {
1413 	struct g_consumer *cp;
1414 	struct g_provider *pp;
1415 	struct g_geom *gp;
1416 	struct g_part_entry *entry, *tmp;
1417 	struct g_part_table *table;
1418 	int error, reprobe;
1419 
1420 	gp = gpp->gpp_geom;
1421 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1422 	g_topology_assert();
1423 
1424 	table = gp->softc;
1425 	if (!table->gpt_opened) {
1426 		gctl_error(req, "%d", EPERM);
1427 		return (EPERM);
1428 	}
1429 
1430 	cp = LIST_FIRST(&gp->consumer);
1431 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1432 		entry->gpe_modified = 0;
1433 		if (entry->gpe_created) {
1434 			pp = entry->gpe_pp;
1435 			if (pp != NULL) {
1436 				pp->private = NULL;
1437 				entry->gpe_pp = NULL;
1438 				g_wither_provider(pp, ENXIO);
1439 			}
1440 			entry->gpe_deleted = 1;
1441 		}
1442 		if (entry->gpe_deleted) {
1443 			LIST_REMOVE(entry, gpe_entry);
1444 			g_free(entry);
1445 		}
1446 	}
1447 
1448 	g_topology_unlock();
1449 
1450 	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1451 	    table->gpt_created) ? 1 : 0;
1452 
1453 	if (reprobe) {
1454 		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1455 			if (entry->gpe_internal)
1456 				continue;
1457 			error = EBUSY;
1458 			goto fail;
1459 		}
1460 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1461 			LIST_REMOVE(entry, gpe_entry);
1462 			g_free(entry);
1463 		}
1464 		error = g_part_probe(gp, cp, table->gpt_depth);
1465 		if (error) {
1466 			g_topology_lock();
1467 			g_access(cp, -1, -1, -1);
1468 			g_part_wither(gp, error);
1469 			return (0);
1470 		}
1471 		table = gp->softc;
1472 
1473 		/*
1474 		 * Synthesize a disk geometry. Some partitioning schemes
1475 		 * depend on it and since some file systems need it even
1476 		 * when the partitition scheme doesn't, we do it here in
1477 		 * scheme-independent code.
1478 		 */
1479 		pp = cp->provider;
1480 		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1481 	}
1482 
1483 	error = G_PART_READ(table, cp);
1484 	if (error)
1485 		goto fail;
1486 	error = g_part_check_integrity(table, cp);
1487 	if (error)
1488 		goto fail;
1489 
1490 	g_topology_lock();
1491 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1492 		if (!entry->gpe_internal)
1493 			g_part_new_provider(gp, table, entry);
1494 	}
1495 
1496 	table->gpt_opened = 0;
1497 	g_access(cp, -1, -1, -1);
1498 	return (0);
1499 
1500 fail:
1501 	g_topology_lock();
1502 	gctl_error(req, "%d", error);
1503 	return (error);
1504 }
1505 
1506 static void
1507 g_part_wither(struct g_geom *gp, int error)
1508 {
1509 	struct g_part_entry *entry;
1510 	struct g_part_table *table;
1511 
1512 	table = gp->softc;
1513 	if (table != NULL) {
1514 		G_PART_DESTROY(table, NULL);
1515 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1516 			LIST_REMOVE(entry, gpe_entry);
1517 			g_free(entry);
1518 		}
1519 		if (gp->softc != NULL) {
1520 			kobj_delete((kobj_t)gp->softc, M_GEOM);
1521 			gp->softc = NULL;
1522 		}
1523 	}
1524 	g_wither_geom(gp, error);
1525 }
1526 
1527 /*
1528  * Class methods.
1529  */
1530 
1531 static void
1532 g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1533 {
1534 	struct g_part_parms gpp;
1535 	struct g_part_table *table;
1536 	struct gctl_req_arg *ap;
1537 	enum g_part_ctl ctlreq;
1538 	unsigned int i, mparms, oparms, parm;
1539 	int auto_commit, close_on_error;
1540 	int error, modifies;
1541 
1542 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1543 	g_topology_assert();
1544 
1545 	ctlreq = G_PART_CTL_NONE;
1546 	modifies = 1;
1547 	mparms = 0;
1548 	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1549 	switch (*verb) {
1550 	case 'a':
1551 		if (!strcmp(verb, "add")) {
1552 			ctlreq = G_PART_CTL_ADD;
1553 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1554 			    G_PART_PARM_START | G_PART_PARM_TYPE;
1555 			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1556 		}
1557 		break;
1558 	case 'b':
1559 		if (!strcmp(verb, "bootcode")) {
1560 			ctlreq = G_PART_CTL_BOOTCODE;
1561 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1562 		}
1563 		break;
1564 	case 'c':
1565 		if (!strcmp(verb, "commit")) {
1566 			ctlreq = G_PART_CTL_COMMIT;
1567 			mparms |= G_PART_PARM_GEOM;
1568 			modifies = 0;
1569 		} else if (!strcmp(verb, "create")) {
1570 			ctlreq = G_PART_CTL_CREATE;
1571 			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1572 			oparms |= G_PART_PARM_ENTRIES;
1573 		}
1574 		break;
1575 	case 'd':
1576 		if (!strcmp(verb, "delete")) {
1577 			ctlreq = G_PART_CTL_DELETE;
1578 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1579 		} else if (!strcmp(verb, "destroy")) {
1580 			ctlreq = G_PART_CTL_DESTROY;
1581 			mparms |= G_PART_PARM_GEOM;
1582 			oparms |= G_PART_PARM_FORCE;
1583 		}
1584 		break;
1585 	case 'm':
1586 		if (!strcmp(verb, "modify")) {
1587 			ctlreq = G_PART_CTL_MODIFY;
1588 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1589 			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1590 		} else if (!strcmp(verb, "move")) {
1591 			ctlreq = G_PART_CTL_MOVE;
1592 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1593 		}
1594 		break;
1595 	case 'r':
1596 		if (!strcmp(verb, "recover")) {
1597 			ctlreq = G_PART_CTL_RECOVER;
1598 			mparms |= G_PART_PARM_GEOM;
1599 		} else if (!strcmp(verb, "resize")) {
1600 			ctlreq = G_PART_CTL_RESIZE;
1601 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1602 			    G_PART_PARM_SIZE;
1603 		}
1604 		break;
1605 	case 's':
1606 		if (!strcmp(verb, "set")) {
1607 			ctlreq = G_PART_CTL_SET;
1608 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM;
1609 			oparms |= G_PART_PARM_INDEX;
1610 		}
1611 		break;
1612 	case 'u':
1613 		if (!strcmp(verb, "undo")) {
1614 			ctlreq = G_PART_CTL_UNDO;
1615 			mparms |= G_PART_PARM_GEOM;
1616 			modifies = 0;
1617 		} else if (!strcmp(verb, "unset")) {
1618 			ctlreq = G_PART_CTL_UNSET;
1619 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM;
1620 			oparms |= G_PART_PARM_INDEX;
1621 		}
1622 		break;
1623 	}
1624 	if (ctlreq == G_PART_CTL_NONE) {
1625 		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1626 		return;
1627 	}
1628 
1629 	bzero(&gpp, sizeof(gpp));
1630 	for (i = 0; i < req->narg; i++) {
1631 		ap = &req->arg[i];
1632 		parm = 0;
1633 		switch (ap->name[0]) {
1634 		case 'a':
1635 			if (!strcmp(ap->name, "arg0")) {
1636 				parm = mparms &
1637 				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1638 			}
1639 			if (!strcmp(ap->name, "attrib"))
1640 				parm = G_PART_PARM_ATTRIB;
1641 			break;
1642 		case 'b':
1643 			if (!strcmp(ap->name, "bootcode"))
1644 				parm = G_PART_PARM_BOOTCODE;
1645 			break;
1646 		case 'c':
1647 			if (!strcmp(ap->name, "class"))
1648 				continue;
1649 			break;
1650 		case 'e':
1651 			if (!strcmp(ap->name, "entries"))
1652 				parm = G_PART_PARM_ENTRIES;
1653 			break;
1654 		case 'f':
1655 			if (!strcmp(ap->name, "flags"))
1656 				parm = G_PART_PARM_FLAGS;
1657 			else if (!strcmp(ap->name, "force"))
1658 				parm = G_PART_PARM_FORCE;
1659 			break;
1660 		case 'i':
1661 			if (!strcmp(ap->name, "index"))
1662 				parm = G_PART_PARM_INDEX;
1663 			break;
1664 		case 'l':
1665 			if (!strcmp(ap->name, "label"))
1666 				parm = G_PART_PARM_LABEL;
1667 			break;
1668 		case 'o':
1669 			if (!strcmp(ap->name, "output"))
1670 				parm = G_PART_PARM_OUTPUT;
1671 			break;
1672 		case 's':
1673 			if (!strcmp(ap->name, "scheme"))
1674 				parm = G_PART_PARM_SCHEME;
1675 			else if (!strcmp(ap->name, "size"))
1676 				parm = G_PART_PARM_SIZE;
1677 			else if (!strcmp(ap->name, "start"))
1678 				parm = G_PART_PARM_START;
1679 			break;
1680 		case 't':
1681 			if (!strcmp(ap->name, "type"))
1682 				parm = G_PART_PARM_TYPE;
1683 			break;
1684 		case 'v':
1685 			if (!strcmp(ap->name, "verb"))
1686 				continue;
1687 			else if (!strcmp(ap->name, "version"))
1688 				parm = G_PART_PARM_VERSION;
1689 			break;
1690 		}
1691 		if ((parm & (mparms | oparms)) == 0) {
1692 			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1693 			return;
1694 		}
1695 		switch (parm) {
1696 		case G_PART_PARM_ATTRIB:
1697 			error = g_part_parm_str(req, ap->name,
1698 			    &gpp.gpp_attrib);
1699 			break;
1700 		case G_PART_PARM_BOOTCODE:
1701 			error = g_part_parm_bootcode(req, ap->name,
1702 			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1703 			break;
1704 		case G_PART_PARM_ENTRIES:
1705 			error = g_part_parm_intmax(req, ap->name,
1706 			    &gpp.gpp_entries);
1707 			break;
1708 		case G_PART_PARM_FLAGS:
1709 			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1710 			break;
1711 		case G_PART_PARM_FORCE:
1712 			error = g_part_parm_uint32(req, ap->name,
1713 			    &gpp.gpp_force);
1714 			break;
1715 		case G_PART_PARM_GEOM:
1716 			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1717 			break;
1718 		case G_PART_PARM_INDEX:
1719 			error = g_part_parm_intmax(req, ap->name,
1720 			    &gpp.gpp_index);
1721 			break;
1722 		case G_PART_PARM_LABEL:
1723 			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1724 			break;
1725 		case G_PART_PARM_OUTPUT:
1726 			error = 0;	/* Write-only parameter */
1727 			break;
1728 		case G_PART_PARM_PROVIDER:
1729 			error = g_part_parm_provider(req, ap->name,
1730 			    &gpp.gpp_provider);
1731 			break;
1732 		case G_PART_PARM_SCHEME:
1733 			error = g_part_parm_scheme(req, ap->name,
1734 			    &gpp.gpp_scheme);
1735 			break;
1736 		case G_PART_PARM_SIZE:
1737 			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1738 			break;
1739 		case G_PART_PARM_START:
1740 			error = g_part_parm_quad(req, ap->name,
1741 			    &gpp.gpp_start);
1742 			break;
1743 		case G_PART_PARM_TYPE:
1744 			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1745 			break;
1746 		case G_PART_PARM_VERSION:
1747 			error = g_part_parm_uint32(req, ap->name,
1748 			    &gpp.gpp_version);
1749 			break;
1750 		default:
1751 			error = EDOOFUS;
1752 			gctl_error(req, "%d %s", error, ap->name);
1753 			break;
1754 		}
1755 		if (error != 0) {
1756 			if (error == ENOATTR) {
1757 				gctl_error(req, "%d param '%s'", error,
1758 				    ap->name);
1759 			}
1760 			return;
1761 		}
1762 		gpp.gpp_parms |= parm;
1763 	}
1764 	if ((gpp.gpp_parms & mparms) != mparms) {
1765 		parm = mparms - (gpp.gpp_parms & mparms);
1766 		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1767 		return;
1768 	}
1769 
1770 	/* Obtain permissions if possible/necessary. */
1771 	close_on_error = 0;
1772 	table = NULL;
1773 	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1774 		table = gpp.gpp_geom->softc;
1775 		if (table != NULL && table->gpt_corrupt &&
1776 		    ctlreq != G_PART_CTL_DESTROY &&
1777 		    ctlreq != G_PART_CTL_RECOVER) {
1778 			gctl_error(req, "%d table '%s' is corrupt",
1779 			    EPERM, gpp.gpp_geom->name);
1780 			return;
1781 		}
1782 		if (table != NULL && !table->gpt_opened) {
1783 			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1784 			    1, 1, 1);
1785 			if (error) {
1786 				gctl_error(req, "%d geom '%s'", error,
1787 				    gpp.gpp_geom->name);
1788 				return;
1789 			}
1790 			table->gpt_opened = 1;
1791 			close_on_error = 1;
1792 		}
1793 	}
1794 
1795 	/* Allow the scheme to check or modify the parameters. */
1796 	if (table != NULL) {
1797 		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1798 		if (error) {
1799 			gctl_error(req, "%d pre-check failed", error);
1800 			goto out;
1801 		}
1802 	} else
1803 		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1804 
1805 	switch (ctlreq) {
1806 	case G_PART_CTL_NONE:
1807 		panic("%s", __func__);
1808 	case G_PART_CTL_ADD:
1809 		error = g_part_ctl_add(req, &gpp);
1810 		break;
1811 	case G_PART_CTL_BOOTCODE:
1812 		error = g_part_ctl_bootcode(req, &gpp);
1813 		break;
1814 	case G_PART_CTL_COMMIT:
1815 		error = g_part_ctl_commit(req, &gpp);
1816 		break;
1817 	case G_PART_CTL_CREATE:
1818 		error = g_part_ctl_create(req, &gpp);
1819 		break;
1820 	case G_PART_CTL_DELETE:
1821 		error = g_part_ctl_delete(req, &gpp);
1822 		break;
1823 	case G_PART_CTL_DESTROY:
1824 		error = g_part_ctl_destroy(req, &gpp);
1825 		break;
1826 	case G_PART_CTL_MODIFY:
1827 		error = g_part_ctl_modify(req, &gpp);
1828 		break;
1829 	case G_PART_CTL_MOVE:
1830 		error = g_part_ctl_move(req, &gpp);
1831 		break;
1832 	case G_PART_CTL_RECOVER:
1833 		error = g_part_ctl_recover(req, &gpp);
1834 		break;
1835 	case G_PART_CTL_RESIZE:
1836 		error = g_part_ctl_resize(req, &gpp);
1837 		break;
1838 	case G_PART_CTL_SET:
1839 		error = g_part_ctl_setunset(req, &gpp, 1);
1840 		break;
1841 	case G_PART_CTL_UNDO:
1842 		error = g_part_ctl_undo(req, &gpp);
1843 		break;
1844 	case G_PART_CTL_UNSET:
1845 		error = g_part_ctl_setunset(req, &gpp, 0);
1846 		break;
1847 	}
1848 
1849 	/* Implement automatic commit. */
1850 	if (!error) {
1851 		auto_commit = (modifies &&
1852 		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1853 		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1854 		if (auto_commit) {
1855 			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1856 			    __func__));
1857 			error = g_part_ctl_commit(req, &gpp);
1858 		}
1859 	}
1860 
1861  out:
1862 	if (error && close_on_error) {
1863 		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1864 		table->gpt_opened = 0;
1865 	}
1866 }
1867 
1868 static int
1869 g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1870     struct g_geom *gp)
1871 {
1872 
1873 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1874 	g_topology_assert();
1875 
1876 	g_part_wither(gp, EINVAL);
1877 	return (0);
1878 }
1879 
1880 static struct g_geom *
1881 g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1882 {
1883 	struct g_consumer *cp;
1884 	struct g_geom *gp;
1885 	struct g_part_entry *entry;
1886 	struct g_part_table *table;
1887 	struct root_hold_token *rht;
1888 	int attr, depth;
1889 	int error;
1890 
1891 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1892 	g_topology_assert();
1893 
1894 	/* Skip providers that are already open for writing. */
1895 	if (pp->acw > 0)
1896 		return (NULL);
1897 
1898 	/*
1899 	 * Create a GEOM with consumer and hook it up to the provider.
1900 	 * With that we become part of the topology. Optain read access
1901 	 * to the provider.
1902 	 */
1903 	gp = g_new_geomf(mp, "%s", pp->name);
1904 	cp = g_new_consumer(gp);
1905 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
1906 	error = g_attach(cp, pp);
1907 	if (error == 0)
1908 		error = g_access(cp, 1, 0, 0);
1909 	if (error != 0) {
1910 		if (cp->provider)
1911 			g_detach(cp);
1912 		g_destroy_consumer(cp);
1913 		g_destroy_geom(gp);
1914 		return (NULL);
1915 	}
1916 
1917 	rht = root_mount_hold(mp->name);
1918 	g_topology_unlock();
1919 
1920 	/*
1921 	 * Short-circuit the whole probing galore when there's no
1922 	 * media present.
1923 	 */
1924 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1925 		error = ENODEV;
1926 		goto fail;
1927 	}
1928 
1929 	/* Make sure we can nest and if so, determine our depth. */
1930 	error = g_getattr("PART::isleaf", cp, &attr);
1931 	if (!error && attr) {
1932 		error = ENODEV;
1933 		goto fail;
1934 	}
1935 	error = g_getattr("PART::depth", cp, &attr);
1936 	depth = (!error) ? attr + 1 : 0;
1937 
1938 	error = g_part_probe(gp, cp, depth);
1939 	if (error)
1940 		goto fail;
1941 
1942 	table = gp->softc;
1943 
1944 	/*
1945 	 * Synthesize a disk geometry. Some partitioning schemes
1946 	 * depend on it and since some file systems need it even
1947 	 * when the partitition scheme doesn't, we do it here in
1948 	 * scheme-independent code.
1949 	 */
1950 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1951 
1952 	error = G_PART_READ(table, cp);
1953 	if (error)
1954 		goto fail;
1955 	error = g_part_check_integrity(table, cp);
1956 	if (error)
1957 		goto fail;
1958 
1959 	g_topology_lock();
1960 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1961 		if (!entry->gpe_internal)
1962 			g_part_new_provider(gp, table, entry);
1963 	}
1964 
1965 	root_mount_rel(rht);
1966 	g_access(cp, -1, 0, 0);
1967 	return (gp);
1968 
1969  fail:
1970 	g_topology_lock();
1971 	root_mount_rel(rht);
1972 	g_access(cp, -1, 0, 0);
1973 	g_detach(cp);
1974 	g_destroy_consumer(cp);
1975 	g_destroy_geom(gp);
1976 	return (NULL);
1977 }
1978 
1979 /*
1980  * Geom methods.
1981  */
1982 
1983 static int
1984 g_part_access(struct g_provider *pp, int dr, int dw, int de)
1985 {
1986 	struct g_consumer *cp;
1987 
1988 	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1989 	    dw, de));
1990 
1991 	cp = LIST_FIRST(&pp->geom->consumer);
1992 
1993 	/* We always gain write-exclusive access. */
1994 	return (g_access(cp, dr, dw, dw + de));
1995 }
1996 
1997 static void
1998 g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1999     struct g_consumer *cp, struct g_provider *pp)
2000 {
2001 	char buf[64];
2002 	struct g_part_entry *entry;
2003 	struct g_part_table *table;
2004 
2005 	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
2006 	table = gp->softc;
2007 
2008 	if (indent == NULL) {
2009 		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
2010 		entry = pp->private;
2011 		if (entry == NULL)
2012 			return;
2013 		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
2014 		    (uintmax_t)entry->gpe_offset,
2015 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2016 		/*
2017 		 * libdisk compatibility quirk - the scheme dumps the
2018 		 * slicer name and partition type in a way that is
2019 		 * compatible with libdisk. When libdisk is not used
2020 		 * anymore, this should go away.
2021 		 */
2022 		G_PART_DUMPCONF(table, entry, sb, indent);
2023 	} else if (cp != NULL) {	/* Consumer configuration. */
2024 		KASSERT(pp == NULL, ("%s", __func__));
2025 		/* none */
2026 	} else if (pp != NULL) {	/* Provider configuration. */
2027 		entry = pp->private;
2028 		if (entry == NULL)
2029 			return;
2030 		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
2031 		    (uintmax_t)entry->gpe_start);
2032 		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
2033 		    (uintmax_t)entry->gpe_end);
2034 		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
2035 		    entry->gpe_index);
2036 		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2037 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2038 		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
2039 		    (uintmax_t)entry->gpe_offset);
2040 		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
2041 		    (uintmax_t)pp->mediasize);
2042 		G_PART_DUMPCONF(table, entry, sb, indent);
2043 	} else {			/* Geom configuration. */
2044 		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2045 		    table->gpt_scheme->name);
2046 		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2047 		    table->gpt_entries);
2048 		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2049 		    (uintmax_t)table->gpt_first);
2050 		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2051 		    (uintmax_t)table->gpt_last);
2052 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2053 		    table->gpt_sectors);
2054 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2055 		    table->gpt_heads);
2056 		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2057 		    table->gpt_corrupt ? "CORRUPT": "OK");
2058 		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2059 		    table->gpt_opened ? "true": "false");
2060 		G_PART_DUMPCONF(table, NULL, sb, indent);
2061 	}
2062 }
2063 
2064 /*-
2065  * This start routine is only called for non-trivial requests, all the
2066  * trivial ones are handled autonomously by the slice code.
2067  * For requests we handle here, we must call the g_io_deliver() on the
2068  * bio, and return non-zero to indicate to the slice code that we did so.
2069  * This code executes in the "DOWN" I/O path, this means:
2070  *    * No sleeping.
2071  *    * Don't grab the topology lock.
2072  *    * Don't call biowait, g_getattr(), g_setattr() or g_read_data()
2073  */
2074 static int
2075 g_part_ioctl(struct g_provider *pp, u_long cmd, void *data, int fflag, struct thread *td)
2076 {
2077 	struct g_part_table *table;
2078 
2079 	table = pp->geom->softc;
2080 	return G_PART_IOCTL(table, pp, cmd, data, fflag, td);
2081 }
2082 
2083 static void
2084 g_part_resize(struct g_consumer *cp)
2085 {
2086 	struct g_part_table *table;
2087 
2088 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2089 	g_topology_assert();
2090 
2091 	table = cp->geom->softc;
2092 	if (table->gpt_opened == 0) {
2093 		if (g_access(cp, 1, 1, 1) != 0)
2094 			return;
2095 		table->gpt_opened = 1;
2096 	}
2097 	if (G_PART_RESIZE(table, NULL, NULL) == 0)
2098 		printf("GEOM_PART: %s was automatically resized.\n"
2099 		    "  Use `gpart commit %s` to save changes or "
2100 		    "`gpart undo %s` to revert them.\n", cp->geom->name,
2101 		    cp->geom->name, cp->geom->name);
2102 	if (g_part_check_integrity(table, cp) != 0) {
2103 		g_access(cp, -1, -1, -1);
2104 		table->gpt_opened = 0;
2105 		g_part_wither(table->gpt_gp, ENXIO);
2106 	}
2107 }
2108 
2109 static void
2110 g_part_orphan(struct g_consumer *cp)
2111 {
2112 	struct g_provider *pp;
2113 	struct g_part_table *table;
2114 
2115 	pp = cp->provider;
2116 	KASSERT(pp != NULL, ("%s", __func__));
2117 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2118 	g_topology_assert();
2119 
2120 	KASSERT(pp->error != 0, ("%s", __func__));
2121 	table = cp->geom->softc;
2122 	if (table != NULL && table->gpt_opened)
2123 		g_access(cp, -1, -1, -1);
2124 	g_part_wither(cp->geom, pp->error);
2125 }
2126 
2127 static void
2128 g_part_spoiled(struct g_consumer *cp)
2129 {
2130 
2131 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2132 	g_topology_assert();
2133 
2134 	cp->flags |= G_CF_ORPHAN;
2135 	g_part_wither(cp->geom, ENXIO);
2136 }
2137 
2138 static void
2139 g_part_start(struct bio *bp)
2140 {
2141 	struct bio *bp2;
2142 	struct g_consumer *cp;
2143 	struct g_geom *gp;
2144 	struct g_part_entry *entry;
2145 	struct g_part_table *table;
2146 	struct g_kerneldump *gkd;
2147 	struct g_provider *pp;
2148 	char buf[64];
2149 
2150 	pp = bp->bio_to;
2151 	gp = pp->geom;
2152 	table = gp->softc;
2153 	cp = LIST_FIRST(&gp->consumer);
2154 
2155 	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2156 	    pp->name));
2157 
2158 	entry = pp->private;
2159 	if (entry == NULL) {
2160 		g_io_deliver(bp, ENXIO);
2161 		return;
2162 	}
2163 
2164 	switch(bp->bio_cmd) {
2165 	case BIO_DELETE:
2166 	case BIO_READ:
2167 	case BIO_WRITE:
2168 		if (bp->bio_offset >= pp->mediasize) {
2169 			g_io_deliver(bp, EIO);
2170 			return;
2171 		}
2172 		bp2 = g_clone_bio(bp);
2173 		if (bp2 == NULL) {
2174 			g_io_deliver(bp, ENOMEM);
2175 			return;
2176 		}
2177 		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2178 			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2179 		bp2->bio_done = g_std_done;
2180 		bp2->bio_offset += entry->gpe_offset;
2181 		g_io_request(bp2, cp);
2182 		return;
2183 	case BIO_FLUSH:
2184 		break;
2185 	case BIO_GETATTR:
2186 		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2187 			return;
2188 		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2189 			return;
2190 		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2191 			return;
2192 		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2193 			return;
2194 		if (g_handleattr_str(bp, "PART::scheme",
2195 		    table->gpt_scheme->name))
2196 			return;
2197 		if (g_handleattr_str(bp, "PART::type",
2198 		    G_PART_TYPE(table, entry, buf, sizeof(buf))))
2199 			return;
2200 		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2201 			/*
2202 			 * Check that the partition is suitable for kernel
2203 			 * dumps. Typically only swap partitions should be
2204 			 * used. If the request comes from the nested scheme
2205 			 * we allow dumping there as well.
2206 			 */
2207 			if ((bp->bio_from == NULL ||
2208 			    bp->bio_from->geom->class != &g_part_class) &&
2209 			    G_PART_DUMPTO(table, entry) == 0) {
2210 				g_io_deliver(bp, ENODEV);
2211 				printf("GEOM_PART: Partition '%s' not suitable"
2212 				    " for kernel dumps (wrong type?)\n",
2213 				    pp->name);
2214 				return;
2215 			}
2216 			gkd = (struct g_kerneldump *)bp->bio_data;
2217 			if (gkd->offset >= pp->mediasize) {
2218 				g_io_deliver(bp, EIO);
2219 				return;
2220 			}
2221 			if (gkd->offset + gkd->length > pp->mediasize)
2222 				gkd->length = pp->mediasize - gkd->offset;
2223 			gkd->offset += entry->gpe_offset;
2224 		}
2225 		break;
2226 	default:
2227 		g_io_deliver(bp, EOPNOTSUPP);
2228 		return;
2229 	}
2230 
2231 	bp2 = g_clone_bio(bp);
2232 	if (bp2 == NULL) {
2233 		g_io_deliver(bp, ENOMEM);
2234 		return;
2235 	}
2236 	bp2->bio_done = g_std_done;
2237 	g_io_request(bp2, cp);
2238 }
2239 
2240 static void
2241 g_part_init(struct g_class *mp)
2242 {
2243 
2244 	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2245 }
2246 
2247 static void
2248 g_part_fini(struct g_class *mp)
2249 {
2250 
2251 	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2252 }
2253 
2254 static void
2255 g_part_unload_event(void *arg, int flag)
2256 {
2257 	struct g_consumer *cp;
2258 	struct g_geom *gp;
2259 	struct g_provider *pp;
2260 	struct g_part_scheme *scheme;
2261 	struct g_part_table *table;
2262 	uintptr_t *xchg;
2263 	int acc, error;
2264 
2265 	if (flag == EV_CANCEL)
2266 		return;
2267 
2268 	xchg = arg;
2269 	error = 0;
2270 	scheme = (void *)(*xchg);
2271 
2272 	g_topology_assert();
2273 
2274 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2275 		table = gp->softc;
2276 		if (table->gpt_scheme != scheme)
2277 			continue;
2278 
2279 		acc = 0;
2280 		LIST_FOREACH(pp, &gp->provider, provider)
2281 			acc += pp->acr + pp->acw + pp->ace;
2282 		LIST_FOREACH(cp, &gp->consumer, consumer)
2283 			acc += cp->acr + cp->acw + cp->ace;
2284 
2285 		if (!acc)
2286 			g_part_wither(gp, ENOSYS);
2287 		else
2288 			error = EBUSY;
2289 	}
2290 
2291 	if (!error)
2292 		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2293 
2294 	*xchg = error;
2295 }
2296 
2297 int
2298 g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2299 {
2300 	struct g_part_scheme *iter;
2301 	uintptr_t arg;
2302 	int error;
2303 
2304 	error = 0;
2305 	switch (type) {
2306 	case MOD_LOAD:
2307 		TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
2308 			if (scheme == iter) {
2309 				printf("GEOM_PART: scheme %s is already "
2310 				    "registered!\n", scheme->name);
2311 				break;
2312 			}
2313 		}
2314 		if (iter == NULL) {
2315 			TAILQ_INSERT_TAIL(&g_part_schemes, scheme,
2316 			    scheme_list);
2317 			g_retaste(&g_part_class);
2318 		}
2319 		break;
2320 	case MOD_UNLOAD:
2321 		arg = (uintptr_t)scheme;
2322 		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2323 		    NULL);
2324 		if (error == 0)
2325 			error = arg;
2326 		break;
2327 	default:
2328 		error = EOPNOTSUPP;
2329 		break;
2330 	}
2331 
2332 	return (error);
2333 }
2334