xref: /freebsd/sys/geom/part/g_part.c (revision 595e514d0df2bac5b813d35f83e32875dbf16a83)
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/diskmbr.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/kobj.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/queue.h>
41 #include <sys/sbuf.h>
42 #include <sys/sysctl.h>
43 #include <sys/systm.h>
44 #include <sys/uuid.h>
45 #include <geom/geom.h>
46 #include <geom/geom_ctl.h>
47 #include <geom/geom_int.h>
48 #include <geom/part/g_part.h>
49 
50 #include "g_part_if.h"
51 
52 #ifndef _PATH_DEV
53 #define _PATH_DEV "/dev/"
54 #endif
55 
56 static kobj_method_t g_part_null_methods[] = {
57 	{ 0, 0 }
58 };
59 
60 static struct g_part_scheme g_part_null_scheme = {
61 	"(none)",
62 	g_part_null_methods,
63 	sizeof(struct g_part_table),
64 };
65 
66 TAILQ_HEAD(, g_part_scheme) g_part_schemes =
67     TAILQ_HEAD_INITIALIZER(g_part_schemes);
68 
69 struct g_part_alias_list {
70 	const char *lexeme;
71 	enum g_part_alias alias;
72 } g_part_alias_list[G_PART_ALIAS_COUNT] = {
73 	{ "apple-boot", G_PART_ALIAS_APPLE_BOOT },
74 	{ "apple-hfs", G_PART_ALIAS_APPLE_HFS },
75 	{ "apple-label", G_PART_ALIAS_APPLE_LABEL },
76 	{ "apple-raid", G_PART_ALIAS_APPLE_RAID },
77 	{ "apple-raid-offline", G_PART_ALIAS_APPLE_RAID_OFFLINE },
78 	{ "apple-tv-recovery", G_PART_ALIAS_APPLE_TV_RECOVERY },
79 	{ "apple-ufs", G_PART_ALIAS_APPLE_UFS },
80 	{ "bios-boot", G_PART_ALIAS_BIOS_BOOT },
81 	{ "ebr", G_PART_ALIAS_EBR },
82 	{ "efi", G_PART_ALIAS_EFI },
83 	{ "fat16", G_PART_ALIAS_MS_FAT16 },
84 	{ "fat32", G_PART_ALIAS_MS_FAT32 },
85 	{ "freebsd", G_PART_ALIAS_FREEBSD },
86 	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
87 	{ "freebsd-nandfs", G_PART_ALIAS_FREEBSD_NANDFS },
88 	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
89 	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
90 	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
91 	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
92 	{ "linux-data", G_PART_ALIAS_LINUX_DATA },
93 	{ "linux-lvm", G_PART_ALIAS_LINUX_LVM },
94 	{ "linux-raid", G_PART_ALIAS_LINUX_RAID },
95 	{ "linux-swap", G_PART_ALIAS_LINUX_SWAP },
96 	{ "mbr", G_PART_ALIAS_MBR },
97 	{ "ms-basic-data", G_PART_ALIAS_MS_BASIC_DATA },
98 	{ "ms-ldm-data", G_PART_ALIAS_MS_LDM_DATA },
99 	{ "ms-ldm-metadata", G_PART_ALIAS_MS_LDM_METADATA },
100 	{ "ms-reserved", G_PART_ALIAS_MS_RESERVED },
101 	{ "ntfs", G_PART_ALIAS_MS_NTFS },
102 	{ "netbsd-ccd", G_PART_ALIAS_NETBSD_CCD },
103 	{ "netbsd-cgd", G_PART_ALIAS_NETBSD_CGD },
104 	{ "netbsd-ffs", G_PART_ALIAS_NETBSD_FFS },
105 	{ "netbsd-lfs", G_PART_ALIAS_NETBSD_LFS },
106 	{ "netbsd-raid", G_PART_ALIAS_NETBSD_RAID },
107 	{ "netbsd-swap", G_PART_ALIAS_NETBSD_SWAP },
108 	{ "vmware-vmfs", G_PART_ALIAS_VMFS },
109 	{ "vmware-vmkdiag", G_PART_ALIAS_VMKDIAG },
110 	{ "vmware-reserved", G_PART_ALIAS_VMRESERVED },
111 };
112 
113 SYSCTL_DECL(_kern_geom);
114 SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
115     "GEOM_PART stuff");
116 static u_int check_integrity = 1;
117 TUNABLE_INT("kern.geom.part.check_integrity", &check_integrity);
118 SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity,
119     CTLFLAG_RW | CTLFLAG_TUN, &check_integrity, 1,
120     "Enable integrity checking");
121 
122 /*
123  * The GEOM partitioning class.
124  */
125 static g_ctl_req_t g_part_ctlreq;
126 static g_ctl_destroy_geom_t g_part_destroy_geom;
127 static g_fini_t g_part_fini;
128 static g_init_t g_part_init;
129 static g_taste_t g_part_taste;
130 
131 static g_access_t g_part_access;
132 static g_dumpconf_t g_part_dumpconf;
133 static g_orphan_t g_part_orphan;
134 static g_spoiled_t g_part_spoiled;
135 static g_start_t g_part_start;
136 
137 static struct g_class g_part_class = {
138 	.name = "PART",
139 	.version = G_VERSION,
140 	/* Class methods. */
141 	.ctlreq = g_part_ctlreq,
142 	.destroy_geom = g_part_destroy_geom,
143 	.fini = g_part_fini,
144 	.init = g_part_init,
145 	.taste = g_part_taste,
146 	/* Geom methods. */
147 	.access = g_part_access,
148 	.dumpconf = g_part_dumpconf,
149 	.orphan = g_part_orphan,
150 	.spoiled = g_part_spoiled,
151 	.start = g_part_start,
152 };
153 
154 DECLARE_GEOM_CLASS(g_part_class, g_part);
155 MODULE_VERSION(g_part, 0);
156 
157 /*
158  * Support functions.
159  */
160 
161 static void g_part_wither(struct g_geom *, int);
162 
163 const char *
164 g_part_alias_name(enum g_part_alias alias)
165 {
166 	int i;
167 
168 	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
169 		if (g_part_alias_list[i].alias != alias)
170 			continue;
171 		return (g_part_alias_list[i].lexeme);
172 	}
173 
174 	return (NULL);
175 }
176 
177 void
178 g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
179     u_int *bestheads)
180 {
181 	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
182 	off_t chs, cylinders;
183 	u_int heads;
184 	int idx;
185 
186 	*bestchs = 0;
187 	*bestheads = 0;
188 	for (idx = 0; candidate_heads[idx] != 0; idx++) {
189 		heads = candidate_heads[idx];
190 		cylinders = blocks / heads / sectors;
191 		if (cylinders < heads || cylinders < sectors)
192 			break;
193 		if (cylinders > 1023)
194 			continue;
195 		chs = cylinders * heads * sectors;
196 		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
197 			*bestchs = chs;
198 			*bestheads = heads;
199 		}
200 	}
201 }
202 
203 static void
204 g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
205     off_t blocks)
206 {
207 	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
208 	off_t chs, bestchs;
209 	u_int heads, sectors;
210 	int idx;
211 
212 	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
213 	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
214 		table->gpt_fixgeom = 0;
215 		table->gpt_heads = 0;
216 		table->gpt_sectors = 0;
217 		bestchs = 0;
218 		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
219 			sectors = candidate_sectors[idx];
220 			g_part_geometry_heads(blocks, sectors, &chs, &heads);
221 			if (chs == 0)
222 				continue;
223 			/*
224 			 * Prefer a geometry with sectors > 1, but only if
225 			 * it doesn't bump down the number of heads to 1.
226 			 */
227 			if (chs > bestchs || (chs == bestchs && heads > 1 &&
228 			    table->gpt_sectors == 1)) {
229 				bestchs = chs;
230 				table->gpt_heads = heads;
231 				table->gpt_sectors = sectors;
232 			}
233 		}
234 		/*
235 		 * If we didn't find a geometry at all, then the disk is
236 		 * too big. This means we can use the maximum number of
237 		 * heads and sectors.
238 		 */
239 		if (bestchs == 0) {
240 			table->gpt_heads = 255;
241 			table->gpt_sectors = 63;
242 		}
243 	} else {
244 		table->gpt_fixgeom = 1;
245 		table->gpt_heads = heads;
246 		table->gpt_sectors = sectors;
247 	}
248 }
249 
250 #define	DPRINTF(...)	if (bootverbose) {	\
251 	printf("GEOM_PART: " __VA_ARGS__);	\
252 }
253 
254 static int
255 g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
256 {
257 	struct g_part_entry *e1, *e2;
258 	struct g_provider *pp;
259 	off_t offset;
260 	int failed;
261 
262 	failed = 0;
263 	pp = cp->provider;
264 	if (table->gpt_last < table->gpt_first) {
265 		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
266 		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
267 		failed++;
268 	}
269 	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
270 		DPRINTF("last LBA extends beyond mediasize: "
271 		    "%jd > %jd\n", (intmax_t)table->gpt_last,
272 		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
273 		failed++;
274 	}
275 	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
276 		if (e1->gpe_deleted || e1->gpe_internal)
277 			continue;
278 		if (e1->gpe_start < table->gpt_first) {
279 			DPRINTF("partition %d has start offset below first "
280 			    "LBA: %jd < %jd\n", e1->gpe_index,
281 			    (intmax_t)e1->gpe_start,
282 			    (intmax_t)table->gpt_first);
283 			failed++;
284 		}
285 		if (e1->gpe_start > table->gpt_last) {
286 			DPRINTF("partition %d has start offset beyond last "
287 			    "LBA: %jd > %jd\n", e1->gpe_index,
288 			    (intmax_t)e1->gpe_start,
289 			    (intmax_t)table->gpt_last);
290 			failed++;
291 		}
292 		if (e1->gpe_end < e1->gpe_start) {
293 			DPRINTF("partition %d has end offset below start "
294 			    "offset: %jd < %jd\n", e1->gpe_index,
295 			    (intmax_t)e1->gpe_end,
296 			    (intmax_t)e1->gpe_start);
297 			failed++;
298 		}
299 		if (e1->gpe_end > table->gpt_last) {
300 			DPRINTF("partition %d has end offset beyond last "
301 			    "LBA: %jd > %jd\n", e1->gpe_index,
302 			    (intmax_t)e1->gpe_end,
303 			    (intmax_t)table->gpt_last);
304 			failed++;
305 		}
306 		if (pp->stripesize > 0) {
307 			offset = e1->gpe_start * pp->sectorsize;
308 			if (e1->gpe_offset > offset)
309 				offset = e1->gpe_offset;
310 			if ((offset + pp->stripeoffset) % pp->stripesize) {
311 				DPRINTF("partition %d is not aligned on %u "
312 				    "bytes\n", e1->gpe_index, pp->stripesize);
313 				/* Don't treat this as a critical failure */
314 			}
315 		}
316 		e2 = e1;
317 		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
318 			if (e2->gpe_deleted || e2->gpe_internal)
319 				continue;
320 			if (e1->gpe_start >= e2->gpe_start &&
321 			    e1->gpe_start <= e2->gpe_end) {
322 				DPRINTF("partition %d has start offset inside "
323 				    "partition %d: start[%d] %jd >= start[%d] "
324 				    "%jd <= end[%d] %jd\n",
325 				    e1->gpe_index, e2->gpe_index,
326 				    e2->gpe_index, (intmax_t)e2->gpe_start,
327 				    e1->gpe_index, (intmax_t)e1->gpe_start,
328 				    e2->gpe_index, (intmax_t)e2->gpe_end);
329 				failed++;
330 			}
331 			if (e1->gpe_end >= e2->gpe_start &&
332 			    e1->gpe_end <= e2->gpe_end) {
333 				DPRINTF("partition %d has end offset inside "
334 				    "partition %d: start[%d] %jd >= end[%d] "
335 				    "%jd <= end[%d] %jd\n",
336 				    e1->gpe_index, e2->gpe_index,
337 				    e2->gpe_index, (intmax_t)e2->gpe_start,
338 				    e1->gpe_index, (intmax_t)e1->gpe_end,
339 				    e2->gpe_index, (intmax_t)e2->gpe_end);
340 				failed++;
341 			}
342 			if (e1->gpe_start < e2->gpe_start &&
343 			    e1->gpe_end > e2->gpe_end) {
344 				DPRINTF("partition %d contains partition %d: "
345 				    "start[%d] %jd > start[%d] %jd, end[%d] "
346 				    "%jd < end[%d] %jd\n",
347 				    e1->gpe_index, e2->gpe_index,
348 				    e1->gpe_index, (intmax_t)e1->gpe_start,
349 				    e2->gpe_index, (intmax_t)e2->gpe_start,
350 				    e2->gpe_index, (intmax_t)e2->gpe_end,
351 				    e1->gpe_index, (intmax_t)e1->gpe_end);
352 				failed++;
353 			}
354 		}
355 	}
356 	if (failed != 0) {
357 		printf("GEOM_PART: integrity check failed (%s, %s)\n",
358 		    pp->name, table->gpt_scheme->name);
359 		if (check_integrity != 0)
360 			return (EINVAL);
361 		table->gpt_corrupt = 1;
362 	}
363 	return (0);
364 }
365 #undef	DPRINTF
366 
367 struct g_part_entry *
368 g_part_new_entry(struct g_part_table *table, int index, quad_t start,
369     quad_t end)
370 {
371 	struct g_part_entry *entry, *last;
372 
373 	last = NULL;
374 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
375 		if (entry->gpe_index == index)
376 			break;
377 		if (entry->gpe_index > index) {
378 			entry = NULL;
379 			break;
380 		}
381 		last = entry;
382 	}
383 	if (entry == NULL) {
384 		entry = g_malloc(table->gpt_scheme->gps_entrysz,
385 		    M_WAITOK | M_ZERO);
386 		entry->gpe_index = index;
387 		if (last == NULL)
388 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
389 		else
390 			LIST_INSERT_AFTER(last, entry, gpe_entry);
391 	} else
392 		entry->gpe_offset = 0;
393 	entry->gpe_start = start;
394 	entry->gpe_end = end;
395 	return (entry);
396 }
397 
398 static void
399 g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
400     struct g_part_entry *entry)
401 {
402 	struct g_consumer *cp;
403 	struct g_provider *pp;
404 	struct sbuf *sb;
405 	off_t offset;
406 
407 	cp = LIST_FIRST(&gp->consumer);
408 	pp = cp->provider;
409 
410 	offset = entry->gpe_start * pp->sectorsize;
411 	if (entry->gpe_offset < offset)
412 		entry->gpe_offset = offset;
413 
414 	if (entry->gpe_pp == NULL) {
415 		sb = sbuf_new_auto();
416 		G_PART_FULLNAME(table, entry, sb, gp->name);
417 		sbuf_finish(sb);
418 		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
419 		sbuf_delete(sb);
420 		entry->gpe_pp->private = entry;		/* Close the circle. */
421 	}
422 	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
423 	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
424 	    pp->sectorsize;
425 	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
426 	entry->gpe_pp->sectorsize = pp->sectorsize;
427 	entry->gpe_pp->stripesize = pp->stripesize;
428 	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
429 	if (pp->stripesize > 0)
430 		entry->gpe_pp->stripeoffset %= pp->stripesize;
431 	entry->gpe_pp->flags |= pp->flags & G_PF_ACCEPT_UNMAPPED;
432 	g_error_provider(entry->gpe_pp, 0);
433 }
434 
435 static struct g_geom*
436 g_part_find_geom(const char *name)
437 {
438 	struct g_geom *gp;
439 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
440 		if (!strcmp(name, gp->name))
441 			break;
442 	}
443 	return (gp);
444 }
445 
446 static int
447 g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
448 {
449 	struct g_geom *gp;
450 	const char *gname;
451 
452 	gname = gctl_get_asciiparam(req, name);
453 	if (gname == NULL)
454 		return (ENOATTR);
455 	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
456 		gname += sizeof(_PATH_DEV) - 1;
457 	gp = g_part_find_geom(gname);
458 	if (gp == NULL) {
459 		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
460 		return (EINVAL);
461 	}
462 	if ((gp->flags & G_GEOM_WITHER) != 0) {
463 		gctl_error(req, "%d %s", ENXIO, gname);
464 		return (ENXIO);
465 	}
466 	*v = gp;
467 	return (0);
468 }
469 
470 static int
471 g_part_parm_provider(struct gctl_req *req, const char *name,
472     struct g_provider **v)
473 {
474 	struct g_provider *pp;
475 	const char *pname;
476 
477 	pname = gctl_get_asciiparam(req, name);
478 	if (pname == NULL)
479 		return (ENOATTR);
480 	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
481 		pname += sizeof(_PATH_DEV) - 1;
482 	pp = g_provider_by_name(pname);
483 	if (pp == NULL) {
484 		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
485 		return (EINVAL);
486 	}
487 	*v = pp;
488 	return (0);
489 }
490 
491 static int
492 g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
493 {
494 	const char *p;
495 	char *x;
496 	quad_t q;
497 
498 	p = gctl_get_asciiparam(req, name);
499 	if (p == NULL)
500 		return (ENOATTR);
501 	q = strtoq(p, &x, 0);
502 	if (*x != '\0' || q < 0) {
503 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
504 		return (EINVAL);
505 	}
506 	*v = q;
507 	return (0);
508 }
509 
510 static int
511 g_part_parm_scheme(struct gctl_req *req, const char *name,
512     struct g_part_scheme **v)
513 {
514 	struct g_part_scheme *s;
515 	const char *p;
516 
517 	p = gctl_get_asciiparam(req, name);
518 	if (p == NULL)
519 		return (ENOATTR);
520 	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
521 		if (s == &g_part_null_scheme)
522 			continue;
523 		if (!strcasecmp(s->name, p))
524 			break;
525 	}
526 	if (s == NULL) {
527 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
528 		return (EINVAL);
529 	}
530 	*v = s;
531 	return (0);
532 }
533 
534 static int
535 g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
536 {
537 	const char *p;
538 
539 	p = gctl_get_asciiparam(req, name);
540 	if (p == NULL)
541 		return (ENOATTR);
542 	/* An empty label is always valid. */
543 	if (strcmp(name, "label") != 0 && p[0] == '\0') {
544 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
545 		return (EINVAL);
546 	}
547 	*v = p;
548 	return (0);
549 }
550 
551 static int
552 g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
553 {
554 	const intmax_t *p;
555 	int size;
556 
557 	p = gctl_get_param(req, name, &size);
558 	if (p == NULL)
559 		return (ENOATTR);
560 	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
561 		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
562 		return (EINVAL);
563 	}
564 	*v = (u_int)*p;
565 	return (0);
566 }
567 
568 static int
569 g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
570 {
571 	const uint32_t *p;
572 	int size;
573 
574 	p = gctl_get_param(req, name, &size);
575 	if (p == NULL)
576 		return (ENOATTR);
577 	if (size != sizeof(*p) || *p > INT_MAX) {
578 		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
579 		return (EINVAL);
580 	}
581 	*v = (u_int)*p;
582 	return (0);
583 }
584 
585 static int
586 g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
587     unsigned int *s)
588 {
589 	const void *p;
590 	int size;
591 
592 	p = gctl_get_param(req, name, &size);
593 	if (p == NULL)
594 		return (ENOATTR);
595 	*v = p;
596 	*s = size;
597 	return (0);
598 }
599 
600 static int
601 g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
602 {
603 	struct g_part_scheme *iter, *scheme;
604 	struct g_part_table *table;
605 	int pri, probe;
606 
607 	table = gp->softc;
608 	scheme = (table != NULL) ? table->gpt_scheme : NULL;
609 	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
610 	if (pri == 0)
611 		goto done;
612 	if (pri > 0) {	/* error */
613 		scheme = NULL;
614 		pri = INT_MIN;
615 	}
616 
617 	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
618 		if (iter == &g_part_null_scheme)
619 			continue;
620 		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
621 		    M_WAITOK);
622 		table->gpt_gp = gp;
623 		table->gpt_scheme = iter;
624 		table->gpt_depth = depth;
625 		probe = G_PART_PROBE(table, cp);
626 		if (probe <= 0 && probe > pri) {
627 			pri = probe;
628 			scheme = iter;
629 			if (gp->softc != NULL)
630 				kobj_delete((kobj_t)gp->softc, M_GEOM);
631 			gp->softc = table;
632 			if (pri == 0)
633 				goto done;
634 		} else
635 			kobj_delete((kobj_t)table, M_GEOM);
636 	}
637 
638 done:
639 	return ((scheme == NULL) ? ENXIO : 0);
640 }
641 
642 /*
643  * Control request functions.
644  */
645 
646 static int
647 g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
648 {
649 	struct g_geom *gp;
650 	struct g_provider *pp;
651 	struct g_part_entry *delent, *last, *entry;
652 	struct g_part_table *table;
653 	struct sbuf *sb;
654 	quad_t end;
655 	unsigned int index;
656 	int error;
657 
658 	gp = gpp->gpp_geom;
659 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
660 	g_topology_assert();
661 
662 	pp = LIST_FIRST(&gp->consumer)->provider;
663 	table = gp->softc;
664 	end = gpp->gpp_start + gpp->gpp_size - 1;
665 
666 	if (gpp->gpp_start < table->gpt_first ||
667 	    gpp->gpp_start > table->gpt_last) {
668 		gctl_error(req, "%d start '%jd'", EINVAL,
669 		    (intmax_t)gpp->gpp_start);
670 		return (EINVAL);
671 	}
672 	if (end < gpp->gpp_start || end > table->gpt_last) {
673 		gctl_error(req, "%d size '%jd'", EINVAL,
674 		    (intmax_t)gpp->gpp_size);
675 		return (EINVAL);
676 	}
677 	if (gpp->gpp_index > table->gpt_entries) {
678 		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
679 		return (EINVAL);
680 	}
681 
682 	delent = last = NULL;
683 	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
684 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
685 		if (entry->gpe_deleted) {
686 			if (entry->gpe_index == index)
687 				delent = entry;
688 			continue;
689 		}
690 		if (entry->gpe_index == index)
691 			index = entry->gpe_index + 1;
692 		if (entry->gpe_index < index)
693 			last = entry;
694 		if (entry->gpe_internal)
695 			continue;
696 		if (gpp->gpp_start >= entry->gpe_start &&
697 		    gpp->gpp_start <= entry->gpe_end) {
698 			gctl_error(req, "%d start '%jd'", ENOSPC,
699 			    (intmax_t)gpp->gpp_start);
700 			return (ENOSPC);
701 		}
702 		if (end >= entry->gpe_start && end <= entry->gpe_end) {
703 			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
704 			return (ENOSPC);
705 		}
706 		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
707 			gctl_error(req, "%d size '%jd'", ENOSPC,
708 			    (intmax_t)gpp->gpp_size);
709 			return (ENOSPC);
710 		}
711 	}
712 	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
713 		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
714 		return (EEXIST);
715 	}
716 	if (index > table->gpt_entries) {
717 		gctl_error(req, "%d index '%d'", ENOSPC, index);
718 		return (ENOSPC);
719 	}
720 
721 	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
722 	    M_WAITOK | M_ZERO) : delent;
723 	entry->gpe_index = index;
724 	entry->gpe_start = gpp->gpp_start;
725 	entry->gpe_end = end;
726 	error = G_PART_ADD(table, entry, gpp);
727 	if (error) {
728 		gctl_error(req, "%d", error);
729 		if (delent == NULL)
730 			g_free(entry);
731 		return (error);
732 	}
733 	if (delent == NULL) {
734 		if (last == NULL)
735 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
736 		else
737 			LIST_INSERT_AFTER(last, entry, gpe_entry);
738 		entry->gpe_created = 1;
739 	} else {
740 		entry->gpe_deleted = 0;
741 		entry->gpe_modified = 1;
742 	}
743 	g_part_new_provider(gp, table, entry);
744 
745 	/* Provide feedback if so requested. */
746 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
747 		sb = sbuf_new_auto();
748 		G_PART_FULLNAME(table, entry, sb, gp->name);
749 		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
750 			sbuf_printf(sb, " added, but partition is not "
751 			    "aligned on %u bytes\n", pp->stripesize);
752 		else
753 			sbuf_cat(sb, " added\n");
754 		sbuf_finish(sb);
755 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
756 		sbuf_delete(sb);
757 	}
758 	return (0);
759 }
760 
761 static int
762 g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
763 {
764 	struct g_geom *gp;
765 	struct g_part_table *table;
766 	struct sbuf *sb;
767 	int error, sz;
768 
769 	gp = gpp->gpp_geom;
770 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
771 	g_topology_assert();
772 
773 	table = gp->softc;
774 	sz = table->gpt_scheme->gps_bootcodesz;
775 	if (sz == 0) {
776 		error = ENODEV;
777 		goto fail;
778 	}
779 	if (gpp->gpp_codesize > sz) {
780 		error = EFBIG;
781 		goto fail;
782 	}
783 
784 	error = G_PART_BOOTCODE(table, gpp);
785 	if (error)
786 		goto fail;
787 
788 	/* Provide feedback if so requested. */
789 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
790 		sb = sbuf_new_auto();
791 		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
792 		sbuf_finish(sb);
793 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
794 		sbuf_delete(sb);
795 	}
796 	return (0);
797 
798  fail:
799 	gctl_error(req, "%d", error);
800 	return (error);
801 }
802 
803 static int
804 g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
805 {
806 	struct g_consumer *cp;
807 	struct g_geom *gp;
808 	struct g_provider *pp;
809 	struct g_part_entry *entry, *tmp;
810 	struct g_part_table *table;
811 	char *buf;
812 	int error, i;
813 
814 	gp = gpp->gpp_geom;
815 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
816 	g_topology_assert();
817 
818 	table = gp->softc;
819 	if (!table->gpt_opened) {
820 		gctl_error(req, "%d", EPERM);
821 		return (EPERM);
822 	}
823 
824 	g_topology_unlock();
825 
826 	cp = LIST_FIRST(&gp->consumer);
827 	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
828 		pp = cp->provider;
829 		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
830 		while (table->gpt_smhead != 0) {
831 			i = ffs(table->gpt_smhead) - 1;
832 			error = g_write_data(cp, i * pp->sectorsize, buf,
833 			    pp->sectorsize);
834 			if (error) {
835 				g_free(buf);
836 				goto fail;
837 			}
838 			table->gpt_smhead &= ~(1 << i);
839 		}
840 		while (table->gpt_smtail != 0) {
841 			i = ffs(table->gpt_smtail) - 1;
842 			error = g_write_data(cp, pp->mediasize - (i + 1) *
843 			    pp->sectorsize, buf, pp->sectorsize);
844 			if (error) {
845 				g_free(buf);
846 				goto fail;
847 			}
848 			table->gpt_smtail &= ~(1 << i);
849 		}
850 		g_free(buf);
851 	}
852 
853 	if (table->gpt_scheme == &g_part_null_scheme) {
854 		g_topology_lock();
855 		g_access(cp, -1, -1, -1);
856 		g_part_wither(gp, ENXIO);
857 		return (0);
858 	}
859 
860 	error = G_PART_WRITE(table, cp);
861 	if (error)
862 		goto fail;
863 
864 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
865 		if (!entry->gpe_deleted) {
866 			entry->gpe_created = 0;
867 			entry->gpe_modified = 0;
868 			continue;
869 		}
870 		LIST_REMOVE(entry, gpe_entry);
871 		g_free(entry);
872 	}
873 	table->gpt_created = 0;
874 	table->gpt_opened = 0;
875 
876 	g_topology_lock();
877 	g_access(cp, -1, -1, -1);
878 	return (0);
879 
880 fail:
881 	g_topology_lock();
882 	gctl_error(req, "%d", error);
883 	return (error);
884 }
885 
886 static int
887 g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
888 {
889 	struct g_consumer *cp;
890 	struct g_geom *gp;
891 	struct g_provider *pp;
892 	struct g_part_scheme *scheme;
893 	struct g_part_table *null, *table;
894 	struct sbuf *sb;
895 	int attr, error;
896 
897 	pp = gpp->gpp_provider;
898 	scheme = gpp->gpp_scheme;
899 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
900 	g_topology_assert();
901 
902 	/* Check that there isn't already a g_part geom on the provider. */
903 	gp = g_part_find_geom(pp->name);
904 	if (gp != NULL) {
905 		null = gp->softc;
906 		if (null->gpt_scheme != &g_part_null_scheme) {
907 			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
908 			return (EEXIST);
909 		}
910 	} else
911 		null = NULL;
912 
913 	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
914 	    (gpp->gpp_entries < scheme->gps_minent ||
915 	     gpp->gpp_entries > scheme->gps_maxent)) {
916 		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
917 		return (EINVAL);
918 	}
919 
920 	if (null == NULL)
921 		gp = g_new_geomf(&g_part_class, "%s", pp->name);
922 	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
923 	    M_WAITOK);
924 	table = gp->softc;
925 	table->gpt_gp = gp;
926 	table->gpt_scheme = gpp->gpp_scheme;
927 	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
928 	    gpp->gpp_entries : scheme->gps_minent;
929 	LIST_INIT(&table->gpt_entry);
930 	if (null == NULL) {
931 		cp = g_new_consumer(gp);
932 		error = g_attach(cp, pp);
933 		if (error == 0)
934 			error = g_access(cp, 1, 1, 1);
935 		if (error != 0) {
936 			g_part_wither(gp, error);
937 			gctl_error(req, "%d geom '%s'", error, pp->name);
938 			return (error);
939 		}
940 		table->gpt_opened = 1;
941 	} else {
942 		cp = LIST_FIRST(&gp->consumer);
943 		table->gpt_opened = null->gpt_opened;
944 		table->gpt_smhead = null->gpt_smhead;
945 		table->gpt_smtail = null->gpt_smtail;
946 	}
947 
948 	g_topology_unlock();
949 
950 	/* Make sure the provider has media. */
951 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
952 		error = ENODEV;
953 		goto fail;
954 	}
955 
956 	/* Make sure we can nest and if so, determine our depth. */
957 	error = g_getattr("PART::isleaf", cp, &attr);
958 	if (!error && attr) {
959 		error = ENODEV;
960 		goto fail;
961 	}
962 	error = g_getattr("PART::depth", cp, &attr);
963 	table->gpt_depth = (!error) ? attr + 1 : 0;
964 
965 	/*
966 	 * Synthesize a disk geometry. Some partitioning schemes
967 	 * depend on it and since some file systems need it even
968 	 * when the partitition scheme doesn't, we do it here in
969 	 * scheme-independent code.
970 	 */
971 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
972 
973 	error = G_PART_CREATE(table, gpp);
974 	if (error)
975 		goto fail;
976 
977 	g_topology_lock();
978 
979 	table->gpt_created = 1;
980 	if (null != NULL)
981 		kobj_delete((kobj_t)null, M_GEOM);
982 
983 	/*
984 	 * Support automatic commit by filling in the gpp_geom
985 	 * parameter.
986 	 */
987 	gpp->gpp_parms |= G_PART_PARM_GEOM;
988 	gpp->gpp_geom = gp;
989 
990 	/* Provide feedback if so requested. */
991 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
992 		sb = sbuf_new_auto();
993 		sbuf_printf(sb, "%s created\n", gp->name);
994 		sbuf_finish(sb);
995 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
996 		sbuf_delete(sb);
997 	}
998 	return (0);
999 
1000 fail:
1001 	g_topology_lock();
1002 	if (null == NULL) {
1003 		g_access(cp, -1, -1, -1);
1004 		g_part_wither(gp, error);
1005 	} else {
1006 		kobj_delete((kobj_t)gp->softc, M_GEOM);
1007 		gp->softc = null;
1008 	}
1009 	gctl_error(req, "%d provider", error);
1010 	return (error);
1011 }
1012 
1013 static int
1014 g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1015 {
1016 	struct g_geom *gp;
1017 	struct g_provider *pp;
1018 	struct g_part_entry *entry;
1019 	struct g_part_table *table;
1020 	struct sbuf *sb;
1021 
1022 	gp = gpp->gpp_geom;
1023 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1024 	g_topology_assert();
1025 
1026 	table = gp->softc;
1027 
1028 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1029 		if (entry->gpe_deleted || entry->gpe_internal)
1030 			continue;
1031 		if (entry->gpe_index == gpp->gpp_index)
1032 			break;
1033 	}
1034 	if (entry == NULL) {
1035 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1036 		return (ENOENT);
1037 	}
1038 
1039 	pp = entry->gpe_pp;
1040 	if (pp != NULL) {
1041 		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1042 			gctl_error(req, "%d", EBUSY);
1043 			return (EBUSY);
1044 		}
1045 
1046 		pp->private = NULL;
1047 		entry->gpe_pp = NULL;
1048 	}
1049 
1050 	if (pp != NULL)
1051 		g_wither_provider(pp, ENXIO);
1052 
1053 	/* Provide feedback if so requested. */
1054 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1055 		sb = sbuf_new_auto();
1056 		G_PART_FULLNAME(table, entry, sb, gp->name);
1057 		sbuf_cat(sb, " deleted\n");
1058 		sbuf_finish(sb);
1059 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1060 		sbuf_delete(sb);
1061 	}
1062 
1063 	if (entry->gpe_created) {
1064 		LIST_REMOVE(entry, gpe_entry);
1065 		g_free(entry);
1066 	} else {
1067 		entry->gpe_modified = 0;
1068 		entry->gpe_deleted = 1;
1069 	}
1070 	return (0);
1071 }
1072 
1073 static int
1074 g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1075 {
1076 	struct g_consumer *cp;
1077 	struct g_geom *gp;
1078 	struct g_provider *pp;
1079 	struct g_part_entry *entry, *tmp;
1080 	struct g_part_table *null, *table;
1081 	struct sbuf *sb;
1082 	int error;
1083 
1084 	gp = gpp->gpp_geom;
1085 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1086 	g_topology_assert();
1087 
1088 	table = gp->softc;
1089 	/* Check for busy providers. */
1090 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1091 		if (entry->gpe_deleted || entry->gpe_internal)
1092 			continue;
1093 		if (gpp->gpp_force) {
1094 			pp = entry->gpe_pp;
1095 			if (pp == NULL)
1096 				continue;
1097 			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1098 				continue;
1099 		}
1100 		gctl_error(req, "%d", EBUSY);
1101 		return (EBUSY);
1102 	}
1103 
1104 	if (gpp->gpp_force) {
1105 		/* Destroy all providers. */
1106 		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1107 			pp = entry->gpe_pp;
1108 			if (pp != NULL) {
1109 				pp->private = NULL;
1110 				g_wither_provider(pp, ENXIO);
1111 			}
1112 			LIST_REMOVE(entry, gpe_entry);
1113 			g_free(entry);
1114 		}
1115 	}
1116 
1117 	error = G_PART_DESTROY(table, gpp);
1118 	if (error) {
1119 		gctl_error(req, "%d", error);
1120 		return (error);
1121 	}
1122 
1123 	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1124 	    M_WAITOK);
1125 	null = gp->softc;
1126 	null->gpt_gp = gp;
1127 	null->gpt_scheme = &g_part_null_scheme;
1128 	LIST_INIT(&null->gpt_entry);
1129 
1130 	cp = LIST_FIRST(&gp->consumer);
1131 	pp = cp->provider;
1132 	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1133 
1134 	null->gpt_depth = table->gpt_depth;
1135 	null->gpt_opened = table->gpt_opened;
1136 	null->gpt_smhead = table->gpt_smhead;
1137 	null->gpt_smtail = table->gpt_smtail;
1138 
1139 	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1140 		LIST_REMOVE(entry, gpe_entry);
1141 		g_free(entry);
1142 	}
1143 	kobj_delete((kobj_t)table, M_GEOM);
1144 
1145 	/* Provide feedback if so requested. */
1146 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1147 		sb = sbuf_new_auto();
1148 		sbuf_printf(sb, "%s destroyed\n", gp->name);
1149 		sbuf_finish(sb);
1150 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1151 		sbuf_delete(sb);
1152 	}
1153 	return (0);
1154 }
1155 
1156 static int
1157 g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1158 {
1159 	struct g_geom *gp;
1160 	struct g_part_entry *entry;
1161 	struct g_part_table *table;
1162 	struct sbuf *sb;
1163 	int error;
1164 
1165 	gp = gpp->gpp_geom;
1166 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1167 	g_topology_assert();
1168 
1169 	table = gp->softc;
1170 
1171 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1172 		if (entry->gpe_deleted || entry->gpe_internal)
1173 			continue;
1174 		if (entry->gpe_index == gpp->gpp_index)
1175 			break;
1176 	}
1177 	if (entry == NULL) {
1178 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1179 		return (ENOENT);
1180 	}
1181 
1182 	error = G_PART_MODIFY(table, entry, gpp);
1183 	if (error) {
1184 		gctl_error(req, "%d", error);
1185 		return (error);
1186 	}
1187 
1188 	if (!entry->gpe_created)
1189 		entry->gpe_modified = 1;
1190 
1191 	/* Provide feedback if so requested. */
1192 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1193 		sb = sbuf_new_auto();
1194 		G_PART_FULLNAME(table, entry, sb, gp->name);
1195 		sbuf_cat(sb, " modified\n");
1196 		sbuf_finish(sb);
1197 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1198 		sbuf_delete(sb);
1199 	}
1200 	return (0);
1201 }
1202 
1203 static int
1204 g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1205 {
1206 	gctl_error(req, "%d verb 'move'", ENOSYS);
1207 	return (ENOSYS);
1208 }
1209 
1210 static int
1211 g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1212 {
1213 	struct g_part_table *table;
1214 	struct g_geom *gp;
1215 	struct sbuf *sb;
1216 	int error, recovered;
1217 
1218 	gp = gpp->gpp_geom;
1219 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1220 	g_topology_assert();
1221 	table = gp->softc;
1222 	error = recovered = 0;
1223 
1224 	if (table->gpt_corrupt) {
1225 		error = G_PART_RECOVER(table);
1226 		if (error == 0)
1227 			error = g_part_check_integrity(table,
1228 			    LIST_FIRST(&gp->consumer));
1229 		if (error) {
1230 			gctl_error(req, "%d recovering '%s' failed",
1231 			    error, gp->name);
1232 			return (error);
1233 		}
1234 		recovered = 1;
1235 	}
1236 	/* Provide feedback if so requested. */
1237 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1238 		sb = sbuf_new_auto();
1239 		if (recovered)
1240 			sbuf_printf(sb, "%s recovered\n", gp->name);
1241 		else
1242 			sbuf_printf(sb, "%s recovering is not needed\n",
1243 			    gp->name);
1244 		sbuf_finish(sb);
1245 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1246 		sbuf_delete(sb);
1247 	}
1248 	return (0);
1249 }
1250 
1251 static int
1252 g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1253 {
1254 	struct g_geom *gp;
1255 	struct g_provider *pp;
1256 	struct g_part_entry *pe, *entry;
1257 	struct g_part_table *table;
1258 	struct sbuf *sb;
1259 	quad_t end;
1260 	int error;
1261 	off_t mediasize;
1262 
1263 	gp = gpp->gpp_geom;
1264 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1265 	g_topology_assert();
1266 	table = gp->softc;
1267 
1268 	/* check gpp_index */
1269 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1270 		if (entry->gpe_deleted || entry->gpe_internal)
1271 			continue;
1272 		if (entry->gpe_index == gpp->gpp_index)
1273 			break;
1274 	}
1275 	if (entry == NULL) {
1276 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1277 		return (ENOENT);
1278 	}
1279 
1280 	/* check gpp_size */
1281 	end = entry->gpe_start + gpp->gpp_size - 1;
1282 	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1283 		gctl_error(req, "%d size '%jd'", EINVAL,
1284 		    (intmax_t)gpp->gpp_size);
1285 		return (EINVAL);
1286 	}
1287 
1288 	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1289 		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1290 			continue;
1291 		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1292 			gctl_error(req, "%d end '%jd'", ENOSPC,
1293 			    (intmax_t)end);
1294 			return (ENOSPC);
1295 		}
1296 		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1297 			gctl_error(req, "%d size '%jd'", ENOSPC,
1298 			    (intmax_t)gpp->gpp_size);
1299 			return (ENOSPC);
1300 		}
1301 	}
1302 
1303 	pp = entry->gpe_pp;
1304 	if ((g_debugflags & 16) == 0 &&
1305 	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1306 		if (entry->gpe_end - entry->gpe_start + 1 > gpp->gpp_size) {
1307 			/* Deny shrinking of an opened partition. */
1308 			gctl_error(req, "%d", EBUSY);
1309 			return (EBUSY);
1310 		}
1311 	}
1312 
1313 	error = G_PART_RESIZE(table, entry, gpp);
1314 	if (error) {
1315 		gctl_error(req, "%d", error);
1316 		return (error);
1317 	}
1318 
1319 	if (!entry->gpe_created)
1320 		entry->gpe_modified = 1;
1321 
1322 	/* update mediasize of changed provider */
1323 	mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1324 		pp->sectorsize;
1325 	g_resize_provider(pp, mediasize);
1326 
1327 	/* Provide feedback if so requested. */
1328 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1329 		sb = sbuf_new_auto();
1330 		G_PART_FULLNAME(table, entry, sb, gp->name);
1331 		sbuf_cat(sb, " resized\n");
1332 		sbuf_finish(sb);
1333 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1334 		sbuf_delete(sb);
1335 	}
1336 	return (0);
1337 }
1338 
1339 static int
1340 g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1341     unsigned int set)
1342 {
1343 	struct g_geom *gp;
1344 	struct g_part_entry *entry;
1345 	struct g_part_table *table;
1346 	struct sbuf *sb;
1347 	int error;
1348 
1349 	gp = gpp->gpp_geom;
1350 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1351 	g_topology_assert();
1352 
1353 	table = gp->softc;
1354 
1355 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1356 		if (entry->gpe_deleted || entry->gpe_internal)
1357 			continue;
1358 		if (entry->gpe_index == gpp->gpp_index)
1359 			break;
1360 	}
1361 	if (entry == NULL) {
1362 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1363 		return (ENOENT);
1364 	}
1365 
1366 	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1367 	if (error) {
1368 		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1369 		return (error);
1370 	}
1371 
1372 	/* Provide feedback if so requested. */
1373 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1374 		sb = sbuf_new_auto();
1375 		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1376 		    (set) ? "" : "un");
1377 		G_PART_FULLNAME(table, entry, sb, gp->name);
1378 		sbuf_printf(sb, "\n");
1379 		sbuf_finish(sb);
1380 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1381 		sbuf_delete(sb);
1382 	}
1383 	return (0);
1384 }
1385 
1386 static int
1387 g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1388 {
1389 	struct g_consumer *cp;
1390 	struct g_provider *pp;
1391 	struct g_geom *gp;
1392 	struct g_part_entry *entry, *tmp;
1393 	struct g_part_table *table;
1394 	int error, reprobe;
1395 
1396 	gp = gpp->gpp_geom;
1397 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1398 	g_topology_assert();
1399 
1400 	table = gp->softc;
1401 	if (!table->gpt_opened) {
1402 		gctl_error(req, "%d", EPERM);
1403 		return (EPERM);
1404 	}
1405 
1406 	cp = LIST_FIRST(&gp->consumer);
1407 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1408 		entry->gpe_modified = 0;
1409 		if (entry->gpe_created) {
1410 			pp = entry->gpe_pp;
1411 			if (pp != NULL) {
1412 				pp->private = NULL;
1413 				entry->gpe_pp = NULL;
1414 				g_wither_provider(pp, ENXIO);
1415 			}
1416 			entry->gpe_deleted = 1;
1417 		}
1418 		if (entry->gpe_deleted) {
1419 			LIST_REMOVE(entry, gpe_entry);
1420 			g_free(entry);
1421 		}
1422 	}
1423 
1424 	g_topology_unlock();
1425 
1426 	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1427 	    table->gpt_created) ? 1 : 0;
1428 
1429 	if (reprobe) {
1430 		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1431 			if (entry->gpe_internal)
1432 				continue;
1433 			error = EBUSY;
1434 			goto fail;
1435 		}
1436 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1437 			LIST_REMOVE(entry, gpe_entry);
1438 			g_free(entry);
1439 		}
1440 		error = g_part_probe(gp, cp, table->gpt_depth);
1441 		if (error) {
1442 			g_topology_lock();
1443 			g_access(cp, -1, -1, -1);
1444 			g_part_wither(gp, error);
1445 			return (0);
1446 		}
1447 		table = gp->softc;
1448 
1449 		/*
1450 		 * Synthesize a disk geometry. Some partitioning schemes
1451 		 * depend on it and since some file systems need it even
1452 		 * when the partitition scheme doesn't, we do it here in
1453 		 * scheme-independent code.
1454 		 */
1455 		pp = cp->provider;
1456 		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1457 	}
1458 
1459 	error = G_PART_READ(table, cp);
1460 	if (error)
1461 		goto fail;
1462 	error = g_part_check_integrity(table, cp);
1463 	if (error)
1464 		goto fail;
1465 
1466 	g_topology_lock();
1467 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1468 		if (!entry->gpe_internal)
1469 			g_part_new_provider(gp, table, entry);
1470 	}
1471 
1472 	table->gpt_opened = 0;
1473 	g_access(cp, -1, -1, -1);
1474 	return (0);
1475 
1476 fail:
1477 	g_topology_lock();
1478 	gctl_error(req, "%d", error);
1479 	return (error);
1480 }
1481 
1482 static void
1483 g_part_wither(struct g_geom *gp, int error)
1484 {
1485 	struct g_part_entry *entry;
1486 	struct g_part_table *table;
1487 
1488 	table = gp->softc;
1489 	if (table != NULL) {
1490 		G_PART_DESTROY(table, NULL);
1491 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1492 			LIST_REMOVE(entry, gpe_entry);
1493 			g_free(entry);
1494 		}
1495 		if (gp->softc != NULL) {
1496 			kobj_delete((kobj_t)gp->softc, M_GEOM);
1497 			gp->softc = NULL;
1498 		}
1499 	}
1500 	g_wither_geom(gp, error);
1501 }
1502 
1503 /*
1504  * Class methods.
1505  */
1506 
1507 static void
1508 g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1509 {
1510 	struct g_part_parms gpp;
1511 	struct g_part_table *table;
1512 	struct gctl_req_arg *ap;
1513 	enum g_part_ctl ctlreq;
1514 	unsigned int i, mparms, oparms, parm;
1515 	int auto_commit, close_on_error;
1516 	int error, modifies;
1517 
1518 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1519 	g_topology_assert();
1520 
1521 	ctlreq = G_PART_CTL_NONE;
1522 	modifies = 1;
1523 	mparms = 0;
1524 	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1525 	switch (*verb) {
1526 	case 'a':
1527 		if (!strcmp(verb, "add")) {
1528 			ctlreq = G_PART_CTL_ADD;
1529 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1530 			    G_PART_PARM_START | G_PART_PARM_TYPE;
1531 			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1532 		}
1533 		break;
1534 	case 'b':
1535 		if (!strcmp(verb, "bootcode")) {
1536 			ctlreq = G_PART_CTL_BOOTCODE;
1537 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1538 		}
1539 		break;
1540 	case 'c':
1541 		if (!strcmp(verb, "commit")) {
1542 			ctlreq = G_PART_CTL_COMMIT;
1543 			mparms |= G_PART_PARM_GEOM;
1544 			modifies = 0;
1545 		} else if (!strcmp(verb, "create")) {
1546 			ctlreq = G_PART_CTL_CREATE;
1547 			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1548 			oparms |= G_PART_PARM_ENTRIES;
1549 		}
1550 		break;
1551 	case 'd':
1552 		if (!strcmp(verb, "delete")) {
1553 			ctlreq = G_PART_CTL_DELETE;
1554 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1555 		} else if (!strcmp(verb, "destroy")) {
1556 			ctlreq = G_PART_CTL_DESTROY;
1557 			mparms |= G_PART_PARM_GEOM;
1558 			oparms |= G_PART_PARM_FORCE;
1559 		}
1560 		break;
1561 	case 'm':
1562 		if (!strcmp(verb, "modify")) {
1563 			ctlreq = G_PART_CTL_MODIFY;
1564 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1565 			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1566 		} else if (!strcmp(verb, "move")) {
1567 			ctlreq = G_PART_CTL_MOVE;
1568 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1569 		}
1570 		break;
1571 	case 'r':
1572 		if (!strcmp(verb, "recover")) {
1573 			ctlreq = G_PART_CTL_RECOVER;
1574 			mparms |= G_PART_PARM_GEOM;
1575 		} else if (!strcmp(verb, "resize")) {
1576 			ctlreq = G_PART_CTL_RESIZE;
1577 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1578 			    G_PART_PARM_SIZE;
1579 		}
1580 		break;
1581 	case 's':
1582 		if (!strcmp(verb, "set")) {
1583 			ctlreq = G_PART_CTL_SET;
1584 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1585 			    G_PART_PARM_INDEX;
1586 		}
1587 		break;
1588 	case 'u':
1589 		if (!strcmp(verb, "undo")) {
1590 			ctlreq = G_PART_CTL_UNDO;
1591 			mparms |= G_PART_PARM_GEOM;
1592 			modifies = 0;
1593 		} else if (!strcmp(verb, "unset")) {
1594 			ctlreq = G_PART_CTL_UNSET;
1595 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1596 			    G_PART_PARM_INDEX;
1597 		}
1598 		break;
1599 	}
1600 	if (ctlreq == G_PART_CTL_NONE) {
1601 		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1602 		return;
1603 	}
1604 
1605 	bzero(&gpp, sizeof(gpp));
1606 	for (i = 0; i < req->narg; i++) {
1607 		ap = &req->arg[i];
1608 		parm = 0;
1609 		switch (ap->name[0]) {
1610 		case 'a':
1611 			if (!strcmp(ap->name, "arg0")) {
1612 				parm = mparms &
1613 				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1614 			}
1615 			if (!strcmp(ap->name, "attrib"))
1616 				parm = G_PART_PARM_ATTRIB;
1617 			break;
1618 		case 'b':
1619 			if (!strcmp(ap->name, "bootcode"))
1620 				parm = G_PART_PARM_BOOTCODE;
1621 			break;
1622 		case 'c':
1623 			if (!strcmp(ap->name, "class"))
1624 				continue;
1625 			break;
1626 		case 'e':
1627 			if (!strcmp(ap->name, "entries"))
1628 				parm = G_PART_PARM_ENTRIES;
1629 			break;
1630 		case 'f':
1631 			if (!strcmp(ap->name, "flags"))
1632 				parm = G_PART_PARM_FLAGS;
1633 			else if (!strcmp(ap->name, "force"))
1634 				parm = G_PART_PARM_FORCE;
1635 			break;
1636 		case 'i':
1637 			if (!strcmp(ap->name, "index"))
1638 				parm = G_PART_PARM_INDEX;
1639 			break;
1640 		case 'l':
1641 			if (!strcmp(ap->name, "label"))
1642 				parm = G_PART_PARM_LABEL;
1643 			break;
1644 		case 'o':
1645 			if (!strcmp(ap->name, "output"))
1646 				parm = G_PART_PARM_OUTPUT;
1647 			break;
1648 		case 's':
1649 			if (!strcmp(ap->name, "scheme"))
1650 				parm = G_PART_PARM_SCHEME;
1651 			else if (!strcmp(ap->name, "size"))
1652 				parm = G_PART_PARM_SIZE;
1653 			else if (!strcmp(ap->name, "start"))
1654 				parm = G_PART_PARM_START;
1655 			break;
1656 		case 't':
1657 			if (!strcmp(ap->name, "type"))
1658 				parm = G_PART_PARM_TYPE;
1659 			break;
1660 		case 'v':
1661 			if (!strcmp(ap->name, "verb"))
1662 				continue;
1663 			else if (!strcmp(ap->name, "version"))
1664 				parm = G_PART_PARM_VERSION;
1665 			break;
1666 		}
1667 		if ((parm & (mparms | oparms)) == 0) {
1668 			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1669 			return;
1670 		}
1671 		switch (parm) {
1672 		case G_PART_PARM_ATTRIB:
1673 			error = g_part_parm_str(req, ap->name,
1674 			    &gpp.gpp_attrib);
1675 			break;
1676 		case G_PART_PARM_BOOTCODE:
1677 			error = g_part_parm_bootcode(req, ap->name,
1678 			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1679 			break;
1680 		case G_PART_PARM_ENTRIES:
1681 			error = g_part_parm_intmax(req, ap->name,
1682 			    &gpp.gpp_entries);
1683 			break;
1684 		case G_PART_PARM_FLAGS:
1685 			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1686 			break;
1687 		case G_PART_PARM_FORCE:
1688 			error = g_part_parm_uint32(req, ap->name,
1689 			    &gpp.gpp_force);
1690 			break;
1691 		case G_PART_PARM_GEOM:
1692 			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1693 			break;
1694 		case G_PART_PARM_INDEX:
1695 			error = g_part_parm_intmax(req, ap->name,
1696 			    &gpp.gpp_index);
1697 			break;
1698 		case G_PART_PARM_LABEL:
1699 			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1700 			break;
1701 		case G_PART_PARM_OUTPUT:
1702 			error = 0;	/* Write-only parameter */
1703 			break;
1704 		case G_PART_PARM_PROVIDER:
1705 			error = g_part_parm_provider(req, ap->name,
1706 			    &gpp.gpp_provider);
1707 			break;
1708 		case G_PART_PARM_SCHEME:
1709 			error = g_part_parm_scheme(req, ap->name,
1710 			    &gpp.gpp_scheme);
1711 			break;
1712 		case G_PART_PARM_SIZE:
1713 			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1714 			break;
1715 		case G_PART_PARM_START:
1716 			error = g_part_parm_quad(req, ap->name,
1717 			    &gpp.gpp_start);
1718 			break;
1719 		case G_PART_PARM_TYPE:
1720 			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1721 			break;
1722 		case G_PART_PARM_VERSION:
1723 			error = g_part_parm_uint32(req, ap->name,
1724 			    &gpp.gpp_version);
1725 			break;
1726 		default:
1727 			error = EDOOFUS;
1728 			gctl_error(req, "%d %s", error, ap->name);
1729 			break;
1730 		}
1731 		if (error != 0) {
1732 			if (error == ENOATTR) {
1733 				gctl_error(req, "%d param '%s'", error,
1734 				    ap->name);
1735 			}
1736 			return;
1737 		}
1738 		gpp.gpp_parms |= parm;
1739 	}
1740 	if ((gpp.gpp_parms & mparms) != mparms) {
1741 		parm = mparms - (gpp.gpp_parms & mparms);
1742 		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1743 		return;
1744 	}
1745 
1746 	/* Obtain permissions if possible/necessary. */
1747 	close_on_error = 0;
1748 	table = NULL;
1749 	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1750 		table = gpp.gpp_geom->softc;
1751 		if (table != NULL && table->gpt_corrupt &&
1752 		    ctlreq != G_PART_CTL_DESTROY &&
1753 		    ctlreq != G_PART_CTL_RECOVER) {
1754 			gctl_error(req, "%d table '%s' is corrupt",
1755 			    EPERM, gpp.gpp_geom->name);
1756 			return;
1757 		}
1758 		if (table != NULL && !table->gpt_opened) {
1759 			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1760 			    1, 1, 1);
1761 			if (error) {
1762 				gctl_error(req, "%d geom '%s'", error,
1763 				    gpp.gpp_geom->name);
1764 				return;
1765 			}
1766 			table->gpt_opened = 1;
1767 			close_on_error = 1;
1768 		}
1769 	}
1770 
1771 	/* Allow the scheme to check or modify the parameters. */
1772 	if (table != NULL) {
1773 		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1774 		if (error) {
1775 			gctl_error(req, "%d pre-check failed", error);
1776 			goto out;
1777 		}
1778 	} else
1779 		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1780 
1781 	switch (ctlreq) {
1782 	case G_PART_CTL_NONE:
1783 		panic("%s", __func__);
1784 	case G_PART_CTL_ADD:
1785 		error = g_part_ctl_add(req, &gpp);
1786 		break;
1787 	case G_PART_CTL_BOOTCODE:
1788 		error = g_part_ctl_bootcode(req, &gpp);
1789 		break;
1790 	case G_PART_CTL_COMMIT:
1791 		error = g_part_ctl_commit(req, &gpp);
1792 		break;
1793 	case G_PART_CTL_CREATE:
1794 		error = g_part_ctl_create(req, &gpp);
1795 		break;
1796 	case G_PART_CTL_DELETE:
1797 		error = g_part_ctl_delete(req, &gpp);
1798 		break;
1799 	case G_PART_CTL_DESTROY:
1800 		error = g_part_ctl_destroy(req, &gpp);
1801 		break;
1802 	case G_PART_CTL_MODIFY:
1803 		error = g_part_ctl_modify(req, &gpp);
1804 		break;
1805 	case G_PART_CTL_MOVE:
1806 		error = g_part_ctl_move(req, &gpp);
1807 		break;
1808 	case G_PART_CTL_RECOVER:
1809 		error = g_part_ctl_recover(req, &gpp);
1810 		break;
1811 	case G_PART_CTL_RESIZE:
1812 		error = g_part_ctl_resize(req, &gpp);
1813 		break;
1814 	case G_PART_CTL_SET:
1815 		error = g_part_ctl_setunset(req, &gpp, 1);
1816 		break;
1817 	case G_PART_CTL_UNDO:
1818 		error = g_part_ctl_undo(req, &gpp);
1819 		break;
1820 	case G_PART_CTL_UNSET:
1821 		error = g_part_ctl_setunset(req, &gpp, 0);
1822 		break;
1823 	}
1824 
1825 	/* Implement automatic commit. */
1826 	if (!error) {
1827 		auto_commit = (modifies &&
1828 		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1829 		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1830 		if (auto_commit) {
1831 			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1832 			    __func__));
1833 			error = g_part_ctl_commit(req, &gpp);
1834 		}
1835 	}
1836 
1837  out:
1838 	if (error && close_on_error) {
1839 		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1840 		table->gpt_opened = 0;
1841 	}
1842 }
1843 
1844 static int
1845 g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1846     struct g_geom *gp)
1847 {
1848 
1849 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1850 	g_topology_assert();
1851 
1852 	g_part_wither(gp, EINVAL);
1853 	return (0);
1854 }
1855 
1856 static struct g_geom *
1857 g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1858 {
1859 	struct g_consumer *cp;
1860 	struct g_geom *gp;
1861 	struct g_part_entry *entry;
1862 	struct g_part_table *table;
1863 	struct root_hold_token *rht;
1864 	int attr, depth;
1865 	int error;
1866 
1867 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1868 	g_topology_assert();
1869 
1870 	/* Skip providers that are already open for writing. */
1871 	if (pp->acw > 0)
1872 		return (NULL);
1873 
1874 	/*
1875 	 * Create a GEOM with consumer and hook it up to the provider.
1876 	 * With that we become part of the topology. Optain read access
1877 	 * to the provider.
1878 	 */
1879 	gp = g_new_geomf(mp, "%s", pp->name);
1880 	cp = g_new_consumer(gp);
1881 	error = g_attach(cp, pp);
1882 	if (error == 0)
1883 		error = g_access(cp, 1, 0, 0);
1884 	if (error != 0) {
1885 		if (cp->provider)
1886 			g_detach(cp);
1887 		g_destroy_consumer(cp);
1888 		g_destroy_geom(gp);
1889 		return (NULL);
1890 	}
1891 
1892 	rht = root_mount_hold(mp->name);
1893 	g_topology_unlock();
1894 
1895 	/*
1896 	 * Short-circuit the whole probing galore when there's no
1897 	 * media present.
1898 	 */
1899 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1900 		error = ENODEV;
1901 		goto fail;
1902 	}
1903 
1904 	/* Make sure we can nest and if so, determine our depth. */
1905 	error = g_getattr("PART::isleaf", cp, &attr);
1906 	if (!error && attr) {
1907 		error = ENODEV;
1908 		goto fail;
1909 	}
1910 	error = g_getattr("PART::depth", cp, &attr);
1911 	depth = (!error) ? attr + 1 : 0;
1912 
1913 	error = g_part_probe(gp, cp, depth);
1914 	if (error)
1915 		goto fail;
1916 
1917 	table = gp->softc;
1918 
1919 	/*
1920 	 * Synthesize a disk geometry. Some partitioning schemes
1921 	 * depend on it and since some file systems need it even
1922 	 * when the partitition scheme doesn't, we do it here in
1923 	 * scheme-independent code.
1924 	 */
1925 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1926 
1927 	error = G_PART_READ(table, cp);
1928 	if (error)
1929 		goto fail;
1930 	error = g_part_check_integrity(table, cp);
1931 	if (error)
1932 		goto fail;
1933 
1934 	g_topology_lock();
1935 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1936 		if (!entry->gpe_internal)
1937 			g_part_new_provider(gp, table, entry);
1938 	}
1939 
1940 	root_mount_rel(rht);
1941 	g_access(cp, -1, 0, 0);
1942 	return (gp);
1943 
1944  fail:
1945 	g_topology_lock();
1946 	root_mount_rel(rht);
1947 	g_access(cp, -1, 0, 0);
1948 	g_detach(cp);
1949 	g_destroy_consumer(cp);
1950 	g_destroy_geom(gp);
1951 	return (NULL);
1952 }
1953 
1954 /*
1955  * Geom methods.
1956  */
1957 
1958 static int
1959 g_part_access(struct g_provider *pp, int dr, int dw, int de)
1960 {
1961 	struct g_consumer *cp;
1962 
1963 	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1964 	    dw, de));
1965 
1966 	cp = LIST_FIRST(&pp->geom->consumer);
1967 
1968 	/* We always gain write-exclusive access. */
1969 	return (g_access(cp, dr, dw, dw + de));
1970 }
1971 
1972 static void
1973 g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1974     struct g_consumer *cp, struct g_provider *pp)
1975 {
1976 	char buf[64];
1977 	struct g_part_entry *entry;
1978 	struct g_part_table *table;
1979 
1980 	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
1981 	table = gp->softc;
1982 
1983 	if (indent == NULL) {
1984 		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
1985 		entry = pp->private;
1986 		if (entry == NULL)
1987 			return;
1988 		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
1989 		    (uintmax_t)entry->gpe_offset,
1990 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1991 		/*
1992 		 * libdisk compatibility quirk - the scheme dumps the
1993 		 * slicer name and partition type in a way that is
1994 		 * compatible with libdisk. When libdisk is not used
1995 		 * anymore, this should go away.
1996 		 */
1997 		G_PART_DUMPCONF(table, entry, sb, indent);
1998 	} else if (cp != NULL) {	/* Consumer configuration. */
1999 		KASSERT(pp == NULL, ("%s", __func__));
2000 		/* none */
2001 	} else if (pp != NULL) {	/* Provider configuration. */
2002 		entry = pp->private;
2003 		if (entry == NULL)
2004 			return;
2005 		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
2006 		    (uintmax_t)entry->gpe_start);
2007 		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
2008 		    (uintmax_t)entry->gpe_end);
2009 		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
2010 		    entry->gpe_index);
2011 		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2012 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2013 		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
2014 		    (uintmax_t)entry->gpe_offset);
2015 		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
2016 		    (uintmax_t)pp->mediasize);
2017 		G_PART_DUMPCONF(table, entry, sb, indent);
2018 	} else {			/* Geom configuration. */
2019 		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2020 		    table->gpt_scheme->name);
2021 		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2022 		    table->gpt_entries);
2023 		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2024 		    (uintmax_t)table->gpt_first);
2025 		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2026 		    (uintmax_t)table->gpt_last);
2027 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2028 		    table->gpt_sectors);
2029 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2030 		    table->gpt_heads);
2031 		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2032 		    table->gpt_corrupt ? "CORRUPT": "OK");
2033 		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2034 		    table->gpt_opened ? "true": "false");
2035 		G_PART_DUMPCONF(table, NULL, sb, indent);
2036 	}
2037 }
2038 
2039 static void
2040 g_part_orphan(struct g_consumer *cp)
2041 {
2042 	struct g_provider *pp;
2043 	struct g_part_table *table;
2044 
2045 	pp = cp->provider;
2046 	KASSERT(pp != NULL, ("%s", __func__));
2047 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2048 	g_topology_assert();
2049 
2050 	KASSERT(pp->error != 0, ("%s", __func__));
2051 	table = cp->geom->softc;
2052 	if (table != NULL && table->gpt_opened)
2053 		g_access(cp, -1, -1, -1);
2054 	g_part_wither(cp->geom, pp->error);
2055 }
2056 
2057 static void
2058 g_part_spoiled(struct g_consumer *cp)
2059 {
2060 
2061 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2062 	g_topology_assert();
2063 
2064 	cp->flags |= G_CF_ORPHAN;
2065 	g_part_wither(cp->geom, ENXIO);
2066 }
2067 
2068 static void
2069 g_part_start(struct bio *bp)
2070 {
2071 	struct bio *bp2;
2072 	struct g_consumer *cp;
2073 	struct g_geom *gp;
2074 	struct g_part_entry *entry;
2075 	struct g_part_table *table;
2076 	struct g_kerneldump *gkd;
2077 	struct g_provider *pp;
2078 	char buf[64];
2079 
2080 	pp = bp->bio_to;
2081 	gp = pp->geom;
2082 	table = gp->softc;
2083 	cp = LIST_FIRST(&gp->consumer);
2084 
2085 	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2086 	    pp->name));
2087 
2088 	entry = pp->private;
2089 	if (entry == NULL) {
2090 		g_io_deliver(bp, ENXIO);
2091 		return;
2092 	}
2093 
2094 	switch(bp->bio_cmd) {
2095 	case BIO_DELETE:
2096 	case BIO_READ:
2097 	case BIO_WRITE:
2098 		if (bp->bio_offset >= pp->mediasize) {
2099 			g_io_deliver(bp, EIO);
2100 			return;
2101 		}
2102 		bp2 = g_clone_bio(bp);
2103 		if (bp2 == NULL) {
2104 			g_io_deliver(bp, ENOMEM);
2105 			return;
2106 		}
2107 		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2108 			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2109 		bp2->bio_done = g_std_done;
2110 		bp2->bio_offset += entry->gpe_offset;
2111 		g_io_request(bp2, cp);
2112 		return;
2113 	case BIO_FLUSH:
2114 		break;
2115 	case BIO_GETATTR:
2116 		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2117 			return;
2118 		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2119 			return;
2120 		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2121 			return;
2122 		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2123 			return;
2124 		if (g_handleattr_str(bp, "PART::scheme",
2125 		    table->gpt_scheme->name))
2126 			return;
2127 		if (g_handleattr_str(bp, "PART::type",
2128 		    G_PART_TYPE(table, entry, buf, sizeof(buf))))
2129 			return;
2130 		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2131 			/*
2132 			 * Check that the partition is suitable for kernel
2133 			 * dumps. Typically only swap partitions should be
2134 			 * used. If the request comes from the nested scheme
2135 			 * we allow dumping there as well.
2136 			 */
2137 			if ((bp->bio_from == NULL ||
2138 			    bp->bio_from->geom->class != &g_part_class) &&
2139 			    G_PART_DUMPTO(table, entry) == 0) {
2140 				g_io_deliver(bp, ENODEV);
2141 				printf("GEOM_PART: Partition '%s' not suitable"
2142 				    " for kernel dumps (wrong type?)\n",
2143 				    pp->name);
2144 				return;
2145 			}
2146 			gkd = (struct g_kerneldump *)bp->bio_data;
2147 			if (gkd->offset >= pp->mediasize) {
2148 				g_io_deliver(bp, EIO);
2149 				return;
2150 			}
2151 			if (gkd->offset + gkd->length > pp->mediasize)
2152 				gkd->length = pp->mediasize - gkd->offset;
2153 			gkd->offset += entry->gpe_offset;
2154 		}
2155 		break;
2156 	default:
2157 		g_io_deliver(bp, EOPNOTSUPP);
2158 		return;
2159 	}
2160 
2161 	bp2 = g_clone_bio(bp);
2162 	if (bp2 == NULL) {
2163 		g_io_deliver(bp, ENOMEM);
2164 		return;
2165 	}
2166 	bp2->bio_done = g_std_done;
2167 	g_io_request(bp2, cp);
2168 }
2169 
2170 static void
2171 g_part_init(struct g_class *mp)
2172 {
2173 
2174 	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2175 }
2176 
2177 static void
2178 g_part_fini(struct g_class *mp)
2179 {
2180 
2181 	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2182 }
2183 
2184 static void
2185 g_part_unload_event(void *arg, int flag)
2186 {
2187 	struct g_consumer *cp;
2188 	struct g_geom *gp;
2189 	struct g_provider *pp;
2190 	struct g_part_scheme *scheme;
2191 	struct g_part_table *table;
2192 	uintptr_t *xchg;
2193 	int acc, error;
2194 
2195 	if (flag == EV_CANCEL)
2196 		return;
2197 
2198 	xchg = arg;
2199 	error = 0;
2200 	scheme = (void *)(*xchg);
2201 
2202 	g_topology_assert();
2203 
2204 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2205 		table = gp->softc;
2206 		if (table->gpt_scheme != scheme)
2207 			continue;
2208 
2209 		acc = 0;
2210 		LIST_FOREACH(pp, &gp->provider, provider)
2211 			acc += pp->acr + pp->acw + pp->ace;
2212 		LIST_FOREACH(cp, &gp->consumer, consumer)
2213 			acc += cp->acr + cp->acw + cp->ace;
2214 
2215 		if (!acc)
2216 			g_part_wither(gp, ENOSYS);
2217 		else
2218 			error = EBUSY;
2219 	}
2220 
2221 	if (!error)
2222 		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2223 
2224 	*xchg = error;
2225 }
2226 
2227 int
2228 g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2229 {
2230 	struct g_part_scheme *iter;
2231 	uintptr_t arg;
2232 	int error;
2233 
2234 	error = 0;
2235 	switch (type) {
2236 	case MOD_LOAD:
2237 		TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
2238 			if (scheme == iter) {
2239 				printf("GEOM_PART: scheme %s is already "
2240 				    "registered!\n", scheme->name);
2241 				break;
2242 			}
2243 		}
2244 		if (iter == NULL) {
2245 			TAILQ_INSERT_TAIL(&g_part_schemes, scheme,
2246 			    scheme_list);
2247 			g_retaste(&g_part_class);
2248 		}
2249 		break;
2250 	case MOD_UNLOAD:
2251 		arg = (uintptr_t)scheme;
2252 		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2253 		    NULL);
2254 		if (error == 0)
2255 			error = arg;
2256 		break;
2257 	default:
2258 		error = EOPNOTSUPP;
2259 		break;
2260 	}
2261 
2262 	return (error);
2263 }
2264