xref: /freebsd/sys/geom/part/g_part.c (revision 586f63035fbe5e45cfc971037fd76375661ece26)
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 	{ "fat32", G_PART_ALIAS_MS_FAT32 },
84 	{ "freebsd", G_PART_ALIAS_FREEBSD },
85 	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
86 	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
87 	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
88 	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
89 	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
90 	{ "linux-data", G_PART_ALIAS_LINUX_DATA },
91 	{ "linux-lvm", G_PART_ALIAS_LINUX_LVM },
92 	{ "linux-raid", G_PART_ALIAS_LINUX_RAID },
93 	{ "linux-swap", G_PART_ALIAS_LINUX_SWAP },
94 	{ "mbr", G_PART_ALIAS_MBR },
95 	{ "ms-basic-data", G_PART_ALIAS_MS_BASIC_DATA },
96 	{ "ms-ldm-data", G_PART_ALIAS_MS_LDM_DATA },
97 	{ "ms-ldm-metadata", G_PART_ALIAS_MS_LDM_METADATA },
98 	{ "ms-reserved", G_PART_ALIAS_MS_RESERVED },
99 	{ "ntfs", G_PART_ALIAS_MS_NTFS },
100 	{ "netbsd-ccd", G_PART_ALIAS_NETBSD_CCD },
101 	{ "netbsd-cgd", G_PART_ALIAS_NETBSD_CGD },
102 	{ "netbsd-ffs", G_PART_ALIAS_NETBSD_FFS },
103 	{ "netbsd-lfs", G_PART_ALIAS_NETBSD_LFS },
104 	{ "netbsd-raid", G_PART_ALIAS_NETBSD_RAID },
105 	{ "netbsd-swap", G_PART_ALIAS_NETBSD_SWAP },
106 };
107 
108 SYSCTL_DECL(_kern_geom);
109 static SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
110     "GEOM_PART stuff");
111 static u_int check_integrity = 1;
112 TUNABLE_INT("kern.geom.part.check_integrity", &check_integrity);
113 SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity, CTLFLAG_RW,
114     &check_integrity, 1, "Enable integrity checking");
115 
116 /*
117  * The GEOM partitioning class.
118  */
119 static g_ctl_req_t g_part_ctlreq;
120 static g_ctl_destroy_geom_t g_part_destroy_geom;
121 static g_fini_t g_part_fini;
122 static g_init_t g_part_init;
123 static g_taste_t g_part_taste;
124 
125 static g_access_t g_part_access;
126 static g_dumpconf_t g_part_dumpconf;
127 static g_orphan_t g_part_orphan;
128 static g_spoiled_t g_part_spoiled;
129 static g_start_t g_part_start;
130 
131 static struct g_class g_part_class = {
132 	.name = "PART",
133 	.version = G_VERSION,
134 	/* Class methods. */
135 	.ctlreq = g_part_ctlreq,
136 	.destroy_geom = g_part_destroy_geom,
137 	.fini = g_part_fini,
138 	.init = g_part_init,
139 	.taste = g_part_taste,
140 	/* Geom methods. */
141 	.access = g_part_access,
142 	.dumpconf = g_part_dumpconf,
143 	.orphan = g_part_orphan,
144 	.spoiled = g_part_spoiled,
145 	.start = g_part_start,
146 };
147 
148 DECLARE_GEOM_CLASS(g_part_class, g_part);
149 
150 /*
151  * Support functions.
152  */
153 
154 static void g_part_wither(struct g_geom *, int);
155 
156 const char *
157 g_part_alias_name(enum g_part_alias alias)
158 {
159 	int i;
160 
161 	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
162 		if (g_part_alias_list[i].alias != alias)
163 			continue;
164 		return (g_part_alias_list[i].lexeme);
165 	}
166 
167 	return (NULL);
168 }
169 
170 void
171 g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
172     u_int *bestheads)
173 {
174 	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
175 	off_t chs, cylinders;
176 	u_int heads;
177 	int idx;
178 
179 	*bestchs = 0;
180 	*bestheads = 0;
181 	for (idx = 0; candidate_heads[idx] != 0; idx++) {
182 		heads = candidate_heads[idx];
183 		cylinders = blocks / heads / sectors;
184 		if (cylinders < heads || cylinders < sectors)
185 			break;
186 		if (cylinders > 1023)
187 			continue;
188 		chs = cylinders * heads * sectors;
189 		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
190 			*bestchs = chs;
191 			*bestheads = heads;
192 		}
193 	}
194 }
195 
196 static void
197 g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
198     off_t blocks)
199 {
200 	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
201 	off_t chs, bestchs;
202 	u_int heads, sectors;
203 	int idx;
204 
205 	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
206 	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
207 		table->gpt_fixgeom = 0;
208 		table->gpt_heads = 0;
209 		table->gpt_sectors = 0;
210 		bestchs = 0;
211 		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
212 			sectors = candidate_sectors[idx];
213 			g_part_geometry_heads(blocks, sectors, &chs, &heads);
214 			if (chs == 0)
215 				continue;
216 			/*
217 			 * Prefer a geometry with sectors > 1, but only if
218 			 * it doesn't bump down the numbver of heads to 1.
219 			 */
220 			if (chs > bestchs || (chs == bestchs && heads > 1 &&
221 			    table->gpt_sectors == 1)) {
222 				bestchs = chs;
223 				table->gpt_heads = heads;
224 				table->gpt_sectors = sectors;
225 			}
226 		}
227 		/*
228 		 * If we didn't find a geometry at all, then the disk is
229 		 * too big. This means we can use the maximum number of
230 		 * heads and sectors.
231 		 */
232 		if (bestchs == 0) {
233 			table->gpt_heads = 255;
234 			table->gpt_sectors = 63;
235 		}
236 	} else {
237 		table->gpt_fixgeom = 1;
238 		table->gpt_heads = heads;
239 		table->gpt_sectors = sectors;
240 	}
241 }
242 
243 #define	DPRINTF(...)	if (bootverbose) {	\
244 	printf("GEOM_PART: " __VA_ARGS__);	\
245 }
246 
247 static int
248 g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
249 {
250 	struct g_part_entry *e1, *e2;
251 	struct g_provider *pp;
252 	off_t offset;
253 	int failed;
254 
255 	failed = 0;
256 	pp = cp->provider;
257 	if (table->gpt_last < table->gpt_first) {
258 		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
259 		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
260 		failed++;
261 	}
262 	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
263 		DPRINTF("last LBA extends beyond mediasize: "
264 		    "%jd > %jd\n", (intmax_t)table->gpt_last,
265 		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
266 		failed++;
267 	}
268 	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
269 		if (e1->gpe_deleted || e1->gpe_internal)
270 			continue;
271 		if (e1->gpe_start < table->gpt_first) {
272 			DPRINTF("partition %d has start offset below first "
273 			    "LBA: %jd < %jd\n", e1->gpe_index,
274 			    (intmax_t)e1->gpe_start,
275 			    (intmax_t)table->gpt_first);
276 			failed++;
277 		}
278 		if (e1->gpe_start > table->gpt_last) {
279 			DPRINTF("partition %d has start offset beyond last "
280 			    "LBA: %jd > %jd\n", e1->gpe_index,
281 			    (intmax_t)e1->gpe_start,
282 			    (intmax_t)table->gpt_last);
283 			failed++;
284 		}
285 		if (e1->gpe_end < e1->gpe_start) {
286 			DPRINTF("partition %d has end offset below start "
287 			    "offset: %jd < %jd\n", e1->gpe_index,
288 			    (intmax_t)e1->gpe_end,
289 			    (intmax_t)e1->gpe_start);
290 			failed++;
291 		}
292 		if (e1->gpe_end > table->gpt_last) {
293 			DPRINTF("partition %d has end offset beyond last "
294 			    "LBA: %jd > %jd\n", e1->gpe_index,
295 			    (intmax_t)e1->gpe_end,
296 			    (intmax_t)table->gpt_last);
297 			failed++;
298 		}
299 		if (pp->stripesize > 0) {
300 			offset = e1->gpe_start * pp->sectorsize;
301 			if (e1->gpe_offset > offset)
302 				offset = e1->gpe_offset;
303 			if ((offset + pp->stripeoffset) % pp->stripesize) {
304 				DPRINTF("partition %d is not aligned on %u "
305 				    "bytes\n", e1->gpe_index, pp->stripesize);
306 				/* Don't treat this as a critical failure */
307 			}
308 		}
309 		e2 = e1;
310 		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
311 			if (e2->gpe_deleted || e2->gpe_internal)
312 				continue;
313 			if (e1->gpe_start >= e2->gpe_start &&
314 			    e1->gpe_start <= e2->gpe_end) {
315 				DPRINTF("partition %d has start offset inside "
316 				    "partition %d: start[%d] %jd >= start[%d] "
317 				    "%jd <= end[%d] %jd\n",
318 				    e1->gpe_index, e2->gpe_index,
319 				    e2->gpe_index, (intmax_t)e2->gpe_start,
320 				    e1->gpe_index, (intmax_t)e1->gpe_start,
321 				    e2->gpe_index, (intmax_t)e2->gpe_end);
322 				failed++;
323 			}
324 			if (e1->gpe_end >= e2->gpe_start &&
325 			    e1->gpe_end <= e2->gpe_end) {
326 				DPRINTF("partition %d has end offset inside "
327 				    "partition %d: start[%d] %jd >= end[%d] "
328 				    "%jd <= end[%d] %jd\n",
329 				    e1->gpe_index, e2->gpe_index,
330 				    e2->gpe_index, (intmax_t)e2->gpe_start,
331 				    e1->gpe_index, (intmax_t)e1->gpe_end,
332 				    e2->gpe_index, (intmax_t)e2->gpe_end);
333 				failed++;
334 			}
335 			if (e1->gpe_start < e2->gpe_start &&
336 			    e1->gpe_end > e2->gpe_end) {
337 				DPRINTF("partition %d contains partition %d: "
338 				    "start[%d] %jd > start[%d] %jd, end[%d] "
339 				    "%jd < end[%d] %jd\n",
340 				    e1->gpe_index, e2->gpe_index,
341 				    e1->gpe_index, (intmax_t)e1->gpe_start,
342 				    e2->gpe_index, (intmax_t)e2->gpe_start,
343 				    e2->gpe_index, (intmax_t)e2->gpe_end,
344 				    e1->gpe_index, (intmax_t)e1->gpe_end);
345 				failed++;
346 			}
347 		}
348 	}
349 	if (failed != 0) {
350 		printf("GEOM_PART: integrity check failed (%s, %s)\n",
351 		    pp->name, table->gpt_scheme->name);
352 		if (check_integrity != 0)
353 			return (EINVAL);
354 		table->gpt_corrupt = 1;
355 	}
356 	return (0);
357 }
358 #undef	DPRINTF
359 
360 struct g_part_entry *
361 g_part_new_entry(struct g_part_table *table, int index, quad_t start,
362     quad_t end)
363 {
364 	struct g_part_entry *entry, *last;
365 
366 	last = NULL;
367 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
368 		if (entry->gpe_index == index)
369 			break;
370 		if (entry->gpe_index > index) {
371 			entry = NULL;
372 			break;
373 		}
374 		last = entry;
375 	}
376 	if (entry == NULL) {
377 		entry = g_malloc(table->gpt_scheme->gps_entrysz,
378 		    M_WAITOK | M_ZERO);
379 		entry->gpe_index = index;
380 		if (last == NULL)
381 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
382 		else
383 			LIST_INSERT_AFTER(last, entry, gpe_entry);
384 	} else
385 		entry->gpe_offset = 0;
386 	entry->gpe_start = start;
387 	entry->gpe_end = end;
388 	return (entry);
389 }
390 
391 static void
392 g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
393     struct g_part_entry *entry)
394 {
395 	struct g_consumer *cp;
396 	struct g_provider *pp;
397 	struct sbuf *sb;
398 	off_t offset;
399 
400 	cp = LIST_FIRST(&gp->consumer);
401 	pp = cp->provider;
402 
403 	offset = entry->gpe_start * pp->sectorsize;
404 	if (entry->gpe_offset < offset)
405 		entry->gpe_offset = offset;
406 
407 	if (entry->gpe_pp == NULL) {
408 		sb = sbuf_new_auto();
409 		G_PART_FULLNAME(table, entry, sb, gp->name);
410 		sbuf_finish(sb);
411 		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
412 		sbuf_delete(sb);
413 		entry->gpe_pp->private = entry;		/* Close the circle. */
414 	}
415 	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
416 	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
417 	    pp->sectorsize;
418 	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
419 	entry->gpe_pp->sectorsize = pp->sectorsize;
420 	entry->gpe_pp->flags = pp->flags & G_PF_CANDELETE;
421 	entry->gpe_pp->stripesize = pp->stripesize;
422 	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
423 	if (pp->stripesize > 0)
424 		entry->gpe_pp->stripeoffset %= pp->stripesize;
425 	g_error_provider(entry->gpe_pp, 0);
426 }
427 
428 static struct g_geom*
429 g_part_find_geom(const char *name)
430 {
431 	struct g_geom *gp;
432 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
433 		if (!strcmp(name, gp->name))
434 			break;
435 	}
436 	return (gp);
437 }
438 
439 static int
440 g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
441 {
442 	struct g_geom *gp;
443 	const char *gname;
444 
445 	gname = gctl_get_asciiparam(req, name);
446 	if (gname == NULL)
447 		return (ENOATTR);
448 	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
449 		gname += sizeof(_PATH_DEV) - 1;
450 	gp = g_part_find_geom(gname);
451 	if (gp == NULL) {
452 		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
453 		return (EINVAL);
454 	}
455 	if ((gp->flags & G_GEOM_WITHER) != 0) {
456 		gctl_error(req, "%d %s", ENXIO, gname);
457 		return (ENXIO);
458 	}
459 	*v = gp;
460 	return (0);
461 }
462 
463 static int
464 g_part_parm_provider(struct gctl_req *req, const char *name,
465     struct g_provider **v)
466 {
467 	struct g_provider *pp;
468 	const char *pname;
469 
470 	pname = gctl_get_asciiparam(req, name);
471 	if (pname == NULL)
472 		return (ENOATTR);
473 	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
474 		pname += sizeof(_PATH_DEV) - 1;
475 	pp = g_provider_by_name(pname);
476 	if (pp == NULL) {
477 		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
478 		return (EINVAL);
479 	}
480 	*v = pp;
481 	return (0);
482 }
483 
484 static int
485 g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
486 {
487 	const char *p;
488 	char *x;
489 	quad_t q;
490 
491 	p = gctl_get_asciiparam(req, name);
492 	if (p == NULL)
493 		return (ENOATTR);
494 	q = strtoq(p, &x, 0);
495 	if (*x != '\0' || q < 0) {
496 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
497 		return (EINVAL);
498 	}
499 	*v = q;
500 	return (0);
501 }
502 
503 static int
504 g_part_parm_scheme(struct gctl_req *req, const char *name,
505     struct g_part_scheme **v)
506 {
507 	struct g_part_scheme *s;
508 	const char *p;
509 
510 	p = gctl_get_asciiparam(req, name);
511 	if (p == NULL)
512 		return (ENOATTR);
513 	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
514 		if (s == &g_part_null_scheme)
515 			continue;
516 		if (!strcasecmp(s->name, p))
517 			break;
518 	}
519 	if (s == NULL) {
520 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
521 		return (EINVAL);
522 	}
523 	*v = s;
524 	return (0);
525 }
526 
527 static int
528 g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
529 {
530 	const char *p;
531 
532 	p = gctl_get_asciiparam(req, name);
533 	if (p == NULL)
534 		return (ENOATTR);
535 	/* An empty label is always valid. */
536 	if (strcmp(name, "label") != 0 && p[0] == '\0') {
537 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
538 		return (EINVAL);
539 	}
540 	*v = p;
541 	return (0);
542 }
543 
544 static int
545 g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
546 {
547 	const intmax_t *p;
548 	int size;
549 
550 	p = gctl_get_param(req, name, &size);
551 	if (p == NULL)
552 		return (ENOATTR);
553 	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
554 		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
555 		return (EINVAL);
556 	}
557 	*v = (u_int)*p;
558 	return (0);
559 }
560 
561 static int
562 g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
563 {
564 	const uint32_t *p;
565 	int size;
566 
567 	p = gctl_get_param(req, name, &size);
568 	if (p == NULL)
569 		return (ENOATTR);
570 	if (size != sizeof(*p) || *p > INT_MAX) {
571 		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
572 		return (EINVAL);
573 	}
574 	*v = (u_int)*p;
575 	return (0);
576 }
577 
578 static int
579 g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
580     unsigned int *s)
581 {
582 	const void *p;
583 	int size;
584 
585 	p = gctl_get_param(req, name, &size);
586 	if (p == NULL)
587 		return (ENOATTR);
588 	*v = p;
589 	*s = size;
590 	return (0);
591 }
592 
593 static int
594 g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
595 {
596 	struct g_part_scheme *iter, *scheme;
597 	struct g_part_table *table;
598 	int pri, probe;
599 
600 	table = gp->softc;
601 	scheme = (table != NULL) ? table->gpt_scheme : NULL;
602 	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
603 	if (pri == 0)
604 		goto done;
605 	if (pri > 0) {	/* error */
606 		scheme = NULL;
607 		pri = INT_MIN;
608 	}
609 
610 	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
611 		if (iter == &g_part_null_scheme)
612 			continue;
613 		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
614 		    M_WAITOK);
615 		table->gpt_gp = gp;
616 		table->gpt_scheme = iter;
617 		table->gpt_depth = depth;
618 		probe = G_PART_PROBE(table, cp);
619 		if (probe <= 0 && probe > pri) {
620 			pri = probe;
621 			scheme = iter;
622 			if (gp->softc != NULL)
623 				kobj_delete((kobj_t)gp->softc, M_GEOM);
624 			gp->softc = table;
625 			if (pri == 0)
626 				goto done;
627 		} else
628 			kobj_delete((kobj_t)table, M_GEOM);
629 	}
630 
631 done:
632 	return ((scheme == NULL) ? ENXIO : 0);
633 }
634 
635 /*
636  * Control request functions.
637  */
638 
639 static int
640 g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
641 {
642 	struct g_geom *gp;
643 	struct g_provider *pp;
644 	struct g_part_entry *delent, *last, *entry;
645 	struct g_part_table *table;
646 	struct sbuf *sb;
647 	quad_t end;
648 	unsigned int index;
649 	int error;
650 
651 	gp = gpp->gpp_geom;
652 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
653 	g_topology_assert();
654 
655 	pp = LIST_FIRST(&gp->consumer)->provider;
656 	table = gp->softc;
657 	end = gpp->gpp_start + gpp->gpp_size - 1;
658 
659 	if (gpp->gpp_start < table->gpt_first ||
660 	    gpp->gpp_start > table->gpt_last) {
661 		gctl_error(req, "%d start '%jd'", EINVAL,
662 		    (intmax_t)gpp->gpp_start);
663 		return (EINVAL);
664 	}
665 	if (end < gpp->gpp_start || end > table->gpt_last) {
666 		gctl_error(req, "%d size '%jd'", EINVAL,
667 		    (intmax_t)gpp->gpp_size);
668 		return (EINVAL);
669 	}
670 	if (gpp->gpp_index > table->gpt_entries) {
671 		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
672 		return (EINVAL);
673 	}
674 
675 	delent = last = NULL;
676 	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
677 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
678 		if (entry->gpe_deleted) {
679 			if (entry->gpe_index == index)
680 				delent = entry;
681 			continue;
682 		}
683 		if (entry->gpe_index == index)
684 			index = entry->gpe_index + 1;
685 		if (entry->gpe_index < index)
686 			last = entry;
687 		if (entry->gpe_internal)
688 			continue;
689 		if (gpp->gpp_start >= entry->gpe_start &&
690 		    gpp->gpp_start <= entry->gpe_end) {
691 			gctl_error(req, "%d start '%jd'", ENOSPC,
692 			    (intmax_t)gpp->gpp_start);
693 			return (ENOSPC);
694 		}
695 		if (end >= entry->gpe_start && end <= entry->gpe_end) {
696 			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
697 			return (ENOSPC);
698 		}
699 		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
700 			gctl_error(req, "%d size '%jd'", ENOSPC,
701 			    (intmax_t)gpp->gpp_size);
702 			return (ENOSPC);
703 		}
704 	}
705 	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
706 		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
707 		return (EEXIST);
708 	}
709 	if (index > table->gpt_entries) {
710 		gctl_error(req, "%d index '%d'", ENOSPC, index);
711 		return (ENOSPC);
712 	}
713 
714 	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
715 	    M_WAITOK | M_ZERO) : delent;
716 	entry->gpe_index = index;
717 	entry->gpe_start = gpp->gpp_start;
718 	entry->gpe_end = end;
719 	error = G_PART_ADD(table, entry, gpp);
720 	if (error) {
721 		gctl_error(req, "%d", error);
722 		if (delent == NULL)
723 			g_free(entry);
724 		return (error);
725 	}
726 	if (delent == NULL) {
727 		if (last == NULL)
728 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
729 		else
730 			LIST_INSERT_AFTER(last, entry, gpe_entry);
731 		entry->gpe_created = 1;
732 	} else {
733 		entry->gpe_deleted = 0;
734 		entry->gpe_modified = 1;
735 	}
736 	g_part_new_provider(gp, table, entry);
737 
738 	/* Provide feedback if so requested. */
739 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
740 		sb = sbuf_new_auto();
741 		G_PART_FULLNAME(table, entry, sb, gp->name);
742 		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
743 			sbuf_printf(sb, " added, but partition is not "
744 			    "aligned on %u bytes\n", pp->stripesize);
745 		else
746 			sbuf_cat(sb, " added\n");
747 		sbuf_finish(sb);
748 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
749 		sbuf_delete(sb);
750 	}
751 	return (0);
752 }
753 
754 static int
755 g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
756 {
757 	struct g_geom *gp;
758 	struct g_part_table *table;
759 	struct sbuf *sb;
760 	int error, sz;
761 
762 	gp = gpp->gpp_geom;
763 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
764 	g_topology_assert();
765 
766 	table = gp->softc;
767 	sz = table->gpt_scheme->gps_bootcodesz;
768 	if (sz == 0) {
769 		error = ENODEV;
770 		goto fail;
771 	}
772 	if (gpp->gpp_codesize > sz) {
773 		error = EFBIG;
774 		goto fail;
775 	}
776 
777 	error = G_PART_BOOTCODE(table, gpp);
778 	if (error)
779 		goto fail;
780 
781 	/* Provide feedback if so requested. */
782 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
783 		sb = sbuf_new_auto();
784 		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
785 		sbuf_finish(sb);
786 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
787 		sbuf_delete(sb);
788 	}
789 	return (0);
790 
791  fail:
792 	gctl_error(req, "%d", error);
793 	return (error);
794 }
795 
796 static int
797 g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
798 {
799 	struct g_consumer *cp;
800 	struct g_geom *gp;
801 	struct g_provider *pp;
802 	struct g_part_entry *entry, *tmp;
803 	struct g_part_table *table;
804 	char *buf;
805 	int error, i;
806 
807 	gp = gpp->gpp_geom;
808 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
809 	g_topology_assert();
810 
811 	table = gp->softc;
812 	if (!table->gpt_opened) {
813 		gctl_error(req, "%d", EPERM);
814 		return (EPERM);
815 	}
816 
817 	g_topology_unlock();
818 
819 	cp = LIST_FIRST(&gp->consumer);
820 	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
821 		pp = cp->provider;
822 		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
823 		while (table->gpt_smhead != 0) {
824 			i = ffs(table->gpt_smhead) - 1;
825 			error = g_write_data(cp, i * pp->sectorsize, buf,
826 			    pp->sectorsize);
827 			if (error) {
828 				g_free(buf);
829 				goto fail;
830 			}
831 			table->gpt_smhead &= ~(1 << i);
832 		}
833 		while (table->gpt_smtail != 0) {
834 			i = ffs(table->gpt_smtail) - 1;
835 			error = g_write_data(cp, pp->mediasize - (i + 1) *
836 			    pp->sectorsize, buf, pp->sectorsize);
837 			if (error) {
838 				g_free(buf);
839 				goto fail;
840 			}
841 			table->gpt_smtail &= ~(1 << i);
842 		}
843 		g_free(buf);
844 	}
845 
846 	if (table->gpt_scheme == &g_part_null_scheme) {
847 		g_topology_lock();
848 		g_access(cp, -1, -1, -1);
849 		g_part_wither(gp, ENXIO);
850 		return (0);
851 	}
852 
853 	error = G_PART_WRITE(table, cp);
854 	if (error)
855 		goto fail;
856 
857 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
858 		if (!entry->gpe_deleted) {
859 			entry->gpe_created = 0;
860 			entry->gpe_modified = 0;
861 			continue;
862 		}
863 		LIST_REMOVE(entry, gpe_entry);
864 		g_free(entry);
865 	}
866 	table->gpt_created = 0;
867 	table->gpt_opened = 0;
868 
869 	g_topology_lock();
870 	g_access(cp, -1, -1, -1);
871 	return (0);
872 
873 fail:
874 	g_topology_lock();
875 	gctl_error(req, "%d", error);
876 	return (error);
877 }
878 
879 static int
880 g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
881 {
882 	struct g_consumer *cp;
883 	struct g_geom *gp;
884 	struct g_provider *pp;
885 	struct g_part_scheme *scheme;
886 	struct g_part_table *null, *table;
887 	struct sbuf *sb;
888 	int attr, error;
889 
890 	pp = gpp->gpp_provider;
891 	scheme = gpp->gpp_scheme;
892 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
893 	g_topology_assert();
894 
895 	/* Check that there isn't already a g_part geom on the provider. */
896 	gp = g_part_find_geom(pp->name);
897 	if (gp != NULL) {
898 		null = gp->softc;
899 		if (null->gpt_scheme != &g_part_null_scheme) {
900 			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
901 			return (EEXIST);
902 		}
903 	} else
904 		null = NULL;
905 
906 	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
907 	    (gpp->gpp_entries < scheme->gps_minent ||
908 	     gpp->gpp_entries > scheme->gps_maxent)) {
909 		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
910 		return (EINVAL);
911 	}
912 
913 	if (null == NULL)
914 		gp = g_new_geomf(&g_part_class, "%s", pp->name);
915 	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
916 	    M_WAITOK);
917 	table = gp->softc;
918 	table->gpt_gp = gp;
919 	table->gpt_scheme = gpp->gpp_scheme;
920 	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
921 	    gpp->gpp_entries : scheme->gps_minent;
922 	LIST_INIT(&table->gpt_entry);
923 	if (null == NULL) {
924 		cp = g_new_consumer(gp);
925 		error = g_attach(cp, pp);
926 		if (error == 0)
927 			error = g_access(cp, 1, 1, 1);
928 		if (error != 0) {
929 			g_part_wither(gp, error);
930 			gctl_error(req, "%d geom '%s'", error, pp->name);
931 			return (error);
932 		}
933 		table->gpt_opened = 1;
934 	} else {
935 		cp = LIST_FIRST(&gp->consumer);
936 		table->gpt_opened = null->gpt_opened;
937 		table->gpt_smhead = null->gpt_smhead;
938 		table->gpt_smtail = null->gpt_smtail;
939 	}
940 
941 	g_topology_unlock();
942 
943 	/* Make sure the provider has media. */
944 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
945 		error = ENODEV;
946 		goto fail;
947 	}
948 
949 	/* Make sure we can nest and if so, determine our depth. */
950 	error = g_getattr("PART::isleaf", cp, &attr);
951 	if (!error && attr) {
952 		error = ENODEV;
953 		goto fail;
954 	}
955 	error = g_getattr("PART::depth", cp, &attr);
956 	table->gpt_depth = (!error) ? attr + 1 : 0;
957 
958 	/*
959 	 * Synthesize a disk geometry. Some partitioning schemes
960 	 * depend on it and since some file systems need it even
961 	 * when the partitition scheme doesn't, we do it here in
962 	 * scheme-independent code.
963 	 */
964 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
965 
966 	error = G_PART_CREATE(table, gpp);
967 	if (error)
968 		goto fail;
969 
970 	g_topology_lock();
971 
972 	table->gpt_created = 1;
973 	if (null != NULL)
974 		kobj_delete((kobj_t)null, M_GEOM);
975 
976 	/*
977 	 * Support automatic commit by filling in the gpp_geom
978 	 * parameter.
979 	 */
980 	gpp->gpp_parms |= G_PART_PARM_GEOM;
981 	gpp->gpp_geom = gp;
982 
983 	/* Provide feedback if so requested. */
984 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
985 		sb = sbuf_new_auto();
986 		sbuf_printf(sb, "%s created\n", gp->name);
987 		sbuf_finish(sb);
988 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
989 		sbuf_delete(sb);
990 	}
991 	return (0);
992 
993 fail:
994 	g_topology_lock();
995 	if (null == NULL) {
996 		g_access(cp, -1, -1, -1);
997 		g_part_wither(gp, error);
998 	} else {
999 		kobj_delete((kobj_t)gp->softc, M_GEOM);
1000 		gp->softc = null;
1001 	}
1002 	gctl_error(req, "%d provider", error);
1003 	return (error);
1004 }
1005 
1006 static int
1007 g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1008 {
1009 	struct g_geom *gp;
1010 	struct g_provider *pp;
1011 	struct g_part_entry *entry;
1012 	struct g_part_table *table;
1013 	struct sbuf *sb;
1014 
1015 	gp = gpp->gpp_geom;
1016 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1017 	g_topology_assert();
1018 
1019 	table = gp->softc;
1020 
1021 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1022 		if (entry->gpe_deleted || entry->gpe_internal)
1023 			continue;
1024 		if (entry->gpe_index == gpp->gpp_index)
1025 			break;
1026 	}
1027 	if (entry == NULL) {
1028 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1029 		return (ENOENT);
1030 	}
1031 
1032 	pp = entry->gpe_pp;
1033 	if (pp != NULL) {
1034 		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1035 			gctl_error(req, "%d", EBUSY);
1036 			return (EBUSY);
1037 		}
1038 
1039 		pp->private = NULL;
1040 		entry->gpe_pp = NULL;
1041 	}
1042 
1043 	if (pp != NULL)
1044 		g_wither_provider(pp, ENXIO);
1045 
1046 	/* Provide feedback if so requested. */
1047 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1048 		sb = sbuf_new_auto();
1049 		G_PART_FULLNAME(table, entry, sb, gp->name);
1050 		sbuf_cat(sb, " deleted\n");
1051 		sbuf_finish(sb);
1052 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1053 		sbuf_delete(sb);
1054 	}
1055 
1056 	if (entry->gpe_created) {
1057 		LIST_REMOVE(entry, gpe_entry);
1058 		g_free(entry);
1059 	} else {
1060 		entry->gpe_modified = 0;
1061 		entry->gpe_deleted = 1;
1062 	}
1063 	return (0);
1064 }
1065 
1066 static int
1067 g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1068 {
1069 	struct g_consumer *cp;
1070 	struct g_geom *gp;
1071 	struct g_provider *pp;
1072 	struct g_part_entry *entry, *tmp;
1073 	struct g_part_table *null, *table;
1074 	struct sbuf *sb;
1075 	int error;
1076 
1077 	gp = gpp->gpp_geom;
1078 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1079 	g_topology_assert();
1080 
1081 	table = gp->softc;
1082 	/* Check for busy providers. */
1083 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1084 		if (entry->gpe_deleted || entry->gpe_internal)
1085 			continue;
1086 		if (gpp->gpp_force) {
1087 			pp = entry->gpe_pp;
1088 			if (pp == NULL)
1089 				continue;
1090 			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1091 				continue;
1092 		}
1093 		gctl_error(req, "%d", EBUSY);
1094 		return (EBUSY);
1095 	}
1096 
1097 	if (gpp->gpp_force) {
1098 		/* Destroy all providers. */
1099 		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1100 			pp = entry->gpe_pp;
1101 			if (pp != NULL) {
1102 				pp->private = NULL;
1103 				g_wither_provider(pp, ENXIO);
1104 			}
1105 			LIST_REMOVE(entry, gpe_entry);
1106 			g_free(entry);
1107 		}
1108 	}
1109 
1110 	error = G_PART_DESTROY(table, gpp);
1111 	if (error) {
1112 		gctl_error(req, "%d", error);
1113 		return (error);
1114 	}
1115 
1116 	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1117 	    M_WAITOK);
1118 	null = gp->softc;
1119 	null->gpt_gp = gp;
1120 	null->gpt_scheme = &g_part_null_scheme;
1121 	LIST_INIT(&null->gpt_entry);
1122 
1123 	cp = LIST_FIRST(&gp->consumer);
1124 	pp = cp->provider;
1125 	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1126 
1127 	null->gpt_depth = table->gpt_depth;
1128 	null->gpt_opened = table->gpt_opened;
1129 	null->gpt_smhead = table->gpt_smhead;
1130 	null->gpt_smtail = table->gpt_smtail;
1131 
1132 	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1133 		LIST_REMOVE(entry, gpe_entry);
1134 		g_free(entry);
1135 	}
1136 	kobj_delete((kobj_t)table, M_GEOM);
1137 
1138 	/* Provide feedback if so requested. */
1139 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1140 		sb = sbuf_new_auto();
1141 		sbuf_printf(sb, "%s destroyed\n", gp->name);
1142 		sbuf_finish(sb);
1143 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1144 		sbuf_delete(sb);
1145 	}
1146 	return (0);
1147 }
1148 
1149 static int
1150 g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1151 {
1152 	struct g_geom *gp;
1153 	struct g_part_entry *entry;
1154 	struct g_part_table *table;
1155 	struct sbuf *sb;
1156 	int error;
1157 
1158 	gp = gpp->gpp_geom;
1159 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1160 	g_topology_assert();
1161 
1162 	table = gp->softc;
1163 
1164 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1165 		if (entry->gpe_deleted || entry->gpe_internal)
1166 			continue;
1167 		if (entry->gpe_index == gpp->gpp_index)
1168 			break;
1169 	}
1170 	if (entry == NULL) {
1171 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1172 		return (ENOENT);
1173 	}
1174 
1175 	error = G_PART_MODIFY(table, entry, gpp);
1176 	if (error) {
1177 		gctl_error(req, "%d", error);
1178 		return (error);
1179 	}
1180 
1181 	if (!entry->gpe_created)
1182 		entry->gpe_modified = 1;
1183 
1184 	/* Provide feedback if so requested. */
1185 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1186 		sb = sbuf_new_auto();
1187 		G_PART_FULLNAME(table, entry, sb, gp->name);
1188 		sbuf_cat(sb, " modified\n");
1189 		sbuf_finish(sb);
1190 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1191 		sbuf_delete(sb);
1192 	}
1193 	return (0);
1194 }
1195 
1196 static int
1197 g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1198 {
1199 	gctl_error(req, "%d verb 'move'", ENOSYS);
1200 	return (ENOSYS);
1201 }
1202 
1203 static int
1204 g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1205 {
1206 	struct g_part_table *table;
1207 	struct g_geom *gp;
1208 	struct sbuf *sb;
1209 	int error, recovered;
1210 
1211 	gp = gpp->gpp_geom;
1212 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1213 	g_topology_assert();
1214 	table = gp->softc;
1215 	error = recovered = 0;
1216 
1217 	if (table->gpt_corrupt) {
1218 		error = G_PART_RECOVER(table);
1219 		if (error) {
1220 			gctl_error(req, "%d recovering '%s' failed",
1221 			    error, gp->name);
1222 			return (error);
1223 		}
1224 		recovered = 1;
1225 	}
1226 	/* Provide feedback if so requested. */
1227 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1228 		sb = sbuf_new_auto();
1229 		if (recovered)
1230 			sbuf_printf(sb, "%s recovered\n", gp->name);
1231 		else
1232 			sbuf_printf(sb, "%s recovering is not needed\n",
1233 			    gp->name);
1234 		sbuf_finish(sb);
1235 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1236 		sbuf_delete(sb);
1237 	}
1238 	return (0);
1239 }
1240 
1241 static int
1242 g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1243 {
1244 	struct g_geom *gp;
1245 	struct g_provider *pp;
1246 	struct g_part_entry *pe, *entry;
1247 	struct g_part_table *table;
1248 	struct sbuf *sb;
1249 	quad_t end;
1250 	int error;
1251 
1252 	gp = gpp->gpp_geom;
1253 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1254 	g_topology_assert();
1255 	table = gp->softc;
1256 
1257 	/* check gpp_index */
1258 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1259 		if (entry->gpe_deleted || entry->gpe_internal)
1260 			continue;
1261 		if (entry->gpe_index == gpp->gpp_index)
1262 			break;
1263 	}
1264 	if (entry == NULL) {
1265 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1266 		return (ENOENT);
1267 	}
1268 
1269 	/* check gpp_size */
1270 	end = entry->gpe_start + gpp->gpp_size - 1;
1271 	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1272 		gctl_error(req, "%d size '%jd'", EINVAL,
1273 		    (intmax_t)gpp->gpp_size);
1274 		return (EINVAL);
1275 	}
1276 
1277 	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1278 		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1279 			continue;
1280 		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1281 			gctl_error(req, "%d end '%jd'", ENOSPC,
1282 			    (intmax_t)end);
1283 			return (ENOSPC);
1284 		}
1285 		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1286 			gctl_error(req, "%d size '%jd'", ENOSPC,
1287 			    (intmax_t)gpp->gpp_size);
1288 			return (ENOSPC);
1289 		}
1290 	}
1291 
1292 	pp = entry->gpe_pp;
1293 	if ((g_debugflags & 16) == 0 &&
1294 	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1295 		gctl_error(req, "%d", EBUSY);
1296 		return (EBUSY);
1297 	}
1298 
1299 	error = G_PART_RESIZE(table, entry, gpp);
1300 	if (error) {
1301 		gctl_error(req, "%d", error);
1302 		return (error);
1303 	}
1304 
1305 	if (!entry->gpe_created)
1306 		entry->gpe_modified = 1;
1307 
1308 	/* update mediasize of changed provider */
1309 	pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1310 		pp->sectorsize;
1311 
1312 	/* Provide feedback if so requested. */
1313 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1314 		sb = sbuf_new_auto();
1315 		G_PART_FULLNAME(table, entry, sb, gp->name);
1316 		sbuf_cat(sb, " resized\n");
1317 		sbuf_finish(sb);
1318 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1319 		sbuf_delete(sb);
1320 	}
1321 	return (0);
1322 }
1323 
1324 static int
1325 g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1326     unsigned int set)
1327 {
1328 	struct g_geom *gp;
1329 	struct g_part_entry *entry;
1330 	struct g_part_table *table;
1331 	struct sbuf *sb;
1332 	int error;
1333 
1334 	gp = gpp->gpp_geom;
1335 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1336 	g_topology_assert();
1337 
1338 	table = gp->softc;
1339 
1340 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1341 		if (entry->gpe_deleted || entry->gpe_internal)
1342 			continue;
1343 		if (entry->gpe_index == gpp->gpp_index)
1344 			break;
1345 	}
1346 	if (entry == NULL) {
1347 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1348 		return (ENOENT);
1349 	}
1350 
1351 	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1352 	if (error) {
1353 		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1354 		return (error);
1355 	}
1356 
1357 	/* Provide feedback if so requested. */
1358 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1359 		sb = sbuf_new_auto();
1360 		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1361 		    (set) ? "" : "un");
1362 		G_PART_FULLNAME(table, entry, sb, gp->name);
1363 		sbuf_printf(sb, "\n");
1364 		sbuf_finish(sb);
1365 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1366 		sbuf_delete(sb);
1367 	}
1368 	return (0);
1369 }
1370 
1371 static int
1372 g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1373 {
1374 	struct g_consumer *cp;
1375 	struct g_provider *pp;
1376 	struct g_geom *gp;
1377 	struct g_part_entry *entry, *tmp;
1378 	struct g_part_table *table;
1379 	int error, reprobe;
1380 
1381 	gp = gpp->gpp_geom;
1382 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1383 	g_topology_assert();
1384 
1385 	table = gp->softc;
1386 	if (!table->gpt_opened) {
1387 		gctl_error(req, "%d", EPERM);
1388 		return (EPERM);
1389 	}
1390 
1391 	cp = LIST_FIRST(&gp->consumer);
1392 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1393 		entry->gpe_modified = 0;
1394 		if (entry->gpe_created) {
1395 			pp = entry->gpe_pp;
1396 			if (pp != NULL) {
1397 				pp->private = NULL;
1398 				entry->gpe_pp = NULL;
1399 				g_wither_provider(pp, ENXIO);
1400 			}
1401 			entry->gpe_deleted = 1;
1402 		}
1403 		if (entry->gpe_deleted) {
1404 			LIST_REMOVE(entry, gpe_entry);
1405 			g_free(entry);
1406 		}
1407 	}
1408 
1409 	g_topology_unlock();
1410 
1411 	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1412 	    table->gpt_created) ? 1 : 0;
1413 
1414 	if (reprobe) {
1415 		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1416 			if (entry->gpe_internal)
1417 				continue;
1418 			error = EBUSY;
1419 			goto fail;
1420 		}
1421 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1422 			LIST_REMOVE(entry, gpe_entry);
1423 			g_free(entry);
1424 		}
1425 		error = g_part_probe(gp, cp, table->gpt_depth);
1426 		if (error) {
1427 			g_topology_lock();
1428 			g_access(cp, -1, -1, -1);
1429 			g_part_wither(gp, error);
1430 			return (0);
1431 		}
1432 		table = gp->softc;
1433 
1434 		/*
1435 		 * Synthesize a disk geometry. Some partitioning schemes
1436 		 * depend on it and since some file systems need it even
1437 		 * when the partitition scheme doesn't, we do it here in
1438 		 * scheme-independent code.
1439 		 */
1440 		pp = cp->provider;
1441 		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1442 	}
1443 
1444 	error = G_PART_READ(table, cp);
1445 	if (error)
1446 		goto fail;
1447 	error = g_part_check_integrity(table, cp);
1448 	if (error)
1449 		goto fail;
1450 
1451 	g_topology_lock();
1452 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1453 		if (!entry->gpe_internal)
1454 			g_part_new_provider(gp, table, entry);
1455 	}
1456 
1457 	table->gpt_opened = 0;
1458 	g_access(cp, -1, -1, -1);
1459 	return (0);
1460 
1461 fail:
1462 	g_topology_lock();
1463 	gctl_error(req, "%d", error);
1464 	return (error);
1465 }
1466 
1467 static void
1468 g_part_wither(struct g_geom *gp, int error)
1469 {
1470 	struct g_part_entry *entry;
1471 	struct g_part_table *table;
1472 
1473 	table = gp->softc;
1474 	if (table != NULL) {
1475 		G_PART_DESTROY(table, NULL);
1476 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1477 			LIST_REMOVE(entry, gpe_entry);
1478 			g_free(entry);
1479 		}
1480 		if (gp->softc != NULL) {
1481 			kobj_delete((kobj_t)gp->softc, M_GEOM);
1482 			gp->softc = NULL;
1483 		}
1484 	}
1485 	g_wither_geom(gp, error);
1486 }
1487 
1488 /*
1489  * Class methods.
1490  */
1491 
1492 static void
1493 g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1494 {
1495 	struct g_part_parms gpp;
1496 	struct g_part_table *table;
1497 	struct gctl_req_arg *ap;
1498 	enum g_part_ctl ctlreq;
1499 	unsigned int i, mparms, oparms, parm;
1500 	int auto_commit, close_on_error;
1501 	int error, modifies;
1502 
1503 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1504 	g_topology_assert();
1505 
1506 	ctlreq = G_PART_CTL_NONE;
1507 	modifies = 1;
1508 	mparms = 0;
1509 	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1510 	switch (*verb) {
1511 	case 'a':
1512 		if (!strcmp(verb, "add")) {
1513 			ctlreq = G_PART_CTL_ADD;
1514 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1515 			    G_PART_PARM_START | G_PART_PARM_TYPE;
1516 			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1517 		}
1518 		break;
1519 	case 'b':
1520 		if (!strcmp(verb, "bootcode")) {
1521 			ctlreq = G_PART_CTL_BOOTCODE;
1522 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1523 		}
1524 		break;
1525 	case 'c':
1526 		if (!strcmp(verb, "commit")) {
1527 			ctlreq = G_PART_CTL_COMMIT;
1528 			mparms |= G_PART_PARM_GEOM;
1529 			modifies = 0;
1530 		} else if (!strcmp(verb, "create")) {
1531 			ctlreq = G_PART_CTL_CREATE;
1532 			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1533 			oparms |= G_PART_PARM_ENTRIES;
1534 		}
1535 		break;
1536 	case 'd':
1537 		if (!strcmp(verb, "delete")) {
1538 			ctlreq = G_PART_CTL_DELETE;
1539 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1540 		} else if (!strcmp(verb, "destroy")) {
1541 			ctlreq = G_PART_CTL_DESTROY;
1542 			mparms |= G_PART_PARM_GEOM;
1543 			oparms |= G_PART_PARM_FORCE;
1544 		}
1545 		break;
1546 	case 'm':
1547 		if (!strcmp(verb, "modify")) {
1548 			ctlreq = G_PART_CTL_MODIFY;
1549 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1550 			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1551 		} else if (!strcmp(verb, "move")) {
1552 			ctlreq = G_PART_CTL_MOVE;
1553 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1554 		}
1555 		break;
1556 	case 'r':
1557 		if (!strcmp(verb, "recover")) {
1558 			ctlreq = G_PART_CTL_RECOVER;
1559 			mparms |= G_PART_PARM_GEOM;
1560 		} else if (!strcmp(verb, "resize")) {
1561 			ctlreq = G_PART_CTL_RESIZE;
1562 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1563 			    G_PART_PARM_SIZE;
1564 		}
1565 		break;
1566 	case 's':
1567 		if (!strcmp(verb, "set")) {
1568 			ctlreq = G_PART_CTL_SET;
1569 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1570 			    G_PART_PARM_INDEX;
1571 		}
1572 		break;
1573 	case 'u':
1574 		if (!strcmp(verb, "undo")) {
1575 			ctlreq = G_PART_CTL_UNDO;
1576 			mparms |= G_PART_PARM_GEOM;
1577 			modifies = 0;
1578 		} else if (!strcmp(verb, "unset")) {
1579 			ctlreq = G_PART_CTL_UNSET;
1580 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1581 			    G_PART_PARM_INDEX;
1582 		}
1583 		break;
1584 	}
1585 	if (ctlreq == G_PART_CTL_NONE) {
1586 		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1587 		return;
1588 	}
1589 
1590 	bzero(&gpp, sizeof(gpp));
1591 	for (i = 0; i < req->narg; i++) {
1592 		ap = &req->arg[i];
1593 		parm = 0;
1594 		switch (ap->name[0]) {
1595 		case 'a':
1596 			if (!strcmp(ap->name, "arg0")) {
1597 				parm = mparms &
1598 				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1599 			}
1600 			if (!strcmp(ap->name, "attrib"))
1601 				parm = G_PART_PARM_ATTRIB;
1602 			break;
1603 		case 'b':
1604 			if (!strcmp(ap->name, "bootcode"))
1605 				parm = G_PART_PARM_BOOTCODE;
1606 			break;
1607 		case 'c':
1608 			if (!strcmp(ap->name, "class"))
1609 				continue;
1610 			break;
1611 		case 'e':
1612 			if (!strcmp(ap->name, "entries"))
1613 				parm = G_PART_PARM_ENTRIES;
1614 			break;
1615 		case 'f':
1616 			if (!strcmp(ap->name, "flags"))
1617 				parm = G_PART_PARM_FLAGS;
1618 			else if (!strcmp(ap->name, "force"))
1619 				parm = G_PART_PARM_FORCE;
1620 			break;
1621 		case 'i':
1622 			if (!strcmp(ap->name, "index"))
1623 				parm = G_PART_PARM_INDEX;
1624 			break;
1625 		case 'l':
1626 			if (!strcmp(ap->name, "label"))
1627 				parm = G_PART_PARM_LABEL;
1628 			break;
1629 		case 'o':
1630 			if (!strcmp(ap->name, "output"))
1631 				parm = G_PART_PARM_OUTPUT;
1632 			break;
1633 		case 's':
1634 			if (!strcmp(ap->name, "scheme"))
1635 				parm = G_PART_PARM_SCHEME;
1636 			else if (!strcmp(ap->name, "size"))
1637 				parm = G_PART_PARM_SIZE;
1638 			else if (!strcmp(ap->name, "start"))
1639 				parm = G_PART_PARM_START;
1640 			break;
1641 		case 't':
1642 			if (!strcmp(ap->name, "type"))
1643 				parm = G_PART_PARM_TYPE;
1644 			break;
1645 		case 'v':
1646 			if (!strcmp(ap->name, "verb"))
1647 				continue;
1648 			else if (!strcmp(ap->name, "version"))
1649 				parm = G_PART_PARM_VERSION;
1650 			break;
1651 		}
1652 		if ((parm & (mparms | oparms)) == 0) {
1653 			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1654 			return;
1655 		}
1656 		switch (parm) {
1657 		case G_PART_PARM_ATTRIB:
1658 			error = g_part_parm_str(req, ap->name,
1659 			    &gpp.gpp_attrib);
1660 			break;
1661 		case G_PART_PARM_BOOTCODE:
1662 			error = g_part_parm_bootcode(req, ap->name,
1663 			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1664 			break;
1665 		case G_PART_PARM_ENTRIES:
1666 			error = g_part_parm_intmax(req, ap->name,
1667 			    &gpp.gpp_entries);
1668 			break;
1669 		case G_PART_PARM_FLAGS:
1670 			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1671 			break;
1672 		case G_PART_PARM_FORCE:
1673 			error = g_part_parm_uint32(req, ap->name,
1674 			    &gpp.gpp_force);
1675 			break;
1676 		case G_PART_PARM_GEOM:
1677 			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1678 			break;
1679 		case G_PART_PARM_INDEX:
1680 			error = g_part_parm_intmax(req, ap->name,
1681 			    &gpp.gpp_index);
1682 			break;
1683 		case G_PART_PARM_LABEL:
1684 			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1685 			break;
1686 		case G_PART_PARM_OUTPUT:
1687 			error = 0;	/* Write-only parameter */
1688 			break;
1689 		case G_PART_PARM_PROVIDER:
1690 			error = g_part_parm_provider(req, ap->name,
1691 			    &gpp.gpp_provider);
1692 			break;
1693 		case G_PART_PARM_SCHEME:
1694 			error = g_part_parm_scheme(req, ap->name,
1695 			    &gpp.gpp_scheme);
1696 			break;
1697 		case G_PART_PARM_SIZE:
1698 			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1699 			break;
1700 		case G_PART_PARM_START:
1701 			error = g_part_parm_quad(req, ap->name,
1702 			    &gpp.gpp_start);
1703 			break;
1704 		case G_PART_PARM_TYPE:
1705 			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1706 			break;
1707 		case G_PART_PARM_VERSION:
1708 			error = g_part_parm_uint32(req, ap->name,
1709 			    &gpp.gpp_version);
1710 			break;
1711 		default:
1712 			error = EDOOFUS;
1713 			gctl_error(req, "%d %s", error, ap->name);
1714 			break;
1715 		}
1716 		if (error != 0) {
1717 			if (error == ENOATTR) {
1718 				gctl_error(req, "%d param '%s'", error,
1719 				    ap->name);
1720 			}
1721 			return;
1722 		}
1723 		gpp.gpp_parms |= parm;
1724 	}
1725 	if ((gpp.gpp_parms & mparms) != mparms) {
1726 		parm = mparms - (gpp.gpp_parms & mparms);
1727 		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1728 		return;
1729 	}
1730 
1731 	/* Obtain permissions if possible/necessary. */
1732 	close_on_error = 0;
1733 	table = NULL;
1734 	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1735 		table = gpp.gpp_geom->softc;
1736 		if (table != NULL && table->gpt_corrupt &&
1737 		    ctlreq != G_PART_CTL_DESTROY &&
1738 		    ctlreq != G_PART_CTL_RECOVER) {
1739 			gctl_error(req, "%d table '%s' is corrupt",
1740 			    EPERM, gpp.gpp_geom->name);
1741 			return;
1742 		}
1743 		if (table != NULL && !table->gpt_opened) {
1744 			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1745 			    1, 1, 1);
1746 			if (error) {
1747 				gctl_error(req, "%d geom '%s'", error,
1748 				    gpp.gpp_geom->name);
1749 				return;
1750 			}
1751 			table->gpt_opened = 1;
1752 			close_on_error = 1;
1753 		}
1754 	}
1755 
1756 	/* Allow the scheme to check or modify the parameters. */
1757 	if (table != NULL) {
1758 		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1759 		if (error) {
1760 			gctl_error(req, "%d pre-check failed", error);
1761 			goto out;
1762 		}
1763 	} else
1764 		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1765 
1766 	switch (ctlreq) {
1767 	case G_PART_CTL_NONE:
1768 		panic("%s", __func__);
1769 	case G_PART_CTL_ADD:
1770 		error = g_part_ctl_add(req, &gpp);
1771 		break;
1772 	case G_PART_CTL_BOOTCODE:
1773 		error = g_part_ctl_bootcode(req, &gpp);
1774 		break;
1775 	case G_PART_CTL_COMMIT:
1776 		error = g_part_ctl_commit(req, &gpp);
1777 		break;
1778 	case G_PART_CTL_CREATE:
1779 		error = g_part_ctl_create(req, &gpp);
1780 		break;
1781 	case G_PART_CTL_DELETE:
1782 		error = g_part_ctl_delete(req, &gpp);
1783 		break;
1784 	case G_PART_CTL_DESTROY:
1785 		error = g_part_ctl_destroy(req, &gpp);
1786 		break;
1787 	case G_PART_CTL_MODIFY:
1788 		error = g_part_ctl_modify(req, &gpp);
1789 		break;
1790 	case G_PART_CTL_MOVE:
1791 		error = g_part_ctl_move(req, &gpp);
1792 		break;
1793 	case G_PART_CTL_RECOVER:
1794 		error = g_part_ctl_recover(req, &gpp);
1795 		break;
1796 	case G_PART_CTL_RESIZE:
1797 		error = g_part_ctl_resize(req, &gpp);
1798 		break;
1799 	case G_PART_CTL_SET:
1800 		error = g_part_ctl_setunset(req, &gpp, 1);
1801 		break;
1802 	case G_PART_CTL_UNDO:
1803 		error = g_part_ctl_undo(req, &gpp);
1804 		break;
1805 	case G_PART_CTL_UNSET:
1806 		error = g_part_ctl_setunset(req, &gpp, 0);
1807 		break;
1808 	}
1809 
1810 	/* Implement automatic commit. */
1811 	if (!error) {
1812 		auto_commit = (modifies &&
1813 		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1814 		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1815 		if (auto_commit) {
1816 			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1817 			    __func__));
1818 			error = g_part_ctl_commit(req, &gpp);
1819 		}
1820 	}
1821 
1822  out:
1823 	if (error && close_on_error) {
1824 		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1825 		table->gpt_opened = 0;
1826 	}
1827 }
1828 
1829 static int
1830 g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1831     struct g_geom *gp)
1832 {
1833 
1834 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1835 	g_topology_assert();
1836 
1837 	g_part_wither(gp, EINVAL);
1838 	return (0);
1839 }
1840 
1841 static struct g_geom *
1842 g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1843 {
1844 	struct g_consumer *cp;
1845 	struct g_geom *gp;
1846 	struct g_part_entry *entry;
1847 	struct g_part_table *table;
1848 	struct root_hold_token *rht;
1849 	int attr, depth;
1850 	int error;
1851 
1852 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1853 	g_topology_assert();
1854 
1855 	/* Skip providers that are already open for writing. */
1856 	if (pp->acw > 0)
1857 		return (NULL);
1858 
1859 	/*
1860 	 * Create a GEOM with consumer and hook it up to the provider.
1861 	 * With that we become part of the topology. Optain read access
1862 	 * to the provider.
1863 	 */
1864 	gp = g_new_geomf(mp, "%s", pp->name);
1865 	cp = g_new_consumer(gp);
1866 	error = g_attach(cp, pp);
1867 	if (error == 0)
1868 		error = g_access(cp, 1, 0, 0);
1869 	if (error != 0) {
1870 		g_part_wither(gp, error);
1871 		return (NULL);
1872 	}
1873 
1874 	rht = root_mount_hold(mp->name);
1875 	g_topology_unlock();
1876 
1877 	/*
1878 	 * Short-circuit the whole probing galore when there's no
1879 	 * media present.
1880 	 */
1881 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1882 		error = ENODEV;
1883 		goto fail;
1884 	}
1885 
1886 	/* Make sure we can nest and if so, determine our depth. */
1887 	error = g_getattr("PART::isleaf", cp, &attr);
1888 	if (!error && attr) {
1889 		error = ENODEV;
1890 		goto fail;
1891 	}
1892 	error = g_getattr("PART::depth", cp, &attr);
1893 	depth = (!error) ? attr + 1 : 0;
1894 
1895 	error = g_part_probe(gp, cp, depth);
1896 	if (error)
1897 		goto fail;
1898 
1899 	table = gp->softc;
1900 
1901 	/*
1902 	 * Synthesize a disk geometry. Some partitioning schemes
1903 	 * depend on it and since some file systems need it even
1904 	 * when the partitition scheme doesn't, we do it here in
1905 	 * scheme-independent code.
1906 	 */
1907 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1908 
1909 	error = G_PART_READ(table, cp);
1910 	if (error)
1911 		goto fail;
1912 	error = g_part_check_integrity(table, cp);
1913 	if (error)
1914 		goto fail;
1915 
1916 	g_topology_lock();
1917 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1918 		if (!entry->gpe_internal)
1919 			g_part_new_provider(gp, table, entry);
1920 	}
1921 
1922 	root_mount_rel(rht);
1923 	g_access(cp, -1, 0, 0);
1924 	return (gp);
1925 
1926  fail:
1927 	g_topology_lock();
1928 	root_mount_rel(rht);
1929 	g_access(cp, -1, 0, 0);
1930 	g_part_wither(gp, error);
1931 	return (NULL);
1932 }
1933 
1934 /*
1935  * Geom methods.
1936  */
1937 
1938 static int
1939 g_part_access(struct g_provider *pp, int dr, int dw, int de)
1940 {
1941 	struct g_consumer *cp;
1942 
1943 	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1944 	    dw, de));
1945 
1946 	cp = LIST_FIRST(&pp->geom->consumer);
1947 
1948 	/* We always gain write-exclusive access. */
1949 	return (g_access(cp, dr, dw, dw + de));
1950 }
1951 
1952 static void
1953 g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1954     struct g_consumer *cp, struct g_provider *pp)
1955 {
1956 	char buf[64];
1957 	struct g_part_entry *entry;
1958 	struct g_part_table *table;
1959 
1960 	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
1961 	table = gp->softc;
1962 
1963 	if (indent == NULL) {
1964 		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
1965 		entry = pp->private;
1966 		if (entry == NULL)
1967 			return;
1968 		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
1969 		    (uintmax_t)entry->gpe_offset,
1970 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1971 		/*
1972 		 * libdisk compatibility quirk - the scheme dumps the
1973 		 * slicer name and partition type in a way that is
1974 		 * compatible with libdisk. When libdisk is not used
1975 		 * anymore, this should go away.
1976 		 */
1977 		G_PART_DUMPCONF(table, entry, sb, indent);
1978 	} else if (cp != NULL) {	/* Consumer configuration. */
1979 		KASSERT(pp == NULL, ("%s", __func__));
1980 		/* none */
1981 	} else if (pp != NULL) {	/* Provider configuration. */
1982 		entry = pp->private;
1983 		if (entry == NULL)
1984 			return;
1985 		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
1986 		    (uintmax_t)entry->gpe_start);
1987 		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
1988 		    (uintmax_t)entry->gpe_end);
1989 		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
1990 		    entry->gpe_index);
1991 		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1992 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1993 		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
1994 		    (uintmax_t)entry->gpe_offset);
1995 		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
1996 		    (uintmax_t)pp->mediasize);
1997 		G_PART_DUMPCONF(table, entry, sb, indent);
1998 	} else {			/* Geom configuration. */
1999 		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2000 		    table->gpt_scheme->name);
2001 		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2002 		    table->gpt_entries);
2003 		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2004 		    (uintmax_t)table->gpt_first);
2005 		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2006 		    (uintmax_t)table->gpt_last);
2007 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2008 		    table->gpt_sectors);
2009 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2010 		    table->gpt_heads);
2011 		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2012 		    table->gpt_corrupt ? "CORRUPT": "OK");
2013 		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2014 		    table->gpt_opened ? "true": "false");
2015 		G_PART_DUMPCONF(table, NULL, sb, indent);
2016 	}
2017 }
2018 
2019 static void
2020 g_part_orphan(struct g_consumer *cp)
2021 {
2022 	struct g_provider *pp;
2023 	struct g_part_table *table;
2024 
2025 	pp = cp->provider;
2026 	KASSERT(pp != NULL, ("%s", __func__));
2027 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2028 	g_topology_assert();
2029 
2030 	KASSERT(pp->error != 0, ("%s", __func__));
2031 	table = cp->geom->softc;
2032 	if (table != NULL && table->gpt_opened)
2033 		g_access(cp, -1, -1, -1);
2034 	g_part_wither(cp->geom, pp->error);
2035 }
2036 
2037 static void
2038 g_part_spoiled(struct g_consumer *cp)
2039 {
2040 
2041 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2042 	g_topology_assert();
2043 
2044 	g_part_wither(cp->geom, ENXIO);
2045 }
2046 
2047 static void
2048 g_part_start(struct bio *bp)
2049 {
2050 	struct bio *bp2;
2051 	struct g_consumer *cp;
2052 	struct g_geom *gp;
2053 	struct g_part_entry *entry;
2054 	struct g_part_table *table;
2055 	struct g_kerneldump *gkd;
2056 	struct g_provider *pp;
2057 
2058 	pp = bp->bio_to;
2059 	gp = pp->geom;
2060 	table = gp->softc;
2061 	cp = LIST_FIRST(&gp->consumer);
2062 
2063 	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2064 	    pp->name));
2065 
2066 	entry = pp->private;
2067 	if (entry == NULL) {
2068 		g_io_deliver(bp, ENXIO);
2069 		return;
2070 	}
2071 
2072 	switch(bp->bio_cmd) {
2073 	case BIO_DELETE:
2074 	case BIO_READ:
2075 	case BIO_WRITE:
2076 		if (bp->bio_offset >= pp->mediasize) {
2077 			g_io_deliver(bp, EIO);
2078 			return;
2079 		}
2080 		bp2 = g_clone_bio(bp);
2081 		if (bp2 == NULL) {
2082 			g_io_deliver(bp, ENOMEM);
2083 			return;
2084 		}
2085 		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2086 			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2087 		bp2->bio_done = g_std_done;
2088 		bp2->bio_offset += entry->gpe_offset;
2089 		g_io_request(bp2, cp);
2090 		return;
2091 	case BIO_FLUSH:
2092 		break;
2093 	case BIO_GETATTR:
2094 		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2095 			return;
2096 		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2097 			return;
2098 		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2099 			return;
2100 		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2101 			return;
2102 		if (g_handleattr_str(bp, "PART::scheme",
2103 		    table->gpt_scheme->name))
2104 			return;
2105 		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2106 			/*
2107 			 * Check that the partition is suitable for kernel
2108 			 * dumps. Typically only swap partitions should be
2109 			 * used.
2110 			 */
2111 			if (!G_PART_DUMPTO(table, entry)) {
2112 				g_io_deliver(bp, ENODEV);
2113 				printf("GEOM_PART: Partition '%s' not suitable"
2114 				    " for kernel dumps (wrong type?)\n",
2115 				    pp->name);
2116 				return;
2117 			}
2118 			gkd = (struct g_kerneldump *)bp->bio_data;
2119 			if (gkd->offset >= pp->mediasize) {
2120 				g_io_deliver(bp, EIO);
2121 				return;
2122 			}
2123 			if (gkd->offset + gkd->length > pp->mediasize)
2124 				gkd->length = pp->mediasize - gkd->offset;
2125 			gkd->offset += entry->gpe_offset;
2126 		}
2127 		break;
2128 	default:
2129 		g_io_deliver(bp, EOPNOTSUPP);
2130 		return;
2131 	}
2132 
2133 	bp2 = g_clone_bio(bp);
2134 	if (bp2 == NULL) {
2135 		g_io_deliver(bp, ENOMEM);
2136 		return;
2137 	}
2138 	bp2->bio_done = g_std_done;
2139 	g_io_request(bp2, cp);
2140 }
2141 
2142 static void
2143 g_part_init(struct g_class *mp)
2144 {
2145 
2146 	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2147 }
2148 
2149 static void
2150 g_part_fini(struct g_class *mp)
2151 {
2152 
2153 	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2154 }
2155 
2156 static void
2157 g_part_unload_event(void *arg, int flag)
2158 {
2159 	struct g_consumer *cp;
2160 	struct g_geom *gp;
2161 	struct g_provider *pp;
2162 	struct g_part_scheme *scheme;
2163 	struct g_part_table *table;
2164 	uintptr_t *xchg;
2165 	int acc, error;
2166 
2167 	if (flag == EV_CANCEL)
2168 		return;
2169 
2170 	xchg = arg;
2171 	error = 0;
2172 	scheme = (void *)(*xchg);
2173 
2174 	g_topology_assert();
2175 
2176 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2177 		table = gp->softc;
2178 		if (table->gpt_scheme != scheme)
2179 			continue;
2180 
2181 		acc = 0;
2182 		LIST_FOREACH(pp, &gp->provider, provider)
2183 			acc += pp->acr + pp->acw + pp->ace;
2184 		LIST_FOREACH(cp, &gp->consumer, consumer)
2185 			acc += cp->acr + cp->acw + cp->ace;
2186 
2187 		if (!acc)
2188 			g_part_wither(gp, ENOSYS);
2189 		else
2190 			error = EBUSY;
2191 	}
2192 
2193 	if (!error)
2194 		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2195 
2196 	*xchg = error;
2197 }
2198 
2199 int
2200 g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2201 {
2202 	uintptr_t arg;
2203 	int error;
2204 
2205 	switch (type) {
2206 	case MOD_LOAD:
2207 		TAILQ_INSERT_TAIL(&g_part_schemes, scheme, scheme_list);
2208 
2209 		error = g_retaste(&g_part_class);
2210 		if (error)
2211 			TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2212 		break;
2213 	case MOD_UNLOAD:
2214 		arg = (uintptr_t)scheme;
2215 		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2216 		    NULL);
2217 		if (!error)
2218 			error = (arg == (uintptr_t)scheme) ? EDOOFUS : arg;
2219 		break;
2220 	default:
2221 		error = EOPNOTSUPP;
2222 		break;
2223 	}
2224 
2225 	return (error);
2226 }
2227