xref: /freebsd/sys/geom/part/g_part.c (revision 6486b015fc84e96725fef22b0e3363351399ae83)
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 	{ "vmware-vmfs", G_PART_ALIAS_VMFS },
107 	{ "vmware-vmkdiag", G_PART_ALIAS_VMKDIAG },
108 	{ "vmware-reserved", G_PART_ALIAS_VMRESERVED },
109 };
110 
111 SYSCTL_DECL(_kern_geom);
112 SYSCTL_NODE(_kern_geom, OID_AUTO, part, CTLFLAG_RW, 0,
113     "GEOM_PART stuff");
114 static u_int check_integrity = 1;
115 TUNABLE_INT("kern.geom.part.check_integrity", &check_integrity);
116 SYSCTL_UINT(_kern_geom_part, OID_AUTO, check_integrity,
117     CTLFLAG_RW | CTLFLAG_TUN, &check_integrity, 1,
118     "Enable integrity checking");
119 
120 /*
121  * The GEOM partitioning class.
122  */
123 static g_ctl_req_t g_part_ctlreq;
124 static g_ctl_destroy_geom_t g_part_destroy_geom;
125 static g_fini_t g_part_fini;
126 static g_init_t g_part_init;
127 static g_taste_t g_part_taste;
128 
129 static g_access_t g_part_access;
130 static g_dumpconf_t g_part_dumpconf;
131 static g_orphan_t g_part_orphan;
132 static g_spoiled_t g_part_spoiled;
133 static g_start_t g_part_start;
134 
135 static struct g_class g_part_class = {
136 	.name = "PART",
137 	.version = G_VERSION,
138 	/* Class methods. */
139 	.ctlreq = g_part_ctlreq,
140 	.destroy_geom = g_part_destroy_geom,
141 	.fini = g_part_fini,
142 	.init = g_part_init,
143 	.taste = g_part_taste,
144 	/* Geom methods. */
145 	.access = g_part_access,
146 	.dumpconf = g_part_dumpconf,
147 	.orphan = g_part_orphan,
148 	.spoiled = g_part_spoiled,
149 	.start = g_part_start,
150 };
151 
152 DECLARE_GEOM_CLASS(g_part_class, g_part);
153 MODULE_VERSION(g_part, 0);
154 
155 /*
156  * Support functions.
157  */
158 
159 static void g_part_wither(struct g_geom *, int);
160 
161 const char *
162 g_part_alias_name(enum g_part_alias alias)
163 {
164 	int i;
165 
166 	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
167 		if (g_part_alias_list[i].alias != alias)
168 			continue;
169 		return (g_part_alias_list[i].lexeme);
170 	}
171 
172 	return (NULL);
173 }
174 
175 void
176 g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
177     u_int *bestheads)
178 {
179 	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
180 	off_t chs, cylinders;
181 	u_int heads;
182 	int idx;
183 
184 	*bestchs = 0;
185 	*bestheads = 0;
186 	for (idx = 0; candidate_heads[idx] != 0; idx++) {
187 		heads = candidate_heads[idx];
188 		cylinders = blocks / heads / sectors;
189 		if (cylinders < heads || cylinders < sectors)
190 			break;
191 		if (cylinders > 1023)
192 			continue;
193 		chs = cylinders * heads * sectors;
194 		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
195 			*bestchs = chs;
196 			*bestheads = heads;
197 		}
198 	}
199 }
200 
201 static void
202 g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
203     off_t blocks)
204 {
205 	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
206 	off_t chs, bestchs;
207 	u_int heads, sectors;
208 	int idx;
209 
210 	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
211 	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
212 		table->gpt_fixgeom = 0;
213 		table->gpt_heads = 0;
214 		table->gpt_sectors = 0;
215 		bestchs = 0;
216 		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
217 			sectors = candidate_sectors[idx];
218 			g_part_geometry_heads(blocks, sectors, &chs, &heads);
219 			if (chs == 0)
220 				continue;
221 			/*
222 			 * Prefer a geometry with sectors > 1, but only if
223 			 * it doesn't bump down the number of heads to 1.
224 			 */
225 			if (chs > bestchs || (chs == bestchs && heads > 1 &&
226 			    table->gpt_sectors == 1)) {
227 				bestchs = chs;
228 				table->gpt_heads = heads;
229 				table->gpt_sectors = sectors;
230 			}
231 		}
232 		/*
233 		 * If we didn't find a geometry at all, then the disk is
234 		 * too big. This means we can use the maximum number of
235 		 * heads and sectors.
236 		 */
237 		if (bestchs == 0) {
238 			table->gpt_heads = 255;
239 			table->gpt_sectors = 63;
240 		}
241 	} else {
242 		table->gpt_fixgeom = 1;
243 		table->gpt_heads = heads;
244 		table->gpt_sectors = sectors;
245 	}
246 }
247 
248 #define	DPRINTF(...)	if (bootverbose) {	\
249 	printf("GEOM_PART: " __VA_ARGS__);	\
250 }
251 
252 static int
253 g_part_check_integrity(struct g_part_table *table, struct g_consumer *cp)
254 {
255 	struct g_part_entry *e1, *e2;
256 	struct g_provider *pp;
257 	off_t offset;
258 	int failed;
259 
260 	failed = 0;
261 	pp = cp->provider;
262 	if (table->gpt_last < table->gpt_first) {
263 		DPRINTF("last LBA is below first LBA: %jd < %jd\n",
264 		    (intmax_t)table->gpt_last, (intmax_t)table->gpt_first);
265 		failed++;
266 	}
267 	if (table->gpt_last > pp->mediasize / pp->sectorsize - 1) {
268 		DPRINTF("last LBA extends beyond mediasize: "
269 		    "%jd > %jd\n", (intmax_t)table->gpt_last,
270 		    (intmax_t)pp->mediasize / pp->sectorsize - 1);
271 		failed++;
272 	}
273 	LIST_FOREACH(e1, &table->gpt_entry, gpe_entry) {
274 		if (e1->gpe_deleted || e1->gpe_internal)
275 			continue;
276 		if (e1->gpe_start < table->gpt_first) {
277 			DPRINTF("partition %d has start offset below first "
278 			    "LBA: %jd < %jd\n", e1->gpe_index,
279 			    (intmax_t)e1->gpe_start,
280 			    (intmax_t)table->gpt_first);
281 			failed++;
282 		}
283 		if (e1->gpe_start > table->gpt_last) {
284 			DPRINTF("partition %d has start offset beyond last "
285 			    "LBA: %jd > %jd\n", e1->gpe_index,
286 			    (intmax_t)e1->gpe_start,
287 			    (intmax_t)table->gpt_last);
288 			failed++;
289 		}
290 		if (e1->gpe_end < e1->gpe_start) {
291 			DPRINTF("partition %d has end offset below start "
292 			    "offset: %jd < %jd\n", e1->gpe_index,
293 			    (intmax_t)e1->gpe_end,
294 			    (intmax_t)e1->gpe_start);
295 			failed++;
296 		}
297 		if (e1->gpe_end > table->gpt_last) {
298 			DPRINTF("partition %d has end offset beyond last "
299 			    "LBA: %jd > %jd\n", e1->gpe_index,
300 			    (intmax_t)e1->gpe_end,
301 			    (intmax_t)table->gpt_last);
302 			failed++;
303 		}
304 		if (pp->stripesize > 0) {
305 			offset = e1->gpe_start * pp->sectorsize;
306 			if (e1->gpe_offset > offset)
307 				offset = e1->gpe_offset;
308 			if ((offset + pp->stripeoffset) % pp->stripesize) {
309 				DPRINTF("partition %d is not aligned on %u "
310 				    "bytes\n", e1->gpe_index, pp->stripesize);
311 				/* Don't treat this as a critical failure */
312 			}
313 		}
314 		e2 = e1;
315 		while ((e2 = LIST_NEXT(e2, gpe_entry)) != NULL) {
316 			if (e2->gpe_deleted || e2->gpe_internal)
317 				continue;
318 			if (e1->gpe_start >= e2->gpe_start &&
319 			    e1->gpe_start <= e2->gpe_end) {
320 				DPRINTF("partition %d has start offset inside "
321 				    "partition %d: start[%d] %jd >= start[%d] "
322 				    "%jd <= end[%d] %jd\n",
323 				    e1->gpe_index, e2->gpe_index,
324 				    e2->gpe_index, (intmax_t)e2->gpe_start,
325 				    e1->gpe_index, (intmax_t)e1->gpe_start,
326 				    e2->gpe_index, (intmax_t)e2->gpe_end);
327 				failed++;
328 			}
329 			if (e1->gpe_end >= e2->gpe_start &&
330 			    e1->gpe_end <= e2->gpe_end) {
331 				DPRINTF("partition %d has end offset inside "
332 				    "partition %d: start[%d] %jd >= end[%d] "
333 				    "%jd <= end[%d] %jd\n",
334 				    e1->gpe_index, e2->gpe_index,
335 				    e2->gpe_index, (intmax_t)e2->gpe_start,
336 				    e1->gpe_index, (intmax_t)e1->gpe_end,
337 				    e2->gpe_index, (intmax_t)e2->gpe_end);
338 				failed++;
339 			}
340 			if (e1->gpe_start < e2->gpe_start &&
341 			    e1->gpe_end > e2->gpe_end) {
342 				DPRINTF("partition %d contains partition %d: "
343 				    "start[%d] %jd > start[%d] %jd, end[%d] "
344 				    "%jd < end[%d] %jd\n",
345 				    e1->gpe_index, e2->gpe_index,
346 				    e1->gpe_index, (intmax_t)e1->gpe_start,
347 				    e2->gpe_index, (intmax_t)e2->gpe_start,
348 				    e2->gpe_index, (intmax_t)e2->gpe_end,
349 				    e1->gpe_index, (intmax_t)e1->gpe_end);
350 				failed++;
351 			}
352 		}
353 	}
354 	if (failed != 0) {
355 		printf("GEOM_PART: integrity check failed (%s, %s)\n",
356 		    pp->name, table->gpt_scheme->name);
357 		if (check_integrity != 0)
358 			return (EINVAL);
359 		table->gpt_corrupt = 1;
360 	}
361 	return (0);
362 }
363 #undef	DPRINTF
364 
365 struct g_part_entry *
366 g_part_new_entry(struct g_part_table *table, int index, quad_t start,
367     quad_t end)
368 {
369 	struct g_part_entry *entry, *last;
370 
371 	last = NULL;
372 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
373 		if (entry->gpe_index == index)
374 			break;
375 		if (entry->gpe_index > index) {
376 			entry = NULL;
377 			break;
378 		}
379 		last = entry;
380 	}
381 	if (entry == NULL) {
382 		entry = g_malloc(table->gpt_scheme->gps_entrysz,
383 		    M_WAITOK | M_ZERO);
384 		entry->gpe_index = index;
385 		if (last == NULL)
386 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
387 		else
388 			LIST_INSERT_AFTER(last, entry, gpe_entry);
389 	} else
390 		entry->gpe_offset = 0;
391 	entry->gpe_start = start;
392 	entry->gpe_end = end;
393 	return (entry);
394 }
395 
396 static void
397 g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
398     struct g_part_entry *entry)
399 {
400 	struct g_consumer *cp;
401 	struct g_provider *pp;
402 	struct sbuf *sb;
403 	off_t offset;
404 
405 	cp = LIST_FIRST(&gp->consumer);
406 	pp = cp->provider;
407 
408 	offset = entry->gpe_start * pp->sectorsize;
409 	if (entry->gpe_offset < offset)
410 		entry->gpe_offset = offset;
411 
412 	if (entry->gpe_pp == NULL) {
413 		sb = sbuf_new_auto();
414 		G_PART_FULLNAME(table, entry, sb, gp->name);
415 		sbuf_finish(sb);
416 		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
417 		sbuf_delete(sb);
418 		entry->gpe_pp->private = entry;		/* Close the circle. */
419 	}
420 	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
421 	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
422 	    pp->sectorsize;
423 	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
424 	entry->gpe_pp->sectorsize = pp->sectorsize;
425 	entry->gpe_pp->flags = pp->flags & G_PF_CANDELETE;
426 	entry->gpe_pp->stripesize = pp->stripesize;
427 	entry->gpe_pp->stripeoffset = pp->stripeoffset + entry->gpe_offset;
428 	if (pp->stripesize > 0)
429 		entry->gpe_pp->stripeoffset %= pp->stripesize;
430 	g_error_provider(entry->gpe_pp, 0);
431 }
432 
433 static struct g_geom*
434 g_part_find_geom(const char *name)
435 {
436 	struct g_geom *gp;
437 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
438 		if (!strcmp(name, gp->name))
439 			break;
440 	}
441 	return (gp);
442 }
443 
444 static int
445 g_part_parm_geom(struct gctl_req *req, const char *name, struct g_geom **v)
446 {
447 	struct g_geom *gp;
448 	const char *gname;
449 
450 	gname = gctl_get_asciiparam(req, name);
451 	if (gname == NULL)
452 		return (ENOATTR);
453 	if (strncmp(gname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
454 		gname += sizeof(_PATH_DEV) - 1;
455 	gp = g_part_find_geom(gname);
456 	if (gp == NULL) {
457 		gctl_error(req, "%d %s '%s'", EINVAL, name, gname);
458 		return (EINVAL);
459 	}
460 	if ((gp->flags & G_GEOM_WITHER) != 0) {
461 		gctl_error(req, "%d %s", ENXIO, gname);
462 		return (ENXIO);
463 	}
464 	*v = gp;
465 	return (0);
466 }
467 
468 static int
469 g_part_parm_provider(struct gctl_req *req, const char *name,
470     struct g_provider **v)
471 {
472 	struct g_provider *pp;
473 	const char *pname;
474 
475 	pname = gctl_get_asciiparam(req, name);
476 	if (pname == NULL)
477 		return (ENOATTR);
478 	if (strncmp(pname, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
479 		pname += sizeof(_PATH_DEV) - 1;
480 	pp = g_provider_by_name(pname);
481 	if (pp == NULL) {
482 		gctl_error(req, "%d %s '%s'", EINVAL, name, pname);
483 		return (EINVAL);
484 	}
485 	*v = pp;
486 	return (0);
487 }
488 
489 static int
490 g_part_parm_quad(struct gctl_req *req, const char *name, quad_t *v)
491 {
492 	const char *p;
493 	char *x;
494 	quad_t q;
495 
496 	p = gctl_get_asciiparam(req, name);
497 	if (p == NULL)
498 		return (ENOATTR);
499 	q = strtoq(p, &x, 0);
500 	if (*x != '\0' || q < 0) {
501 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
502 		return (EINVAL);
503 	}
504 	*v = q;
505 	return (0);
506 }
507 
508 static int
509 g_part_parm_scheme(struct gctl_req *req, const char *name,
510     struct g_part_scheme **v)
511 {
512 	struct g_part_scheme *s;
513 	const char *p;
514 
515 	p = gctl_get_asciiparam(req, name);
516 	if (p == NULL)
517 		return (ENOATTR);
518 	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
519 		if (s == &g_part_null_scheme)
520 			continue;
521 		if (!strcasecmp(s->name, p))
522 			break;
523 	}
524 	if (s == NULL) {
525 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
526 		return (EINVAL);
527 	}
528 	*v = s;
529 	return (0);
530 }
531 
532 static int
533 g_part_parm_str(struct gctl_req *req, const char *name, const char **v)
534 {
535 	const char *p;
536 
537 	p = gctl_get_asciiparam(req, name);
538 	if (p == NULL)
539 		return (ENOATTR);
540 	/* An empty label is always valid. */
541 	if (strcmp(name, "label") != 0 && p[0] == '\0') {
542 		gctl_error(req, "%d %s '%s'", EINVAL, name, p);
543 		return (EINVAL);
544 	}
545 	*v = p;
546 	return (0);
547 }
548 
549 static int
550 g_part_parm_intmax(struct gctl_req *req, const char *name, u_int *v)
551 {
552 	const intmax_t *p;
553 	int size;
554 
555 	p = gctl_get_param(req, name, &size);
556 	if (p == NULL)
557 		return (ENOATTR);
558 	if (size != sizeof(*p) || *p < 0 || *p > INT_MAX) {
559 		gctl_error(req, "%d %s '%jd'", EINVAL, name, *p);
560 		return (EINVAL);
561 	}
562 	*v = (u_int)*p;
563 	return (0);
564 }
565 
566 static int
567 g_part_parm_uint32(struct gctl_req *req, const char *name, u_int *v)
568 {
569 	const uint32_t *p;
570 	int size;
571 
572 	p = gctl_get_param(req, name, &size);
573 	if (p == NULL)
574 		return (ENOATTR);
575 	if (size != sizeof(*p) || *p > INT_MAX) {
576 		gctl_error(req, "%d %s '%u'", EINVAL, name, (unsigned int)*p);
577 		return (EINVAL);
578 	}
579 	*v = (u_int)*p;
580 	return (0);
581 }
582 
583 static int
584 g_part_parm_bootcode(struct gctl_req *req, const char *name, const void **v,
585     unsigned int *s)
586 {
587 	const void *p;
588 	int size;
589 
590 	p = gctl_get_param(req, name, &size);
591 	if (p == NULL)
592 		return (ENOATTR);
593 	*v = p;
594 	*s = size;
595 	return (0);
596 }
597 
598 static int
599 g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
600 {
601 	struct g_part_scheme *iter, *scheme;
602 	struct g_part_table *table;
603 	int pri, probe;
604 
605 	table = gp->softc;
606 	scheme = (table != NULL) ? table->gpt_scheme : NULL;
607 	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
608 	if (pri == 0)
609 		goto done;
610 	if (pri > 0) {	/* error */
611 		scheme = NULL;
612 		pri = INT_MIN;
613 	}
614 
615 	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
616 		if (iter == &g_part_null_scheme)
617 			continue;
618 		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
619 		    M_WAITOK);
620 		table->gpt_gp = gp;
621 		table->gpt_scheme = iter;
622 		table->gpt_depth = depth;
623 		probe = G_PART_PROBE(table, cp);
624 		if (probe <= 0 && probe > pri) {
625 			pri = probe;
626 			scheme = iter;
627 			if (gp->softc != NULL)
628 				kobj_delete((kobj_t)gp->softc, M_GEOM);
629 			gp->softc = table;
630 			if (pri == 0)
631 				goto done;
632 		} else
633 			kobj_delete((kobj_t)table, M_GEOM);
634 	}
635 
636 done:
637 	return ((scheme == NULL) ? ENXIO : 0);
638 }
639 
640 /*
641  * Control request functions.
642  */
643 
644 static int
645 g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
646 {
647 	struct g_geom *gp;
648 	struct g_provider *pp;
649 	struct g_part_entry *delent, *last, *entry;
650 	struct g_part_table *table;
651 	struct sbuf *sb;
652 	quad_t end;
653 	unsigned int index;
654 	int error;
655 
656 	gp = gpp->gpp_geom;
657 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
658 	g_topology_assert();
659 
660 	pp = LIST_FIRST(&gp->consumer)->provider;
661 	table = gp->softc;
662 	end = gpp->gpp_start + gpp->gpp_size - 1;
663 
664 	if (gpp->gpp_start < table->gpt_first ||
665 	    gpp->gpp_start > table->gpt_last) {
666 		gctl_error(req, "%d start '%jd'", EINVAL,
667 		    (intmax_t)gpp->gpp_start);
668 		return (EINVAL);
669 	}
670 	if (end < gpp->gpp_start || end > table->gpt_last) {
671 		gctl_error(req, "%d size '%jd'", EINVAL,
672 		    (intmax_t)gpp->gpp_size);
673 		return (EINVAL);
674 	}
675 	if (gpp->gpp_index > table->gpt_entries) {
676 		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
677 		return (EINVAL);
678 	}
679 
680 	delent = last = NULL;
681 	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
682 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
683 		if (entry->gpe_deleted) {
684 			if (entry->gpe_index == index)
685 				delent = entry;
686 			continue;
687 		}
688 		if (entry->gpe_index == index)
689 			index = entry->gpe_index + 1;
690 		if (entry->gpe_index < index)
691 			last = entry;
692 		if (entry->gpe_internal)
693 			continue;
694 		if (gpp->gpp_start >= entry->gpe_start &&
695 		    gpp->gpp_start <= entry->gpe_end) {
696 			gctl_error(req, "%d start '%jd'", ENOSPC,
697 			    (intmax_t)gpp->gpp_start);
698 			return (ENOSPC);
699 		}
700 		if (end >= entry->gpe_start && end <= entry->gpe_end) {
701 			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
702 			return (ENOSPC);
703 		}
704 		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
705 			gctl_error(req, "%d size '%jd'", ENOSPC,
706 			    (intmax_t)gpp->gpp_size);
707 			return (ENOSPC);
708 		}
709 	}
710 	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
711 		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
712 		return (EEXIST);
713 	}
714 	if (index > table->gpt_entries) {
715 		gctl_error(req, "%d index '%d'", ENOSPC, index);
716 		return (ENOSPC);
717 	}
718 
719 	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
720 	    M_WAITOK | M_ZERO) : delent;
721 	entry->gpe_index = index;
722 	entry->gpe_start = gpp->gpp_start;
723 	entry->gpe_end = end;
724 	error = G_PART_ADD(table, entry, gpp);
725 	if (error) {
726 		gctl_error(req, "%d", error);
727 		if (delent == NULL)
728 			g_free(entry);
729 		return (error);
730 	}
731 	if (delent == NULL) {
732 		if (last == NULL)
733 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
734 		else
735 			LIST_INSERT_AFTER(last, entry, gpe_entry);
736 		entry->gpe_created = 1;
737 	} else {
738 		entry->gpe_deleted = 0;
739 		entry->gpe_modified = 1;
740 	}
741 	g_part_new_provider(gp, table, entry);
742 
743 	/* Provide feedback if so requested. */
744 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
745 		sb = sbuf_new_auto();
746 		G_PART_FULLNAME(table, entry, sb, gp->name);
747 		if (pp->stripesize > 0 && entry->gpe_pp->stripeoffset != 0)
748 			sbuf_printf(sb, " added, but partition is not "
749 			    "aligned on %u bytes\n", pp->stripesize);
750 		else
751 			sbuf_cat(sb, " added\n");
752 		sbuf_finish(sb);
753 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
754 		sbuf_delete(sb);
755 	}
756 	return (0);
757 }
758 
759 static int
760 g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
761 {
762 	struct g_geom *gp;
763 	struct g_part_table *table;
764 	struct sbuf *sb;
765 	int error, sz;
766 
767 	gp = gpp->gpp_geom;
768 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
769 	g_topology_assert();
770 
771 	table = gp->softc;
772 	sz = table->gpt_scheme->gps_bootcodesz;
773 	if (sz == 0) {
774 		error = ENODEV;
775 		goto fail;
776 	}
777 	if (gpp->gpp_codesize > sz) {
778 		error = EFBIG;
779 		goto fail;
780 	}
781 
782 	error = G_PART_BOOTCODE(table, gpp);
783 	if (error)
784 		goto fail;
785 
786 	/* Provide feedback if so requested. */
787 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
788 		sb = sbuf_new_auto();
789 		sbuf_printf(sb, "bootcode written to %s\n", gp->name);
790 		sbuf_finish(sb);
791 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
792 		sbuf_delete(sb);
793 	}
794 	return (0);
795 
796  fail:
797 	gctl_error(req, "%d", error);
798 	return (error);
799 }
800 
801 static int
802 g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
803 {
804 	struct g_consumer *cp;
805 	struct g_geom *gp;
806 	struct g_provider *pp;
807 	struct g_part_entry *entry, *tmp;
808 	struct g_part_table *table;
809 	char *buf;
810 	int error, i;
811 
812 	gp = gpp->gpp_geom;
813 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
814 	g_topology_assert();
815 
816 	table = gp->softc;
817 	if (!table->gpt_opened) {
818 		gctl_error(req, "%d", EPERM);
819 		return (EPERM);
820 	}
821 
822 	g_topology_unlock();
823 
824 	cp = LIST_FIRST(&gp->consumer);
825 	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
826 		pp = cp->provider;
827 		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
828 		while (table->gpt_smhead != 0) {
829 			i = ffs(table->gpt_smhead) - 1;
830 			error = g_write_data(cp, i * pp->sectorsize, buf,
831 			    pp->sectorsize);
832 			if (error) {
833 				g_free(buf);
834 				goto fail;
835 			}
836 			table->gpt_smhead &= ~(1 << i);
837 		}
838 		while (table->gpt_smtail != 0) {
839 			i = ffs(table->gpt_smtail) - 1;
840 			error = g_write_data(cp, pp->mediasize - (i + 1) *
841 			    pp->sectorsize, buf, pp->sectorsize);
842 			if (error) {
843 				g_free(buf);
844 				goto fail;
845 			}
846 			table->gpt_smtail &= ~(1 << i);
847 		}
848 		g_free(buf);
849 	}
850 
851 	if (table->gpt_scheme == &g_part_null_scheme) {
852 		g_topology_lock();
853 		g_access(cp, -1, -1, -1);
854 		g_part_wither(gp, ENXIO);
855 		return (0);
856 	}
857 
858 	error = G_PART_WRITE(table, cp);
859 	if (error)
860 		goto fail;
861 
862 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
863 		if (!entry->gpe_deleted) {
864 			entry->gpe_created = 0;
865 			entry->gpe_modified = 0;
866 			continue;
867 		}
868 		LIST_REMOVE(entry, gpe_entry);
869 		g_free(entry);
870 	}
871 	table->gpt_created = 0;
872 	table->gpt_opened = 0;
873 
874 	g_topology_lock();
875 	g_access(cp, -1, -1, -1);
876 	return (0);
877 
878 fail:
879 	g_topology_lock();
880 	gctl_error(req, "%d", error);
881 	return (error);
882 }
883 
884 static int
885 g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
886 {
887 	struct g_consumer *cp;
888 	struct g_geom *gp;
889 	struct g_provider *pp;
890 	struct g_part_scheme *scheme;
891 	struct g_part_table *null, *table;
892 	struct sbuf *sb;
893 	int attr, error;
894 
895 	pp = gpp->gpp_provider;
896 	scheme = gpp->gpp_scheme;
897 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
898 	g_topology_assert();
899 
900 	/* Check that there isn't already a g_part geom on the provider. */
901 	gp = g_part_find_geom(pp->name);
902 	if (gp != NULL) {
903 		null = gp->softc;
904 		if (null->gpt_scheme != &g_part_null_scheme) {
905 			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
906 			return (EEXIST);
907 		}
908 	} else
909 		null = NULL;
910 
911 	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
912 	    (gpp->gpp_entries < scheme->gps_minent ||
913 	     gpp->gpp_entries > scheme->gps_maxent)) {
914 		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
915 		return (EINVAL);
916 	}
917 
918 	if (null == NULL)
919 		gp = g_new_geomf(&g_part_class, "%s", pp->name);
920 	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
921 	    M_WAITOK);
922 	table = gp->softc;
923 	table->gpt_gp = gp;
924 	table->gpt_scheme = gpp->gpp_scheme;
925 	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
926 	    gpp->gpp_entries : scheme->gps_minent;
927 	LIST_INIT(&table->gpt_entry);
928 	if (null == NULL) {
929 		cp = g_new_consumer(gp);
930 		error = g_attach(cp, pp);
931 		if (error == 0)
932 			error = g_access(cp, 1, 1, 1);
933 		if (error != 0) {
934 			g_part_wither(gp, error);
935 			gctl_error(req, "%d geom '%s'", error, pp->name);
936 			return (error);
937 		}
938 		table->gpt_opened = 1;
939 	} else {
940 		cp = LIST_FIRST(&gp->consumer);
941 		table->gpt_opened = null->gpt_opened;
942 		table->gpt_smhead = null->gpt_smhead;
943 		table->gpt_smtail = null->gpt_smtail;
944 	}
945 
946 	g_topology_unlock();
947 
948 	/* Make sure the provider has media. */
949 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
950 		error = ENODEV;
951 		goto fail;
952 	}
953 
954 	/* Make sure we can nest and if so, determine our depth. */
955 	error = g_getattr("PART::isleaf", cp, &attr);
956 	if (!error && attr) {
957 		error = ENODEV;
958 		goto fail;
959 	}
960 	error = g_getattr("PART::depth", cp, &attr);
961 	table->gpt_depth = (!error) ? attr + 1 : 0;
962 
963 	/*
964 	 * Synthesize a disk geometry. Some partitioning schemes
965 	 * depend on it and since some file systems need it even
966 	 * when the partitition scheme doesn't, we do it here in
967 	 * scheme-independent code.
968 	 */
969 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
970 
971 	error = G_PART_CREATE(table, gpp);
972 	if (error)
973 		goto fail;
974 
975 	g_topology_lock();
976 
977 	table->gpt_created = 1;
978 	if (null != NULL)
979 		kobj_delete((kobj_t)null, M_GEOM);
980 
981 	/*
982 	 * Support automatic commit by filling in the gpp_geom
983 	 * parameter.
984 	 */
985 	gpp->gpp_parms |= G_PART_PARM_GEOM;
986 	gpp->gpp_geom = gp;
987 
988 	/* Provide feedback if so requested. */
989 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
990 		sb = sbuf_new_auto();
991 		sbuf_printf(sb, "%s created\n", gp->name);
992 		sbuf_finish(sb);
993 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
994 		sbuf_delete(sb);
995 	}
996 	return (0);
997 
998 fail:
999 	g_topology_lock();
1000 	if (null == NULL) {
1001 		g_access(cp, -1, -1, -1);
1002 		g_part_wither(gp, error);
1003 	} else {
1004 		kobj_delete((kobj_t)gp->softc, M_GEOM);
1005 		gp->softc = null;
1006 	}
1007 	gctl_error(req, "%d provider", error);
1008 	return (error);
1009 }
1010 
1011 static int
1012 g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
1013 {
1014 	struct g_geom *gp;
1015 	struct g_provider *pp;
1016 	struct g_part_entry *entry;
1017 	struct g_part_table *table;
1018 	struct sbuf *sb;
1019 
1020 	gp = gpp->gpp_geom;
1021 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1022 	g_topology_assert();
1023 
1024 	table = gp->softc;
1025 
1026 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1027 		if (entry->gpe_deleted || entry->gpe_internal)
1028 			continue;
1029 		if (entry->gpe_index == gpp->gpp_index)
1030 			break;
1031 	}
1032 	if (entry == NULL) {
1033 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1034 		return (ENOENT);
1035 	}
1036 
1037 	pp = entry->gpe_pp;
1038 	if (pp != NULL) {
1039 		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
1040 			gctl_error(req, "%d", EBUSY);
1041 			return (EBUSY);
1042 		}
1043 
1044 		pp->private = NULL;
1045 		entry->gpe_pp = NULL;
1046 	}
1047 
1048 	if (pp != NULL)
1049 		g_wither_provider(pp, ENXIO);
1050 
1051 	/* Provide feedback if so requested. */
1052 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1053 		sb = sbuf_new_auto();
1054 		G_PART_FULLNAME(table, entry, sb, gp->name);
1055 		sbuf_cat(sb, " deleted\n");
1056 		sbuf_finish(sb);
1057 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1058 		sbuf_delete(sb);
1059 	}
1060 
1061 	if (entry->gpe_created) {
1062 		LIST_REMOVE(entry, gpe_entry);
1063 		g_free(entry);
1064 	} else {
1065 		entry->gpe_modified = 0;
1066 		entry->gpe_deleted = 1;
1067 	}
1068 	return (0);
1069 }
1070 
1071 static int
1072 g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
1073 {
1074 	struct g_consumer *cp;
1075 	struct g_geom *gp;
1076 	struct g_provider *pp;
1077 	struct g_part_entry *entry, *tmp;
1078 	struct g_part_table *null, *table;
1079 	struct sbuf *sb;
1080 	int error;
1081 
1082 	gp = gpp->gpp_geom;
1083 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1084 	g_topology_assert();
1085 
1086 	table = gp->softc;
1087 	/* Check for busy providers. */
1088 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1089 		if (entry->gpe_deleted || entry->gpe_internal)
1090 			continue;
1091 		if (gpp->gpp_force) {
1092 			pp = entry->gpe_pp;
1093 			if (pp == NULL)
1094 				continue;
1095 			if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
1096 				continue;
1097 		}
1098 		gctl_error(req, "%d", EBUSY);
1099 		return (EBUSY);
1100 	}
1101 
1102 	if (gpp->gpp_force) {
1103 		/* Destroy all providers. */
1104 		LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1105 			pp = entry->gpe_pp;
1106 			if (pp != NULL) {
1107 				pp->private = NULL;
1108 				g_wither_provider(pp, ENXIO);
1109 			}
1110 			LIST_REMOVE(entry, gpe_entry);
1111 			g_free(entry);
1112 		}
1113 	}
1114 
1115 	error = G_PART_DESTROY(table, gpp);
1116 	if (error) {
1117 		gctl_error(req, "%d", error);
1118 		return (error);
1119 	}
1120 
1121 	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
1122 	    M_WAITOK);
1123 	null = gp->softc;
1124 	null->gpt_gp = gp;
1125 	null->gpt_scheme = &g_part_null_scheme;
1126 	LIST_INIT(&null->gpt_entry);
1127 
1128 	cp = LIST_FIRST(&gp->consumer);
1129 	pp = cp->provider;
1130 	null->gpt_last = pp->mediasize / pp->sectorsize - 1;
1131 
1132 	null->gpt_depth = table->gpt_depth;
1133 	null->gpt_opened = table->gpt_opened;
1134 	null->gpt_smhead = table->gpt_smhead;
1135 	null->gpt_smtail = table->gpt_smtail;
1136 
1137 	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1138 		LIST_REMOVE(entry, gpe_entry);
1139 		g_free(entry);
1140 	}
1141 	kobj_delete((kobj_t)table, M_GEOM);
1142 
1143 	/* Provide feedback if so requested. */
1144 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1145 		sb = sbuf_new_auto();
1146 		sbuf_printf(sb, "%s destroyed\n", gp->name);
1147 		sbuf_finish(sb);
1148 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1149 		sbuf_delete(sb);
1150 	}
1151 	return (0);
1152 }
1153 
1154 static int
1155 g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
1156 {
1157 	struct g_geom *gp;
1158 	struct g_part_entry *entry;
1159 	struct g_part_table *table;
1160 	struct sbuf *sb;
1161 	int error;
1162 
1163 	gp = gpp->gpp_geom;
1164 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1165 	g_topology_assert();
1166 
1167 	table = gp->softc;
1168 
1169 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1170 		if (entry->gpe_deleted || entry->gpe_internal)
1171 			continue;
1172 		if (entry->gpe_index == gpp->gpp_index)
1173 			break;
1174 	}
1175 	if (entry == NULL) {
1176 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1177 		return (ENOENT);
1178 	}
1179 
1180 	error = G_PART_MODIFY(table, entry, gpp);
1181 	if (error) {
1182 		gctl_error(req, "%d", error);
1183 		return (error);
1184 	}
1185 
1186 	if (!entry->gpe_created)
1187 		entry->gpe_modified = 1;
1188 
1189 	/* Provide feedback if so requested. */
1190 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1191 		sb = sbuf_new_auto();
1192 		G_PART_FULLNAME(table, entry, sb, gp->name);
1193 		sbuf_cat(sb, " modified\n");
1194 		sbuf_finish(sb);
1195 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1196 		sbuf_delete(sb);
1197 	}
1198 	return (0);
1199 }
1200 
1201 static int
1202 g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
1203 {
1204 	gctl_error(req, "%d verb 'move'", ENOSYS);
1205 	return (ENOSYS);
1206 }
1207 
1208 static int
1209 g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
1210 {
1211 	struct g_part_table *table;
1212 	struct g_geom *gp;
1213 	struct sbuf *sb;
1214 	int error, recovered;
1215 
1216 	gp = gpp->gpp_geom;
1217 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1218 	g_topology_assert();
1219 	table = gp->softc;
1220 	error = recovered = 0;
1221 
1222 	if (table->gpt_corrupt) {
1223 		error = G_PART_RECOVER(table);
1224 		if (error == 0)
1225 			error = g_part_check_integrity(table,
1226 			    LIST_FIRST(&gp->consumer));
1227 		if (error) {
1228 			gctl_error(req, "%d recovering '%s' failed",
1229 			    error, gp->name);
1230 			return (error);
1231 		}
1232 		recovered = 1;
1233 	}
1234 	/* Provide feedback if so requested. */
1235 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1236 		sb = sbuf_new_auto();
1237 		if (recovered)
1238 			sbuf_printf(sb, "%s recovered\n", gp->name);
1239 		else
1240 			sbuf_printf(sb, "%s recovering is not needed\n",
1241 			    gp->name);
1242 		sbuf_finish(sb);
1243 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1244 		sbuf_delete(sb);
1245 	}
1246 	return (0);
1247 }
1248 
1249 static int
1250 g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
1251 {
1252 	struct g_geom *gp;
1253 	struct g_provider *pp;
1254 	struct g_part_entry *pe, *entry;
1255 	struct g_part_table *table;
1256 	struct sbuf *sb;
1257 	quad_t end;
1258 	int error;
1259 
1260 	gp = gpp->gpp_geom;
1261 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1262 	g_topology_assert();
1263 	table = gp->softc;
1264 
1265 	/* check gpp_index */
1266 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1267 		if (entry->gpe_deleted || entry->gpe_internal)
1268 			continue;
1269 		if (entry->gpe_index == gpp->gpp_index)
1270 			break;
1271 	}
1272 	if (entry == NULL) {
1273 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1274 		return (ENOENT);
1275 	}
1276 
1277 	/* check gpp_size */
1278 	end = entry->gpe_start + gpp->gpp_size - 1;
1279 	if (gpp->gpp_size < 1 || end > table->gpt_last) {
1280 		gctl_error(req, "%d size '%jd'", EINVAL,
1281 		    (intmax_t)gpp->gpp_size);
1282 		return (EINVAL);
1283 	}
1284 
1285 	LIST_FOREACH(pe, &table->gpt_entry, gpe_entry) {
1286 		if (pe->gpe_deleted || pe->gpe_internal || pe == entry)
1287 			continue;
1288 		if (end >= pe->gpe_start && end <= pe->gpe_end) {
1289 			gctl_error(req, "%d end '%jd'", ENOSPC,
1290 			    (intmax_t)end);
1291 			return (ENOSPC);
1292 		}
1293 		if (entry->gpe_start < pe->gpe_start && end > pe->gpe_end) {
1294 			gctl_error(req, "%d size '%jd'", ENOSPC,
1295 			    (intmax_t)gpp->gpp_size);
1296 			return (ENOSPC);
1297 		}
1298 	}
1299 
1300 	pp = entry->gpe_pp;
1301 	if ((g_debugflags & 16) == 0 &&
1302 	    (pp->acr > 0 || pp->acw > 0 || pp->ace > 0)) {
1303 		gctl_error(req, "%d", EBUSY);
1304 		return (EBUSY);
1305 	}
1306 
1307 	error = G_PART_RESIZE(table, entry, gpp);
1308 	if (error) {
1309 		gctl_error(req, "%d", error);
1310 		return (error);
1311 	}
1312 
1313 	if (!entry->gpe_created)
1314 		entry->gpe_modified = 1;
1315 
1316 	/* update mediasize of changed provider */
1317 	pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
1318 		pp->sectorsize;
1319 
1320 	/* Provide feedback if so requested. */
1321 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1322 		sb = sbuf_new_auto();
1323 		G_PART_FULLNAME(table, entry, sb, gp->name);
1324 		sbuf_cat(sb, " resized\n");
1325 		sbuf_finish(sb);
1326 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1327 		sbuf_delete(sb);
1328 	}
1329 	return (0);
1330 }
1331 
1332 static int
1333 g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
1334     unsigned int set)
1335 {
1336 	struct g_geom *gp;
1337 	struct g_part_entry *entry;
1338 	struct g_part_table *table;
1339 	struct sbuf *sb;
1340 	int error;
1341 
1342 	gp = gpp->gpp_geom;
1343 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1344 	g_topology_assert();
1345 
1346 	table = gp->softc;
1347 
1348 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1349 		if (entry->gpe_deleted || entry->gpe_internal)
1350 			continue;
1351 		if (entry->gpe_index == gpp->gpp_index)
1352 			break;
1353 	}
1354 	if (entry == NULL) {
1355 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
1356 		return (ENOENT);
1357 	}
1358 
1359 	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
1360 	if (error) {
1361 		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
1362 		return (error);
1363 	}
1364 
1365 	/* Provide feedback if so requested. */
1366 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1367 		sb = sbuf_new_auto();
1368 		sbuf_printf(sb, "%s %sset on ", gpp->gpp_attrib,
1369 		    (set) ? "" : "un");
1370 		G_PART_FULLNAME(table, entry, sb, gp->name);
1371 		sbuf_printf(sb, "\n");
1372 		sbuf_finish(sb);
1373 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1374 		sbuf_delete(sb);
1375 	}
1376 	return (0);
1377 }
1378 
1379 static int
1380 g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1381 {
1382 	struct g_consumer *cp;
1383 	struct g_provider *pp;
1384 	struct g_geom *gp;
1385 	struct g_part_entry *entry, *tmp;
1386 	struct g_part_table *table;
1387 	int error, reprobe;
1388 
1389 	gp = gpp->gpp_geom;
1390 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1391 	g_topology_assert();
1392 
1393 	table = gp->softc;
1394 	if (!table->gpt_opened) {
1395 		gctl_error(req, "%d", EPERM);
1396 		return (EPERM);
1397 	}
1398 
1399 	cp = LIST_FIRST(&gp->consumer);
1400 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1401 		entry->gpe_modified = 0;
1402 		if (entry->gpe_created) {
1403 			pp = entry->gpe_pp;
1404 			if (pp != NULL) {
1405 				pp->private = NULL;
1406 				entry->gpe_pp = NULL;
1407 				g_wither_provider(pp, ENXIO);
1408 			}
1409 			entry->gpe_deleted = 1;
1410 		}
1411 		if (entry->gpe_deleted) {
1412 			LIST_REMOVE(entry, gpe_entry);
1413 			g_free(entry);
1414 		}
1415 	}
1416 
1417 	g_topology_unlock();
1418 
1419 	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1420 	    table->gpt_created) ? 1 : 0;
1421 
1422 	if (reprobe) {
1423 		LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1424 			if (entry->gpe_internal)
1425 				continue;
1426 			error = EBUSY;
1427 			goto fail;
1428 		}
1429 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1430 			LIST_REMOVE(entry, gpe_entry);
1431 			g_free(entry);
1432 		}
1433 		error = g_part_probe(gp, cp, table->gpt_depth);
1434 		if (error) {
1435 			g_topology_lock();
1436 			g_access(cp, -1, -1, -1);
1437 			g_part_wither(gp, error);
1438 			return (0);
1439 		}
1440 		table = gp->softc;
1441 
1442 		/*
1443 		 * Synthesize a disk geometry. Some partitioning schemes
1444 		 * depend on it and since some file systems need it even
1445 		 * when the partitition scheme doesn't, we do it here in
1446 		 * scheme-independent code.
1447 		 */
1448 		pp = cp->provider;
1449 		g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1450 	}
1451 
1452 	error = G_PART_READ(table, cp);
1453 	if (error)
1454 		goto fail;
1455 	error = g_part_check_integrity(table, cp);
1456 	if (error)
1457 		goto fail;
1458 
1459 	g_topology_lock();
1460 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1461 		if (!entry->gpe_internal)
1462 			g_part_new_provider(gp, table, entry);
1463 	}
1464 
1465 	table->gpt_opened = 0;
1466 	g_access(cp, -1, -1, -1);
1467 	return (0);
1468 
1469 fail:
1470 	g_topology_lock();
1471 	gctl_error(req, "%d", error);
1472 	return (error);
1473 }
1474 
1475 static void
1476 g_part_wither(struct g_geom *gp, int error)
1477 {
1478 	struct g_part_entry *entry;
1479 	struct g_part_table *table;
1480 
1481 	table = gp->softc;
1482 	if (table != NULL) {
1483 		G_PART_DESTROY(table, NULL);
1484 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1485 			LIST_REMOVE(entry, gpe_entry);
1486 			g_free(entry);
1487 		}
1488 		if (gp->softc != NULL) {
1489 			kobj_delete((kobj_t)gp->softc, M_GEOM);
1490 			gp->softc = NULL;
1491 		}
1492 	}
1493 	g_wither_geom(gp, error);
1494 }
1495 
1496 /*
1497  * Class methods.
1498  */
1499 
1500 static void
1501 g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1502 {
1503 	struct g_part_parms gpp;
1504 	struct g_part_table *table;
1505 	struct gctl_req_arg *ap;
1506 	enum g_part_ctl ctlreq;
1507 	unsigned int i, mparms, oparms, parm;
1508 	int auto_commit, close_on_error;
1509 	int error, modifies;
1510 
1511 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1512 	g_topology_assert();
1513 
1514 	ctlreq = G_PART_CTL_NONE;
1515 	modifies = 1;
1516 	mparms = 0;
1517 	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1518 	switch (*verb) {
1519 	case 'a':
1520 		if (!strcmp(verb, "add")) {
1521 			ctlreq = G_PART_CTL_ADD;
1522 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1523 			    G_PART_PARM_START | G_PART_PARM_TYPE;
1524 			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1525 		}
1526 		break;
1527 	case 'b':
1528 		if (!strcmp(verb, "bootcode")) {
1529 			ctlreq = G_PART_CTL_BOOTCODE;
1530 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1531 		}
1532 		break;
1533 	case 'c':
1534 		if (!strcmp(verb, "commit")) {
1535 			ctlreq = G_PART_CTL_COMMIT;
1536 			mparms |= G_PART_PARM_GEOM;
1537 			modifies = 0;
1538 		} else if (!strcmp(verb, "create")) {
1539 			ctlreq = G_PART_CTL_CREATE;
1540 			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1541 			oparms |= G_PART_PARM_ENTRIES;
1542 		}
1543 		break;
1544 	case 'd':
1545 		if (!strcmp(verb, "delete")) {
1546 			ctlreq = G_PART_CTL_DELETE;
1547 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1548 		} else if (!strcmp(verb, "destroy")) {
1549 			ctlreq = G_PART_CTL_DESTROY;
1550 			mparms |= G_PART_PARM_GEOM;
1551 			oparms |= G_PART_PARM_FORCE;
1552 		}
1553 		break;
1554 	case 'm':
1555 		if (!strcmp(verb, "modify")) {
1556 			ctlreq = G_PART_CTL_MODIFY;
1557 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1558 			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1559 		} else if (!strcmp(verb, "move")) {
1560 			ctlreq = G_PART_CTL_MOVE;
1561 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1562 		}
1563 		break;
1564 	case 'r':
1565 		if (!strcmp(verb, "recover")) {
1566 			ctlreq = G_PART_CTL_RECOVER;
1567 			mparms |= G_PART_PARM_GEOM;
1568 		} else if (!strcmp(verb, "resize")) {
1569 			ctlreq = G_PART_CTL_RESIZE;
1570 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX |
1571 			    G_PART_PARM_SIZE;
1572 		}
1573 		break;
1574 	case 's':
1575 		if (!strcmp(verb, "set")) {
1576 			ctlreq = G_PART_CTL_SET;
1577 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1578 			    G_PART_PARM_INDEX;
1579 		}
1580 		break;
1581 	case 'u':
1582 		if (!strcmp(verb, "undo")) {
1583 			ctlreq = G_PART_CTL_UNDO;
1584 			mparms |= G_PART_PARM_GEOM;
1585 			modifies = 0;
1586 		} else if (!strcmp(verb, "unset")) {
1587 			ctlreq = G_PART_CTL_UNSET;
1588 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1589 			    G_PART_PARM_INDEX;
1590 		}
1591 		break;
1592 	}
1593 	if (ctlreq == G_PART_CTL_NONE) {
1594 		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1595 		return;
1596 	}
1597 
1598 	bzero(&gpp, sizeof(gpp));
1599 	for (i = 0; i < req->narg; i++) {
1600 		ap = &req->arg[i];
1601 		parm = 0;
1602 		switch (ap->name[0]) {
1603 		case 'a':
1604 			if (!strcmp(ap->name, "arg0")) {
1605 				parm = mparms &
1606 				    (G_PART_PARM_GEOM | G_PART_PARM_PROVIDER);
1607 			}
1608 			if (!strcmp(ap->name, "attrib"))
1609 				parm = G_PART_PARM_ATTRIB;
1610 			break;
1611 		case 'b':
1612 			if (!strcmp(ap->name, "bootcode"))
1613 				parm = G_PART_PARM_BOOTCODE;
1614 			break;
1615 		case 'c':
1616 			if (!strcmp(ap->name, "class"))
1617 				continue;
1618 			break;
1619 		case 'e':
1620 			if (!strcmp(ap->name, "entries"))
1621 				parm = G_PART_PARM_ENTRIES;
1622 			break;
1623 		case 'f':
1624 			if (!strcmp(ap->name, "flags"))
1625 				parm = G_PART_PARM_FLAGS;
1626 			else if (!strcmp(ap->name, "force"))
1627 				parm = G_PART_PARM_FORCE;
1628 			break;
1629 		case 'i':
1630 			if (!strcmp(ap->name, "index"))
1631 				parm = G_PART_PARM_INDEX;
1632 			break;
1633 		case 'l':
1634 			if (!strcmp(ap->name, "label"))
1635 				parm = G_PART_PARM_LABEL;
1636 			break;
1637 		case 'o':
1638 			if (!strcmp(ap->name, "output"))
1639 				parm = G_PART_PARM_OUTPUT;
1640 			break;
1641 		case 's':
1642 			if (!strcmp(ap->name, "scheme"))
1643 				parm = G_PART_PARM_SCHEME;
1644 			else if (!strcmp(ap->name, "size"))
1645 				parm = G_PART_PARM_SIZE;
1646 			else if (!strcmp(ap->name, "start"))
1647 				parm = G_PART_PARM_START;
1648 			break;
1649 		case 't':
1650 			if (!strcmp(ap->name, "type"))
1651 				parm = G_PART_PARM_TYPE;
1652 			break;
1653 		case 'v':
1654 			if (!strcmp(ap->name, "verb"))
1655 				continue;
1656 			else if (!strcmp(ap->name, "version"))
1657 				parm = G_PART_PARM_VERSION;
1658 			break;
1659 		}
1660 		if ((parm & (mparms | oparms)) == 0) {
1661 			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1662 			return;
1663 		}
1664 		switch (parm) {
1665 		case G_PART_PARM_ATTRIB:
1666 			error = g_part_parm_str(req, ap->name,
1667 			    &gpp.gpp_attrib);
1668 			break;
1669 		case G_PART_PARM_BOOTCODE:
1670 			error = g_part_parm_bootcode(req, ap->name,
1671 			    &gpp.gpp_codeptr, &gpp.gpp_codesize);
1672 			break;
1673 		case G_PART_PARM_ENTRIES:
1674 			error = g_part_parm_intmax(req, ap->name,
1675 			    &gpp.gpp_entries);
1676 			break;
1677 		case G_PART_PARM_FLAGS:
1678 			error = g_part_parm_str(req, ap->name, &gpp.gpp_flags);
1679 			break;
1680 		case G_PART_PARM_FORCE:
1681 			error = g_part_parm_uint32(req, ap->name,
1682 			    &gpp.gpp_force);
1683 			break;
1684 		case G_PART_PARM_GEOM:
1685 			error = g_part_parm_geom(req, ap->name, &gpp.gpp_geom);
1686 			break;
1687 		case G_PART_PARM_INDEX:
1688 			error = g_part_parm_intmax(req, ap->name,
1689 			    &gpp.gpp_index);
1690 			break;
1691 		case G_PART_PARM_LABEL:
1692 			error = g_part_parm_str(req, ap->name, &gpp.gpp_label);
1693 			break;
1694 		case G_PART_PARM_OUTPUT:
1695 			error = 0;	/* Write-only parameter */
1696 			break;
1697 		case G_PART_PARM_PROVIDER:
1698 			error = g_part_parm_provider(req, ap->name,
1699 			    &gpp.gpp_provider);
1700 			break;
1701 		case G_PART_PARM_SCHEME:
1702 			error = g_part_parm_scheme(req, ap->name,
1703 			    &gpp.gpp_scheme);
1704 			break;
1705 		case G_PART_PARM_SIZE:
1706 			error = g_part_parm_quad(req, ap->name, &gpp.gpp_size);
1707 			break;
1708 		case G_PART_PARM_START:
1709 			error = g_part_parm_quad(req, ap->name,
1710 			    &gpp.gpp_start);
1711 			break;
1712 		case G_PART_PARM_TYPE:
1713 			error = g_part_parm_str(req, ap->name, &gpp.gpp_type);
1714 			break;
1715 		case G_PART_PARM_VERSION:
1716 			error = g_part_parm_uint32(req, ap->name,
1717 			    &gpp.gpp_version);
1718 			break;
1719 		default:
1720 			error = EDOOFUS;
1721 			gctl_error(req, "%d %s", error, ap->name);
1722 			break;
1723 		}
1724 		if (error != 0) {
1725 			if (error == ENOATTR) {
1726 				gctl_error(req, "%d param '%s'", error,
1727 				    ap->name);
1728 			}
1729 			return;
1730 		}
1731 		gpp.gpp_parms |= parm;
1732 	}
1733 	if ((gpp.gpp_parms & mparms) != mparms) {
1734 		parm = mparms - (gpp.gpp_parms & mparms);
1735 		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1736 		return;
1737 	}
1738 
1739 	/* Obtain permissions if possible/necessary. */
1740 	close_on_error = 0;
1741 	table = NULL;
1742 	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1743 		table = gpp.gpp_geom->softc;
1744 		if (table != NULL && table->gpt_corrupt &&
1745 		    ctlreq != G_PART_CTL_DESTROY &&
1746 		    ctlreq != G_PART_CTL_RECOVER) {
1747 			gctl_error(req, "%d table '%s' is corrupt",
1748 			    EPERM, gpp.gpp_geom->name);
1749 			return;
1750 		}
1751 		if (table != NULL && !table->gpt_opened) {
1752 			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1753 			    1, 1, 1);
1754 			if (error) {
1755 				gctl_error(req, "%d geom '%s'", error,
1756 				    gpp.gpp_geom->name);
1757 				return;
1758 			}
1759 			table->gpt_opened = 1;
1760 			close_on_error = 1;
1761 		}
1762 	}
1763 
1764 	/* Allow the scheme to check or modify the parameters. */
1765 	if (table != NULL) {
1766 		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1767 		if (error) {
1768 			gctl_error(req, "%d pre-check failed", error);
1769 			goto out;
1770 		}
1771 	} else
1772 		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1773 
1774 	switch (ctlreq) {
1775 	case G_PART_CTL_NONE:
1776 		panic("%s", __func__);
1777 	case G_PART_CTL_ADD:
1778 		error = g_part_ctl_add(req, &gpp);
1779 		break;
1780 	case G_PART_CTL_BOOTCODE:
1781 		error = g_part_ctl_bootcode(req, &gpp);
1782 		break;
1783 	case G_PART_CTL_COMMIT:
1784 		error = g_part_ctl_commit(req, &gpp);
1785 		break;
1786 	case G_PART_CTL_CREATE:
1787 		error = g_part_ctl_create(req, &gpp);
1788 		break;
1789 	case G_PART_CTL_DELETE:
1790 		error = g_part_ctl_delete(req, &gpp);
1791 		break;
1792 	case G_PART_CTL_DESTROY:
1793 		error = g_part_ctl_destroy(req, &gpp);
1794 		break;
1795 	case G_PART_CTL_MODIFY:
1796 		error = g_part_ctl_modify(req, &gpp);
1797 		break;
1798 	case G_PART_CTL_MOVE:
1799 		error = g_part_ctl_move(req, &gpp);
1800 		break;
1801 	case G_PART_CTL_RECOVER:
1802 		error = g_part_ctl_recover(req, &gpp);
1803 		break;
1804 	case G_PART_CTL_RESIZE:
1805 		error = g_part_ctl_resize(req, &gpp);
1806 		break;
1807 	case G_PART_CTL_SET:
1808 		error = g_part_ctl_setunset(req, &gpp, 1);
1809 		break;
1810 	case G_PART_CTL_UNDO:
1811 		error = g_part_ctl_undo(req, &gpp);
1812 		break;
1813 	case G_PART_CTL_UNSET:
1814 		error = g_part_ctl_setunset(req, &gpp, 0);
1815 		break;
1816 	}
1817 
1818 	/* Implement automatic commit. */
1819 	if (!error) {
1820 		auto_commit = (modifies &&
1821 		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1822 		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1823 		if (auto_commit) {
1824 			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, ("%s",
1825 			    __func__));
1826 			error = g_part_ctl_commit(req, &gpp);
1827 		}
1828 	}
1829 
1830  out:
1831 	if (error && close_on_error) {
1832 		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1833 		table->gpt_opened = 0;
1834 	}
1835 }
1836 
1837 static int
1838 g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1839     struct g_geom *gp)
1840 {
1841 
1842 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1843 	g_topology_assert();
1844 
1845 	g_part_wither(gp, EINVAL);
1846 	return (0);
1847 }
1848 
1849 static struct g_geom *
1850 g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1851 {
1852 	struct g_consumer *cp;
1853 	struct g_geom *gp;
1854 	struct g_part_entry *entry;
1855 	struct g_part_table *table;
1856 	struct root_hold_token *rht;
1857 	int attr, depth;
1858 	int error;
1859 
1860 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1861 	g_topology_assert();
1862 
1863 	/* Skip providers that are already open for writing. */
1864 	if (pp->acw > 0)
1865 		return (NULL);
1866 
1867 	/*
1868 	 * Create a GEOM with consumer and hook it up to the provider.
1869 	 * With that we become part of the topology. Optain read access
1870 	 * to the provider.
1871 	 */
1872 	gp = g_new_geomf(mp, "%s", pp->name);
1873 	cp = g_new_consumer(gp);
1874 	error = g_attach(cp, pp);
1875 	if (error == 0)
1876 		error = g_access(cp, 1, 0, 0);
1877 	if (error != 0) {
1878 		g_part_wither(gp, error);
1879 		return (NULL);
1880 	}
1881 
1882 	rht = root_mount_hold(mp->name);
1883 	g_topology_unlock();
1884 
1885 	/*
1886 	 * Short-circuit the whole probing galore when there's no
1887 	 * media present.
1888 	 */
1889 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1890 		error = ENODEV;
1891 		goto fail;
1892 	}
1893 
1894 	/* Make sure we can nest and if so, determine our depth. */
1895 	error = g_getattr("PART::isleaf", cp, &attr);
1896 	if (!error && attr) {
1897 		error = ENODEV;
1898 		goto fail;
1899 	}
1900 	error = g_getattr("PART::depth", cp, &attr);
1901 	depth = (!error) ? attr + 1 : 0;
1902 
1903 	error = g_part_probe(gp, cp, depth);
1904 	if (error)
1905 		goto fail;
1906 
1907 	table = gp->softc;
1908 
1909 	/*
1910 	 * Synthesize a disk geometry. Some partitioning schemes
1911 	 * depend on it and since some file systems need it even
1912 	 * when the partitition scheme doesn't, we do it here in
1913 	 * scheme-independent code.
1914 	 */
1915 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1916 
1917 	error = G_PART_READ(table, cp);
1918 	if (error)
1919 		goto fail;
1920 	error = g_part_check_integrity(table, cp);
1921 	if (error)
1922 		goto fail;
1923 
1924 	g_topology_lock();
1925 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1926 		if (!entry->gpe_internal)
1927 			g_part_new_provider(gp, table, entry);
1928 	}
1929 
1930 	root_mount_rel(rht);
1931 	g_access(cp, -1, 0, 0);
1932 	return (gp);
1933 
1934  fail:
1935 	g_topology_lock();
1936 	root_mount_rel(rht);
1937 	g_access(cp, -1, 0, 0);
1938 	g_part_wither(gp, error);
1939 	return (NULL);
1940 }
1941 
1942 /*
1943  * Geom methods.
1944  */
1945 
1946 static int
1947 g_part_access(struct g_provider *pp, int dr, int dw, int de)
1948 {
1949 	struct g_consumer *cp;
1950 
1951 	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1952 	    dw, de));
1953 
1954 	cp = LIST_FIRST(&pp->geom->consumer);
1955 
1956 	/* We always gain write-exclusive access. */
1957 	return (g_access(cp, dr, dw, dw + de));
1958 }
1959 
1960 static void
1961 g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1962     struct g_consumer *cp, struct g_provider *pp)
1963 {
1964 	char buf[64];
1965 	struct g_part_entry *entry;
1966 	struct g_part_table *table;
1967 
1968 	KASSERT(sb != NULL && gp != NULL, ("%s", __func__));
1969 	table = gp->softc;
1970 
1971 	if (indent == NULL) {
1972 		KASSERT(cp == NULL && pp != NULL, ("%s", __func__));
1973 		entry = pp->private;
1974 		if (entry == NULL)
1975 			return;
1976 		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
1977 		    (uintmax_t)entry->gpe_offset,
1978 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1979 		/*
1980 		 * libdisk compatibility quirk - the scheme dumps the
1981 		 * slicer name and partition type in a way that is
1982 		 * compatible with libdisk. When libdisk is not used
1983 		 * anymore, this should go away.
1984 		 */
1985 		G_PART_DUMPCONF(table, entry, sb, indent);
1986 	} else if (cp != NULL) {	/* Consumer configuration. */
1987 		KASSERT(pp == NULL, ("%s", __func__));
1988 		/* none */
1989 	} else if (pp != NULL) {	/* Provider configuration. */
1990 		entry = pp->private;
1991 		if (entry == NULL)
1992 			return;
1993 		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
1994 		    (uintmax_t)entry->gpe_start);
1995 		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
1996 		    (uintmax_t)entry->gpe_end);
1997 		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
1998 		    entry->gpe_index);
1999 		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
2000 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
2001 		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
2002 		    (uintmax_t)entry->gpe_offset);
2003 		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
2004 		    (uintmax_t)pp->mediasize);
2005 		G_PART_DUMPCONF(table, entry, sb, indent);
2006 	} else {			/* Geom configuration. */
2007 		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
2008 		    table->gpt_scheme->name);
2009 		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
2010 		    table->gpt_entries);
2011 		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
2012 		    (uintmax_t)table->gpt_first);
2013 		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
2014 		    (uintmax_t)table->gpt_last);
2015 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
2016 		    table->gpt_sectors);
2017 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
2018 		    table->gpt_heads);
2019 		sbuf_printf(sb, "%s<state>%s</state>\n", indent,
2020 		    table->gpt_corrupt ? "CORRUPT": "OK");
2021 		sbuf_printf(sb, "%s<modified>%s</modified>\n", indent,
2022 		    table->gpt_opened ? "true": "false");
2023 		G_PART_DUMPCONF(table, NULL, sb, indent);
2024 	}
2025 }
2026 
2027 static void
2028 g_part_orphan(struct g_consumer *cp)
2029 {
2030 	struct g_provider *pp;
2031 	struct g_part_table *table;
2032 
2033 	pp = cp->provider;
2034 	KASSERT(pp != NULL, ("%s", __func__));
2035 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
2036 	g_topology_assert();
2037 
2038 	KASSERT(pp->error != 0, ("%s", __func__));
2039 	table = cp->geom->softc;
2040 	if (table != NULL && table->gpt_opened)
2041 		g_access(cp, -1, -1, -1);
2042 	g_part_wither(cp->geom, pp->error);
2043 }
2044 
2045 static void
2046 g_part_spoiled(struct g_consumer *cp)
2047 {
2048 
2049 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
2050 	g_topology_assert();
2051 
2052 	g_part_wither(cp->geom, ENXIO);
2053 }
2054 
2055 static void
2056 g_part_start(struct bio *bp)
2057 {
2058 	struct bio *bp2;
2059 	struct g_consumer *cp;
2060 	struct g_geom *gp;
2061 	struct g_part_entry *entry;
2062 	struct g_part_table *table;
2063 	struct g_kerneldump *gkd;
2064 	struct g_provider *pp;
2065 	char buf[64];
2066 
2067 	pp = bp->bio_to;
2068 	gp = pp->geom;
2069 	table = gp->softc;
2070 	cp = LIST_FIRST(&gp->consumer);
2071 
2072 	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
2073 	    pp->name));
2074 
2075 	entry = pp->private;
2076 	if (entry == NULL) {
2077 		g_io_deliver(bp, ENXIO);
2078 		return;
2079 	}
2080 
2081 	switch(bp->bio_cmd) {
2082 	case BIO_DELETE:
2083 	case BIO_READ:
2084 	case BIO_WRITE:
2085 		if (bp->bio_offset >= pp->mediasize) {
2086 			g_io_deliver(bp, EIO);
2087 			return;
2088 		}
2089 		bp2 = g_clone_bio(bp);
2090 		if (bp2 == NULL) {
2091 			g_io_deliver(bp, ENOMEM);
2092 			return;
2093 		}
2094 		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
2095 			bp2->bio_length = pp->mediasize - bp2->bio_offset;
2096 		bp2->bio_done = g_std_done;
2097 		bp2->bio_offset += entry->gpe_offset;
2098 		g_io_request(bp2, cp);
2099 		return;
2100 	case BIO_FLUSH:
2101 		break;
2102 	case BIO_GETATTR:
2103 		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
2104 			return;
2105 		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
2106 			return;
2107 		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
2108 			return;
2109 		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
2110 			return;
2111 		if (g_handleattr_str(bp, "PART::scheme",
2112 		    table->gpt_scheme->name))
2113 			return;
2114 		if (g_handleattr_str(bp, "PART::type",
2115 		    G_PART_TYPE(table, entry, buf, sizeof(buf))))
2116 			return;
2117 		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
2118 			/*
2119 			 * Check that the partition is suitable for kernel
2120 			 * dumps. Typically only swap partitions should be
2121 			 * used. If the request comes from the nested scheme
2122 			 * we allow dumping there as well.
2123 			 */
2124 			if ((bp->bio_from == NULL ||
2125 			    bp->bio_from->geom->class != &g_part_class) &&
2126 			    G_PART_DUMPTO(table, entry) == 0) {
2127 				g_io_deliver(bp, ENODEV);
2128 				printf("GEOM_PART: Partition '%s' not suitable"
2129 				    " for kernel dumps (wrong type?)\n",
2130 				    pp->name);
2131 				return;
2132 			}
2133 			gkd = (struct g_kerneldump *)bp->bio_data;
2134 			if (gkd->offset >= pp->mediasize) {
2135 				g_io_deliver(bp, EIO);
2136 				return;
2137 			}
2138 			if (gkd->offset + gkd->length > pp->mediasize)
2139 				gkd->length = pp->mediasize - gkd->offset;
2140 			gkd->offset += entry->gpe_offset;
2141 		}
2142 		break;
2143 	default:
2144 		g_io_deliver(bp, EOPNOTSUPP);
2145 		return;
2146 	}
2147 
2148 	bp2 = g_clone_bio(bp);
2149 	if (bp2 == NULL) {
2150 		g_io_deliver(bp, ENOMEM);
2151 		return;
2152 	}
2153 	bp2->bio_done = g_std_done;
2154 	g_io_request(bp2, cp);
2155 }
2156 
2157 static void
2158 g_part_init(struct g_class *mp)
2159 {
2160 
2161 	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
2162 }
2163 
2164 static void
2165 g_part_fini(struct g_class *mp)
2166 {
2167 
2168 	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
2169 }
2170 
2171 static void
2172 g_part_unload_event(void *arg, int flag)
2173 {
2174 	struct g_consumer *cp;
2175 	struct g_geom *gp;
2176 	struct g_provider *pp;
2177 	struct g_part_scheme *scheme;
2178 	struct g_part_table *table;
2179 	uintptr_t *xchg;
2180 	int acc, error;
2181 
2182 	if (flag == EV_CANCEL)
2183 		return;
2184 
2185 	xchg = arg;
2186 	error = 0;
2187 	scheme = (void *)(*xchg);
2188 
2189 	g_topology_assert();
2190 
2191 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
2192 		table = gp->softc;
2193 		if (table->gpt_scheme != scheme)
2194 			continue;
2195 
2196 		acc = 0;
2197 		LIST_FOREACH(pp, &gp->provider, provider)
2198 			acc += pp->acr + pp->acw + pp->ace;
2199 		LIST_FOREACH(cp, &gp->consumer, consumer)
2200 			acc += cp->acr + cp->acw + cp->ace;
2201 
2202 		if (!acc)
2203 			g_part_wither(gp, ENOSYS);
2204 		else
2205 			error = EBUSY;
2206 	}
2207 
2208 	if (!error)
2209 		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
2210 
2211 	*xchg = error;
2212 }
2213 
2214 int
2215 g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
2216 {
2217 	struct g_part_scheme *iter;
2218 	uintptr_t arg;
2219 	int error;
2220 
2221 	error = 0;
2222 	switch (type) {
2223 	case MOD_LOAD:
2224 		TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
2225 			if (scheme == iter) {
2226 				printf("GEOM_PART: scheme %s is already "
2227 				    "registered!\n", scheme->name);
2228 				break;
2229 			}
2230 		}
2231 		if (iter == NULL) {
2232 			TAILQ_INSERT_TAIL(&g_part_schemes, scheme,
2233 			    scheme_list);
2234 			g_retaste(&g_part_class);
2235 		}
2236 		break;
2237 	case MOD_UNLOAD:
2238 		arg = (uintptr_t)scheme;
2239 		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
2240 		    NULL);
2241 		if (error == 0)
2242 			error = arg;
2243 		break;
2244 	default:
2245 		error = EOPNOTSUPP;
2246 		break;
2247 	}
2248 
2249 	return (error);
2250 }
2251