xref: /freebsd/sys/geom/part/g_part.c (revision dc60165b73e4c4d829a2cb9fed5cce585e93d9a9)
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/disk.h>
33 #include <sys/diskmbr.h>
34 #include <sys/endian.h>
35 #include <sys/kernel.h>
36 #include <sys/kobj.h>
37 #include <sys/limits.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/queue.h>
42 #include <sys/sbuf.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-hfs", G_PART_ALIAS_APPLE_HFS },
74 	{ "efi", G_PART_ALIAS_EFI },
75 	{ "freebsd", G_PART_ALIAS_FREEBSD },
76 	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
77 	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
78 	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
79 	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
80 	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
81 	{ "mbr", G_PART_ALIAS_MBR }
82 };
83 
84 /*
85  * The GEOM partitioning class.
86  */
87 static g_ctl_req_t g_part_ctlreq;
88 static g_ctl_destroy_geom_t g_part_destroy_geom;
89 static g_fini_t g_part_fini;
90 static g_init_t g_part_init;
91 static g_taste_t g_part_taste;
92 
93 static g_access_t g_part_access;
94 static g_dumpconf_t g_part_dumpconf;
95 static g_ioctl_t g_part_ioctl;
96 static g_orphan_t g_part_orphan;
97 static g_spoiled_t g_part_spoiled;
98 static g_start_t g_part_start;
99 
100 static struct g_class g_part_class = {
101 	.name = "PART",
102 	.version = G_VERSION,
103 	/* Class methods. */
104 	.ctlreq = g_part_ctlreq,
105 	.destroy_geom = g_part_destroy_geom,
106 	.fini = g_part_fini,
107 	.init = g_part_init,
108 	.taste = g_part_taste,
109 	/* Geom methods. */
110 	.access = g_part_access,
111 	.dumpconf = g_part_dumpconf,
112 	.ioctl = g_part_ioctl,
113 	.orphan = g_part_orphan,
114 	.spoiled = g_part_spoiled,
115 	.start = g_part_start,
116 };
117 
118 DECLARE_GEOM_CLASS(g_part_class, g_part);
119 
120 /*
121  * Support functions.
122  */
123 
124 static void g_part_wither(struct g_geom *, int);
125 
126 const char *
127 g_part_alias_name(enum g_part_alias alias)
128 {
129 	int i;
130 
131 	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
132 		if (g_part_alias_list[i].alias != alias)
133 			continue;
134 		return (g_part_alias_list[i].lexeme);
135 	}
136 
137 	return (NULL);
138 }
139 
140 void
141 g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
142     u_int *bestheads)
143 {
144 	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
145 	off_t chs, cylinders;
146 	u_int heads;
147 	int idx;
148 
149 	*bestchs = 0;
150 	*bestheads = 0;
151 	for (idx = 0; candidate_heads[idx] != 0; idx++) {
152 		heads = candidate_heads[idx];
153 		cylinders = blocks / heads / sectors;
154 		if (cylinders < heads || cylinders < sectors)
155 			break;
156 		if (cylinders > 1023)
157 			continue;
158 		chs = cylinders * heads * sectors;
159 		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
160 			*bestchs = chs;
161 			*bestheads = heads;
162 		}
163 	}
164 }
165 
166 static void
167 g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
168     off_t blocks)
169 {
170 	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
171 	off_t chs, bestchs;
172 	u_int heads, sectors;
173 	int idx;
174 
175 	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
176 	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
177 		table->gpt_fixgeom = 0;
178 		table->gpt_heads = 0;
179 		table->gpt_sectors = 0;
180 		bestchs = 0;
181 		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
182 			sectors = candidate_sectors[idx];
183 			g_part_geometry_heads(blocks, sectors, &chs, &heads);
184 			if (chs == 0)
185 				continue;
186 			/*
187 			 * Prefer a geometry with sectors > 1, but only if
188 			 * it doesn't bump down the numbver of heads to 1.
189 			 */
190 			if (chs > bestchs || (chs == bestchs && heads > 1 &&
191 			    table->gpt_sectors == 1)) {
192 				bestchs = chs;
193 				table->gpt_heads = heads;
194 				table->gpt_sectors = sectors;
195 			}
196 		}
197 		/*
198 		 * If we didn't find a geometry at all, then the disk is
199 		 * too big. This means we can use the maximum number of
200 		 * heads and sectors.
201 		 */
202 		if (bestchs == 0) {
203 			table->gpt_heads = 255;
204 			table->gpt_sectors = 63;
205 		}
206 	} else {
207 		table->gpt_fixgeom = 1;
208 		table->gpt_heads = heads;
209 		table->gpt_sectors = sectors;
210 	}
211 }
212 
213 struct g_part_entry *
214 g_part_new_entry(struct g_part_table *table, int index, quad_t start,
215     quad_t end)
216 {
217 	struct g_part_entry *entry, *last;
218 
219 	last = NULL;
220 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
221 		if (entry->gpe_index == index)
222 			break;
223 		if (entry->gpe_index > index) {
224 			entry = NULL;
225 			break;
226 		}
227 		last = entry;
228 	}
229 	if (entry == NULL) {
230 		entry = g_malloc(table->gpt_scheme->gps_entrysz,
231 		    M_WAITOK | M_ZERO);
232 		entry->gpe_index = index;
233 		if (last == NULL)
234 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
235 		else
236 			LIST_INSERT_AFTER(last, entry, gpe_entry);
237 	} else
238 		entry->gpe_offset = 0;
239 	entry->gpe_start = start;
240 	entry->gpe_end = end;
241 	return (entry);
242 }
243 
244 static void
245 g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
246     struct g_part_entry *entry)
247 {
248 	char buf[32];
249 	struct g_consumer *cp;
250 	struct g_provider *pp;
251 	off_t offset;
252 
253 	cp = LIST_FIRST(&gp->consumer);
254 	pp = cp->provider;
255 
256 	offset = entry->gpe_start * pp->sectorsize;
257 	if (entry->gpe_offset < offset)
258 		entry->gpe_offset = offset;
259 
260 	if (entry->gpe_pp == NULL) {
261 		entry->gpe_pp = g_new_providerf(gp, "%s%s", gp->name,
262 		    G_PART_NAME(table, entry, buf, sizeof(buf)));
263 		entry->gpe_pp->private = entry;		/* Close the circle. */
264 	}
265 	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
266 	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
267 	    pp->sectorsize;
268 	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
269 	entry->gpe_pp->sectorsize = pp->sectorsize;
270 	entry->gpe_pp->flags = pp->flags & G_PF_CANDELETE;
271 	if (pp->stripesize > 0) {
272 		entry->gpe_pp->stripesize = pp->stripesize;
273 		entry->gpe_pp->stripeoffset = (pp->stripeoffset +
274 		    entry->gpe_offset) % pp->stripesize;
275 	}
276 	g_error_provider(entry->gpe_pp, 0);
277 }
278 
279 static int
280 g_part_parm_geom(const char *rawname, struct g_geom **v)
281 {
282 	struct g_geom *gp;
283 	const char *pname;
284 
285 	if (strncmp(rawname, _PATH_DEV, strlen(_PATH_DEV)) == 0)
286 		pname = rawname + strlen(_PATH_DEV);
287 	else
288 		pname = rawname;
289 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
290 		if (!strcmp(pname, gp->name))
291 			break;
292 	}
293 	if (gp == NULL)
294 		return (EINVAL);
295 	*v = gp;
296 	return (0);
297 }
298 
299 static int
300 g_part_parm_provider(const char *pname, struct g_provider **v)
301 {
302 	struct g_provider *pp;
303 
304 	if (strncmp(pname, _PATH_DEV, strlen(_PATH_DEV)) == 0)
305 		pp = g_provider_by_name(pname + strlen(_PATH_DEV));
306 	else
307 		pp = g_provider_by_name(pname);
308 	if (pp == NULL)
309 		return (EINVAL);
310 	*v = pp;
311 	return (0);
312 }
313 
314 static int
315 g_part_parm_quad(const char *p, quad_t *v)
316 {
317 	char *x;
318 	quad_t q;
319 
320 	q = strtoq(p, &x, 0);
321 	if (*x != '\0' || q < 0)
322 		return (EINVAL);
323 	*v = q;
324 	return (0);
325 }
326 
327 static int
328 g_part_parm_scheme(const char *p, struct g_part_scheme **v)
329 {
330 	struct g_part_scheme *s;
331 
332 	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
333 		if (s == &g_part_null_scheme)
334 			continue;
335 		if (!strcasecmp(s->name, p))
336 			break;
337 	}
338 	if (s == NULL)
339 		return (EINVAL);
340 	*v = s;
341 	return (0);
342 }
343 
344 static int
345 g_part_parm_str(const char *p, const char **v)
346 {
347 
348 	if (p[0] == '\0')
349 		return (EINVAL);
350 	*v = p;
351 	return (0);
352 }
353 
354 static int
355 g_part_parm_uint(const char *p, u_int *v)
356 {
357 	char *x;
358 	long l;
359 
360 	l = strtol(p, &x, 0);
361 	if (*x != '\0' || l < 0 || l > INT_MAX)
362 		return (EINVAL);
363 	*v = (unsigned int)l;
364 	return (0);
365 }
366 
367 static int
368 g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
369 {
370 	struct g_part_scheme *iter, *scheme;
371 	struct g_part_table *table;
372 	int pri, probe;
373 
374 	table = gp->softc;
375 	scheme = (table != NULL) ? table->gpt_scheme : NULL;
376 	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
377 	if (pri == 0)
378 		goto done;
379 	if (pri > 0) {	/* error */
380 		scheme = NULL;
381 		pri = INT_MIN;
382 	}
383 
384 	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
385 		if (iter == &g_part_null_scheme)
386 			continue;
387 		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
388 		    M_WAITOK);
389 		table->gpt_gp = gp;
390 		table->gpt_scheme = iter;
391 		table->gpt_depth = depth;
392 		probe = G_PART_PROBE(table, cp);
393 		if (probe <= 0 && probe > pri) {
394 			pri = probe;
395 			scheme = iter;
396 			if (gp->softc != NULL)
397 				kobj_delete((kobj_t)gp->softc, M_GEOM);
398 			gp->softc = table;
399 			if (pri == 0)
400 				goto done;
401 		} else
402 			kobj_delete((kobj_t)table, M_GEOM);
403 	}
404 
405 done:
406 	return ((scheme == NULL) ? ENXIO : 0);
407 }
408 
409 /*
410  * Control request functions.
411  */
412 
413 static int
414 g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
415 {
416 	char buf[32];
417 	struct g_geom *gp;
418 	struct g_provider *pp;
419 	struct g_part_entry *delent, *last, *entry;
420 	struct g_part_table *table;
421 	struct sbuf *sb;
422 	quad_t end;
423 	unsigned int index;
424 	int error;
425 
426 	gp = gpp->gpp_geom;
427 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
428 	g_topology_assert();
429 
430 	pp = LIST_FIRST(&gp->consumer)->provider;
431 	table = gp->softc;
432 	end = gpp->gpp_start + gpp->gpp_size - 1;
433 
434 	if (gpp->gpp_start < table->gpt_first ||
435 	    gpp->gpp_start > table->gpt_last) {
436 		gctl_error(req, "%d start '%jd'", EINVAL,
437 		    (intmax_t)gpp->gpp_start);
438 		return (EINVAL);
439 	}
440 	if (end < gpp->gpp_start || end > table->gpt_last) {
441 		gctl_error(req, "%d size '%jd'", EINVAL,
442 		    (intmax_t)gpp->gpp_size);
443 		return (EINVAL);
444 	}
445 	if (gpp->gpp_index > table->gpt_entries) {
446 		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
447 		return (EINVAL);
448 	}
449 
450 	delent = last = NULL;
451 	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
452 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
453 		if (entry->gpe_deleted) {
454 			if (entry->gpe_index == index)
455 				delent = entry;
456 			continue;
457 		}
458 		if (entry->gpe_index == index)
459 			index = entry->gpe_index + 1;
460 		if (entry->gpe_index < index)
461 			last = entry;
462 		if (entry->gpe_internal)
463 			continue;
464 		if (gpp->gpp_start >= entry->gpe_start &&
465 		    gpp->gpp_start <= entry->gpe_end) {
466 			gctl_error(req, "%d start '%jd'", ENOSPC,
467 			    (intmax_t)gpp->gpp_start);
468 			return (ENOSPC);
469 		}
470 		if (end >= entry->gpe_start && end <= entry->gpe_end) {
471 			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
472 			return (ENOSPC);
473 		}
474 		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
475 			gctl_error(req, "%d size '%jd'", ENOSPC,
476 			    (intmax_t)gpp->gpp_size);
477 			return (ENOSPC);
478 		}
479 	}
480 	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
481 		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
482 		return (EEXIST);
483 	}
484 
485 	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
486 	    M_WAITOK | M_ZERO) : delent;
487 	entry->gpe_index = index;
488 	entry->gpe_start = gpp->gpp_start;
489 	entry->gpe_end = end;
490 	error = G_PART_ADD(table, entry, gpp);
491 	if (error) {
492 		gctl_error(req, "%d", error);
493 		if (delent == NULL)
494 			g_free(entry);
495 		return (error);
496 	}
497 	if (delent == NULL) {
498 		if (last == NULL)
499 			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
500 		else
501 			LIST_INSERT_AFTER(last, entry, gpe_entry);
502 		entry->gpe_created = 1;
503 	} else {
504 		entry->gpe_deleted = 0;
505 		entry->gpe_modified = 1;
506 	}
507 	g_part_new_provider(gp, table, entry);
508 
509 	/* Provide feedback if so requested. */
510 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
511 		sb = sbuf_new_auto();
512 		sbuf_printf(sb, "%s%s added\n", gp->name,
513 		    G_PART_NAME(table, entry, buf, sizeof(buf)));
514 		sbuf_finish(sb);
515 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
516 		sbuf_delete(sb);
517 	}
518 	return (0);
519 }
520 
521 static int
522 g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
523 {
524 	struct g_geom *gp;
525 	struct g_part_table *table;
526 	struct sbuf *sb;
527 	int error, sz;
528 
529 	gp = gpp->gpp_geom;
530 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
531 	g_topology_assert();
532 
533 	table = gp->softc;
534 	sz = table->gpt_scheme->gps_bootcodesz;
535 	if (sz == 0) {
536 		error = ENODEV;
537 		goto fail;
538 	}
539 	if (gpp->gpp_codesize > sz) {
540 		error = EFBIG;
541 		goto fail;
542 	}
543 
544 	error = G_PART_BOOTCODE(table, gpp);
545 	if (error)
546 		goto fail;
547 
548 	/* Provide feedback if so requested. */
549 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
550 		sb = sbuf_new_auto();
551 		sbuf_printf(sb, "%s has bootcode\n", gp->name);
552 		sbuf_finish(sb);
553 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
554 		sbuf_delete(sb);
555 	}
556 	return (0);
557 
558  fail:
559 	gctl_error(req, "%d", error);
560 	return (error);
561 }
562 
563 static int
564 g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
565 {
566 	struct g_consumer *cp;
567 	struct g_geom *gp;
568 	struct g_provider *pp;
569 	struct g_part_entry *entry, *tmp;
570 	struct g_part_table *table;
571 	char *buf;
572 	int error, i;
573 
574 	gp = gpp->gpp_geom;
575 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
576 	g_topology_assert();
577 
578 	table = gp->softc;
579 	if (!table->gpt_opened) {
580 		gctl_error(req, "%d", EPERM);
581 		return (EPERM);
582 	}
583 
584 	g_topology_unlock();
585 
586 	cp = LIST_FIRST(&gp->consumer);
587 	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
588 		pp = cp->provider;
589 		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
590 		while (table->gpt_smhead != 0) {
591 			i = ffs(table->gpt_smhead) - 1;
592 			error = g_write_data(cp, i * pp->sectorsize, buf,
593 			    pp->sectorsize);
594 			if (error) {
595 				g_free(buf);
596 				goto fail;
597 			}
598 			table->gpt_smhead &= ~(1 << i);
599 		}
600 		while (table->gpt_smtail != 0) {
601 			i = ffs(table->gpt_smtail) - 1;
602 			error = g_write_data(cp, pp->mediasize - (i + 1) *
603 			    pp->sectorsize, buf, pp->sectorsize);
604 			if (error) {
605 				g_free(buf);
606 				goto fail;
607 			}
608 			table->gpt_smtail &= ~(1 << i);
609 		}
610 		g_free(buf);
611 	}
612 
613 	if (table->gpt_scheme == &g_part_null_scheme) {
614 		g_topology_lock();
615 		g_access(cp, -1, -1, -1);
616 		g_part_wither(gp, ENXIO);
617 		return (0);
618 	}
619 
620 	error = G_PART_WRITE(table, cp);
621 	if (error)
622 		goto fail;
623 
624 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
625 		if (!entry->gpe_deleted) {
626 			entry->gpe_created = 0;
627 			entry->gpe_modified = 0;
628 			continue;
629 		}
630 		LIST_REMOVE(entry, gpe_entry);
631 		g_free(entry);
632 	}
633 	table->gpt_created = 0;
634 	table->gpt_opened = 0;
635 
636 	g_topology_lock();
637 	g_access(cp, -1, -1, -1);
638 	return (0);
639 
640 fail:
641 	g_topology_lock();
642 	gctl_error(req, "%d", error);
643 	return (error);
644 }
645 
646 static int
647 g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
648 {
649 	struct g_consumer *cp;
650 	struct g_geom *gp;
651 	struct g_provider *pp;
652 	struct g_part_scheme *scheme;
653 	struct g_part_table *null, *table;
654 	struct sbuf *sb;
655 	int attr, error;
656 
657 	pp = gpp->gpp_provider;
658 	scheme = gpp->gpp_scheme;
659 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
660 	g_topology_assert();
661 
662 	/* Check that there isn't already a g_part geom on the provider. */
663 	error = g_part_parm_geom(pp->name, &gp);
664 	if (!error) {
665 		null = gp->softc;
666 		if (null->gpt_scheme != &g_part_null_scheme) {
667 			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
668 			return (EEXIST);
669 		}
670 	} else
671 		null = NULL;
672 
673 	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
674 	    (gpp->gpp_entries < scheme->gps_minent ||
675 	     gpp->gpp_entries > scheme->gps_maxent)) {
676 		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
677 		return (EINVAL);
678 	}
679 
680 	if (null == NULL)
681 		gp = g_new_geomf(&g_part_class, "%s", pp->name);
682 	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
683 	    M_WAITOK);
684 	table = gp->softc;
685 	table->gpt_gp = gp;
686 	table->gpt_scheme = gpp->gpp_scheme;
687 	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
688 	    gpp->gpp_entries : scheme->gps_minent;
689 	LIST_INIT(&table->gpt_entry);
690 	if (null == NULL) {
691 		cp = g_new_consumer(gp);
692 		error = g_attach(cp, pp);
693 		if (error == 0)
694 			error = g_access(cp, 1, 1, 1);
695 		if (error != 0) {
696 			g_part_wither(gp, error);
697 			gctl_error(req, "%d geom '%s'", error, pp->name);
698 			return (error);
699 		}
700 		table->gpt_opened = 1;
701 	} else {
702 		cp = LIST_FIRST(&gp->consumer);
703 		table->gpt_opened = null->gpt_opened;
704 		table->gpt_smhead = null->gpt_smhead;
705 		table->gpt_smtail = null->gpt_smtail;
706 	}
707 
708 	g_topology_unlock();
709 
710 	/* Make sure the provider has media. */
711 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
712 		error = ENODEV;
713 		goto fail;
714 	}
715 
716 	/* Make sure we can nest and if so, determine our depth. */
717 	error = g_getattr("PART::isleaf", cp, &attr);
718 	if (!error && attr) {
719 		error = ENODEV;
720 		goto fail;
721 	}
722 	error = g_getattr("PART::depth", cp, &attr);
723 	table->gpt_depth = (!error) ? attr + 1 : 0;
724 
725 	/*
726 	 * Synthesize a disk geometry. Some partitioning schemes
727 	 * depend on it and since some file systems need it even
728 	 * when the partitition scheme doesn't, we do it here in
729 	 * scheme-independent code.
730 	 */
731 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
732 
733 	error = G_PART_CREATE(table, gpp);
734 	if (error)
735 		goto fail;
736 
737 	g_topology_lock();
738 
739 	table->gpt_created = 1;
740 	if (null != NULL)
741 		kobj_delete((kobj_t)null, M_GEOM);
742 
743 	/*
744 	 * Support automatic commit by filling in the gpp_geom
745 	 * parameter.
746 	 */
747 	gpp->gpp_parms |= G_PART_PARM_GEOM;
748 	gpp->gpp_geom = gp;
749 
750 	/* Provide feedback if so requested. */
751 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
752 		sb = sbuf_new_auto();
753 		sbuf_printf(sb, "%s created\n", gp->name);
754 		sbuf_finish(sb);
755 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
756 		sbuf_delete(sb);
757 	}
758 	return (0);
759 
760 fail:
761 	g_topology_lock();
762 	if (null == NULL) {
763 		g_access(cp, -1, -1, -1);
764 		g_part_wither(gp, error);
765 	} else {
766 		kobj_delete((kobj_t)gp->softc, M_GEOM);
767 		gp->softc = null;
768 	}
769 	gctl_error(req, "%d provider", error);
770 	return (error);
771 }
772 
773 static int
774 g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
775 {
776 	char buf[32];
777 	struct g_geom *gp;
778 	struct g_provider *pp;
779 	struct g_part_entry *entry;
780 	struct g_part_table *table;
781 	struct sbuf *sb;
782 
783 	gp = gpp->gpp_geom;
784 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
785 	g_topology_assert();
786 
787 	table = gp->softc;
788 
789 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
790 		if (entry->gpe_deleted || entry->gpe_internal)
791 			continue;
792 		if (entry->gpe_index == gpp->gpp_index)
793 			break;
794 	}
795 	if (entry == NULL) {
796 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
797 		return (ENOENT);
798 	}
799 
800 	pp = entry->gpe_pp;
801 	if (pp != NULL) {
802 		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
803 			gctl_error(req, "%d", EBUSY);
804 			return (EBUSY);
805 		}
806 
807 		pp->private = NULL;
808 		entry->gpe_pp = NULL;
809 	}
810 
811 	if (entry->gpe_created) {
812 		LIST_REMOVE(entry, gpe_entry);
813 		g_free(entry);
814 	} else {
815 		entry->gpe_modified = 0;
816 		entry->gpe_deleted = 1;
817 	}
818 
819 	if (pp != NULL)
820 		g_wither_provider(pp, ENXIO);
821 
822 	/* Provide feedback if so requested. */
823 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
824 		sb = sbuf_new_auto();
825 		sbuf_printf(sb, "%s%s deleted\n", gp->name,
826 		    G_PART_NAME(table, entry, buf, sizeof(buf)));
827 		sbuf_finish(sb);
828 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
829 		sbuf_delete(sb);
830 	}
831 	return (0);
832 }
833 
834 static int
835 g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
836 {
837 	struct g_geom *gp;
838 	struct g_part_entry *entry;
839 	struct g_part_table *null, *table;
840 	struct sbuf *sb;
841 	int error;
842 
843 	gp = gpp->gpp_geom;
844 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
845 	g_topology_assert();
846 
847 	table = gp->softc;
848 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
849 		if (entry->gpe_deleted || entry->gpe_internal)
850 			continue;
851 		gctl_error(req, "%d", EBUSY);
852 		return (EBUSY);
853 	}
854 
855 	error = G_PART_DESTROY(table, gpp);
856 	if (error) {
857 		gctl_error(req, "%d", error);
858 		return (error);
859 	}
860 
861 	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
862 	    M_WAITOK);
863 	null = gp->softc;
864 	null->gpt_gp = gp;
865 	null->gpt_scheme = &g_part_null_scheme;
866 	LIST_INIT(&null->gpt_entry);
867 	null->gpt_depth = table->gpt_depth;
868 	null->gpt_opened = table->gpt_opened;
869 	null->gpt_smhead = table->gpt_smhead;
870 	null->gpt_smtail = table->gpt_smtail;
871 
872 	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
873 		LIST_REMOVE(entry, gpe_entry);
874 		g_free(entry);
875 	}
876 	kobj_delete((kobj_t)table, M_GEOM);
877 
878 	/* Provide feedback if so requested. */
879 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
880 		sb = sbuf_new_auto();
881 		sbuf_printf(sb, "%s destroyed\n", gp->name);
882 		sbuf_finish(sb);
883 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
884 		sbuf_delete(sb);
885 	}
886 	return (0);
887 }
888 
889 static int
890 g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
891 {
892 	char buf[32];
893 	struct g_geom *gp;
894 	struct g_part_entry *entry;
895 	struct g_part_table *table;
896 	struct sbuf *sb;
897 	int error;
898 
899 	gp = gpp->gpp_geom;
900 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
901 	g_topology_assert();
902 
903 	table = gp->softc;
904 
905 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
906 		if (entry->gpe_deleted || entry->gpe_internal)
907 			continue;
908 		if (entry->gpe_index == gpp->gpp_index)
909 			break;
910 	}
911 	if (entry == NULL) {
912 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
913 		return (ENOENT);
914 	}
915 
916 	error = G_PART_MODIFY(table, entry, gpp);
917 	if (error) {
918 		gctl_error(req, "%d", error);
919 		return (error);
920 	}
921 
922 	if (!entry->gpe_created)
923 		entry->gpe_modified = 1;
924 
925 	/* Provide feedback if so requested. */
926 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
927 		sb = sbuf_new_auto();
928 		sbuf_printf(sb, "%s%s modified\n", gp->name,
929 		    G_PART_NAME(table, entry, buf, sizeof(buf)));
930 		sbuf_finish(sb);
931 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
932 		sbuf_delete(sb);
933 	}
934 	return (0);
935 }
936 
937 static int
938 g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
939 {
940 	gctl_error(req, "%d verb 'move'", ENOSYS);
941 	return (ENOSYS);
942 }
943 
944 static int
945 g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
946 {
947 	gctl_error(req, "%d verb 'recover'", ENOSYS);
948 	return (ENOSYS);
949 }
950 
951 static int
952 g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
953 {
954 	gctl_error(req, "%d verb 'resize'", ENOSYS);
955 	return (ENOSYS);
956 }
957 
958 static int
959 g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
960     unsigned int set)
961 {
962 	char buf[32];
963 	struct g_geom *gp;
964 	struct g_part_entry *entry;
965 	struct g_part_table *table;
966 	struct sbuf *sb;
967 	int error;
968 
969 	gp = gpp->gpp_geom;
970 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
971 	g_topology_assert();
972 
973 	table = gp->softc;
974 
975 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
976 		if (entry->gpe_deleted || entry->gpe_internal)
977 			continue;
978 		if (entry->gpe_index == gpp->gpp_index)
979 			break;
980 	}
981 	if (entry == NULL) {
982 		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
983 		return (ENOENT);
984 	}
985 
986 	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
987 	if (error) {
988 		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
989 		return (error);
990 	}
991 
992 	/* Provide feedback if so requested. */
993 	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
994 		sb = sbuf_new_auto();
995 		sbuf_printf(sb, "%s%s has %s %sset\n", gp->name,
996 		    G_PART_NAME(table, entry, buf, sizeof(buf)),
997 		    gpp->gpp_attrib, (set) ? "" : "un");
998 		sbuf_finish(sb);
999 		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1000 		sbuf_delete(sb);
1001 	}
1002 	return (0);
1003 }
1004 
1005 static int
1006 g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1007 {
1008 	struct g_consumer *cp;
1009 	struct g_provider *pp;
1010 	struct g_geom *gp;
1011 	struct g_part_entry *entry, *tmp;
1012 	struct g_part_table *table;
1013 	int error, reprobe;
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 	if (!table->gpt_opened) {
1021 		gctl_error(req, "%d", EPERM);
1022 		return (EPERM);
1023 	}
1024 
1025 	cp = LIST_FIRST(&gp->consumer);
1026 	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1027 		entry->gpe_modified = 0;
1028 		if (entry->gpe_created) {
1029 			pp = entry->gpe_pp;
1030 			if (pp != NULL) {
1031 				pp->private = NULL;
1032 				entry->gpe_pp = NULL;
1033 				g_wither_provider(pp, ENXIO);
1034 			}
1035 			entry->gpe_deleted = 1;
1036 		}
1037 		if (entry->gpe_deleted) {
1038 			LIST_REMOVE(entry, gpe_entry);
1039 			g_free(entry);
1040 		}
1041 	}
1042 
1043 	g_topology_unlock();
1044 
1045 	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1046 	    table->gpt_created) ? 1 : 0;
1047 
1048 	if (reprobe) {
1049 		if (!LIST_EMPTY(&table->gpt_entry)) {
1050 			error = EBUSY;
1051 			goto fail;
1052 		}
1053 		error = g_part_probe(gp, cp, table->gpt_depth);
1054 		if (error) {
1055 			g_topology_lock();
1056 			g_access(cp, -1, -1, -1);
1057 			g_part_wither(gp, error);
1058 			return (0);
1059 		}
1060 		table = gp->softc;
1061 	}
1062 
1063 	error = G_PART_READ(table, cp);
1064 	if (error)
1065 		goto fail;
1066 
1067 	g_topology_lock();
1068 
1069 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1070 		if (!entry->gpe_internal)
1071 			g_part_new_provider(gp, table, entry);
1072 	}
1073 
1074 	table->gpt_opened = 0;
1075 	g_access(cp, -1, -1, -1);
1076 	return (0);
1077 
1078 fail:
1079 	g_topology_lock();
1080 	gctl_error(req, "%d", error);
1081 	return (error);
1082 }
1083 
1084 static void
1085 g_part_wither(struct g_geom *gp, int error)
1086 {
1087 	struct g_part_entry *entry;
1088 	struct g_part_table *table;
1089 
1090 	table = gp->softc;
1091 	if (table != NULL) {
1092 		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1093 			LIST_REMOVE(entry, gpe_entry);
1094 			g_free(entry);
1095 		}
1096 		if (gp->softc != NULL) {
1097 			kobj_delete((kobj_t)gp->softc, M_GEOM);
1098 			gp->softc = NULL;
1099 		}
1100 	}
1101 	g_wither_geom(gp, error);
1102 }
1103 
1104 /*
1105  * Class methods.
1106  */
1107 
1108 static void
1109 g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1110 {
1111 	struct g_part_parms gpp;
1112 	struct g_part_table *table;
1113 	struct gctl_req_arg *ap;
1114 	const char *p;
1115 	enum g_part_ctl ctlreq;
1116 	unsigned int i, mparms, oparms, parm;
1117 	int auto_commit, close_on_error;
1118 	int error, len, modifies;
1119 
1120 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1121 	g_topology_assert();
1122 
1123 	ctlreq = G_PART_CTL_NONE;
1124 	modifies = 1;
1125 	mparms = 0;
1126 	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1127 	switch (*verb) {
1128 	case 'a':
1129 		if (!strcmp(verb, "add")) {
1130 			ctlreq = G_PART_CTL_ADD;
1131 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1132 			    G_PART_PARM_START | G_PART_PARM_TYPE;
1133 			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1134 		}
1135 		break;
1136 	case 'b':
1137 		if (!strcmp(verb, "bootcode")) {
1138 			ctlreq = G_PART_CTL_BOOTCODE;
1139 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1140 		}
1141 		break;
1142 	case 'c':
1143 		if (!strcmp(verb, "commit")) {
1144 			ctlreq = G_PART_CTL_COMMIT;
1145 			mparms |= G_PART_PARM_GEOM;
1146 			modifies = 0;
1147 		} else if (!strcmp(verb, "create")) {
1148 			ctlreq = G_PART_CTL_CREATE;
1149 			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1150 			oparms |= G_PART_PARM_ENTRIES;
1151 		}
1152 		break;
1153 	case 'd':
1154 		if (!strcmp(verb, "delete")) {
1155 			ctlreq = G_PART_CTL_DELETE;
1156 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1157 		} else if (!strcmp(verb, "destroy")) {
1158 			ctlreq = G_PART_CTL_DESTROY;
1159 			mparms |= G_PART_PARM_GEOM;
1160 		}
1161 		break;
1162 	case 'm':
1163 		if (!strcmp(verb, "modify")) {
1164 			ctlreq = G_PART_CTL_MODIFY;
1165 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1166 			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1167 		} else if (!strcmp(verb, "move")) {
1168 			ctlreq = G_PART_CTL_MOVE;
1169 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1170 		}
1171 		break;
1172 	case 'r':
1173 		if (!strcmp(verb, "recover")) {
1174 			ctlreq = G_PART_CTL_RECOVER;
1175 			mparms |= G_PART_PARM_GEOM;
1176 		} else if (!strcmp(verb, "resize")) {
1177 			ctlreq = G_PART_CTL_RESIZE;
1178 			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1179 		}
1180 		break;
1181 	case 's':
1182 		if (!strcmp(verb, "set")) {
1183 			ctlreq = G_PART_CTL_SET;
1184 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1185 			    G_PART_PARM_INDEX;
1186 		}
1187 		break;
1188 	case 'u':
1189 		if (!strcmp(verb, "undo")) {
1190 			ctlreq = G_PART_CTL_UNDO;
1191 			mparms |= G_PART_PARM_GEOM;
1192 			modifies = 0;
1193 		} else if (!strcmp(verb, "unset")) {
1194 			ctlreq = G_PART_CTL_UNSET;
1195 			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1196 			    G_PART_PARM_INDEX;
1197 		}
1198 		break;
1199 	}
1200 	if (ctlreq == G_PART_CTL_NONE) {
1201 		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1202 		return;
1203 	}
1204 
1205 	bzero(&gpp, sizeof(gpp));
1206 	for (i = 0; i < req->narg; i++) {
1207 		ap = &req->arg[i];
1208 		parm = 0;
1209 		switch (ap->name[0]) {
1210 		case 'a':
1211 			if (!strcmp(ap->name, "attrib"))
1212 				parm = G_PART_PARM_ATTRIB;
1213 			break;
1214 		case 'b':
1215 			if (!strcmp(ap->name, "bootcode"))
1216 				parm = G_PART_PARM_BOOTCODE;
1217 			break;
1218 		case 'c':
1219 			if (!strcmp(ap->name, "class"))
1220 				continue;
1221 			break;
1222 		case 'e':
1223 			if (!strcmp(ap->name, "entries"))
1224 				parm = G_PART_PARM_ENTRIES;
1225 			break;
1226 		case 'f':
1227 			if (!strcmp(ap->name, "flags"))
1228 				parm = G_PART_PARM_FLAGS;
1229 			break;
1230 		case 'g':
1231 			if (!strcmp(ap->name, "geom"))
1232 				parm = G_PART_PARM_GEOM;
1233 			break;
1234 		case 'i':
1235 			if (!strcmp(ap->name, "index"))
1236 				parm = G_PART_PARM_INDEX;
1237 			break;
1238 		case 'l':
1239 			if (!strcmp(ap->name, "label"))
1240 				parm = G_PART_PARM_LABEL;
1241 			break;
1242 		case 'o':
1243 			if (!strcmp(ap->name, "output"))
1244 				parm = G_PART_PARM_OUTPUT;
1245 			break;
1246 		case 'p':
1247 			if (!strcmp(ap->name, "provider"))
1248 				parm = G_PART_PARM_PROVIDER;
1249 			break;
1250 		case 's':
1251 			if (!strcmp(ap->name, "scheme"))
1252 				parm = G_PART_PARM_SCHEME;
1253 			else if (!strcmp(ap->name, "size"))
1254 				parm = G_PART_PARM_SIZE;
1255 			else if (!strcmp(ap->name, "start"))
1256 				parm = G_PART_PARM_START;
1257 			break;
1258 		case 't':
1259 			if (!strcmp(ap->name, "type"))
1260 				parm = G_PART_PARM_TYPE;
1261 			break;
1262 		case 'v':
1263 			if (!strcmp(ap->name, "verb"))
1264 				continue;
1265 			else if (!strcmp(ap->name, "version"))
1266 				parm = G_PART_PARM_VERSION;
1267 			break;
1268 		}
1269 		if ((parm & (mparms | oparms)) == 0) {
1270 			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1271 			return;
1272 		}
1273 		if (parm == G_PART_PARM_BOOTCODE)
1274 			p = gctl_get_param(req, ap->name, &len);
1275 		else
1276 			p = gctl_get_asciiparam(req, ap->name);
1277 		if (p == NULL) {
1278 			gctl_error(req, "%d param '%s'", ENOATTR, ap->name);
1279 			return;
1280 		}
1281 		switch (parm) {
1282 		case G_PART_PARM_ATTRIB:
1283 			error = g_part_parm_str(p, &gpp.gpp_attrib);
1284 			break;
1285 		case G_PART_PARM_BOOTCODE:
1286 			gpp.gpp_codeptr = p;
1287 			gpp.gpp_codesize = len;
1288 			error = 0;
1289 			break;
1290 		case G_PART_PARM_ENTRIES:
1291 			error = g_part_parm_uint(p, &gpp.gpp_entries);
1292 			break;
1293 		case G_PART_PARM_FLAGS:
1294 			if (p[0] == '\0')
1295 				continue;
1296 			error = g_part_parm_str(p, &gpp.gpp_flags);
1297 			break;
1298 		case G_PART_PARM_GEOM:
1299 			error = g_part_parm_geom(p, &gpp.gpp_geom);
1300 			break;
1301 		case G_PART_PARM_INDEX:
1302 			error = g_part_parm_uint(p, &gpp.gpp_index);
1303 			break;
1304 		case G_PART_PARM_LABEL:
1305 			/* An empty label is always valid. */
1306 			gpp.gpp_label = p;
1307 			error = 0;
1308 			break;
1309 		case G_PART_PARM_OUTPUT:
1310 			error = 0;	/* Write-only parameter */
1311 			break;
1312 		case G_PART_PARM_PROVIDER:
1313 			error = g_part_parm_provider(p, &gpp.gpp_provider);
1314 			break;
1315 		case G_PART_PARM_SCHEME:
1316 			error = g_part_parm_scheme(p, &gpp.gpp_scheme);
1317 			break;
1318 		case G_PART_PARM_SIZE:
1319 			error = g_part_parm_quad(p, &gpp.gpp_size);
1320 			break;
1321 		case G_PART_PARM_START:
1322 			error = g_part_parm_quad(p, &gpp.gpp_start);
1323 			break;
1324 		case G_PART_PARM_TYPE:
1325 			error = g_part_parm_str(p, &gpp.gpp_type);
1326 			break;
1327 		case G_PART_PARM_VERSION:
1328 			error = g_part_parm_uint(p, &gpp.gpp_version);
1329 			break;
1330 		default:
1331 			error = EDOOFUS;
1332 			break;
1333 		}
1334 		if (error) {
1335 			gctl_error(req, "%d %s '%s'", error, ap->name, p);
1336 			return;
1337 		}
1338 		gpp.gpp_parms |= parm;
1339 	}
1340 	if ((gpp.gpp_parms & mparms) != mparms) {
1341 		parm = mparms - (gpp.gpp_parms & mparms);
1342 		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1343 		return;
1344 	}
1345 
1346 	/* Obtain permissions if possible/necessary. */
1347 	close_on_error = 0;
1348 	table = NULL;
1349 	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1350 		table = gpp.gpp_geom->softc;
1351 		if (table != NULL && !table->gpt_opened) {
1352 			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1353 			    1, 1, 1);
1354 			if (error) {
1355 				gctl_error(req, "%d geom '%s'", error,
1356 				    gpp.gpp_geom->name);
1357 				return;
1358 			}
1359 			table->gpt_opened = 1;
1360 			close_on_error = 1;
1361 		}
1362 	}
1363 
1364 	/* Allow the scheme to check or modify the parameters. */
1365 	if (table != NULL) {
1366 		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1367 		if (error) {
1368 			gctl_error(req, "%d pre-check failed", error);
1369 			goto out;
1370 		}
1371 	} else
1372 		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1373 
1374 	switch (ctlreq) {
1375 	case G_PART_CTL_NONE:
1376 		panic("%s", __func__);
1377 	case G_PART_CTL_ADD:
1378 		error = g_part_ctl_add(req, &gpp);
1379 		break;
1380 	case G_PART_CTL_BOOTCODE:
1381 		error = g_part_ctl_bootcode(req, &gpp);
1382 		break;
1383 	case G_PART_CTL_COMMIT:
1384 		error = g_part_ctl_commit(req, &gpp);
1385 		break;
1386 	case G_PART_CTL_CREATE:
1387 		error = g_part_ctl_create(req, &gpp);
1388 		break;
1389 	case G_PART_CTL_DELETE:
1390 		error = g_part_ctl_delete(req, &gpp);
1391 		break;
1392 	case G_PART_CTL_DESTROY:
1393 		error = g_part_ctl_destroy(req, &gpp);
1394 		break;
1395 	case G_PART_CTL_MODIFY:
1396 		error = g_part_ctl_modify(req, &gpp);
1397 		break;
1398 	case G_PART_CTL_MOVE:
1399 		error = g_part_ctl_move(req, &gpp);
1400 		break;
1401 	case G_PART_CTL_RECOVER:
1402 		error = g_part_ctl_recover(req, &gpp);
1403 		break;
1404 	case G_PART_CTL_RESIZE:
1405 		error = g_part_ctl_resize(req, &gpp);
1406 		break;
1407 	case G_PART_CTL_SET:
1408 		error = g_part_ctl_setunset(req, &gpp, 1);
1409 		break;
1410 	case G_PART_CTL_UNDO:
1411 		error = g_part_ctl_undo(req, &gpp);
1412 		break;
1413 	case G_PART_CTL_UNSET:
1414 		error = g_part_ctl_setunset(req, &gpp, 0);
1415 		break;
1416 	}
1417 
1418 	/* Implement automatic commit. */
1419 	if (!error) {
1420 		auto_commit = (modifies &&
1421 		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1422 		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1423 		if (auto_commit) {
1424 			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, (__func__));
1425 			error = g_part_ctl_commit(req, &gpp);
1426 		}
1427 	}
1428 
1429  out:
1430 	if (error && close_on_error) {
1431 		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1432 		table->gpt_opened = 0;
1433 	}
1434 }
1435 
1436 static int
1437 g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1438     struct g_geom *gp)
1439 {
1440 
1441 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1442 	g_topology_assert();
1443 
1444 	g_part_wither(gp, EINVAL);
1445 	return (0);
1446 }
1447 
1448 static struct g_geom *
1449 g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1450 {
1451 	struct g_consumer *cp;
1452 	struct g_geom *gp;
1453 	struct g_part_entry *entry;
1454 	struct g_part_table *table;
1455 	struct root_hold_token *rht;
1456 	int attr, depth;
1457 	int error;
1458 
1459 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1460 	g_topology_assert();
1461 
1462 	/*
1463 	 * Create a GEOM with consumer and hook it up to the provider.
1464 	 * With that we become part of the topology. Optain read access
1465 	 * to the provider.
1466 	 */
1467 	gp = g_new_geomf(mp, "%s", pp->name);
1468 	cp = g_new_consumer(gp);
1469 	error = g_attach(cp, pp);
1470 	if (error == 0)
1471 		error = g_access(cp, 1, 0, 0);
1472 	if (error != 0) {
1473 		g_part_wither(gp, error);
1474 		return (NULL);
1475 	}
1476 
1477 	rht = root_mount_hold(mp->name, M_WAITOK);
1478 	g_topology_unlock();
1479 
1480 	/*
1481 	 * Short-circuit the whole probing galore when there's no
1482 	 * media present.
1483 	 */
1484 	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1485 		error = ENODEV;
1486 		goto fail;
1487 	}
1488 
1489 	/* Make sure we can nest and if so, determine our depth. */
1490 	error = g_getattr("PART::isleaf", cp, &attr);
1491 	if (!error && attr) {
1492 		error = ENODEV;
1493 		goto fail;
1494 	}
1495 	error = g_getattr("PART::depth", cp, &attr);
1496 	depth = (!error) ? attr + 1 : 0;
1497 
1498 	error = g_part_probe(gp, cp, depth);
1499 	if (error)
1500 		goto fail;
1501 
1502 	table = gp->softc;
1503 
1504 	/*
1505 	 * Synthesize a disk geometry. Some partitioning schemes
1506 	 * depend on it and since some file systems need it even
1507 	 * when the partitition scheme doesn't, we do it here in
1508 	 * scheme-independent code.
1509 	 */
1510 	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1511 
1512 	error = G_PART_READ(table, cp);
1513 	if (error)
1514 		goto fail;
1515 
1516 	g_topology_lock();
1517 	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1518 		if (!entry->gpe_internal)
1519 			g_part_new_provider(gp, table, entry);
1520 	}
1521 
1522 	root_mount_rel(rht);
1523 	g_access(cp, -1, 0, 0);
1524 	return (gp);
1525 
1526  fail:
1527 	g_topology_lock();
1528 	root_mount_rel(rht);
1529 	g_access(cp, -1, 0, 0);
1530 	g_part_wither(gp, error);
1531 	return (NULL);
1532 }
1533 
1534 /*
1535  * Geom methods.
1536  */
1537 
1538 static int
1539 g_part_access(struct g_provider *pp, int dr, int dw, int de)
1540 {
1541 	struct g_consumer *cp;
1542 
1543 	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1544 	    dw, de));
1545 
1546 	cp = LIST_FIRST(&pp->geom->consumer);
1547 
1548 	/* We always gain write-exclusive access. */
1549 	return (g_access(cp, dr, dw, dw + de));
1550 }
1551 
1552 static void
1553 g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1554     struct g_consumer *cp, struct g_provider *pp)
1555 {
1556 	char buf[64];
1557 	struct g_part_entry *entry;
1558 	struct g_part_table *table;
1559 
1560 	KASSERT(sb != NULL && gp != NULL, (__func__));
1561 	table = gp->softc;
1562 
1563 	if (indent == NULL) {
1564 		KASSERT(cp == NULL && pp != NULL, (__func__));
1565 		entry = pp->private;
1566 		if (entry == NULL)
1567 			return;
1568 		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
1569 		    (uintmax_t)entry->gpe_offset,
1570 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1571 		/*
1572 		 * libdisk compatibility quirk - the scheme dumps the
1573 		 * slicer name and partition type in a way that is
1574 		 * compatible with libdisk. When libdisk is not used
1575 		 * anymore, this should go away.
1576 		 */
1577 		G_PART_DUMPCONF(table, entry, sb, indent);
1578 	} else if (cp != NULL) {	/* Consumer configuration. */
1579 		KASSERT(pp == NULL, (__func__));
1580 		/* none */
1581 	} else if (pp != NULL) {	/* Provider configuration. */
1582 		entry = pp->private;
1583 		if (entry == NULL)
1584 			return;
1585 		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
1586 		    (uintmax_t)entry->gpe_start);
1587 		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
1588 		    (uintmax_t)entry->gpe_end);
1589 		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
1590 		    entry->gpe_index);
1591 		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1592 		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1593 		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
1594 		    (uintmax_t)entry->gpe_offset);
1595 		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
1596 		    (uintmax_t)pp->mediasize);
1597 		G_PART_DUMPCONF(table, entry, sb, indent);
1598 	} else {			/* Geom configuration. */
1599 		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
1600 		    table->gpt_scheme->name);
1601 		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
1602 		    table->gpt_entries);
1603 		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
1604 		    (uintmax_t)table->gpt_first);
1605 		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
1606 		    (uintmax_t)table->gpt_last);
1607 		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
1608 		    table->gpt_sectors);
1609 		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
1610 		    table->gpt_heads);
1611 		G_PART_DUMPCONF(table, NULL, sb, indent);
1612 	}
1613 }
1614 
1615 static int
1616 g_part_ioctl(struct g_provider *pp, u_long cmd, void *data, int fflag,
1617     struct thread *td)
1618 {
1619 	struct g_geom *gp;
1620 	struct g_part_table *table;
1621 	struct g_part_entry *entry;
1622 	int error;
1623 
1624 	gp = pp->geom;
1625 	table = gp->softc;
1626 	entry = pp->private;
1627 
1628 	switch (cmd) {
1629 	case DIOCGPROVIDERALIAS:
1630 		error = G_PART_DEVALIAS(table, entry, data, MAXPATHLEN);
1631 		break;
1632 	default:
1633 		error = ENOTTY;
1634 		break;
1635 	}
1636 
1637 	return (error);
1638 }
1639 
1640 static void
1641 g_part_orphan(struct g_consumer *cp)
1642 {
1643 	struct g_provider *pp;
1644 
1645 	pp = cp->provider;
1646 	KASSERT(pp != NULL, (__func__));
1647 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
1648 	g_topology_assert();
1649 
1650 	KASSERT(pp->error != 0, (__func__));
1651 	g_part_wither(cp->geom, pp->error);
1652 }
1653 
1654 static void
1655 g_part_spoiled(struct g_consumer *cp)
1656 {
1657 
1658 	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
1659 	g_topology_assert();
1660 
1661 	g_part_wither(cp->geom, ENXIO);
1662 }
1663 
1664 static void
1665 g_part_start(struct bio *bp)
1666 {
1667 	struct bio *bp2;
1668 	struct g_consumer *cp;
1669 	struct g_geom *gp;
1670 	struct g_part_entry *entry;
1671 	struct g_part_table *table;
1672 	struct g_kerneldump *gkd;
1673 	struct g_provider *pp;
1674 
1675 	pp = bp->bio_to;
1676 	gp = pp->geom;
1677 	table = gp->softc;
1678 	cp = LIST_FIRST(&gp->consumer);
1679 
1680 	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
1681 	    pp->name));
1682 
1683 	entry = pp->private;
1684 	if (entry == NULL) {
1685 		g_io_deliver(bp, ENXIO);
1686 		return;
1687 	}
1688 
1689 	switch(bp->bio_cmd) {
1690 	case BIO_DELETE:
1691 	case BIO_READ:
1692 	case BIO_WRITE:
1693 		if (bp->bio_offset >= pp->mediasize) {
1694 			g_io_deliver(bp, EIO);
1695 			return;
1696 		}
1697 		bp2 = g_clone_bio(bp);
1698 		if (bp2 == NULL) {
1699 			g_io_deliver(bp, ENOMEM);
1700 			return;
1701 		}
1702 		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
1703 			bp2->bio_length = pp->mediasize - bp2->bio_offset;
1704 		bp2->bio_done = g_std_done;
1705 		bp2->bio_offset += entry->gpe_offset;
1706 		g_io_request(bp2, cp);
1707 		return;
1708 	case BIO_FLUSH:
1709 		break;
1710 	case BIO_GETATTR:
1711 		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
1712 			return;
1713 		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
1714 			return;
1715 		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
1716 			return;
1717 		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
1718 			return;
1719 		if (g_handleattr_str(bp, "PART::scheme",
1720 		    table->gpt_scheme->name))
1721 			return;
1722 		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
1723 			/*
1724 			 * Check that the partition is suitable for kernel
1725 			 * dumps. Typically only swap partitions should be
1726 			 * used.
1727 			 */
1728 			if (!G_PART_DUMPTO(table, entry)) {
1729 				g_io_deliver(bp, ENODEV);
1730 				printf("GEOM_PART: Partition '%s' not suitable"
1731 				    " for kernel dumps (wrong type?)\n",
1732 				    pp->name);
1733 				return;
1734 			}
1735 			gkd = (struct g_kerneldump *)bp->bio_data;
1736 			if (gkd->offset >= pp->mediasize) {
1737 				g_io_deliver(bp, EIO);
1738 				return;
1739 			}
1740 			if (gkd->offset + gkd->length > pp->mediasize)
1741 				gkd->length = pp->mediasize - gkd->offset;
1742 			gkd->offset += entry->gpe_offset;
1743 		}
1744 		break;
1745 	default:
1746 		g_io_deliver(bp, EOPNOTSUPP);
1747 		return;
1748 	}
1749 
1750 	bp2 = g_clone_bio(bp);
1751 	if (bp2 == NULL) {
1752 		g_io_deliver(bp, ENOMEM);
1753 		return;
1754 	}
1755 	bp2->bio_done = g_std_done;
1756 	g_io_request(bp2, cp);
1757 }
1758 
1759 static void
1760 g_part_init(struct g_class *mp)
1761 {
1762 
1763 	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
1764 }
1765 
1766 static void
1767 g_part_fini(struct g_class *mp)
1768 {
1769 
1770 	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
1771 }
1772 
1773 static void
1774 g_part_unload_event(void *arg, int flag)
1775 {
1776 	struct g_consumer *cp;
1777 	struct g_geom *gp;
1778 	struct g_provider *pp;
1779 	struct g_part_scheme *scheme;
1780 	struct g_part_table *table;
1781 	uintptr_t *xchg;
1782 	int acc, error;
1783 
1784 	if (flag == EV_CANCEL)
1785 		return;
1786 
1787 	xchg = arg;
1788 	error = 0;
1789 	scheme = (void *)(*xchg);
1790 
1791 	g_topology_assert();
1792 
1793 	LIST_FOREACH(gp, &g_part_class.geom, geom) {
1794 		table = gp->softc;
1795 		if (table->gpt_scheme != scheme)
1796 			continue;
1797 
1798 		acc = 0;
1799 		LIST_FOREACH(pp, &gp->provider, provider)
1800 			acc += pp->acr + pp->acw + pp->ace;
1801 		LIST_FOREACH(cp, &gp->consumer, consumer)
1802 			acc += cp->acr + cp->acw + cp->ace;
1803 
1804 		if (!acc)
1805 			g_part_wither(gp, ENOSYS);
1806 		else
1807 			error = EBUSY;
1808 	}
1809 
1810 	if (!error)
1811 		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
1812 
1813 	*xchg = error;
1814 }
1815 
1816 int
1817 g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
1818 {
1819 	uintptr_t arg;
1820 	int error;
1821 
1822 	switch (type) {
1823 	case MOD_LOAD:
1824 		TAILQ_INSERT_TAIL(&g_part_schemes, scheme, scheme_list);
1825 
1826 		error = g_retaste(&g_part_class);
1827 		if (error)
1828 			TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
1829 		break;
1830 	case MOD_UNLOAD:
1831 		arg = (uintptr_t)scheme;
1832 		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
1833 		    NULL);
1834 		if (!error)
1835 			error = (arg == (uintptr_t)scheme) ? EDOOFUS : arg;
1836 		break;
1837 	default:
1838 		error = EOPNOTSUPP;
1839 		break;
1840 	}
1841 
1842 	return (error);
1843 }
1844