g_part.c revision 200534
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: head/sys/geom/part/g_part.c 200534 2009-12-14 20:04:06Z rpaulo $");
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/systm.h>
43#include <sys/uuid.h>
44#include <geom/geom.h>
45#include <geom/geom_ctl.h>
46#include <geom/geom_int.h>
47#include <geom/part/g_part.h>
48
49#include "g_part_if.h"
50
51#ifndef _PATH_DEV
52#define _PATH_DEV "/dev/"
53#endif
54
55static kobj_method_t g_part_null_methods[] = {
56	{ 0, 0 }
57};
58
59static struct g_part_scheme g_part_null_scheme = {
60	"(none)",
61	g_part_null_methods,
62	sizeof(struct g_part_table),
63};
64
65TAILQ_HEAD(, g_part_scheme) g_part_schemes =
66    TAILQ_HEAD_INITIALIZER(g_part_schemes);
67
68struct g_part_alias_list {
69	const char *lexeme;
70	enum g_part_alias alias;
71} g_part_alias_list[G_PART_ALIAS_COUNT] = {
72	{ "apple-boot", G_PART_ALIAS_APPLE_BOOT },
73	{ "apple-hfs", G_PART_ALIAS_APPLE_HFS },
74	{ "apple-label", G_PART_ALIAS_APPLE_LABEL },
75	{ "apple-raid", G_PART_ALIAS_APPLE_RAID },
76	{ "apple-raid-offline", G_PART_ALIAS_APPLE_RAID_OFFLINE },
77	{ "apple-tv-recovery", G_PART_ALIAS_APPLE_TV_RECOVERY },
78	{ "apple-ufs", G_PART_ALIAS_APPLE_UFS },
79	{ "efi", G_PART_ALIAS_EFI },
80	{ "freebsd", G_PART_ALIAS_FREEBSD },
81	{ "freebsd-boot", G_PART_ALIAS_FREEBSD_BOOT },
82	{ "freebsd-swap", G_PART_ALIAS_FREEBSD_SWAP },
83	{ "freebsd-ufs", G_PART_ALIAS_FREEBSD_UFS },
84	{ "freebsd-vinum", G_PART_ALIAS_FREEBSD_VINUM },
85	{ "freebsd-zfs", G_PART_ALIAS_FREEBSD_ZFS },
86	{ "linux-data", G_PART_ALIAS_LINUX_DATA },
87	{ "linux-lvm", G_PART_ALIAS_LINUX_LVM },
88	{ "linux-raid", G_PART_ALIAS_LINUX_RAID },
89	{ "linux-swap", G_PART_ALIAS_LINUX_SWAP },
90	{ "mbr", G_PART_ALIAS_MBR }
91};
92
93/*
94 * The GEOM partitioning class.
95 */
96static g_ctl_req_t g_part_ctlreq;
97static g_ctl_destroy_geom_t g_part_destroy_geom;
98static g_fini_t g_part_fini;
99static g_init_t g_part_init;
100static g_taste_t g_part_taste;
101
102static g_access_t g_part_access;
103static g_dumpconf_t g_part_dumpconf;
104static g_orphan_t g_part_orphan;
105static g_spoiled_t g_part_spoiled;
106static g_start_t g_part_start;
107
108static struct g_class g_part_class = {
109	.name = "PART",
110	.version = G_VERSION,
111	/* Class methods. */
112	.ctlreq = g_part_ctlreq,
113	.destroy_geom = g_part_destroy_geom,
114	.fini = g_part_fini,
115	.init = g_part_init,
116	.taste = g_part_taste,
117	/* Geom methods. */
118	.access = g_part_access,
119	.dumpconf = g_part_dumpconf,
120	.orphan = g_part_orphan,
121	.spoiled = g_part_spoiled,
122	.start = g_part_start,
123};
124
125DECLARE_GEOM_CLASS(g_part_class, g_part);
126
127/*
128 * Support functions.
129 */
130
131static void g_part_wither(struct g_geom *, int);
132
133const char *
134g_part_alias_name(enum g_part_alias alias)
135{
136	int i;
137
138	for (i = 0; i < G_PART_ALIAS_COUNT; i++) {
139		if (g_part_alias_list[i].alias != alias)
140			continue;
141		return (g_part_alias_list[i].lexeme);
142	}
143
144	return (NULL);
145}
146
147void
148g_part_geometry_heads(off_t blocks, u_int sectors, off_t *bestchs,
149    u_int *bestheads)
150{
151	static u_int candidate_heads[] = { 1, 2, 16, 32, 64, 128, 255, 0 };
152	off_t chs, cylinders;
153	u_int heads;
154	int idx;
155
156	*bestchs = 0;
157	*bestheads = 0;
158	for (idx = 0; candidate_heads[idx] != 0; idx++) {
159		heads = candidate_heads[idx];
160		cylinders = blocks / heads / sectors;
161		if (cylinders < heads || cylinders < sectors)
162			break;
163		if (cylinders > 1023)
164			continue;
165		chs = cylinders * heads * sectors;
166		if (chs > *bestchs || (chs == *bestchs && *bestheads == 1)) {
167			*bestchs = chs;
168			*bestheads = heads;
169		}
170	}
171}
172
173static void
174g_part_geometry(struct g_part_table *table, struct g_consumer *cp,
175    off_t blocks)
176{
177	static u_int candidate_sectors[] = { 1, 9, 17, 33, 63, 0 };
178	off_t chs, bestchs;
179	u_int heads, sectors;
180	int idx;
181
182	if (g_getattr("GEOM::fwsectors", cp, &sectors) != 0 || sectors == 0 ||
183	    g_getattr("GEOM::fwheads", cp, &heads) != 0 || heads == 0) {
184		table->gpt_fixgeom = 0;
185		table->gpt_heads = 0;
186		table->gpt_sectors = 0;
187		bestchs = 0;
188		for (idx = 0; candidate_sectors[idx] != 0; idx++) {
189			sectors = candidate_sectors[idx];
190			g_part_geometry_heads(blocks, sectors, &chs, &heads);
191			if (chs == 0)
192				continue;
193			/*
194			 * Prefer a geometry with sectors > 1, but only if
195			 * it doesn't bump down the numbver of heads to 1.
196			 */
197			if (chs > bestchs || (chs == bestchs && heads > 1 &&
198			    table->gpt_sectors == 1)) {
199				bestchs = chs;
200				table->gpt_heads = heads;
201				table->gpt_sectors = sectors;
202			}
203		}
204		/*
205		 * If we didn't find a geometry at all, then the disk is
206		 * too big. This means we can use the maximum number of
207		 * heads and sectors.
208		 */
209		if (bestchs == 0) {
210			table->gpt_heads = 255;
211			table->gpt_sectors = 63;
212		}
213	} else {
214		table->gpt_fixgeom = 1;
215		table->gpt_heads = heads;
216		table->gpt_sectors = sectors;
217	}
218}
219
220struct g_part_entry *
221g_part_new_entry(struct g_part_table *table, int index, quad_t start,
222    quad_t end)
223{
224	struct g_part_entry *entry, *last;
225
226	last = NULL;
227	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
228		if (entry->gpe_index == index)
229			break;
230		if (entry->gpe_index > index) {
231			entry = NULL;
232			break;
233		}
234		last = entry;
235	}
236	if (entry == NULL) {
237		entry = g_malloc(table->gpt_scheme->gps_entrysz,
238		    M_WAITOK | M_ZERO);
239		entry->gpe_index = index;
240		if (last == NULL)
241			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
242		else
243			LIST_INSERT_AFTER(last, entry, gpe_entry);
244	} else
245		entry->gpe_offset = 0;
246	entry->gpe_start = start;
247	entry->gpe_end = end;
248	return (entry);
249}
250
251static void
252g_part_new_provider(struct g_geom *gp, struct g_part_table *table,
253    struct g_part_entry *entry)
254{
255	struct g_consumer *cp;
256	struct g_provider *pp;
257	struct sbuf *sb;
258	off_t offset;
259
260	cp = LIST_FIRST(&gp->consumer);
261	pp = cp->provider;
262
263	offset = entry->gpe_start * pp->sectorsize;
264	if (entry->gpe_offset < offset)
265		entry->gpe_offset = offset;
266
267	if (entry->gpe_pp == NULL) {
268		sb = sbuf_new_auto();
269		G_PART_FULLNAME(table, entry, sb, gp->name);
270		sbuf_finish(sb);
271		entry->gpe_pp = g_new_providerf(gp, "%s", sbuf_data(sb));
272		sbuf_delete(sb);
273		entry->gpe_pp->private = entry;		/* Close the circle. */
274	}
275	entry->gpe_pp->index = entry->gpe_index - 1;	/* index is 1-based. */
276	entry->gpe_pp->mediasize = (entry->gpe_end - entry->gpe_start + 1) *
277	    pp->sectorsize;
278	entry->gpe_pp->mediasize -= entry->gpe_offset - offset;
279	entry->gpe_pp->sectorsize = pp->sectorsize;
280	entry->gpe_pp->flags = pp->flags & G_PF_CANDELETE;
281	if (pp->stripesize > 0) {
282		entry->gpe_pp->stripesize = pp->stripesize;
283		entry->gpe_pp->stripeoffset = (pp->stripeoffset +
284		    entry->gpe_offset) % pp->stripesize;
285	}
286	g_error_provider(entry->gpe_pp, 0);
287}
288
289static int
290g_part_parm_geom(const char *rawname, struct g_geom **v)
291{
292	struct g_geom *gp;
293	const char *pname;
294
295	if (strncmp(rawname, _PATH_DEV, strlen(_PATH_DEV)) == 0)
296		pname = rawname + strlen(_PATH_DEV);
297	else
298		pname = rawname;
299	LIST_FOREACH(gp, &g_part_class.geom, geom) {
300		if (!strcmp(pname, gp->name))
301			break;
302	}
303	if (gp == NULL)
304		return (EINVAL);
305	*v = gp;
306	return (0);
307}
308
309static int
310g_part_parm_provider(const char *pname, struct g_provider **v)
311{
312	struct g_provider *pp;
313
314	if (strncmp(pname, _PATH_DEV, strlen(_PATH_DEV)) == 0)
315		pp = g_provider_by_name(pname + strlen(_PATH_DEV));
316	else
317		pp = g_provider_by_name(pname);
318	if (pp == NULL)
319		return (EINVAL);
320	*v = pp;
321	return (0);
322}
323
324static int
325g_part_parm_quad(const char *p, quad_t *v)
326{
327	char *x;
328	quad_t q;
329
330	q = strtoq(p, &x, 0);
331	if (*x != '\0' || q < 0)
332		return (EINVAL);
333	*v = q;
334	return (0);
335}
336
337static int
338g_part_parm_scheme(const char *p, struct g_part_scheme **v)
339{
340	struct g_part_scheme *s;
341
342	TAILQ_FOREACH(s, &g_part_schemes, scheme_list) {
343		if (s == &g_part_null_scheme)
344			continue;
345		if (!strcasecmp(s->name, p))
346			break;
347	}
348	if (s == NULL)
349		return (EINVAL);
350	*v = s;
351	return (0);
352}
353
354static int
355g_part_parm_str(const char *p, const char **v)
356{
357
358	if (p[0] == '\0')
359		return (EINVAL);
360	*v = p;
361	return (0);
362}
363
364static int
365g_part_parm_uint(const char *p, u_int *v)
366{
367	char *x;
368	long l;
369
370	l = strtol(p, &x, 0);
371	if (*x != '\0' || l < 0 || l > INT_MAX)
372		return (EINVAL);
373	*v = (unsigned int)l;
374	return (0);
375}
376
377static int
378g_part_probe(struct g_geom *gp, struct g_consumer *cp, int depth)
379{
380	struct g_part_scheme *iter, *scheme;
381	struct g_part_table *table;
382	int pri, probe;
383
384	table = gp->softc;
385	scheme = (table != NULL) ? table->gpt_scheme : NULL;
386	pri = (scheme != NULL) ? G_PART_PROBE(table, cp) : INT_MIN;
387	if (pri == 0)
388		goto done;
389	if (pri > 0) {	/* error */
390		scheme = NULL;
391		pri = INT_MIN;
392	}
393
394	TAILQ_FOREACH(iter, &g_part_schemes, scheme_list) {
395		if (iter == &g_part_null_scheme)
396			continue;
397		table = (void *)kobj_create((kobj_class_t)iter, M_GEOM,
398		    M_WAITOK);
399		table->gpt_gp = gp;
400		table->gpt_scheme = iter;
401		table->gpt_depth = depth;
402		probe = G_PART_PROBE(table, cp);
403		if (probe <= 0 && probe > pri) {
404			pri = probe;
405			scheme = iter;
406			if (gp->softc != NULL)
407				kobj_delete((kobj_t)gp->softc, M_GEOM);
408			gp->softc = table;
409			if (pri == 0)
410				goto done;
411		} else
412			kobj_delete((kobj_t)table, M_GEOM);
413	}
414
415done:
416	return ((scheme == NULL) ? ENXIO : 0);
417}
418
419/*
420 * Control request functions.
421 */
422
423static int
424g_part_ctl_add(struct gctl_req *req, struct g_part_parms *gpp)
425{
426	struct g_geom *gp;
427	struct g_provider *pp;
428	struct g_part_entry *delent, *last, *entry;
429	struct g_part_table *table;
430	struct sbuf *sb;
431	quad_t end;
432	unsigned int index;
433	int error;
434
435	gp = gpp->gpp_geom;
436	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
437	g_topology_assert();
438
439	pp = LIST_FIRST(&gp->consumer)->provider;
440	table = gp->softc;
441	end = gpp->gpp_start + gpp->gpp_size - 1;
442
443	if (gpp->gpp_start < table->gpt_first ||
444	    gpp->gpp_start > table->gpt_last) {
445		gctl_error(req, "%d start '%jd'", EINVAL,
446		    (intmax_t)gpp->gpp_start);
447		return (EINVAL);
448	}
449	if (end < gpp->gpp_start || end > table->gpt_last) {
450		gctl_error(req, "%d size '%jd'", EINVAL,
451		    (intmax_t)gpp->gpp_size);
452		return (EINVAL);
453	}
454	if (gpp->gpp_index > table->gpt_entries) {
455		gctl_error(req, "%d index '%d'", EINVAL, gpp->gpp_index);
456		return (EINVAL);
457	}
458
459	delent = last = NULL;
460	index = (gpp->gpp_index > 0) ? gpp->gpp_index : 1;
461	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
462		if (entry->gpe_deleted) {
463			if (entry->gpe_index == index)
464				delent = entry;
465			continue;
466		}
467		if (entry->gpe_index == index)
468			index = entry->gpe_index + 1;
469		if (entry->gpe_index < index)
470			last = entry;
471		if (entry->gpe_internal)
472			continue;
473		if (gpp->gpp_start >= entry->gpe_start &&
474		    gpp->gpp_start <= entry->gpe_end) {
475			gctl_error(req, "%d start '%jd'", ENOSPC,
476			    (intmax_t)gpp->gpp_start);
477			return (ENOSPC);
478		}
479		if (end >= entry->gpe_start && end <= entry->gpe_end) {
480			gctl_error(req, "%d end '%jd'", ENOSPC, (intmax_t)end);
481			return (ENOSPC);
482		}
483		if (gpp->gpp_start < entry->gpe_start && end > entry->gpe_end) {
484			gctl_error(req, "%d size '%jd'", ENOSPC,
485			    (intmax_t)gpp->gpp_size);
486			return (ENOSPC);
487		}
488	}
489	if (gpp->gpp_index > 0 && index != gpp->gpp_index) {
490		gctl_error(req, "%d index '%d'", EEXIST, gpp->gpp_index);
491		return (EEXIST);
492	}
493	if (index > table->gpt_entries) {
494		gctl_error(req, "%d index '%d'", ENOSPC, index);
495		return (ENOSPC);
496	}
497
498	entry = (delent == NULL) ? g_malloc(table->gpt_scheme->gps_entrysz,
499	    M_WAITOK | M_ZERO) : delent;
500	entry->gpe_index = index;
501	entry->gpe_start = gpp->gpp_start;
502	entry->gpe_end = end;
503	error = G_PART_ADD(table, entry, gpp);
504	if (error) {
505		gctl_error(req, "%d", error);
506		if (delent == NULL)
507			g_free(entry);
508		return (error);
509	}
510	if (delent == NULL) {
511		if (last == NULL)
512			LIST_INSERT_HEAD(&table->gpt_entry, entry, gpe_entry);
513		else
514			LIST_INSERT_AFTER(last, entry, gpe_entry);
515		entry->gpe_created = 1;
516	} else {
517		entry->gpe_deleted = 0;
518		entry->gpe_modified = 1;
519	}
520	g_part_new_provider(gp, table, entry);
521
522	/* Provide feedback if so requested. */
523	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
524		sb = sbuf_new_auto();
525		G_PART_FULLNAME(table, entry, sb, gp->name);
526		sbuf_cat(sb, " added\n");
527		sbuf_finish(sb);
528		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
529		sbuf_delete(sb);
530	}
531	return (0);
532}
533
534static int
535g_part_ctl_bootcode(struct gctl_req *req, struct g_part_parms *gpp)
536{
537	struct g_geom *gp;
538	struct g_part_table *table;
539	struct sbuf *sb;
540	int error, sz;
541
542	gp = gpp->gpp_geom;
543	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
544	g_topology_assert();
545
546	table = gp->softc;
547	sz = table->gpt_scheme->gps_bootcodesz;
548	if (sz == 0) {
549		error = ENODEV;
550		goto fail;
551	}
552	if (gpp->gpp_codesize > sz) {
553		error = EFBIG;
554		goto fail;
555	}
556
557	error = G_PART_BOOTCODE(table, gpp);
558	if (error)
559		goto fail;
560
561	/* Provide feedback if so requested. */
562	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
563		sb = sbuf_new_auto();
564		sbuf_printf(sb, "%s has bootcode\n", gp->name);
565		sbuf_finish(sb);
566		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
567		sbuf_delete(sb);
568	}
569	return (0);
570
571 fail:
572	gctl_error(req, "%d", error);
573	return (error);
574}
575
576static int
577g_part_ctl_commit(struct gctl_req *req, struct g_part_parms *gpp)
578{
579	struct g_consumer *cp;
580	struct g_geom *gp;
581	struct g_provider *pp;
582	struct g_part_entry *entry, *tmp;
583	struct g_part_table *table;
584	char *buf;
585	int error, i;
586
587	gp = gpp->gpp_geom;
588	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
589	g_topology_assert();
590
591	table = gp->softc;
592	if (!table->gpt_opened) {
593		gctl_error(req, "%d", EPERM);
594		return (EPERM);
595	}
596
597	g_topology_unlock();
598
599	cp = LIST_FIRST(&gp->consumer);
600	if ((table->gpt_smhead | table->gpt_smtail) != 0) {
601		pp = cp->provider;
602		buf = g_malloc(pp->sectorsize, M_WAITOK | M_ZERO);
603		while (table->gpt_smhead != 0) {
604			i = ffs(table->gpt_smhead) - 1;
605			error = g_write_data(cp, i * pp->sectorsize, buf,
606			    pp->sectorsize);
607			if (error) {
608				g_free(buf);
609				goto fail;
610			}
611			table->gpt_smhead &= ~(1 << i);
612		}
613		while (table->gpt_smtail != 0) {
614			i = ffs(table->gpt_smtail) - 1;
615			error = g_write_data(cp, pp->mediasize - (i + 1) *
616			    pp->sectorsize, buf, pp->sectorsize);
617			if (error) {
618				g_free(buf);
619				goto fail;
620			}
621			table->gpt_smtail &= ~(1 << i);
622		}
623		g_free(buf);
624	}
625
626	if (table->gpt_scheme == &g_part_null_scheme) {
627		g_topology_lock();
628		g_access(cp, -1, -1, -1);
629		g_part_wither(gp, ENXIO);
630		return (0);
631	}
632
633	error = G_PART_WRITE(table, cp);
634	if (error)
635		goto fail;
636
637	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
638		if (!entry->gpe_deleted) {
639			entry->gpe_created = 0;
640			entry->gpe_modified = 0;
641			continue;
642		}
643		LIST_REMOVE(entry, gpe_entry);
644		g_free(entry);
645	}
646	table->gpt_created = 0;
647	table->gpt_opened = 0;
648
649	g_topology_lock();
650	g_access(cp, -1, -1, -1);
651	return (0);
652
653fail:
654	g_topology_lock();
655	gctl_error(req, "%d", error);
656	return (error);
657}
658
659static int
660g_part_ctl_create(struct gctl_req *req, struct g_part_parms *gpp)
661{
662	struct g_consumer *cp;
663	struct g_geom *gp;
664	struct g_provider *pp;
665	struct g_part_scheme *scheme;
666	struct g_part_table *null, *table;
667	struct sbuf *sb;
668	int attr, error;
669
670	pp = gpp->gpp_provider;
671	scheme = gpp->gpp_scheme;
672	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
673	g_topology_assert();
674
675	/* Check that there isn't already a g_part geom on the provider. */
676	error = g_part_parm_geom(pp->name, &gp);
677	if (!error) {
678		null = gp->softc;
679		if (null->gpt_scheme != &g_part_null_scheme) {
680			gctl_error(req, "%d geom '%s'", EEXIST, pp->name);
681			return (EEXIST);
682		}
683	} else
684		null = NULL;
685
686	if ((gpp->gpp_parms & G_PART_PARM_ENTRIES) &&
687	    (gpp->gpp_entries < scheme->gps_minent ||
688	     gpp->gpp_entries > scheme->gps_maxent)) {
689		gctl_error(req, "%d entries '%d'", EINVAL, gpp->gpp_entries);
690		return (EINVAL);
691	}
692
693	if (null == NULL)
694		gp = g_new_geomf(&g_part_class, "%s", pp->name);
695	gp->softc = kobj_create((kobj_class_t)gpp->gpp_scheme, M_GEOM,
696	    M_WAITOK);
697	table = gp->softc;
698	table->gpt_gp = gp;
699	table->gpt_scheme = gpp->gpp_scheme;
700	table->gpt_entries = (gpp->gpp_parms & G_PART_PARM_ENTRIES) ?
701	    gpp->gpp_entries : scheme->gps_minent;
702	LIST_INIT(&table->gpt_entry);
703	if (null == NULL) {
704		cp = g_new_consumer(gp);
705		error = g_attach(cp, pp);
706		if (error == 0)
707			error = g_access(cp, 1, 1, 1);
708		if (error != 0) {
709			g_part_wither(gp, error);
710			gctl_error(req, "%d geom '%s'", error, pp->name);
711			return (error);
712		}
713		table->gpt_opened = 1;
714	} else {
715		cp = LIST_FIRST(&gp->consumer);
716		table->gpt_opened = null->gpt_opened;
717		table->gpt_smhead = null->gpt_smhead;
718		table->gpt_smtail = null->gpt_smtail;
719	}
720
721	g_topology_unlock();
722
723	/* Make sure the provider has media. */
724	if (pp->mediasize == 0 || pp->sectorsize == 0) {
725		error = ENODEV;
726		goto fail;
727	}
728
729	/* Make sure we can nest and if so, determine our depth. */
730	error = g_getattr("PART::isleaf", cp, &attr);
731	if (!error && attr) {
732		error = ENODEV;
733		goto fail;
734	}
735	error = g_getattr("PART::depth", cp, &attr);
736	table->gpt_depth = (!error) ? attr + 1 : 0;
737
738	/*
739	 * Synthesize a disk geometry. Some partitioning schemes
740	 * depend on it and since some file systems need it even
741	 * when the partitition scheme doesn't, we do it here in
742	 * scheme-independent code.
743	 */
744	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
745
746	error = G_PART_CREATE(table, gpp);
747	if (error)
748		goto fail;
749
750	g_topology_lock();
751
752	table->gpt_created = 1;
753	if (null != NULL)
754		kobj_delete((kobj_t)null, M_GEOM);
755
756	/*
757	 * Support automatic commit by filling in the gpp_geom
758	 * parameter.
759	 */
760	gpp->gpp_parms |= G_PART_PARM_GEOM;
761	gpp->gpp_geom = gp;
762
763	/* Provide feedback if so requested. */
764	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
765		sb = sbuf_new_auto();
766		sbuf_printf(sb, "%s created\n", gp->name);
767		sbuf_finish(sb);
768		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
769		sbuf_delete(sb);
770	}
771	return (0);
772
773fail:
774	g_topology_lock();
775	if (null == NULL) {
776		g_access(cp, -1, -1, -1);
777		g_part_wither(gp, error);
778	} else {
779		kobj_delete((kobj_t)gp->softc, M_GEOM);
780		gp->softc = null;
781	}
782	gctl_error(req, "%d provider", error);
783	return (error);
784}
785
786static int
787g_part_ctl_delete(struct gctl_req *req, struct g_part_parms *gpp)
788{
789	struct g_geom *gp;
790	struct g_provider *pp;
791	struct g_part_entry *entry;
792	struct g_part_table *table;
793	struct sbuf *sb;
794
795	gp = gpp->gpp_geom;
796	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
797	g_topology_assert();
798
799	table = gp->softc;
800
801	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
802		if (entry->gpe_deleted || entry->gpe_internal)
803			continue;
804		if (entry->gpe_index == gpp->gpp_index)
805			break;
806	}
807	if (entry == NULL) {
808		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
809		return (ENOENT);
810	}
811
812	pp = entry->gpe_pp;
813	if (pp != NULL) {
814		if (pp->acr > 0 || pp->acw > 0 || pp->ace > 0) {
815			gctl_error(req, "%d", EBUSY);
816			return (EBUSY);
817		}
818
819		pp->private = NULL;
820		entry->gpe_pp = NULL;
821	}
822
823	if (entry->gpe_created) {
824		LIST_REMOVE(entry, gpe_entry);
825		g_free(entry);
826	} else {
827		entry->gpe_modified = 0;
828		entry->gpe_deleted = 1;
829	}
830
831	if (pp != NULL)
832		g_wither_provider(pp, ENXIO);
833
834	/* Provide feedback if so requested. */
835	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
836		sb = sbuf_new_auto();
837		G_PART_FULLNAME(table, entry, sb, gp->name);
838		sbuf_cat(sb, " deleted\n");
839		sbuf_finish(sb);
840		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
841		sbuf_delete(sb);
842	}
843	return (0);
844}
845
846static int
847g_part_ctl_destroy(struct gctl_req *req, struct g_part_parms *gpp)
848{
849	struct g_geom *gp;
850	struct g_part_entry *entry;
851	struct g_part_table *null, *table;
852	struct sbuf *sb;
853	int error;
854
855	gp = gpp->gpp_geom;
856	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
857	g_topology_assert();
858
859	table = gp->softc;
860	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
861		if (entry->gpe_deleted || entry->gpe_internal)
862			continue;
863		gctl_error(req, "%d", EBUSY);
864		return (EBUSY);
865	}
866
867	error = G_PART_DESTROY(table, gpp);
868	if (error) {
869		gctl_error(req, "%d", error);
870		return (error);
871	}
872
873	gp->softc = kobj_create((kobj_class_t)&g_part_null_scheme, M_GEOM,
874	    M_WAITOK);
875	null = gp->softc;
876	null->gpt_gp = gp;
877	null->gpt_scheme = &g_part_null_scheme;
878	LIST_INIT(&null->gpt_entry);
879	null->gpt_depth = table->gpt_depth;
880	null->gpt_opened = table->gpt_opened;
881	null->gpt_smhead = table->gpt_smhead;
882	null->gpt_smtail = table->gpt_smtail;
883
884	while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
885		LIST_REMOVE(entry, gpe_entry);
886		g_free(entry);
887	}
888	kobj_delete((kobj_t)table, M_GEOM);
889
890	/* Provide feedback if so requested. */
891	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
892		sb = sbuf_new_auto();
893		sbuf_printf(sb, "%s destroyed\n", gp->name);
894		sbuf_finish(sb);
895		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
896		sbuf_delete(sb);
897	}
898	return (0);
899}
900
901static int
902g_part_ctl_modify(struct gctl_req *req, struct g_part_parms *gpp)
903{
904	struct g_geom *gp;
905	struct g_part_entry *entry;
906	struct g_part_table *table;
907	struct sbuf *sb;
908	int error;
909
910	gp = gpp->gpp_geom;
911	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
912	g_topology_assert();
913
914	table = gp->softc;
915
916	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
917		if (entry->gpe_deleted || entry->gpe_internal)
918			continue;
919		if (entry->gpe_index == gpp->gpp_index)
920			break;
921	}
922	if (entry == NULL) {
923		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
924		return (ENOENT);
925	}
926
927	error = G_PART_MODIFY(table, entry, gpp);
928	if (error) {
929		gctl_error(req, "%d", error);
930		return (error);
931	}
932
933	if (!entry->gpe_created)
934		entry->gpe_modified = 1;
935
936	/* Provide feedback if so requested. */
937	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
938		sb = sbuf_new_auto();
939		G_PART_FULLNAME(table, entry, sb, gp->name);
940		sbuf_cat(sb, " modified\n");
941		sbuf_finish(sb);
942		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
943		sbuf_delete(sb);
944	}
945	return (0);
946}
947
948static int
949g_part_ctl_move(struct gctl_req *req, struct g_part_parms *gpp)
950{
951	gctl_error(req, "%d verb 'move'", ENOSYS);
952	return (ENOSYS);
953}
954
955static int
956g_part_ctl_recover(struct gctl_req *req, struct g_part_parms *gpp)
957{
958	gctl_error(req, "%d verb 'recover'", ENOSYS);
959	return (ENOSYS);
960}
961
962static int
963g_part_ctl_resize(struct gctl_req *req, struct g_part_parms *gpp)
964{
965	gctl_error(req, "%d verb 'resize'", ENOSYS);
966	return (ENOSYS);
967}
968
969static int
970g_part_ctl_setunset(struct gctl_req *req, struct g_part_parms *gpp,
971    unsigned int set)
972{
973	struct g_geom *gp;
974	struct g_part_entry *entry;
975	struct g_part_table *table;
976	struct sbuf *sb;
977	int error;
978
979	gp = gpp->gpp_geom;
980	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
981	g_topology_assert();
982
983	table = gp->softc;
984
985	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
986		if (entry->gpe_deleted || entry->gpe_internal)
987			continue;
988		if (entry->gpe_index == gpp->gpp_index)
989			break;
990	}
991	if (entry == NULL) {
992		gctl_error(req, "%d index '%d'", ENOENT, gpp->gpp_index);
993		return (ENOENT);
994	}
995
996	error = G_PART_SETUNSET(table, entry, gpp->gpp_attrib, set);
997	if (error) {
998		gctl_error(req, "%d attrib '%s'", error, gpp->gpp_attrib);
999		return (error);
1000	}
1001
1002	/* Provide feedback if so requested. */
1003	if (gpp->gpp_parms & G_PART_PARM_OUTPUT) {
1004		sb = sbuf_new_auto();
1005		G_PART_FULLNAME(table, entry, sb, gp->name);
1006		sbuf_printf(sb, " has %s %sset\n", gpp->gpp_attrib,
1007		    (set) ? "" : "un");
1008		sbuf_finish(sb);
1009		gctl_set_param(req, "output", sbuf_data(sb), sbuf_len(sb) + 1);
1010		sbuf_delete(sb);
1011	}
1012	return (0);
1013}
1014
1015static int
1016g_part_ctl_undo(struct gctl_req *req, struct g_part_parms *gpp)
1017{
1018	struct g_consumer *cp;
1019	struct g_provider *pp;
1020	struct g_geom *gp;
1021	struct g_part_entry *entry, *tmp;
1022	struct g_part_table *table;
1023	int error, reprobe;
1024
1025	gp = gpp->gpp_geom;
1026	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, gp->name));
1027	g_topology_assert();
1028
1029	table = gp->softc;
1030	if (!table->gpt_opened) {
1031		gctl_error(req, "%d", EPERM);
1032		return (EPERM);
1033	}
1034
1035	cp = LIST_FIRST(&gp->consumer);
1036	LIST_FOREACH_SAFE(entry, &table->gpt_entry, gpe_entry, tmp) {
1037		entry->gpe_modified = 0;
1038		if (entry->gpe_created) {
1039			pp = entry->gpe_pp;
1040			if (pp != NULL) {
1041				pp->private = NULL;
1042				entry->gpe_pp = NULL;
1043				g_wither_provider(pp, ENXIO);
1044			}
1045			entry->gpe_deleted = 1;
1046		}
1047		if (entry->gpe_deleted) {
1048			LIST_REMOVE(entry, gpe_entry);
1049			g_free(entry);
1050		}
1051	}
1052
1053	g_topology_unlock();
1054
1055	reprobe = (table->gpt_scheme == &g_part_null_scheme ||
1056	    table->gpt_created) ? 1 : 0;
1057
1058	if (reprobe) {
1059		if (!LIST_EMPTY(&table->gpt_entry)) {
1060			error = EBUSY;
1061			goto fail;
1062		}
1063		error = g_part_probe(gp, cp, table->gpt_depth);
1064		if (error) {
1065			g_topology_lock();
1066			g_access(cp, -1, -1, -1);
1067			g_part_wither(gp, error);
1068			return (0);
1069		}
1070		table = gp->softc;
1071	}
1072
1073	error = G_PART_READ(table, cp);
1074	if (error)
1075		goto fail;
1076
1077	g_topology_lock();
1078
1079	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1080		if (!entry->gpe_internal)
1081			g_part_new_provider(gp, table, entry);
1082	}
1083
1084	table->gpt_opened = 0;
1085	g_access(cp, -1, -1, -1);
1086	return (0);
1087
1088fail:
1089	g_topology_lock();
1090	gctl_error(req, "%d", error);
1091	return (error);
1092}
1093
1094static void
1095g_part_wither(struct g_geom *gp, int error)
1096{
1097	struct g_part_entry *entry;
1098	struct g_part_table *table;
1099
1100	table = gp->softc;
1101	if (table != NULL) {
1102		while ((entry = LIST_FIRST(&table->gpt_entry)) != NULL) {
1103			LIST_REMOVE(entry, gpe_entry);
1104			g_free(entry);
1105		}
1106		if (gp->softc != NULL) {
1107			kobj_delete((kobj_t)gp->softc, M_GEOM);
1108			gp->softc = NULL;
1109		}
1110	}
1111	g_wither_geom(gp, error);
1112}
1113
1114/*
1115 * Class methods.
1116 */
1117
1118static void
1119g_part_ctlreq(struct gctl_req *req, struct g_class *mp, const char *verb)
1120{
1121	struct g_part_parms gpp;
1122	struct g_part_table *table;
1123	struct gctl_req_arg *ap;
1124	const char *p;
1125	enum g_part_ctl ctlreq;
1126	unsigned int i, mparms, oparms, parm;
1127	int auto_commit, close_on_error;
1128	int error, len, modifies;
1129
1130	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, verb));
1131	g_topology_assert();
1132
1133	ctlreq = G_PART_CTL_NONE;
1134	modifies = 1;
1135	mparms = 0;
1136	oparms = G_PART_PARM_FLAGS | G_PART_PARM_OUTPUT | G_PART_PARM_VERSION;
1137	switch (*verb) {
1138	case 'a':
1139		if (!strcmp(verb, "add")) {
1140			ctlreq = G_PART_CTL_ADD;
1141			mparms |= G_PART_PARM_GEOM | G_PART_PARM_SIZE |
1142			    G_PART_PARM_START | G_PART_PARM_TYPE;
1143			oparms |= G_PART_PARM_INDEX | G_PART_PARM_LABEL;
1144		}
1145		break;
1146	case 'b':
1147		if (!strcmp(verb, "bootcode")) {
1148			ctlreq = G_PART_CTL_BOOTCODE;
1149			mparms |= G_PART_PARM_GEOM | G_PART_PARM_BOOTCODE;
1150		}
1151		break;
1152	case 'c':
1153		if (!strcmp(verb, "commit")) {
1154			ctlreq = G_PART_CTL_COMMIT;
1155			mparms |= G_PART_PARM_GEOM;
1156			modifies = 0;
1157		} else if (!strcmp(verb, "create")) {
1158			ctlreq = G_PART_CTL_CREATE;
1159			mparms |= G_PART_PARM_PROVIDER | G_PART_PARM_SCHEME;
1160			oparms |= G_PART_PARM_ENTRIES;
1161		}
1162		break;
1163	case 'd':
1164		if (!strcmp(verb, "delete")) {
1165			ctlreq = G_PART_CTL_DELETE;
1166			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1167		} else if (!strcmp(verb, "destroy")) {
1168			ctlreq = G_PART_CTL_DESTROY;
1169			mparms |= G_PART_PARM_GEOM;
1170		}
1171		break;
1172	case 'm':
1173		if (!strcmp(verb, "modify")) {
1174			ctlreq = G_PART_CTL_MODIFY;
1175			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1176			oparms |= G_PART_PARM_LABEL | G_PART_PARM_TYPE;
1177		} else if (!strcmp(verb, "move")) {
1178			ctlreq = G_PART_CTL_MOVE;
1179			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1180		}
1181		break;
1182	case 'r':
1183		if (!strcmp(verb, "recover")) {
1184			ctlreq = G_PART_CTL_RECOVER;
1185			mparms |= G_PART_PARM_GEOM;
1186		} else if (!strcmp(verb, "resize")) {
1187			ctlreq = G_PART_CTL_RESIZE;
1188			mparms |= G_PART_PARM_GEOM | G_PART_PARM_INDEX;
1189		}
1190		break;
1191	case 's':
1192		if (!strcmp(verb, "set")) {
1193			ctlreq = G_PART_CTL_SET;
1194			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1195			    G_PART_PARM_INDEX;
1196		}
1197		break;
1198	case 'u':
1199		if (!strcmp(verb, "undo")) {
1200			ctlreq = G_PART_CTL_UNDO;
1201			mparms |= G_PART_PARM_GEOM;
1202			modifies = 0;
1203		} else if (!strcmp(verb, "unset")) {
1204			ctlreq = G_PART_CTL_UNSET;
1205			mparms |= G_PART_PARM_ATTRIB | G_PART_PARM_GEOM |
1206			    G_PART_PARM_INDEX;
1207		}
1208		break;
1209	}
1210	if (ctlreq == G_PART_CTL_NONE) {
1211		gctl_error(req, "%d verb '%s'", EINVAL, verb);
1212		return;
1213	}
1214
1215	bzero(&gpp, sizeof(gpp));
1216	for (i = 0; i < req->narg; i++) {
1217		ap = &req->arg[i];
1218		parm = 0;
1219		switch (ap->name[0]) {
1220		case 'a':
1221			if (!strcmp(ap->name, "attrib"))
1222				parm = G_PART_PARM_ATTRIB;
1223			break;
1224		case 'b':
1225			if (!strcmp(ap->name, "bootcode"))
1226				parm = G_PART_PARM_BOOTCODE;
1227			break;
1228		case 'c':
1229			if (!strcmp(ap->name, "class"))
1230				continue;
1231			break;
1232		case 'e':
1233			if (!strcmp(ap->name, "entries"))
1234				parm = G_PART_PARM_ENTRIES;
1235			break;
1236		case 'f':
1237			if (!strcmp(ap->name, "flags"))
1238				parm = G_PART_PARM_FLAGS;
1239			break;
1240		case 'g':
1241			if (!strcmp(ap->name, "geom"))
1242				parm = G_PART_PARM_GEOM;
1243			break;
1244		case 'i':
1245			if (!strcmp(ap->name, "index"))
1246				parm = G_PART_PARM_INDEX;
1247			break;
1248		case 'l':
1249			if (!strcmp(ap->name, "label"))
1250				parm = G_PART_PARM_LABEL;
1251			break;
1252		case 'o':
1253			if (!strcmp(ap->name, "output"))
1254				parm = G_PART_PARM_OUTPUT;
1255			break;
1256		case 'p':
1257			if (!strcmp(ap->name, "provider"))
1258				parm = G_PART_PARM_PROVIDER;
1259			break;
1260		case 's':
1261			if (!strcmp(ap->name, "scheme"))
1262				parm = G_PART_PARM_SCHEME;
1263			else if (!strcmp(ap->name, "size"))
1264				parm = G_PART_PARM_SIZE;
1265			else if (!strcmp(ap->name, "start"))
1266				parm = G_PART_PARM_START;
1267			break;
1268		case 't':
1269			if (!strcmp(ap->name, "type"))
1270				parm = G_PART_PARM_TYPE;
1271			break;
1272		case 'v':
1273			if (!strcmp(ap->name, "verb"))
1274				continue;
1275			else if (!strcmp(ap->name, "version"))
1276				parm = G_PART_PARM_VERSION;
1277			break;
1278		}
1279		if ((parm & (mparms | oparms)) == 0) {
1280			gctl_error(req, "%d param '%s'", EINVAL, ap->name);
1281			return;
1282		}
1283		if (parm == G_PART_PARM_BOOTCODE)
1284			p = gctl_get_param(req, ap->name, &len);
1285		else
1286			p = gctl_get_asciiparam(req, ap->name);
1287		if (p == NULL) {
1288			gctl_error(req, "%d param '%s'", ENOATTR, ap->name);
1289			return;
1290		}
1291		switch (parm) {
1292		case G_PART_PARM_ATTRIB:
1293			error = g_part_parm_str(p, &gpp.gpp_attrib);
1294			break;
1295		case G_PART_PARM_BOOTCODE:
1296			gpp.gpp_codeptr = p;
1297			gpp.gpp_codesize = len;
1298			error = 0;
1299			break;
1300		case G_PART_PARM_ENTRIES:
1301			error = g_part_parm_uint(p, &gpp.gpp_entries);
1302			break;
1303		case G_PART_PARM_FLAGS:
1304			if (p[0] == '\0')
1305				continue;
1306			error = g_part_parm_str(p, &gpp.gpp_flags);
1307			break;
1308		case G_PART_PARM_GEOM:
1309			error = g_part_parm_geom(p, &gpp.gpp_geom);
1310			break;
1311		case G_PART_PARM_INDEX:
1312			error = g_part_parm_uint(p, &gpp.gpp_index);
1313			break;
1314		case G_PART_PARM_LABEL:
1315			/* An empty label is always valid. */
1316			gpp.gpp_label = p;
1317			error = 0;
1318			break;
1319		case G_PART_PARM_OUTPUT:
1320			error = 0;	/* Write-only parameter */
1321			break;
1322		case G_PART_PARM_PROVIDER:
1323			error = g_part_parm_provider(p, &gpp.gpp_provider);
1324			break;
1325		case G_PART_PARM_SCHEME:
1326			error = g_part_parm_scheme(p, &gpp.gpp_scheme);
1327			break;
1328		case G_PART_PARM_SIZE:
1329			error = g_part_parm_quad(p, &gpp.gpp_size);
1330			break;
1331		case G_PART_PARM_START:
1332			error = g_part_parm_quad(p, &gpp.gpp_start);
1333			break;
1334		case G_PART_PARM_TYPE:
1335			error = g_part_parm_str(p, &gpp.gpp_type);
1336			break;
1337		case G_PART_PARM_VERSION:
1338			error = g_part_parm_uint(p, &gpp.gpp_version);
1339			break;
1340		default:
1341			error = EDOOFUS;
1342			break;
1343		}
1344		if (error) {
1345			gctl_error(req, "%d %s '%s'", error, ap->name, p);
1346			return;
1347		}
1348		gpp.gpp_parms |= parm;
1349	}
1350	if ((gpp.gpp_parms & mparms) != mparms) {
1351		parm = mparms - (gpp.gpp_parms & mparms);
1352		gctl_error(req, "%d param '%x'", ENOATTR, parm);
1353		return;
1354	}
1355
1356	/* Obtain permissions if possible/necessary. */
1357	close_on_error = 0;
1358	table = NULL;
1359	if (modifies && (gpp.gpp_parms & G_PART_PARM_GEOM)) {
1360		table = gpp.gpp_geom->softc;
1361		if (table != NULL && !table->gpt_opened) {
1362			error = g_access(LIST_FIRST(&gpp.gpp_geom->consumer),
1363			    1, 1, 1);
1364			if (error) {
1365				gctl_error(req, "%d geom '%s'", error,
1366				    gpp.gpp_geom->name);
1367				return;
1368			}
1369			table->gpt_opened = 1;
1370			close_on_error = 1;
1371		}
1372	}
1373
1374	/* Allow the scheme to check or modify the parameters. */
1375	if (table != NULL) {
1376		error = G_PART_PRECHECK(table, ctlreq, &gpp);
1377		if (error) {
1378			gctl_error(req, "%d pre-check failed", error);
1379			goto out;
1380		}
1381	} else
1382		error = EDOOFUS;	/* Prevent bogus uninit. warning. */
1383
1384	switch (ctlreq) {
1385	case G_PART_CTL_NONE:
1386		panic("%s", __func__);
1387	case G_PART_CTL_ADD:
1388		error = g_part_ctl_add(req, &gpp);
1389		break;
1390	case G_PART_CTL_BOOTCODE:
1391		error = g_part_ctl_bootcode(req, &gpp);
1392		break;
1393	case G_PART_CTL_COMMIT:
1394		error = g_part_ctl_commit(req, &gpp);
1395		break;
1396	case G_PART_CTL_CREATE:
1397		error = g_part_ctl_create(req, &gpp);
1398		break;
1399	case G_PART_CTL_DELETE:
1400		error = g_part_ctl_delete(req, &gpp);
1401		break;
1402	case G_PART_CTL_DESTROY:
1403		error = g_part_ctl_destroy(req, &gpp);
1404		break;
1405	case G_PART_CTL_MODIFY:
1406		error = g_part_ctl_modify(req, &gpp);
1407		break;
1408	case G_PART_CTL_MOVE:
1409		error = g_part_ctl_move(req, &gpp);
1410		break;
1411	case G_PART_CTL_RECOVER:
1412		error = g_part_ctl_recover(req, &gpp);
1413		break;
1414	case G_PART_CTL_RESIZE:
1415		error = g_part_ctl_resize(req, &gpp);
1416		break;
1417	case G_PART_CTL_SET:
1418		error = g_part_ctl_setunset(req, &gpp, 1);
1419		break;
1420	case G_PART_CTL_UNDO:
1421		error = g_part_ctl_undo(req, &gpp);
1422		break;
1423	case G_PART_CTL_UNSET:
1424		error = g_part_ctl_setunset(req, &gpp, 0);
1425		break;
1426	}
1427
1428	/* Implement automatic commit. */
1429	if (!error) {
1430		auto_commit = (modifies &&
1431		    (gpp.gpp_parms & G_PART_PARM_FLAGS) &&
1432		    strchr(gpp.gpp_flags, 'C') != NULL) ? 1 : 0;
1433		if (auto_commit) {
1434			KASSERT(gpp.gpp_parms & G_PART_PARM_GEOM, (__func__));
1435			error = g_part_ctl_commit(req, &gpp);
1436		}
1437	}
1438
1439 out:
1440	if (error && close_on_error) {
1441		g_access(LIST_FIRST(&gpp.gpp_geom->consumer), -1, -1, -1);
1442		table->gpt_opened = 0;
1443	}
1444}
1445
1446static int
1447g_part_destroy_geom(struct gctl_req *req, struct g_class *mp,
1448    struct g_geom *gp)
1449{
1450
1451	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, gp->name));
1452	g_topology_assert();
1453
1454	g_part_wither(gp, EINVAL);
1455	return (0);
1456}
1457
1458static struct g_geom *
1459g_part_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1460{
1461	struct g_consumer *cp;
1462	struct g_geom *gp;
1463	struct g_part_entry *entry;
1464	struct g_part_table *table;
1465	struct root_hold_token *rht;
1466	int attr, depth;
1467	int error;
1468
1469	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s,%s)", __func__, mp->name, pp->name));
1470	g_topology_assert();
1471
1472	/* Skip providers that are already open for writing. */
1473	if (pp->acw > 0)
1474		return (NULL);
1475
1476	/*
1477	 * Create a GEOM with consumer and hook it up to the provider.
1478	 * With that we become part of the topology. Optain read access
1479	 * to the provider.
1480	 */
1481	gp = g_new_geomf(mp, "%s", pp->name);
1482	cp = g_new_consumer(gp);
1483	error = g_attach(cp, pp);
1484	if (error == 0)
1485		error = g_access(cp, 1, 0, 0);
1486	if (error != 0) {
1487		g_part_wither(gp, error);
1488		return (NULL);
1489	}
1490
1491	rht = root_mount_hold(mp->name);
1492	g_topology_unlock();
1493
1494	/*
1495	 * Short-circuit the whole probing galore when there's no
1496	 * media present.
1497	 */
1498	if (pp->mediasize == 0 || pp->sectorsize == 0) {
1499		error = ENODEV;
1500		goto fail;
1501	}
1502
1503	/* Make sure we can nest and if so, determine our depth. */
1504	error = g_getattr("PART::isleaf", cp, &attr);
1505	if (!error && attr) {
1506		error = ENODEV;
1507		goto fail;
1508	}
1509	error = g_getattr("PART::depth", cp, &attr);
1510	depth = (!error) ? attr + 1 : 0;
1511
1512	error = g_part_probe(gp, cp, depth);
1513	if (error)
1514		goto fail;
1515
1516	table = gp->softc;
1517
1518	/*
1519	 * Synthesize a disk geometry. Some partitioning schemes
1520	 * depend on it and since some file systems need it even
1521	 * when the partitition scheme doesn't, we do it here in
1522	 * scheme-independent code.
1523	 */
1524	g_part_geometry(table, cp, pp->mediasize / pp->sectorsize);
1525
1526	error = G_PART_READ(table, cp);
1527	if (error)
1528		goto fail;
1529
1530	g_topology_lock();
1531	LIST_FOREACH(entry, &table->gpt_entry, gpe_entry) {
1532		if (!entry->gpe_internal)
1533			g_part_new_provider(gp, table, entry);
1534	}
1535
1536	root_mount_rel(rht);
1537	g_access(cp, -1, 0, 0);
1538	return (gp);
1539
1540 fail:
1541	g_topology_lock();
1542	root_mount_rel(rht);
1543	g_access(cp, -1, 0, 0);
1544	g_part_wither(gp, error);
1545	return (NULL);
1546}
1547
1548/*
1549 * Geom methods.
1550 */
1551
1552static int
1553g_part_access(struct g_provider *pp, int dr, int dw, int de)
1554{
1555	struct g_consumer *cp;
1556
1557	G_PART_TRACE((G_T_ACCESS, "%s(%s,%d,%d,%d)", __func__, pp->name, dr,
1558	    dw, de));
1559
1560	cp = LIST_FIRST(&pp->geom->consumer);
1561
1562	/* We always gain write-exclusive access. */
1563	return (g_access(cp, dr, dw, dw + de));
1564}
1565
1566static void
1567g_part_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1568    struct g_consumer *cp, struct g_provider *pp)
1569{
1570	char buf[64];
1571	struct g_part_entry *entry;
1572	struct g_part_table *table;
1573
1574	KASSERT(sb != NULL && gp != NULL, (__func__));
1575	table = gp->softc;
1576
1577	if (indent == NULL) {
1578		KASSERT(cp == NULL && pp != NULL, (__func__));
1579		entry = pp->private;
1580		if (entry == NULL)
1581			return;
1582		sbuf_printf(sb, " i %u o %ju ty %s", entry->gpe_index,
1583		    (uintmax_t)entry->gpe_offset,
1584		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1585		/*
1586		 * libdisk compatibility quirk - the scheme dumps the
1587		 * slicer name and partition type in a way that is
1588		 * compatible with libdisk. When libdisk is not used
1589		 * anymore, this should go away.
1590		 */
1591		G_PART_DUMPCONF(table, entry, sb, indent);
1592	} else if (cp != NULL) {	/* Consumer configuration. */
1593		KASSERT(pp == NULL, (__func__));
1594		/* none */
1595	} else if (pp != NULL) {	/* Provider configuration. */
1596		entry = pp->private;
1597		if (entry == NULL)
1598			return;
1599		sbuf_printf(sb, "%s<start>%ju</start>\n", indent,
1600		    (uintmax_t)entry->gpe_start);
1601		sbuf_printf(sb, "%s<end>%ju</end>\n", indent,
1602		    (uintmax_t)entry->gpe_end);
1603		sbuf_printf(sb, "%s<index>%u</index>\n", indent,
1604		    entry->gpe_index);
1605		sbuf_printf(sb, "%s<type>%s</type>\n", indent,
1606		    G_PART_TYPE(table, entry, buf, sizeof(buf)));
1607		sbuf_printf(sb, "%s<offset>%ju</offset>\n", indent,
1608		    (uintmax_t)entry->gpe_offset);
1609		sbuf_printf(sb, "%s<length>%ju</length>\n", indent,
1610		    (uintmax_t)pp->mediasize);
1611		G_PART_DUMPCONF(table, entry, sb, indent);
1612	} else {			/* Geom configuration. */
1613		sbuf_printf(sb, "%s<scheme>%s</scheme>\n", indent,
1614		    table->gpt_scheme->name);
1615		sbuf_printf(sb, "%s<entries>%u</entries>\n", indent,
1616		    table->gpt_entries);
1617		sbuf_printf(sb, "%s<first>%ju</first>\n", indent,
1618		    (uintmax_t)table->gpt_first);
1619		sbuf_printf(sb, "%s<last>%ju</last>\n", indent,
1620		    (uintmax_t)table->gpt_last);
1621		sbuf_printf(sb, "%s<fwsectors>%u</fwsectors>\n", indent,
1622		    table->gpt_sectors);
1623		sbuf_printf(sb, "%s<fwheads>%u</fwheads>\n", indent,
1624		    table->gpt_heads);
1625		G_PART_DUMPCONF(table, NULL, sb, indent);
1626	}
1627}
1628
1629static void
1630g_part_orphan(struct g_consumer *cp)
1631{
1632	struct g_provider *pp;
1633
1634	pp = cp->provider;
1635	KASSERT(pp != NULL, (__func__));
1636	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, pp->name));
1637	g_topology_assert();
1638
1639	KASSERT(pp->error != 0, (__func__));
1640	g_part_wither(cp->geom, pp->error);
1641}
1642
1643static void
1644g_part_spoiled(struct g_consumer *cp)
1645{
1646
1647	G_PART_TRACE((G_T_TOPOLOGY, "%s(%s)", __func__, cp->provider->name));
1648	g_topology_assert();
1649
1650	g_part_wither(cp->geom, ENXIO);
1651}
1652
1653static void
1654g_part_start(struct bio *bp)
1655{
1656	struct bio *bp2;
1657	struct g_consumer *cp;
1658	struct g_geom *gp;
1659	struct g_part_entry *entry;
1660	struct g_part_table *table;
1661	struct g_kerneldump *gkd;
1662	struct g_provider *pp;
1663
1664	pp = bp->bio_to;
1665	gp = pp->geom;
1666	table = gp->softc;
1667	cp = LIST_FIRST(&gp->consumer);
1668
1669	G_PART_TRACE((G_T_BIO, "%s: cmd=%d, provider=%s", __func__, bp->bio_cmd,
1670	    pp->name));
1671
1672	entry = pp->private;
1673	if (entry == NULL) {
1674		g_io_deliver(bp, ENXIO);
1675		return;
1676	}
1677
1678	switch(bp->bio_cmd) {
1679	case BIO_DELETE:
1680	case BIO_READ:
1681	case BIO_WRITE:
1682		if (bp->bio_offset >= pp->mediasize) {
1683			g_io_deliver(bp, EIO);
1684			return;
1685		}
1686		bp2 = g_clone_bio(bp);
1687		if (bp2 == NULL) {
1688			g_io_deliver(bp, ENOMEM);
1689			return;
1690		}
1691		if (bp2->bio_offset + bp2->bio_length > pp->mediasize)
1692			bp2->bio_length = pp->mediasize - bp2->bio_offset;
1693		bp2->bio_done = g_std_done;
1694		bp2->bio_offset += entry->gpe_offset;
1695		g_io_request(bp2, cp);
1696		return;
1697	case BIO_FLUSH:
1698		break;
1699	case BIO_GETATTR:
1700		if (g_handleattr_int(bp, "GEOM::fwheads", table->gpt_heads))
1701			return;
1702		if (g_handleattr_int(bp, "GEOM::fwsectors", table->gpt_sectors))
1703			return;
1704		if (g_handleattr_int(bp, "PART::isleaf", table->gpt_isleaf))
1705			return;
1706		if (g_handleattr_int(bp, "PART::depth", table->gpt_depth))
1707			return;
1708		if (g_handleattr_str(bp, "PART::scheme",
1709		    table->gpt_scheme->name))
1710			return;
1711		if (!strcmp("GEOM::kerneldump", bp->bio_attribute)) {
1712			/*
1713			 * Check that the partition is suitable for kernel
1714			 * dumps. Typically only swap partitions should be
1715			 * used.
1716			 */
1717			if (!G_PART_DUMPTO(table, entry)) {
1718				g_io_deliver(bp, ENODEV);
1719				printf("GEOM_PART: Partition '%s' not suitable"
1720				    " for kernel dumps (wrong type?)\n",
1721				    pp->name);
1722				return;
1723			}
1724			gkd = (struct g_kerneldump *)bp->bio_data;
1725			if (gkd->offset >= pp->mediasize) {
1726				g_io_deliver(bp, EIO);
1727				return;
1728			}
1729			if (gkd->offset + gkd->length > pp->mediasize)
1730				gkd->length = pp->mediasize - gkd->offset;
1731			gkd->offset += entry->gpe_offset;
1732		}
1733		break;
1734	default:
1735		g_io_deliver(bp, EOPNOTSUPP);
1736		return;
1737	}
1738
1739	bp2 = g_clone_bio(bp);
1740	if (bp2 == NULL) {
1741		g_io_deliver(bp, ENOMEM);
1742		return;
1743	}
1744	bp2->bio_done = g_std_done;
1745	g_io_request(bp2, cp);
1746}
1747
1748static void
1749g_part_init(struct g_class *mp)
1750{
1751
1752	TAILQ_INSERT_HEAD(&g_part_schemes, &g_part_null_scheme, scheme_list);
1753}
1754
1755static void
1756g_part_fini(struct g_class *mp)
1757{
1758
1759	TAILQ_REMOVE(&g_part_schemes, &g_part_null_scheme, scheme_list);
1760}
1761
1762static void
1763g_part_unload_event(void *arg, int flag)
1764{
1765	struct g_consumer *cp;
1766	struct g_geom *gp;
1767	struct g_provider *pp;
1768	struct g_part_scheme *scheme;
1769	struct g_part_table *table;
1770	uintptr_t *xchg;
1771	int acc, error;
1772
1773	if (flag == EV_CANCEL)
1774		return;
1775
1776	xchg = arg;
1777	error = 0;
1778	scheme = (void *)(*xchg);
1779
1780	g_topology_assert();
1781
1782	LIST_FOREACH(gp, &g_part_class.geom, geom) {
1783		table = gp->softc;
1784		if (table->gpt_scheme != scheme)
1785			continue;
1786
1787		acc = 0;
1788		LIST_FOREACH(pp, &gp->provider, provider)
1789			acc += pp->acr + pp->acw + pp->ace;
1790		LIST_FOREACH(cp, &gp->consumer, consumer)
1791			acc += cp->acr + cp->acw + cp->ace;
1792
1793		if (!acc)
1794			g_part_wither(gp, ENOSYS);
1795		else
1796			error = EBUSY;
1797	}
1798
1799	if (!error)
1800		TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
1801
1802	*xchg = error;
1803}
1804
1805int
1806g_part_modevent(module_t mod, int type, struct g_part_scheme *scheme)
1807{
1808	uintptr_t arg;
1809	int error;
1810
1811	switch (type) {
1812	case MOD_LOAD:
1813		TAILQ_INSERT_TAIL(&g_part_schemes, scheme, scheme_list);
1814
1815		error = g_retaste(&g_part_class);
1816		if (error)
1817			TAILQ_REMOVE(&g_part_schemes, scheme, scheme_list);
1818		break;
1819	case MOD_UNLOAD:
1820		arg = (uintptr_t)scheme;
1821		error = g_waitfor_event(g_part_unload_event, &arg, M_WAITOK,
1822		    NULL);
1823		if (!error)
1824			error = (arg == (uintptr_t)scheme) ? EDOOFUS : arg;
1825		break;
1826	default:
1827		error = EOPNOTSUPP;
1828		break;
1829	}
1830
1831	return (error);
1832}
1833