xref: /freebsd/sys/geom/geom_subr.c (revision 1e413cf93298b5b97441a21d9a50fdcd0ee9945e)
1 /*-
2  * Copyright (c) 2002 Poul-Henning Kamp
3  * Copyright (c) 2002 Networks Associates Technology, Inc.
4  * All rights reserved.
5  *
6  * This software was developed for the FreeBSD Project by Poul-Henning Kamp
7  * and NAI Labs, the Security Research Division of Network Associates, Inc.
8  * under DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
9  * DARPA CHATS research program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. The names of the authors may not be used to endorse or promote
20  *    products derived from this software without specific prior written
21  *    permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
38 
39 #include "opt_ddb.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/devicestat.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
46 #include <sys/bio.h>
47 #include <sys/sysctl.h>
48 #include <sys/proc.h>
49 #include <sys/kthread.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/errno.h>
53 #include <sys/sbuf.h>
54 #include <geom/geom.h>
55 #include <geom/geom_int.h>
56 #include <machine/stdarg.h>
57 
58 #ifdef DDB
59 #include <ddb/ddb.h>
60 #endif
61 
62 struct class_list_head g_classes = LIST_HEAD_INITIALIZER(g_classes);
63 static struct g_tailq_head geoms = TAILQ_HEAD_INITIALIZER(geoms);
64 char *g_wait_event, *g_wait_up, *g_wait_down, *g_wait_sim;
65 
66 struct g_hh00 {
67 	struct g_class	*mp;
68 	int		error;
69 	int		post;
70 };
71 
72 /*
73  * This event offers a new class a chance to taste all preexisting providers.
74  */
75 static void
76 g_load_class(void *arg, int flag)
77 {
78 	struct g_hh00 *hh;
79 	struct g_class *mp2, *mp;
80 	struct g_geom *gp;
81 	struct g_provider *pp;
82 
83 	g_topology_assert();
84 	if (flag == EV_CANCEL)	/* XXX: can't happen ? */
85 		return;
86 	if (g_shutdown)
87 		return;
88 
89 	hh = arg;
90 	mp = hh->mp;
91 	hh->error = 0;
92 	if (hh->post) {
93 		g_free(hh);
94 		hh = NULL;
95 	}
96 	g_trace(G_T_TOPOLOGY, "g_load_class(%s)", mp->name);
97 	KASSERT(mp->name != NULL && *mp->name != '\0',
98 	    ("GEOM class has no name"));
99 	LIST_FOREACH(mp2, &g_classes, class) {
100 		if (mp2 == mp) {
101 			printf("The GEOM class %s is already loaded.\n",
102 			    mp2->name);
103 			if (hh != NULL)
104 				hh->error = EEXIST;
105 			return;
106 		} else if (strcmp(mp2->name, mp->name) == 0) {
107 			printf("A GEOM class %s is already loaded.\n",
108 			    mp2->name);
109 			if (hh != NULL)
110 				hh->error = EEXIST;
111 			return;
112 		}
113 	}
114 
115 	LIST_INIT(&mp->geom);
116 	LIST_INSERT_HEAD(&g_classes, mp, class);
117 	if (mp->init != NULL)
118 		mp->init(mp);
119 	if (mp->taste == NULL)
120 		return;
121 	LIST_FOREACH(mp2, &g_classes, class) {
122 		if (mp == mp2)
123 			continue;
124 		LIST_FOREACH(gp, &mp2->geom, geom) {
125 			LIST_FOREACH(pp, &gp->provider, provider) {
126 				mp->taste(mp, pp, 0);
127 				g_topology_assert();
128 			}
129 		}
130 	}
131 }
132 
133 static void
134 g_unload_class(void *arg, int flag)
135 {
136 	struct g_hh00 *hh;
137 	struct g_class *mp;
138 	struct g_geom *gp;
139 	struct g_provider *pp;
140 	struct g_consumer *cp;
141 	int error;
142 
143 	g_topology_assert();
144 	hh = arg;
145 	mp = hh->mp;
146 	G_VALID_CLASS(mp);
147 	g_trace(G_T_TOPOLOGY, "g_unload_class(%s)", mp->name);
148 
149 	/*
150 	 * We allow unloading if we have no geoms, or a class
151 	 * method we can use to get rid of them.
152 	 */
153 	if (!LIST_EMPTY(&mp->geom) && mp->destroy_geom == NULL) {
154 		hh->error = EOPNOTSUPP;
155 		return;
156 	}
157 
158 	/* We refuse to unload if anything is open */
159 	LIST_FOREACH(gp, &mp->geom, geom) {
160 		LIST_FOREACH(pp, &gp->provider, provider)
161 			if (pp->acr || pp->acw || pp->ace) {
162 				hh->error = EBUSY;
163 				return;
164 			}
165 		LIST_FOREACH(cp, &gp->consumer, consumer)
166 			if (cp->acr || cp->acw || cp->ace) {
167 				hh->error = EBUSY;
168 				return;
169 			}
170 	}
171 
172 	/* Bar new entries */
173 	mp->taste = NULL;
174 	mp->config = NULL;
175 
176 	error = 0;
177 	for (;;) {
178 		gp = LIST_FIRST(&mp->geom);
179 		if (gp == NULL)
180 			break;
181 		error = mp->destroy_geom(NULL, mp, gp);
182 		if (error != 0)
183 			break;
184 	}
185 	if (error == 0) {
186 		if (mp->fini != NULL)
187 			mp->fini(mp);
188 		LIST_REMOVE(mp, class);
189 	}
190 	hh->error = error;
191 	return;
192 }
193 
194 int
195 g_modevent(module_t mod, int type, void *data)
196 {
197 	struct g_hh00 *hh;
198 	int error;
199 	static int g_ignition;
200 	struct g_class *mp;
201 
202 	mp = data;
203 	if (mp->version != G_VERSION) {
204 		printf("GEOM class %s has Wrong version %x\n",
205 		    mp->name, mp->version);
206 		return (EINVAL);
207 	}
208 	if (!g_ignition) {
209 		g_ignition++;
210 		g_init();
211 	}
212 	hh = g_malloc(sizeof *hh, M_WAITOK | M_ZERO);
213 	hh->mp = data;
214 	error = EOPNOTSUPP;
215 	switch (type) {
216 	case MOD_LOAD:
217 		g_trace(G_T_TOPOLOGY, "g_modevent(%s, LOAD)", hh->mp->name);
218 		/*
219 		 * Once the system is not cold, MOD_LOAD calls will be
220 		 * from the userland and the g_event thread will be able
221 		 * to acknowledge their completion.
222 		 */
223 		if (cold) {
224 			hh->post = 1;
225 			error = g_post_event(g_load_class, hh, M_WAITOK, NULL);
226 		} else {
227 			error = g_waitfor_event(g_load_class, hh, M_WAITOK,
228 			    NULL);
229 			if (error == 0)
230 				error = hh->error;
231 			g_free(hh);
232 		}
233 		break;
234 	case MOD_UNLOAD:
235 		g_trace(G_T_TOPOLOGY, "g_modevent(%s, UNLOAD)", hh->mp->name);
236 		error = g_waitfor_event(g_unload_class, hh, M_WAITOK, NULL);
237 		if (error == 0)
238 			error = hh->error;
239 		if (error == 0) {
240 			KASSERT(LIST_EMPTY(&hh->mp->geom),
241 			    ("Unloaded class (%s) still has geom", hh->mp->name));
242 		}
243 		g_free(hh);
244 		break;
245 	default:
246 		g_free(hh);
247 		break;
248 	}
249 	return (error);
250 }
251 
252 struct g_geom *
253 g_new_geomf(struct g_class *mp, const char *fmt, ...)
254 {
255 	struct g_geom *gp;
256 	va_list ap;
257 	struct sbuf *sb;
258 
259 	g_topology_assert();
260 	G_VALID_CLASS(mp);
261 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
262 	va_start(ap, fmt);
263 	sbuf_vprintf(sb, fmt, ap);
264 	va_end(ap);
265 	sbuf_finish(sb);
266 	gp = g_malloc(sizeof *gp, M_WAITOK | M_ZERO);
267 	gp->name = g_malloc(sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
268 	gp->class = mp;
269 	gp->rank = 1;
270 	LIST_INIT(&gp->consumer);
271 	LIST_INIT(&gp->provider);
272 	LIST_INSERT_HEAD(&mp->geom, gp, geom);
273 	TAILQ_INSERT_HEAD(&geoms, gp, geoms);
274 	strcpy(gp->name, sbuf_data(sb));
275 	sbuf_delete(sb);
276 	/* Fill in defaults from class */
277 	gp->start = mp->start;
278 	gp->spoiled = mp->spoiled;
279 	gp->dumpconf = mp->dumpconf;
280 	gp->access = mp->access;
281 	gp->orphan = mp->orphan;
282 	gp->ioctl = mp->ioctl;
283 	return (gp);
284 }
285 
286 void
287 g_destroy_geom(struct g_geom *gp)
288 {
289 
290 	g_topology_assert();
291 	G_VALID_GEOM(gp);
292 	g_trace(G_T_TOPOLOGY, "g_destroy_geom(%p(%s))", gp, gp->name);
293 	KASSERT(LIST_EMPTY(&gp->consumer),
294 	    ("g_destroy_geom(%s) with consumer(s) [%p]",
295 	    gp->name, LIST_FIRST(&gp->consumer)));
296 	KASSERT(LIST_EMPTY(&gp->provider),
297 	    ("g_destroy_geom(%s) with provider(s) [%p]",
298 	    gp->name, LIST_FIRST(&gp->provider)));
299 	g_cancel_event(gp);
300 	LIST_REMOVE(gp, geom);
301 	TAILQ_REMOVE(&geoms, gp, geoms);
302 	g_free(gp->name);
303 	g_free(gp);
304 }
305 
306 /*
307  * This function is called (repeatedly) until the has withered away.
308  */
309 void
310 g_wither_geom(struct g_geom *gp, int error)
311 {
312 	struct g_provider *pp;
313 
314 	g_topology_assert();
315 	G_VALID_GEOM(gp);
316 	g_trace(G_T_TOPOLOGY, "g_wither_geom(%p(%s))", gp, gp->name);
317 	if (!(gp->flags & G_GEOM_WITHER)) {
318 		gp->flags |= G_GEOM_WITHER;
319 		LIST_FOREACH(pp, &gp->provider, provider)
320 			if (!(pp->flags & G_PF_ORPHAN))
321 				g_orphan_provider(pp, error);
322 	}
323 	g_do_wither();
324 }
325 
326 /*
327  * Convenience function to destroy a particular provider.
328  */
329 void
330 g_wither_provider(struct g_provider *pp, int error)
331 {
332 
333 	pp->flags |= G_PF_WITHER;
334 	if (!(pp->flags & G_PF_ORPHAN))
335 		g_orphan_provider(pp, error);
336 }
337 
338 /*
339  * This function is called (repeatedly) until the has withered away.
340  */
341 void
342 g_wither_geom_close(struct g_geom *gp, int error)
343 {
344 	struct g_consumer *cp;
345 
346 	g_topology_assert();
347 	G_VALID_GEOM(gp);
348 	g_trace(G_T_TOPOLOGY, "g_wither_geom_close(%p(%s))", gp, gp->name);
349 	LIST_FOREACH(cp, &gp->consumer, consumer)
350 		if (cp->acr || cp->acw || cp->ace)
351 			g_access(cp, -cp->acr, -cp->acw, -cp->ace);
352 	g_wither_geom(gp, error);
353 }
354 
355 /*
356  * This function is called (repeatedly) until we cant wash away more
357  * withered bits at present.  Return value contains two bits.  Bit 0
358  * set means "withering stuff we can't wash now", bit 1 means "call
359  * me again, there may be stuff I didn't get the first time around.
360  */
361 int
362 g_wither_washer()
363 {
364 	struct g_class *mp;
365 	struct g_geom *gp, *gp2;
366 	struct g_provider *pp, *pp2;
367 	struct g_consumer *cp, *cp2;
368 	int result;
369 
370 	result = 0;
371 	g_topology_assert();
372 	LIST_FOREACH(mp, &g_classes, class) {
373 		LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
374 			LIST_FOREACH_SAFE(pp, &gp->provider, provider, pp2) {
375 				if (!(pp->flags & G_PF_WITHER))
376 					continue;
377 				if (LIST_EMPTY(&pp->consumers))
378 					g_destroy_provider(pp);
379 				else
380 					result |= 1;
381 			}
382 			if (!(gp->flags & G_GEOM_WITHER))
383 				continue;
384 			LIST_FOREACH_SAFE(pp, &gp->provider, provider, pp2) {
385 				if (LIST_EMPTY(&pp->consumers))
386 					g_destroy_provider(pp);
387 				else
388 					result |= 1;
389 			}
390 			LIST_FOREACH_SAFE(cp, &gp->consumer, consumer, cp2) {
391 				if (cp->acr || cp->acw || cp->ace) {
392 					result |= 1;
393 					continue;
394 				}
395 				if (cp->provider != NULL)
396 					g_detach(cp);
397 				g_destroy_consumer(cp);
398 				result |= 2;
399 			}
400 			if (LIST_EMPTY(&gp->provider) &&
401 			    LIST_EMPTY(&gp->consumer))
402 				g_destroy_geom(gp);
403 			else
404 				result |= 1;
405 		}
406 	}
407 	return (result);
408 }
409 
410 struct g_consumer *
411 g_new_consumer(struct g_geom *gp)
412 {
413 	struct g_consumer *cp;
414 
415 	g_topology_assert();
416 	G_VALID_GEOM(gp);
417 	KASSERT(!(gp->flags & G_GEOM_WITHER),
418 	    ("g_new_consumer on WITHERing geom(%s) (class %s)",
419 	    gp->name, gp->class->name));
420 	KASSERT(gp->orphan != NULL,
421 	    ("g_new_consumer on geom(%s) (class %s) without orphan",
422 	    gp->name, gp->class->name));
423 
424 	cp = g_malloc(sizeof *cp, M_WAITOK | M_ZERO);
425 	cp->geom = gp;
426 	cp->stat = devstat_new_entry(cp, -1, 0, DEVSTAT_ALL_SUPPORTED,
427 	    DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
428 	LIST_INSERT_HEAD(&gp->consumer, cp, consumer);
429 	return(cp);
430 }
431 
432 void
433 g_destroy_consumer(struct g_consumer *cp)
434 {
435 	struct g_geom *gp;
436 
437 	g_topology_assert();
438 	G_VALID_CONSUMER(cp);
439 	g_trace(G_T_TOPOLOGY, "g_destroy_consumer(%p)", cp);
440 	KASSERT (cp->provider == NULL, ("g_destroy_consumer but attached"));
441 	KASSERT (cp->acr == 0, ("g_destroy_consumer with acr"));
442 	KASSERT (cp->acw == 0, ("g_destroy_consumer with acw"));
443 	KASSERT (cp->ace == 0, ("g_destroy_consumer with ace"));
444 	g_cancel_event(cp);
445 	gp = cp->geom;
446 	LIST_REMOVE(cp, consumer);
447 	devstat_remove_entry(cp->stat);
448 	g_free(cp);
449 	if (gp->flags & G_GEOM_WITHER)
450 		g_do_wither();
451 }
452 
453 static void
454 g_new_provider_event(void *arg, int flag)
455 {
456 	struct g_class *mp;
457 	struct g_provider *pp;
458 	struct g_consumer *cp;
459 	int i;
460 
461 	g_topology_assert();
462 	if (flag == EV_CANCEL)
463 		return;
464 	if (g_shutdown)
465 		return;
466 	pp = arg;
467 	G_VALID_PROVIDER(pp);
468 	LIST_FOREACH(mp, &g_classes, class) {
469 		if (mp->taste == NULL)
470 			continue;
471 		i = 1;
472 		LIST_FOREACH(cp, &pp->consumers, consumers)
473 			if (cp->geom->class == mp)
474 				i = 0;
475 		if (!i)
476 			continue;
477 		mp->taste(mp, pp, 0);
478 		g_topology_assert();
479 	}
480 }
481 
482 
483 struct g_provider *
484 g_new_providerf(struct g_geom *gp, const char *fmt, ...)
485 {
486 	struct g_provider *pp;
487 	struct sbuf *sb;
488 	va_list ap;
489 
490 	g_topology_assert();
491 	G_VALID_GEOM(gp);
492 	KASSERT(gp->access != NULL,
493 	    ("new provider on geom(%s) without ->access (class %s)",
494 	    gp->name, gp->class->name));
495 	KASSERT(gp->start != NULL,
496 	    ("new provider on geom(%s) without ->start (class %s)",
497 	    gp->name, gp->class->name));
498 	KASSERT(!(gp->flags & G_GEOM_WITHER),
499 	    ("new provider on WITHERing geom(%s) (class %s)",
500 	    gp->name, gp->class->name));
501 	sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
502 	va_start(ap, fmt);
503 	sbuf_vprintf(sb, fmt, ap);
504 	va_end(ap);
505 	sbuf_finish(sb);
506 	pp = g_malloc(sizeof *pp + sbuf_len(sb) + 1, M_WAITOK | M_ZERO);
507 	pp->name = (char *)(pp + 1);
508 	strcpy(pp->name, sbuf_data(sb));
509 	sbuf_delete(sb);
510 	LIST_INIT(&pp->consumers);
511 	pp->error = ENXIO;
512 	pp->geom = gp;
513 	pp->stat = devstat_new_entry(pp, -1, 0, DEVSTAT_ALL_SUPPORTED,
514 	    DEVSTAT_TYPE_DIRECT, DEVSTAT_PRIORITY_MAX);
515 	LIST_INSERT_HEAD(&gp->provider, pp, provider);
516 	g_post_event(g_new_provider_event, pp, M_WAITOK, pp, gp, NULL);
517 	return (pp);
518 }
519 
520 void
521 g_error_provider(struct g_provider *pp, int error)
522 {
523 
524 	/* G_VALID_PROVIDER(pp);  We may not have g_topology */
525 	pp->error = error;
526 }
527 
528 struct g_provider *
529 g_provider_by_name(char const *arg)
530 {
531 	struct g_class *cp;
532 	struct g_geom *gp;
533 	struct g_provider *pp;
534 
535 	LIST_FOREACH(cp, &g_classes, class) {
536 		LIST_FOREACH(gp, &cp->geom, geom) {
537 			LIST_FOREACH(pp, &gp->provider, provider) {
538 				if (!strcmp(arg, pp->name))
539 					return (pp);
540 			}
541 		}
542 	}
543 	return (NULL);
544 }
545 
546 void
547 g_destroy_provider(struct g_provider *pp)
548 {
549 	struct g_geom *gp;
550 
551 	g_topology_assert();
552 	G_VALID_PROVIDER(pp);
553 	KASSERT(LIST_EMPTY(&pp->consumers),
554 	    ("g_destroy_provider but attached"));
555 	KASSERT (pp->acr == 0, ("g_destroy_provider with acr"));
556 	KASSERT (pp->acw == 0, ("g_destroy_provider with acw"));
557 	KASSERT (pp->acw == 0, ("g_destroy_provider with ace"));
558 	g_cancel_event(pp);
559 	LIST_REMOVE(pp, provider);
560 	gp = pp->geom;
561 	devstat_remove_entry(pp->stat);
562 	g_free(pp);
563 	if ((gp->flags & G_GEOM_WITHER))
564 		g_do_wither();
565 }
566 
567 /*
568  * We keep the "geoms" list sorted by topological order (== increasing
569  * numerical rank) at all times.
570  * When an attach is done, the attaching geoms rank is invalidated
571  * and it is moved to the tail of the list.
572  * All geoms later in the sequence has their ranks reevaluated in
573  * sequence.  If we cannot assign rank to a geom because it's
574  * prerequisites do not have rank, we move that element to the tail
575  * of the sequence with invalid rank as well.
576  * At some point we encounter our original geom and if we stil fail
577  * to assign it a rank, there must be a loop and we fail back to
578  * g_attach() which detach again and calls redo_rank again
579  * to fix up the damage.
580  * It would be much simpler code wise to do it recursively, but we
581  * can't risk that on the kernel stack.
582  */
583 
584 static int
585 redo_rank(struct g_geom *gp)
586 {
587 	struct g_consumer *cp;
588 	struct g_geom *gp1, *gp2;
589 	int n, m;
590 
591 	g_topology_assert();
592 	G_VALID_GEOM(gp);
593 
594 	/* Invalidate this geoms rank and move it to the tail */
595 	gp1 = TAILQ_NEXT(gp, geoms);
596 	if (gp1 != NULL) {
597 		gp->rank = 0;
598 		TAILQ_REMOVE(&geoms, gp, geoms);
599 		TAILQ_INSERT_TAIL(&geoms, gp, geoms);
600 	} else {
601 		gp1 = gp;
602 	}
603 
604 	/* re-rank the rest of the sequence */
605 	for (; gp1 != NULL; gp1 = gp2) {
606 		gp1->rank = 0;
607 		m = 1;
608 		LIST_FOREACH(cp, &gp1->consumer, consumer) {
609 			if (cp->provider == NULL)
610 				continue;
611 			n = cp->provider->geom->rank;
612 			if (n == 0) {
613 				m = 0;
614 				break;
615 			} else if (n >= m)
616 				m = n + 1;
617 		}
618 		gp1->rank = m;
619 		gp2 = TAILQ_NEXT(gp1, geoms);
620 
621 		/* got a rank, moving on */
622 		if (m != 0)
623 			continue;
624 
625 		/* no rank to original geom means loop */
626 		if (gp == gp1)
627 			return (ELOOP);
628 
629 		/* no rank, put it at the end move on */
630 		TAILQ_REMOVE(&geoms, gp1, geoms);
631 		TAILQ_INSERT_TAIL(&geoms, gp1, geoms);
632 	}
633 	return (0);
634 }
635 
636 int
637 g_attach(struct g_consumer *cp, struct g_provider *pp)
638 {
639 	int error;
640 
641 	g_topology_assert();
642 	G_VALID_CONSUMER(cp);
643 	G_VALID_PROVIDER(pp);
644 	KASSERT(cp->provider == NULL, ("attach but attached"));
645 	cp->provider = pp;
646 	LIST_INSERT_HEAD(&pp->consumers, cp, consumers);
647 	error = redo_rank(cp->geom);
648 	if (error) {
649 		LIST_REMOVE(cp, consumers);
650 		cp->provider = NULL;
651 		redo_rank(cp->geom);
652 	}
653 	return (error);
654 }
655 
656 void
657 g_detach(struct g_consumer *cp)
658 {
659 	struct g_provider *pp;
660 
661 	g_topology_assert();
662 	G_VALID_CONSUMER(cp);
663 	g_trace(G_T_TOPOLOGY, "g_detach(%p)", cp);
664 	KASSERT(cp->provider != NULL, ("detach but not attached"));
665 	KASSERT(cp->acr == 0, ("detach but nonzero acr"));
666 	KASSERT(cp->acw == 0, ("detach but nonzero acw"));
667 	KASSERT(cp->ace == 0, ("detach but nonzero ace"));
668 	KASSERT(cp->nstart == cp->nend,
669 	    ("detach with active requests"));
670 	pp = cp->provider;
671 	LIST_REMOVE(cp, consumers);
672 	cp->provider = NULL;
673 	if (pp->geom->flags & G_GEOM_WITHER)
674 		g_do_wither();
675 	else if (pp->flags & G_PF_WITHER)
676 		g_do_wither();
677 	redo_rank(cp->geom);
678 }
679 
680 /*
681  * g_access()
682  *
683  * Access-check with delta values.  The question asked is "can provider
684  * "cp" change the access counters by the relative amounts dc[rwe] ?"
685  */
686 
687 int
688 g_access(struct g_consumer *cp, int dcr, int dcw, int dce)
689 {
690 	struct g_provider *pp;
691 	int pr,pw,pe;
692 	int error;
693 
694 	g_topology_assert();
695 	G_VALID_CONSUMER(cp);
696 	pp = cp->provider;
697 	KASSERT(pp != NULL, ("access but not attached"));
698 	G_VALID_PROVIDER(pp);
699 
700 	g_trace(G_T_ACCESS, "g_access(%p(%s), %d, %d, %d)",
701 	    cp, pp->name, dcr, dcw, dce);
702 
703 	KASSERT(cp->acr + dcr >= 0, ("access resulting in negative acr"));
704 	KASSERT(cp->acw + dcw >= 0, ("access resulting in negative acw"));
705 	KASSERT(cp->ace + dce >= 0, ("access resulting in negative ace"));
706 	KASSERT(dcr != 0 || dcw != 0 || dce != 0, ("NOP access request"));
707 	KASSERT(pp->geom->access != NULL, ("NULL geom->access"));
708 
709 	/*
710 	 * If our class cares about being spoiled, and we have been, we
711 	 * are probably just ahead of the event telling us that.  Fail
712 	 * now rather than having to unravel this later.
713 	 */
714 	if (cp->geom->spoiled != NULL && cp->spoiled &&
715 	    (dcr > 0 || dcw > 0 || dce > 0))
716 		return (ENXIO);
717 
718 	/*
719 	 * Figure out what counts the provider would have had, if this
720 	 * consumer had (r0w0e0) at this time.
721 	 */
722 	pr = pp->acr - cp->acr;
723 	pw = pp->acw - cp->acw;
724 	pe = pp->ace - cp->ace;
725 
726 	g_trace(G_T_ACCESS,
727     "open delta:[r%dw%de%d] old:[r%dw%de%d] provider:[r%dw%de%d] %p(%s)",
728 	    dcr, dcw, dce,
729 	    cp->acr, cp->acw, cp->ace,
730 	    pp->acr, pp->acw, pp->ace,
731 	    pp, pp->name);
732 
733 	/* If foot-shooting is enabled, any open on rank#1 is OK */
734 	if ((g_debugflags & 16) && pp->geom->rank == 1)
735 		;
736 	/* If we try exclusive but already write: fail */
737 	else if (dce > 0 && pw > 0)
738 		return (EPERM);
739 	/* If we try write but already exclusive: fail */
740 	else if (dcw > 0 && pe > 0)
741 		return (EPERM);
742 	/* If we try to open more but provider is error'ed: fail */
743 	else if ((dcr > 0 || dcw > 0 || dce > 0) && pp->error != 0)
744 		return (pp->error);
745 
746 	/* Ok then... */
747 
748 	error = pp->geom->access(pp, dcr, dcw, dce);
749 	KASSERT(dcr > 0 || dcw > 0 || dce > 0 || error == 0,
750 	    ("Geom provider %s::%s failed closing ->access()",
751 	    pp->geom->class->name, pp->name));
752 	if (!error) {
753 		/*
754 		 * If we open first write, spoil any partner consumers.
755 		 * If we close last write and provider is not errored,
756 		 * trigger re-taste.
757 		 */
758 		if (pp->acw == 0 && dcw != 0)
759 			g_spoil(pp, cp);
760 		else if (pp->acw != 0 && pp->acw == -dcw && pp->error == 0 &&
761 		    !(pp->geom->flags & G_GEOM_WITHER))
762 			g_post_event(g_new_provider_event, pp, M_WAITOK,
763 			    pp, NULL);
764 
765 		pp->acr += dcr;
766 		pp->acw += dcw;
767 		pp->ace += dce;
768 		cp->acr += dcr;
769 		cp->acw += dcw;
770 		cp->ace += dce;
771 		if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)
772 			KASSERT(pp->sectorsize > 0,
773 			    ("Provider %s lacks sectorsize", pp->name));
774 	}
775 	return (error);
776 }
777 
778 int
779 g_handleattr_int(struct bio *bp, const char *attribute, int val)
780 {
781 
782 	return (g_handleattr(bp, attribute, &val, sizeof val));
783 }
784 
785 int
786 g_handleattr_off_t(struct bio *bp, const char *attribute, off_t val)
787 {
788 
789 	return (g_handleattr(bp, attribute, &val, sizeof val));
790 }
791 
792 int
793 g_handleattr_str(struct bio *bp, const char *attribute, char *str)
794 {
795 
796 	return (g_handleattr(bp, attribute, str, 0));
797 }
798 
799 int
800 g_handleattr(struct bio *bp, const char *attribute, void *val, int len)
801 {
802 	int error = 0;
803 
804 	if (strcmp(bp->bio_attribute, attribute))
805 		return (0);
806 	if (len == 0) {
807 		bzero(bp->bio_data, bp->bio_length);
808 		if (strlcpy(bp->bio_data, val, bp->bio_length) >=
809 		    bp->bio_length) {
810 			printf("%s: %s bio_length %jd len %zu -> EFAULT\n",
811 			    __func__, bp->bio_to->name,
812 			    (intmax_t)bp->bio_length, strlen(val));
813 			error = EFAULT;
814 		}
815 	} else if (bp->bio_length == len) {
816 		bcopy(val, bp->bio_data, len);
817 		bp->bio_completed = len;
818 	} else {
819 		printf("%s: %s bio_length %jd len %d -> EFAULT\n", __func__,
820 		    bp->bio_to->name, (intmax_t)bp->bio_length, len);
821 		error = EFAULT;
822 	}
823 	g_io_deliver(bp, error);
824 	return (1);
825 }
826 
827 int
828 g_std_access(struct g_provider *pp,
829 	int dr __unused, int dw __unused, int de __unused)
830 {
831 
832 	g_topology_assert();
833 	G_VALID_PROVIDER(pp);
834         return (0);
835 }
836 
837 void
838 g_std_done(struct bio *bp)
839 {
840 	struct bio *bp2;
841 
842 	bp2 = bp->bio_parent;
843 	if (bp2->bio_error == 0)
844 		bp2->bio_error = bp->bio_error;
845 	bp2->bio_completed += bp->bio_completed;
846 	g_destroy_bio(bp);
847 	bp2->bio_inbed++;
848 	if (bp2->bio_children == bp2->bio_inbed)
849 		g_io_deliver(bp2, bp2->bio_error);
850 }
851 
852 /* XXX: maybe this is only g_slice_spoiled */
853 
854 void
855 g_std_spoiled(struct g_consumer *cp)
856 {
857 	struct g_geom *gp;
858 	struct g_provider *pp;
859 
860 	g_topology_assert();
861 	G_VALID_CONSUMER(cp);
862 	g_trace(G_T_TOPOLOGY, "g_std_spoiled(%p)", cp);
863 	g_detach(cp);
864 	gp = cp->geom;
865 	LIST_FOREACH(pp, &gp->provider, provider)
866 		g_orphan_provider(pp, ENXIO);
867 	g_destroy_consumer(cp);
868 	if (LIST_EMPTY(&gp->provider) && LIST_EMPTY(&gp->consumer))
869 		g_destroy_geom(gp);
870 	else
871 		gp->flags |= G_GEOM_WITHER;
872 }
873 
874 /*
875  * Spoiling happens when a provider is opened for writing, but consumers
876  * which are configured by in-band data are attached (slicers for instance).
877  * Since the write might potentially change the in-band data, such consumers
878  * need to re-evaluate their existence after the writing session closes.
879  * We do this by (offering to) tear them down when the open for write happens
880  * in return for a re-taste when it closes again.
881  * Together with the fact that such consumers grab an 'e' bit whenever they
882  * are open, regardless of mode, this ends up DTRT.
883  */
884 
885 static void
886 g_spoil_event(void *arg, int flag)
887 {
888 	struct g_provider *pp;
889 	struct g_consumer *cp, *cp2;
890 
891 	g_topology_assert();
892 	if (flag == EV_CANCEL)
893 		return;
894 	pp = arg;
895 	G_VALID_PROVIDER(pp);
896 	for (cp = LIST_FIRST(&pp->consumers); cp != NULL; cp = cp2) {
897 		cp2 = LIST_NEXT(cp, consumers);
898 		if (!cp->spoiled)
899 			continue;
900 		cp->spoiled = 0;
901 		if (cp->geom->spoiled == NULL)
902 			continue;
903 		cp->geom->spoiled(cp);
904 		g_topology_assert();
905 	}
906 }
907 
908 void
909 g_spoil(struct g_provider *pp, struct g_consumer *cp)
910 {
911 	struct g_consumer *cp2;
912 
913 	g_topology_assert();
914 	G_VALID_PROVIDER(pp);
915 	G_VALID_CONSUMER(cp);
916 
917 	LIST_FOREACH(cp2, &pp->consumers, consumers) {
918 		if (cp2 == cp)
919 			continue;
920 /*
921 		KASSERT(cp2->acr == 0, ("spoiling cp->acr = %d", cp2->acr));
922 		KASSERT(cp2->acw == 0, ("spoiling cp->acw = %d", cp2->acw));
923 */
924 		KASSERT(cp2->ace == 0, ("spoiling cp->ace = %d", cp2->ace));
925 		cp2->spoiled++;
926 	}
927 	g_post_event(g_spoil_event, pp, M_WAITOK, pp, NULL);
928 }
929 
930 int
931 g_getattr__(const char *attr, struct g_consumer *cp, void *var, int len)
932 {
933 	int error, i;
934 
935 	i = len;
936 	error = g_io_getattr(attr, cp, &i, var);
937 	if (error)
938 		return (error);
939 	if (i != len)
940 		return (EINVAL);
941 	return (0);
942 }
943 
944 #if defined(DIAGNOSTIC) || defined(DDB)
945 /*
946  * This function walks (topologically unsafely) the mesh and return a
947  * non-zero integer if it finds the argument pointer is an object.
948  * The return value indicates which type of object it is belived to be.
949  * If topology is not locked, this function is potentially dangerous,
950  * but since it is for debugging purposes and can be useful for instance
951  * from DDB, we do not assert topology lock is held.
952  */
953 int
954 g_valid_obj(void const *ptr)
955 {
956 	struct g_class *mp;
957 	struct g_geom *gp;
958 	struct g_consumer *cp;
959 	struct g_provider *pp;
960 
961 	LIST_FOREACH(mp, &g_classes, class) {
962 		if (ptr == mp)
963 			return (1);
964 		LIST_FOREACH(gp, &mp->geom, geom) {
965 			if (ptr == gp)
966 				return (2);
967 			LIST_FOREACH(cp, &gp->consumer, consumer)
968 				if (ptr == cp)
969 					return (3);
970 			LIST_FOREACH(pp, &gp->provider, provider)
971 				if (ptr == pp)
972 					return (4);
973 		}
974 	}
975 	return(0);
976 }
977 #endif
978 
979 #ifdef DDB
980 
981 #define	gprintf(...)	do {						\
982 	printf("%*s", indent, "");					\
983 	printf(__VA_ARGS__);						\
984 } while (0)
985 #define	gprintln(...)	do {						\
986 	gprintf(__VA_ARGS__);						\
987 	printf("\n");							\
988 } while (0)
989 
990 #define	ADDFLAG(obj, flag, sflag)	do {				\
991 	if ((obj)->flags & (flag)) {					\
992 		if (comma)						\
993 			strlcat(str, ",", size);			\
994 		strlcat(str, (sflag), size);				\
995 		comma = 1;						\
996 	}								\
997 } while (0)
998 
999 static char *
1000 provider_flags_to_string(struct g_provider *pp, char *str, size_t size)
1001 {
1002 	int comma = 0;
1003 
1004 	bzero(str, size);
1005 	if (pp->flags == 0) {
1006 		strlcpy(str, "NONE", size);
1007 		return (str);
1008 	}
1009 	ADDFLAG(pp, G_PF_CANDELETE, "G_PF_CANDELETE");
1010 	ADDFLAG(pp, G_PF_WITHER, "G_PF_WITHER");
1011 	ADDFLAG(pp, G_PF_ORPHAN, "G_PF_ORPHAN");
1012 	return (str);
1013 }
1014 
1015 static char *
1016 geom_flags_to_string(struct g_geom *gp, char *str, size_t size)
1017 {
1018 	int comma = 0;
1019 
1020 	bzero(str, size);
1021 	if (gp->flags == 0) {
1022 		strlcpy(str, "NONE", size);
1023 		return (str);
1024 	}
1025 	ADDFLAG(gp, G_GEOM_WITHER, "G_GEOM_WITHER");
1026 	return (str);
1027 }
1028 static void
1029 db_show_geom_consumer(int indent, struct g_consumer *cp)
1030 {
1031 
1032 	if (indent == 0) {
1033 		gprintln("consumer: %p", cp);
1034 		gprintln("  class:    %s (%p)", cp->geom->class->name,
1035 		    cp->geom->class);
1036 		gprintln("  geom:     %s (%p)", cp->geom->name, cp->geom);
1037 		if (cp->provider == NULL)
1038 			gprintln("  provider: none");
1039 		else {
1040 			gprintln("  provider: %s (%p)", cp->provider->name,
1041 			    cp->provider);
1042 		}
1043 		gprintln("  access:   r%dw%de%d", cp->acr, cp->acw, cp->ace);
1044 		gprintln("  spoiled:  %d", cp->spoiled);
1045 		gprintln("  nstart:   %u", cp->nstart);
1046 		gprintln("  nend:     %u", cp->nend);
1047 	} else {
1048 		gprintf("consumer: %p (%s), access=r%dw%de%d", cp,
1049 		    cp->provider != NULL ? cp->provider->name : "none",
1050 		    cp->acr, cp->acw, cp->ace);
1051 		if (cp->spoiled)
1052 			printf(", spoiled=%d", cp->spoiled);
1053 		printf("\n");
1054 	}
1055 }
1056 
1057 static void
1058 db_show_geom_provider(int indent, struct g_provider *pp)
1059 {
1060 	struct g_consumer *cp;
1061 	char flags[64];
1062 
1063 	if (indent == 0) {
1064 		gprintln("provider: %s (%p)", pp->name, pp);
1065 		gprintln("  class:        %s (%p)", pp->geom->class->name,
1066 		    pp->geom->class);
1067 		gprintln("  geom:         %s (%p)", pp->geom->name, pp->geom);
1068 		gprintln("  mediasize:    %jd", (intmax_t)pp->mediasize);
1069 		gprintln("  sectorsize:   %u", pp->sectorsize);
1070 		gprintln("  stripesize:   %u", pp->stripesize);
1071 		gprintln("  stripeoffset: %u", pp->stripeoffset);
1072 		gprintln("  access:       r%dw%de%d", pp->acr, pp->acw,
1073 		    pp->ace);
1074 		gprintln("  flags:        %s (0x%04x)",
1075 		    provider_flags_to_string(pp, flags, sizeof(flags)),
1076 		    pp->flags);
1077 		gprintln("  error:        %d", pp->error);
1078 		gprintln("  nstart:       %u", pp->nstart);
1079 		gprintln("  nend:         %u", pp->nend);
1080 		if (LIST_EMPTY(&pp->consumers))
1081 			gprintln("  consumers:    none");
1082 	} else {
1083 		gprintf("provider: %s (%p), access=r%dw%de%d",
1084 		    pp->name, pp, pp->acr, pp->acw, pp->ace);
1085 		if (pp->flags != 0) {
1086 			printf(", flags=%s (0x%04x)",
1087 			    provider_flags_to_string(pp, flags, sizeof(flags)),
1088 			    pp->flags);
1089 		}
1090 		printf("\n");
1091 	}
1092 	if (!LIST_EMPTY(&pp->consumers)) {
1093 		LIST_FOREACH(cp, &pp->consumers, consumers)
1094 			db_show_geom_consumer(indent + 2, cp);
1095 	}
1096 }
1097 
1098 static void
1099 db_show_geom_geom(int indent, struct g_geom *gp)
1100 {
1101 	struct g_provider *pp;
1102 	struct g_consumer *cp;
1103 	char flags[64];
1104 
1105 	if (indent == 0) {
1106 		gprintln("geom: %s (%p)", gp->name, gp);
1107 		gprintln("  class:     %s (%p)", gp->class->name, gp->class);
1108 		gprintln("  flags:     %s (0x%04x)",
1109 		    geom_flags_to_string(gp, flags, sizeof(flags)), gp->flags);
1110 		gprintln("  rank:      %d", gp->rank);
1111 		if (LIST_EMPTY(&gp->provider))
1112 			gprintln("  providers: none");
1113 		if (LIST_EMPTY(&gp->consumer))
1114 			gprintln("  consumers: none");
1115 	} else {
1116 		gprintf("geom: %s (%p), rank=%d", gp->name, gp, gp->rank);
1117 		if (gp->flags != 0) {
1118 			printf(", flags=%s (0x%04x)",
1119 			    geom_flags_to_string(gp, flags, sizeof(flags)),
1120 			    gp->flags);
1121 		}
1122 		printf("\n");
1123 	}
1124 	if (!LIST_EMPTY(&gp->provider)) {
1125 		LIST_FOREACH(pp, &gp->provider, provider)
1126 			db_show_geom_provider(indent + 2, pp);
1127 	}
1128 	if (!LIST_EMPTY(&gp->consumer)) {
1129 		LIST_FOREACH(cp, &gp->consumer, consumer)
1130 			db_show_geom_consumer(indent + 2, cp);
1131 	}
1132 }
1133 
1134 static void
1135 db_show_geom_class(struct g_class *mp)
1136 {
1137 	struct g_geom *gp;
1138 
1139 	printf("class: %s (%p)\n", mp->name, mp);
1140 	LIST_FOREACH(gp, &mp->geom, geom)
1141 		db_show_geom_geom(2, gp);
1142 }
1143 
1144 /*
1145  * Print the GEOM topology or the given object.
1146  */
1147 DB_SHOW_COMMAND(geom, db_show_geom)
1148 {
1149 	struct g_class *mp;
1150 
1151 	if (!have_addr) {
1152 		/* No address given, print the entire topology. */
1153 		LIST_FOREACH(mp, &g_classes, class) {
1154 			db_show_geom_class(mp);
1155 			printf("\n");
1156 		}
1157 	} else {
1158 		switch (g_valid_obj((void *)addr)) {
1159 		case 1:
1160 			db_show_geom_class((struct g_class *)addr);
1161 			break;
1162 		case 2:
1163 			db_show_geom_geom(0, (struct g_geom *)addr);
1164 			break;
1165 		case 3:
1166 			db_show_geom_consumer(0, (struct g_consumer *)addr);
1167 			break;
1168 		case 4:
1169 			db_show_geom_provider(0, (struct g_provider *)addr);
1170 			break;
1171 		default:
1172 			printf("Not a GEOM object.\n");
1173 			break;
1174 		}
1175 	}
1176 }
1177 
1178 #undef	gprintf
1179 #undef	gprintln
1180 #undef	ADDFLAG
1181 
1182 #endif	/* DDB */
1183