xref: /freebsd-src/sys/geom/concat/g_concat.c (revision d409305fa3838fb39b38c26fc085fb729b8766d5)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2004-2005 Pawel Jakub Dawidek <pjd@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/lock.h>
37 #include <sys/mutex.h>
38 #include <sys/sx.h>
39 #include <sys/bio.h>
40 #include <sys/sbuf.h>
41 #include <sys/sysctl.h>
42 #include <sys/malloc.h>
43 #include <geom/geom.h>
44 #include <geom/geom_dbg.h>
45 #include <geom/concat/g_concat.h>
46 
47 FEATURE(geom_concat, "GEOM concatenation support");
48 
49 static MALLOC_DEFINE(M_CONCAT, "concat_data", "GEOM_CONCAT Data");
50 
51 SYSCTL_DECL(_kern_geom);
52 static SYSCTL_NODE(_kern_geom, OID_AUTO, concat, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
53     "GEOM_CONCAT stuff");
54 static u_int g_concat_debug = 0;
55 SYSCTL_UINT(_kern_geom_concat, OID_AUTO, debug, CTLFLAG_RWTUN, &g_concat_debug, 0,
56     "Debug level");
57 
58 static int g_concat_destroy(struct g_concat_softc *sc, boolean_t force);
59 static int g_concat_destroy_geom(struct gctl_req *req, struct g_class *mp,
60     struct g_geom *gp);
61 
62 static g_taste_t g_concat_taste;
63 static g_ctl_req_t g_concat_config;
64 static g_dumpconf_t g_concat_dumpconf;
65 
66 struct g_class g_concat_class = {
67 	.name = G_CONCAT_CLASS_NAME,
68 	.version = G_VERSION,
69 	.ctlreq = g_concat_config,
70 	.taste = g_concat_taste,
71 	.destroy_geom = g_concat_destroy_geom
72 };
73 
74 /*
75  * Greatest Common Divisor.
76  */
77 static u_int
78 gcd(u_int a, u_int b)
79 {
80 	u_int c;
81 
82 	while (b != 0) {
83 		c = a;
84 		a = b;
85 		b = (c % b);
86 	}
87 	return (a);
88 }
89 
90 /*
91  * Least Common Multiple.
92  */
93 static u_int
94 lcm(u_int a, u_int b)
95 {
96 
97 	return ((a * b) / gcd(a, b));
98 }
99 
100 /*
101  * Return the number of valid disks.
102  */
103 static u_int
104 g_concat_nvalid(struct g_concat_softc *sc)
105 {
106 	u_int no;
107 	struct g_concat_disk *disk;
108 
109 	sx_assert(&sc->sc_disks_lock, SA_LOCKED);
110 
111 	no = 0;
112 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
113 		if (disk->d_consumer != NULL)
114 			no++;
115 	}
116 
117 	return (no);
118 }
119 
120 static void
121 g_concat_remove_disk(struct g_concat_disk *disk)
122 {
123 	struct g_consumer *cp;
124 	struct g_concat_softc *sc;
125 
126 	g_topology_assert();
127 	KASSERT(disk->d_consumer != NULL, ("Non-valid disk in %s.", __func__));
128 	sc = disk->d_softc;
129 	cp = disk->d_consumer;
130 
131 	if (!disk->d_removed) {
132 		G_CONCAT_DEBUG(0, "Disk %s removed from %s.",
133 		    cp->provider->name, sc->sc_name);
134 		disk->d_removed = 1;
135 	}
136 
137 	if (sc->sc_provider != NULL) {
138 		G_CONCAT_DEBUG(0, "Device %s deactivated.",
139 		    sc->sc_provider->name);
140 		g_wither_provider(sc->sc_provider, ENXIO);
141 		sc->sc_provider = NULL;
142 	}
143 
144 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
145 		return;
146 	disk->d_consumer = NULL;
147 	g_detach(cp);
148 	g_destroy_consumer(cp);
149 	/* If there are no valid disks anymore, remove device. */
150 	if (LIST_EMPTY(&sc->sc_geom->consumer))
151 		g_concat_destroy(sc, 1);
152 }
153 
154 static void
155 g_concat_orphan(struct g_consumer *cp)
156 {
157 	struct g_concat_softc *sc;
158 	struct g_concat_disk *disk;
159 	struct g_geom *gp;
160 
161 	g_topology_assert();
162 	gp = cp->geom;
163 	sc = gp->softc;
164 	if (sc == NULL)
165 		return;
166 
167 	disk = cp->private;
168 	if (disk == NULL)	/* Possible? */
169 		return;
170 	g_concat_remove_disk(disk);
171 }
172 
173 static int
174 g_concat_access(struct g_provider *pp, int dr, int dw, int de)
175 {
176 	struct g_consumer *cp1, *cp2, *tmp;
177 	struct g_concat_disk *disk;
178 	struct g_geom *gp;
179 	struct g_concat_softc *sc;
180 	int error;
181 
182 	g_topology_assert();
183 	gp = pp->geom;
184 	sc = gp->softc;
185 
186 	/* On first open, grab an extra "exclusive" bit */
187 	if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
188 		de++;
189 	/* ... and let go of it on last close */
190 	if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0)
191 		de--;
192 
193 	sx_slock(&sc->sc_disks_lock);
194 	LIST_FOREACH_SAFE(cp1, &gp->consumer, consumer, tmp) {
195 		error = g_access(cp1, dr, dw, de);
196 		if (error != 0)
197 			goto fail;
198 		disk = cp1->private;
199 		if (cp1->acr == 0 && cp1->acw == 0 && cp1->ace == 0 &&
200 		    disk->d_removed) {
201 			g_concat_remove_disk(disk); /* May destroy geom. */
202 		}
203 	}
204 	sx_sunlock(&sc->sc_disks_lock);
205 	return (0);
206 
207 fail:
208 	sx_sunlock(&sc->sc_disks_lock);
209 	LIST_FOREACH(cp2, &gp->consumer, consumer) {
210 		if (cp1 == cp2)
211 			break;
212 		g_access(cp2, -dr, -dw, -de);
213 	}
214 	return (error);
215 }
216 
217 static void
218 g_concat_candelete(struct bio *bp)
219 {
220 	struct g_concat_softc *sc;
221 	struct g_concat_disk *disk;
222 	int val;
223 
224 	sc = bp->bio_to->geom->softc;
225 	sx_assert(&sc->sc_disks_lock, SX_LOCKED);
226 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
227 		if (!disk->d_removed && disk->d_candelete)
228 			break;
229 	}
230 	val = disk != NULL;
231 	g_handleattr(bp, "GEOM::candelete", &val, sizeof(val));
232 }
233 
234 static void
235 g_concat_kernel_dump(struct bio *bp)
236 {
237 	struct g_concat_softc *sc;
238 	struct g_concat_disk *disk;
239 	struct bio *cbp;
240 	struct g_kerneldump *gkd;
241 
242 	sc = bp->bio_to->geom->softc;
243 	gkd = (struct g_kerneldump *)bp->bio_data;
244 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
245 		if (disk->d_start <= gkd->offset &&
246 		    disk->d_end > gkd->offset)
247 			break;
248 	}
249 	if (disk == NULL) {
250 		g_io_deliver(bp, EOPNOTSUPP);
251 		return;
252 	}
253 
254 	gkd->offset -= disk->d_start;
255 	if (gkd->length > disk->d_end - disk->d_start - gkd->offset)
256 		gkd->length = disk->d_end - disk->d_start - gkd->offset;
257 	cbp = g_clone_bio(bp);
258 	if (cbp == NULL) {
259 		g_io_deliver(bp, ENOMEM);
260 		return;
261 	}
262 	cbp->bio_done = g_std_done;
263 	g_io_request(cbp, disk->d_consumer);
264 	G_CONCAT_DEBUG(1, "Kernel dump will go to %s.",
265 	    disk->d_consumer->provider->name);
266 }
267 
268 static void
269 g_concat_done(struct bio *bp)
270 {
271 	struct g_concat_softc *sc;
272 	struct bio *pbp;
273 
274 	pbp = bp->bio_parent;
275 	sc = pbp->bio_to->geom->softc;
276 	mtx_lock(&sc->sc_completion_lock);
277 	if (pbp->bio_error == 0)
278 		pbp->bio_error = bp->bio_error;
279 	pbp->bio_completed += bp->bio_completed;
280 	pbp->bio_inbed++;
281 	if (pbp->bio_children == pbp->bio_inbed) {
282 		mtx_unlock(&sc->sc_completion_lock);
283 		g_io_deliver(pbp, pbp->bio_error);
284 	} else
285 		mtx_unlock(&sc->sc_completion_lock);
286 	g_destroy_bio(bp);
287 }
288 
289 /*
290  * Called for both BIO_FLUSH and BIO_SPEEDUP. Just pass the call down
291  */
292 static void
293 g_concat_passdown(struct g_concat_softc *sc, struct bio *bp)
294 {
295 	struct bio_queue_head queue;
296 	struct g_consumer *cp;
297 	struct bio *cbp;
298 	struct g_concat_disk *disk;
299 
300 	sx_assert(&sc->sc_disks_lock, SX_LOCKED);
301 
302 	bioq_init(&queue);
303 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
304 		cbp = g_clone_bio(bp);
305 		if (cbp == NULL) {
306 			while ((cbp = bioq_takefirst(&queue)) != NULL)
307 				g_destroy_bio(cbp);
308 			if (bp->bio_error == 0)
309 				bp->bio_error = ENOMEM;
310 			g_io_deliver(bp, bp->bio_error);
311 			return;
312 		}
313 		bioq_insert_tail(&queue, cbp);
314 		cbp->bio_done = g_concat_done;
315 		cbp->bio_caller1 = disk->d_consumer;
316 		cbp->bio_to = disk->d_consumer->provider;
317 	}
318 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
319 		G_CONCAT_LOGREQ(cbp, "Sending request.");
320 		cp = cbp->bio_caller1;
321 		cbp->bio_caller1 = NULL;
322 		g_io_request(cbp, cp);
323 	}
324 }
325 
326 static void
327 g_concat_start(struct bio *bp)
328 {
329 	struct bio_queue_head queue;
330 	struct g_concat_softc *sc;
331 	struct g_concat_disk *disk;
332 	struct g_provider *pp;
333 	off_t offset, end, length, off, len;
334 	struct bio *cbp;
335 	char *addr;
336 
337 	pp = bp->bio_to;
338 	sc = pp->geom->softc;
339 	/*
340 	 * If sc == NULL, provider's error should be set and g_concat_start()
341 	 * should not be called at all.
342 	 */
343 	KASSERT(sc != NULL,
344 	    ("Provider's error should be set (error=%d)(device=%s).",
345 	    bp->bio_to->error, bp->bio_to->name));
346 
347 	G_CONCAT_LOGREQ(bp, "Request received.");
348 	sx_slock(&sc->sc_disks_lock);
349 
350 	switch (bp->bio_cmd) {
351 	case BIO_READ:
352 	case BIO_WRITE:
353 	case BIO_DELETE:
354 		break;
355 	case BIO_SPEEDUP:
356 	case BIO_FLUSH:
357 		g_concat_passdown(sc, bp);
358 		goto end;
359 	case BIO_GETATTR:
360 		if (strcmp("GEOM::kerneldump", bp->bio_attribute) == 0) {
361 			g_concat_kernel_dump(bp);
362 			goto end;
363 		} else if (strcmp("GEOM::candelete", bp->bio_attribute) == 0) {
364 			g_concat_candelete(bp);
365 			goto end;
366 		}
367 		/* To which provider it should be delivered? */
368 		/* FALLTHROUGH */
369 	default:
370 		g_io_deliver(bp, EOPNOTSUPP);
371 		goto end;
372 	}
373 
374 	offset = bp->bio_offset;
375 	length = bp->bio_length;
376 	if ((bp->bio_flags & BIO_UNMAPPED) != 0)
377 		addr = NULL;
378 	else
379 		addr = bp->bio_data;
380 	end = offset + length;
381 
382 	bioq_init(&queue);
383 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
384 		if (disk->d_end <= offset)
385 			continue;
386 		if (disk->d_start >= end)
387 			break;
388 
389 		off = offset - disk->d_start;
390 		len = MIN(length, disk->d_end - offset);
391 		length -= len;
392 		offset += len;
393 
394 		cbp = g_clone_bio(bp);
395 		if (cbp == NULL) {
396 			while ((cbp = bioq_takefirst(&queue)) != NULL)
397 				g_destroy_bio(cbp);
398 			if (bp->bio_error == 0)
399 				bp->bio_error = ENOMEM;
400 			g_io_deliver(bp, bp->bio_error);
401 			goto end;
402 		}
403 		bioq_insert_tail(&queue, cbp);
404 		/*
405 		 * Fill in the component buf structure.
406 		 */
407 		if (len == bp->bio_length)
408 			cbp->bio_done = g_std_done;
409 		else
410 			cbp->bio_done = g_concat_done;
411 		cbp->bio_offset = off;
412 		cbp->bio_length = len;
413 		if ((bp->bio_flags & BIO_UNMAPPED) != 0) {
414 			cbp->bio_ma_offset += (uintptr_t)addr;
415 			cbp->bio_ma += cbp->bio_ma_offset / PAGE_SIZE;
416 			cbp->bio_ma_offset %= PAGE_SIZE;
417 			cbp->bio_ma_n = round_page(cbp->bio_ma_offset +
418 			    cbp->bio_length) / PAGE_SIZE;
419 		} else
420 			cbp->bio_data = addr;
421 		addr += len;
422 		cbp->bio_to = disk->d_consumer->provider;
423 		cbp->bio_caller1 = disk;
424 
425 		if (length == 0)
426 			break;
427 	}
428 	KASSERT(length == 0,
429 	    ("Length is still greater than 0 (class=%s, name=%s).",
430 	    bp->bio_to->geom->class->name, bp->bio_to->geom->name));
431 	while ((cbp = bioq_takefirst(&queue)) != NULL) {
432 		G_CONCAT_LOGREQ(cbp, "Sending request.");
433 		disk = cbp->bio_caller1;
434 		cbp->bio_caller1 = NULL;
435 		g_io_request(cbp, disk->d_consumer);
436 	}
437 end:
438 	sx_sunlock(&sc->sc_disks_lock);
439 }
440 
441 static void
442 g_concat_check_and_run(struct g_concat_softc *sc)
443 {
444 	struct g_concat_disk *disk;
445 	struct g_provider *dp, *pp;
446 	u_int sectorsize = 0;
447 	off_t start;
448 	int error;
449 
450 	g_topology_assert();
451 	if (g_concat_nvalid(sc) != sc->sc_ndisks)
452 		return;
453 
454 	pp = g_new_providerf(sc->sc_geom, "concat/%s", sc->sc_name);
455 	pp->flags |= G_PF_DIRECT_SEND | G_PF_DIRECT_RECEIVE |
456 	    G_PF_ACCEPT_UNMAPPED;
457 	start = 0;
458 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
459 		dp = disk->d_consumer->provider;
460 		disk->d_start = start;
461 		disk->d_end = disk->d_start + dp->mediasize;
462 		if (sc->sc_type == G_CONCAT_TYPE_AUTOMATIC)
463 			disk->d_end -= dp->sectorsize;
464 		start = disk->d_end;
465 		error = g_access(disk->d_consumer, 1, 0, 0);
466 		if (error == 0) {
467 			error = g_getattr("GEOM::candelete", disk->d_consumer,
468 			    &disk->d_candelete);
469 			if (error != 0)
470 				disk->d_candelete = 0;
471 			(void)g_access(disk->d_consumer, -1, 0, 0);
472 		} else
473 			G_CONCAT_DEBUG(1, "Failed to access disk %s, error %d.",
474 			    dp->name, error);
475 		if (disk == TAILQ_FIRST(&sc->sc_disks))
476 			sectorsize = dp->sectorsize;
477 		else
478 			sectorsize = lcm(sectorsize, dp->sectorsize);
479 
480 		/* A provider underneath us doesn't support unmapped */
481 		if ((dp->flags & G_PF_ACCEPT_UNMAPPED) == 0) {
482 			G_CONCAT_DEBUG(1, "Cancelling unmapped "
483 			    "because of %s.", dp->name);
484 			pp->flags &= ~G_PF_ACCEPT_UNMAPPED;
485 		}
486 	}
487 	pp->sectorsize = sectorsize;
488 	/* We have sc->sc_disks[sc->sc_ndisks - 1].d_end in 'start'. */
489 	pp->mediasize = start;
490 	dp = TAILQ_FIRST(&sc->sc_disks)->d_consumer->provider;
491 	pp->stripesize = dp->stripesize;
492 	pp->stripeoffset = dp->stripeoffset;
493 	sc->sc_provider = pp;
494 	g_error_provider(pp, 0);
495 
496 	G_CONCAT_DEBUG(0, "Device %s activated.", sc->sc_provider->name);
497 }
498 
499 static int
500 g_concat_read_metadata(struct g_consumer *cp, struct g_concat_metadata *md)
501 {
502 	struct g_provider *pp;
503 	u_char *buf;
504 	int error;
505 
506 	g_topology_assert();
507 
508 	error = g_access(cp, 1, 0, 0);
509 	if (error != 0)
510 		return (error);
511 	pp = cp->provider;
512 	g_topology_unlock();
513 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
514 	    &error);
515 	g_topology_lock();
516 	g_access(cp, -1, 0, 0);
517 	if (buf == NULL)
518 		return (error);
519 
520 	/* Decode metadata. */
521 	concat_metadata_decode(buf, md);
522 	g_free(buf);
523 
524 	return (0);
525 }
526 
527 /*
528  * Add disk to given device.
529  */
530 static int
531 g_concat_add_disk(struct g_concat_softc *sc, struct g_provider *pp, u_int no)
532 {
533 	struct g_concat_disk *disk;
534 	struct g_consumer *cp, *fcp;
535 	struct g_geom *gp;
536 	int error;
537 
538 	g_topology_assert();
539 
540 	sx_slock(&sc->sc_disks_lock);
541 
542 	/* Metadata corrupted? */
543 	if (no >= sc->sc_ndisks) {
544 		sx_sunlock(&sc->sc_disks_lock);
545 		return (EINVAL);
546 	}
547 
548 	for (disk = TAILQ_FIRST(&sc->sc_disks); no > 0; no--) {
549 		disk = TAILQ_NEXT(disk, d_next);
550 	}
551 
552 	/* Check if disk is not already attached. */
553 	if (disk->d_consumer != NULL) {
554 		sx_sunlock(&sc->sc_disks_lock);
555 		return (EEXIST);
556 	}
557 
558 	gp = sc->sc_geom;
559 	fcp = LIST_FIRST(&gp->consumer);
560 
561 	cp = g_new_consumer(gp);
562 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
563 	error = g_attach(cp, pp);
564 	if (error != 0) {
565 		sx_sunlock(&sc->sc_disks_lock);
566 		g_destroy_consumer(cp);
567 		return (error);
568 	}
569 
570 	if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) {
571 		error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
572 		if (error != 0) {
573 			sx_sunlock(&sc->sc_disks_lock);
574 			g_detach(cp);
575 			g_destroy_consumer(cp);
576 			return (error);
577 		}
578 	}
579 	if (sc->sc_type == G_CONCAT_TYPE_AUTOMATIC) {
580 		struct g_concat_metadata md;
581 
582 		// temporarily give up the lock to avoid lock order violation
583 		// due to topology unlock in g_concat_read_metadata
584 		sx_sunlock(&sc->sc_disks_lock);
585 		/* Re-read metadata. */
586 		error = g_concat_read_metadata(cp, &md);
587 		sx_slock(&sc->sc_disks_lock);
588 
589 		if (error != 0)
590 			goto fail;
591 
592 		if (strcmp(md.md_magic, G_CONCAT_MAGIC) != 0 ||
593 		    strcmp(md.md_name, sc->sc_name) != 0 ||
594 		    md.md_id != sc->sc_id) {
595 			G_CONCAT_DEBUG(0, "Metadata on %s changed.", pp->name);
596 			goto fail;
597 		}
598 	}
599 
600 	cp->private = disk;
601 	disk->d_consumer = cp;
602 	disk->d_softc = sc;
603 	disk->d_start = 0;	/* not yet */
604 	disk->d_end = 0;	/* not yet */
605 	disk->d_removed = 0;
606 
607 	G_CONCAT_DEBUG(0, "Disk %s attached to %s.", pp->name, sc->sc_name);
608 
609 	g_concat_check_and_run(sc);
610 	sx_sunlock(&sc->sc_disks_lock); // need lock for check_and_run
611 
612 	return (0);
613 fail:
614 	sx_sunlock(&sc->sc_disks_lock);
615 	if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0))
616 		g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace);
617 	g_detach(cp);
618 	g_destroy_consumer(cp);
619 	return (error);
620 }
621 
622 static struct g_geom *
623 g_concat_create(struct g_class *mp, const struct g_concat_metadata *md,
624     u_int type)
625 {
626 	struct g_concat_softc *sc;
627 	struct g_concat_disk *disk;
628 	struct g_geom *gp;
629 	u_int no;
630 
631 	G_CONCAT_DEBUG(1, "Creating device %s (id=%u).", md->md_name,
632 	    md->md_id);
633 
634 	/* One disks is minimum. */
635 	if (md->md_all < 1)
636 		return (NULL);
637 
638 	/* Check for duplicate unit */
639 	LIST_FOREACH(gp, &mp->geom, geom) {
640 		sc = gp->softc;
641 		if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) {
642 			G_CONCAT_DEBUG(0, "Device %s already configured.",
643 			    gp->name);
644 			return (NULL);
645 		}
646 	}
647 	gp = g_new_geomf(mp, "%s", md->md_name);
648 	sc = malloc(sizeof(*sc), M_CONCAT, M_WAITOK | M_ZERO);
649 	gp->start = g_concat_start;
650 	gp->spoiled = g_concat_orphan;
651 	gp->orphan = g_concat_orphan;
652 	gp->access = g_concat_access;
653 	gp->dumpconf = g_concat_dumpconf;
654 
655 	sc->sc_id = md->md_id;
656 	sc->sc_ndisks = md->md_all;
657 	TAILQ_INIT(&sc->sc_disks);
658 	for (no = 0; no < sc->sc_ndisks; no++) {
659 		disk = malloc(sizeof(*disk), M_CONCAT, M_WAITOK | M_ZERO);
660 		TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
661 	}
662 	sc->sc_type = type;
663 	mtx_init(&sc->sc_completion_lock, "gconcat lock", NULL, MTX_DEF);
664 	sx_init(&sc->sc_disks_lock, "gconcat append lock");
665 
666 	gp->softc = sc;
667 	sc->sc_geom = gp;
668 	sc->sc_provider = NULL;
669 
670 	G_CONCAT_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id);
671 
672 	return (gp);
673 }
674 
675 static int
676 g_concat_destroy(struct g_concat_softc *sc, boolean_t force)
677 {
678 	struct g_provider *pp;
679 	struct g_consumer *cp, *cp1;
680 	struct g_geom *gp;
681 	struct g_concat_disk *disk;
682 
683 	g_topology_assert();
684 
685 	if (sc == NULL)
686 		return (ENXIO);
687 
688 	pp = sc->sc_provider;
689 	if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
690 		if (force) {
691 			G_CONCAT_DEBUG(0, "Device %s is still open, so it "
692 			    "can't be definitely removed.", pp->name);
693 		} else {
694 			G_CONCAT_DEBUG(1,
695 			    "Device %s is still open (r%dw%de%d).", pp->name,
696 			    pp->acr, pp->acw, pp->ace);
697 			return (EBUSY);
698 		}
699 	}
700 
701 	gp = sc->sc_geom;
702 	LIST_FOREACH_SAFE(cp, &gp->consumer, consumer, cp1) {
703 		g_concat_remove_disk(cp->private);
704 		if (cp1 == NULL)
705 			return (0);	/* Recursion happened. */
706 	}
707 	if (!LIST_EMPTY(&gp->consumer))
708 		return (EINPROGRESS);
709 
710 	gp->softc = NULL;
711 	KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)",
712 	    gp->name));
713 	while ((disk = TAILQ_FIRST(&sc->sc_disks)) != NULL) {
714 		TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
715 		free(disk, M_CONCAT);
716 	}
717 	mtx_destroy(&sc->sc_completion_lock);
718 	sx_destroy(&sc->sc_disks_lock);
719 	free(sc, M_CONCAT);
720 
721 	G_CONCAT_DEBUG(0, "Device %s destroyed.", gp->name);
722 	g_wither_geom(gp, ENXIO);
723 	return (0);
724 }
725 
726 static int
727 g_concat_destroy_geom(struct gctl_req *req __unused,
728     struct g_class *mp __unused, struct g_geom *gp)
729 {
730 	struct g_concat_softc *sc;
731 
732 	sc = gp->softc;
733 	return (g_concat_destroy(sc, 0));
734 }
735 
736 static struct g_geom *
737 g_concat_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
738 {
739 	struct g_concat_metadata md;
740 	struct g_concat_softc *sc;
741 	struct g_consumer *cp;
742 	struct g_geom *gp;
743 	int error;
744 
745 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
746 	g_topology_assert();
747 
748 	/* Skip providers that are already open for writing. */
749 	if (pp->acw > 0)
750 		return (NULL);
751 
752 	G_CONCAT_DEBUG(3, "Tasting %s.", pp->name);
753 
754 	gp = g_new_geomf(mp, "concat:taste");
755 	gp->start = g_concat_start;
756 	gp->access = g_concat_access;
757 	gp->orphan = g_concat_orphan;
758 	cp = g_new_consumer(gp);
759 	error = g_attach(cp, pp);
760 	if (error == 0) {
761 		error = g_concat_read_metadata(cp, &md);
762 		g_detach(cp);
763 	}
764 	g_destroy_consumer(cp);
765 	g_destroy_geom(gp);
766 	if (error != 0)
767 		return (NULL);
768 	gp = NULL;
769 
770 	if (strcmp(md.md_magic, G_CONCAT_MAGIC) != 0)
771 		return (NULL);
772 	if (md.md_version > G_CONCAT_VERSION) {
773 		printf("geom_concat.ko module is too old to handle %s.\n",
774 		    pp->name);
775 		return (NULL);
776 	}
777 	/*
778 	 * Backward compatibility:
779 	 */
780 	/* There was no md_provider field in earlier versions of metadata. */
781 	if (md.md_version < 3)
782 		bzero(md.md_provider, sizeof(md.md_provider));
783 	/* There was no md_provsize field in earlier versions of metadata. */
784 	if (md.md_version < 4)
785 		md.md_provsize = pp->mediasize;
786 
787 	if (md.md_provider[0] != '\0' &&
788 	    !g_compare_names(md.md_provider, pp->name))
789 		return (NULL);
790 	if (md.md_provsize != pp->mediasize)
791 		return (NULL);
792 
793 	/*
794 	 * Let's check if device already exists.
795 	 */
796 	sc = NULL;
797 	LIST_FOREACH(gp, &mp->geom, geom) {
798 		sc = gp->softc;
799 		if (sc == NULL)
800 			continue;
801 		if (sc->sc_type != G_CONCAT_TYPE_AUTOMATIC)
802 			continue;
803 		if (strcmp(md.md_name, sc->sc_name) != 0)
804 			continue;
805 		if (md.md_id != sc->sc_id)
806 			continue;
807 		break;
808 	}
809 	if (gp != NULL) {
810 		G_CONCAT_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
811 		error = g_concat_add_disk(sc, pp, md.md_no);
812 		if (error != 0) {
813 			G_CONCAT_DEBUG(0,
814 			    "Cannot add disk %s to %s (error=%d).", pp->name,
815 			    gp->name, error);
816 			return (NULL);
817 		}
818 	} else {
819 		gp = g_concat_create(mp, &md, G_CONCAT_TYPE_AUTOMATIC);
820 		if (gp == NULL) {
821 			G_CONCAT_DEBUG(0, "Cannot create device %s.",
822 			    md.md_name);
823 			return (NULL);
824 		}
825 		sc = gp->softc;
826 		G_CONCAT_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
827 		error = g_concat_add_disk(sc, pp, md.md_no);
828 		if (error != 0) {
829 			G_CONCAT_DEBUG(0,
830 			    "Cannot add disk %s to %s (error=%d).", pp->name,
831 			    gp->name, error);
832 			g_concat_destroy(sc, 1);
833 			return (NULL);
834 		}
835 	}
836 
837 	return (gp);
838 }
839 
840 static void
841 g_concat_ctl_create(struct gctl_req *req, struct g_class *mp)
842 {
843 	u_int attached, no;
844 	struct g_concat_metadata md;
845 	struct g_provider *pp;
846 	struct g_concat_softc *sc;
847 	struct g_geom *gp;
848 	struct sbuf *sb;
849 	const char *name;
850 	char param[16];
851 	int *nargs;
852 
853 	g_topology_assert();
854 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
855 	if (nargs == NULL) {
856 		gctl_error(req, "No '%s' argument.", "nargs");
857 		return;
858 	}
859 	if (*nargs < 2) {
860 		gctl_error(req, "Too few arguments.");
861 		return;
862 	}
863 
864 	strlcpy(md.md_magic, G_CONCAT_MAGIC, sizeof(md.md_magic));
865 	md.md_version = G_CONCAT_VERSION;
866 	name = gctl_get_asciiparam(req, "arg0");
867 	if (name == NULL) {
868 		gctl_error(req, "No 'arg%u' argument.", 0);
869 		return;
870 	}
871 	strlcpy(md.md_name, name, sizeof(md.md_name));
872 	md.md_id = arc4random();
873 	md.md_no = 0;
874 	md.md_all = *nargs - 1;
875 	bzero(md.md_provider, sizeof(md.md_provider));
876 	/* This field is not important here. */
877 	md.md_provsize = 0;
878 
879 	/* Check all providers are valid */
880 	for (no = 1; no < *nargs; no++) {
881 		snprintf(param, sizeof(param), "arg%u", no);
882 		pp = gctl_get_provider(req, param);
883 		if (pp == NULL)
884 			return;
885 	}
886 
887 	gp = g_concat_create(mp, &md, G_CONCAT_TYPE_MANUAL);
888 	if (gp == NULL) {
889 		gctl_error(req, "Can't configure %s.", md.md_name);
890 		return;
891 	}
892 
893 	sc = gp->softc;
894 	sb = sbuf_new_auto();
895 	sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name);
896 	for (attached = 0, no = 1; no < *nargs; no++) {
897 		snprintf(param, sizeof(param), "arg%u", no);
898 		pp = gctl_get_provider(req, param);
899 		if (pp == NULL) {
900 			name = gctl_get_asciiparam(req, param);
901 			MPASS(name != NULL);
902 			sbuf_printf(sb, " %s", name);
903 			continue;
904 		}
905 		if (g_concat_add_disk(sc, pp, no - 1) != 0) {
906 			G_CONCAT_DEBUG(1, "Disk %u (%s) not attached to %s.",
907 			    no, pp->name, gp->name);
908 			sbuf_printf(sb, " %s", pp->name);
909 			continue;
910 		}
911 		attached++;
912 	}
913 	sbuf_finish(sb);
914 	if (md.md_all != attached) {
915 		g_concat_destroy(gp->softc, 1);
916 		gctl_error(req, "%s", sbuf_data(sb));
917 	}
918 	sbuf_delete(sb);
919 }
920 
921 static struct g_concat_softc *
922 g_concat_find_device(struct g_class *mp, const char *name)
923 {
924 	struct g_concat_softc *sc;
925 	struct g_geom *gp;
926 
927 	if (strncmp(name, _PATH_DEV, strlen(_PATH_DEV)) == 0)
928 		name += strlen(_PATH_DEV);
929 
930 	LIST_FOREACH(gp, &mp->geom, geom) {
931 		sc = gp->softc;
932 		if (sc == NULL)
933 			continue;
934 		if (strcmp(sc->sc_name, name) == 0)
935 			return (sc);
936 	}
937 	return (NULL);
938 }
939 
940 static void
941 g_concat_ctl_destroy(struct gctl_req *req, struct g_class *mp)
942 {
943 	struct g_concat_softc *sc;
944 	int *force, *nargs, error;
945 	const char *name;
946 	char param[16];
947 	u_int i;
948 
949 	g_topology_assert();
950 
951 	nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
952 	if (nargs == NULL) {
953 		gctl_error(req, "No '%s' argument.", "nargs");
954 		return;
955 	}
956 	if (*nargs <= 0) {
957 		gctl_error(req, "Missing device(s).");
958 		return;
959 	}
960 	force = gctl_get_paraml(req, "force", sizeof(*force));
961 	if (force == NULL) {
962 		gctl_error(req, "No '%s' argument.", "force");
963 		return;
964 	}
965 
966 	for (i = 0; i < (u_int)*nargs; i++) {
967 		snprintf(param, sizeof(param), "arg%u", i);
968 		name = gctl_get_asciiparam(req, param);
969 		if (name == NULL) {
970 			gctl_error(req, "No 'arg%u' argument.", i);
971 			return;
972 		}
973 		sc = g_concat_find_device(mp, name);
974 		if (sc == NULL) {
975 			gctl_error(req, "No such device: %s.", name);
976 			return;
977 		}
978 		error = g_concat_destroy(sc, *force);
979 		if (error != 0) {
980 			gctl_error(req, "Cannot destroy device %s (error=%d).",
981 			    sc->sc_name, error);
982 			return;
983 		}
984 	}
985 }
986 
987 static void
988 g_concat_config(struct gctl_req *req, struct g_class *mp, const char *verb)
989 {
990 	uint32_t *version;
991 
992 	g_topology_assert();
993 
994 	version = gctl_get_paraml(req, "version", sizeof(*version));
995 	if (version == NULL) {
996 		gctl_error(req, "No '%s' argument.", "version");
997 		return;
998 	}
999 	if (*version != G_CONCAT_VERSION) {
1000 		gctl_error(req, "Userland and kernel parts are out of sync.");
1001 		return;
1002 	}
1003 
1004 	if (strcmp(verb, "create") == 0) {
1005 		g_concat_ctl_create(req, mp);
1006 		return;
1007 	} else if (strcmp(verb, "destroy") == 0 ||
1008 	    strcmp(verb, "stop") == 0) {
1009 		g_concat_ctl_destroy(req, mp);
1010 		return;
1011 	}
1012 	gctl_error(req, "Unknown verb.");
1013 }
1014 
1015 static void
1016 g_concat_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1017     struct g_consumer *cp, struct g_provider *pp)
1018 {
1019 	struct g_concat_softc *sc;
1020 
1021 	g_topology_assert();
1022 	sc = gp->softc;
1023 	if (sc == NULL)
1024 		return;
1025 
1026 	sx_slock(&sc->sc_disks_lock);
1027 	if (pp != NULL) {
1028 		/* Nothing here. */
1029 	} else if (cp != NULL) {
1030 		struct g_concat_disk *disk;
1031 
1032 		disk = cp->private;
1033 		if (disk == NULL)
1034 			goto end;
1035 		sbuf_printf(sb, "%s<End>%jd</End>\n", indent,
1036 		    (intmax_t)disk->d_end);
1037 		sbuf_printf(sb, "%s<Start>%jd</Start>\n", indent,
1038 		    (intmax_t)disk->d_start);
1039 	} else {
1040 		sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
1041 		sbuf_printf(sb, "%s<Type>", indent);
1042 		switch (sc->sc_type) {
1043 		case G_CONCAT_TYPE_AUTOMATIC:
1044 			sbuf_cat(sb, "AUTOMATIC");
1045 			break;
1046 		case G_CONCAT_TYPE_MANUAL:
1047 			sbuf_cat(sb, "MANUAL");
1048 			break;
1049 		default:
1050 			sbuf_cat(sb, "UNKNOWN");
1051 			break;
1052 		}
1053 		sbuf_cat(sb, "</Type>\n");
1054 		sbuf_printf(sb, "%s<Status>Total=%u, Online=%u</Status>\n",
1055 		    indent, sc->sc_ndisks, g_concat_nvalid(sc));
1056 		sbuf_printf(sb, "%s<State>", indent);
1057 		if (sc->sc_provider != NULL && sc->sc_provider->error == 0)
1058 			sbuf_cat(sb, "UP");
1059 		else
1060 			sbuf_cat(sb, "DOWN");
1061 		sbuf_cat(sb, "</State>\n");
1062 	}
1063 end:
1064 	sx_sunlock(&sc->sc_disks_lock);
1065 }
1066 
1067 DECLARE_GEOM_CLASS(g_concat_class, g_concat);
1068 MODULE_VERSION(geom_concat, 0);
1069