1/*-
2 * Copyright (c) 2003 Pawel Jakub Dawidek <pjd@FreeBSD.org>
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 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/geom/stripe/g_stripe.c 131878 2004-07-09 14:30:09Z pjd $");
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/kernel.h>
33#include <sys/module.h>
34#include <sys/lock.h>
35#include <sys/mutex.h>
36#include <sys/bio.h>
37#include <sys/sysctl.h>
38#include <sys/malloc.h>
39#include <vm/uma.h>
40#include <geom/geom.h>
41#include <geom/stripe/g_stripe.h>
42
43
44#define MAX_IO_SIZE (DFLTPHYS * 2)
45static MALLOC_DEFINE(M_STRIPE, "stripe data", "GEOM_STRIPE Data");
46
47static uma_zone_t g_stripe_zone;
48
49static int g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force);
50static int g_stripe_destroy_geom(struct gctl_req *req, struct g_class *mp,
51 struct g_geom *gp);
52
53static g_taste_t g_stripe_taste;
54static g_ctl_req_t g_stripe_config;
55static g_dumpconf_t g_stripe_dumpconf;
56static g_init_t g_stripe_init;
57static g_fini_t g_stripe_fini;
58
59struct g_class g_stripe_class = {
60 .name = G_STRIPE_CLASS_NAME,
61 .ctlreq = g_stripe_config,
62 .taste = g_stripe_taste,
63 .destroy_geom = g_stripe_destroy_geom,
64 .init = g_stripe_init,
65 .fini = g_stripe_fini
66};
67
68SYSCTL_DECL(_kern_geom);
69SYSCTL_NODE(_kern_geom, OID_AUTO, stripe, CTLFLAG_RW, 0, "GEOM_STRIPE stuff");
70static u_int g_stripe_debug = 0;
71SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, debug, CTLFLAG_RW, &g_stripe_debug, 0,
72 "Debug level");
73static int g_stripe_fast = 1;
74TUNABLE_INT("kern.geom.stripe.fast", &g_stripe_fast);
75static int
76g_sysctl_stripe_fast(SYSCTL_HANDLER_ARGS)
77{
78 int error, fast;
79
80 fast = g_stripe_fast;
81 error = sysctl_handle_int(oidp, &fast, sizeof(fast), req);
82 if (error == 0 && req->newptr != NULL)
83 g_stripe_fast = fast;
84 return (error);
85}
86SYSCTL_PROC(_kern_geom_stripe, OID_AUTO, fast, CTLTYPE_INT | CTLFLAG_RW,
87 NULL, 0, g_sysctl_stripe_fast, "I", "Fast, but memory-consuming mode");
88static u_int g_stripe_maxmem = MAX_IO_SIZE * 10;
89TUNABLE_INT("kern.geom.stripe.maxmem", &g_stripe_maxmem);
90SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, maxmem, CTLFLAG_RD, &g_stripe_maxmem,
91 0, "Maximum memory that could be allocated in \"fast\" mode (in bytes)");
92
93/*
94 * Greatest Common Divisor.
95 */
96static u_int
97gcd(u_int a, u_int b)
98{
99 u_int c;
100
101 while (b != 0) {
102 c = a;
103 a = b;
104 b = (c % b);
105 }
106 return (a);
107}
108
109/*
110 * Least Common Multiple.
111 */
112static u_int
113lcm(u_int a, u_int b)
114{
115
116 return ((a * b) / gcd(a, b));
117}
118
119static void
120g_stripe_init(struct g_class *mp __unused)
121{
122
123 g_stripe_zone = uma_zcreate("g_stripe_zone", MAX_IO_SIZE, NULL, NULL,
124 NULL, NULL, 0, 0);
125 g_stripe_maxmem -= g_stripe_maxmem % MAX_IO_SIZE;
126 uma_zone_set_max(g_stripe_zone, g_stripe_maxmem / MAX_IO_SIZE);
127}
128
129static void
130g_stripe_fini(struct g_class *mp __unused)
131{
132
133 uma_zdestroy(g_stripe_zone);
134}
135
136/*
137 * Return the number of valid disks.
138 */
139static u_int
140g_stripe_nvalid(struct g_stripe_softc *sc)
141{
142 u_int i, no;
143
144 no = 0;
145 for (i = 0; i < sc->sc_ndisks; i++) {
146 if (sc->sc_disks[i] != NULL)
147 no++;
148 }
149
150 return (no);
151}
152
153static void
154g_stripe_remove_disk(struct g_consumer *cp)
155{
156 struct g_stripe_softc *sc;
157 u_int no;
158
159 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__));
160 sc = (struct g_stripe_softc *)cp->private;
161 KASSERT(sc != NULL, ("NULL sc in %s.", __func__));
162 no = cp->index;
163
164 G_STRIPE_DEBUG(0, "Disk %s removed from %s.", cp->provider->name,
165 sc->sc_geom->name);
166
167 sc->sc_disks[no] = NULL;
168 if (sc->sc_provider != NULL) {
169 g_orphan_provider(sc->sc_provider, ENXIO);
170 sc->sc_provider = NULL;
171 G_STRIPE_DEBUG(0, "Device %s removed.", sc->sc_geom->name);
172 }
173
174 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
175 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
176 g_detach(cp);
177 g_destroy_consumer(cp);
178}
179
180static void
181g_stripe_orphan(struct g_consumer *cp)
182{
183 struct g_stripe_softc *sc;
184 struct g_geom *gp;
185
186 g_topology_assert();
187 gp = cp->geom;
188 sc = gp->softc;
189 if (sc == NULL)
190 return;
191
192 g_stripe_remove_disk(cp);
193 /* If there are no valid disks anymore, remove device. */
194 if (g_stripe_nvalid(sc) == 0)
195 g_stripe_destroy(sc, 1);
196}
197
198static int
199g_stripe_access(struct g_provider *pp, int dr, int dw, int de)
200{
201 struct g_consumer *cp1, *cp2;
202 struct g_stripe_softc *sc;
203 struct g_geom *gp;
204 int error;
205
206 gp = pp->geom;
207 sc = gp->softc;
208
209 if (sc == NULL) {
210 /*
211 * It looks like geom is being withered.
212 * In that case we allow only negative requests.
213 */
214 KASSERT(dr <= 0 && dw <= 0 && de <= 0,
215 ("Positive access request (device=%s).", pp->name));
216 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 &&
217 (pp->ace + de) == 0) {
218 G_STRIPE_DEBUG(0, "Device %s definitely destroyed.",
219 gp->name);
220 }
221 return (0);
222 }
223
224 /* On first open, grab an extra "exclusive" bit */
225 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
226 de++;
227 /* ... and let go of it on last close */
228 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0)
229 de--;
230
231 error = ENXIO;
232 LIST_FOREACH(cp1, &gp->consumer, consumer) {
233 error = g_access(cp1, dr, dw, de);
234 if (error == 0)
235 continue;
236 /*
237 * If we fail here, backout all previous changes.
238 */
239 LIST_FOREACH(cp2, &gp->consumer, consumer) {
240 if (cp1 == cp2)
241 return (error);
242 g_access(cp2, -dr, -dw, -de);
243 }
244 /* NOTREACHED */
245 }
246
247 return (error);
248}
249
250static void
251g_stripe_copy(struct g_stripe_softc *sc, char *src, char *dst, off_t offset,
252 off_t length, int mode)
253{
254 u_int stripesize;
255 size_t len;
256
257 stripesize = sc->sc_stripesize;
258 len = (size_t)(stripesize - (offset & (stripesize - 1)));
259 do {
260 bcopy(src, dst, len);
261 if (mode) {
262 dst += len + stripesize * (sc->sc_ndisks - 1);
263 src += len;
264 } else {
265 dst += len;
266 src += len + stripesize * (sc->sc_ndisks - 1);
267 }
268 length -= len;
269 KASSERT(length >= 0,
270 ("Length < 0 (stripesize=%zu, offset=%jd, length=%jd).",
271 (size_t)stripesize, (intmax_t)offset, (intmax_t)length));
272 if (length > stripesize)
273 len = stripesize;
274 else
275 len = length;
276 } while (length > 0);
277}
278
279static void
280g_stripe_done(struct bio *bp)
281{
282 struct g_stripe_softc *sc;
283 struct bio *pbp;
284
285 pbp = bp->bio_parent;
286 sc = pbp->bio_to->geom->softc;
287 if (pbp->bio_error == 0)
288 pbp->bio_error = bp->bio_error;
289 pbp->bio_completed += bp->bio_completed;
290 if (bp->bio_cmd == BIO_READ && bp->bio_driver1 != NULL) {
291 g_stripe_copy(sc, bp->bio_data, bp->bio_driver1, bp->bio_offset,
292 bp->bio_length, 1);
293 bp->bio_data = bp->bio_driver1;
294 bp->bio_driver1 = NULL;
295 }
296 g_destroy_bio(bp);
297 pbp->bio_inbed++;
298 if (pbp->bio_children == pbp->bio_inbed) {
299 if (pbp->bio_caller1 != NULL)
300 uma_zfree(g_stripe_zone, pbp->bio_caller1);
301 g_io_deliver(pbp, pbp->bio_error);
302 }
303}
304
305static int
306g_stripe_start_fast(struct bio *bp, u_int no, off_t offset, off_t length)
307{
308 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
309 u_int nparts = 0, stripesize;
310 struct g_stripe_softc *sc;
311 char *addr, *data = NULL;
312 struct bio *cbp;
313 int error;
314
315 sc = bp->bio_to->geom->softc;
316
317 addr = bp->bio_data;
318 stripesize = sc->sc_stripesize;
319
320 cbp = g_clone_bio(bp);
321 if (cbp == NULL) {
322 error = ENOMEM;
323 goto failure;
324 }
325 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
326 nparts++;
327 /*
328 * Fill in the component buf structure.
329 */
330 cbp->bio_done = g_stripe_done;
331 cbp->bio_offset = offset;
332 cbp->bio_data = addr;
333 cbp->bio_driver1 = NULL;
334 cbp->bio_length = length;
335 cbp->bio_driver2 = sc->sc_disks[no];
336
337 /* offset -= offset % stripesize; */
338 offset -= offset & (stripesize - 1);
339 addr += length;
340 length = bp->bio_length - length;
341 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
342 if (no > sc->sc_ndisks - 1) {
343 no = 0;
344 offset += stripesize;
345 }
346 if (nparts >= sc->sc_ndisks) {
347 cbp = TAILQ_NEXT(cbp, bio_queue);
348 if (cbp == NULL)
349 cbp = TAILQ_FIRST(&queue);
350 nparts++;
351 /*
352 * Update bio structure.
353 */
354 /*
355 * MIN() is in case when
356 * (bp->bio_length % sc->sc_stripesize) != 0.
357 */
358 cbp->bio_length += MIN(stripesize, length);
359 if (cbp->bio_driver1 == NULL) {
360 cbp->bio_driver1 = cbp->bio_data;
361 cbp->bio_data = NULL;
362 if (data == NULL) {
363 data = uma_zalloc(g_stripe_zone,
364 M_NOWAIT);
365 if (data == NULL) {
366 error = ENOMEM;
367 goto failure;
368 }
369 }
370 }
371 } else {
372 cbp = g_clone_bio(bp);
373 if (cbp == NULL) {
374 error = ENOMEM;
375 goto failure;
376 }
377 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
378 nparts++;
379 /*
380 * Fill in the component buf structure.
381 */
382 cbp->bio_done = g_stripe_done;
383 cbp->bio_offset = offset;
384 cbp->bio_data = addr;
385 cbp->bio_driver1 = NULL;
386 /*
387 * MIN() is in case when
388 * (bp->bio_length % sc->sc_stripesize) != 0.
389 */
390 cbp->bio_length = MIN(stripesize, length);
391 cbp->bio_driver2 = sc->sc_disks[no];
392 }
393 }
394 if (data != NULL)
395 bp->bio_caller1 = data;
396 /*
397 * Fire off all allocated requests!
398 */
399 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
400 struct g_consumer *cp;
401
402 TAILQ_REMOVE(&queue, cbp, bio_queue);
403 cp = cbp->bio_driver2;
404 cbp->bio_driver2 = NULL;
405 cbp->bio_to = cp->provider;
406 if (cbp->bio_driver1 != NULL) {
407 cbp->bio_data = data;
408 if (bp->bio_cmd == BIO_WRITE) {
409 g_stripe_copy(sc, cbp->bio_driver1, data,
410 cbp->bio_offset, cbp->bio_length, 0);
411 }
412 data += cbp->bio_length;
413 }
414 G_STRIPE_LOGREQ(cbp, "Sending request.");
415 g_io_request(cbp, cp);
416 }
417 return (0);
418failure:
419 if (data != NULL)
420 uma_zfree(g_stripe_zone, data);
421 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
422 TAILQ_REMOVE(&queue, cbp, bio_queue);
423 if (cbp->bio_driver1 != NULL) {
424 cbp->bio_data = cbp->bio_driver1;
425 cbp->bio_driver1 = NULL;
426 }
427 g_destroy_bio(cbp);
428 }
429 return (error);
430}
431
432static int
433g_stripe_start_economic(struct bio *bp, u_int no, off_t offset, off_t length)
434{
435 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
436 struct g_stripe_softc *sc;
437 uint32_t stripesize;
438 struct bio *cbp;
439 char *addr;
440 int error;
441
442 sc = bp->bio_to->geom->softc;
443
444 addr = bp->bio_data;
445 stripesize = sc->sc_stripesize;
446
447 cbp = g_clone_bio(bp);
448 if (cbp == NULL) {
449 error = ENOMEM;
450 goto failure;
451 }
452 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
453 /*
454 * Fill in the component buf structure.
455 */
456 cbp->bio_done = g_std_done;
457 cbp->bio_offset = offset;
458 cbp->bio_data = addr;
459 cbp->bio_length = length;
460 cbp->bio_driver2 = sc->sc_disks[no];
461
462 /* offset -= offset % stripesize; */
463 offset -= offset & (stripesize - 1);
464 addr += length;
465 length = bp->bio_length - length;
466 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
467 if (no > sc->sc_ndisks - 1) {
468 no = 0;
469 offset += stripesize;
470 }
471 cbp = g_clone_bio(bp);
472 if (cbp == NULL) {
473 error = ENOMEM;
474 goto failure;
475 }
476 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
477
478 /*
479 * Fill in the component buf structure.
480 */
481 cbp->bio_done = g_std_done;
482 cbp->bio_offset = offset;
483 cbp->bio_data = addr;
484 /*
485 * MIN() is in case when
486 * (bp->bio_length % sc->sc_stripesize) != 0.
487 */
488 cbp->bio_length = MIN(stripesize, length);
489
490 cbp->bio_driver2 = sc->sc_disks[no];
491 }
492 /*
493 * Fire off all allocated requests!
494 */
495 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
496 struct g_consumer *cp;
497
498 TAILQ_REMOVE(&queue, cbp, bio_queue);
499 cp = cbp->bio_driver2;
500 cbp->bio_driver2 = NULL;
501 cbp->bio_to = cp->provider;
502 G_STRIPE_LOGREQ(cbp, "Sending request.");
503 g_io_request(cbp, cp);
504 }
505 return (0);
506failure:
507 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
508 TAILQ_REMOVE(&queue, cbp, bio_queue);
509 g_destroy_bio(cbp);
510 }
511 return (error);
512}
513
514static void
515g_stripe_start(struct bio *bp)
516{
517 off_t offset, start, length, nstripe;
518 struct g_stripe_softc *sc;
519 u_int no, stripesize;
520 int error, fast = 0;
521
522 sc = bp->bio_to->geom->softc;
523 /*
524 * If sc == NULL, provider's error should be set and g_stripe_start()
525 * should not be called at all.
526 */
527 KASSERT(sc != NULL,
528 ("Provider's error should be set (error=%d)(device=%s).",
529 bp->bio_to->error, bp->bio_to->name));
530
531 G_STRIPE_LOGREQ(bp, "Request received.");
532
533 switch (bp->bio_cmd) {
534 case BIO_READ:
535 case BIO_WRITE:
536 case BIO_DELETE:
537 /*
538 * Only those requests are supported.
539 */
540 break;
541 case BIO_GETATTR:
542 /* To which provider it should be delivered? */
543 default:
544 g_io_deliver(bp, EOPNOTSUPP);
545 return;
546 }
547
548 stripesize = sc->sc_stripesize;
549
550 /*
551 * Calculations are quite messy, but fast I hope.
552 */
553
554 /* Stripe number. */
555 /* nstripe = bp->bio_offset / stripesize; */
556 nstripe = bp->bio_offset >> (off_t)sc->sc_stripebits;
557 /* Disk number. */
558 no = nstripe % sc->sc_ndisks;
559 /* Start position in stripe. */
560 /* start = bp->bio_offset % stripesize; */
561 start = bp->bio_offset & (stripesize - 1);
562 /* Start position in disk. */
563 /* offset = (nstripe / sc->sc_ndisks) * stripesize + start; */
564 offset = ((nstripe / sc->sc_ndisks) << sc->sc_stripebits) + start;
565 /* Length of data to operate. */
566 length = MIN(bp->bio_length, stripesize - start);
567
568 /*
569 * Do use "fast" mode when:
570 * 1. "Fast" mode is ON.
571 * and
572 * 2. Request size is less than or equal to MAX_IO_SIZE (128kB),
573 * which should always be true.
574 * and
575 * 3. Request size is bigger than stripesize * ndisks. If it isn't,
576 * there will be no need to send more than one I/O request to
577 * a provider, so there is nothing to optmize.
578 */
579 if (g_stripe_fast && bp->bio_length <= MAX_IO_SIZE &&
580 bp->bio_length >= stripesize * sc->sc_ndisks) {
581 fast = 1;
582 }
583 error = 0;
584 if (fast)
585 error = g_stripe_start_fast(bp, no, offset, length);
586 /*
587 * Do use "economic" when:
588 * 1. "Economic" mode is ON.
589 * or
590 * 2. "Fast" mode failed. It can only failed if there is no memory.
591 */
592 if (!fast || error != 0)
593 error = g_stripe_start_economic(bp, no, offset, length);
594 if (error != 0) {
595 if (bp->bio_error == 0)
596 bp->bio_error = error;
597 g_io_deliver(bp, bp->bio_error);
598 }
599}
600
601static void
602g_stripe_check_and_run(struct g_stripe_softc *sc)
603{
604 off_t mediasize, ms;
605 u_int no, sectorsize = 0;
606
607 if (g_stripe_nvalid(sc) != sc->sc_ndisks)
608 return;
609
610 sc->sc_provider = g_new_providerf(sc->sc_geom, "%s", sc->sc_geom->name);
611 /*
612 * Find the smallest disk.
613 */
614 mediasize = sc->sc_disks[0]->provider->mediasize;
615 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
616 mediasize -= sc->sc_disks[0]->provider->sectorsize;
617 mediasize -= mediasize % sc->sc_stripesize;
618 sectorsize = sc->sc_disks[0]->provider->sectorsize;
619 for (no = 1; no < sc->sc_ndisks; no++) {
620 ms = sc->sc_disks[no]->provider->mediasize;
621 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
622 ms -= sc->sc_disks[no]->provider->sectorsize;
623 ms -= ms % sc->sc_stripesize;
624 if (ms < mediasize)
625 mediasize = ms;
626 sectorsize = lcm(sectorsize,
627 sc->sc_disks[no]->provider->sectorsize);
628 }
629 sc->sc_provider->sectorsize = sectorsize;
630 sc->sc_provider->mediasize = mediasize * sc->sc_ndisks;
631 g_error_provider(sc->sc_provider, 0);
632
633 G_STRIPE_DEBUG(0, "Device %s activated.", sc->sc_geom->name);
634}
635
636static int
637g_stripe_read_metadata(struct g_consumer *cp, struct g_stripe_metadata *md)
638{
639 struct g_provider *pp;
640 u_char *buf;
641 int error;
642
643 g_topology_assert();
644
645 error = g_access(cp, 1, 0, 0);
646 if (error != 0)
647 return (error);
648 pp = cp->provider;
649 g_topology_unlock();
650 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
651 &error);
652 g_topology_lock();
653 g_access(cp, -1, 0, 0);
654 if (buf == NULL)
655 return (error);
656
657 /* Decode metadata. */
658 stripe_metadata_decode(buf, md);
659 g_free(buf);
660
661 return (0);
662}
663
664/*
665 * Add disk to given device.
666 */
667static int
668g_stripe_add_disk(struct g_stripe_softc *sc, struct g_provider *pp, u_int no)
669{
670 struct g_consumer *cp, *fcp;
671 struct g_geom *gp;
672 int error;
673
674 /* Metadata corrupted? */
675 if (no >= sc->sc_ndisks)
676 return (EINVAL);
677
678 /* Check if disk is not already attached. */
679 if (sc->sc_disks[no] != NULL)
680 return (EEXIST);
681
682 gp = sc->sc_geom;
683 fcp = LIST_FIRST(&gp->consumer);
684
685 cp = g_new_consumer(gp);
686 error = g_attach(cp, pp);
687 if (error != 0) {
688 g_destroy_consumer(cp);
689 return (error);
690 }
691
692 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) {
693 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
694 if (error != 0) {
695 g_detach(cp);
696 g_destroy_consumer(cp);
697 return (error);
698 }
699 }
700 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) {
701 struct g_stripe_metadata md;
702
703 /* Reread metadata. */
704 error = g_stripe_read_metadata(cp, &md);
705 if (error != 0)
706 goto fail;
707
708 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0 ||
709 strcmp(md.md_name, sc->sc_name) != 0 ||
710 md.md_id != sc->sc_id) {
711 G_STRIPE_DEBUG(0, "Metadata on %s changed.", pp->name);
712 goto fail;
713 }
714 }
715
716 cp->private = sc;
717 cp->index = no;
718 sc->sc_disks[no] = cp;
719
720 G_STRIPE_DEBUG(0, "Disk %s attached to %s.", pp->name, gp->name);
721
722 g_stripe_check_and_run(sc);
723
724 return (0);
725fail:
726 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0))
727 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace);
728 g_detach(cp);
729 g_destroy_consumer(cp);
730 return (error);
731}
732
733static struct g_geom *
734g_stripe_create(struct g_class *mp, const struct g_stripe_metadata *md,
735 u_int type)
736{
737 struct g_stripe_softc *sc;
738 struct g_geom *gp;
739 u_int no;
740
741 G_STRIPE_DEBUG(1, "Creating device %s.stripe (id=%u).", md->md_name,
742 md->md_id);
743
744 /* Two disks is minimum. */
745 if (md->md_all <= 1) {
746 G_STRIPE_DEBUG(0, "Too few disks defined for %s.stripe.",
747 md->md_name);
748 return (NULL);
749 }
750#if 0
751 /* Stripe size have to be grater than or equal to sector size. */
752 if (md->md_stripesize < sectorsize) {
753 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.stripe.",
754 md->md_name);
755 return (NULL);
756 }
757#endif
758 /* Stripe size have to be power of 2. */
759 if (!powerof2(md->md_stripesize)) {
760 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.stripe.",
761 md->md_name);
762 return (NULL);
763 }
764
765 /* Check for duplicate unit */
766 LIST_FOREACH(gp, &mp->geom, geom) {
767 sc = gp->softc;
768 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) {
769 G_STRIPE_DEBUG(0, "Device %s already configured.",
770 gp->name);
771 return (NULL);
772 }
773 }
774 gp = g_new_geomf(mp, "%s.stripe", md->md_name);
775 gp->softc = NULL; /* for a moment */
776
777 sc = malloc(sizeof(*sc), M_STRIPE, M_NOWAIT | M_ZERO);
778 if (sc == NULL) {
779 G_STRIPE_DEBUG(0, "Can't allocate memory for device %s.",
780 gp->name);
781 g_destroy_geom(gp);
782 return (NULL);
783 }
784
785 gp->start = g_stripe_start;
786 gp->spoiled = g_stripe_orphan;
787 gp->orphan = g_stripe_orphan;
788 gp->access = g_stripe_access;
789 gp->dumpconf = g_stripe_dumpconf;
790
791 strlcpy(sc->sc_name, md->md_name, sizeof(sc->sc_name));
792 sc->sc_id = md->md_id;
793 sc->sc_stripesize = md->md_stripesize;
794 sc->sc_stripebits = BITCOUNT(sc->sc_stripesize - 1);
795 sc->sc_ndisks = md->md_all;
796 sc->sc_disks = malloc(sizeof(struct g_consumer *) * sc->sc_ndisks,
797 M_STRIPE, M_WAITOK | M_ZERO);
798 for (no = 0; no < sc->sc_ndisks; no++)
799 sc->sc_disks[no] = NULL;
800 sc->sc_type = type;
801
802 gp->softc = sc;
803 sc->sc_geom = gp;
804 sc->sc_provider = NULL;
805
806 G_STRIPE_DEBUG(0, "Device %s created (id=%u).", gp->name, sc->sc_id);
807
808 return (gp);
809}
810
811static int
812g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force)
813{
814 struct g_provider *pp;
815 struct g_geom *gp;
816 u_int no;
817
818 g_topology_assert();
819
820 if (sc == NULL)
821 return (ENXIO);
822
823 pp = sc->sc_provider;
824 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
825 if (force) {
826 G_STRIPE_DEBUG(0, "Device %s is still open, so it "
827 "can't be definitely removed.", pp->name);
828 } else {
829 G_STRIPE_DEBUG(1,
830 "Device %s is still open (r%dw%de%d).", pp->name,
831 pp->acr, pp->acw, pp->ace);
832 return (EBUSY);
833 }
834 }
835
836 for (no = 0; no < sc->sc_ndisks; no++) {
837 if (sc->sc_disks[no] != NULL)
838 g_stripe_remove_disk(sc->sc_disks[no]);
839 }
840
841 gp = sc->sc_geom;
842 gp->softc = NULL;
843 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)",
844 gp->name));
845 free(sc->sc_disks, M_STRIPE);
846 free(sc, M_STRIPE);
847
848 pp = LIST_FIRST(&gp->provider);
849 if (pp == NULL || (pp->acr == 0 && pp->acw == 0 && pp->ace == 0))
850 G_STRIPE_DEBUG(0, "Device %s destroyed.", gp->name);
851
852 g_wither_geom(gp, ENXIO);
853
854 return (0);
855}
856
857static int
858g_stripe_destroy_geom(struct gctl_req *req __unused,
859 struct g_class *mp __unused, struct g_geom *gp)
860{
861 struct g_stripe_softc *sc;
862
863 sc = gp->softc;
864 return (g_stripe_destroy(sc, 0));
865}
866
867static struct g_geom *
868g_stripe_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
869{
870 struct g_stripe_metadata md;
871 struct g_stripe_softc *sc;
872 struct g_consumer *cp;
873 struct g_geom *gp;
874 int error;
875
876 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
877 g_topology_assert();
878
879 G_STRIPE_DEBUG(3, "Tasting %s.", pp->name);
880
881 gp = g_new_geomf(mp, "stripe:taste");
882 gp->start = g_stripe_start;
883 gp->access = g_stripe_access;
884 gp->orphan = g_stripe_orphan;
885 cp = g_new_consumer(gp);
886 g_attach(cp, pp);
887
888 error = g_stripe_read_metadata(cp, &md);
889 g_wither_geom(gp, ENXIO);
890 if (error != 0)
891 return (NULL);
892 gp = NULL;
893
894 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0)
895 return (NULL);
896 if (md.md_version > G_STRIPE_VERSION) {
897 printf("geom_stripe.ko module is too old to handle %s.\n",
898 pp->name);
899 return (NULL);
900 }
901
902 /*
903 * Let's check if device already exists.
904 */
905 sc = NULL;
906 LIST_FOREACH(gp, &mp->geom, geom) {
907 sc = gp->softc;
908 if (sc == NULL)
909 continue;
910 if (sc->sc_type != G_STRIPE_TYPE_AUTOMATIC)
911 continue;
912 if (strcmp(md.md_name, sc->sc_name) != 0)
913 continue;
914 if (md.md_id != sc->sc_id)
915 continue;
916 break;
917 }
918 if (gp != NULL) {
919 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
920 error = g_stripe_add_disk(sc, pp, md.md_no);
921 if (error != 0) {
922 G_STRIPE_DEBUG(0,
923 "Cannot add disk %s to %s (error=%d).", pp->name,
924 gp->name, error);
925 return (NULL);
926 }
927 } else {
928 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_AUTOMATIC);
929 if (gp == NULL) {
930 G_STRIPE_DEBUG(0, "Cannot create device %s.stripe.",
931 md.md_name);
932 return (NULL);
933 }
934 sc = gp->softc;
935 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
936 error = g_stripe_add_disk(sc, pp, md.md_no);
937 if (error != 0) {
938 G_STRIPE_DEBUG(0,
939 "Cannot add disk %s to %s (error=%d).", pp->name,
940 gp->name, error);
941 g_stripe_destroy(sc, 1);
942 return (NULL);
943 }
944 }
945
946 return (gp);
947}
948
949static void
950g_stripe_ctl_create(struct gctl_req *req, struct g_class *mp)
951{
952 u_int attached, no;
953 struct g_stripe_metadata md;
954 struct g_provider *pp;
955 struct g_stripe_softc *sc;
956 struct g_geom *gp;
957 struct sbuf *sb;
958 intmax_t *stripesize;
959 const char *name;
960 char param[16];
961 int *nargs;
962
963 g_topology_assert();
964 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
965 if (nargs == NULL) {
966 gctl_error(req, "No '%s' argument.", "nargs");
967 return;
968 }
969 if (*nargs <= 2) {
970 gctl_error(req, "Too few arguments.");
971 return;
972 }
973
974 strlcpy(md.md_magic, G_STRIPE_MAGIC, sizeof(md.md_magic));
975 md.md_version = G_STRIPE_VERSION;
976 name = gctl_get_asciiparam(req, "arg0");
977 if (name == NULL) {
978 gctl_error(req, "No 'arg%u' argument.", 0);
979 return;
980 }
981 strlcpy(md.md_name, name, sizeof(md.md_name));
982 md.md_id = arc4random();
983 md.md_no = 0;
984 md.md_all = *nargs - 1;
985 stripesize = gctl_get_paraml(req, "stripesize", sizeof(*stripesize));
986 if (stripesize == NULL) {
987 gctl_error(req, "No '%s' argument.", "stripesize");
988 return;
989 }
990 md.md_stripesize = *stripesize;
991
992 /* Check all providers are valid */
993 for (no = 1; no < *nargs; no++) {
994 snprintf(param, sizeof(param), "arg%u", no);
995 name = gctl_get_asciiparam(req, param);
996 if (name == NULL) {
997 gctl_error(req, "No 'arg%u' argument.", no);
998 return;
999 }
1000 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1001 name += strlen("/dev/");
1002 pp = g_provider_by_name(name);
1003 if (pp == NULL) {
1004 G_STRIPE_DEBUG(1, "Disk %s is invalid.", name);
1005 gctl_error(req, "Disk %s is invalid.", name);
1006 return;
1007 }
1008 }
1009
1010 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_MANUAL);
1011 if (gp == NULL) {
1012 gctl_error(req, "Can't configure %s.stripe.", md.md_name);
1013 return;
1014 }
1015
1016 sc = gp->softc;
1017 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1018 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name);
1019 for (attached = 0, no = 1; no < *nargs; no++) {
1020 snprintf(param, sizeof(param), "arg%u", no);
1021 name = gctl_get_asciiparam(req, param);
1022 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1023 name += strlen("/dev/");
1024 pp = g_provider_by_name(name);
1025 KASSERT(pp != NULL, ("Provider %s disappear?!", name));
1026 if (g_stripe_add_disk(sc, pp, no - 1) != 0) {
1027 G_STRIPE_DEBUG(1, "Disk %u (%s) not attached to %s.",
1028 no, pp->name, gp->name);
1029 sbuf_printf(sb, " %s", pp->name);
1030 continue;
1031 }
1032 attached++;
1033 }
1034 sbuf_finish(sb);
1035 if (md.md_all != attached) {
1036 g_stripe_destroy(gp->softc, 1);
1037 gctl_error(req, "%s", sbuf_data(sb));
1038 }
1039 sbuf_delete(sb);
1040}
1041
1042static struct g_stripe_softc *
1043g_stripe_find_device(struct g_class *mp, const char *name)
1044{
1045 struct g_stripe_softc *sc;
1046 struct g_geom *gp;
1047
1048 LIST_FOREACH(gp, &mp->geom, geom) {
1049 sc = gp->softc;
1050 if (sc == NULL)
1051 continue;
1052 if (strcmp(gp->name, name) == 0 ||
1053 strcmp(sc->sc_name, name) == 0) {
1054 return (sc);
1055 }
1056 }
1057 return (NULL);
1058}
1059
1060static void
1061g_stripe_ctl_destroy(struct gctl_req *req, struct g_class *mp)
1062{
1063 struct g_stripe_softc *sc;
1064 int *force, *nargs, error;
1065 const char *name;
1066 char param[16];
1067 u_int i;
1068
1069 g_topology_assert();
1070
1071 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1072 if (nargs == NULL) {
1073 gctl_error(req, "No '%s' argument.", "nargs");
1074 return;
1075 }
1076 if (*nargs <= 0) {
1077 gctl_error(req, "Missing device(s).");
1078 return;
1079 }
1080 force = gctl_get_paraml(req, "force", sizeof(*force));
1081 if (force == NULL) {
1082 gctl_error(req, "No '%s' argument.", "force");
1083 return;
1084 }
1085
1086 for (i = 0; i < (u_int)*nargs; i++) {
1087 snprintf(param, sizeof(param), "arg%u", i);
1088 name = gctl_get_asciiparam(req, param);
1089 if (name == NULL) {
1090 gctl_error(req, "No 'arg%u' argument.", i);
1091 return;
1092 }
1093 sc = g_stripe_find_device(mp, name);
1094 if (sc == NULL) {
1095 gctl_error(req, "No such device: %s.", name);
1096 return;
1097 }
1098 error = g_stripe_destroy(sc, *force);
1099 if (error != 0) {
1100 gctl_error(req, "Cannot destroy device %s (error=%d).",
1101 sc->sc_geom->name, error);
1102 return;
1103 }
1104 }
1105}
1106
1107static void
1108g_stripe_config(struct gctl_req *req, struct g_class *mp, const char *verb)
1109{
1110 uint32_t *version;
1111
1112 g_topology_assert();
1113
1114 version = gctl_get_paraml(req, "version", sizeof(*version));
1115 if (version == NULL) {
1116 gctl_error(req, "No '%s' argument.", "version");
1117 return;
1118 }
1119 if (*version != G_STRIPE_VERSION) {
1120 gctl_error(req, "Userland and kernel parts are out of sync.");
1121 return;
1122 }
1123
1124 if (strcmp(verb, "create") == 0) {
1125 g_stripe_ctl_create(req, mp);
1126 return;
1127 } else if (strcmp(verb, "destroy") == 0 ||
1128 strcmp(verb, "stop") == 0) {
1129 g_stripe_ctl_destroy(req, mp);
1130 return;
1131 }
1132
1133 gctl_error(req, "Unknown verb.");
1134}
1135
1136static void
1137g_stripe_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1138 struct g_consumer *cp, struct g_provider *pp)
1139{
1140 struct g_stripe_softc *sc;
1141
1142 sc = gp->softc;
1143 if (sc == NULL || pp == NULL)
1144 return;
1145 sbuf_printf(sb, "%s<id>%u</id>\n", indent, (u_int)sc->sc_id);
1146 sbuf_printf(sb, "%s<stripesize>%u</stripesize>\n", indent,
1147 (u_int)sc->sc_stripesize);
1148 switch (sc->sc_type) {
1149 case G_STRIPE_TYPE_AUTOMATIC:
1150 sbuf_printf(sb, "%s<type>%s</type>\n", indent, "automatic");
1151 break;
1152 case G_STRIPE_TYPE_MANUAL:
1153 sbuf_printf(sb, "%s<type>%s</type>\n", indent, "manual");
1154 break;
1155 default:
1156 sbuf_printf(sb, "%s<type>%s</type>\n", indent, "unknown");
1157 break;
1158 }
1159 sbuf_printf(sb, "%s<providers>", indent);
1160 LIST_FOREACH(cp, &gp->consumer, consumer) {
1161 if (cp->provider == NULL)
1162 continue;
1163 sbuf_printf(sb, "%s", cp->provider->name);
1164 if (LIST_NEXT(cp, consumer) != NULL)
1165 sbuf_printf(sb, " ");
1166 }
1167 sbuf_printf(sb, "</providers>\n");
1168 sbuf_printf(sb, "%s<status>total=%u, online=%u</status>\n", indent,
1169 sc->sc_ndisks, g_stripe_nvalid(sc));
1170 if (pp->error == 0)
1171 sbuf_printf(sb, "%s<state>UP</state>\n", indent);
1172 else
1173 sbuf_printf(sb, "%s<state>DOWN</state>\n", indent);
1174}
1175
1176DECLARE_GEOM_CLASS(g_stripe_class, g_stripe);
2 * Copyright (c) 2003 Pawel Jakub Dawidek <pjd@FreeBSD.org>
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 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/geom/stripe/g_stripe.c 131878 2004-07-09 14:30:09Z pjd $");
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/kernel.h>
33#include <sys/module.h>
34#include <sys/lock.h>
35#include <sys/mutex.h>
36#include <sys/bio.h>
37#include <sys/sysctl.h>
38#include <sys/malloc.h>
39#include <vm/uma.h>
40#include <geom/geom.h>
41#include <geom/stripe/g_stripe.h>
42
43
44#define MAX_IO_SIZE (DFLTPHYS * 2)
45static MALLOC_DEFINE(M_STRIPE, "stripe data", "GEOM_STRIPE Data");
46
47static uma_zone_t g_stripe_zone;
48
49static int g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force);
50static int g_stripe_destroy_geom(struct gctl_req *req, struct g_class *mp,
51 struct g_geom *gp);
52
53static g_taste_t g_stripe_taste;
54static g_ctl_req_t g_stripe_config;
55static g_dumpconf_t g_stripe_dumpconf;
56static g_init_t g_stripe_init;
57static g_fini_t g_stripe_fini;
58
59struct g_class g_stripe_class = {
60 .name = G_STRIPE_CLASS_NAME,
61 .ctlreq = g_stripe_config,
62 .taste = g_stripe_taste,
63 .destroy_geom = g_stripe_destroy_geom,
64 .init = g_stripe_init,
65 .fini = g_stripe_fini
66};
67
68SYSCTL_DECL(_kern_geom);
69SYSCTL_NODE(_kern_geom, OID_AUTO, stripe, CTLFLAG_RW, 0, "GEOM_STRIPE stuff");
70static u_int g_stripe_debug = 0;
71SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, debug, CTLFLAG_RW, &g_stripe_debug, 0,
72 "Debug level");
73static int g_stripe_fast = 1;
74TUNABLE_INT("kern.geom.stripe.fast", &g_stripe_fast);
75static int
76g_sysctl_stripe_fast(SYSCTL_HANDLER_ARGS)
77{
78 int error, fast;
79
80 fast = g_stripe_fast;
81 error = sysctl_handle_int(oidp, &fast, sizeof(fast), req);
82 if (error == 0 && req->newptr != NULL)
83 g_stripe_fast = fast;
84 return (error);
85}
86SYSCTL_PROC(_kern_geom_stripe, OID_AUTO, fast, CTLTYPE_INT | CTLFLAG_RW,
87 NULL, 0, g_sysctl_stripe_fast, "I", "Fast, but memory-consuming mode");
88static u_int g_stripe_maxmem = MAX_IO_SIZE * 10;
89TUNABLE_INT("kern.geom.stripe.maxmem", &g_stripe_maxmem);
90SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, maxmem, CTLFLAG_RD, &g_stripe_maxmem,
91 0, "Maximum memory that could be allocated in \"fast\" mode (in bytes)");
92
93/*
94 * Greatest Common Divisor.
95 */
96static u_int
97gcd(u_int a, u_int b)
98{
99 u_int c;
100
101 while (b != 0) {
102 c = a;
103 a = b;
104 b = (c % b);
105 }
106 return (a);
107}
108
109/*
110 * Least Common Multiple.
111 */
112static u_int
113lcm(u_int a, u_int b)
114{
115
116 return ((a * b) / gcd(a, b));
117}
118
119static void
120g_stripe_init(struct g_class *mp __unused)
121{
122
123 g_stripe_zone = uma_zcreate("g_stripe_zone", MAX_IO_SIZE, NULL, NULL,
124 NULL, NULL, 0, 0);
125 g_stripe_maxmem -= g_stripe_maxmem % MAX_IO_SIZE;
126 uma_zone_set_max(g_stripe_zone, g_stripe_maxmem / MAX_IO_SIZE);
127}
128
129static void
130g_stripe_fini(struct g_class *mp __unused)
131{
132
133 uma_zdestroy(g_stripe_zone);
134}
135
136/*
137 * Return the number of valid disks.
138 */
139static u_int
140g_stripe_nvalid(struct g_stripe_softc *sc)
141{
142 u_int i, no;
143
144 no = 0;
145 for (i = 0; i < sc->sc_ndisks; i++) {
146 if (sc->sc_disks[i] != NULL)
147 no++;
148 }
149
150 return (no);
151}
152
153static void
154g_stripe_remove_disk(struct g_consumer *cp)
155{
156 struct g_stripe_softc *sc;
157 u_int no;
158
159 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__));
160 sc = (struct g_stripe_softc *)cp->private;
161 KASSERT(sc != NULL, ("NULL sc in %s.", __func__));
162 no = cp->index;
163
164 G_STRIPE_DEBUG(0, "Disk %s removed from %s.", cp->provider->name,
165 sc->sc_geom->name);
166
167 sc->sc_disks[no] = NULL;
168 if (sc->sc_provider != NULL) {
169 g_orphan_provider(sc->sc_provider, ENXIO);
170 sc->sc_provider = NULL;
171 G_STRIPE_DEBUG(0, "Device %s removed.", sc->sc_geom->name);
172 }
173
174 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
175 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
176 g_detach(cp);
177 g_destroy_consumer(cp);
178}
179
180static void
181g_stripe_orphan(struct g_consumer *cp)
182{
183 struct g_stripe_softc *sc;
184 struct g_geom *gp;
185
186 g_topology_assert();
187 gp = cp->geom;
188 sc = gp->softc;
189 if (sc == NULL)
190 return;
191
192 g_stripe_remove_disk(cp);
193 /* If there are no valid disks anymore, remove device. */
194 if (g_stripe_nvalid(sc) == 0)
195 g_stripe_destroy(sc, 1);
196}
197
198static int
199g_stripe_access(struct g_provider *pp, int dr, int dw, int de)
200{
201 struct g_consumer *cp1, *cp2;
202 struct g_stripe_softc *sc;
203 struct g_geom *gp;
204 int error;
205
206 gp = pp->geom;
207 sc = gp->softc;
208
209 if (sc == NULL) {
210 /*
211 * It looks like geom is being withered.
212 * In that case we allow only negative requests.
213 */
214 KASSERT(dr <= 0 && dw <= 0 && de <= 0,
215 ("Positive access request (device=%s).", pp->name));
216 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 &&
217 (pp->ace + de) == 0) {
218 G_STRIPE_DEBUG(0, "Device %s definitely destroyed.",
219 gp->name);
220 }
221 return (0);
222 }
223
224 /* On first open, grab an extra "exclusive" bit */
225 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
226 de++;
227 /* ... and let go of it on last close */
228 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0)
229 de--;
230
231 error = ENXIO;
232 LIST_FOREACH(cp1, &gp->consumer, consumer) {
233 error = g_access(cp1, dr, dw, de);
234 if (error == 0)
235 continue;
236 /*
237 * If we fail here, backout all previous changes.
238 */
239 LIST_FOREACH(cp2, &gp->consumer, consumer) {
240 if (cp1 == cp2)
241 return (error);
242 g_access(cp2, -dr, -dw, -de);
243 }
244 /* NOTREACHED */
245 }
246
247 return (error);
248}
249
250static void
251g_stripe_copy(struct g_stripe_softc *sc, char *src, char *dst, off_t offset,
252 off_t length, int mode)
253{
254 u_int stripesize;
255 size_t len;
256
257 stripesize = sc->sc_stripesize;
258 len = (size_t)(stripesize - (offset & (stripesize - 1)));
259 do {
260 bcopy(src, dst, len);
261 if (mode) {
262 dst += len + stripesize * (sc->sc_ndisks - 1);
263 src += len;
264 } else {
265 dst += len;
266 src += len + stripesize * (sc->sc_ndisks - 1);
267 }
268 length -= len;
269 KASSERT(length >= 0,
270 ("Length < 0 (stripesize=%zu, offset=%jd, length=%jd).",
271 (size_t)stripesize, (intmax_t)offset, (intmax_t)length));
272 if (length > stripesize)
273 len = stripesize;
274 else
275 len = length;
276 } while (length > 0);
277}
278
279static void
280g_stripe_done(struct bio *bp)
281{
282 struct g_stripe_softc *sc;
283 struct bio *pbp;
284
285 pbp = bp->bio_parent;
286 sc = pbp->bio_to->geom->softc;
287 if (pbp->bio_error == 0)
288 pbp->bio_error = bp->bio_error;
289 pbp->bio_completed += bp->bio_completed;
290 if (bp->bio_cmd == BIO_READ && bp->bio_driver1 != NULL) {
291 g_stripe_copy(sc, bp->bio_data, bp->bio_driver1, bp->bio_offset,
292 bp->bio_length, 1);
293 bp->bio_data = bp->bio_driver1;
294 bp->bio_driver1 = NULL;
295 }
296 g_destroy_bio(bp);
297 pbp->bio_inbed++;
298 if (pbp->bio_children == pbp->bio_inbed) {
299 if (pbp->bio_caller1 != NULL)
300 uma_zfree(g_stripe_zone, pbp->bio_caller1);
301 g_io_deliver(pbp, pbp->bio_error);
302 }
303}
304
305static int
306g_stripe_start_fast(struct bio *bp, u_int no, off_t offset, off_t length)
307{
308 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
309 u_int nparts = 0, stripesize;
310 struct g_stripe_softc *sc;
311 char *addr, *data = NULL;
312 struct bio *cbp;
313 int error;
314
315 sc = bp->bio_to->geom->softc;
316
317 addr = bp->bio_data;
318 stripesize = sc->sc_stripesize;
319
320 cbp = g_clone_bio(bp);
321 if (cbp == NULL) {
322 error = ENOMEM;
323 goto failure;
324 }
325 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
326 nparts++;
327 /*
328 * Fill in the component buf structure.
329 */
330 cbp->bio_done = g_stripe_done;
331 cbp->bio_offset = offset;
332 cbp->bio_data = addr;
333 cbp->bio_driver1 = NULL;
334 cbp->bio_length = length;
335 cbp->bio_driver2 = sc->sc_disks[no];
336
337 /* offset -= offset % stripesize; */
338 offset -= offset & (stripesize - 1);
339 addr += length;
340 length = bp->bio_length - length;
341 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
342 if (no > sc->sc_ndisks - 1) {
343 no = 0;
344 offset += stripesize;
345 }
346 if (nparts >= sc->sc_ndisks) {
347 cbp = TAILQ_NEXT(cbp, bio_queue);
348 if (cbp == NULL)
349 cbp = TAILQ_FIRST(&queue);
350 nparts++;
351 /*
352 * Update bio structure.
353 */
354 /*
355 * MIN() is in case when
356 * (bp->bio_length % sc->sc_stripesize) != 0.
357 */
358 cbp->bio_length += MIN(stripesize, length);
359 if (cbp->bio_driver1 == NULL) {
360 cbp->bio_driver1 = cbp->bio_data;
361 cbp->bio_data = NULL;
362 if (data == NULL) {
363 data = uma_zalloc(g_stripe_zone,
364 M_NOWAIT);
365 if (data == NULL) {
366 error = ENOMEM;
367 goto failure;
368 }
369 }
370 }
371 } else {
372 cbp = g_clone_bio(bp);
373 if (cbp == NULL) {
374 error = ENOMEM;
375 goto failure;
376 }
377 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
378 nparts++;
379 /*
380 * Fill in the component buf structure.
381 */
382 cbp->bio_done = g_stripe_done;
383 cbp->bio_offset = offset;
384 cbp->bio_data = addr;
385 cbp->bio_driver1 = NULL;
386 /*
387 * MIN() is in case when
388 * (bp->bio_length % sc->sc_stripesize) != 0.
389 */
390 cbp->bio_length = MIN(stripesize, length);
391 cbp->bio_driver2 = sc->sc_disks[no];
392 }
393 }
394 if (data != NULL)
395 bp->bio_caller1 = data;
396 /*
397 * Fire off all allocated requests!
398 */
399 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
400 struct g_consumer *cp;
401
402 TAILQ_REMOVE(&queue, cbp, bio_queue);
403 cp = cbp->bio_driver2;
404 cbp->bio_driver2 = NULL;
405 cbp->bio_to = cp->provider;
406 if (cbp->bio_driver1 != NULL) {
407 cbp->bio_data = data;
408 if (bp->bio_cmd == BIO_WRITE) {
409 g_stripe_copy(sc, cbp->bio_driver1, data,
410 cbp->bio_offset, cbp->bio_length, 0);
411 }
412 data += cbp->bio_length;
413 }
414 G_STRIPE_LOGREQ(cbp, "Sending request.");
415 g_io_request(cbp, cp);
416 }
417 return (0);
418failure:
419 if (data != NULL)
420 uma_zfree(g_stripe_zone, data);
421 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
422 TAILQ_REMOVE(&queue, cbp, bio_queue);
423 if (cbp->bio_driver1 != NULL) {
424 cbp->bio_data = cbp->bio_driver1;
425 cbp->bio_driver1 = NULL;
426 }
427 g_destroy_bio(cbp);
428 }
429 return (error);
430}
431
432static int
433g_stripe_start_economic(struct bio *bp, u_int no, off_t offset, off_t length)
434{
435 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
436 struct g_stripe_softc *sc;
437 uint32_t stripesize;
438 struct bio *cbp;
439 char *addr;
440 int error;
441
442 sc = bp->bio_to->geom->softc;
443
444 addr = bp->bio_data;
445 stripesize = sc->sc_stripesize;
446
447 cbp = g_clone_bio(bp);
448 if (cbp == NULL) {
449 error = ENOMEM;
450 goto failure;
451 }
452 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
453 /*
454 * Fill in the component buf structure.
455 */
456 cbp->bio_done = g_std_done;
457 cbp->bio_offset = offset;
458 cbp->bio_data = addr;
459 cbp->bio_length = length;
460 cbp->bio_driver2 = sc->sc_disks[no];
461
462 /* offset -= offset % stripesize; */
463 offset -= offset & (stripesize - 1);
464 addr += length;
465 length = bp->bio_length - length;
466 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
467 if (no > sc->sc_ndisks - 1) {
468 no = 0;
469 offset += stripesize;
470 }
471 cbp = g_clone_bio(bp);
472 if (cbp == NULL) {
473 error = ENOMEM;
474 goto failure;
475 }
476 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
477
478 /*
479 * Fill in the component buf structure.
480 */
481 cbp->bio_done = g_std_done;
482 cbp->bio_offset = offset;
483 cbp->bio_data = addr;
484 /*
485 * MIN() is in case when
486 * (bp->bio_length % sc->sc_stripesize) != 0.
487 */
488 cbp->bio_length = MIN(stripesize, length);
489
490 cbp->bio_driver2 = sc->sc_disks[no];
491 }
492 /*
493 * Fire off all allocated requests!
494 */
495 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
496 struct g_consumer *cp;
497
498 TAILQ_REMOVE(&queue, cbp, bio_queue);
499 cp = cbp->bio_driver2;
500 cbp->bio_driver2 = NULL;
501 cbp->bio_to = cp->provider;
502 G_STRIPE_LOGREQ(cbp, "Sending request.");
503 g_io_request(cbp, cp);
504 }
505 return (0);
506failure:
507 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
508 TAILQ_REMOVE(&queue, cbp, bio_queue);
509 g_destroy_bio(cbp);
510 }
511 return (error);
512}
513
514static void
515g_stripe_start(struct bio *bp)
516{
517 off_t offset, start, length, nstripe;
518 struct g_stripe_softc *sc;
519 u_int no, stripesize;
520 int error, fast = 0;
521
522 sc = bp->bio_to->geom->softc;
523 /*
524 * If sc == NULL, provider's error should be set and g_stripe_start()
525 * should not be called at all.
526 */
527 KASSERT(sc != NULL,
528 ("Provider's error should be set (error=%d)(device=%s).",
529 bp->bio_to->error, bp->bio_to->name));
530
531 G_STRIPE_LOGREQ(bp, "Request received.");
532
533 switch (bp->bio_cmd) {
534 case BIO_READ:
535 case BIO_WRITE:
536 case BIO_DELETE:
537 /*
538 * Only those requests are supported.
539 */
540 break;
541 case BIO_GETATTR:
542 /* To which provider it should be delivered? */
543 default:
544 g_io_deliver(bp, EOPNOTSUPP);
545 return;
546 }
547
548 stripesize = sc->sc_stripesize;
549
550 /*
551 * Calculations are quite messy, but fast I hope.
552 */
553
554 /* Stripe number. */
555 /* nstripe = bp->bio_offset / stripesize; */
556 nstripe = bp->bio_offset >> (off_t)sc->sc_stripebits;
557 /* Disk number. */
558 no = nstripe % sc->sc_ndisks;
559 /* Start position in stripe. */
560 /* start = bp->bio_offset % stripesize; */
561 start = bp->bio_offset & (stripesize - 1);
562 /* Start position in disk. */
563 /* offset = (nstripe / sc->sc_ndisks) * stripesize + start; */
564 offset = ((nstripe / sc->sc_ndisks) << sc->sc_stripebits) + start;
565 /* Length of data to operate. */
566 length = MIN(bp->bio_length, stripesize - start);
567
568 /*
569 * Do use "fast" mode when:
570 * 1. "Fast" mode is ON.
571 * and
572 * 2. Request size is less than or equal to MAX_IO_SIZE (128kB),
573 * which should always be true.
574 * and
575 * 3. Request size is bigger than stripesize * ndisks. If it isn't,
576 * there will be no need to send more than one I/O request to
577 * a provider, so there is nothing to optmize.
578 */
579 if (g_stripe_fast && bp->bio_length <= MAX_IO_SIZE &&
580 bp->bio_length >= stripesize * sc->sc_ndisks) {
581 fast = 1;
582 }
583 error = 0;
584 if (fast)
585 error = g_stripe_start_fast(bp, no, offset, length);
586 /*
587 * Do use "economic" when:
588 * 1. "Economic" mode is ON.
589 * or
590 * 2. "Fast" mode failed. It can only failed if there is no memory.
591 */
592 if (!fast || error != 0)
593 error = g_stripe_start_economic(bp, no, offset, length);
594 if (error != 0) {
595 if (bp->bio_error == 0)
596 bp->bio_error = error;
597 g_io_deliver(bp, bp->bio_error);
598 }
599}
600
601static void
602g_stripe_check_and_run(struct g_stripe_softc *sc)
603{
604 off_t mediasize, ms;
605 u_int no, sectorsize = 0;
606
607 if (g_stripe_nvalid(sc) != sc->sc_ndisks)
608 return;
609
610 sc->sc_provider = g_new_providerf(sc->sc_geom, "%s", sc->sc_geom->name);
611 /*
612 * Find the smallest disk.
613 */
614 mediasize = sc->sc_disks[0]->provider->mediasize;
615 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
616 mediasize -= sc->sc_disks[0]->provider->sectorsize;
617 mediasize -= mediasize % sc->sc_stripesize;
618 sectorsize = sc->sc_disks[0]->provider->sectorsize;
619 for (no = 1; no < sc->sc_ndisks; no++) {
620 ms = sc->sc_disks[no]->provider->mediasize;
621 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
622 ms -= sc->sc_disks[no]->provider->sectorsize;
623 ms -= ms % sc->sc_stripesize;
624 if (ms < mediasize)
625 mediasize = ms;
626 sectorsize = lcm(sectorsize,
627 sc->sc_disks[no]->provider->sectorsize);
628 }
629 sc->sc_provider->sectorsize = sectorsize;
630 sc->sc_provider->mediasize = mediasize * sc->sc_ndisks;
631 g_error_provider(sc->sc_provider, 0);
632
633 G_STRIPE_DEBUG(0, "Device %s activated.", sc->sc_geom->name);
634}
635
636static int
637g_stripe_read_metadata(struct g_consumer *cp, struct g_stripe_metadata *md)
638{
639 struct g_provider *pp;
640 u_char *buf;
641 int error;
642
643 g_topology_assert();
644
645 error = g_access(cp, 1, 0, 0);
646 if (error != 0)
647 return (error);
648 pp = cp->provider;
649 g_topology_unlock();
650 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
651 &error);
652 g_topology_lock();
653 g_access(cp, -1, 0, 0);
654 if (buf == NULL)
655 return (error);
656
657 /* Decode metadata. */
658 stripe_metadata_decode(buf, md);
659 g_free(buf);
660
661 return (0);
662}
663
664/*
665 * Add disk to given device.
666 */
667static int
668g_stripe_add_disk(struct g_stripe_softc *sc, struct g_provider *pp, u_int no)
669{
670 struct g_consumer *cp, *fcp;
671 struct g_geom *gp;
672 int error;
673
674 /* Metadata corrupted? */
675 if (no >= sc->sc_ndisks)
676 return (EINVAL);
677
678 /* Check if disk is not already attached. */
679 if (sc->sc_disks[no] != NULL)
680 return (EEXIST);
681
682 gp = sc->sc_geom;
683 fcp = LIST_FIRST(&gp->consumer);
684
685 cp = g_new_consumer(gp);
686 error = g_attach(cp, pp);
687 if (error != 0) {
688 g_destroy_consumer(cp);
689 return (error);
690 }
691
692 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) {
693 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
694 if (error != 0) {
695 g_detach(cp);
696 g_destroy_consumer(cp);
697 return (error);
698 }
699 }
700 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) {
701 struct g_stripe_metadata md;
702
703 /* Reread metadata. */
704 error = g_stripe_read_metadata(cp, &md);
705 if (error != 0)
706 goto fail;
707
708 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0 ||
709 strcmp(md.md_name, sc->sc_name) != 0 ||
710 md.md_id != sc->sc_id) {
711 G_STRIPE_DEBUG(0, "Metadata on %s changed.", pp->name);
712 goto fail;
713 }
714 }
715
716 cp->private = sc;
717 cp->index = no;
718 sc->sc_disks[no] = cp;
719
720 G_STRIPE_DEBUG(0, "Disk %s attached to %s.", pp->name, gp->name);
721
722 g_stripe_check_and_run(sc);
723
724 return (0);
725fail:
726 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0))
727 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace);
728 g_detach(cp);
729 g_destroy_consumer(cp);
730 return (error);
731}
732
733static struct g_geom *
734g_stripe_create(struct g_class *mp, const struct g_stripe_metadata *md,
735 u_int type)
736{
737 struct g_stripe_softc *sc;
738 struct g_geom *gp;
739 u_int no;
740
741 G_STRIPE_DEBUG(1, "Creating device %s.stripe (id=%u).", md->md_name,
742 md->md_id);
743
744 /* Two disks is minimum. */
745 if (md->md_all <= 1) {
746 G_STRIPE_DEBUG(0, "Too few disks defined for %s.stripe.",
747 md->md_name);
748 return (NULL);
749 }
750#if 0
751 /* Stripe size have to be grater than or equal to sector size. */
752 if (md->md_stripesize < sectorsize) {
753 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.stripe.",
754 md->md_name);
755 return (NULL);
756 }
757#endif
758 /* Stripe size have to be power of 2. */
759 if (!powerof2(md->md_stripesize)) {
760 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.stripe.",
761 md->md_name);
762 return (NULL);
763 }
764
765 /* Check for duplicate unit */
766 LIST_FOREACH(gp, &mp->geom, geom) {
767 sc = gp->softc;
768 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) {
769 G_STRIPE_DEBUG(0, "Device %s already configured.",
770 gp->name);
771 return (NULL);
772 }
773 }
774 gp = g_new_geomf(mp, "%s.stripe", md->md_name);
775 gp->softc = NULL; /* for a moment */
776
777 sc = malloc(sizeof(*sc), M_STRIPE, M_NOWAIT | M_ZERO);
778 if (sc == NULL) {
779 G_STRIPE_DEBUG(0, "Can't allocate memory for device %s.",
780 gp->name);
781 g_destroy_geom(gp);
782 return (NULL);
783 }
784
785 gp->start = g_stripe_start;
786 gp->spoiled = g_stripe_orphan;
787 gp->orphan = g_stripe_orphan;
788 gp->access = g_stripe_access;
789 gp->dumpconf = g_stripe_dumpconf;
790
791 strlcpy(sc->sc_name, md->md_name, sizeof(sc->sc_name));
792 sc->sc_id = md->md_id;
793 sc->sc_stripesize = md->md_stripesize;
794 sc->sc_stripebits = BITCOUNT(sc->sc_stripesize - 1);
795 sc->sc_ndisks = md->md_all;
796 sc->sc_disks = malloc(sizeof(struct g_consumer *) * sc->sc_ndisks,
797 M_STRIPE, M_WAITOK | M_ZERO);
798 for (no = 0; no < sc->sc_ndisks; no++)
799 sc->sc_disks[no] = NULL;
800 sc->sc_type = type;
801
802 gp->softc = sc;
803 sc->sc_geom = gp;
804 sc->sc_provider = NULL;
805
806 G_STRIPE_DEBUG(0, "Device %s created (id=%u).", gp->name, sc->sc_id);
807
808 return (gp);
809}
810
811static int
812g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force)
813{
814 struct g_provider *pp;
815 struct g_geom *gp;
816 u_int no;
817
818 g_topology_assert();
819
820 if (sc == NULL)
821 return (ENXIO);
822
823 pp = sc->sc_provider;
824 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
825 if (force) {
826 G_STRIPE_DEBUG(0, "Device %s is still open, so it "
827 "can't be definitely removed.", pp->name);
828 } else {
829 G_STRIPE_DEBUG(1,
830 "Device %s is still open (r%dw%de%d).", pp->name,
831 pp->acr, pp->acw, pp->ace);
832 return (EBUSY);
833 }
834 }
835
836 for (no = 0; no < sc->sc_ndisks; no++) {
837 if (sc->sc_disks[no] != NULL)
838 g_stripe_remove_disk(sc->sc_disks[no]);
839 }
840
841 gp = sc->sc_geom;
842 gp->softc = NULL;
843 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)",
844 gp->name));
845 free(sc->sc_disks, M_STRIPE);
846 free(sc, M_STRIPE);
847
848 pp = LIST_FIRST(&gp->provider);
849 if (pp == NULL || (pp->acr == 0 && pp->acw == 0 && pp->ace == 0))
850 G_STRIPE_DEBUG(0, "Device %s destroyed.", gp->name);
851
852 g_wither_geom(gp, ENXIO);
853
854 return (0);
855}
856
857static int
858g_stripe_destroy_geom(struct gctl_req *req __unused,
859 struct g_class *mp __unused, struct g_geom *gp)
860{
861 struct g_stripe_softc *sc;
862
863 sc = gp->softc;
864 return (g_stripe_destroy(sc, 0));
865}
866
867static struct g_geom *
868g_stripe_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
869{
870 struct g_stripe_metadata md;
871 struct g_stripe_softc *sc;
872 struct g_consumer *cp;
873 struct g_geom *gp;
874 int error;
875
876 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
877 g_topology_assert();
878
879 G_STRIPE_DEBUG(3, "Tasting %s.", pp->name);
880
881 gp = g_new_geomf(mp, "stripe:taste");
882 gp->start = g_stripe_start;
883 gp->access = g_stripe_access;
884 gp->orphan = g_stripe_orphan;
885 cp = g_new_consumer(gp);
886 g_attach(cp, pp);
887
888 error = g_stripe_read_metadata(cp, &md);
889 g_wither_geom(gp, ENXIO);
890 if (error != 0)
891 return (NULL);
892 gp = NULL;
893
894 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0)
895 return (NULL);
896 if (md.md_version > G_STRIPE_VERSION) {
897 printf("geom_stripe.ko module is too old to handle %s.\n",
898 pp->name);
899 return (NULL);
900 }
901
902 /*
903 * Let's check if device already exists.
904 */
905 sc = NULL;
906 LIST_FOREACH(gp, &mp->geom, geom) {
907 sc = gp->softc;
908 if (sc == NULL)
909 continue;
910 if (sc->sc_type != G_STRIPE_TYPE_AUTOMATIC)
911 continue;
912 if (strcmp(md.md_name, sc->sc_name) != 0)
913 continue;
914 if (md.md_id != sc->sc_id)
915 continue;
916 break;
917 }
918 if (gp != NULL) {
919 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
920 error = g_stripe_add_disk(sc, pp, md.md_no);
921 if (error != 0) {
922 G_STRIPE_DEBUG(0,
923 "Cannot add disk %s to %s (error=%d).", pp->name,
924 gp->name, error);
925 return (NULL);
926 }
927 } else {
928 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_AUTOMATIC);
929 if (gp == NULL) {
930 G_STRIPE_DEBUG(0, "Cannot create device %s.stripe.",
931 md.md_name);
932 return (NULL);
933 }
934 sc = gp->softc;
935 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
936 error = g_stripe_add_disk(sc, pp, md.md_no);
937 if (error != 0) {
938 G_STRIPE_DEBUG(0,
939 "Cannot add disk %s to %s (error=%d).", pp->name,
940 gp->name, error);
941 g_stripe_destroy(sc, 1);
942 return (NULL);
943 }
944 }
945
946 return (gp);
947}
948
949static void
950g_stripe_ctl_create(struct gctl_req *req, struct g_class *mp)
951{
952 u_int attached, no;
953 struct g_stripe_metadata md;
954 struct g_provider *pp;
955 struct g_stripe_softc *sc;
956 struct g_geom *gp;
957 struct sbuf *sb;
958 intmax_t *stripesize;
959 const char *name;
960 char param[16];
961 int *nargs;
962
963 g_topology_assert();
964 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
965 if (nargs == NULL) {
966 gctl_error(req, "No '%s' argument.", "nargs");
967 return;
968 }
969 if (*nargs <= 2) {
970 gctl_error(req, "Too few arguments.");
971 return;
972 }
973
974 strlcpy(md.md_magic, G_STRIPE_MAGIC, sizeof(md.md_magic));
975 md.md_version = G_STRIPE_VERSION;
976 name = gctl_get_asciiparam(req, "arg0");
977 if (name == NULL) {
978 gctl_error(req, "No 'arg%u' argument.", 0);
979 return;
980 }
981 strlcpy(md.md_name, name, sizeof(md.md_name));
982 md.md_id = arc4random();
983 md.md_no = 0;
984 md.md_all = *nargs - 1;
985 stripesize = gctl_get_paraml(req, "stripesize", sizeof(*stripesize));
986 if (stripesize == NULL) {
987 gctl_error(req, "No '%s' argument.", "stripesize");
988 return;
989 }
990 md.md_stripesize = *stripesize;
991
992 /* Check all providers are valid */
993 for (no = 1; no < *nargs; no++) {
994 snprintf(param, sizeof(param), "arg%u", no);
995 name = gctl_get_asciiparam(req, param);
996 if (name == NULL) {
997 gctl_error(req, "No 'arg%u' argument.", no);
998 return;
999 }
1000 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1001 name += strlen("/dev/");
1002 pp = g_provider_by_name(name);
1003 if (pp == NULL) {
1004 G_STRIPE_DEBUG(1, "Disk %s is invalid.", name);
1005 gctl_error(req, "Disk %s is invalid.", name);
1006 return;
1007 }
1008 }
1009
1010 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_MANUAL);
1011 if (gp == NULL) {
1012 gctl_error(req, "Can't configure %s.stripe.", md.md_name);
1013 return;
1014 }
1015
1016 sc = gp->softc;
1017 sb = sbuf_new(NULL, NULL, 0, SBUF_AUTOEXTEND);
1018 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name);
1019 for (attached = 0, no = 1; no < *nargs; no++) {
1020 snprintf(param, sizeof(param), "arg%u", no);
1021 name = gctl_get_asciiparam(req, param);
1022 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1023 name += strlen("/dev/");
1024 pp = g_provider_by_name(name);
1025 KASSERT(pp != NULL, ("Provider %s disappear?!", name));
1026 if (g_stripe_add_disk(sc, pp, no - 1) != 0) {
1027 G_STRIPE_DEBUG(1, "Disk %u (%s) not attached to %s.",
1028 no, pp->name, gp->name);
1029 sbuf_printf(sb, " %s", pp->name);
1030 continue;
1031 }
1032 attached++;
1033 }
1034 sbuf_finish(sb);
1035 if (md.md_all != attached) {
1036 g_stripe_destroy(gp->softc, 1);
1037 gctl_error(req, "%s", sbuf_data(sb));
1038 }
1039 sbuf_delete(sb);
1040}
1041
1042static struct g_stripe_softc *
1043g_stripe_find_device(struct g_class *mp, const char *name)
1044{
1045 struct g_stripe_softc *sc;
1046 struct g_geom *gp;
1047
1048 LIST_FOREACH(gp, &mp->geom, geom) {
1049 sc = gp->softc;
1050 if (sc == NULL)
1051 continue;
1052 if (strcmp(gp->name, name) == 0 ||
1053 strcmp(sc->sc_name, name) == 0) {
1054 return (sc);
1055 }
1056 }
1057 return (NULL);
1058}
1059
1060static void
1061g_stripe_ctl_destroy(struct gctl_req *req, struct g_class *mp)
1062{
1063 struct g_stripe_softc *sc;
1064 int *force, *nargs, error;
1065 const char *name;
1066 char param[16];
1067 u_int i;
1068
1069 g_topology_assert();
1070
1071 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1072 if (nargs == NULL) {
1073 gctl_error(req, "No '%s' argument.", "nargs");
1074 return;
1075 }
1076 if (*nargs <= 0) {
1077 gctl_error(req, "Missing device(s).");
1078 return;
1079 }
1080 force = gctl_get_paraml(req, "force", sizeof(*force));
1081 if (force == NULL) {
1082 gctl_error(req, "No '%s' argument.", "force");
1083 return;
1084 }
1085
1086 for (i = 0; i < (u_int)*nargs; i++) {
1087 snprintf(param, sizeof(param), "arg%u", i);
1088 name = gctl_get_asciiparam(req, param);
1089 if (name == NULL) {
1090 gctl_error(req, "No 'arg%u' argument.", i);
1091 return;
1092 }
1093 sc = g_stripe_find_device(mp, name);
1094 if (sc == NULL) {
1095 gctl_error(req, "No such device: %s.", name);
1096 return;
1097 }
1098 error = g_stripe_destroy(sc, *force);
1099 if (error != 0) {
1100 gctl_error(req, "Cannot destroy device %s (error=%d).",
1101 sc->sc_geom->name, error);
1102 return;
1103 }
1104 }
1105}
1106
1107static void
1108g_stripe_config(struct gctl_req *req, struct g_class *mp, const char *verb)
1109{
1110 uint32_t *version;
1111
1112 g_topology_assert();
1113
1114 version = gctl_get_paraml(req, "version", sizeof(*version));
1115 if (version == NULL) {
1116 gctl_error(req, "No '%s' argument.", "version");
1117 return;
1118 }
1119 if (*version != G_STRIPE_VERSION) {
1120 gctl_error(req, "Userland and kernel parts are out of sync.");
1121 return;
1122 }
1123
1124 if (strcmp(verb, "create") == 0) {
1125 g_stripe_ctl_create(req, mp);
1126 return;
1127 } else if (strcmp(verb, "destroy") == 0 ||
1128 strcmp(verb, "stop") == 0) {
1129 g_stripe_ctl_destroy(req, mp);
1130 return;
1131 }
1132
1133 gctl_error(req, "Unknown verb.");
1134}
1135
1136static void
1137g_stripe_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1138 struct g_consumer *cp, struct g_provider *pp)
1139{
1140 struct g_stripe_softc *sc;
1141
1142 sc = gp->softc;
1143 if (sc == NULL || pp == NULL)
1144 return;
1145 sbuf_printf(sb, "%s<id>%u</id>\n", indent, (u_int)sc->sc_id);
1146 sbuf_printf(sb, "%s<stripesize>%u</stripesize>\n", indent,
1147 (u_int)sc->sc_stripesize);
1148 switch (sc->sc_type) {
1149 case G_STRIPE_TYPE_AUTOMATIC:
1150 sbuf_printf(sb, "%s<type>%s</type>\n", indent, "automatic");
1151 break;
1152 case G_STRIPE_TYPE_MANUAL:
1153 sbuf_printf(sb, "%s<type>%s</type>\n", indent, "manual");
1154 break;
1155 default:
1156 sbuf_printf(sb, "%s<type>%s</type>\n", indent, "unknown");
1157 break;
1158 }
1159 sbuf_printf(sb, "%s<providers>", indent);
1160 LIST_FOREACH(cp, &gp->consumer, consumer) {
1161 if (cp->provider == NULL)
1162 continue;
1163 sbuf_printf(sb, "%s", cp->provider->name);
1164 if (LIST_NEXT(cp, consumer) != NULL)
1165 sbuf_printf(sb, " ");
1166 }
1167 sbuf_printf(sb, "</providers>\n");
1168 sbuf_printf(sb, "%s<status>total=%u, online=%u</status>\n", indent,
1169 sc->sc_ndisks, g_stripe_nvalid(sc));
1170 if (pp->error == 0)
1171 sbuf_printf(sb, "%s<state>UP</state>\n", indent);
1172 else
1173 sbuf_printf(sb, "%s<state>DOWN</state>\n", indent);
1174}
1175
1176DECLARE_GEOM_CLASS(g_stripe_class, g_stripe);