1 /*-
2 * Copyright (c) 2004-2005 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: releng/8.1/sys/geom/stripe/g_stripe.c 201579 2010-01-05 13:46:39Z mav $");
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 static MALLOC_DEFINE(M_STRIPE, "stripe_data", "GEOM_STRIPE Data");
45
46 static uma_zone_t g_stripe_zone;
47
48 static int g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force);
49 static int g_stripe_destroy_geom(struct gctl_req *req, struct g_class *mp,
50 struct g_geom *gp);
51
52 static g_taste_t g_stripe_taste;
53 static g_ctl_req_t g_stripe_config;
54 static g_dumpconf_t g_stripe_dumpconf;
55 static g_init_t g_stripe_init;
56 static g_fini_t g_stripe_fini;
57
58 struct g_class g_stripe_class = {
59 .name = G_STRIPE_CLASS_NAME,
60 .version = G_VERSION,
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
68 SYSCTL_DECL(_kern_geom);
69 SYSCTL_NODE(_kern_geom, OID_AUTO, stripe, CTLFLAG_RW, 0, "GEOM_STRIPE stuff");
70 static u_int g_stripe_debug = 0;
71 TUNABLE_INT("kern.geom.stripe.debug", &g_stripe_debug);
72 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, debug, CTLFLAG_RW, &g_stripe_debug, 0,
73 "Debug level");
74 static int g_stripe_fast = 0;
75 TUNABLE_INT("kern.geom.stripe.fast", &g_stripe_fast);
76 static int
77 g_sysctl_stripe_fast(SYSCTL_HANDLER_ARGS)
78 {
79 int error, fast;
80
81 fast = g_stripe_fast;
82 error = sysctl_handle_int(oidp, &fast, 0, req);
83 if (error == 0 && req->newptr != NULL)
84 g_stripe_fast = fast;
85 return (error);
86 }
87 SYSCTL_PROC(_kern_geom_stripe, OID_AUTO, fast, CTLTYPE_INT | CTLFLAG_RW,
88 NULL, 0, g_sysctl_stripe_fast, "I", "Fast, but memory-consuming, mode");
89 static u_int g_stripe_maxmem = MAXPHYS * 100;
90 TUNABLE_INT("kern.geom.stripe.maxmem", &g_stripe_maxmem);
91 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, maxmem, CTLFLAG_RD, &g_stripe_maxmem,
92 0, "Maximum memory that can be allocated in \"fast\" mode (in bytes)");
93 static u_int g_stripe_fast_failed = 0;
94 SYSCTL_UINT(_kern_geom_stripe, OID_AUTO, fast_failed, CTLFLAG_RD,
95 &g_stripe_fast_failed, 0, "How many times \"fast\" mode failed");
96
97 /*
98 * Greatest Common Divisor.
99 */
100 static u_int
101 gcd(u_int a, u_int b)
102 {
103 u_int c;
104
105 while (b != 0) {
106 c = a;
107 a = b;
108 b = (c % b);
109 }
110 return (a);
111 }
112
113 /*
114 * Least Common Multiple.
115 */
116 static u_int
117 lcm(u_int a, u_int b)
118 {
119
120 return ((a * b) / gcd(a, b));
121 }
122
123 static void
124 g_stripe_init(struct g_class *mp __unused)
125 {
126
127 g_stripe_zone = uma_zcreate("g_stripe_zone", MAXPHYS, NULL, NULL,
128 NULL, NULL, 0, 0);
129 g_stripe_maxmem -= g_stripe_maxmem % MAXPHYS;
130 uma_zone_set_max(g_stripe_zone, g_stripe_maxmem / MAXPHYS);
131 }
132
133 static void
134 g_stripe_fini(struct g_class *mp __unused)
135 {
136
137 uma_zdestroy(g_stripe_zone);
138 }
139
140 /*
141 * Return the number of valid disks.
142 */
143 static u_int
144 g_stripe_nvalid(struct g_stripe_softc *sc)
145 {
146 u_int i, no;
147
148 no = 0;
149 for (i = 0; i < sc->sc_ndisks; i++) {
150 if (sc->sc_disks[i] != NULL)
151 no++;
152 }
153
154 return (no);
155 }
156
157 static void
158 g_stripe_remove_disk(struct g_consumer *cp)
159 {
160 struct g_stripe_softc *sc;
161 u_int no;
162
163 KASSERT(cp != NULL, ("Non-valid disk in %s.", __func__));
164 sc = (struct g_stripe_softc *)cp->private;
165 KASSERT(sc != NULL, ("NULL sc in %s.", __func__));
166 no = cp->index;
167
168 G_STRIPE_DEBUG(0, "Disk %s removed from %s.", cp->provider->name,
169 sc->sc_name);
170
171 sc->sc_disks[no] = NULL;
172 if (sc->sc_provider != NULL) {
173 sc->sc_provider->flags |= G_PF_WITHER;
174 g_orphan_provider(sc->sc_provider, ENXIO);
175 sc->sc_provider = NULL;
176 G_STRIPE_DEBUG(0, "Device %s removed.", sc->sc_name);
177 }
178
179 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
180 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
181 g_detach(cp);
182 g_destroy_consumer(cp);
183 }
184
185 static void
186 g_stripe_orphan(struct g_consumer *cp)
187 {
188 struct g_stripe_softc *sc;
189 struct g_geom *gp;
190
191 g_topology_assert();
192 gp = cp->geom;
193 sc = gp->softc;
194 if (sc == NULL)
195 return;
196
197 g_stripe_remove_disk(cp);
198 /* If there are no valid disks anymore, remove device. */
199 if (g_stripe_nvalid(sc) == 0)
200 g_stripe_destroy(sc, 1);
201 }
202
203 static int
204 g_stripe_access(struct g_provider *pp, int dr, int dw, int de)
205 {
206 struct g_consumer *cp1, *cp2;
207 struct g_stripe_softc *sc;
208 struct g_geom *gp;
209 int error;
210
211 gp = pp->geom;
212 sc = gp->softc;
213
214 if (sc == NULL) {
215 /*
216 * It looks like geom is being withered.
217 * In that case we allow only negative requests.
218 */
219 KASSERT(dr <= 0 && dw <= 0 && de <= 0,
220 ("Positive access request (device=%s).", pp->name));
221 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 &&
222 (pp->ace + de) == 0) {
223 G_STRIPE_DEBUG(0, "Device %s definitely destroyed.",
224 gp->name);
225 }
226 return (0);
227 }
228
229 /* On first open, grab an extra "exclusive" bit */
230 if (pp->acr == 0 && pp->acw == 0 && pp->ace == 0)
231 de++;
232 /* ... and let go of it on last close */
233 if ((pp->acr + dr) == 0 && (pp->acw + dw) == 0 && (pp->ace + de) == 0)
234 de--;
235
236 error = ENXIO;
237 LIST_FOREACH(cp1, &gp->consumer, consumer) {
238 error = g_access(cp1, dr, dw, de);
239 if (error == 0)
240 continue;
241 /*
242 * If we fail here, backout all previous changes.
243 */
244 LIST_FOREACH(cp2, &gp->consumer, consumer) {
245 if (cp1 == cp2)
246 return (error);
247 g_access(cp2, -dr, -dw, -de);
248 }
249 /* NOTREACHED */
250 }
251
252 return (error);
253 }
254
255 static void
256 g_stripe_copy(struct g_stripe_softc *sc, char *src, char *dst, off_t offset,
257 off_t length, int mode)
258 {
259 u_int stripesize;
260 size_t len;
261
262 stripesize = sc->sc_stripesize;
263 len = (size_t)(stripesize - (offset & (stripesize - 1)));
264 do {
265 bcopy(src, dst, len);
266 if (mode) {
267 dst += len + stripesize * (sc->sc_ndisks - 1);
268 src += len;
269 } else {
270 dst += len;
271 src += len + stripesize * (sc->sc_ndisks - 1);
272 }
273 length -= len;
274 KASSERT(length >= 0,
275 ("Length < 0 (stripesize=%zu, offset=%jd, length=%jd).",
276 (size_t)stripesize, (intmax_t)offset, (intmax_t)length));
277 if (length > stripesize)
278 len = stripesize;
279 else
280 len = length;
281 } while (length > 0);
282 }
283
284 static void
285 g_stripe_done(struct bio *bp)
286 {
287 struct g_stripe_softc *sc;
288 struct bio *pbp;
289
290 pbp = bp->bio_parent;
291 sc = pbp->bio_to->geom->softc;
292 if (pbp->bio_error == 0)
293 pbp->bio_error = bp->bio_error;
294 pbp->bio_completed += bp->bio_completed;
295 if (bp->bio_cmd == BIO_READ && bp->bio_caller1 != NULL) {
296 g_stripe_copy(sc, bp->bio_data, bp->bio_caller1, bp->bio_offset,
297 bp->bio_length, 1);
298 bp->bio_data = bp->bio_caller1;
299 bp->bio_caller1 = NULL;
300 }
301 g_destroy_bio(bp);
302 pbp->bio_inbed++;
303 if (pbp->bio_children == pbp->bio_inbed) {
304 if (pbp->bio_driver1 != NULL)
305 uma_zfree(g_stripe_zone, pbp->bio_driver1);
306 g_io_deliver(pbp, pbp->bio_error);
307 }
308 }
309
310 static int
311 g_stripe_start_fast(struct bio *bp, u_int no, off_t offset, off_t length)
312 {
313 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
314 u_int nparts = 0, stripesize;
315 struct g_stripe_softc *sc;
316 char *addr, *data = NULL;
317 struct bio *cbp;
318 int error;
319
320 sc = bp->bio_to->geom->softc;
321
322 addr = bp->bio_data;
323 stripesize = sc->sc_stripesize;
324
325 cbp = g_clone_bio(bp);
326 if (cbp == NULL) {
327 error = ENOMEM;
328 goto failure;
329 }
330 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
331 nparts++;
332 /*
333 * Fill in the component buf structure.
334 */
335 cbp->bio_done = g_stripe_done;
336 cbp->bio_offset = offset;
337 cbp->bio_data = addr;
338 cbp->bio_caller1 = NULL;
339 cbp->bio_length = length;
340 cbp->bio_caller2 = sc->sc_disks[no];
341
342 /* offset -= offset % stripesize; */
343 offset -= offset & (stripesize - 1);
344 addr += length;
345 length = bp->bio_length - length;
346 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
347 if (no > sc->sc_ndisks - 1) {
348 no = 0;
349 offset += stripesize;
350 }
351 if (nparts >= sc->sc_ndisks) {
352 cbp = TAILQ_NEXT(cbp, bio_queue);
353 if (cbp == NULL)
354 cbp = TAILQ_FIRST(&queue);
355 nparts++;
356 /*
357 * Update bio structure.
358 */
359 /*
360 * MIN() is in case when
361 * (bp->bio_length % sc->sc_stripesize) != 0.
362 */
363 cbp->bio_length += MIN(stripesize, length);
364 if (cbp->bio_caller1 == NULL) {
365 cbp->bio_caller1 = cbp->bio_data;
366 cbp->bio_data = NULL;
367 if (data == NULL) {
368 data = uma_zalloc(g_stripe_zone,
369 M_NOWAIT);
370 if (data == NULL) {
371 error = ENOMEM;
372 goto failure;
373 }
374 }
375 }
376 } else {
377 cbp = g_clone_bio(bp);
378 if (cbp == NULL) {
379 error = ENOMEM;
380 goto failure;
381 }
382 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
383 nparts++;
384 /*
385 * Fill in the component buf structure.
386 */
387 cbp->bio_done = g_stripe_done;
388 cbp->bio_offset = offset;
389 cbp->bio_data = addr;
390 cbp->bio_caller1 = NULL;
391 /*
392 * MIN() is in case when
393 * (bp->bio_length % sc->sc_stripesize) != 0.
394 */
395 cbp->bio_length = MIN(stripesize, length);
396 cbp->bio_caller2 = sc->sc_disks[no];
397 }
398 }
399 if (data != NULL)
400 bp->bio_driver1 = data;
401 /*
402 * Fire off all allocated requests!
403 */
404 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
405 struct g_consumer *cp;
406
407 TAILQ_REMOVE(&queue, cbp, bio_queue);
408 cp = cbp->bio_caller2;
409 cbp->bio_caller2 = NULL;
410 cbp->bio_to = cp->provider;
411 if (cbp->bio_caller1 != NULL) {
412 cbp->bio_data = data;
413 if (bp->bio_cmd == BIO_WRITE) {
414 g_stripe_copy(sc, cbp->bio_caller1, data,
415 cbp->bio_offset, cbp->bio_length, 0);
416 }
417 data += cbp->bio_length;
418 }
419 G_STRIPE_LOGREQ(cbp, "Sending request.");
420 g_io_request(cbp, cp);
421 }
422 return (0);
423 failure:
424 if (data != NULL)
425 uma_zfree(g_stripe_zone, data);
426 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
427 TAILQ_REMOVE(&queue, cbp, bio_queue);
428 if (cbp->bio_caller1 != NULL) {
429 cbp->bio_data = cbp->bio_caller1;
430 cbp->bio_caller1 = NULL;
431 }
432 bp->bio_children--;
433 g_destroy_bio(cbp);
434 }
435 return (error);
436 }
437
438 static int
439 g_stripe_start_economic(struct bio *bp, u_int no, off_t offset, off_t length)
440 {
441 TAILQ_HEAD(, bio) queue = TAILQ_HEAD_INITIALIZER(queue);
442 struct g_stripe_softc *sc;
443 uint32_t stripesize;
444 struct bio *cbp;
445 char *addr;
446 int error;
447
448 sc = bp->bio_to->geom->softc;
449
450 addr = bp->bio_data;
451 stripesize = sc->sc_stripesize;
452
453 cbp = g_clone_bio(bp);
454 if (cbp == NULL) {
455 error = ENOMEM;
456 goto failure;
457 }
458 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
459 /*
460 * Fill in the component buf structure.
461 */
462 cbp->bio_done = g_std_done;
463 cbp->bio_offset = offset;
464 cbp->bio_data = addr;
465 cbp->bio_length = length;
466 cbp->bio_caller2 = sc->sc_disks[no];
467
468 /* offset -= offset % stripesize; */
469 offset -= offset & (stripesize - 1);
470 addr += length;
471 length = bp->bio_length - length;
472 for (no++; length > 0; no++, length -= stripesize, addr += stripesize) {
473 if (no > sc->sc_ndisks - 1) {
474 no = 0;
475 offset += stripesize;
476 }
477 cbp = g_clone_bio(bp);
478 if (cbp == NULL) {
479 error = ENOMEM;
480 goto failure;
481 }
482 TAILQ_INSERT_TAIL(&queue, cbp, bio_queue);
483
484 /*
485 * Fill in the component buf structure.
486 */
487 cbp->bio_done = g_std_done;
488 cbp->bio_offset = offset;
489 cbp->bio_data = addr;
490 /*
491 * MIN() is in case when
492 * (bp->bio_length % sc->sc_stripesize) != 0.
493 */
494 cbp->bio_length = MIN(stripesize, length);
495
496 cbp->bio_caller2 = sc->sc_disks[no];
497 }
498 /*
499 * Fire off all allocated requests!
500 */
501 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
502 struct g_consumer *cp;
503
504 TAILQ_REMOVE(&queue, cbp, bio_queue);
505 cp = cbp->bio_caller2;
506 cbp->bio_caller2 = NULL;
507 cbp->bio_to = cp->provider;
508 G_STRIPE_LOGREQ(cbp, "Sending request.");
509 g_io_request(cbp, cp);
510 }
511 return (0);
512 failure:
513 while ((cbp = TAILQ_FIRST(&queue)) != NULL) {
514 TAILQ_REMOVE(&queue, cbp, bio_queue);
515 bp->bio_children--;
516 g_destroy_bio(cbp);
517 }
518 return (error);
519 }
520
521 static void
522 g_stripe_flush(struct g_stripe_softc *sc, struct bio *bp)
523 {
524 struct bio_queue_head queue;
525 struct g_consumer *cp;
526 struct bio *cbp;
527 u_int no;
528
529 bioq_init(&queue);
530 for (no = 0; no < sc->sc_ndisks; no++) {
531 cbp = g_clone_bio(bp);
532 if (cbp == NULL) {
533 for (cbp = bioq_first(&queue); cbp != NULL;
534 cbp = bioq_first(&queue)) {
535 bioq_remove(&queue, cbp);
536 g_destroy_bio(cbp);
537 }
538 if (bp->bio_error == 0)
539 bp->bio_error = ENOMEM;
540 g_io_deliver(bp, bp->bio_error);
541 return;
542 }
543 bioq_insert_tail(&queue, cbp);
544 cbp->bio_done = g_std_done;
545 cbp->bio_caller1 = sc->sc_disks[no];
546 cbp->bio_to = sc->sc_disks[no]->provider;
547 }
548 for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
549 bioq_remove(&queue, cbp);
550 G_STRIPE_LOGREQ(cbp, "Sending request.");
551 cp = cbp->bio_caller1;
552 cbp->bio_caller1 = NULL;
553 g_io_request(cbp, cp);
554 }
555 }
556
557 static void
558 g_stripe_start(struct bio *bp)
559 {
560 off_t offset, start, length, nstripe;
561 struct g_stripe_softc *sc;
562 u_int no, stripesize;
563 int error, fast = 0;
564
565 sc = bp->bio_to->geom->softc;
566 /*
567 * If sc == NULL, provider's error should be set and g_stripe_start()
568 * should not be called at all.
569 */
570 KASSERT(sc != NULL,
571 ("Provider's error should be set (error=%d)(device=%s).",
572 bp->bio_to->error, bp->bio_to->name));
573
574 G_STRIPE_LOGREQ(bp, "Request received.");
575
576 switch (bp->bio_cmd) {
577 case BIO_READ:
578 case BIO_WRITE:
579 case BIO_DELETE:
580 break;
581 case BIO_FLUSH:
582 g_stripe_flush(sc, bp);
583 return;
584 case BIO_GETATTR:
585 /* To which provider it should be delivered? */
586 default:
587 g_io_deliver(bp, EOPNOTSUPP);
588 return;
589 }
590
591 stripesize = sc->sc_stripesize;
592
593 /*
594 * Calculations are quite messy, but fast I hope.
595 */
596
597 /* Stripe number. */
598 /* nstripe = bp->bio_offset / stripesize; */
599 nstripe = bp->bio_offset >> (off_t)sc->sc_stripebits;
600 /* Disk number. */
601 no = nstripe % sc->sc_ndisks;
602 /* Start position in stripe. */
603 /* start = bp->bio_offset % stripesize; */
604 start = bp->bio_offset & (stripesize - 1);
605 /* Start position in disk. */
606 /* offset = (nstripe / sc->sc_ndisks) * stripesize + start; */
607 offset = ((nstripe / sc->sc_ndisks) << sc->sc_stripebits) + start;
608 /* Length of data to operate. */
609 length = MIN(bp->bio_length, stripesize - start);
610
611 /*
612 * Do use "fast" mode when:
613 * 1. "Fast" mode is ON.
614 * and
615 * 2. Request size is less than or equal to MAXPHYS,
616 * which should always be true.
617 * and
618 * 3. Request size is bigger than stripesize * ndisks. If it isn't,
619 * there will be no need to send more than one I/O request to
620 * a provider, so there is nothing to optmize.
621 */
622 if (g_stripe_fast && bp->bio_length <= MAXPHYS &&
623 bp->bio_length >= stripesize * sc->sc_ndisks) {
624 fast = 1;
625 }
626 error = 0;
627 if (fast) {
628 error = g_stripe_start_fast(bp, no, offset, length);
629 if (error != 0)
630 g_stripe_fast_failed++;
631 }
632 /*
633 * Do use "economic" when:
634 * 1. "Economic" mode is ON.
635 * or
636 * 2. "Fast" mode failed. It can only failed if there is no memory.
637 */
638 if (!fast || error != 0)
639 error = g_stripe_start_economic(bp, no, offset, length);
640 if (error != 0) {
641 if (bp->bio_error == 0)
642 bp->bio_error = error;
643 g_io_deliver(bp, bp->bio_error);
644 }
645 }
646
647 static void
648 g_stripe_check_and_run(struct g_stripe_softc *sc)
649 {
650 off_t mediasize, ms;
651 u_int no, sectorsize = 0;
652
653 if (g_stripe_nvalid(sc) != sc->sc_ndisks)
654 return;
655
656 sc->sc_provider = g_new_providerf(sc->sc_geom, "stripe/%s",
657 sc->sc_name);
658 /*
659 * Find the smallest disk.
660 */
661 mediasize = sc->sc_disks[0]->provider->mediasize;
662 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
663 mediasize -= sc->sc_disks[0]->provider->sectorsize;
664 mediasize -= mediasize % sc->sc_stripesize;
665 sectorsize = sc->sc_disks[0]->provider->sectorsize;
666 for (no = 1; no < sc->sc_ndisks; no++) {
667 ms = sc->sc_disks[no]->provider->mediasize;
668 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC)
669 ms -= sc->sc_disks[no]->provider->sectorsize;
670 ms -= ms % sc->sc_stripesize;
671 if (ms < mediasize)
672 mediasize = ms;
673 sectorsize = lcm(sectorsize,
674 sc->sc_disks[no]->provider->sectorsize);
675 }
676 sc->sc_provider->sectorsize = sectorsize;
677 sc->sc_provider->mediasize = mediasize * sc->sc_ndisks;
678 sc->sc_provider->stripesize = sc->sc_stripesize;
679 sc->sc_provider->stripeoffset = 0;
680 g_error_provider(sc->sc_provider, 0);
681
682 G_STRIPE_DEBUG(0, "Device %s activated.", sc->sc_name);
683 }
684
685 static int
686 g_stripe_read_metadata(struct g_consumer *cp, struct g_stripe_metadata *md)
687 {
688 struct g_provider *pp;
689 u_char *buf;
690 int error;
691
692 g_topology_assert();
693
694 error = g_access(cp, 1, 0, 0);
695 if (error != 0)
696 return (error);
697 pp = cp->provider;
698 g_topology_unlock();
699 buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
700 &error);
701 g_topology_lock();
702 g_access(cp, -1, 0, 0);
703 if (buf == NULL)
704 return (error);
705
706 /* Decode metadata. */
707 stripe_metadata_decode(buf, md);
708 g_free(buf);
709
710 return (0);
711 }
712
713 /*
714 * Add disk to given device.
715 */
716 static int
717 g_stripe_add_disk(struct g_stripe_softc *sc, struct g_provider *pp, u_int no)
718 {
719 struct g_consumer *cp, *fcp;
720 struct g_geom *gp;
721 int error;
722
723 /* Metadata corrupted? */
724 if (no >= sc->sc_ndisks)
725 return (EINVAL);
726
727 /* Check if disk is not already attached. */
728 if (sc->sc_disks[no] != NULL)
729 return (EEXIST);
730
731 gp = sc->sc_geom;
732 fcp = LIST_FIRST(&gp->consumer);
733
734 cp = g_new_consumer(gp);
735 error = g_attach(cp, pp);
736 if (error != 0) {
737 g_destroy_consumer(cp);
738 return (error);
739 }
740
741 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0)) {
742 error = g_access(cp, fcp->acr, fcp->acw, fcp->ace);
743 if (error != 0) {
744 g_detach(cp);
745 g_destroy_consumer(cp);
746 return (error);
747 }
748 }
749 if (sc->sc_type == G_STRIPE_TYPE_AUTOMATIC) {
750 struct g_stripe_metadata md;
751
752 /* Reread metadata. */
753 error = g_stripe_read_metadata(cp, &md);
754 if (error != 0)
755 goto fail;
756
757 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0 ||
758 strcmp(md.md_name, sc->sc_name) != 0 ||
759 md.md_id != sc->sc_id) {
760 G_STRIPE_DEBUG(0, "Metadata on %s changed.", pp->name);
761 goto fail;
762 }
763 }
764
765 cp->private = sc;
766 cp->index = no;
767 sc->sc_disks[no] = cp;
768
769 G_STRIPE_DEBUG(0, "Disk %s attached to %s.", pp->name, sc->sc_name);
770
771 g_stripe_check_and_run(sc);
772
773 return (0);
774 fail:
775 if (fcp != NULL && (fcp->acr > 0 || fcp->acw > 0 || fcp->ace > 0))
776 g_access(cp, -fcp->acr, -fcp->acw, -fcp->ace);
777 g_detach(cp);
778 g_destroy_consumer(cp);
779 return (error);
780 }
781
782 static struct g_geom *
783 g_stripe_create(struct g_class *mp, const struct g_stripe_metadata *md,
784 u_int type)
785 {
786 struct g_stripe_softc *sc;
787 struct g_geom *gp;
788 u_int no;
789
790 G_STRIPE_DEBUG(1, "Creating device %s (id=%u).", md->md_name,
791 md->md_id);
792
793 /* Two disks is minimum. */
794 if (md->md_all < 2) {
795 G_STRIPE_DEBUG(0, "Too few disks defined for %s.", md->md_name);
796 return (NULL);
797 }
798 #if 0
799 /* Stripe size have to be grater than or equal to sector size. */
800 if (md->md_stripesize < sectorsize) {
801 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name);
802 return (NULL);
803 }
804 #endif
805 /* Stripe size have to be power of 2. */
806 if (!powerof2(md->md_stripesize)) {
807 G_STRIPE_DEBUG(0, "Invalid stripe size for %s.", md->md_name);
808 return (NULL);
809 }
810
811 /* Check for duplicate unit */
812 LIST_FOREACH(gp, &mp->geom, geom) {
813 sc = gp->softc;
814 if (sc != NULL && strcmp(sc->sc_name, md->md_name) == 0) {
815 G_STRIPE_DEBUG(0, "Device %s already configured.",
816 sc->sc_name);
817 return (NULL);
818 }
819 }
820 gp = g_new_geomf(mp, "%s", md->md_name);
821 gp->softc = NULL; /* for a moment */
822
823 sc = malloc(sizeof(*sc), M_STRIPE, M_WAITOK | M_ZERO);
824 gp->start = g_stripe_start;
825 gp->spoiled = g_stripe_orphan;
826 gp->orphan = g_stripe_orphan;
827 gp->access = g_stripe_access;
828 gp->dumpconf = g_stripe_dumpconf;
829
830 sc->sc_id = md->md_id;
831 sc->sc_stripesize = md->md_stripesize;
832 sc->sc_stripebits = bitcount32(sc->sc_stripesize - 1);
833 sc->sc_ndisks = md->md_all;
834 sc->sc_disks = malloc(sizeof(struct g_consumer *) * sc->sc_ndisks,
835 M_STRIPE, M_WAITOK | M_ZERO);
836 for (no = 0; no < sc->sc_ndisks; no++)
837 sc->sc_disks[no] = NULL;
838 sc->sc_type = type;
839
840 gp->softc = sc;
841 sc->sc_geom = gp;
842 sc->sc_provider = NULL;
843
844 G_STRIPE_DEBUG(0, "Device %s created (id=%u).", sc->sc_name, sc->sc_id);
845
846 return (gp);
847 }
848
849 static int
850 g_stripe_destroy(struct g_stripe_softc *sc, boolean_t force)
851 {
852 struct g_provider *pp;
853 struct g_geom *gp;
854 u_int no;
855
856 g_topology_assert();
857
858 if (sc == NULL)
859 return (ENXIO);
860
861 pp = sc->sc_provider;
862 if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
863 if (force) {
864 G_STRIPE_DEBUG(0, "Device %s is still open, so it "
865 "can't be definitely removed.", pp->name);
866 } else {
867 G_STRIPE_DEBUG(1,
868 "Device %s is still open (r%dw%de%d).", pp->name,
869 pp->acr, pp->acw, pp->ace);
870 return (EBUSY);
871 }
872 }
873
874 for (no = 0; no < sc->sc_ndisks; no++) {
875 if (sc->sc_disks[no] != NULL)
876 g_stripe_remove_disk(sc->sc_disks[no]);
877 }
878
879 gp = sc->sc_geom;
880 gp->softc = NULL;
881 KASSERT(sc->sc_provider == NULL, ("Provider still exists? (device=%s)",
882 gp->name));
883 free(sc->sc_disks, M_STRIPE);
884 free(sc, M_STRIPE);
885
886 pp = LIST_FIRST(&gp->provider);
887 if (pp == NULL || (pp->acr == 0 && pp->acw == 0 && pp->ace == 0))
888 G_STRIPE_DEBUG(0, "Device %s destroyed.", gp->name);
889
890 g_wither_geom(gp, ENXIO);
891
892 return (0);
893 }
894
895 static int
896 g_stripe_destroy_geom(struct gctl_req *req __unused,
897 struct g_class *mp __unused, struct g_geom *gp)
898 {
899 struct g_stripe_softc *sc;
900
901 sc = gp->softc;
902 return (g_stripe_destroy(sc, 0));
903 }
904
905 static struct g_geom *
906 g_stripe_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
907 {
908 struct g_stripe_metadata md;
909 struct g_stripe_softc *sc;
910 struct g_consumer *cp;
911 struct g_geom *gp;
912 int error;
913
914 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
915 g_topology_assert();
916
917 /* Skip providers that are already open for writing. */
918 if (pp->acw > 0)
919 return (NULL);
920
921 G_STRIPE_DEBUG(3, "Tasting %s.", pp->name);
922
923 gp = g_new_geomf(mp, "stripe:taste");
924 gp->start = g_stripe_start;
925 gp->access = g_stripe_access;
926 gp->orphan = g_stripe_orphan;
927 cp = g_new_consumer(gp);
928 g_attach(cp, pp);
929 error = g_stripe_read_metadata(cp, &md);
930 g_detach(cp);
931 g_destroy_consumer(cp);
932 g_destroy_geom(gp);
933 if (error != 0)
934 return (NULL);
935 gp = NULL;
936
937 if (strcmp(md.md_magic, G_STRIPE_MAGIC) != 0)
938 return (NULL);
939 if (md.md_version > G_STRIPE_VERSION) {
940 printf("geom_stripe.ko module is too old to handle %s.\n",
941 pp->name);
942 return (NULL);
943 }
944 /*
945 * Backward compatibility:
946 */
947 /* There was no md_provider field in earlier versions of metadata. */
948 if (md.md_version < 2)
949 bzero(md.md_provider, sizeof(md.md_provider));
950 /* There was no md_provsize field in earlier versions of metadata. */
951 if (md.md_version < 3)
952 md.md_provsize = pp->mediasize;
953
954 if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0)
955 return (NULL);
956 if (md.md_provsize != pp->mediasize)
957 return (NULL);
958
959 /*
960 * Let's check if device already exists.
961 */
962 sc = NULL;
963 LIST_FOREACH(gp, &mp->geom, geom) {
964 sc = gp->softc;
965 if (sc == NULL)
966 continue;
967 if (sc->sc_type != G_STRIPE_TYPE_AUTOMATIC)
968 continue;
969 if (strcmp(md.md_name, sc->sc_name) != 0)
970 continue;
971 if (md.md_id != sc->sc_id)
972 continue;
973 break;
974 }
975 if (gp != NULL) {
976 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
977 error = g_stripe_add_disk(sc, pp, md.md_no);
978 if (error != 0) {
979 G_STRIPE_DEBUG(0,
980 "Cannot add disk %s to %s (error=%d).", pp->name,
981 gp->name, error);
982 return (NULL);
983 }
984 } else {
985 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_AUTOMATIC);
986 if (gp == NULL) {
987 G_STRIPE_DEBUG(0, "Cannot create device %s.",
988 md.md_name);
989 return (NULL);
990 }
991 sc = gp->softc;
992 G_STRIPE_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
993 error = g_stripe_add_disk(sc, pp, md.md_no);
994 if (error != 0) {
995 G_STRIPE_DEBUG(0,
996 "Cannot add disk %s to %s (error=%d).", pp->name,
997 gp->name, error);
998 g_stripe_destroy(sc, 1);
999 return (NULL);
1000 }
1001 }
1002
1003 return (gp);
1004 }
1005
1006 static void
1007 g_stripe_ctl_create(struct gctl_req *req, struct g_class *mp)
1008 {
1009 u_int attached, no;
1010 struct g_stripe_metadata md;
1011 struct g_provider *pp;
1012 struct g_stripe_softc *sc;
1013 struct g_geom *gp;
1014 struct sbuf *sb;
1015 intmax_t *stripesize;
1016 const char *name;
1017 char param[16];
1018 int *nargs;
1019
1020 g_topology_assert();
1021 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1022 if (nargs == NULL) {
1023 gctl_error(req, "No '%s' argument.", "nargs");
1024 return;
1025 }
1026 if (*nargs <= 2) {
1027 gctl_error(req, "Too few arguments.");
1028 return;
1029 }
1030
1031 strlcpy(md.md_magic, G_STRIPE_MAGIC, sizeof(md.md_magic));
1032 md.md_version = G_STRIPE_VERSION;
1033 name = gctl_get_asciiparam(req, "arg0");
1034 if (name == NULL) {
1035 gctl_error(req, "No 'arg%u' argument.", 0);
1036 return;
1037 }
1038 strlcpy(md.md_name, name, sizeof(md.md_name));
1039 md.md_id = arc4random();
1040 md.md_no = 0;
1041 md.md_all = *nargs - 1;
1042 stripesize = gctl_get_paraml(req, "stripesize", sizeof(*stripesize));
1043 if (stripesize == NULL) {
1044 gctl_error(req, "No '%s' argument.", "stripesize");
1045 return;
1046 }
1047 md.md_stripesize = *stripesize;
1048 bzero(md.md_provider, sizeof(md.md_provider));
1049 /* This field is not important here. */
1050 md.md_provsize = 0;
1051
1052 /* Check all providers are valid */
1053 for (no = 1; no < *nargs; no++) {
1054 snprintf(param, sizeof(param), "arg%u", no);
1055 name = gctl_get_asciiparam(req, param);
1056 if (name == NULL) {
1057 gctl_error(req, "No 'arg%u' argument.", no);
1058 return;
1059 }
1060 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1061 name += strlen("/dev/");
1062 pp = g_provider_by_name(name);
1063 if (pp == NULL) {
1064 G_STRIPE_DEBUG(1, "Disk %s is invalid.", name);
1065 gctl_error(req, "Disk %s is invalid.", name);
1066 return;
1067 }
1068 }
1069
1070 gp = g_stripe_create(mp, &md, G_STRIPE_TYPE_MANUAL);
1071 if (gp == NULL) {
1072 gctl_error(req, "Can't configure %s.", md.md_name);
1073 return;
1074 }
1075
1076 sc = gp->softc;
1077 sb = sbuf_new_auto();
1078 sbuf_printf(sb, "Can't attach disk(s) to %s:", gp->name);
1079 for (attached = 0, no = 1; no < *nargs; no++) {
1080 snprintf(param, sizeof(param), "arg%u", no);
1081 name = gctl_get_asciiparam(req, param);
1082 if (name == NULL) {
1083 gctl_error(req, "No 'arg%u' argument.", no);
1084 continue;
1085 }
1086 if (strncmp(name, "/dev/", strlen("/dev/")) == 0)
1087 name += strlen("/dev/");
1088 pp = g_provider_by_name(name);
1089 KASSERT(pp != NULL, ("Provider %s disappear?!", name));
1090 if (g_stripe_add_disk(sc, pp, no - 1) != 0) {
1091 G_STRIPE_DEBUG(1, "Disk %u (%s) not attached to %s.",
1092 no, pp->name, gp->name);
1093 sbuf_printf(sb, " %s", pp->name);
1094 continue;
1095 }
1096 attached++;
1097 }
1098 sbuf_finish(sb);
1099 if (md.md_all != attached) {
1100 g_stripe_destroy(gp->softc, 1);
1101 gctl_error(req, "%s", sbuf_data(sb));
1102 }
1103 sbuf_delete(sb);
1104 }
1105
1106 static struct g_stripe_softc *
1107 g_stripe_find_device(struct g_class *mp, const char *name)
1108 {
1109 struct g_stripe_softc *sc;
1110 struct g_geom *gp;
1111
1112 LIST_FOREACH(gp, &mp->geom, geom) {
1113 sc = gp->softc;
1114 if (sc == NULL)
1115 continue;
1116 if (strcmp(sc->sc_name, name) == 0)
1117 return (sc);
1118 }
1119 return (NULL);
1120 }
1121
1122 static void
1123 g_stripe_ctl_destroy(struct gctl_req *req, struct g_class *mp)
1124 {
1125 struct g_stripe_softc *sc;
1126 int *force, *nargs, error;
1127 const char *name;
1128 char param[16];
1129 u_int i;
1130
1131 g_topology_assert();
1132
1133 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1134 if (nargs == NULL) {
1135 gctl_error(req, "No '%s' argument.", "nargs");
1136 return;
1137 }
1138 if (*nargs <= 0) {
1139 gctl_error(req, "Missing device(s).");
1140 return;
1141 }
1142 force = gctl_get_paraml(req, "force", sizeof(*force));
1143 if (force == NULL) {
1144 gctl_error(req, "No '%s' argument.", "force");
1145 return;
1146 }
1147
1148 for (i = 0; i < (u_int)*nargs; i++) {
1149 snprintf(param, sizeof(param), "arg%u", i);
1150 name = gctl_get_asciiparam(req, param);
1151 if (name == NULL) {
1152 gctl_error(req, "No 'arg%u' argument.", i);
1153 return;
1154 }
1155 sc = g_stripe_find_device(mp, name);
1156 if (sc == NULL) {
1157 gctl_error(req, "No such device: %s.", name);
1158 return;
1159 }
1160 error = g_stripe_destroy(sc, *force);
1161 if (error != 0) {
1162 gctl_error(req, "Cannot destroy device %s (error=%d).",
1163 sc->sc_name, error);
1164 return;
1165 }
1166 }
1167 }
1168
1169 static void
1170 g_stripe_config(struct gctl_req *req, struct g_class *mp, const char *verb)
1171 {
1172 uint32_t *version;
1173
1174 g_topology_assert();
1175
1176 version = gctl_get_paraml(req, "version", sizeof(*version));
1177 if (version == NULL) {
1178 gctl_error(req, "No '%s' argument.", "version");
1179 return;
1180 }
1181 if (*version != G_STRIPE_VERSION) {
1182 gctl_error(req, "Userland and kernel parts are out of sync.");
1183 return;
1184 }
1185
1186 if (strcmp(verb, "create") == 0) {
1187 g_stripe_ctl_create(req, mp);
1188 return;
1189 } else if (strcmp(verb, "destroy") == 0 ||
1190 strcmp(verb, "stop") == 0) {
1191 g_stripe_ctl_destroy(req, mp);
1192 return;
1193 }
1194
1195 gctl_error(req, "Unknown verb.");
1196 }
1197
1198 static void
1199 g_stripe_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
1200 struct g_consumer *cp, struct g_provider *pp)
1201 {
1202 struct g_stripe_softc *sc;
1203
1204 sc = gp->softc;
1205 if (sc == NULL)
1206 return;
1207 if (pp != NULL) {
1208 /* Nothing here. */
1209 } else if (cp != NULL) {
1210 sbuf_printf(sb, "%s<Number>%u</Number>\n", indent,
1211 (u_int)cp->index);
1212 } else {
1213 sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
1214 sbuf_printf(sb, "%s<Stripesize>%u</Stripesize>\n", indent,
1215 (u_int)sc->sc_stripesize);
1216 sbuf_printf(sb, "%s<Type>", indent);
1217 switch (sc->sc_type) {
1218 case G_STRIPE_TYPE_AUTOMATIC:
1219 sbuf_printf(sb, "AUTOMATIC");
1220 break;
1221 case G_STRIPE_TYPE_MANUAL:
1222 sbuf_printf(sb, "MANUAL");
1223 break;
1224 default:
1225 sbuf_printf(sb, "UNKNOWN");
1226 break;
1227 }
1228 sbuf_printf(sb, "</Type>\n");
1229 sbuf_printf(sb, "%s<Status>Total=%u, Online=%u</Status>\n",
1230 indent, sc->sc_ndisks, g_stripe_nvalid(sc));
1231 sbuf_printf(sb, "%s<State>", indent);
1232 if (sc->sc_provider != NULL && sc->sc_provider->error == 0)
1233 sbuf_printf(sb, "UP");
1234 else
1235 sbuf_printf(sb, "DOWN");
1236 sbuf_printf(sb, "</State>\n");
1237 }
1238 }
1239
1240 DECLARE_GEOM_CLASS(g_stripe_class, g_stripe);
Cache object: e9bc4c09b923674cb60e84b301408649
|