1 /*-
2 * Copyright (c) 2004, 2007 Lukas Ertl
3 * Copyright (c) 2007, 2009 Ulf Lilleengen
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25 * SUCH DAMAGE.
26 */
27
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD: releng/8.3/sys/geom/vinum/geom_vinum_plex.c 191856 2009-05-06 19:34:32Z lulf $");
30
31 #include <sys/param.h>
32 #include <sys/bio.h>
33 #include <sys/lock.h>
34 #include <sys/malloc.h>
35 #include <sys/systm.h>
36
37 #include <geom/geom.h>
38 #include <geom/vinum/geom_vinum_var.h>
39 #include <geom/vinum/geom_vinum_raid5.h>
40 #include <geom/vinum/geom_vinum.h>
41
42 static int gv_check_parity(struct gv_plex *, struct bio *,
43 struct gv_raid5_packet *);
44 static int gv_normal_parity(struct gv_plex *, struct bio *,
45 struct gv_raid5_packet *);
46 static void gv_plex_flush(struct gv_plex *);
47 static int gv_plex_offset(struct gv_plex *, off_t, off_t, off_t *, off_t *,
48 int *, int);
49 static int gv_plex_normal_request(struct gv_plex *, struct bio *, off_t,
50 off_t, caddr_t);
51 static void gv_post_bio(struct gv_softc *, struct bio *);
52
53 void
54 gv_plex_start(struct gv_plex *p, struct bio *bp)
55 {
56 struct bio *cbp;
57 struct gv_sd *s;
58 struct gv_raid5_packet *wp;
59 caddr_t addr;
60 off_t bcount, boff, len;
61
62 bcount = bp->bio_length;
63 addr = bp->bio_data;
64 boff = bp->bio_offset;
65
66 /* Walk over the whole length of the request, we might split it up. */
67 while (bcount > 0) {
68 wp = NULL;
69
70 /*
71 * RAID5 plexes need special treatment, as a single request
72 * might involve several read/write sub-requests.
73 */
74 if (p->org == GV_PLEX_RAID5) {
75 wp = gv_raid5_start(p, bp, addr, boff, bcount);
76 if (wp == NULL)
77 return;
78
79 len = wp->length;
80
81 if (TAILQ_EMPTY(&wp->bits))
82 g_free(wp);
83 else if (wp->lockbase != -1)
84 TAILQ_INSERT_TAIL(&p->packets, wp, list);
85
86 /*
87 * Requests to concatenated and striped plexes go straight
88 * through.
89 */
90 } else {
91 len = gv_plex_normal_request(p, bp, boff, bcount, addr);
92 }
93 if (len < 0)
94 return;
95
96 bcount -= len;
97 addr += len;
98 boff += len;
99 }
100
101 /*
102 * Fire off all sub-requests. We get the correct consumer (== drive)
103 * to send each request to via the subdisk that was stored in
104 * cbp->bio_caller1.
105 */
106 cbp = bioq_takefirst(p->bqueue);
107 while (cbp != NULL) {
108 /*
109 * RAID5 sub-requests need to come in correct order, otherwise
110 * we trip over the parity, as it might be overwritten by
111 * another sub-request. We abuse cbp->bio_caller2 to mark
112 * potential overlap situations.
113 */
114 if (cbp->bio_caller2 != NULL && gv_stripe_active(p, cbp)) {
115 /* Park the bio on the waiting queue. */
116 cbp->bio_pflags |= GV_BIO_ONHOLD;
117 bioq_disksort(p->wqueue, cbp);
118 } else {
119 s = cbp->bio_caller1;
120 g_io_request(cbp, s->drive_sc->consumer);
121 }
122 cbp = bioq_takefirst(p->bqueue);
123 }
124 }
125
126 static int
127 gv_plex_offset(struct gv_plex *p, off_t boff, off_t bcount, off_t *real_off,
128 off_t *real_len, int *sdno, int growing)
129 {
130 struct gv_sd *s;
131 int i, sdcount;
132 off_t len_left, stripeend, stripeno, stripestart;
133
134 switch (p->org) {
135 case GV_PLEX_CONCAT:
136 /*
137 * Find the subdisk where this request starts. The subdisks in
138 * this list must be ordered by plex_offset.
139 */
140 i = 0;
141 LIST_FOREACH(s, &p->subdisks, in_plex) {
142 if (s->plex_offset <= boff &&
143 s->plex_offset + s->size > boff) {
144 *sdno = i;
145 break;
146 }
147 i++;
148 }
149 if (s == NULL || s->drive_sc == NULL)
150 return (GV_ERR_NOTFOUND);
151
152 /* Calculate corresponding offsets on disk. */
153 *real_off = boff - s->plex_offset;
154 len_left = s->size - (*real_off);
155 KASSERT(len_left >= 0, ("gv_plex_offset: len_left < 0"));
156 *real_len = (bcount > len_left) ? len_left : bcount;
157 break;
158
159 case GV_PLEX_STRIPED:
160 /* The number of the stripe where the request starts. */
161 stripeno = boff / p->stripesize;
162 KASSERT(stripeno >= 0, ("gv_plex_offset: stripeno < 0"));
163
164 /* Take growing subdisks into account when calculating. */
165 sdcount = gv_sdcount(p, (boff >= p->synced));
166
167 if (!(boff + bcount <= p->synced) &&
168 (p->flags & GV_PLEX_GROWING) &&
169 !growing)
170 return (GV_ERR_ISBUSY);
171 *sdno = stripeno % sdcount;
172
173 KASSERT(sdno >= 0, ("gv_plex_offset: sdno < 0"));
174 stripestart = (stripeno / sdcount) *
175 p->stripesize;
176 KASSERT(stripestart >= 0, ("gv_plex_offset: stripestart < 0"));
177 stripeend = stripestart + p->stripesize;
178 *real_off = boff - (stripeno * p->stripesize) +
179 stripestart;
180 len_left = stripeend - *real_off;
181 KASSERT(len_left >= 0, ("gv_plex_offset: len_left < 0"));
182
183 *real_len = (bcount <= len_left) ? bcount : len_left;
184 break;
185
186 default:
187 return (GV_ERR_PLEXORG);
188 }
189 return (0);
190 }
191
192 /*
193 * Prepare a normal plex request.
194 */
195 static int
196 gv_plex_normal_request(struct gv_plex *p, struct bio *bp, off_t boff,
197 off_t bcount, caddr_t addr)
198 {
199 struct gv_sd *s;
200 struct bio *cbp;
201 off_t real_len, real_off;
202 int i, err, sdno;
203
204 s = NULL;
205 sdno = -1;
206 real_len = real_off = 0;
207
208 err = ENXIO;
209
210 if (p == NULL || LIST_EMPTY(&p->subdisks))
211 goto bad;
212
213 err = gv_plex_offset(p, boff, bcount, &real_off,
214 &real_len, &sdno, (bp->bio_pflags & GV_BIO_GROW));
215 /* If the request was blocked, put it into wait. */
216 if (err == GV_ERR_ISBUSY) {
217 bioq_disksort(p->rqueue, bp);
218 return (-1); /* "Fail", and delay request. */
219 }
220 if (err) {
221 err = ENXIO;
222 goto bad;
223 }
224 err = ENXIO;
225
226 /* Find the right subdisk. */
227 i = 0;
228 LIST_FOREACH(s, &p->subdisks, in_plex) {
229 if (i == sdno)
230 break;
231 i++;
232 }
233
234 /* Subdisk not found. */
235 if (s == NULL || s->drive_sc == NULL)
236 goto bad;
237
238 /* Now check if we can handle the request on this subdisk. */
239 switch (s->state) {
240 case GV_SD_UP:
241 /* If the subdisk is up, just continue. */
242 break;
243 case GV_SD_DOWN:
244 if (bp->bio_pflags & GV_BIO_INTERNAL)
245 G_VINUM_DEBUG(0, "subdisk must be in the stale state in"
246 " order to perform administrative requests");
247 goto bad;
248 case GV_SD_STALE:
249 if (!(bp->bio_pflags & GV_BIO_SYNCREQ)) {
250 G_VINUM_DEBUG(0, "subdisk stale, unable to perform "
251 "regular requests");
252 goto bad;
253 }
254
255 G_VINUM_DEBUG(1, "sd %s is initializing", s->name);
256 gv_set_sd_state(s, GV_SD_INITIALIZING, GV_SETSTATE_FORCE);
257 break;
258 case GV_SD_INITIALIZING:
259 if (bp->bio_cmd == BIO_READ)
260 goto bad;
261 break;
262 default:
263 /* All other subdisk states mean it's not accessible. */
264 goto bad;
265 }
266
267 /* Clone the bio and adjust the offsets and sizes. */
268 cbp = g_clone_bio(bp);
269 if (cbp == NULL) {
270 err = ENOMEM;
271 goto bad;
272 }
273 cbp->bio_offset = real_off + s->drive_offset;
274 cbp->bio_length = real_len;
275 cbp->bio_data = addr;
276 cbp->bio_done = gv_done;
277 cbp->bio_caller1 = s;
278
279 /* Store the sub-requests now and let others issue them. */
280 bioq_insert_tail(p->bqueue, cbp);
281 return (real_len);
282 bad:
283 G_VINUM_LOGREQ(0, bp, "plex request failed.");
284 /* Building the sub-request failed. If internal BIO, do not deliver. */
285 if (bp->bio_pflags & GV_BIO_INTERNAL) {
286 if (bp->bio_pflags & GV_BIO_MALLOC)
287 g_free(bp->bio_data);
288 g_destroy_bio(bp);
289 p->flags &= ~(GV_PLEX_SYNCING | GV_PLEX_REBUILDING |
290 GV_PLEX_GROWING);
291 return (-1);
292 }
293 g_io_deliver(bp, err);
294 return (-1);
295 }
296
297 /*
298 * Handle a completed request to a striped or concatenated plex.
299 */
300 void
301 gv_plex_normal_done(struct gv_plex *p, struct bio *bp)
302 {
303 struct bio *pbp;
304
305 pbp = bp->bio_parent;
306 if (pbp->bio_error == 0)
307 pbp->bio_error = bp->bio_error;
308 g_destroy_bio(bp);
309 pbp->bio_inbed++;
310 if (pbp->bio_children == pbp->bio_inbed) {
311 /* Just set it to length since multiple plexes will
312 * screw things up. */
313 pbp->bio_completed = pbp->bio_length;
314 if (pbp->bio_pflags & GV_BIO_SYNCREQ)
315 gv_sync_complete(p, pbp);
316 else if (pbp->bio_pflags & GV_BIO_GROW)
317 gv_grow_complete(p, pbp);
318 else
319 g_io_deliver(pbp, pbp->bio_error);
320 }
321 }
322
323 /*
324 * Handle a completed request to a RAID-5 plex.
325 */
326 void
327 gv_plex_raid5_done(struct gv_plex *p, struct bio *bp)
328 {
329 struct gv_softc *sc;
330 struct bio *cbp, *pbp;
331 struct gv_bioq *bq, *bq2;
332 struct gv_raid5_packet *wp;
333 off_t completed;
334 int i;
335
336 completed = 0;
337 sc = p->vinumconf;
338 wp = bp->bio_caller2;
339
340 switch (bp->bio_parent->bio_cmd) {
341 case BIO_READ:
342 if (wp == NULL) {
343 completed = bp->bio_completed;
344 break;
345 }
346
347 TAILQ_FOREACH_SAFE(bq, &wp->bits, queue, bq2) {
348 if (bq->bp != bp)
349 continue;
350 TAILQ_REMOVE(&wp->bits, bq, queue);
351 g_free(bq);
352 for (i = 0; i < wp->length; i++)
353 wp->data[i] ^= bp->bio_data[i];
354 break;
355 }
356 if (TAILQ_EMPTY(&wp->bits)) {
357 completed = wp->length;
358 if (wp->lockbase != -1) {
359 TAILQ_REMOVE(&p->packets, wp, list);
360 /* Bring the waiting bios back into the game. */
361 pbp = bioq_takefirst(p->wqueue);
362 while (pbp != NULL) {
363 gv_post_bio(sc, pbp);
364 pbp = bioq_takefirst(p->wqueue);
365 }
366 }
367 g_free(wp);
368 }
369
370 break;
371
372 case BIO_WRITE:
373 /* XXX can this ever happen? */
374 if (wp == NULL) {
375 completed = bp->bio_completed;
376 break;
377 }
378
379 /* Check if we need to handle parity data. */
380 TAILQ_FOREACH_SAFE(bq, &wp->bits, queue, bq2) {
381 if (bq->bp != bp)
382 continue;
383 TAILQ_REMOVE(&wp->bits, bq, queue);
384 g_free(bq);
385 cbp = wp->parity;
386 if (cbp != NULL) {
387 for (i = 0; i < wp->length; i++)
388 cbp->bio_data[i] ^= bp->bio_data[i];
389 }
390 break;
391 }
392
393 /* Handle parity data. */
394 if (TAILQ_EMPTY(&wp->bits)) {
395 if (bp->bio_parent->bio_pflags & GV_BIO_CHECK)
396 i = gv_check_parity(p, bp, wp);
397 else
398 i = gv_normal_parity(p, bp, wp);
399
400 /* All of our sub-requests have finished. */
401 if (i) {
402 completed = wp->length;
403 TAILQ_REMOVE(&p->packets, wp, list);
404 /* Bring the waiting bios back into the game. */
405 pbp = bioq_takefirst(p->wqueue);
406 while (pbp != NULL) {
407 gv_post_bio(sc, pbp);
408 pbp = bioq_takefirst(p->wqueue);
409 }
410 g_free(wp);
411 }
412 }
413
414 break;
415 }
416
417 pbp = bp->bio_parent;
418 if (pbp->bio_error == 0)
419 pbp->bio_error = bp->bio_error;
420 pbp->bio_completed += completed;
421
422 /* When the original request is finished, we deliver it. */
423 pbp->bio_inbed++;
424 if (pbp->bio_inbed == pbp->bio_children) {
425 /* Hand it over for checking or delivery. */
426 if (pbp->bio_cmd == BIO_WRITE &&
427 (pbp->bio_pflags & GV_BIO_CHECK)) {
428 gv_parity_complete(p, pbp);
429 } else if (pbp->bio_cmd == BIO_WRITE &&
430 (pbp->bio_pflags & GV_BIO_REBUILD)) {
431 gv_rebuild_complete(p, pbp);
432 } else if (pbp->bio_pflags & GV_BIO_INIT) {
433 gv_init_complete(p, pbp);
434 } else if (pbp->bio_pflags & GV_BIO_SYNCREQ) {
435 gv_sync_complete(p, pbp);
436 } else if (pbp->bio_pflags & GV_BIO_GROW) {
437 gv_grow_complete(p, pbp);
438 } else {
439 g_io_deliver(pbp, pbp->bio_error);
440 }
441 }
442
443 /* Clean up what we allocated. */
444 if (bp->bio_cflags & GV_BIO_MALLOC)
445 g_free(bp->bio_data);
446 g_destroy_bio(bp);
447 }
448
449 static int
450 gv_check_parity(struct gv_plex *p, struct bio *bp, struct gv_raid5_packet *wp)
451 {
452 struct bio *pbp;
453 struct gv_sd *s;
454 int err, finished, i;
455
456 err = 0;
457 finished = 1;
458
459 if (wp->waiting != NULL) {
460 pbp = wp->waiting;
461 wp->waiting = NULL;
462 s = pbp->bio_caller1;
463 g_io_request(pbp, s->drive_sc->consumer);
464 finished = 0;
465
466 } else if (wp->parity != NULL) {
467 pbp = wp->parity;
468 wp->parity = NULL;
469
470 /* Check if the parity is correct. */
471 for (i = 0; i < wp->length; i++) {
472 if (bp->bio_data[i] != pbp->bio_data[i]) {
473 err = 1;
474 break;
475 }
476 }
477
478 /* The parity is not correct... */
479 if (err) {
480 bp->bio_parent->bio_error = EAGAIN;
481
482 /* ... but we rebuild it. */
483 if (bp->bio_parent->bio_pflags & GV_BIO_PARITY) {
484 s = pbp->bio_caller1;
485 g_io_request(pbp, s->drive_sc->consumer);
486 finished = 0;
487 }
488 }
489
490 /*
491 * Clean up the BIO we would have used for rebuilding the
492 * parity.
493 */
494 if (finished) {
495 bp->bio_parent->bio_inbed++;
496 g_destroy_bio(pbp);
497 }
498
499 }
500
501 return (finished);
502 }
503
504 static int
505 gv_normal_parity(struct gv_plex *p, struct bio *bp, struct gv_raid5_packet *wp)
506 {
507 struct bio *cbp, *pbp;
508 struct gv_sd *s;
509 int finished, i;
510
511 finished = 1;
512
513 if (wp->waiting != NULL) {
514 pbp = wp->waiting;
515 wp->waiting = NULL;
516 cbp = wp->parity;
517 for (i = 0; i < wp->length; i++)
518 cbp->bio_data[i] ^= pbp->bio_data[i];
519 s = pbp->bio_caller1;
520 g_io_request(pbp, s->drive_sc->consumer);
521 finished = 0;
522
523 } else if (wp->parity != NULL) {
524 cbp = wp->parity;
525 wp->parity = NULL;
526 s = cbp->bio_caller1;
527 g_io_request(cbp, s->drive_sc->consumer);
528 finished = 0;
529 }
530
531 return (finished);
532 }
533
534 /* Flush the queue with delayed requests. */
535 static void
536 gv_plex_flush(struct gv_plex *p)
537 {
538 struct gv_softc *sc;
539 struct bio *bp;
540
541 sc = p->vinumconf;
542 bp = bioq_takefirst(p->rqueue);
543 while (bp != NULL) {
544 gv_plex_start(p, bp);
545 bp = bioq_takefirst(p->rqueue);
546 }
547 }
548
549 static void
550 gv_post_bio(struct gv_softc *sc, struct bio *bp)
551 {
552
553 KASSERT(sc != NULL, ("NULL sc"));
554 KASSERT(bp != NULL, ("NULL bp"));
555 mtx_lock(&sc->bqueue_mtx);
556 bioq_disksort(sc->bqueue_down, bp);
557 wakeup(sc);
558 mtx_unlock(&sc->bqueue_mtx);
559 }
560
561 int
562 gv_sync_request(struct gv_plex *from, struct gv_plex *to, off_t offset,
563 off_t length, int type, caddr_t data)
564 {
565 struct gv_softc *sc;
566 struct bio *bp;
567
568 KASSERT(from != NULL, ("NULL from"));
569 KASSERT(to != NULL, ("NULL to"));
570 sc = from->vinumconf;
571 KASSERT(sc != NULL, ("NULL sc"));
572
573 bp = g_new_bio();
574 if (bp == NULL) {
575 G_VINUM_DEBUG(0, "sync from '%s' failed at offset "
576 " %jd; out of memory", from->name, offset);
577 return (ENOMEM);
578 }
579 bp->bio_length = length;
580 bp->bio_done = gv_done;
581 bp->bio_pflags |= GV_BIO_SYNCREQ;
582 bp->bio_offset = offset;
583 bp->bio_caller1 = from;
584 bp->bio_caller2 = to;
585 bp->bio_cmd = type;
586 if (data == NULL)
587 data = g_malloc(length, M_WAITOK);
588 bp->bio_pflags |= GV_BIO_MALLOC; /* Free on the next run. */
589 bp->bio_data = data;
590
591 /* Send down next. */
592 gv_post_bio(sc, bp);
593 //gv_plex_start(from, bp);
594 return (0);
595 }
596
597 /*
598 * Handle a finished plex sync bio.
599 */
600 int
601 gv_sync_complete(struct gv_plex *to, struct bio *bp)
602 {
603 struct gv_plex *from, *p;
604 struct gv_sd *s;
605 struct gv_volume *v;
606 struct gv_softc *sc;
607 off_t offset;
608 int err;
609
610 g_topology_assert_not();
611
612 err = 0;
613 KASSERT(to != NULL, ("NULL to"));
614 KASSERT(bp != NULL, ("NULL bp"));
615 from = bp->bio_caller2;
616 KASSERT(from != NULL, ("NULL from"));
617 v = to->vol_sc;
618 KASSERT(v != NULL, ("NULL v"));
619 sc = v->vinumconf;
620 KASSERT(sc != NULL, ("NULL sc"));
621
622 /* If it was a read, write it. */
623 if (bp->bio_cmd == BIO_READ) {
624 err = gv_sync_request(from, to, bp->bio_offset, bp->bio_length,
625 BIO_WRITE, bp->bio_data);
626 /* If it was a write, read the next one. */
627 } else if (bp->bio_cmd == BIO_WRITE) {
628 if (bp->bio_pflags & GV_BIO_MALLOC)
629 g_free(bp->bio_data);
630 to->synced += bp->bio_length;
631 /* If we're finished, clean up. */
632 if (bp->bio_offset + bp->bio_length >= from->size) {
633 G_VINUM_DEBUG(1, "syncing of %s from %s completed",
634 to->name, from->name);
635 /* Update our state. */
636 LIST_FOREACH(s, &to->subdisks, in_plex)
637 gv_set_sd_state(s, GV_SD_UP, 0);
638 gv_update_plex_state(to);
639 to->flags &= ~GV_PLEX_SYNCING;
640 to->synced = 0;
641 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
642 } else {
643 offset = bp->bio_offset + bp->bio_length;
644 err = gv_sync_request(from, to, offset,
645 MIN(bp->bio_length, from->size - offset),
646 BIO_READ, NULL);
647 }
648 }
649 g_destroy_bio(bp);
650 /* Clean up if there was an error. */
651 if (err) {
652 to->flags &= ~GV_PLEX_SYNCING;
653 G_VINUM_DEBUG(0, "error syncing plexes: error code %d", err);
654 }
655
656 /* Check if all plexes are synced, and lower refcounts. */
657 g_topology_lock();
658 LIST_FOREACH(p, &v->plexes, in_volume) {
659 if (p->flags & GV_PLEX_SYNCING) {
660 g_topology_unlock();
661 return (-1);
662 }
663 }
664 /* If we came here, all plexes are synced, and we're free. */
665 gv_access(v->provider, -1, -1, 0);
666 g_topology_unlock();
667 G_VINUM_DEBUG(1, "plex sync completed");
668 gv_volume_flush(v);
669 return (0);
670 }
671
672 /*
673 * Create a new bio struct for the next grow request.
674 */
675 int
676 gv_grow_request(struct gv_plex *p, off_t offset, off_t length, int type,
677 caddr_t data)
678 {
679 struct gv_softc *sc;
680 struct bio *bp;
681
682 KASSERT(p != NULL, ("gv_grow_request: NULL p"));
683 sc = p->vinumconf;
684 KASSERT(sc != NULL, ("gv_grow_request: NULL sc"));
685
686 bp = g_new_bio();
687 if (bp == NULL) {
688 G_VINUM_DEBUG(0, "grow of %s failed creating bio: "
689 "out of memory", p->name);
690 return (ENOMEM);
691 }
692
693 bp->bio_cmd = type;
694 bp->bio_done = gv_done;
695 bp->bio_error = 0;
696 bp->bio_caller1 = p;
697 bp->bio_offset = offset;
698 bp->bio_length = length;
699 bp->bio_pflags |= GV_BIO_GROW;
700 if (data == NULL)
701 data = g_malloc(length, M_WAITOK);
702 bp->bio_pflags |= GV_BIO_MALLOC;
703 bp->bio_data = data;
704
705 gv_post_bio(sc, bp);
706 //gv_plex_start(p, bp);
707 return (0);
708 }
709
710 /*
711 * Finish handling of a bio to a growing plex.
712 */
713 void
714 gv_grow_complete(struct gv_plex *p, struct bio *bp)
715 {
716 struct gv_softc *sc;
717 struct gv_sd *s;
718 struct gv_volume *v;
719 off_t origsize, offset;
720 int sdcount, err;
721
722 v = p->vol_sc;
723 KASSERT(v != NULL, ("gv_grow_complete: NULL v"));
724 sc = v->vinumconf;
725 KASSERT(sc != NULL, ("gv_grow_complete: NULL sc"));
726 err = 0;
727
728 /* If it was a read, write it. */
729 if (bp->bio_cmd == BIO_READ) {
730 p->synced += bp->bio_length;
731 err = gv_grow_request(p, bp->bio_offset, bp->bio_length,
732 BIO_WRITE, bp->bio_data);
733 /* If it was a write, read next. */
734 } else if (bp->bio_cmd == BIO_WRITE) {
735 if (bp->bio_pflags & GV_BIO_MALLOC)
736 g_free(bp->bio_data);
737
738 /* Find the real size of the plex. */
739 sdcount = gv_sdcount(p, 1);
740 s = LIST_FIRST(&p->subdisks);
741 KASSERT(s != NULL, ("NULL s"));
742 origsize = (s->size * (sdcount - 1));
743 if (bp->bio_offset + bp->bio_length >= origsize) {
744 G_VINUM_DEBUG(1, "growing of %s completed", p->name);
745 p->flags &= ~GV_PLEX_GROWING;
746 LIST_FOREACH(s, &p->subdisks, in_plex) {
747 s->flags &= ~GV_SD_GROW;
748 gv_set_sd_state(s, GV_SD_UP, 0);
749 }
750 p->size = gv_plex_size(p);
751 gv_update_vol_size(v, gv_vol_size(v));
752 gv_set_plex_state(p, GV_PLEX_UP, 0);
753 g_topology_lock();
754 gv_access(v->provider, -1, -1, 0);
755 g_topology_unlock();
756 p->synced = 0;
757 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
758 /* Issue delayed requests. */
759 gv_plex_flush(p);
760 } else {
761 offset = bp->bio_offset + bp->bio_length;
762 err = gv_grow_request(p, offset,
763 MIN(bp->bio_length, origsize - offset),
764 BIO_READ, NULL);
765 }
766 }
767 g_destroy_bio(bp);
768
769 if (err) {
770 p->flags &= ~GV_PLEX_GROWING;
771 G_VINUM_DEBUG(0, "error growing plex: error code %d", err);
772 }
773 }
774
775
776 /*
777 * Create an initialization BIO and send it off to the consumer. Assume that
778 * we're given initialization data as parameter.
779 */
780 void
781 gv_init_request(struct gv_sd *s, off_t start, caddr_t data, off_t length)
782 {
783 struct gv_drive *d;
784 struct g_consumer *cp;
785 struct bio *bp, *cbp;
786
787 KASSERT(s != NULL, ("gv_init_request: NULL s"));
788 d = s->drive_sc;
789 KASSERT(d != NULL, ("gv_init_request: NULL d"));
790 cp = d->consumer;
791 KASSERT(cp != NULL, ("gv_init_request: NULL cp"));
792
793 bp = g_new_bio();
794 if (bp == NULL) {
795 G_VINUM_DEBUG(0, "subdisk '%s' init: write failed at offset %jd"
796 " (drive offset %jd); out of memory", s->name,
797 (intmax_t)s->initialized, (intmax_t)start);
798 return; /* XXX: Error codes. */
799 }
800 bp->bio_cmd = BIO_WRITE;
801 bp->bio_data = data;
802 bp->bio_done = gv_done;
803 bp->bio_error = 0;
804 bp->bio_length = length;
805 bp->bio_pflags |= GV_BIO_INIT;
806 bp->bio_offset = start;
807 bp->bio_caller1 = s;
808
809 /* Then ofcourse, we have to clone it. */
810 cbp = g_clone_bio(bp);
811 if (cbp == NULL) {
812 G_VINUM_DEBUG(0, "subdisk '%s' init: write failed at offset %jd"
813 " (drive offset %jd); out of memory", s->name,
814 (intmax_t)s->initialized, (intmax_t)start);
815 return; /* XXX: Error codes. */
816 }
817 cbp->bio_done = gv_done;
818 cbp->bio_caller1 = s;
819 /* Send it off to the consumer. */
820 g_io_request(cbp, cp);
821 }
822
823 /*
824 * Handle a finished initialization BIO.
825 */
826 void
827 gv_init_complete(struct gv_plex *p, struct bio *bp)
828 {
829 struct gv_softc *sc;
830 struct gv_drive *d;
831 struct g_consumer *cp;
832 struct gv_sd *s;
833 off_t start, length;
834 caddr_t data;
835 int error;
836
837 s = bp->bio_caller1;
838 start = bp->bio_offset;
839 length = bp->bio_length;
840 error = bp->bio_error;
841 data = bp->bio_data;
842
843 KASSERT(s != NULL, ("gv_init_complete: NULL s"));
844 d = s->drive_sc;
845 KASSERT(d != NULL, ("gv_init_complete: NULL d"));
846 cp = d->consumer;
847 KASSERT(cp != NULL, ("gv_init_complete: NULL cp"));
848 sc = p->vinumconf;
849 KASSERT(sc != NULL, ("gv_init_complete: NULL sc"));
850
851 g_destroy_bio(bp);
852
853 /*
854 * First we need to find out if it was okay, and abort if it's not.
855 * Then we need to free previous buffers, find out the correct subdisk,
856 * as well as getting the correct starting point and length of the BIO.
857 */
858 if (start >= s->drive_offset + s->size) {
859 /* Free the data we initialized. */
860 if (data != NULL)
861 g_free(data);
862 g_topology_assert_not();
863 g_topology_lock();
864 g_access(cp, 0, -1, 0);
865 g_topology_unlock();
866 if (error) {
867 gv_set_sd_state(s, GV_SD_STALE, GV_SETSTATE_FORCE |
868 GV_SETSTATE_CONFIG);
869 } else {
870 gv_set_sd_state(s, GV_SD_UP, GV_SETSTATE_CONFIG);
871 s->initialized = 0;
872 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
873 G_VINUM_DEBUG(1, "subdisk '%s' init: finished "
874 "successfully", s->name);
875 }
876 return;
877 }
878 s->initialized += length;
879 start += length;
880 gv_init_request(s, start, data, length);
881 }
882
883 /*
884 * Create a new bio struct for the next parity rebuild. Used both by internal
885 * rebuild of degraded plexes as well as user initiated rebuilds/checks.
886 */
887 void
888 gv_parity_request(struct gv_plex *p, int flags, off_t offset)
889 {
890 struct gv_softc *sc;
891 struct bio *bp;
892
893 KASSERT(p != NULL, ("gv_parity_request: NULL p"));
894 sc = p->vinumconf;
895 KASSERT(sc != NULL, ("gv_parity_request: NULL sc"));
896
897 bp = g_new_bio();
898 if (bp == NULL) {
899 G_VINUM_DEBUG(0, "rebuild of %s failed creating bio: "
900 "out of memory", p->name);
901 return;
902 }
903
904 bp->bio_cmd = BIO_WRITE;
905 bp->bio_done = gv_done;
906 bp->bio_error = 0;
907 bp->bio_length = p->stripesize;
908 bp->bio_caller1 = p;
909
910 /*
911 * Check if it's a rebuild of a degraded plex or a user request of
912 * parity rebuild.
913 */
914 if (flags & GV_BIO_REBUILD)
915 bp->bio_data = g_malloc(GV_DFLT_SYNCSIZE, M_WAITOK);
916 else if (flags & GV_BIO_CHECK)
917 bp->bio_data = g_malloc(p->stripesize, M_WAITOK | M_ZERO);
918 else {
919 G_VINUM_DEBUG(0, "invalid flags given in rebuild");
920 return;
921 }
922
923 bp->bio_pflags = flags;
924 bp->bio_pflags |= GV_BIO_MALLOC;
925
926 /* We still have more parity to build. */
927 bp->bio_offset = offset;
928 gv_post_bio(sc, bp);
929 //gv_plex_start(p, bp); /* Send it down to the plex. */
930 }
931
932 /*
933 * Handle a finished parity write.
934 */
935 void
936 gv_parity_complete(struct gv_plex *p, struct bio *bp)
937 {
938 struct gv_softc *sc;
939 int error, flags;
940
941 error = bp->bio_error;
942 flags = bp->bio_pflags;
943 flags &= ~GV_BIO_MALLOC;
944
945 sc = p->vinumconf;
946 KASSERT(sc != NULL, ("gv_parity_complete: NULL sc"));
947
948 /* Clean up what we allocated. */
949 if (bp->bio_pflags & GV_BIO_MALLOC)
950 g_free(bp->bio_data);
951 g_destroy_bio(bp);
952
953 if (error == EAGAIN) {
954 G_VINUM_DEBUG(0, "parity incorrect at offset 0x%jx",
955 (intmax_t)p->synced);
956 }
957
958 /* Any error is fatal, except EAGAIN when we're rebuilding. */
959 if (error && !(error == EAGAIN && (flags & GV_BIO_PARITY))) {
960 /* Make sure we don't have the lock. */
961 g_topology_assert_not();
962 g_topology_lock();
963 gv_access(p->vol_sc->provider, -1, -1, 0);
964 g_topology_unlock();
965 G_VINUM_DEBUG(0, "parity check on %s failed at 0x%jx "
966 "errno %d", p->name, (intmax_t)p->synced, error);
967 return;
968 } else {
969 p->synced += p->stripesize;
970 }
971
972 if (p->synced >= p->size) {
973 /* Make sure we don't have the lock. */
974 g_topology_assert_not();
975 g_topology_lock();
976 gv_access(p->vol_sc->provider, -1, -1, 0);
977 g_topology_unlock();
978 /* We're finished. */
979 G_VINUM_DEBUG(1, "parity operation on %s finished", p->name);
980 p->synced = 0;
981 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
982 return;
983 }
984
985 /* Send down next. It will determine if we need to itself. */
986 gv_parity_request(p, flags, p->synced);
987 }
988
989 /*
990 * Handle a finished plex rebuild bio.
991 */
992 void
993 gv_rebuild_complete(struct gv_plex *p, struct bio *bp)
994 {
995 struct gv_softc *sc;
996 struct gv_sd *s;
997 int error, flags;
998 off_t offset;
999
1000 error = bp->bio_error;
1001 flags = bp->bio_pflags;
1002 offset = bp->bio_offset;
1003 flags &= ~GV_BIO_MALLOC;
1004 sc = p->vinumconf;
1005 KASSERT(sc != NULL, ("gv_rebuild_complete: NULL sc"));
1006
1007 /* Clean up what we allocated. */
1008 if (bp->bio_pflags & GV_BIO_MALLOC)
1009 g_free(bp->bio_data);
1010 g_destroy_bio(bp);
1011
1012 if (error) {
1013 g_topology_assert_not();
1014 g_topology_lock();
1015 gv_access(p->vol_sc->provider, -1, -1, 0);
1016 g_topology_unlock();
1017
1018 G_VINUM_DEBUG(0, "rebuild of %s failed at offset %jd errno: %d",
1019 p->name, (intmax_t)offset, error);
1020 p->flags &= ~GV_PLEX_REBUILDING;
1021 p->synced = 0;
1022 gv_plex_flush(p); /* Flush out remaining rebuild BIOs. */
1023 return;
1024 }
1025
1026 offset += (p->stripesize * (gv_sdcount(p, 1) - 1));
1027 if (offset >= p->size) {
1028 /* We're finished. */
1029 g_topology_assert_not();
1030 g_topology_lock();
1031 gv_access(p->vol_sc->provider, -1, -1, 0);
1032 g_topology_unlock();
1033
1034 G_VINUM_DEBUG(1, "rebuild of %s finished", p->name);
1035 gv_save_config(p->vinumconf);
1036 p->flags &= ~GV_PLEX_REBUILDING;
1037 p->synced = 0;
1038 /* Try to up all subdisks. */
1039 LIST_FOREACH(s, &p->subdisks, in_plex)
1040 gv_update_sd_state(s);
1041 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
1042 gv_plex_flush(p); /* Flush out remaining rebuild BIOs. */
1043 return;
1044 }
1045
1046 /* Send down next. It will determine if we need to itself. */
1047 gv_parity_request(p, flags, offset);
1048 }
Cache object: a27975a7942922a27fec7ef67e2a70ae
|