2 * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
3 * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
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.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS 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 AUTHORS 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
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
36 #include <sys/limits.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/systm.h>
41 #include <geom/geom.h>
42 #include "geom/raid/g_raid.h"
43 #include "g_raid_md_if.h"
45 static MALLOC_DEFINE(M_MD_PROMISE, "md_promise_data", "GEOM_RAID Promise metadata");
47 #define PROMISE_MAX_DISKS 8
48 #define PROMISE_MAX_SUBDISKS 2
49 #define PROMISE_META_OFFSET 14
51 struct promise_raid_disk {
52 uint8_t flags; /* Subdisk status. */
53 #define PROMISE_F_VALID 0x01
54 #define PROMISE_F_ONLINE 0x02
55 #define PROMISE_F_ASSIGNED 0x04
56 #define PROMISE_F_SPARE 0x08
57 #define PROMISE_F_DUPLICATE 0x10
58 #define PROMISE_F_REDIR 0x20
59 #define PROMISE_F_DOWN 0x40
60 #define PROMISE_F_READY 0x80
62 uint8_t number; /* Position in a volume. */
63 uint8_t channel; /* ATA channel number. */
64 uint8_t device; /* ATA device number. */
65 uint64_t id __packed; /* Subdisk ID. */
68 struct promise_raid_conf {
70 #define PROMISE_MAGIC "Promise Technology, Inc."
71 #define FREEBSD_MAGIC "FreeBSD ATA driver RAID "
75 #define PROMISE_MAGIC0(x) (((uint64_t)(x.channel) << 48) | \
76 ((uint64_t)(x.device != 0) << 56))
82 #define PROMISE_I_VALID 0x00000080
84 struct promise_raid_disk disk; /* This subdisk info. */
85 uint32_t disk_offset; /* Subdisk offset. */
86 uint32_t disk_sectors; /* Subdisk size */
87 uint32_t disk_rebuild; /* Rebuild position. */
88 uint16_t generation; /* Generation number. */
89 uint8_t status; /* Volume status. */
90 #define PROMISE_S_VALID 0x01
91 #define PROMISE_S_ONLINE 0x02
92 #define PROMISE_S_INITED 0x04
93 #define PROMISE_S_READY 0x08
94 #define PROMISE_S_DEGRADED 0x10
95 #define PROMISE_S_MARKED 0x20
96 #define PROMISE_S_MIGRATING 0x40
97 #define PROMISE_S_FUNCTIONAL 0x80
99 uint8_t type; /* Voluem type. */
100 #define PROMISE_T_RAID0 0x00
101 #define PROMISE_T_RAID1 0x01
102 #define PROMISE_T_RAID3 0x02
103 #define PROMISE_T_RAID5 0x04
104 #define PROMISE_T_SPAN 0x08
105 #define PROMISE_T_JBOD 0x10
107 uint8_t total_disks; /* Disks in this volume. */
108 uint8_t stripe_shift; /* Strip size. */
109 uint8_t array_width; /* Number of RAID0 stripes. */
110 uint8_t array_number; /* Global volume number. */
111 uint32_t total_sectors; /* Volume size. */
112 uint16_t cylinders; /* Volume geometry: C. */
113 uint8_t heads; /* Volume geometry: H. */
114 uint8_t sectors; /* Volume geometry: S. */
115 uint64_t volume_id __packed; /* Volume ID, */
116 struct promise_raid_disk disks[PROMISE_MAX_DISKS];
117 /* Subdisks in this volume. */
118 char name[32]; /* Volume label. */
121 uint32_t magic_3; /* Something related to rebuild. */
122 uint64_t rebuild_lba64; /* Per-volume rebuild position. */
125 uint32_t total_sectors_high;
129 uint32_t magic_8[31];
130 uint32_t backup_time;
132 uint32_t disk_offset_high;
133 uint32_t disk_sectors_high;
134 uint32_t disk_rebuild_high;
136 uint32_t magic_11[3];
137 uint32_t filler3[284];
141 struct g_raid_md_promise_perdisk {
144 struct promise_raid_conf *pd_meta[PROMISE_MAX_SUBDISKS];
147 struct g_raid_md_promise_pervolume {
148 struct promise_raid_conf *pv_meta;
150 uint16_t pv_generation;
151 int pv_disks_present;
153 struct callout pv_start_co; /* STARTING state timer. */
156 static g_raid_md_create_t g_raid_md_create_promise;
157 static g_raid_md_taste_t g_raid_md_taste_promise;
158 static g_raid_md_event_t g_raid_md_event_promise;
159 static g_raid_md_volume_event_t g_raid_md_volume_event_promise;
160 static g_raid_md_ctl_t g_raid_md_ctl_promise;
161 static g_raid_md_write_t g_raid_md_write_promise;
162 static g_raid_md_fail_disk_t g_raid_md_fail_disk_promise;
163 static g_raid_md_free_disk_t g_raid_md_free_disk_promise;
164 static g_raid_md_free_volume_t g_raid_md_free_volume_promise;
165 static g_raid_md_free_t g_raid_md_free_promise;
167 static kobj_method_t g_raid_md_promise_methods[] = {
168 KOBJMETHOD(g_raid_md_create, g_raid_md_create_promise),
169 KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_promise),
170 KOBJMETHOD(g_raid_md_event, g_raid_md_event_promise),
171 KOBJMETHOD(g_raid_md_volume_event, g_raid_md_volume_event_promise),
172 KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_promise),
173 KOBJMETHOD(g_raid_md_write, g_raid_md_write_promise),
174 KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_promise),
175 KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_promise),
176 KOBJMETHOD(g_raid_md_free_volume, g_raid_md_free_volume_promise),
177 KOBJMETHOD(g_raid_md_free, g_raid_md_free_promise),
181 static struct g_raid_md_class g_raid_md_promise_class = {
183 g_raid_md_promise_methods,
184 sizeof(struct g_raid_md_object),
191 g_raid_md_promise_print(struct promise_raid_conf *meta)
195 if (g_raid_debug < 1)
198 printf("********* ATA Promise Metadata *********\n");
199 printf("promise_id <%.24s>\n", meta->promise_id);
200 printf("disk %02x %02x %02x %02x %016jx\n",
201 meta->disk.flags, meta->disk.number, meta->disk.channel,
202 meta->disk.device, meta->disk.id);
203 printf("disk_offset %u\n", meta->disk_offset);
204 printf("disk_sectors %u\n", meta->disk_sectors);
205 printf("disk_rebuild %u\n", meta->disk_rebuild);
206 printf("generation %u\n", meta->generation);
207 printf("status 0x%02x\n", meta->status);
208 printf("type %u\n", meta->type);
209 printf("total_disks %u\n", meta->total_disks);
210 printf("stripe_shift %u\n", meta->stripe_shift);
211 printf("array_width %u\n", meta->array_width);
212 printf("array_number %u\n", meta->array_number);
213 printf("total_sectors %u\n", meta->total_sectors);
214 printf("cylinders %u\n", meta->cylinders);
215 printf("heads %u\n", meta->heads);
216 printf("sectors %u\n", meta->sectors);
217 printf("volume_id 0x%016jx\n", meta->volume_id);
219 for (i = 0; i < PROMISE_MAX_DISKS; i++ ) {
220 printf(" %02x %02x %02x %02x %016jx\n",
221 meta->disks[i].flags, meta->disks[i].number,
222 meta->disks[i].channel, meta->disks[i].device,
225 printf("name <%.32s>\n", meta->name);
226 printf("magic_3 0x%08x\n", meta->magic_3);
227 printf("rebuild_lba64 %ju\n", meta->rebuild_lba64);
228 printf("magic_4 0x%08x\n", meta->magic_4);
229 printf("magic_5 0x%08x\n", meta->magic_5);
230 printf("total_sectors_high 0x%08x\n", meta->total_sectors_high);
231 printf("sector_size %u\n", meta->sector_size);
232 printf("backup_time %d\n", meta->backup_time);
233 printf("disk_offset_high 0x%08x\n", meta->disk_offset_high);
234 printf("disk_sectors_high 0x%08x\n", meta->disk_sectors_high);
235 printf("disk_rebuild_high 0x%08x\n", meta->disk_rebuild_high);
236 printf("=================================================\n");
239 static struct promise_raid_conf *
240 promise_meta_copy(struct promise_raid_conf *meta)
242 struct promise_raid_conf *nmeta;
244 nmeta = malloc(sizeof(*nmeta), M_MD_PROMISE, M_WAITOK);
245 memcpy(nmeta, meta, sizeof(*nmeta));
250 promise_meta_find_disk(struct promise_raid_conf *meta, uint64_t id)
254 for (pos = 0; pos < meta->total_disks; pos++) {
255 if (meta->disks[pos].id == id)
262 promise_meta_unused_range(struct promise_raid_conf **metaarr, int nsd,
263 off_t sectors, off_t *off, off_t *size)
265 off_t coff, csize, tmp;
275 for (j = 0; j < nsd; j++) {
276 tmp = ((off_t)metaarr[j]->disk_offset_high << 32) +
277 metaarr[j]->disk_offset;
279 csize = MIN(csize, tmp - coff);
287 coff = ((off_t)metaarr[i]->disk_offset_high << 32) +
288 metaarr[i]->disk_offset +
289 ((off_t)metaarr[i]->disk_sectors_high << 32) +
290 metaarr[i]->disk_sectors;
291 csize = sectors - coff;
294 return ((*size > 0) ? 1 : 0);
298 promise_meta_translate_disk(struct g_raid_volume *vol, int md_disk_pos)
302 if (md_disk_pos >= 0 && vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
303 width = vol->v_disks_count / 2;
304 disk_pos = (md_disk_pos / width) +
305 (md_disk_pos % width) * width;
307 disk_pos = md_disk_pos;
312 promise_meta_get_name(struct promise_raid_conf *meta, char *buf)
316 strncpy(buf, meta->name, 32);
318 for (i = 31; i >= 0; i--) {
326 promise_meta_put_name(struct promise_raid_conf *meta, char *buf)
329 memset(meta->name, 0x20, 32);
330 memcpy(meta->name, buf, MIN(strlen(buf), 32));
334 promise_meta_read(struct g_consumer *cp, struct promise_raid_conf **metaarr)
336 struct g_provider *pp;
337 struct promise_raid_conf *meta;
339 int error, i, subdisks;
340 uint32_t checksum, *ptr;
345 if (pp->sectorsize * 4 > MAXPHYS) {
346 G_RAID_DEBUG(1, "%s: Blocksize is too big.", pp->name);
350 /* Read metadata block. */
351 buf = g_read_data(cp, pp->mediasize - pp->sectorsize *
352 (63 - subdisks * PROMISE_META_OFFSET),
353 pp->sectorsize * 4, &error);
355 G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
359 meta = (struct promise_raid_conf *)buf;
361 /* Check if this is an Promise RAID struct */
362 if (strncmp(meta->promise_id, PROMISE_MAGIC, strlen(PROMISE_MAGIC)) &&
363 strncmp(meta->promise_id, FREEBSD_MAGIC, strlen(FREEBSD_MAGIC))) {
366 "Promise signature check failed on %s", pp->name);
370 meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK);
371 memcpy(meta, buf, MIN(sizeof(*meta), pp->sectorsize * 4));
374 /* Check metadata checksum. */
375 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
377 if (checksum != meta->checksum) {
378 G_RAID_DEBUG(1, "Promise checksum check failed on %s", pp->name);
379 free(meta, M_MD_PROMISE);
383 if ((meta->integrity & PROMISE_I_VALID) == 0) {
384 G_RAID_DEBUG(1, "Promise metadata is invalid on %s", pp->name);
385 free(meta, M_MD_PROMISE);
389 if (meta->total_disks > PROMISE_MAX_DISKS) {
390 G_RAID_DEBUG(1, "Wrong number of disks on %s (%d)",
391 pp->name, meta->total_disks);
392 free(meta, M_MD_PROMISE);
396 /* Remove filler garbage from fields used in newer metadata. */
397 if (meta->disk_offset_high == 0x8b8c8d8e &&
398 meta->disk_sectors_high == 0x8788898a &&
399 meta->disk_rebuild_high == 0x83848586) {
400 meta->disk_offset_high = 0;
401 meta->disk_sectors_high = 0;
402 if (meta->disk_rebuild == UINT32_MAX)
403 meta->disk_rebuild_high = UINT32_MAX;
405 meta->disk_rebuild_high = 0;
406 if (meta->total_sectors_high == 0x15161718) {
407 meta->total_sectors_high = 0;
408 meta->backup_time = 0;
409 if (meta->rebuild_lba64 == 0x2122232425262728)
410 meta->rebuild_lba64 = UINT64_MAX;
413 if (meta->sector_size < 1 || meta->sector_size > 8)
414 meta->sector_size = 1;
416 /* Save this part and look for next. */
420 if (subdisks < PROMISE_MAX_SUBDISKS)
427 promise_meta_write(struct g_consumer *cp,
428 struct promise_raid_conf **metaarr, int nsd)
430 struct g_provider *pp;
431 struct promise_raid_conf *meta;
434 int error, i, subdisk, fake;
435 uint32_t checksum, *ptr;
441 buf = malloc(pp->sectorsize * 4, M_MD_PROMISE, M_WAITOK | M_ZERO);
444 meta = metaarr[subdisk];
445 } else if (!fake && promise_meta_unused_range(metaarr, nsd,
446 cp->provider->mediasize / cp->provider->sectorsize,
448 /* Optionally add record for unused space. */
449 meta = (struct promise_raid_conf *)buf;
450 memcpy(&meta->promise_id[0], PROMISE_MAGIC,
451 sizeof(PROMISE_MAGIC) - 1);
452 meta->dummy_0 = 0x00020000;
453 meta->integrity = PROMISE_I_VALID;
454 meta->disk.flags = PROMISE_F_ONLINE | PROMISE_F_VALID;
455 meta->disk.number = 0xff;
456 arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
457 meta->disk_offset_high = off >> 32;
458 meta->disk_offset = (uint32_t)off;
459 meta->disk_sectors_high = size >> 32;
460 meta->disk_sectors = (uint32_t)size;
461 meta->disk_rebuild_high = UINT32_MAX;
462 meta->disk_rebuild = UINT32_MAX;
466 /* Recalculate checksum for case if metadata were changed. */
468 for (checksum = 0, ptr = (uint32_t *)meta, i = 0; i < 511; i++)
470 meta->checksum = checksum;
471 memcpy(buf, meta, MIN(pp->sectorsize * 4, sizeof(*meta)));
473 error = g_write_data(cp, pp->mediasize - pp->sectorsize *
474 (63 - subdisk * PROMISE_META_OFFSET),
475 buf, pp->sectorsize * 4);
477 G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
480 free(buf, M_MD_PROMISE);
483 if (subdisk < PROMISE_MAX_SUBDISKS)
490 promise_meta_erase(struct g_consumer *cp)
492 struct g_provider *pp;
497 buf = malloc(4 * pp->sectorsize, M_MD_PROMISE, M_WAITOK | M_ZERO);
498 for (subdisk = 0; subdisk < PROMISE_MAX_SUBDISKS; subdisk++) {
499 error = g_write_data(cp, pp->mediasize - pp->sectorsize *
500 (63 - subdisk * PROMISE_META_OFFSET),
501 buf, 4 * pp->sectorsize);
503 G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
507 free(buf, M_MD_PROMISE);
512 promise_meta_write_spare(struct g_consumer *cp)
514 struct promise_raid_conf *meta;
518 meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
519 memcpy(&meta->promise_id[0], PROMISE_MAGIC, sizeof(PROMISE_MAGIC) - 1);
520 meta->dummy_0 = 0x00020000;
521 meta->integrity = PROMISE_I_VALID;
522 meta->disk.flags = PROMISE_F_SPARE | PROMISE_F_ONLINE | PROMISE_F_VALID;
523 meta->disk.number = 0xff;
524 arc4rand(&meta->disk.id, sizeof(meta->disk.id), 0);
525 tmp = cp->provider->mediasize / cp->provider->sectorsize - 131072;
526 meta->disk_sectors_high = tmp >> 32;
527 meta->disk_sectors = (uint32_t)tmp;
528 meta->disk_rebuild_high = UINT32_MAX;
529 meta->disk_rebuild = UINT32_MAX;
530 error = promise_meta_write(cp, &meta, 1);
531 free(meta, M_MD_PROMISE);
535 static struct g_raid_volume *
536 g_raid_md_promise_get_volume(struct g_raid_softc *sc, uint64_t id)
538 struct g_raid_volume *vol;
539 struct g_raid_md_promise_pervolume *pv;
541 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
550 g_raid_md_promise_purge_volumes(struct g_raid_softc *sc)
552 struct g_raid_volume *vol, *tvol;
553 struct g_raid_md_promise_pervolume *pv;
557 TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tvol) {
559 if (!pv->pv_started || vol->v_stopping)
561 for (i = 0; i < vol->v_disks_count; i++) {
562 if (vol->v_subdisks[i].sd_state != G_RAID_SUBDISK_S_NONE)
565 if (i >= vol->v_disks_count) {
566 g_raid_destroy_volume(vol);
574 g_raid_md_promise_purge_disks(struct g_raid_softc *sc)
576 struct g_raid_disk *disk, *tdisk;
577 struct g_raid_volume *vol;
578 struct g_raid_md_promise_perdisk *pd;
582 TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tdisk) {
583 if (disk->d_state == G_RAID_DISK_S_SPARE)
585 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
587 /* Scan for deleted volumes. */
588 for (i = 0; i < pd->pd_subdisks; ) {
589 vol = g_raid_md_promise_get_volume(sc,
590 pd->pd_meta[i]->volume_id);
591 if (vol != NULL && !vol->v_stopping) {
595 free(pd->pd_meta[i], M_MD_PROMISE);
596 for (j = i; j < pd->pd_subdisks - 1; j++)
597 pd->pd_meta[j] = pd->pd_meta[j + 1];
598 pd->pd_meta[pd->pd_subdisks - 1] = NULL;
603 /* If there is no metadata left - erase and delete disk. */
604 if (pd->pd_subdisks == 0) {
605 promise_meta_erase(disk->d_consumer);
606 g_raid_destroy_disk(disk);
614 g_raid_md_promise_supported(int level, int qual, int disks, int force)
617 if (disks > PROMISE_MAX_DISKS)
620 case G_RAID_VOLUME_RL_RAID0:
623 if (!force && disks < 2)
626 case G_RAID_VOLUME_RL_RAID1:
629 if (!force && (disks != 2))
632 case G_RAID_VOLUME_RL_RAID1E:
637 if (!force && (disks != 4))
640 case G_RAID_VOLUME_RL_SINGLE:
644 case G_RAID_VOLUME_RL_CONCAT:
648 case G_RAID_VOLUME_RL_RAID5:
651 if (qual != G_RAID_VOLUME_RLQ_R5LA)
657 if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
663 g_raid_md_promise_start_disk(struct g_raid_disk *disk, int sdn,
664 struct g_raid_volume *vol)
666 struct g_raid_softc *sc;
667 struct g_raid_subdisk *sd;
668 struct g_raid_md_promise_perdisk *pd;
669 struct g_raid_md_promise_pervolume *pv;
670 struct promise_raid_conf *meta;
671 off_t eoff, esize, size;
672 int disk_pos, md_disk_pos, i, resurrection = 0;
675 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
681 /* Find disk position in metadata by its serial. */
682 md_disk_pos = promise_meta_find_disk(meta, pd->pd_meta[sdn]->disk.id);
683 /* For RAID0+1 we need to translate order. */
684 disk_pos = promise_meta_translate_disk(vol, md_disk_pos);
690 G_RAID_DEBUG1(1, sc, "Disk %s is not part of the volume %s",
691 g_raid_get_diskname(disk), vol->v_name);
692 /* Failed stale disk is useless for us. */
694 pd->pd_meta[sdn]->disk.flags & PROMISE_F_DOWN) {
695 g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE_FAILED);
698 /* If we were given specific metadata subdisk - erase it. */
700 free(pd->pd_meta[sdn], M_MD_PROMISE);
701 for (i = sdn; i < pd->pd_subdisks - 1; i++)
702 pd->pd_meta[i] = pd->pd_meta[i + 1];
703 pd->pd_meta[pd->pd_subdisks - 1] = NULL;
706 /* If we are in the start process, that's all for now. */
710 * If we have already started - try to get use of the disk.
711 * Try to replace OFFLINE disks first, then FAILED.
713 promise_meta_unused_range(pd->pd_meta, pd->pd_subdisks,
714 disk->d_consumer->provider->mediasize /
715 disk->d_consumer->provider->sectorsize,
718 G_RAID_DEBUG1(1, sc, "No free space on disk %s",
719 g_raid_get_diskname(disk));
723 for (i = 0; i < vol->v_disks_count; i++) {
724 sd = &vol->v_subdisks[i];
725 if (sd->sd_state != G_RAID_SUBDISK_S_NONE)
727 if (sd->sd_state <= G_RAID_SUBDISK_S_FAILED &&
729 vol->v_subdisks[i].sd_state < sd->sd_state))
733 vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT &&
734 (off_t)esize * 512 < size) {
735 G_RAID_DEBUG1(1, sc, "Disk %s free space "
736 "is too small (%ju < %ju)",
737 g_raid_get_diskname(disk),
738 (off_t)esize * 512, size);
742 if (vol->v_raid_level != G_RAID_VOLUME_RL_CONCAT)
744 /* For RAID0+1 we need to translate order. */
745 md_disk_pos = promise_meta_translate_disk(vol, disk_pos);
748 if (pd->pd_subdisks == 0) {
749 g_raid_change_disk_state(disk,
750 G_RAID_DISK_S_SPARE);
754 G_RAID_DEBUG1(1, sc, "Disk %s takes pos %d in the volume %s",
755 g_raid_get_diskname(disk), disk_pos, vol->v_name);
759 sd = &vol->v_subdisks[disk_pos];
761 if (resurrection && sd->sd_disk != NULL) {
762 g_raid_change_disk_state(sd->sd_disk,
763 G_RAID_DISK_S_STALE_FAILED);
764 TAILQ_REMOVE(&sd->sd_disk->d_subdisks,
767 vol->v_subdisks[disk_pos].sd_disk = disk;
768 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
770 /* Welcome the new disk. */
772 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
773 else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN)
774 g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
776 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
779 sd->sd_offset = (off_t)eoff * 512;
780 sd->sd_size = (off_t)esize * 512;
782 sd->sd_offset = (((off_t)pd->pd_meta[sdn]->disk_offset_high
783 << 32) + pd->pd_meta[sdn]->disk_offset) * 512;
784 sd->sd_size = (((off_t)pd->pd_meta[sdn]->disk_sectors_high
785 << 32) + pd->pd_meta[sdn]->disk_sectors) * 512;
789 /* Stale disk, almost same as new. */
790 g_raid_change_subdisk_state(sd,
791 G_RAID_SUBDISK_S_NEW);
792 } else if (meta->disks[md_disk_pos].flags & PROMISE_F_DOWN) {
794 g_raid_change_subdisk_state(sd,
795 G_RAID_SUBDISK_S_FAILED);
796 } else if (meta->disks[md_disk_pos].flags & PROMISE_F_REDIR) {
797 /* Rebuilding disk. */
798 g_raid_change_subdisk_state(sd,
799 G_RAID_SUBDISK_S_REBUILD);
800 if (pd->pd_meta[sdn]->generation != meta->generation)
801 sd->sd_rebuild_pos = 0;
804 (((off_t)pd->pd_meta[sdn]->disk_rebuild_high << 32) +
805 pd->pd_meta[sdn]->disk_rebuild) * 512;
807 } else if (!(meta->disks[md_disk_pos].flags & PROMISE_F_ONLINE)) {
808 /* Rebuilding disk. */
809 g_raid_change_subdisk_state(sd,
810 G_RAID_SUBDISK_S_NEW);
811 } else if (pd->pd_meta[sdn]->generation != meta->generation ||
812 (meta->status & PROMISE_S_MARKED)) {
813 /* Stale disk or dirty volume (unclean shutdown). */
814 g_raid_change_subdisk_state(sd,
815 G_RAID_SUBDISK_S_STALE);
817 /* Up to date disk. */
818 g_raid_change_subdisk_state(sd,
819 G_RAID_SUBDISK_S_ACTIVE);
821 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
822 G_RAID_EVENT_SUBDISK);
824 return (resurrection);
828 g_raid_md_promise_refill(struct g_raid_softc *sc)
830 struct g_raid_volume *vol;
831 struct g_raid_subdisk *sd;
832 struct g_raid_disk *disk;
833 struct g_raid_md_object *md;
834 struct g_raid_md_promise_perdisk *pd;
835 struct g_raid_md_promise_pervolume *pv;
836 int update, updated, i, bad;
841 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
843 if (!pv->pv_started || vol->v_stopping)
846 /* Search for subdisk that needs replacement. */
848 for (i = 0; i < vol->v_disks_count; i++) {
849 sd = &vol->v_subdisks[i];
850 if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
851 sd->sd_state == G_RAID_SUBDISK_S_FAILED)
857 G_RAID_DEBUG1(1, sc, "Volume %s is not complete, "
858 "trying to refill.", vol->v_name);
860 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
862 if (disk->d_state < G_RAID_DISK_S_SPARE)
864 /* Skip already used by this volume. */
865 for (i = 0; i < vol->v_disks_count; i++) {
866 sd = &vol->v_subdisks[i];
867 if (sd->sd_disk == disk)
870 if (i < vol->v_disks_count)
873 /* Try to use disk if it has empty extents. */
874 pd = disk->d_md_data;
875 if (pd->pd_subdisks < PROMISE_MAX_SUBDISKS) {
877 g_raid_md_promise_start_disk(disk, -1, vol);
882 g_raid_md_write_promise(md, vol, NULL, disk);
892 g_raid_md_promise_start(struct g_raid_volume *vol)
894 struct g_raid_softc *sc;
895 struct g_raid_subdisk *sd;
896 struct g_raid_disk *disk;
897 struct g_raid_md_object *md;
898 struct g_raid_md_promise_perdisk *pd;
899 struct g_raid_md_promise_pervolume *pv;
900 struct promise_raid_conf *meta;
908 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
909 if (meta->type == PROMISE_T_RAID0)
910 vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
911 else if (meta->type == PROMISE_T_RAID1) {
912 if (meta->array_width == 1)
913 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
915 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
916 } else if (meta->type == PROMISE_T_RAID3)
917 vol->v_raid_level = G_RAID_VOLUME_RL_RAID3;
918 else if (meta->type == PROMISE_T_RAID5) {
919 vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
920 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
921 } else if (meta->type == PROMISE_T_SPAN)
922 vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
923 else if (meta->type == PROMISE_T_JBOD)
924 vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
926 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
927 vol->v_strip_size = 512 << meta->stripe_shift; //ZZZ
928 vol->v_disks_count = meta->total_disks;
929 vol->v_mediasize = (off_t)meta->total_sectors * 512; //ZZZ
930 if (meta->total_sectors_high < 256) /* If value looks sane. */
932 ((off_t)meta->total_sectors_high << 32) * 512; //ZZZ
933 vol->v_sectorsize = 512 * meta->sector_size;
934 for (i = 0; i < vol->v_disks_count; i++) {
935 sd = &vol->v_subdisks[i];
936 sd->sd_offset = (((off_t)meta->disk_offset_high << 32) +
937 meta->disk_offset) * 512;
938 sd->sd_size = (((off_t)meta->disk_sectors_high << 32) +
939 meta->disk_sectors) * 512;
941 g_raid_start_volume(vol);
943 /* Make all disks found till the moment take their places. */
944 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
945 pd = disk->d_md_data;
946 for (i = 0; i < pd->pd_subdisks; i++) {
947 if (pd->pd_meta[i]->volume_id == meta->volume_id)
948 g_raid_md_promise_start_disk(disk, i, vol);
953 callout_stop(&pv->pv_start_co);
954 G_RAID_DEBUG1(0, sc, "Volume started.");
955 g_raid_md_write_promise(md, vol, NULL, NULL);
957 /* Pickup any STALE/SPARE disks to refill array if needed. */
958 g_raid_md_promise_refill(sc);
960 g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
964 g_raid_promise_go(void *arg)
966 struct g_raid_volume *vol;
967 struct g_raid_softc *sc;
968 struct g_raid_md_promise_pervolume *pv;
973 if (!pv->pv_started) {
974 G_RAID_DEBUG1(0, sc, "Force volume start due to timeout.");
975 g_raid_event_send(vol, G_RAID_VOLUME_E_STARTMD,
976 G_RAID_EVENT_VOLUME);
981 g_raid_md_promise_new_disk(struct g_raid_disk *disk)
983 struct g_raid_softc *sc;
984 struct g_raid_md_object *md;
985 struct promise_raid_conf *pdmeta;
986 struct g_raid_md_promise_perdisk *pd;
987 struct g_raid_md_promise_pervolume *pv;
988 struct g_raid_volume *vol;
994 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
996 if (pd->pd_subdisks == 0) {
997 g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
998 g_raid_md_promise_refill(sc);
1002 for (i = 0; i < pd->pd_subdisks; i++) {
1003 pdmeta = pd->pd_meta[i];
1005 /* Look for volume with matching ID. */
1006 vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1008 promise_meta_get_name(pdmeta, buf);
1009 vol = g_raid_create_volume(sc, buf, pdmeta->array_number);
1010 pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1011 pv->pv_id = pdmeta->volume_id;
1012 vol->v_md_data = pv;
1013 callout_init(&pv->pv_start_co, 1);
1014 callout_reset(&pv->pv_start_co,
1015 g_raid_start_timeout * hz,
1016 g_raid_promise_go, vol);
1018 pv = vol->v_md_data;
1020 /* If we haven't started yet - check metadata freshness. */
1021 if (pv->pv_meta == NULL || !pv->pv_started) {
1022 if (pv->pv_meta == NULL ||
1023 ((int16_t)(pdmeta->generation - pv->pv_generation)) > 0) {
1024 G_RAID_DEBUG1(1, sc, "Newer disk");
1025 if (pv->pv_meta != NULL)
1026 free(pv->pv_meta, M_MD_PROMISE);
1027 pv->pv_meta = promise_meta_copy(pdmeta);
1028 pv->pv_generation = pv->pv_meta->generation;
1029 pv->pv_disks_present = 1;
1030 } else if (pdmeta->generation == pv->pv_generation) {
1031 pv->pv_disks_present++;
1032 G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
1033 pv->pv_disks_present,
1034 pv->pv_meta->total_disks);
1036 G_RAID_DEBUG1(1, sc, "Older disk");
1041 for (i = 0; i < pd->pd_subdisks; i++) {
1042 pdmeta = pd->pd_meta[i];
1044 /* Look for volume with matching ID. */
1045 vol = g_raid_md_promise_get_volume(sc, pdmeta->volume_id);
1048 pv = vol->v_md_data;
1050 if (pv->pv_started) {
1051 if (g_raid_md_promise_start_disk(disk, i, vol))
1052 g_raid_md_write_promise(md, vol, NULL, NULL);
1054 /* If we collected all needed disks - start array. */
1055 if (pv->pv_disks_present == pv->pv_meta->total_disks)
1056 g_raid_md_promise_start(vol);
1062 g_raid_md_create_promise(struct g_raid_md_object *md, struct g_class *mp,
1065 struct g_geom *geom;
1066 struct g_raid_softc *sc;
1068 /* Search for existing node. */
1069 LIST_FOREACH(geom, &mp->geom, geom) {
1073 if (sc->sc_stopping != 0)
1075 if (sc->sc_md->mdo_class != md->mdo_class)
1081 return (G_RAID_MD_TASTE_EXISTING);
1084 /* Create new one if not found. */
1085 sc = g_raid_create_node(mp, "Promise", md);
1087 return (G_RAID_MD_TASTE_FAIL);
1090 return (G_RAID_MD_TASTE_NEW);
1094 g_raid_md_taste_promise(struct g_raid_md_object *md, struct g_class *mp,
1095 struct g_consumer *cp, struct g_geom **gp)
1097 struct g_consumer *rcp;
1098 struct g_provider *pp;
1099 struct g_raid_softc *sc;
1100 struct g_raid_disk *disk;
1101 struct promise_raid_conf *meta, *metaarr[4];
1102 struct g_raid_md_promise_perdisk *pd;
1103 struct g_geom *geom;
1104 int i, j, result, len, subdisks;
1108 G_RAID_DEBUG(1, "Tasting Promise on %s", cp->provider->name);
1111 /* Read metadata from device. */
1113 g_topology_unlock();
1115 len = sizeof(vendor);
1116 if (pp->geom->rank == 1)
1117 g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
1118 subdisks = promise_meta_read(cp, metaarr);
1120 if (subdisks == 0) {
1121 if (g_raid_aggressive_spare) {
1122 if (vendor == 0x105a || vendor == 0x1002) {
1124 "No Promise metadata, forcing spare.");
1128 "Promise/ATI vendor mismatch "
1129 "0x%04x != 0x105a/0x1002",
1133 return (G_RAID_MD_TASTE_FAIL);
1136 /* Metadata valid. Print it. */
1137 for (i = 0; i < subdisks; i++)
1138 g_raid_md_promise_print(metaarr[i]);
1140 /* Purge meaningless (empty/spare) records. */
1141 for (i = 0; i < subdisks; ) {
1142 if (metaarr[i]->disk.flags & PROMISE_F_ASSIGNED) {
1146 free(metaarr[i], M_MD_PROMISE);
1147 for (j = i; j < subdisks - 1; j++)
1148 metaarr[i] = metaarr[j + 1];
1149 metaarr[subdisks - 1] = NULL;
1154 /* Search for matching node. */
1156 LIST_FOREACH(geom, &mp->geom, geom) {
1160 if (sc->sc_stopping != 0)
1162 if (sc->sc_md->mdo_class != md->mdo_class)
1167 /* Found matching node. */
1169 G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
1170 result = G_RAID_MD_TASTE_EXISTING;
1172 } else { /* Not found matching node -- create one. */
1173 result = G_RAID_MD_TASTE_NEW;
1174 snprintf(name, sizeof(name), "Promise");
1175 sc = g_raid_create_node(mp, name, md);
1180 /* There is no return after this point, so we close passed consumer. */
1181 g_access(cp, -1, 0, 0);
1183 rcp = g_new_consumer(geom);
1184 rcp->flags |= G_CF_DIRECT_RECEIVE;
1186 if (g_access(rcp, 1, 1, 1) != 0)
1189 g_topology_unlock();
1190 sx_xlock(&sc->sc_lock);
1192 pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1193 pd->pd_subdisks = subdisks;
1194 for (i = 0; i < subdisks; i++)
1195 pd->pd_meta[i] = metaarr[i];
1196 disk = g_raid_create_disk(sc);
1197 disk->d_md_data = (void *)pd;
1198 disk->d_consumer = rcp;
1199 rcp->private = disk;
1201 g_raid_get_disk_info(disk);
1203 g_raid_md_promise_new_disk(disk);
1205 sx_xunlock(&sc->sc_lock);
1212 g_raid_md_event_promise(struct g_raid_md_object *md,
1213 struct g_raid_disk *disk, u_int event)
1215 struct g_raid_softc *sc;
1221 case G_RAID_DISK_E_DISCONNECTED:
1223 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1224 g_raid_destroy_disk(disk);
1225 g_raid_md_promise_purge_volumes(sc);
1227 /* Write updated metadata to all disks. */
1228 g_raid_md_write_promise(md, NULL, NULL, NULL);
1230 /* Check if anything left. */
1231 if (g_raid_ndisks(sc, -1) == 0)
1232 g_raid_destroy_node(sc, 0);
1234 g_raid_md_promise_refill(sc);
1241 g_raid_md_volume_event_promise(struct g_raid_md_object *md,
1242 struct g_raid_volume *vol, u_int event)
1244 struct g_raid_md_promise_pervolume *pv;
1246 pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1248 case G_RAID_VOLUME_E_STARTMD:
1249 if (!pv->pv_started)
1250 g_raid_md_promise_start(vol);
1257 g_raid_md_ctl_promise(struct g_raid_md_object *md,
1258 struct gctl_req *req)
1260 struct g_raid_softc *sc;
1261 struct g_raid_volume *vol, *vol1;
1262 struct g_raid_subdisk *sd;
1263 struct g_raid_disk *disk, *disks[PROMISE_MAX_DISKS];
1264 struct g_raid_md_promise_perdisk *pd;
1265 struct g_raid_md_promise_pervolume *pv;
1266 struct g_consumer *cp;
1267 struct g_provider *pp;
1269 const char *nodename, *verb, *volname, *levelname, *diskname;
1272 off_t esize, offs[PROMISE_MAX_DISKS], size, sectorsize, strip;
1273 intmax_t *sizearg, *striparg;
1274 int numdisks, i, len, level, qual;
1278 verb = gctl_get_param(req, "verb", NULL);
1279 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1281 if (strcmp(verb, "label") == 0) {
1284 gctl_error(req, "Invalid number of arguments.");
1287 volname = gctl_get_asciiparam(req, "arg1");
1288 if (volname == NULL) {
1289 gctl_error(req, "No volume name.");
1292 levelname = gctl_get_asciiparam(req, "arg2");
1293 if (levelname == NULL) {
1294 gctl_error(req, "No RAID level.");
1297 if (strcasecmp(levelname, "RAID5") == 0)
1298 levelname = "RAID5-LA";
1299 if (g_raid_volume_str2level(levelname, &level, &qual)) {
1300 gctl_error(req, "Unknown RAID level '%s'.", levelname);
1303 numdisks = *nargs - 3;
1304 force = gctl_get_paraml(req, "force", sizeof(*force));
1305 if (!g_raid_md_promise_supported(level, qual, numdisks,
1306 force ? *force : 0)) {
1307 gctl_error(req, "Unsupported RAID level "
1308 "(0x%02x/0x%02x), or number of disks (%d).",
1309 level, qual, numdisks);
1313 /* Search for disks, connect them and probe. */
1316 bzero(disks, sizeof(disks));
1317 bzero(offs, sizeof(offs));
1318 for (i = 0; i < numdisks; i++) {
1319 snprintf(arg, sizeof(arg), "arg%d", i + 3);
1320 diskname = gctl_get_asciiparam(req, arg);
1321 if (diskname == NULL) {
1322 gctl_error(req, "No disk name (%s).", arg);
1326 if (strcmp(diskname, "NONE") == 0)
1329 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1330 if (disk->d_consumer != NULL &&
1331 disk->d_consumer->provider != NULL &&
1332 strcmp(disk->d_consumer->provider->name,
1337 if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
1338 gctl_error(req, "Disk '%s' is in a "
1339 "wrong state (%s).", diskname,
1340 g_raid_disk_state2str(disk->d_state));
1344 pd = disk->d_md_data;
1345 if (pd->pd_subdisks >= PROMISE_MAX_SUBDISKS) {
1346 gctl_error(req, "Disk '%s' already "
1347 "used by %d volumes.",
1348 diskname, pd->pd_subdisks);
1352 pp = disk->d_consumer->provider;
1354 promise_meta_unused_range(pd->pd_meta,
1356 pp->mediasize / pp->sectorsize,
1358 size = MIN(size, (off_t)esize * pp->sectorsize);
1359 sectorsize = MAX(sectorsize, pp->sectorsize);
1364 cp = g_raid_open_consumer(sc, diskname);
1366 gctl_error(req, "Can't open disk '%s'.",
1368 g_topology_unlock();
1373 pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1374 disk = g_raid_create_disk(sc);
1375 disk->d_md_data = (void *)pd;
1376 disk->d_consumer = cp;
1379 g_topology_unlock();
1381 g_raid_get_disk_info(disk);
1383 /* Reserve some space for metadata. */
1384 size = MIN(size, pp->mediasize - 131072llu * pp->sectorsize);
1385 sectorsize = MAX(sectorsize, pp->sectorsize);
1388 for (i = 0; i < numdisks; i++) {
1389 if (disks[i] != NULL &&
1390 disks[i]->d_state == G_RAID_DISK_S_NONE)
1391 g_raid_destroy_disk(disks[i]);
1396 if (sectorsize <= 0) {
1397 gctl_error(req, "Can't get sector size.");
1401 /* Handle size argument. */
1402 len = sizeof(*sizearg);
1403 sizearg = gctl_get_param(req, "size", &len);
1404 if (sizearg != NULL && len == sizeof(*sizearg) &&
1406 if (*sizearg > size) {
1407 gctl_error(req, "Size too big %lld > %lld.",
1408 (long long)*sizearg, (long long)size);
1414 /* Handle strip argument. */
1416 len = sizeof(*striparg);
1417 striparg = gctl_get_param(req, "strip", &len);
1418 if (striparg != NULL && len == sizeof(*striparg) &&
1420 if (*striparg < sectorsize) {
1421 gctl_error(req, "Strip size too small.");
1424 if (*striparg % sectorsize != 0) {
1425 gctl_error(req, "Incorrect strip size.");
1431 /* Round size down to strip or sector. */
1432 if (level == G_RAID_VOLUME_RL_RAID1 ||
1433 level == G_RAID_VOLUME_RL_SINGLE ||
1434 level == G_RAID_VOLUME_RL_CONCAT)
1435 size -= (size % sectorsize);
1436 else if (level == G_RAID_VOLUME_RL_RAID1E &&
1437 (numdisks & 1) != 0)
1438 size -= (size % (2 * strip));
1440 size -= (size % strip);
1442 gctl_error(req, "Size too small.");
1446 /* We have all we need, create things: volume, ... */
1447 pv = malloc(sizeof(*pv), M_MD_PROMISE, M_WAITOK | M_ZERO);
1448 arc4rand(&pv->pv_id, sizeof(pv->pv_id), 0);
1449 pv->pv_generation = 0;
1451 vol = g_raid_create_volume(sc, volname, -1);
1452 vol->v_md_data = pv;
1453 vol->v_raid_level = level;
1454 vol->v_raid_level_qualifier = qual;
1455 vol->v_strip_size = strip;
1456 vol->v_disks_count = numdisks;
1457 if (level == G_RAID_VOLUME_RL_RAID0 ||
1458 level == G_RAID_VOLUME_RL_CONCAT ||
1459 level == G_RAID_VOLUME_RL_SINGLE)
1460 vol->v_mediasize = size * numdisks;
1461 else if (level == G_RAID_VOLUME_RL_RAID1)
1462 vol->v_mediasize = size;
1463 else if (level == G_RAID_VOLUME_RL_RAID3 ||
1464 level == G_RAID_VOLUME_RL_RAID5)
1465 vol->v_mediasize = size * (numdisks - 1);
1467 vol->v_mediasize = ((size * numdisks) / strip / 2) *
1470 vol->v_sectorsize = sectorsize;
1471 g_raid_start_volume(vol);
1473 /* , and subdisks. */
1474 for (i = 0; i < numdisks; i++) {
1476 sd = &vol->v_subdisks[i];
1478 sd->sd_offset = (off_t)offs[i] * 512;
1482 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1483 g_raid_change_disk_state(disk,
1484 G_RAID_DISK_S_ACTIVE);
1485 g_raid_change_subdisk_state(sd,
1486 G_RAID_SUBDISK_S_ACTIVE);
1487 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1488 G_RAID_EVENT_SUBDISK);
1491 /* Write metadata based on created entities. */
1492 G_RAID_DEBUG1(0, sc, "Array started.");
1493 g_raid_md_write_promise(md, vol, NULL, NULL);
1495 /* Pickup any STALE/SPARE disks to refill array if needed. */
1496 g_raid_md_promise_refill(sc);
1498 g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1499 G_RAID_EVENT_VOLUME);
1502 if (strcmp(verb, "add") == 0) {
1504 gctl_error(req, "`add` command is not applicable, "
1505 "use `label` instead.");
1508 if (strcmp(verb, "delete") == 0) {
1510 nodename = gctl_get_asciiparam(req, "arg0");
1511 if (nodename != NULL && strcasecmp(sc->sc_name, nodename) != 0)
1514 /* Full node destruction. */
1515 if (*nargs == 1 && nodename != NULL) {
1516 /* Check if some volume is still open. */
1517 force = gctl_get_paraml(req, "force", sizeof(*force));
1518 if (force != NULL && *force == 0 &&
1519 g_raid_nopens(sc) != 0) {
1520 gctl_error(req, "Some volume is still open.");
1524 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1525 if (disk->d_consumer)
1526 promise_meta_erase(disk->d_consumer);
1528 g_raid_destroy_node(sc, 0);
1532 /* Destroy specified volume. If it was last - all node. */
1534 gctl_error(req, "Invalid number of arguments.");
1537 volname = gctl_get_asciiparam(req,
1538 nodename != NULL ? "arg1" : "arg0");
1539 if (volname == NULL) {
1540 gctl_error(req, "No volume name.");
1544 /* Search for volume. */
1545 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1546 if (strcmp(vol->v_name, volname) == 0)
1548 pp = vol->v_provider;
1551 if (strcmp(pp->name, volname) == 0)
1553 if (strncmp(pp->name, "raid/", 5) == 0 &&
1554 strcmp(pp->name + 5, volname) == 0)
1558 i = strtol(volname, &tmp, 10);
1559 if (verb != volname && tmp[0] == 0) {
1560 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1561 if (vol->v_global_id == i)
1567 gctl_error(req, "Volume '%s' not found.", volname);
1571 /* Check if volume is still open. */
1572 force = gctl_get_paraml(req, "force", sizeof(*force));
1573 if (force != NULL && *force == 0 &&
1574 vol->v_provider_open != 0) {
1575 gctl_error(req, "Volume is still open.");
1579 /* Destroy volume and potentially node. */
1581 TAILQ_FOREACH(vol1, &sc->sc_volumes, v_next)
1584 g_raid_destroy_volume(vol);
1585 g_raid_md_promise_purge_disks(sc);
1586 g_raid_md_write_promise(md, NULL, NULL, NULL);
1588 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1589 if (disk->d_consumer)
1590 promise_meta_erase(disk->d_consumer);
1592 g_raid_destroy_node(sc, 0);
1596 if (strcmp(verb, "remove") == 0 ||
1597 strcmp(verb, "fail") == 0) {
1599 gctl_error(req, "Invalid number of arguments.");
1602 for (i = 1; i < *nargs; i++) {
1603 snprintf(arg, sizeof(arg), "arg%d", i);
1604 diskname = gctl_get_asciiparam(req, arg);
1605 if (diskname == NULL) {
1606 gctl_error(req, "No disk name (%s).", arg);
1610 if (strncmp(diskname, "/dev/", 5) == 0)
1613 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1614 if (disk->d_consumer != NULL &&
1615 disk->d_consumer->provider != NULL &&
1616 strcmp(disk->d_consumer->provider->name,
1621 gctl_error(req, "Disk '%s' not found.",
1627 if (strcmp(verb, "fail") == 0) {
1628 g_raid_md_fail_disk_promise(md, NULL, disk);
1632 /* Erase metadata on deleting disk and destroy it. */
1633 promise_meta_erase(disk->d_consumer);
1634 g_raid_destroy_disk(disk);
1636 g_raid_md_promise_purge_volumes(sc);
1638 /* Write updated metadata to remaining disks. */
1639 g_raid_md_write_promise(md, NULL, NULL, NULL);
1641 /* Check if anything left. */
1642 if (g_raid_ndisks(sc, -1) == 0)
1643 g_raid_destroy_node(sc, 0);
1645 g_raid_md_promise_refill(sc);
1648 if (strcmp(verb, "insert") == 0) {
1650 gctl_error(req, "Invalid number of arguments.");
1653 for (i = 1; i < *nargs; i++) {
1654 /* Get disk name. */
1655 snprintf(arg, sizeof(arg), "arg%d", i);
1656 diskname = gctl_get_asciiparam(req, arg);
1657 if (diskname == NULL) {
1658 gctl_error(req, "No disk name (%s).", arg);
1663 /* Try to find provider with specified name. */
1665 cp = g_raid_open_consumer(sc, diskname);
1667 gctl_error(req, "Can't open disk '%s'.",
1669 g_topology_unlock();
1674 g_topology_unlock();
1676 pd = malloc(sizeof(*pd), M_MD_PROMISE, M_WAITOK | M_ZERO);
1678 disk = g_raid_create_disk(sc);
1679 disk->d_consumer = cp;
1680 disk->d_md_data = (void *)pd;
1683 g_raid_get_disk_info(disk);
1685 /* Welcome the "new" disk. */
1686 g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
1687 promise_meta_write_spare(cp);
1688 g_raid_md_promise_refill(sc);
1696 g_raid_md_write_promise(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1697 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1699 struct g_raid_softc *sc;
1700 struct g_raid_volume *vol;
1701 struct g_raid_subdisk *sd;
1702 struct g_raid_disk *disk;
1703 struct g_raid_md_promise_perdisk *pd;
1704 struct g_raid_md_promise_pervolume *pv;
1705 struct promise_raid_conf *meta;
1706 off_t rebuild_lba64;
1707 int i, j, pos, rebuild;
1711 if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1714 /* Generate new per-volume metadata for affected volumes. */
1715 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1716 if (vol->v_stopping)
1719 /* Skip volumes not related to specified targets. */
1720 if (tvol != NULL && vol != tvol)
1722 if (tsd != NULL && vol != tsd->sd_volume)
1724 if (tdisk != NULL) {
1725 for (i = 0; i < vol->v_disks_count; i++) {
1726 if (vol->v_subdisks[i].sd_disk == tdisk)
1729 if (i >= vol->v_disks_count)
1733 pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1734 pv->pv_generation++;
1736 meta = malloc(sizeof(*meta), M_MD_PROMISE, M_WAITOK | M_ZERO);
1737 if (pv->pv_meta != NULL)
1738 memcpy(meta, pv->pv_meta, sizeof(*meta));
1739 memcpy(meta->promise_id, PROMISE_MAGIC,
1740 sizeof(PROMISE_MAGIC) - 1);
1741 meta->dummy_0 = 0x00020000;
1742 meta->integrity = PROMISE_I_VALID;
1744 meta->generation = pv->pv_generation;
1745 meta->status = PROMISE_S_VALID | PROMISE_S_ONLINE |
1746 PROMISE_S_INITED | PROMISE_S_READY;
1747 if (vol->v_state <= G_RAID_VOLUME_S_DEGRADED)
1748 meta->status |= PROMISE_S_DEGRADED;
1750 meta->status |= PROMISE_S_MARKED; /* XXX: INVENTED! */
1751 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0 ||
1752 vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1753 meta->type = PROMISE_T_RAID0;
1754 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1755 vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1756 meta->type = PROMISE_T_RAID1;
1757 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3)
1758 meta->type = PROMISE_T_RAID3;
1759 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1760 meta->type = PROMISE_T_RAID5;
1761 else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT)
1762 meta->type = PROMISE_T_SPAN;
1764 meta->type = PROMISE_T_JBOD;
1765 meta->total_disks = vol->v_disks_count;
1766 meta->stripe_shift = ffs(vol->v_strip_size / 1024);
1767 meta->array_width = vol->v_disks_count;
1768 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1769 vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1770 meta->array_width /= 2;
1771 meta->array_number = vol->v_global_id;
1772 meta->total_sectors = vol->v_mediasize / 512;
1773 meta->total_sectors_high = (vol->v_mediasize / 512) >> 32;
1774 meta->sector_size = vol->v_sectorsize / 512;
1775 meta->cylinders = meta->total_sectors / (255 * 63) - 1;
1778 meta->volume_id = pv->pv_id;
1779 rebuild_lba64 = UINT64_MAX;
1781 for (i = 0; i < vol->v_disks_count; i++) {
1782 sd = &vol->v_subdisks[i];
1783 /* For RAID0+1 we need to translate order. */
1784 pos = promise_meta_translate_disk(vol, i);
1785 meta->disks[pos].flags = PROMISE_F_VALID |
1787 if (sd->sd_state == G_RAID_SUBDISK_S_NONE) {
1788 meta->disks[pos].flags |= 0;
1789 } else if (sd->sd_state == G_RAID_SUBDISK_S_FAILED) {
1790 meta->disks[pos].flags |=
1791 PROMISE_F_DOWN | PROMISE_F_REDIR;
1792 } else if (sd->sd_state <= G_RAID_SUBDISK_S_REBUILD) {
1793 meta->disks[pos].flags |=
1794 PROMISE_F_ONLINE | PROMISE_F_REDIR;
1795 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1796 rebuild_lba64 = MIN(rebuild_lba64,
1797 sd->sd_rebuild_pos / 512);
1802 meta->disks[pos].flags |= PROMISE_F_ONLINE;
1803 if (sd->sd_state < G_RAID_SUBDISK_S_ACTIVE) {
1804 meta->status |= PROMISE_S_MARKED;
1805 if (sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
1806 rebuild_lba64 = MIN(rebuild_lba64,
1807 sd->sd_rebuild_pos / 512);
1812 if (pv->pv_meta != NULL) {
1813 meta->disks[pos].id = pv->pv_meta->disks[pos].id;
1815 meta->disks[pos].number = i * 2;
1816 arc4rand(&meta->disks[pos].id,
1817 sizeof(meta->disks[pos].id), 0);
1820 promise_meta_put_name(meta, vol->v_name);
1822 /* Try to mimic AMD BIOS rebuild/resync behavior. */
1823 if (rebuild_lba64 != UINT64_MAX) {
1825 meta->magic_3 = 0x03040010UL; /* Rebuild? */
1827 meta->magic_3 = 0x03040008UL; /* Resync? */
1828 /* Translate from per-disk to per-volume LBA. */
1829 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1830 vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E) {
1831 rebuild_lba64 *= meta->array_width;
1832 } else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1833 vol->v_raid_level == G_RAID_VOLUME_RL_RAID5) {
1834 rebuild_lba64 *= meta->array_width - 1;
1838 meta->magic_3 = 0x03000000UL;
1839 meta->rebuild_lba64 = rebuild_lba64;
1840 meta->magic_4 = 0x04010101UL;
1842 /* Replace per-volume metadata with new. */
1843 if (pv->pv_meta != NULL)
1844 free(pv->pv_meta, M_MD_PROMISE);
1847 /* Copy new metadata to the disks, adding or replacing old. */
1848 for (i = 0; i < vol->v_disks_count; i++) {
1849 sd = &vol->v_subdisks[i];
1853 /* For RAID0+1 we need to translate order. */
1854 pos = promise_meta_translate_disk(vol, i);
1855 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1856 for (j = 0; j < pd->pd_subdisks; j++) {
1857 if (pd->pd_meta[j]->volume_id == meta->volume_id)
1860 if (j == pd->pd_subdisks)
1862 if (pd->pd_meta[j] != NULL)
1863 free(pd->pd_meta[j], M_MD_PROMISE);
1864 pd->pd_meta[j] = promise_meta_copy(meta);
1865 pd->pd_meta[j]->disk = meta->disks[pos];
1866 pd->pd_meta[j]->disk.number = pos;
1867 pd->pd_meta[j]->disk_offset_high =
1868 (sd->sd_offset / 512) >> 32;
1869 pd->pd_meta[j]->disk_offset = sd->sd_offset / 512;
1870 pd->pd_meta[j]->disk_sectors_high =
1871 (sd->sd_size / 512) >> 32;
1872 pd->pd_meta[j]->disk_sectors = sd->sd_size / 512;
1873 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD) {
1874 pd->pd_meta[j]->disk_rebuild_high =
1875 (sd->sd_rebuild_pos / 512) >> 32;
1876 pd->pd_meta[j]->disk_rebuild =
1877 sd->sd_rebuild_pos / 512;
1878 } else if (sd->sd_state < G_RAID_SUBDISK_S_REBUILD) {
1879 pd->pd_meta[j]->disk_rebuild_high = 0;
1880 pd->pd_meta[j]->disk_rebuild = 0;
1882 pd->pd_meta[j]->disk_rebuild_high = UINT32_MAX;
1883 pd->pd_meta[j]->disk_rebuild = UINT32_MAX;
1889 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1890 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1891 if (disk->d_state != G_RAID_DISK_S_ACTIVE)
1893 if (!pd->pd_updated)
1895 G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1896 g_raid_get_diskname(disk));
1897 for (i = 0; i < pd->pd_subdisks; i++)
1898 g_raid_md_promise_print(pd->pd_meta[i]);
1899 promise_meta_write(disk->d_consumer,
1900 pd->pd_meta, pd->pd_subdisks);
1908 g_raid_md_fail_disk_promise(struct g_raid_md_object *md,
1909 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1911 struct g_raid_softc *sc;
1912 struct g_raid_md_promise_perdisk *pd;
1913 struct g_raid_subdisk *sd;
1917 pd = (struct g_raid_md_promise_perdisk *)tdisk->d_md_data;
1919 /* We can't fail disk that is not a part of array now. */
1920 if (tdisk->d_state != G_RAID_DISK_S_ACTIVE)
1924 * Mark disk as failed in metadata and try to write that metadata
1925 * to the disk itself to prevent it's later resurrection as STALE.
1927 if (pd->pd_subdisks > 0 && tdisk->d_consumer != NULL)
1928 G_RAID_DEBUG(1, "Writing Promise metadata to %s",
1929 g_raid_get_diskname(tdisk));
1930 for (i = 0; i < pd->pd_subdisks; i++) {
1931 pd->pd_meta[i]->disk.flags |=
1932 PROMISE_F_DOWN | PROMISE_F_REDIR;
1933 pos = pd->pd_meta[i]->disk.number;
1934 if (pos >= 0 && pos < PROMISE_MAX_DISKS) {
1935 pd->pd_meta[i]->disks[pos].flags |=
1936 PROMISE_F_DOWN | PROMISE_F_REDIR;
1938 g_raid_md_promise_print(pd->pd_meta[i]);
1940 if (tdisk->d_consumer != NULL)
1941 promise_meta_write(tdisk->d_consumer,
1942 pd->pd_meta, pd->pd_subdisks);
1944 /* Change states. */
1945 g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1946 TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1947 g_raid_change_subdisk_state(sd,
1948 G_RAID_SUBDISK_S_FAILED);
1949 g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1950 G_RAID_EVENT_SUBDISK);
1953 /* Write updated metadata to remaining disks. */
1954 g_raid_md_write_promise(md, NULL, NULL, tdisk);
1956 g_raid_md_promise_refill(sc);
1961 g_raid_md_free_disk_promise(struct g_raid_md_object *md,
1962 struct g_raid_disk *disk)
1964 struct g_raid_md_promise_perdisk *pd;
1967 pd = (struct g_raid_md_promise_perdisk *)disk->d_md_data;
1968 for (i = 0; i < pd->pd_subdisks; i++) {
1969 if (pd->pd_meta[i] != NULL) {
1970 free(pd->pd_meta[i], M_MD_PROMISE);
1971 pd->pd_meta[i] = NULL;
1974 free(pd, M_MD_PROMISE);
1975 disk->d_md_data = NULL;
1980 g_raid_md_free_volume_promise(struct g_raid_md_object *md,
1981 struct g_raid_volume *vol)
1983 struct g_raid_md_promise_pervolume *pv;
1985 pv = (struct g_raid_md_promise_pervolume *)vol->v_md_data;
1986 if (pv && pv->pv_meta != NULL) {
1987 free(pv->pv_meta, M_MD_PROMISE);
1990 if (pv && !pv->pv_started) {
1992 callout_stop(&pv->pv_start_co);
1994 free(pv, M_MD_PROMISE);
1995 vol->v_md_data = NULL;
2000 g_raid_md_free_promise(struct g_raid_md_object *md)
2006 G_RAID_MD_DECLARE(promise, "Promise");