1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2011 Alexander Motin <mav@FreeBSD.org>
5 * Copyright (c) 2000 - 2008 Søren Schmidt <sos@FreeBSD.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30 #include <sys/param.h>
31 #include <sys/bio.h>
32 #include <sys/endian.h>
33 #include <sys/kernel.h>
34 #include <sys/kobj.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/systm.h>
40 #include <sys/taskqueue.h>
41 #include <geom/geom.h>
42 #include <geom/geom_dbg.h>
43 #include "geom/raid/g_raid.h"
44 #include "g_raid_md_if.h"
45
46 static MALLOC_DEFINE(M_MD_NVIDIA, "md_nvidia_data", "GEOM_RAID NVIDIA metadata");
47
48 struct nvidia_raid_conf {
49 uint8_t nvidia_id[8];
50 #define NVIDIA_MAGIC "NVIDIA "
51
52 uint32_t config_size;
53 uint32_t checksum;
54 uint16_t version;
55 uint8_t disk_number;
56 uint8_t dummy_0;
57 uint32_t total_sectors;
58 uint32_t sector_size;
59 uint8_t name[16];
60 uint8_t revision[4];
61 uint32_t disk_status;
62
63 uint32_t magic_0;
64 #define NVIDIA_MAGIC0 0x00640044
65
66 uint64_t volume_id[2];
67 uint8_t state;
68 #define NVIDIA_S_IDLE 0
69 #define NVIDIA_S_INIT 2
70 #define NVIDIA_S_REBUILD 3
71 #define NVIDIA_S_UPGRADE 4
72 #define NVIDIA_S_SYNC 5
73 uint8_t array_width;
74 uint8_t total_disks;
75 uint8_t orig_array_width;
76 uint16_t type;
77 #define NVIDIA_T_RAID0 0x0080
78 #define NVIDIA_T_RAID1 0x0081
79 #define NVIDIA_T_RAID3 0x0083
80 #define NVIDIA_T_RAID5 0x0085 /* RLQ = 00/02? */
81 #define NVIDIA_T_RAID5_SYM 0x0095 /* RLQ = 03 */
82 #define NVIDIA_T_RAID10 0x008a
83 #define NVIDIA_T_RAID01 0x8180
84 #define NVIDIA_T_CONCAT 0x00ff
85
86 uint16_t dummy_3;
87 uint32_t strip_sectors;
88 uint32_t strip_bytes;
89 uint32_t strip_shift;
90 uint32_t strip_mask;
91 uint32_t stripe_sectors;
92 uint32_t stripe_bytes;
93 uint32_t rebuild_lba;
94 uint32_t orig_type;
95 uint32_t orig_total_sectors;
96 uint32_t status;
97 #define NVIDIA_S_BOOTABLE 0x00000001
98 #define NVIDIA_S_DEGRADED 0x00000002
99
100 uint32_t filler[98];
101 } __packed;
102
103 struct g_raid_md_nvidia_perdisk {
104 struct nvidia_raid_conf *pd_meta;
105 int pd_disk_pos;
106 off_t pd_disk_size;
107 };
108
109 struct g_raid_md_nvidia_object {
110 struct g_raid_md_object mdio_base;
111 uint64_t mdio_volume_id[2];
112 struct nvidia_raid_conf *mdio_meta;
113 struct callout mdio_start_co; /* STARTING state timer. */
114 int mdio_total_disks;
115 int mdio_disks_present;
116 int mdio_started;
117 int mdio_incomplete;
118 struct root_hold_token *mdio_rootmount; /* Root mount delay token. */
119 };
120
121 static g_raid_md_create_t g_raid_md_create_nvidia;
122 static g_raid_md_taste_t g_raid_md_taste_nvidia;
123 static g_raid_md_event_t g_raid_md_event_nvidia;
124 static g_raid_md_ctl_t g_raid_md_ctl_nvidia;
125 static g_raid_md_write_t g_raid_md_write_nvidia;
126 static g_raid_md_fail_disk_t g_raid_md_fail_disk_nvidia;
127 static g_raid_md_free_disk_t g_raid_md_free_disk_nvidia;
128 static g_raid_md_free_t g_raid_md_free_nvidia;
129
130 static kobj_method_t g_raid_md_nvidia_methods[] = {
131 KOBJMETHOD(g_raid_md_create, g_raid_md_create_nvidia),
132 KOBJMETHOD(g_raid_md_taste, g_raid_md_taste_nvidia),
133 KOBJMETHOD(g_raid_md_event, g_raid_md_event_nvidia),
134 KOBJMETHOD(g_raid_md_ctl, g_raid_md_ctl_nvidia),
135 KOBJMETHOD(g_raid_md_write, g_raid_md_write_nvidia),
136 KOBJMETHOD(g_raid_md_fail_disk, g_raid_md_fail_disk_nvidia),
137 KOBJMETHOD(g_raid_md_free_disk, g_raid_md_free_disk_nvidia),
138 KOBJMETHOD(g_raid_md_free, g_raid_md_free_nvidia),
139 { 0, 0 }
140 };
141
142 static struct g_raid_md_class g_raid_md_nvidia_class = {
143 "NVIDIA",
144 g_raid_md_nvidia_methods,
145 sizeof(struct g_raid_md_nvidia_object),
146 .mdc_enable = 1,
147 .mdc_priority = 100
148 };
149
150 static int NVIDIANodeID = 1;
151
152 static void
g_raid_md_nvidia_print(struct nvidia_raid_conf * meta)153 g_raid_md_nvidia_print(struct nvidia_raid_conf *meta)
154 {
155
156 if (g_raid_debug < 1)
157 return;
158
159 printf("********* ATA NVIDIA RAID Metadata *********\n");
160 printf("nvidia_id <%.8s>\n", meta->nvidia_id);
161 printf("config_size %u\n", meta->config_size);
162 printf("checksum 0x%08x\n", meta->checksum);
163 printf("version 0x%04x\n", meta->version);
164 printf("disk_number %d\n", meta->disk_number);
165 printf("dummy_0 0x%02x\n", meta->dummy_0);
166 printf("total_sectors %u\n", meta->total_sectors);
167 printf("sector_size %u\n", meta->sector_size);
168 printf("name <%.16s>\n", meta->name);
169 printf("revision 0x%02x%02x%02x%02x\n",
170 meta->revision[0], meta->revision[1],
171 meta->revision[2], meta->revision[3]);
172 printf("disk_status 0x%08x\n", meta->disk_status);
173 printf("magic_0 0x%08x\n", meta->magic_0);
174 printf("volume_id 0x%016jx%016jx\n",
175 meta->volume_id[1], meta->volume_id[0]);
176 printf("state 0x%02x\n", meta->state);
177 printf("array_width %u\n", meta->array_width);
178 printf("total_disks %u\n", meta->total_disks);
179 printf("orig_array_width %u\n", meta->orig_array_width);
180 printf("type 0x%04x\n", meta->type);
181 printf("dummy_3 0x%04x\n", meta->dummy_3);
182 printf("strip_sectors %u\n", meta->strip_sectors);
183 printf("strip_bytes %u\n", meta->strip_bytes);
184 printf("strip_shift %u\n", meta->strip_shift);
185 printf("strip_mask 0x%08x\n", meta->strip_mask);
186 printf("stripe_sectors %u\n", meta->stripe_sectors);
187 printf("stripe_bytes %u\n", meta->stripe_bytes);
188 printf("rebuild_lba %u\n", meta->rebuild_lba);
189 printf("orig_type 0x%04x\n", meta->orig_type);
190 printf("orig_total_sectors %u\n", meta->orig_total_sectors);
191 printf("status 0x%08x\n", meta->status);
192 printf("=================================================\n");
193 }
194
195 static struct nvidia_raid_conf *
nvidia_meta_copy(struct nvidia_raid_conf * meta)196 nvidia_meta_copy(struct nvidia_raid_conf *meta)
197 {
198 struct nvidia_raid_conf *nmeta;
199
200 nmeta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
201 memcpy(nmeta, meta, sizeof(*meta));
202 return (nmeta);
203 }
204
205 static int
nvidia_meta_translate_disk(struct nvidia_raid_conf * meta,int md_disk_pos)206 nvidia_meta_translate_disk(struct nvidia_raid_conf *meta, int md_disk_pos)
207 {
208 int disk_pos;
209
210 if (md_disk_pos >= 0 && meta->type == NVIDIA_T_RAID01) {
211 disk_pos = (md_disk_pos / meta->array_width) +
212 (md_disk_pos % meta->array_width) * meta->array_width;
213 } else
214 disk_pos = md_disk_pos;
215 return (disk_pos);
216 }
217
218 static void
nvidia_meta_get_name(struct nvidia_raid_conf * meta,char * buf)219 nvidia_meta_get_name(struct nvidia_raid_conf *meta, char *buf)
220 {
221 int i;
222
223 strncpy(buf, meta->name, 16);
224 buf[16] = 0;
225 for (i = 15; i >= 0; i--) {
226 if (buf[i] > 0x20)
227 break;
228 buf[i] = 0;
229 }
230 }
231
232 static void
nvidia_meta_put_name(struct nvidia_raid_conf * meta,char * buf)233 nvidia_meta_put_name(struct nvidia_raid_conf *meta, char *buf)
234 {
235
236 memset(meta->name, 0x20, 16);
237 memcpy(meta->name, buf, MIN(strlen(buf), 16));
238 }
239
240 static struct nvidia_raid_conf *
nvidia_meta_read(struct g_consumer * cp)241 nvidia_meta_read(struct g_consumer *cp)
242 {
243 struct g_provider *pp;
244 struct nvidia_raid_conf *meta;
245 char *buf;
246 int error, i;
247 uint32_t checksum, *ptr;
248
249 pp = cp->provider;
250 if (pp->sectorsize < sizeof(*meta))
251 return (NULL);
252 /* Read the anchor sector. */
253 buf = g_read_data(cp,
254 pp->mediasize - 2 * pp->sectorsize, pp->sectorsize, &error);
255 if (buf == NULL) {
256 G_RAID_DEBUG(1, "Cannot read metadata from %s (error=%d).",
257 pp->name, error);
258 return (NULL);
259 }
260 meta = (struct nvidia_raid_conf *)buf;
261
262 /* Check if this is an NVIDIA RAID struct */
263 if (strncmp(meta->nvidia_id, NVIDIA_MAGIC, strlen(NVIDIA_MAGIC))) {
264 G_RAID_DEBUG(1, "NVIDIA signature check failed on %s", pp->name);
265 g_free(buf);
266 return (NULL);
267 }
268 if (meta->config_size > 128 ||
269 meta->config_size < 30) {
270 G_RAID_DEBUG(1, "NVIDIA metadata size looks wrong: %d",
271 meta->config_size);
272 g_free(buf);
273 return (NULL);
274 }
275 meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK);
276 memcpy(meta, buf, min(sizeof(*meta), pp->sectorsize));
277 g_free(buf);
278
279 /* Check metadata checksum. */
280 for (checksum = 0, ptr = (uint32_t *)meta,
281 i = 0; i < meta->config_size; i++)
282 checksum += *ptr++;
283 if (checksum != 0) {
284 G_RAID_DEBUG(1, "NVIDIA checksum check failed on %s", pp->name);
285 free(meta, M_MD_NVIDIA);
286 return (NULL);
287 }
288
289 /* Check volume state. */
290 if (meta->state != NVIDIA_S_IDLE && meta->state != NVIDIA_S_INIT &&
291 meta->state != NVIDIA_S_REBUILD && meta->state != NVIDIA_S_SYNC) {
292 G_RAID_DEBUG(1, "NVIDIA unknown state on %s (0x%02x)",
293 pp->name, meta->state);
294 free(meta, M_MD_NVIDIA);
295 return (NULL);
296 }
297
298 /* Check raid type. */
299 if (meta->type != NVIDIA_T_RAID0 && meta->type != NVIDIA_T_RAID1 &&
300 meta->type != NVIDIA_T_RAID3 && meta->type != NVIDIA_T_RAID5 &&
301 meta->type != NVIDIA_T_RAID5_SYM &&
302 meta->type != NVIDIA_T_RAID01 && meta->type != NVIDIA_T_CONCAT) {
303 G_RAID_DEBUG(1, "NVIDIA unknown RAID level on %s (0x%02x)",
304 pp->name, meta->type);
305 free(meta, M_MD_NVIDIA);
306 return (NULL);
307 }
308
309 return (meta);
310 }
311
312 static int
nvidia_meta_write(struct g_consumer * cp,struct nvidia_raid_conf * meta)313 nvidia_meta_write(struct g_consumer *cp, struct nvidia_raid_conf *meta)
314 {
315 struct g_provider *pp;
316 char *buf;
317 int error, i;
318 uint32_t checksum, *ptr;
319
320 pp = cp->provider;
321
322 /* Recalculate checksum for case if metadata were changed. */
323 meta->checksum = 0;
324 for (checksum = 0, ptr = (uint32_t *)meta,
325 i = 0; i < meta->config_size; i++)
326 checksum += *ptr++;
327 meta->checksum -= checksum;
328
329 /* Create and fill buffer. */
330 buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
331 memcpy(buf, meta, sizeof(*meta));
332
333 /* Write metadata. */
334 error = g_write_data(cp,
335 pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
336 if (error != 0) {
337 G_RAID_DEBUG(1, "Cannot write metadata to %s (error=%d).",
338 pp->name, error);
339 }
340
341 free(buf, M_MD_NVIDIA);
342 return (error);
343 }
344
345 static int
nvidia_meta_erase(struct g_consumer * cp)346 nvidia_meta_erase(struct g_consumer *cp)
347 {
348 struct g_provider *pp;
349 char *buf;
350 int error;
351
352 pp = cp->provider;
353 buf = malloc(pp->sectorsize, M_MD_NVIDIA, M_WAITOK | M_ZERO);
354 error = g_write_data(cp,
355 pp->mediasize - 2 * pp->sectorsize, buf, pp->sectorsize);
356 if (error != 0) {
357 G_RAID_DEBUG(1, "Cannot erase metadata on %s (error=%d).",
358 pp->name, error);
359 }
360 free(buf, M_MD_NVIDIA);
361 return (error);
362 }
363
364 static struct g_raid_disk *
g_raid_md_nvidia_get_disk(struct g_raid_softc * sc,int id)365 g_raid_md_nvidia_get_disk(struct g_raid_softc *sc, int id)
366 {
367 struct g_raid_disk *disk;
368 struct g_raid_md_nvidia_perdisk *pd;
369
370 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
371 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
372 if (pd->pd_disk_pos == id)
373 break;
374 }
375 return (disk);
376 }
377
378 static int
g_raid_md_nvidia_supported(int level,int qual,int disks,int force)379 g_raid_md_nvidia_supported(int level, int qual, int disks, int force)
380 {
381
382 switch (level) {
383 case G_RAID_VOLUME_RL_RAID0:
384 if (disks < 1)
385 return (0);
386 if (!force && (disks < 2 || disks > 6))
387 return (0);
388 break;
389 case G_RAID_VOLUME_RL_RAID1:
390 if (disks < 1)
391 return (0);
392 if (!force && (disks != 2))
393 return (0);
394 break;
395 case G_RAID_VOLUME_RL_RAID1E:
396 if (disks < 2)
397 return (0);
398 if (disks % 2 != 0)
399 return (0);
400 if (!force && (disks < 4))
401 return (0);
402 break;
403 case G_RAID_VOLUME_RL_SINGLE:
404 if (disks != 1)
405 return (0);
406 break;
407 case G_RAID_VOLUME_RL_CONCAT:
408 if (disks < 2)
409 return (0);
410 break;
411 case G_RAID_VOLUME_RL_RAID5:
412 if (disks < 3)
413 return (0);
414 if (qual != G_RAID_VOLUME_RLQ_R5LA &&
415 qual != G_RAID_VOLUME_RLQ_R5LS)
416 return (0);
417 break;
418 default:
419 return (0);
420 }
421 if (level != G_RAID_VOLUME_RL_RAID5 && qual != G_RAID_VOLUME_RLQ_NONE)
422 return (0);
423 return (1);
424 }
425
426 static int
g_raid_md_nvidia_start_disk(struct g_raid_disk * disk)427 g_raid_md_nvidia_start_disk(struct g_raid_disk *disk)
428 {
429 struct g_raid_softc *sc;
430 struct g_raid_subdisk *sd, *tmpsd;
431 struct g_raid_disk *olddisk, *tmpdisk;
432 struct g_raid_md_object *md;
433 struct g_raid_md_nvidia_object *mdi;
434 struct g_raid_md_nvidia_perdisk *pd, *oldpd;
435 struct nvidia_raid_conf *meta;
436 int disk_pos, resurrection = 0;
437
438 sc = disk->d_softc;
439 md = sc->sc_md;
440 mdi = (struct g_raid_md_nvidia_object *)md;
441 meta = mdi->mdio_meta;
442 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
443 olddisk = NULL;
444
445 /* Find disk position in metadata by its serial. */
446 if (pd->pd_meta != NULL) {
447 disk_pos = pd->pd_meta->disk_number;
448 if (disk_pos >= meta->total_disks || mdi->mdio_started)
449 disk_pos = -3;
450 } else
451 disk_pos = -3;
452 /* For RAID0+1 we need to translate order. */
453 disk_pos = nvidia_meta_translate_disk(meta, disk_pos);
454 if (disk_pos < 0) {
455 G_RAID_DEBUG1(1, sc, "Unknown, probably new or stale disk");
456 /* If we are in the start process, that's all for now. */
457 if (!mdi->mdio_started)
458 goto nofit;
459 /*
460 * If we have already started - try to get use of the disk.
461 * Try to replace OFFLINE disks first, then FAILED.
462 */
463 TAILQ_FOREACH(tmpdisk, &sc->sc_disks, d_next) {
464 if (tmpdisk->d_state != G_RAID_DISK_S_OFFLINE &&
465 tmpdisk->d_state != G_RAID_DISK_S_FAILED)
466 continue;
467 /* Make sure this disk is big enough. */
468 TAILQ_FOREACH(sd, &tmpdisk->d_subdisks, sd_next) {
469 if (sd->sd_offset + sd->sd_size + 2 * 512 >
470 pd->pd_disk_size) {
471 G_RAID_DEBUG1(1, sc,
472 "Disk too small (%ju < %ju)",
473 pd->pd_disk_size,
474 sd->sd_offset + sd->sd_size + 512);
475 break;
476 }
477 }
478 if (sd != NULL)
479 continue;
480 if (tmpdisk->d_state == G_RAID_DISK_S_OFFLINE) {
481 olddisk = tmpdisk;
482 break;
483 } else if (olddisk == NULL)
484 olddisk = tmpdisk;
485 }
486 if (olddisk == NULL) {
487 nofit:
488 g_raid_change_disk_state(disk, G_RAID_DISK_S_SPARE);
489 return (1);
490 }
491 oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
492 disk_pos = oldpd->pd_disk_pos;
493 resurrection = 1;
494 }
495
496 if (olddisk == NULL) {
497 /* Find placeholder by position. */
498 olddisk = g_raid_md_nvidia_get_disk(sc, disk_pos);
499 if (olddisk == NULL)
500 panic("No disk at position %d!", disk_pos);
501 if (olddisk->d_state != G_RAID_DISK_S_OFFLINE) {
502 G_RAID_DEBUG1(1, sc, "More than one disk for pos %d",
503 disk_pos);
504 g_raid_change_disk_state(disk, G_RAID_DISK_S_STALE);
505 return (0);
506 }
507 oldpd = (struct g_raid_md_nvidia_perdisk *)olddisk->d_md_data;
508 }
509
510 /* Replace failed disk or placeholder with new disk. */
511 TAILQ_FOREACH_SAFE(sd, &olddisk->d_subdisks, sd_next, tmpsd) {
512 TAILQ_REMOVE(&olddisk->d_subdisks, sd, sd_next);
513 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
514 sd->sd_disk = disk;
515 }
516 oldpd->pd_disk_pos = -2;
517 pd->pd_disk_pos = disk_pos;
518
519 /* If it was placeholder -- destroy it. */
520 if (olddisk->d_state == G_RAID_DISK_S_OFFLINE) {
521 g_raid_destroy_disk(olddisk);
522 } else {
523 /* Otherwise, make it STALE_FAILED. */
524 g_raid_change_disk_state(olddisk, G_RAID_DISK_S_STALE_FAILED);
525 }
526
527 /* Welcome the new disk. */
528 if (resurrection)
529 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
530 else// if (pd->pd_meta->disk_status == NVIDIA_S_CURRENT ||
531 //pd->pd_meta->disk_status == NVIDIA_S_REBUILD)
532 g_raid_change_disk_state(disk, G_RAID_DISK_S_ACTIVE);
533 // else
534 // g_raid_change_disk_state(disk, G_RAID_DISK_S_FAILED);
535 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
536 /*
537 * Different disks may have different sizes,
538 * in concat mode. Update from real disk size.
539 */
540 if (meta->type == NVIDIA_T_CONCAT)
541 sd->sd_size = pd->pd_disk_size - 0x800 * 512;
542
543 if (resurrection) {
544 /* New or ex-spare disk. */
545 g_raid_change_subdisk_state(sd,
546 G_RAID_SUBDISK_S_NEW);
547 } else if (meta->state == NVIDIA_S_REBUILD &&
548 (pd->pd_meta->disk_status & 0x100)) {
549 /* Rebuilding disk. */
550 g_raid_change_subdisk_state(sd,
551 G_RAID_SUBDISK_S_REBUILD);
552 sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
553 meta->array_width * pd->pd_meta->sector_size;
554 } else if (meta->state == NVIDIA_S_SYNC) {
555 /* Resyncing/dirty disk. */
556 g_raid_change_subdisk_state(sd,
557 G_RAID_SUBDISK_S_RESYNC);
558 sd->sd_rebuild_pos = (off_t)pd->pd_meta->rebuild_lba /
559 meta->array_width * pd->pd_meta->sector_size;
560 } else {
561 /* Up to date disk. */
562 g_raid_change_subdisk_state(sd,
563 G_RAID_SUBDISK_S_ACTIVE);
564 }
565 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
566 G_RAID_EVENT_SUBDISK);
567 }
568
569 /* Update status of our need for spare. */
570 if (mdi->mdio_started) {
571 mdi->mdio_incomplete =
572 (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
573 mdi->mdio_total_disks);
574 }
575
576 return (resurrection);
577 }
578
579 static void
g_disk_md_nvidia_retaste(void * arg,int pending)580 g_disk_md_nvidia_retaste(void *arg, int pending)
581 {
582
583 G_RAID_DEBUG(1, "Array is not complete, trying to retaste.");
584 g_retaste(&g_raid_class);
585 free(arg, M_MD_NVIDIA);
586 }
587
588 static void
g_raid_md_nvidia_refill(struct g_raid_softc * sc)589 g_raid_md_nvidia_refill(struct g_raid_softc *sc)
590 {
591 struct g_raid_md_object *md;
592 struct g_raid_md_nvidia_object *mdi;
593 struct g_raid_disk *disk;
594 struct task *task;
595 int update, na;
596
597 md = sc->sc_md;
598 mdi = (struct g_raid_md_nvidia_object *)md;
599 update = 0;
600 do {
601 /* Make sure we miss anything. */
602 na = g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE);
603 if (na == mdi->mdio_total_disks)
604 break;
605
606 G_RAID_DEBUG1(1, md->mdo_softc,
607 "Array is not complete (%d of %d), "
608 "trying to refill.", na, mdi->mdio_total_disks);
609
610 /* Try to get use some of STALE disks. */
611 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
612 if (disk->d_state == G_RAID_DISK_S_STALE) {
613 update += g_raid_md_nvidia_start_disk(disk);
614 if (disk->d_state == G_RAID_DISK_S_ACTIVE)
615 break;
616 }
617 }
618 if (disk != NULL)
619 continue;
620
621 /* Try to get use some of SPARE disks. */
622 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
623 if (disk->d_state == G_RAID_DISK_S_SPARE) {
624 update += g_raid_md_nvidia_start_disk(disk);
625 if (disk->d_state == G_RAID_DISK_S_ACTIVE)
626 break;
627 }
628 }
629 } while (disk != NULL);
630
631 /* Write new metadata if we changed something. */
632 if (update)
633 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
634
635 /* Update status of our need for spare. */
636 mdi->mdio_incomplete = (g_raid_ndisks(sc, G_RAID_DISK_S_ACTIVE) <
637 mdi->mdio_total_disks);
638
639 /* Request retaste hoping to find spare. */
640 if (mdi->mdio_incomplete) {
641 task = malloc(sizeof(struct task),
642 M_MD_NVIDIA, M_WAITOK | M_ZERO);
643 TASK_INIT(task, 0, g_disk_md_nvidia_retaste, task);
644 taskqueue_enqueue(taskqueue_swi, task);
645 }
646 }
647
648 static void
g_raid_md_nvidia_start(struct g_raid_softc * sc)649 g_raid_md_nvidia_start(struct g_raid_softc *sc)
650 {
651 struct g_raid_md_object *md;
652 struct g_raid_md_nvidia_object *mdi;
653 struct g_raid_md_nvidia_perdisk *pd;
654 struct nvidia_raid_conf *meta;
655 struct g_raid_volume *vol;
656 struct g_raid_subdisk *sd;
657 struct g_raid_disk *disk;
658 off_t size;
659 int j, disk_pos;
660 char buf[17];
661
662 md = sc->sc_md;
663 mdi = (struct g_raid_md_nvidia_object *)md;
664 meta = mdi->mdio_meta;
665
666 /* Create volumes and subdisks. */
667 nvidia_meta_get_name(meta, buf);
668 vol = g_raid_create_volume(sc, buf, -1);
669 vol->v_mediasize = (off_t)meta->total_sectors * 512;
670 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_NONE;
671 if (meta->type == NVIDIA_T_RAID0) {
672 vol->v_raid_level = G_RAID_VOLUME_RL_RAID0;
673 size = vol->v_mediasize / mdi->mdio_total_disks;
674 } else if (meta->type == NVIDIA_T_RAID1) {
675 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1;
676 size = vol->v_mediasize;
677 } else if (meta->type == NVIDIA_T_RAID01) {
678 vol->v_raid_level = G_RAID_VOLUME_RL_RAID1E;
679 size = vol->v_mediasize / (mdi->mdio_total_disks / 2);
680 } else if (meta->type == NVIDIA_T_CONCAT) {
681 if (mdi->mdio_total_disks == 1)
682 vol->v_raid_level = G_RAID_VOLUME_RL_SINGLE;
683 else
684 vol->v_raid_level = G_RAID_VOLUME_RL_CONCAT;
685 size = 0;
686 } else if (meta->type == NVIDIA_T_RAID5) {
687 vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
688 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LA;
689 size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
690 } else if (meta->type == NVIDIA_T_RAID5_SYM) {
691 vol->v_raid_level = G_RAID_VOLUME_RL_RAID5;
692 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_R5LS;
693 size = vol->v_mediasize / (mdi->mdio_total_disks - 1);
694 } else {
695 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
696 size = 0;
697 }
698 vol->v_strip_size = meta->strip_sectors * 512; //ZZZ
699 vol->v_disks_count = mdi->mdio_total_disks;
700 vol->v_sectorsize = 512; //ZZZ
701 for (j = 0; j < vol->v_disks_count; j++) {
702 sd = &vol->v_subdisks[j];
703 sd->sd_offset = 0;
704 sd->sd_size = size;
705 }
706 g_raid_start_volume(vol);
707
708 /* Create disk placeholders to store data for later writing. */
709 for (disk_pos = 0; disk_pos < mdi->mdio_total_disks; disk_pos++) {
710 pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
711 pd->pd_disk_pos = disk_pos;
712 disk = g_raid_create_disk(sc);
713 disk->d_md_data = (void *)pd;
714 disk->d_state = G_RAID_DISK_S_OFFLINE;
715 sd = &vol->v_subdisks[disk_pos];
716 sd->sd_disk = disk;
717 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
718 }
719
720 /* Make all disks found till the moment take their places. */
721 do {
722 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
723 if (disk->d_state == G_RAID_DISK_S_NONE) {
724 g_raid_md_nvidia_start_disk(disk);
725 break;
726 }
727 }
728 } while (disk != NULL);
729
730 mdi->mdio_started = 1;
731 G_RAID_DEBUG1(0, sc, "Array started.");
732 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
733
734 /* Pickup any STALE/SPARE disks to refill array if needed. */
735 g_raid_md_nvidia_refill(sc);
736
737 g_raid_event_send(vol, G_RAID_VOLUME_E_START, G_RAID_EVENT_VOLUME);
738
739 callout_stop(&mdi->mdio_start_co);
740 G_RAID_DEBUG1(1, sc, "root_mount_rel %p", mdi->mdio_rootmount);
741 root_mount_rel(mdi->mdio_rootmount);
742 mdi->mdio_rootmount = NULL;
743 }
744
745 static void
g_raid_md_nvidia_new_disk(struct g_raid_disk * disk)746 g_raid_md_nvidia_new_disk(struct g_raid_disk *disk)
747 {
748 struct g_raid_softc *sc;
749 struct g_raid_md_object *md;
750 struct g_raid_md_nvidia_object *mdi;
751 struct nvidia_raid_conf *pdmeta;
752 struct g_raid_md_nvidia_perdisk *pd;
753
754 sc = disk->d_softc;
755 md = sc->sc_md;
756 mdi = (struct g_raid_md_nvidia_object *)md;
757 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
758 pdmeta = pd->pd_meta;
759
760 if (mdi->mdio_started) {
761 if (g_raid_md_nvidia_start_disk(disk))
762 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
763 } else {
764 if (mdi->mdio_meta == NULL ||
765 mdi->mdio_meta->disk_number >= mdi->mdio_meta->total_disks) {
766 G_RAID_DEBUG1(1, sc, "Newer disk");
767 if (mdi->mdio_meta != NULL)
768 free(mdi->mdio_meta, M_MD_NVIDIA);
769 mdi->mdio_meta = nvidia_meta_copy(pdmeta);
770 mdi->mdio_total_disks = pdmeta->total_disks;
771 mdi->mdio_disks_present = 1;
772 } else if (pdmeta->disk_number < mdi->mdio_meta->total_disks) {
773 mdi->mdio_disks_present++;
774 G_RAID_DEBUG1(1, sc, "Matching disk (%d of %d up)",
775 mdi->mdio_disks_present,
776 mdi->mdio_total_disks);
777 } else
778 G_RAID_DEBUG1(1, sc, "Spare disk");
779
780 /* If we collected all needed disks - start array. */
781 if (mdi->mdio_disks_present == mdi->mdio_total_disks)
782 g_raid_md_nvidia_start(sc);
783 }
784 }
785
786 static void
g_raid_nvidia_go(void * arg)787 g_raid_nvidia_go(void *arg)
788 {
789 struct g_raid_softc *sc;
790 struct g_raid_md_object *md;
791 struct g_raid_md_nvidia_object *mdi;
792
793 sc = arg;
794 md = sc->sc_md;
795 mdi = (struct g_raid_md_nvidia_object *)md;
796 if (!mdi->mdio_started) {
797 G_RAID_DEBUG1(0, sc, "Force array start due to timeout.");
798 g_raid_event_send(sc, G_RAID_NODE_E_START, 0);
799 }
800 }
801
802 static int
g_raid_md_create_nvidia(struct g_raid_md_object * md,struct g_class * mp,struct g_geom ** gp)803 g_raid_md_create_nvidia(struct g_raid_md_object *md, struct g_class *mp,
804 struct g_geom **gp)
805 {
806 struct g_raid_softc *sc;
807 struct g_raid_md_nvidia_object *mdi;
808 char name[32];
809
810 mdi = (struct g_raid_md_nvidia_object *)md;
811 arc4rand(&mdi->mdio_volume_id, 16, 0);
812 snprintf(name, sizeof(name), "NVIDIA-%d",
813 atomic_fetchadd_int(&NVIDIANodeID, 1));
814 sc = g_raid_create_node(mp, name, md);
815 if (sc == NULL)
816 return (G_RAID_MD_TASTE_FAIL);
817 md->mdo_softc = sc;
818 *gp = sc->sc_geom;
819 return (G_RAID_MD_TASTE_NEW);
820 }
821
822 static int
g_raid_md_taste_nvidia(struct g_raid_md_object * md,struct g_class * mp,struct g_consumer * cp,struct g_geom ** gp)823 g_raid_md_taste_nvidia(struct g_raid_md_object *md, struct g_class *mp,
824 struct g_consumer *cp, struct g_geom **gp)
825 {
826 struct g_consumer *rcp;
827 struct g_provider *pp;
828 struct g_raid_md_nvidia_object *mdi, *mdi1;
829 struct g_raid_softc *sc;
830 struct g_raid_disk *disk;
831 struct nvidia_raid_conf *meta;
832 struct g_raid_md_nvidia_perdisk *pd;
833 struct g_geom *geom;
834 int result, spare, len;
835 char name[32];
836 uint16_t vendor;
837
838 G_RAID_DEBUG(1, "Tasting NVIDIA on %s", cp->provider->name);
839 mdi = (struct g_raid_md_nvidia_object *)md;
840 pp = cp->provider;
841
842 /* Read metadata from device. */
843 meta = NULL;
844 g_topology_unlock();
845 vendor = 0xffff;
846 len = sizeof(vendor);
847 if (pp->geom->rank == 1)
848 g_io_getattr("GEOM::hba_vendor", cp, &len, &vendor);
849 meta = nvidia_meta_read(cp);
850 g_topology_lock();
851 if (meta == NULL) {
852 if (g_raid_aggressive_spare) {
853 if (vendor == 0x10de) {
854 G_RAID_DEBUG(1,
855 "No NVIDIA metadata, forcing spare.");
856 spare = 2;
857 goto search;
858 } else {
859 G_RAID_DEBUG(1,
860 "NVIDIA vendor mismatch 0x%04x != 0x10de",
861 vendor);
862 }
863 }
864 return (G_RAID_MD_TASTE_FAIL);
865 }
866
867 /* Metadata valid. Print it. */
868 g_raid_md_nvidia_print(meta);
869 G_RAID_DEBUG(1, "NVIDIA disk position %d", meta->disk_number);
870 spare = 0;//(meta->type == NVIDIA_T_SPARE) ? 1 : 0;
871
872 search:
873 /* Search for matching node. */
874 sc = NULL;
875 mdi1 = NULL;
876 LIST_FOREACH(geom, &mp->geom, geom) {
877 sc = geom->softc;
878 if (sc == NULL)
879 continue;
880 if (sc->sc_stopping != 0)
881 continue;
882 if (sc->sc_md->mdo_class != md->mdo_class)
883 continue;
884 mdi1 = (struct g_raid_md_nvidia_object *)sc->sc_md;
885 if (spare) {
886 if (mdi1->mdio_incomplete)
887 break;
888 } else {
889 if (memcmp(&mdi1->mdio_volume_id,
890 &meta->volume_id, 16) == 0)
891 break;
892 }
893 }
894
895 /* Found matching node. */
896 if (geom != NULL) {
897 G_RAID_DEBUG(1, "Found matching array %s", sc->sc_name);
898 result = G_RAID_MD_TASTE_EXISTING;
899
900 } else if (spare) { /* Not found needy node -- left for later. */
901 G_RAID_DEBUG(1, "Spare is not needed at this time");
902 goto fail1;
903
904 } else { /* Not found matching node -- create one. */
905 result = G_RAID_MD_TASTE_NEW;
906 memcpy(&mdi->mdio_volume_id, &meta->volume_id, 16);
907 snprintf(name, sizeof(name), "NVIDIA-%d",
908 atomic_fetchadd_int(&NVIDIANodeID, 1));
909 sc = g_raid_create_node(mp, name, md);
910 md->mdo_softc = sc;
911 geom = sc->sc_geom;
912 callout_init(&mdi->mdio_start_co, 1);
913 callout_reset(&mdi->mdio_start_co, g_raid_start_timeout * hz,
914 g_raid_nvidia_go, sc);
915 mdi->mdio_rootmount = root_mount_hold("GRAID-NVIDIA");
916 G_RAID_DEBUG1(1, sc, "root_mount_hold %p", mdi->mdio_rootmount);
917 }
918
919 /* There is no return after this point, so we close passed consumer. */
920 g_access(cp, -1, 0, 0);
921
922 rcp = g_new_consumer(geom);
923 rcp->flags |= G_CF_DIRECT_RECEIVE;
924 g_attach(rcp, pp);
925 if (g_access(rcp, 1, 1, 1) != 0)
926 ; //goto fail1;
927
928 g_topology_unlock();
929 sx_xlock(&sc->sc_lock);
930
931 pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
932 pd->pd_meta = meta;
933 if (spare == 2) {
934 pd->pd_disk_pos = -3;
935 } else {
936 pd->pd_disk_pos = -1;
937 }
938 pd->pd_disk_size = pp->mediasize;
939 disk = g_raid_create_disk(sc);
940 disk->d_md_data = (void *)pd;
941 disk->d_consumer = rcp;
942 rcp->private = disk;
943
944 g_raid_get_disk_info(disk);
945
946 g_raid_md_nvidia_new_disk(disk);
947
948 sx_xunlock(&sc->sc_lock);
949 g_topology_lock();
950 *gp = geom;
951 return (result);
952 fail1:
953 free(meta, M_MD_NVIDIA);
954 return (G_RAID_MD_TASTE_FAIL);
955 }
956
957 static int
g_raid_md_event_nvidia(struct g_raid_md_object * md,struct g_raid_disk * disk,u_int event)958 g_raid_md_event_nvidia(struct g_raid_md_object *md,
959 struct g_raid_disk *disk, u_int event)
960 {
961 struct g_raid_softc *sc;
962 struct g_raid_subdisk *sd;
963 struct g_raid_md_nvidia_object *mdi;
964 struct g_raid_md_nvidia_perdisk *pd;
965
966 sc = md->mdo_softc;
967 mdi = (struct g_raid_md_nvidia_object *)md;
968 if (disk == NULL) {
969 switch (event) {
970 case G_RAID_NODE_E_START:
971 if (!mdi->mdio_started) {
972 /* Bump volume ID to drop missing disks. */
973 arc4rand(&mdi->mdio_volume_id, 16, 0);
974 g_raid_md_nvidia_start(sc);
975 }
976 return (0);
977 }
978 return (-1);
979 }
980 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
981 switch (event) {
982 case G_RAID_DISK_E_DISCONNECTED:
983 /* If disk was assigned, just update statuses. */
984 if (pd->pd_disk_pos >= 0) {
985 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
986 if (disk->d_consumer) {
987 g_raid_kill_consumer(sc, disk->d_consumer);
988 disk->d_consumer = NULL;
989 }
990 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
991 g_raid_change_subdisk_state(sd,
992 G_RAID_SUBDISK_S_NONE);
993 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
994 G_RAID_EVENT_SUBDISK);
995 }
996 } else {
997 /* Otherwise -- delete. */
998 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
999 g_raid_destroy_disk(disk);
1000 }
1001
1002 if (mdi->mdio_started) {
1003 /* Bump volume ID to prevent disk resurrection. */
1004 if (pd->pd_disk_pos >= 0)
1005 arc4rand(&mdi->mdio_volume_id, 16, 0);
1006
1007 /* Write updated metadata to all disks. */
1008 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1009 }
1010
1011 /* Check if anything left except placeholders. */
1012 if (g_raid_ndisks(sc, -1) ==
1013 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1014 g_raid_destroy_node(sc, 0);
1015 else
1016 g_raid_md_nvidia_refill(sc);
1017 return (0);
1018 }
1019 return (-2);
1020 }
1021
1022 static int
g_raid_md_ctl_nvidia(struct g_raid_md_object * md,struct gctl_req * req)1023 g_raid_md_ctl_nvidia(struct g_raid_md_object *md,
1024 struct gctl_req *req)
1025 {
1026 struct g_raid_softc *sc;
1027 struct g_raid_volume *vol;
1028 struct g_raid_subdisk *sd;
1029 struct g_raid_disk *disk;
1030 struct g_raid_md_nvidia_object *mdi;
1031 struct g_raid_md_nvidia_perdisk *pd;
1032 struct g_consumer *cp;
1033 struct g_provider *pp;
1034 char arg[16];
1035 const char *verb, *volname, *levelname, *diskname;
1036 int *nargs, *force;
1037 off_t size, sectorsize, strip, volsize;
1038 intmax_t *sizearg, *striparg;
1039 int numdisks, i, len, level, qual, update;
1040 int error;
1041
1042 sc = md->mdo_softc;
1043 mdi = (struct g_raid_md_nvidia_object *)md;
1044 verb = gctl_get_param(req, "verb", NULL);
1045 nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs));
1046 error = 0;
1047 if (strcmp(verb, "label") == 0) {
1048 if (*nargs < 4) {
1049 gctl_error(req, "Invalid number of arguments.");
1050 return (-1);
1051 }
1052 volname = gctl_get_asciiparam(req, "arg1");
1053 if (volname == NULL) {
1054 gctl_error(req, "No volume name.");
1055 return (-2);
1056 }
1057 levelname = gctl_get_asciiparam(req, "arg2");
1058 if (levelname == NULL) {
1059 gctl_error(req, "No RAID level.");
1060 return (-3);
1061 }
1062 if (strcasecmp(levelname, "RAID5") == 0)
1063 levelname = "RAID5-LS";
1064 if (g_raid_volume_str2level(levelname, &level, &qual)) {
1065 gctl_error(req, "Unknown RAID level '%s'.", levelname);
1066 return (-4);
1067 }
1068 numdisks = *nargs - 3;
1069 force = gctl_get_paraml(req, "force", sizeof(*force));
1070 if (!g_raid_md_nvidia_supported(level, qual, numdisks,
1071 force ? *force : 0)) {
1072 gctl_error(req, "Unsupported RAID level "
1073 "(0x%02x/0x%02x), or number of disks (%d).",
1074 level, qual, numdisks);
1075 return (-5);
1076 }
1077
1078 /* Search for disks, connect them and probe. */
1079 size = 0x7fffffffffffffffllu;
1080 sectorsize = 0;
1081 for (i = 0; i < numdisks; i++) {
1082 snprintf(arg, sizeof(arg), "arg%d", i + 3);
1083 diskname = gctl_get_asciiparam(req, arg);
1084 if (diskname == NULL) {
1085 gctl_error(req, "No disk name (%s).", arg);
1086 error = -6;
1087 break;
1088 }
1089 if (strcmp(diskname, "NONE") == 0) {
1090 cp = NULL;
1091 pp = NULL;
1092 } else {
1093 g_topology_lock();
1094 cp = g_raid_open_consumer(sc, diskname);
1095 if (cp == NULL) {
1096 gctl_error(req, "Can't open '%s'.",
1097 diskname);
1098 g_topology_unlock();
1099 error = -7;
1100 break;
1101 }
1102 pp = cp->provider;
1103 }
1104 pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1105 pd->pd_disk_pos = i;
1106 disk = g_raid_create_disk(sc);
1107 disk->d_md_data = (void *)pd;
1108 disk->d_consumer = cp;
1109 if (cp == NULL)
1110 continue;
1111 cp->private = disk;
1112 g_topology_unlock();
1113
1114 g_raid_get_disk_info(disk);
1115
1116 pd->pd_disk_size = pp->mediasize;
1117 if (size > pp->mediasize)
1118 size = pp->mediasize;
1119 if (sectorsize < pp->sectorsize)
1120 sectorsize = pp->sectorsize;
1121 }
1122 if (error != 0)
1123 return (error);
1124
1125 if (sectorsize <= 0) {
1126 gctl_error(req, "Can't get sector size.");
1127 return (-8);
1128 }
1129
1130 /* Reserve space for metadata. */
1131 size -= 2 * sectorsize;
1132
1133 /* Handle size argument. */
1134 len = sizeof(*sizearg);
1135 sizearg = gctl_get_param(req, "size", &len);
1136 if (sizearg != NULL && len == sizeof(*sizearg) &&
1137 *sizearg > 0) {
1138 if (*sizearg > size) {
1139 gctl_error(req, "Size too big %lld > %lld.",
1140 (long long)*sizearg, (long long)size);
1141 return (-9);
1142 }
1143 size = *sizearg;
1144 }
1145
1146 /* Handle strip argument. */
1147 strip = 131072;
1148 len = sizeof(*striparg);
1149 striparg = gctl_get_param(req, "strip", &len);
1150 if (striparg != NULL && len == sizeof(*striparg) &&
1151 *striparg > 0) {
1152 if (*striparg < sectorsize) {
1153 gctl_error(req, "Strip size too small.");
1154 return (-10);
1155 }
1156 if (*striparg % sectorsize != 0) {
1157 gctl_error(req, "Incorrect strip size.");
1158 return (-11);
1159 }
1160 if (strip > 65535 * sectorsize) {
1161 gctl_error(req, "Strip size too big.");
1162 return (-12);
1163 }
1164 strip = *striparg;
1165 }
1166
1167 /* Round size down to strip or sector. */
1168 if (level == G_RAID_VOLUME_RL_RAID1)
1169 size -= (size % sectorsize);
1170 else if (level == G_RAID_VOLUME_RL_RAID1E &&
1171 (numdisks & 1) != 0)
1172 size -= (size % (2 * strip));
1173 else
1174 size -= (size % strip);
1175 if (size <= 0) {
1176 gctl_error(req, "Size too small.");
1177 return (-13);
1178 }
1179
1180 if (level == G_RAID_VOLUME_RL_RAID0 ||
1181 level == G_RAID_VOLUME_RL_CONCAT ||
1182 level == G_RAID_VOLUME_RL_SINGLE)
1183 volsize = size * numdisks;
1184 else if (level == G_RAID_VOLUME_RL_RAID1)
1185 volsize = size;
1186 else if (level == G_RAID_VOLUME_RL_RAID5)
1187 volsize = size * (numdisks - 1);
1188 else { /* RAID1E */
1189 volsize = ((size * numdisks) / strip / 2) *
1190 strip;
1191 }
1192 if (volsize > 0xffffffffllu * sectorsize) {
1193 gctl_error(req, "Size too big.");
1194 return (-14);
1195 }
1196
1197 /* We have all we need, create things: volume, ... */
1198 mdi->mdio_total_disks = numdisks;
1199 mdi->mdio_started = 1;
1200 vol = g_raid_create_volume(sc, volname, -1);
1201 vol->v_md_data = (void *)(intptr_t)0;
1202 vol->v_raid_level = level;
1203 vol->v_raid_level_qualifier = qual;
1204 vol->v_strip_size = strip;
1205 vol->v_disks_count = numdisks;
1206 vol->v_mediasize = volsize;
1207 vol->v_sectorsize = sectorsize;
1208 g_raid_start_volume(vol);
1209
1210 /* , and subdisks. */
1211 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1212 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1213 sd = &vol->v_subdisks[pd->pd_disk_pos];
1214 sd->sd_disk = disk;
1215 sd->sd_offset = 0;
1216 sd->sd_size = size;
1217 TAILQ_INSERT_TAIL(&disk->d_subdisks, sd, sd_next);
1218 if (sd->sd_disk->d_consumer != NULL) {
1219 g_raid_change_disk_state(disk,
1220 G_RAID_DISK_S_ACTIVE);
1221 g_raid_change_subdisk_state(sd,
1222 G_RAID_SUBDISK_S_ACTIVE);
1223 g_raid_event_send(sd, G_RAID_SUBDISK_E_NEW,
1224 G_RAID_EVENT_SUBDISK);
1225 } else {
1226 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1227 }
1228 }
1229
1230 /* Write metadata based on created entities. */
1231 G_RAID_DEBUG1(0, sc, "Array started.");
1232 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1233
1234 /* Pickup any STALE/SPARE disks to refill array if needed. */
1235 g_raid_md_nvidia_refill(sc);
1236
1237 g_raid_event_send(vol, G_RAID_VOLUME_E_START,
1238 G_RAID_EVENT_VOLUME);
1239 return (0);
1240 }
1241 if (strcmp(verb, "delete") == 0) {
1242 /* Check if some volume is still open. */
1243 force = gctl_get_paraml(req, "force", sizeof(*force));
1244 if (force != NULL && *force == 0 &&
1245 g_raid_nopens(sc) != 0) {
1246 gctl_error(req, "Some volume is still open.");
1247 return (-4);
1248 }
1249
1250 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1251 if (disk->d_consumer)
1252 nvidia_meta_erase(disk->d_consumer);
1253 }
1254 g_raid_destroy_node(sc, 0);
1255 return (0);
1256 }
1257 if (strcmp(verb, "remove") == 0 ||
1258 strcmp(verb, "fail") == 0) {
1259 if (*nargs < 2) {
1260 gctl_error(req, "Invalid number of arguments.");
1261 return (-1);
1262 }
1263 for (i = 1; i < *nargs; i++) {
1264 snprintf(arg, sizeof(arg), "arg%d", i);
1265 diskname = gctl_get_asciiparam(req, arg);
1266 if (diskname == NULL) {
1267 gctl_error(req, "No disk name (%s).", arg);
1268 error = -2;
1269 break;
1270 }
1271 if (strncmp(diskname, _PATH_DEV, 5) == 0)
1272 diskname += 5;
1273
1274 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1275 if (disk->d_consumer != NULL &&
1276 disk->d_consumer->provider != NULL &&
1277 strcmp(disk->d_consumer->provider->name,
1278 diskname) == 0)
1279 break;
1280 }
1281 if (disk == NULL) {
1282 gctl_error(req, "Disk '%s' not found.",
1283 diskname);
1284 error = -3;
1285 break;
1286 }
1287
1288 if (strcmp(verb, "fail") == 0) {
1289 g_raid_md_fail_disk_nvidia(md, NULL, disk);
1290 continue;
1291 }
1292
1293 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1294
1295 /* Erase metadata on deleting disk. */
1296 nvidia_meta_erase(disk->d_consumer);
1297
1298 /* If disk was assigned, just update statuses. */
1299 if (pd->pd_disk_pos >= 0) {
1300 g_raid_change_disk_state(disk, G_RAID_DISK_S_OFFLINE);
1301 g_raid_kill_consumer(sc, disk->d_consumer);
1302 disk->d_consumer = NULL;
1303 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
1304 g_raid_change_subdisk_state(sd,
1305 G_RAID_SUBDISK_S_NONE);
1306 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1307 G_RAID_EVENT_SUBDISK);
1308 }
1309 } else {
1310 /* Otherwise -- delete. */
1311 g_raid_change_disk_state(disk, G_RAID_DISK_S_NONE);
1312 g_raid_destroy_disk(disk);
1313 }
1314 }
1315
1316 /* Write updated metadata to remaining disks. */
1317 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1318
1319 /* Check if anything left except placeholders. */
1320 if (g_raid_ndisks(sc, -1) ==
1321 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1322 g_raid_destroy_node(sc, 0);
1323 else
1324 g_raid_md_nvidia_refill(sc);
1325 return (error);
1326 }
1327 if (strcmp(verb, "insert") == 0) {
1328 if (*nargs < 2) {
1329 gctl_error(req, "Invalid number of arguments.");
1330 return (-1);
1331 }
1332 update = 0;
1333 for (i = 1; i < *nargs; i++) {
1334 /* Get disk name. */
1335 snprintf(arg, sizeof(arg), "arg%d", i);
1336 diskname = gctl_get_asciiparam(req, arg);
1337 if (diskname == NULL) {
1338 gctl_error(req, "No disk name (%s).", arg);
1339 error = -3;
1340 break;
1341 }
1342
1343 /* Try to find provider with specified name. */
1344 g_topology_lock();
1345 cp = g_raid_open_consumer(sc, diskname);
1346 if (cp == NULL) {
1347 gctl_error(req, "Can't open disk '%s'.",
1348 diskname);
1349 g_topology_unlock();
1350 error = -4;
1351 break;
1352 }
1353 pp = cp->provider;
1354
1355 pd = malloc(sizeof(*pd), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1356 pd->pd_disk_pos = -3;
1357 pd->pd_disk_size = pp->mediasize;
1358
1359 disk = g_raid_create_disk(sc);
1360 disk->d_consumer = cp;
1361 disk->d_md_data = (void *)pd;
1362 cp->private = disk;
1363 g_topology_unlock();
1364
1365 g_raid_get_disk_info(disk);
1366
1367 /* Welcome the "new" disk. */
1368 update += g_raid_md_nvidia_start_disk(disk);
1369 if (disk->d_state != G_RAID_DISK_S_SPARE &&
1370 disk->d_state != G_RAID_DISK_S_ACTIVE) {
1371 gctl_error(req, "Disk '%s' doesn't fit.",
1372 diskname);
1373 g_raid_destroy_disk(disk);
1374 error = -8;
1375 break;
1376 }
1377 }
1378
1379 /* Write new metadata if we changed something. */
1380 if (update)
1381 g_raid_md_write_nvidia(md, NULL, NULL, NULL);
1382 return (error);
1383 }
1384 gctl_error(req, "Command '%s' is not supported.", verb);
1385 return (-100);
1386 }
1387
1388 static int
g_raid_md_write_nvidia(struct g_raid_md_object * md,struct g_raid_volume * tvol,struct g_raid_subdisk * tsd,struct g_raid_disk * tdisk)1389 g_raid_md_write_nvidia(struct g_raid_md_object *md, struct g_raid_volume *tvol,
1390 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1391 {
1392 struct g_raid_softc *sc;
1393 struct g_raid_volume *vol;
1394 struct g_raid_subdisk *sd;
1395 struct g_raid_disk *disk;
1396 struct g_raid_md_nvidia_object *mdi;
1397 struct g_raid_md_nvidia_perdisk *pd;
1398 struct nvidia_raid_conf *meta;
1399 int i, spares;
1400
1401 sc = md->mdo_softc;
1402 mdi = (struct g_raid_md_nvidia_object *)md;
1403
1404 if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1405 return (0);
1406
1407 /* There is only one volume. */
1408 vol = TAILQ_FIRST(&sc->sc_volumes);
1409
1410 /* Fill global fields. */
1411 meta = malloc(sizeof(*meta), M_MD_NVIDIA, M_WAITOK | M_ZERO);
1412 if (mdi->mdio_meta)
1413 memcpy(meta, mdi->mdio_meta, sizeof(*meta));
1414 memcpy(meta->nvidia_id, NVIDIA_MAGIC, sizeof(NVIDIA_MAGIC) - 1);
1415 meta->config_size = 30;
1416 meta->version = 0x0064;
1417 meta->total_sectors = vol->v_mediasize / vol->v_sectorsize;
1418 meta->sector_size = vol->v_sectorsize;
1419 nvidia_meta_put_name(meta, vol->v_name);
1420 meta->magic_0 = NVIDIA_MAGIC0;
1421 memcpy(&meta->volume_id, &mdi->mdio_volume_id, 16);
1422 meta->state = NVIDIA_S_IDLE;
1423 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1424 meta->array_width = 1;
1425 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1426 meta->array_width = vol->v_disks_count / 2;
1427 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID5)
1428 meta->array_width = vol->v_disks_count - 1;
1429 else
1430 meta->array_width = vol->v_disks_count;
1431 meta->total_disks = vol->v_disks_count;
1432 meta->orig_array_width = meta->array_width;
1433 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID0)
1434 meta->type = NVIDIA_T_RAID0;
1435 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1)
1436 meta->type = NVIDIA_T_RAID1;
1437 else if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1E)
1438 meta->type = NVIDIA_T_RAID01;
1439 else if (vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT ||
1440 vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE)
1441 meta->type = NVIDIA_T_CONCAT;
1442 else if (vol->v_raid_level_qualifier == G_RAID_VOLUME_RLQ_R5LA)
1443 meta->type = NVIDIA_T_RAID5;
1444 else
1445 meta->type = NVIDIA_T_RAID5_SYM;
1446 meta->strip_sectors = vol->v_strip_size / vol->v_sectorsize;
1447 meta->strip_bytes = vol->v_strip_size;
1448 meta->strip_shift = ffs(meta->strip_sectors) - 1;
1449 meta->strip_mask = meta->strip_sectors - 1;
1450 meta->stripe_sectors = meta->strip_sectors * meta->orig_array_width;
1451 meta->stripe_bytes = meta->stripe_sectors * vol->v_sectorsize;
1452 meta->rebuild_lba = 0;
1453 meta->orig_type = meta->type;
1454 meta->orig_total_sectors = meta->total_sectors;
1455 meta->status = 0;
1456
1457 for (i = 0; i < vol->v_disks_count; i++) {
1458 sd = &vol->v_subdisks[i];
1459 if ((sd->sd_state == G_RAID_SUBDISK_S_STALE ||
1460 sd->sd_state == G_RAID_SUBDISK_S_RESYNC ||
1461 vol->v_dirty) &&
1462 meta->state != NVIDIA_S_REBUILD)
1463 meta->state = NVIDIA_S_SYNC;
1464 else if (sd->sd_state == G_RAID_SUBDISK_S_NEW ||
1465 sd->sd_state == G_RAID_SUBDISK_S_REBUILD)
1466 meta->state = NVIDIA_S_REBUILD;
1467 }
1468
1469 /* We are done. Print meta data and store them to disks. */
1470 if (mdi->mdio_meta != NULL)
1471 free(mdi->mdio_meta, M_MD_NVIDIA);
1472 mdi->mdio_meta = meta;
1473 spares = 0;
1474 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
1475 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1476 if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1477 disk->d_state != G_RAID_DISK_S_SPARE)
1478 continue;
1479 if (pd->pd_meta != NULL) {
1480 free(pd->pd_meta, M_MD_NVIDIA);
1481 pd->pd_meta = NULL;
1482 }
1483 pd->pd_meta = nvidia_meta_copy(meta);
1484 if ((sd = TAILQ_FIRST(&disk->d_subdisks)) != NULL) {
1485 /* For RAID0+1 we need to translate order. */
1486 pd->pd_meta->disk_number =
1487 nvidia_meta_translate_disk(meta, sd->sd_pos);
1488 if (sd->sd_state != G_RAID_SUBDISK_S_ACTIVE) {
1489 pd->pd_meta->disk_status = 0x100;
1490 pd->pd_meta->rebuild_lba =
1491 sd->sd_rebuild_pos / vol->v_sectorsize *
1492 meta->array_width;
1493 }
1494 } else
1495 pd->pd_meta->disk_number = meta->total_disks + spares++;
1496 G_RAID_DEBUG(1, "Writing NVIDIA metadata to %s",
1497 g_raid_get_diskname(disk));
1498 g_raid_md_nvidia_print(pd->pd_meta);
1499 nvidia_meta_write(disk->d_consumer, pd->pd_meta);
1500 }
1501 return (0);
1502 }
1503
1504 static int
g_raid_md_fail_disk_nvidia(struct g_raid_md_object * md,struct g_raid_subdisk * tsd,struct g_raid_disk * tdisk)1505 g_raid_md_fail_disk_nvidia(struct g_raid_md_object *md,
1506 struct g_raid_subdisk *tsd, struct g_raid_disk *tdisk)
1507 {
1508 struct g_raid_softc *sc;
1509 struct g_raid_md_nvidia_perdisk *pd;
1510 struct g_raid_subdisk *sd;
1511
1512 sc = md->mdo_softc;
1513 pd = (struct g_raid_md_nvidia_perdisk *)tdisk->d_md_data;
1514
1515 /* We can't fail disk that is not a part of array now. */
1516 if (pd->pd_disk_pos < 0)
1517 return (-1);
1518
1519 /* Erase metadata to prevent disks's later resurrection. */
1520 if (tdisk->d_consumer)
1521 nvidia_meta_erase(tdisk->d_consumer);
1522
1523 /* Change states. */
1524 g_raid_change_disk_state(tdisk, G_RAID_DISK_S_FAILED);
1525 TAILQ_FOREACH(sd, &tdisk->d_subdisks, sd_next) {
1526 g_raid_change_subdisk_state(sd,
1527 G_RAID_SUBDISK_S_FAILED);
1528 g_raid_event_send(sd, G_RAID_SUBDISK_E_FAILED,
1529 G_RAID_EVENT_SUBDISK);
1530 }
1531
1532 /* Write updated metadata to remaining disks. */
1533 g_raid_md_write_nvidia(md, NULL, NULL, tdisk);
1534
1535 /* Check if anything left except placeholders. */
1536 if (g_raid_ndisks(sc, -1) ==
1537 g_raid_ndisks(sc, G_RAID_DISK_S_OFFLINE))
1538 g_raid_destroy_node(sc, 0);
1539 else
1540 g_raid_md_nvidia_refill(sc);
1541 return (0);
1542 }
1543
1544 static int
g_raid_md_free_disk_nvidia(struct g_raid_md_object * md,struct g_raid_disk * disk)1545 g_raid_md_free_disk_nvidia(struct g_raid_md_object *md,
1546 struct g_raid_disk *disk)
1547 {
1548 struct g_raid_md_nvidia_perdisk *pd;
1549
1550 pd = (struct g_raid_md_nvidia_perdisk *)disk->d_md_data;
1551 if (pd->pd_meta != NULL) {
1552 free(pd->pd_meta, M_MD_NVIDIA);
1553 pd->pd_meta = NULL;
1554 }
1555 free(pd, M_MD_NVIDIA);
1556 disk->d_md_data = NULL;
1557 return (0);
1558 }
1559
1560 static int
g_raid_md_free_nvidia(struct g_raid_md_object * md)1561 g_raid_md_free_nvidia(struct g_raid_md_object *md)
1562 {
1563 struct g_raid_md_nvidia_object *mdi;
1564
1565 mdi = (struct g_raid_md_nvidia_object *)md;
1566 if (!mdi->mdio_started) {
1567 mdi->mdio_started = 0;
1568 callout_stop(&mdi->mdio_start_co);
1569 G_RAID_DEBUG1(1, md->mdo_softc,
1570 "root_mount_rel %p", mdi->mdio_rootmount);
1571 root_mount_rel(mdi->mdio_rootmount);
1572 mdi->mdio_rootmount = NULL;
1573 }
1574 if (mdi->mdio_meta != NULL) {
1575 free(mdi->mdio_meta, M_MD_NVIDIA);
1576 mdi->mdio_meta = NULL;
1577 }
1578 return (0);
1579 }
1580
1581 G_RAID_MD_DECLARE(nvidia, "NVIDIA");
1582