fix mismerge in ll_rw_blk.c
[linux-2.6] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/devfs_fs_kernel.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/suspend.h>
44
45 #include <linux/init.h>
46
47 #include <linux/file.h>
48
49 #ifdef CONFIG_KMOD
50 #include <linux/kmod.h>
51 #endif
52
53 #include <asm/unaligned.h>
54
55 #define MAJOR_NR MD_MAJOR
56 #define MD_DRIVER
57
58 /* 63 partitions with the alternate major number (mdp) */
59 #define MdpMinorShift 6
60
61 #define DEBUG 0
62 #define dprintk(x...) ((void)(DEBUG && printk(x)))
63
64
65 #ifndef MODULE
66 static void autostart_arrays (int part);
67 #endif
68
69 static mdk_personality_t *pers[MAX_PERSONALITY];
70 static DEFINE_SPINLOCK(pers_lock);
71
72 /*
73  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
74  * is 1000 KB/sec, so the extra system load does not show up that much.
75  * Increase it if you want to have more _guaranteed_ speed. Note that
76  * the RAID driver will use the maximum available bandwith if the IO
77  * subsystem is idle. There is also an 'absolute maximum' reconstruction
78  * speed limit - in case reconstruction slows down your system despite
79  * idle IO detection.
80  *
81  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
82  */
83
84 static int sysctl_speed_limit_min = 1000;
85 static int sysctl_speed_limit_max = 200000;
86
87 static struct ctl_table_header *raid_table_header;
88
89 static ctl_table raid_table[] = {
90         {
91                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
92                 .procname       = "speed_limit_min",
93                 .data           = &sysctl_speed_limit_min,
94                 .maxlen         = sizeof(int),
95                 .mode           = 0644,
96                 .proc_handler   = &proc_dointvec,
97         },
98         {
99                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
100                 .procname       = "speed_limit_max",
101                 .data           = &sysctl_speed_limit_max,
102                 .maxlen         = sizeof(int),
103                 .mode           = 0644,
104                 .proc_handler   = &proc_dointvec,
105         },
106         { .ctl_name = 0 }
107 };
108
109 static ctl_table raid_dir_table[] = {
110         {
111                 .ctl_name       = DEV_RAID,
112                 .procname       = "raid",
113                 .maxlen         = 0,
114                 .mode           = 0555,
115                 .child          = raid_table,
116         },
117         { .ctl_name = 0 }
118 };
119
120 static ctl_table raid_root_table[] = {
121         {
122                 .ctl_name       = CTL_DEV,
123                 .procname       = "dev",
124                 .maxlen         = 0,
125                 .mode           = 0555,
126                 .child          = raid_dir_table,
127         },
128         { .ctl_name = 0 }
129 };
130
131 static struct block_device_operations md_fops;
132
133 /*
134  * Enables to iterate over all existing md arrays
135  * all_mddevs_lock protects this list.
136  */
137 static LIST_HEAD(all_mddevs);
138 static DEFINE_SPINLOCK(all_mddevs_lock);
139
140
141 /*
142  * iterates through all used mddevs in the system.
143  * We take care to grab the all_mddevs_lock whenever navigating
144  * the list, and to always hold a refcount when unlocked.
145  * Any code which breaks out of this loop while own
146  * a reference to the current mddev and must mddev_put it.
147  */
148 #define ITERATE_MDDEV(mddev,tmp)                                        \
149                                                                         \
150         for (({ spin_lock(&all_mddevs_lock);                            \
151                 tmp = all_mddevs.next;                                  \
152                 mddev = NULL;});                                        \
153              ({ if (tmp != &all_mddevs)                                 \
154                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
155                 spin_unlock(&all_mddevs_lock);                          \
156                 if (mddev) mddev_put(mddev);                            \
157                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
158                 tmp != &all_mddevs;});                                  \
159              ({ spin_lock(&all_mddevs_lock);                            \
160                 tmp = tmp->next;})                                      \
161                 )
162
163
164 static int md_fail_request (request_queue_t *q, struct bio *bio)
165 {
166         bio_io_error(bio, bio->bi_size);
167         return 0;
168 }
169
170 static inline mddev_t *mddev_get(mddev_t *mddev)
171 {
172         atomic_inc(&mddev->active);
173         return mddev;
174 }
175
176 static void mddev_put(mddev_t *mddev)
177 {
178         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
179                 return;
180         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
181                 list_del(&mddev->all_mddevs);
182                 blk_put_queue(mddev->queue);
183                 kfree(mddev);
184         }
185         spin_unlock(&all_mddevs_lock);
186 }
187
188 static mddev_t * mddev_find(dev_t unit)
189 {
190         mddev_t *mddev, *new = NULL;
191
192  retry:
193         spin_lock(&all_mddevs_lock);
194         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
195                 if (mddev->unit == unit) {
196                         mddev_get(mddev);
197                         spin_unlock(&all_mddevs_lock);
198                         kfree(new);
199                         return mddev;
200                 }
201
202         if (new) {
203                 list_add(&new->all_mddevs, &all_mddevs);
204                 spin_unlock(&all_mddevs_lock);
205                 return new;
206         }
207         spin_unlock(&all_mddevs_lock);
208
209         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
210         if (!new)
211                 return NULL;
212
213         memset(new, 0, sizeof(*new));
214
215         new->unit = unit;
216         if (MAJOR(unit) == MD_MAJOR)
217                 new->md_minor = MINOR(unit);
218         else
219                 new->md_minor = MINOR(unit) >> MdpMinorShift;
220
221         init_MUTEX(&new->reconfig_sem);
222         INIT_LIST_HEAD(&new->disks);
223         INIT_LIST_HEAD(&new->all_mddevs);
224         init_timer(&new->safemode_timer);
225         atomic_set(&new->active, 1);
226         spin_lock_init(&new->write_lock);
227         init_waitqueue_head(&new->sb_wait);
228
229         new->queue = blk_alloc_queue(GFP_KERNEL);
230         if (!new->queue) {
231                 kfree(new);
232                 return NULL;
233         }
234
235         blk_queue_make_request(new->queue, md_fail_request);
236
237         goto retry;
238 }
239
240 static inline int mddev_lock(mddev_t * mddev)
241 {
242         return down_interruptible(&mddev->reconfig_sem);
243 }
244
245 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
246 {
247         down(&mddev->reconfig_sem);
248 }
249
250 static inline int mddev_trylock(mddev_t * mddev)
251 {
252         return down_trylock(&mddev->reconfig_sem);
253 }
254
255 static inline void mddev_unlock(mddev_t * mddev)
256 {
257         up(&mddev->reconfig_sem);
258
259         md_wakeup_thread(mddev->thread);
260 }
261
262 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
263 {
264         mdk_rdev_t * rdev;
265         struct list_head *tmp;
266
267         ITERATE_RDEV(mddev,rdev,tmp) {
268                 if (rdev->desc_nr == nr)
269                         return rdev;
270         }
271         return NULL;
272 }
273
274 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
275 {
276         struct list_head *tmp;
277         mdk_rdev_t *rdev;
278
279         ITERATE_RDEV(mddev,rdev,tmp) {
280                 if (rdev->bdev->bd_dev == dev)
281                         return rdev;
282         }
283         return NULL;
284 }
285
286 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
287 {
288         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
289         return MD_NEW_SIZE_BLOCKS(size);
290 }
291
292 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
293 {
294         sector_t size;
295
296         size = rdev->sb_offset;
297
298         if (chunk_size)
299                 size &= ~((sector_t)chunk_size/1024 - 1);
300         return size;
301 }
302
303 static int alloc_disk_sb(mdk_rdev_t * rdev)
304 {
305         if (rdev->sb_page)
306                 MD_BUG();
307
308         rdev->sb_page = alloc_page(GFP_KERNEL);
309         if (!rdev->sb_page) {
310                 printk(KERN_ALERT "md: out of memory.\n");
311                 return -EINVAL;
312         }
313
314         return 0;
315 }
316
317 static void free_disk_sb(mdk_rdev_t * rdev)
318 {
319         if (rdev->sb_page) {
320                 page_cache_release(rdev->sb_page);
321                 rdev->sb_loaded = 0;
322                 rdev->sb_page = NULL;
323                 rdev->sb_offset = 0;
324                 rdev->size = 0;
325         }
326 }
327
328
329 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
330 {
331         mdk_rdev_t *rdev = bio->bi_private;
332         if (bio->bi_size)
333                 return 1;
334
335         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
336                 md_error(rdev->mddev, rdev);
337
338         if (atomic_dec_and_test(&rdev->mddev->pending_writes))
339                 wake_up(&rdev->mddev->sb_wait);
340         bio_put(bio);
341         return 0;
342 }
343
344 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
345                    sector_t sector, int size, struct page *page)
346 {
347         /* write first size bytes of page to sector of rdev
348          * Increment mddev->pending_writes before returning
349          * and decrement it on completion, waking up sb_wait
350          * if zero is reached.
351          * If an error occurred, call md_error
352          */
353         struct bio *bio = bio_alloc(GFP_NOIO, 1);
354
355         bio->bi_bdev = rdev->bdev;
356         bio->bi_sector = sector;
357         bio_add_page(bio, page, size, 0);
358         bio->bi_private = rdev;
359         bio->bi_end_io = super_written;
360         atomic_inc(&mddev->pending_writes);
361         submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
362 }
363
364 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
365 {
366         if (bio->bi_size)
367                 return 1;
368
369         complete((struct completion*)bio->bi_private);
370         return 0;
371 }
372
373 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
374                    struct page *page, int rw)
375 {
376         struct bio *bio = bio_alloc(GFP_NOIO, 1);
377         struct completion event;
378         int ret;
379
380         rw |= (1 << BIO_RW_SYNC);
381
382         bio->bi_bdev = bdev;
383         bio->bi_sector = sector;
384         bio_add_page(bio, page, size, 0);
385         init_completion(&event);
386         bio->bi_private = &event;
387         bio->bi_end_io = bi_complete;
388         submit_bio(rw, bio);
389         wait_for_completion(&event);
390
391         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
392         bio_put(bio);
393         return ret;
394 }
395
396 static int read_disk_sb(mdk_rdev_t * rdev)
397 {
398         char b[BDEVNAME_SIZE];
399         if (!rdev->sb_page) {
400                 MD_BUG();
401                 return -EINVAL;
402         }
403         if (rdev->sb_loaded)
404                 return 0;
405
406
407         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
408                 goto fail;
409         rdev->sb_loaded = 1;
410         return 0;
411
412 fail:
413         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
414                 bdevname(rdev->bdev,b));
415         return -EINVAL;
416 }
417
418 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
419 {
420         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
421                 (sb1->set_uuid1 == sb2->set_uuid1) &&
422                 (sb1->set_uuid2 == sb2->set_uuid2) &&
423                 (sb1->set_uuid3 == sb2->set_uuid3))
424
425                 return 1;
426
427         return 0;
428 }
429
430
431 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
432 {
433         int ret;
434         mdp_super_t *tmp1, *tmp2;
435
436         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
437         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
438
439         if (!tmp1 || !tmp2) {
440                 ret = 0;
441                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
442                 goto abort;
443         }
444
445         *tmp1 = *sb1;
446         *tmp2 = *sb2;
447
448         /*
449          * nr_disks is not constant
450          */
451         tmp1->nr_disks = 0;
452         tmp2->nr_disks = 0;
453
454         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
455                 ret = 0;
456         else
457                 ret = 1;
458
459 abort:
460         kfree(tmp1);
461         kfree(tmp2);
462         return ret;
463 }
464
465 static unsigned int calc_sb_csum(mdp_super_t * sb)
466 {
467         unsigned int disk_csum, csum;
468
469         disk_csum = sb->sb_csum;
470         sb->sb_csum = 0;
471         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
472         sb->sb_csum = disk_csum;
473         return csum;
474 }
475
476
477 /*
478  * Handle superblock details.
479  * We want to be able to handle multiple superblock formats
480  * so we have a common interface to them all, and an array of
481  * different handlers.
482  * We rely on user-space to write the initial superblock, and support
483  * reading and updating of superblocks.
484  * Interface methods are:
485  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
486  *      loads and validates a superblock on dev.
487  *      if refdev != NULL, compare superblocks on both devices
488  *    Return:
489  *      0 - dev has a superblock that is compatible with refdev
490  *      1 - dev has a superblock that is compatible and newer than refdev
491  *          so dev should be used as the refdev in future
492  *     -EINVAL superblock incompatible or invalid
493  *     -othererror e.g. -EIO
494  *
495  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
496  *      Verify that dev is acceptable into mddev.
497  *       The first time, mddev->raid_disks will be 0, and data from
498  *       dev should be merged in.  Subsequent calls check that dev
499  *       is new enough.  Return 0 or -EINVAL
500  *
501  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
502  *     Update the superblock for rdev with data in mddev
503  *     This does not write to disc.
504  *
505  */
506
507 struct super_type  {
508         char            *name;
509         struct module   *owner;
510         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
511         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
512         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
513 };
514
515 /*
516  * load_super for 0.90.0 
517  */
518 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
519 {
520         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
521         mdp_super_t *sb;
522         int ret;
523         sector_t sb_offset;
524
525         /*
526          * Calculate the position of the superblock,
527          * it's at the end of the disk.
528          *
529          * It also happens to be a multiple of 4Kb.
530          */
531         sb_offset = calc_dev_sboffset(rdev->bdev);
532         rdev->sb_offset = sb_offset;
533
534         ret = read_disk_sb(rdev);
535         if (ret) return ret;
536
537         ret = -EINVAL;
538
539         bdevname(rdev->bdev, b);
540         sb = (mdp_super_t*)page_address(rdev->sb_page);
541
542         if (sb->md_magic != MD_SB_MAGIC) {
543                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
544                        b);
545                 goto abort;
546         }
547
548         if (sb->major_version != 0 ||
549             sb->minor_version != 90) {
550                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
551                         sb->major_version, sb->minor_version,
552                         b);
553                 goto abort;
554         }
555
556         if (sb->raid_disks <= 0)
557                 goto abort;
558
559         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
560                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
561                         b);
562                 goto abort;
563         }
564
565         rdev->preferred_minor = sb->md_minor;
566         rdev->data_offset = 0;
567
568         if (sb->level == LEVEL_MULTIPATH)
569                 rdev->desc_nr = -1;
570         else
571                 rdev->desc_nr = sb->this_disk.number;
572
573         if (refdev == 0)
574                 ret = 1;
575         else {
576                 __u64 ev1, ev2;
577                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
578                 if (!uuid_equal(refsb, sb)) {
579                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
580                                 b, bdevname(refdev->bdev,b2));
581                         goto abort;
582                 }
583                 if (!sb_equal(refsb, sb)) {
584                         printk(KERN_WARNING "md: %s has same UUID"
585                                " but different superblock to %s\n",
586                                b, bdevname(refdev->bdev, b2));
587                         goto abort;
588                 }
589                 ev1 = md_event(sb);
590                 ev2 = md_event(refsb);
591                 if (ev1 > ev2)
592                         ret = 1;
593                 else 
594                         ret = 0;
595         }
596         rdev->size = calc_dev_size(rdev, sb->chunk_size);
597
598  abort:
599         return ret;
600 }
601
602 /*
603  * validate_super for 0.90.0
604  */
605 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
606 {
607         mdp_disk_t *desc;
608         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
609
610         rdev->raid_disk = -1;
611         rdev->in_sync = 0;
612         if (mddev->raid_disks == 0) {
613                 mddev->major_version = 0;
614                 mddev->minor_version = sb->minor_version;
615                 mddev->patch_version = sb->patch_version;
616                 mddev->persistent = ! sb->not_persistent;
617                 mddev->chunk_size = sb->chunk_size;
618                 mddev->ctime = sb->ctime;
619                 mddev->utime = sb->utime;
620                 mddev->level = sb->level;
621                 mddev->layout = sb->layout;
622                 mddev->raid_disks = sb->raid_disks;
623                 mddev->size = sb->size;
624                 mddev->events = md_event(sb);
625                 mddev->bitmap_offset = 0;
626
627                 if (sb->state & (1<<MD_SB_CLEAN))
628                         mddev->recovery_cp = MaxSector;
629                 else {
630                         if (sb->events_hi == sb->cp_events_hi && 
631                                 sb->events_lo == sb->cp_events_lo) {
632                                 mddev->recovery_cp = sb->recovery_cp;
633                         } else
634                                 mddev->recovery_cp = 0;
635                 }
636
637                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
638                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
639                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
640                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
641
642                 mddev->max_disks = MD_SB_DISKS;
643
644                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
645                     mddev->bitmap_file == NULL) {
646                         if (mddev->level != 1) {
647                                 /* FIXME use a better test */
648                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
649                                 return -EINVAL;
650                         }
651                         mddev->bitmap_offset = (MD_SB_BYTES >> 9);
652                 }
653
654         } else if (mddev->pers == NULL) {
655                 /* Insist on good event counter while assembling */
656                 __u64 ev1 = md_event(sb);
657                 ++ev1;
658                 if (ev1 < mddev->events) 
659                         return -EINVAL;
660         } else if (mddev->bitmap) {
661                 /* if adding to array with a bitmap, then we can accept an
662                  * older device ... but not too old.
663                  */
664                 __u64 ev1 = md_event(sb);
665                 if (ev1 < mddev->bitmap->events_cleared)
666                         return 0;
667         } else /* just a hot-add of a new device, leave raid_disk at -1 */
668                 return 0;
669
670         if (mddev->level != LEVEL_MULTIPATH) {
671                 rdev->faulty = 0;
672                 desc = sb->disks + rdev->desc_nr;
673
674                 if (desc->state & (1<<MD_DISK_FAULTY))
675                         rdev->faulty = 1;
676                 else if (desc->state & (1<<MD_DISK_SYNC) &&
677                          desc->raid_disk < mddev->raid_disks) {
678                         rdev->in_sync = 1;
679                         rdev->raid_disk = desc->raid_disk;
680                 }
681         } else /* MULTIPATH are always insync */
682                 rdev->in_sync = 1;
683         return 0;
684 }
685
686 /*
687  * sync_super for 0.90.0
688  */
689 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
690 {
691         mdp_super_t *sb;
692         struct list_head *tmp;
693         mdk_rdev_t *rdev2;
694         int next_spare = mddev->raid_disks;
695
696         /* make rdev->sb match mddev data..
697          *
698          * 1/ zero out disks
699          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
700          * 3/ any empty disks < next_spare become removed
701          *
702          * disks[0] gets initialised to REMOVED because
703          * we cannot be sure from other fields if it has
704          * been initialised or not.
705          */
706         int i;
707         int active=0, working=0,failed=0,spare=0,nr_disks=0;
708
709         sb = (mdp_super_t*)page_address(rdev->sb_page);
710
711         memset(sb, 0, sizeof(*sb));
712
713         sb->md_magic = MD_SB_MAGIC;
714         sb->major_version = mddev->major_version;
715         sb->minor_version = mddev->minor_version;
716         sb->patch_version = mddev->patch_version;
717         sb->gvalid_words  = 0; /* ignored */
718         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
719         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
720         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
721         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
722
723         sb->ctime = mddev->ctime;
724         sb->level = mddev->level;
725         sb->size  = mddev->size;
726         sb->raid_disks = mddev->raid_disks;
727         sb->md_minor = mddev->md_minor;
728         sb->not_persistent = !mddev->persistent;
729         sb->utime = mddev->utime;
730         sb->state = 0;
731         sb->events_hi = (mddev->events>>32);
732         sb->events_lo = (u32)mddev->events;
733
734         if (mddev->in_sync)
735         {
736                 sb->recovery_cp = mddev->recovery_cp;
737                 sb->cp_events_hi = (mddev->events>>32);
738                 sb->cp_events_lo = (u32)mddev->events;
739                 if (mddev->recovery_cp == MaxSector)
740                         sb->state = (1<< MD_SB_CLEAN);
741         } else
742                 sb->recovery_cp = 0;
743
744         sb->layout = mddev->layout;
745         sb->chunk_size = mddev->chunk_size;
746
747         if (mddev->bitmap && mddev->bitmap_file == NULL)
748                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
749
750         sb->disks[0].state = (1<<MD_DISK_REMOVED);
751         ITERATE_RDEV(mddev,rdev2,tmp) {
752                 mdp_disk_t *d;
753                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
754                         rdev2->desc_nr = rdev2->raid_disk;
755                 else
756                         rdev2->desc_nr = next_spare++;
757                 d = &sb->disks[rdev2->desc_nr];
758                 nr_disks++;
759                 d->number = rdev2->desc_nr;
760                 d->major = MAJOR(rdev2->bdev->bd_dev);
761                 d->minor = MINOR(rdev2->bdev->bd_dev);
762                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
763                         d->raid_disk = rdev2->raid_disk;
764                 else
765                         d->raid_disk = rdev2->desc_nr; /* compatibility */
766                 if (rdev2->faulty) {
767                         d->state = (1<<MD_DISK_FAULTY);
768                         failed++;
769                 } else if (rdev2->in_sync) {
770                         d->state = (1<<MD_DISK_ACTIVE);
771                         d->state |= (1<<MD_DISK_SYNC);
772                         active++;
773                         working++;
774                 } else {
775                         d->state = 0;
776                         spare++;
777                         working++;
778                 }
779         }
780         
781         /* now set the "removed" and "faulty" bits on any missing devices */
782         for (i=0 ; i < mddev->raid_disks ; i++) {
783                 mdp_disk_t *d = &sb->disks[i];
784                 if (d->state == 0 && d->number == 0) {
785                         d->number = i;
786                         d->raid_disk = i;
787                         d->state = (1<<MD_DISK_REMOVED);
788                         d->state |= (1<<MD_DISK_FAULTY);
789                         failed++;
790                 }
791         }
792         sb->nr_disks = nr_disks;
793         sb->active_disks = active;
794         sb->working_disks = working;
795         sb->failed_disks = failed;
796         sb->spare_disks = spare;
797
798         sb->this_disk = sb->disks[rdev->desc_nr];
799         sb->sb_csum = calc_sb_csum(sb);
800 }
801
802 /*
803  * version 1 superblock
804  */
805
806 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
807 {
808         unsigned int disk_csum, csum;
809         unsigned long long newcsum;
810         int size = 256 + le32_to_cpu(sb->max_dev)*2;
811         unsigned int *isuper = (unsigned int*)sb;
812         int i;
813
814         disk_csum = sb->sb_csum;
815         sb->sb_csum = 0;
816         newcsum = 0;
817         for (i=0; size>=4; size -= 4 )
818                 newcsum += le32_to_cpu(*isuper++);
819
820         if (size == 2)
821                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
822
823         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
824         sb->sb_csum = disk_csum;
825         return cpu_to_le32(csum);
826 }
827
828 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
829 {
830         struct mdp_superblock_1 *sb;
831         int ret;
832         sector_t sb_offset;
833         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
834
835         /*
836          * Calculate the position of the superblock.
837          * It is always aligned to a 4K boundary and
838          * depeding on minor_version, it can be:
839          * 0: At least 8K, but less than 12K, from end of device
840          * 1: At start of device
841          * 2: 4K from start of device.
842          */
843         switch(minor_version) {
844         case 0:
845                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
846                 sb_offset -= 8*2;
847                 sb_offset &= ~(sector_t)(4*2-1);
848                 /* convert from sectors to K */
849                 sb_offset /= 2;
850                 break;
851         case 1:
852                 sb_offset = 0;
853                 break;
854         case 2:
855                 sb_offset = 4;
856                 break;
857         default:
858                 return -EINVAL;
859         }
860         rdev->sb_offset = sb_offset;
861
862         ret = read_disk_sb(rdev);
863         if (ret) return ret;
864
865
866         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
867
868         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
869             sb->major_version != cpu_to_le32(1) ||
870             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
871             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
872             sb->feature_map != 0)
873                 return -EINVAL;
874
875         if (calc_sb_1_csum(sb) != sb->sb_csum) {
876                 printk("md: invalid superblock checksum on %s\n",
877                         bdevname(rdev->bdev,b));
878                 return -EINVAL;
879         }
880         if (le64_to_cpu(sb->data_size) < 10) {
881                 printk("md: data_size too small on %s\n",
882                        bdevname(rdev->bdev,b));
883                 return -EINVAL;
884         }
885         rdev->preferred_minor = 0xffff;
886         rdev->data_offset = le64_to_cpu(sb->data_offset);
887
888         if (refdev == 0)
889                 return 1;
890         else {
891                 __u64 ev1, ev2;
892                 struct mdp_superblock_1 *refsb = 
893                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
894
895                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
896                     sb->level != refsb->level ||
897                     sb->layout != refsb->layout ||
898                     sb->chunksize != refsb->chunksize) {
899                         printk(KERN_WARNING "md: %s has strangely different"
900                                 " superblock to %s\n",
901                                 bdevname(rdev->bdev,b),
902                                 bdevname(refdev->bdev,b2));
903                         return -EINVAL;
904                 }
905                 ev1 = le64_to_cpu(sb->events);
906                 ev2 = le64_to_cpu(refsb->events);
907
908                 if (ev1 > ev2)
909                         return 1;
910         }
911         if (minor_version) 
912                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
913         else
914                 rdev->size = rdev->sb_offset;
915         if (rdev->size < le64_to_cpu(sb->data_size)/2)
916                 return -EINVAL;
917         rdev->size = le64_to_cpu(sb->data_size)/2;
918         if (le32_to_cpu(sb->chunksize))
919                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
920         return 0;
921 }
922
923 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
924 {
925         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
926
927         rdev->raid_disk = -1;
928         rdev->in_sync = 0;
929         if (mddev->raid_disks == 0) {
930                 mddev->major_version = 1;
931                 mddev->patch_version = 0;
932                 mddev->persistent = 1;
933                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
934                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
935                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
936                 mddev->level = le32_to_cpu(sb->level);
937                 mddev->layout = le32_to_cpu(sb->layout);
938                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
939                 mddev->size = le64_to_cpu(sb->size)/2;
940                 mddev->events = le64_to_cpu(sb->events);
941                 mddev->bitmap_offset = 0;
942                 
943                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
944                 memcpy(mddev->uuid, sb->set_uuid, 16);
945
946                 mddev->max_disks =  (4096-256)/2;
947
948                 if ((le32_to_cpu(sb->feature_map) & 1) &&
949                     mddev->bitmap_file == NULL ) {
950                         if (mddev->level != 1) {
951                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
952                                 return -EINVAL;
953                         }
954                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
955                 }
956         } else if (mddev->pers == NULL) {
957                 /* Insist of good event counter while assembling */
958                 __u64 ev1 = le64_to_cpu(sb->events);
959                 ++ev1;
960                 if (ev1 < mddev->events)
961                         return -EINVAL;
962         } else if (mddev->bitmap) {
963                 /* If adding to array with a bitmap, then we can accept an
964                  * older device, but not too old.
965                  */
966                 __u64 ev1 = le64_to_cpu(sb->events);
967                 if (ev1 < mddev->bitmap->events_cleared)
968                         return 0;
969         } else /* just a hot-add of a new device, leave raid_disk at -1 */
970                 return 0;
971
972         if (mddev->level != LEVEL_MULTIPATH) {
973                 int role;
974                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
975                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
976                 switch(role) {
977                 case 0xffff: /* spare */
978                         rdev->faulty = 0;
979                         break;
980                 case 0xfffe: /* faulty */
981                         rdev->faulty = 1;
982                         break;
983                 default:
984                         rdev->in_sync = 1;
985                         rdev->faulty = 0;
986                         rdev->raid_disk = role;
987                         break;
988                 }
989         } else /* MULTIPATH are always insync */
990                 rdev->in_sync = 1;
991
992         return 0;
993 }
994
995 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
996 {
997         struct mdp_superblock_1 *sb;
998         struct list_head *tmp;
999         mdk_rdev_t *rdev2;
1000         int max_dev, i;
1001         /* make rdev->sb match mddev and rdev data. */
1002
1003         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1004
1005         sb->feature_map = 0;
1006         sb->pad0 = 0;
1007         memset(sb->pad1, 0, sizeof(sb->pad1));
1008         memset(sb->pad2, 0, sizeof(sb->pad2));
1009         memset(sb->pad3, 0, sizeof(sb->pad3));
1010
1011         sb->utime = cpu_to_le64((__u64)mddev->utime);
1012         sb->events = cpu_to_le64(mddev->events);
1013         if (mddev->in_sync)
1014                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1015         else
1016                 sb->resync_offset = cpu_to_le64(0);
1017
1018         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1019                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1020                 sb->feature_map = cpu_to_le32(1);
1021         }
1022
1023         max_dev = 0;
1024         ITERATE_RDEV(mddev,rdev2,tmp)
1025                 if (rdev2->desc_nr+1 > max_dev)
1026                         max_dev = rdev2->desc_nr+1;
1027         
1028         sb->max_dev = cpu_to_le32(max_dev);
1029         for (i=0; i<max_dev;i++)
1030                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1031         
1032         ITERATE_RDEV(mddev,rdev2,tmp) {
1033                 i = rdev2->desc_nr;
1034                 if (rdev2->faulty)
1035                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1036                 else if (rdev2->in_sync)
1037                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1038                 else
1039                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1040         }
1041
1042         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1043         sb->sb_csum = calc_sb_1_csum(sb);
1044 }
1045
1046
1047 static struct super_type super_types[] = {
1048         [0] = {
1049                 .name   = "0.90.0",
1050                 .owner  = THIS_MODULE,
1051                 .load_super     = super_90_load,
1052                 .validate_super = super_90_validate,
1053                 .sync_super     = super_90_sync,
1054         },
1055         [1] = {
1056                 .name   = "md-1",
1057                 .owner  = THIS_MODULE,
1058                 .load_super     = super_1_load,
1059                 .validate_super = super_1_validate,
1060                 .sync_super     = super_1_sync,
1061         },
1062 };
1063         
1064 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1065 {
1066         struct list_head *tmp;
1067         mdk_rdev_t *rdev;
1068
1069         ITERATE_RDEV(mddev,rdev,tmp)
1070                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1071                         return rdev;
1072
1073         return NULL;
1074 }
1075
1076 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1077 {
1078         struct list_head *tmp;
1079         mdk_rdev_t *rdev;
1080
1081         ITERATE_RDEV(mddev1,rdev,tmp)
1082                 if (match_dev_unit(mddev2, rdev))
1083                         return 1;
1084
1085         return 0;
1086 }
1087
1088 static LIST_HEAD(pending_raid_disks);
1089
1090 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1091 {
1092         mdk_rdev_t *same_pdev;
1093         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1094
1095         if (rdev->mddev) {
1096                 MD_BUG();
1097                 return -EINVAL;
1098         }
1099         same_pdev = match_dev_unit(mddev, rdev);
1100         if (same_pdev)
1101                 printk(KERN_WARNING
1102                         "%s: WARNING: %s appears to be on the same physical"
1103                         " disk as %s. True\n     protection against single-disk"
1104                         " failure might be compromised.\n",
1105                         mdname(mddev), bdevname(rdev->bdev,b),
1106                         bdevname(same_pdev->bdev,b2));
1107
1108         /* Verify rdev->desc_nr is unique.
1109          * If it is -1, assign a free number, else
1110          * check number is not in use
1111          */
1112         if (rdev->desc_nr < 0) {
1113                 int choice = 0;
1114                 if (mddev->pers) choice = mddev->raid_disks;
1115                 while (find_rdev_nr(mddev, choice))
1116                         choice++;
1117                 rdev->desc_nr = choice;
1118         } else {
1119                 if (find_rdev_nr(mddev, rdev->desc_nr))
1120                         return -EBUSY;
1121         }
1122                         
1123         list_add(&rdev->same_set, &mddev->disks);
1124         rdev->mddev = mddev;
1125         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1126         return 0;
1127 }
1128
1129 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1130 {
1131         char b[BDEVNAME_SIZE];
1132         if (!rdev->mddev) {
1133                 MD_BUG();
1134                 return;
1135         }
1136         list_del_init(&rdev->same_set);
1137         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1138         rdev->mddev = NULL;
1139 }
1140
1141 /*
1142  * prevent the device from being mounted, repartitioned or
1143  * otherwise reused by a RAID array (or any other kernel
1144  * subsystem), by bd_claiming the device.
1145  */
1146 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1147 {
1148         int err = 0;
1149         struct block_device *bdev;
1150         char b[BDEVNAME_SIZE];
1151
1152         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1153         if (IS_ERR(bdev)) {
1154                 printk(KERN_ERR "md: could not open %s.\n",
1155                         __bdevname(dev, b));
1156                 return PTR_ERR(bdev);
1157         }
1158         err = bd_claim(bdev, rdev);
1159         if (err) {
1160                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1161                         bdevname(bdev, b));
1162                 blkdev_put(bdev);
1163                 return err;
1164         }
1165         rdev->bdev = bdev;
1166         return err;
1167 }
1168
1169 static void unlock_rdev(mdk_rdev_t *rdev)
1170 {
1171         struct block_device *bdev = rdev->bdev;
1172         rdev->bdev = NULL;
1173         if (!bdev)
1174                 MD_BUG();
1175         bd_release(bdev);
1176         blkdev_put(bdev);
1177 }
1178
1179 void md_autodetect_dev(dev_t dev);
1180
1181 static void export_rdev(mdk_rdev_t * rdev)
1182 {
1183         char b[BDEVNAME_SIZE];
1184         printk(KERN_INFO "md: export_rdev(%s)\n",
1185                 bdevname(rdev->bdev,b));
1186         if (rdev->mddev)
1187                 MD_BUG();
1188         free_disk_sb(rdev);
1189         list_del_init(&rdev->same_set);
1190 #ifndef MODULE
1191         md_autodetect_dev(rdev->bdev->bd_dev);
1192 #endif
1193         unlock_rdev(rdev);
1194         kfree(rdev);
1195 }
1196
1197 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1198 {
1199         unbind_rdev_from_array(rdev);
1200         export_rdev(rdev);
1201 }
1202
1203 static void export_array(mddev_t *mddev)
1204 {
1205         struct list_head *tmp;
1206         mdk_rdev_t *rdev;
1207
1208         ITERATE_RDEV(mddev,rdev,tmp) {
1209                 if (!rdev->mddev) {
1210                         MD_BUG();
1211                         continue;
1212                 }
1213                 kick_rdev_from_array(rdev);
1214         }
1215         if (!list_empty(&mddev->disks))
1216                 MD_BUG();
1217         mddev->raid_disks = 0;
1218         mddev->major_version = 0;
1219 }
1220
1221 static void print_desc(mdp_disk_t *desc)
1222 {
1223         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1224                 desc->major,desc->minor,desc->raid_disk,desc->state);
1225 }
1226
1227 static void print_sb(mdp_super_t *sb)
1228 {
1229         int i;
1230
1231         printk(KERN_INFO 
1232                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1233                 sb->major_version, sb->minor_version, sb->patch_version,
1234                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1235                 sb->ctime);
1236         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1237                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1238                 sb->md_minor, sb->layout, sb->chunk_size);
1239         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1240                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1241                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1242                 sb->failed_disks, sb->spare_disks,
1243                 sb->sb_csum, (unsigned long)sb->events_lo);
1244
1245         printk(KERN_INFO);
1246         for (i = 0; i < MD_SB_DISKS; i++) {
1247                 mdp_disk_t *desc;
1248
1249                 desc = sb->disks + i;
1250                 if (desc->number || desc->major || desc->minor ||
1251                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1252                         printk("     D %2d: ", i);
1253                         print_desc(desc);
1254                 }
1255         }
1256         printk(KERN_INFO "md:     THIS: ");
1257         print_desc(&sb->this_disk);
1258
1259 }
1260
1261 static void print_rdev(mdk_rdev_t *rdev)
1262 {
1263         char b[BDEVNAME_SIZE];
1264         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1265                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1266                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1267         if (rdev->sb_loaded) {
1268                 printk(KERN_INFO "md: rdev superblock:\n");
1269                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1270         } else
1271                 printk(KERN_INFO "md: no rdev superblock!\n");
1272 }
1273
1274 void md_print_devices(void)
1275 {
1276         struct list_head *tmp, *tmp2;
1277         mdk_rdev_t *rdev;
1278         mddev_t *mddev;
1279         char b[BDEVNAME_SIZE];
1280
1281         printk("\n");
1282         printk("md:     **********************************\n");
1283         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1284         printk("md:     **********************************\n");
1285         ITERATE_MDDEV(mddev,tmp) {
1286
1287                 if (mddev->bitmap)
1288                         bitmap_print_sb(mddev->bitmap);
1289                 else
1290                         printk("%s: ", mdname(mddev));
1291                 ITERATE_RDEV(mddev,rdev,tmp2)
1292                         printk("<%s>", bdevname(rdev->bdev,b));
1293                 printk("\n");
1294
1295                 ITERATE_RDEV(mddev,rdev,tmp2)
1296                         print_rdev(rdev);
1297         }
1298         printk("md:     **********************************\n");
1299         printk("\n");
1300 }
1301
1302
1303 static void sync_sbs(mddev_t * mddev)
1304 {
1305         mdk_rdev_t *rdev;
1306         struct list_head *tmp;
1307
1308         ITERATE_RDEV(mddev,rdev,tmp) {
1309                 super_types[mddev->major_version].
1310                         sync_super(mddev, rdev);
1311                 rdev->sb_loaded = 1;
1312         }
1313 }
1314
1315 static void md_update_sb(mddev_t * mddev)
1316 {
1317         int err;
1318         struct list_head *tmp;
1319         mdk_rdev_t *rdev;
1320         int sync_req;
1321
1322 repeat:
1323         spin_lock(&mddev->write_lock);
1324         sync_req = mddev->in_sync;
1325         mddev->utime = get_seconds();
1326         mddev->events ++;
1327
1328         if (!mddev->events) {
1329                 /*
1330                  * oops, this 64-bit counter should never wrap.
1331                  * Either we are in around ~1 trillion A.C., assuming
1332                  * 1 reboot per second, or we have a bug:
1333                  */
1334                 MD_BUG();
1335                 mddev->events --;
1336         }
1337         mddev->sb_dirty = 2;
1338         sync_sbs(mddev);
1339
1340         /*
1341          * do not write anything to disk if using
1342          * nonpersistent superblocks
1343          */
1344         if (!mddev->persistent) {
1345                 mddev->sb_dirty = 0;
1346                 spin_unlock(&mddev->write_lock);
1347                 wake_up(&mddev->sb_wait);
1348                 return;
1349         }
1350         spin_unlock(&mddev->write_lock);
1351
1352         dprintk(KERN_INFO 
1353                 "md: updating %s RAID superblock on device (in sync %d)\n",
1354                 mdname(mddev),mddev->in_sync);
1355
1356         err = bitmap_update_sb(mddev->bitmap);
1357         ITERATE_RDEV(mddev,rdev,tmp) {
1358                 char b[BDEVNAME_SIZE];
1359                 dprintk(KERN_INFO "md: ");
1360                 if (rdev->faulty)
1361                         dprintk("(skipping faulty ");
1362
1363                 dprintk("%s ", bdevname(rdev->bdev,b));
1364                 if (!rdev->faulty) {
1365                         md_super_write(mddev,rdev,
1366                                        rdev->sb_offset<<1, MD_SB_BYTES,
1367                                        rdev->sb_page);
1368                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1369                                 bdevname(rdev->bdev,b),
1370                                 (unsigned long long)rdev->sb_offset);
1371
1372                 } else
1373                         dprintk(")\n");
1374                 if (mddev->level == LEVEL_MULTIPATH)
1375                         /* only need to write one superblock... */
1376                         break;
1377         }
1378         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1379         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1380
1381         spin_lock(&mddev->write_lock);
1382         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1383                 /* have to write it out again */
1384                 spin_unlock(&mddev->write_lock);
1385                 goto repeat;
1386         }
1387         mddev->sb_dirty = 0;
1388         spin_unlock(&mddev->write_lock);
1389         wake_up(&mddev->sb_wait);
1390
1391 }
1392
1393 /*
1394  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1395  *
1396  * mark the device faulty if:
1397  *
1398  *   - the device is nonexistent (zero size)
1399  *   - the device has no valid superblock
1400  *
1401  * a faulty rdev _never_ has rdev->sb set.
1402  */
1403 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1404 {
1405         char b[BDEVNAME_SIZE];
1406         int err;
1407         mdk_rdev_t *rdev;
1408         sector_t size;
1409
1410         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1411         if (!rdev) {
1412                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1413                 return ERR_PTR(-ENOMEM);
1414         }
1415         memset(rdev, 0, sizeof(*rdev));
1416
1417         if ((err = alloc_disk_sb(rdev)))
1418                 goto abort_free;
1419
1420         err = lock_rdev(rdev, newdev);
1421         if (err)
1422                 goto abort_free;
1423
1424         rdev->desc_nr = -1;
1425         rdev->faulty = 0;
1426         rdev->in_sync = 0;
1427         rdev->data_offset = 0;
1428         atomic_set(&rdev->nr_pending, 0);
1429
1430         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1431         if (!size) {
1432                 printk(KERN_WARNING 
1433                         "md: %s has zero or unknown size, marking faulty!\n",
1434                         bdevname(rdev->bdev,b));
1435                 err = -EINVAL;
1436                 goto abort_free;
1437         }
1438
1439         if (super_format >= 0) {
1440                 err = super_types[super_format].
1441                         load_super(rdev, NULL, super_minor);
1442                 if (err == -EINVAL) {
1443                         printk(KERN_WARNING 
1444                                 "md: %s has invalid sb, not importing!\n",
1445                                 bdevname(rdev->bdev,b));
1446                         goto abort_free;
1447                 }
1448                 if (err < 0) {
1449                         printk(KERN_WARNING 
1450                                 "md: could not read %s's sb, not importing!\n",
1451                                 bdevname(rdev->bdev,b));
1452                         goto abort_free;
1453                 }
1454         }
1455         INIT_LIST_HEAD(&rdev->same_set);
1456
1457         return rdev;
1458
1459 abort_free:
1460         if (rdev->sb_page) {
1461                 if (rdev->bdev)
1462                         unlock_rdev(rdev);
1463                 free_disk_sb(rdev);
1464         }
1465         kfree(rdev);
1466         return ERR_PTR(err);
1467 }
1468
1469 /*
1470  * Check a full RAID array for plausibility
1471  */
1472
1473
1474 static void analyze_sbs(mddev_t * mddev)
1475 {
1476         int i;
1477         struct list_head *tmp;
1478         mdk_rdev_t *rdev, *freshest;
1479         char b[BDEVNAME_SIZE];
1480
1481         freshest = NULL;
1482         ITERATE_RDEV(mddev,rdev,tmp)
1483                 switch (super_types[mddev->major_version].
1484                         load_super(rdev, freshest, mddev->minor_version)) {
1485                 case 1:
1486                         freshest = rdev;
1487                         break;
1488                 case 0:
1489                         break;
1490                 default:
1491                         printk( KERN_ERR \
1492                                 "md: fatal superblock inconsistency in %s"
1493                                 " -- removing from array\n", 
1494                                 bdevname(rdev->bdev,b));
1495                         kick_rdev_from_array(rdev);
1496                 }
1497
1498
1499         super_types[mddev->major_version].
1500                 validate_super(mddev, freshest);
1501
1502         i = 0;
1503         ITERATE_RDEV(mddev,rdev,tmp) {
1504                 if (rdev != freshest)
1505                         if (super_types[mddev->major_version].
1506                             validate_super(mddev, rdev)) {
1507                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1508                                         " from array!\n",
1509                                         bdevname(rdev->bdev,b));
1510                                 kick_rdev_from_array(rdev);
1511                                 continue;
1512                         }
1513                 if (mddev->level == LEVEL_MULTIPATH) {
1514                         rdev->desc_nr = i++;
1515                         rdev->raid_disk = rdev->desc_nr;
1516                         rdev->in_sync = 1;
1517                 }
1518         }
1519
1520
1521
1522         if (mddev->recovery_cp != MaxSector &&
1523             mddev->level >= 1)
1524                 printk(KERN_ERR "md: %s: raid array is not clean"
1525                        " -- starting background reconstruction\n",
1526                        mdname(mddev));
1527
1528 }
1529
1530 int mdp_major = 0;
1531
1532 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1533 {
1534         static DECLARE_MUTEX(disks_sem);
1535         mddev_t *mddev = mddev_find(dev);
1536         struct gendisk *disk;
1537         int partitioned = (MAJOR(dev) != MD_MAJOR);
1538         int shift = partitioned ? MdpMinorShift : 0;
1539         int unit = MINOR(dev) >> shift;
1540
1541         if (!mddev)
1542                 return NULL;
1543
1544         down(&disks_sem);
1545         if (mddev->gendisk) {
1546                 up(&disks_sem);
1547                 mddev_put(mddev);
1548                 return NULL;
1549         }
1550         disk = alloc_disk(1 << shift);
1551         if (!disk) {
1552                 up(&disks_sem);
1553                 mddev_put(mddev);
1554                 return NULL;
1555         }
1556         disk->major = MAJOR(dev);
1557         disk->first_minor = unit << shift;
1558         if (partitioned) {
1559                 sprintf(disk->disk_name, "md_d%d", unit);
1560                 sprintf(disk->devfs_name, "md/d%d", unit);
1561         } else {
1562                 sprintf(disk->disk_name, "md%d", unit);
1563                 sprintf(disk->devfs_name, "md/%d", unit);
1564         }
1565         disk->fops = &md_fops;
1566         disk->private_data = mddev;
1567         disk->queue = mddev->queue;
1568         add_disk(disk);
1569         mddev->gendisk = disk;
1570         up(&disks_sem);
1571         return NULL;
1572 }
1573
1574 void md_wakeup_thread(mdk_thread_t *thread);
1575
1576 static void md_safemode_timeout(unsigned long data)
1577 {
1578         mddev_t *mddev = (mddev_t *) data;
1579
1580         mddev->safemode = 1;
1581         md_wakeup_thread(mddev->thread);
1582 }
1583
1584
1585 static int do_md_run(mddev_t * mddev)
1586 {
1587         int pnum, err;
1588         int chunk_size;
1589         struct list_head *tmp;
1590         mdk_rdev_t *rdev;
1591         struct gendisk *disk;
1592         char b[BDEVNAME_SIZE];
1593
1594         if (list_empty(&mddev->disks))
1595                 /* cannot run an array with no devices.. */
1596                 return -EINVAL;
1597
1598         if (mddev->pers)
1599                 return -EBUSY;
1600
1601         /*
1602          * Analyze all RAID superblock(s)
1603          */
1604         if (!mddev->raid_disks)
1605                 analyze_sbs(mddev);
1606
1607         chunk_size = mddev->chunk_size;
1608         pnum = level_to_pers(mddev->level);
1609
1610         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1611                 if (!chunk_size) {
1612                         /*
1613                          * 'default chunksize' in the old md code used to
1614                          * be PAGE_SIZE, baaad.
1615                          * we abort here to be on the safe side. We don't
1616                          * want to continue the bad practice.
1617                          */
1618                         printk(KERN_ERR 
1619                                 "no chunksize specified, see 'man raidtab'\n");
1620                         return -EINVAL;
1621                 }
1622                 if (chunk_size > MAX_CHUNK_SIZE) {
1623                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1624                                 chunk_size, MAX_CHUNK_SIZE);
1625                         return -EINVAL;
1626                 }
1627                 /*
1628                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1629                  */
1630                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1631                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1632                         return -EINVAL;
1633                 }
1634                 if (chunk_size < PAGE_SIZE) {
1635                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1636                                 chunk_size, PAGE_SIZE);
1637                         return -EINVAL;
1638                 }
1639
1640                 /* devices must have minimum size of one chunk */
1641                 ITERATE_RDEV(mddev,rdev,tmp) {
1642                         if (rdev->faulty)
1643                                 continue;
1644                         if (rdev->size < chunk_size / 1024) {
1645                                 printk(KERN_WARNING
1646                                         "md: Dev %s smaller than chunk_size:"
1647                                         " %lluk < %dk\n",
1648                                         bdevname(rdev->bdev,b),
1649                                         (unsigned long long)rdev->size,
1650                                         chunk_size / 1024);
1651                                 return -EINVAL;
1652                         }
1653                 }
1654         }
1655
1656 #ifdef CONFIG_KMOD
1657         if (!pers[pnum])
1658         {
1659                 request_module("md-personality-%d", pnum);
1660         }
1661 #endif
1662
1663         /*
1664          * Drop all container device buffers, from now on
1665          * the only valid external interface is through the md
1666          * device.
1667          * Also find largest hardsector size
1668          */
1669         ITERATE_RDEV(mddev,rdev,tmp) {
1670                 if (rdev->faulty)
1671                         continue;
1672                 sync_blockdev(rdev->bdev);
1673                 invalidate_bdev(rdev->bdev, 0);
1674         }
1675
1676         md_probe(mddev->unit, NULL, NULL);
1677         disk = mddev->gendisk;
1678         if (!disk)
1679                 return -ENOMEM;
1680
1681         spin_lock(&pers_lock);
1682         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1683                 spin_unlock(&pers_lock);
1684                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1685                        pnum);
1686                 return -EINVAL;
1687         }
1688
1689         mddev->pers = pers[pnum];
1690         spin_unlock(&pers_lock);
1691
1692         mddev->recovery = 0;
1693         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1694
1695         /* before we start the array running, initialise the bitmap */
1696         err = bitmap_create(mddev);
1697         if (err)
1698                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1699                         mdname(mddev), err);
1700         else
1701                 err = mddev->pers->run(mddev);
1702         if (err) {
1703                 printk(KERN_ERR "md: pers->run() failed ...\n");
1704                 module_put(mddev->pers->owner);
1705                 mddev->pers = NULL;
1706                 bitmap_destroy(mddev);
1707                 return err;
1708         }
1709         atomic_set(&mddev->writes_pending,0);
1710         mddev->safemode = 0;
1711         mddev->safemode_timer.function = md_safemode_timeout;
1712         mddev->safemode_timer.data = (unsigned long) mddev;
1713         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1714         mddev->in_sync = 1;
1715         
1716         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1717         md_wakeup_thread(mddev->thread);
1718         
1719         if (mddev->sb_dirty)
1720                 md_update_sb(mddev);
1721
1722         set_capacity(disk, mddev->array_size<<1);
1723
1724         /* If we call blk_queue_make_request here, it will
1725          * re-initialise max_sectors etc which may have been
1726          * refined inside -> run.  So just set the bits we need to set.
1727          * Most initialisation happended when we called
1728          * blk_queue_make_request(..., md_fail_request)
1729          * earlier.
1730          */
1731         mddev->queue->queuedata = mddev;
1732         mddev->queue->make_request_fn = mddev->pers->make_request;
1733
1734         mddev->changed = 1;
1735         return 0;
1736 }
1737
1738 static int restart_array(mddev_t *mddev)
1739 {
1740         struct gendisk *disk = mddev->gendisk;
1741         int err;
1742
1743         /*
1744          * Complain if it has no devices
1745          */
1746         err = -ENXIO;
1747         if (list_empty(&mddev->disks))
1748                 goto out;
1749
1750         if (mddev->pers) {
1751                 err = -EBUSY;
1752                 if (!mddev->ro)
1753                         goto out;
1754
1755                 mddev->safemode = 0;
1756                 mddev->ro = 0;
1757                 set_disk_ro(disk, 0);
1758
1759                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1760                         mdname(mddev));
1761                 /*
1762                  * Kick recovery or resync if necessary
1763                  */
1764                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1765                 md_wakeup_thread(mddev->thread);
1766                 err = 0;
1767         } else {
1768                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1769                         mdname(mddev));
1770                 err = -EINVAL;
1771         }
1772
1773 out:
1774         return err;
1775 }
1776
1777 static int do_md_stop(mddev_t * mddev, int ro)
1778 {
1779         int err = 0;
1780         struct gendisk *disk = mddev->gendisk;
1781
1782         if (mddev->pers) {
1783                 if (atomic_read(&mddev->active)>2) {
1784                         printk("md: %s still in use.\n",mdname(mddev));
1785                         return -EBUSY;
1786                 }
1787
1788                 if (mddev->sync_thread) {
1789                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1790                         md_unregister_thread(mddev->sync_thread);
1791                         mddev->sync_thread = NULL;
1792                 }
1793
1794                 del_timer_sync(&mddev->safemode_timer);
1795
1796                 invalidate_partition(disk, 0);
1797
1798                 if (ro) {
1799                         err  = -ENXIO;
1800                         if (mddev->ro)
1801                                 goto out;
1802                         mddev->ro = 1;
1803                 } else {
1804                         bitmap_flush(mddev);
1805                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1806                         if (mddev->ro)
1807                                 set_disk_ro(disk, 0);
1808                         blk_queue_make_request(mddev->queue, md_fail_request);
1809                         mddev->pers->stop(mddev);
1810                         module_put(mddev->pers->owner);
1811                         mddev->pers = NULL;
1812                         if (mddev->ro)
1813                                 mddev->ro = 0;
1814                 }
1815                 if (!mddev->in_sync) {
1816                         /* mark array as shutdown cleanly */
1817                         mddev->in_sync = 1;
1818                         md_update_sb(mddev);
1819                 }
1820                 if (ro)
1821                         set_disk_ro(disk, 1);
1822         }
1823
1824         bitmap_destroy(mddev);
1825         if (mddev->bitmap_file) {
1826                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1827                 fput(mddev->bitmap_file);
1828                 mddev->bitmap_file = NULL;
1829         }
1830         mddev->bitmap_offset = 0;
1831
1832         /*
1833          * Free resources if final stop
1834          */
1835         if (!ro) {
1836                 struct gendisk *disk;
1837                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1838
1839                 export_array(mddev);
1840
1841                 mddev->array_size = 0;
1842                 disk = mddev->gendisk;
1843                 if (disk)
1844                         set_capacity(disk, 0);
1845                 mddev->changed = 1;
1846         } else
1847                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1848                         mdname(mddev));
1849         err = 0;
1850 out:
1851         return err;
1852 }
1853
1854 static void autorun_array(mddev_t *mddev)
1855 {
1856         mdk_rdev_t *rdev;
1857         struct list_head *tmp;
1858         int err;
1859
1860         if (list_empty(&mddev->disks))
1861                 return;
1862
1863         printk(KERN_INFO "md: running: ");
1864
1865         ITERATE_RDEV(mddev,rdev,tmp) {
1866                 char b[BDEVNAME_SIZE];
1867                 printk("<%s>", bdevname(rdev->bdev,b));
1868         }
1869         printk("\n");
1870
1871         err = do_md_run (mddev);
1872         if (err) {
1873                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1874                 do_md_stop (mddev, 0);
1875         }
1876 }
1877
1878 /*
1879  * lets try to run arrays based on all disks that have arrived
1880  * until now. (those are in pending_raid_disks)
1881  *
1882  * the method: pick the first pending disk, collect all disks with
1883  * the same UUID, remove all from the pending list and put them into
1884  * the 'same_array' list. Then order this list based on superblock
1885  * update time (freshest comes first), kick out 'old' disks and
1886  * compare superblocks. If everything's fine then run it.
1887  *
1888  * If "unit" is allocated, then bump its reference count
1889  */
1890 static void autorun_devices(int part)
1891 {
1892         struct list_head candidates;
1893         struct list_head *tmp;
1894         mdk_rdev_t *rdev0, *rdev;
1895         mddev_t *mddev;
1896         char b[BDEVNAME_SIZE];
1897
1898         printk(KERN_INFO "md: autorun ...\n");
1899         while (!list_empty(&pending_raid_disks)) {
1900                 dev_t dev;
1901                 rdev0 = list_entry(pending_raid_disks.next,
1902                                          mdk_rdev_t, same_set);
1903
1904                 printk(KERN_INFO "md: considering %s ...\n",
1905                         bdevname(rdev0->bdev,b));
1906                 INIT_LIST_HEAD(&candidates);
1907                 ITERATE_RDEV_PENDING(rdev,tmp)
1908                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1909                                 printk(KERN_INFO "md:  adding %s ...\n",
1910                                         bdevname(rdev->bdev,b));
1911                                 list_move(&rdev->same_set, &candidates);
1912                         }
1913                 /*
1914                  * now we have a set of devices, with all of them having
1915                  * mostly sane superblocks. It's time to allocate the
1916                  * mddev.
1917                  */
1918                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1919                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1920                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1921                         break;
1922                 }
1923                 if (part)
1924                         dev = MKDEV(mdp_major,
1925                                     rdev0->preferred_minor << MdpMinorShift);
1926                 else
1927                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1928
1929                 md_probe(dev, NULL, NULL);
1930                 mddev = mddev_find(dev);
1931                 if (!mddev) {
1932                         printk(KERN_ERR 
1933                                 "md: cannot allocate memory for md drive.\n");
1934                         break;
1935                 }
1936                 if (mddev_lock(mddev)) 
1937                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1938                                mdname(mddev));
1939                 else if (mddev->raid_disks || mddev->major_version
1940                          || !list_empty(&mddev->disks)) {
1941                         printk(KERN_WARNING 
1942                                 "md: %s already running, cannot run %s\n",
1943                                 mdname(mddev), bdevname(rdev0->bdev,b));
1944                         mddev_unlock(mddev);
1945                 } else {
1946                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1947                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1948                                 list_del_init(&rdev->same_set);
1949                                 if (bind_rdev_to_array(rdev, mddev))
1950                                         export_rdev(rdev);
1951                         }
1952                         autorun_array(mddev);
1953                         mddev_unlock(mddev);
1954                 }
1955                 /* on success, candidates will be empty, on error
1956                  * it won't...
1957                  */
1958                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1959                         export_rdev(rdev);
1960                 mddev_put(mddev);
1961         }
1962         printk(KERN_INFO "md: ... autorun DONE.\n");
1963 }
1964
1965 /*
1966  * import RAID devices based on one partition
1967  * if possible, the array gets run as well.
1968  */
1969
1970 static int autostart_array(dev_t startdev)
1971 {
1972         char b[BDEVNAME_SIZE];
1973         int err = -EINVAL, i;
1974         mdp_super_t *sb = NULL;
1975         mdk_rdev_t *start_rdev = NULL, *rdev;
1976
1977         start_rdev = md_import_device(startdev, 0, 0);
1978         if (IS_ERR(start_rdev))
1979                 return err;
1980
1981
1982         /* NOTE: this can only work for 0.90.0 superblocks */
1983         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1984         if (sb->major_version != 0 ||
1985             sb->minor_version != 90 ) {
1986                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1987                 export_rdev(start_rdev);
1988                 return err;
1989         }
1990
1991         if (start_rdev->faulty) {
1992                 printk(KERN_WARNING 
1993                         "md: can not autostart based on faulty %s!\n",
1994                         bdevname(start_rdev->bdev,b));
1995                 export_rdev(start_rdev);
1996                 return err;
1997         }
1998         list_add(&start_rdev->same_set, &pending_raid_disks);
1999
2000         for (i = 0; i < MD_SB_DISKS; i++) {
2001                 mdp_disk_t *desc = sb->disks + i;
2002                 dev_t dev = MKDEV(desc->major, desc->minor);
2003
2004                 if (!dev)
2005                         continue;
2006                 if (dev == startdev)
2007                         continue;
2008                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2009                         continue;
2010                 rdev = md_import_device(dev, 0, 0);
2011                 if (IS_ERR(rdev))
2012                         continue;
2013
2014                 list_add(&rdev->same_set, &pending_raid_disks);
2015         }
2016
2017         /*
2018          * possibly return codes
2019          */
2020         autorun_devices(0);
2021         return 0;
2022
2023 }
2024
2025
2026 static int get_version(void __user * arg)
2027 {
2028         mdu_version_t ver;
2029
2030         ver.major = MD_MAJOR_VERSION;
2031         ver.minor = MD_MINOR_VERSION;
2032         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2033
2034         if (copy_to_user(arg, &ver, sizeof(ver)))
2035                 return -EFAULT;
2036
2037         return 0;
2038 }
2039
2040 static int get_array_info(mddev_t * mddev, void __user * arg)
2041 {
2042         mdu_array_info_t info;
2043         int nr,working,active,failed,spare;
2044         mdk_rdev_t *rdev;
2045         struct list_head *tmp;
2046
2047         nr=working=active=failed=spare=0;
2048         ITERATE_RDEV(mddev,rdev,tmp) {
2049                 nr++;
2050                 if (rdev->faulty)
2051                         failed++;
2052                 else {
2053                         working++;
2054                         if (rdev->in_sync)
2055                                 active++;       
2056                         else
2057                                 spare++;
2058                 }
2059         }
2060
2061         info.major_version = mddev->major_version;
2062         info.minor_version = mddev->minor_version;
2063         info.patch_version = MD_PATCHLEVEL_VERSION;
2064         info.ctime         = mddev->ctime;
2065         info.level         = mddev->level;
2066         info.size          = mddev->size;
2067         info.nr_disks      = nr;
2068         info.raid_disks    = mddev->raid_disks;
2069         info.md_minor      = mddev->md_minor;
2070         info.not_persistent= !mddev->persistent;
2071
2072         info.utime         = mddev->utime;
2073         info.state         = 0;
2074         if (mddev->in_sync)
2075                 info.state = (1<<MD_SB_CLEAN);
2076         info.active_disks  = active;
2077         info.working_disks = working;
2078         info.failed_disks  = failed;
2079         info.spare_disks   = spare;
2080
2081         info.layout        = mddev->layout;
2082         info.chunk_size    = mddev->chunk_size;
2083
2084         if (copy_to_user(arg, &info, sizeof(info)))
2085                 return -EFAULT;
2086
2087         return 0;
2088 }
2089
2090 static int get_bitmap_file(mddev_t * mddev, void * arg)
2091 {
2092         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2093         char *ptr, *buf = NULL;
2094         int err = -ENOMEM;
2095
2096         file = kmalloc(sizeof(*file), GFP_KERNEL);
2097         if (!file)
2098                 goto out;
2099
2100         /* bitmap disabled, zero the first byte and copy out */
2101         if (!mddev->bitmap || !mddev->bitmap->file) {
2102                 file->pathname[0] = '\0';
2103                 goto copy_out;
2104         }
2105
2106         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2107         if (!buf)
2108                 goto out;
2109
2110         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2111         if (!ptr)
2112                 goto out;
2113
2114         strcpy(file->pathname, ptr);
2115
2116 copy_out:
2117         err = 0;
2118         if (copy_to_user(arg, file, sizeof(*file)))
2119                 err = -EFAULT;
2120 out:
2121         kfree(buf);
2122         kfree(file);
2123         return err;
2124 }
2125
2126 static int get_disk_info(mddev_t * mddev, void __user * arg)
2127 {
2128         mdu_disk_info_t info;
2129         unsigned int nr;
2130         mdk_rdev_t *rdev;
2131
2132         if (copy_from_user(&info, arg, sizeof(info)))
2133                 return -EFAULT;
2134
2135         nr = info.number;
2136
2137         rdev = find_rdev_nr(mddev, nr);
2138         if (rdev) {
2139                 info.major = MAJOR(rdev->bdev->bd_dev);
2140                 info.minor = MINOR(rdev->bdev->bd_dev);
2141                 info.raid_disk = rdev->raid_disk;
2142                 info.state = 0;
2143                 if (rdev->faulty)
2144                         info.state |= (1<<MD_DISK_FAULTY);
2145                 else if (rdev->in_sync) {
2146                         info.state |= (1<<MD_DISK_ACTIVE);
2147                         info.state |= (1<<MD_DISK_SYNC);
2148                 }
2149         } else {
2150                 info.major = info.minor = 0;
2151                 info.raid_disk = -1;
2152                 info.state = (1<<MD_DISK_REMOVED);
2153         }
2154
2155         if (copy_to_user(arg, &info, sizeof(info)))
2156                 return -EFAULT;
2157
2158         return 0;
2159 }
2160
2161 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2162 {
2163         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2164         mdk_rdev_t *rdev;
2165         dev_t dev = MKDEV(info->major,info->minor);
2166
2167         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2168                 return -EOVERFLOW;
2169
2170         if (!mddev->raid_disks) {
2171                 int err;
2172                 /* expecting a device which has a superblock */
2173                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2174                 if (IS_ERR(rdev)) {
2175                         printk(KERN_WARNING 
2176                                 "md: md_import_device returned %ld\n",
2177                                 PTR_ERR(rdev));
2178                         return PTR_ERR(rdev);
2179                 }
2180                 if (!list_empty(&mddev->disks)) {
2181                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2182                                                         mdk_rdev_t, same_set);
2183                         int err = super_types[mddev->major_version]
2184                                 .load_super(rdev, rdev0, mddev->minor_version);
2185                         if (err < 0) {
2186                                 printk(KERN_WARNING 
2187                                         "md: %s has different UUID to %s\n",
2188                                         bdevname(rdev->bdev,b), 
2189                                         bdevname(rdev0->bdev,b2));
2190                                 export_rdev(rdev);
2191                                 return -EINVAL;
2192                         }
2193                 }
2194                 err = bind_rdev_to_array(rdev, mddev);
2195                 if (err)
2196                         export_rdev(rdev);
2197                 return err;
2198         }
2199
2200         /*
2201          * add_new_disk can be used once the array is assembled
2202          * to add "hot spares".  They must already have a superblock
2203          * written
2204          */
2205         if (mddev->pers) {
2206                 int err;
2207                 if (!mddev->pers->hot_add_disk) {
2208                         printk(KERN_WARNING 
2209                                 "%s: personality does not support diskops!\n",
2210                                mdname(mddev));
2211                         return -EINVAL;
2212                 }
2213                 rdev = md_import_device(dev, mddev->major_version,
2214                                         mddev->minor_version);
2215                 if (IS_ERR(rdev)) {
2216                         printk(KERN_WARNING 
2217                                 "md: md_import_device returned %ld\n",
2218                                 PTR_ERR(rdev));
2219                         return PTR_ERR(rdev);
2220                 }
2221                 /* set save_raid_disk if appropriate */
2222                 if (!mddev->persistent) {
2223                         if (info->state & (1<<MD_DISK_SYNC)  &&
2224                             info->raid_disk < mddev->raid_disks)
2225                                 rdev->raid_disk = info->raid_disk;
2226                         else
2227                                 rdev->raid_disk = -1;
2228                 } else
2229                         super_types[mddev->major_version].
2230                                 validate_super(mddev, rdev);
2231                 rdev->saved_raid_disk = rdev->raid_disk;
2232
2233                 rdev->in_sync = 0; /* just to be sure */
2234                 rdev->raid_disk = -1;
2235                 err = bind_rdev_to_array(rdev, mddev);
2236                 if (err)
2237                         export_rdev(rdev);
2238
2239                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2240                 md_wakeup_thread(mddev->thread);
2241                 return err;
2242         }
2243
2244         /* otherwise, add_new_disk is only allowed
2245          * for major_version==0 superblocks
2246          */
2247         if (mddev->major_version != 0) {
2248                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2249                        mdname(mddev));
2250                 return -EINVAL;
2251         }
2252
2253         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2254                 int err;
2255                 rdev = md_import_device (dev, -1, 0);
2256                 if (IS_ERR(rdev)) {
2257                         printk(KERN_WARNING 
2258                                 "md: error, md_import_device() returned %ld\n",
2259                                 PTR_ERR(rdev));
2260                         return PTR_ERR(rdev);
2261                 }
2262                 rdev->desc_nr = info->number;
2263                 if (info->raid_disk < mddev->raid_disks)
2264                         rdev->raid_disk = info->raid_disk;
2265                 else
2266                         rdev->raid_disk = -1;
2267
2268                 rdev->faulty = 0;
2269                 if (rdev->raid_disk < mddev->raid_disks)
2270                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2271                 else
2272                         rdev->in_sync = 0;
2273
2274                 err = bind_rdev_to_array(rdev, mddev);
2275                 if (err) {
2276                         export_rdev(rdev);
2277                         return err;
2278                 }
2279
2280                 if (!mddev->persistent) {
2281                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2282                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2283                 } else 
2284                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2285                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2286
2287                 if (!mddev->size || (mddev->size > rdev->size))
2288                         mddev->size = rdev->size;
2289         }
2290
2291         return 0;
2292 }
2293
2294 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2295 {
2296         char b[BDEVNAME_SIZE];
2297         mdk_rdev_t *rdev;
2298
2299         if (!mddev->pers)
2300                 return -ENODEV;
2301
2302         rdev = find_rdev(mddev, dev);
2303         if (!rdev)
2304                 return -ENXIO;
2305
2306         if (rdev->raid_disk >= 0)
2307                 goto busy;
2308
2309         kick_rdev_from_array(rdev);
2310         md_update_sb(mddev);
2311
2312         return 0;
2313 busy:
2314         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2315                 bdevname(rdev->bdev,b), mdname(mddev));
2316         return -EBUSY;
2317 }
2318
2319 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2320 {
2321         char b[BDEVNAME_SIZE];
2322         int err;
2323         unsigned int size;
2324         mdk_rdev_t *rdev;
2325
2326         if (!mddev->pers)
2327                 return -ENODEV;
2328
2329         if (mddev->major_version != 0) {
2330                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2331                         " version-0 superblocks.\n",
2332                         mdname(mddev));
2333                 return -EINVAL;
2334         }
2335         if (!mddev->pers->hot_add_disk) {
2336                 printk(KERN_WARNING 
2337                         "%s: personality does not support diskops!\n",
2338                         mdname(mddev));
2339                 return -EINVAL;
2340         }
2341
2342         rdev = md_import_device (dev, -1, 0);
2343         if (IS_ERR(rdev)) {
2344                 printk(KERN_WARNING 
2345                         "md: error, md_import_device() returned %ld\n",
2346                         PTR_ERR(rdev));
2347                 return -EINVAL;
2348         }
2349
2350         if (mddev->persistent)
2351                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2352         else
2353                 rdev->sb_offset =
2354                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2355
2356         size = calc_dev_size(rdev, mddev->chunk_size);
2357         rdev->size = size;
2358
2359         if (size < mddev->size) {
2360                 printk(KERN_WARNING 
2361                         "%s: disk size %llu blocks < array size %llu\n",
2362                         mdname(mddev), (unsigned long long)size,
2363                         (unsigned long long)mddev->size);
2364                 err = -ENOSPC;
2365                 goto abort_export;
2366         }
2367
2368         if (rdev->faulty) {
2369                 printk(KERN_WARNING 
2370                         "md: can not hot-add faulty %s disk to %s!\n",
2371                         bdevname(rdev->bdev,b), mdname(mddev));
2372                 err = -EINVAL;
2373                 goto abort_export;
2374         }
2375         rdev->in_sync = 0;
2376         rdev->desc_nr = -1;
2377         bind_rdev_to_array(rdev, mddev);
2378
2379         /*
2380          * The rest should better be atomic, we can have disk failures
2381          * noticed in interrupt contexts ...
2382          */
2383
2384         if (rdev->desc_nr == mddev->max_disks) {
2385                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2386                         mdname(mddev));
2387                 err = -EBUSY;
2388                 goto abort_unbind_export;
2389         }
2390
2391         rdev->raid_disk = -1;
2392
2393         md_update_sb(mddev);
2394
2395         /*
2396          * Kick recovery, maybe this spare has to be added to the
2397          * array immediately.
2398          */
2399         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2400         md_wakeup_thread(mddev->thread);
2401
2402         return 0;
2403
2404 abort_unbind_export:
2405         unbind_rdev_from_array(rdev);
2406
2407 abort_export:
2408         export_rdev(rdev);
2409         return err;
2410 }
2411
2412 /* similar to deny_write_access, but accounts for our holding a reference
2413  * to the file ourselves */
2414 static int deny_bitmap_write_access(struct file * file)
2415 {
2416         struct inode *inode = file->f_mapping->host;
2417
2418         spin_lock(&inode->i_lock);
2419         if (atomic_read(&inode->i_writecount) > 1) {
2420                 spin_unlock(&inode->i_lock);
2421                 return -ETXTBSY;
2422         }
2423         atomic_set(&inode->i_writecount, -1);
2424         spin_unlock(&inode->i_lock);
2425
2426         return 0;
2427 }
2428
2429 static int set_bitmap_file(mddev_t *mddev, int fd)
2430 {
2431         int err;
2432
2433         if (mddev->pers)
2434                 return -EBUSY;
2435
2436         mddev->bitmap_file = fget(fd);
2437
2438         if (mddev->bitmap_file == NULL) {
2439                 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2440                         mdname(mddev));
2441                 return -EBADF;
2442         }
2443
2444         err = deny_bitmap_write_access(mddev->bitmap_file);
2445         if (err) {
2446                 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2447                         mdname(mddev));
2448                 fput(mddev->bitmap_file);
2449                 mddev->bitmap_file = NULL;
2450         } else
2451                 mddev->bitmap_offset = 0; /* file overrides offset */
2452         return err;
2453 }
2454
2455 /*
2456  * set_array_info is used two different ways
2457  * The original usage is when creating a new array.
2458  * In this usage, raid_disks is > 0 and it together with
2459  *  level, size, not_persistent,layout,chunksize determine the
2460  *  shape of the array.
2461  *  This will always create an array with a type-0.90.0 superblock.
2462  * The newer usage is when assembling an array.
2463  *  In this case raid_disks will be 0, and the major_version field is
2464  *  use to determine which style super-blocks are to be found on the devices.
2465  *  The minor and patch _version numbers are also kept incase the
2466  *  super_block handler wishes to interpret them.
2467  */
2468 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2469 {
2470
2471         if (info->raid_disks == 0) {
2472                 /* just setting version number for superblock loading */
2473                 if (info->major_version < 0 ||
2474                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2475                     super_types[info->major_version].name == NULL) {
2476                         /* maybe try to auto-load a module? */
2477                         printk(KERN_INFO 
2478                                 "md: superblock version %d not known\n",
2479                                 info->major_version);
2480                         return -EINVAL;
2481                 }
2482                 mddev->major_version = info->major_version;
2483                 mddev->minor_version = info->minor_version;
2484                 mddev->patch_version = info->patch_version;
2485                 return 0;
2486         }
2487         mddev->major_version = MD_MAJOR_VERSION;
2488         mddev->minor_version = MD_MINOR_VERSION;
2489         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2490         mddev->ctime         = get_seconds();
2491
2492         mddev->level         = info->level;
2493         mddev->size          = info->size;
2494         mddev->raid_disks    = info->raid_disks;
2495         /* don't set md_minor, it is determined by which /dev/md* was
2496          * openned
2497          */
2498         if (info->state & (1<<MD_SB_CLEAN))
2499                 mddev->recovery_cp = MaxSector;
2500         else
2501                 mddev->recovery_cp = 0;
2502         mddev->persistent    = ! info->not_persistent;
2503
2504         mddev->layout        = info->layout;
2505         mddev->chunk_size    = info->chunk_size;
2506
2507         mddev->max_disks     = MD_SB_DISKS;
2508
2509         mddev->sb_dirty      = 1;
2510
2511         /*
2512          * Generate a 128 bit UUID
2513          */
2514         get_random_bytes(mddev->uuid, 16);
2515
2516         return 0;
2517 }
2518
2519 /*
2520  * update_array_info is used to change the configuration of an
2521  * on-line array.
2522  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2523  * fields in the info are checked against the array.
2524  * Any differences that cannot be handled will cause an error.
2525  * Normally, only one change can be managed at a time.
2526  */
2527 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2528 {
2529         int rv = 0;
2530         int cnt = 0;
2531
2532         if (mddev->major_version != info->major_version ||
2533             mddev->minor_version != info->minor_version ||
2534 /*          mddev->patch_version != info->patch_version || */
2535             mddev->ctime         != info->ctime         ||
2536             mddev->level         != info->level         ||
2537 /*          mddev->layout        != info->layout        || */
2538             !mddev->persistent   != info->not_persistent||
2539             mddev->chunk_size    != info->chunk_size    )
2540                 return -EINVAL;
2541         /* Check there is only one change */
2542         if (mddev->size != info->size) cnt++;
2543         if (mddev->raid_disks != info->raid_disks) cnt++;
2544         if (mddev->layout != info->layout) cnt++;
2545         if (cnt == 0) return 0;
2546         if (cnt > 1) return -EINVAL;
2547
2548         if (mddev->layout != info->layout) {
2549                 /* Change layout
2550                  * we don't need to do anything at the md level, the
2551                  * personality will take care of it all.
2552                  */
2553                 if (mddev->pers->reconfig == NULL)
2554                         return -EINVAL;
2555                 else
2556                         return mddev->pers->reconfig(mddev, info->layout, -1);
2557         }
2558         if (mddev->size != info->size) {
2559                 mdk_rdev_t * rdev;
2560                 struct list_head *tmp;
2561                 if (mddev->pers->resize == NULL)
2562                         return -EINVAL;
2563                 /* The "size" is the amount of each device that is used.
2564                  * This can only make sense for arrays with redundancy.
2565                  * linear and raid0 always use whatever space is available
2566                  * We can only consider changing the size if no resync
2567                  * or reconstruction is happening, and if the new size
2568                  * is acceptable. It must fit before the sb_offset or,
2569                  * if that is <data_offset, it must fit before the
2570                  * size of each device.
2571                  * If size is zero, we find the largest size that fits.
2572                  */
2573                 if (mddev->sync_thread)
2574                         return -EBUSY;
2575                 ITERATE_RDEV(mddev,rdev,tmp) {
2576                         sector_t avail;
2577                         int fit = (info->size == 0);
2578                         if (rdev->sb_offset > rdev->data_offset)
2579                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2580                         else
2581                                 avail = get_capacity(rdev->bdev->bd_disk)
2582                                         - rdev->data_offset;
2583                         if (fit && (info->size == 0 || info->size > avail/2))
2584                                 info->size = avail/2;
2585                         if (avail < ((sector_t)info->size << 1))
2586                                 return -ENOSPC;
2587                 }
2588                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2589                 if (!rv) {
2590                         struct block_device *bdev;
2591
2592                         bdev = bdget_disk(mddev->gendisk, 0);
2593                         if (bdev) {
2594                                 down(&bdev->bd_inode->i_sem);
2595                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2596                                 up(&bdev->bd_inode->i_sem);
2597                                 bdput(bdev);
2598                         }
2599                 }
2600         }
2601         if (mddev->raid_disks    != info->raid_disks) {
2602                 /* change the number of raid disks */
2603                 if (mddev->pers->reshape == NULL)
2604                         return -EINVAL;
2605                 if (info->raid_disks <= 0 ||
2606                     info->raid_disks >= mddev->max_disks)
2607                         return -EINVAL;
2608                 if (mddev->sync_thread)
2609                         return -EBUSY;
2610                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2611                 if (!rv) {
2612                         struct block_device *bdev;
2613
2614                         bdev = bdget_disk(mddev->gendisk, 0);
2615                         if (bdev) {
2616                                 down(&bdev->bd_inode->i_sem);
2617                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2618                                 up(&bdev->bd_inode->i_sem);
2619                                 bdput(bdev);
2620                         }
2621                 }
2622         }
2623         md_update_sb(mddev);
2624         return rv;
2625 }
2626
2627 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2628 {
2629         mdk_rdev_t *rdev;
2630
2631         if (mddev->pers == NULL)
2632                 return -ENODEV;
2633
2634         rdev = find_rdev(mddev, dev);
2635         if (!rdev)
2636                 return -ENODEV;
2637
2638         md_error(mddev, rdev);
2639         return 0;
2640 }
2641
2642 static int md_ioctl(struct inode *inode, struct file *file,
2643                         unsigned int cmd, unsigned long arg)
2644 {
2645         int err = 0;
2646         void __user *argp = (void __user *)arg;
2647         struct hd_geometry __user *loc = argp;
2648         mddev_t *mddev = NULL;
2649
2650         if (!capable(CAP_SYS_ADMIN))
2651                 return -EACCES;
2652
2653         /*
2654          * Commands dealing with the RAID driver but not any
2655          * particular array:
2656          */
2657         switch (cmd)
2658         {
2659                 case RAID_VERSION:
2660                         err = get_version(argp);
2661                         goto done;
2662
2663                 case PRINT_RAID_DEBUG:
2664                         err = 0;
2665                         md_print_devices();
2666                         goto done;
2667
2668 #ifndef MODULE
2669                 case RAID_AUTORUN:
2670                         err = 0;
2671                         autostart_arrays(arg);
2672                         goto done;
2673 #endif
2674                 default:;
2675         }
2676
2677         /*
2678          * Commands creating/starting a new array:
2679          */
2680
2681         mddev = inode->i_bdev->bd_disk->private_data;
2682
2683         if (!mddev) {
2684                 BUG();
2685                 goto abort;
2686         }
2687
2688
2689         if (cmd == START_ARRAY) {
2690                 /* START_ARRAY doesn't need to lock the array as autostart_array
2691                  * does the locking, and it could even be a different array
2692                  */
2693                 static int cnt = 3;
2694                 if (cnt > 0 ) {
2695                         printk(KERN_WARNING
2696                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2697                                "This will not be supported beyond 2.6\n",
2698                                current->comm, current->pid);
2699                         cnt--;
2700                 }
2701                 err = autostart_array(new_decode_dev(arg));
2702                 if (err) {
2703                         printk(KERN_WARNING "md: autostart failed!\n");
2704                         goto abort;
2705                 }
2706                 goto done;
2707         }
2708
2709         err = mddev_lock(mddev);
2710         if (err) {
2711                 printk(KERN_INFO 
2712                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2713                         err, cmd);
2714                 goto abort;
2715         }
2716
2717         switch (cmd)
2718         {
2719                 case SET_ARRAY_INFO:
2720                         {
2721                                 mdu_array_info_t info;
2722                                 if (!arg)
2723                                         memset(&info, 0, sizeof(info));
2724                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2725                                         err = -EFAULT;
2726                                         goto abort_unlock;
2727                                 }
2728                                 if (mddev->pers) {
2729                                         err = update_array_info(mddev, &info);
2730                                         if (err) {
2731                                                 printk(KERN_WARNING "md: couldn't update"
2732                                                        " array info. %d\n", err);
2733                                                 goto abort_unlock;
2734                                         }
2735                                         goto done_unlock;
2736                                 }
2737                                 if (!list_empty(&mddev->disks)) {
2738                                         printk(KERN_WARNING
2739                                                "md: array %s already has disks!\n",
2740                                                mdname(mddev));
2741                                         err = -EBUSY;
2742                                         goto abort_unlock;
2743                                 }
2744                                 if (mddev->raid_disks) {
2745                                         printk(KERN_WARNING
2746                                                "md: array %s already initialised!\n",
2747                                                mdname(mddev));
2748                                         err = -EBUSY;
2749                                         goto abort_unlock;
2750                                 }
2751                                 err = set_array_info(mddev, &info);
2752                                 if (err) {
2753                                         printk(KERN_WARNING "md: couldn't set"
2754                                                " array info. %d\n", err);
2755                                         goto abort_unlock;
2756                                 }
2757                         }
2758                         goto done_unlock;
2759
2760                 default:;
2761         }
2762
2763         /*
2764          * Commands querying/configuring an existing array:
2765          */
2766         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2767          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2768         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2769                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2770                 err = -ENODEV;
2771                 goto abort_unlock;
2772         }
2773
2774         /*
2775          * Commands even a read-only array can execute:
2776          */
2777         switch (cmd)
2778         {
2779                 case GET_ARRAY_INFO:
2780                         err = get_array_info(mddev, argp);
2781                         goto done_unlock;
2782
2783                 case GET_BITMAP_FILE:
2784                         err = get_bitmap_file(mddev, (void *)arg);
2785                         goto done_unlock;
2786
2787                 case GET_DISK_INFO:
2788                         err = get_disk_info(mddev, argp);
2789                         goto done_unlock;
2790
2791                 case RESTART_ARRAY_RW:
2792                         err = restart_array(mddev);
2793                         goto done_unlock;
2794
2795                 case STOP_ARRAY:
2796                         err = do_md_stop (mddev, 0);
2797                         goto done_unlock;
2798
2799                 case STOP_ARRAY_RO:
2800                         err = do_md_stop (mddev, 1);
2801                         goto done_unlock;
2802
2803         /*
2804          * We have a problem here : there is no easy way to give a CHS
2805          * virtual geometry. We currently pretend that we have a 2 heads
2806          * 4 sectors (with a BIG number of cylinders...). This drives
2807          * dosfs just mad... ;-)
2808          */
2809                 case HDIO_GETGEO:
2810                         if (!loc) {
2811                                 err = -EINVAL;
2812                                 goto abort_unlock;
2813                         }
2814                         err = put_user (2, (char __user *) &loc->heads);
2815                         if (err)
2816                                 goto abort_unlock;
2817                         err = put_user (4, (char __user *) &loc->sectors);
2818                         if (err)
2819                                 goto abort_unlock;
2820                         err = put_user(get_capacity(mddev->gendisk)/8,
2821                                         (short __user *) &loc->cylinders);
2822                         if (err)
2823                                 goto abort_unlock;
2824                         err = put_user (get_start_sect(inode->i_bdev),
2825                                                 (long __user *) &loc->start);
2826                         goto done_unlock;
2827         }
2828
2829         /*
2830          * The remaining ioctls are changing the state of the
2831          * superblock, so we do not allow read-only arrays
2832          * here:
2833          */
2834         if (mddev->ro) {
2835                 err = -EROFS;
2836                 goto abort_unlock;
2837         }
2838
2839         switch (cmd)
2840         {
2841                 case ADD_NEW_DISK:
2842                 {
2843                         mdu_disk_info_t info;
2844                         if (copy_from_user(&info, argp, sizeof(info)))
2845                                 err = -EFAULT;
2846                         else
2847                                 err = add_new_disk(mddev, &info);
2848                         goto done_unlock;
2849                 }
2850
2851                 case HOT_REMOVE_DISK:
2852                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2853                         goto done_unlock;
2854
2855                 case HOT_ADD_DISK:
2856                         err = hot_add_disk(mddev, new_decode_dev(arg));
2857                         goto done_unlock;
2858
2859                 case SET_DISK_FAULTY:
2860                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2861                         goto done_unlock;
2862
2863                 case RUN_ARRAY:
2864                         err = do_md_run (mddev);
2865                         goto done_unlock;
2866
2867                 case SET_BITMAP_FILE:
2868                         err = set_bitmap_file(mddev, (int)arg);
2869                         goto done_unlock;
2870
2871                 default:
2872                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2873                                 printk(KERN_WARNING "md: %s(pid %d) used"
2874                                         " obsolete MD ioctl, upgrade your"
2875                                         " software to use new ictls.\n",
2876                                         current->comm, current->pid);
2877                         err = -EINVAL;
2878                         goto abort_unlock;
2879         }
2880
2881 done_unlock:
2882 abort_unlock:
2883         mddev_unlock(mddev);
2884
2885         return err;
2886 done:
2887         if (err)
2888                 MD_BUG();
2889 abort:
2890         return err;
2891 }
2892
2893 static int md_open(struct inode *inode, struct file *file)
2894 {
2895         /*
2896          * Succeed if we can lock the mddev, which confirms that
2897          * it isn't being stopped right now.
2898          */
2899         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2900         int err;
2901
2902         if ((err = mddev_lock(mddev)))
2903                 goto out;
2904
2905         err = 0;
2906         mddev_get(mddev);
2907         mddev_unlock(mddev);
2908
2909         check_disk_change(inode->i_bdev);
2910  out:
2911         return err;
2912 }
2913
2914 static int md_release(struct inode *inode, struct file * file)
2915 {
2916         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2917
2918         if (!mddev)
2919                 BUG();
2920         mddev_put(mddev);
2921
2922         return 0;
2923 }
2924
2925 static int md_media_changed(struct gendisk *disk)
2926 {
2927         mddev_t *mddev = disk->private_data;
2928
2929         return mddev->changed;
2930 }
2931
2932 static int md_revalidate(struct gendisk *disk)
2933 {
2934         mddev_t *mddev = disk->private_data;
2935
2936         mddev->changed = 0;
2937         return 0;
2938 }
2939 static struct block_device_operations md_fops =
2940 {
2941         .owner          = THIS_MODULE,
2942         .open           = md_open,
2943         .release        = md_release,
2944         .ioctl          = md_ioctl,
2945         .media_changed  = md_media_changed,
2946         .revalidate_disk= md_revalidate,
2947 };
2948
2949 static int md_thread(void * arg)
2950 {
2951         mdk_thread_t *thread = arg;
2952
2953         lock_kernel();
2954
2955         /*
2956          * Detach thread
2957          */
2958
2959         daemonize(thread->name, mdname(thread->mddev));
2960
2961         current->exit_signal = SIGCHLD;
2962         allow_signal(SIGKILL);
2963         thread->tsk = current;
2964
2965         /*
2966          * md_thread is a 'system-thread', it's priority should be very
2967          * high. We avoid resource deadlocks individually in each
2968          * raid personality. (RAID5 does preallocation) We also use RR and
2969          * the very same RT priority as kswapd, thus we will never get
2970          * into a priority inversion deadlock.
2971          *
2972          * we definitely have to have equal or higher priority than
2973          * bdflush, otherwise bdflush will deadlock if there are too
2974          * many dirty RAID5 blocks.
2975          */
2976         unlock_kernel();
2977
2978         complete(thread->event);
2979         while (thread->run) {
2980                 void (*run)(mddev_t *);
2981
2982                 wait_event_interruptible_timeout(thread->wqueue,
2983                                                  test_bit(THREAD_WAKEUP, &thread->flags),
2984                                                  thread->timeout);
2985                 try_to_freeze();
2986
2987                 clear_bit(THREAD_WAKEUP, &thread->flags);
2988
2989                 run = thread->run;
2990                 if (run)
2991                         run(thread->mddev);
2992
2993                 if (signal_pending(current))
2994                         flush_signals(current);
2995         }
2996         complete(thread->event);
2997         return 0;
2998 }
2999
3000 void md_wakeup_thread(mdk_thread_t *thread)
3001 {
3002         if (thread) {
3003                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3004                 set_bit(THREAD_WAKEUP, &thread->flags);
3005                 wake_up(&thread->wqueue);
3006         }
3007 }
3008
3009 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3010                                  const char *name)
3011 {
3012         mdk_thread_t *thread;
3013         int ret;
3014         struct completion event;
3015
3016         thread = (mdk_thread_t *) kmalloc
3017                                 (sizeof(mdk_thread_t), GFP_KERNEL);
3018         if (!thread)
3019                 return NULL;
3020
3021         memset(thread, 0, sizeof(mdk_thread_t));
3022         init_waitqueue_head(&thread->wqueue);
3023
3024         init_completion(&event);
3025         thread->event = &event;
3026         thread->run = run;
3027         thread->mddev = mddev;
3028         thread->name = name;
3029         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3030         ret = kernel_thread(md_thread, thread, 0);
3031         if (ret < 0) {
3032                 kfree(thread);
3033                 return NULL;
3034         }
3035         wait_for_completion(&event);
3036         return thread;
3037 }
3038
3039 void md_unregister_thread(mdk_thread_t *thread)
3040 {
3041         struct completion event;
3042
3043         init_completion(&event);
3044
3045         thread->event = &event;
3046
3047         /* As soon as ->run is set to NULL, the task could disappear,
3048          * so we need to hold tasklist_lock until we have sent the signal
3049          */
3050         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3051         read_lock(&tasklist_lock);
3052         thread->run = NULL;
3053         send_sig(SIGKILL, thread->tsk, 1);
3054         read_unlock(&tasklist_lock);
3055         wait_for_completion(&event);
3056         kfree(thread);
3057 }
3058
3059 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3060 {
3061         if (!mddev) {
3062                 MD_BUG();
3063                 return;
3064         }
3065
3066         if (!rdev || rdev->faulty)
3067                 return;
3068 /*
3069         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3070                 mdname(mddev),
3071                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3072                 __builtin_return_address(0),__builtin_return_address(1),
3073                 __builtin_return_address(2),__builtin_return_address(3));
3074 */
3075         if (!mddev->pers->error_handler)
3076                 return;
3077         mddev->pers->error_handler(mddev,rdev);
3078         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3079         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3080         md_wakeup_thread(mddev->thread);
3081 }
3082
3083 /* seq_file implementation /proc/mdstat */
3084
3085 static void status_unused(struct seq_file *seq)
3086 {
3087         int i = 0;
3088         mdk_rdev_t *rdev;
3089         struct list_head *tmp;
3090
3091         seq_printf(seq, "unused devices: ");
3092
3093         ITERATE_RDEV_PENDING(rdev,tmp) {
3094                 char b[BDEVNAME_SIZE];
3095                 i++;
3096                 seq_printf(seq, "%s ",
3097                               bdevname(rdev->bdev,b));
3098         }
3099         if (!i)
3100                 seq_printf(seq, "<none>");
3101
3102         seq_printf(seq, "\n");
3103 }
3104
3105
3106 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3107 {
3108         unsigned long max_blocks, resync, res, dt, db, rt;
3109
3110         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3111
3112         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3113                 max_blocks = mddev->resync_max_sectors >> 1;
3114         else
3115                 max_blocks = mddev->size;
3116
3117         /*
3118          * Should not happen.
3119          */
3120         if (!max_blocks) {
3121                 MD_BUG();
3122                 return;
3123         }
3124         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3125         {
3126                 int i, x = res/50, y = 20-x;
3127                 seq_printf(seq, "[");
3128                 for (i = 0; i < x; i++)
3129                         seq_printf(seq, "=");
3130                 seq_printf(seq, ">");
3131                 for (i = 0; i < y; i++)
3132                         seq_printf(seq, ".");
3133                 seq_printf(seq, "] ");
3134         }
3135         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3136                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3137                        "resync" : "recovery"),
3138                       res/10, res % 10, resync, max_blocks);
3139
3140         /*
3141          * We do not want to overflow, so the order of operands and
3142          * the * 100 / 100 trick are important. We do a +1 to be
3143          * safe against division by zero. We only estimate anyway.
3144          *
3145          * dt: time from mark until now
3146          * db: blocks written from mark until now
3147          * rt: remaining time
3148          */
3149         dt = ((jiffies - mddev->resync_mark) / HZ);
3150         if (!dt) dt++;
3151         db = resync - (mddev->resync_mark_cnt/2);
3152         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3153
3154         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3155
3156         seq_printf(seq, " speed=%ldK/sec", db/dt);
3157 }
3158
3159 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3160 {
3161         struct list_head *tmp;
3162         loff_t l = *pos;
3163         mddev_t *mddev;
3164
3165         if (l >= 0x10000)
3166                 return NULL;
3167         if (!l--)
3168                 /* header */
3169                 return (void*)1;
3170
3171         spin_lock(&all_mddevs_lock);
3172         list_for_each(tmp,&all_mddevs)
3173                 if (!l--) {
3174                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3175                         mddev_get(mddev);
3176                         spin_unlock(&all_mddevs_lock);
3177                         return mddev;
3178                 }
3179         spin_unlock(&all_mddevs_lock);
3180         if (!l--)
3181                 return (void*)2;/* tail */
3182         return NULL;
3183 }
3184
3185 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3186 {
3187         struct list_head *tmp;
3188         mddev_t *next_mddev, *mddev = v;
3189         
3190         ++*pos;
3191         if (v == (void*)2)
3192                 return NULL;
3193
3194         spin_lock(&all_mddevs_lock);
3195         if (v == (void*)1)
3196                 tmp = all_mddevs.next;
3197         else
3198                 tmp = mddev->all_mddevs.next;
3199         if (tmp != &all_mddevs)
3200                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3201         else {
3202                 next_mddev = (void*)2;
3203                 *pos = 0x10000;
3204         }               
3205         spin_unlock(&all_mddevs_lock);
3206
3207         if (v != (void*)1)
3208                 mddev_put(mddev);
3209         return next_mddev;
3210
3211 }
3212
3213 static void md_seq_stop(struct seq_file *seq, void *v)
3214 {
3215         mddev_t *mddev = v;
3216
3217         if (mddev && v != (void*)1 && v != (void*)2)
3218                 mddev_put(mddev);
3219 }
3220
3221 static int md_seq_show(struct seq_file *seq, void *v)
3222 {
3223         mddev_t *mddev = v;
3224         sector_t size;
3225         struct list_head *tmp2;
3226         mdk_rdev_t *rdev;
3227         int i;
3228         struct bitmap *bitmap;
3229
3230         if (v == (void*)1) {
3231                 seq_printf(seq, "Personalities : ");
3232                 spin_lock(&pers_lock);
3233                 for (i = 0; i < MAX_PERSONALITY; i++)
3234                         if (pers[i])
3235                                 seq_printf(seq, "[%s] ", pers[i]->name);
3236
3237                 spin_unlock(&pers_lock);
3238                 seq_printf(seq, "\n");
3239                 return 0;
3240         }
3241         if (v == (void*)2) {
3242                 status_unused(seq);
3243                 return 0;
3244         }
3245
3246         if (mddev_lock(mddev)!=0) 
3247                 return -EINTR;
3248         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3249                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3250                                                 mddev->pers ? "" : "in");
3251                 if (mddev->pers) {
3252                         if (mddev->ro)
3253                                 seq_printf(seq, " (read-only)");
3254                         seq_printf(seq, " %s", mddev->pers->name);
3255                 }
3256
3257                 size = 0;
3258                 ITERATE_RDEV(mddev,rdev,tmp2) {
3259                         char b[BDEVNAME_SIZE];
3260                         seq_printf(seq, " %s[%d]",
3261                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3262                         if (rdev->faulty) {
3263                                 seq_printf(seq, "(F)");
3264                                 continue;
3265                         }
3266                         size += rdev->size;
3267                 }
3268
3269                 if (!list_empty(&mddev->disks)) {
3270                         if (mddev->pers)
3271                                 seq_printf(seq, "\n      %llu blocks",
3272                                         (unsigned long long)mddev->array_size);
3273                         else
3274                                 seq_printf(seq, "\n      %llu blocks",
3275                                         (unsigned long long)size);
3276                 }
3277
3278                 if (mddev->pers) {
3279                         mddev->pers->status (seq, mddev);
3280                         seq_printf(seq, "\n      ");
3281                         if (mddev->curr_resync > 2) {
3282                                 status_resync (seq, mddev);
3283                                 seq_printf(seq, "\n      ");
3284                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3285                                 seq_printf(seq, "       resync=DELAYED\n      ");
3286                 } else
3287                         seq_printf(seq, "\n       ");
3288
3289                 if ((bitmap = mddev->bitmap)) {
3290                         unsigned long chunk_kb;
3291                         unsigned long flags;
3292                         spin_lock_irqsave(&bitmap->lock, flags);
3293                         chunk_kb = bitmap->chunksize >> 10;
3294                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3295                                 "%lu%s chunk",
3296                                 bitmap->pages - bitmap->missing_pages,
3297                                 bitmap->pages,
3298                                 (bitmap->pages - bitmap->missing_pages)
3299                                         << (PAGE_SHIFT - 10),
3300                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3301                                 chunk_kb ? "KB" : "B");
3302                         if (bitmap->file) {
3303                                 seq_printf(seq, ", file: ");
3304                                 seq_path(seq, bitmap->file->f_vfsmnt,
3305                                          bitmap->file->f_dentry," \t\n");
3306                         }
3307
3308                         seq_printf(seq, "\n");
3309                         spin_unlock_irqrestore(&bitmap->lock, flags);
3310                 }
3311
3312                 seq_printf(seq, "\n");
3313         }
3314         mddev_unlock(mddev);
3315         
3316         return 0;
3317 }
3318
3319 static struct seq_operations md_seq_ops = {
3320         .start  = md_seq_start,
3321         .next   = md_seq_next,
3322         .stop   = md_seq_stop,
3323         .show   = md_seq_show,
3324 };
3325
3326 static int md_seq_open(struct inode *inode, struct file *file)
3327 {
3328         int error;
3329
3330         error = seq_open(file, &md_seq_ops);
3331         return error;
3332 }
3333
3334 static struct file_operations md_seq_fops = {
3335         .open           = md_seq_open,
3336         .read           = seq_read,
3337         .llseek         = seq_lseek,
3338         .release        = seq_release,
3339 };
3340
3341 int register_md_personality(int pnum, mdk_personality_t *p)
3342 {
3343         if (pnum >= MAX_PERSONALITY) {
3344                 printk(KERN_ERR
3345                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3346                        p->name, pnum, MAX_PERSONALITY-1);
3347                 return -EINVAL;
3348         }
3349
3350         spin_lock(&pers_lock);
3351         if (pers[pnum]) {
3352                 spin_unlock(&pers_lock);
3353                 return -EBUSY;
3354         }
3355
3356         pers[pnum] = p;
3357         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3358         spin_unlock(&pers_lock);
3359         return 0;
3360 }
3361
3362 int unregister_md_personality(int pnum)
3363 {
3364         if (pnum >= MAX_PERSONALITY)
3365                 return -EINVAL;
3366
3367         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3368         spin_lock(&pers_lock);
3369         pers[pnum] = NULL;
3370         spin_unlock(&pers_lock);
3371         return 0;
3372 }
3373
3374 static int is_mddev_idle(mddev_t *mddev)
3375 {
3376         mdk_rdev_t * rdev;
3377         struct list_head *tmp;
3378         int idle;
3379         unsigned long curr_events;
3380
3381         idle = 1;
3382         ITERATE_RDEV(mddev,rdev,tmp) {
3383                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3384                 curr_events = disk_stat_read(disk, read_sectors) + 
3385                                 disk_stat_read(disk, write_sectors) - 
3386                                 atomic_read(&disk->sync_io);
3387                 /* Allow some slack between valud of curr_events and last_events,
3388                  * as there are some uninteresting races.
3389                  * Note: the following is an unsigned comparison.
3390                  */
3391                 if ((curr_events - rdev->last_events + 32) > 64) {
3392                         rdev->last_events = curr_events;
3393                         idle = 0;
3394                 }
3395         }
3396         return idle;
3397 }
3398
3399 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3400 {
3401         /* another "blocks" (512byte) blocks have been synced */
3402         atomic_sub(blocks, &mddev->recovery_active);
3403         wake_up(&mddev->recovery_wait);
3404         if (!ok) {
3405                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3406                 md_wakeup_thread(mddev->thread);
3407                 // stop recovery, signal do_sync ....
3408         }
3409 }
3410
3411
3412 /* md_write_start(mddev, bi)
3413  * If we need to update some array metadata (e.g. 'active' flag
3414  * in superblock) before writing, schedule a superblock update
3415  * and wait for it to complete.
3416  */
3417 void md_write_start(mddev_t *mddev, struct bio *bi)
3418 {
3419         DEFINE_WAIT(w);
3420         if (bio_data_dir(bi) != WRITE)
3421                 return;
3422
3423         atomic_inc(&mddev->writes_pending);
3424         if (mddev->in_sync) {
3425                 spin_lock(&mddev->write_lock);
3426                 if (mddev->in_sync) {
3427                         mddev->in_sync = 0;
3428                         mddev->sb_dirty = 1;
3429                         md_wakeup_thread(mddev->thread);
3430                 }
3431                 spin_unlock(&mddev->write_lock);
3432         }
3433         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3434 }
3435
3436 void md_write_end(mddev_t *mddev)
3437 {
3438         if (atomic_dec_and_test(&mddev->writes_pending)) {
3439                 if (mddev->safemode == 2)
3440                         md_wakeup_thread(mddev->thread);
3441                 else
3442                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3443         }
3444 }
3445
3446 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3447
3448 #define SYNC_MARKS      10
3449 #define SYNC_MARK_STEP  (3*HZ)
3450 static void md_do_sync(mddev_t *mddev)
3451 {
3452         mddev_t *mddev2;
3453         unsigned int currspeed = 0,
3454                  window;
3455         sector_t max_sectors,j, io_sectors;
3456         unsigned long mark[SYNC_MARKS];
3457         sector_t mark_cnt[SYNC_MARKS];
3458         int last_mark,m;
3459         struct list_head *tmp;
3460         sector_t last_check;
3461         int skipped = 0;
3462
3463         /* just incase thread restarts... */
3464         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3465                 return;
3466
3467         /* we overload curr_resync somewhat here.
3468          * 0 == not engaged in resync at all
3469          * 2 == checking that there is no conflict with another sync
3470          * 1 == like 2, but have yielded to allow conflicting resync to
3471          *              commense
3472          * other == active in resync - this many blocks
3473          *
3474          * Before starting a resync we must have set curr_resync to
3475          * 2, and then checked that every "conflicting" array has curr_resync
3476          * less than ours.  When we find one that is the same or higher
3477          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3478          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3479          * This will mean we have to start checking from the beginning again.
3480          *
3481          */
3482
3483         do {
3484                 mddev->curr_resync = 2;
3485
3486         try_again:
3487                 if (signal_pending(current)) {
3488                         flush_signals(current);
3489                         goto skip;
3490                 }
3491                 ITERATE_MDDEV(mddev2,tmp) {
3492                         if (mddev2 == mddev)
3493                                 continue;
3494                         if (mddev2->curr_resync && 
3495                             match_mddev_units(mddev,mddev2)) {
3496                                 DEFINE_WAIT(wq);
3497                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3498                                         /* arbitrarily yield */
3499                                         mddev->curr_resync = 1;
3500                                         wake_up(&resync_wait);
3501                                 }
3502                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3503                                         /* no need to wait here, we can wait the next
3504                                          * time 'round when curr_resync == 2
3505                                          */
3506                                         continue;
3507                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3508                                 if (!signal_pending(current)
3509                                     && mddev2->curr_resync >= mddev->curr_resync) {
3510                                         printk(KERN_INFO "md: delaying resync of %s"
3511                                                " until %s has finished resync (they"
3512                                                " share one or more physical units)\n",
3513                                                mdname(mddev), mdname(mddev2));
3514                                         mddev_put(mddev2);
3515                                         schedule();
3516                                         finish_wait(&resync_wait, &wq);
3517                                         goto try_again;
3518                                 }
3519                                 finish_wait(&resync_wait, &wq);
3520                         }
3521                 }
3522         } while (mddev->curr_resync < 2);
3523
3524         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3525                 /* resync follows the size requested by the personality,
3526                  * which defaults to physical size, but can be virtual size
3527                  */
3528                 max_sectors = mddev->resync_max_sectors;
3529         else
3530                 /* recovery follows the physical size of devices */
3531                 max_sectors = mddev->size << 1;
3532
3533         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3534         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3535                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3536         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3537                "(but not more than %d KB/sec) for reconstruction.\n",
3538                sysctl_speed_limit_max);
3539
3540         is_mddev_idle(mddev); /* this also initializes IO event counters */
3541         /* we don't use the checkpoint if there's a bitmap */
3542         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3543                 j = mddev->recovery_cp;
3544         else
3545                 j = 0;
3546         io_sectors = 0;
3547         for (m = 0; m < SYNC_MARKS; m++) {
3548                 mark[m] = jiffies;
3549                 mark_cnt[m] = io_sectors;
3550         }
3551         last_mark = 0;
3552         mddev->resync_mark = mark[last_mark];
3553         mddev->resync_mark_cnt = mark_cnt[last_mark];
3554
3555         /*
3556          * Tune reconstruction:
3557          */
3558         window = 32*(PAGE_SIZE/512);
3559         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3560                 window/2,(unsigned long long) max_sectors/2);
3561
3562         atomic_set(&mddev->recovery_active, 0);
3563         init_waitqueue_head(&mddev->recovery_wait);
3564         last_check = 0;
3565
3566         if (j>2) {
3567                 printk(KERN_INFO 
3568                         "md: resuming recovery of %s from checkpoint.\n",
3569                         mdname(mddev));
3570                 mddev->curr_resync = j;
3571         }
3572
3573         while (j < max_sectors) {
3574                 sector_t sectors;
3575
3576                 skipped = 0;
3577                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3578                                             currspeed < sysctl_speed_limit_min);
3579                 if (sectors == 0) {
3580                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3581                         goto out;
3582                 }
3583
3584                 if (!skipped) { /* actual IO requested */
3585                         io_sectors += sectors;
3586                         atomic_add(sectors, &mddev->recovery_active);
3587                 }
3588
3589                 j += sectors;
3590                 if (j>1) mddev->curr_resync = j;
3591
3592
3593                 if (last_check + window > io_sectors || j == max_sectors)
3594                         continue;
3595
3596                 last_check = io_sectors;
3597
3598                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3599                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3600                         break;
3601
3602         repeat:
3603                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3604                         /* step marks */
3605                         int next = (last_mark+1) % SYNC_MARKS;
3606
3607                         mddev->resync_mark = mark[next];
3608                         mddev->resync_mark_cnt = mark_cnt[next];
3609                         mark[next] = jiffies;
3610                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3611                         last_mark = next;
3612                 }
3613
3614
3615                 if (signal_pending(current)) {
3616                         /*
3617                          * got a signal, exit.
3618                          */
3619                         printk(KERN_INFO 
3620                                 "md: md_do_sync() got signal ... exiting\n");
3621                         flush_signals(current);
3622                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3623                         goto out;
3624                 }
3625
3626                 /*
3627                  * this loop exits only if either when we are slower than
3628                  * the 'hard' speed limit, or the system was IO-idle for
3629                  * a jiffy.
3630                  * the system might be non-idle CPU-wise, but we only care
3631                  * about not overloading the IO subsystem. (things like an
3632                  * e2fsck being done on the RAID array should execute fast)
3633                  */
3634                 mddev->queue->unplug_fn(mddev->queue);
3635                 cond_resched();
3636
3637                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3638                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3639
3640                 if (currspeed > sysctl_speed_limit_min) {
3641                         if ((currspeed > sysctl_speed_limit_max) ||
3642                                         !is_mddev_idle(mddev)) {
3643                                 msleep_interruptible(250);
3644                                 goto repeat;
3645                         }
3646                 }
3647         }
3648         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3649         /*
3650          * this also signals 'finished resyncing' to md_stop
3651          */
3652  out:
3653         mddev->queue->unplug_fn(mddev->queue);
3654
3655         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3656
3657         /* tell personality that we are finished */
3658         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3659
3660         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3661             mddev->curr_resync > 2 &&
3662             mddev->curr_resync >= mddev->recovery_cp) {
3663                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3664                         printk(KERN_INFO 
3665                                 "md: checkpointing recovery of %s.\n",
3666                                 mdname(mddev));
3667                         mddev->recovery_cp = mddev->curr_resync;
3668                 } else
3669                         mddev->recovery_cp = MaxSector;
3670         }
3671
3672  skip:
3673         mddev->curr_resync = 0;
3674         wake_up(&resync_wait);
3675         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3676         md_wakeup_thread(mddev->thread);
3677 }
3678
3679
3680 /*
3681  * This routine is regularly called by all per-raid-array threads to
3682  * deal with generic issues like resync and super-block update.
3683  * Raid personalities that don't have a thread (linear/raid0) do not
3684  * need this as they never do any recovery or update the superblock.
3685  *
3686  * It does not do any resync itself, but rather "forks" off other threads
3687  * to do that as needed.
3688  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3689  * "->recovery" and create a thread at ->sync_thread.
3690  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3691  * and wakeups up this thread which will reap the thread and finish up.
3692  * This thread also removes any faulty devices (with nr_pending == 0).
3693  *
3694  * The overall approach is:
3695  *  1/ if the superblock needs updating, update it.
3696  *  2/ If a recovery thread is running, don't do anything else.
3697  *  3/ If recovery has finished, clean up, possibly marking spares active.
3698  *  4/ If there are any faulty devices, remove them.
3699  *  5/ If array is degraded, try to add spares devices
3700  *  6/ If array has spares or is not in-sync, start a resync thread.
3701  */
3702 void md_check_recovery(mddev_t *mddev)
3703 {
3704         mdk_rdev_t *rdev;
3705         struct list_head *rtmp;
3706
3707
3708         if (mddev->bitmap)
3709                 bitmap_daemon_work(mddev->bitmap);
3710
3711         if (mddev->ro)
3712                 return;
3713
3714         if (signal_pending(current)) {
3715                 if (mddev->pers->sync_request) {
3716                         printk(KERN_INFO "md: %s in immediate safe mode\n",
3717                                mdname(mddev));
3718                         mddev->safemode = 2;
3719                 }
3720                 flush_signals(current);
3721         }
3722
3723         if ( ! (
3724                 mddev->sb_dirty ||
3725                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3726                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3727                 (mddev->safemode == 1) ||
3728                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3729                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3730                 ))
3731                 return;
3732
3733         if (mddev_trylock(mddev)==0) {
3734                 int spares =0;
3735
3736                 spin_lock(&mddev->write_lock);
3737                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3738                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3739                         mddev->in_sync = 1;
3740                         mddev->sb_dirty = 1;
3741                 }
3742                 if (mddev->safemode == 1)
3743                         mddev->safemode = 0;
3744                 spin_unlock(&mddev->write_lock);
3745
3746                 if (mddev->sb_dirty)
3747                         md_update_sb(mddev);
3748
3749
3750                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3751                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3752                         /* resync/recovery still happening */
3753                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3754                         goto unlock;
3755                 }
3756                 if (mddev->sync_thread) {
3757                         /* resync has finished, collect result */
3758                         md_unregister_thread(mddev->sync_thread);
3759                         mddev->sync_thread = NULL;
3760                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3761                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3762                                 /* success...*/
3763                                 /* activate any spares */
3764                                 mddev->pers->spare_active(mddev);
3765                         }
3766                         md_update_sb(mddev);
3767
3768                         /* if array is no-longer degraded, then any saved_raid_disk
3769                          * information must be scrapped
3770                          */
3771                         if (!mddev->degraded)
3772                                 ITERATE_RDEV(mddev,rdev,rtmp)
3773                                         rdev->saved_raid_disk = -1;
3774
3775                         mddev->recovery = 0;
3776                         /* flag recovery needed just to double check */
3777                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3778                         goto unlock;
3779                 }
3780                 if (mddev->recovery)
3781                         /* probably just the RECOVERY_NEEDED flag */
3782                         mddev->recovery = 0;
3783
3784                 /* no recovery is running.
3785                  * remove any failed drives, then
3786                  * add spares if possible.
3787                  * Spare are also removed and re-added, to allow
3788                  * the personality to fail the re-add.
3789                  */
3790                 ITERATE_RDEV(mddev,rdev,rtmp)
3791                         if (rdev->raid_disk >= 0 &&
3792                             (rdev->faulty || ! rdev->in_sync) &&
3793                             atomic_read(&rdev->nr_pending)==0) {
3794                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3795                                         rdev->raid_disk = -1;
3796                         }
3797
3798                 if (mddev->degraded) {
3799                         ITERATE_RDEV(mddev,rdev,rtmp)
3800                                 if (rdev->raid_disk < 0
3801                                     && !rdev->faulty) {
3802                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3803                                                 spares++;
3804                                         else
3805                                                 break;
3806                                 }
3807                 }
3808
3809                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3810                         /* nothing we can do ... */
3811                         goto unlock;
3812                 }
3813                 if (mddev->pers->sync_request) {
3814                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3815                         if (!spares)
3816                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3817                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3818                                 /* We are adding a device or devices to an array
3819                                  * which has the bitmap stored on all devices.
3820                                  * So make sure all bitmap pages get written
3821                                  */
3822                                 bitmap_write_all(mddev->bitmap);
3823                         }
3824                         mddev->sync_thread = md_register_thread(md_do_sync,
3825                                                                 mddev,
3826                                                                 "%s_resync");
3827                         if (!mddev->sync_thread) {
3828                                 printk(KERN_ERR "%s: could not start resync"
3829                                         " thread...\n", 
3830                                         mdname(mddev));
3831                                 /* leave the spares where they are, it shouldn't hurt */
3832                                 mddev->recovery = 0;
3833                         } else {
3834                                 md_wakeup_thread(mddev->sync_thread);
3835                         }
3836                 }
3837         unlock:
3838                 mddev_unlock(mddev);
3839         }
3840 }
3841
3842 static int md_notify_reboot(struct notifier_block *this,
3843                             unsigned long code, void *x)
3844 {
3845         struct list_head *tmp;
3846         mddev_t *mddev;
3847
3848         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3849
3850                 printk(KERN_INFO "md: stopping all md devices.\n");
3851
3852                 ITERATE_MDDEV(mddev,tmp)
3853                         if (mddev_trylock(mddev)==0)
3854                                 do_md_stop (mddev, 1);
3855                 /*
3856                  * certain more exotic SCSI devices are known to be
3857                  * volatile wrt too early system reboots. While the
3858                  * right place to handle this issue is the given
3859                  * driver, we do want to have a safe RAID driver ...
3860                  */
3861                 mdelay(1000*1);
3862         }
3863         return NOTIFY_DONE;
3864 }
3865
3866 static struct notifier_block md_notifier = {
3867         .notifier_call  = md_notify_reboot,
3868         .next           = NULL,
3869         .priority       = INT_MAX, /* before any real devices */
3870 };
3871
3872 static void md_geninit(void)
3873 {
3874         struct proc_dir_entry *p;
3875
3876         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3877
3878         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3879         if (p)
3880                 p->proc_fops = &md_seq_fops;
3881 }
3882
3883 static int __init md_init(void)
3884 {
3885         int minor;
3886
3887         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3888                         " MD_SB_DISKS=%d\n",
3889                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3890                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3891         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3892                         BITMAP_MINOR);
3893
3894         if (register_blkdev(MAJOR_NR, "md"))
3895                 return -1;
3896         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3897                 unregister_blkdev(MAJOR_NR, "md");
3898                 return -1;
3899         }
3900         devfs_mk_dir("md");
3901         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3902                                 md_probe, NULL, NULL);
3903         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3904                             md_probe, NULL, NULL);
3905
3906         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3907                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3908                                 S_IFBLK|S_IRUSR|S_IWUSR,
3909                                 "md/%d", minor);
3910
3911         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3912                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3913                               S_IFBLK|S_IRUSR|S_IWUSR,
3914                               "md/mdp%d", minor);
3915
3916
3917         register_reboot_notifier(&md_notifier);
3918         raid_table_header = register_sysctl_table(raid_root_table, 1);
3919
3920         md_geninit();
3921         return (0);
3922 }
3923
3924
3925 #ifndef MODULE
3926
3927 /*
3928  * Searches all registered partitions for autorun RAID arrays
3929  * at boot time.
3930  */
3931 static dev_t detected_devices[128];
3932 static int dev_cnt;
3933
3934 void md_autodetect_dev(dev_t dev)
3935 {
3936         if (dev_cnt >= 0 && dev_cnt < 127)
3937                 detected_devices[dev_cnt++] = dev;
3938 }
3939
3940
3941 static void autostart_arrays(int part)
3942 {
3943         mdk_rdev_t *rdev;
3944         int i;
3945
3946         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3947
3948         for (i = 0; i < dev_cnt; i++) {
3949                 dev_t dev = detected_devices[i];
3950
3951                 rdev = md_import_device(dev,0, 0);
3952                 if (IS_ERR(rdev))
3953                         continue;
3954
3955                 if (rdev->faulty) {
3956                         MD_BUG();
3957                         continue;
3958                 }
3959                 list_add(&rdev->same_set, &pending_raid_disks);
3960         }
3961         dev_cnt = 0;
3962
3963         autorun_devices(part);
3964 }
3965
3966 #endif
3967
3968 static __exit void md_exit(void)
3969 {
3970         mddev_t *mddev;
3971         struct list_head *tmp;
3972         int i;
3973         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3974         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3975         for (i=0; i < MAX_MD_DEVS; i++)
3976                 devfs_remove("md/%d", i);
3977         for (i=0; i < MAX_MD_DEVS; i++)
3978                 devfs_remove("md/d%d", i);
3979
3980         devfs_remove("md");
3981
3982         unregister_blkdev(MAJOR_NR,"md");
3983         unregister_blkdev(mdp_major, "mdp");
3984         unregister_reboot_notifier(&md_notifier);
3985         unregister_sysctl_table(raid_table_header);
3986         remove_proc_entry("mdstat", NULL);
3987         ITERATE_MDDEV(mddev,tmp) {
3988                 struct gendisk *disk = mddev->gendisk;
3989                 if (!disk)
3990                         continue;
3991                 export_array(mddev);
3992                 del_gendisk(disk);
3993                 put_disk(disk);
3994                 mddev->gendisk = NULL;
3995                 mddev_put(mddev);
3996         }
3997 }
3998
3999 module_init(md_init)
4000 module_exit(md_exit)
4001
4002 EXPORT_SYMBOL(register_md_personality);
4003 EXPORT_SYMBOL(unregister_md_personality);
4004 EXPORT_SYMBOL(md_error);
4005 EXPORT_SYMBOL(md_done_sync);
4006 EXPORT_SYMBOL(md_write_start);
4007 EXPORT_SYMBOL(md_write_end);
4008 EXPORT_SYMBOL(md_register_thread);
4009 EXPORT_SYMBOL(md_unregister_thread);
4010 EXPORT_SYMBOL(md_wakeup_thread);
4011 EXPORT_SYMBOL(md_print_devices);
4012 EXPORT_SYMBOL(md_check_recovery);
4013 MODULE_LICENSE("GPL");
4014 MODULE_ALIAS("md");
4015 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);