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