Kobject: convert drivers/* from kobject_unregister() to kobject_put()
[linux-2.6.git] / 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/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
47
48 #include <linux/init.h>
49
50 #include <linux/file.h>
51
52 #ifdef CONFIG_KMOD
53 #include <linux/kmod.h>
54 #endif
55
56 #include <asm/unaligned.h>
57
58 #define MAJOR_NR MD_MAJOR
59 #define MD_DRIVER
60
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
63
64 #define DEBUG 0
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68 #ifndef MODULE
69 static void autostart_arrays (int part);
70 #endif
71
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
74
75 static void md_print_devices(void);
76
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
79 /*
80  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81  * is 1000 KB/sec, so the extra system load does not show up that much.
82  * Increase it if you want to have more _guaranteed_ speed. Note that
83  * the RAID driver will use the maximum available bandwidth if the IO
84  * subsystem is idle. There is also an 'absolute maximum' reconstruction
85  * speed limit - in case reconstruction slows down your system despite
86  * idle IO detection.
87  *
88  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89  * or /sys/block/mdX/md/sync_speed_{min,max}
90  */
91
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
95 {
96         return mddev->sync_speed_min ?
97                 mddev->sync_speed_min : sysctl_speed_limit_min;
98 }
99
100 static inline int speed_max(mddev_t *mddev)
101 {
102         return mddev->sync_speed_max ?
103                 mddev->sync_speed_max : sysctl_speed_limit_max;
104 }
105
106 static struct ctl_table_header *raid_table_header;
107
108 static ctl_table raid_table[] = {
109         {
110                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
111                 .procname       = "speed_limit_min",
112                 .data           = &sysctl_speed_limit_min,
113                 .maxlen         = sizeof(int),
114                 .mode           = S_IRUGO|S_IWUSR,
115                 .proc_handler   = &proc_dointvec,
116         },
117         {
118                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
119                 .procname       = "speed_limit_max",
120                 .data           = &sysctl_speed_limit_max,
121                 .maxlen         = sizeof(int),
122                 .mode           = S_IRUGO|S_IWUSR,
123                 .proc_handler   = &proc_dointvec,
124         },
125         { .ctl_name = 0 }
126 };
127
128 static ctl_table raid_dir_table[] = {
129         {
130                 .ctl_name       = DEV_RAID,
131                 .procname       = "raid",
132                 .maxlen         = 0,
133                 .mode           = S_IRUGO|S_IXUGO,
134                 .child          = raid_table,
135         },
136         { .ctl_name = 0 }
137 };
138
139 static ctl_table raid_root_table[] = {
140         {
141                 .ctl_name       = CTL_DEV,
142                 .procname       = "dev",
143                 .maxlen         = 0,
144                 .mode           = 0555,
145                 .child          = raid_dir_table,
146         },
147         { .ctl_name = 0 }
148 };
149
150 static struct block_device_operations md_fops;
151
152 static int start_readonly;
153
154 /*
155  * We have a system wide 'event count' that is incremented
156  * on any 'interesting' event, and readers of /proc/mdstat
157  * can use 'poll' or 'select' to find out when the event
158  * count increases.
159  *
160  * Events are:
161  *  start array, stop array, error, add device, remove device,
162  *  start build, activate spare
163  */
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
167 {
168         atomic_inc(&md_event_count);
169         wake_up(&md_event_waiters);
170         sysfs_notify(&mddev->kobj, NULL, "sync_action");
171 }
172 EXPORT_SYMBOL_GPL(md_new_event);
173
174 /* Alternate version that can be called from interrupts
175  * when calling sysfs_notify isn't needed.
176  */
177 static void md_new_event_inintr(mddev_t *mddev)
178 {
179         atomic_inc(&md_event_count);
180         wake_up(&md_event_waiters);
181 }
182
183 /*
184  * Enables to iterate over all existing md arrays
185  * all_mddevs_lock protects this list.
186  */
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191 /*
192  * iterates through all used mddevs in the system.
193  * We take care to grab the all_mddevs_lock whenever navigating
194  * the list, and to always hold a refcount when unlocked.
195  * Any code which breaks out of this loop while own
196  * a reference to the current mddev and must mddev_put it.
197  */
198 #define ITERATE_MDDEV(mddev,tmp)                                        \
199                                                                         \
200         for (({ spin_lock(&all_mddevs_lock);                            \
201                 tmp = all_mddevs.next;                                  \
202                 mddev = NULL;});                                        \
203              ({ if (tmp != &all_mddevs)                                 \
204                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205                 spin_unlock(&all_mddevs_lock);                          \
206                 if (mddev) mddev_put(mddev);                            \
207                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
208                 tmp != &all_mddevs;});                                  \
209              ({ spin_lock(&all_mddevs_lock);                            \
210                 tmp = tmp->next;})                                      \
211                 )
212
213
214 static int md_fail_request (struct request_queue *q, struct bio *bio)
215 {
216         bio_io_error(bio);
217         return 0;
218 }
219
220 static inline mddev_t *mddev_get(mddev_t *mddev)
221 {
222         atomic_inc(&mddev->active);
223         return mddev;
224 }
225
226 static void mddev_put(mddev_t *mddev)
227 {
228         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229                 return;
230         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231                 list_del(&mddev->all_mddevs);
232                 spin_unlock(&all_mddevs_lock);
233                 blk_cleanup_queue(mddev->queue);
234                 kobject_put(&mddev->kobj);
235         } else
236                 spin_unlock(&all_mddevs_lock);
237 }
238
239 static mddev_t * mddev_find(dev_t unit)
240 {
241         mddev_t *mddev, *new = NULL;
242
243  retry:
244         spin_lock(&all_mddevs_lock);
245         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246                 if (mddev->unit == unit) {
247                         mddev_get(mddev);
248                         spin_unlock(&all_mddevs_lock);
249                         kfree(new);
250                         return mddev;
251                 }
252
253         if (new) {
254                 list_add(&new->all_mddevs, &all_mddevs);
255                 spin_unlock(&all_mddevs_lock);
256                 return new;
257         }
258         spin_unlock(&all_mddevs_lock);
259
260         new = kzalloc(sizeof(*new), GFP_KERNEL);
261         if (!new)
262                 return NULL;
263
264         new->unit = unit;
265         if (MAJOR(unit) == MD_MAJOR)
266                 new->md_minor = MINOR(unit);
267         else
268                 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
270         mutex_init(&new->reconfig_mutex);
271         INIT_LIST_HEAD(&new->disks);
272         INIT_LIST_HEAD(&new->all_mddevs);
273         init_timer(&new->safemode_timer);
274         atomic_set(&new->active, 1);
275         spin_lock_init(&new->write_lock);
276         init_waitqueue_head(&new->sb_wait);
277         new->reshape_position = MaxSector;
278
279         new->queue = blk_alloc_queue(GFP_KERNEL);
280         if (!new->queue) {
281                 kfree(new);
282                 return NULL;
283         }
284         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
285
286         blk_queue_make_request(new->queue, md_fail_request);
287
288         goto retry;
289 }
290
291 static inline int mddev_lock(mddev_t * mddev)
292 {
293         return mutex_lock_interruptible(&mddev->reconfig_mutex);
294 }
295
296 static inline int mddev_trylock(mddev_t * mddev)
297 {
298         return mutex_trylock(&mddev->reconfig_mutex);
299 }
300
301 static inline void mddev_unlock(mddev_t * mddev)
302 {
303         mutex_unlock(&mddev->reconfig_mutex);
304
305         md_wakeup_thread(mddev->thread);
306 }
307
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
309 {
310         mdk_rdev_t * rdev;
311         struct list_head *tmp;
312
313         ITERATE_RDEV(mddev,rdev,tmp) {
314                 if (rdev->desc_nr == nr)
315                         return rdev;
316         }
317         return NULL;
318 }
319
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
321 {
322         struct list_head *tmp;
323         mdk_rdev_t *rdev;
324
325         ITERATE_RDEV(mddev,rdev,tmp) {
326                 if (rdev->bdev->bd_dev == dev)
327                         return rdev;
328         }
329         return NULL;
330 }
331
332 static struct mdk_personality *find_pers(int level, char *clevel)
333 {
334         struct mdk_personality *pers;
335         list_for_each_entry(pers, &pers_list, list) {
336                 if (level != LEVEL_NONE && pers->level == level)
337                         return pers;
338                 if (strcmp(pers->name, clevel)==0)
339                         return pers;
340         }
341         return NULL;
342 }
343
344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
345 {
346         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
347         return MD_NEW_SIZE_BLOCKS(size);
348 }
349
350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
351 {
352         sector_t size;
353
354         size = rdev->sb_offset;
355
356         if (chunk_size)
357                 size &= ~((sector_t)chunk_size/1024 - 1);
358         return size;
359 }
360
361 static int alloc_disk_sb(mdk_rdev_t * rdev)
362 {
363         if (rdev->sb_page)
364                 MD_BUG();
365
366         rdev->sb_page = alloc_page(GFP_KERNEL);
367         if (!rdev->sb_page) {
368                 printk(KERN_ALERT "md: out of memory.\n");
369                 return -EINVAL;
370         }
371
372         return 0;
373 }
374
375 static void free_disk_sb(mdk_rdev_t * rdev)
376 {
377         if (rdev->sb_page) {
378                 put_page(rdev->sb_page);
379                 rdev->sb_loaded = 0;
380                 rdev->sb_page = NULL;
381                 rdev->sb_offset = 0;
382                 rdev->size = 0;
383         }
384 }
385
386
387 static void super_written(struct bio *bio, int error)
388 {
389         mdk_rdev_t *rdev = bio->bi_private;
390         mddev_t *mddev = rdev->mddev;
391
392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
393                 printk("md: super_written gets error=%d, uptodate=%d\n",
394                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
395                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
396                 md_error(mddev, rdev);
397         }
398
399         if (atomic_dec_and_test(&mddev->pending_writes))
400                 wake_up(&mddev->sb_wait);
401         bio_put(bio);
402 }
403
404 static void super_written_barrier(struct bio *bio, int error)
405 {
406         struct bio *bio2 = bio->bi_private;
407         mdk_rdev_t *rdev = bio2->bi_private;
408         mddev_t *mddev = rdev->mddev;
409
410         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
411             error == -EOPNOTSUPP) {
412                 unsigned long flags;
413                 /* barriers don't appear to be supported :-( */
414                 set_bit(BarriersNotsupp, &rdev->flags);
415                 mddev->barriers_work = 0;
416                 spin_lock_irqsave(&mddev->write_lock, flags);
417                 bio2->bi_next = mddev->biolist;
418                 mddev->biolist = bio2;
419                 spin_unlock_irqrestore(&mddev->write_lock, flags);
420                 wake_up(&mddev->sb_wait);
421                 bio_put(bio);
422         } else {
423                 bio_put(bio2);
424                 bio->bi_private = rdev;
425                 super_written(bio, error);
426         }
427 }
428
429 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
430                    sector_t sector, int size, struct page *page)
431 {
432         /* write first size bytes of page to sector of rdev
433          * Increment mddev->pending_writes before returning
434          * and decrement it on completion, waking up sb_wait
435          * if zero is reached.
436          * If an error occurred, call md_error
437          *
438          * As we might need to resubmit the request if BIO_RW_BARRIER
439          * causes ENOTSUPP, we allocate a spare bio...
440          */
441         struct bio *bio = bio_alloc(GFP_NOIO, 1);
442         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
443
444         bio->bi_bdev = rdev->bdev;
445         bio->bi_sector = sector;
446         bio_add_page(bio, page, size, 0);
447         bio->bi_private = rdev;
448         bio->bi_end_io = super_written;
449         bio->bi_rw = rw;
450
451         atomic_inc(&mddev->pending_writes);
452         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
453                 struct bio *rbio;
454                 rw |= (1<<BIO_RW_BARRIER);
455                 rbio = bio_clone(bio, GFP_NOIO);
456                 rbio->bi_private = bio;
457                 rbio->bi_end_io = super_written_barrier;
458                 submit_bio(rw, rbio);
459         } else
460                 submit_bio(rw, bio);
461 }
462
463 void md_super_wait(mddev_t *mddev)
464 {
465         /* wait for all superblock writes that were scheduled to complete.
466          * if any had to be retried (due to BARRIER problems), retry them
467          */
468         DEFINE_WAIT(wq);
469         for(;;) {
470                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
471                 if (atomic_read(&mddev->pending_writes)==0)
472                         break;
473                 while (mddev->biolist) {
474                         struct bio *bio;
475                         spin_lock_irq(&mddev->write_lock);
476                         bio = mddev->biolist;
477                         mddev->biolist = bio->bi_next ;
478                         bio->bi_next = NULL;
479                         spin_unlock_irq(&mddev->write_lock);
480                         submit_bio(bio->bi_rw, bio);
481                 }
482                 schedule();
483         }
484         finish_wait(&mddev->sb_wait, &wq);
485 }
486
487 static void bi_complete(struct bio *bio, int error)
488 {
489         complete((struct completion*)bio->bi_private);
490 }
491
492 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
493                    struct page *page, int rw)
494 {
495         struct bio *bio = bio_alloc(GFP_NOIO, 1);
496         struct completion event;
497         int ret;
498
499         rw |= (1 << BIO_RW_SYNC);
500
501         bio->bi_bdev = bdev;
502         bio->bi_sector = sector;
503         bio_add_page(bio, page, size, 0);
504         init_completion(&event);
505         bio->bi_private = &event;
506         bio->bi_end_io = bi_complete;
507         submit_bio(rw, bio);
508         wait_for_completion(&event);
509
510         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
511         bio_put(bio);
512         return ret;
513 }
514 EXPORT_SYMBOL_GPL(sync_page_io);
515
516 static int read_disk_sb(mdk_rdev_t * rdev, int size)
517 {
518         char b[BDEVNAME_SIZE];
519         if (!rdev->sb_page) {
520                 MD_BUG();
521                 return -EINVAL;
522         }
523         if (rdev->sb_loaded)
524                 return 0;
525
526
527         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
528                 goto fail;
529         rdev->sb_loaded = 1;
530         return 0;
531
532 fail:
533         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
534                 bdevname(rdev->bdev,b));
535         return -EINVAL;
536 }
537
538 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
539 {
540         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
541                 (sb1->set_uuid1 == sb2->set_uuid1) &&
542                 (sb1->set_uuid2 == sb2->set_uuid2) &&
543                 (sb1->set_uuid3 == sb2->set_uuid3))
544
545                 return 1;
546
547         return 0;
548 }
549
550
551 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
552 {
553         int ret;
554         mdp_super_t *tmp1, *tmp2;
555
556         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
557         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
558
559         if (!tmp1 || !tmp2) {
560                 ret = 0;
561                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
562                 goto abort;
563         }
564
565         *tmp1 = *sb1;
566         *tmp2 = *sb2;
567
568         /*
569          * nr_disks is not constant
570          */
571         tmp1->nr_disks = 0;
572         tmp2->nr_disks = 0;
573
574         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
575                 ret = 0;
576         else
577                 ret = 1;
578
579 abort:
580         kfree(tmp1);
581         kfree(tmp2);
582         return ret;
583 }
584
585
586 static u32 md_csum_fold(u32 csum)
587 {
588         csum = (csum & 0xffff) + (csum >> 16);
589         return (csum & 0xffff) + (csum >> 16);
590 }
591
592 static unsigned int calc_sb_csum(mdp_super_t * sb)
593 {
594         u64 newcsum = 0;
595         u32 *sb32 = (u32*)sb;
596         int i;
597         unsigned int disk_csum, csum;
598
599         disk_csum = sb->sb_csum;
600         sb->sb_csum = 0;
601
602         for (i = 0; i < MD_SB_BYTES/4 ; i++)
603                 newcsum += sb32[i];
604         csum = (newcsum & 0xffffffff) + (newcsum>>32);
605
606
607 #ifdef CONFIG_ALPHA
608         /* This used to use csum_partial, which was wrong for several
609          * reasons including that different results are returned on
610          * different architectures.  It isn't critical that we get exactly
611          * the same return value as before (we always csum_fold before
612          * testing, and that removes any differences).  However as we
613          * know that csum_partial always returned a 16bit value on
614          * alphas, do a fold to maximise conformity to previous behaviour.
615          */
616         sb->sb_csum = md_csum_fold(disk_csum);
617 #else
618         sb->sb_csum = disk_csum;
619 #endif
620         return csum;
621 }
622
623
624 /*
625  * Handle superblock details.
626  * We want to be able to handle multiple superblock formats
627  * so we have a common interface to them all, and an array of
628  * different handlers.
629  * We rely on user-space to write the initial superblock, and support
630  * reading and updating of superblocks.
631  * Interface methods are:
632  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
633  *      loads and validates a superblock on dev.
634  *      if refdev != NULL, compare superblocks on both devices
635  *    Return:
636  *      0 - dev has a superblock that is compatible with refdev
637  *      1 - dev has a superblock that is compatible and newer than refdev
638  *          so dev should be used as the refdev in future
639  *     -EINVAL superblock incompatible or invalid
640  *     -othererror e.g. -EIO
641  *
642  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
643  *      Verify that dev is acceptable into mddev.
644  *       The first time, mddev->raid_disks will be 0, and data from
645  *       dev should be merged in.  Subsequent calls check that dev
646  *       is new enough.  Return 0 or -EINVAL
647  *
648  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
649  *     Update the superblock for rdev with data in mddev
650  *     This does not write to disc.
651  *
652  */
653
654 struct super_type  {
655         char            *name;
656         struct module   *owner;
657         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
658         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
659         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
660 };
661
662 /*
663  * load_super for 0.90.0 
664  */
665 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
666 {
667         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
668         mdp_super_t *sb;
669         int ret;
670         sector_t sb_offset;
671
672         /*
673          * Calculate the position of the superblock,
674          * it's at the end of the disk.
675          *
676          * It also happens to be a multiple of 4Kb.
677          */
678         sb_offset = calc_dev_sboffset(rdev->bdev);
679         rdev->sb_offset = sb_offset;
680
681         ret = read_disk_sb(rdev, MD_SB_BYTES);
682         if (ret) return ret;
683
684         ret = -EINVAL;
685
686         bdevname(rdev->bdev, b);
687         sb = (mdp_super_t*)page_address(rdev->sb_page);
688
689         if (sb->md_magic != MD_SB_MAGIC) {
690                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
691                        b);
692                 goto abort;
693         }
694
695         if (sb->major_version != 0 ||
696             sb->minor_version < 90 ||
697             sb->minor_version > 91) {
698                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
699                         sb->major_version, sb->minor_version,
700                         b);
701                 goto abort;
702         }
703
704         if (sb->raid_disks <= 0)
705                 goto abort;
706
707         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
708                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
709                         b);
710                 goto abort;
711         }
712
713         rdev->preferred_minor = sb->md_minor;
714         rdev->data_offset = 0;
715         rdev->sb_size = MD_SB_BYTES;
716
717         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
718                 if (sb->level != 1 && sb->level != 4
719                     && sb->level != 5 && sb->level != 6
720                     && sb->level != 10) {
721                         /* FIXME use a better test */
722                         printk(KERN_WARNING
723                                "md: bitmaps not supported for this level.\n");
724                         goto abort;
725                 }
726         }
727
728         if (sb->level == LEVEL_MULTIPATH)
729                 rdev->desc_nr = -1;
730         else
731                 rdev->desc_nr = sb->this_disk.number;
732
733         if (refdev == 0)
734                 ret = 1;
735         else {
736                 __u64 ev1, ev2;
737                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
738                 if (!uuid_equal(refsb, sb)) {
739                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
740                                 b, bdevname(refdev->bdev,b2));
741                         goto abort;
742                 }
743                 if (!sb_equal(refsb, sb)) {
744                         printk(KERN_WARNING "md: %s has same UUID"
745                                " but different superblock to %s\n",
746                                b, bdevname(refdev->bdev, b2));
747                         goto abort;
748                 }
749                 ev1 = md_event(sb);
750                 ev2 = md_event(refsb);
751                 if (ev1 > ev2)
752                         ret = 1;
753                 else 
754                         ret = 0;
755         }
756         rdev->size = calc_dev_size(rdev, sb->chunk_size);
757
758         if (rdev->size < sb->size && sb->level > 1)
759                 /* "this cannot possibly happen" ... */
760                 ret = -EINVAL;
761
762  abort:
763         return ret;
764 }
765
766 /*
767  * validate_super for 0.90.0
768  */
769 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
770 {
771         mdp_disk_t *desc;
772         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
773         __u64 ev1 = md_event(sb);
774
775         rdev->raid_disk = -1;
776         rdev->flags = 0;
777         if (mddev->raid_disks == 0) {
778                 mddev->major_version = 0;
779                 mddev->minor_version = sb->minor_version;
780                 mddev->patch_version = sb->patch_version;
781                 mddev->persistent = ! sb->not_persistent;
782                 mddev->chunk_size = sb->chunk_size;
783                 mddev->ctime = sb->ctime;
784                 mddev->utime = sb->utime;
785                 mddev->level = sb->level;
786                 mddev->clevel[0] = 0;
787                 mddev->layout = sb->layout;
788                 mddev->raid_disks = sb->raid_disks;
789                 mddev->size = sb->size;
790                 mddev->events = ev1;
791                 mddev->bitmap_offset = 0;
792                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
793
794                 if (mddev->minor_version >= 91) {
795                         mddev->reshape_position = sb->reshape_position;
796                         mddev->delta_disks = sb->delta_disks;
797                         mddev->new_level = sb->new_level;
798                         mddev->new_layout = sb->new_layout;
799                         mddev->new_chunk = sb->new_chunk;
800                 } else {
801                         mddev->reshape_position = MaxSector;
802                         mddev->delta_disks = 0;
803                         mddev->new_level = mddev->level;
804                         mddev->new_layout = mddev->layout;
805                         mddev->new_chunk = mddev->chunk_size;
806                 }
807
808                 if (sb->state & (1<<MD_SB_CLEAN))
809                         mddev->recovery_cp = MaxSector;
810                 else {
811                         if (sb->events_hi == sb->cp_events_hi && 
812                                 sb->events_lo == sb->cp_events_lo) {
813                                 mddev->recovery_cp = sb->recovery_cp;
814                         } else
815                                 mddev->recovery_cp = 0;
816                 }
817
818                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
819                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
820                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
821                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
822
823                 mddev->max_disks = MD_SB_DISKS;
824
825                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
826                     mddev->bitmap_file == NULL)
827                         mddev->bitmap_offset = mddev->default_bitmap_offset;
828
829         } else if (mddev->pers == NULL) {
830                 /* Insist on good event counter while assembling */
831                 ++ev1;
832                 if (ev1 < mddev->events) 
833                         return -EINVAL;
834         } else if (mddev->bitmap) {
835                 /* if adding to array with a bitmap, then we can accept an
836                  * older device ... but not too old.
837                  */
838                 if (ev1 < mddev->bitmap->events_cleared)
839                         return 0;
840         } else {
841                 if (ev1 < mddev->events)
842                         /* just a hot-add of a new device, leave raid_disk at -1 */
843                         return 0;
844         }
845
846         if (mddev->level != LEVEL_MULTIPATH) {
847                 desc = sb->disks + rdev->desc_nr;
848
849                 if (desc->state & (1<<MD_DISK_FAULTY))
850                         set_bit(Faulty, &rdev->flags);
851                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
852                             desc->raid_disk < mddev->raid_disks */) {
853                         set_bit(In_sync, &rdev->flags);
854                         rdev->raid_disk = desc->raid_disk;
855                 }
856                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
857                         set_bit(WriteMostly, &rdev->flags);
858         } else /* MULTIPATH are always insync */
859                 set_bit(In_sync, &rdev->flags);
860         return 0;
861 }
862
863 /*
864  * sync_super for 0.90.0
865  */
866 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
867 {
868         mdp_super_t *sb;
869         struct list_head *tmp;
870         mdk_rdev_t *rdev2;
871         int next_spare = mddev->raid_disks;
872
873
874         /* make rdev->sb match mddev data..
875          *
876          * 1/ zero out disks
877          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
878          * 3/ any empty disks < next_spare become removed
879          *
880          * disks[0] gets initialised to REMOVED because
881          * we cannot be sure from other fields if it has
882          * been initialised or not.
883          */
884         int i;
885         int active=0, working=0,failed=0,spare=0,nr_disks=0;
886
887         rdev->sb_size = MD_SB_BYTES;
888
889         sb = (mdp_super_t*)page_address(rdev->sb_page);
890
891         memset(sb, 0, sizeof(*sb));
892
893         sb->md_magic = MD_SB_MAGIC;
894         sb->major_version = mddev->major_version;
895         sb->patch_version = mddev->patch_version;
896         sb->gvalid_words  = 0; /* ignored */
897         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
898         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
899         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
900         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
901
902         sb->ctime = mddev->ctime;
903         sb->level = mddev->level;
904         sb->size  = mddev->size;
905         sb->raid_disks = mddev->raid_disks;
906         sb->md_minor = mddev->md_minor;
907         sb->not_persistent = !mddev->persistent;
908         sb->utime = mddev->utime;
909         sb->state = 0;
910         sb->events_hi = (mddev->events>>32);
911         sb->events_lo = (u32)mddev->events;
912
913         if (mddev->reshape_position == MaxSector)
914                 sb->minor_version = 90;
915         else {
916                 sb->minor_version = 91;
917                 sb->reshape_position = mddev->reshape_position;
918                 sb->new_level = mddev->new_level;
919                 sb->delta_disks = mddev->delta_disks;
920                 sb->new_layout = mddev->new_layout;
921                 sb->new_chunk = mddev->new_chunk;
922         }
923         mddev->minor_version = sb->minor_version;
924         if (mddev->in_sync)
925         {
926                 sb->recovery_cp = mddev->recovery_cp;
927                 sb->cp_events_hi = (mddev->events>>32);
928                 sb->cp_events_lo = (u32)mddev->events;
929                 if (mddev->recovery_cp == MaxSector)
930                         sb->state = (1<< MD_SB_CLEAN);
931         } else
932                 sb->recovery_cp = 0;
933
934         sb->layout = mddev->layout;
935         sb->chunk_size = mddev->chunk_size;
936
937         if (mddev->bitmap && mddev->bitmap_file == NULL)
938                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
939
940         sb->disks[0].state = (1<<MD_DISK_REMOVED);
941         ITERATE_RDEV(mddev,rdev2,tmp) {
942                 mdp_disk_t *d;
943                 int desc_nr;
944                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
945                     && !test_bit(Faulty, &rdev2->flags))
946                         desc_nr = rdev2->raid_disk;
947                 else
948                         desc_nr = next_spare++;
949                 rdev2->desc_nr = desc_nr;
950                 d = &sb->disks[rdev2->desc_nr];
951                 nr_disks++;
952                 d->number = rdev2->desc_nr;
953                 d->major = MAJOR(rdev2->bdev->bd_dev);
954                 d->minor = MINOR(rdev2->bdev->bd_dev);
955                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
956                     && !test_bit(Faulty, &rdev2->flags))
957                         d->raid_disk = rdev2->raid_disk;
958                 else
959                         d->raid_disk = rdev2->desc_nr; /* compatibility */
960                 if (test_bit(Faulty, &rdev2->flags))
961                         d->state = (1<<MD_DISK_FAULTY);
962                 else if (test_bit(In_sync, &rdev2->flags)) {
963                         d->state = (1<<MD_DISK_ACTIVE);
964                         d->state |= (1<<MD_DISK_SYNC);
965                         active++;
966                         working++;
967                 } else {
968                         d->state = 0;
969                         spare++;
970                         working++;
971                 }
972                 if (test_bit(WriteMostly, &rdev2->flags))
973                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
974         }
975         /* now set the "removed" and "faulty" bits on any missing devices */
976         for (i=0 ; i < mddev->raid_disks ; i++) {
977                 mdp_disk_t *d = &sb->disks[i];
978                 if (d->state == 0 && d->number == 0) {
979                         d->number = i;
980                         d->raid_disk = i;
981                         d->state = (1<<MD_DISK_REMOVED);
982                         d->state |= (1<<MD_DISK_FAULTY);
983                         failed++;
984                 }
985         }
986         sb->nr_disks = nr_disks;
987         sb->active_disks = active;
988         sb->working_disks = working;
989         sb->failed_disks = failed;
990         sb->spare_disks = spare;
991
992         sb->this_disk = sb->disks[rdev->desc_nr];
993         sb->sb_csum = calc_sb_csum(sb);
994 }
995
996 /*
997  * version 1 superblock
998  */
999
1000 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1001 {
1002         __le32 disk_csum;
1003         u32 csum;
1004         unsigned long long newcsum;
1005         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1006         __le32 *isuper = (__le32*)sb;
1007         int i;
1008
1009         disk_csum = sb->sb_csum;
1010         sb->sb_csum = 0;
1011         newcsum = 0;
1012         for (i=0; size>=4; size -= 4 )
1013                 newcsum += le32_to_cpu(*isuper++);
1014
1015         if (size == 2)
1016                 newcsum += le16_to_cpu(*(__le16*) isuper);
1017
1018         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1019         sb->sb_csum = disk_csum;
1020         return cpu_to_le32(csum);
1021 }
1022
1023 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1024 {
1025         struct mdp_superblock_1 *sb;
1026         int ret;
1027         sector_t sb_offset;
1028         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1029         int bmask;
1030
1031         /*
1032          * Calculate the position of the superblock.
1033          * It is always aligned to a 4K boundary and
1034          * depeding on minor_version, it can be:
1035          * 0: At least 8K, but less than 12K, from end of device
1036          * 1: At start of device
1037          * 2: 4K from start of device.
1038          */
1039         switch(minor_version) {
1040         case 0:
1041                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1042                 sb_offset -= 8*2;
1043                 sb_offset &= ~(sector_t)(4*2-1);
1044                 /* convert from sectors to K */
1045                 sb_offset /= 2;
1046                 break;
1047         case 1:
1048                 sb_offset = 0;
1049                 break;
1050         case 2:
1051                 sb_offset = 4;
1052                 break;
1053         default:
1054                 return -EINVAL;
1055         }
1056         rdev->sb_offset = sb_offset;
1057
1058         /* superblock is rarely larger than 1K, but it can be larger,
1059          * and it is safe to read 4k, so we do that
1060          */
1061         ret = read_disk_sb(rdev, 4096);
1062         if (ret) return ret;
1063
1064
1065         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1066
1067         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1068             sb->major_version != cpu_to_le32(1) ||
1069             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1070             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1071             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1072                 return -EINVAL;
1073
1074         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1075                 printk("md: invalid superblock checksum on %s\n",
1076                         bdevname(rdev->bdev,b));
1077                 return -EINVAL;
1078         }
1079         if (le64_to_cpu(sb->data_size) < 10) {
1080                 printk("md: data_size too small on %s\n",
1081                        bdevname(rdev->bdev,b));
1082                 return -EINVAL;
1083         }
1084         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1085                 if (sb->level != cpu_to_le32(1) &&
1086                     sb->level != cpu_to_le32(4) &&
1087                     sb->level != cpu_to_le32(5) &&
1088                     sb->level != cpu_to_le32(6) &&
1089                     sb->level != cpu_to_le32(10)) {
1090                         printk(KERN_WARNING
1091                                "md: bitmaps not supported for this level.\n");
1092                         return -EINVAL;
1093                 }
1094         }
1095
1096         rdev->preferred_minor = 0xffff;
1097         rdev->data_offset = le64_to_cpu(sb->data_offset);
1098         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1099
1100         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1101         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1102         if (rdev->sb_size & bmask)
1103                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1104
1105         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1106                 rdev->desc_nr = -1;
1107         else
1108                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1109
1110         if (refdev == 0)
1111                 ret = 1;
1112         else {
1113                 __u64 ev1, ev2;
1114                 struct mdp_superblock_1 *refsb = 
1115                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1116
1117                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1118                     sb->level != refsb->level ||
1119                     sb->layout != refsb->layout ||
1120                     sb->chunksize != refsb->chunksize) {
1121                         printk(KERN_WARNING "md: %s has strangely different"
1122                                 " superblock to %s\n",
1123                                 bdevname(rdev->bdev,b),
1124                                 bdevname(refdev->bdev,b2));
1125                         return -EINVAL;
1126                 }
1127                 ev1 = le64_to_cpu(sb->events);
1128                 ev2 = le64_to_cpu(refsb->events);
1129
1130                 if (ev1 > ev2)
1131                         ret = 1;
1132                 else
1133                         ret = 0;
1134         }
1135         if (minor_version) 
1136                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1137         else
1138                 rdev->size = rdev->sb_offset;
1139         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1140                 return -EINVAL;
1141         rdev->size = le64_to_cpu(sb->data_size)/2;
1142         if (le32_to_cpu(sb->chunksize))
1143                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1144
1145         if (le64_to_cpu(sb->size) > rdev->size*2)
1146                 return -EINVAL;
1147         return ret;
1148 }
1149
1150 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1151 {
1152         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1153         __u64 ev1 = le64_to_cpu(sb->events);
1154
1155         rdev->raid_disk = -1;
1156         rdev->flags = 0;
1157         if (mddev->raid_disks == 0) {
1158                 mddev->major_version = 1;
1159                 mddev->patch_version = 0;
1160                 mddev->persistent = 1;
1161                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1162                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1163                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1164                 mddev->level = le32_to_cpu(sb->level);
1165                 mddev->clevel[0] = 0;
1166                 mddev->layout = le32_to_cpu(sb->layout);
1167                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1168                 mddev->size = le64_to_cpu(sb->size)/2;
1169                 mddev->events = ev1;
1170                 mddev->bitmap_offset = 0;
1171                 mddev->default_bitmap_offset = 1024 >> 9;
1172                 
1173                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1174                 memcpy(mddev->uuid, sb->set_uuid, 16);
1175
1176                 mddev->max_disks =  (4096-256)/2;
1177
1178                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1179                     mddev->bitmap_file == NULL )
1180                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1181
1182                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1183                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1184                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1185                         mddev->new_level = le32_to_cpu(sb->new_level);
1186                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1187                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1188                 } else {
1189                         mddev->reshape_position = MaxSector;
1190                         mddev->delta_disks = 0;
1191                         mddev->new_level = mddev->level;
1192                         mddev->new_layout = mddev->layout;
1193                         mddev->new_chunk = mddev->chunk_size;
1194                 }
1195
1196         } else if (mddev->pers == NULL) {
1197                 /* Insist of good event counter while assembling */
1198                 ++ev1;
1199                 if (ev1 < mddev->events)
1200                         return -EINVAL;
1201         } else if (mddev->bitmap) {
1202                 /* If adding to array with a bitmap, then we can accept an
1203                  * older device, but not too old.
1204                  */
1205                 if (ev1 < mddev->bitmap->events_cleared)
1206                         return 0;
1207         } else {
1208                 if (ev1 < mddev->events)
1209                         /* just a hot-add of a new device, leave raid_disk at -1 */
1210                         return 0;
1211         }
1212         if (mddev->level != LEVEL_MULTIPATH) {
1213                 int role;
1214                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1215                 switch(role) {
1216                 case 0xffff: /* spare */
1217                         break;
1218                 case 0xfffe: /* faulty */
1219                         set_bit(Faulty, &rdev->flags);
1220                         break;
1221                 default:
1222                         if ((le32_to_cpu(sb->feature_map) &
1223                              MD_FEATURE_RECOVERY_OFFSET))
1224                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1225                         else
1226                                 set_bit(In_sync, &rdev->flags);
1227                         rdev->raid_disk = role;
1228                         break;
1229                 }
1230                 if (sb->devflags & WriteMostly1)
1231                         set_bit(WriteMostly, &rdev->flags);
1232         } else /* MULTIPATH are always insync */
1233                 set_bit(In_sync, &rdev->flags);
1234
1235         return 0;
1236 }
1237
1238 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1239 {
1240         struct mdp_superblock_1 *sb;
1241         struct list_head *tmp;
1242         mdk_rdev_t *rdev2;
1243         int max_dev, i;
1244         /* make rdev->sb match mddev and rdev data. */
1245
1246         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1247
1248         sb->feature_map = 0;
1249         sb->pad0 = 0;
1250         sb->recovery_offset = cpu_to_le64(0);
1251         memset(sb->pad1, 0, sizeof(sb->pad1));
1252         memset(sb->pad2, 0, sizeof(sb->pad2));
1253         memset(sb->pad3, 0, sizeof(sb->pad3));
1254
1255         sb->utime = cpu_to_le64((__u64)mddev->utime);
1256         sb->events = cpu_to_le64(mddev->events);
1257         if (mddev->in_sync)
1258                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1259         else
1260                 sb->resync_offset = cpu_to_le64(0);
1261
1262         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1263
1264         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1265         sb->size = cpu_to_le64(mddev->size<<1);
1266
1267         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1268                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1269                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1270         }
1271
1272         if (rdev->raid_disk >= 0 &&
1273             !test_bit(In_sync, &rdev->flags) &&
1274             rdev->recovery_offset > 0) {
1275                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1276                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1277         }
1278
1279         if (mddev->reshape_position != MaxSector) {
1280                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1281                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1282                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1283                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1284                 sb->new_level = cpu_to_le32(mddev->new_level);
1285                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1286         }
1287
1288         max_dev = 0;
1289         ITERATE_RDEV(mddev,rdev2,tmp)
1290                 if (rdev2->desc_nr+1 > max_dev)
1291                         max_dev = rdev2->desc_nr+1;
1292
1293         if (max_dev > le32_to_cpu(sb->max_dev))
1294                 sb->max_dev = cpu_to_le32(max_dev);
1295         for (i=0; i<max_dev;i++)
1296                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1297         
1298         ITERATE_RDEV(mddev,rdev2,tmp) {
1299                 i = rdev2->desc_nr;
1300                 if (test_bit(Faulty, &rdev2->flags))
1301                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1302                 else if (test_bit(In_sync, &rdev2->flags))
1303                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1304                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1305                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1306                 else
1307                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1308         }
1309
1310         sb->sb_csum = calc_sb_1_csum(sb);
1311 }
1312
1313
1314 static struct super_type super_types[] = {
1315         [0] = {
1316                 .name   = "0.90.0",
1317                 .owner  = THIS_MODULE,
1318                 .load_super     = super_90_load,
1319                 .validate_super = super_90_validate,
1320                 .sync_super     = super_90_sync,
1321         },
1322         [1] = {
1323                 .name   = "md-1",
1324                 .owner  = THIS_MODULE,
1325                 .load_super     = super_1_load,
1326                 .validate_super = super_1_validate,
1327                 .sync_super     = super_1_sync,
1328         },
1329 };
1330
1331 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1332 {
1333         struct list_head *tmp, *tmp2;
1334         mdk_rdev_t *rdev, *rdev2;
1335
1336         ITERATE_RDEV(mddev1,rdev,tmp)
1337                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1338                         if (rdev->bdev->bd_contains ==
1339                             rdev2->bdev->bd_contains)
1340                                 return 1;
1341
1342         return 0;
1343 }
1344
1345 static LIST_HEAD(pending_raid_disks);
1346
1347 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1348 {
1349         char b[BDEVNAME_SIZE];
1350         struct kobject *ko;
1351         char *s;
1352         int err;
1353
1354         if (rdev->mddev) {
1355                 MD_BUG();
1356                 return -EINVAL;
1357         }
1358         /* make sure rdev->size exceeds mddev->size */
1359         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1360                 if (mddev->pers) {
1361                         /* Cannot change size, so fail
1362                          * If mddev->level <= 0, then we don't care
1363                          * about aligning sizes (e.g. linear)
1364                          */
1365                         if (mddev->level > 0)
1366                                 return -ENOSPC;
1367                 } else
1368                         mddev->size = rdev->size;
1369         }
1370
1371         /* Verify rdev->desc_nr is unique.
1372          * If it is -1, assign a free number, else
1373          * check number is not in use
1374          */
1375         if (rdev->desc_nr < 0) {
1376                 int choice = 0;
1377                 if (mddev->pers) choice = mddev->raid_disks;
1378                 while (find_rdev_nr(mddev, choice))
1379                         choice++;
1380                 rdev->desc_nr = choice;
1381         } else {
1382                 if (find_rdev_nr(mddev, rdev->desc_nr))
1383                         return -EBUSY;
1384         }
1385         bdevname(rdev->bdev,b);
1386         while ( (s=strchr(b, '/')) != NULL)
1387                 *s = '!';
1388
1389         rdev->mddev = mddev;
1390         printk(KERN_INFO "md: bind<%s>\n", b);
1391
1392         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1393                 goto fail;
1394
1395         if (rdev->bdev->bd_part)
1396                 ko = &rdev->bdev->bd_part->dev.kobj;
1397         else
1398                 ko = &rdev->bdev->bd_disk->dev.kobj;
1399         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1400                 kobject_del(&rdev->kobj);
1401                 goto fail;
1402         }
1403         list_add(&rdev->same_set, &mddev->disks);
1404         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1405         return 0;
1406
1407  fail:
1408         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1409                b, mdname(mddev));
1410         return err;
1411 }
1412
1413 static void delayed_delete(struct work_struct *ws)
1414 {
1415         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1416         kobject_del(&rdev->kobj);
1417 }
1418
1419 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1420 {
1421         char b[BDEVNAME_SIZE];
1422         if (!rdev->mddev) {
1423                 MD_BUG();
1424                 return;
1425         }
1426         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1427         list_del_init(&rdev->same_set);
1428         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1429         rdev->mddev = NULL;
1430         sysfs_remove_link(&rdev->kobj, "block");
1431
1432         /* We need to delay this, otherwise we can deadlock when
1433          * writing to 'remove' to "dev/state"
1434          */
1435         INIT_WORK(&rdev->del_work, delayed_delete);
1436         schedule_work(&rdev->del_work);
1437 }
1438
1439 /*
1440  * prevent the device from being mounted, repartitioned or
1441  * otherwise reused by a RAID array (or any other kernel
1442  * subsystem), by bd_claiming the device.
1443  */
1444 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1445 {
1446         int err = 0;
1447         struct block_device *bdev;
1448         char b[BDEVNAME_SIZE];
1449
1450         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1451         if (IS_ERR(bdev)) {
1452                 printk(KERN_ERR "md: could not open %s.\n",
1453                         __bdevname(dev, b));
1454                 return PTR_ERR(bdev);
1455         }
1456         err = bd_claim(bdev, rdev);
1457         if (err) {
1458                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1459                         bdevname(bdev, b));
1460                 blkdev_put(bdev);
1461                 return err;
1462         }
1463         rdev->bdev = bdev;
1464         return err;
1465 }
1466
1467 static void unlock_rdev(mdk_rdev_t *rdev)
1468 {
1469         struct block_device *bdev = rdev->bdev;
1470         rdev->bdev = NULL;
1471         if (!bdev)
1472                 MD_BUG();
1473         bd_release(bdev);
1474         blkdev_put(bdev);
1475 }
1476
1477 void md_autodetect_dev(dev_t dev);
1478
1479 static void export_rdev(mdk_rdev_t * rdev)
1480 {
1481         char b[BDEVNAME_SIZE];
1482         printk(KERN_INFO "md: export_rdev(%s)\n",
1483                 bdevname(rdev->bdev,b));
1484         if (rdev->mddev)
1485                 MD_BUG();
1486         free_disk_sb(rdev);
1487         list_del_init(&rdev->same_set);
1488 #ifndef MODULE
1489         md_autodetect_dev(rdev->bdev->bd_dev);
1490 #endif
1491         unlock_rdev(rdev);
1492         kobject_put(&rdev->kobj);
1493 }
1494
1495 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1496 {
1497         unbind_rdev_from_array(rdev);
1498         export_rdev(rdev);
1499 }
1500
1501 static void export_array(mddev_t *mddev)
1502 {
1503         struct list_head *tmp;
1504         mdk_rdev_t *rdev;
1505
1506         ITERATE_RDEV(mddev,rdev,tmp) {
1507                 if (!rdev->mddev) {
1508                         MD_BUG();
1509                         continue;
1510                 }
1511                 kick_rdev_from_array(rdev);
1512         }
1513         if (!list_empty(&mddev->disks))
1514                 MD_BUG();
1515         mddev->raid_disks = 0;
1516         mddev->major_version = 0;
1517 }
1518
1519 static void print_desc(mdp_disk_t *desc)
1520 {
1521         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1522                 desc->major,desc->minor,desc->raid_disk,desc->state);
1523 }
1524
1525 static void print_sb(mdp_super_t *sb)
1526 {
1527         int i;
1528
1529         printk(KERN_INFO 
1530                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1531                 sb->major_version, sb->minor_version, sb->patch_version,
1532                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1533                 sb->ctime);
1534         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1535                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1536                 sb->md_minor, sb->layout, sb->chunk_size);
1537         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1538                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1539                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1540                 sb->failed_disks, sb->spare_disks,
1541                 sb->sb_csum, (unsigned long)sb->events_lo);
1542
1543         printk(KERN_INFO);
1544         for (i = 0; i < MD_SB_DISKS; i++) {
1545                 mdp_disk_t *desc;
1546
1547                 desc = sb->disks + i;
1548                 if (desc->number || desc->major || desc->minor ||
1549                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1550                         printk("     D %2d: ", i);
1551                         print_desc(desc);
1552                 }
1553         }
1554         printk(KERN_INFO "md:     THIS: ");
1555         print_desc(&sb->this_disk);
1556
1557 }
1558
1559 static void print_rdev(mdk_rdev_t *rdev)
1560 {
1561         char b[BDEVNAME_SIZE];
1562         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1563                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1564                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1565                 rdev->desc_nr);
1566         if (rdev->sb_loaded) {
1567                 printk(KERN_INFO "md: rdev superblock:\n");
1568                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1569         } else
1570                 printk(KERN_INFO "md: no rdev superblock!\n");
1571 }
1572
1573 static void md_print_devices(void)
1574 {
1575         struct list_head *tmp, *tmp2;
1576         mdk_rdev_t *rdev;
1577         mddev_t *mddev;
1578         char b[BDEVNAME_SIZE];
1579
1580         printk("\n");
1581         printk("md:     **********************************\n");
1582         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1583         printk("md:     **********************************\n");
1584         ITERATE_MDDEV(mddev,tmp) {
1585
1586                 if (mddev->bitmap)
1587                         bitmap_print_sb(mddev->bitmap);
1588                 else
1589                         printk("%s: ", mdname(mddev));
1590                 ITERATE_RDEV(mddev,rdev,tmp2)
1591                         printk("<%s>", bdevname(rdev->bdev,b));
1592                 printk("\n");
1593
1594                 ITERATE_RDEV(mddev,rdev,tmp2)
1595                         print_rdev(rdev);
1596         }
1597         printk("md:     **********************************\n");
1598         printk("\n");
1599 }
1600
1601
1602 static void sync_sbs(mddev_t * mddev, int nospares)
1603 {
1604         /* Update each superblock (in-memory image), but
1605          * if we are allowed to, skip spares which already
1606          * have the right event counter, or have one earlier
1607          * (which would mean they aren't being marked as dirty
1608          * with the rest of the array)
1609          */
1610         mdk_rdev_t *rdev;
1611         struct list_head *tmp;
1612
1613         ITERATE_RDEV(mddev,rdev,tmp) {
1614                 if (rdev->sb_events == mddev->events ||
1615                     (nospares &&
1616                      rdev->raid_disk < 0 &&
1617                      (rdev->sb_events&1)==0 &&
1618                      rdev->sb_events+1 == mddev->events)) {
1619                         /* Don't update this superblock */
1620                         rdev->sb_loaded = 2;
1621                 } else {
1622                         super_types[mddev->major_version].
1623                                 sync_super(mddev, rdev);
1624                         rdev->sb_loaded = 1;
1625                 }
1626         }
1627 }
1628
1629 static void md_update_sb(mddev_t * mddev, int force_change)
1630 {
1631         struct list_head *tmp;
1632         mdk_rdev_t *rdev;
1633         int sync_req;
1634         int nospares = 0;
1635
1636 repeat:
1637         spin_lock_irq(&mddev->write_lock);
1638
1639         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1640         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1641                 force_change = 1;
1642         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1643                 /* just a clean<-> dirty transition, possibly leave spares alone,
1644                  * though if events isn't the right even/odd, we will have to do
1645                  * spares after all
1646                  */
1647                 nospares = 1;
1648         if (force_change)
1649                 nospares = 0;
1650         if (mddev->degraded)
1651                 /* If the array is degraded, then skipping spares is both
1652                  * dangerous and fairly pointless.
1653                  * Dangerous because a device that was removed from the array
1654                  * might have a event_count that still looks up-to-date,
1655                  * so it can be re-added without a resync.
1656                  * Pointless because if there are any spares to skip,
1657                  * then a recovery will happen and soon that array won't
1658                  * be degraded any more and the spare can go back to sleep then.
1659                  */
1660                 nospares = 0;
1661
1662         sync_req = mddev->in_sync;
1663         mddev->utime = get_seconds();
1664
1665         /* If this is just a dirty<->clean transition, and the array is clean
1666          * and 'events' is odd, we can roll back to the previous clean state */
1667         if (nospares
1668             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1669             && (mddev->events & 1)
1670             && mddev->events != 1)
1671                 mddev->events--;
1672         else {
1673                 /* otherwise we have to go forward and ... */
1674                 mddev->events ++;
1675                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1676                         /* .. if the array isn't clean, insist on an odd 'events' */
1677                         if ((mddev->events&1)==0) {
1678                                 mddev->events++;
1679                                 nospares = 0;
1680                         }
1681                 } else {
1682                         /* otherwise insist on an even 'events' (for clean states) */
1683                         if ((mddev->events&1)) {
1684                                 mddev->events++;
1685                                 nospares = 0;
1686                         }
1687                 }
1688         }
1689
1690         if (!mddev->events) {
1691                 /*
1692                  * oops, this 64-bit counter should never wrap.
1693                  * Either we are in around ~1 trillion A.C., assuming
1694                  * 1 reboot per second, or we have a bug:
1695                  */
1696                 MD_BUG();
1697                 mddev->events --;
1698         }
1699         sync_sbs(mddev, nospares);
1700
1701         /*
1702          * do not write anything to disk if using
1703          * nonpersistent superblocks
1704          */
1705         if (!mddev->persistent) {
1706                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1707                 spin_unlock_irq(&mddev->write_lock);
1708                 wake_up(&mddev->sb_wait);
1709                 return;
1710         }
1711         spin_unlock_irq(&mddev->write_lock);
1712
1713         dprintk(KERN_INFO 
1714                 "md: updating %s RAID superblock on device (in sync %d)\n",
1715                 mdname(mddev),mddev->in_sync);
1716
1717         bitmap_update_sb(mddev->bitmap);
1718         ITERATE_RDEV(mddev,rdev,tmp) {
1719                 char b[BDEVNAME_SIZE];
1720                 dprintk(KERN_INFO "md: ");
1721                 if (rdev->sb_loaded != 1)
1722                         continue; /* no noise on spare devices */
1723                 if (test_bit(Faulty, &rdev->flags))
1724                         dprintk("(skipping faulty ");
1725
1726                 dprintk("%s ", bdevname(rdev->bdev,b));
1727                 if (!test_bit(Faulty, &rdev->flags)) {
1728                         md_super_write(mddev,rdev,
1729                                        rdev->sb_offset<<1, rdev->sb_size,
1730                                        rdev->sb_page);
1731                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1732                                 bdevname(rdev->bdev,b),
1733                                 (unsigned long long)rdev->sb_offset);
1734                         rdev->sb_events = mddev->events;
1735
1736                 } else
1737                         dprintk(")\n");
1738                 if (mddev->level == LEVEL_MULTIPATH)
1739                         /* only need to write one superblock... */
1740                         break;
1741         }
1742         md_super_wait(mddev);
1743         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1744
1745         spin_lock_irq(&mddev->write_lock);
1746         if (mddev->in_sync != sync_req ||
1747             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1748                 /* have to write it out again */
1749                 spin_unlock_irq(&mddev->write_lock);
1750                 goto repeat;
1751         }
1752         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1753         spin_unlock_irq(&mddev->write_lock);
1754         wake_up(&mddev->sb_wait);
1755
1756 }
1757
1758 /* words written to sysfs files may, or my not, be \n terminated.
1759  * We want to accept with case. For this we use cmd_match.
1760  */
1761 static int cmd_match(const char *cmd, const char *str)
1762 {
1763         /* See if cmd, written into a sysfs file, matches
1764          * str.  They must either be the same, or cmd can
1765          * have a trailing newline
1766          */
1767         while (*cmd && *str && *cmd == *str) {
1768                 cmd++;
1769                 str++;
1770         }
1771         if (*cmd == '\n')
1772                 cmd++;
1773         if (*str || *cmd)
1774                 return 0;
1775         return 1;
1776 }
1777
1778 struct rdev_sysfs_entry {
1779         struct attribute attr;
1780         ssize_t (*show)(mdk_rdev_t *, char *);
1781         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1782 };
1783
1784 static ssize_t
1785 state_show(mdk_rdev_t *rdev, char *page)
1786 {
1787         char *sep = "";
1788         int len=0;
1789
1790         if (test_bit(Faulty, &rdev->flags)) {
1791                 len+= sprintf(page+len, "%sfaulty",sep);
1792                 sep = ",";
1793         }
1794         if (test_bit(In_sync, &rdev->flags)) {
1795                 len += sprintf(page+len, "%sin_sync",sep);
1796                 sep = ",";
1797         }
1798         if (test_bit(WriteMostly, &rdev->flags)) {
1799                 len += sprintf(page+len, "%swrite_mostly",sep);
1800                 sep = ",";
1801         }
1802         if (!test_bit(Faulty, &rdev->flags) &&
1803             !test_bit(In_sync, &rdev->flags)) {
1804                 len += sprintf(page+len, "%sspare", sep);
1805                 sep = ",";
1806         }
1807         return len+sprintf(page+len, "\n");
1808 }
1809
1810 static ssize_t
1811 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1812 {
1813         /* can write
1814          *  faulty  - simulates and error
1815          *  remove  - disconnects the device
1816          *  writemostly - sets write_mostly
1817          *  -writemostly - clears write_mostly
1818          */
1819         int err = -EINVAL;
1820         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1821                 md_error(rdev->mddev, rdev);
1822                 err = 0;
1823         } else if (cmd_match(buf, "remove")) {
1824                 if (rdev->raid_disk >= 0)
1825                         err = -EBUSY;
1826                 else {
1827                         mddev_t *mddev = rdev->mddev;
1828                         kick_rdev_from_array(rdev);
1829                         if (mddev->pers)
1830                                 md_update_sb(mddev, 1);
1831                         md_new_event(mddev);
1832                         err = 0;
1833                 }
1834         } else if (cmd_match(buf, "writemostly")) {
1835                 set_bit(WriteMostly, &rdev->flags);
1836                 err = 0;
1837         } else if (cmd_match(buf, "-writemostly")) {
1838                 clear_bit(WriteMostly, &rdev->flags);
1839                 err = 0;
1840         }
1841         return err ? err : len;
1842 }
1843 static struct rdev_sysfs_entry rdev_state =
1844 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1845
1846 static ssize_t
1847 super_show(mdk_rdev_t *rdev, char *page)
1848 {
1849         if (rdev->sb_loaded && rdev->sb_size) {
1850                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1851                 return rdev->sb_size;
1852         } else
1853                 return 0;
1854 }
1855 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1856
1857 static ssize_t
1858 errors_show(mdk_rdev_t *rdev, char *page)
1859 {
1860         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1861 }
1862
1863 static ssize_t
1864 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1865 {
1866         char *e;
1867         unsigned long n = simple_strtoul(buf, &e, 10);
1868         if (*buf && (*e == 0 || *e == '\n')) {
1869                 atomic_set(&rdev->corrected_errors, n);
1870                 return len;
1871         }
1872         return -EINVAL;
1873 }
1874 static struct rdev_sysfs_entry rdev_errors =
1875 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1876
1877 static ssize_t
1878 slot_show(mdk_rdev_t *rdev, char *page)
1879 {
1880         if (rdev->raid_disk < 0)
1881                 return sprintf(page, "none\n");
1882         else
1883                 return sprintf(page, "%d\n", rdev->raid_disk);
1884 }
1885
1886 static ssize_t
1887 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1888 {
1889         char *e;
1890         int slot = simple_strtoul(buf, &e, 10);
1891         if (strncmp(buf, "none", 4)==0)
1892                 slot = -1;
1893         else if (e==buf || (*e && *e!= '\n'))
1894                 return -EINVAL;
1895         if (rdev->mddev->pers)
1896                 /* Cannot set slot in active array (yet) */
1897                 return -EBUSY;
1898         if (slot >= rdev->mddev->raid_disks)
1899                 return -ENOSPC;
1900         rdev->raid_disk = slot;
1901         /* assume it is working */
1902         rdev->flags = 0;
1903         set_bit(In_sync, &rdev->flags);
1904         return len;
1905 }
1906
1907
1908 static struct rdev_sysfs_entry rdev_slot =
1909 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1910
1911 static ssize_t
1912 offset_show(mdk_rdev_t *rdev, char *page)
1913 {
1914         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1915 }
1916
1917 static ssize_t
1918 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1919 {
1920         char *e;
1921         unsigned long long offset = simple_strtoull(buf, &e, 10);
1922         if (e==buf || (*e && *e != '\n'))
1923                 return -EINVAL;
1924         if (rdev->mddev->pers)
1925                 return -EBUSY;
1926         rdev->data_offset = offset;
1927         return len;
1928 }
1929
1930 static struct rdev_sysfs_entry rdev_offset =
1931 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1932
1933 static ssize_t
1934 rdev_size_show(mdk_rdev_t *rdev, char *page)
1935 {
1936         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1937 }
1938
1939 static ssize_t
1940 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1941 {
1942         char *e;
1943         unsigned long long size = simple_strtoull(buf, &e, 10);
1944         if (e==buf || (*e && *e != '\n'))
1945                 return -EINVAL;
1946         if (rdev->mddev->pers)
1947                 return -EBUSY;
1948         rdev->size = size;
1949         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1950                 rdev->mddev->size = size;
1951         return len;
1952 }
1953
1954 static struct rdev_sysfs_entry rdev_size =
1955 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1956
1957 static struct attribute *rdev_default_attrs[] = {
1958         &rdev_state.attr,
1959         &rdev_super.attr,
1960         &rdev_errors.attr,
1961         &rdev_slot.attr,
1962         &rdev_offset.attr,
1963         &rdev_size.attr,
1964         NULL,
1965 };
1966 static ssize_t
1967 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1968 {
1969         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1970         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1971
1972         if (!entry->show)
1973                 return -EIO;
1974         return entry->show(rdev, page);
1975 }
1976
1977 static ssize_t
1978 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1979               const char *page, size_t length)
1980 {
1981         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1982         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1983
1984         if (!entry->store)
1985                 return -EIO;
1986         if (!capable(CAP_SYS_ADMIN))
1987                 return -EACCES;
1988         return entry->store(rdev, page, length);
1989 }
1990
1991 static void rdev_free(struct kobject *ko)
1992 {
1993         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1994         kfree(rdev);
1995 }
1996 static struct sysfs_ops rdev_sysfs_ops = {
1997         .show           = rdev_attr_show,
1998         .store          = rdev_attr_store,
1999 };
2000 static struct kobj_type rdev_ktype = {
2001         .release        = rdev_free,
2002         .sysfs_ops      = &rdev_sysfs_ops,
2003         .default_attrs  = rdev_default_attrs,
2004 };
2005
2006 /*
2007  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2008  *
2009  * mark the device faulty if:
2010  *
2011  *   - the device is nonexistent (zero size)
2012  *   - the device has no valid superblock
2013  *
2014  * a faulty rdev _never_ has rdev->sb set.
2015  */
2016 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2017 {
2018         char b[BDEVNAME_SIZE];
2019         int err;
2020         mdk_rdev_t *rdev;
2021         sector_t size;
2022
2023         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2024         if (!rdev) {
2025                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2026                 return ERR_PTR(-ENOMEM);
2027         }
2028
2029         if ((err = alloc_disk_sb(rdev)))
2030                 goto abort_free;
2031
2032         err = lock_rdev(rdev, newdev);
2033         if (err)
2034                 goto abort_free;
2035
2036         kobject_init(&rdev->kobj, &rdev_ktype);
2037
2038         rdev->desc_nr = -1;
2039         rdev->saved_raid_disk = -1;
2040         rdev->raid_disk = -1;
2041         rdev->flags = 0;
2042         rdev->data_offset = 0;
2043         rdev->sb_events = 0;
2044         atomic_set(&rdev->nr_pending, 0);
2045         atomic_set(&rdev->read_errors, 0);
2046         atomic_set(&rdev->corrected_errors, 0);
2047
2048         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2049         if (!size) {
2050                 printk(KERN_WARNING 
2051                         "md: %s has zero or unknown size, marking faulty!\n",
2052                         bdevname(rdev->bdev,b));
2053                 err = -EINVAL;
2054                 goto abort_free;
2055         }
2056
2057         if (super_format >= 0) {
2058                 err = super_types[super_format].
2059                         load_super(rdev, NULL, super_minor);
2060                 if (err == -EINVAL) {
2061                         printk(KERN_WARNING
2062                                 "md: %s does not have a valid v%d.%d "
2063                                "superblock, not importing!\n",
2064                                 bdevname(rdev->bdev,b),
2065                                super_format, super_minor);
2066                         goto abort_free;
2067                 }
2068                 if (err < 0) {
2069                         printk(KERN_WARNING 
2070                                 "md: could not read %s's sb, not importing!\n",
2071                                 bdevname(rdev->bdev,b));
2072                         goto abort_free;
2073                 }
2074         }
2075         INIT_LIST_HEAD(&rdev->same_set);
2076
2077         return rdev;
2078
2079 abort_free:
2080         if (rdev->sb_page) {
2081                 if (rdev->bdev)
2082                         unlock_rdev(rdev);
2083                 free_disk_sb(rdev);
2084         }
2085         kfree(rdev);
2086         return ERR_PTR(err);
2087 }
2088
2089 /*
2090  * Check a full RAID array for plausibility
2091  */
2092
2093
2094 static void analyze_sbs(mddev_t * mddev)
2095 {
2096         int i;
2097         struct list_head *tmp;
2098         mdk_rdev_t *rdev, *freshest;
2099         char b[BDEVNAME_SIZE];
2100
2101         freshest = NULL;
2102         ITERATE_RDEV(mddev,rdev,tmp)
2103                 switch (super_types[mddev->major_version].
2104                         load_super(rdev, freshest, mddev->minor_version)) {
2105                 case 1:
2106                         freshest = rdev;
2107                         break;
2108                 case 0:
2109                         break;
2110                 default:
2111                         printk( KERN_ERR \
2112                                 "md: fatal superblock inconsistency in %s"
2113                                 " -- removing from array\n", 
2114                                 bdevname(rdev->bdev,b));
2115                         kick_rdev_from_array(rdev);
2116                 }
2117
2118
2119         super_types[mddev->major_version].
2120                 validate_super(mddev, freshest);
2121
2122         i = 0;
2123         ITERATE_RDEV(mddev,rdev,tmp) {
2124                 if (rdev != freshest)
2125                         if (super_types[mddev->major_version].
2126                             validate_super(mddev, rdev)) {
2127                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2128                                         " from array!\n",
2129                                         bdevname(rdev->bdev,b));
2130                                 kick_rdev_from_array(rdev);
2131                                 continue;
2132                         }
2133                 if (mddev->level == LEVEL_MULTIPATH) {
2134                         rdev->desc_nr = i++;
2135                         rdev->raid_disk = rdev->desc_nr;
2136                         set_bit(In_sync, &rdev->flags);
2137                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2138                         rdev->raid_disk = -1;
2139                         clear_bit(In_sync, &rdev->flags);
2140                 }
2141         }
2142
2143
2144
2145         if (mddev->recovery_cp != MaxSector &&
2146             mddev->level >= 1)
2147                 printk(KERN_ERR "md: %s: raid array is not clean"
2148                        " -- starting background reconstruction\n",
2149                        mdname(mddev));
2150
2151 }
2152
2153 static ssize_t
2154 safe_delay_show(mddev_t *mddev, char *page)
2155 {
2156         int msec = (mddev->safemode_delay*1000)/HZ;
2157         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2158 }
2159 static ssize_t
2160 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2161 {
2162         int scale=1;
2163         int dot=0;
2164         int i;
2165         unsigned long msec;
2166         char buf[30];
2167         char *e;
2168         /* remove a period, and count digits after it */
2169         if (len >= sizeof(buf))
2170                 return -EINVAL;
2171         strlcpy(buf, cbuf, len);
2172         buf[len] = 0;
2173         for (i=0; i<len; i++) {
2174                 if (dot) {
2175                         if (isdigit(buf[i])) {
2176                                 buf[i-1] = buf[i];
2177                                 scale *= 10;
2178                         }
2179                         buf[i] = 0;
2180                 } else if (buf[i] == '.') {
2181                         dot=1;
2182                         buf[i] = 0;
2183                 }
2184         }
2185         msec = simple_strtoul(buf, &e, 10);
2186         if (e == buf || (*e && *e != '\n'))
2187                 return -EINVAL;
2188         msec = (msec * 1000) / scale;
2189         if (msec == 0)
2190                 mddev->safemode_delay = 0;
2191         else {
2192                 mddev->safemode_delay = (msec*HZ)/1000;
2193                 if (mddev->safemode_delay == 0)
2194                         mddev->safemode_delay = 1;
2195         }
2196         return len;
2197 }
2198 static struct md_sysfs_entry md_safe_delay =
2199 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2200
2201 static ssize_t
2202 level_show(mddev_t *mddev, char *page)
2203 {
2204         struct mdk_personality *p = mddev->pers;
2205         if (p)
2206                 return sprintf(page, "%s\n", p->name);
2207         else if (mddev->clevel[0])
2208                 return sprintf(page, "%s\n", mddev->clevel);
2209         else if (mddev->level != LEVEL_NONE)
2210                 return sprintf(page, "%d\n", mddev->level);
2211         else
2212                 return 0;
2213 }
2214
2215 static ssize_t
2216 level_store(mddev_t *mddev, const char *buf, size_t len)
2217 {
2218         int rv = len;
2219         if (mddev->pers)
2220                 return -EBUSY;
2221         if (len == 0)
2222                 return 0;
2223         if (len >= sizeof(mddev->clevel))
2224                 return -ENOSPC;
2225         strncpy(mddev->clevel, buf, len);
2226         if (mddev->clevel[len-1] == '\n')
2227                 len--;
2228         mddev->clevel[len] = 0;
2229         mddev->level = LEVEL_NONE;
2230         return rv;
2231 }
2232
2233 static struct md_sysfs_entry md_level =
2234 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2235
2236
2237 static ssize_t
2238 layout_show(mddev_t *mddev, char *page)
2239 {
2240         /* just a number, not meaningful for all levels */
2241         if (mddev->reshape_position != MaxSector &&
2242             mddev->layout != mddev->new_layout)
2243                 return sprintf(page, "%d (%d)\n",
2244                                mddev->new_layout, mddev->layout);
2245         return sprintf(page, "%d\n", mddev->layout);
2246 }
2247
2248 static ssize_t
2249 layout_store(mddev_t *mddev, const char *buf, size_t len)
2250 {
2251         char *e;
2252         unsigned long n = simple_strtoul(buf, &e, 10);
2253
2254         if (!*buf || (*e && *e != '\n'))
2255                 return -EINVAL;
2256
2257         if (mddev->pers)
2258                 return -EBUSY;
2259         if (mddev->reshape_position != MaxSector)
2260                 mddev->new_layout = n;
2261         else
2262                 mddev->layout = n;
2263         return len;
2264 }
2265 static struct md_sysfs_entry md_layout =
2266 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2267
2268
2269 static ssize_t
2270 raid_disks_show(mddev_t *mddev, char *page)
2271 {
2272         if (mddev->raid_disks == 0)
2273                 return 0;
2274         if (mddev->reshape_position != MaxSector &&
2275             mddev->delta_disks != 0)
2276                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2277                                mddev->raid_disks - mddev->delta_disks);
2278         return sprintf(page, "%d\n", mddev->raid_disks);
2279 }
2280
2281 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2282
2283 static ssize_t
2284 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2285 {
2286         char *e;
2287         int rv = 0;
2288         unsigned long n = simple_strtoul(buf, &e, 10);
2289
2290         if (!*buf || (*e && *e != '\n'))
2291                 return -EINVAL;
2292
2293         if (mddev->pers)
2294                 rv = update_raid_disks(mddev, n);
2295         else if (mddev->reshape_position != MaxSector) {
2296                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2297                 mddev->delta_disks = n - olddisks;
2298                 mddev->raid_disks = n;
2299         } else
2300                 mddev->raid_disks = n;
2301         return rv ? rv : len;
2302 }
2303 static struct md_sysfs_entry md_raid_disks =
2304 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2305
2306 static ssize_t
2307 chunk_size_show(mddev_t *mddev, char *page)
2308 {
2309         if (mddev->reshape_position != MaxSector &&
2310             mddev->chunk_size != mddev->new_chunk)
2311                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2312                                mddev->chunk_size);
2313         return sprintf(page, "%d\n", mddev->chunk_size);
2314 }
2315
2316 static ssize_t
2317 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2318 {
2319         /* can only set chunk_size if array is not yet active */
2320         char *e;
2321         unsigned long n = simple_strtoul(buf, &e, 10);
2322
2323         if (!*buf || (*e && *e != '\n'))
2324                 return -EINVAL;
2325
2326         if (mddev->pers)
2327                 return -EBUSY;
2328         else if (mddev->reshape_position != MaxSector)
2329                 mddev->new_chunk = n;
2330         else
2331                 mddev->chunk_size = n;
2332         return len;
2333 }
2334 static struct md_sysfs_entry md_chunk_size =
2335 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2336
2337 static ssize_t
2338 resync_start_show(mddev_t *mddev, char *page)
2339 {
2340         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2341 }
2342
2343 static ssize_t
2344 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2345 {
2346         /* can only set chunk_size if array is not yet active */
2347         char *e;
2348         unsigned long long n = simple_strtoull(buf, &e, 10);
2349
2350         if (mddev->pers)
2351                 return -EBUSY;
2352         if (!*buf || (*e && *e != '\n'))
2353                 return -EINVAL;
2354
2355         mddev->recovery_cp = n;
2356         return len;
2357 }
2358 static struct md_sysfs_entry md_resync_start =
2359 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2360
2361 /*
2362  * The array state can be:
2363  *
2364  * clear
2365  *     No devices, no size, no level
2366  *     Equivalent to STOP_ARRAY ioctl
2367  * inactive
2368  *     May have some settings, but array is not active
2369  *        all IO results in error
2370  *     When written, doesn't tear down array, but just stops it
2371  * suspended (not supported yet)
2372  *     All IO requests will block. The array can be reconfigured.
2373  *     Writing this, if accepted, will block until array is quiessent
2374  * readonly
2375  *     no resync can happen.  no superblocks get written.
2376  *     write requests fail
2377  * read-auto
2378  *     like readonly, but behaves like 'clean' on a write request.
2379  *
2380  * clean - no pending writes, but otherwise active.
2381  *     When written to inactive array, starts without resync
2382  *     If a write request arrives then
2383  *       if metadata is known, mark 'dirty' and switch to 'active'.
2384  *       if not known, block and switch to write-pending
2385  *     If written to an active array that has pending writes, then fails.
2386  * active
2387  *     fully active: IO and resync can be happening.
2388  *     When written to inactive array, starts with resync
2389  *
2390  * write-pending
2391  *     clean, but writes are blocked waiting for 'active' to be written.
2392  *
2393  * active-idle
2394  *     like active, but no writes have been seen for a while (100msec).
2395  *
2396  */
2397 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2398                    write_pending, active_idle, bad_word};
2399 static char *array_states[] = {
2400         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2401         "write-pending", "active-idle", NULL };
2402
2403 static int match_word(const char *word, char **list)
2404 {
2405         int n;
2406         for (n=0; list[n]; n++)
2407                 if (cmd_match(word, list[n]))
2408                         break;
2409         return n;
2410 }
2411
2412 static ssize_t
2413 array_state_show(mddev_t *mddev, char *page)
2414 {
2415         enum array_state st = inactive;
2416
2417         if (mddev->pers)
2418                 switch(mddev->ro) {
2419                 case 1:
2420                         st = readonly;
2421                         break;
2422                 case 2:
2423                         st = read_auto;
2424                         break;
2425                 case 0:
2426                         if (mddev->in_sync)
2427                                 st = clean;
2428                         else if (mddev->safemode)
2429                                 st = active_idle;
2430                         else
2431                                 st = active;
2432                 }
2433         else {
2434                 if (list_empty(&mddev->disks) &&
2435                     mddev->raid_disks == 0 &&
2436                     mddev->size == 0)
2437                         st = clear;
2438                 else
2439                         st = inactive;
2440         }
2441         return sprintf(page, "%s\n", array_states[st]);
2442 }
2443
2444 static int do_md_stop(mddev_t * mddev, int ro);
2445 static int do_md_run(mddev_t * mddev);
2446 static int restart_array(mddev_t *mddev);
2447
2448 static ssize_t
2449 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2450 {
2451         int err = -EINVAL;
2452         enum array_state st = match_word(buf, array_states);
2453         switch(st) {
2454         case bad_word:
2455                 break;
2456         case clear:
2457                 /* stopping an active array */
2458                 if (mddev->pers) {
2459                         if (atomic_read(&mddev->active) > 1)
2460                                 return -EBUSY;
2461                         err = do_md_stop(mddev, 0);
2462                 }
2463                 break;
2464         case inactive:
2465                 /* stopping an active array */
2466                 if (mddev->pers) {
2467                         if (atomic_read(&mddev->active) > 1)
2468                                 return -EBUSY;
2469                         err = do_md_stop(mddev, 2);
2470                 }
2471                 break;
2472         case suspended:
2473                 break; /* not supported yet */
2474         case readonly:
2475                 if (mddev->pers)
2476                         err = do_md_stop(mddev, 1);
2477                 else {
2478                         mddev->ro = 1;
2479                         err = do_md_run(mddev);
2480                 }
2481                 break;
2482         case read_auto:
2483                 /* stopping an active array */
2484                 if (mddev->pers) {
2485                         err = do_md_stop(mddev, 1);
2486                         if (err == 0)
2487                                 mddev->ro = 2; /* FIXME mark devices writable */
2488                 } else {
2489                         mddev->ro = 2;
2490                         err = do_md_run(mddev);
2491                 }
2492                 break;
2493         case clean:
2494                 if (mddev->pers) {
2495                         restart_array(mddev);
2496                         spin_lock_irq(&mddev->write_lock);
2497                         if (atomic_read(&mddev->writes_pending) == 0) {
2498                                 mddev->in_sync = 1;
2499                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2500                         }
2501                         spin_unlock_irq(&mddev->write_lock);
2502                 } else {
2503                         mddev->ro = 0;
2504                         mddev->recovery_cp = MaxSector;
2505                         err = do_md_run(mddev);
2506                 }
2507                 break;
2508         case active:
2509                 if (mddev->pers) {
2510                         restart_array(mddev);
2511                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2512                         wake_up(&mddev->sb_wait);
2513                         err = 0;
2514                 } else {
2515                         mddev->ro = 0;
2516                         err = do_md_run(mddev);
2517                 }
2518                 break;
2519         case write_pending:
2520         case active_idle:
2521                 /* these cannot be set */
2522                 break;
2523         }
2524         if (err)
2525                 return err;
2526         else
2527                 return len;
2528 }
2529 static struct md_sysfs_entry md_array_state =
2530 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2531
2532 static ssize_t
2533 null_show(mddev_t *mddev, char *page)
2534 {
2535         return -EINVAL;
2536 }
2537
2538 static ssize_t
2539 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2540 {
2541         /* buf must be %d:%d\n? giving major and minor numbers */
2542         /* The new device is added to the array.
2543          * If the array has a persistent superblock, we read the
2544          * superblock to initialise info and check validity.
2545          * Otherwise, only checking done is that in bind_rdev_to_array,
2546          * which mainly checks size.
2547          */
2548         char *e;
2549         int major = simple_strtoul(buf, &e, 10);
2550         int minor;
2551         dev_t dev;
2552         mdk_rdev_t *rdev;
2553         int err;
2554
2555         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2556                 return -EINVAL;
2557         minor = simple_strtoul(e+1, &e, 10);
2558         if (*e && *e != '\n')
2559                 return -EINVAL;
2560         dev = MKDEV(major, minor);
2561         if (major != MAJOR(dev) ||
2562             minor != MINOR(dev))
2563                 return -EOVERFLOW;
2564
2565
2566         if (mddev->persistent) {
2567                 rdev = md_import_device(dev, mddev->major_version,
2568                                         mddev->minor_version);
2569                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2570                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2571                                                        mdk_rdev_t, same_set);
2572                         err = super_types[mddev->major_version]
2573                                 .load_super(rdev, rdev0, mddev->minor_version);
2574                         if (err < 0)
2575                                 goto out;
2576                 }
2577         } else
2578                 rdev = md_import_device(dev, -1, -1);
2579
2580         if (IS_ERR(rdev))
2581                 return PTR_ERR(rdev);
2582         err = bind_rdev_to_array(rdev, mddev);
2583  out:
2584         if (err)
2585                 export_rdev(rdev);
2586         return err ? err : len;
2587 }
2588
2589 static struct md_sysfs_entry md_new_device =
2590 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2591
2592 static ssize_t
2593 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2594 {
2595         char *end;
2596         unsigned long chunk, end_chunk;
2597
2598         if (!mddev->bitmap)
2599                 goto out;
2600         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2601         while (*buf) {
2602                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2603                 if (buf == end) break;
2604                 if (*end == '-') { /* range */
2605                         buf = end + 1;
2606                         end_chunk = simple_strtoul(buf, &end, 0);
2607                         if (buf == end) break;
2608                 }
2609                 if (*end && !isspace(*end)) break;
2610                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2611                 buf = end;
2612                 while (isspace(*buf)) buf++;
2613         }
2614         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2615 out:
2616         return len;
2617 }
2618
2619 static struct md_sysfs_entry md_bitmap =
2620 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2621
2622 static ssize_t
2623 size_show(mddev_t *mddev, char *page)
2624 {
2625         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2626 }
2627
2628 static int update_size(mddev_t *mddev, unsigned long size);
2629
2630 static ssize_t
2631 size_store(mddev_t *mddev, const char *buf, size_t len)
2632 {
2633         /* If array is inactive, we can reduce the component size, but
2634          * not increase it (except from 0).
2635          * If array is active, we can try an on-line resize
2636          */
2637         char *e;
2638         int err = 0;
2639         unsigned long long size = simple_strtoull(buf, &e, 10);
2640         if (!*buf || *buf == '\n' ||
2641             (*e && *e != '\n'))
2642                 return -EINVAL;
2643
2644         if (mddev->pers) {
2645                 err = update_size(mddev, size);
2646                 md_update_sb(mddev, 1);
2647         } else {
2648                 if (mddev->size == 0 ||
2649                     mddev->size > size)
2650                         mddev->size = size;
2651                 else
2652                         err = -ENOSPC;
2653         }
2654         return err ? err : len;
2655 }
2656
2657 static struct md_sysfs_entry md_size =
2658 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2659
2660
2661 /* Metdata version.
2662  * This is either 'none' for arrays with externally managed metadata,
2663  * or N.M for internally known formats
2664  */
2665 static ssize_t
2666 metadata_show(mddev_t *mddev, char *page)
2667 {
2668         if (mddev->persistent)
2669                 return sprintf(page, "%d.%d\n",
2670                                mddev->major_version, mddev->minor_version);
2671         else
2672                 return sprintf(page, "none\n");
2673 }
2674
2675 static ssize_t
2676 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2677 {
2678         int major, minor;
2679         char *e;
2680         if (!list_empty(&mddev->disks))
2681                 return -EBUSY;
2682
2683         if (cmd_match(buf, "none")) {
2684                 mddev->persistent = 0;
2685                 mddev->major_version = 0;
2686                 mddev->minor_version = 90;
2687                 return len;
2688         }
2689         major = simple_strtoul(buf, &e, 10);
2690         if (e==buf || *e != '.')
2691                 return -EINVAL;
2692         buf = e+1;
2693         minor = simple_strtoul(buf, &e, 10);
2694         if (e==buf || (*e && *e != '\n') )
2695                 return -EINVAL;
2696         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2697                 return -ENOENT;
2698         mddev->major_version = major;
2699         mddev->minor_version = minor;
2700         mddev->persistent = 1;
2701         return len;
2702 }
2703
2704 static struct md_sysfs_entry md_metadata =
2705 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2706
2707 static ssize_t
2708 action_show(mddev_t *mddev, char *page)
2709 {
2710         char *type = "idle";
2711         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2712             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2713                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2714                         type = "reshape";
2715                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2716                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2717                                 type = "resync";
2718                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2719                                 type = "check";
2720                         else
2721                                 type = "repair";
2722                 } else
2723                         type = "recover";
2724         }
2725         return sprintf(page, "%s\n", type);
2726 }
2727
2728 static ssize_t
2729 action_store(mddev_t *mddev, const char *page, size_t len)
2730 {
2731         if (!mddev->pers || !mddev->pers->sync_request)
2732                 return -EINVAL;
2733
2734         if (cmd_match(page, "idle")) {
2735                 if (mddev->sync_thread) {
2736                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2737                         md_unregister_thread(mddev->sync_thread);
2738                         mddev->sync_thread = NULL;
2739                         mddev->recovery = 0;
2740                 }
2741         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2742                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2743                 return -EBUSY;
2744         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2745                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2746         else if (cmd_match(page, "reshape")) {
2747                 int err;
2748                 if (mddev->pers->start_reshape == NULL)
2749                         return -EINVAL;
2750                 err = mddev->pers->start_reshape(mddev);
2751                 if (err)
2752                         return err;
2753         } else {
2754                 if (cmd_match(page, "check"))
2755                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2756                 else if (!cmd_match(page, "repair"))
2757                         return -EINVAL;
2758                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2759                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2760         }
2761         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2762         md_wakeup_thread(mddev->thread);
2763         return len;
2764 }
2765
2766 static ssize_t
2767 mismatch_cnt_show(mddev_t *mddev, char *page)
2768 {
2769         return sprintf(page, "%llu\n",
2770                        (unsigned long long) mddev->resync_mismatches);
2771 }
2772
2773 static struct md_sysfs_entry md_scan_mode =
2774 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2775
2776
2777 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2778
2779 static ssize_t
2780 sync_min_show(mddev_t *mddev, char *page)
2781 {
2782         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2783                        mddev->sync_speed_min ? "local": "system");
2784 }
2785
2786 static ssize_t
2787 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2788 {
2789         int min;
2790         char *e;
2791         if (strncmp(buf, "system", 6)==0) {
2792                 mddev->sync_speed_min = 0;
2793                 return len;
2794         }
2795         min = simple_strtoul(buf, &e, 10);
2796         if (buf == e || (*e && *e != '\n') || min <= 0)
2797                 return -EINVAL;
2798         mddev->sync_speed_min = min;
2799         return len;
2800 }
2801
2802 static struct md_sysfs_entry md_sync_min =
2803 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2804
2805 static ssize_t
2806 sync_max_show(mddev_t *mddev, char *page)
2807 {
2808         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2809                        mddev->sync_speed_max ? "local": "system");
2810 }
2811
2812 static ssize_t
2813 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2814 {
2815         int max;
2816         char *e;
2817         if (strncmp(buf, "system", 6)==0) {
2818                 mddev->sync_speed_max = 0;
2819                 return len;
2820         }
2821         max = simple_strtoul(buf, &e, 10);
2822         if (buf == e || (*e && *e != '\n') || max <= 0)
2823                 return -EINVAL;
2824         mddev->sync_speed_max = max;
2825         return len;
2826 }
2827
2828 static struct md_sysfs_entry md_sync_max =
2829 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2830
2831 static ssize_t
2832 degraded_show(mddev_t *mddev, char *page)
2833 {
2834         return sprintf(page, "%d\n", mddev->degraded);
2835 }
2836 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2837
2838 static ssize_t
2839 sync_speed_show(mddev_t *mddev, char *page)
2840 {
2841         unsigned long resync, dt, db;
2842         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2843         dt = ((jiffies - mddev->resync_mark) / HZ);
2844         if (!dt) dt++;
2845         db = resync - (mddev->resync_mark_cnt);
2846         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2847 }
2848
2849 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2850
2851 static ssize_t
2852 sync_completed_show(mddev_t *mddev, char *page)
2853 {
2854         unsigned long max_blocks, resync;
2855
2856         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2857                 max_blocks = mddev->resync_max_sectors;
2858         else
2859                 max_blocks = mddev->size << 1;
2860
2861         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2862         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2863 }
2864
2865 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2866
2867 static ssize_t
2868 suspend_lo_show(mddev_t *mddev, char *page)
2869 {
2870         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2871 }
2872
2873 static ssize_t
2874 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2875 {
2876         char *e;
2877         unsigned long long new = simple_strtoull(buf, &e, 10);
2878
2879         if (mddev->pers->quiesce == NULL)
2880                 return -EINVAL;
2881         if (buf == e || (*e && *e != '\n'))
2882                 return -EINVAL;
2883         if (new >= mddev->suspend_hi ||
2884             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2885                 mddev->suspend_lo = new;
2886                 mddev->pers->quiesce(mddev, 2);
2887                 return len;
2888         } else
2889                 return -EINVAL;
2890 }
2891 static struct md_sysfs_entry md_suspend_lo =
2892 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2893
2894
2895 static ssize_t
2896 suspend_hi_show(mddev_t *mddev, char *page)
2897 {
2898         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2899 }
2900
2901 static ssize_t
2902 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2903 {
2904         char *e;
2905         unsigned long long new = simple_strtoull(buf, &e, 10);
2906
2907         if (mddev->pers->quiesce == NULL)
2908                 return -EINVAL;
2909         if (buf == e || (*e && *e != '\n'))
2910                 return -EINVAL;
2911         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2912             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2913                 mddev->suspend_hi = new;
2914                 mddev->pers->quiesce(mddev, 1);
2915                 mddev->pers->quiesce(mddev, 0);
2916                 return len;
2917         } else
2918                 return -EINVAL;
2919 }
2920 static struct md_sysfs_entry md_suspend_hi =
2921 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2922
2923 static ssize_t
2924 reshape_position_show(mddev_t *mddev, char *page)
2925 {
2926         if (mddev->reshape_position != MaxSector)
2927                 return sprintf(page, "%llu\n",
2928                                (unsigned long long)mddev->reshape_position);
2929         strcpy(page, "none\n");
2930         return 5;
2931 }
2932
2933 static ssize_t
2934 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2935 {
2936         char *e;
2937         unsigned long long new = simple_strtoull(buf, &e, 10);
2938         if (mddev->pers)
2939                 return -EBUSY;
2940         if (buf == e || (*e && *e != '\n'))
2941                 return -EINVAL;
2942         mddev->reshape_position = new;
2943         mddev->delta_disks = 0;
2944         mddev->new_level = mddev->level;
2945         mddev->new_layout = mddev->layout;
2946         mddev->new_chunk = mddev->chunk_size;
2947         return len;
2948 }
2949
2950 static struct md_sysfs_entry md_reshape_position =
2951 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2952        reshape_position_store);
2953
2954
2955 static struct attribute *md_default_attrs[] = {
2956         &md_level.attr,
2957         &md_layout.attr,
2958         &md_raid_disks.attr,
2959         &md_chunk_size.attr,
2960         &md_size.attr,
2961         &md_resync_start.attr,
2962         &md_metadata.attr,
2963         &md_new_device.attr,
2964         &md_safe_delay.attr,
2965         &md_array_state.attr,
2966         &md_reshape_position.attr,
2967         NULL,
2968 };
2969
2970 static struct attribute *md_redundancy_attrs[] = {
2971         &md_scan_mode.attr,
2972         &md_mismatches.attr,
2973         &md_sync_min.attr,
2974         &md_sync_max.attr,
2975         &md_sync_speed.attr,
2976         &md_sync_completed.attr,
2977         &md_suspend_lo.attr,
2978         &md_suspend_hi.attr,
2979         &md_bitmap.attr,
2980         &md_degraded.attr,
2981         NULL,
2982 };
2983 static struct attribute_group md_redundancy_group = {
2984         .name = NULL,
2985         .attrs = md_redundancy_attrs,
2986 };
2987
2988
2989 static ssize_t
2990 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2991 {
2992         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2993         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2994         ssize_t rv;
2995
2996         if (!entry->show)
2997                 return -EIO;
2998         rv = mddev_lock(mddev);
2999         if (!rv) {
3000                 rv = entry->show(mddev, page);
3001                 mddev_unlock(mddev);
3002         }
3003         return rv;
3004 }
3005
3006 static ssize_t
3007 md_attr_store(struct kobject *kobj, struct attribute *attr,
3008               const char *page, size_t length)
3009 {
3010         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3011         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3012         ssize_t rv;
3013
3014         if (!entry->store)
3015                 return -EIO;
3016         if (!capable(CAP_SYS_ADMIN))
3017                 return -EACCES;
3018         rv = mddev_lock(mddev);
3019         if (!rv) {
3020                 rv = entry->store(mddev, page, length);
3021                 mddev_unlock(mddev);
3022         }
3023         return rv;
3024 }
3025
3026 static void md_free(struct kobject *ko)
3027 {
3028         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3029         kfree(mddev);
3030 }
3031
3032 static struct sysfs_ops md_sysfs_ops = {
3033         .show   = md_attr_show,
3034         .store  = md_attr_store,
3035 };
3036 static struct kobj_type md_ktype = {
3037         .release        = md_free,
3038         .sysfs_ops      = &md_sysfs_ops,
3039         .default_attrs  = md_default_attrs,
3040 };
3041
3042 int mdp_major = 0;
3043
3044 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3045 {
3046         static DEFINE_MUTEX(disks_mutex);
3047         mddev_t *mddev = mddev_find(dev);
3048         struct gendisk *disk;
3049         int partitioned = (MAJOR(dev) != MD_MAJOR);
3050         int shift = partitioned ? MdpMinorShift : 0;
3051         int unit = MINOR(dev) >> shift;
3052         int error;
3053
3054         if (!mddev)
3055                 return NULL;
3056
3057         mutex_lock(&disks_mutex);
3058         if (mddev->gendisk) {
3059                 mutex_unlock(&disks_mutex);
3060                 mddev_put(mddev);
3061                 return NULL;
3062         }
3063         disk = alloc_disk(1 << shift);
3064         if (!disk) {
3065                 mutex_unlock(&disks_mutex);
3066                 mddev_put(mddev);
3067                 return NULL;
3068         }
3069         disk->major = MAJOR(dev);
3070         disk->first_minor = unit << shift;
3071         if (partitioned)
3072                 sprintf(disk->disk_name, "md_d%d", unit);
3073         else
3074                 sprintf(disk->disk_name, "md%d", unit);
3075         disk->fops = &md_fops;
3076         disk->private_data = mddev;
3077         disk->queue = mddev->queue;
3078         add_disk(disk);
3079         mddev->gendisk = disk;
3080         mutex_unlock(&disks_mutex);
3081         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3082                                      "%s", "md");
3083         if (error)
3084                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3085                        disk->disk_name);
3086         else
3087                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3088         return NULL;
3089 }
3090
3091 static void md_safemode_timeout(unsigned long data)
3092 {
3093         mddev_t *mddev = (mddev_t *) data;
3094
3095         mddev->safemode = 1;
3096         md_wakeup_thread(mddev->thread);
3097 }
3098
3099 static int start_dirty_degraded;
3100
3101 static int do_md_run(mddev_t * mddev)
3102 {
3103         int err;
3104         int chunk_size;
3105         struct list_head *tmp;
3106         mdk_rdev_t *rdev;
3107         struct gendisk *disk;
3108         struct mdk_personality *pers;
3109         char b[BDEVNAME_SIZE];
3110
3111         if (list_empty(&mddev->disks))
3112                 /* cannot run an array with no devices.. */
3113                 return -EINVAL;
3114
3115         if (mddev->pers)
3116                 return -EBUSY;
3117
3118         /*
3119          * Analyze all RAID superblock(s)
3120          */
3121         if (!mddev->raid_disks)
3122                 analyze_sbs(mddev);
3123
3124         chunk_size = mddev->chunk_size;
3125
3126         if (chunk_size) {
3127                 if (chunk_size > MAX_CHUNK_SIZE) {
3128                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3129                                 chunk_size, MAX_CHUNK_SIZE);
3130                         return -EINVAL;
3131                 }
3132                 /*
3133                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3134                  */
3135                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3136                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3137                         return -EINVAL;
3138                 }
3139                 if (chunk_size < PAGE_SIZE) {
3140                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3141                                 chunk_size, PAGE_SIZE);
3142                         return -EINVAL;
3143                 }
3144
3145                 /* devices must have minimum size of one chunk */
3146                 ITERATE_RDEV(mddev,rdev,tmp) {
3147                         if (test_bit(Faulty, &rdev->flags))
3148                                 continue;
3149                         if (rdev->size < chunk_size / 1024) {
3150                                 printk(KERN_WARNING
3151                                         "md: Dev %s smaller than chunk_size:"
3152                                         " %lluk < %dk\n",
3153                                         bdevname(rdev->bdev,b),
3154                                         (unsigned long long)rdev->size,
3155                                         chunk_size / 1024);
3156                                 return -EINVAL;
3157                         }
3158                 }
3159         }
3160
3161 #ifdef CONFIG_KMOD
3162         if (mddev->level != LEVEL_NONE)
3163                 request_module("md-level-%d", mddev->level);
3164         else if (mddev->clevel[0])
3165                 request_module("md-%s", mddev->clevel);
3166 #endif
3167
3168         /*
3169          * Drop all container device buffers, from now on
3170          * the only valid external interface is through the md
3171          * device.
3172          */
3173         ITERATE_RDEV(mddev,rdev,tmp) {
3174                 if (test_bit(Faulty, &rdev->flags))
3175                         continue;
3176                 sync_blockdev(rdev->bdev);
3177                 invalidate_bdev(rdev->bdev);
3178
3179                 /* perform some consistency tests on the device.
3180                  * We don't want the data to overlap the metadata,
3181                  * Internal Bitmap issues has handled elsewhere.
3182                  */
3183                 if (rdev->data_offset < rdev->sb_offset) {
3184                         if (mddev->size &&
3185                             rdev->data_offset + mddev->size*2
3186                             > rdev->sb_offset*2) {
3187                                 printk("md: %s: data overlaps metadata\n",
3188                                        mdname(mddev));
3189                                 return -EINVAL;
3190                         }
3191                 } else {
3192                         if (rdev->sb_offset*2 + rdev->sb_size/512
3193                             > rdev->data_offset) {
3194                                 printk("md: %s: metadata overlaps data\n",
3195                                        mdname(mddev));
3196                                 return -EINVAL;
3197                         }
3198                 }
3199         }
3200
3201         md_probe(mddev->unit, NULL, NULL);
3202         disk = mddev->gendisk;
3203         if (!disk)
3204                 return -ENOMEM;
3205
3206         spin_lock(&pers_lock);
3207         pers = find_pers(mddev->level, mddev->clevel);
3208         if (!pers || !try_module_get(pers->owner)) {
3209                 spin_unlock(&pers_lock);
3210                 if (mddev->level != LEVEL_NONE)
3211                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3212                                mddev->level);
3213                 else
3214                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3215                                mddev->clevel);
3216                 return -EINVAL;
3217         }
3218         mddev->pers = pers;
3219         spin_unlock(&pers_lock);
3220         mddev->level = pers->level;
3221         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3222
3223         if (mddev->reshape_position != MaxSector &&
3224             pers->start_reshape == NULL) {
3225                 /* This personality cannot handle reshaping... */
3226                 mddev->pers = NULL;
3227                 module_put(pers->owner);
3228                 return -EINVAL;
3229         }
3230
3231         if (pers->sync_request) {
3232                 /* Warn if this is a potentially silly
3233                  * configuration.
3234                  */
3235                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3236                 mdk_rdev_t *rdev2;
3237                 struct list_head *tmp2;
3238                 int warned = 0;
3239                 ITERATE_RDEV(mddev, rdev, tmp) {
3240                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3241                                 if (rdev < rdev2 &&
3242                                     rdev->bdev->bd_contains ==
3243                                     rdev2->bdev->bd_contains) {
3244                                         printk(KERN_WARNING
3245                                                "%s: WARNING: %s appears to be"
3246                                                " on the same physical disk as"
3247                                                " %s.\n",
3248                                                mdname(mddev),
3249                                                bdevname(rdev->bdev,b),
3250                                                bdevname(rdev2->bdev,b2));
3251                                         warned = 1;
3252                                 }
3253                         }
3254                 }
3255                 if (warned)
3256                         printk(KERN_WARNING
3257                                "True protection against single-disk"
3258                                " failure might be compromised.\n");
3259         }
3260
3261         mddev->recovery = 0;
3262         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3263         mddev->barriers_work = 1;
3264         mddev->ok_start_degraded = start_dirty_degraded;
3265
3266         if (start_readonly)
3267                 mddev->ro = 2; /* read-only, but switch on first write */
3268
3269         err = mddev->pers->run(mddev);
3270         if (!err && mddev->pers->sync_request) {
3271                 err = bitmap_create(mddev);
3272                 if (err) {
3273                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3274                                mdname(mddev), err);
3275                         mddev->pers->stop(mddev);
3276                 }
3277         }
3278         if (err) {
3279                 printk(KERN_ERR "md: pers->run() failed ...\n");
3280                 module_put(mddev->pers->owner);
3281                 mddev->pers = NULL;
3282                 bitmap_destroy(mddev);
3283                 return err;
3284         }
3285         if (mddev->pers->sync_request) {
3286                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3287                         printk(KERN_WARNING
3288                                "md: cannot register extra attributes for %s\n",
3289                                mdname(mddev));
3290         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3291                 mddev->ro = 0;
3292
3293         atomic_set(&mddev->writes_pending,0);
3294         mddev->safemode = 0;
3295         mddev->safemode_timer.function = md_safemode_timeout;
3296         mddev->safemode_timer.data = (unsigned long) mddev;
3297         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3298         mddev->in_sync = 1;
3299
3300         ITERATE_RDEV(mddev,rdev,tmp)
3301                 if (rdev->raid_disk >= 0) {
3302                         char nm[20];
3303                         sprintf(nm, "rd%d", rdev->raid_disk);
3304                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3305                                 printk("md: cannot register %s for %s\n",
3306                                        nm, mdname(mddev));
3307                 }
3308         
3309         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3310         
3311         if (mddev->flags)
3312                 md_update_sb(mddev, 0);
3313
3314         set_capacity(disk, mddev->array_size<<1);
3315
3316         /* If we call blk_queue_make_request here, it will
3317          * re-initialise max_sectors etc which may have been
3318          * refined inside -> run.  So just set the bits we need to set.
3319          * Most initialisation happended when we called
3320          * blk_queue_make_request(..., md_fail_request)
3321          * earlier.
3322          */
3323         mddev->queue->queuedata = mddev;
3324         mddev->queue->make_request_fn = mddev->pers->make_request;
3325
3326         /* If there is a partially-recovered drive we need to
3327          * start recovery here.  If we leave it to md_check_recovery,
3328          * it will remove the drives and not do the right thing
3329          */
3330         if (mddev->degraded && !mddev->sync_thread) {
3331                 struct list_head *rtmp;
3332                 int spares = 0;
3333                 ITERATE_RDEV(mddev,rdev,rtmp)
3334                         if (rdev->raid_disk >= 0 &&
3335                             !test_bit(In_sync, &rdev->flags) &&
3336                             !test_bit(Faulty, &rdev->flags))
3337                                 /* complete an interrupted recovery */
3338                                 spares++;
3339                 if (spares && mddev->pers->sync_request) {
3340                         mddev->recovery = 0;
3341                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3342                         mddev->sync_thread = md_register_thread(md_do_sync,
3343                                                                 mddev,
3344                                                                 "%s_resync");
3345                         if (!mddev->sync_thread) {
3346                                 printk(KERN_ERR "%s: could not start resync"
3347                                        " thread...\n",
3348                                        mdname(mddev));
3349                                 /* leave the spares where they are, it shouldn't hurt */
3350                                 mddev->recovery = 0;
3351                         }
3352                 }
3353         }
3354         md_wakeup_thread(mddev->thread);
3355         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3356
3357         mddev->changed = 1;
3358         md_new_event(mddev);
3359         kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3360         return 0;
3361 }
3362
3363 static int restart_array(mddev_t *mddev)
3364 {
3365         struct gendisk *disk = mddev->gendisk;
3366         int err;
3367
3368         /*
3369          * Complain if it has no devices
3370          */
3371         err = -ENXIO;
3372         if (list_empty(&mddev->disks))
3373                 goto out;
3374
3375         if (mddev->pers) {
3376                 err = -EBUSY;
3377                 if (!mddev->ro)
3378                         goto out;
3379
3380                 mddev->safemode = 0;
3381                 mddev->ro = 0;
3382                 set_disk_ro(disk, 0);
3383
3384                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3385                         mdname(mddev));
3386                 /*
3387                  * Kick recovery or resync if necessary
3388                  */
3389                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3390                 md_wakeup_thread(mddev->thread);
3391                 md_wakeup_thread(mddev->sync_thread);
3392                 err = 0;
3393         } else
3394                 err = -EINVAL;
3395
3396 out:
3397         return err;
3398 }
3399
3400 /* similar to deny_write_access, but accounts for our holding a reference
3401  * to the file ourselves */
3402 static int deny_bitmap_write_access(struct file * file)
3403 {
3404         struct inode *inode = file->f_mapping->host;
3405
3406         spin_lock(&inode->i_lock);
3407         if (atomic_read(&inode->i_writecount) > 1) {
3408                 spin_unlock(&inode->i_lock);
3409                 return -ETXTBSY;
3410         }
3411         atomic_set(&inode->i_writecount, -1);
3412         spin_unlock(&inode->i_lock);
3413
3414         return 0;
3415 }
3416
3417 static void restore_bitmap_write_access(struct file *file)
3418 {
3419         struct inode *inode = file->f_mapping->host;
3420
3421         spin_lock(&inode->i_lock);
3422         atomic_set(&inode->i_writecount, 1);
3423         spin_unlock(&inode->i_lock);
3424 }
3425
3426 /* mode:
3427  *   0 - completely stop and dis-assemble array
3428  *   1 - switch to readonly
3429  *   2 - stop but do not disassemble array
3430  */
3431 static int do_md_stop(mddev_t * mddev, int mode)
3432 {
3433         int err = 0;
3434         struct gendisk *disk = mddev->gendisk;
3435
3436         if (mddev->pers) {
3437                 if (atomic_read(&mddev->active)>2) {
3438                         printk("md: %s still in use.\n",mdname(mddev));
3439                         return -EBUSY;
3440                 }
3441
3442                 if (mddev->sync_thread) {
3443                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3444                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3445                         md_unregister_thread(mddev->sync_thread);
3446                         mddev->sync_thread = NULL;
3447                 }
3448
3449                 del_timer_sync(&mddev->safemode_timer);
3450
3451                 invalidate_partition(disk, 0);
3452
3453                 switch(mode) {
3454                 case 1: /* readonly */
3455                         err  = -ENXIO;
3456                         if (mddev->ro==1)
3457                                 goto out;
3458                         mddev->ro = 1;
3459                         break;
3460                 case 0: /* disassemble */
3461                 case 2: /* stop */
3462                         bitmap_flush(mddev);
3463                         md_super_wait(mddev);
3464                         if (mddev->ro)
3465                                 set_disk_ro(disk, 0);
3466                         blk_queue_make_request(mddev->queue, md_fail_request);
3467                         mddev->pers->stop(mddev);
3468                         mddev->queue->merge_bvec_fn = NULL;
3469                         mddev->queue->unplug_fn = NULL;
3470                         mddev->queue->backing_dev_info.congested_fn = NULL;
3471                         if (mddev->pers->sync_request)
3472                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3473
3474                         module_put(mddev->pers->owner);
3475                         mddev->pers = NULL;
3476
3477                         set_capacity(disk, 0);
3478                         mddev->changed = 1;
3479
3480                         if (mddev->ro)
3481                                 mddev->ro = 0;
3482                 }
3483                 if (!mddev->in_sync || mddev->flags) {
3484                         /* mark array as shutdown cleanly */
3485                         mddev->in_sync = 1;
3486                         md_update_sb(mddev, 1);
3487                 }
3488                 if (mode == 1)
3489                         set_disk_ro(disk, 1);
3490                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3491         }
3492
3493         /*
3494          * Free resources if final stop
3495          */
3496         if (mode == 0) {
3497                 mdk_rdev_t *rdev;
3498                 struct list_head *tmp;
3499
3500                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3501
3502                 bitmap_destroy(mddev);
3503                 if (mddev->bitmap_file) {
3504                         restore_bitmap_write_access(mddev->bitmap_file);
3505                         fput(mddev->bitmap_file);
3506                         mddev->bitmap_file = NULL;
3507                 }
3508                 mddev->bitmap_offset = 0;
3509
3510                 ITERATE_RDEV(mddev,rdev,tmp)
3511                         if (rdev->raid_disk >= 0) {
3512                                 char nm[20];
3513                                 sprintf(nm, "rd%d", rdev->raid_disk);
3514                                 sysfs_remove_link(&mddev->kobj, nm);
3515                         }
3516
3517                 /* make sure all delayed_delete calls have finished */
3518                 flush_scheduled_work();
3519
3520                 export_array(mddev);
3521
3522                 mddev->array_size = 0;
3523                 mddev->size = 0;
3524                 mddev->raid_disks = 0;
3525                 mddev->recovery_cp = 0;
3526                 mddev->reshape_position = MaxSector;
3527
3528         } else if (mddev->pers)
3529                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3530                         mdname(mddev));
3531         err = 0;
3532         md_new_event(mddev);
3533 out:
3534         return err;
3535 }
3536
3537 #ifndef MODULE
3538 static void autorun_array(mddev_t *mddev)
3539 {
3540         mdk_rdev_t *rdev;
3541         struct list_head *tmp;
3542         int err;
3543
3544         if (list_empty(&mddev->disks))
3545                 return;
3546
3547         printk(KERN_INFO "md: running: ");
3548
3549         ITERATE_RDEV(mddev,rdev,tmp) {
3550                 char b[BDEVNAME_SIZE];
3551                 printk("<%s>", bdevname(rdev->bdev,b));
3552         }
3553         printk("\n");
3554
3555         err = do_md_run (mddev);
3556         if (err) {
3557                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3558                 do_md_stop (mddev, 0);
3559         }
3560 }
3561
3562 /*
3563  * lets try to run arrays based on all disks that have arrived
3564  * until now. (those are in pending_raid_disks)
3565  *
3566  * the method: pick the first pending disk, collect all disks with
3567  * the same UUID, remove all from the pending list and put them into
3568  * the 'same_array' list. Then order this list based on superblock
3569  * update time (freshest comes first), kick out 'old' disks and
3570  * compare superblocks. If everything's fine then run it.
3571  *
3572  * If "unit" is allocated, then bump its reference count
3573  */
3574 static void autorun_devices(int part)
3575 {
3576         struct list_head *tmp;
3577         mdk_rdev_t *rdev0, *rdev;
3578         mddev_t *mddev;
3579         char b[BDEVNAME_SIZE];
3580
3581         printk(KERN_INFO "md: autorun ...\n");
3582         while (!list_empty(&pending_raid_disks)) {
3583                 int unit;
3584                 dev_t dev;
3585                 LIST_HEAD(candidates);
3586                 rdev0 = list_entry(pending_raid_disks.next,
3587                                          mdk_rdev_t, same_set);
3588
3589                 printk(KERN_INFO "md: considering %s ...\n",
3590                         bdevname(rdev0->bdev,b));
3591                 INIT_LIST_HEAD(&candidates);
3592                 ITERATE_RDEV_PENDING(rdev,tmp)
3593                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3594                                 printk(KERN_INFO "md:  adding %s ...\n",
3595                                         bdevname(rdev->bdev,b));
3596                                 list_move(&rdev->same_set, &candidates);
3597                         }
3598                 /*
3599                  * now we have a set of devices, with all of them having
3600                  * mostly sane superblocks. It's time to allocate the
3601                  * mddev.
3602                  */
3603                 if (part) {
3604                         dev = MKDEV(mdp_major,
3605                                     rdev0->preferred_minor << MdpMinorShift);
3606                         unit = MINOR(dev) >> MdpMinorShift;
3607                 } else {
3608                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3609                         unit = MINOR(dev);
3610                 }
3611                 if (rdev0->preferred_minor != unit) {
3612                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3613                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3614                         break;
3615                 }
3616
3617                 md_probe(dev, NULL, NULL);
3618                 mddev = mddev_find(dev);
3619                 if (!mddev) {
3620                         printk(KERN_ERR 
3621                                 "md: cannot allocate memory for md drive.\n");
3622                         break;
3623                 }
3624                 if (mddev_lock(mddev)) 
3625                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3626                                mdname(mddev));
3627                 else if (mddev->raid_disks || mddev->major_version
3628                          || !list_empty(&mddev->disks)) {
3629                         printk(KERN_WARNING 
3630                                 "md: %s already running, cannot run %s\n",
3631                                 mdname(mddev), bdevname(rdev0->bdev,b));
3632                         mddev_unlock(mddev);
3633                 } else {
3634                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3635                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3636                                 list_del_init(&rdev->same_set);
3637                                 if (bind_rdev_to_array(rdev, mddev))
3638                                         export_rdev(rdev);
3639                         }
3640                         autorun_array(mddev);
3641                         mddev_unlock(mddev);
3642                 }
3643                 /* on success, candidates will be empty, on error
3644                  * it won't...
3645                  */
3646                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3647                         export_rdev(rdev);
3648                 mddev_put(mddev);
3649         }
3650         printk(KERN_INFO "md: ... autorun DONE.\n");
3651 }
3652 #endif /* !MODULE */
3653
3654 static int get_version(void __user * arg)
3655 {
3656         mdu_version_t ver;
3657
3658         ver.major = MD_MAJOR_VERSION;
3659         ver.minor = MD_MINOR_VERSION;
3660         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3661
3662         if (copy_to_user(arg, &ver, sizeof(ver)))
3663                 return -EFAULT;
3664
3665         return 0;
3666 }
3667
3668 static int get_array_info(mddev_t * mddev, void __user * arg)
3669 {
3670         mdu_array_info_t info;
3671         int nr,working,active,failed,spare;
3672         mdk_rdev_t *rdev;
3673         struct list_head *tmp;
3674
3675         nr=working=active=failed=spare=0;
3676         ITERATE_RDEV(mddev,rdev,tmp) {
3677                 nr++;
3678                 if (test_bit(Faulty, &rdev->flags))
3679                         failed++;
3680                 else {
3681                         working++;
3682                         if (test_bit(In_sync, &rdev->flags))
3683                                 active++;       
3684                         else
3685                                 spare++;
3686                 }
3687         }
3688
3689         info.major_version = mddev->major_version;
3690         info.minor_version = mddev->minor_version;
3691         info.patch_version = MD_PATCHLEVEL_VERSION;
3692         info.ctime         = mddev->ctime;
3693         info.level         = mddev->level;
3694         info.size          = mddev->size;
3695         if (info.size != mddev->size) /* overflow */
3696                 info.size = -1;
3697         info.nr_disks      = nr;
3698         info.raid_disks    = mddev->raid_disks;
3699         info.md_minor      = mddev->md_minor;
3700         info.not_persistent= !mddev->persistent;
3701
3702         info.utime         = mddev->utime;
3703         info.state         = 0;
3704         if (mddev->in_sync)
3705                 info.state = (1<<MD_SB_CLEAN);
3706         if (mddev->bitmap && mddev->bitmap_offset)
3707                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3708         info.active_disks  = active;
3709         info.working_disks = working;
3710         info.failed_disks  = failed;
3711         info.spare_disks   = spare;
3712
3713         info.layout        = mddev->layout;
3714         info.chunk_size    = mddev->chunk_size;
3715
3716         if (copy_to_user(arg, &info, sizeof(info)))
3717                 return -EFAULT;
3718
3719         return 0;
3720 }
3721
3722 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3723 {
3724         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3725         char *ptr, *buf = NULL;
3726         int err = -ENOMEM;
3727
3728         md_allow_write(mddev);
3729
3730         file = kmalloc(sizeof(*file), GFP_KERNEL);
3731         if (!file)
3732                 goto out;
3733
3734         /* bitmap disabled, zero the first byte and copy out */
3735         if (!mddev->bitmap || !mddev->bitmap->file) {
3736                 file->pathname[0] = '\0';
3737                 goto copy_out;
3738         }
3739
3740         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3741         if (!buf)
3742                 goto out;
3743
3744         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3745         if (!ptr)
3746                 goto out;
3747
3748         strcpy(file->pathname, ptr);
3749
3750 copy_out:
3751         err = 0;
3752         if (copy_to_user(arg, file, sizeof(*file)))
3753                 err = -EFAULT;
3754 out:
3755         kfree(buf);
3756         kfree(file);
3757         return err;
3758 }
3759
3760 static int get_disk_info(mddev_t * mddev, void __user * arg)
3761 {
3762         mdu_disk_info_t info;
3763         unsigned int nr;
3764         mdk_rdev_t *rdev;
3765
3766         if (copy_from_user(&info, arg, sizeof(info)))
3767                 return -EFAULT;
3768
3769         nr = info.number;
3770
3771         rdev = find_rdev_nr(mddev, nr);
3772         if (rdev) {
3773                 info.major = MAJOR(rdev->bdev->bd_dev);
3774                 info.minor = MINOR(rdev->bdev->bd_dev);
3775                 info.raid_disk = rdev->raid_disk;
3776                 info.state = 0;
3777                 if (test_bit(Faulty, &rdev->flags))
3778                         info.state |= (1<<MD_DISK_FAULTY);
3779                 else if (test_bit(In_sync, &rdev->flags)) {
3780                         info.state |= (1<<MD_DISK_ACTIVE);
3781                         info.state |= (1<<MD_DISK_SYNC);
3782                 }
3783                 if (test_bit(WriteMostly, &rdev->flags))
3784                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3785         } else {
3786                 info.major = info.minor = 0;
3787                 info.raid_disk = -1;
3788                 info.state = (1<<MD_DISK_REMOVED);
3789         }
3790
3791         if (copy_to_user(arg, &info, sizeof(info)))
3792                 return -EFAULT;
3793
3794         return 0;
3795 }
3796
3797 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3798 {
3799         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3800         mdk_rdev_t *rdev;
3801         dev_t dev = MKDEV(info->major,info->minor);
3802
3803         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3804                 return -EOVERFLOW;
3805
3806         if (!mddev->raid_disks) {
3807                 int err;
3808                 /* expecting a device which has a superblock */
3809                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3810                 if (IS_ERR(rdev)) {
3811                         printk(KERN_WARNING 
3812                                 "md: md_import_device returned %ld\n",
3813                                 PTR_ERR(rdev));
3814                         return PTR_ERR(rdev);
3815                 }
3816                 if (!list_empty(&mddev->disks)) {
3817                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3818                                                         mdk_rdev_t, same_set);
3819                         int err = super_types[mddev->major_version]
3820                                 .load_super(rdev, rdev0, mddev->minor_version);
3821                         if (err < 0) {
3822                                 printk(KERN_WARNING 
3823                                         "md: %s has different UUID to %s\n",
3824                                         bdevname(rdev->bdev,b), 
3825                                         bdevname(rdev0->bdev,b2));
3826                                 export_rdev(rdev);
3827                                 return -EINVAL;
3828                         }
3829                 }
3830                 err = bind_rdev_to_array(rdev, mddev);
3831                 if (err)
3832                         export_rdev(rdev);
3833                 return err;
3834         }
3835
3836         /*
3837          * add_new_disk can be used once the array is assembled
3838          * to add "hot spares".  They must already have a superblock
3839          * written
3840          */
3841         if (mddev->pers) {
3842                 int err;
3843                 if (!mddev->pers->hot_add_disk) {
3844                         printk(KERN_WARNING 
3845                                 "%s: personality does not support diskops!\n",
3846                                mdname(mddev));
3847                         return -EINVAL;
3848                 }
3849                 if (mddev->persistent)
3850                         rdev = md_import_device(dev, mddev->major_version,
3851                                                 mddev->minor_version);
3852                 else
3853                         rdev = md_import_device(dev, -1, -1);
3854                 if (IS_ERR(rdev)) {
3855                         printk(KERN_WARNING 
3856                                 "md: md_import_device returned %ld\n",
3857                                 PTR_ERR(rdev));
3858                         return PTR_ERR(rdev);
3859                 }
3860                 /* set save_raid_disk if appropriate */
3861                 if (!mddev->persistent) {
3862                         if (info->state & (1<<MD_DISK_SYNC)  &&
3863                             info->raid_disk < mddev->raid_disks)
3864                                 rdev->raid_disk = info->raid_disk;
3865                         else
3866                                 rdev->raid_disk = -1;
3867                 } else
3868                         super_types[mddev->major_version].
3869                                 validate_super(mddev, rdev);
3870                 rdev->saved_raid_disk = rdev->raid_disk;
3871
3872                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3873                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3874                         set_bit(WriteMostly, &rdev->flags);
3875
3876                 rdev->raid_disk = -1;
3877                 err = bind_rdev_to_array(rdev, mddev);
3878                 if (!err && !mddev->pers->hot_remove_disk) {
3879                         /* If there is hot_add_disk but no hot_remove_disk
3880                          * then added disks for geometry changes,
3881                          * and should be added immediately.
3882                          */
3883                         super_types[mddev->major_version].
3884                                 validate_super(mddev, rdev);
3885                         err = mddev->pers->hot_add_disk(mddev, rdev);
3886                         if (err)
3887                                 unbind_rdev_from_array(rdev);
3888                 }
3889                 if (err)
3890                         export_rdev(rdev);
3891
3892                 md_update_sb(mddev, 1);
3893                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3894                 md_wakeup_thread(mddev->thread);
3895                 return err;
3896         }
3897
3898         /* otherwise, add_new_disk is only allowed
3899          * for major_version==0 superblocks
3900          */
3901         if (mddev->major_version != 0) {
3902                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3903                        mdname(mddev));
3904                 return -EINVAL;
3905         }
3906
3907         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3908                 int err;
3909                 rdev = md_import_device (dev, -1, 0);
3910                 if (IS_ERR(rdev)) {
3911                         printk(KERN_WARNING 
3912                                 "md: error, md_import_device() returned %ld\n",
3913                                 PTR_ERR(rdev));
3914                         return PTR_ERR(rdev);
3915                 }
3916                 rdev->desc_nr = info->number;
3917                 if (info->raid_disk < mddev->raid_disks)
3918                         rdev->raid_disk = info->raid_disk;
3919                 else
3920                         rdev->raid_disk = -1;
3921
3922                 rdev->flags = 0;
3923
3924                 if (rdev->raid_disk < mddev->raid_disks)
3925                         if (info->state & (1<<MD_DISK_SYNC))
3926                                 set_bit(In_sync, &rdev->flags);
3927
3928                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3929                         set_bit(WriteMostly, &rdev->flags);
3930
3931                 if (!mddev->persistent) {
3932                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3933                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3934                 } else 
3935                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3936                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3937
3938                 err = bind_rdev_to_array(rdev, mddev);
3939                 if (err) {
3940                         export_rdev(rdev);
3941                         return err;
3942                 }
3943         }
3944
3945         return 0;
3946 }
3947
3948 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3949 {
3950         char b[BDEVNAME_SIZE];
3951         mdk_rdev_t *rdev;
3952
3953         if (!mddev->pers)
3954                 return -ENODEV;
3955
3956         rdev = find_rdev(mddev, dev);
3957         if (!rdev)
3958                 return -ENXIO;
3959
3960         if (rdev->raid_disk >= 0)
3961                 goto busy;
3962
3963         kick_rdev_from_array(rdev);
3964         md_update_sb(mddev, 1);
3965         md_new_event(mddev);
3966
3967         return 0;
3968 busy:
3969         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3970                 bdevname(rdev->bdev,b), mdname(mddev));
3971         return -EBUSY;
3972 }
3973
3974 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3975 {
3976         char b[BDEVNAME_SIZE];
3977         int err;
3978         unsigned int size;
3979         mdk_rdev_t *rdev;
3980
3981         if (!mddev->pers)
3982                 return -ENODEV;
3983
3984         if (mddev->major_version != 0) {
3985                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3986                         " version-0 superblocks.\n",
3987                         mdname(mddev));
3988                 return -EINVAL;
3989         }
3990         if (!mddev->pers->hot_add_disk) {
3991                 printk(KERN_WARNING 
3992                         "%s: personality does not support diskops!\n",
3993                         mdname(mddev));
3994                 return -EINVAL;
3995         }
3996
3997         rdev = md_import_device (dev, -1, 0);
3998         if (IS_ERR(rdev)) {
3999                 printk(KERN_WARNING 
4000                         "md: error, md_import_device() returned %ld\n",
4001                         PTR_ERR(rdev));
4002                 return -EINVAL;
4003         }
4004
4005         if (mddev->persistent)
4006                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4007         else
4008                 rdev->sb_offset =
4009                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4010
4011         size = calc_dev_size(rdev, mddev->chunk_size);
4012         rdev->size = size;
4013
4014         if (test_bit(Faulty, &rdev->flags)) {
4015                 printk(KERN_WARNING 
4016                         "md: can not hot-add faulty %s disk to %s!\n",
4017                         bdevname(rdev->bdev,b), mdname(mddev));
4018                 err = -EINVAL;
4019                 goto abort_export;
4020         }
4021         clear_bit(In_sync, &rdev->flags);
4022         rdev->desc_nr = -1;
4023         rdev->saved_raid_disk = -1;
4024         err = bind_rdev_to_array(rdev, mddev);
4025         if (err)
4026                 goto abort_export;
4027
4028         /*
4029          * The rest should better be atomic, we can have disk failures
4030          * noticed in interrupt contexts ...
4031          */
4032
4033         if (rdev->desc_nr == mddev->max_disks) {
4034                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4035                         mdname(mddev));
4036                 err = -EBUSY;
4037                 goto abort_unbind_export;
4038         }
4039
4040         rdev->raid_disk = -1;
4041
4042         md_update_sb(mddev, 1);
4043
4044         /*
4045          * Kick recovery, maybe this spare has to be added to the
4046          * array immediately.
4047          */
4048         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4049         md_wakeup_thread(mddev->thread);
4050         md_new_event(mddev);
4051         return 0;
4052
4053 abort_unbind_export:
4054         unbind_rdev_from_array(rdev);
4055
4056 abort_export:
4057         export_rdev(rdev);
4058         return err;
4059 }
4060
4061 static int set_bitmap_file(mddev_t *mddev, int fd)
4062 {
4063         int err;
4064
4065         if (mddev->pers) {
4066                 if (!mddev->pers->quiesce)
4067                         return -EBUSY;
4068                 if (mddev->recovery || mddev->sync_thread)
4069                         return -EBUSY;
4070                 /* we should be able to change the bitmap.. */
4071         }
4072
4073
4074         if (fd >= 0) {
4075                 if (mddev->bitmap)
4076                         return -EEXIST; /* cannot add when bitmap is present */
4077                 mddev->bitmap_file = fget(fd);
4078
4079                 if (mddev->bitmap_file == NULL) {
4080                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4081                                mdname(mddev));
4082                         return -EBADF;
4083                 }
4084
4085                 err = deny_bitmap_write_access(mddev->bitmap_file);
4086                 if (err) {
4087                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4088                                mdname(mddev));
4089                         fput(mddev->bitmap_file);
4090                         mddev->bitmap_file = NULL;
4091                         return err;
4092                 }
4093                 mddev->bitmap_offset = 0; /* file overrides offset */
4094         } else if (mddev->bitmap == NULL)
4095                 return -ENOENT; /* cannot remove what isn't there */
4096         err = 0;
4097         if (mddev->pers) {
4098                 mddev->pers->quiesce(mddev, 1);
4099                 if (fd >= 0)
4100                         err = bitmap_create(mddev);
4101                 if (fd < 0 || err) {
4102                         bitmap_destroy(mddev);
4103                         fd = -1; /* make sure to put the file */
4104                 }
4105                 mddev->pers->quiesce(mddev, 0);
4106         }
4107         if (fd < 0) {
4108                 if (mddev->bitmap_file) {
4109                         restore_bitmap_write_access(mddev->bitmap_file);
4110                         fput(mddev->bitmap_file);
4111                 }
4112                 mddev->bitmap_file = NULL;
4113         }
4114
4115         return err;
4116 }
4117
4118 /*
4119  * set_array_info is used two different ways
4120  * The original usage is when creating a new array.
4121  * In this usage, raid_disks is > 0 and it together with
4122  *  level, size, not_persistent,layout,chunksize determine the
4123  *  shape of the array.
4124  *  This will always create an array with a type-0.90.0 superblock.
4125  * The newer usage is when assembling an array.
4126  *  In this case raid_disks will be 0, and the major_version field is
4127  *  use to determine which style super-blocks are to be found on the devices.
4128  *  The minor and patch _version numbers are also kept incase the
4129  *  super_block handler wishes to interpret them.
4130  */
4131 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4132 {
4133
4134         if (info->raid_disks == 0) {
4135                 /* just setting version number for superblock loading */
4136                 if (info->major_version < 0 ||
4137                     info->major_version >= ARRAY_SIZE(super_types) ||
4138                     super_types[info->major_version].name == NULL) {
4139                         /* maybe try to auto-load a module? */
4140                         printk(KERN_INFO 
4141                                 "md: superblock version %d not known\n",
4142                                 info->major_version);
4143                         return -EINVAL;
4144                 }
4145                 mddev->major_version = info->major_version;
4146                 mddev->minor_version = info->minor_version;
4147                 mddev->patch_version = info->patch_version;
4148                 mddev->persistent = !info->not_persistent;
4149                 return 0;
4150         }
4151         mddev->major_version = MD_MAJOR_VERSION;
4152         mddev->minor_version = MD_MINOR_VERSION;
4153         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4154         mddev->ctime         = get_seconds();
4155
4156         mddev->level         = info->level;
4157         mddev->clevel[0]     = 0;
4158         mddev->size          = info->size;
4159         mddev->raid_disks    = info->raid_disks;
4160         /* don't set md_minor, it is determined by which /dev/md* was
4161          * openned
4162          */
4163         if (info->state & (1<<MD_SB_CLEAN))
4164                 mddev->recovery_cp = MaxSector;
4165         else
4166                 mddev->recovery_cp = 0;
4167         mddev->persistent    = ! info->not_persistent;
4168
4169         mddev->layout        = info->layout;
4170         mddev->chunk_size    = info->chunk_size;
4171
4172         mddev->max_disks     = MD_SB_DISKS;
4173
4174         mddev->flags         = 0;
4175         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4176
4177         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4178         mddev->bitmap_offset = 0;
4179
4180         mddev->reshape_position = MaxSector;
4181
4182         /*
4183          * Generate a 128 bit UUID
4184          */
4185         get_random_bytes(mddev->uuid, 16);
4186
4187         mddev->new_level = mddev->level;
4188         mddev->new_chunk = mddev->chunk_size;
4189         mddev->new_layout = mddev->layout;
4190         mddev->delta_disks = 0;
4191
4192         return 0;
4193 }
4194
4195 static int update_size(mddev_t *mddev, unsigned long size)
4196 {
4197         mdk_rdev_t * rdev;
4198         int rv;
4199         struct list_head *tmp;
4200         int fit = (size == 0);
4201
4202         if (mddev->pers->resize == NULL)
4203                 return -EINVAL;
4204         /* The "size" is the amount of each device that is used.
4205          * This can only make sense for arrays with redundancy.
4206          * linear and raid0 always use whatever space is available
4207          * We can only consider changing the size if no resync
4208          * or reconstruction is happening, and if the new size
4209          * is acceptable. It must fit before the sb_offset or,
4210          * if that is <data_offset, it must fit before the
4211          * size of each device.
4212          * If size is zero, we find the largest size that fits.
4213          */
4214         if (mddev->sync_thread)
4215                 return -EBUSY;
4216         ITERATE_RDEV(mddev,rdev,tmp) {
4217                 sector_t avail;
4218                 avail = rdev->size * 2;
4219
4220                 if (fit && (size == 0 || size > avail/2))
4221                         size = avail/2;
4222                 if (avail < ((sector_t)size << 1))
4223                         return -ENOSPC;
4224         }
4225         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4226         if (!rv) {
4227                 struct block_device *bdev;
4228
4229                 bdev = bdget_disk(mddev->gendisk, 0);
4230                 if (bdev) {
4231                         mutex_lock(&bdev->bd_inode->i_mutex);
4232                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4233                         mutex_unlock(&bdev->bd_inode->i_mutex);
4234                         bdput(bdev);
4235                 }
4236         }
4237         return rv;
4238 }
4239
4240 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4241 {
4242         int rv;
4243         /* change the number of raid disks */
4244         if (mddev->pers->check_reshape == NULL)
4245                 return -EINVAL;
4246         if (raid_disks <= 0 ||
4247             raid_disks >= mddev->max_disks)
4248                 return -EINVAL;
4249         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4250                 return -EBUSY;
4251         mddev->delta_disks = raid_disks - mddev->raid_disks;
4252
4253         rv = mddev->pers->check_reshape(mddev);
4254         return rv;
4255 }
4256
4257
4258 /*
4259  * update_array_info is used to change the configuration of an
4260  * on-line array.
4261  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4262  * fields in the info are checked against the array.
4263  * Any differences that cannot be handled will cause an error.
4264  * Normally, only one change can be managed at a time.
4265  */
4266 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4267 {
4268         int rv = 0;
4269         int cnt = 0;
4270         int state = 0;
4271
4272         /* calculate expected state,ignoring low bits */
4273         if (mddev->bitmap && mddev->bitmap_offset)
4274                 state |= (1 << MD_SB_BITMAP_PRESENT);
4275
4276         if (mddev->major_version != info->major_version ||
4277             mddev->minor_version != info->minor_version ||
4278 /*          mddev->patch_version != info->patch_version || */
4279             mddev->ctime         != info->ctime         ||
4280             mddev->level         != info->level         ||
4281 /*          mddev->layout        != info->layout        || */
4282             !mddev->persistent   != info->not_persistent||
4283             mddev->chunk_size    != info->chunk_size    ||
4284             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4285             ((state^info->state) & 0xfffffe00)
4286                 )
4287                 return -EINVAL;
4288         /* Check there is only one change */
4289         if (info->size >= 0 && mddev->size != info->size) cnt++;
4290         if (mddev->raid_disks != info->raid_disks) cnt++;
4291         if (mddev->layout != info->layout) cnt++;
4292         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4293         if (cnt == 0) return 0;
4294         if (cnt > 1) return -EINVAL;
4295
4296         if (mddev->layout != info->layout) {
4297                 /* Change layout
4298                  * we don't need to do anything at the md level, the
4299                  * personality will take care of it all.
4300                  */
4301                 if (mddev->pers->reconfig == NULL)
4302                         return -EINVAL;
4303                 else
4304                         return mddev->pers->reconfig(mddev, info->layout, -1);
4305         }
4306         if (info->size >= 0 && mddev->size != info->size)
4307                 rv = update_size(mddev, info->size);
4308
4309         if (mddev->raid_disks    != info->raid_disks)
4310                 rv = update_raid_disks(mddev, info->raid_disks);
4311
4312         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4313                 if (mddev->pers->quiesce == NULL)
4314                         return -EINVAL;
4315                 if (mddev->recovery || mddev->sync_thread)
4316                         return -EBUSY;
4317                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4318                         /* add the bitmap */
4319                         if (mddev->bitmap)
4320                                 return -EEXIST;
4321                         if (mddev->default_bitmap_offset == 0)
4322                                 return -EINVAL;
4323                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4324                         mddev->pers->quiesce(mddev, 1);
4325                         rv = bitmap_create(mddev);
4326                         if (rv)
4327                                 bitmap_destroy(mddev);
4328                         mddev->pers->quiesce(mddev, 0);
4329                 } else {
4330                         /* remove the bitmap */
4331                         if (!mddev->bitmap)
4332                                 return -ENOENT;
4333                         if (mddev->bitmap->file)
4334                                 return -EINVAL;
4335                         mddev->pers->quiesce(mddev, 1);
4336                         bitmap_destroy(mddev);
4337                         mddev->pers->quiesce(mddev, 0);
4338                         mddev->bitmap_offset = 0;
4339                 }
4340         }
4341         md_update_sb(mddev, 1);
4342         return rv;
4343 }
4344
4345 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4346 {
4347         mdk_rdev_t *rdev;
4348
4349         if (mddev->pers == NULL)
4350                 return -ENODEV;
4351
4352         rdev = find_rdev(mddev, dev);
4353         if (!rdev)
4354                 return -ENODEV;
4355
4356         md_error(mddev, rdev);
4357         return 0;
4358 }
4359
4360 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4361 {
4362         mddev_t *mddev = bdev->bd_disk->private_data;
4363
4364         geo->heads = 2;
4365         geo->sectors = 4;
4366         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4367         return 0;
4368 }
4369
4370 static int md_ioctl(struct inode *inode, struct file *file,
4371                         unsigned int cmd, unsigned long arg)
4372 {
4373         int err = 0;
4374         void __user *argp = (void __user *)arg;
4375         mddev_t *mddev = NULL;
4376
4377         if (!capable(CAP_SYS_ADMIN))
4378                 return -EACCES;
4379
4380         /*
4381          * Commands dealing with the RAID driver but not any
4382          * particular array:
4383          */
4384         switch (cmd)
4385         {
4386                 case RAID_VERSION:
4387                         err = get_version(argp);
4388                         goto done;
4389
4390                 case PRINT_RAID_DEBUG:
4391                         err = 0;
4392                         md_print_devices();
4393                         goto done;
4394
4395 #ifndef MODULE
4396                 case RAID_AUTORUN:
4397                         err = 0;
4398                         autostart_arrays(arg);
4399                         goto done;
4400 #endif
4401                 default:;
4402         }
4403
4404         /*
4405          * Commands creating/starting a new array:
4406          */
4407
4408         mddev = inode->i_bdev->bd_disk->private_data;
4409
4410         if (!mddev) {
4411                 BUG();
4412                 goto abort;
4413         }
4414
4415         err = mddev_lock(mddev);
4416         if (err) {
4417                 printk(KERN_INFO 
4418                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4419                         err, cmd);
4420                 goto abort;
4421         }
4422
4423         switch (cmd)
4424         {
4425                 case SET_ARRAY_INFO:
4426                         {
4427                                 mdu_array_info_t info;
4428                                 if (!arg)
4429                                         memset(&info, 0, sizeof(info));
4430                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4431                                         err = -EFAULT;
4432                                         goto abort_unlock;
4433                                 }
4434                                 if (mddev->pers) {
4435                                         err = update_array_info(mddev, &info);
4436                                         if (err) {
4437                                                 printk(KERN_WARNING "md: couldn't update"
4438                                                        " array info. %d\n", err);
4439                                                 goto abort_unlock;
4440                                         }
4441                                         goto done_unlock;
4442                                 }
4443                                 if (!list_empty(&mddev->disks)) {
4444                                         printk(KERN_WARNING
4445                                                "md: array %s already has disks!\n",
4446                                                mdname(mddev));
4447                                         err = -EBUSY;
4448                                         goto abort_unlock;
4449                                 }
4450                                 if (mddev->raid_disks) {
4451                                         printk(KERN_WARNING
4452                                                "md: array %s already initialised!\n",
4453                                                mdname(mddev));
4454                                         err = -EBUSY;
4455                                         goto abort_unlock;
4456                                 }
4457                                 err = set_array_info(mddev, &info);
4458                                 if (err) {
4459                                         printk(KERN_WARNING "md: couldn't set"
4460                                                " array info. %d\n", err);
4461                                         goto abort_unlock;
4462                                 }
4463                         }
4464                         goto done_unlock;
4465
4466                 default:;
4467         }
4468
4469         /*
4470          * Commands querying/configuring an existing array:
4471          */
4472         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4473          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4474         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4475                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4476                         && cmd != GET_BITMAP_FILE) {
4477                 err = -ENODEV;
4478                 goto abort_unlock;
4479         }
4480
4481         /*
4482          * Commands even a read-only array can execute:
4483          */
4484         switch (cmd)
4485         {
4486                 case GET_ARRAY_INFO:
4487                         err = get_array_info(mddev, argp);
4488                         goto done_unlock;
4489
4490                 case GET_BITMAP_FILE:
4491                         err = get_bitmap_file(mddev, argp);
4492                         goto done_unlock;
4493
4494                 case GET_DISK_INFO:
4495                         err = get_disk_info(mddev, argp);
4496                         goto done_unlock;
4497
4498                 case RESTART_ARRAY_RW:
4499                         err = restart_array(mddev);
4500                         goto done_unlock;
4501
4502                 case STOP_ARRAY:
4503                         err = do_md_stop (mddev, 0);
4504                         goto done_unlock;
4505
4506                 case STOP_ARRAY_RO:
4507                         err = do_md_stop (mddev, 1);
4508                         goto done_unlock;
4509
4510         /*
4511          * We have a problem here : there is no easy way to give a CHS
4512          * virtual geometry. We currently pretend that we have a 2 heads
4513          * 4 sectors (with a BIG number of cylinders...). This drives
4514          * dosfs just mad... ;-)
4515          */
4516         }
4517
4518         /*
4519          * The remaining ioctls are changing the state of the
4520          * superblock, so we do not allow them on read-only arrays.
4521          * However non-MD ioctls (e.g. get-size) will still come through
4522          * here and hit the 'default' below, so only disallow
4523          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4524          */
4525         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4526             mddev->ro && mddev->pers) {
4527                 if (mddev->ro == 2) {
4528                         mddev->ro = 0;
4529                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4530                 md_wakeup_thread(mddev->thread);
4531
4532                 } else {
4533                         err = -EROFS;
4534                         goto abort_unlock;
4535                 }
4536         }
4537
4538         switch (cmd)
4539         {
4540                 case ADD_NEW_DISK:
4541                 {
4542                         mdu_disk_info_t info;
4543                         if (copy_from_user(&info, argp, sizeof(info)))
4544                                 err = -EFAULT;
4545                         else
4546                                 err = add_new_disk(mddev, &info);
4547                         goto done_unlock;
4548                 }
4549
4550                 case HOT_REMOVE_DISK:
4551                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4552                         goto done_unlock;
4553
4554                 case HOT_ADD_DISK:
4555                         err = hot_add_disk(mddev, new_decode_dev(arg));
4556                         goto done_unlock;
4557
4558                 case SET_DISK_FAULTY:
4559                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4560                         goto done_unlock;
4561
4562                 case RUN_ARRAY:
4563                         err = do_md_run (mddev);
4564                         goto done_unlock;
4565
4566                 case SET_BITMAP_FILE:
4567                         err = set_bitmap_file(mddev, (int)arg);
4568                         goto done_unlock;
4569
4570                 default:
4571                         err = -EINVAL;
4572                         goto abort_unlock;
4573         }
4574
4575 done_unlock:
4576 abort_unlock:
4577         mddev_unlock(mddev);
4578
4579         return err;
4580 done:
4581         if (err)
4582                 MD_BUG();
4583 abort:
4584         return err;
4585 }
4586
4587 static int md_open(struct inode *inode, struct file *file)
4588 {
4589         /*
4590          * Succeed if we can lock the mddev, which confirms that
4591          * it isn't being stopped right now.
4592          */
4593         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4594         int err;
4595
4596         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4597                 goto out;
4598
4599         err = 0;
4600         mddev_get(mddev);
4601         mddev_unlock(mddev);
4602
4603         check_disk_change(inode->i_bdev);
4604  out:
4605         return err;
4606 }
4607
4608 static int md_release(struct inode *inode, struct file * file)
4609 {
4610         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4611
4612         BUG_ON(!mddev);
4613         mddev_put(mddev);
4614
4615         return 0;
4616 }
4617
4618 static int md_media_changed(struct gendisk *disk)
4619 {
4620         mddev_t *mddev = disk->private_data;
4621
4622         return mddev->changed;
4623 }
4624
4625 static int md_revalidate(struct gendisk *disk)
4626 {
4627         mddev_t *mddev = disk->private_data;
4628
4629         mddev->changed = 0;
4630         return 0;
4631 }
4632 static struct block_device_operations md_fops =
4633 {
4634         .owner          = THIS_MODULE,
4635         .open           = md_open,
4636         .release        = md_release,
4637         .ioctl          = md_ioctl,
4638         .getgeo         = md_getgeo,
4639         .media_changed  = md_media_changed,
4640         .revalidate_disk= md_revalidate,
4641 };
4642
4643 static int md_thread(void * arg)
4644 {
4645         mdk_thread_t *thread = arg;
4646
4647         /*
4648          * md_thread is a 'system-thread', it's priority should be very
4649          * high. We avoid resource deadlocks individually in each
4650          * raid personality. (RAID5 does preallocation) We also use RR and
4651          * the very same RT priority as kswapd, thus we will never get
4652          * into a priority inversion deadlock.
4653          *
4654          * we definitely have to have equal or higher priority than
4655          * bdflush, otherwise bdflush will deadlock if there are too
4656          * many dirty RAID5 blocks.
4657          */
4658
4659         allow_signal(SIGKILL);
4660         while (!kthread_should_stop()) {
4661
4662                 /* We need to wait INTERRUPTIBLE so that
4663                  * we don't add to the load-average.
4664                  * That means we need to be sure no signals are
4665                  * pending
4666                  */
4667                 if (signal_pending(current))
4668                         flush_signals(current);
4669
4670                 wait_event_interruptible_timeout
4671                         (thread->wqueue,
4672                          test_bit(THREAD_WAKEUP, &thread->flags)
4673                          || kthread_should_stop(),
4674                          thread->timeout);
4675
4676                 clear_bit(THREAD_WAKEUP, &thread->flags);
4677
4678                 thread->run(thread->mddev);
4679         }
4680
4681         return 0;
4682 }
4683
4684 void md_wakeup_thread(mdk_thread_t *thread)
4685 {
4686         if (thread) {
4687                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4688                 set_bit(THREAD_WAKEUP, &thread->flags);
4689                 wake_up(&thread->wqueue);
4690         }
4691 }
4692
4693 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4694                                  const char *name)
4695 {
4696         mdk_thread_t *thread;
4697
4698         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4699         if (!thread)
4700                 return NULL;
4701
4702         init_waitqueue_head(&thread->wqueue);
4703
4704         thread->run = run;
4705         thread->mddev = mddev;
4706         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4707         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4708         if (IS_ERR(thread->tsk)) {
4709                 kfree(thread);
4710                 return NULL;
4711         }
4712         return thread;
4713 }
4714
4715 void md_unregister_thread(mdk_thread_t *thread)
4716 {
4717         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4718
4719         kthread_stop(thread->tsk);
4720         kfree(thread);
4721 }
4722
4723 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4724 {
4725         if (!mddev) {
4726                 MD_BUG();
4727                 return;
4728         }
4729
4730         if (!rdev || test_bit(Faulty, &rdev->flags))
4731                 return;
4732 /*
4733         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4734                 mdname(mddev),
4735                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4736                 __builtin_return_address(0),__builtin_return_address(1),
4737                 __builtin_return_address(2),__builtin_return_address(3));
4738 */
4739         if (!mddev->pers)
4740                 return;
4741         if (!mddev->pers->error_handler)
4742                 return;
4743         mddev->pers->error_handler(mddev,rdev);
4744         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4745         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4746         md_wakeup_thread(mddev->thread);
4747         md_new_event_inintr(mddev);
4748 }
4749
4750 /* seq_file implementation /proc/mdstat */
4751
4752 static void status_unused(struct seq_file *seq)
4753 {
4754         int i = 0;
4755         mdk_rdev_t *rdev;
4756         struct list_head *tmp;
4757
4758         seq_printf(seq, "unused devices: ");
4759
4760         ITERATE_RDEV_PENDING(rdev,tmp) {
4761                 char b[BDEVNAME_SIZE];
4762                 i++;
4763                 seq_printf(seq, "%s ",
4764                               bdevname(rdev->bdev,b));
4765         }
4766         if (!i)
4767                 seq_printf(seq, "<none>");
4768
4769         seq_printf(seq, "\n");
4770 }
4771
4772
4773 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4774 {
4775         sector_t max_blocks, resync, res;
4776         unsigned long dt, db, rt;
4777         int scale;
4778         unsigned int per_milli;
4779
4780         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4781
4782         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4783                 max_blocks = mddev->resync_max_sectors >> 1;
4784         else
4785                 max_blocks = mddev->size;
4786
4787         /*
4788          * Should not happen.
4789          */
4790         if (!max_blocks) {
4791                 MD_BUG();
4792                 return;
4793         }
4794         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4795          * in a sector_t, and (max_blocks>>scale) will fit in a
4796          * u32, as those are the requirements for sector_div.
4797          * Thus 'scale' must be at least 10
4798          */
4799         scale = 10;
4800         if (sizeof(sector_t) > sizeof(unsigned long)) {
4801                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4802                         scale++;
4803         }
4804         res = (resync>>scale)*1000;
4805         sector_div(res, (u32)((max_blocks>>scale)+1));
4806
4807         per_milli = res;
4808         {
4809                 int i, x = per_milli/50, y = 20-x;
4810                 seq_printf(seq, "[");
4811                 for (i = 0; i < x; i++)
4812                         seq_printf(seq, "=");
4813                 seq_printf(seq, ">");
4814                 for (i = 0; i < y; i++)
4815                         seq_printf(seq, ".");
4816                 seq_printf(seq, "] ");
4817         }
4818         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4819                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4820                     "reshape" :
4821                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4822                      "check" :
4823                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4824                       "resync" : "recovery"))),
4825                    per_milli/10, per_milli % 10,
4826                    (unsigned long long) resync,
4827                    (unsigned long long) max_blocks);
4828
4829         /*
4830          * We do not want to overflow, so the order of operands and
4831          * the * 100 / 100 trick are important. We do a +1 to be
4832          * safe against division by zero. We only estimate anyway.
4833          *
4834          * dt: time from mark until now
4835          * db: blocks written from mark until now
4836          * rt: remaining time
4837          */
4838         dt = ((jiffies - mddev->resync_mark) / HZ);
4839         if (!dt) dt++;
4840         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4841                 - mddev->resync_mark_cnt;
4842         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4843
4844         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4845
4846         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4847 }
4848
4849 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4850 {
4851         struct list_head *tmp;
4852         loff_t l = *pos;
4853         mddev_t *mddev;
4854
4855         if (l >= 0x10000)
4856                 return NULL;
4857         if (!l--)
4858                 /* header */
4859                 return (void*)1;
4860
4861         spin_lock(&all_mddevs_lock);
4862         list_for_each(tmp,&all_mddevs)
4863                 if (!l--) {
4864                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4865                         mddev_get(mddev);
4866                         spin_unlock(&all_mddevs_lock);
4867                         return mddev;
4868                 }
4869         spin_unlock(&all_mddevs_lock);
4870         if (!l--)
4871                 return (void*)2;/* tail */
4872         return NULL;
4873 }
4874
4875 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4876 {
4877         struct list_head *tmp;
4878         mddev_t *next_mddev, *mddev = v;
4879         
4880         ++*pos;
4881         if (v == (void*)2)
4882                 return NULL;
4883
4884         spin_lock(&all_mddevs_lock);
4885         if (v == (void*)1)
4886                 tmp = all_mddevs.next;
4887         else
4888                 tmp = mddev->all_mddevs.next;
4889         if (tmp != &all_mddevs)
4890                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4891         else {
4892                 next_mddev = (void*)2;
4893                 *pos = 0x10000;
4894         }               
4895         spin_unlock(&all_mddevs_lock);
4896
4897         if (v != (void*)1)
4898                 mddev_put(mddev);
4899         return next_mddev;
4900
4901 }
4902
4903 static void md_seq_stop(struct seq_file *seq, void *v)
4904 {
4905         mddev_t *mddev = v;
4906
4907         if (mddev && v != (void*)1 && v != (void*)2)
4908                 mddev_put(mddev);
4909 }
4910
4911 struct mdstat_info {
4912         int event;
4913 };
4914
4915 static int md_seq_show(struct seq_file *seq, void *v)
4916 {
4917         mddev_t *mddev = v;
4918         sector_t size;
4919         struct list_head *tmp2;
4920         mdk_rdev_t *rdev;
4921         struct mdstat_info *mi = seq->private;
4922         struct bitmap *bitmap;
4923
4924         if (v == (void*)1) {
4925                 struct mdk_personality *pers;
4926                 seq_printf(seq, "Personalities : ");
4927                 spin_lock(&pers_lock);
4928                 list_for_each_entry(pers, &pers_list, list)
4929                         seq_printf(seq, "[%s] ", pers->name);
4930
4931                 spin_unlock(&pers_lock);
4932                 seq_printf(seq, "\n");
4933                 mi->event = atomic_read(&md_event_count);
4934                 return 0;
4935         }
4936         if (v == (void*)2) {
4937                 status_unused(seq);
4938                 return 0;
4939         }
4940
4941         if (mddev_lock(mddev) < 0)
4942                 return -EINTR;
4943
4944         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4945                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4946                                                 mddev->pers ? "" : "in");
4947                 if (mddev->pers) {
4948                         if (mddev->ro==1)
4949                                 seq_printf(seq, " (read-only)");
4950                         if (mddev->ro==2)
4951                                 seq_printf(seq, "(auto-read-only)");
4952                         seq_printf(seq, " %s", mddev->pers->name);
4953                 }
4954
4955                 size = 0;
4956                 ITERATE_RDEV(mddev,rdev,tmp2) {
4957                         char b[BDEVNAME_SIZE];
4958                         seq_printf(seq, " %s[%d]",
4959                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4960                         if (test_bit(WriteMostly, &rdev->flags))
4961                                 seq_printf(seq, "(W)");
4962                         if (test_bit(Faulty, &rdev->flags)) {
4963                                 seq_printf(seq, "(F)");
4964                                 continue;
4965                         } else if (rdev->raid_disk < 0)
4966                                 seq_printf(seq, "(S)"); /* spare */
4967                         size += rdev->size;
4968                 }
4969
4970                 if (!list_empty(&mddev->disks)) {
4971                         if (mddev->pers)
4972                                 seq_printf(seq, "\n      %llu blocks",
4973                                         (unsigned long long)mddev->array_size);
4974                         else
4975                                 seq_printf(seq, "\n      %llu blocks",
4976                                         (unsigned long long)size);
4977                 }
4978                 if (mddev->persistent) {
4979                         if (mddev->major_version != 0 ||
4980                             mddev->minor_version != 90) {
4981                                 seq_printf(seq," super %d.%d",
4982                                            mddev->major_version,
4983                                            mddev->minor_version);
4984                         }
4985                 } else
4986                         seq_printf(seq, " super non-persistent");
4987
4988                 if (mddev->pers) {
4989                         mddev->pers->status (seq, mddev);
4990                         seq_printf(seq, "\n      ");
4991                         if (mddev->pers->sync_request) {
4992                                 if (mddev->curr_resync > 2) {
4993                                         status_resync (seq, mddev);
4994                                         seq_printf(seq, "\n      ");
4995                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4996                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4997                                 else if (mddev->recovery_cp < MaxSector)
4998                                         seq_printf(seq, "\tresync=PENDING\n      ");
4999                         }
5000                 } else
5001                         seq_printf(seq, "\n       ");
5002
5003                 if ((bitmap = mddev->bitmap)) {
5004                         unsigned long chunk_kb;
5005                         unsigned long flags;
5006                         spin_lock_irqsave(&bitmap->lock, flags);
5007                         chunk_kb = bitmap->chunksize >> 10;
5008                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5009                                 "%lu%s chunk",
5010                                 bitmap->pages - bitmap->missing_pages,
5011                                 bitmap->pages,
5012                                 (bitmap->pages - bitmap->missing_pages)
5013                                         << (PAGE_SHIFT - 10),
5014                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5015                                 chunk_kb ? "KB" : "B");
5016                         if (bitmap->file) {
5017                                 seq_printf(seq, ", file: ");
5018                                 seq_path(seq, bitmap->file->f_path.mnt,
5019                                          bitmap->file->f_path.dentry," \t\n");
5020                         }
5021
5022                         seq_printf(seq, "\n");
5023                         spin_unlock_irqrestore(&bitmap->lock, flags);
5024                 }
5025
5026                 seq_printf(seq, "\n");
5027         }
5028         mddev_unlock(mddev);
5029         
5030         return 0;
5031 }
5032
5033 static struct seq_operations md_seq_ops = {
5034         .start  = md_seq_start,
5035         .next   = md_seq_next,
5036         .stop   = md_seq_stop,
5037         .show   = md_seq_show,
5038 };
5039
5040 static int md_seq_open(struct inode *inode, struct file *file)
5041 {
5042         int error;
5043         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5044         if (mi == NULL)
5045                 return -ENOMEM;
5046
5047         error = seq_open(file, &md_seq_ops);
5048         if (error)
5049                 kfree(mi);
5050         else {
5051                 struct seq_file *p = file->private_data;
5052                 p->private = mi;
5053                 mi->event = atomic_read(&md_event_count);
5054         }
5055         return error;
5056 }
5057
5058 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5059 {
5060         struct seq_file *m = filp->private_data;
5061         struct mdstat_info *mi = m->private;
5062         int mask;
5063
5064         poll_wait(filp, &md_event_waiters, wait);
5065
5066         /* always allow read */
5067         mask = POLLIN | POLLRDNORM;
5068
5069         if (mi->event != atomic_read(&md_event_count))
5070                 mask |= POLLERR | POLLPRI;
5071         return mask;
5072 }
5073
5074 static const struct file_operations md_seq_fops = {
5075         .owner          = THIS_MODULE,
5076         .open           = md_seq_open,
5077         .read           = seq_read,
5078         .llseek         = seq_lseek,
5079         .release        = seq_release_private,
5080         .poll           = mdstat_poll,
5081 };
5082
5083 int register_md_personality(struct mdk_personality *p)
5084 {
5085         spin_lock(&pers_lock);
5086         list_add_tail(&p->list, &pers_list);
5087         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5088         spin_unlock(&pers_lock);
5089         return 0;
5090 }
5091
5092 int unregister_md_personality(struct mdk_personality *p)
5093 {
5094         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5095         spin_lock(&pers_lock);
5096         list_del_init(&p->list);
5097         spin_unlock(&pers_lock);
5098         return 0;
5099 }
5100
5101 static int is_mddev_idle(mddev_t *mddev)
5102 {
5103         mdk_rdev_t * rdev;
5104         struct list_head *tmp;
5105         int idle;
5106         long curr_events;
5107
5108         idle = 1;
5109         ITERATE_RDEV(mddev,rdev,tmp) {
5110                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5111                 curr_events = disk_stat_read(disk, sectors[0]) + 
5112                                 disk_stat_read(disk, sectors[1]) - 
5113                                 atomic_read(&disk->sync_io);
5114                 /* sync IO will cause sync_io to increase before the disk_stats
5115                  * as sync_io is counted when a request starts, and
5116                  * disk_stats is counted when it completes.
5117                  * So resync activity will cause curr_events to be smaller than
5118                  * when there was no such activity.
5119                  * non-sync IO will cause disk_stat to increase without
5120                  * increasing sync_io so curr_events will (eventually)
5121                  * be larger than it was before.  Once it becomes
5122                  * substantially larger, the test below will cause
5123                  * the array to appear non-idle, and resync will slow
5124                  * down.
5125                  * If there is a lot of outstanding resync activity when
5126                  * we set last_event to curr_events, then all that activity
5127                  * completing might cause the array to appear non-idle
5128                  * and resync will be slowed down even though there might
5129                  * not have been non-resync activity.  This will only
5130                  * happen once though.  'last_events' will soon reflect
5131                  * the state where there is little or no outstanding
5132                  * resync requests, and further resync activity will
5133                  * always make curr_events less than last_events.
5134                  *
5135                  */
5136                 if (curr_events - rdev->last_events > 4096) {
5137                         rdev->last_events = curr_events;
5138                         idle = 0;
5139                 }
5140         }
5141         return idle;
5142 }
5143
5144 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5145 {
5146         /* another "blocks" (512byte) blocks have been synced */
5147         atomic_sub(blocks, &mddev->recovery_active);
5148         wake_up(&mddev->recovery_wait);
5149         if (!ok) {
5150                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5151                 md_wakeup_thread(mddev->thread);
5152                 // stop recovery, signal do_sync ....
5153         }
5154 }
5155
5156
5157 /* md_write_start(mddev, bi)
5158  * If we need to update some array metadata (e.g. 'active' flag
5159  * in superblock) before writing, schedule a superblock update
5160  * and wait for it to complete.
5161  */
5162 void md_write_start(mddev_t *mddev, struct bio *bi)
5163 {
5164         if (bio_data_dir(bi) != WRITE)
5165                 return;
5166
5167         BUG_ON(mddev->ro == 1);
5168         if (mddev->ro == 2) {
5169                 /* need to switch to read/write */
5170                 mddev->ro = 0;
5171                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5172                 md_wakeup_thread(mddev->thread);
5173         }
5174         atomic_inc(&mddev->writes_pending);
5175         if (mddev->in_sync) {
5176                 spin_lock_irq(&mddev->write_lock);
5177                 if (mddev->in_sync) {
5178                         mddev->in_sync = 0;
5179                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5180                         md_wakeup_thread(mddev->thread);
5181                 }
5182                 spin_unlock_irq(&mddev->write_lock);
5183         }
5184         wait_event(mddev->sb_wait, mddev->flags==0);
5185 }
5186
5187 void md_write_end(mddev_t *mddev)
5188 {
5189         if (atomic_dec_and_test(&mddev->writes_pending)) {
5190                 if (mddev->safemode == 2)
5191                         md_wakeup_thread(mddev->thread);
5192                 else if (mddev->safemode_delay)
5193                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5194         }
5195 }
5196
5197 /* md_allow_write(mddev)
5198  * Calling this ensures that the array is marked 'active' so that writes
5199  * may proceed without blocking.  It is important to call this before
5200  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5201  * Must be called with mddev_lock held.
5202  */
5203 void md_allow_write(mddev_t *mddev)
5204 {
5205         if (!mddev->pers)
5206                 return;
5207         if (mddev->ro)
5208                 return;
5209
5210         spin_lock_irq(&mddev->write_lock);
5211         if (mddev->in_sync) {
5212                 mddev->in_sync = 0;
5213                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5214                 if (mddev->safemode_delay &&
5215                     mddev->safemode == 0)
5216                         mddev->safemode = 1;
5217                 spin_unlock_irq(&mddev->write_lock);
5218                 md_update_sb(mddev, 0);
5219         } else
5220                 spin_unlock_irq(&mddev->write_lock);
5221 }
5222 EXPORT_SYMBOL_GPL(md_allow_write);
5223
5224 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5225
5226 #define SYNC_MARKS      10
5227 #define SYNC_MARK_STEP  (3*HZ)
5228 void md_do_sync(mddev_t *mddev)
5229 {
5230         mddev_t *mddev2;
5231         unsigned int currspeed = 0,
5232                  window;
5233         sector_t max_sectors,j, io_sectors;
5234         unsigned long mark[SYNC_MARKS];
5235         sector_t mark_cnt[SYNC_MARKS];
5236         int last_mark,m;
5237         struct list_head *tmp;
5238         sector_t last_check;
5239         int skipped = 0;
5240         struct list_head *rtmp;
5241         mdk_rdev_t *rdev;
5242         char *desc;
5243
5244         /* just incase thread restarts... */
5245         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5246                 return;
5247         if (mddev->ro) /* never try to sync a read-only array */
5248                 return;
5249
5250         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5251                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5252                         desc = "data-check";
5253                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5254                         desc = "requested-resync";
5255                 else
5256                         desc = "resync";
5257         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5258                 desc = "reshape";
5259         else
5260                 desc = "recovery";
5261
5262         /* we overload curr_resync somewhat here.
5263          * 0 == not engaged in resync at all
5264          * 2 == checking that there is no conflict with another sync
5265          * 1 == like 2, but have yielded to allow conflicting resync to
5266          *              commense
5267          * other == active in resync - this many blocks
5268          *
5269          * Before starting a resync we must have set curr_resync to
5270          * 2, and then checked that every "conflicting" array has curr_resync
5271          * less than ours.  When we find one that is the same or higher
5272          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5273          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5274          * This will mean we have to start checking from the beginning again.
5275          *
5276          */
5277
5278         do {
5279                 mddev->curr_resync = 2;
5280
5281         try_again:
5282                 if (kthread_should_stop()) {
5283                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5284                         goto skip;
5285                 }
5286                 ITERATE_MDDEV(mddev2,tmp) {
5287                         if (mddev2 == mddev)
5288                                 continue;
5289                         if (mddev2->curr_resync && 
5290                             match_mddev_units(mddev,mddev2)) {
5291                                 DEFINE_WAIT(wq);
5292                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5293                                         /* arbitrarily yield */
5294                                         mddev->curr_resync = 1;
5295                                         wake_up(&resync_wait);
5296                                 }
5297                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5298                                         /* no need to wait here, we can wait the next
5299                                          * time 'round when curr_resync == 2
5300                                          */
5301                                         continue;
5302                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5303                                 if (!kthread_should_stop() &&
5304                                     mddev2->curr_resync >= mddev->curr_resync) {
5305                                         printk(KERN_INFO "md: delaying %s of %s"
5306                                                " until %s has finished (they"
5307                                                " share one or more physical units)\n",
5308                                                desc, mdname(mddev), mdname(mddev2));
5309                                         mddev_put(mddev2);
5310                                         schedule();
5311                                         finish_wait(&resync_wait, &wq);
5312                                         goto try_again;
5313                                 }
5314                                 finish_wait(&resync_wait, &wq);
5315                         }
5316                 }
5317         } while (mddev->curr_resync < 2);
5318
5319         j = 0;
5320         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5321                 /* resync follows the size requested by the personality,
5322                  * which defaults to physical size, but can be virtual size
5323                  */
5324                 max_sectors = mddev->resync_max_sectors;
5325                 mddev->resync_mismatches = 0;
5326                 /* we don't use the checkpoint if there's a bitmap */
5327                 if (!mddev->bitmap &&
5328                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5329                         j = mddev->recovery_cp;
5330         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5331                 max_sectors = mddev->size << 1;
5332         else {
5333                 /* recovery follows the physical size of devices */
5334                 max_sectors = mddev->size << 1;
5335                 j = MaxSector;
5336                 ITERATE_RDEV(mddev,rdev,rtmp)
5337                         if (rdev->raid_disk >= 0 &&
5338                             !test_bit(Faulty, &rdev->flags) &&
5339                             !test_bit(In_sync, &rdev->flags) &&
5340                             rdev->recovery_offset < j)
5341                                 j = rdev->recovery_offset;
5342         }
5343
5344         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5345         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5346                 " %d KB/sec/disk.\n", speed_min(mddev));
5347         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5348                "(but not more than %d KB/sec) for %s.\n",
5349                speed_max(mddev), desc);
5350
5351         is_mddev_idle(mddev); /* this also initializes IO event counters */
5352
5353         io_sectors = 0;
5354         for (m = 0; m < SYNC_MARKS; m++) {
5355                 mark[m] = jiffies;
5356                 mark_cnt[m] = io_sectors;
5357         }
5358         last_mark = 0;
5359         mddev->resync_mark = mark[last_mark];
5360         mddev->resync_mark_cnt = mark_cnt[last_mark];
5361
5362         /*
5363          * Tune reconstruction:
5364          */
5365         window = 32*(PAGE_SIZE/512);
5366         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5367                 window/2,(unsigned long long) max_sectors/2);
5368
5369         atomic_set(&mddev->recovery_active, 0);
5370         init_waitqueue_head(&mddev->recovery_wait);
5371         last_check = 0;
5372
5373         if (j>2) {
5374                 printk(KERN_INFO 
5375                        "md: resuming %s of %s from checkpoint.\n",
5376                        desc, mdname(mddev));
5377                 mddev->curr_resync = j;
5378         }
5379
5380         while (j < max_sectors) {
5381                 sector_t sectors;
5382
5383                 skipped = 0;
5384                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5385                                             currspeed < speed_min(mddev));
5386                 if (sectors == 0) {
5387                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5388                         goto out;
5389                 }
5390
5391                 if (!skipped) { /* actual IO requested */
5392                         io_sectors += sectors;
5393                         atomic_add(sectors, &mddev->recovery_active);
5394                 }
5395
5396                 j += sectors;
5397                 if (j>1) mddev->curr_resync = j;
5398                 mddev->curr_mark_cnt = io_sectors;
5399                 if (last_check == 0)
5400                         /* this is the earliers that rebuilt will be
5401                          * visible in /proc/mdstat
5402                          */
5403                         md_new_event(mddev);
5404
5405                 if (last_check + window > io_sectors || j == max_sectors)
5406                         continue;
5407
5408                 last_check = io_sectors;
5409
5410                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5411                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5412                         break;
5413
5414         repeat:
5415                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5416                         /* step marks */
5417                         int next = (last_mark+1) % SYNC_MARKS;
5418
5419                         mddev->resync_mark = mark[next];
5420                         mddev->resync_mark_cnt = mark_cnt[next];
5421                         mark[next] = jiffies;
5422                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5423                         last_mark = next;
5424                 }
5425
5426
5427                 if (kthread_should_stop()) {
5428                         /*
5429                          * got a signal, exit.
5430                          */
5431                         printk(KERN_INFO 
5432                                 "md: md_do_sync() got signal ... exiting\n");
5433                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5434                         goto out;
5435                 }
5436
5437                 /*
5438                  * this loop exits only if either when we are slower than
5439                  * the 'hard' speed limit, or the system was IO-idle for
5440                  * a jiffy.
5441                  * the system might be non-idle CPU-wise, but we only care
5442                  * about not overloading the IO subsystem. (things like an
5443                  * e2fsck being done on the RAID array should execute fast)
5444                  */
5445                 blk_unplug(mddev->queue);
5446                 cond_resched();
5447
5448                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5449                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5450
5451                 if (currspeed > speed_min(mddev)) {
5452                         if ((currspeed > speed_max(mddev)) ||
5453                                         !is_mddev_idle(mddev)) {
5454                                 msleep(500);
5455                                 goto repeat;
5456                         }
5457                 }
5458         }
5459         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5460         /*
5461          * this also signals 'finished resyncing' to md_stop
5462          */
5463  out:
5464         blk_unplug(mddev->queue);
5465
5466         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5467
5468         /* tell personality that we are finished */
5469         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5470
5471         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5472             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5473             mddev->curr_resync > 2) {
5474                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5475                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5476                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5477                                         printk(KERN_INFO
5478                                                "md: checkpointing %s of %s.\n",
5479                                                desc, mdname(mddev));
5480                                         mddev->recovery_cp = mddev->curr_resync;
5481                                 }
5482                         } else
5483                                 mddev->recovery_cp = MaxSector;
5484                 } else {
5485                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5486                                 mddev->curr_resync = MaxSector;
5487                         ITERATE_RDEV(mddev,rdev,rtmp)
5488                                 if (rdev->raid_disk >= 0 &&
5489                                     !test_bit(Faulty, &rdev->flags) &&
5490                                     !test_bit(In_sync, &rdev->flags) &&
5491                                     rdev->recovery_offset < mddev->curr_resync)
5492                                         rdev->recovery_offset = mddev->curr_resync;
5493                 }
5494         }
5495         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5496
5497  skip:
5498         mddev->curr_resync = 0;
5499         wake_up(&resync_wait);
5500         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5501         md_wakeup_thread(mddev->thread);
5502 }
5503 EXPORT_SYMBOL_GPL(md_do_sync);
5504
5505
5506 static int remove_and_add_spares(mddev_t *mddev)
5507 {
5508         mdk_rdev_t *rdev;
5509         struct list_head *rtmp;
5510         int spares = 0;
5511
5512         ITERATE_RDEV(mddev,rdev,rtmp)
5513                 if (rdev->raid_disk >= 0 &&
5514                     (test_bit(Faulty, &rdev->flags) ||
5515                      ! test_bit(In_sync, &rdev->flags)) &&
5516                     atomic_read(&rdev->nr_pending)==0) {
5517                         if (mddev->pers->hot_remove_disk(
5518                                     mddev, rdev->raid_disk)==0) {
5519                                 char nm[20];
5520                                 sprintf(nm,"rd%d", rdev->raid_disk);
5521                                 sysfs_remove_link(&mddev->kobj, nm);
5522                                 rdev->raid_disk = -1;
5523                         }
5524                 }
5525
5526         if (mddev->degraded) {
5527                 ITERATE_RDEV(mddev,rdev,rtmp)
5528                         if (rdev->raid_disk < 0
5529                             && !test_bit(Faulty, &rdev->flags)) {
5530                                 rdev->recovery_offset = 0;
5531                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5532                                         char nm[20];
5533                                         sprintf(nm, "rd%d", rdev->raid_disk);
5534                                         if (sysfs_create_link(&mddev->kobj,
5535                                                               &rdev->kobj, nm))
5536                                                 printk(KERN_WARNING
5537                                                        "md: cannot register "
5538                                                        "%s for %s\n",
5539                                                        nm, mdname(mddev));
5540                                         spares++;
5541                                         md_new_event(mddev);
5542                                 } else
5543                                         break;
5544                         }
5545         }
5546         return spares;
5547 }
5548 /*
5549  * This routine is regularly called by all per-raid-array threads to
5550  * deal with generic issues like resync and super-block update.
5551  * Raid personalities that don't have a thread (linear/raid0) do not
5552  * need this as they never do any recovery or update the superblock.
5553  *
5554  * It does not do any resync itself, but rather "forks" off other threads
5555  * to do that as needed.
5556  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5557  * "->recovery" and create a thread at ->sync_thread.
5558  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5559  * and wakeups up this thread which will reap the thread and finish up.
5560  * This thread also removes any faulty devices (with nr_pending == 0).
5561  *
5562  * The overall approach is:
5563  *  1/ if the superblock needs updating, update it.
5564  *  2/ If a recovery thread is running, don't do anything else.
5565  *  3/ If recovery has finished, clean up, possibly marking spares active.
5566  *  4/ If there are any faulty devices, remove them.
5567  *  5/ If array is degraded, try to add spares devices
5568  *  6/ If array has spares or is not in-sync, start a resync thread.
5569  */
5570 void md_check_recovery(mddev_t *mddev)
5571 {
5572         mdk_rdev_t *rdev;
5573         struct list_head *rtmp;
5574
5575
5576         if (mddev->bitmap)
5577                 bitmap_daemon_work(mddev->bitmap);
5578
5579         if (mddev->ro)
5580                 return;
5581
5582         if (signal_pending(current)) {
5583                 if (mddev->pers->sync_request) {
5584                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5585                                mdname(mddev));
5586                         mddev->safemode = 2;
5587                 }
5588                 flush_signals(current);
5589         }
5590
5591         if ( ! (
5592                 mddev->flags ||
5593                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5594                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5595                 (mddev->safemode == 1) ||
5596                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5597                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5598                 ))
5599                 return;
5600
5601         if (mddev_trylock(mddev)) {
5602                 int spares = 0;
5603
5604                 spin_lock_irq(&mddev->write_lock);
5605                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5606                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5607                         mddev->in_sync = 1;
5608                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5609                 }
5610                 if (mddev->safemode == 1)
5611                         mddev->safemode = 0;
5612                 spin_unlock_irq(&mddev->write_lock);
5613
5614                 if (mddev->flags)
5615                         md_update_sb(mddev, 0);
5616
5617
5618                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5619                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5620                         /* resync/recovery still happening */
5621                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5622                         goto unlock;
5623                 }
5624                 if (mddev->sync_thread) {
5625                         /* resync has finished, collect result */
5626                         md_unregister_thread(mddev->sync_thread);
5627                         mddev->sync_thread = NULL;
5628                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5629                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5630                                 /* success...*/
5631                                 /* activate any spares */
5632                                 mddev->pers->spare_active(mddev);
5633                         }
5634                         md_update_sb(mddev, 1);
5635
5636                         /* if array is no-longer degraded, then any saved_raid_disk
5637                          * information must be scrapped
5638                          */
5639                         if (!mddev->degraded)
5640                                 ITERATE_RDEV(mddev,rdev,rtmp)
5641                                         rdev->saved_raid_disk = -1;
5642
5643                         mddev->recovery = 0;
5644                         /* flag recovery needed just to double check */
5645                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5646                         md_new_event(mddev);
5647                         goto unlock;
5648                 }
5649                 /* Clear some bits that don't mean anything, but
5650                  * might be left set
5651                  */
5652                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5653                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5654                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5655                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5656
5657                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5658                         goto unlock;
5659                 /* no recovery is running.
5660                  * remove any failed drives, then
5661                  * add spares if possible.
5662                  * Spare are also removed and re-added, to allow
5663                  * the personality to fail the re-add.
5664                  */
5665
5666                 if (mddev->reshape_position != MaxSector) {
5667                         if (mddev->pers->check_reshape(mddev) != 0)
5668                                 /* Cannot proceed */
5669                                 goto unlock;
5670                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5671                 } else if ((spares = remove_and_add_spares(mddev))) {
5672                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5673                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5674                 } else if (mddev->recovery_cp < MaxSector) {
5675                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5676                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5677                         /* nothing to be done ... */
5678                         goto unlock;
5679
5680                 if (mddev->pers->sync_request) {
5681                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5682                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5683                                 /* We are adding a device or devices to an array
5684                                  * which has the bitmap stored on all devices.
5685                                  * So make sure all bitmap pages get written
5686                                  */
5687                                 bitmap_write_all(mddev->bitmap);
5688                         }
5689                         mddev->sync_thread = md_register_thread(md_do_sync,
5690                                                                 mddev,
5691                                                                 "%s_resync");
5692                         if (!mddev->sync_thread) {
5693                                 printk(KERN_ERR "%s: could not start resync"
5694                                         " thread...\n", 
5695                                         mdname(mddev));
5696                                 /* leave the spares where they are, it shouldn't hurt */
5697                                 mddev->recovery = 0;
5698                         } else
5699                                 md_wakeup_thread(mddev->sync_thread);
5700                         md_new_event(mddev);
5701                 }
5702         unlock:
5703                 mddev_unlock(mddev);
5704         }
5705 }
5706
5707 static int md_notify_reboot(struct notifier_block *this,
5708                             unsigned long code, void *x)
5709 {
5710         struct list_head *tmp;
5711         mddev_t *mddev;
5712
5713         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5714
5715                 printk(KERN_INFO "md: stopping all md devices.\n");
5716
5717                 ITERATE_MDDEV(mddev,tmp)
5718                         if (mddev_trylock(mddev)) {
5719                                 do_md_stop (mddev, 1);
5720                                 mddev_unlock(mddev);
5721                         }
5722                 /*
5723                  * certain more exotic SCSI devices are known to be
5724                  * volatile wrt too early system reboots. While the
5725                  * right place to handle this issue is the given
5726                  * driver, we do want to have a safe RAID driver ...
5727                  */
5728                 mdelay(1000*1);
5729         }
5730         return NOTIFY_DONE;
5731 }
5732
5733 static struct notifier_block md_notifier = {
5734         .notifier_call  = md_notify_reboot,
5735         .next           = NULL,
5736         .priority       = INT_MAX, /* before any real devices */
5737 };
5738
5739 static void md_geninit(void)
5740 {
5741         struct proc_dir_entry *p;
5742
5743         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5744
5745         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5746         if (p)
5747                 p->proc_fops = &md_seq_fops;
5748 }
5749
5750 static int __init md_init(void)
5751 {
5752         if (register_blkdev(MAJOR_NR, "md"))
5753                 return -1;
5754         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5755                 unregister_blkdev(MAJOR_NR, "md");
5756                 return -1;
5757         }
5758         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5759                             md_probe, NULL, NULL);
5760         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5761                             md_probe, NULL, NULL);
5762
5763         register_reboot_notifier(&md_notifier);
5764         raid_table_header = register_sysctl_table(raid_root_table);
5765
5766         md_geninit();
5767         return (0);
5768 }
5769
5770
5771 #ifndef MODULE
5772
5773 /*
5774  * Searches all registered partitions for autorun RAID arrays
5775  * at boot time.
5776  */
5777
5778 static LIST_HEAD(all_detected_devices);
5779 struct detected_devices_node {
5780         struct list_head list;
5781         dev_t dev;
5782 };
5783
5784 void md_autodetect_dev(dev_t dev)
5785 {
5786         struct detected_devices_node *node_detected_dev;
5787
5788         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5789         if (node_detected_dev) {
5790                 node_detected_dev->dev = dev;
5791                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5792         } else {
5793                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5794                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5795         }
5796 }
5797
5798
5799 static void autostart_arrays(int part)
5800 {
5801         mdk_rdev_t *rdev;
5802         struct detected_devices_node *node_detected_dev;
5803         dev_t dev;
5804         int i_scanned, i_passed;
5805
5806         i_scanned = 0;
5807         i_passed = 0;
5808
5809         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5810
5811         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
5812                 i_scanned++;
5813                 node_detected_dev = list_entry(all_detected_devices.next,
5814                                         struct detected_devices_node, list);
5815                 list_del(&node_detected_dev->list);
5816                 dev = node_detected_dev->dev;
5817                 kfree(node_detected_dev);
5818                 rdev = md_import_device(dev,0, 90);
5819                 if (IS_ERR(rdev))
5820                         continue;
5821
5822                 if (test_bit(Faulty, &rdev->flags)) {
5823                         MD_BUG();
5824                         continue;
5825                 }
5826                 list_add(&rdev->same_set, &pending_raid_disks);
5827                 i_passed++;
5828         }
5829
5830         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
5831                                                 i_scanned, i_passed);
5832
5833         autorun_devices(part);
5834 }
5835
5836 #endif /* !MODULE */
5837
5838 static __exit void md_exit(void)
5839 {
5840         mddev_t *mddev;
5841         struct list_head *tmp;
5842
5843         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5844         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5845
5846         unregister_blkdev(MAJOR_NR,"md");
5847         unregister_blkdev(mdp_major, "mdp");
5848         unregister_reboot_notifier(&md_notifier);
5849         unregister_sysctl_table(raid_table_header);
5850         remove_proc_entry("mdstat", NULL);
5851         ITERATE_MDDEV(mddev,tmp) {
5852                 struct gendisk *disk = mddev->gendisk;
5853                 if (!disk)
5854                         continue;
5855                 export_array(mddev);
5856                 del_gendisk(disk);
5857                 put_disk(disk);
5858                 mddev->gendisk = NULL;
5859                 mddev_put(mddev);
5860         }
5861 }
5862
5863 subsys_initcall(md_init);
5864 module_exit(md_exit)
5865
5866 static int get_ro(char *buffer, struct kernel_param *kp)
5867 {
5868         return sprintf(buffer, "%d", start_readonly);
5869 }
5870 static int set_ro(const char *val, struct kernel_param *kp)
5871 {
5872         char *e;
5873         int num = simple_strtoul(val, &e, 10);
5874         if (*val && (*e == '\0' || *e == '\n')) {
5875                 start_readonly = num;
5876                 return 0;
5877         }
5878         return -EINVAL;
5879 }
5880
5881 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5882 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5883
5884
5885 EXPORT_SYMBOL(register_md_personality);
5886 EXPORT_SYMBOL(unregister_md_personality);
5887 EXPORT_SYMBOL(md_error);
5888 EXPORT_SYMBOL(md_done_sync);
5889 EXPORT_SYMBOL(md_write_start);
5890 EXPORT_SYMBOL(md_write_end);
5891 EXPORT_SYMBOL(md_register_thread);
5892 EXPORT_SYMBOL(md_unregister_thread);
5893 EXPORT_SYMBOL(md_wakeup_thread);
5894 EXPORT_SYMBOL(md_check_recovery);
5895 MODULE_LICENSE("GPL");
5896 MODULE_ALIAS("md");
5897 MODULE_ALIAS_BLOCKDEV_MAJOR(