[PATCH] md: helper function to match commands written to sysfs files
[linux-3.10.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/config.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/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
46
47 #include <linux/init.h>
48
49 #include <linux/file.h>
50
51 #ifdef CONFIG_KMOD
52 #include <linux/kmod.h>
53 #endif
54
55 #include <asm/unaligned.h>
56
57 #define MAJOR_NR MD_MAJOR
58 #define MD_DRIVER
59
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
62
63 #define DEBUG 0
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67 #ifndef MODULE
68 static void autostart_arrays (int part);
69 #endif
70
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
73
74 /*
75  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76  * is 1000 KB/sec, so the extra system load does not show up that much.
77  * Increase it if you want to have more _guaranteed_ speed. Note that
78  * the RAID driver will use the maximum available bandwidth if the IO
79  * subsystem is idle. There is also an 'absolute maximum' reconstruction
80  * speed limit - in case reconstruction slows down your system despite
81  * idle IO detection.
82  *
83  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84  */
85
86 static int sysctl_speed_limit_min = 1000;
87 static int sysctl_speed_limit_max = 200000;
88
89 static struct ctl_table_header *raid_table_header;
90
91 static ctl_table raid_table[] = {
92         {
93                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
94                 .procname       = "speed_limit_min",
95                 .data           = &sysctl_speed_limit_min,
96                 .maxlen         = sizeof(int),
97                 .mode           = 0644,
98                 .proc_handler   = &proc_dointvec,
99         },
100         {
101                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
102                 .procname       = "speed_limit_max",
103                 .data           = &sysctl_speed_limit_max,
104                 .maxlen         = sizeof(int),
105                 .mode           = 0644,
106                 .proc_handler   = &proc_dointvec,
107         },
108         { .ctl_name = 0 }
109 };
110
111 static ctl_table raid_dir_table[] = {
112         {
113                 .ctl_name       = DEV_RAID,
114                 .procname       = "raid",
115                 .maxlen         = 0,
116                 .mode           = 0555,
117                 .child          = raid_table,
118         },
119         { .ctl_name = 0 }
120 };
121
122 static ctl_table raid_root_table[] = {
123         {
124                 .ctl_name       = CTL_DEV,
125                 .procname       = "dev",
126                 .maxlen         = 0,
127                 .mode           = 0555,
128                 .child          = raid_dir_table,
129         },
130         { .ctl_name = 0 }
131 };
132
133 static struct block_device_operations md_fops;
134
135 static int start_readonly;
136
137 /*
138  * We have a system wide 'event count' that is incremented
139  * on any 'interesting' event, and readers of /proc/mdstat
140  * can use 'poll' or 'select' to find out when the event
141  * count increases.
142  *
143  * Events are:
144  *  start array, stop array, error, add device, remove device,
145  *  start build, activate spare
146  */
147 DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
148 static atomic_t md_event_count;
149 void md_new_event(mddev_t *mddev)
150 {
151         atomic_inc(&md_event_count);
152         wake_up(&md_event_waiters);
153 }
154
155 /*
156  * Enables to iterate over all existing md arrays
157  * all_mddevs_lock protects this list.
158  */
159 static LIST_HEAD(all_mddevs);
160 static DEFINE_SPINLOCK(all_mddevs_lock);
161
162
163 /*
164  * iterates through all used mddevs in the system.
165  * We take care to grab the all_mddevs_lock whenever navigating
166  * the list, and to always hold a refcount when unlocked.
167  * Any code which breaks out of this loop while own
168  * a reference to the current mddev and must mddev_put it.
169  */
170 #define ITERATE_MDDEV(mddev,tmp)                                        \
171                                                                         \
172         for (({ spin_lock(&all_mddevs_lock);                            \
173                 tmp = all_mddevs.next;                                  \
174                 mddev = NULL;});                                        \
175              ({ if (tmp != &all_mddevs)                                 \
176                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
177                 spin_unlock(&all_mddevs_lock);                          \
178                 if (mddev) mddev_put(mddev);                            \
179                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
180                 tmp != &all_mddevs;});                                  \
181              ({ spin_lock(&all_mddevs_lock);                            \
182                 tmp = tmp->next;})                                      \
183                 )
184
185
186 static int md_fail_request (request_queue_t *q, struct bio *bio)
187 {
188         bio_io_error(bio, bio->bi_size);
189         return 0;
190 }
191
192 static inline mddev_t *mddev_get(mddev_t *mddev)
193 {
194         atomic_inc(&mddev->active);
195         return mddev;
196 }
197
198 static void mddev_put(mddev_t *mddev)
199 {
200         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
201                 return;
202         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
203                 list_del(&mddev->all_mddevs);
204                 blk_put_queue(mddev->queue);
205                 kobject_unregister(&mddev->kobj);
206         }
207         spin_unlock(&all_mddevs_lock);
208 }
209
210 static mddev_t * mddev_find(dev_t unit)
211 {
212         mddev_t *mddev, *new = NULL;
213
214  retry:
215         spin_lock(&all_mddevs_lock);
216         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
217                 if (mddev->unit == unit) {
218                         mddev_get(mddev);
219                         spin_unlock(&all_mddevs_lock);
220                         kfree(new);
221                         return mddev;
222                 }
223
224         if (new) {
225                 list_add(&new->all_mddevs, &all_mddevs);
226                 spin_unlock(&all_mddevs_lock);
227                 return new;
228         }
229         spin_unlock(&all_mddevs_lock);
230
231         new = kzalloc(sizeof(*new), GFP_KERNEL);
232         if (!new)
233                 return NULL;
234
235         new->unit = unit;
236         if (MAJOR(unit) == MD_MAJOR)
237                 new->md_minor = MINOR(unit);
238         else
239                 new->md_minor = MINOR(unit) >> MdpMinorShift;
240
241         init_MUTEX(&new->reconfig_sem);
242         INIT_LIST_HEAD(&new->disks);
243         INIT_LIST_HEAD(&new->all_mddevs);
244         init_timer(&new->safemode_timer);
245         atomic_set(&new->active, 1);
246         spin_lock_init(&new->write_lock);
247         init_waitqueue_head(&new->sb_wait);
248
249         new->queue = blk_alloc_queue(GFP_KERNEL);
250         if (!new->queue) {
251                 kfree(new);
252                 return NULL;
253         }
254
255         blk_queue_make_request(new->queue, md_fail_request);
256
257         goto retry;
258 }
259
260 static inline int mddev_lock(mddev_t * mddev)
261 {
262         return down_interruptible(&mddev->reconfig_sem);
263 }
264
265 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
266 {
267         down(&mddev->reconfig_sem);
268 }
269
270 static inline int mddev_trylock(mddev_t * mddev)
271 {
272         return down_trylock(&mddev->reconfig_sem);
273 }
274
275 static inline void mddev_unlock(mddev_t * mddev)
276 {
277         up(&mddev->reconfig_sem);
278
279         md_wakeup_thread(mddev->thread);
280 }
281
282 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
283 {
284         mdk_rdev_t * rdev;
285         struct list_head *tmp;
286
287         ITERATE_RDEV(mddev,rdev,tmp) {
288                 if (rdev->desc_nr == nr)
289                         return rdev;
290         }
291         return NULL;
292 }
293
294 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
295 {
296         struct list_head *tmp;
297         mdk_rdev_t *rdev;
298
299         ITERATE_RDEV(mddev,rdev,tmp) {
300                 if (rdev->bdev->bd_dev == dev)
301                         return rdev;
302         }
303         return NULL;
304 }
305
306 static struct mdk_personality *find_pers(int level)
307 {
308         struct mdk_personality *pers;
309         list_for_each_entry(pers, &pers_list, list)
310                 if (pers->level == level)
311                         return pers;
312         return NULL;
313 }
314
315 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
316 {
317         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
318         return MD_NEW_SIZE_BLOCKS(size);
319 }
320
321 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
322 {
323         sector_t size;
324
325         size = rdev->sb_offset;
326
327         if (chunk_size)
328                 size &= ~((sector_t)chunk_size/1024 - 1);
329         return size;
330 }
331
332 static int alloc_disk_sb(mdk_rdev_t * rdev)
333 {
334         if (rdev->sb_page)
335                 MD_BUG();
336
337         rdev->sb_page = alloc_page(GFP_KERNEL);
338         if (!rdev->sb_page) {
339                 printk(KERN_ALERT "md: out of memory.\n");
340                 return -EINVAL;
341         }
342
343         return 0;
344 }
345
346 static void free_disk_sb(mdk_rdev_t * rdev)
347 {
348         if (rdev->sb_page) {
349                 put_page(rdev->sb_page);
350                 rdev->sb_loaded = 0;
351                 rdev->sb_page = NULL;
352                 rdev->sb_offset = 0;
353                 rdev->size = 0;
354         }
355 }
356
357
358 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
359 {
360         mdk_rdev_t *rdev = bio->bi_private;
361         mddev_t *mddev = rdev->mddev;
362         if (bio->bi_size)
363                 return 1;
364
365         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
366                 md_error(mddev, rdev);
367
368         if (atomic_dec_and_test(&mddev->pending_writes))
369                 wake_up(&mddev->sb_wait);
370         bio_put(bio);
371         return 0;
372 }
373
374 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
375 {
376         struct bio *bio2 = bio->bi_private;
377         mdk_rdev_t *rdev = bio2->bi_private;
378         mddev_t *mddev = rdev->mddev;
379         if (bio->bi_size)
380                 return 1;
381
382         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
383             error == -EOPNOTSUPP) {
384                 unsigned long flags;
385                 /* barriers don't appear to be supported :-( */
386                 set_bit(BarriersNotsupp, &rdev->flags);
387                 mddev->barriers_work = 0;
388                 spin_lock_irqsave(&mddev->write_lock, flags);
389                 bio2->bi_next = mddev->biolist;
390                 mddev->biolist = bio2;
391                 spin_unlock_irqrestore(&mddev->write_lock, flags);
392                 wake_up(&mddev->sb_wait);
393                 bio_put(bio);
394                 return 0;
395         }
396         bio_put(bio2);
397         bio->bi_private = rdev;
398         return super_written(bio, bytes_done, error);
399 }
400
401 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
402                    sector_t sector, int size, struct page *page)
403 {
404         /* write first size bytes of page to sector of rdev
405          * Increment mddev->pending_writes before returning
406          * and decrement it on completion, waking up sb_wait
407          * if zero is reached.
408          * If an error occurred, call md_error
409          *
410          * As we might need to resubmit the request if BIO_RW_BARRIER
411          * causes ENOTSUPP, we allocate a spare bio...
412          */
413         struct bio *bio = bio_alloc(GFP_NOIO, 1);
414         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
415
416         bio->bi_bdev = rdev->bdev;
417         bio->bi_sector = sector;
418         bio_add_page(bio, page, size, 0);
419         bio->bi_private = rdev;
420         bio->bi_end_io = super_written;
421         bio->bi_rw = rw;
422
423         atomic_inc(&mddev->pending_writes);
424         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
425                 struct bio *rbio;
426                 rw |= (1<<BIO_RW_BARRIER);
427                 rbio = bio_clone(bio, GFP_NOIO);
428                 rbio->bi_private = bio;
429                 rbio->bi_end_io = super_written_barrier;
430                 submit_bio(rw, rbio);
431         } else
432                 submit_bio(rw, bio);
433 }
434
435 void md_super_wait(mddev_t *mddev)
436 {
437         /* wait for all superblock writes that were scheduled to complete.
438          * if any had to be retried (due to BARRIER problems), retry them
439          */
440         DEFINE_WAIT(wq);
441         for(;;) {
442                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
443                 if (atomic_read(&mddev->pending_writes)==0)
444                         break;
445                 while (mddev->biolist) {
446                         struct bio *bio;
447                         spin_lock_irq(&mddev->write_lock);
448                         bio = mddev->biolist;
449                         mddev->biolist = bio->bi_next ;
450                         bio->bi_next = NULL;
451                         spin_unlock_irq(&mddev->write_lock);
452                         submit_bio(bio->bi_rw, bio);
453                 }
454                 schedule();
455         }
456         finish_wait(&mddev->sb_wait, &wq);
457 }
458
459 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
460 {
461         if (bio->bi_size)
462                 return 1;
463
464         complete((struct completion*)bio->bi_private);
465         return 0;
466 }
467
468 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
469                    struct page *page, int rw)
470 {
471         struct bio *bio = bio_alloc(GFP_NOIO, 1);
472         struct completion event;
473         int ret;
474
475         rw |= (1 << BIO_RW_SYNC);
476
477         bio->bi_bdev = bdev;
478         bio->bi_sector = sector;
479         bio_add_page(bio, page, size, 0);
480         init_completion(&event);
481         bio->bi_private = &event;
482         bio->bi_end_io = bi_complete;
483         submit_bio(rw, bio);
484         wait_for_completion(&event);
485
486         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
487         bio_put(bio);
488         return ret;
489 }
490 EXPORT_SYMBOL_GPL(sync_page_io);
491
492 static int read_disk_sb(mdk_rdev_t * rdev, int size)
493 {
494         char b[BDEVNAME_SIZE];
495         if (!rdev->sb_page) {
496                 MD_BUG();
497                 return -EINVAL;
498         }
499         if (rdev->sb_loaded)
500                 return 0;
501
502
503         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
504                 goto fail;
505         rdev->sb_loaded = 1;
506         return 0;
507
508 fail:
509         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
510                 bdevname(rdev->bdev,b));
511         return -EINVAL;
512 }
513
514 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
515 {
516         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
517                 (sb1->set_uuid1 == sb2->set_uuid1) &&
518                 (sb1->set_uuid2 == sb2->set_uuid2) &&
519                 (sb1->set_uuid3 == sb2->set_uuid3))
520
521                 return 1;
522
523         return 0;
524 }
525
526
527 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
528 {
529         int ret;
530         mdp_super_t *tmp1, *tmp2;
531
532         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
533         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
534
535         if (!tmp1 || !tmp2) {
536                 ret = 0;
537                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
538                 goto abort;
539         }
540
541         *tmp1 = *sb1;
542         *tmp2 = *sb2;
543
544         /*
545          * nr_disks is not constant
546          */
547         tmp1->nr_disks = 0;
548         tmp2->nr_disks = 0;
549
550         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
551                 ret = 0;
552         else
553                 ret = 1;
554
555 abort:
556         kfree(tmp1);
557         kfree(tmp2);
558         return ret;
559 }
560
561 static unsigned int calc_sb_csum(mdp_super_t * sb)
562 {
563         unsigned int disk_csum, csum;
564
565         disk_csum = sb->sb_csum;
566         sb->sb_csum = 0;
567         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
568         sb->sb_csum = disk_csum;
569         return csum;
570 }
571
572
573 /*
574  * Handle superblock details.
575  * We want to be able to handle multiple superblock formats
576  * so we have a common interface to them all, and an array of
577  * different handlers.
578  * We rely on user-space to write the initial superblock, and support
579  * reading and updating of superblocks.
580  * Interface methods are:
581  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
582  *      loads and validates a superblock on dev.
583  *      if refdev != NULL, compare superblocks on both devices
584  *    Return:
585  *      0 - dev has a superblock that is compatible with refdev
586  *      1 - dev has a superblock that is compatible and newer than refdev
587  *          so dev should be used as the refdev in future
588  *     -EINVAL superblock incompatible or invalid
589  *     -othererror e.g. -EIO
590  *
591  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
592  *      Verify that dev is acceptable into mddev.
593  *       The first time, mddev->raid_disks will be 0, and data from
594  *       dev should be merged in.  Subsequent calls check that dev
595  *       is new enough.  Return 0 or -EINVAL
596  *
597  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
598  *     Update the superblock for rdev with data in mddev
599  *     This does not write to disc.
600  *
601  */
602
603 struct super_type  {
604         char            *name;
605         struct module   *owner;
606         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
607         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
608         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
609 };
610
611 /*
612  * load_super for 0.90.0 
613  */
614 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
615 {
616         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
617         mdp_super_t *sb;
618         int ret;
619         sector_t sb_offset;
620
621         /*
622          * Calculate the position of the superblock,
623          * it's at the end of the disk.
624          *
625          * It also happens to be a multiple of 4Kb.
626          */
627         sb_offset = calc_dev_sboffset(rdev->bdev);
628         rdev->sb_offset = sb_offset;
629
630         ret = read_disk_sb(rdev, MD_SB_BYTES);
631         if (ret) return ret;
632
633         ret = -EINVAL;
634
635         bdevname(rdev->bdev, b);
636         sb = (mdp_super_t*)page_address(rdev->sb_page);
637
638         if (sb->md_magic != MD_SB_MAGIC) {
639                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
640                        b);
641                 goto abort;
642         }
643
644         if (sb->major_version != 0 ||
645             sb->minor_version != 90) {
646                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
647                         sb->major_version, sb->minor_version,
648                         b);
649                 goto abort;
650         }
651
652         if (sb->raid_disks <= 0)
653                 goto abort;
654
655         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
656                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
657                         b);
658                 goto abort;
659         }
660
661         rdev->preferred_minor = sb->md_minor;
662         rdev->data_offset = 0;
663         rdev->sb_size = MD_SB_BYTES;
664
665         if (sb->level == LEVEL_MULTIPATH)
666                 rdev->desc_nr = -1;
667         else
668                 rdev->desc_nr = sb->this_disk.number;
669
670         if (refdev == 0)
671                 ret = 1;
672         else {
673                 __u64 ev1, ev2;
674                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
675                 if (!uuid_equal(refsb, sb)) {
676                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
677                                 b, bdevname(refdev->bdev,b2));
678                         goto abort;
679                 }
680                 if (!sb_equal(refsb, sb)) {
681                         printk(KERN_WARNING "md: %s has same UUID"
682                                " but different superblock to %s\n",
683                                b, bdevname(refdev->bdev, b2));
684                         goto abort;
685                 }
686                 ev1 = md_event(sb);
687                 ev2 = md_event(refsb);
688                 if (ev1 > ev2)
689                         ret = 1;
690                 else 
691                         ret = 0;
692         }
693         rdev->size = calc_dev_size(rdev, sb->chunk_size);
694
695  abort:
696         return ret;
697 }
698
699 /*
700  * validate_super for 0.90.0
701  */
702 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
703 {
704         mdp_disk_t *desc;
705         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
706
707         rdev->raid_disk = -1;
708         rdev->flags = 0;
709         if (mddev->raid_disks == 0) {
710                 mddev->major_version = 0;
711                 mddev->minor_version = sb->minor_version;
712                 mddev->patch_version = sb->patch_version;
713                 mddev->persistent = ! sb->not_persistent;
714                 mddev->chunk_size = sb->chunk_size;
715                 mddev->ctime = sb->ctime;
716                 mddev->utime = sb->utime;
717                 mddev->level = sb->level;
718                 mddev->layout = sb->layout;
719                 mddev->raid_disks = sb->raid_disks;
720                 mddev->size = sb->size;
721                 mddev->events = md_event(sb);
722                 mddev->bitmap_offset = 0;
723                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
724
725                 if (sb->state & (1<<MD_SB_CLEAN))
726                         mddev->recovery_cp = MaxSector;
727                 else {
728                         if (sb->events_hi == sb->cp_events_hi && 
729                                 sb->events_lo == sb->cp_events_lo) {
730                                 mddev->recovery_cp = sb->recovery_cp;
731                         } else
732                                 mddev->recovery_cp = 0;
733                 }
734
735                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
736                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
737                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
738                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
739
740                 mddev->max_disks = MD_SB_DISKS;
741
742                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
743                     mddev->bitmap_file == NULL) {
744                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
745                             && mddev->level != 10) {
746                                 /* FIXME use a better test */
747                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
748                                 return -EINVAL;
749                         }
750                         mddev->bitmap_offset = mddev->default_bitmap_offset;
751                 }
752
753         } else if (mddev->pers == NULL) {
754                 /* Insist on good event counter while assembling */
755                 __u64 ev1 = md_event(sb);
756                 ++ev1;
757                 if (ev1 < mddev->events) 
758                         return -EINVAL;
759         } else if (mddev->bitmap) {
760                 /* if adding to array with a bitmap, then we can accept an
761                  * older device ... but not too old.
762                  */
763                 __u64 ev1 = md_event(sb);
764                 if (ev1 < mddev->bitmap->events_cleared)
765                         return 0;
766         } else /* just a hot-add of a new device, leave raid_disk at -1 */
767                 return 0;
768
769         if (mddev->level != LEVEL_MULTIPATH) {
770                 desc = sb->disks + rdev->desc_nr;
771
772                 if (desc->state & (1<<MD_DISK_FAULTY))
773                         set_bit(Faulty, &rdev->flags);
774                 else if (desc->state & (1<<MD_DISK_SYNC) &&
775                          desc->raid_disk < mddev->raid_disks) {
776                         set_bit(In_sync, &rdev->flags);
777                         rdev->raid_disk = desc->raid_disk;
778                 }
779                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
780                         set_bit(WriteMostly, &rdev->flags);
781         } else /* MULTIPATH are always insync */
782                 set_bit(In_sync, &rdev->flags);
783         return 0;
784 }
785
786 /*
787  * sync_super for 0.90.0
788  */
789 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
790 {
791         mdp_super_t *sb;
792         struct list_head *tmp;
793         mdk_rdev_t *rdev2;
794         int next_spare = mddev->raid_disks;
795
796
797         /* make rdev->sb match mddev data..
798          *
799          * 1/ zero out disks
800          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
801          * 3/ any empty disks < next_spare become removed
802          *
803          * disks[0] gets initialised to REMOVED because
804          * we cannot be sure from other fields if it has
805          * been initialised or not.
806          */
807         int i;
808         int active=0, working=0,failed=0,spare=0,nr_disks=0;
809
810         rdev->sb_size = MD_SB_BYTES;
811
812         sb = (mdp_super_t*)page_address(rdev->sb_page);
813
814         memset(sb, 0, sizeof(*sb));
815
816         sb->md_magic = MD_SB_MAGIC;
817         sb->major_version = mddev->major_version;
818         sb->minor_version = mddev->minor_version;
819         sb->patch_version = mddev->patch_version;
820         sb->gvalid_words  = 0; /* ignored */
821         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
822         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
823         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
824         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
825
826         sb->ctime = mddev->ctime;
827         sb->level = mddev->level;
828         sb->size  = mddev->size;
829         sb->raid_disks = mddev->raid_disks;
830         sb->md_minor = mddev->md_minor;
831         sb->not_persistent = !mddev->persistent;
832         sb->utime = mddev->utime;
833         sb->state = 0;
834         sb->events_hi = (mddev->events>>32);
835         sb->events_lo = (u32)mddev->events;
836
837         if (mddev->in_sync)
838         {
839                 sb->recovery_cp = mddev->recovery_cp;
840                 sb->cp_events_hi = (mddev->events>>32);
841                 sb->cp_events_lo = (u32)mddev->events;
842                 if (mddev->recovery_cp == MaxSector)
843                         sb->state = (1<< MD_SB_CLEAN);
844         } else
845                 sb->recovery_cp = 0;
846
847         sb->layout = mddev->layout;
848         sb->chunk_size = mddev->chunk_size;
849
850         if (mddev->bitmap && mddev->bitmap_file == NULL)
851                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
852
853         sb->disks[0].state = (1<<MD_DISK_REMOVED);
854         ITERATE_RDEV(mddev,rdev2,tmp) {
855                 mdp_disk_t *d;
856                 int desc_nr;
857                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
858                     && !test_bit(Faulty, &rdev2->flags))
859                         desc_nr = rdev2->raid_disk;
860                 else
861                         desc_nr = next_spare++;
862                 rdev2->desc_nr = desc_nr;
863                 d = &sb->disks[rdev2->desc_nr];
864                 nr_disks++;
865                 d->number = rdev2->desc_nr;
866                 d->major = MAJOR(rdev2->bdev->bd_dev);
867                 d->minor = MINOR(rdev2->bdev->bd_dev);
868                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
869                     && !test_bit(Faulty, &rdev2->flags))
870                         d->raid_disk = rdev2->raid_disk;
871                 else
872                         d->raid_disk = rdev2->desc_nr; /* compatibility */
873                 if (test_bit(Faulty, &rdev2->flags)) {
874                         d->state = (1<<MD_DISK_FAULTY);
875                         failed++;
876                 } else if (test_bit(In_sync, &rdev2->flags)) {
877                         d->state = (1<<MD_DISK_ACTIVE);
878                         d->state |= (1<<MD_DISK_SYNC);
879                         active++;
880                         working++;
881                 } else {
882                         d->state = 0;
883                         spare++;
884                         working++;
885                 }
886                 if (test_bit(WriteMostly, &rdev2->flags))
887                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
888         }
889         /* now set the "removed" and "faulty" bits on any missing devices */
890         for (i=0 ; i < mddev->raid_disks ; i++) {
891                 mdp_disk_t *d = &sb->disks[i];
892                 if (d->state == 0 && d->number == 0) {
893                         d->number = i;
894                         d->raid_disk = i;
895                         d->state = (1<<MD_DISK_REMOVED);
896                         d->state |= (1<<MD_DISK_FAULTY);
897                         failed++;
898                 }
899         }
900         sb->nr_disks = nr_disks;
901         sb->active_disks = active;
902         sb->working_disks = working;
903         sb->failed_disks = failed;
904         sb->spare_disks = spare;
905
906         sb->this_disk = sb->disks[rdev->desc_nr];
907         sb->sb_csum = calc_sb_csum(sb);
908 }
909
910 /*
911  * version 1 superblock
912  */
913
914 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
915 {
916         unsigned int disk_csum, csum;
917         unsigned long long newcsum;
918         int size = 256 + le32_to_cpu(sb->max_dev)*2;
919         unsigned int *isuper = (unsigned int*)sb;
920         int i;
921
922         disk_csum = sb->sb_csum;
923         sb->sb_csum = 0;
924         newcsum = 0;
925         for (i=0; size>=4; size -= 4 )
926                 newcsum += le32_to_cpu(*isuper++);
927
928         if (size == 2)
929                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
930
931         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
932         sb->sb_csum = disk_csum;
933         return cpu_to_le32(csum);
934 }
935
936 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
937 {
938         struct mdp_superblock_1 *sb;
939         int ret;
940         sector_t sb_offset;
941         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
942         int bmask;
943
944         /*
945          * Calculate the position of the superblock.
946          * It is always aligned to a 4K boundary and
947          * depeding on minor_version, it can be:
948          * 0: At least 8K, but less than 12K, from end of device
949          * 1: At start of device
950          * 2: 4K from start of device.
951          */
952         switch(minor_version) {
953         case 0:
954                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
955                 sb_offset -= 8*2;
956                 sb_offset &= ~(sector_t)(4*2-1);
957                 /* convert from sectors to K */
958                 sb_offset /= 2;
959                 break;
960         case 1:
961                 sb_offset = 0;
962                 break;
963         case 2:
964                 sb_offset = 4;
965                 break;
966         default:
967                 return -EINVAL;
968         }
969         rdev->sb_offset = sb_offset;
970
971         /* superblock is rarely larger than 1K, but it can be larger,
972          * and it is safe to read 4k, so we do that
973          */
974         ret = read_disk_sb(rdev, 4096);
975         if (ret) return ret;
976
977
978         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
979
980         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
981             sb->major_version != cpu_to_le32(1) ||
982             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
983             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
984             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
985                 return -EINVAL;
986
987         if (calc_sb_1_csum(sb) != sb->sb_csum) {
988                 printk("md: invalid superblock checksum on %s\n",
989                         bdevname(rdev->bdev,b));
990                 return -EINVAL;
991         }
992         if (le64_to_cpu(sb->data_size) < 10) {
993                 printk("md: data_size too small on %s\n",
994                        bdevname(rdev->bdev,b));
995                 return -EINVAL;
996         }
997         rdev->preferred_minor = 0xffff;
998         rdev->data_offset = le64_to_cpu(sb->data_offset);
999
1000         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1001         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1002         if (rdev->sb_size & bmask)
1003                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1004
1005         if (refdev == 0)
1006                 return 1;
1007         else {
1008                 __u64 ev1, ev2;
1009                 struct mdp_superblock_1 *refsb = 
1010                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1011
1012                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1013                     sb->level != refsb->level ||
1014                     sb->layout != refsb->layout ||
1015                     sb->chunksize != refsb->chunksize) {
1016                         printk(KERN_WARNING "md: %s has strangely different"
1017                                 " superblock to %s\n",
1018                                 bdevname(rdev->bdev,b),
1019                                 bdevname(refdev->bdev,b2));
1020                         return -EINVAL;
1021                 }
1022                 ev1 = le64_to_cpu(sb->events);
1023                 ev2 = le64_to_cpu(refsb->events);
1024
1025                 if (ev1 > ev2)
1026                         return 1;
1027         }
1028         if (minor_version) 
1029                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1030         else
1031                 rdev->size = rdev->sb_offset;
1032         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1033                 return -EINVAL;
1034         rdev->size = le64_to_cpu(sb->data_size)/2;
1035         if (le32_to_cpu(sb->chunksize))
1036                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1037         return 0;
1038 }
1039
1040 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1041 {
1042         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1043
1044         rdev->raid_disk = -1;
1045         rdev->flags = 0;
1046         if (mddev->raid_disks == 0) {
1047                 mddev->major_version = 1;
1048                 mddev->patch_version = 0;
1049                 mddev->persistent = 1;
1050                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1051                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1052                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1053                 mddev->level = le32_to_cpu(sb->level);
1054                 mddev->layout = le32_to_cpu(sb->layout);
1055                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1056                 mddev->size = le64_to_cpu(sb->size)/2;
1057                 mddev->events = le64_to_cpu(sb->events);
1058                 mddev->bitmap_offset = 0;
1059                 mddev->default_bitmap_offset = 1024;
1060                 
1061                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1062                 memcpy(mddev->uuid, sb->set_uuid, 16);
1063
1064                 mddev->max_disks =  (4096-256)/2;
1065
1066                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1067                     mddev->bitmap_file == NULL ) {
1068                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1069                             && mddev->level != 10) {
1070                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1071                                 return -EINVAL;
1072                         }
1073                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1074                 }
1075         } else if (mddev->pers == NULL) {
1076                 /* Insist of good event counter while assembling */
1077                 __u64 ev1 = le64_to_cpu(sb->events);
1078                 ++ev1;
1079                 if (ev1 < mddev->events)
1080                         return -EINVAL;
1081         } else if (mddev->bitmap) {
1082                 /* If adding to array with a bitmap, then we can accept an
1083                  * older device, but not too old.
1084                  */
1085                 __u64 ev1 = le64_to_cpu(sb->events);
1086                 if (ev1 < mddev->bitmap->events_cleared)
1087                         return 0;
1088         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1089                 return 0;
1090
1091         if (mddev->level != LEVEL_MULTIPATH) {
1092                 int role;
1093                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1094                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1095                 switch(role) {
1096                 case 0xffff: /* spare */
1097                         break;
1098                 case 0xfffe: /* faulty */
1099                         set_bit(Faulty, &rdev->flags);
1100                         break;
1101                 default:
1102                         set_bit(In_sync, &rdev->flags);
1103                         rdev->raid_disk = role;
1104                         break;
1105                 }
1106                 if (sb->devflags & WriteMostly1)
1107                         set_bit(WriteMostly, &rdev->flags);
1108         } else /* MULTIPATH are always insync */
1109                 set_bit(In_sync, &rdev->flags);
1110
1111         return 0;
1112 }
1113
1114 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1115 {
1116         struct mdp_superblock_1 *sb;
1117         struct list_head *tmp;
1118         mdk_rdev_t *rdev2;
1119         int max_dev, i;
1120         /* make rdev->sb match mddev and rdev data. */
1121
1122         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1123
1124         sb->feature_map = 0;
1125         sb->pad0 = 0;
1126         memset(sb->pad1, 0, sizeof(sb->pad1));
1127         memset(sb->pad2, 0, sizeof(sb->pad2));
1128         memset(sb->pad3, 0, sizeof(sb->pad3));
1129
1130         sb->utime = cpu_to_le64((__u64)mddev->utime);
1131         sb->events = cpu_to_le64(mddev->events);
1132         if (mddev->in_sync)
1133                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1134         else
1135                 sb->resync_offset = cpu_to_le64(0);
1136
1137         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1138                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1139                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1140         }
1141
1142         max_dev = 0;
1143         ITERATE_RDEV(mddev,rdev2,tmp)
1144                 if (rdev2->desc_nr+1 > max_dev)
1145                         max_dev = rdev2->desc_nr+1;
1146         
1147         sb->max_dev = cpu_to_le32(max_dev);
1148         for (i=0; i<max_dev;i++)
1149                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1150         
1151         ITERATE_RDEV(mddev,rdev2,tmp) {
1152                 i = rdev2->desc_nr;
1153                 if (test_bit(Faulty, &rdev2->flags))
1154                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1155                 else if (test_bit(In_sync, &rdev2->flags))
1156                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1157                 else
1158                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1159         }
1160
1161         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1162         sb->sb_csum = calc_sb_1_csum(sb);
1163 }
1164
1165
1166 static struct super_type super_types[] = {
1167         [0] = {
1168                 .name   = "0.90.0",
1169                 .owner  = THIS_MODULE,
1170                 .load_super     = super_90_load,
1171                 .validate_super = super_90_validate,
1172                 .sync_super     = super_90_sync,
1173         },
1174         [1] = {
1175                 .name   = "md-1",
1176                 .owner  = THIS_MODULE,
1177                 .load_super     = super_1_load,
1178                 .validate_super = super_1_validate,
1179                 .sync_super     = super_1_sync,
1180         },
1181 };
1182         
1183 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1184 {
1185         struct list_head *tmp;
1186         mdk_rdev_t *rdev;
1187
1188         ITERATE_RDEV(mddev,rdev,tmp)
1189                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1190                         return rdev;
1191
1192         return NULL;
1193 }
1194
1195 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1196 {
1197         struct list_head *tmp;
1198         mdk_rdev_t *rdev;
1199
1200         ITERATE_RDEV(mddev1,rdev,tmp)
1201                 if (match_dev_unit(mddev2, rdev))
1202                         return 1;
1203
1204         return 0;
1205 }
1206
1207 static LIST_HEAD(pending_raid_disks);
1208
1209 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1210 {
1211         mdk_rdev_t *same_pdev;
1212         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1213         struct kobject *ko;
1214
1215         if (rdev->mddev) {
1216                 MD_BUG();
1217                 return -EINVAL;
1218         }
1219         same_pdev = match_dev_unit(mddev, rdev);
1220         if (same_pdev)
1221                 printk(KERN_WARNING
1222                         "%s: WARNING: %s appears to be on the same physical"
1223                         " disk as %s. True\n     protection against single-disk"
1224                         " failure might be compromised.\n",
1225                         mdname(mddev), bdevname(rdev->bdev,b),
1226                         bdevname(same_pdev->bdev,b2));
1227
1228         /* Verify rdev->desc_nr is unique.
1229          * If it is -1, assign a free number, else
1230          * check number is not in use
1231          */
1232         if (rdev->desc_nr < 0) {
1233                 int choice = 0;
1234                 if (mddev->pers) choice = mddev->raid_disks;
1235                 while (find_rdev_nr(mddev, choice))
1236                         choice++;
1237                 rdev->desc_nr = choice;
1238         } else {
1239                 if (find_rdev_nr(mddev, rdev->desc_nr))
1240                         return -EBUSY;
1241         }
1242         bdevname(rdev->bdev,b);
1243         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1244                 return -ENOMEM;
1245                         
1246         list_add(&rdev->same_set, &mddev->disks);
1247         rdev->mddev = mddev;
1248         printk(KERN_INFO "md: bind<%s>\n", b);
1249
1250         rdev->kobj.parent = &mddev->kobj;
1251         kobject_add(&rdev->kobj);
1252
1253         if (rdev->bdev->bd_part)
1254                 ko = &rdev->bdev->bd_part->kobj;
1255         else
1256                 ko = &rdev->bdev->bd_disk->kobj;
1257         sysfs_create_link(&rdev->kobj, ko, "block");
1258         return 0;
1259 }
1260
1261 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1262 {
1263         char b[BDEVNAME_SIZE];
1264         if (!rdev->mddev) {
1265                 MD_BUG();
1266                 return;
1267         }
1268         list_del_init(&rdev->same_set);
1269         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1270         rdev->mddev = NULL;
1271         sysfs_remove_link(&rdev->kobj, "block");
1272         kobject_del(&rdev->kobj);
1273 }
1274
1275 /*
1276  * prevent the device from being mounted, repartitioned or
1277  * otherwise reused by a RAID array (or any other kernel
1278  * subsystem), by bd_claiming the device.
1279  */
1280 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1281 {
1282         int err = 0;
1283         struct block_device *bdev;
1284         char b[BDEVNAME_SIZE];
1285
1286         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1287         if (IS_ERR(bdev)) {
1288                 printk(KERN_ERR "md: could not open %s.\n",
1289                         __bdevname(dev, b));
1290                 return PTR_ERR(bdev);
1291         }
1292         err = bd_claim(bdev, rdev);
1293         if (err) {
1294                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1295                         bdevname(bdev, b));
1296                 blkdev_put(bdev);
1297                 return err;
1298         }
1299         rdev->bdev = bdev;
1300         return err;
1301 }
1302
1303 static void unlock_rdev(mdk_rdev_t *rdev)
1304 {
1305         struct block_device *bdev = rdev->bdev;
1306         rdev->bdev = NULL;
1307         if (!bdev)
1308                 MD_BUG();
1309         bd_release(bdev);
1310         blkdev_put(bdev);
1311 }
1312
1313 void md_autodetect_dev(dev_t dev);
1314
1315 static void export_rdev(mdk_rdev_t * rdev)
1316 {
1317         char b[BDEVNAME_SIZE];
1318         printk(KERN_INFO "md: export_rdev(%s)\n",
1319                 bdevname(rdev->bdev,b));
1320         if (rdev->mddev)
1321                 MD_BUG();
1322         free_disk_sb(rdev);
1323         list_del_init(&rdev->same_set);
1324 #ifndef MODULE
1325         md_autodetect_dev(rdev->bdev->bd_dev);
1326 #endif
1327         unlock_rdev(rdev);
1328         kobject_put(&rdev->kobj);
1329 }
1330
1331 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1332 {
1333         unbind_rdev_from_array(rdev);
1334         export_rdev(rdev);
1335 }
1336
1337 static void export_array(mddev_t *mddev)
1338 {
1339         struct list_head *tmp;
1340         mdk_rdev_t *rdev;
1341
1342         ITERATE_RDEV(mddev,rdev,tmp) {
1343                 if (!rdev->mddev) {
1344                         MD_BUG();
1345                         continue;
1346                 }
1347                 kick_rdev_from_array(rdev);
1348         }
1349         if (!list_empty(&mddev->disks))
1350                 MD_BUG();
1351         mddev->raid_disks = 0;
1352         mddev->major_version = 0;
1353 }
1354
1355 static void print_desc(mdp_disk_t *desc)
1356 {
1357         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1358                 desc->major,desc->minor,desc->raid_disk,desc->state);
1359 }
1360
1361 static void print_sb(mdp_super_t *sb)
1362 {
1363         int i;
1364
1365         printk(KERN_INFO 
1366                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1367                 sb->major_version, sb->minor_version, sb->patch_version,
1368                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1369                 sb->ctime);
1370         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1371                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1372                 sb->md_minor, sb->layout, sb->chunk_size);
1373         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1374                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1375                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1376                 sb->failed_disks, sb->spare_disks,
1377                 sb->sb_csum, (unsigned long)sb->events_lo);
1378
1379         printk(KERN_INFO);
1380         for (i = 0; i < MD_SB_DISKS; i++) {
1381                 mdp_disk_t *desc;
1382
1383                 desc = sb->disks + i;
1384                 if (desc->number || desc->major || desc->minor ||
1385                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1386                         printk("     D %2d: ", i);
1387                         print_desc(desc);
1388                 }
1389         }
1390         printk(KERN_INFO "md:     THIS: ");
1391         print_desc(&sb->this_disk);
1392
1393 }
1394
1395 static void print_rdev(mdk_rdev_t *rdev)
1396 {
1397         char b[BDEVNAME_SIZE];
1398         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1399                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1400                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1401                 rdev->desc_nr);
1402         if (rdev->sb_loaded) {
1403                 printk(KERN_INFO "md: rdev superblock:\n");
1404                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1405         } else
1406                 printk(KERN_INFO "md: no rdev superblock!\n");
1407 }
1408
1409 void md_print_devices(void)
1410 {
1411         struct list_head *tmp, *tmp2;
1412         mdk_rdev_t *rdev;
1413         mddev_t *mddev;
1414         char b[BDEVNAME_SIZE];
1415
1416         printk("\n");
1417         printk("md:     **********************************\n");
1418         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1419         printk("md:     **********************************\n");
1420         ITERATE_MDDEV(mddev,tmp) {
1421
1422                 if (mddev->bitmap)
1423                         bitmap_print_sb(mddev->bitmap);
1424                 else
1425                         printk("%s: ", mdname(mddev));
1426                 ITERATE_RDEV(mddev,rdev,tmp2)
1427                         printk("<%s>", bdevname(rdev->bdev,b));
1428                 printk("\n");
1429
1430                 ITERATE_RDEV(mddev,rdev,tmp2)
1431                         print_rdev(rdev);
1432         }
1433         printk("md:     **********************************\n");
1434         printk("\n");
1435 }
1436
1437
1438 static void sync_sbs(mddev_t * mddev)
1439 {
1440         mdk_rdev_t *rdev;
1441         struct list_head *tmp;
1442
1443         ITERATE_RDEV(mddev,rdev,tmp) {
1444                 super_types[mddev->major_version].
1445                         sync_super(mddev, rdev);
1446                 rdev->sb_loaded = 1;
1447         }
1448 }
1449
1450 static void md_update_sb(mddev_t * mddev)
1451 {
1452         int err;
1453         struct list_head *tmp;
1454         mdk_rdev_t *rdev;
1455         int sync_req;
1456
1457 repeat:
1458         spin_lock_irq(&mddev->write_lock);
1459         sync_req = mddev->in_sync;
1460         mddev->utime = get_seconds();
1461         mddev->events ++;
1462
1463         if (!mddev->events) {
1464                 /*
1465                  * oops, this 64-bit counter should never wrap.
1466                  * Either we are in around ~1 trillion A.C., assuming
1467                  * 1 reboot per second, or we have a bug:
1468                  */
1469                 MD_BUG();
1470                 mddev->events --;
1471         }
1472         mddev->sb_dirty = 2;
1473         sync_sbs(mddev);
1474
1475         /*
1476          * do not write anything to disk if using
1477          * nonpersistent superblocks
1478          */
1479         if (!mddev->persistent) {
1480                 mddev->sb_dirty = 0;
1481                 spin_unlock_irq(&mddev->write_lock);
1482                 wake_up(&mddev->sb_wait);
1483                 return;
1484         }
1485         spin_unlock_irq(&mddev->write_lock);
1486
1487         dprintk(KERN_INFO 
1488                 "md: updating %s RAID superblock on device (in sync %d)\n",
1489                 mdname(mddev),mddev->in_sync);
1490
1491         err = bitmap_update_sb(mddev->bitmap);
1492         ITERATE_RDEV(mddev,rdev,tmp) {
1493                 char b[BDEVNAME_SIZE];
1494                 dprintk(KERN_INFO "md: ");
1495                 if (test_bit(Faulty, &rdev->flags))
1496                         dprintk("(skipping faulty ");
1497
1498                 dprintk("%s ", bdevname(rdev->bdev,b));
1499                 if (!test_bit(Faulty, &rdev->flags)) {
1500                         md_super_write(mddev,rdev,
1501                                        rdev->sb_offset<<1, rdev->sb_size,
1502                                        rdev->sb_page);
1503                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1504                                 bdevname(rdev->bdev,b),
1505                                 (unsigned long long)rdev->sb_offset);
1506
1507                 } else
1508                         dprintk(")\n");
1509                 if (mddev->level == LEVEL_MULTIPATH)
1510                         /* only need to write one superblock... */
1511                         break;
1512         }
1513         md_super_wait(mddev);
1514         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1515
1516         spin_lock_irq(&mddev->write_lock);
1517         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1518                 /* have to write it out again */
1519                 spin_unlock_irq(&mddev->write_lock);
1520                 goto repeat;
1521         }
1522         mddev->sb_dirty = 0;
1523         spin_unlock_irq(&mddev->write_lock);
1524         wake_up(&mddev->sb_wait);
1525
1526 }
1527
1528 /* words written to sysfs files may, or my not, be \n terminated.
1529  * We want to accept with case. For this we use cmd_match.
1530  */
1531 static int cmd_match(const char *cmd, const char *str)
1532 {
1533         /* See if cmd, written into a sysfs file, matches
1534          * str.  They must either be the same, or cmd can
1535          * have a trailing newline
1536          */
1537         while (*cmd && *str && *cmd == *str) {
1538                 cmd++;
1539                 str++;
1540         }
1541         if (*cmd == '\n')
1542                 cmd++;
1543         if (*str || *cmd)
1544                 return 0;
1545         return 1;
1546 }
1547
1548 struct rdev_sysfs_entry {
1549         struct attribute attr;
1550         ssize_t (*show)(mdk_rdev_t *, char *);
1551         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1552 };
1553
1554 static ssize_t
1555 state_show(mdk_rdev_t *rdev, char *page)
1556 {
1557         char *sep = "";
1558         int len=0;
1559
1560         if (test_bit(Faulty, &rdev->flags)) {
1561                 len+= sprintf(page+len, "%sfaulty",sep);
1562                 sep = ",";
1563         }
1564         if (test_bit(In_sync, &rdev->flags)) {
1565                 len += sprintf(page+len, "%sin_sync",sep);
1566                 sep = ",";
1567         }
1568         if (!test_bit(Faulty, &rdev->flags) &&
1569             !test_bit(In_sync, &rdev->flags)) {
1570                 len += sprintf(page+len, "%sspare", sep);
1571                 sep = ",";
1572         }
1573         return len+sprintf(page+len, "\n");
1574 }
1575
1576 static struct rdev_sysfs_entry
1577 rdev_state = __ATTR_RO(state);
1578
1579 static ssize_t
1580 super_show(mdk_rdev_t *rdev, char *page)
1581 {
1582         if (rdev->sb_loaded && rdev->sb_size) {
1583                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1584                 return rdev->sb_size;
1585         } else
1586                 return 0;
1587 }
1588 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1589
1590 static struct attribute *rdev_default_attrs[] = {
1591         &rdev_state.attr,
1592         &rdev_super.attr,
1593         NULL,
1594 };
1595 static ssize_t
1596 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1597 {
1598         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1599         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1600
1601         if (!entry->show)
1602                 return -EIO;
1603         return entry->show(rdev, page);
1604 }
1605
1606 static ssize_t
1607 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1608               const char *page, size_t length)
1609 {
1610         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1611         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1612
1613         if (!entry->store)
1614                 return -EIO;
1615         return entry->store(rdev, page, length);
1616 }
1617
1618 static void rdev_free(struct kobject *ko)
1619 {
1620         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1621         kfree(rdev);
1622 }
1623 static struct sysfs_ops rdev_sysfs_ops = {
1624         .show           = rdev_attr_show,
1625         .store          = rdev_attr_store,
1626 };
1627 static struct kobj_type rdev_ktype = {
1628         .release        = rdev_free,
1629         .sysfs_ops      = &rdev_sysfs_ops,
1630         .default_attrs  = rdev_default_attrs,
1631 };
1632
1633 /*
1634  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1635  *
1636  * mark the device faulty if:
1637  *
1638  *   - the device is nonexistent (zero size)
1639  *   - the device has no valid superblock
1640  *
1641  * a faulty rdev _never_ has rdev->sb set.
1642  */
1643 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1644 {
1645         char b[BDEVNAME_SIZE];
1646         int err;
1647         mdk_rdev_t *rdev;
1648         sector_t size;
1649
1650         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1651         if (!rdev) {
1652                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1653                 return ERR_PTR(-ENOMEM);
1654         }
1655
1656         if ((err = alloc_disk_sb(rdev)))
1657                 goto abort_free;
1658
1659         err = lock_rdev(rdev, newdev);
1660         if (err)
1661                 goto abort_free;
1662
1663         rdev->kobj.parent = NULL;
1664         rdev->kobj.ktype = &rdev_ktype;
1665         kobject_init(&rdev->kobj);
1666
1667         rdev->desc_nr = -1;
1668         rdev->flags = 0;
1669         rdev->data_offset = 0;
1670         atomic_set(&rdev->nr_pending, 0);
1671         atomic_set(&rdev->read_errors, 0);
1672
1673         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1674         if (!size) {
1675                 printk(KERN_WARNING 
1676                         "md: %s has zero or unknown size, marking faulty!\n",
1677                         bdevname(rdev->bdev,b));
1678                 err = -EINVAL;
1679                 goto abort_free;
1680         }
1681
1682         if (super_format >= 0) {
1683                 err = super_types[super_format].
1684                         load_super(rdev, NULL, super_minor);
1685                 if (err == -EINVAL) {
1686                         printk(KERN_WARNING 
1687                                 "md: %s has invalid sb, not importing!\n",
1688                                 bdevname(rdev->bdev,b));
1689                         goto abort_free;
1690                 }
1691                 if (err < 0) {
1692                         printk(KERN_WARNING 
1693                                 "md: could not read %s's sb, not importing!\n",
1694                                 bdevname(rdev->bdev,b));
1695                         goto abort_free;
1696                 }
1697         }
1698         INIT_LIST_HEAD(&rdev->same_set);
1699
1700         return rdev;
1701
1702 abort_free:
1703         if (rdev->sb_page) {
1704                 if (rdev->bdev)
1705                         unlock_rdev(rdev);
1706                 free_disk_sb(rdev);
1707         }
1708         kfree(rdev);
1709         return ERR_PTR(err);
1710 }
1711
1712 /*
1713  * Check a full RAID array for plausibility
1714  */
1715
1716
1717 static void analyze_sbs(mddev_t * mddev)
1718 {
1719         int i;
1720         struct list_head *tmp;
1721         mdk_rdev_t *rdev, *freshest;
1722         char b[BDEVNAME_SIZE];
1723
1724         freshest = NULL;
1725         ITERATE_RDEV(mddev,rdev,tmp)
1726                 switch (super_types[mddev->major_version].
1727                         load_super(rdev, freshest, mddev->minor_version)) {
1728                 case 1:
1729                         freshest = rdev;
1730                         break;
1731                 case 0:
1732                         break;
1733                 default:
1734                         printk( KERN_ERR \
1735                                 "md: fatal superblock inconsistency in %s"
1736                                 " -- removing from array\n", 
1737                                 bdevname(rdev->bdev,b));
1738                         kick_rdev_from_array(rdev);
1739                 }
1740
1741
1742         super_types[mddev->major_version].
1743                 validate_super(mddev, freshest);
1744
1745         i = 0;
1746         ITERATE_RDEV(mddev,rdev,tmp) {
1747                 if (rdev != freshest)
1748                         if (super_types[mddev->major_version].
1749                             validate_super(mddev, rdev)) {
1750                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1751                                         " from array!\n",
1752                                         bdevname(rdev->bdev,b));
1753                                 kick_rdev_from_array(rdev);
1754                                 continue;
1755                         }
1756                 if (mddev->level == LEVEL_MULTIPATH) {
1757                         rdev->desc_nr = i++;
1758                         rdev->raid_disk = rdev->desc_nr;
1759                         set_bit(In_sync, &rdev->flags);
1760                 }
1761         }
1762
1763
1764
1765         if (mddev->recovery_cp != MaxSector &&
1766             mddev->level >= 1)
1767                 printk(KERN_ERR "md: %s: raid array is not clean"
1768                        " -- starting background reconstruction\n",
1769                        mdname(mddev));
1770
1771 }
1772
1773 static ssize_t
1774 level_show(mddev_t *mddev, char *page)
1775 {
1776         struct mdk_personality *p = mddev->pers;
1777         if (p == NULL && mddev->raid_disks == 0)
1778                 return 0;
1779         if (mddev->level >= 0)
1780                 return sprintf(page, "raid%d\n", mddev->level);
1781         else
1782                 return sprintf(page, "%s\n", p->name);
1783 }
1784
1785 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1786
1787 static ssize_t
1788 raid_disks_show(mddev_t *mddev, char *page)
1789 {
1790         if (mddev->raid_disks == 0)
1791                 return 0;
1792         return sprintf(page, "%d\n", mddev->raid_disks);
1793 }
1794
1795 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1796
1797 static ssize_t
1798 action_show(mddev_t *mddev, char *page)
1799 {
1800         char *type = "idle";
1801         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1802             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1803                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1804                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1805                                 type = "resync";
1806                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1807                                 type = "check";
1808                         else
1809                                 type = "repair";
1810                 } else
1811                         type = "recover";
1812         }
1813         return sprintf(page, "%s\n", type);
1814 }
1815
1816 static ssize_t
1817 action_store(mddev_t *mddev, const char *page, size_t len)
1818 {
1819         if (!mddev->pers || !mddev->pers->sync_request)
1820                 return -EINVAL;
1821
1822         if (cmd_match(page, "idle")) {
1823                 if (mddev->sync_thread) {
1824                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1825                         md_unregister_thread(mddev->sync_thread);
1826                         mddev->sync_thread = NULL;
1827                         mddev->recovery = 0;
1828                 }
1829                 return len;
1830         }
1831
1832         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1833             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1834                 return -EBUSY;
1835         if (cmd_match(page, "resync") || cmd_match(page, "recover"))
1836                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1837         else {
1838                 if (cmd_match(page, "check"))
1839                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1840                 else if (cmd_match(page, "repair"))
1841                         return -EINVAL;
1842                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1843                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1844                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1845         }
1846         md_wakeup_thread(mddev->thread);
1847         return len;
1848 }
1849
1850 static ssize_t
1851 mismatch_cnt_show(mddev_t *mddev, char *page)
1852 {
1853         return sprintf(page, "%llu\n",
1854                        (unsigned long long) mddev->resync_mismatches);
1855 }
1856
1857 static struct md_sysfs_entry
1858 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1859
1860
1861 static struct md_sysfs_entry
1862 md_mismatches = __ATTR_RO(mismatch_cnt);
1863
1864 static struct attribute *md_default_attrs[] = {
1865         &md_level.attr,
1866         &md_raid_disks.attr,
1867         NULL,
1868 };
1869
1870 static struct attribute *md_redundancy_attrs[] = {
1871         &md_scan_mode.attr,
1872         &md_mismatches.attr,
1873         NULL,
1874 };
1875 static struct attribute_group md_redundancy_group = {
1876         .name = NULL,
1877         .attrs = md_redundancy_attrs,
1878 };
1879
1880
1881 static ssize_t
1882 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1883 {
1884         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1885         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1886         ssize_t rv;
1887
1888         if (!entry->show)
1889                 return -EIO;
1890         mddev_lock(mddev);
1891         rv = entry->show(mddev, page);
1892         mddev_unlock(mddev);
1893         return rv;
1894 }
1895
1896 static ssize_t
1897 md_attr_store(struct kobject *kobj, struct attribute *attr,
1898               const char *page, size_t length)
1899 {
1900         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1901         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1902         ssize_t rv;
1903
1904         if (!entry->store)
1905                 return -EIO;
1906         mddev_lock(mddev);
1907         rv = entry->store(mddev, page, length);
1908         mddev_unlock(mddev);
1909         return rv;
1910 }
1911
1912 static void md_free(struct kobject *ko)
1913 {
1914         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1915         kfree(mddev);
1916 }
1917
1918 static struct sysfs_ops md_sysfs_ops = {
1919         .show   = md_attr_show,
1920         .store  = md_attr_store,
1921 };
1922 static struct kobj_type md_ktype = {
1923         .release        = md_free,
1924         .sysfs_ops      = &md_sysfs_ops,
1925         .default_attrs  = md_default_attrs,
1926 };
1927
1928 int mdp_major = 0;
1929
1930 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1931 {
1932         static DECLARE_MUTEX(disks_sem);
1933         mddev_t *mddev = mddev_find(dev);
1934         struct gendisk *disk;
1935         int partitioned = (MAJOR(dev) != MD_MAJOR);
1936         int shift = partitioned ? MdpMinorShift : 0;
1937         int unit = MINOR(dev) >> shift;
1938
1939         if (!mddev)
1940                 return NULL;
1941
1942         down(&disks_sem);
1943         if (mddev->gendisk) {
1944                 up(&disks_sem);
1945                 mddev_put(mddev);
1946                 return NULL;
1947         }
1948         disk = alloc_disk(1 << shift);
1949         if (!disk) {
1950                 up(&disks_sem);
1951                 mddev_put(mddev);
1952                 return NULL;
1953         }
1954         disk->major = MAJOR(dev);
1955         disk->first_minor = unit << shift;
1956         if (partitioned) {
1957                 sprintf(disk->disk_name, "md_d%d", unit);
1958                 sprintf(disk->devfs_name, "md/d%d", unit);
1959         } else {
1960                 sprintf(disk->disk_name, "md%d", unit);
1961                 sprintf(disk->devfs_name, "md/%d", unit);
1962         }
1963         disk->fops = &md_fops;
1964         disk->private_data = mddev;
1965         disk->queue = mddev->queue;
1966         add_disk(disk);
1967         mddev->gendisk = disk;
1968         up(&disks_sem);
1969         mddev->kobj.parent = &disk->kobj;
1970         mddev->kobj.k_name = NULL;
1971         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1972         mddev->kobj.ktype = &md_ktype;
1973         kobject_register(&mddev->kobj);
1974         return NULL;
1975 }
1976
1977 void md_wakeup_thread(mdk_thread_t *thread);
1978
1979 static void md_safemode_timeout(unsigned long data)
1980 {
1981         mddev_t *mddev = (mddev_t *) data;
1982
1983         mddev->safemode = 1;
1984         md_wakeup_thread(mddev->thread);
1985 }
1986
1987 static int start_dirty_degraded;
1988
1989 static int do_md_run(mddev_t * mddev)
1990 {
1991         int err;
1992         int chunk_size;
1993         struct list_head *tmp;
1994         mdk_rdev_t *rdev;
1995         struct gendisk *disk;
1996         struct mdk_personality *pers;
1997         char b[BDEVNAME_SIZE];
1998
1999         if (list_empty(&mddev->disks))
2000                 /* cannot run an array with no devices.. */
2001                 return -EINVAL;
2002
2003         if (mddev->pers)
2004                 return -EBUSY;
2005
2006         /*
2007          * Analyze all RAID superblock(s)
2008          */
2009         if (!mddev->raid_disks)
2010                 analyze_sbs(mddev);
2011
2012         chunk_size = mddev->chunk_size;
2013
2014         if (chunk_size) {
2015                 if (chunk_size > MAX_CHUNK_SIZE) {
2016                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2017                                 chunk_size, MAX_CHUNK_SIZE);
2018                         return -EINVAL;
2019                 }
2020                 /*
2021                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2022                  */
2023                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2024                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2025                         return -EINVAL;
2026                 }
2027                 if (chunk_size < PAGE_SIZE) {
2028                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2029                                 chunk_size, PAGE_SIZE);
2030                         return -EINVAL;
2031                 }
2032
2033                 /* devices must have minimum size of one chunk */
2034                 ITERATE_RDEV(mddev,rdev,tmp) {
2035                         if (test_bit(Faulty, &rdev->flags))
2036                                 continue;
2037                         if (rdev->size < chunk_size / 1024) {
2038                                 printk(KERN_WARNING
2039                                         "md: Dev %s smaller than chunk_size:"
2040                                         " %lluk < %dk\n",
2041                                         bdevname(rdev->bdev,b),
2042                                         (unsigned long long)rdev->size,
2043                                         chunk_size / 1024);
2044                                 return -EINVAL;
2045                         }
2046                 }
2047         }
2048
2049 #ifdef CONFIG_KMOD
2050         request_module("md-level-%d", mddev->level);
2051 #endif
2052
2053         /*
2054          * Drop all container device buffers, from now on
2055          * the only valid external interface is through the md
2056          * device.
2057          * Also find largest hardsector size
2058          */
2059         ITERATE_RDEV(mddev,rdev,tmp) {
2060                 if (test_bit(Faulty, &rdev->flags))
2061                         continue;
2062                 sync_blockdev(rdev->bdev);
2063                 invalidate_bdev(rdev->bdev, 0);
2064         }
2065
2066         md_probe(mddev->unit, NULL, NULL);
2067         disk = mddev->gendisk;
2068         if (!disk)
2069                 return -ENOMEM;
2070
2071         spin_lock(&pers_lock);
2072         pers = find_pers(mddev->level);
2073         if (!pers || !try_module_get(pers->owner)) {
2074                 spin_unlock(&pers_lock);
2075                 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2076                        mddev->level);
2077                 return -EINVAL;
2078         }
2079         mddev->pers = pers;
2080         spin_unlock(&pers_lock);
2081
2082         mddev->recovery = 0;
2083         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2084         mddev->barriers_work = 1;
2085         mddev->ok_start_degraded = start_dirty_degraded;
2086
2087         if (start_readonly)
2088                 mddev->ro = 2; /* read-only, but switch on first write */
2089
2090         err = mddev->pers->run(mddev);
2091         if (!err && mddev->pers->sync_request) {
2092                 err = bitmap_create(mddev);
2093                 if (err) {
2094                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2095                                mdname(mddev), err);
2096                         mddev->pers->stop(mddev);
2097                 }
2098         }
2099         if (err) {
2100                 printk(KERN_ERR "md: pers->run() failed ...\n");
2101                 module_put(mddev->pers->owner);
2102                 mddev->pers = NULL;
2103                 bitmap_destroy(mddev);
2104                 return err;
2105         }
2106         if (mddev->pers->sync_request)
2107                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2108         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2109                 mddev->ro = 0;
2110
2111         atomic_set(&mddev->writes_pending,0);
2112         mddev->safemode = 0;
2113         mddev->safemode_timer.function = md_safemode_timeout;
2114         mddev->safemode_timer.data = (unsigned long) mddev;
2115         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2116         mddev->in_sync = 1;
2117
2118         ITERATE_RDEV(mddev,rdev,tmp)
2119                 if (rdev->raid_disk >= 0) {
2120                         char nm[20];
2121                         sprintf(nm, "rd%d", rdev->raid_disk);
2122                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2123                 }
2124         
2125         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2126         md_wakeup_thread(mddev->thread);
2127         
2128         if (mddev->sb_dirty)
2129                 md_update_sb(mddev);
2130
2131         set_capacity(disk, mddev->array_size<<1);
2132
2133         /* If we call blk_queue_make_request here, it will
2134          * re-initialise max_sectors etc which may have been
2135          * refined inside -> run.  So just set the bits we need to set.
2136          * Most initialisation happended when we called
2137          * blk_queue_make_request(..., md_fail_request)
2138          * earlier.
2139          */
2140         mddev->queue->queuedata = mddev;
2141         mddev->queue->make_request_fn = mddev->pers->make_request;
2142
2143         mddev->changed = 1;
2144         md_new_event(mddev);
2145         return 0;
2146 }
2147
2148 static int restart_array(mddev_t *mddev)
2149 {
2150         struct gendisk *disk = mddev->gendisk;
2151         int err;
2152
2153         /*
2154          * Complain if it has no devices
2155          */
2156         err = -ENXIO;
2157         if (list_empty(&mddev->disks))
2158                 goto out;
2159
2160         if (mddev->pers) {
2161                 err = -EBUSY;
2162                 if (!mddev->ro)
2163                         goto out;
2164
2165                 mddev->safemode = 0;
2166                 mddev->ro = 0;
2167                 set_disk_ro(disk, 0);
2168
2169                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2170                         mdname(mddev));
2171                 /*
2172                  * Kick recovery or resync if necessary
2173                  */
2174                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2175                 md_wakeup_thread(mddev->thread);
2176                 err = 0;
2177         } else {
2178                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2179                         mdname(mddev));
2180                 err = -EINVAL;
2181         }
2182
2183 out:
2184         return err;
2185 }
2186
2187 static int do_md_stop(mddev_t * mddev, int ro)
2188 {
2189         int err = 0;
2190         struct gendisk *disk = mddev->gendisk;
2191
2192         if (mddev->pers) {
2193                 if (atomic_read(&mddev->active)>2) {
2194                         printk("md: %s still in use.\n",mdname(mddev));
2195                         return -EBUSY;
2196                 }
2197
2198                 if (mddev->sync_thread) {
2199                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2200                         md_unregister_thread(mddev->sync_thread);
2201                         mddev->sync_thread = NULL;
2202                 }
2203
2204                 del_timer_sync(&mddev->safemode_timer);
2205
2206                 invalidate_partition(disk, 0);
2207
2208                 if (ro) {
2209                         err  = -ENXIO;
2210                         if (mddev->ro==1)
2211                                 goto out;
2212                         mddev->ro = 1;
2213                 } else {
2214                         bitmap_flush(mddev);
2215                         md_super_wait(mddev);
2216                         if (mddev->ro)
2217                                 set_disk_ro(disk, 0);
2218                         blk_queue_make_request(mddev->queue, md_fail_request);
2219                         mddev->pers->stop(mddev);
2220                         if (mddev->pers->sync_request)
2221                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2222
2223                         module_put(mddev->pers->owner);
2224                         mddev->pers = NULL;
2225                         if (mddev->ro)
2226                                 mddev->ro = 0;
2227                 }
2228                 if (!mddev->in_sync) {
2229                         /* mark array as shutdown cleanly */
2230                         mddev->in_sync = 1;
2231                         md_update_sb(mddev);
2232                 }
2233                 if (ro)
2234                         set_disk_ro(disk, 1);
2235         }
2236
2237         bitmap_destroy(mddev);
2238         if (mddev->bitmap_file) {
2239                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2240                 fput(mddev->bitmap_file);
2241                 mddev->bitmap_file = NULL;
2242         }
2243         mddev->bitmap_offset = 0;
2244
2245         /*
2246          * Free resources if final stop
2247          */
2248         if (!ro) {
2249                 mdk_rdev_t *rdev;
2250                 struct list_head *tmp;
2251                 struct gendisk *disk;
2252                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2253
2254                 ITERATE_RDEV(mddev,rdev,tmp)
2255                         if (rdev->raid_disk >= 0) {
2256                                 char nm[20];
2257                                 sprintf(nm, "rd%d", rdev->raid_disk);
2258                                 sysfs_remove_link(&mddev->kobj, nm);
2259                         }
2260
2261                 export_array(mddev);
2262
2263                 mddev->array_size = 0;
2264                 disk = mddev->gendisk;
2265                 if (disk)
2266                         set_capacity(disk, 0);
2267                 mddev->changed = 1;
2268         } else
2269                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2270                         mdname(mddev));
2271         err = 0;
2272         md_new_event(mddev);
2273 out:
2274         return err;
2275 }
2276
2277 static void autorun_array(mddev_t *mddev)
2278 {
2279         mdk_rdev_t *rdev;
2280         struct list_head *tmp;
2281         int err;
2282
2283         if (list_empty(&mddev->disks))
2284                 return;
2285
2286         printk(KERN_INFO "md: running: ");
2287
2288         ITERATE_RDEV(mddev,rdev,tmp) {
2289                 char b[BDEVNAME_SIZE];
2290                 printk("<%s>", bdevname(rdev->bdev,b));
2291         }
2292         printk("\n");
2293
2294         err = do_md_run (mddev);
2295         if (err) {
2296                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2297                 do_md_stop (mddev, 0);
2298         }
2299 }
2300
2301 /*
2302  * lets try to run arrays based on all disks that have arrived
2303  * until now. (those are in pending_raid_disks)
2304  *
2305  * the method: pick the first pending disk, collect all disks with
2306  * the same UUID, remove all from the pending list and put them into
2307  * the 'same_array' list. Then order this list based on superblock
2308  * update time (freshest comes first), kick out 'old' disks and
2309  * compare superblocks. If everything's fine then run it.
2310  *
2311  * If "unit" is allocated, then bump its reference count
2312  */
2313 static void autorun_devices(int part)
2314 {
2315         struct list_head candidates;
2316         struct list_head *tmp;
2317         mdk_rdev_t *rdev0, *rdev;
2318         mddev_t *mddev;
2319         char b[BDEVNAME_SIZE];
2320
2321         printk(KERN_INFO "md: autorun ...\n");
2322         while (!list_empty(&pending_raid_disks)) {
2323                 dev_t dev;
2324                 rdev0 = list_entry(pending_raid_disks.next,
2325                                          mdk_rdev_t, same_set);
2326
2327                 printk(KERN_INFO "md: considering %s ...\n",
2328                         bdevname(rdev0->bdev,b));
2329                 INIT_LIST_HEAD(&candidates);
2330                 ITERATE_RDEV_PENDING(rdev,tmp)
2331                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2332                                 printk(KERN_INFO "md:  adding %s ...\n",
2333                                         bdevname(rdev->bdev,b));
2334                                 list_move(&rdev->same_set, &candidates);
2335                         }
2336                 /*
2337                  * now we have a set of devices, with all of them having
2338                  * mostly sane superblocks. It's time to allocate the
2339                  * mddev.
2340                  */
2341                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2342                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2343                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2344                         break;
2345                 }
2346                 if (part)
2347                         dev = MKDEV(mdp_major,
2348                                     rdev0->preferred_minor << MdpMinorShift);
2349                 else
2350                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2351
2352                 md_probe(dev, NULL, NULL);
2353                 mddev = mddev_find(dev);
2354                 if (!mddev) {
2355                         printk(KERN_ERR 
2356                                 "md: cannot allocate memory for md drive.\n");
2357                         break;
2358                 }
2359                 if (mddev_lock(mddev)) 
2360                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2361                                mdname(mddev));
2362                 else if (mddev->raid_disks || mddev->major_version
2363                          || !list_empty(&mddev->disks)) {
2364                         printk(KERN_WARNING 
2365                                 "md: %s already running, cannot run %s\n",
2366                                 mdname(mddev), bdevname(rdev0->bdev,b));
2367                         mddev_unlock(mddev);
2368                 } else {
2369                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2370                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2371                                 list_del_init(&rdev->same_set);
2372                                 if (bind_rdev_to_array(rdev, mddev))
2373                                         export_rdev(rdev);
2374                         }
2375                         autorun_array(mddev);
2376                         mddev_unlock(mddev);
2377                 }
2378                 /* on success, candidates will be empty, on error
2379                  * it won't...
2380                  */
2381                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2382                         export_rdev(rdev);
2383                 mddev_put(mddev);
2384         }
2385         printk(KERN_INFO "md: ... autorun DONE.\n");
2386 }
2387
2388 /*
2389  * import RAID devices based on one partition
2390  * if possible, the array gets run as well.
2391  */
2392
2393 static int autostart_array(dev_t startdev)
2394 {
2395         char b[BDEVNAME_SIZE];
2396         int err = -EINVAL, i;
2397         mdp_super_t *sb = NULL;
2398         mdk_rdev_t *start_rdev = NULL, *rdev;
2399
2400         start_rdev = md_import_device(startdev, 0, 0);
2401         if (IS_ERR(start_rdev))
2402                 return err;
2403
2404
2405         /* NOTE: this can only work for 0.90.0 superblocks */
2406         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2407         if (sb->major_version != 0 ||
2408             sb->minor_version != 90 ) {
2409                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2410                 export_rdev(start_rdev);
2411                 return err;
2412         }
2413
2414         if (test_bit(Faulty, &start_rdev->flags)) {
2415                 printk(KERN_WARNING 
2416                         "md: can not autostart based on faulty %s!\n",
2417                         bdevname(start_rdev->bdev,b));
2418                 export_rdev(start_rdev);
2419                 return err;
2420         }
2421         list_add(&start_rdev->same_set, &pending_raid_disks);
2422
2423         for (i = 0; i < MD_SB_DISKS; i++) {
2424                 mdp_disk_t *desc = sb->disks + i;
2425                 dev_t dev = MKDEV(desc->major, desc->minor);
2426
2427                 if (!dev)
2428                         continue;
2429                 if (dev == startdev)
2430                         continue;
2431                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2432                         continue;
2433                 rdev = md_import_device(dev, 0, 0);
2434                 if (IS_ERR(rdev))
2435                         continue;
2436
2437                 list_add(&rdev->same_set, &pending_raid_disks);
2438         }
2439
2440         /*
2441          * possibly return codes
2442          */
2443         autorun_devices(0);
2444         return 0;
2445
2446 }
2447
2448
2449 static int get_version(void __user * arg)
2450 {
2451         mdu_version_t ver;
2452
2453         ver.major = MD_MAJOR_VERSION;
2454         ver.minor = MD_MINOR_VERSION;
2455         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2456
2457         if (copy_to_user(arg, &ver, sizeof(ver)))
2458                 return -EFAULT;
2459
2460         return 0;
2461 }
2462
2463 static int get_array_info(mddev_t * mddev, void __user * arg)
2464 {
2465         mdu_array_info_t info;
2466         int nr,working,active,failed,spare;
2467         mdk_rdev_t *rdev;
2468         struct list_head *tmp;
2469
2470         nr=working=active=failed=spare=0;
2471         ITERATE_RDEV(mddev,rdev,tmp) {
2472                 nr++;
2473                 if (test_bit(Faulty, &rdev->flags))
2474                         failed++;
2475                 else {
2476                         working++;
2477                         if (test_bit(In_sync, &rdev->flags))
2478                                 active++;       
2479                         else
2480                                 spare++;
2481                 }
2482         }
2483
2484         info.major_version = mddev->major_version;
2485         info.minor_version = mddev->minor_version;
2486         info.patch_version = MD_PATCHLEVEL_VERSION;
2487         info.ctime         = mddev->ctime;
2488         info.level         = mddev->level;
2489         info.size          = mddev->size;
2490         info.nr_disks      = nr;
2491         info.raid_disks    = mddev->raid_disks;
2492         info.md_minor      = mddev->md_minor;
2493         info.not_persistent= !mddev->persistent;
2494
2495         info.utime         = mddev->utime;
2496         info.state         = 0;
2497         if (mddev->in_sync)
2498                 info.state = (1<<MD_SB_CLEAN);
2499         if (mddev->bitmap && mddev->bitmap_offset)
2500                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2501         info.active_disks  = active;
2502         info.working_disks = working;
2503         info.failed_disks  = failed;
2504         info.spare_disks   = spare;
2505
2506         info.layout        = mddev->layout;
2507         info.chunk_size    = mddev->chunk_size;
2508
2509         if (copy_to_user(arg, &info, sizeof(info)))
2510                 return -EFAULT;
2511
2512         return 0;
2513 }
2514
2515 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2516 {
2517         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2518         char *ptr, *buf = NULL;
2519         int err = -ENOMEM;
2520
2521         file = kmalloc(sizeof(*file), GFP_KERNEL);
2522         if (!file)
2523                 goto out;
2524
2525         /* bitmap disabled, zero the first byte and copy out */
2526         if (!mddev->bitmap || !mddev->bitmap->file) {
2527                 file->pathname[0] = '\0';
2528                 goto copy_out;
2529         }
2530
2531         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2532         if (!buf)
2533                 goto out;
2534
2535         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2536         if (!ptr)
2537                 goto out;
2538
2539         strcpy(file->pathname, ptr);
2540
2541 copy_out:
2542         err = 0;
2543         if (copy_to_user(arg, file, sizeof(*file)))
2544                 err = -EFAULT;
2545 out:
2546         kfree(buf);
2547         kfree(file);
2548         return err;
2549 }
2550
2551 static int get_disk_info(mddev_t * mddev, void __user * arg)
2552 {
2553         mdu_disk_info_t info;
2554         unsigned int nr;
2555         mdk_rdev_t *rdev;
2556
2557         if (copy_from_user(&info, arg, sizeof(info)))
2558                 return -EFAULT;
2559
2560         nr = info.number;
2561
2562         rdev = find_rdev_nr(mddev, nr);
2563         if (rdev) {
2564                 info.major = MAJOR(rdev->bdev->bd_dev);
2565                 info.minor = MINOR(rdev->bdev->bd_dev);
2566                 info.raid_disk = rdev->raid_disk;
2567                 info.state = 0;
2568                 if (test_bit(Faulty, &rdev->flags))
2569                         info.state |= (1<<MD_DISK_FAULTY);
2570                 else if (test_bit(In_sync, &rdev->flags)) {
2571                         info.state |= (1<<MD_DISK_ACTIVE);
2572                         info.state |= (1<<MD_DISK_SYNC);
2573                 }
2574                 if (test_bit(WriteMostly, &rdev->flags))
2575                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2576         } else {
2577                 info.major = info.minor = 0;
2578                 info.raid_disk = -1;
2579                 info.state = (1<<MD_DISK_REMOVED);
2580         }
2581
2582         if (copy_to_user(arg, &info, sizeof(info)))
2583                 return -EFAULT;
2584
2585         return 0;
2586 }
2587
2588 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2589 {
2590         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2591         mdk_rdev_t *rdev;
2592         dev_t dev = MKDEV(info->major,info->minor);
2593
2594         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2595                 return -EOVERFLOW;
2596
2597         if (!mddev->raid_disks) {
2598                 int err;
2599                 /* expecting a device which has a superblock */
2600                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2601                 if (IS_ERR(rdev)) {
2602                         printk(KERN_WARNING 
2603                                 "md: md_import_device returned %ld\n",
2604                                 PTR_ERR(rdev));
2605                         return PTR_ERR(rdev);
2606                 }
2607                 if (!list_empty(&mddev->disks)) {
2608                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2609                                                         mdk_rdev_t, same_set);
2610                         int err = super_types[mddev->major_version]
2611                                 .load_super(rdev, rdev0, mddev->minor_version);
2612                         if (err < 0) {
2613                                 printk(KERN_WARNING 
2614                                         "md: %s has different UUID to %s\n",
2615                                         bdevname(rdev->bdev,b), 
2616                                         bdevname(rdev0->bdev,b2));
2617                                 export_rdev(rdev);
2618                                 return -EINVAL;
2619                         }
2620                 }
2621                 err = bind_rdev_to_array(rdev, mddev);
2622                 if (err)
2623                         export_rdev(rdev);
2624                 return err;
2625         }
2626
2627         /*
2628          * add_new_disk can be used once the array is assembled
2629          * to add "hot spares".  They must already have a superblock
2630          * written
2631          */
2632         if (mddev->pers) {
2633                 int err;
2634                 if (!mddev->pers->hot_add_disk) {
2635                         printk(KERN_WARNING 
2636                                 "%s: personality does not support diskops!\n",
2637                                mdname(mddev));
2638                         return -EINVAL;
2639                 }
2640                 if (mddev->persistent)
2641                         rdev = md_import_device(dev, mddev->major_version,
2642                                                 mddev->minor_version);
2643                 else
2644                         rdev = md_import_device(dev, -1, -1);
2645                 if (IS_ERR(rdev)) {
2646                         printk(KERN_WARNING 
2647                                 "md: md_import_device returned %ld\n",
2648                                 PTR_ERR(rdev));
2649                         return PTR_ERR(rdev);
2650                 }
2651                 /* set save_raid_disk if appropriate */
2652                 if (!mddev->persistent) {
2653                         if (info->state & (1<<MD_DISK_SYNC)  &&
2654                             info->raid_disk < mddev->raid_disks)
2655                                 rdev->raid_disk = info->raid_disk;
2656                         else
2657                                 rdev->raid_disk = -1;
2658                 } else
2659                         super_types[mddev->major_version].
2660                                 validate_super(mddev, rdev);
2661                 rdev->saved_raid_disk = rdev->raid_disk;
2662
2663                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2664                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2665                         set_bit(WriteMostly, &rdev->flags);
2666
2667                 rdev->raid_disk = -1;
2668                 err = bind_rdev_to_array(rdev, mddev);
2669                 if (err)
2670                         export_rdev(rdev);
2671
2672                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2673                 md_wakeup_thread(mddev->thread);
2674                 return err;
2675         }
2676
2677         /* otherwise, add_new_disk is only allowed
2678          * for major_version==0 superblocks
2679          */
2680         if (mddev->major_version != 0) {
2681                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2682                        mdname(mddev));
2683                 return -EINVAL;
2684         }
2685
2686         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2687                 int err;
2688                 rdev = md_import_device (dev, -1, 0);
2689                 if (IS_ERR(rdev)) {
2690                         printk(KERN_WARNING 
2691                                 "md: error, md_import_device() returned %ld\n",
2692                                 PTR_ERR(rdev));
2693                         return PTR_ERR(rdev);
2694                 }
2695                 rdev->desc_nr = info->number;
2696                 if (info->raid_disk < mddev->raid_disks)
2697                         rdev->raid_disk = info->raid_disk;
2698                 else
2699                         rdev->raid_disk = -1;
2700
2701                 rdev->flags = 0;
2702
2703                 if (rdev->raid_disk < mddev->raid_disks)
2704                         if (info->state & (1<<MD_DISK_SYNC))
2705                                 set_bit(In_sync, &rdev->flags);
2706
2707                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2708                         set_bit(WriteMostly, &rdev->flags);
2709
2710                 err = bind_rdev_to_array(rdev, mddev);
2711                 if (err) {
2712                         export_rdev(rdev);
2713                         return err;
2714                 }
2715
2716                 if (!mddev->persistent) {
2717                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2718                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2719                 } else 
2720                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2721                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2722
2723                 if (!mddev->size || (mddev->size > rdev->size))
2724                         mddev->size = rdev->size;
2725         }
2726
2727         return 0;
2728 }
2729
2730 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2731 {
2732         char b[BDEVNAME_SIZE];
2733         mdk_rdev_t *rdev;
2734
2735         if (!mddev->pers)
2736                 return -ENODEV;
2737
2738         rdev = find_rdev(mddev, dev);
2739         if (!rdev)
2740                 return -ENXIO;
2741
2742         if (rdev->raid_disk >= 0)
2743                 goto busy;
2744
2745         kick_rdev_from_array(rdev);
2746         md_update_sb(mddev);
2747         md_new_event(mddev);
2748
2749         return 0;
2750 busy:
2751         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2752                 bdevname(rdev->bdev,b), mdname(mddev));
2753         return -EBUSY;
2754 }
2755
2756 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2757 {
2758         char b[BDEVNAME_SIZE];
2759         int err;
2760         unsigned int size;
2761         mdk_rdev_t *rdev;
2762
2763         if (!mddev->pers)
2764                 return -ENODEV;
2765
2766         if (mddev->major_version != 0) {
2767                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2768                         " version-0 superblocks.\n",
2769                         mdname(mddev));
2770                 return -EINVAL;
2771         }
2772         if (!mddev->pers->hot_add_disk) {
2773                 printk(KERN_WARNING 
2774                         "%s: personality does not support diskops!\n",
2775                         mdname(mddev));
2776                 return -EINVAL;
2777         }
2778
2779         rdev = md_import_device (dev, -1, 0);
2780         if (IS_ERR(rdev)) {
2781                 printk(KERN_WARNING 
2782                         "md: error, md_import_device() returned %ld\n",
2783                         PTR_ERR(rdev));
2784                 return -EINVAL;
2785         }
2786
2787         if (mddev->persistent)
2788                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2789         else
2790                 rdev->sb_offset =
2791                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2792
2793         size = calc_dev_size(rdev, mddev->chunk_size);
2794         rdev->size = size;
2795
2796         if (size < mddev->size) {
2797                 printk(KERN_WARNING 
2798                         "%s: disk size %llu blocks < array size %llu\n",
2799                         mdname(mddev), (unsigned long long)size,
2800                         (unsigned long long)mddev->size);
2801                 err = -ENOSPC;
2802                 goto abort_export;
2803         }
2804
2805         if (test_bit(Faulty, &rdev->flags)) {
2806                 printk(KERN_WARNING 
2807                         "md: can not hot-add faulty %s disk to %s!\n",
2808                         bdevname(rdev->bdev,b), mdname(mddev));
2809                 err = -EINVAL;
2810                 goto abort_export;
2811         }
2812         clear_bit(In_sync, &rdev->flags);
2813         rdev->desc_nr = -1;
2814         bind_rdev_to_array(rdev, mddev);
2815
2816         /*
2817          * The rest should better be atomic, we can have disk failures
2818          * noticed in interrupt contexts ...
2819          */
2820
2821         if (rdev->desc_nr == mddev->max_disks) {
2822                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2823                         mdname(mddev));
2824                 err = -EBUSY;
2825                 goto abort_unbind_export;
2826         }
2827
2828         rdev->raid_disk = -1;
2829
2830         md_update_sb(mddev);
2831
2832         /*
2833          * Kick recovery, maybe this spare has to be added to the
2834          * array immediately.
2835          */
2836         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2837         md_wakeup_thread(mddev->thread);
2838         md_new_event(mddev);
2839         return 0;
2840
2841 abort_unbind_export:
2842         unbind_rdev_from_array(rdev);
2843
2844 abort_export:
2845         export_rdev(rdev);
2846         return err;
2847 }
2848
2849 /* similar to deny_write_access, but accounts for our holding a reference
2850  * to the file ourselves */
2851 static int deny_bitmap_write_access(struct file * file)
2852 {
2853         struct inode *inode = file->f_mapping->host;
2854
2855         spin_lock(&inode->i_lock);
2856         if (atomic_read(&inode->i_writecount) > 1) {
2857                 spin_unlock(&inode->i_lock);
2858                 return -ETXTBSY;
2859         }
2860         atomic_set(&inode->i_writecount, -1);
2861         spin_unlock(&inode->i_lock);
2862
2863         return 0;
2864 }
2865
2866 static int set_bitmap_file(mddev_t *mddev, int fd)
2867 {
2868         int err;
2869
2870         if (mddev->pers) {
2871                 if (!mddev->pers->quiesce)
2872                         return -EBUSY;
2873                 if (mddev->recovery || mddev->sync_thread)
2874                         return -EBUSY;
2875                 /* we should be able to change the bitmap.. */
2876         }
2877
2878
2879         if (fd >= 0) {
2880                 if (mddev->bitmap)
2881                         return -EEXIST; /* cannot add when bitmap is present */
2882                 mddev->bitmap_file = fget(fd);
2883
2884                 if (mddev->bitmap_file == NULL) {
2885                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2886                                mdname(mddev));
2887                         return -EBADF;
2888                 }
2889
2890                 err = deny_bitmap_write_access(mddev->bitmap_file);
2891                 if (err) {
2892                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2893                                mdname(mddev));
2894                         fput(mddev->bitmap_file);
2895                         mddev->bitmap_file = NULL;
2896                         return err;
2897                 }
2898                 mddev->bitmap_offset = 0; /* file overrides offset */
2899         } else if (mddev->bitmap == NULL)
2900                 return -ENOENT; /* cannot remove what isn't there */
2901         err = 0;
2902         if (mddev->pers) {
2903                 mddev->pers->quiesce(mddev, 1);
2904                 if (fd >= 0)
2905                         err = bitmap_create(mddev);
2906                 if (fd < 0 || err)
2907                         bitmap_destroy(mddev);
2908                 mddev->pers->quiesce(mddev, 0);
2909         } else if (fd < 0) {
2910                 if (mddev->bitmap_file)
2911                         fput(mddev->bitmap_file);
2912                 mddev->bitmap_file = NULL;
2913         }
2914
2915         return err;
2916 }
2917
2918 /*
2919  * set_array_info is used two different ways
2920  * The original usage is when creating a new array.
2921  * In this usage, raid_disks is > 0 and it together with
2922  *  level, size, not_persistent,layout,chunksize determine the
2923  *  shape of the array.
2924  *  This will always create an array with a type-0.90.0 superblock.
2925  * The newer usage is when assembling an array.
2926  *  In this case raid_disks will be 0, and the major_version field is
2927  *  use to determine which style super-blocks are to be found on the devices.
2928  *  The minor and patch _version numbers are also kept incase the
2929  *  super_block handler wishes to interpret them.
2930  */
2931 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2932 {
2933
2934         if (info->raid_disks == 0) {
2935                 /* just setting version number for superblock loading */
2936                 if (info->major_version < 0 ||
2937                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2938                     super_types[info->major_version].name == NULL) {
2939                         /* maybe try to auto-load a module? */
2940                         printk(KERN_INFO 
2941                                 "md: superblock version %d not known\n",
2942                                 info->major_version);
2943                         return -EINVAL;
2944                 }
2945                 mddev->major_version = info->major_version;
2946                 mddev->minor_version = info->minor_version;
2947                 mddev->patch_version = info->patch_version;
2948                 return 0;
2949         }
2950         mddev->major_version = MD_MAJOR_VERSION;
2951         mddev->minor_version = MD_MINOR_VERSION;
2952         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2953         mddev->ctime         = get_seconds();
2954
2955         mddev->level         = info->level;
2956         mddev->size          = info->size;
2957         mddev->raid_disks    = info->raid_disks;
2958         /* don't set md_minor, it is determined by which /dev/md* was
2959          * openned
2960          */
2961         if (info->state & (1<<MD_SB_CLEAN))
2962                 mddev->recovery_cp = MaxSector;
2963         else
2964                 mddev->recovery_cp = 0;
2965         mddev->persistent    = ! info->not_persistent;
2966
2967         mddev->layout        = info->layout;
2968         mddev->chunk_size    = info->chunk_size;
2969
2970         mddev->max_disks     = MD_SB_DISKS;
2971
2972         mddev->sb_dirty      = 1;
2973
2974         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2975         mddev->bitmap_offset = 0;
2976
2977         /*
2978          * Generate a 128 bit UUID
2979          */
2980         get_random_bytes(mddev->uuid, 16);
2981
2982         return 0;
2983 }
2984
2985 /*
2986  * update_array_info is used to change the configuration of an
2987  * on-line array.
2988  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2989  * fields in the info are checked against the array.
2990  * Any differences that cannot be handled will cause an error.
2991  * Normally, only one change can be managed at a time.
2992  */
2993 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2994 {
2995         int rv = 0;
2996         int cnt = 0;
2997         int state = 0;
2998
2999         /* calculate expected state,ignoring low bits */
3000         if (mddev->bitmap && mddev->bitmap_offset)
3001                 state |= (1 << MD_SB_BITMAP_PRESENT);
3002
3003         if (mddev->major_version != info->major_version ||
3004             mddev->minor_version != info->minor_version ||
3005 /*          mddev->patch_version != info->patch_version || */
3006             mddev->ctime         != info->ctime         ||
3007             mddev->level         != info->level         ||
3008 /*          mddev->layout        != info->layout        || */
3009             !mddev->persistent   != info->not_persistent||
3010             mddev->chunk_size    != info->chunk_size    ||
3011             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3012             ((state^info->state) & 0xfffffe00)
3013                 )
3014                 return -EINVAL;
3015         /* Check there is only one change */
3016         if (mddev->size != info->size) cnt++;
3017         if (mddev->raid_disks != info->raid_disks) cnt++;
3018         if (mddev->layout != info->layout) cnt++;
3019         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3020         if (cnt == 0) return 0;
3021         if (cnt > 1) return -EINVAL;
3022
3023         if (mddev->layout != info->layout) {
3024                 /* Change layout
3025                  * we don't need to do anything at the md level, the
3026                  * personality will take care of it all.
3027                  */
3028                 if (mddev->pers->reconfig == NULL)
3029                         return -EINVAL;
3030                 else
3031                         return mddev->pers->reconfig(mddev, info->layout, -1);
3032         }
3033         if (mddev->size != info->size) {
3034                 mdk_rdev_t * rdev;
3035                 struct list_head *tmp;
3036                 if (mddev->pers->resize == NULL)
3037                         return -EINVAL;
3038                 /* The "size" is the amount of each device that is used.
3039                  * This can only make sense for arrays with redundancy.
3040                  * linear and raid0 always use whatever space is available
3041                  * We can only consider changing the size if no resync
3042                  * or reconstruction is happening, and if the new size
3043                  * is acceptable. It must fit before the sb_offset or,
3044                  * if that is <data_offset, it must fit before the
3045                  * size of each device.
3046                  * If size is zero, we find the largest size that fits.
3047                  */
3048                 if (mddev->sync_thread)
3049                         return -EBUSY;
3050                 ITERATE_RDEV(mddev,rdev,tmp) {
3051                         sector_t avail;
3052                         int fit = (info->size == 0);
3053                         if (rdev->sb_offset > rdev->data_offset)
3054                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3055                         else
3056                                 avail = get_capacity(rdev->bdev->bd_disk)
3057                                         - rdev->data_offset;
3058                         if (fit && (info->size == 0 || info->size > avail/2))
3059                                 info->size = avail/2;
3060                         if (avail < ((sector_t)info->size << 1))
3061                                 return -ENOSPC;
3062                 }
3063                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3064                 if (!rv) {
3065                         struct block_device *bdev;
3066
3067                         bdev = bdget_disk(mddev->gendisk, 0);
3068                         if (bdev) {
3069                                 down(&bdev->bd_inode->i_sem);
3070                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3071                                 up(&bdev->bd_inode->i_sem);
3072                                 bdput(bdev);
3073                         }
3074                 }
3075         }
3076         if (mddev->raid_disks    != info->raid_disks) {
3077                 /* change the number of raid disks */
3078                 if (mddev->pers->reshape == NULL)
3079                         return -EINVAL;
3080                 if (info->raid_disks <= 0 ||
3081                     info->raid_disks >= mddev->max_disks)
3082                         return -EINVAL;
3083                 if (mddev->sync_thread)
3084                         return -EBUSY;
3085                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3086                 if (!rv) {
3087                         struct block_device *bdev;
3088
3089                         bdev = bdget_disk(mddev->gendisk, 0);
3090                         if (bdev) {
3091                                 down(&bdev->bd_inode->i_sem);
3092                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3093                                 up(&bdev->bd_inode->i_sem);
3094                                 bdput(bdev);
3095                         }
3096                 }
3097         }
3098         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3099                 if (mddev->pers->quiesce == NULL)
3100                         return -EINVAL;
3101                 if (mddev->recovery || mddev->sync_thread)
3102                         return -EBUSY;
3103                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3104                         /* add the bitmap */
3105                         if (mddev->bitmap)
3106                                 return -EEXIST;
3107                         if (mddev->default_bitmap_offset == 0)
3108                                 return -EINVAL;
3109                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3110                         mddev->pers->quiesce(mddev, 1);
3111                         rv = bitmap_create(mddev);
3112                         if (rv)
3113                                 bitmap_destroy(mddev);
3114                         mddev->pers->quiesce(mddev, 0);
3115                 } else {
3116                         /* remove the bitmap */
3117                         if (!mddev->bitmap)
3118                                 return -ENOENT;
3119                         if (mddev->bitmap->file)
3120                                 return -EINVAL;
3121                         mddev->pers->quiesce(mddev, 1);
3122                         bitmap_destroy(mddev);
3123                         mddev->pers->quiesce(mddev, 0);
3124                         mddev->bitmap_offset = 0;
3125                 }
3126         }
3127         md_update_sb(mddev);
3128         return rv;
3129 }
3130
3131 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3132 {
3133         mdk_rdev_t *rdev;
3134
3135         if (mddev->pers == NULL)
3136                 return -ENODEV;
3137
3138         rdev = find_rdev(mddev, dev);
3139         if (!rdev)
3140                 return -ENODEV;
3141
3142         md_error(mddev, rdev);
3143         return 0;
3144 }
3145
3146 static int md_ioctl(struct inode *inode, struct file *file,
3147                         unsigned int cmd, unsigned long arg)
3148 {
3149         int err = 0;
3150         void __user *argp = (void __user *)arg;
3151         struct hd_geometry __user *loc = argp;
3152         mddev_t *mddev = NULL;
3153
3154         if (!capable(CAP_SYS_ADMIN))
3155                 return -EACCES;
3156
3157         /*
3158          * Commands dealing with the RAID driver but not any
3159          * particular array:
3160          */
3161         switch (cmd)
3162         {
3163                 case RAID_VERSION:
3164                         err = get_version(argp);
3165                         goto done;
3166
3167                 case PRINT_RAID_DEBUG:
3168                         err = 0;
3169                         md_print_devices();
3170                         goto done;
3171
3172 #ifndef MODULE
3173                 case RAID_AUTORUN:
3174                         err = 0;
3175                         autostart_arrays(arg);
3176                         goto done;
3177 #endif
3178                 default:;
3179         }
3180
3181         /*
3182          * Commands creating/starting a new array:
3183          */
3184
3185         mddev = inode->i_bdev->bd_disk->private_data;
3186
3187         if (!mddev) {
3188                 BUG();
3189                 goto abort;
3190         }
3191
3192
3193         if (cmd == START_ARRAY) {
3194                 /* START_ARRAY doesn't need to lock the array as autostart_array
3195                  * does the locking, and it could even be a different array
3196                  */
3197                 static int cnt = 3;
3198                 if (cnt > 0 ) {
3199                         printk(KERN_WARNING
3200                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3201                                "This will not be supported beyond July 2006\n",
3202                                current->comm, current->pid);
3203                         cnt--;
3204                 }
3205                 err = autostart_array(new_decode_dev(arg));
3206                 if (err) {
3207                         printk(KERN_WARNING "md: autostart failed!\n");
3208                         goto abort;
3209                 }
3210                 goto done;
3211         }
3212
3213         err = mddev_lock(mddev);
3214         if (err) {
3215                 printk(KERN_INFO 
3216                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3217                         err, cmd);
3218                 goto abort;
3219         }
3220
3221         switch (cmd)
3222         {
3223                 case SET_ARRAY_INFO:
3224                         {
3225                                 mdu_array_info_t info;
3226                                 if (!arg)
3227                                         memset(&info, 0, sizeof(info));
3228                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3229                                         err = -EFAULT;
3230                                         goto abort_unlock;
3231                                 }
3232                                 if (mddev->pers) {
3233                                         err = update_array_info(mddev, &info);
3234                                         if (err) {
3235                                                 printk(KERN_WARNING "md: couldn't update"
3236                                                        " array info. %d\n", err);
3237                                                 goto abort_unlock;
3238                                         }
3239                                         goto done_unlock;
3240                                 }
3241                                 if (!list_empty(&mddev->disks)) {
3242                                         printk(KERN_WARNING
3243                                                "md: array %s already has disks!\n",
3244                                                mdname(mddev));
3245                                         err = -EBUSY;
3246                                         goto abort_unlock;
3247                                 }
3248                                 if (mddev->raid_disks) {
3249                                         printk(KERN_WARNING
3250                                                "md: array %s already initialised!\n",
3251                                                mdname(mddev));
3252                                         err = -EBUSY;
3253                                         goto abort_unlock;
3254                                 }
3255                                 err = set_array_info(mddev, &info);
3256                                 if (err) {
3257                                         printk(KERN_WARNING "md: couldn't set"
3258                                                " array info. %d\n", err);
3259                                         goto abort_unlock;
3260                                 }
3261                         }
3262                         goto done_unlock;
3263
3264                 default:;
3265         }
3266
3267         /*
3268          * Commands querying/configuring an existing array:
3269          */
3270         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3271          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3272         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3273                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3274                 err = -ENODEV;
3275                 goto abort_unlock;
3276         }
3277
3278         /*
3279          * Commands even a read-only array can execute:
3280          */
3281         switch (cmd)
3282         {
3283                 case GET_ARRAY_INFO:
3284                         err = get_array_info(mddev, argp);
3285                         goto done_unlock;
3286
3287                 case GET_BITMAP_FILE:
3288                         err = get_bitmap_file(mddev, argp);
3289                         goto done_unlock;
3290
3291                 case GET_DISK_INFO:
3292                         err = get_disk_info(mddev, argp);
3293                         goto done_unlock;
3294
3295                 case RESTART_ARRAY_RW:
3296                         err = restart_array(mddev);
3297                         goto done_unlock;
3298
3299                 case STOP_ARRAY:
3300                         err = do_md_stop (mddev, 0);
3301                         goto done_unlock;
3302
3303                 case STOP_ARRAY_RO:
3304                         err = do_md_stop (mddev, 1);
3305                         goto done_unlock;
3306
3307         /*
3308          * We have a problem here : there is no easy way to give a CHS
3309          * virtual geometry. We currently pretend that we have a 2 heads
3310          * 4 sectors (with a BIG number of cylinders...). This drives
3311          * dosfs just mad... ;-)
3312          */
3313                 case HDIO_GETGEO:
3314                         if (!loc) {
3315                                 err = -EINVAL;
3316                                 goto abort_unlock;
3317                         }
3318                         err = put_user (2, (char __user *) &loc->heads);
3319                         if (err)
3320                                 goto abort_unlock;
3321                         err = put_user (4, (char __user *) &loc->sectors);
3322                         if (err)
3323                                 goto abort_unlock;
3324                         err = put_user(get_capacity(mddev->gendisk)/8,
3325                                         (short __user *) &loc->cylinders);
3326                         if (err)
3327                                 goto abort_unlock;
3328                         err = put_user (get_start_sect(inode->i_bdev),
3329                                                 (long __user *) &loc->start);
3330                         goto done_unlock;
3331         }
3332
3333         /*
3334          * The remaining ioctls are changing the state of the
3335          * superblock, so we do not allow them on read-only arrays.
3336          * However non-MD ioctls (e.g. get-size) will still come through
3337          * here and hit the 'default' below, so only disallow
3338          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3339          */
3340         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3341             mddev->ro && mddev->pers) {
3342                 if (mddev->ro == 2) {
3343                         mddev->ro = 0;
3344                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3345                 md_wakeup_thread(mddev->thread);
3346
3347                 } else {
3348                         err = -EROFS;
3349                         goto abort_unlock;
3350                 }
3351         }
3352
3353         switch (cmd)
3354         {
3355                 case ADD_NEW_DISK:
3356                 {
3357                         mdu_disk_info_t info;
3358                         if (copy_from_user(&info, argp, sizeof(info)))
3359                                 err = -EFAULT;
3360                         else
3361                                 err = add_new_disk(mddev, &info);
3362                         goto done_unlock;
3363                 }
3364
3365                 case HOT_REMOVE_DISK:
3366                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3367                         goto done_unlock;
3368
3369                 case HOT_ADD_DISK:
3370                         err = hot_add_disk(mddev, new_decode_dev(arg));
3371                         goto done_unlock;
3372
3373                 case SET_DISK_FAULTY:
3374                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3375                         goto done_unlock;
3376
3377                 case RUN_ARRAY:
3378                         err = do_md_run (mddev);
3379                         goto done_unlock;
3380
3381                 case SET_BITMAP_FILE:
3382                         err = set_bitmap_file(mddev, (int)arg);
3383                         goto done_unlock;
3384
3385                 default:
3386                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3387                                 printk(KERN_WARNING "md: %s(pid %d) used"
3388                                         " obsolete MD ioctl, upgrade your"
3389                                         " software to use new ictls.\n",
3390                                         current->comm, current->pid);
3391                         err = -EINVAL;
3392                         goto abort_unlock;
3393         }
3394
3395 done_unlock:
3396 abort_unlock:
3397         mddev_unlock(mddev);
3398
3399         return err;
3400 done:
3401         if (err)
3402                 MD_BUG();
3403 abort:
3404         return err;
3405 }
3406
3407 static int md_open(struct inode *inode, struct file *file)
3408 {
3409         /*
3410          * Succeed if we can lock the mddev, which confirms that
3411          * it isn't being stopped right now.
3412          */
3413         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3414         int err;
3415
3416         if ((err = mddev_lock(mddev)))
3417                 goto out;
3418
3419         err = 0;
3420         mddev_get(mddev);
3421         mddev_unlock(mddev);
3422
3423         check_disk_change(inode->i_bdev);
3424  out:
3425         return err;
3426 }
3427
3428 static int md_release(struct inode *inode, struct file * file)
3429 {
3430         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3431
3432         if (!mddev)
3433                 BUG();
3434         mddev_put(mddev);
3435
3436         return 0;
3437 }
3438
3439 static int md_media_changed(struct gendisk *disk)
3440 {
3441         mddev_t *mddev = disk->private_data;
3442
3443         return mddev->changed;
3444 }
3445
3446 static int md_revalidate(struct gendisk *disk)
3447 {
3448         mddev_t *mddev = disk->private_data;
3449
3450         mddev->changed = 0;
3451         return 0;
3452 }
3453 static struct block_device_operations md_fops =
3454 {
3455         .owner          = THIS_MODULE,
3456         .open           = md_open,
3457         .release        = md_release,
3458         .ioctl          = md_ioctl,
3459         .media_changed  = md_media_changed,
3460         .revalidate_disk= md_revalidate,
3461 };
3462
3463 static int md_thread(void * arg)
3464 {
3465         mdk_thread_t *thread = arg;
3466
3467         /*
3468          * md_thread is a 'system-thread', it's priority should be very
3469          * high. We avoid resource deadlocks individually in each
3470          * raid personality. (RAID5 does preallocation) We also use RR and
3471          * the very same RT priority as kswapd, thus we will never get
3472          * into a priority inversion deadlock.
3473          *
3474          * we definitely have to have equal or higher priority than
3475          * bdflush, otherwise bdflush will deadlock if there are too
3476          * many dirty RAID5 blocks.
3477          */
3478
3479         allow_signal(SIGKILL);
3480         while (!kthread_should_stop()) {
3481
3482                 /* We need to wait INTERRUPTIBLE so that
3483                  * we don't add to the load-average.
3484                  * That means we need to be sure no signals are
3485                  * pending
3486                  */
3487                 if (signal_pending(current))
3488                         flush_signals(current);
3489
3490                 wait_event_interruptible_timeout
3491                         (thread->wqueue,
3492                          test_bit(THREAD_WAKEUP, &thread->flags)
3493                          || kthread_should_stop(),
3494                          thread->timeout);
3495                 try_to_freeze();
3496
3497                 clear_bit(THREAD_WAKEUP, &thread->flags);
3498
3499                 thread->run(thread->mddev);
3500         }
3501
3502         return 0;
3503 }
3504
3505 void md_wakeup_thread(mdk_thread_t *thread)
3506 {
3507         if (thread) {
3508                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3509                 set_bit(THREAD_WAKEUP, &thread->flags);
3510                 wake_up(&thread->wqueue);
3511         }
3512 }
3513
3514 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3515                                  const char *name)
3516 {
3517         mdk_thread_t *thread;
3518
3519         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3520         if (!thread)
3521                 return NULL;
3522
3523         init_waitqueue_head(&thread->wqueue);
3524
3525         thread->run = run;
3526         thread->mddev = mddev;
3527         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3528         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3529         if (IS_ERR(thread->tsk)) {
3530                 kfree(thread);
3531                 return NULL;
3532         }
3533         return thread;
3534 }
3535
3536 void md_unregister_thread(mdk_thread_t *thread)
3537 {
3538         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3539
3540         kthread_stop(thread->tsk);
3541         kfree(thread);
3542 }
3543
3544 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3545 {
3546         if (!mddev) {
3547                 MD_BUG();
3548                 return;
3549         }
3550
3551         if (!rdev || test_bit(Faulty, &rdev->flags))
3552                 return;
3553 /*
3554         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3555                 mdname(mddev),
3556                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3557                 __builtin_return_address(0),__builtin_return_address(1),
3558                 __builtin_return_address(2),__builtin_return_address(3));
3559 */
3560         if (!mddev->pers->error_handler)
3561                 return;
3562         mddev->pers->error_handler(mddev,rdev);
3563         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3564         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3565         md_wakeup_thread(mddev->thread);
3566         md_new_event(mddev);
3567 }
3568
3569 /* seq_file implementation /proc/mdstat */
3570
3571 static void status_unused(struct seq_file *seq)
3572 {
3573         int i = 0;
3574         mdk_rdev_t *rdev;
3575         struct list_head *tmp;
3576
3577         seq_printf(seq, "unused devices: ");
3578
3579         ITERATE_RDEV_PENDING(rdev,tmp) {
3580                 char b[BDEVNAME_SIZE];
3581                 i++;
3582                 seq_printf(seq, "%s ",
3583                               bdevname(rdev->bdev,b));
3584         }
3585         if (!i)
3586                 seq_printf(seq, "<none>");
3587
3588         seq_printf(seq, "\n");
3589 }
3590
3591
3592 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3593 {
3594         unsigned long max_blocks, resync, res, dt, db, rt;
3595
3596         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3597
3598         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3599                 max_blocks = mddev->resync_max_sectors >> 1;
3600         else
3601                 max_blocks = mddev->size;
3602
3603         /*
3604          * Should not happen.
3605          */
3606         if (!max_blocks) {
3607                 MD_BUG();
3608                 return;
3609         }
3610         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3611         {
3612                 int i, x = res/50, y = 20-x;
3613                 seq_printf(seq, "[");
3614                 for (i = 0; i < x; i++)
3615                         seq_printf(seq, "=");
3616                 seq_printf(seq, ">");
3617                 for (i = 0; i < y; i++)
3618                         seq_printf(seq, ".");
3619                 seq_printf(seq, "] ");
3620         }
3621         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3622                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3623                        "resync" : "recovery"),
3624                       res/10, res % 10, resync, max_blocks);
3625
3626         /*
3627          * We do not want to overflow, so the order of operands and
3628          * the * 100 / 100 trick are important. We do a +1 to be
3629          * safe against division by zero. We only estimate anyway.
3630          *
3631          * dt: time from mark until now
3632          * db: blocks written from mark until now
3633          * rt: remaining time
3634          */
3635         dt = ((jiffies - mddev->resync_mark) / HZ);
3636         if (!dt) dt++;
3637         db = resync - (mddev->resync_mark_cnt/2);
3638         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3639
3640         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3641
3642         seq_printf(seq, " speed=%ldK/sec", db/dt);
3643 }
3644
3645 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3646 {
3647         struct list_head *tmp;
3648         loff_t l = *pos;
3649         mddev_t *mddev;
3650
3651         if (l >= 0x10000)
3652                 return NULL;
3653         if (!l--)
3654                 /* header */
3655                 return (void*)1;
3656
3657         spin_lock(&all_mddevs_lock);
3658         list_for_each(tmp,&all_mddevs)
3659                 if (!l--) {
3660                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3661                         mddev_get(mddev);
3662                         spin_unlock(&all_mddevs_lock);
3663                         return mddev;
3664                 }
3665         spin_unlock(&all_mddevs_lock);
3666         if (!l--)
3667                 return (void*)2;/* tail */
3668         return NULL;
3669 }
3670
3671 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3672 {
3673         struct list_head *tmp;
3674         mddev_t *next_mddev, *mddev = v;
3675         
3676         ++*pos;
3677         if (v == (void*)2)
3678                 return NULL;
3679
3680         spin_lock(&all_mddevs_lock);
3681         if (v == (void*)1)
3682                 tmp = all_mddevs.next;
3683         else
3684                 tmp = mddev->all_mddevs.next;
3685         if (tmp != &all_mddevs)
3686                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3687         else {
3688                 next_mddev = (void*)2;
3689                 *pos = 0x10000;
3690         }               
3691         spin_unlock(&all_mddevs_lock);
3692
3693         if (v != (void*)1)
3694                 mddev_put(mddev);
3695         return next_mddev;
3696
3697 }
3698
3699 static void md_seq_stop(struct seq_file *seq, void *v)
3700 {
3701         mddev_t *mddev = v;
3702
3703         if (mddev && v != (void*)1 && v != (void*)2)
3704                 mddev_put(mddev);
3705 }
3706
3707 struct mdstat_info {
3708         int event;
3709 };
3710
3711 static int md_seq_show(struct seq_file *seq, void *v)
3712 {
3713         mddev_t *mddev = v;
3714         sector_t size;
3715         struct list_head *tmp2;
3716         mdk_rdev_t *rdev;
3717         struct mdstat_info *mi = seq->private;
3718         struct bitmap *bitmap;
3719
3720         if (v == (void*)1) {
3721                 struct mdk_personality *pers;
3722                 seq_printf(seq, "Personalities : ");
3723                 spin_lock(&pers_lock);
3724                 list_for_each_entry(pers, &pers_list, list)
3725                         seq_printf(seq, "[%s] ", pers->name);
3726
3727                 spin_unlock(&pers_lock);
3728                 seq_printf(seq, "\n");
3729                 mi->event = atomic_read(&md_event_count);
3730                 return 0;
3731         }
3732         if (v == (void*)2) {
3733                 status_unused(seq);
3734                 return 0;
3735         }
3736
3737         if (mddev_lock(mddev)!=0) 
3738                 return -EINTR;
3739         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3740                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3741                                                 mddev->pers ? "" : "in");
3742                 if (mddev->pers) {
3743                         if (mddev->ro==1)
3744                                 seq_printf(seq, " (read-only)");
3745                         if (mddev->ro==2)
3746                                 seq_printf(seq, "(auto-read-only)");
3747                         seq_printf(seq, " %s", mddev->pers->name);
3748                 }
3749
3750                 size = 0;
3751                 ITERATE_RDEV(mddev,rdev,tmp2) {
3752                         char b[BDEVNAME_SIZE];
3753                         seq_printf(seq, " %s[%d]",
3754                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3755                         if (test_bit(WriteMostly, &rdev->flags))
3756                                 seq_printf(seq, "(W)");
3757                         if (test_bit(Faulty, &rdev->flags)) {
3758                                 seq_printf(seq, "(F)");
3759                                 continue;
3760                         } else if (rdev->raid_disk < 0)
3761                                 seq_printf(seq, "(S)"); /* spare */
3762                         size += rdev->size;
3763                 }
3764
3765                 if (!list_empty(&mddev->disks)) {
3766                         if (mddev->pers)
3767                                 seq_printf(seq, "\n      %llu blocks",
3768                                         (unsigned long long)mddev->array_size);
3769                         else
3770                                 seq_printf(seq, "\n      %llu blocks",
3771                                         (unsigned long long)size);
3772                 }
3773                 if (mddev->persistent) {
3774                         if (mddev->major_version != 0 ||
3775                             mddev->minor_version != 90) {
3776                                 seq_printf(seq," super %d.%d",
3777                                            mddev->major_version,
3778                                            mddev->minor_version);
3779                         }
3780                 } else
3781                         seq_printf(seq, " super non-persistent");
3782
3783                 if (mddev->pers) {
3784                         mddev->pers->status (seq, mddev);
3785                         seq_printf(seq, "\n      ");
3786                         if (mddev->pers->sync_request) {
3787                                 if (mddev->curr_resync > 2) {
3788                                         status_resync (seq, mddev);
3789                                         seq_printf(seq, "\n      ");
3790                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3791                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3792                                 else if (mddev->recovery_cp < MaxSector)
3793                                         seq_printf(seq, "\tresync=PENDING\n      ");
3794                         }
3795                 } else
3796                         seq_printf(seq, "\n       ");
3797
3798                 if ((bitmap = mddev->bitmap)) {
3799                         unsigned long chunk_kb;
3800                         unsigned long flags;
3801                         spin_lock_irqsave(&bitmap->lock, flags);
3802                         chunk_kb = bitmap->chunksize >> 10;
3803                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3804                                 "%lu%s chunk",
3805                                 bitmap->pages - bitmap->missing_pages,
3806                                 bitmap->pages,
3807                                 (bitmap->pages - bitmap->missing_pages)
3808                                         << (PAGE_SHIFT - 10),
3809                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3810                                 chunk_kb ? "KB" : "B");
3811                         if (bitmap->file) {
3812                                 seq_printf(seq, ", file: ");
3813                                 seq_path(seq, bitmap->file->f_vfsmnt,
3814                                          bitmap->file->f_dentry," \t\n");
3815                         }
3816
3817                         seq_printf(seq, "\n");
3818                         spin_unlock_irqrestore(&bitmap->lock, flags);
3819                 }
3820
3821                 seq_printf(seq, "\n");
3822         }
3823         mddev_unlock(mddev);
3824         
3825         return 0;
3826 }
3827
3828 static struct seq_operations md_seq_ops = {
3829         .start  = md_seq_start,
3830         .next   = md_seq_next,
3831         .stop   = md_seq_stop,
3832         .show   = md_seq_show,
3833 };
3834
3835 static int md_seq_open(struct inode *inode, struct file *file)
3836 {
3837         int error;
3838         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3839         if (mi == NULL)
3840                 return -ENOMEM;
3841
3842         error = seq_open(file, &md_seq_ops);
3843         if (error)
3844                 kfree(mi);
3845         else {
3846                 struct seq_file *p = file->private_data;
3847                 p->private = mi;
3848                 mi->event = atomic_read(&md_event_count);
3849         }
3850         return error;
3851 }
3852
3853 static int md_seq_release(struct inode *inode, struct file *file)
3854 {
3855         struct seq_file *m = file->private_data;
3856         struct mdstat_info *mi = m->private;
3857         m->private = NULL;
3858         kfree(mi);
3859         return seq_release(inode, file);
3860 }
3861
3862 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
3863 {
3864         struct seq_file *m = filp->private_data;
3865         struct mdstat_info *mi = m->private;
3866         int mask;
3867
3868         poll_wait(filp, &md_event_waiters, wait);
3869
3870         /* always allow read */
3871         mask = POLLIN | POLLRDNORM;
3872
3873         if (mi->event != atomic_read(&md_event_count))
3874                 mask |= POLLERR | POLLPRI;
3875         return mask;
3876 }
3877
3878 static struct file_operations md_seq_fops = {
3879         .open           = md_seq_open,
3880         .read           = seq_read,
3881         .llseek         = seq_lseek,
3882         .release        = md_seq_release,
3883         .poll           = mdstat_poll,
3884 };
3885
3886 int register_md_personality(struct mdk_personality *p)
3887 {
3888         spin_lock(&pers_lock);
3889         list_add_tail(&p->list, &pers_list);
3890         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
3891         spin_unlock(&pers_lock);
3892         return 0;
3893 }
3894
3895 int unregister_md_personality(struct mdk_personality *p)
3896 {
3897         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
3898         spin_lock(&pers_lock);
3899         list_del_init(&p->list);
3900         spin_unlock(&pers_lock);
3901         return 0;
3902 }
3903
3904 static int is_mddev_idle(mddev_t *mddev)
3905 {
3906         mdk_rdev_t * rdev;
3907         struct list_head *tmp;
3908         int idle;
3909         unsigned long curr_events;
3910
3911         idle = 1;
3912         ITERATE_RDEV(mddev,rdev,tmp) {
3913                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3914                 curr_events = disk_stat_read(disk, sectors[0]) + 
3915                                 disk_stat_read(disk, sectors[1]) - 
3916                                 atomic_read(&disk->sync_io);
3917                 /* The difference between curr_events and last_events
3918                  * will be affected by any new non-sync IO (making
3919                  * curr_events bigger) and any difference in the amount of
3920                  * in-flight syncio (making current_events bigger or smaller)
3921                  * The amount in-flight is currently limited to
3922                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3923                  * which is at most 4096 sectors.
3924                  * These numbers are fairly fragile and should be made
3925                  * more robust, probably by enforcing the
3926                  * 'window size' that md_do_sync sort-of uses.
3927                  *
3928                  * Note: the following is an unsigned comparison.
3929                  */
3930                 if ((curr_events - rdev->last_events + 4096) > 8192) {
3931                         rdev->last_events = curr_events;
3932                         idle = 0;
3933                 }
3934         }
3935         return idle;
3936 }
3937
3938 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3939 {
3940         /* another "blocks" (512byte) blocks have been synced */
3941         atomic_sub(blocks, &mddev->recovery_active);
3942         wake_up(&mddev->recovery_wait);
3943         if (!ok) {
3944                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3945                 md_wakeup_thread(mddev->thread);
3946                 // stop recovery, signal do_sync ....
3947         }
3948 }
3949
3950
3951 /* md_write_start(mddev, bi)
3952  * If we need to update some array metadata (e.g. 'active' flag
3953  * in superblock) before writing, schedule a superblock update
3954  * and wait for it to complete.
3955  */
3956 void md_write_start(mddev_t *mddev, struct bio *bi)
3957 {
3958         if (bio_data_dir(bi) != WRITE)
3959                 return;
3960
3961         BUG_ON(mddev->ro == 1);
3962         if (mddev->ro == 2) {
3963                 /* need to switch to read/write */
3964                 mddev->ro = 0;
3965                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3966                 md_wakeup_thread(mddev->thread);
3967         }
3968         atomic_inc(&mddev->writes_pending);
3969         if (mddev->in_sync) {
3970                 spin_lock_irq(&mddev->write_lock);
3971                 if (mddev->in_sync) {
3972                         mddev->in_sync = 0;
3973                         mddev->sb_dirty = 1;
3974                         md_wakeup_thread(mddev->thread);
3975                 }
3976                 spin_unlock_irq(&mddev->write_lock);
3977         }
3978         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3979 }
3980
3981 void md_write_end(mddev_t *mddev)
3982 {
3983         if (atomic_dec_and_test(&mddev->writes_pending)) {
3984                 if (mddev->safemode == 2)
3985                         md_wakeup_thread(mddev->thread);
3986                 else
3987                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3988         }
3989 }
3990
3991 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3992
3993 #define SYNC_MARKS      10
3994 #define SYNC_MARK_STEP  (3*HZ)
3995 static void md_do_sync(mddev_t *mddev)
3996 {
3997         mddev_t *mddev2;
3998         unsigned int currspeed = 0,
3999                  window;
4000         sector_t max_sectors,j, io_sectors;
4001         unsigned long mark[SYNC_MARKS];
4002         sector_t mark_cnt[SYNC_MARKS];
4003         int last_mark,m;
4004         struct list_head *tmp;
4005         sector_t last_check;
4006         int skipped = 0;
4007
4008         /* just incase thread restarts... */
4009         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4010                 return;
4011
4012         /* we overload curr_resync somewhat here.
4013          * 0 == not engaged in resync at all
4014          * 2 == checking that there is no conflict with another sync
4015          * 1 == like 2, but have yielded to allow conflicting resync to
4016          *              commense
4017          * other == active in resync - this many blocks
4018          *
4019          * Before starting a resync we must have set curr_resync to
4020          * 2, and then checked that every "conflicting" array has curr_resync
4021          * less than ours.  When we find one that is the same or higher
4022          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4023          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4024          * This will mean we have to start checking from the beginning again.
4025          *
4026          */
4027
4028         do {
4029                 mddev->curr_resync = 2;
4030
4031         try_again:
4032                 if (kthread_should_stop()) {
4033                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4034                         goto skip;
4035                 }
4036                 ITERATE_MDDEV(mddev2,tmp) {
4037                         if (mddev2 == mddev)
4038                                 continue;
4039                         if (mddev2->curr_resync && 
4040                             match_mddev_units(mddev,mddev2)) {
4041                                 DEFINE_WAIT(wq);
4042                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4043                                         /* arbitrarily yield */
4044                                         mddev->curr_resync = 1;
4045                                         wake_up(&resync_wait);
4046                                 }
4047                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4048                                         /* no need to wait here, we can wait the next
4049                                          * time 'round when curr_resync == 2
4050                                          */
4051                                         continue;
4052                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4053                                 if (!kthread_should_stop() &&
4054                                     mddev2->curr_resync >= mddev->curr_resync) {
4055                                         printk(KERN_INFO "md: delaying resync of %s"
4056                                                " until %s has finished resync (they"
4057                                                " share one or more physical units)\n",
4058                                                mdname(mddev), mdname(mddev2));
4059                                         mddev_put(mddev2);
4060                                         schedule();
4061                                         finish_wait(&resync_wait, &wq);
4062                                         goto try_again;
4063                                 }
4064                                 finish_wait(&resync_wait, &wq);
4065                         }
4066                 }
4067         } while (mddev->curr_resync < 2);
4068
4069         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4070                 /* resync follows the size requested by the personality,
4071                  * which defaults to physical size, but can be virtual size
4072                  */
4073                 max_sectors = mddev->resync_max_sectors;
4074                 mddev->resync_mismatches = 0;
4075         } else
4076                 /* recovery follows the physical size of devices */
4077                 max_sectors = mddev->size << 1;
4078
4079         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4080         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4081                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4082         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4083                "(but not more than %d KB/sec) for reconstruction.\n",
4084                sysctl_speed_limit_max);
4085
4086         is_mddev_idle(mddev); /* this also initializes IO event counters */
4087         /* we don't use the checkpoint if there's a bitmap */
4088         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4089             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4090                 j = mddev->recovery_cp;
4091         else
4092                 j = 0;
4093         io_sectors = 0;
4094         for (m = 0; m < SYNC_MARKS; m++) {
4095                 mark[m] = jiffies;
4096                 mark_cnt[m] = io_sectors;
4097         }
4098         last_mark = 0;
4099         mddev->resync_mark = mark[last_mark];
4100         mddev->resync_mark_cnt = mark_cnt[last_mark];
4101
4102         /*
4103          * Tune reconstruction:
4104          */
4105         window = 32*(PAGE_SIZE/512);
4106         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4107                 window/2,(unsigned long long) max_sectors/2);
4108
4109         atomic_set(&mddev->recovery_active, 0);
4110         init_waitqueue_head(&mddev->recovery_wait);
4111         last_check = 0;
4112
4113         if (j>2) {
4114                 printk(KERN_INFO 
4115                         "md: resuming recovery of %s from checkpoint.\n",
4116                         mdname(mddev));
4117                 mddev->curr_resync = j;
4118         }
4119
4120         while (j < max_sectors) {
4121                 sector_t sectors;
4122
4123                 skipped = 0;
4124                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4125                                             currspeed < sysctl_speed_limit_min);
4126                 if (sectors == 0) {
4127                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4128                         goto out;
4129                 }
4130
4131                 if (!skipped) { /* actual IO requested */
4132                         io_sectors += sectors;
4133                         atomic_add(sectors, &mddev->recovery_active);
4134                 }
4135
4136                 j += sectors;
4137                 if (j>1) mddev->curr_resync = j;
4138                 if (last_check == 0)
4139                         /* this is the earliers that rebuilt will be
4140                          * visible in /proc/mdstat
4141                          */
4142                         md_new_event(mddev);
4143
4144                 if (last_check + window > io_sectors || j == max_sectors)
4145                         continue;
4146
4147                 last_check = io_sectors;
4148
4149                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4150                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4151                         break;
4152
4153         repeat:
4154                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4155                         /* step marks */
4156                         int next = (last_mark+1) % SYNC_MARKS;
4157
4158                         mddev->resync_mark = mark[next];
4159                         mddev->resync_mark_cnt = mark_cnt[next];
4160                         mark[next] = jiffies;
4161                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4162                         last_mark = next;
4163                 }
4164
4165
4166                 if (kthread_should_stop()) {
4167                         /*
4168                          * got a signal, exit.
4169                          */
4170                         printk(KERN_INFO 
4171                                 "md: md_do_sync() got signal ... exiting\n");
4172                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4173                         goto out;
4174                 }
4175
4176                 /*
4177                  * this loop exits only if either when we are slower than
4178                  * the 'hard' speed limit, or the system was IO-idle for
4179                  * a jiffy.
4180                  * the system might be non-idle CPU-wise, but we only care
4181                  * about not overloading the IO subsystem. (things like an
4182                  * e2fsck being done on the RAID array should execute fast)
4183                  */
4184                 mddev->queue->unplug_fn(mddev->queue);
4185                 cond_resched();
4186
4187                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4188                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4189
4190                 if (currspeed > sysctl_speed_limit_min) {
4191                         if ((currspeed > sysctl_speed_limit_max) ||
4192                                         !is_mddev_idle(mddev)) {
4193                                 msleep(500);
4194                                 goto repeat;
4195                         }
4196                 }
4197         }
4198         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4199         /*
4200          * this also signals 'finished resyncing' to md_stop
4201          */
4202  out:
4203         mddev->queue->unplug_fn(mddev->queue);
4204
4205         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4206
4207         /* tell personality that we are finished */
4208         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4209
4210         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4211             mddev->curr_resync > 2 &&
4212             mddev->curr_resync >= mddev->recovery_cp) {
4213                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4214                         printk(KERN_INFO 
4215                                 "md: checkpointing recovery of %s.\n",
4216                                 mdname(mddev));
4217                         mddev->recovery_cp = mddev->curr_resync;
4218                 } else
4219                         mddev->recovery_cp = MaxSector;
4220         }
4221
4222  skip:
4223         mddev->curr_resync = 0;
4224         wake_up(&resync_wait);
4225         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4226         md_wakeup_thread(mddev->thread);
4227 }
4228
4229
4230 /*
4231  * This routine is regularly called by all per-raid-array threads to
4232  * deal with generic issues like resync and super-block update.
4233  * Raid personalities that don't have a thread (linear/raid0) do not
4234  * need this as they never do any recovery or update the superblock.
4235  *
4236  * It does not do any resync itself, but rather "forks" off other threads
4237  * to do that as needed.
4238  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4239  * "->recovery" and create a thread at ->sync_thread.
4240  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4241  * and wakeups up this thread which will reap the thread and finish up.
4242  * This thread also removes any faulty devices (with nr_pending == 0).
4243  *
4244  * The overall approach is:
4245  *  1/ if the superblock needs updating, update it.
4246  *  2/ If a recovery thread is running, don't do anything else.
4247  *  3/ If recovery has finished, clean up, possibly marking spares active.
4248  *  4/ If there are any faulty devices, remove them.
4249  *  5/ If array is degraded, try to add spares devices
4250  *  6/ If array has spares or is not in-sync, start a resync thread.
4251  */
4252 void md_check_recovery(mddev_t *mddev)
4253 {
4254         mdk_rdev_t *rdev;
4255         struct list_head *rtmp;
4256
4257
4258         if (mddev->bitmap)
4259                 bitmap_daemon_work(mddev->bitmap);
4260
4261         if (mddev->ro)
4262                 return;
4263
4264         if (signal_pending(current)) {
4265                 if (mddev->pers->sync_request) {
4266                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4267                                mdname(mddev));
4268                         mddev->safemode = 2;
4269                 }
4270                 flush_signals(current);
4271         }
4272
4273         if ( ! (
4274                 mddev->sb_dirty ||
4275                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4276                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4277                 (mddev->safemode == 1) ||
4278                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4279                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4280                 ))
4281                 return;
4282
4283         if (mddev_trylock(mddev)==0) {
4284                 int spares =0;
4285
4286                 spin_lock_irq(&mddev->write_lock);
4287                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4288                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4289                         mddev->in_sync = 1;
4290                         mddev->sb_dirty = 1;
4291                 }
4292                 if (mddev->safemode == 1)
4293                         mddev->safemode = 0;
4294                 spin_unlock_irq(&mddev->write_lock);
4295
4296                 if (mddev->sb_dirty)
4297                         md_update_sb(mddev);
4298
4299
4300                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4301                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4302                         /* resync/recovery still happening */
4303                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4304                         goto unlock;
4305                 }
4306                 if (mddev->sync_thread) {
4307                         /* resync has finished, collect result */
4308                         md_unregister_thread(mddev->sync_thread);
4309                         mddev->sync_thread = NULL;
4310                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4311                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4312                                 /* success...*/
4313                                 /* activate any spares */
4314                                 mddev->pers->spare_active(mddev);
4315                         }
4316                         md_update_sb(mddev);
4317
4318                         /* if array is no-longer degraded, then any saved_raid_disk
4319                          * information must be scrapped
4320                          */
4321                         if (!mddev->degraded)
4322                                 ITERATE_RDEV(mddev,rdev,rtmp)
4323                                         rdev->saved_raid_disk = -1;
4324
4325                         mddev->recovery = 0;
4326                         /* flag recovery needed just to double check */
4327                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4328                         md_new_event(mddev);
4329                         goto unlock;
4330                 }
4331                 /* Clear some bits that don't mean anything, but
4332                  * might be left set
4333                  */
4334                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4335                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4336                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4337                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4338
4339                 /* no recovery is running.
4340                  * remove any failed drives, then
4341                  * add spares if possible.
4342                  * Spare are also removed and re-added, to allow
4343                  * the personality to fail the re-add.
4344                  */
4345                 ITERATE_RDEV(mddev,rdev,rtmp)
4346                         if (rdev->raid_disk >= 0 &&
4347                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4348                             atomic_read(&rdev->nr_pending)==0) {
4349                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4350                                         char nm[20];
4351                                         sprintf(nm,"rd%d", rdev->raid_disk);
4352                                         sysfs_remove_link(&mddev->kobj, nm);
4353                                         rdev->raid_disk = -1;
4354                                 }
4355                         }
4356
4357                 if (mddev->degraded) {
4358                         ITERATE_RDEV(mddev,rdev,rtmp)
4359                                 if (rdev->raid_disk < 0
4360                                     && !test_bit(Faulty, &rdev->flags)) {
4361                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4362                                                 char nm[20];
4363                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4364                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4365                                                 spares++;
4366                                                 md_new_event(mddev);
4367                                         } else
4368                                                 break;
4369                                 }
4370                 }
4371
4372                 if (spares) {
4373                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4374                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4375                 } else if (mddev->recovery_cp < MaxSector) {
4376                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4377                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4378                         /* nothing to be done ... */
4379                         goto unlock;
4380
4381                 if (mddev->pers->sync_request) {
4382                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4383                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4384                                 /* We are adding a device or devices to an array
4385                                  * which has the bitmap stored on all devices.
4386                                  * So make sure all bitmap pages get written
4387                                  */
4388                                 bitmap_write_all(mddev->bitmap);
4389                         }
4390                         mddev->sync_thread = md_register_thread(md_do_sync,
4391                                                                 mddev,
4392                                                                 "%s_resync");
4393                         if (!mddev->sync_thread) {
4394                                 printk(KERN_ERR "%s: could not start resync"
4395                                         " thread...\n", 
4396                                         mdname(mddev));
4397                                 /* leave the spares where they are, it shouldn't hurt */
4398                                 mddev->recovery = 0;
4399                         } else
4400                                 md_wakeup_thread(mddev->sync_thread);
4401                         md_new_event(mddev);
4402                 }
4403         unlock:
4404                 mddev_unlock(mddev);
4405         }
4406 }
4407
4408 static int md_notify_reboot(struct notifier_block *this,
4409                             unsigned long code, void *x)
4410 {
4411         struct list_head *tmp;
4412         mddev_t *mddev;
4413
4414         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4415
4416                 printk(KERN_INFO "md: stopping all md devices.\n");
4417
4418                 ITERATE_MDDEV(mddev,tmp)
4419                         if (mddev_trylock(mddev)==0)
4420                                 do_md_stop (mddev, 1);
4421                 /*
4422                  * certain more exotic SCSI devices are known to be
4423                  * volatile wrt too early system reboots. While the
4424                  * right place to handle this issue is the given
4425                  * driver, we do want to have a safe RAID driver ...
4426                  */
4427                 mdelay(1000*1);
4428         }
4429         return NOTIFY_DONE;
4430 }
4431
4432 static struct notifier_block md_notifier = {
4433         .notifier_call  = md_notify_reboot,
4434         .next           = NULL,
4435         .priority       = INT_MAX, /* before any real devices */
4436 };
4437
4438 static void md_geninit(void)
4439 {
4440         struct proc_dir_entry *p;
4441
4442         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4443
4444         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4445         if (p)
4446                 p->proc_fops = &md_seq_fops;
4447 }
4448
4449 static int __init md_init(void)
4450 {
4451         int minor;
4452
4453         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4454                         " MD_SB_DISKS=%d\n",
4455                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4456                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4457         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4458                         BITMAP_MINOR);
4459
4460         if (register_blkdev(MAJOR_NR, "md"))
4461                 return -1;
4462         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4463                 unregister_blkdev(MAJOR_NR, "md");
4464                 return -1;
4465         }
4466         devfs_mk_dir("md");
4467         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4468                                 md_probe, NULL, NULL);
4469         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4470                             md_probe, NULL, NULL);
4471
4472         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4473                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4474                                 S_IFBLK|S_IRUSR|S_IWUSR,
4475                                 "md/%d", minor);
4476
4477         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4478                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4479                               S_IFBLK|S_IRUSR|S_IWUSR,
4480                               "md/mdp%d", minor);
4481
4482
4483         register_reboot_notifier(&md_notifier);
4484         raid_table_header = register_sysctl_table(raid_root_table, 1);
4485
4486         md_geninit();
4487         return (0);
4488 }
4489
4490
4491 #ifndef MODULE
4492
4493 /*
4494  * Searches all registered partitions for autorun RAID arrays
4495  * at boot time.
4496  */
4497 static dev_t detected_devices[128];
4498 static int dev_cnt;
4499
4500 void md_autodetect_dev(dev_t dev)
4501 {
4502         if (dev_cnt >= 0 && dev_cnt < 127)
4503                 detected_devices[dev_cnt++] = dev;
4504 }
4505
4506
4507 static void autostart_arrays(int part)
4508 {
4509         mdk_rdev_t *rdev;
4510         int i;
4511
4512         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4513
4514         for (i = 0; i < dev_cnt; i++) {
4515                 dev_t dev = detected_devices[i];
4516
4517                 rdev = md_import_device(dev,0, 0);
4518                 if (IS_ERR(rdev))
4519                         continue;
4520
4521                 if (test_bit(Faulty, &rdev->flags)) {
4522                         MD_BUG();
4523                         continue;
4524                 }
4525                 list_add(&rdev->same_set, &pending_raid_disks);
4526         }
4527         dev_cnt = 0;
4528
4529         autorun_devices(part);
4530 }
4531
4532 #endif
4533
4534 static __exit void md_exit(void)
4535 {
4536         mddev_t *mddev;
4537         struct list_head *tmp;
4538         int i;
4539         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4540         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4541         for (i=0; i < MAX_MD_DEVS; i++)
4542                 devfs_remove("md/%d", i);
4543         for (i=0; i < MAX_MD_DEVS; i++)
4544                 devfs_remove("md/d%d", i);
4545
4546         devfs_remove("md");
4547
4548         unregister_blkdev(MAJOR_NR,"md");
4549         unregister_blkdev(mdp_major, "mdp");
4550         unregister_reboot_notifier(&md_notifier);
4551         unregister_sysctl_table(raid_table_header);
4552         remove_proc_entry("mdstat", NULL);
4553         ITERATE_MDDEV(mddev,tmp) {
4554                 struct gendisk *disk = mddev->gendisk;
4555                 if (!disk)
4556                         continue;
4557                 export_array(mddev);
4558                 del_gendisk(disk);
4559                 put_disk(disk);
4560                 mddev->gendisk = NULL;
4561                 mddev_put(mddev);
4562         }
4563 }
4564
4565 module_init(md_init)
4566 module_exit(md_exit)
4567
4568 static int get_ro(char *buffer, struct kernel_param *kp)
4569 {
4570         return sprintf(buffer, "%d", start_readonly);
4571 }
4572 static int set_ro(const char *val, struct kernel_param *kp)
4573 {
4574         char *e;
4575         int num = simple_strtoul(val, &e, 10);
4576         if (*val && (*e == '\0' || *e == '\n')) {
4577                 start_readonly = num;
4578                 return 0;;
4579         }
4580         return -EINVAL;
4581 }
4582
4583 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4584 module_param(start_dirty_degraded, int, 0644);
4585
4586
4587 EXPORT_SYMBOL(register_md_personality);
4588 EXPORT_SYMBOL(unregister_md_personality);
4589 EXPORT_SYMBOL(md_error);
4590 EXPORT_SYMBOL(md_done_sync);
4591 EXPORT_SYMBOL(md_write_start);
4592 EXPORT_SYMBOL(md_write_end);
4593 EXPORT_SYMBOL(md_register_thread);
4594 EXPORT_SYMBOL(md_unregister_thread);
4595 EXPORT_SYMBOL(md_wakeup_thread);
4596 EXPORT_SYMBOL(md_print_devices);
4597 EXPORT_SYMBOL(md_check_recovery);
4598 MODULE_LICENSE("GPL");
4599 MODULE_ALIAS("md");
4600 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);