0b3081aa4d6343f9471ef92749558112044a2d74
[linux-2.6.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/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 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
148 static atomic_t md_event_count;
149 static 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 static 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         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1830                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1831                 return -EBUSY;
1832         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
1833                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1834         else {
1835                 if (cmd_match(page, "check"))
1836                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1837                 else if (cmd_match(page, "repair"))
1838                         return -EINVAL;
1839                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1840                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1841         }
1842         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1843         md_wakeup_thread(mddev->thread);
1844         return len;
1845 }
1846
1847 static ssize_t
1848 mismatch_cnt_show(mddev_t *mddev, char *page)
1849 {
1850         return sprintf(page, "%llu\n",
1851                        (unsigned long long) mddev->resync_mismatches);
1852 }
1853
1854 static struct md_sysfs_entry
1855 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1856
1857
1858 static struct md_sysfs_entry
1859 md_mismatches = __ATTR_RO(mismatch_cnt);
1860
1861 static struct attribute *md_default_attrs[] = {
1862         &md_level.attr,
1863         &md_raid_disks.attr,
1864         NULL,
1865 };
1866
1867 static struct attribute *md_redundancy_attrs[] = {
1868         &md_scan_mode.attr,
1869         &md_mismatches.attr,
1870         NULL,
1871 };
1872 static struct attribute_group md_redundancy_group = {
1873         .name = NULL,
1874         .attrs = md_redundancy_attrs,
1875 };
1876
1877
1878 static ssize_t
1879 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1880 {
1881         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1882         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1883         ssize_t rv;
1884
1885         if (!entry->show)
1886                 return -EIO;
1887         mddev_lock(mddev);
1888         rv = entry->show(mddev, page);
1889         mddev_unlock(mddev);
1890         return rv;
1891 }
1892
1893 static ssize_t
1894 md_attr_store(struct kobject *kobj, struct attribute *attr,
1895               const char *page, size_t length)
1896 {
1897         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1898         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1899         ssize_t rv;
1900
1901         if (!entry->store)
1902                 return -EIO;
1903         mddev_lock(mddev);
1904         rv = entry->store(mddev, page, length);
1905         mddev_unlock(mddev);
1906         return rv;
1907 }
1908
1909 static void md_free(struct kobject *ko)
1910 {
1911         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1912         kfree(mddev);
1913 }
1914
1915 static struct sysfs_ops md_sysfs_ops = {
1916         .show   = md_attr_show,
1917         .store  = md_attr_store,
1918 };
1919 static struct kobj_type md_ktype = {
1920         .release        = md_free,
1921         .sysfs_ops      = &md_sysfs_ops,
1922         .default_attrs  = md_default_attrs,
1923 };
1924
1925 int mdp_major = 0;
1926
1927 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1928 {
1929         static DECLARE_MUTEX(disks_sem);
1930         mddev_t *mddev = mddev_find(dev);
1931         struct gendisk *disk;
1932         int partitioned = (MAJOR(dev) != MD_MAJOR);
1933         int shift = partitioned ? MdpMinorShift : 0;
1934         int unit = MINOR(dev) >> shift;
1935
1936         if (!mddev)
1937                 return NULL;
1938
1939         down(&disks_sem);
1940         if (mddev->gendisk) {
1941                 up(&disks_sem);
1942                 mddev_put(mddev);
1943                 return NULL;
1944         }
1945         disk = alloc_disk(1 << shift);
1946         if (!disk) {
1947                 up(&disks_sem);
1948                 mddev_put(mddev);
1949                 return NULL;
1950         }
1951         disk->major = MAJOR(dev);
1952         disk->first_minor = unit << shift;
1953         if (partitioned) {
1954                 sprintf(disk->disk_name, "md_d%d", unit);
1955                 sprintf(disk->devfs_name, "md/d%d", unit);
1956         } else {
1957                 sprintf(disk->disk_name, "md%d", unit);
1958                 sprintf(disk->devfs_name, "md/%d", unit);
1959         }
1960         disk->fops = &md_fops;
1961         disk->private_data = mddev;
1962         disk->queue = mddev->queue;
1963         add_disk(disk);
1964         mddev->gendisk = disk;
1965         up(&disks_sem);
1966         mddev->kobj.parent = &disk->kobj;
1967         mddev->kobj.k_name = NULL;
1968         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1969         mddev->kobj.ktype = &md_ktype;
1970         kobject_register(&mddev->kobj);
1971         return NULL;
1972 }
1973
1974 void md_wakeup_thread(mdk_thread_t *thread);
1975
1976 static void md_safemode_timeout(unsigned long data)
1977 {
1978         mddev_t *mddev = (mddev_t *) data;
1979
1980         mddev->safemode = 1;
1981         md_wakeup_thread(mddev->thread);
1982 }
1983
1984 static int start_dirty_degraded;
1985
1986 static int do_md_run(mddev_t * mddev)
1987 {
1988         int err;
1989         int chunk_size;
1990         struct list_head *tmp;
1991         mdk_rdev_t *rdev;
1992         struct gendisk *disk;
1993         struct mdk_personality *pers;
1994         char b[BDEVNAME_SIZE];
1995
1996         if (list_empty(&mddev->disks))
1997                 /* cannot run an array with no devices.. */
1998                 return -EINVAL;
1999
2000         if (mddev->pers)
2001                 return -EBUSY;
2002
2003         /*
2004          * Analyze all RAID superblock(s)
2005          */
2006         if (!mddev->raid_disks)
2007                 analyze_sbs(mddev);
2008
2009         chunk_size = mddev->chunk_size;
2010
2011         if (chunk_size) {
2012                 if (chunk_size > MAX_CHUNK_SIZE) {
2013                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2014                                 chunk_size, MAX_CHUNK_SIZE);
2015                         return -EINVAL;
2016                 }
2017                 /*
2018                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2019                  */
2020                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2021                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2022                         return -EINVAL;
2023                 }
2024                 if (chunk_size < PAGE_SIZE) {
2025                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2026                                 chunk_size, PAGE_SIZE);
2027                         return -EINVAL;
2028                 }
2029
2030                 /* devices must have minimum size of one chunk */
2031                 ITERATE_RDEV(mddev,rdev,tmp) {
2032                         if (test_bit(Faulty, &rdev->flags))
2033                                 continue;
2034                         if (rdev->size < chunk_size / 1024) {
2035                                 printk(KERN_WARNING
2036                                         "md: Dev %s smaller than chunk_size:"
2037                                         " %lluk < %dk\n",
2038                                         bdevname(rdev->bdev,b),
2039                                         (unsigned long long)rdev->size,
2040                                         chunk_size / 1024);
2041                                 return -EINVAL;
2042                         }
2043                 }
2044         }
2045
2046 #ifdef CONFIG_KMOD
2047         request_module("md-level-%d", mddev->level);
2048 #endif
2049
2050         /*
2051          * Drop all container device buffers, from now on
2052          * the only valid external interface is through the md
2053          * device.
2054          * Also find largest hardsector size
2055          */
2056         ITERATE_RDEV(mddev,rdev,tmp) {
2057                 if (test_bit(Faulty, &rdev->flags))
2058                         continue;
2059                 sync_blockdev(rdev->bdev);
2060                 invalidate_bdev(rdev->bdev, 0);
2061         }
2062
2063         md_probe(mddev->unit, NULL, NULL);
2064         disk = mddev->gendisk;
2065         if (!disk)
2066                 return -ENOMEM;
2067
2068         spin_lock(&pers_lock);
2069         pers = find_pers(mddev->level);
2070         if (!pers || !try_module_get(pers->owner)) {
2071                 spin_unlock(&pers_lock);
2072                 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2073                        mddev->level);
2074                 return -EINVAL;
2075         }
2076         mddev->pers = pers;
2077         spin_unlock(&pers_lock);
2078
2079         mddev->recovery = 0;
2080         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2081         mddev->barriers_work = 1;
2082         mddev->ok_start_degraded = start_dirty_degraded;
2083
2084         if (start_readonly)
2085                 mddev->ro = 2; /* read-only, but switch on first write */
2086
2087         err = mddev->pers->run(mddev);
2088         if (!err && mddev->pers->sync_request) {
2089                 err = bitmap_create(mddev);
2090                 if (err) {
2091                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2092                                mdname(mddev), err);
2093                         mddev->pers->stop(mddev);
2094                 }
2095         }
2096         if (err) {
2097                 printk(KERN_ERR "md: pers->run() failed ...\n");
2098                 module_put(mddev->pers->owner);
2099                 mddev->pers = NULL;
2100                 bitmap_destroy(mddev);
2101                 return err;
2102         }
2103         if (mddev->pers->sync_request)
2104                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2105         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2106                 mddev->ro = 0;
2107
2108         atomic_set(&mddev->writes_pending,0);
2109         mddev->safemode = 0;
2110         mddev->safemode_timer.function = md_safemode_timeout;
2111         mddev->safemode_timer.data = (unsigned long) mddev;
2112         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2113         mddev->in_sync = 1;
2114
2115         ITERATE_RDEV(mddev,rdev,tmp)
2116                 if (rdev->raid_disk >= 0) {
2117                         char nm[20];
2118                         sprintf(nm, "rd%d", rdev->raid_disk);
2119                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2120                 }
2121         
2122         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2123         md_wakeup_thread(mddev->thread);
2124         
2125         if (mddev->sb_dirty)
2126                 md_update_sb(mddev);
2127
2128         set_capacity(disk, mddev->array_size<<1);
2129
2130         /* If we call blk_queue_make_request here, it will
2131          * re-initialise max_sectors etc which may have been
2132          * refined inside -> run.  So just set the bits we need to set.
2133          * Most initialisation happended when we called
2134          * blk_queue_make_request(..., md_fail_request)
2135          * earlier.
2136          */
2137         mddev->queue->queuedata = mddev;
2138         mddev->queue->make_request_fn = mddev->pers->make_request;
2139
2140         mddev->changed = 1;
2141         md_new_event(mddev);
2142         return 0;
2143 }
2144
2145 static int restart_array(mddev_t *mddev)
2146 {
2147         struct gendisk *disk = mddev->gendisk;
2148         int err;
2149
2150         /*
2151          * Complain if it has no devices
2152          */
2153         err = -ENXIO;
2154         if (list_empty(&mddev->disks))
2155                 goto out;
2156
2157         if (mddev->pers) {
2158                 err = -EBUSY;
2159                 if (!mddev->ro)
2160                         goto out;
2161
2162                 mddev->safemode = 0;
2163                 mddev->ro = 0;
2164                 set_disk_ro(disk, 0);
2165
2166                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2167                         mdname(mddev));
2168                 /*
2169                  * Kick recovery or resync if necessary
2170                  */
2171                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2172                 md_wakeup_thread(mddev->thread);
2173                 err = 0;
2174         } else {
2175                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2176                         mdname(mddev));
2177                 err = -EINVAL;
2178         }
2179
2180 out:
2181         return err;
2182 }
2183
2184 static int do_md_stop(mddev_t * mddev, int ro)
2185 {
2186         int err = 0;
2187         struct gendisk *disk = mddev->gendisk;
2188
2189         if (mddev->pers) {
2190                 if (atomic_read(&mddev->active)>2) {
2191                         printk("md: %s still in use.\n",mdname(mddev));
2192                         return -EBUSY;
2193                 }
2194
2195                 if (mddev->sync_thread) {
2196                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2197                         md_unregister_thread(mddev->sync_thread);
2198                         mddev->sync_thread = NULL;
2199                 }
2200
2201                 del_timer_sync(&mddev->safemode_timer);
2202
2203                 invalidate_partition(disk, 0);
2204
2205                 if (ro) {
2206                         err  = -ENXIO;
2207                         if (mddev->ro==1)
2208                                 goto out;
2209                         mddev->ro = 1;
2210                 } else {
2211                         bitmap_flush(mddev);
2212                         md_super_wait(mddev);
2213                         if (mddev->ro)
2214                                 set_disk_ro(disk, 0);
2215                         blk_queue_make_request(mddev->queue, md_fail_request);
2216                         mddev->pers->stop(mddev);
2217                         if (mddev->pers->sync_request)
2218                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2219
2220                         module_put(mddev->pers->owner);
2221                         mddev->pers = NULL;
2222                         if (mddev->ro)
2223                                 mddev->ro = 0;
2224                 }
2225                 if (!mddev->in_sync) {
2226                         /* mark array as shutdown cleanly */
2227                         mddev->in_sync = 1;
2228                         md_update_sb(mddev);
2229                 }
2230                 if (ro)
2231                         set_disk_ro(disk, 1);
2232         }
2233
2234         bitmap_destroy(mddev);
2235         if (mddev->bitmap_file) {
2236                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2237                 fput(mddev->bitmap_file);
2238                 mddev->bitmap_file = NULL;
2239         }
2240         mddev->bitmap_offset = 0;
2241
2242         /*
2243          * Free resources if final stop
2244          */
2245         if (!ro) {
2246                 mdk_rdev_t *rdev;
2247                 struct list_head *tmp;
2248                 struct gendisk *disk;
2249                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2250
2251                 ITERATE_RDEV(mddev,rdev,tmp)
2252                         if (rdev->raid_disk >= 0) {
2253                                 char nm[20];
2254                                 sprintf(nm, "rd%d", rdev->raid_disk);
2255                                 sysfs_remove_link(&mddev->kobj, nm);
2256                         }
2257
2258                 export_array(mddev);
2259
2260                 mddev->array_size = 0;
2261                 disk = mddev->gendisk;
2262                 if (disk)
2263                         set_capacity(disk, 0);
2264                 mddev->changed = 1;
2265         } else
2266                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2267                         mdname(mddev));
2268         err = 0;
2269         md_new_event(mddev);
2270 out:
2271         return err;
2272 }
2273
2274 static void autorun_array(mddev_t *mddev)
2275 {
2276         mdk_rdev_t *rdev;
2277         struct list_head *tmp;
2278         int err;
2279
2280         if (list_empty(&mddev->disks))
2281                 return;
2282
2283         printk(KERN_INFO "md: running: ");
2284
2285         ITERATE_RDEV(mddev,rdev,tmp) {
2286                 char b[BDEVNAME_SIZE];
2287                 printk("<%s>", bdevname(rdev->bdev,b));
2288         }
2289         printk("\n");
2290
2291         err = do_md_run (mddev);
2292         if (err) {
2293                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2294                 do_md_stop (mddev, 0);
2295         }
2296 }
2297
2298 /*
2299  * lets try to run arrays based on all disks that have arrived
2300  * until now. (those are in pending_raid_disks)
2301  *
2302  * the method: pick the first pending disk, collect all disks with
2303  * the same UUID, remove all from the pending list and put them into
2304  * the 'same_array' list. Then order this list based on superblock
2305  * update time (freshest comes first), kick out 'old' disks and
2306  * compare superblocks. If everything's fine then run it.
2307  *
2308  * If "unit" is allocated, then bump its reference count
2309  */
2310 static void autorun_devices(int part)
2311 {
2312         struct list_head candidates;
2313         struct list_head *tmp;
2314         mdk_rdev_t *rdev0, *rdev;
2315         mddev_t *mddev;
2316         char b[BDEVNAME_SIZE];
2317
2318         printk(KERN_INFO "md: autorun ...\n");
2319         while (!list_empty(&pending_raid_disks)) {
2320                 dev_t dev;
2321                 rdev0 = list_entry(pending_raid_disks.next,
2322                                          mdk_rdev_t, same_set);
2323
2324                 printk(KERN_INFO "md: considering %s ...\n",
2325                         bdevname(rdev0->bdev,b));
2326                 INIT_LIST_HEAD(&candidates);
2327                 ITERATE_RDEV_PENDING(rdev,tmp)
2328                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2329                                 printk(KERN_INFO "md:  adding %s ...\n",
2330                                         bdevname(rdev->bdev,b));
2331                                 list_move(&rdev->same_set, &candidates);
2332                         }
2333                 /*
2334                  * now we have a set of devices, with all of them having
2335                  * mostly sane superblocks. It's time to allocate the
2336                  * mddev.
2337                  */
2338                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2339                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2340                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2341                         break;
2342                 }
2343                 if (part)
2344                         dev = MKDEV(mdp_major,
2345                                     rdev0->preferred_minor << MdpMinorShift);
2346                 else
2347                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2348
2349                 md_probe(dev, NULL, NULL);
2350                 mddev = mddev_find(dev);
2351                 if (!mddev) {
2352                         printk(KERN_ERR 
2353                                 "md: cannot allocate memory for md drive.\n");
2354                         break;
2355                 }
2356                 if (mddev_lock(mddev)) 
2357                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2358                                mdname(mddev));
2359                 else if (mddev->raid_disks || mddev->major_version
2360                          || !list_empty(&mddev->disks)) {
2361                         printk(KERN_WARNING 
2362                                 "md: %s already running, cannot run %s\n",
2363                                 mdname(mddev), bdevname(rdev0->bdev,b));
2364                         mddev_unlock(mddev);
2365                 } else {
2366                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2367                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2368                                 list_del_init(&rdev->same_set);
2369                                 if (bind_rdev_to_array(rdev, mddev))
2370                                         export_rdev(rdev);
2371                         }
2372                         autorun_array(mddev);
2373                         mddev_unlock(mddev);
2374                 }
2375                 /* on success, candidates will be empty, on error
2376                  * it won't...
2377                  */
2378                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2379                         export_rdev(rdev);
2380                 mddev_put(mddev);
2381         }
2382         printk(KERN_INFO "md: ... autorun DONE.\n");
2383 }
2384
2385 /*
2386  * import RAID devices based on one partition
2387  * if possible, the array gets run as well.
2388  */
2389
2390 static int autostart_array(dev_t startdev)
2391 {
2392         char b[BDEVNAME_SIZE];
2393         int err = -EINVAL, i;
2394         mdp_super_t *sb = NULL;
2395         mdk_rdev_t *start_rdev = NULL, *rdev;
2396
2397         start_rdev = md_import_device(startdev, 0, 0);
2398         if (IS_ERR(start_rdev))
2399                 return err;
2400
2401
2402         /* NOTE: this can only work for 0.90.0 superblocks */
2403         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2404         if (sb->major_version != 0 ||
2405             sb->minor_version != 90 ) {
2406                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2407                 export_rdev(start_rdev);
2408                 return err;
2409         }
2410
2411         if (test_bit(Faulty, &start_rdev->flags)) {
2412                 printk(KERN_WARNING 
2413                         "md: can not autostart based on faulty %s!\n",
2414                         bdevname(start_rdev->bdev,b));
2415                 export_rdev(start_rdev);
2416                 return err;
2417         }
2418         list_add(&start_rdev->same_set, &pending_raid_disks);
2419
2420         for (i = 0; i < MD_SB_DISKS; i++) {
2421                 mdp_disk_t *desc = sb->disks + i;
2422                 dev_t dev = MKDEV(desc->major, desc->minor);
2423
2424                 if (!dev)
2425                         continue;
2426                 if (dev == startdev)
2427                         continue;
2428                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2429                         continue;
2430                 rdev = md_import_device(dev, 0, 0);
2431                 if (IS_ERR(rdev))
2432                         continue;
2433
2434                 list_add(&rdev->same_set, &pending_raid_disks);
2435         }
2436
2437         /*
2438          * possibly return codes
2439          */
2440         autorun_devices(0);
2441         return 0;
2442
2443 }
2444
2445
2446 static int get_version(void __user * arg)
2447 {
2448         mdu_version_t ver;
2449
2450         ver.major = MD_MAJOR_VERSION;
2451         ver.minor = MD_MINOR_VERSION;
2452         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2453
2454         if (copy_to_user(arg, &ver, sizeof(ver)))
2455                 return -EFAULT;
2456
2457         return 0;
2458 }
2459
2460 static int get_array_info(mddev_t * mddev, void __user * arg)
2461 {
2462         mdu_array_info_t info;
2463         int nr,working,active,failed,spare;
2464         mdk_rdev_t *rdev;
2465         struct list_head *tmp;
2466
2467         nr=working=active=failed=spare=0;
2468         ITERATE_RDEV(mddev,rdev,tmp) {
2469                 nr++;
2470                 if (test_bit(Faulty, &rdev->flags))
2471                         failed++;
2472                 else {
2473                         working++;
2474                         if (test_bit(In_sync, &rdev->flags))
2475                                 active++;       
2476                         else
2477                                 spare++;
2478                 }
2479         }
2480
2481         info.major_version = mddev->major_version;
2482         info.minor_version = mddev->minor_version;
2483         info.patch_version = MD_PATCHLEVEL_VERSION;
2484         info.ctime         = mddev->ctime;
2485         info.level         = mddev->level;
2486         info.size          = mddev->size;
2487         info.nr_disks      = nr;
2488         info.raid_disks    = mddev->raid_disks;
2489         info.md_minor      = mddev->md_minor;
2490         info.not_persistent= !mddev->persistent;
2491
2492         info.utime         = mddev->utime;
2493         info.state         = 0;
2494         if (mddev->in_sync)
2495                 info.state = (1<<MD_SB_CLEAN);
2496         if (mddev->bitmap && mddev->bitmap_offset)
2497                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2498         info.active_disks  = active;
2499         info.working_disks = working;
2500         info.failed_disks  = failed;
2501         info.spare_disks   = spare;
2502
2503         info.layout        = mddev->layout;
2504         info.chunk_size    = mddev->chunk_size;
2505
2506         if (copy_to_user(arg, &info, sizeof(info)))
2507                 return -EFAULT;
2508
2509         return 0;
2510 }
2511
2512 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2513 {
2514         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2515         char *ptr, *buf = NULL;
2516         int err = -ENOMEM;
2517
2518         file = kmalloc(sizeof(*file), GFP_KERNEL);
2519         if (!file)
2520                 goto out;
2521
2522         /* bitmap disabled, zero the first byte and copy out */
2523         if (!mddev->bitmap || !mddev->bitmap->file) {
2524                 file->pathname[0] = '\0';
2525                 goto copy_out;
2526         }
2527
2528         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2529         if (!buf)
2530                 goto out;
2531
2532         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2533         if (!ptr)
2534                 goto out;
2535
2536         strcpy(file->pathname, ptr);
2537
2538 copy_out:
2539         err = 0;
2540         if (copy_to_user(arg, file, sizeof(*file)))
2541                 err = -EFAULT;
2542 out:
2543         kfree(buf);
2544         kfree(file);
2545         return err;
2546 }
2547
2548 static int get_disk_info(mddev_t * mddev, void __user * arg)
2549 {
2550         mdu_disk_info_t info;
2551         unsigned int nr;
2552         mdk_rdev_t *rdev;
2553
2554         if (copy_from_user(&info, arg, sizeof(info)))
2555                 return -EFAULT;
2556
2557         nr = info.number;
2558
2559         rdev = find_rdev_nr(mddev, nr);
2560         if (rdev) {
2561                 info.major = MAJOR(rdev->bdev->bd_dev);
2562                 info.minor = MINOR(rdev->bdev->bd_dev);
2563                 info.raid_disk = rdev->raid_disk;
2564                 info.state = 0;
2565                 if (test_bit(Faulty, &rdev->flags))
2566                         info.state |= (1<<MD_DISK_FAULTY);
2567                 else if (test_bit(In_sync, &rdev->flags)) {
2568                         info.state |= (1<<MD_DISK_ACTIVE);
2569                         info.state |= (1<<MD_DISK_SYNC);
2570                 }
2571                 if (test_bit(WriteMostly, &rdev->flags))
2572                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2573         } else {
2574                 info.major = info.minor = 0;
2575                 info.raid_disk = -1;
2576                 info.state = (1<<MD_DISK_REMOVED);
2577         }
2578
2579         if (copy_to_user(arg, &info, sizeof(info)))
2580                 return -EFAULT;
2581
2582         return 0;
2583 }
2584
2585 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2586 {
2587         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2588         mdk_rdev_t *rdev;
2589         dev_t dev = MKDEV(info->major,info->minor);
2590
2591         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2592                 return -EOVERFLOW;
2593
2594         if (!mddev->raid_disks) {
2595                 int err;
2596                 /* expecting a device which has a superblock */
2597                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2598                 if (IS_ERR(rdev)) {
2599                         printk(KERN_WARNING 
2600                                 "md: md_import_device returned %ld\n",
2601                                 PTR_ERR(rdev));
2602                         return PTR_ERR(rdev);
2603                 }
2604                 if (!list_empty(&mddev->disks)) {
2605                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2606                                                         mdk_rdev_t, same_set);
2607                         int err = super_types[mddev->major_version]
2608                                 .load_super(rdev, rdev0, mddev->minor_version);
2609                         if (err < 0) {
2610                                 printk(KERN_WARNING 
2611                                         "md: %s has different UUID to %s\n",
2612                                         bdevname(rdev->bdev,b), 
2613                                         bdevname(rdev0->bdev,b2));
2614                                 export_rdev(rdev);
2615                                 return -EINVAL;
2616                         }
2617                 }
2618                 err = bind_rdev_to_array(rdev, mddev);
2619                 if (err)
2620                         export_rdev(rdev);
2621                 return err;
2622         }
2623
2624         /*
2625          * add_new_disk can be used once the array is assembled
2626          * to add "hot spares".  They must already have a superblock
2627          * written
2628          */
2629         if (mddev->pers) {
2630                 int err;
2631                 if (!mddev->pers->hot_add_disk) {
2632                         printk(KERN_WARNING 
2633                                 "%s: personality does not support diskops!\n",
2634                                mdname(mddev));
2635                         return -EINVAL;
2636                 }
2637                 if (mddev->persistent)
2638                         rdev = md_import_device(dev, mddev->major_version,
2639                                                 mddev->minor_version);
2640                 else
2641                         rdev = md_import_device(dev, -1, -1);
2642                 if (IS_ERR(rdev)) {
2643                         printk(KERN_WARNING 
2644                                 "md: md_import_device returned %ld\n",
2645                                 PTR_ERR(rdev));
2646                         return PTR_ERR(rdev);
2647                 }
2648                 /* set save_raid_disk if appropriate */
2649                 if (!mddev->persistent) {
2650                         if (info->state & (1<<MD_DISK_SYNC)  &&
2651                             info->raid_disk < mddev->raid_disks)
2652                                 rdev->raid_disk = info->raid_disk;
2653                         else
2654                                 rdev->raid_disk = -1;
2655                 } else
2656                         super_types[mddev->major_version].
2657                                 validate_super(mddev, rdev);
2658                 rdev->saved_raid_disk = rdev->raid_disk;
2659
2660                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2661                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2662                         set_bit(WriteMostly, &rdev->flags);
2663
2664                 rdev->raid_disk = -1;
2665                 err = bind_rdev_to_array(rdev, mddev);
2666                 if (err)
2667                         export_rdev(rdev);
2668
2669                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2670                 md_wakeup_thread(mddev->thread);
2671                 return err;
2672         }
2673
2674         /* otherwise, add_new_disk is only allowed
2675          * for major_version==0 superblocks
2676          */
2677         if (mddev->major_version != 0) {
2678                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2679                        mdname(mddev));
2680                 return -EINVAL;
2681         }
2682
2683         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2684                 int err;
2685                 rdev = md_import_device (dev, -1, 0);
2686                 if (IS_ERR(rdev)) {
2687                         printk(KERN_WARNING 
2688                                 "md: error, md_import_device() returned %ld\n",
2689                                 PTR_ERR(rdev));
2690                         return PTR_ERR(rdev);
2691                 }
2692                 rdev->desc_nr = info->number;
2693                 if (info->raid_disk < mddev->raid_disks)
2694                         rdev->raid_disk = info->raid_disk;
2695                 else
2696                         rdev->raid_disk = -1;
2697
2698                 rdev->flags = 0;
2699
2700                 if (rdev->raid_disk < mddev->raid_disks)
2701                         if (info->state & (1<<MD_DISK_SYNC))
2702                                 set_bit(In_sync, &rdev->flags);
2703
2704                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2705                         set_bit(WriteMostly, &rdev->flags);
2706
2707                 err = bind_rdev_to_array(rdev, mddev);
2708                 if (err) {
2709                         export_rdev(rdev);
2710                         return err;
2711                 }
2712
2713                 if (!mddev->persistent) {
2714                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2715                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2716                 } else 
2717                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2718                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2719
2720                 if (!mddev->size || (mddev->size > rdev->size))
2721                         mddev->size = rdev->size;
2722         }
2723
2724         return 0;
2725 }
2726
2727 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2728 {
2729         char b[BDEVNAME_SIZE];
2730         mdk_rdev_t *rdev;
2731
2732         if (!mddev->pers)
2733                 return -ENODEV;
2734
2735         rdev = find_rdev(mddev, dev);
2736         if (!rdev)
2737                 return -ENXIO;
2738
2739         if (rdev->raid_disk >= 0)
2740                 goto busy;
2741
2742         kick_rdev_from_array(rdev);
2743         md_update_sb(mddev);
2744         md_new_event(mddev);
2745
2746         return 0;
2747 busy:
2748         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2749                 bdevname(rdev->bdev,b), mdname(mddev));
2750         return -EBUSY;
2751 }
2752
2753 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2754 {
2755         char b[BDEVNAME_SIZE];
2756         int err;
2757         unsigned int size;
2758         mdk_rdev_t *rdev;
2759
2760         if (!mddev->pers)
2761                 return -ENODEV;
2762
2763         if (mddev->major_version != 0) {
2764                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2765                         " version-0 superblocks.\n",
2766                         mdname(mddev));
2767                 return -EINVAL;
2768         }
2769         if (!mddev->pers->hot_add_disk) {
2770                 printk(KERN_WARNING 
2771                         "%s: personality does not support diskops!\n",
2772                         mdname(mddev));
2773                 return -EINVAL;
2774         }
2775
2776         rdev = md_import_device (dev, -1, 0);
2777         if (IS_ERR(rdev)) {
2778                 printk(KERN_WARNING 
2779                         "md: error, md_import_device() returned %ld\n",
2780                         PTR_ERR(rdev));
2781                 return -EINVAL;
2782         }
2783
2784         if (mddev->persistent)
2785                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2786         else
2787                 rdev->sb_offset =
2788                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2789
2790         size = calc_dev_size(rdev, mddev->chunk_size);
2791         rdev->size = size;
2792
2793         if (size < mddev->size) {
2794                 printk(KERN_WARNING 
2795                         "%s: disk size %llu blocks < array size %llu\n",
2796                         mdname(mddev), (unsigned long long)size,
2797                         (unsigned long long)mddev->size);
2798                 err = -ENOSPC;
2799                 goto abort_export;
2800         }
2801
2802         if (test_bit(Faulty, &rdev->flags)) {
2803                 printk(KERN_WARNING 
2804                         "md: can not hot-add faulty %s disk to %s!\n",
2805                         bdevname(rdev->bdev,b), mdname(mddev));
2806                 err = -EINVAL;
2807                 goto abort_export;
2808         }
2809         clear_bit(In_sync, &rdev->flags);
2810         rdev->desc_nr = -1;
2811         bind_rdev_to_array(rdev, mddev);
2812
2813         /*
2814          * The rest should better be atomic, we can have disk failures
2815          * noticed in interrupt contexts ...
2816          */
2817
2818         if (rdev->desc_nr == mddev->max_disks) {
2819                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2820                         mdname(mddev));
2821                 err = -EBUSY;
2822                 goto abort_unbind_export;
2823         }
2824
2825         rdev->raid_disk = -1;
2826
2827         md_update_sb(mddev);
2828
2829         /*
2830          * Kick recovery, maybe this spare has to be added to the
2831          * array immediately.
2832          */
2833         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2834         md_wakeup_thread(mddev->thread);
2835         md_new_event(mddev);
2836         return 0;
2837
2838 abort_unbind_export:
2839         unbind_rdev_from_array(rdev);
2840
2841 abort_export:
2842         export_rdev(rdev);
2843         return err;
2844 }
2845
2846 /* similar to deny_write_access, but accounts for our holding a reference
2847  * to the file ourselves */
2848 static int deny_bitmap_write_access(struct file * file)
2849 {
2850         struct inode *inode = file->f_mapping->host;
2851
2852         spin_lock(&inode->i_lock);
2853         if (atomic_read(&inode->i_writecount) > 1) {
2854                 spin_unlock(&inode->i_lock);
2855                 return -ETXTBSY;
2856         }
2857         atomic_set(&inode->i_writecount, -1);
2858         spin_unlock(&inode->i_lock);
2859
2860         return 0;
2861 }
2862
2863 static int set_bitmap_file(mddev_t *mddev, int fd)
2864 {
2865         int err;
2866
2867         if (mddev->pers) {
2868                 if (!mddev->pers->quiesce)
2869                         return -EBUSY;
2870                 if (mddev->recovery || mddev->sync_thread)
2871                         return -EBUSY;
2872                 /* we should be able to change the bitmap.. */
2873         }
2874
2875
2876         if (fd >= 0) {
2877                 if (mddev->bitmap)
2878                         return -EEXIST; /* cannot add when bitmap is present */
2879                 mddev->bitmap_file = fget(fd);
2880
2881                 if (mddev->bitmap_file == NULL) {
2882                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2883                                mdname(mddev));
2884                         return -EBADF;
2885                 }
2886
2887                 err = deny_bitmap_write_access(mddev->bitmap_file);
2888                 if (err) {
2889                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2890                                mdname(mddev));
2891                         fput(mddev->bitmap_file);
2892                         mddev->bitmap_file = NULL;
2893                         return err;
2894                 }
2895                 mddev->bitmap_offset = 0; /* file overrides offset */
2896         } else if (mddev->bitmap == NULL)
2897                 return -ENOENT; /* cannot remove what isn't there */
2898         err = 0;
2899         if (mddev->pers) {
2900                 mddev->pers->quiesce(mddev, 1);
2901                 if (fd >= 0)
2902                         err = bitmap_create(mddev);
2903                 if (fd < 0 || err)
2904                         bitmap_destroy(mddev);
2905                 mddev->pers->quiesce(mddev, 0);
2906         } else if (fd < 0) {
2907                 if (mddev->bitmap_file)
2908                         fput(mddev->bitmap_file);
2909                 mddev->bitmap_file = NULL;
2910         }
2911
2912         return err;
2913 }
2914
2915 /*
2916  * set_array_info is used two different ways
2917  * The original usage is when creating a new array.
2918  * In this usage, raid_disks is > 0 and it together with
2919  *  level, size, not_persistent,layout,chunksize determine the
2920  *  shape of the array.
2921  *  This will always create an array with a type-0.90.0 superblock.
2922  * The newer usage is when assembling an array.
2923  *  In this case raid_disks will be 0, and the major_version field is
2924  *  use to determine which style super-blocks are to be found on the devices.
2925  *  The minor and patch _version numbers are also kept incase the
2926  *  super_block handler wishes to interpret them.
2927  */
2928 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2929 {
2930
2931         if (info->raid_disks == 0) {
2932                 /* just setting version number for superblock loading */
2933                 if (info->major_version < 0 ||
2934                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2935                     super_types[info->major_version].name == NULL) {
2936                         /* maybe try to auto-load a module? */
2937                         printk(KERN_INFO 
2938                                 "md: superblock version %d not known\n",
2939                                 info->major_version);
2940                         return -EINVAL;
2941                 }
2942                 mddev->major_version = info->major_version;
2943                 mddev->minor_version = info->minor_version;
2944                 mddev->patch_version = info->patch_version;
2945                 return 0;
2946         }
2947         mddev->major_version = MD_MAJOR_VERSION;
2948         mddev->minor_version = MD_MINOR_VERSION;
2949         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2950         mddev->ctime         = get_seconds();
2951
2952         mddev->level         = info->level;
2953         mddev->size          = info->size;
2954         mddev->raid_disks    = info->raid_disks;
2955         /* don't set md_minor, it is determined by which /dev/md* was
2956          * openned
2957          */
2958         if (info->state & (1<<MD_SB_CLEAN))
2959                 mddev->recovery_cp = MaxSector;
2960         else
2961                 mddev->recovery_cp = 0;
2962         mddev->persistent    = ! info->not_persistent;
2963
2964         mddev->layout        = info->layout;
2965         mddev->chunk_size    = info->chunk_size;
2966
2967         mddev->max_disks     = MD_SB_DISKS;
2968
2969         mddev->sb_dirty      = 1;
2970
2971         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2972         mddev->bitmap_offset = 0;
2973
2974         /*
2975          * Generate a 128 bit UUID
2976          */
2977         get_random_bytes(mddev->uuid, 16);
2978
2979         return 0;
2980 }
2981
2982 /*
2983  * update_array_info is used to change the configuration of an
2984  * on-line array.
2985  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2986  * fields in the info are checked against the array.
2987  * Any differences that cannot be handled will cause an error.
2988  * Normally, only one change can be managed at a time.
2989  */
2990 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2991 {
2992         int rv = 0;
2993         int cnt = 0;
2994         int state = 0;
2995
2996         /* calculate expected state,ignoring low bits */
2997         if (mddev->bitmap && mddev->bitmap_offset)
2998                 state |= (1 << MD_SB_BITMAP_PRESENT);
2999
3000         if (mddev->major_version != info->major_version ||
3001             mddev->minor_version != info->minor_version ||
3002 /*          mddev->patch_version != info->patch_version || */
3003             mddev->ctime         != info->ctime         ||
3004             mddev->level         != info->level         ||
3005 /*          mddev->layout        != info->layout        || */
3006             !mddev->persistent   != info->not_persistent||
3007             mddev->chunk_size    != info->chunk_size    ||
3008             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3009             ((state^info->state) & 0xfffffe00)
3010                 )
3011                 return -EINVAL;
3012         /* Check there is only one change */
3013         if (mddev->size != info->size) cnt++;
3014         if (mddev->raid_disks != info->raid_disks) cnt++;
3015         if (mddev->layout != info->layout) cnt++;
3016         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3017         if (cnt == 0) return 0;
3018         if (cnt > 1) return -EINVAL;
3019
3020         if (mddev->layout != info->layout) {
3021                 /* Change layout
3022                  * we don't need to do anything at the md level, the
3023                  * personality will take care of it all.
3024                  */
3025                 if (mddev->pers->reconfig == NULL)
3026                         return -EINVAL;
3027                 else
3028                         return mddev->pers->reconfig(mddev, info->layout, -1);
3029         }
3030         if (mddev->size != info->size) {
3031                 mdk_rdev_t * rdev;
3032                 struct list_head *tmp;
3033                 if (mddev->pers->resize == NULL)
3034                         return -EINVAL;
3035                 /* The "size" is the amount of each device that is used.
3036                  * This can only make sense for arrays with redundancy.
3037                  * linear and raid0 always use whatever space is available
3038                  * We can only consider changing the size if no resync
3039                  * or reconstruction is happening, and if the new size
3040                  * is acceptable. It must fit before the sb_offset or,
3041                  * if that is <data_offset, it must fit before the
3042                  * size of each device.
3043                  * If size is zero, we find the largest size that fits.
3044                  */
3045                 if (mddev->sync_thread)
3046                         return -EBUSY;
3047                 ITERATE_RDEV(mddev,rdev,tmp) {
3048                         sector_t avail;
3049                         int fit = (info->size == 0);
3050                         if (rdev->sb_offset > rdev->data_offset)
3051                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
3052                         else
3053                                 avail = get_capacity(rdev->bdev->bd_disk)
3054                                         - rdev->data_offset;
3055                         if (fit && (info->size == 0 || info->size > avail/2))
3056                                 info->size = avail/2;
3057                         if (avail < ((sector_t)info->size << 1))
3058                                 return -ENOSPC;
3059                 }
3060                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3061                 if (!rv) {
3062                         struct block_device *bdev;
3063
3064                         bdev = bdget_disk(mddev->gendisk, 0);
3065                         if (bdev) {
3066                                 down(&bdev->bd_inode->i_sem);
3067                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3068                                 up(&bdev->bd_inode->i_sem);
3069                                 bdput(bdev);
3070                         }
3071                 }
3072         }
3073         if (mddev->raid_disks    != info->raid_disks) {
3074                 /* change the number of raid disks */
3075                 if (mddev->pers->reshape == NULL)
3076                         return -EINVAL;
3077                 if (info->raid_disks <= 0 ||
3078                     info->raid_disks >= mddev->max_disks)
3079                         return -EINVAL;
3080                 if (mddev->sync_thread)
3081                         return -EBUSY;
3082                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3083                 if (!rv) {
3084                         struct block_device *bdev;
3085
3086                         bdev = bdget_disk(mddev->gendisk, 0);
3087                         if (bdev) {
3088                                 down(&bdev->bd_inode->i_sem);
3089                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3090                                 up(&bdev->bd_inode->i_sem);
3091                                 bdput(bdev);
3092                         }
3093                 }
3094         }
3095         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3096                 if (mddev->pers->quiesce == NULL)
3097                         return -EINVAL;
3098                 if (mddev->recovery || mddev->sync_thread)
3099                         return -EBUSY;
3100                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3101                         /* add the bitmap */
3102                         if (mddev->bitmap)
3103                                 return -EEXIST;
3104                         if (mddev->default_bitmap_offset == 0)
3105                                 return -EINVAL;
3106                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3107                         mddev->pers->quiesce(mddev, 1);
3108                         rv = bitmap_create(mddev);
3109                         if (rv)
3110                                 bitmap_destroy(mddev);
3111                         mddev->pers->quiesce(mddev, 0);
3112                 } else {
3113                         /* remove the bitmap */
3114                         if (!mddev->bitmap)
3115                                 return -ENOENT;
3116                         if (mddev->bitmap->file)
3117                                 return -EINVAL;
3118                         mddev->pers->quiesce(mddev, 1);
3119                         bitmap_destroy(mddev);
3120                         mddev->pers->quiesce(mddev, 0);
3121                         mddev->bitmap_offset = 0;
3122                 }
3123         }
3124         md_update_sb(mddev);
3125         return rv;
3126 }
3127
3128 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3129 {
3130         mdk_rdev_t *rdev;
3131
3132         if (mddev->pers == NULL)
3133                 return -ENODEV;
3134
3135         rdev = find_rdev(mddev, dev);
3136         if (!rdev)
3137                 return -ENODEV;
3138
3139         md_error(mddev, rdev);
3140         return 0;
3141 }
3142
3143 static int md_ioctl(struct inode *inode, struct file *file,
3144                         unsigned int cmd, unsigned long arg)
3145 {
3146         int err = 0;
3147         void __user *argp = (void __user *)arg;
3148         struct hd_geometry __user *loc = argp;
3149         mddev_t *mddev = NULL;
3150
3151         if (!capable(CAP_SYS_ADMIN))
3152                 return -EACCES;
3153
3154         /*
3155          * Commands dealing with the RAID driver but not any
3156          * particular array:
3157          */
3158         switch (cmd)
3159         {
3160                 case RAID_VERSION:
3161                         err = get_version(argp);
3162                         goto done;
3163
3164                 case PRINT_RAID_DEBUG:
3165                         err = 0;
3166                         md_print_devices();
3167                         goto done;
3168
3169 #ifndef MODULE
3170                 case RAID_AUTORUN:
3171                         err = 0;
3172                         autostart_arrays(arg);
3173                         goto done;
3174 #endif
3175                 default:;
3176         }
3177
3178         /*
3179          * Commands creating/starting a new array:
3180          */
3181
3182         mddev = inode->i_bdev->bd_disk->private_data;
3183
3184         if (!mddev) {
3185                 BUG();
3186                 goto abort;
3187         }
3188
3189
3190         if (cmd == START_ARRAY) {
3191                 /* START_ARRAY doesn't need to lock the array as autostart_array
3192                  * does the locking, and it could even be a different array
3193                  */
3194                 static int cnt = 3;
3195                 if (cnt > 0 ) {
3196                         printk(KERN_WARNING
3197                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3198                                "This will not be supported beyond July 2006\n",
3199                                current->comm, current->pid);
3200                         cnt--;
3201                 }
3202                 err = autostart_array(new_decode_dev(arg));
3203                 if (err) {
3204                         printk(KERN_WARNING "md: autostart failed!\n");
3205                         goto abort;
3206                 }
3207                 goto done;
3208         }
3209
3210         err = mddev_lock(mddev);
3211         if (err) {
3212                 printk(KERN_INFO 
3213                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3214                         err, cmd);
3215                 goto abort;
3216         }
3217
3218         switch (cmd)
3219         {
3220                 case SET_ARRAY_INFO:
3221                         {
3222                                 mdu_array_info_t info;
3223                                 if (!arg)
3224                                         memset(&info, 0, sizeof(info));
3225                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3226                                         err = -EFAULT;
3227                                         goto abort_unlock;
3228                                 }
3229                                 if (mddev->pers) {
3230                                         err = update_array_info(mddev, &info);
3231                                         if (err) {
3232                                                 printk(KERN_WARNING "md: couldn't update"
3233                                                        " array info. %d\n", err);
3234                                                 goto abort_unlock;
3235                                         }
3236                                         goto done_unlock;
3237                                 }
3238                                 if (!list_empty(&mddev->disks)) {
3239                                         printk(KERN_WARNING
3240                                                "md: array %s already has disks!\n",
3241                                                mdname(mddev));
3242                                         err = -EBUSY;
3243                                         goto abort_unlock;
3244                                 }
3245                                 if (mddev->raid_disks) {
3246                                         printk(KERN_WARNING
3247                                                "md: array %s already initialised!\n",
3248                                                mdname(mddev));
3249                                         err = -EBUSY;
3250                                         goto abort_unlock;
3251                                 }
3252                                 err = set_array_info(mddev, &info);
3253                                 if (err) {
3254                                         printk(KERN_WARNING "md: couldn't set"
3255                                                " array info. %d\n", err);
3256                                         goto abort_unlock;
3257                                 }
3258                         }
3259                         goto done_unlock;
3260
3261                 default:;
3262         }
3263
3264         /*
3265          * Commands querying/configuring an existing array:
3266          */
3267         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3268          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3269         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3270                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3271                 err = -ENODEV;
3272                 goto abort_unlock;
3273         }
3274
3275         /*
3276          * Commands even a read-only array can execute:
3277          */
3278         switch (cmd)
3279         {
3280                 case GET_ARRAY_INFO:
3281                         err = get_array_info(mddev, argp);
3282                         goto done_unlock;
3283
3284                 case GET_BITMAP_FILE:
3285                         err = get_bitmap_file(mddev, argp);
3286                         goto done_unlock;
3287
3288                 case GET_DISK_INFO:
3289                         err = get_disk_info(mddev, argp);
3290                         goto done_unlock;
3291
3292                 case RESTART_ARRAY_RW:
3293                         err = restart_array(mddev);
3294                         goto done_unlock;
3295
3296                 case STOP_ARRAY:
3297                         err = do_md_stop (mddev, 0);
3298                         goto done_unlock;
3299
3300                 case STOP_ARRAY_RO:
3301                         err = do_md_stop (mddev, 1);
3302                         goto done_unlock;
3303
3304         /*
3305          * We have a problem here : there is no easy way to give a CHS
3306          * virtual geometry. We currently pretend that we have a 2 heads
3307          * 4 sectors (with a BIG number of cylinders...). This drives
3308          * dosfs just mad... ;-)
3309          */
3310                 case HDIO_GETGEO:
3311                         if (!loc) {
3312                                 err = -EINVAL;
3313                                 goto abort_unlock;
3314                         }
3315                         err = put_user (2, (char __user *) &loc->heads);
3316                         if (err)
3317                                 goto abort_unlock;
3318                         err = put_user (4, (char __user *) &loc->sectors);
3319                         if (err)
3320                                 goto abort_unlock;
3321                         err = put_user(get_capacity(mddev->gendisk)/8,
3322                                         (short __user *) &loc->cylinders);
3323                         if (err)
3324                                 goto abort_unlock;
3325                         err = put_user (get_start_sect(inode->i_bdev),
3326                                                 (long __user *) &loc->start);
3327                         goto done_unlock;
3328         }
3329
3330         /*
3331          * The remaining ioctls are changing the state of the
3332          * superblock, so we do not allow them on read-only arrays.
3333          * However non-MD ioctls (e.g. get-size) will still come through
3334          * here and hit the 'default' below, so only disallow
3335          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3336          */
3337         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3338             mddev->ro && mddev->pers) {
3339                 if (mddev->ro == 2) {
3340                         mddev->ro = 0;
3341                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3342                 md_wakeup_thread(mddev->thread);
3343
3344                 } else {
3345                         err = -EROFS;
3346                         goto abort_unlock;
3347                 }
3348         }
3349
3350         switch (cmd)
3351         {
3352                 case ADD_NEW_DISK:
3353                 {
3354                         mdu_disk_info_t info;
3355                         if (copy_from_user(&info, argp, sizeof(info)))
3356                                 err = -EFAULT;
3357                         else
3358                                 err = add_new_disk(mddev, &info);
3359                         goto done_unlock;
3360                 }
3361
3362                 case HOT_REMOVE_DISK:
3363                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3364                         goto done_unlock;
3365
3366                 case HOT_ADD_DISK:
3367                         err = hot_add_disk(mddev, new_decode_dev(arg));
3368                         goto done_unlock;
3369
3370                 case SET_DISK_FAULTY:
3371                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3372                         goto done_unlock;
3373
3374                 case RUN_ARRAY:
3375                         err = do_md_run (mddev);
3376                         goto done_unlock;
3377
3378                 case SET_BITMAP_FILE:
3379                         err = set_bitmap_file(mddev, (int)arg);
3380                         goto done_unlock;
3381
3382                 default:
3383                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3384                                 printk(KERN_WARNING "md: %s(pid %d) used"
3385                                         " obsolete MD ioctl, upgrade your"
3386                                         " software to use new ictls.\n",
3387                                         current->comm, current->pid);
3388                         err = -EINVAL;
3389                         goto abort_unlock;
3390         }
3391
3392 done_unlock:
3393 abort_unlock:
3394         mddev_unlock(mddev);
3395
3396         return err;
3397 done:
3398         if (err)
3399                 MD_BUG();
3400 abort:
3401         return err;
3402 }
3403
3404 static int md_open(struct inode *inode, struct file *file)
3405 {
3406         /*
3407          * Succeed if we can lock the mddev, which confirms that
3408          * it isn't being stopped right now.
3409          */
3410         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3411         int err;
3412
3413         if ((err = mddev_lock(mddev)))
3414                 goto out;
3415
3416         err = 0;
3417         mddev_get(mddev);
3418         mddev_unlock(mddev);
3419
3420         check_disk_change(inode->i_bdev);
3421  out:
3422         return err;
3423 }
3424
3425 static int md_release(struct inode *inode, struct file * file)
3426 {
3427         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3428
3429         if (!mddev)
3430                 BUG();
3431         mddev_put(mddev);
3432
3433         return 0;
3434 }
3435
3436 static int md_media_changed(struct gendisk *disk)
3437 {
3438         mddev_t *mddev = disk->private_data;
3439
3440         return mddev->changed;
3441 }
3442
3443 static int md_revalidate(struct gendisk *disk)
3444 {
3445         mddev_t *mddev = disk->private_data;
3446
3447         mddev->changed = 0;
3448         return 0;
3449 }
3450 static struct block_device_operations md_fops =
3451 {
3452         .owner          = THIS_MODULE,
3453         .open           = md_open,
3454         .release        = md_release,
3455         .ioctl          = md_ioctl,
3456         .media_changed  = md_media_changed,
3457         .revalidate_disk= md_revalidate,
3458 };
3459
3460 static int md_thread(void * arg)
3461 {
3462         mdk_thread_t *thread = arg;
3463
3464         /*
3465          * md_thread is a 'system-thread', it's priority should be very
3466          * high. We avoid resource deadlocks individually in each
3467          * raid personality. (RAID5 does preallocation) We also use RR and
3468          * the very same RT priority as kswapd, thus we will never get
3469          * into a priority inversion deadlock.
3470          *
3471          * we definitely have to have equal or higher priority than
3472          * bdflush, otherwise bdflush will deadlock if there are too
3473          * many dirty RAID5 blocks.
3474          */
3475
3476         allow_signal(SIGKILL);
3477         while (!kthread_should_stop()) {
3478
3479                 /* We need to wait INTERRUPTIBLE so that
3480                  * we don't add to the load-average.
3481                  * That means we need to be sure no signals are
3482                  * pending
3483                  */
3484                 if (signal_pending(current))
3485                         flush_signals(current);
3486
3487                 wait_event_interruptible_timeout
3488                         (thread->wqueue,
3489                          test_bit(THREAD_WAKEUP, &thread->flags)
3490                          || kthread_should_stop(),
3491                          thread->timeout);
3492                 try_to_freeze();
3493
3494                 clear_bit(THREAD_WAKEUP, &thread->flags);
3495
3496                 thread->run(thread->mddev);
3497         }
3498
3499         return 0;
3500 }
3501
3502 void md_wakeup_thread(mdk_thread_t *thread)
3503 {
3504         if (thread) {
3505                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3506                 set_bit(THREAD_WAKEUP, &thread->flags);
3507                 wake_up(&thread->wqueue);
3508         }
3509 }
3510
3511 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3512                                  const char *name)
3513 {
3514         mdk_thread_t *thread;
3515
3516         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3517         if (!thread)
3518                 return NULL;
3519
3520         init_waitqueue_head(&thread->wqueue);
3521
3522         thread->run = run;
3523         thread->mddev = mddev;
3524         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3525         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3526         if (IS_ERR(thread->tsk)) {
3527                 kfree(thread);
3528                 return NULL;
3529         }
3530         return thread;
3531 }
3532
3533 void md_unregister_thread(mdk_thread_t *thread)
3534 {
3535         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3536
3537         kthread_stop(thread->tsk);
3538         kfree(thread);
3539 }
3540
3541 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3542 {
3543         if (!mddev) {
3544                 MD_BUG();
3545                 return;
3546         }
3547
3548         if (!rdev || test_bit(Faulty, &rdev->flags))
3549                 return;
3550 /*
3551         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3552                 mdname(mddev),
3553                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3554                 __builtin_return_address(0),__builtin_return_address(1),
3555                 __builtin_return_address(2),__builtin_return_address(3));
3556 */
3557         if (!mddev->pers->error_handler)
3558                 return;
3559         mddev->pers->error_handler(mddev,rdev);
3560         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3561         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3562         md_wakeup_thread(mddev->thread);
3563         md_new_event(mddev);
3564 }
3565
3566 /* seq_file implementation /proc/mdstat */
3567
3568 static void status_unused(struct seq_file *seq)
3569 {
3570         int i = 0;
3571         mdk_rdev_t *rdev;
3572         struct list_head *tmp;
3573
3574         seq_printf(seq, "unused devices: ");
3575
3576         ITERATE_RDEV_PENDING(rdev,tmp) {
3577                 char b[BDEVNAME_SIZE];
3578                 i++;
3579                 seq_printf(seq, "%s ",
3580                               bdevname(rdev->bdev,b));
3581         }
3582         if (!i)
3583                 seq_printf(seq, "<none>");
3584
3585         seq_printf(seq, "\n");
3586 }
3587
3588
3589 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3590 {
3591         unsigned long max_blocks, resync, res, dt, db, rt;
3592
3593         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3594
3595         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3596                 max_blocks = mddev->resync_max_sectors >> 1;
3597         else
3598                 max_blocks = mddev->size;
3599
3600         /*
3601          * Should not happen.
3602          */
3603         if (!max_blocks) {
3604                 MD_BUG();
3605                 return;
3606         }
3607         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3608         {
3609                 int i, x = res/50, y = 20-x;
3610                 seq_printf(seq, "[");
3611                 for (i = 0; i < x; i++)
3612                         seq_printf(seq, "=");
3613                 seq_printf(seq, ">");
3614                 for (i = 0; i < y; i++)
3615                         seq_printf(seq, ".");
3616                 seq_printf(seq, "] ");
3617         }
3618         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3619                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3620                        "resync" : "recovery"),
3621                       res/10, res % 10, resync, max_blocks);
3622
3623         /*
3624          * We do not want to overflow, so the order of operands and
3625          * the * 100 / 100 trick are important. We do a +1 to be
3626          * safe against division by zero. We only estimate anyway.
3627          *
3628          * dt: time from mark until now
3629          * db: blocks written from mark until now
3630          * rt: remaining time
3631          */
3632         dt = ((jiffies - mddev->resync_mark) / HZ);
3633         if (!dt) dt++;
3634         db = resync - (mddev->resync_mark_cnt/2);
3635         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3636
3637         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3638
3639         seq_printf(seq, " speed=%ldK/sec", db/dt);
3640 }
3641
3642 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3643 {
3644         struct list_head *tmp;
3645         loff_t l = *pos;
3646         mddev_t *mddev;
3647
3648         if (l >= 0x10000)
3649                 return NULL;
3650         if (!l--)
3651                 /* header */
3652                 return (void*)1;
3653
3654         spin_lock(&all_mddevs_lock);
3655         list_for_each(tmp,&all_mddevs)
3656                 if (!l--) {
3657                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3658                         mddev_get(mddev);
3659                         spin_unlock(&all_mddevs_lock);
3660                         return mddev;
3661                 }
3662         spin_unlock(&all_mddevs_lock);
3663         if (!l--)
3664                 return (void*)2;/* tail */
3665         return NULL;
3666 }
3667
3668 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3669 {
3670         struct list_head *tmp;
3671         mddev_t *next_mddev, *mddev = v;
3672         
3673         ++*pos;
3674         if (v == (void*)2)
3675                 return NULL;
3676
3677         spin_lock(&all_mddevs_lock);
3678         if (v == (void*)1)
3679                 tmp = all_mddevs.next;
3680         else
3681                 tmp = mddev->all_mddevs.next;
3682         if (tmp != &all_mddevs)
3683                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3684         else {
3685                 next_mddev = (void*)2;
3686                 *pos = 0x10000;
3687         }               
3688         spin_unlock(&all_mddevs_lock);
3689
3690         if (v != (void*)1)
3691                 mddev_put(mddev);
3692         return next_mddev;
3693
3694 }
3695
3696 static void md_seq_stop(struct seq_file *seq, void *v)
3697 {
3698         mddev_t *mddev = v;
3699
3700         if (mddev && v != (void*)1 && v != (void*)2)
3701                 mddev_put(mddev);
3702 }
3703
3704 struct mdstat_info {
3705         int event;
3706 };
3707
3708 static int md_seq_show(struct seq_file *seq, void *v)
3709 {
3710         mddev_t *mddev = v;
3711         sector_t size;
3712         struct list_head *tmp2;
3713         mdk_rdev_t *rdev;
3714         struct mdstat_info *mi = seq->private;
3715         struct bitmap *bitmap;
3716
3717         if (v == (void*)1) {
3718                 struct mdk_personality *pers;
3719                 seq_printf(seq, "Personalities : ");
3720                 spin_lock(&pers_lock);
3721                 list_for_each_entry(pers, &pers_list, list)
3722                         seq_printf(seq, "[%s] ", pers->name);
3723
3724                 spin_unlock(&pers_lock);
3725                 seq_printf(seq, "\n");
3726                 mi->event = atomic_read(&md_event_count);
3727                 return 0;
3728         }
3729         if (v == (void*)2) {
3730                 status_unused(seq);
3731                 return 0;
3732         }
3733
3734         if (mddev_lock(mddev)!=0) 
3735                 return -EINTR;
3736         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3737                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3738                                                 mddev->pers ? "" : "in");
3739                 if (mddev->pers) {
3740                         if (mddev->ro==1)
3741                                 seq_printf(seq, " (read-only)");
3742                         if (mddev->ro==2)
3743                                 seq_printf(seq, "(auto-read-only)");
3744                         seq_printf(seq, " %s", mddev->pers->name);
3745                 }
3746
3747                 size = 0;
3748                 ITERATE_RDEV(mddev,rdev,tmp2) {
3749                         char b[BDEVNAME_SIZE];
3750                         seq_printf(seq, " %s[%d]",
3751                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3752                         if (test_bit(WriteMostly, &rdev->flags))
3753                                 seq_printf(seq, "(W)");
3754                         if (test_bit(Faulty, &rdev->flags)) {
3755                                 seq_printf(seq, "(F)");
3756                                 continue;
3757                         } else if (rdev->raid_disk < 0)
3758                                 seq_printf(seq, "(S)"); /* spare */
3759                         size += rdev->size;
3760                 }
3761
3762                 if (!list_empty(&mddev->disks)) {
3763                         if (mddev->pers)
3764                                 seq_printf(seq, "\n      %llu blocks",
3765                                         (unsigned long long)mddev->array_size);
3766                         else
3767                                 seq_printf(seq, "\n      %llu blocks",
3768                                         (unsigned long long)size);
3769                 }
3770                 if (mddev->persistent) {
3771                         if (mddev->major_version != 0 ||
3772                             mddev->minor_version != 90) {
3773                                 seq_printf(seq," super %d.%d",
3774                                            mddev->major_version,
3775                                            mddev->minor_version);
3776                         }
3777                 } else
3778                         seq_printf(seq, " super non-persistent");
3779
3780                 if (mddev->pers) {
3781                         mddev->pers->status (seq, mddev);
3782                         seq_printf(seq, "\n      ");
3783                         if (mddev->pers->sync_request) {
3784                                 if (mddev->curr_resync > 2) {
3785                                         status_resync (seq, mddev);
3786                                         seq_printf(seq, "\n      ");
3787                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3788                                         seq_printf(seq, "\tresync=DELAYED\n      ");
3789                                 else if (mddev->recovery_cp < MaxSector)
3790                                         seq_printf(seq, "\tresync=PENDING\n      ");
3791                         }
3792                 } else
3793                         seq_printf(seq, "\n       ");
3794
3795                 if ((bitmap = mddev->bitmap)) {
3796                         unsigned long chunk_kb;
3797                         unsigned long flags;
3798                         spin_lock_irqsave(&bitmap->lock, flags);
3799                         chunk_kb = bitmap->chunksize >> 10;
3800                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3801                                 "%lu%s chunk",
3802                                 bitmap->pages - bitmap->missing_pages,
3803                                 bitmap->pages,
3804                                 (bitmap->pages - bitmap->missing_pages)
3805                                         << (PAGE_SHIFT - 10),
3806                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3807                                 chunk_kb ? "KB" : "B");
3808                         if (bitmap->file) {
3809                                 seq_printf(seq, ", file: ");
3810                                 seq_path(seq, bitmap->file->f_vfsmnt,
3811                                          bitmap->file->f_dentry," \t\n");
3812                         }
3813
3814                         seq_printf(seq, "\n");
3815                         spin_unlock_irqrestore(&bitmap->lock, flags);
3816                 }
3817
3818                 seq_printf(seq, "\n");
3819         }
3820         mddev_unlock(mddev);
3821         
3822         return 0;
3823 }
3824
3825 static struct seq_operations md_seq_ops = {
3826         .start  = md_seq_start,
3827         .next   = md_seq_next,
3828         .stop   = md_seq_stop,
3829         .show   = md_seq_show,
3830 };
3831
3832 static int md_seq_open(struct inode *inode, struct file *file)
3833 {
3834         int error;
3835         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3836         if (mi == NULL)
3837                 return -ENOMEM;
3838
3839         error = seq_open(file, &md_seq_ops);
3840         if (error)
3841                 kfree(mi);
3842         else {
3843                 struct seq_file *p = file->private_data;
3844                 p->private = mi;
3845                 mi->event = atomic_read(&md_event_count);
3846         }
3847         return error;
3848 }
3849
3850 static int md_seq_release(struct inode *inode, struct file *file)
3851 {
3852         struct seq_file *m = file->private_data;
3853         struct mdstat_info *mi = m->private;
3854         m->private = NULL;
3855         kfree(mi);
3856         return seq_release(inode, file);
3857 }
3858
3859 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
3860 {
3861         struct seq_file *m = filp->private_data;
3862         struct mdstat_info *mi = m->private;
3863         int mask;
3864
3865         poll_wait(filp, &md_event_waiters, wait);
3866
3867         /* always allow read */
3868         mask = POLLIN | POLLRDNORM;
3869
3870         if (mi->event != atomic_read(&md_event_count))
3871                 mask |= POLLERR | POLLPRI;
3872         return mask;
3873 }
3874
3875 static struct file_operations md_seq_fops = {
3876         .open           = md_seq_open,
3877         .read           = seq_read,
3878         .llseek         = seq_lseek,
3879         .release        = md_seq_release,
3880         .poll           = mdstat_poll,
3881 };
3882
3883 int register_md_personality(struct mdk_personality *p)
3884 {
3885         spin_lock(&pers_lock);
3886         list_add_tail(&p->list, &pers_list);
3887         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
3888         spin_unlock(&pers_lock);
3889         return 0;
3890 }
3891
3892 int unregister_md_personality(struct mdk_personality *p)
3893 {
3894         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
3895         spin_lock(&pers_lock);
3896         list_del_init(&p->list);
3897         spin_unlock(&pers_lock);
3898         return 0;
3899 }
3900
3901 static int is_mddev_idle(mddev_t *mddev)
3902 {
3903         mdk_rdev_t * rdev;
3904         struct list_head *tmp;
3905         int idle;
3906         unsigned long curr_events;
3907
3908         idle = 1;
3909         ITERATE_RDEV(mddev,rdev,tmp) {
3910                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3911                 curr_events = disk_stat_read(disk, sectors[0]) + 
3912                                 disk_stat_read(disk, sectors[1]) - 
3913                                 atomic_read(&disk->sync_io);
3914                 /* The difference between curr_events and last_events
3915                  * will be affected by any new non-sync IO (making
3916                  * curr_events bigger) and any difference in the amount of
3917                  * in-flight syncio (making current_events bigger or smaller)
3918                  * The amount in-flight is currently limited to
3919                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3920                  * which is at most 4096 sectors.
3921                  * These numbers are fairly fragile and should be made
3922                  * more robust, probably by enforcing the
3923                  * 'window size' that md_do_sync sort-of uses.
3924                  *
3925                  * Note: the following is an unsigned comparison.
3926                  */
3927                 if ((curr_events - rdev->last_events + 4096) > 8192) {
3928                         rdev->last_events = curr_events;
3929                         idle = 0;
3930                 }
3931         }
3932         return idle;
3933 }
3934
3935 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3936 {
3937         /* another "blocks" (512byte) blocks have been synced */
3938         atomic_sub(blocks, &mddev->recovery_active);
3939         wake_up(&mddev->recovery_wait);
3940         if (!ok) {
3941                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3942                 md_wakeup_thread(mddev->thread);
3943                 // stop recovery, signal do_sync ....
3944         }
3945 }
3946
3947
3948 /* md_write_start(mddev, bi)
3949  * If we need to update some array metadata (e.g. 'active' flag
3950  * in superblock) before writing, schedule a superblock update
3951  * and wait for it to complete.
3952  */
3953 void md_write_start(mddev_t *mddev, struct bio *bi)
3954 {
3955         if (bio_data_dir(bi) != WRITE)
3956                 return;
3957
3958         BUG_ON(mddev->ro == 1);
3959         if (mddev->ro == 2) {
3960                 /* need to switch to read/write */
3961                 mddev->ro = 0;
3962                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3963                 md_wakeup_thread(mddev->thread);
3964         }
3965         atomic_inc(&mddev->writes_pending);
3966         if (mddev->in_sync) {
3967                 spin_lock_irq(&mddev->write_lock);
3968                 if (mddev->in_sync) {
3969                         mddev->in_sync = 0;
3970                         mddev->sb_dirty = 1;
3971                         md_wakeup_thread(mddev->thread);
3972                 }
3973                 spin_unlock_irq(&mddev->write_lock);
3974         }
3975         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3976 }
3977
3978 void md_write_end(mddev_t *mddev)
3979 {
3980         if (atomic_dec_and_test(&mddev->writes_pending)) {
3981                 if (mddev->safemode == 2)
3982                         md_wakeup_thread(mddev->thread);
3983                 else
3984                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3985         }
3986 }
3987
3988 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3989
3990 #define SYNC_MARKS      10
3991 #define SYNC_MARK_STEP  (3*HZ)
3992 static void md_do_sync(mddev_t *mddev)
3993 {
3994         mddev_t *mddev2;
3995         unsigned int currspeed = 0,
3996                  window;
3997         sector_t max_sectors,j, io_sectors;
3998         unsigned long mark[SYNC_MARKS];
3999         sector_t mark_cnt[SYNC_MARKS];
4000         int last_mark,m;
4001         struct list_head *tmp;
4002         sector_t last_check;
4003         int skipped = 0;
4004
4005         /* just incase thread restarts... */
4006         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4007                 return;
4008
4009         /* we overload curr_resync somewhat here.
4010          * 0 == not engaged in resync at all
4011          * 2 == checking that there is no conflict with another sync
4012          * 1 == like 2, but have yielded to allow conflicting resync to
4013          *              commense
4014          * other == active in resync - this many blocks
4015          *
4016          * Before starting a resync we must have set curr_resync to
4017          * 2, and then checked that every "conflicting" array has curr_resync
4018          * less than ours.  When we find one that is the same or higher
4019          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4020          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4021          * This will mean we have to start checking from the beginning again.
4022          *
4023          */
4024
4025         do {
4026                 mddev->curr_resync = 2;
4027
4028         try_again:
4029                 if (kthread_should_stop()) {
4030                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4031                         goto skip;
4032                 }
4033                 ITERATE_MDDEV(mddev2,tmp) {
4034                         if (mddev2 == mddev)
4035                                 continue;
4036                         if (mddev2->curr_resync && 
4037                             match_mddev_units(mddev,mddev2)) {
4038                                 DEFINE_WAIT(wq);
4039                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4040                                         /* arbitrarily yield */
4041                                         mddev->curr_resync = 1;
4042                                         wake_up(&resync_wait);
4043                                 }
4044                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4045                                         /* no need to wait here, we can wait the next
4046                                          * time 'round when curr_resync == 2
4047                                          */
4048                                         continue;
4049                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4050                                 if (!kthread_should_stop() &&
4051                                     mddev2->curr_resync >= mddev->curr_resync) {
4052                                         printk(KERN_INFO "md: delaying resync of %s"
4053                                                " until %s has finished resync (they"
4054                                                " share one or more physical units)\n",
4055                                                mdname(mddev), mdname(mddev2));
4056                                         mddev_put(mddev2);
4057                                         schedule();
4058                                         finish_wait(&resync_wait, &wq);
4059                                         goto try_again;
4060                                 }
4061                                 finish_wait(&resync_wait, &wq);
4062                         }
4063                 }
4064         } while (mddev->curr_resync < 2);
4065
4066         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4067                 /* resync follows the size requested by the personality,
4068                  * which defaults to physical size, but can be virtual size
4069                  */
4070                 max_sectors = mddev->resync_max_sectors;
4071                 mddev->resync_mismatches = 0;
4072         } else
4073                 /* recovery follows the physical size of devices */
4074                 max_sectors = mddev->size << 1;
4075
4076         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4077         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4078                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4079         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4080                "(but not more than %d KB/sec) for reconstruction.\n",
4081                sysctl_speed_limit_max);
4082
4083         is_mddev_idle(mddev); /* this also initializes IO event counters */
4084         /* we don't use the checkpoint if there's a bitmap */
4085         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4086             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4087                 j = mddev->recovery_cp;
4088         else
4089                 j = 0;
4090         io_sectors = 0;
4091         for (m = 0; m < SYNC_MARKS; m++) {
4092                 mark[m] = jiffies;
4093                 mark_cnt[m] = io_sectors;
4094         }
4095         last_mark = 0;
4096         mddev->resync_mark = mark[last_mark];
4097         mddev->resync_mark_cnt = mark_cnt[last_mark];
4098
4099         /*
4100          * Tune reconstruction:
4101          */
4102         window = 32*(PAGE_SIZE/512);
4103         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4104                 window/2,(unsigned long long) max_sectors/2);
4105
4106         atomic_set(&mddev->recovery_active, 0);
4107         init_waitqueue_head(&mddev->recovery_wait);
4108         last_check = 0;
4109
4110         if (j>2) {
4111                 printk(KERN_INFO 
4112                         "md: resuming recovery of %s from checkpoint.\n",
4113                         mdname(mddev));
4114                 mddev->curr_resync = j;
4115         }
4116
4117         while (j < max_sectors) {
4118                 sector_t sectors;
4119
4120                 skipped = 0;
4121                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4122                                             currspeed < sysctl_speed_limit_min);
4123                 if (sectors == 0) {
4124                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4125                         goto out;
4126                 }
4127
4128                 if (!skipped) { /* actual IO requested */
4129                         io_sectors += sectors;
4130                         atomic_add(sectors, &mddev->recovery_active);
4131                 }
4132
4133                 j += sectors;
4134                 if (j>1) mddev->curr_resync = j;
4135                 if (last_check == 0)
4136                         /* this is the earliers that rebuilt will be
4137                          * visible in /proc/mdstat
4138                          */
4139                         md_new_event(mddev);
4140
4141                 if (last_check + window > io_sectors || j == max_sectors)
4142                         continue;
4143
4144                 last_check = io_sectors;
4145
4146                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4147                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4148                         break;
4149
4150         repeat:
4151                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4152                         /* step marks */
4153                         int next = (last_mark+1) % SYNC_MARKS;
4154
4155                         mddev->resync_mark = mark[next];
4156                         mddev->resync_mark_cnt = mark_cnt[next];
4157                         mark[next] = jiffies;
4158                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4159                         last_mark = next;
4160                 }
4161
4162
4163                 if (kthread_should_stop()) {
4164                         /*
4165                          * got a signal, exit.
4166                          */
4167                         printk(KERN_INFO 
4168                                 "md: md_do_sync() got signal ... exiting\n");
4169                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4170                         goto out;
4171                 }
4172
4173                 /*
4174                  * this loop exits only if either when we are slower than
4175                  * the 'hard' speed limit, or the system was IO-idle for
4176                  * a jiffy.
4177                  * the system might be non-idle CPU-wise, but we only care
4178                  * about not overloading the IO subsystem. (things like an
4179                  * e2fsck being done on the RAID array should execute fast)
4180                  */
4181                 mddev->queue->unplug_fn(mddev->queue);
4182                 cond_resched();
4183
4184                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4185                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4186
4187                 if (currspeed > sysctl_speed_limit_min) {
4188                         if ((currspeed > sysctl_speed_limit_max) ||
4189                                         !is_mddev_idle(mddev)) {
4190                                 msleep(500);
4191                                 goto repeat;
4192                         }
4193                 }
4194         }
4195         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4196         /*
4197          * this also signals 'finished resyncing' to md_stop
4198          */
4199  out:
4200         mddev->queue->unplug_fn(mddev->queue);
4201
4202         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4203
4204         /* tell personality that we are finished */
4205         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4206
4207         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4208             mddev->curr_resync > 2 &&
4209             mddev->curr_resync >= mddev->recovery_cp) {
4210                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4211                         printk(KERN_INFO 
4212                                 "md: checkpointing recovery of %s.\n",
4213                                 mdname(mddev));
4214                         mddev->recovery_cp = mddev->curr_resync;
4215                 } else
4216                         mddev->recovery_cp = MaxSector;
4217         }
4218
4219  skip:
4220         mddev->curr_resync = 0;
4221         wake_up(&resync_wait);
4222         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4223         md_wakeup_thread(mddev->thread);
4224 }
4225
4226
4227 /*
4228  * This routine is regularly called by all per-raid-array threads to
4229  * deal with generic issues like resync and super-block update.
4230  * Raid personalities that don't have a thread (linear/raid0) do not
4231  * need this as they never do any recovery or update the superblock.
4232  *
4233  * It does not do any resync itself, but rather "forks" off other threads
4234  * to do that as needed.
4235  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4236  * "->recovery" and create a thread at ->sync_thread.
4237  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4238  * and wakeups up this thread which will reap the thread and finish up.
4239  * This thread also removes any faulty devices (with nr_pending == 0).
4240  *
4241  * The overall approach is:
4242  *  1/ if the superblock needs updating, update it.
4243  *  2/ If a recovery thread is running, don't do anything else.
4244  *  3/ If recovery has finished, clean up, possibly marking spares active.
4245  *  4/ If there are any faulty devices, remove them.
4246  *  5/ If array is degraded, try to add spares devices
4247  *  6/ If array has spares or is not in-sync, start a resync thread.
4248  */
4249 void md_check_recovery(mddev_t *mddev)
4250 {
4251         mdk_rdev_t *rdev;
4252         struct list_head *rtmp;
4253
4254
4255         if (mddev->bitmap)
4256                 bitmap_daemon_work(mddev->bitmap);
4257
4258         if (mddev->ro)
4259                 return;
4260
4261         if (signal_pending(current)) {
4262                 if (mddev->pers->sync_request) {
4263                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4264                                mdname(mddev));
4265                         mddev->safemode = 2;
4266                 }
4267                 flush_signals(current);
4268         }
4269
4270         if ( ! (
4271                 mddev->sb_dirty ||
4272                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4273                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4274                 (mddev->safemode == 1) ||
4275                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4276                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4277                 ))
4278                 return;
4279
4280         if (mddev_trylock(mddev)==0) {
4281                 int spares =0;
4282
4283                 spin_lock_irq(&mddev->write_lock);
4284                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4285                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4286                         mddev->in_sync = 1;
4287                         mddev->sb_dirty = 1;
4288                 }
4289                 if (mddev->safemode == 1)
4290                         mddev->safemode = 0;
4291                 spin_unlock_irq(&mddev->write_lock);
4292
4293                 if (mddev->sb_dirty)
4294                         md_update_sb(mddev);
4295
4296
4297                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4298                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4299                         /* resync/recovery still happening */
4300                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4301                         goto unlock;
4302                 }
4303                 if (mddev->sync_thread) {
4304                         /* resync has finished, collect result */
4305                         md_unregister_thread(mddev->sync_thread);
4306                         mddev->sync_thread = NULL;
4307                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4308                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4309                                 /* success...*/
4310                                 /* activate any spares */
4311                                 mddev->pers->spare_active(mddev);
4312                         }
4313                         md_update_sb(mddev);
4314
4315                         /* if array is no-longer degraded, then any saved_raid_disk
4316                          * information must be scrapped
4317                          */
4318                         if (!mddev->degraded)
4319                                 ITERATE_RDEV(mddev,rdev,rtmp)
4320                                         rdev->saved_raid_disk = -1;
4321
4322                         mddev->recovery = 0;
4323                         /* flag recovery needed just to double check */
4324                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4325                         md_new_event(mddev);
4326                         goto unlock;
4327                 }
4328                 /* Clear some bits that don't mean anything, but
4329                  * might be left set
4330                  */
4331                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4332                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4333                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4334                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4335
4336                 /* no recovery is running.
4337                  * remove any failed drives, then
4338                  * add spares if possible.
4339                  * Spare are also removed and re-added, to allow
4340                  * the personality to fail the re-add.
4341                  */
4342                 ITERATE_RDEV(mddev,rdev,rtmp)
4343                         if (rdev->raid_disk >= 0 &&
4344                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4345                             atomic_read(&rdev->nr_pending)==0) {
4346                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4347                                         char nm[20];
4348                                         sprintf(nm,"rd%d", rdev->raid_disk);
4349                                         sysfs_remove_link(&mddev->kobj, nm);
4350                                         rdev->raid_disk = -1;
4351                                 }
4352                         }
4353
4354                 if (mddev->degraded) {
4355                         ITERATE_RDEV(mddev,rdev,rtmp)
4356                                 if (rdev->raid_disk < 0
4357                                     && !test_bit(Faulty, &rdev->flags)) {
4358                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4359                                                 char nm[20];
4360                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4361                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4362                                                 spares++;
4363                                                 md_new_event(mddev);
4364                                         } else
4365                                                 break;
4366                                 }
4367                 }
4368
4369                 if (spares) {
4370                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4371                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4372                 } else if (mddev->recovery_cp < MaxSector) {
4373                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4374                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4375                         /* nothing to be done ... */
4376                         goto unlock;
4377
4378                 if (mddev->pers->sync_request) {
4379                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4380                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4381                                 /* We are adding a device or devices to an array
4382                                  * which has the bitmap stored on all devices.
4383                                  * So make sure all bitmap pages get written
4384                                  */
4385                                 bitmap_write_all(mddev->bitmap);
4386                         }
4387                         mddev->sync_thread = md_register_thread(md_do_sync,
4388                                                                 mddev,
4389                                                                 "%s_resync");
4390                         if (!mddev->sync_thread) {
4391                                 printk(KERN_ERR "%s: could not start resync"
4392                                         " thread...\n", 
4393                                         mdname(mddev));
4394                                 /* leave the spares where they are, it shouldn't hurt */
4395                                 mddev->recovery = 0;
4396                         } else
4397                                 md_wakeup_thread(mddev->sync_thread);
4398                         md_new_event(mddev);
4399                 }
4400         unlock:
4401                 mddev_unlock(mddev);
4402         }
4403 }
4404
4405 static int md_notify_reboot(struct notifier_block *this,
4406                             unsigned long code, void *x)
4407 {
4408         struct list_head *tmp;
4409         mddev_t *mddev;
4410
4411         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4412
4413                 printk(KERN_INFO "md: stopping all md devices.\n");
4414
4415                 ITERATE_MDDEV(mddev,tmp)
4416                         if (mddev_trylock(mddev)==0)
4417                                 do_md_stop (mddev, 1);
4418                 /*
4419                  * certain more exotic SCSI devices are known to be
4420                  * volatile wrt too early system reboots. While the
4421                  * right place to handle this issue is the given
4422                  * driver, we do want to have a safe RAID driver ...
4423                  */
4424                 mdelay(1000*1);
4425         }
4426         return NOTIFY_DONE;
4427 }
4428
4429 static struct notifier_block md_notifier = {
4430         .notifier_call  = md_notify_reboot,
4431         .next           = NULL,
4432         .priority       = INT_MAX, /* before any real devices */
4433 };
4434
4435 static void md_geninit(void)
4436 {
4437         struct proc_dir_entry *p;
4438
4439         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4440
4441         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4442         if (p)
4443                 p->proc_fops = &md_seq_fops;
4444 }
4445
4446 static int __init md_init(void)
4447 {
4448         int minor;
4449
4450         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4451                         " MD_SB_DISKS=%d\n",
4452                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4453                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4454         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4455                         BITMAP_MINOR);
4456
4457         if (register_blkdev(MAJOR_NR, "md"))
4458                 return -1;
4459         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4460                 unregister_blkdev(MAJOR_NR, "md");
4461                 return -1;
4462         }
4463         devfs_mk_dir("md");
4464         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4465                                 md_probe, NULL, NULL);
4466         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4467                             md_probe, NULL, NULL);
4468
4469         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4470                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4471                                 S_IFBLK|S_IRUSR|S_IWUSR,
4472                                 "md/%d", minor);
4473
4474         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4475                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4476                               S_IFBLK|S_IRUSR|S_IWUSR,
4477                               "md/mdp%d", minor);
4478
4479
4480         register_reboot_notifier(&md_notifier);
4481         raid_table_header = register_sysctl_table(raid_root_table, 1);
4482
4483         md_geninit();
4484         return (0);
4485 }
4486
4487
4488 #ifndef MODULE
4489
4490 /*
4491  * Searches all registered partitions for autorun RAID arrays
4492  * at boot time.
4493  */
4494 static dev_t detected_devices[128];
4495 static int dev_cnt;
4496
4497 void md_autodetect_dev(dev_t dev)
4498 {
4499         if (dev_cnt >= 0 && dev_cnt < 127)
4500                 detected_devices[dev_cnt++] = dev;
4501 }
4502
4503
4504 static void autostart_arrays(int part)
4505 {
4506         mdk_rdev_t *rdev;
4507         int i;
4508
4509         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4510
4511         for (i = 0; i < dev_cnt; i++) {
4512                 dev_t dev = detected_devices[i];
4513
4514                 rdev = md_import_device(dev,0, 0);
4515                 if (IS_ERR(rdev))
4516                         continue;
4517
4518                 if (test_bit(Faulty, &rdev->flags)) {
4519                         MD_BUG();
4520                         continue;
4521                 }
4522                 list_add(&rdev->same_set, &pending_raid_disks);
4523         }
4524         dev_cnt = 0;
4525
4526         autorun_devices(part);
4527 }
4528
4529 #endif
4530
4531 static __exit void md_exit(void)
4532 {
4533         mddev_t *mddev;
4534         struct list_head *tmp;
4535         int i;
4536         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4537         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4538         for (i=0; i < MAX_MD_DEVS; i++)
4539                 devfs_remove("md/%d", i);
4540         for (i=0; i < MAX_MD_DEVS; i++)
4541                 devfs_remove("md/d%d", i);
4542
4543         devfs_remove("md");
4544
4545         unregister_blkdev(MAJOR_NR,"md");
4546         unregister_blkdev(mdp_major, "mdp");
4547         unregister_reboot_notifier(&md_notifier);
4548         unregister_sysctl_table(raid_table_header);
4549         remove_proc_entry("mdstat", NULL);
4550         ITERATE_MDDEV(mddev,tmp) {
4551                 struct gendisk *disk = mddev->gendisk;
4552                 if (!disk)
4553                         continue;
4554                 export_array(mddev);
4555                 del_gendisk(disk);
4556                 put_disk(disk);
4557                 mddev->gendisk = NULL;
4558                 mddev_put(mddev);
4559         }
4560 }
4561
4562 module_init(md_init)
4563 module_exit(md_exit)
4564
4565 static int get_ro(char *buffer, struct kernel_param *kp)
4566 {
4567         return sprintf(buffer, "%d", start_readonly);
4568 }
4569 static int set_ro(const char *val, struct kernel_param *kp)
4570 {
4571         char *e;
4572         int num = simple_strtoul(val, &e, 10);
4573         if (*val && (*e == '\0' || *e == '\n')) {
4574                 start_readonly = num;
4575                 return 0;;
4576         }
4577         return -EINVAL;
4578 }
4579
4580 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4581 module_param(start_dirty_degraded, int, 0644);
4582
4583
4584 EXPORT_SYMBOL(register_md_personality);
4585 EXPORT_SYMBOL(unregister_md_personality);
4586 EXPORT_SYMBOL(md_error);
4587 EXPORT_SYMBOL(md_done_sync);
4588 EXPORT_SYMBOL(md_write_start);
4589 EXPORT_SYMBOL(md_write_end);
4590 EXPORT_SYMBOL(md_register_thread);
4591 EXPORT_SYMBOL(md_unregister_thread);
4592 EXPORT_SYMBOL(md_wakeup_thread);
4593 EXPORT_SYMBOL(md_print_devices);
4594 EXPORT_SYMBOL(md_check_recovery);
4595 MODULE_LICENSE("GPL");
4596 MODULE_ALIAS("md");
4597 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);