c5030863d004b95050fc238e7a6128b4a302ba99
[linux-2.6.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
47
48 #include <linux/init.h>
49
50 #include <linux/file.h>
51
52 #ifdef CONFIG_KMOD
53 #include <linux/kmod.h>
54 #endif
55
56 #include <asm/unaligned.h>
57
58 #define MAJOR_NR MD_MAJOR
59 #define MD_DRIVER
60
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
63
64 #define DEBUG 0
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68 #ifndef MODULE
69 static void autostart_arrays (int part);
70 #endif
71
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
74
75 static void md_print_devices(void);
76
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
79 /*
80  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81  * is 1000 KB/sec, so the extra system load does not show up that much.
82  * Increase it if you want to have more _guaranteed_ speed. Note that
83  * the RAID driver will use the maximum available bandwidth if the IO
84  * subsystem is idle. There is also an 'absolute maximum' reconstruction
85  * speed limit - in case reconstruction slows down your system despite
86  * idle IO detection.
87  *
88  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89  * or /sys/block/mdX/md/sync_speed_{min,max}
90  */
91
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
95 {
96         return mddev->sync_speed_min ?
97                 mddev->sync_speed_min : sysctl_speed_limit_min;
98 }
99
100 static inline int speed_max(mddev_t *mddev)
101 {
102         return mddev->sync_speed_max ?
103                 mddev->sync_speed_max : sysctl_speed_limit_max;
104 }
105
106 static struct ctl_table_header *raid_table_header;
107
108 static ctl_table raid_table[] = {
109         {
110                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
111                 .procname       = "speed_limit_min",
112                 .data           = &sysctl_speed_limit_min,
113                 .maxlen         = sizeof(int),
114                 .mode           = S_IRUGO|S_IWUSR,
115                 .proc_handler   = &proc_dointvec,
116         },
117         {
118                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
119                 .procname       = "speed_limit_max",
120                 .data           = &sysctl_speed_limit_max,
121                 .maxlen         = sizeof(int),
122                 .mode           = S_IRUGO|S_IWUSR,
123                 .proc_handler   = &proc_dointvec,
124         },
125         { .ctl_name = 0 }
126 };
127
128 static ctl_table raid_dir_table[] = {
129         {
130                 .ctl_name       = DEV_RAID,
131                 .procname       = "raid",
132                 .maxlen         = 0,
133                 .mode           = S_IRUGO|S_IXUGO,
134                 .child          = raid_table,
135         },
136         { .ctl_name = 0 }
137 };
138
139 static ctl_table raid_root_table[] = {
140         {
141                 .ctl_name       = CTL_DEV,
142                 .procname       = "dev",
143                 .maxlen         = 0,
144                 .mode           = 0555,
145                 .child          = raid_dir_table,
146         },
147         { .ctl_name = 0 }
148 };
149
150 static struct block_device_operations md_fops;
151
152 static int start_readonly;
153
154 /*
155  * We have a system wide 'event count' that is incremented
156  * on any 'interesting' event, and readers of /proc/mdstat
157  * can use 'poll' or 'select' to find out when the event
158  * count increases.
159  *
160  * Events are:
161  *  start array, stop array, error, add device, remove device,
162  *  start build, activate spare
163  */
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
167 {
168         atomic_inc(&md_event_count);
169         wake_up(&md_event_waiters);
170         sysfs_notify(&mddev->kobj, NULL, "sync_action");
171 }
172 EXPORT_SYMBOL_GPL(md_new_event);
173
174 /* Alternate version that can be called from interrupts
175  * when calling sysfs_notify isn't needed.
176  */
177 static void md_new_event_inintr(mddev_t *mddev)
178 {
179         atomic_inc(&md_event_count);
180         wake_up(&md_event_waiters);
181 }
182
183 /*
184  * Enables to iterate over all existing md arrays
185  * all_mddevs_lock protects this list.
186  */
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191 /*
192  * iterates through all used mddevs in the system.
193  * We take care to grab the all_mddevs_lock whenever navigating
194  * the list, and to always hold a refcount when unlocked.
195  * Any code which breaks out of this loop while own
196  * a reference to the current mddev and must mddev_put it.
197  */
198 #define ITERATE_MDDEV(mddev,tmp)                                        \
199                                                                         \
200         for (({ spin_lock(&all_mddevs_lock);                            \
201                 tmp = all_mddevs.next;                                  \
202                 mddev = NULL;});                                        \
203              ({ if (tmp != &all_mddevs)                                 \
204                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205                 spin_unlock(&all_mddevs_lock);                          \
206                 if (mddev) mddev_put(mddev);                            \
207                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
208                 tmp != &all_mddevs;});                                  \
209              ({ spin_lock(&all_mddevs_lock);                            \
210                 tmp = tmp->next;})                                      \
211                 )
212
213
214 static int md_fail_request (struct request_queue *q, struct bio *bio)
215 {
216         bio_io_error(bio);
217         return 0;
218 }
219
220 static inline mddev_t *mddev_get(mddev_t *mddev)
221 {
222         atomic_inc(&mddev->active);
223         return mddev;
224 }
225
226 static void mddev_put(mddev_t *mddev)
227 {
228         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229                 return;
230         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231                 list_del(&mddev->all_mddevs);
232                 spin_unlock(&all_mddevs_lock);
233                 blk_cleanup_queue(mddev->queue);
234                 kobject_unregister(&mddev->kobj);
235         } else
236                 spin_unlock(&all_mddevs_lock);
237 }
238
239 static mddev_t * mddev_find(dev_t unit)
240 {
241         mddev_t *mddev, *new = NULL;
242
243  retry:
244         spin_lock(&all_mddevs_lock);
245         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246                 if (mddev->unit == unit) {
247                         mddev_get(mddev);
248                         spin_unlock(&all_mddevs_lock);
249                         kfree(new);
250                         return mddev;
251                 }
252
253         if (new) {
254                 list_add(&new->all_mddevs, &all_mddevs);
255                 spin_unlock(&all_mddevs_lock);
256                 return new;
257         }
258         spin_unlock(&all_mddevs_lock);
259
260         new = kzalloc(sizeof(*new), GFP_KERNEL);
261         if (!new)
262                 return NULL;
263
264         new->unit = unit;
265         if (MAJOR(unit) == MD_MAJOR)
266                 new->md_minor = MINOR(unit);
267         else
268                 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
270         mutex_init(&new->reconfig_mutex);
271         INIT_LIST_HEAD(&new->disks);
272         INIT_LIST_HEAD(&new->all_mddevs);
273         init_timer(&new->safemode_timer);
274         atomic_set(&new->active, 1);
275         spin_lock_init(&new->write_lock);
276         init_waitqueue_head(&new->sb_wait);
277         new->reshape_position = MaxSector;
278
279         new->queue = blk_alloc_queue(GFP_KERNEL);
280         if (!new->queue) {
281                 kfree(new);
282                 return NULL;
283         }
284         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
285
286         blk_queue_make_request(new->queue, md_fail_request);
287
288         goto retry;
289 }
290
291 static inline int mddev_lock(mddev_t * mddev)
292 {
293         return mutex_lock_interruptible(&mddev->reconfig_mutex);
294 }
295
296 static inline int mddev_trylock(mddev_t * mddev)
297 {
298         return mutex_trylock(&mddev->reconfig_mutex);
299 }
300
301 static inline void mddev_unlock(mddev_t * mddev)
302 {
303         mutex_unlock(&mddev->reconfig_mutex);
304
305         md_wakeup_thread(mddev->thread);
306 }
307
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
309 {
310         mdk_rdev_t * rdev;
311         struct list_head *tmp;
312
313         ITERATE_RDEV(mddev,rdev,tmp) {
314                 if (rdev->desc_nr == nr)
315                         return rdev;
316         }
317         return NULL;
318 }
319
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
321 {
322         struct list_head *tmp;
323         mdk_rdev_t *rdev;
324
325         ITERATE_RDEV(mddev,rdev,tmp) {
326                 if (rdev->bdev->bd_dev == dev)
327                         return rdev;
328         }
329         return NULL;
330 }
331
332 static struct mdk_personality *find_pers(int level, char *clevel)
333 {
334         struct mdk_personality *pers;
335         list_for_each_entry(pers, &pers_list, list) {
336                 if (level != LEVEL_NONE && pers->level == level)
337                         return pers;
338                 if (strcmp(pers->name, clevel)==0)
339                         return pers;
340         }
341         return NULL;
342 }
343
344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
345 {
346         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
347         return MD_NEW_SIZE_BLOCKS(size);
348 }
349
350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
351 {
352         sector_t size;
353
354         size = rdev->sb_offset;
355
356         if (chunk_size)
357                 size &= ~((sector_t)chunk_size/1024 - 1);
358         return size;
359 }
360
361 static int alloc_disk_sb(mdk_rdev_t * rdev)
362 {
363         if (rdev->sb_page)
364                 MD_BUG();
365
366         rdev->sb_page = alloc_page(GFP_KERNEL);
367         if (!rdev->sb_page) {
368                 printk(KERN_ALERT "md: out of memory.\n");
369                 return -EINVAL;
370         }
371
372         return 0;
373 }
374
375 static void free_disk_sb(mdk_rdev_t * rdev)
376 {
377         if (rdev->sb_page) {
378                 put_page(rdev->sb_page);
379                 rdev->sb_loaded = 0;
380                 rdev->sb_page = NULL;
381                 rdev->sb_offset = 0;
382                 rdev->size = 0;
383         }
384 }
385
386
387 static void super_written(struct bio *bio, int error)
388 {
389         mdk_rdev_t *rdev = bio->bi_private;
390         mddev_t *mddev = rdev->mddev;
391
392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
393                 printk("md: super_written gets error=%d, uptodate=%d\n",
394                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
395                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
396                 md_error(mddev, rdev);
397         }
398
399         if (atomic_dec_and_test(&mddev->pending_writes))
400                 wake_up(&mddev->sb_wait);
401         bio_put(bio);
402 }
403
404 static void super_written_barrier(struct bio *bio, int error)
405 {
406         struct bio *bio2 = bio->bi_private;
407         mdk_rdev_t *rdev = bio2->bi_private;
408         mddev_t *mddev = rdev->mddev;
409
410         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
411             error == -EOPNOTSUPP) {
412                 unsigned long flags;
413                 /* barriers don't appear to be supported :-( */
414                 set_bit(BarriersNotsupp, &rdev->flags);
415                 mddev->barriers_work = 0;
416                 spin_lock_irqsave(&mddev->write_lock, flags);
417                 bio2->bi_next = mddev->biolist;
418                 mddev->biolist = bio2;
419                 spin_unlock_irqrestore(&mddev->write_lock, flags);
420                 wake_up(&mddev->sb_wait);
421                 bio_put(bio);
422         } else {
423                 bio_put(bio2);
424                 bio->bi_private = rdev;
425                 super_written(bio, error);
426         }
427 }
428
429 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
430                    sector_t sector, int size, struct page *page)
431 {
432         /* write first size bytes of page to sector of rdev
433          * Increment mddev->pending_writes before returning
434          * and decrement it on completion, waking up sb_wait
435          * if zero is reached.
436          * If an error occurred, call md_error
437          *
438          * As we might need to resubmit the request if BIO_RW_BARRIER
439          * causes ENOTSUPP, we allocate a spare bio...
440          */
441         struct bio *bio = bio_alloc(GFP_NOIO, 1);
442         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
443
444         bio->bi_bdev = rdev->bdev;
445         bio->bi_sector = sector;
446         bio_add_page(bio, page, size, 0);
447         bio->bi_private = rdev;
448         bio->bi_end_io = super_written;
449         bio->bi_rw = rw;
450
451         atomic_inc(&mddev->pending_writes);
452         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
453                 struct bio *rbio;
454                 rw |= (1<<BIO_RW_BARRIER);
455                 rbio = bio_clone(bio, GFP_NOIO);
456                 rbio->bi_private = bio;
457                 rbio->bi_end_io = super_written_barrier;
458                 submit_bio(rw, rbio);
459         } else
460                 submit_bio(rw, bio);
461 }
462
463 void md_super_wait(mddev_t *mddev)
464 {
465         /* wait for all superblock writes that were scheduled to complete.
466          * if any had to be retried (due to BARRIER problems), retry them
467          */
468         DEFINE_WAIT(wq);
469         for(;;) {
470                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
471                 if (atomic_read(&mddev->pending_writes)==0)
472                         break;
473                 while (mddev->biolist) {
474                         struct bio *bio;
475                         spin_lock_irq(&mddev->write_lock);
476                         bio = mddev->biolist;
477                         mddev->biolist = bio->bi_next ;
478                         bio->bi_next = NULL;
479                         spin_unlock_irq(&mddev->write_lock);
480                         submit_bio(bio->bi_rw, bio);
481                 }
482                 schedule();
483         }
484         finish_wait(&mddev->sb_wait, &wq);
485 }
486
487 static void bi_complete(struct bio *bio, int error)
488 {
489         complete((struct completion*)bio->bi_private);
490 }
491
492 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
493                    struct page *page, int rw)
494 {
495         struct bio *bio = bio_alloc(GFP_NOIO, 1);
496         struct completion event;
497         int ret;
498
499         rw |= (1 << BIO_RW_SYNC);
500
501         bio->bi_bdev = bdev;
502         bio->bi_sector = sector;
503         bio_add_page(bio, page, size, 0);
504         init_completion(&event);
505         bio->bi_private = &event;
506         bio->bi_end_io = bi_complete;
507         submit_bio(rw, bio);
508         wait_for_completion(&event);
509
510         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
511         bio_put(bio);
512         return ret;
513 }
514 EXPORT_SYMBOL_GPL(sync_page_io);
515
516 static int read_disk_sb(mdk_rdev_t * rdev, int size)
517 {
518         char b[BDEVNAME_SIZE];
519         if (!rdev->sb_page) {
520                 MD_BUG();
521                 return -EINVAL;
522         }
523         if (rdev->sb_loaded)
524                 return 0;
525
526
527         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
528                 goto fail;
529         rdev->sb_loaded = 1;
530         return 0;
531
532 fail:
533         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
534                 bdevname(rdev->bdev,b));
535         return -EINVAL;
536 }
537
538 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
539 {
540         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
541                 (sb1->set_uuid1 == sb2->set_uuid1) &&
542                 (sb1->set_uuid2 == sb2->set_uuid2) &&
543                 (sb1->set_uuid3 == sb2->set_uuid3))
544
545                 return 1;
546
547         return 0;
548 }
549
550
551 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
552 {
553         int ret;
554         mdp_super_t *tmp1, *tmp2;
555
556         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
557         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
558
559         if (!tmp1 || !tmp2) {
560                 ret = 0;
561                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
562                 goto abort;
563         }
564
565         *tmp1 = *sb1;
566         *tmp2 = *sb2;
567
568         /*
569          * nr_disks is not constant
570          */
571         tmp1->nr_disks = 0;
572         tmp2->nr_disks = 0;
573
574         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
575                 ret = 0;
576         else
577                 ret = 1;
578
579 abort:
580         kfree(tmp1);
581         kfree(tmp2);
582         return ret;
583 }
584
585
586 static u32 md_csum_fold(u32 csum)
587 {
588         csum = (csum & 0xffff) + (csum >> 16);
589         return (csum & 0xffff) + (csum >> 16);
590 }
591
592 static unsigned int calc_sb_csum(mdp_super_t * sb)
593 {
594         u64 newcsum = 0;
595         u32 *sb32 = (u32*)sb;
596         int i;
597         unsigned int disk_csum, csum;
598
599         disk_csum = sb->sb_csum;
600         sb->sb_csum = 0;
601
602         for (i = 0; i < MD_SB_BYTES/4 ; i++)
603                 newcsum += sb32[i];
604         csum = (newcsum & 0xffffffff) + (newcsum>>32);
605
606
607 #ifdef CONFIG_ALPHA
608         /* This used to use csum_partial, which was wrong for several
609          * reasons including that different results are returned on
610          * different architectures.  It isn't critical that we get exactly
611          * the same return value as before (we always csum_fold before
612          * testing, and that removes any differences).  However as we
613          * know that csum_partial always returned a 16bit value on
614          * alphas, do a fold to maximise conformity to previous behaviour.
615          */
616         sb->sb_csum = md_csum_fold(disk_csum);
617 #else
618         sb->sb_csum = disk_csum;
619 #endif
620         return csum;
621 }
622
623
624 /*
625  * Handle superblock details.
626  * We want to be able to handle multiple superblock formats
627  * so we have a common interface to them all, and an array of
628  * different handlers.
629  * We rely on user-space to write the initial superblock, and support
630  * reading and updating of superblocks.
631  * Interface methods are:
632  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
633  *      loads and validates a superblock on dev.
634  *      if refdev != NULL, compare superblocks on both devices
635  *    Return:
636  *      0 - dev has a superblock that is compatible with refdev
637  *      1 - dev has a superblock that is compatible and newer than refdev
638  *          so dev should be used as the refdev in future
639  *     -EINVAL superblock incompatible or invalid
640  *     -othererror e.g. -EIO
641  *
642  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
643  *      Verify that dev is acceptable into mddev.
644  *       The first time, mddev->raid_disks will be 0, and data from
645  *       dev should be merged in.  Subsequent calls check that dev
646  *       is new enough.  Return 0 or -EINVAL
647  *
648  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
649  *     Update the superblock for rdev with data in mddev
650  *     This does not write to disc.
651  *
652  */
653
654 struct super_type  {
655         char            *name;
656         struct module   *owner;
657         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
658         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
659         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
660 };
661
662 /*
663  * load_super for 0.90.0 
664  */
665 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
666 {
667         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
668         mdp_super_t *sb;
669         int ret;
670         sector_t sb_offset;
671
672         /*
673          * Calculate the position of the superblock,
674          * it's at the end of the disk.
675          *
676          * It also happens to be a multiple of 4Kb.
677          */
678         sb_offset = calc_dev_sboffset(rdev->bdev);
679         rdev->sb_offset = sb_offset;
680
681         ret = read_disk_sb(rdev, MD_SB_BYTES);
682         if (ret) return ret;
683
684         ret = -EINVAL;
685
686         bdevname(rdev->bdev, b);
687         sb = (mdp_super_t*)page_address(rdev->sb_page);
688
689         if (sb->md_magic != MD_SB_MAGIC) {
690                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
691                        b);
692                 goto abort;
693         }
694
695         if (sb->major_version != 0 ||
696             sb->minor_version < 90 ||
697             sb->minor_version > 91) {
698                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
699                         sb->major_version, sb->minor_version,
700                         b);
701                 goto abort;
702         }
703
704         if (sb->raid_disks <= 0)
705                 goto abort;
706
707         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
708                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
709                         b);
710                 goto abort;
711         }
712
713         rdev->preferred_minor = sb->md_minor;
714         rdev->data_offset = 0;
715         rdev->sb_size = MD_SB_BYTES;
716
717         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
718                 if (sb->level != 1 && sb->level != 4
719                     && sb->level != 5 && sb->level != 6
720                     && sb->level != 10) {
721                         /* FIXME use a better test */
722                         printk(KERN_WARNING
723                                "md: bitmaps not supported for this level.\n");
724                         goto abort;
725                 }
726         }
727
728         if (sb->level == LEVEL_MULTIPATH)
729                 rdev->desc_nr = -1;
730         else
731                 rdev->desc_nr = sb->this_disk.number;
732
733         if (refdev == 0)
734                 ret = 1;
735         else {
736                 __u64 ev1, ev2;
737                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
738                 if (!uuid_equal(refsb, sb)) {
739                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
740                                 b, bdevname(refdev->bdev,b2));
741                         goto abort;
742                 }
743                 if (!sb_equal(refsb, sb)) {
744                         printk(KERN_WARNING "md: %s has same UUID"
745                                " but different superblock to %s\n",
746                                b, bdevname(refdev->bdev, b2));
747                         goto abort;
748                 }
749                 ev1 = md_event(sb);
750                 ev2 = md_event(refsb);
751                 if (ev1 > ev2)
752                         ret = 1;
753                 else 
754                         ret = 0;
755         }
756         rdev->size = calc_dev_size(rdev, sb->chunk_size);
757
758         if (rdev->size < sb->size && sb->level > 1)
759                 /* "this cannot possibly happen" ... */
760                 ret = -EINVAL;
761
762  abort:
763         return ret;
764 }
765
766 /*
767  * validate_super for 0.90.0
768  */
769 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
770 {
771         mdp_disk_t *desc;
772         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
773         __u64 ev1 = md_event(sb);
774
775         rdev->raid_disk = -1;
776         rdev->flags = 0;
777         if (mddev->raid_disks == 0) {
778                 mddev->major_version = 0;
779                 mddev->minor_version = sb->minor_version;
780                 mddev->patch_version = sb->patch_version;
781                 mddev->persistent = ! sb->not_persistent;
782                 mddev->chunk_size = sb->chunk_size;
783                 mddev->ctime = sb->ctime;
784                 mddev->utime = sb->utime;
785                 mddev->level = sb->level;
786                 mddev->clevel[0] = 0;
787                 mddev->layout = sb->layout;
788                 mddev->raid_disks = sb->raid_disks;
789                 mddev->size = sb->size;
790                 mddev->events = ev1;
791                 mddev->bitmap_offset = 0;
792                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
793
794                 if (mddev->minor_version >= 91) {
795                         mddev->reshape_position = sb->reshape_position;
796                         mddev->delta_disks = sb->delta_disks;
797                         mddev->new_level = sb->new_level;
798                         mddev->new_layout = sb->new_layout;
799                         mddev->new_chunk = sb->new_chunk;
800                 } else {
801                         mddev->reshape_position = MaxSector;
802                         mddev->delta_disks = 0;
803                         mddev->new_level = mddev->level;
804                         mddev->new_layout = mddev->layout;
805                         mddev->new_chunk = mddev->chunk_size;
806                 }
807
808                 if (sb->state & (1<<MD_SB_CLEAN))
809                         mddev->recovery_cp = MaxSector;
810                 else {
811                         if (sb->events_hi == sb->cp_events_hi && 
812                                 sb->events_lo == sb->cp_events_lo) {
813                                 mddev->recovery_cp = sb->recovery_cp;
814                         } else
815                                 mddev->recovery_cp = 0;
816                 }
817
818                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
819                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
820                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
821                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
822
823                 mddev->max_disks = MD_SB_DISKS;
824
825                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
826                     mddev->bitmap_file == NULL)
827                         mddev->bitmap_offset = mddev->default_bitmap_offset;
828
829         } else if (mddev->pers == NULL) {
830                 /* Insist on good event counter while assembling */
831                 ++ev1;
832                 if (ev1 < mddev->events) 
833                         return -EINVAL;
834         } else if (mddev->bitmap) {
835                 /* if adding to array with a bitmap, then we can accept an
836                  * older device ... but not too old.
837                  */
838                 if (ev1 < mddev->bitmap->events_cleared)
839                         return 0;
840         } else {
841                 if (ev1 < mddev->events)
842                         /* just a hot-add of a new device, leave raid_disk at -1 */
843                         return 0;
844         }
845
846         if (mddev->level != LEVEL_MULTIPATH) {
847                 desc = sb->disks + rdev->desc_nr;
848
849                 if (desc->state & (1<<MD_DISK_FAULTY))
850                         set_bit(Faulty, &rdev->flags);
851                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
852                             desc->raid_disk < mddev->raid_disks */) {
853                         set_bit(In_sync, &rdev->flags);
854                         rdev->raid_disk = desc->raid_disk;
855                 }
856                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
857                         set_bit(WriteMostly, &rdev->flags);
858         } else /* MULTIPATH are always insync */
859                 set_bit(In_sync, &rdev->flags);
860         return 0;
861 }
862
863 /*
864  * sync_super for 0.90.0
865  */
866 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
867 {
868         mdp_super_t *sb;
869         struct list_head *tmp;
870         mdk_rdev_t *rdev2;
871         int next_spare = mddev->raid_disks;
872
873
874         /* make rdev->sb match mddev data..
875          *
876          * 1/ zero out disks
877          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
878          * 3/ any empty disks < next_spare become removed
879          *
880          * disks[0] gets initialised to REMOVED because
881          * we cannot be sure from other fields if it has
882          * been initialised or not.
883          */
884         int i;
885         int active=0, working=0,failed=0,spare=0,nr_disks=0;
886
887         rdev->sb_size = MD_SB_BYTES;
888
889         sb = (mdp_super_t*)page_address(rdev->sb_page);
890
891         memset(sb, 0, sizeof(*sb));
892
893         sb->md_magic = MD_SB_MAGIC;
894         sb->major_version = mddev->major_version;
895         sb->patch_version = mddev->patch_version;
896         sb->gvalid_words  = 0; /* ignored */
897         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
898         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
899         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
900         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
901
902         sb->ctime = mddev->ctime;
903         sb->level = mddev->level;
904         sb->size  = mddev->size;
905         sb->raid_disks = mddev->raid_disks;
906         sb->md_minor = mddev->md_minor;
907         sb->not_persistent = !mddev->persistent;
908         sb->utime = mddev->utime;
909         sb->state = 0;
910         sb->events_hi = (mddev->events>>32);
911         sb->events_lo = (u32)mddev->events;
912
913         if (mddev->reshape_position == MaxSector)
914                 sb->minor_version = 90;
915         else {
916                 sb->minor_version = 91;
917                 sb->reshape_position = mddev->reshape_position;
918                 sb->new_level = mddev->new_level;
919                 sb->delta_disks = mddev->delta_disks;
920                 sb->new_layout = mddev->new_layout;
921                 sb->new_chunk = mddev->new_chunk;
922         }
923         mddev->minor_version = sb->minor_version;
924         if (mddev->in_sync)
925         {
926                 sb->recovery_cp = mddev->recovery_cp;
927                 sb->cp_events_hi = (mddev->events>>32);
928                 sb->cp_events_lo = (u32)mddev->events;
929                 if (mddev->recovery_cp == MaxSector)
930                         sb->state = (1<< MD_SB_CLEAN);
931         } else
932                 sb->recovery_cp = 0;
933
934         sb->layout = mddev->layout;
935         sb->chunk_size = mddev->chunk_size;
936
937         if (mddev->bitmap && mddev->bitmap_file == NULL)
938                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
939
940         sb->disks[0].state = (1<<MD_DISK_REMOVED);
941         ITERATE_RDEV(mddev,rdev2,tmp) {
942                 mdp_disk_t *d;
943                 int desc_nr;
944                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
945                     && !test_bit(Faulty, &rdev2->flags))
946                         desc_nr = rdev2->raid_disk;
947                 else
948                         desc_nr = next_spare++;
949                 rdev2->desc_nr = desc_nr;
950                 d = &sb->disks[rdev2->desc_nr];
951                 nr_disks++;
952                 d->number = rdev2->desc_nr;
953                 d->major = MAJOR(rdev2->bdev->bd_dev);
954                 d->minor = MINOR(rdev2->bdev->bd_dev);
955                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
956                     && !test_bit(Faulty, &rdev2->flags))
957                         d->raid_disk = rdev2->raid_disk;
958                 else
959                         d->raid_disk = rdev2->desc_nr; /* compatibility */
960                 if (test_bit(Faulty, &rdev2->flags))
961                         d->state = (1<<MD_DISK_FAULTY);
962                 else if (test_bit(In_sync, &rdev2->flags)) {
963                         d->state = (1<<MD_DISK_ACTIVE);
964                         d->state |= (1<<MD_DISK_SYNC);
965                         active++;
966                         working++;
967                 } else {
968                         d->state = 0;
969                         spare++;
970                         working++;
971                 }
972                 if (test_bit(WriteMostly, &rdev2->flags))
973                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
974         }
975         /* now set the "removed" and "faulty" bits on any missing devices */
976         for (i=0 ; i < mddev->raid_disks ; i++) {
977                 mdp_disk_t *d = &sb->disks[i];
978                 if (d->state == 0 && d->number == 0) {
979                         d->number = i;
980                         d->raid_disk = i;
981                         d->state = (1<<MD_DISK_REMOVED);
982                         d->state |= (1<<MD_DISK_FAULTY);
983                         failed++;
984                 }
985         }
986         sb->nr_disks = nr_disks;
987         sb->active_disks = active;
988         sb->working_disks = working;
989         sb->failed_disks = failed;
990         sb->spare_disks = spare;
991
992         sb->this_disk = sb->disks[rdev->desc_nr];
993         sb->sb_csum = calc_sb_csum(sb);
994 }
995
996 /*
997  * version 1 superblock
998  */
999
1000 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1001 {
1002         __le32 disk_csum;
1003         u32 csum;
1004         unsigned long long newcsum;
1005         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1006         __le32 *isuper = (__le32*)sb;
1007         int i;
1008
1009         disk_csum = sb->sb_csum;
1010         sb->sb_csum = 0;
1011         newcsum = 0;
1012         for (i=0; size>=4; size -= 4 )
1013                 newcsum += le32_to_cpu(*isuper++);
1014
1015         if (size == 2)
1016                 newcsum += le16_to_cpu(*(__le16*) isuper);
1017
1018         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1019         sb->sb_csum = disk_csum;
1020         return cpu_to_le32(csum);
1021 }
1022
1023 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1024 {
1025         struct mdp_superblock_1 *sb;
1026         int ret;
1027         sector_t sb_offset;
1028         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1029         int bmask;
1030
1031         /*
1032          * Calculate the position of the superblock.
1033          * It is always aligned to a 4K boundary and
1034          * depeding on minor_version, it can be:
1035          * 0: At least 8K, but less than 12K, from end of device
1036          * 1: At start of device
1037          * 2: 4K from start of device.
1038          */
1039         switch(minor_version) {
1040         case 0:
1041                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1042                 sb_offset -= 8*2;
1043                 sb_offset &= ~(sector_t)(4*2-1);
1044                 /* convert from sectors to K */
1045                 sb_offset /= 2;
1046                 break;
1047         case 1:
1048                 sb_offset = 0;
1049                 break;
1050         case 2:
1051                 sb_offset = 4;
1052                 break;
1053         default:
1054                 return -EINVAL;
1055         }
1056         rdev->sb_offset = sb_offset;
1057
1058         /* superblock is rarely larger than 1K, but it can be larger,
1059          * and it is safe to read 4k, so we do that
1060          */
1061         ret = read_disk_sb(rdev, 4096);
1062         if (ret) return ret;
1063
1064
1065         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1066
1067         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1068             sb->major_version != cpu_to_le32(1) ||
1069             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1070             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1071             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1072                 return -EINVAL;
1073
1074         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1075                 printk("md: invalid superblock checksum on %s\n",
1076                         bdevname(rdev->bdev,b));
1077                 return -EINVAL;
1078         }
1079         if (le64_to_cpu(sb->data_size) < 10) {
1080                 printk("md: data_size too small on %s\n",
1081                        bdevname(rdev->bdev,b));
1082                 return -EINVAL;
1083         }
1084         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1085                 if (sb->level != cpu_to_le32(1) &&
1086                     sb->level != cpu_to_le32(4) &&
1087                     sb->level != cpu_to_le32(5) &&
1088                     sb->level != cpu_to_le32(6) &&
1089                     sb->level != cpu_to_le32(10)) {
1090                         printk(KERN_WARNING
1091                                "md: bitmaps not supported for this level.\n");
1092                         return -EINVAL;
1093                 }
1094         }
1095
1096         rdev->preferred_minor = 0xffff;
1097         rdev->data_offset = le64_to_cpu(sb->data_offset);
1098         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1099
1100         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1101         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1102         if (rdev->sb_size & bmask)
1103                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1104
1105         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1106                 rdev->desc_nr = -1;
1107         else
1108                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1109
1110         if (refdev == 0)
1111                 ret = 1;
1112         else {
1113                 __u64 ev1, ev2;
1114                 struct mdp_superblock_1 *refsb = 
1115                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1116
1117                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1118                     sb->level != refsb->level ||
1119                     sb->layout != refsb->layout ||
1120                     sb->chunksize != refsb->chunksize) {
1121                         printk(KERN_WARNING "md: %s has strangely different"
1122                                 " superblock to %s\n",
1123                                 bdevname(rdev->bdev,b),
1124                                 bdevname(refdev->bdev,b2));
1125                         return -EINVAL;
1126                 }
1127                 ev1 = le64_to_cpu(sb->events);
1128                 ev2 = le64_to_cpu(refsb->events);
1129
1130                 if (ev1 > ev2)
1131                         ret = 1;
1132                 else
1133                         ret = 0;
1134         }
1135         if (minor_version) 
1136                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1137         else
1138                 rdev->size = rdev->sb_offset;
1139         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1140                 return -EINVAL;
1141         rdev->size = le64_to_cpu(sb->data_size)/2;
1142         if (le32_to_cpu(sb->chunksize))
1143                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1144
1145         if (le64_to_cpu(sb->size) > rdev->size*2)
1146                 return -EINVAL;
1147         return ret;
1148 }
1149
1150 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1151 {
1152         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1153         __u64 ev1 = le64_to_cpu(sb->events);
1154
1155         rdev->raid_disk = -1;
1156         rdev->flags = 0;
1157         if (mddev->raid_disks == 0) {
1158                 mddev->major_version = 1;
1159                 mddev->patch_version = 0;
1160                 mddev->persistent = 1;
1161                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1162                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1163                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1164                 mddev->level = le32_to_cpu(sb->level);
1165                 mddev->clevel[0] = 0;
1166                 mddev->layout = le32_to_cpu(sb->layout);
1167                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1168                 mddev->size = le64_to_cpu(sb->size)/2;
1169                 mddev->events = ev1;
1170                 mddev->bitmap_offset = 0;
1171                 mddev->default_bitmap_offset = 1024 >> 9;
1172                 
1173                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1174                 memcpy(mddev->uuid, sb->set_uuid, 16);
1175
1176                 mddev->max_disks =  (4096-256)/2;
1177
1178                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1179                     mddev->bitmap_file == NULL )
1180                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1181
1182                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1183                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1184                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1185                         mddev->new_level = le32_to_cpu(sb->new_level);
1186                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1187                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1188                 } else {
1189                         mddev->reshape_position = MaxSector;
1190                         mddev->delta_disks = 0;
1191                         mddev->new_level = mddev->level;
1192                         mddev->new_layout = mddev->layout;
1193                         mddev->new_chunk = mddev->chunk_size;
1194                 }
1195
1196         } else if (mddev->pers == NULL) {
1197                 /* Insist of good event counter while assembling */
1198                 ++ev1;
1199                 if (ev1 < mddev->events)
1200                         return -EINVAL;
1201         } else if (mddev->bitmap) {
1202                 /* If adding to array with a bitmap, then we can accept an
1203                  * older device, but not too old.
1204                  */
1205                 if (ev1 < mddev->bitmap->events_cleared)
1206                         return 0;
1207         } else {
1208                 if (ev1 < mddev->events)
1209                         /* just a hot-add of a new device, leave raid_disk at -1 */
1210                         return 0;
1211         }
1212         if (mddev->level != LEVEL_MULTIPATH) {
1213                 int role;
1214                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1215                 switch(role) {
1216                 case 0xffff: /* spare */
1217                         break;
1218                 case 0xfffe: /* faulty */
1219                         set_bit(Faulty, &rdev->flags);
1220                         break;
1221                 default:
1222                         if ((le32_to_cpu(sb->feature_map) &
1223                              MD_FEATURE_RECOVERY_OFFSET))
1224                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1225                         else
1226                                 set_bit(In_sync, &rdev->flags);
1227                         rdev->raid_disk = role;
1228                         break;
1229                 }
1230                 if (sb->devflags & WriteMostly1)
1231                         set_bit(WriteMostly, &rdev->flags);
1232         } else /* MULTIPATH are always insync */
1233                 set_bit(In_sync, &rdev->flags);
1234
1235         return 0;
1236 }
1237
1238 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1239 {
1240         struct mdp_superblock_1 *sb;
1241         struct list_head *tmp;
1242         mdk_rdev_t *rdev2;
1243         int max_dev, i;
1244         /* make rdev->sb match mddev and rdev data. */
1245
1246         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1247
1248         sb->feature_map = 0;
1249         sb->pad0 = 0;
1250         sb->recovery_offset = cpu_to_le64(0);
1251         memset(sb->pad1, 0, sizeof(sb->pad1));
1252         memset(sb->pad2, 0, sizeof(sb->pad2));
1253         memset(sb->pad3, 0, sizeof(sb->pad3));
1254
1255         sb->utime = cpu_to_le64((__u64)mddev->utime);
1256         sb->events = cpu_to_le64(mddev->events);
1257         if (mddev->in_sync)
1258                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1259         else
1260                 sb->resync_offset = cpu_to_le64(0);
1261
1262         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1263
1264         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1265         sb->size = cpu_to_le64(mddev->size<<1);
1266
1267         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1268                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1269                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1270         }
1271
1272         if (rdev->raid_disk >= 0 &&
1273             !test_bit(In_sync, &rdev->flags) &&
1274             rdev->recovery_offset > 0) {
1275                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1276                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1277         }
1278
1279         if (mddev->reshape_position != MaxSector) {
1280                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1281                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1282                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1283                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1284                 sb->new_level = cpu_to_le32(mddev->new_level);
1285                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1286         }
1287
1288         max_dev = 0;
1289         ITERATE_RDEV(mddev,rdev2,tmp)
1290                 if (rdev2->desc_nr+1 > max_dev)
1291                         max_dev = rdev2->desc_nr+1;
1292
1293         if (max_dev > le32_to_cpu(sb->max_dev))
1294                 sb->max_dev = cpu_to_le32(max_dev);
1295         for (i=0; i<max_dev;i++)
1296                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1297         
1298         ITERATE_RDEV(mddev,rdev2,tmp) {
1299                 i = rdev2->desc_nr;
1300                 if (test_bit(Faulty, &rdev2->flags))
1301                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1302                 else if (test_bit(In_sync, &rdev2->flags))
1303                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1304                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1305                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1306                 else
1307                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1308         }
1309
1310         sb->sb_csum = calc_sb_1_csum(sb);
1311 }
1312
1313
1314 static struct super_type super_types[] = {
1315         [0] = {
1316                 .name   = "0.90.0",
1317                 .owner  = THIS_MODULE,
1318                 .load_super     = super_90_load,
1319                 .validate_super = super_90_validate,
1320                 .sync_super     = super_90_sync,
1321         },
1322         [1] = {
1323                 .name   = "md-1",
1324                 .owner  = THIS_MODULE,
1325                 .load_super     = super_1_load,
1326                 .validate_super = super_1_validate,
1327                 .sync_super     = super_1_sync,
1328         },
1329 };
1330
1331 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1332 {
1333         struct list_head *tmp, *tmp2;
1334         mdk_rdev_t *rdev, *rdev2;
1335
1336         ITERATE_RDEV(mddev1,rdev,tmp)
1337                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1338                         if (rdev->bdev->bd_contains ==
1339                             rdev2->bdev->bd_contains)
1340                                 return 1;
1341
1342         return 0;
1343 }
1344
1345 static LIST_HEAD(pending_raid_disks);
1346
1347 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1348 {
1349         char b[BDEVNAME_SIZE];
1350         struct kobject *ko;
1351         char *s;
1352         int err;
1353
1354         if (rdev->mddev) {
1355                 MD_BUG();
1356                 return -EINVAL;
1357         }
1358         /* make sure rdev->size exceeds mddev->size */
1359         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1360                 if (mddev->pers) {
1361                         /* Cannot change size, so fail
1362                          * If mddev->level <= 0, then we don't care
1363                          * about aligning sizes (e.g. linear)
1364                          */
1365                         if (mddev->level > 0)
1366                                 return -ENOSPC;
1367                 } else
1368                         mddev->size = rdev->size;
1369         }
1370
1371         /* Verify rdev->desc_nr is unique.
1372          * If it is -1, assign a free number, else
1373          * check number is not in use
1374          */
1375         if (rdev->desc_nr < 0) {
1376                 int choice = 0;
1377                 if (mddev->pers) choice = mddev->raid_disks;
1378                 while (find_rdev_nr(mddev, choice))
1379                         choice++;
1380                 rdev->desc_nr = choice;
1381         } else {
1382                 if (find_rdev_nr(mddev, rdev->desc_nr))
1383                         return -EBUSY;
1384         }
1385         bdevname(rdev->bdev,b);
1386         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1387                 return -ENOMEM;
1388         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1389                 *s = '!';
1390                         
1391         rdev->mddev = mddev;
1392         printk(KERN_INFO "md: bind<%s>\n", b);
1393
1394         rdev->kobj.parent = &mddev->kobj;
1395         if ((err = kobject_add(&rdev->kobj)))
1396                 goto fail;
1397
1398         if (rdev->bdev->bd_part)
1399                 ko = &rdev->bdev->bd_part->kobj;
1400         else
1401                 ko = &rdev->bdev->bd_disk->kobj;
1402         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1403                 kobject_del(&rdev->kobj);
1404                 goto fail;
1405         }
1406         list_add(&rdev->same_set, &mddev->disks);
1407         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1408         return 0;
1409
1410  fail:
1411         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1412                b, mdname(mddev));
1413         return err;
1414 }
1415
1416 static void delayed_delete(struct work_struct *ws)
1417 {
1418         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1419         kobject_del(&rdev->kobj);
1420 }
1421
1422 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1423 {
1424         char b[BDEVNAME_SIZE];
1425         if (!rdev->mddev) {
1426                 MD_BUG();
1427                 return;
1428         }
1429         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1430         list_del_init(&rdev->same_set);
1431         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1432         rdev->mddev = NULL;
1433         sysfs_remove_link(&rdev->kobj, "block");
1434
1435         /* We need to delay this, otherwise we can deadlock when
1436          * writing to 'remove' to "dev/state"
1437          */
1438         INIT_WORK(&rdev->del_work, delayed_delete);
1439         schedule_work(&rdev->del_work);
1440 }
1441
1442 /*
1443  * prevent the device from being mounted, repartitioned or
1444  * otherwise reused by a RAID array (or any other kernel
1445  * subsystem), by bd_claiming the device.
1446  */
1447 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1448 {
1449         int err = 0;
1450         struct block_device *bdev;
1451         char b[BDEVNAME_SIZE];
1452
1453         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1454         if (IS_ERR(bdev)) {
1455                 printk(KERN_ERR "md: could not open %s.\n",
1456                         __bdevname(dev, b));
1457                 return PTR_ERR(bdev);
1458         }
1459         err = bd_claim(bdev, rdev);
1460         if (err) {
1461                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1462                         bdevname(bdev, b));
1463                 blkdev_put(bdev);
1464                 return err;
1465         }
1466         rdev->bdev = bdev;
1467         return err;
1468 }
1469
1470 static void unlock_rdev(mdk_rdev_t *rdev)
1471 {
1472         struct block_device *bdev = rdev->bdev;
1473         rdev->bdev = NULL;
1474         if (!bdev)
1475                 MD_BUG();
1476         bd_release(bdev);
1477         blkdev_put(bdev);
1478 }
1479
1480 void md_autodetect_dev(dev_t dev);
1481
1482 static void export_rdev(mdk_rdev_t * rdev)
1483 {
1484         char b[BDEVNAME_SIZE];
1485         printk(KERN_INFO "md: export_rdev(%s)\n",
1486                 bdevname(rdev->bdev,b));
1487         if (rdev->mddev)
1488                 MD_BUG();
1489         free_disk_sb(rdev);
1490         list_del_init(&rdev->same_set);
1491 #ifndef MODULE
1492         md_autodetect_dev(rdev->bdev->bd_dev);
1493 #endif
1494         unlock_rdev(rdev);
1495         kobject_put(&rdev->kobj);
1496 }
1497
1498 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1499 {
1500         unbind_rdev_from_array(rdev);
1501         export_rdev(rdev);
1502 }
1503
1504 static void export_array(mddev_t *mddev)
1505 {
1506         struct list_head *tmp;
1507         mdk_rdev_t *rdev;
1508
1509         ITERATE_RDEV(mddev,rdev,tmp) {
1510                 if (!rdev->mddev) {
1511                         MD_BUG();
1512                         continue;
1513                 }
1514                 kick_rdev_from_array(rdev);
1515         }
1516         if (!list_empty(&mddev->disks))
1517                 MD_BUG();
1518         mddev->raid_disks = 0;
1519         mddev->major_version = 0;
1520 }
1521
1522 static void print_desc(mdp_disk_t *desc)
1523 {
1524         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1525                 desc->major,desc->minor,desc->raid_disk,desc->state);
1526 }
1527
1528 static void print_sb(mdp_super_t *sb)
1529 {
1530         int i;
1531
1532         printk(KERN_INFO 
1533                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1534                 sb->major_version, sb->minor_version, sb->patch_version,
1535                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1536                 sb->ctime);
1537         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1538                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1539                 sb->md_minor, sb->layout, sb->chunk_size);
1540         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1541                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1542                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1543                 sb->failed_disks, sb->spare_disks,
1544                 sb->sb_csum, (unsigned long)sb->events_lo);
1545
1546         printk(KERN_INFO);
1547         for (i = 0; i < MD_SB_DISKS; i++) {
1548                 mdp_disk_t *desc;
1549
1550                 desc = sb->disks + i;
1551                 if (desc->number || desc->major || desc->minor ||
1552                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1553                         printk("     D %2d: ", i);
1554                         print_desc(desc);
1555                 }
1556         }
1557         printk(KERN_INFO "md:     THIS: ");
1558         print_desc(&sb->this_disk);
1559
1560 }
1561
1562 static void print_rdev(mdk_rdev_t *rdev)
1563 {
1564         char b[BDEVNAME_SIZE];
1565         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1566                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1567                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1568                 rdev->desc_nr);
1569         if (rdev->sb_loaded) {
1570                 printk(KERN_INFO "md: rdev superblock:\n");
1571                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1572         } else
1573                 printk(KERN_INFO "md: no rdev superblock!\n");
1574 }
1575
1576 static void md_print_devices(void)
1577 {
1578         struct list_head *tmp, *tmp2;
1579         mdk_rdev_t *rdev;
1580         mddev_t *mddev;
1581         char b[BDEVNAME_SIZE];
1582
1583         printk("\n");
1584         printk("md:     **********************************\n");
1585         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1586         printk("md:     **********************************\n");
1587         ITERATE_MDDEV(mddev,tmp) {
1588
1589                 if (mddev->bitmap)
1590                         bitmap_print_sb(mddev->bitmap);
1591                 else
1592                         printk("%s: ", mdname(mddev));
1593                 ITERATE_RDEV(mddev,rdev,tmp2)
1594                         printk("<%s>", bdevname(rdev->bdev,b));
1595                 printk("\n");
1596
1597                 ITERATE_RDEV(mddev,rdev,tmp2)
1598                         print_rdev(rdev);
1599         }
1600         printk("md:     **********************************\n");
1601         printk("\n");
1602 }
1603
1604
1605 static void sync_sbs(mddev_t * mddev, int nospares)
1606 {
1607         /* Update each superblock (in-memory image), but
1608          * if we are allowed to, skip spares which already
1609          * have the right event counter, or have one earlier
1610          * (which would mean they aren't being marked as dirty
1611          * with the rest of the array)
1612          */
1613         mdk_rdev_t *rdev;
1614         struct list_head *tmp;
1615
1616         ITERATE_RDEV(mddev,rdev,tmp) {
1617                 if (rdev->sb_events == mddev->events ||
1618                     (nospares &&
1619                      rdev->raid_disk < 0 &&
1620                      (rdev->sb_events&1)==0 &&
1621                      rdev->sb_events+1 == mddev->events)) {
1622                         /* Don't update this superblock */
1623                         rdev->sb_loaded = 2;
1624                 } else {
1625                         super_types[mddev->major_version].
1626                                 sync_super(mddev, rdev);
1627                         rdev->sb_loaded = 1;
1628                 }
1629         }
1630 }
1631
1632 static void md_update_sb(mddev_t * mddev, int force_change)
1633 {
1634         struct list_head *tmp;
1635         mdk_rdev_t *rdev;
1636         int sync_req;
1637         int nospares = 0;
1638
1639 repeat:
1640         spin_lock_irq(&mddev->write_lock);
1641
1642         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1643         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1644                 force_change = 1;
1645         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1646                 /* just a clean<-> dirty transition, possibly leave spares alone,
1647                  * though if events isn't the right even/odd, we will have to do
1648                  * spares after all
1649                  */
1650                 nospares = 1;
1651         if (force_change)
1652                 nospares = 0;
1653         if (mddev->degraded)
1654                 /* If the array is degraded, then skipping spares is both
1655                  * dangerous and fairly pointless.
1656                  * Dangerous because a device that was removed from the array
1657                  * might have a event_count that still looks up-to-date,
1658                  * so it can be re-added without a resync.
1659                  * Pointless because if there are any spares to skip,
1660                  * then a recovery will happen and soon that array won't
1661                  * be degraded any more and the spare can go back to sleep then.
1662                  */
1663                 nospares = 0;
1664
1665         sync_req = mddev->in_sync;
1666         mddev->utime = get_seconds();
1667
1668         /* If this is just a dirty<->clean transition, and the array is clean
1669          * and 'events' is odd, we can roll back to the previous clean state */
1670         if (nospares
1671             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1672             && (mddev->events & 1)
1673             && mddev->events != 1)
1674                 mddev->events--;
1675         else {
1676                 /* otherwise we have to go forward and ... */
1677                 mddev->events ++;
1678                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1679                         /* .. if the array isn't clean, insist on an odd 'events' */
1680                         if ((mddev->events&1)==0) {
1681                                 mddev->events++;
1682                                 nospares = 0;
1683                         }
1684                 } else {
1685                         /* otherwise insist on an even 'events' (for clean states) */
1686                         if ((mddev->events&1)) {
1687                                 mddev->events++;
1688                                 nospares = 0;
1689                         }
1690                 }
1691         }
1692
1693         if (!mddev->events) {
1694                 /*
1695                  * oops, this 64-bit counter should never wrap.
1696                  * Either we are in around ~1 trillion A.C., assuming
1697                  * 1 reboot per second, or we have a bug:
1698                  */
1699                 MD_BUG();
1700                 mddev->events --;
1701         }
1702         sync_sbs(mddev, nospares);
1703
1704         /*
1705          * do not write anything to disk if using
1706          * nonpersistent superblocks
1707          */
1708         if (!mddev->persistent) {
1709                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1710                 spin_unlock_irq(&mddev->write_lock);
1711                 wake_up(&mddev->sb_wait);
1712                 return;
1713         }
1714         spin_unlock_irq(&mddev->write_lock);
1715
1716         dprintk(KERN_INFO 
1717                 "md: updating %s RAID superblock on device (in sync %d)\n",
1718                 mdname(mddev),mddev->in_sync);
1719
1720         bitmap_update_sb(mddev->bitmap);
1721         ITERATE_RDEV(mddev,rdev,tmp) {
1722                 char b[BDEVNAME_SIZE];
1723                 dprintk(KERN_INFO "md: ");
1724                 if (rdev->sb_loaded != 1)
1725                         continue; /* no noise on spare devices */
1726                 if (test_bit(Faulty, &rdev->flags))
1727                         dprintk("(skipping faulty ");
1728
1729                 dprintk("%s ", bdevname(rdev->bdev,b));
1730                 if (!test_bit(Faulty, &rdev->flags)) {
1731                         md_super_write(mddev,rdev,
1732                                        rdev->sb_offset<<1, rdev->sb_size,
1733                                        rdev->sb_page);
1734                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1735                                 bdevname(rdev->bdev,b),
1736                                 (unsigned long long)rdev->sb_offset);
1737                         rdev->sb_events = mddev->events;
1738
1739                 } else
1740                         dprintk(")\n");
1741                 if (mddev->level == LEVEL_MULTIPATH)
1742                         /* only need to write one superblock... */
1743                         break;
1744         }
1745         md_super_wait(mddev);
1746         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1747
1748         spin_lock_irq(&mddev->write_lock);
1749         if (mddev->in_sync != sync_req ||
1750             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1751                 /* have to write it out again */
1752                 spin_unlock_irq(&mddev->write_lock);
1753                 goto repeat;
1754         }
1755         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1756         spin_unlock_irq(&mddev->write_lock);
1757         wake_up(&mddev->sb_wait);
1758
1759 }
1760
1761 /* words written to sysfs files may, or my not, be \n terminated.
1762  * We want to accept with case. For this we use cmd_match.
1763  */
1764 static int cmd_match(const char *cmd, const char *str)
1765 {
1766         /* See if cmd, written into a sysfs file, matches
1767          * str.  They must either be the same, or cmd can
1768          * have a trailing newline
1769          */
1770         while (*cmd && *str && *cmd == *str) {
1771                 cmd++;
1772                 str++;
1773         }
1774         if (*cmd == '\n')
1775                 cmd++;
1776         if (*str || *cmd)
1777                 return 0;
1778         return 1;
1779 }
1780
1781 struct rdev_sysfs_entry {
1782         struct attribute attr;
1783         ssize_t (*show)(mdk_rdev_t *, char *);
1784         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1785 };
1786
1787 static ssize_t
1788 state_show(mdk_rdev_t *rdev, char *page)
1789 {
1790         char *sep = "";
1791         int len=0;
1792
1793         if (test_bit(Faulty, &rdev->flags)) {
1794                 len+= sprintf(page+len, "%sfaulty",sep);
1795                 sep = ",";
1796         }
1797         if (test_bit(In_sync, &rdev->flags)) {
1798                 len += sprintf(page+len, "%sin_sync",sep);
1799                 sep = ",";
1800         }
1801         if (test_bit(WriteMostly, &rdev->flags)) {
1802                 len += sprintf(page+len, "%swrite_mostly",sep);
1803                 sep = ",";
1804         }
1805         if (!test_bit(Faulty, &rdev->flags) &&
1806             !test_bit(In_sync, &rdev->flags)) {
1807                 len += sprintf(page+len, "%sspare", sep);
1808                 sep = ",";
1809         }
1810         return len+sprintf(page+len, "\n");
1811 }
1812
1813 static ssize_t
1814 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1815 {
1816         /* can write
1817          *  faulty  - simulates and error
1818          *  remove  - disconnects the device
1819          *  writemostly - sets write_mostly
1820          *  -writemostly - clears write_mostly
1821          */
1822         int err = -EINVAL;
1823         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1824                 md_error(rdev->mddev, rdev);
1825                 err = 0;
1826         } else if (cmd_match(buf, "remove")) {
1827                 if (rdev->raid_disk >= 0)
1828                         err = -EBUSY;
1829                 else {
1830                         mddev_t *mddev = rdev->mddev;
1831                         kick_rdev_from_array(rdev);
1832                         if (mddev->pers)
1833                                 md_update_sb(mddev, 1);
1834                         md_new_event(mddev);
1835                         err = 0;
1836                 }
1837         } else if (cmd_match(buf, "writemostly")) {
1838                 set_bit(WriteMostly, &rdev->flags);
1839                 err = 0;
1840         } else if (cmd_match(buf, "-writemostly")) {
1841                 clear_bit(WriteMostly, &rdev->flags);
1842                 err = 0;
1843         }
1844         return err ? err : len;
1845 }
1846 static struct rdev_sysfs_entry rdev_state =
1847 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1848
1849 static ssize_t
1850 super_show(mdk_rdev_t *rdev, char *page)
1851 {
1852         if (rdev->sb_loaded && rdev->sb_size) {
1853                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1854                 return rdev->sb_size;
1855         } else
1856                 return 0;
1857 }
1858 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1859
1860 static ssize_t
1861 errors_show(mdk_rdev_t *rdev, char *page)
1862 {
1863         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1864 }
1865
1866 static ssize_t
1867 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1868 {
1869         char *e;
1870         unsigned long n = simple_strtoul(buf, &e, 10);
1871         if (*buf && (*e == 0 || *e == '\n')) {
1872                 atomic_set(&rdev->corrected_errors, n);
1873                 return len;
1874         }
1875         return -EINVAL;
1876 }
1877 static struct rdev_sysfs_entry rdev_errors =
1878 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1879
1880 static ssize_t
1881 slot_show(mdk_rdev_t *rdev, char *page)
1882 {
1883         if (rdev->raid_disk < 0)
1884                 return sprintf(page, "none\n");
1885         else
1886                 return sprintf(page, "%d\n", rdev->raid_disk);
1887 }
1888
1889 static ssize_t
1890 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1891 {
1892         char *e;
1893         int slot = simple_strtoul(buf, &e, 10);
1894         if (strncmp(buf, "none", 4)==0)
1895                 slot = -1;
1896         else if (e==buf || (*e && *e!= '\n'))
1897                 return -EINVAL;
1898         if (rdev->mddev->pers)
1899                 /* Cannot set slot in active array (yet) */
1900                 return -EBUSY;
1901         if (slot >= rdev->mddev->raid_disks)
1902                 return -ENOSPC;
1903         rdev->raid_disk = slot;
1904         /* assume it is working */
1905         rdev->flags = 0;
1906         set_bit(In_sync, &rdev->flags);
1907         return len;
1908 }
1909
1910
1911 static struct rdev_sysfs_entry rdev_slot =
1912 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1913
1914 static ssize_t
1915 offset_show(mdk_rdev_t *rdev, char *page)
1916 {
1917         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1918 }
1919
1920 static ssize_t
1921 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1922 {
1923         char *e;
1924         unsigned long long offset = simple_strtoull(buf, &e, 10);
1925         if (e==buf || (*e && *e != '\n'))
1926                 return -EINVAL;
1927         if (rdev->mddev->pers)
1928                 return -EBUSY;
1929         rdev->data_offset = offset;
1930         return len;
1931 }
1932
1933 static struct rdev_sysfs_entry rdev_offset =
1934 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1935
1936 static ssize_t
1937 rdev_size_show(mdk_rdev_t *rdev, char *page)
1938 {
1939         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1940 }
1941
1942 static ssize_t
1943 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1944 {
1945         char *e;
1946         unsigned long long size = simple_strtoull(buf, &e, 10);
1947         if (e==buf || (*e && *e != '\n'))
1948                 return -EINVAL;
1949         if (rdev->mddev->pers)
1950                 return -EBUSY;
1951         rdev->size = size;
1952         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1953                 rdev->mddev->size = size;
1954         return len;
1955 }
1956
1957 static struct rdev_sysfs_entry rdev_size =
1958 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1959
1960 static struct attribute *rdev_default_attrs[] = {
1961         &rdev_state.attr,
1962         &rdev_super.attr,
1963         &rdev_errors.attr,
1964         &rdev_slot.attr,
1965         &rdev_offset.attr,
1966         &rdev_size.attr,
1967         NULL,
1968 };
1969 static ssize_t
1970 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1971 {
1972         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1973         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1974
1975         if (!entry->show)
1976                 return -EIO;
1977         return entry->show(rdev, page);
1978 }
1979
1980 static ssize_t
1981 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1982               const char *page, size_t length)
1983 {
1984         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1985         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1986
1987         if (!entry->store)
1988                 return -EIO;
1989         if (!capable(CAP_SYS_ADMIN))
1990                 return -EACCES;
1991         return entry->store(rdev, page, length);
1992 }
1993
1994 static void rdev_free(struct kobject *ko)
1995 {
1996         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1997         kfree(rdev);
1998 }
1999 static struct sysfs_ops rdev_sysfs_ops = {
2000         .show           = rdev_attr_show,
2001         .store          = rdev_attr_store,
2002 };
2003 static struct kobj_type rdev_ktype = {
2004         .release        = rdev_free,
2005         .sysfs_ops      = &rdev_sysfs_ops,
2006         .default_attrs  = rdev_default_attrs,
2007 };
2008
2009 /*
2010  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2011  *
2012  * mark the device faulty if:
2013  *
2014  *   - the device is nonexistent (zero size)
2015  *   - the device has no valid superblock
2016  *
2017  * a faulty rdev _never_ has rdev->sb set.
2018  */
2019 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2020 {
2021         char b[BDEVNAME_SIZE];
2022         int err;
2023         mdk_rdev_t *rdev;
2024         sector_t size;
2025
2026         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2027         if (!rdev) {
2028                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2029                 return ERR_PTR(-ENOMEM);
2030         }
2031
2032         if ((err = alloc_disk_sb(rdev)))
2033                 goto abort_free;
2034
2035         err = lock_rdev(rdev, newdev);
2036         if (err)
2037                 goto abort_free;
2038
2039         rdev->kobj.parent = NULL;
2040         rdev->kobj.ktype = &rdev_ktype;
2041         kobject_init(&rdev->kobj);
2042
2043         rdev->desc_nr = -1;
2044         rdev->saved_raid_disk = -1;
2045         rdev->raid_disk = -1;
2046         rdev->flags = 0;
2047         rdev->data_offset = 0;
2048         rdev->sb_events = 0;
2049         atomic_set(&rdev->nr_pending, 0);
2050         atomic_set(&rdev->read_errors, 0);
2051         atomic_set(&rdev->corrected_errors, 0);
2052
2053         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2054         if (!size) {
2055                 printk(KERN_WARNING 
2056                         "md: %s has zero or unknown size, marking faulty!\n",
2057                         bdevname(rdev->bdev,b));
2058                 err = -EINVAL;
2059                 goto abort_free;
2060         }
2061
2062         if (super_format >= 0) {
2063                 err = super_types[super_format].
2064                         load_super(rdev, NULL, super_minor);
2065                 if (err == -EINVAL) {
2066                         printk(KERN_WARNING
2067                                 "md: %s does not have a valid v%d.%d "
2068                                "superblock, not importing!\n",
2069                                 bdevname(rdev->bdev,b),
2070                                super_format, super_minor);
2071                         goto abort_free;
2072                 }
2073                 if (err < 0) {
2074                         printk(KERN_WARNING 
2075                                 "md: could not read %s's sb, not importing!\n",
2076                                 bdevname(rdev->bdev,b));
2077                         goto abort_free;
2078                 }
2079         }
2080         INIT_LIST_HEAD(&rdev->same_set);
2081
2082         return rdev;
2083
2084 abort_free:
2085         if (rdev->sb_page) {
2086                 if (rdev->bdev)
2087                         unlock_rdev(rdev);
2088                 free_disk_sb(rdev);
2089         }
2090         kfree(rdev);
2091         return ERR_PTR(err);
2092 }
2093
2094 /*
2095  * Check a full RAID array for plausibility
2096  */
2097
2098
2099 static void analyze_sbs(mddev_t * mddev)
2100 {
2101         int i;
2102         struct list_head *tmp;
2103         mdk_rdev_t *rdev, *freshest;
2104         char b[BDEVNAME_SIZE];
2105
2106         freshest = NULL;
2107         ITERATE_RDEV(mddev,rdev,tmp)
2108                 switch (super_types[mddev->major_version].
2109                         load_super(rdev, freshest, mddev->minor_version)) {
2110                 case 1:
2111                         freshest = rdev;
2112                         break;
2113                 case 0:
2114                         break;
2115                 default:
2116                         printk( KERN_ERR \
2117                                 "md: fatal superblock inconsistency in %s"
2118                                 " -- removing from array\n", 
2119                                 bdevname(rdev->bdev,b));
2120                         kick_rdev_from_array(rdev);
2121                 }
2122
2123
2124         super_types[mddev->major_version].
2125                 validate_super(mddev, freshest);
2126
2127         i = 0;
2128         ITERATE_RDEV(mddev,rdev,tmp) {
2129                 if (rdev != freshest)
2130                         if (super_types[mddev->major_version].
2131                             validate_super(mddev, rdev)) {
2132                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2133                                         " from array!\n",
2134                                         bdevname(rdev->bdev,b));
2135                                 kick_rdev_from_array(rdev);
2136                                 continue;
2137                         }
2138                 if (mddev->level == LEVEL_MULTIPATH) {
2139                         rdev->desc_nr = i++;
2140                         rdev->raid_disk = rdev->desc_nr;
2141                         set_bit(In_sync, &rdev->flags);
2142                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2143                         rdev->raid_disk = -1;
2144                         clear_bit(In_sync, &rdev->flags);
2145                 }
2146         }
2147
2148
2149
2150         if (mddev->recovery_cp != MaxSector &&
2151             mddev->level >= 1)
2152                 printk(KERN_ERR "md: %s: raid array is not clean"
2153                        " -- starting background reconstruction\n",
2154                        mdname(mddev));
2155
2156 }
2157
2158 static ssize_t
2159 safe_delay_show(mddev_t *mddev, char *page)
2160 {
2161         int msec = (mddev->safemode_delay*1000)/HZ;
2162         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2163 }
2164 static ssize_t
2165 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2166 {
2167         int scale=1;
2168         int dot=0;
2169         int i;
2170         unsigned long msec;
2171         char buf[30];
2172         char *e;
2173         /* remove a period, and count digits after it */
2174         if (len >= sizeof(buf))
2175                 return -EINVAL;
2176         strlcpy(buf, cbuf, len);
2177         buf[len] = 0;
2178         for (i=0; i<len; i++) {
2179                 if (dot) {
2180                         if (isdigit(buf[i])) {
2181                                 buf[i-1] = buf[i];
2182                                 scale *= 10;
2183                         }
2184                         buf[i] = 0;
2185                 } else if (buf[i] == '.') {
2186                         dot=1;
2187                         buf[i] = 0;
2188                 }
2189         }
2190         msec = simple_strtoul(buf, &e, 10);
2191         if (e == buf || (*e && *e != '\n'))
2192                 return -EINVAL;
2193         msec = (msec * 1000) / scale;
2194         if (msec == 0)
2195                 mddev->safemode_delay = 0;
2196         else {
2197                 mddev->safemode_delay = (msec*HZ)/1000;
2198                 if (mddev->safemode_delay == 0)
2199                         mddev->safemode_delay = 1;
2200         }
2201         return len;
2202 }
2203 static struct md_sysfs_entry md_safe_delay =
2204 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2205
2206 static ssize_t
2207 level_show(mddev_t *mddev, char *page)
2208 {
2209         struct mdk_personality *p = mddev->pers;
2210         if (p)
2211                 return sprintf(page, "%s\n", p->name);
2212         else if (mddev->clevel[0])
2213                 return sprintf(page, "%s\n", mddev->clevel);
2214         else if (mddev->level != LEVEL_NONE)
2215                 return sprintf(page, "%d\n", mddev->level);
2216         else
2217                 return 0;
2218 }
2219
2220 static ssize_t
2221 level_store(mddev_t *mddev, const char *buf, size_t len)
2222 {
2223         int rv = len;
2224         if (mddev->pers)
2225                 return -EBUSY;
2226         if (len == 0)
2227                 return 0;
2228         if (len >= sizeof(mddev->clevel))
2229                 return -ENOSPC;
2230         strncpy(mddev->clevel, buf, len);
2231         if (mddev->clevel[len-1] == '\n')
2232                 len--;
2233         mddev->clevel[len] = 0;
2234         mddev->level = LEVEL_NONE;
2235         return rv;
2236 }
2237
2238 static struct md_sysfs_entry md_level =
2239 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2240
2241
2242 static ssize_t
2243 layout_show(mddev_t *mddev, char *page)
2244 {
2245         /* just a number, not meaningful for all levels */
2246         if (mddev->reshape_position != MaxSector &&
2247             mddev->layout != mddev->new_layout)
2248                 return sprintf(page, "%d (%d)\n",
2249                                mddev->new_layout, mddev->layout);
2250         return sprintf(page, "%d\n", mddev->layout);
2251 }
2252
2253 static ssize_t
2254 layout_store(mddev_t *mddev, const char *buf, size_t len)
2255 {
2256         char *e;
2257         unsigned long n = simple_strtoul(buf, &e, 10);
2258
2259         if (!*buf || (*e && *e != '\n'))
2260                 return -EINVAL;
2261
2262         if (mddev->pers)
2263                 return -EBUSY;
2264         if (mddev->reshape_position != MaxSector)
2265                 mddev->new_layout = n;
2266         else
2267                 mddev->layout = n;
2268         return len;
2269 }
2270 static struct md_sysfs_entry md_layout =
2271 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2272
2273
2274 static ssize_t
2275 raid_disks_show(mddev_t *mddev, char *page)
2276 {
2277         if (mddev->raid_disks == 0)
2278                 return 0;
2279         if (mddev->reshape_position != MaxSector &&
2280             mddev->delta_disks != 0)
2281                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2282                                mddev->raid_disks - mddev->delta_disks);
2283         return sprintf(page, "%d\n", mddev->raid_disks);
2284 }
2285
2286 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2287
2288 static ssize_t
2289 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2290 {
2291         char *e;
2292         int rv = 0;
2293         unsigned long n = simple_strtoul(buf, &e, 10);
2294
2295         if (!*buf || (*e && *e != '\n'))
2296                 return -EINVAL;
2297
2298         if (mddev->pers)
2299                 rv = update_raid_disks(mddev, n);
2300         else if (mddev->reshape_position != MaxSector) {
2301                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2302                 mddev->delta_disks = n - olddisks;
2303                 mddev->raid_disks = n;
2304         } else
2305                 mddev->raid_disks = n;
2306         return rv ? rv : len;
2307 }
2308 static struct md_sysfs_entry md_raid_disks =
2309 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2310
2311 static ssize_t
2312 chunk_size_show(mddev_t *mddev, char *page)
2313 {
2314         if (mddev->reshape_position != MaxSector &&
2315             mddev->chunk_size != mddev->new_chunk)
2316                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2317                                mddev->chunk_size);
2318         return sprintf(page, "%d\n", mddev->chunk_size);
2319 }
2320
2321 static ssize_t
2322 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2323 {
2324         /* can only set chunk_size if array is not yet active */
2325         char *e;
2326         unsigned long n = simple_strtoul(buf, &e, 10);
2327
2328         if (!*buf || (*e && *e != '\n'))
2329                 return -EINVAL;
2330
2331         if (mddev->pers)
2332                 return -EBUSY;
2333         else if (mddev->reshape_position != MaxSector)
2334                 mddev->new_chunk = n;
2335         else
2336                 mddev->chunk_size = n;
2337         return len;
2338 }
2339 static struct md_sysfs_entry md_chunk_size =
2340 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2341
2342 static ssize_t
2343 resync_start_show(mddev_t *mddev, char *page)
2344 {
2345         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2346 }
2347
2348 static ssize_t
2349 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2350 {
2351         /* can only set chunk_size if array is not yet active */
2352         char *e;
2353         unsigned long long n = simple_strtoull(buf, &e, 10);
2354
2355         if (mddev->pers)
2356                 return -EBUSY;
2357         if (!*buf || (*e && *e != '\n'))
2358                 return -EINVAL;
2359
2360         mddev->recovery_cp = n;
2361         return len;
2362 }
2363 static struct md_sysfs_entry md_resync_start =
2364 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2365
2366 /*
2367  * The array state can be:
2368  *
2369  * clear
2370  *     No devices, no size, no level
2371  *     Equivalent to STOP_ARRAY ioctl
2372  * inactive
2373  *     May have some settings, but array is not active
2374  *        all IO results in error
2375  *     When written, doesn't tear down array, but just stops it
2376  * suspended (not supported yet)
2377  *     All IO requests will block. The array can be reconfigured.
2378  *     Writing this, if accepted, will block until array is quiessent
2379  * readonly
2380  *     no resync can happen.  no superblocks get written.
2381  *     write requests fail
2382  * read-auto
2383  *     like readonly, but behaves like 'clean' on a write request.
2384  *
2385  * clean - no pending writes, but otherwise active.
2386  *     When written to inactive array, starts without resync
2387  *     If a write request arrives then
2388  *       if metadata is known, mark 'dirty' and switch to 'active'.
2389  *       if not known, block and switch to write-pending
2390  *     If written to an active array that has pending writes, then fails.
2391  * active
2392  *     fully active: IO and resync can be happening.
2393  *     When written to inactive array, starts with resync
2394  *
2395  * write-pending
2396  *     clean, but writes are blocked waiting for 'active' to be written.
2397  *
2398  * active-idle
2399  *     like active, but no writes have been seen for a while (100msec).
2400  *
2401  */
2402 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2403                    write_pending, active_idle, bad_word};
2404 static char *array_states[] = {
2405         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2406         "write-pending", "active-idle", NULL };
2407
2408 static int match_word(const char *word, char **list)
2409 {
2410         int n;
2411         for (n=0; list[n]; n++)
2412                 if (cmd_match(word, list[n]))
2413                         break;
2414         return n;
2415 }
2416
2417 static ssize_t
2418 array_state_show(mddev_t *mddev, char *page)
2419 {
2420         enum array_state st = inactive;
2421
2422         if (mddev->pers)
2423                 switch(mddev->ro) {
2424                 case 1:
2425                         st = readonly;
2426                         break;
2427                 case 2:
2428                         st = read_auto;
2429                         break;
2430                 case 0:
2431                         if (mddev->in_sync)
2432                                 st = clean;
2433                         else if (mddev->safemode)
2434                                 st = active_idle;
2435                         else
2436                                 st = active;
2437                 }
2438         else {
2439                 if (list_empty(&mddev->disks) &&
2440                     mddev->raid_disks == 0 &&
2441                     mddev->size == 0)
2442                         st = clear;
2443                 else
2444                         st = inactive;
2445         }
2446         return sprintf(page, "%s\n", array_states[st]);
2447 }
2448
2449 static int do_md_stop(mddev_t * mddev, int ro);
2450 static int do_md_run(mddev_t * mddev);
2451 static int restart_array(mddev_t *mddev);
2452
2453 static ssize_t
2454 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2455 {
2456         int err = -EINVAL;
2457         enum array_state st = match_word(buf, array_states);
2458         switch(st) {
2459         case bad_word:
2460                 break;
2461         case clear:
2462                 /* stopping an active array */
2463                 if (mddev->pers) {
2464                         if (atomic_read(&mddev->active) > 1)
2465                                 return -EBUSY;
2466                         err = do_md_stop(mddev, 0);
2467                 }
2468                 break;
2469         case inactive:
2470                 /* stopping an active array */
2471                 if (mddev->pers) {
2472                         if (atomic_read(&mddev->active) > 1)
2473                                 return -EBUSY;
2474                         err = do_md_stop(mddev, 2);
2475                 }
2476                 break;
2477         case suspended:
2478                 break; /* not supported yet */
2479         case readonly:
2480                 if (mddev->pers)
2481                         err = do_md_stop(mddev, 1);
2482                 else {
2483                         mddev->ro = 1;
2484                         err = do_md_run(mddev);
2485                 }
2486                 break;
2487         case read_auto:
2488                 /* stopping an active array */
2489                 if (mddev->pers) {
2490                         err = do_md_stop(mddev, 1);
2491                         if (err == 0)
2492                                 mddev->ro = 2; /* FIXME mark devices writable */
2493                 } else {
2494                         mddev->ro = 2;
2495                         err = do_md_run(mddev);
2496                 }
2497                 break;
2498         case clean:
2499                 if (mddev->pers) {
2500                         restart_array(mddev);
2501                         spin_lock_irq(&mddev->write_lock);
2502                         if (atomic_read(&mddev->writes_pending) == 0) {
2503                                 mddev->in_sync = 1;
2504                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2505                         }
2506                         spin_unlock_irq(&mddev->write_lock);
2507                 } else {
2508                         mddev->ro = 0;
2509                         mddev->recovery_cp = MaxSector;
2510                         err = do_md_run(mddev);
2511                 }
2512                 break;
2513         case active:
2514                 if (mddev->pers) {
2515                         restart_array(mddev);
2516                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2517                         wake_up(&mddev->sb_wait);
2518                         err = 0;
2519                 } else {
2520                         mddev->ro = 0;
2521                         err = do_md_run(mddev);
2522                 }
2523                 break;
2524         case write_pending:
2525         case active_idle:
2526                 /* these cannot be set */
2527                 break;
2528         }
2529         if (err)
2530                 return err;
2531         else
2532                 return len;
2533 }
2534 static struct md_sysfs_entry md_array_state =
2535 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2536
2537 static ssize_t
2538 null_show(mddev_t *mddev, char *page)
2539 {
2540         return -EINVAL;
2541 }
2542
2543 static ssize_t
2544 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2545 {
2546         /* buf must be %d:%d\n? giving major and minor numbers */
2547         /* The new device is added to the array.
2548          * If the array has a persistent superblock, we read the
2549          * superblock to initialise info and check validity.
2550          * Otherwise, only checking done is that in bind_rdev_to_array,
2551          * which mainly checks size.
2552          */
2553         char *e;
2554         int major = simple_strtoul(buf, &e, 10);
2555         int minor;
2556         dev_t dev;
2557         mdk_rdev_t *rdev;
2558         int err;
2559
2560         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2561                 return -EINVAL;
2562         minor = simple_strtoul(e+1, &e, 10);
2563         if (*e && *e != '\n')
2564                 return -EINVAL;
2565         dev = MKDEV(major, minor);
2566         if (major != MAJOR(dev) ||
2567             minor != MINOR(dev))
2568                 return -EOVERFLOW;
2569
2570
2571         if (mddev->persistent) {
2572                 rdev = md_import_device(dev, mddev->major_version,
2573                                         mddev->minor_version);
2574                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2575                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2576                                                        mdk_rdev_t, same_set);
2577                         err = super_types[mddev->major_version]
2578                                 .load_super(rdev, rdev0, mddev->minor_version);
2579                         if (err < 0)
2580                                 goto out;
2581                 }
2582         } else
2583                 rdev = md_import_device(dev, -1, -1);
2584
2585         if (IS_ERR(rdev))
2586                 return PTR_ERR(rdev);
2587         err = bind_rdev_to_array(rdev, mddev);
2588  out:
2589         if (err)
2590                 export_rdev(rdev);
2591         return err ? err : len;
2592 }
2593
2594 static struct md_sysfs_entry md_new_device =
2595 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2596
2597 static ssize_t
2598 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2599 {
2600         char *end;
2601         unsigned long chunk, end_chunk;
2602
2603         if (!mddev->bitmap)
2604                 goto out;
2605         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2606         while (*buf) {
2607                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2608                 if (buf == end) break;
2609                 if (*end == '-') { /* range */
2610                         buf = end + 1;
2611                         end_chunk = simple_strtoul(buf, &end, 0);
2612                         if (buf == end) break;
2613                 }
2614                 if (*end && !isspace(*end)) break;
2615                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2616                 buf = end;
2617                 while (isspace(*buf)) buf++;
2618         }
2619         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2620 out:
2621         return len;
2622 }
2623
2624 static struct md_sysfs_entry md_bitmap =
2625 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2626
2627 static ssize_t
2628 size_show(mddev_t *mddev, char *page)
2629 {
2630         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2631 }
2632
2633 static int update_size(mddev_t *mddev, unsigned long size);
2634
2635 static ssize_t
2636 size_store(mddev_t *mddev, const char *buf, size_t len)
2637 {
2638         /* If array is inactive, we can reduce the component size, but
2639          * not increase it (except from 0).
2640          * If array is active, we can try an on-line resize
2641          */
2642         char *e;
2643         int err = 0;
2644         unsigned long long size = simple_strtoull(buf, &e, 10);
2645         if (!*buf || *buf == '\n' ||
2646             (*e && *e != '\n'))
2647                 return -EINVAL;
2648
2649         if (mddev->pers) {
2650                 err = update_size(mddev, size);
2651                 md_update_sb(mddev, 1);
2652         } else {
2653                 if (mddev->size == 0 ||
2654                     mddev->size > size)
2655                         mddev->size = size;
2656                 else
2657                         err = -ENOSPC;
2658         }
2659         return err ? err : len;
2660 }
2661
2662 static struct md_sysfs_entry md_size =
2663 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2664
2665
2666 /* Metdata version.
2667  * This is either 'none' for arrays with externally managed metadata,
2668  * or N.M for internally known formats
2669  */
2670 static ssize_t
2671 metadata_show(mddev_t *mddev, char *page)
2672 {
2673         if (mddev->persistent)
2674                 return sprintf(page, "%d.%d\n",
2675                                mddev->major_version, mddev->minor_version);
2676         else
2677                 return sprintf(page, "none\n");
2678 }
2679
2680 static ssize_t
2681 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2682 {
2683         int major, minor;
2684         char *e;
2685         if (!list_empty(&mddev->disks))
2686                 return -EBUSY;
2687
2688         if (cmd_match(buf, "none")) {
2689                 mddev->persistent = 0;
2690                 mddev->major_version = 0;
2691                 mddev->minor_version = 90;
2692                 return len;
2693         }
2694         major = simple_strtoul(buf, &e, 10);
2695         if (e==buf || *e != '.')
2696                 return -EINVAL;
2697         buf = e+1;
2698         minor = simple_strtoul(buf, &e, 10);
2699         if (e==buf || (*e && *e != '\n') )
2700                 return -EINVAL;
2701         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2702                 return -ENOENT;
2703         mddev->major_version = major;
2704         mddev->minor_version = minor;
2705         mddev->persistent = 1;
2706         return len;
2707 }
2708
2709 static struct md_sysfs_entry md_metadata =
2710 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2711
2712 static ssize_t
2713 action_show(mddev_t *mddev, char *page)
2714 {
2715         char *type = "idle";
2716         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2717             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2718                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2719                         type = "reshape";
2720                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2721                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2722                                 type = "resync";
2723                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2724                                 type = "check";
2725                         else
2726                                 type = "repair";
2727                 } else
2728                         type = "recover";
2729         }
2730         return sprintf(page, "%s\n", type);
2731 }
2732
2733 static ssize_t
2734 action_store(mddev_t *mddev, const char *page, size_t len)
2735 {
2736         if (!mddev->pers || !mddev->pers->sync_request)
2737                 return -EINVAL;
2738
2739         if (cmd_match(page, "idle")) {
2740                 if (mddev->sync_thread) {
2741                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2742                         md_unregister_thread(mddev->sync_thread);
2743                         mddev->sync_thread = NULL;
2744                         mddev->recovery = 0;
2745                 }
2746         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2747                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2748                 return -EBUSY;
2749         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2750                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2751         else if (cmd_match(page, "reshape")) {
2752                 int err;
2753                 if (mddev->pers->start_reshape == NULL)
2754                         return -EINVAL;
2755                 err = mddev->pers->start_reshape(mddev);
2756                 if (err)
2757                         return err;
2758         } else {
2759                 if (cmd_match(page, "check"))
2760                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2761                 else if (!cmd_match(page, "repair"))
2762                         return -EINVAL;
2763                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2764                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2765         }
2766         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2767         md_wakeup_thread(mddev->thread);
2768         return len;
2769 }
2770
2771 static ssize_t
2772 mismatch_cnt_show(mddev_t *mddev, char *page)
2773 {
2774         return sprintf(page, "%llu\n",
2775                        (unsigned long long) mddev->resync_mismatches);
2776 }
2777
2778 static struct md_sysfs_entry md_scan_mode =
2779 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2780
2781
2782 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2783
2784 static ssize_t
2785 sync_min_show(mddev_t *mddev, char *page)
2786 {
2787         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2788                        mddev->sync_speed_min ? "local": "system");
2789 }
2790
2791 static ssize_t
2792 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2793 {
2794         int min;
2795         char *e;
2796         if (strncmp(buf, "system", 6)==0) {
2797                 mddev->sync_speed_min = 0;
2798                 return len;
2799         }
2800         min = simple_strtoul(buf, &e, 10);
2801         if (buf == e || (*e && *e != '\n') || min <= 0)
2802                 return -EINVAL;
2803         mddev->sync_speed_min = min;
2804         return len;
2805 }
2806
2807 static struct md_sysfs_entry md_sync_min =
2808 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2809
2810 static ssize_t
2811 sync_max_show(mddev_t *mddev, char *page)
2812 {
2813         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2814                        mddev->sync_speed_max ? "local": "system");
2815 }
2816
2817 static ssize_t
2818 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2819 {
2820         int max;
2821         char *e;
2822         if (strncmp(buf, "system", 6)==0) {
2823                 mddev->sync_speed_max = 0;
2824                 return len;
2825         }
2826         max = simple_strtoul(buf, &e, 10);
2827         if (buf == e || (*e && *e != '\n') || max <= 0)
2828                 return -EINVAL;
2829         mddev->sync_speed_max = max;
2830         return len;
2831 }
2832
2833 static struct md_sysfs_entry md_sync_max =
2834 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2835
2836 static ssize_t
2837 degraded_show(mddev_t *mddev, char *page)
2838 {
2839         return sprintf(page, "%d\n", mddev->degraded);
2840 }
2841 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2842
2843 static ssize_t
2844 sync_speed_show(mddev_t *mddev, char *page)
2845 {
2846         unsigned long resync, dt, db;
2847         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2848         dt = ((jiffies - mddev->resync_mark) / HZ);
2849         if (!dt) dt++;
2850         db = resync - (mddev->resync_mark_cnt);
2851         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2852 }
2853
2854 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2855
2856 static ssize_t
2857 sync_completed_show(mddev_t *mddev, char *page)
2858 {
2859         unsigned long max_blocks, resync;
2860
2861         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2862                 max_blocks = mddev->resync_max_sectors;
2863         else
2864                 max_blocks = mddev->size << 1;
2865
2866         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2867         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2868 }
2869
2870 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2871
2872 static ssize_t
2873 suspend_lo_show(mddev_t *mddev, char *page)
2874 {
2875         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2876 }
2877
2878 static ssize_t
2879 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2880 {
2881         char *e;
2882         unsigned long long new = simple_strtoull(buf, &e, 10);
2883
2884         if (mddev->pers->quiesce == NULL)
2885                 return -EINVAL;
2886         if (buf == e || (*e && *e != '\n'))
2887                 return -EINVAL;
2888         if (new >= mddev->suspend_hi ||
2889             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2890                 mddev->suspend_lo = new;
2891                 mddev->pers->quiesce(mddev, 2);
2892                 return len;
2893         } else
2894                 return -EINVAL;
2895 }
2896 static struct md_sysfs_entry md_suspend_lo =
2897 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2898
2899
2900 static ssize_t
2901 suspend_hi_show(mddev_t *mddev, char *page)
2902 {
2903         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2904 }
2905
2906 static ssize_t
2907 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2908 {
2909         char *e;
2910         unsigned long long new = simple_strtoull(buf, &e, 10);
2911
2912         if (mddev->pers->quiesce == NULL)
2913                 return -EINVAL;
2914         if (buf == e || (*e && *e != '\n'))
2915                 return -EINVAL;
2916         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2917             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2918                 mddev->suspend_hi = new;
2919                 mddev->pers->quiesce(mddev, 1);
2920                 mddev->pers->quiesce(mddev, 0);
2921                 return len;
2922         } else
2923                 return -EINVAL;
2924 }
2925 static struct md_sysfs_entry md_suspend_hi =
2926 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2927
2928 static ssize_t
2929 reshape_position_show(mddev_t *mddev, char *page)
2930 {
2931         if (mddev->reshape_position != MaxSector)
2932                 return sprintf(page, "%llu\n",
2933                                (unsigned long long)mddev->reshape_position);
2934         strcpy(page, "none\n");
2935         return 5;
2936 }
2937
2938 static ssize_t
2939 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2940 {
2941         char *e;
2942         unsigned long long new = simple_strtoull(buf, &e, 10);
2943         if (mddev->pers)
2944                 return -EBUSY;
2945         if (buf == e || (*e && *e != '\n'))
2946                 return -EINVAL;
2947         mddev->reshape_position = new;
2948         mddev->delta_disks = 0;
2949         mddev->new_level = mddev->level;
2950         mddev->new_layout = mddev->layout;
2951         mddev->new_chunk = mddev->chunk_size;
2952         return len;
2953 }
2954
2955 static struct md_sysfs_entry md_reshape_position =
2956 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2957        reshape_position_store);
2958
2959
2960 static struct attribute *md_default_attrs[] = {
2961         &md_level.attr,
2962         &md_layout.attr,
2963         &md_raid_disks.attr,
2964         &md_chunk_size.attr,
2965         &md_size.attr,
2966         &md_resync_start.attr,
2967         &md_metadata.attr,
2968         &md_new_device.attr,
2969         &md_safe_delay.attr,
2970         &md_array_state.attr,
2971         &md_reshape_position.attr,
2972         NULL,
2973 };
2974
2975 static struct attribute *md_redundancy_attrs[] = {
2976         &md_scan_mode.attr,
2977         &md_mismatches.attr,
2978         &md_sync_min.attr,
2979         &md_sync_max.attr,
2980         &md_sync_speed.attr,
2981         &md_sync_completed.attr,
2982         &md_suspend_lo.attr,
2983         &md_suspend_hi.attr,
2984         &md_bitmap.attr,
2985         &md_degraded.attr,
2986         NULL,
2987 };
2988 static struct attribute_group md_redundancy_group = {
2989         .name = NULL,
2990         .attrs = md_redundancy_attrs,
2991 };
2992
2993
2994 static ssize_t
2995 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2996 {
2997         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2998         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2999         ssize_t rv;
3000
3001         if (!entry->show)
3002                 return -EIO;
3003         rv = mddev_lock(mddev);
3004         if (!rv) {
3005                 rv = entry->show(mddev, page);
3006                 mddev_unlock(mddev);
3007         }
3008         return rv;
3009 }
3010
3011 static ssize_t
3012 md_attr_store(struct kobject *kobj, struct attribute *attr,
3013               const char *page, size_t length)
3014 {
3015         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3016         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3017         ssize_t rv;
3018
3019         if (!entry->store)
3020                 return -EIO;
3021         if (!capable(CAP_SYS_ADMIN))
3022                 return -EACCES;
3023         rv = mddev_lock(mddev);
3024         if (!rv) {
3025                 rv = entry->store(mddev, page, length);
3026                 mddev_unlock(mddev);
3027         }
3028         return rv;
3029 }
3030
3031 static void md_free(struct kobject *ko)
3032 {
3033         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3034         kfree(mddev);
3035 }
3036
3037 static struct sysfs_ops md_sysfs_ops = {
3038         .show   = md_attr_show,
3039         .store  = md_attr_store,
3040 };
3041 static struct kobj_type md_ktype = {
3042         .release        = md_free,
3043         .sysfs_ops      = &md_sysfs_ops,
3044         .default_attrs  = md_default_attrs,
3045 };
3046
3047 int mdp_major = 0;
3048
3049 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3050 {
3051         static DEFINE_MUTEX(disks_mutex);
3052         mddev_t *mddev = mddev_find(dev);
3053         struct gendisk *disk;
3054         int partitioned = (MAJOR(dev) != MD_MAJOR);
3055         int shift = partitioned ? MdpMinorShift : 0;
3056         int unit = MINOR(dev) >> shift;
3057         int error;
3058
3059         if (!mddev)
3060                 return NULL;
3061
3062         mutex_lock(&disks_mutex);
3063         if (mddev->gendisk) {
3064                 mutex_unlock(&disks_mutex);
3065                 mddev_put(mddev);
3066                 return NULL;
3067         }
3068         disk = alloc_disk(1 << shift);
3069         if (!disk) {
3070                 mutex_unlock(&disks_mutex);
3071                 mddev_put(mddev);
3072                 return NULL;
3073         }
3074         disk->major = MAJOR(dev);
3075         disk->first_minor = unit << shift;
3076         if (partitioned)
3077                 sprintf(disk->disk_name, "md_d%d", unit);
3078         else
3079                 sprintf(disk->disk_name, "md%d", unit);
3080         disk->fops = &md_fops;
3081         disk->private_data = mddev;
3082         disk->queue = mddev->queue;
3083         add_disk(disk);
3084         mddev->gendisk = disk;
3085         mutex_unlock(&disks_mutex);
3086         error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->kobj,
3087                                      "%s", "md");
3088         if (error)
3089                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3090                        disk->disk_name);
3091         else
3092                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3093         return NULL;
3094 }
3095
3096 static void md_safemode_timeout(unsigned long data)
3097 {
3098         mddev_t *mddev = (mddev_t *) data;
3099
3100         mddev->safemode = 1;
3101         md_wakeup_thread(mddev->thread);
3102 }
3103
3104 static int start_dirty_degraded;
3105
3106 static int do_md_run(mddev_t * mddev)
3107 {
3108         int err;
3109         int chunk_size;
3110         struct list_head *tmp;
3111         mdk_rdev_t *rdev;
3112         struct gendisk *disk;
3113         struct mdk_personality *pers;
3114         char b[BDEVNAME_SIZE];
3115
3116         if (list_empty(&mddev->disks))
3117                 /* cannot run an array with no devices.. */
3118                 return -EINVAL;
3119
3120         if (mddev->pers)
3121                 return -EBUSY;
3122
3123         /*
3124          * Analyze all RAID superblock(s)
3125          */
3126         if (!mddev->raid_disks)
3127                 analyze_sbs(mddev);
3128
3129         chunk_size = mddev->chunk_size;
3130
3131         if (chunk_size) {
3132                 if (chunk_size > MAX_CHUNK_SIZE) {
3133                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3134                                 chunk_size, MAX_CHUNK_SIZE);
3135                         return -EINVAL;
3136                 }
3137                 /*
3138                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3139                  */
3140                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3141                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3142                         return -EINVAL;
3143                 }
3144                 if (chunk_size < PAGE_SIZE) {
3145                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3146                                 chunk_size, PAGE_SIZE);
3147                         return -EINVAL;
3148                 }
3149
3150                 /* devices must have minimum size of one chunk */
3151                 ITERATE_RDEV(mddev,rdev,tmp) {
3152                         if (test_bit(Faulty, &rdev->flags))
3153                                 continue;
3154                         if (rdev->size < chunk_size / 1024) {
3155                                 printk(KERN_WARNING
3156                                         "md: Dev %s smaller than chunk_size:"
3157                                         " %lluk < %dk\n",
3158                                         bdevname(rdev->bdev,b),
3159                                         (unsigned long long)rdev->size,
3160                                         chunk_size / 1024);
3161                                 return -EINVAL;
3162                         }
3163                 }
3164         }
3165
3166 #ifdef CONFIG_KMOD
3167         if (mddev->level != LEVEL_NONE)
3168                 request_module("md-level-%d", mddev->level);
3169         else if (mddev->clevel[0])
3170                 request_module("md-%s", mddev->clevel);
3171 #endif
3172
3173         /*
3174          * Drop all container device buffers, from now on
3175          * the only valid external interface is through the md
3176          * device.
3177          */
3178         ITERATE_RDEV(mddev,rdev,tmp) {
3179                 if (test_bit(Faulty, &rdev->flags))
3180                         continue;
3181                 sync_blockdev(rdev->bdev);
3182                 invalidate_bdev(rdev->bdev);
3183
3184                 /* perform some consistency tests on the device.
3185                  * We don't want the data to overlap the metadata,
3186                  * Internal Bitmap issues has handled elsewhere.
3187                  */
3188                 if (rdev->data_offset < rdev->sb_offset) {
3189                         if (mddev->size &&
3190                             rdev->data_offset + mddev->size*2
3191                             > rdev->sb_offset*2) {
3192                                 printk("md: %s: data overlaps metadata\n",
3193                                        mdname(mddev));
3194                                 return -EINVAL;
3195                         }
3196                 } else {
3197                         if (rdev->sb_offset*2 + rdev->sb_size/512
3198                             > rdev->data_offset) {
3199                                 printk("md: %s: metadata overlaps data\n",
3200                                        mdname(mddev));
3201                                 return -EINVAL;
3202                         }
3203                 }
3204         }
3205
3206         md_probe(mddev->unit, NULL, NULL);
3207         disk = mddev->gendisk;
3208         if (!disk)
3209                 return -ENOMEM;
3210
3211         spin_lock(&pers_lock);
3212         pers = find_pers(mddev->level, mddev->clevel);
3213         if (!pers || !try_module_get(pers->owner)) {
3214                 spin_unlock(&pers_lock);
3215                 if (mddev->level != LEVEL_NONE)
3216                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3217                                mddev->level);
3218                 else
3219                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3220                                mddev->clevel);
3221                 return -EINVAL;
3222         }
3223         mddev->pers = pers;
3224         spin_unlock(&pers_lock);
3225         mddev->level = pers->level;
3226         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3227
3228         if (mddev->reshape_position != MaxSector &&
3229             pers->start_reshape == NULL) {
3230                 /* This personality cannot handle reshaping... */
3231                 mddev->pers = NULL;
3232                 module_put(pers->owner);
3233                 return -EINVAL;
3234         }
3235
3236         if (pers->sync_request) {
3237                 /* Warn if this is a potentially silly
3238                  * configuration.
3239                  */
3240                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3241                 mdk_rdev_t *rdev2;
3242                 struct list_head *tmp2;
3243                 int warned = 0;
3244                 ITERATE_RDEV(mddev, rdev, tmp) {
3245                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3246                                 if (rdev < rdev2 &&
3247                                     rdev->bdev->bd_contains ==
3248                                     rdev2->bdev->bd_contains) {
3249                                         printk(KERN_WARNING
3250                                                "%s: WARNING: %s appears to be"
3251                                                " on the same physical disk as"
3252                                                " %s.\n",
3253                                                mdname(mddev),
3254                                                bdevname(rdev->bdev,b),
3255                                                bdevname(rdev2->bdev,b2));
3256                                         warned = 1;
3257                                 }
3258                         }
3259                 }
3260                 if (warned)
3261                         printk(KERN_WARNING
3262                                "True protection against single-disk"
3263                                " failure might be compromised.\n");
3264         }
3265
3266         mddev->recovery = 0;
3267         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3268         mddev->barriers_work = 1;
3269         mddev->ok_start_degraded = start_dirty_degraded;
3270
3271         if (start_readonly)
3272                 mddev->ro = 2; /* read-only, but switch on first write */
3273
3274         err = mddev->pers->run(mddev);
3275         if (!err && mddev->pers->sync_request) {
3276                 err = bitmap_create(mddev);
3277                 if (err) {
3278                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3279                                mdname(mddev), err);
3280                         mddev->pers->stop(mddev);
3281                 }
3282         }
3283         if (err) {
3284                 printk(KERN_ERR "md: pers->run() failed ...\n");
3285                 module_put(mddev->pers->owner);
3286                 mddev->pers = NULL;
3287                 bitmap_destroy(mddev);
3288                 return err;
3289         }
3290         if (mddev->pers->sync_request) {
3291                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3292                         printk(KERN_WARNING
3293                                "md: cannot register extra attributes for %s\n",
3294                                mdname(mddev));
3295         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3296                 mddev->ro = 0;
3297
3298         atomic_set(&mddev->writes_pending,0);
3299         mddev->safemode = 0;
3300         mddev->safemode_timer.function = md_safemode_timeout;
3301         mddev->safemode_timer.data = (unsigned long) mddev;
3302         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3303         mddev->in_sync = 1;
3304
3305         ITERATE_RDEV(mddev,rdev,tmp)
3306                 if (rdev->raid_disk >= 0) {
3307                         char nm[20];
3308                         sprintf(nm, "rd%d", rdev->raid_disk);
3309                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3310                                 printk("md: cannot register %s for %s\n",
3311                                        nm, mdname(mddev));
3312                 }
3313         
3314         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3315         
3316         if (mddev->flags)
3317                 md_update_sb(mddev, 0);
3318
3319         set_capacity(disk, mddev->array_size<<1);
3320
3321         /* If we call blk_queue_make_request here, it will
3322          * re-initialise max_sectors etc which may have been
3323          * refined inside -> run.  So just set the bits we need to set.
3324          * Most initialisation happended when we called
3325          * blk_queue_make_request(..., md_fail_request)
3326          * earlier.
3327          */
3328         mddev->queue->queuedata = mddev;
3329         mddev->queue->make_request_fn = mddev->pers->make_request;
3330
3331         /* If there is a partially-recovered drive we need to
3332          * start recovery here.  If we leave it to md_check_recovery,
3333          * it will remove the drives and not do the right thing
3334          */
3335         if (mddev->degraded && !mddev->sync_thread) {
3336                 struct list_head *rtmp;
3337                 int spares = 0;
3338                 ITERATE_RDEV(mddev,rdev,rtmp)
3339                         if (rdev->raid_disk >= 0 &&
3340                             !test_bit(In_sync, &rdev->flags) &&
3341                             !test_bit(Faulty, &rdev->flags))
3342                                 /* complete an interrupted recovery */
3343                                 spares++;
3344                 if (spares && mddev->pers->sync_request) {
3345                         mddev->recovery = 0;
3346                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3347                         mddev->sync_thread = md_register_thread(md_do_sync,
3348                                                                 mddev,
3349                                                                 "%s_resync");
3350                         if (!mddev->sync_thread) {
3351                                 printk(KERN_ERR "%s: could not start resync"
3352                                        " thread...\n",
3353                                        mdname(mddev));
3354                                 /* leave the spares where they are, it shouldn't hurt */
3355                                 mddev->recovery = 0;
3356                         }
3357                 }
3358         }
3359         md_wakeup_thread(mddev->thread);
3360         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3361
3362         mddev->changed = 1;
3363         md_new_event(mddev);
3364         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3365         return 0;
3366 }
3367
3368 static int restart_array(mddev_t *mddev)
3369 {
3370         struct gendisk *disk = mddev->gendisk;
3371         int err;
3372
3373         /*
3374          * Complain if it has no devices
3375          */
3376         err = -ENXIO;
3377         if (list_empty(&mddev->disks))
3378                 goto out;
3379
3380         if (mddev->pers) {
3381                 err = -EBUSY;
3382                 if (!mddev->ro)
3383                         goto out;
3384
3385                 mddev->safemode = 0;
3386                 mddev->ro = 0;
3387                 set_disk_ro(disk, 0);
3388
3389                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3390                         mdname(mddev));
3391                 /*
3392                  * Kick recovery or resync if necessary
3393                  */
3394                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3395                 md_wakeup_thread(mddev->thread);
3396                 md_wakeup_thread(mddev->sync_thread);
3397                 err = 0;
3398         } else
3399                 err = -EINVAL;
3400
3401 out:
3402         return err;
3403 }
3404
3405 /* similar to deny_write_access, but accounts for our holding a reference
3406  * to the file ourselves */
3407 static int deny_bitmap_write_access(struct file * file)
3408 {
3409         struct inode *inode = file->f_mapping->host;
3410
3411         spin_lock(&inode->i_lock);
3412         if (atomic_read(&inode->i_writecount) > 1) {
3413                 spin_unlock(&inode->i_lock);
3414                 return -ETXTBSY;
3415         }
3416         atomic_set(&inode->i_writecount, -1);
3417         spin_unlock(&inode->i_lock);
3418
3419         return 0;
3420 }
3421
3422 static void restore_bitmap_write_access(struct file *file)
3423 {
3424         struct inode *inode = file->f_mapping->host;
3425
3426         spin_lock(&inode->i_lock);
3427         atomic_set(&inode->i_writecount, 1);
3428         spin_unlock(&inode->i_lock);
3429 }
3430
3431 /* mode:
3432  *   0 - completely stop and dis-assemble array
3433  *   1 - switch to readonly
3434  *   2 - stop but do not disassemble array
3435  */
3436 static int do_md_stop(mddev_t * mddev, int mode)
3437 {
3438         int err = 0;
3439         struct gendisk *disk = mddev->gendisk;
3440
3441         if (mddev->pers) {
3442                 if (atomic_read(&mddev->active)>2) {
3443                         printk("md: %s still in use.\n",mdname(mddev));
3444                         return -EBUSY;
3445                 }
3446
3447                 if (mddev->sync_thread) {
3448                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3449                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3450                         md_unregister_thread(mddev->sync_thread);
3451                         mddev->sync_thread = NULL;
3452                 }
3453
3454                 del_timer_sync(&mddev->safemode_timer);
3455
3456                 invalidate_partition(disk, 0);
3457
3458                 switch(mode) {
3459                 case 1: /* readonly */
3460                         err  = -ENXIO;
3461                         if (mddev->ro==1)
3462                                 goto out;
3463                         mddev->ro = 1;
3464                         break;
3465                 case 0: /* disassemble */
3466                 case 2: /* stop */
3467                         bitmap_flush(mddev);
3468                         md_super_wait(mddev);
3469                         if (mddev->ro)
3470                                 set_disk_ro(disk, 0);
3471                         blk_queue_make_request(mddev->queue, md_fail_request);
3472                         mddev->pers->stop(mddev);
3473                         mddev->queue->merge_bvec_fn = NULL;
3474                         mddev->queue->unplug_fn = NULL;
3475                         mddev->queue->backing_dev_info.congested_fn = NULL;
3476                         if (mddev->pers->sync_request)
3477                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3478
3479                         module_put(mddev->pers->owner);
3480                         mddev->pers = NULL;
3481
3482                         set_capacity(disk, 0);
3483                         mddev->changed = 1;
3484
3485                         if (mddev->ro)
3486                                 mddev->ro = 0;
3487                 }
3488                 if (!mddev->in_sync || mddev->flags) {
3489                         /* mark array as shutdown cleanly */
3490                         mddev->in_sync = 1;
3491                         md_update_sb(mddev, 1);
3492                 }
3493                 if (mode == 1)
3494                         set_disk_ro(disk, 1);
3495                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3496         }
3497
3498         /*
3499          * Free resources if final stop
3500          */
3501         if (mode == 0) {
3502                 mdk_rdev_t *rdev;
3503                 struct list_head *tmp;
3504
3505                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3506
3507                 bitmap_destroy(mddev);
3508                 if (mddev->bitmap_file) {
3509                         restore_bitmap_write_access(mddev->bitmap_file);
3510                         fput(mddev->bitmap_file);
3511                         mddev->bitmap_file = NULL;
3512                 }
3513                 mddev->bitmap_offset = 0;
3514
3515                 ITERATE_RDEV(mddev,rdev,tmp)
3516                         if (rdev->raid_disk >= 0) {
3517                                 char nm[20];
3518                                 sprintf(nm, "rd%d", rdev->raid_disk);
3519                                 sysfs_remove_link(&mddev->kobj, nm);
3520                         }
3521
3522                 /* make sure all delayed_delete calls have finished */
3523                 flush_scheduled_work();
3524
3525                 export_array(mddev);
3526
3527                 mddev->array_size = 0;
3528                 mddev->size = 0;
3529                 mddev->raid_disks = 0;
3530                 mddev->recovery_cp = 0;
3531                 mddev->reshape_position = MaxSector;
3532
3533         } else if (mddev->pers)
3534                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3535                         mdname(mddev));
3536         err = 0;
3537         md_new_event(mddev);
3538 out:
3539         return err;
3540 }
3541
3542 #ifndef MODULE
3543 static void autorun_array(mddev_t *mddev)
3544 {
3545         mdk_rdev_t *rdev;
3546         struct list_head *tmp;
3547         int err;
3548
3549         if (list_empty(&mddev->disks))
3550                 return;
3551
3552         printk(KERN_INFO "md: running: ");
3553
3554         ITERATE_RDEV(mddev,rdev,tmp) {
3555                 char b[BDEVNAME_SIZE];
3556                 printk("<%s>", bdevname(rdev->bdev,b));
3557         }
3558         printk("\n");
3559
3560         err = do_md_run (mddev);
3561         if (err) {
3562                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3563                 do_md_stop (mddev, 0);
3564         }
3565 }
3566
3567 /*
3568  * lets try to run arrays based on all disks that have arrived
3569  * until now. (those are in pending_raid_disks)
3570  *
3571  * the method: pick the first pending disk, collect all disks with
3572  * the same UUID, remove all from the pending list and put them into
3573  * the 'same_array' list. Then order this list based on superblock
3574  * update time (freshest comes first), kick out 'old' disks and
3575  * compare superblocks. If everything's fine then run it.
3576  *
3577  * If "unit" is allocated, then bump its reference count
3578  */
3579 static void autorun_devices(int part)
3580 {
3581         struct list_head *tmp;
3582         mdk_rdev_t *rdev0, *rdev;
3583         mddev_t *mddev;
3584         char b[BDEVNAME_SIZE];
3585
3586         printk(KERN_INFO "md: autorun ...\n");
3587         while (!list_empty(&pending_raid_disks)) {
3588                 int unit;
3589                 dev_t dev;
3590                 LIST_HEAD(candidates);
3591                 rdev0 = list_entry(pending_raid_disks.next,
3592                                          mdk_rdev_t, same_set);
3593
3594                 printk(KERN_INFO "md: considering %s ...\n",
3595                         bdevname(rdev0->bdev,b));
3596                 INIT_LIST_HEAD(&candidates);
3597                 ITERATE_RDEV_PENDING(rdev,tmp)
3598                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3599                                 printk(KERN_INFO "md:  adding %s ...\n",
3600                                         bdevname(rdev->bdev,b));
3601                                 list_move(&rdev->same_set, &candidates);
3602                         }
3603                 /*
3604                  * now we have a set of devices, with all of them having
3605                  * mostly sane superblocks. It's time to allocate the
3606                  * mddev.
3607                  */
3608                 if (part) {
3609                         dev = MKDEV(mdp_major,
3610                                     rdev0->preferred_minor << MdpMinorShift);
3611                         unit = MINOR(dev) >> MdpMinorShift;
3612                 } else {
3613                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3614                         unit = MINOR(dev);
3615                 }
3616                 if (rdev0->preferred_minor != unit) {
3617                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3618                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3619                         break;
3620                 }
3621
3622                 md_probe(dev, NULL, NULL);
3623                 mddev = mddev_find(dev);
3624                 if (!mddev) {
3625                         printk(KERN_ERR 
3626                                 "md: cannot allocate memory for md drive.\n");
3627                         break;
3628                 }
3629                 if (mddev_lock(mddev)) 
3630                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3631                                mdname(mddev));
3632                 else if (mddev->raid_disks || mddev->major_version
3633                          || !list_empty(&mddev->disks)) {
3634                         printk(KERN_WARNING 
3635                                 "md: %s already running, cannot run %s\n",
3636                                 mdname(mddev), bdevname(rdev0->bdev,b));
3637                         mddev_unlock(mddev);
3638                 } else {
3639                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3640                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3641                                 list_del_init(&rdev->same_set);
3642                                 if (bind_rdev_to_array(rdev, mddev))
3643                                         export_rdev(rdev);
3644                         }
3645                         autorun_array(mddev);
3646                         mddev_unlock(mddev);
3647                 }
3648                 /* on success, candidates will be empty, on error
3649                  * it won't...
3650                  */
3651                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3652                         export_rdev(rdev);
3653                 mddev_put(mddev);
3654         }
3655         printk(KERN_INFO "md: ... autorun DONE.\n");
3656 }
3657 #endif /* !MODULE */
3658
3659 static int get_version(void __user * arg)
3660 {
3661         mdu_version_t ver;
3662
3663         ver.major = MD_MAJOR_VERSION;
3664         ver.minor = MD_MINOR_VERSION;
3665         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3666
3667         if (copy_to_user(arg, &ver, sizeof(ver)))
3668                 return -EFAULT;
3669
3670         return 0;
3671 }
3672
3673 static int get_array_info(mddev_t * mddev, void __user * arg)
3674 {
3675         mdu_array_info_t info;
3676         int nr,working,active,failed,spare;
3677         mdk_rdev_t *rdev;
3678         struct list_head *tmp;
3679
3680         nr=working=active=failed=spare=0;
3681         ITERATE_RDEV(mddev,rdev,tmp) {
3682                 nr++;
3683                 if (test_bit(Faulty, &rdev->flags))
3684                         failed++;
3685                 else {
3686                         working++;
3687                         if (test_bit(In_sync, &rdev->flags))
3688                                 active++;       
3689                         else
3690                                 spare++;
3691                 }
3692         }
3693
3694         info.major_version = mddev->major_version;
3695         info.minor_version = mddev->minor_version;
3696         info.patch_version = MD_PATCHLEVEL_VERSION;
3697         info.ctime         = mddev->ctime;
3698         info.level         = mddev->level;
3699         info.size          = mddev->size;
3700         if (info.size != mddev->size) /* overflow */
3701                 info.size = -1;
3702         info.nr_disks      = nr;
3703         info.raid_disks    = mddev->raid_disks;
3704         info.md_minor      = mddev->md_minor;
3705         info.not_persistent= !mddev->persistent;
3706
3707         info.utime         = mddev->utime;
3708         info.state         = 0;
3709         if (mddev->in_sync)
3710                 info.state = (1<<MD_SB_CLEAN);
3711         if (mddev->bitmap && mddev->bitmap_offset)
3712                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3713         info.active_disks  = active;
3714         info.working_disks = working;
3715         info.failed_disks  = failed;
3716         info.spare_disks   = spare;
3717
3718         info.layout        = mddev->layout;
3719         info.chunk_size    = mddev->chunk_size;
3720
3721         if (copy_to_user(arg, &info, sizeof(info)))
3722                 return -EFAULT;
3723
3724         return 0;
3725 }
3726
3727 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3728 {
3729         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3730         char *ptr, *buf = NULL;
3731         int err = -ENOMEM;
3732
3733         md_allow_write(mddev);
3734
3735         file = kmalloc(sizeof(*file), GFP_KERNEL);
3736         if (!file)
3737                 goto out;
3738
3739         /* bitmap disabled, zero the first byte and copy out */
3740         if (!mddev->bitmap || !mddev->bitmap->file) {
3741                 file->pathname[0] = '\0';
3742                 goto copy_out;
3743         }
3744
3745         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3746         if (!buf)
3747                 goto out;
3748
3749         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3750         if (!ptr)
3751                 goto out;
3752
3753         strcpy(file->pathname, ptr);
3754
3755 copy_out:
3756         err = 0;
3757         if (copy_to_user(arg, file, sizeof(*file)))
3758                 err = -EFAULT;
3759 out:
3760         kfree(buf);
3761         kfree(file);
3762         return err;
3763 }
3764
3765 static int get_disk_info(mddev_t * mddev, void __user * arg)
3766 {
3767         mdu_disk_info_t info;
3768         unsigned int nr;
3769         mdk_rdev_t *rdev;
3770
3771         if (copy_from_user(&info, arg, sizeof(info)))
3772                 return -EFAULT;
3773
3774         nr = info.number;
3775
3776         rdev = find_rdev_nr(mddev, nr);
3777         if (rdev) {
3778                 info.major = MAJOR(rdev->bdev->bd_dev);
3779                 info.minor = MINOR(rdev->bdev->bd_dev);
3780                 info.raid_disk = rdev->raid_disk;
3781                 info.state = 0;
3782                 if (test_bit(Faulty, &rdev->flags))
3783                         info.state |= (1<<MD_DISK_FAULTY);
3784                 else if (test_bit(In_sync, &rdev->flags)) {
3785                         info.state |= (1<<MD_DISK_ACTIVE);
3786                         info.state |= (1<<MD_DISK_SYNC);
3787                 }
3788                 if (test_bit(WriteMostly, &rdev->flags))
3789                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3790         } else {
3791                 info.major = info.minor = 0;
3792                 info.raid_disk = -1;
3793                 info.state = (1<<MD_DISK_REMOVED);
3794         }
3795
3796         if (copy_to_user(arg, &info, sizeof(info)))
3797                 return -EFAULT;
3798
3799         return 0;
3800 }
3801
3802 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3803 {
3804         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3805         mdk_rdev_t *rdev;
3806         dev_t dev = MKDEV(info->major,info->minor);
3807
3808         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3809                 return -EOVERFLOW;
3810
3811         if (!mddev->raid_disks) {
3812                 int err;
3813                 /* expecting a device which has a superblock */
3814                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3815                 if (IS_ERR(rdev)) {
3816                         printk(KERN_WARNING 
3817                                 "md: md_import_device returned %ld\n",
3818                                 PTR_ERR(rdev));
3819                         return PTR_ERR(rdev);
3820                 }
3821                 if (!list_empty(&mddev->disks)) {
3822                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3823                                                         mdk_rdev_t, same_set);
3824                         int err = super_types[mddev->major_version]
3825                                 .load_super(rdev, rdev0, mddev->minor_version);
3826                         if (err < 0) {
3827                                 printk(KERN_WARNING 
3828                                         "md: %s has different UUID to %s\n",
3829                                         bdevname(rdev->bdev,b), 
3830                                         bdevname(rdev0->bdev,b2));
3831                                 export_rdev(rdev);
3832                                 return -EINVAL;
3833                         }
3834                 }
3835                 err = bind_rdev_to_array(rdev, mddev);
3836                 if (err)
3837                         export_rdev(rdev);
3838                 return err;
3839         }
3840
3841         /*
3842          * add_new_disk can be used once the array is assembled
3843          * to add "hot spares".  They must already have a superblock
3844          * written
3845          */
3846         if (mddev->pers) {
3847                 int err;
3848                 if (!mddev->pers->hot_add_disk) {
3849                         printk(KERN_WARNING 
3850                                 "%s: personality does not support diskops!\n",
3851                                mdname(mddev));
3852                         return -EINVAL;
3853                 }
3854                 if (mddev->persistent)
3855                         rdev = md_import_device(dev, mddev->major_version,
3856                                                 mddev->minor_version);
3857                 else
3858                         rdev = md_import_device(dev, -1, -1);
3859                 if (IS_ERR(rdev)) {
3860                         printk(KERN_WARNING 
3861                                 "md: md_import_device returned %ld\n",
3862                                 PTR_ERR(rdev));
3863                         return PTR_ERR(rdev);
3864                 }
3865                 /* set save_raid_disk if appropriate */
3866                 if (!mddev->persistent) {
3867                         if (info->state & (1<<MD_DISK_SYNC)  &&
3868                             info->raid_disk < mddev->raid_disks)
3869                                 rdev->raid_disk = info->raid_disk;
3870                         else
3871                                 rdev->raid_disk = -1;
3872                 } else
3873                         super_types[mddev->major_version].
3874                                 validate_super(mddev, rdev);
3875                 rdev->saved_raid_disk = rdev->raid_disk;
3876
3877                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3878                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3879                         set_bit(WriteMostly, &rdev->flags);
3880
3881                 rdev->raid_disk = -1;
3882                 err = bind_rdev_to_array(rdev, mddev);
3883                 if (!err && !mddev->pers->hot_remove_disk) {
3884                         /* If there is hot_add_disk but no hot_remove_disk
3885                          * then added disks for geometry changes,
3886                          * and should be added immediately.
3887                          */
3888                         super_types[mddev->major_version].
3889                                 validate_super(mddev, rdev);
3890                         err = mddev->pers->hot_add_disk(mddev, rdev);
3891                         if (err)
3892                                 unbind_rdev_from_array(rdev);
3893                 }
3894                 if (err)
3895                         export_rdev(rdev);
3896
3897                 md_update_sb(mddev, 1);
3898                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3899                 md_wakeup_thread(mddev->thread);
3900                 return err;
3901         }
3902
3903         /* otherwise, add_new_disk is only allowed
3904          * for major_version==0 superblocks
3905          */
3906         if (mddev->major_version != 0) {
3907                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3908                        mdname(mddev));
3909                 return -EINVAL;
3910         }
3911
3912         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3913                 int err;
3914                 rdev = md_import_device (dev, -1, 0);
3915                 if (IS_ERR(rdev)) {
3916                         printk(KERN_WARNING 
3917                                 "md: error, md_import_device() returned %ld\n",
3918                                 PTR_ERR(rdev));
3919                         return PTR_ERR(rdev);
3920                 }
3921                 rdev->desc_nr = info->number;
3922                 if (info->raid_disk < mddev->raid_disks)
3923                         rdev->raid_disk = info->raid_disk;
3924                 else
3925                         rdev->raid_disk = -1;
3926
3927                 rdev->flags = 0;
3928
3929                 if (rdev->raid_disk < mddev->raid_disks)
3930                         if (info->state & (1<<MD_DISK_SYNC))
3931                                 set_bit(In_sync, &rdev->flags);
3932
3933                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3934                         set_bit(WriteMostly, &rdev->flags);
3935
3936                 if (!mddev->persistent) {
3937                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3938                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3939                 } else 
3940                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3941                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3942
3943                 err = bind_rdev_to_array(rdev, mddev);
3944                 if (err) {
3945                         export_rdev(rdev);
3946                         return err;
3947                 }
3948         }
3949
3950         return 0;
3951 }
3952
3953 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3954 {
3955         char b[BDEVNAME_SIZE];
3956         mdk_rdev_t *rdev;
3957
3958         if (!mddev->pers)
3959                 return -ENODEV;
3960
3961         rdev = find_rdev(mddev, dev);
3962         if (!rdev)
3963                 return -ENXIO;
3964
3965         if (rdev->raid_disk >= 0)
3966                 goto busy;
3967
3968         kick_rdev_from_array(rdev);
3969         md_update_sb(mddev, 1);
3970         md_new_event(mddev);
3971
3972         return 0;
3973 busy:
3974         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3975                 bdevname(rdev->bdev,b), mdname(mddev));
3976         return -EBUSY;
3977 }
3978
3979 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3980 {
3981         char b[BDEVNAME_SIZE];
3982         int err;
3983         unsigned int size;
3984         mdk_rdev_t *rdev;
3985
3986         if (!mddev->pers)
3987                 return -ENODEV;
3988
3989         if (mddev->major_version != 0) {
3990                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3991                         " version-0 superblocks.\n",
3992                         mdname(mddev));
3993                 return -EINVAL;
3994         }
3995         if (!mddev->pers->hot_add_disk) {
3996                 printk(KERN_WARNING 
3997                         "%s: personality does not support diskops!\n",
3998                         mdname(mddev));
3999                 return -EINVAL;
4000         }
4001
4002         rdev = md_import_device (dev, -1, 0);
4003         if (IS_ERR(rdev)) {
4004                 printk(KERN_WARNING 
4005                         "md: error, md_import_device() returned %ld\n",
4006                         PTR_ERR(rdev));
4007                 return -EINVAL;
4008         }
4009
4010         if (mddev->persistent)
4011                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4012         else
4013                 rdev->sb_offset =
4014                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4015
4016         size = calc_dev_size(rdev, mddev->chunk_size);
4017         rdev->size = size;
4018
4019         if (test_bit(Faulty, &rdev->flags)) {
4020                 printk(KERN_WARNING 
4021                         "md: can not hot-add faulty %s disk to %s!\n",
4022                         bdevname(rdev->bdev,b), mdname(mddev));
4023                 err = -EINVAL;
4024                 goto abort_export;
4025         }
4026         clear_bit(In_sync, &rdev->flags);
4027         rdev->desc_nr = -1;
4028         rdev->saved_raid_disk = -1;
4029         err = bind_rdev_to_array(rdev, mddev);
4030         if (err)
4031                 goto abort_export;
4032
4033         /*
4034          * The rest should better be atomic, we can have disk failures
4035          * noticed in interrupt contexts ...
4036          */
4037
4038         if (rdev->desc_nr == mddev->max_disks) {
4039                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4040                         mdname(mddev));
4041                 err = -EBUSY;
4042                 goto abort_unbind_export;
4043         }
4044
4045         rdev->raid_disk = -1;
4046
4047         md_update_sb(mddev, 1);
4048
4049         /*
4050          * Kick recovery, maybe this spare has to be added to the
4051          * array immediately.
4052          */
4053         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4054         md_wakeup_thread(mddev->thread);
4055         md_new_event(mddev);
4056         return 0;
4057
4058 abort_unbind_export:
4059         unbind_rdev_from_array(rdev);
4060
4061 abort_export:
4062         export_rdev(rdev);
4063         return err;
4064 }
4065
4066 static int set_bitmap_file(mddev_t *mddev, int fd)
4067 {
4068         int err;
4069
4070         if (mddev->pers) {
4071                 if (!mddev->pers->quiesce)
4072                         return -EBUSY;
4073                 if (mddev->recovery || mddev->sync_thread)
4074                         return -EBUSY;
4075                 /* we should be able to change the bitmap.. */
4076         }
4077
4078
4079         if (fd >= 0) {
4080                 if (mddev->bitmap)
4081                         return -EEXIST; /* cannot add when bitmap is present */
4082                 mddev->bitmap_file = fget(fd);
4083
4084                 if (mddev->bitmap_file == NULL) {
4085                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4086                                mdname(mddev));
4087                         return -EBADF;
4088                 }
4089
4090                 err = deny_bitmap_write_access(mddev->bitmap_file);
4091                 if (err) {
4092                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4093                                mdname(mddev));
4094                         fput(mddev->bitmap_file);
4095                         mddev->bitmap_file = NULL;
4096                         return err;
4097                 }
4098                 mddev->bitmap_offset = 0; /* file overrides offset */
4099         } else if (mddev->bitmap == NULL)
4100                 return -ENOENT; /* cannot remove what isn't there */
4101         err = 0;
4102         if (mddev->pers) {
4103                 mddev->pers->quiesce(mddev, 1);
4104                 if (fd >= 0)
4105                         err = bitmap_create(mddev);
4106                 if (fd < 0 || err) {
4107                         bitmap_destroy(mddev);
4108                         fd = -1; /* make sure to put the file */
4109                 }
4110                 mddev->pers->quiesce(mddev, 0);
4111         }
4112         if (fd < 0) {
4113                 if (mddev->bitmap_file) {
4114                         restore_bitmap_write_access(mddev->bitmap_file);
4115                         fput(mddev->bitmap_file);
4116                 }
4117                 mddev->bitmap_file = NULL;
4118         }
4119
4120         return err;
4121 }
4122
4123 /*
4124  * set_array_info is used two different ways
4125  * The original usage is when creating a new array.
4126  * In this usage, raid_disks is > 0 and it together with
4127  *  level, size, not_persistent,layout,chunksize determine the
4128  *  shape of the array.
4129  *  This will always create an array with a type-0.90.0 superblock.
4130  * The newer usage is when assembling an array.
4131  *  In this case raid_disks will be 0, and the major_version field is
4132  *  use to determine which style super-blocks are to be found on the devices.
4133  *  The minor and patch _version numbers are also kept incase the
4134  *  super_block handler wishes to interpret them.
4135  */
4136 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4137 {
4138
4139         if (info->raid_disks == 0) {
4140                 /* just setting version number for superblock loading */
4141                 if (info->major_version < 0 ||
4142                     info->major_version >= ARRAY_SIZE(super_types) ||
4143                     super_types[info->major_version].name == NULL) {
4144                         /* maybe try to auto-load a module? */
4145                         printk(KERN_INFO 
4146                                 "md: superblock version %d not known\n",
4147                                 info->major_version);
4148                         return -EINVAL;
4149                 }
4150                 mddev->major_version = info->major_version;
4151                 mddev->minor_version = info->minor_version;
4152                 mddev->patch_version = info->patch_version;
4153                 mddev->persistent = !info->not_persistent;
4154                 return 0;
4155         }
4156         mddev->major_version = MD_MAJOR_VERSION;
4157         mddev->minor_version = MD_MINOR_VERSION;
4158         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4159         mddev->ctime         = get_seconds();
4160
4161         mddev->level         = info->level;
4162         mddev->clevel[0]     = 0;
4163         mddev->size          = info->size;
4164         mddev->raid_disks    = info->raid_disks;
4165         /* don't set md_minor, it is determined by which /dev/md* was
4166          * openned
4167          */
4168         if (info->state & (1<<MD_SB_CLEAN))
4169                 mddev->recovery_cp = MaxSector;
4170         else
4171                 mddev->recovery_cp = 0;
4172         mddev->persistent    = ! info->not_persistent;
4173
4174         mddev->layout        = info->layout;
4175         mddev->chunk_size    = info->chunk_size;
4176
4177         mddev->max_disks     = MD_SB_DISKS;
4178
4179         mddev->flags         = 0;
4180         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4181
4182         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4183         mddev->bitmap_offset = 0;
4184
4185         mddev->reshape_position = MaxSector;
4186
4187         /*
4188          * Generate a 128 bit UUID
4189          */
4190         get_random_bytes(mddev->uuid, 16);
4191
4192         mddev->new_level = mddev->level;
4193         mddev->new_chunk = mddev->chunk_size;
4194         mddev->new_layout = mddev->layout;
4195         mddev->delta_disks = 0;
4196
4197         return 0;
4198 }
4199
4200 static int update_size(mddev_t *mddev, unsigned long size)
4201 {
4202         mdk_rdev_t * rdev;
4203         int rv;
4204         struct list_head *tmp;
4205         int fit = (size == 0);
4206
4207         if (mddev->pers->resize == NULL)
4208                 return -EINVAL;
4209         /* The "size" is the amount of each device that is used.
4210          * This can only make sense for arrays with redundancy.
4211          * linear and raid0 always use whatever space is available
4212          * We can only consider changing the size if no resync
4213          * or reconstruction is happening, and if the new size
4214          * is acceptable. It must fit before the sb_offset or,
4215          * if that is <data_offset, it must fit before the
4216          * size of each device.
4217          * If size is zero, we find the largest size that fits.
4218          */
4219         if (mddev->sync_thread)
4220                 return -EBUSY;
4221         ITERATE_RDEV(mddev,rdev,tmp) {
4222                 sector_t avail;
4223                 avail = rdev->size * 2;
4224
4225                 if (fit && (size == 0 || size > avail/2))
4226                         size = avail/2;
4227                 if (avail < ((sector_t)size << 1))
4228                         return -ENOSPC;
4229         }
4230         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4231         if (!rv) {
4232                 struct block_device *bdev;
4233
4234                 bdev = bdget_disk(mddev->gendisk, 0);
4235                 if (bdev) {
4236                         mutex_lock(&bdev->bd_inode->i_mutex);
4237                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4238                         mutex_unlock(&bdev->bd_inode->i_mutex);
4239                         bdput(bdev);
4240                 }
4241         }
4242         return rv;
4243 }
4244
4245 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4246 {
4247         int rv;
4248         /* change the number of raid disks */
4249         if (mddev->pers->check_reshape == NULL)
4250                 return -EINVAL;
4251         if (raid_disks <= 0 ||
4252             raid_disks >= mddev->max_disks)
4253                 return -EINVAL;
4254         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4255                 return -EBUSY;
4256         mddev->delta_disks = raid_disks - mddev->raid_disks;
4257
4258         rv = mddev->pers->check_reshape(mddev);
4259         return rv;
4260 }
4261
4262
4263 /*
4264  * update_array_info is used to change the configuration of an
4265  * on-line array.
4266  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4267  * fields in the info are checked against the array.
4268  * Any differences that cannot be handled will cause an error.
4269  * Normally, only one change can be managed at a time.
4270  */
4271 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4272 {
4273         int rv = 0;
4274         int cnt = 0;
4275         int state = 0;
4276
4277         /* calculate expected state,ignoring low bits */
4278         if (mddev->bitmap && mddev->bitmap_offset)
4279                 state |= (1 << MD_SB_BITMAP_PRESENT);
4280
4281         if (mddev->major_version != info->major_version ||
4282             mddev->minor_version != info->minor_version ||
4283 /*          mddev->patch_version != info->patch_version || */
4284             mddev->ctime         != info->ctime         ||
4285             mddev->level         != info->level         ||
4286 /*          mddev->layout        != info->layout        || */
4287             !mddev->persistent   != info->not_persistent||
4288             mddev->chunk_size    != info->chunk_size    ||
4289             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4290             ((state^info->state) & 0xfffffe00)
4291                 )
4292                 return -EINVAL;
4293         /* Check there is only one change */
4294         if (info->size >= 0 && mddev->size != info->size) cnt++;
4295         if (mddev->raid_disks != info->raid_disks) cnt++;
4296         if (mddev->layout != info->layout) cnt++;
4297         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4298         if (cnt == 0) return 0;
4299         if (cnt > 1) return -EINVAL;
4300
4301         if (mddev->layout != info->layout) {
4302                 /* Change layout
4303                  * we don't need to do anything at the md level, the
4304                  * personality will take care of it all.
4305                  */
4306                 if (mddev->pers->reconfig == NULL)
4307                         return -EINVAL;
4308                 else
4309                         return mddev->pers->reconfig(mddev, info->layout, -1);
4310         }
4311         if (info->size >= 0 && mddev->size != info->size)
4312                 rv = update_size(mddev, info->size);
4313
4314         if (mddev->raid_disks    != info->raid_disks)
4315                 rv = update_raid_disks(mddev, info->raid_disks);
4316
4317         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4318                 if (mddev->pers->quiesce == NULL)
4319                         return -EINVAL;
4320                 if (mddev->recovery || mddev->sync_thread)
4321                         return -EBUSY;
4322                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4323                         /* add the bitmap */
4324                         if (mddev->bitmap)
4325                                 return -EEXIST;
4326                         if (mddev->default_bitmap_offset == 0)
4327                                 return -EINVAL;
4328                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4329                         mddev->pers->quiesce(mddev, 1);
4330                         rv = bitmap_create(mddev);
4331                         if (rv)
4332                                 bitmap_destroy(mddev);
4333                         mddev->pers->quiesce(mddev, 0);
4334                 } else {
4335                         /* remove the bitmap */
4336                         if (!mddev->bitmap)
4337                                 return -ENOENT;
4338                         if (mddev->bitmap->file)
4339                                 return -EINVAL;
4340                         mddev->pers->quiesce(mddev, 1);
4341                         bitmap_destroy(mddev);
4342                         mddev->pers->quiesce(mddev, 0);
4343                         mddev->bitmap_offset = 0;
4344                 }
4345         }
4346         md_update_sb(mddev, 1);
4347         return rv;
4348 }
4349
4350 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4351 {
4352         mdk_rdev_t *rdev;
4353
4354         if (mddev->pers == NULL)
4355                 return -ENODEV;
4356
4357         rdev = find_rdev(mddev, dev);
4358         if (!rdev)
4359                 return -ENODEV;
4360
4361         md_error(mddev, rdev);
4362         return 0;
4363 }
4364
4365 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4366 {
4367         mddev_t *mddev = bdev->bd_disk->private_data;
4368
4369         geo->heads = 2;
4370         geo->sectors = 4;
4371         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4372         return 0;
4373 }
4374
4375 static int md_ioctl(struct inode *inode, struct file *file,
4376                         unsigned int cmd, unsigned long arg)
4377 {
4378         int err = 0;
4379         void __user *argp = (void __user *)arg;
4380         mddev_t *mddev = NULL;
4381
4382         if (!capable(CAP_SYS_ADMIN))
4383                 return -EACCES;
4384
4385         /*
4386          * Commands dealing with the RAID driver but not any
4387          * particular array:
4388          */
4389         switch (cmd)
4390         {
4391                 case RAID_VERSION:
4392                         err = get_version(argp);
4393                         goto done;
4394
4395                 case PRINT_RAID_DEBUG:
4396                         err = 0;
4397                         md_print_devices();
4398                         goto done;
4399
4400 #ifndef MODULE
4401                 case RAID_AUTORUN:
4402                         err = 0;
4403                         autostart_arrays(arg);
4404                         goto done;
4405 #endif
4406                 default:;
4407         }
4408
4409         /*
4410          * Commands creating/starting a new array:
4411          */
4412
4413         mddev = inode->i_bdev->bd_disk->private_data;
4414
4415         if (!mddev) {
4416                 BUG();
4417                 goto abort;
4418         }
4419
4420         err = mddev_lock(mddev);
4421         if (err) {
4422                 printk(KERN_INFO 
4423                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4424                         err, cmd);
4425                 goto abort;
4426         }
4427
4428         switch (cmd)
4429         {
4430                 case SET_ARRAY_INFO:
4431                         {
4432                                 mdu_array_info_t info;
4433                                 if (!arg)
4434                                         memset(&info, 0, sizeof(info));
4435                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4436                                         err = -EFAULT;
4437                                         goto abort_unlock;
4438                                 }
4439                                 if (mddev->pers) {
4440                                         err = update_array_info(mddev, &info);
4441                                         if (err) {
4442                                                 printk(KERN_WARNING "md: couldn't update"
4443                                                        " array info. %d\n", err);
4444                                                 goto abort_unlock;
4445                                         }
4446                                         goto done_unlock;
4447                                 }
4448                                 if (!list_empty(&mddev->disks)) {
4449                                         printk(KERN_WARNING
4450                                                "md: array %s already has disks!\n",
4451                                                mdname(mddev));
4452                                         err = -EBUSY;
4453                                         goto abort_unlock;
4454                                 }
4455                                 if (mddev->raid_disks) {
4456                                         printk(KERN_WARNING
4457                                                "md: array %s already initialised!\n",
4458                                                mdname(mddev));
4459                                         err = -EBUSY;
4460                                         goto abort_unlock;
4461                                 }
4462                                 err = set_array_info(mddev, &info);
4463                                 if (err) {
4464                                         printk(KERN_WARNING "md: couldn't set"
4465                                                " array info. %d\n", err);
4466                                         goto abort_unlock;
4467                                 }
4468                         }
4469                         goto done_unlock;
4470
4471                 default:;
4472         }
4473
4474         /*
4475          * Commands querying/configuring an existing array:
4476          */
4477         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4478          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4479         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4480                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4481                         && cmd != GET_BITMAP_FILE) {
4482                 err = -ENODEV;
4483                 goto abort_unlock;
4484         }
4485
4486         /*
4487          * Commands even a read-only array can execute:
4488          */
4489         switch (cmd)
4490         {
4491                 case GET_ARRAY_INFO:
4492                         err = get_array_info(mddev, argp);
4493                         goto done_unlock;
4494
4495                 case GET_BITMAP_FILE:
4496                         err = get_bitmap_file(mddev, argp);
4497                         goto done_unlock;
4498
4499                 case GET_DISK_INFO:
4500                         err = get_disk_info(mddev, argp);
4501                         goto done_unlock;
4502
4503                 case RESTART_ARRAY_RW:
4504                         err = restart_array(mddev);
4505                         goto done_unlock;
4506
4507                 case STOP_ARRAY:
4508                         err = do_md_stop (mddev, 0);
4509                         goto done_unlock;
4510
4511                 case STOP_ARRAY_RO:
4512                         err = do_md_stop (mddev, 1);
4513                         goto done_unlock;
4514
4515         /*
4516          * We have a problem here : there is no easy way to give a CHS
4517          * virtual geometry. We currently pretend that we have a 2 heads
4518          * 4 sectors (with a BIG number of cylinders...). This drives
4519          * dosfs just mad... ;-)
4520          */
4521         }
4522
4523         /*
4524          * The remaining ioctls are changing the state of the
4525          * superblock, so we do not allow them on read-only arrays.
4526          * However non-MD ioctls (e.g. get-size) will still come through
4527          * here and hit the 'default' below, so only disallow
4528          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4529          */
4530         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4531             mddev->ro && mddev->pers) {
4532                 if (mddev->ro == 2) {
4533                         mddev->ro = 0;
4534                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4535                 md_wakeup_thread(mddev->thread);
4536
4537                 } else {
4538                         err = -EROFS;
4539                         goto abort_unlock;
4540                 }
4541         }
4542
4543         switch (cmd)
4544         {
4545                 case ADD_NEW_DISK:
4546                 {
4547                         mdu_disk_info_t info;
4548                         if (copy_from_user(&info, argp, sizeof(info)))
4549                                 err = -EFAULT;
4550                         else
4551                                 err = add_new_disk(mddev, &info);
4552                         goto done_unlock;
4553                 }
4554
4555                 case HOT_REMOVE_DISK:
4556                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4557                         goto done_unlock;
4558
4559                 case HOT_ADD_DISK:
4560                         err = hot_add_disk(mddev, new_decode_dev(arg));
4561                         goto done_unlock;
4562
4563                 case SET_DISK_FAULTY:
4564                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4565                         goto done_unlock;
4566
4567                 case RUN_ARRAY:
4568                         err = do_md_run (mddev);
4569                         goto done_unlock;
4570
4571                 case SET_BITMAP_FILE:
4572                         err = set_bitmap_file(mddev, (int)arg);
4573                         goto done_unlock;
4574
4575                 default:
4576                         err = -EINVAL;
4577                         goto abort_unlock;
4578         }
4579
4580 done_unlock:
4581 abort_unlock:
4582         mddev_unlock(mddev);
4583
4584         return err;
4585 done:
4586         if (err)
4587                 MD_BUG();
4588 abort:
4589         return err;
4590 }
4591
4592 static int md_open(struct inode *inode, struct file *file)
4593 {
4594         /*
4595          * Succeed if we can lock the mddev, which confirms that
4596          * it isn't being stopped right now.
4597          */
4598         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4599         int err;
4600
4601         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4602                 goto out;
4603
4604         err = 0;
4605         mddev_get(mddev);
4606         mddev_unlock(mddev);
4607
4608         check_disk_change(inode->i_bdev);
4609  out:
4610         return err;
4611 }
4612
4613 static int md_release(struct inode *inode, struct file * file)
4614 {
4615         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4616
4617         BUG_ON(!mddev);
4618         mddev_put(mddev);
4619
4620         return 0;
4621 }
4622
4623 static int md_media_changed(struct gendisk *disk)
4624 {
4625         mddev_t *mddev = disk->private_data;
4626
4627         return mddev->changed;
4628 }
4629
4630 static int md_revalidate(struct gendisk *disk)
4631 {
4632         mddev_t *mddev = disk->private_data;
4633
4634         mddev->changed = 0;
4635         return 0;
4636 }
4637 static struct block_device_operations md_fops =
4638 {
4639         .owner          = THIS_MODULE,
4640         .open           = md_open,
4641         .release        = md_release,
4642         .ioctl          = md_ioctl,
4643         .getgeo         = md_getgeo,
4644         .media_changed  = md_media_changed,
4645         .revalidate_disk= md_revalidate,
4646 };
4647
4648 static int md_thread(void * arg)
4649 {
4650         mdk_thread_t *thread = arg;
4651
4652         /*
4653          * md_thread is a 'system-thread', it's priority should be very
4654          * high. We avoid resource deadlocks individually in each
4655          * raid personality. (RAID5 does preallocation) We also use RR and
4656          * the very same RT priority as kswapd, thus we will never get
4657          * into a priority inversion deadlock.
4658          *
4659          * we definitely have to have equal or higher priority than
4660          * bdflush, otherwise bdflush will deadlock if there are too
4661          * many dirty RAID5 blocks.
4662          */
4663
4664         allow_signal(SIGKILL);
4665         while (!kthread_should_stop()) {
4666
4667                 /* We need to wait INTERRUPTIBLE so that
4668                  * we don't add to the load-average.
4669                  * That means we need to be sure no signals are
4670                  * pending
4671                  */
4672                 if (signal_pending(current))
4673                         flush_signals(current);
4674
4675                 wait_event_interruptible_timeout
4676                         (thread->wqueue,
4677                          test_bit(THREAD_WAKEUP, &thread->flags)
4678                          || kthread_should_stop(),
4679                          thread->timeout);
4680
4681                 clear_bit(THREAD_WAKEUP, &thread->flags);
4682
4683                 thread->run(thread->mddev);
4684         }
4685
4686         return 0;
4687 }
4688
4689 void md_wakeup_thread(mdk_thread_t *thread)
4690 {
4691         if (thread) {
4692                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4693                 set_bit(THREAD_WAKEUP, &thread->flags);
4694                 wake_up(&thread->wqueue);
4695         }
4696 }
4697
4698 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4699                                  const char *name)
4700 {
4701         mdk_thread_t *thread;
4702
4703         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4704         if (!thread)
4705                 return NULL;
4706
4707         init_waitqueue_head(&thread->wqueue);
4708
4709         thread->run = run;
4710         thread->mddev = mddev;
4711         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4712         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4713         if (IS_ERR(thread->tsk)) {
4714                 kfree(thread);
4715                 return NULL;
4716         }
4717         return thread;
4718 }
4719
4720 void md_unregister_thread(mdk_thread_t *thread)
4721 {
4722         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4723
4724         kthread_stop(thread->tsk);
4725         kfree(thread);
4726 }
4727
4728 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4729 {
4730         if (!mddev) {
4731                 MD_BUG();
4732                 return;
4733         }
4734
4735         if (!rdev || test_bit(Faulty, &rdev->flags))
4736                 return;
4737 /*
4738         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4739                 mdname(mddev),
4740                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4741                 __builtin_return_address(0),__builtin_return_address(1),
4742                 __builtin_return_address(2),__builtin_return_address(3));
4743 */
4744         if (!mddev->pers)
4745                 return;
4746         if (!mddev->pers->error_handler)
4747                 return;
4748         mddev->pers->error_handler(mddev,rdev);
4749         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4750         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4751         md_wakeup_thread(mddev->thread);
4752         md_new_event_inintr(mddev);
4753 }
4754
4755 /* seq_file implementation /proc/mdstat */
4756
4757 static void status_unused(struct seq_file *seq)
4758 {
4759         int i = 0;
4760         mdk_rdev_t *rdev;
4761         struct list_head *tmp;
4762
4763         seq_printf(seq, "unused devices: ");
4764
4765         ITERATE_RDEV_PENDING(rdev,tmp) {
4766                 char b[BDEVNAME_SIZE];
4767                 i++;
4768                 seq_printf(seq, "%s ",
4769                               bdevname(rdev->bdev,b));
4770         }
4771         if (!i)
4772                 seq_printf(seq, "<none>");
4773
4774         seq_printf(seq, "\n");
4775 }
4776
4777
4778 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4779 {
4780         sector_t max_blocks, resync, res;
4781         unsigned long dt, db, rt;
4782         int scale;
4783         unsigned int per_milli;
4784
4785         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4786
4787         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4788                 max_blocks = mddev->resync_max_sectors >> 1;
4789         else
4790                 max_blocks = mddev->size;
4791
4792         /*
4793          * Should not happen.
4794          */
4795         if (!max_blocks) {
4796                 MD_BUG();
4797                 return;
4798         }
4799         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4800          * in a sector_t, and (max_blocks>>scale) will fit in a
4801          * u32, as those are the requirements for sector_div.
4802          * Thus 'scale' must be at least 10
4803          */
4804         scale = 10;
4805         if (sizeof(sector_t) > sizeof(unsigned long)) {
4806                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4807                         scale++;
4808         }
4809         res = (resync>>scale)*1000;
4810         sector_div(res, (u32)((max_blocks>>scale)+1));
4811
4812         per_milli = res;
4813         {
4814                 int i, x = per_milli/50, y = 20-x;
4815                 seq_printf(seq, "[");
4816                 for (i = 0; i < x; i++)
4817                         seq_printf(seq, "=");
4818                 seq_printf(seq, ">");
4819                 for (i = 0; i < y; i++)
4820                         seq_printf(seq, ".");
4821                 seq_printf(seq, "] ");
4822         }
4823         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4824                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4825                     "reshape" :
4826                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4827                      "check" :
4828                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4829                       "resync" : "recovery"))),
4830                    per_milli/10, per_milli % 10,
4831                    (unsigned long long) resync,
4832                    (unsigned long long) max_blocks);
4833
4834         /*
4835          * We do not want to overflow, so the order of operands and
4836          * the * 100 / 100 trick are important. We do a +1 to be
4837          * safe against division by zero. We only estimate anyway.
4838          *
4839          * dt: time from mark until now
4840          * db: blocks written from mark until now
4841          * rt: remaining time
4842          */
4843         dt = ((jiffies - mddev->resync_mark) / HZ);
4844         if (!dt) dt++;
4845         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4846                 - mddev->resync_mark_cnt;
4847         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4848
4849         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4850
4851         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4852 }
4853
4854 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4855 {
4856         struct list_head *tmp;
4857         loff_t l = *pos;
4858         mddev_t *mddev;
4859
4860         if (l >= 0x10000)
4861                 return NULL;
4862         if (!l--)
4863                 /* header */
4864                 return (void*)1;
4865
4866         spin_lock(&all_mddevs_lock);
4867         list_for_each(tmp,&all_mddevs)
4868                 if (!l--) {
4869                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4870                         mddev_get(mddev);
4871                         spin_unlock(&all_mddevs_lock);
4872                         return mddev;
4873                 }
4874         spin_unlock(&all_mddevs_lock);
4875         if (!l--)
4876                 return (void*)2;/* tail */
4877         return NULL;
4878 }
4879
4880 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4881 {
4882         struct list_head *tmp;
4883         mddev_t *next_mddev, *mddev = v;
4884         
4885         ++*pos;
4886         if (v == (void*)2)
4887                 return NULL;
4888
4889         spin_lock(&all_mddevs_lock);
4890         if (v == (void*)1)
4891                 tmp = all_mddevs.next;
4892         else
4893                 tmp = mddev->all_mddevs.next;
4894         if (tmp != &all_mddevs)
4895                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4896         else {
4897                 next_mddev = (void*)2;
4898                 *pos = 0x10000;
4899         }               
4900         spin_unlock(&all_mddevs_lock);
4901
4902         if (v != (void*)1)
4903                 mddev_put(mddev);
4904         return next_mddev;
4905
4906 }
4907
4908 static void md_seq_stop(struct seq_file *seq, void *v)
4909 {
4910         mddev_t *mddev = v;
4911
4912         if (mddev && v != (void*)1 && v != (void*)2)
4913                 mddev_put(mddev);
4914 }
4915
4916 struct mdstat_info {
4917         int event;
4918 };
4919
4920 static int md_seq_show(struct seq_file *seq, void *v)
4921 {
4922         mddev_t *mddev = v;
4923         sector_t size;
4924         struct list_head *tmp2;
4925         mdk_rdev_t *rdev;
4926         struct mdstat_info *mi = seq->private;
4927         struct bitmap *bitmap;
4928
4929         if (v == (void*)1) {
4930                 struct mdk_personality *pers;
4931                 seq_printf(seq, "Personalities : ");
4932                 spin_lock(&pers_lock);
4933                 list_for_each_entry(pers, &pers_list, list)
4934                         seq_printf(seq, "[%s] ", pers->name);
4935
4936                 spin_unlock(&pers_lock);
4937                 seq_printf(seq, "\n");
4938                 mi->event = atomic_read(&md_event_count);
4939                 return 0;
4940         }
4941         if (v == (void*)2) {
4942                 status_unused(seq);
4943                 return 0;
4944         }
4945
4946         if (mddev_lock(mddev) < 0)
4947                 return -EINTR;
4948
4949         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4950                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4951                                                 mddev->pers ? "" : "in");
4952                 if (mddev->pers) {
4953                         if (mddev->ro==1)
4954                                 seq_printf(seq, " (read-only)");
4955                         if (mddev->ro==2)
4956                                 seq_printf(seq, "(auto-read-only)");
4957                         seq_printf(seq, " %s", mddev->pers->name);
4958                 }
4959
4960                 size = 0;
4961                 ITERATE_RDEV(mddev,rdev,tmp2) {
4962                         char b[BDEVNAME_SIZE];
4963                         seq_printf(seq, " %s[%d]",
4964                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4965                         if (test_bit(WriteMostly, &rdev->flags))
4966                                 seq_printf(seq, "(W)");
4967                         if (test_bit(Faulty, &rdev->flags)) {
4968                                 seq_printf(seq, "(F)");
4969                                 continue;
4970                         } else if (rdev->raid_disk < 0)
4971                                 seq_printf(seq, "(S)"); /* spare */
4972                         size += rdev->size;
4973                 }
4974
4975                 if (!list_empty(&mddev->disks)) {
4976                         if (mddev->pers)
4977                                 seq_printf(seq, "\n      %llu blocks",
4978                                         (unsigned long long)mddev->array_size);
4979                         else
4980                                 seq_printf(seq, "\n      %llu blocks",
4981                                         (unsigned long long)size);
4982                 }
4983                 if (mddev->persistent) {
4984                         if (mddev->major_version != 0 ||
4985                             mddev->minor_version != 90) {
4986                                 seq_printf(seq," super %d.%d",
4987                                            mddev->major_version,
4988                                            mddev->minor_version);
4989                         }
4990                 } else
4991                         seq_printf(seq, " super non-persistent");
4992
4993                 if (mddev->pers) {
4994                         mddev->pers->status (seq, mddev);
4995                         seq_printf(seq, "\n      ");
4996                         if (mddev->pers->sync_request) {
4997                                 if (mddev->curr_resync > 2) {
4998                                         status_resync (seq, mddev);
4999                                         seq_printf(seq, "\n      ");
5000                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5001                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5002                                 else if (mddev->recovery_cp < MaxSector)
5003                                         seq_printf(seq, "\tresync=PENDING\n      ");
5004                         }
5005                 } else
5006                         seq_printf(seq, "\n       ");
5007
5008                 if ((bitmap = mddev->bitmap)) {
5009                         unsigned long chunk_kb;
5010                         unsigned long flags;
5011                         spin_lock_irqsave(&bitmap->lock, flags);
5012                         chunk_kb = bitmap->chunksize >> 10;
5013                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5014                                 "%lu%s chunk",
5015                                 bitmap->pages - bitmap->missing_pages,
5016                                 bitmap->pages,
5017                                 (bitmap->pages - bitmap->missing_pages)
5018                                         << (PAGE_SHIFT - 10),
5019                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5020                                 chunk_kb ? "KB" : "B");
5021                         if (bitmap->file) {
5022                                 seq_printf(seq, ", file: ");
5023                                 seq_path(seq, bitmap->file->f_path.mnt,
5024                                          bitmap->file->f_path.dentry," \t\n");
5025                         }
5026
5027                         seq_printf(seq, "\n");
5028                         spin_unlock_irqrestore(&bitmap->lock, flags);
5029                 }
5030
5031                 seq_printf(seq, "\n");
5032         }
5033         mddev_unlock(mddev);
5034         
5035         return 0;
5036 }
5037
5038 static struct seq_operations md_seq_ops = {
5039         .start  = md_seq_start,
5040         .next   = md_seq_next,
5041         .stop   = md_seq_stop,
5042         .show   = md_seq_show,
5043 };
5044
5045 static int md_seq_open(struct inode *inode, struct file *file)
5046 {
5047         int error;
5048         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5049         if (mi == NULL)
5050                 return -ENOMEM;
5051
5052         error = seq_open(file, &md_seq_ops);
5053         if (error)
5054                 kfree(mi);
5055         else {
5056                 struct seq_file *p = file->private_data;
5057                 p->private = mi;
5058                 mi->event = atomic_read(&md_event_count);
5059         }
5060         return error;
5061 }
5062
5063 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5064 {
5065         struct seq_file *m = filp->private_data;
5066         struct mdstat_info *mi = m->private;
5067         int mask;
5068
5069         poll_wait(filp, &md_event_waiters, wait);
5070
5071         /* always allow read */
5072         mask = POLLIN | POLLRDNORM;
5073
5074         if (mi->event != atomic_read(&md_event_count))
5075                 mask |= POLLERR | POLLPRI;
5076         return mask;
5077 }
5078
5079 static const struct file_operations md_seq_fops = {
5080         .owner          = THIS_MODULE,
5081         .open           = md_seq_open,
5082         .read           = seq_read,
5083         .llseek         = seq_lseek,
5084         .release        = seq_release_private,
5085         .poll           = mdstat_poll,
5086 };
5087
5088 int register_md_personality(struct mdk_personality *p)
5089 {
5090         spin_lock(&pers_lock);
5091         list_add_tail(&p->list, &pers_list);
5092         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5093         spin_unlock(&pers_lock);
5094         return 0;
5095 }
5096
5097 int unregister_md_personality(struct mdk_personality *p)
5098 {
5099         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5100         spin_lock(&pers_lock);
5101         list_del_init(&p->list);
5102         spin_unlock(&pers_lock);
5103         return 0;
5104 }
5105
5106 static int is_mddev_idle(mddev_t *mddev)
5107 {
5108         mdk_rdev_t * rdev;
5109         struct list_head *tmp;
5110         int idle;
5111         long curr_events;
5112
5113         idle = 1;
5114         ITERATE_RDEV(mddev,rdev,tmp) {
5115                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5116                 curr_events = disk_stat_read(disk, sectors[0]) + 
5117                                 disk_stat_read(disk, sectors[1]) - 
5118                                 atomic_read(&disk->sync_io);
5119                 /* sync IO will cause sync_io to increase before the disk_stats
5120                  * as sync_io is counted when a request starts, and
5121                  * disk_stats is counted when it completes.
5122                  * So resync activity will cause curr_events to be smaller than
5123                  * when there was no such activity.
5124                  * non-sync IO will cause disk_stat to increase without
5125                  * increasing sync_io so curr_events will (eventually)
5126                  * be larger than it was before.  Once it becomes
5127                  * substantially larger, the test below will cause
5128                  * the array to appear non-idle, and resync will slow
5129                  * down.
5130                  * If there is a lot of outstanding resync activity when
5131                  * we set last_event to curr_events, then all that activity
5132                  * completing might cause the array to appear non-idle
5133                  * and resync will be slowed down even though there might
5134                  * not have been non-resync activity.  This will only
5135                  * happen once though.  'last_events' will soon reflect
5136                  * the state where there is little or no outstanding
5137                  * resync requests, and further resync activity will
5138                  * always make curr_events less than last_events.
5139                  *
5140                  */
5141                 if (curr_events - rdev->last_events > 4096) {
5142                         rdev->last_events = curr_events;
5143                         idle = 0;
5144                 }
5145         }
5146         return idle;
5147 }
5148
5149 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5150 {
5151         /* another "blocks" (512byte) blocks have been synced */
5152         atomic_sub(blocks, &mddev->recovery_active);
5153         wake_up(&mddev->recovery_wait);
5154         if (!ok) {
5155                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5156                 md_wakeup_thread(mddev->thread);
5157                 // stop recovery, signal do_sync ....
5158         }
5159 }
5160
5161
5162 /* md_write_start(mddev, bi)
5163  * If we need to update some array metadata (e.g. 'active' flag
5164  * in superblock) before writing, schedule a superblock update
5165  * and wait for it to complete.
5166  */
5167 void md_write_start(mddev_t *mddev, struct bio *bi)
5168 {
5169         if (bio_data_dir(bi) != WRITE)
5170                 return;
5171
5172         BUG_ON(mddev->ro == 1);
5173         if (mddev->ro == 2) {
5174                 /* need to switch to read/write */
5175                 mddev->ro = 0;
5176                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5177                 md_wakeup_thread(mddev->thread);
5178         }
5179         atomic_inc(&mddev->writes_pending);
5180         if (mddev->in_sync) {
5181                 spin_lock_irq(&mddev->write_lock);
5182                 if (mddev->in_sync) {
5183                         mddev->in_sync = 0;
5184                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5185                         md_wakeup_thread(mddev->thread);
5186                 }
5187                 spin_unlock_irq(&mddev->write_lock);
5188         }
5189         wait_event(mddev->sb_wait, mddev->flags==0);
5190 }
5191
5192 void md_write_end(mddev_t *mddev)
5193 {
5194         if (atomic_dec_and_test(&mddev->writes_pending)) {
5195                 if (mddev->safemode == 2)
5196                         md_wakeup_thread(mddev->thread);
5197                 else if (mddev->safemode_delay)
5198                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5199         }
5200 }
5201
5202 /* md_allow_write(mddev)
5203  * Calling this ensures that the array is marked 'active' so that writes
5204  * may proceed without blocking.  It is important to call this before
5205  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5206  * Must be called with mddev_lock held.
5207  */
5208 void md_allow_write(mddev_t *mddev)
5209 {
5210         if (!mddev->pers)
5211                 return;
5212         if (mddev->ro)
5213                 return;
5214
5215         spin_lock_irq(&mddev->write_lock);
5216         if (mddev->in_sync) {
5217                 mddev->in_sync = 0;
5218                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5219                 if (mddev->safemode_delay &&
5220                     mddev->safemode == 0)
5221                         mddev->safemode = 1;
5222                 spin_unlock_irq(&mddev->write_lock);
5223                 md_update_sb(mddev, 0);
5224         } else
5225                 spin_unlock_irq(&mddev->write_lock);
5226 }
5227 EXPORT_SYMBOL_GPL(md_allow_write);
5228
5229 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5230
5231 #define SYNC_MARKS      10
5232 #define SYNC_MARK_STEP  (3*HZ)
5233 void md_do_sync(mddev_t *mddev)
5234 {
5235         mddev_t *mddev2;
5236         unsigned int currspeed = 0,
5237                  window;
5238         sector_t max_sectors,j, io_sectors;
5239         unsigned long mark[SYNC_MARKS];
5240         sector_t mark_cnt[SYNC_MARKS];
5241         int last_mark,m;
5242         struct list_head *tmp;
5243         sector_t last_check;
5244         int skipped = 0;
5245         struct list_head *rtmp;
5246         mdk_rdev_t *rdev;
5247         char *desc;
5248
5249         /* just incase thread restarts... */
5250         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5251                 return;
5252         if (mddev->ro) /* never try to sync a read-only array */
5253                 return;
5254
5255         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5256                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5257                         desc = "data-check";
5258                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5259                         desc = "requested-resync";
5260                 else
5261                         desc = "resync";
5262         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5263                 desc = "reshape";
5264         else
5265                 desc = "recovery";
5266
5267         /* we overload curr_resync somewhat here.
5268          * 0 == not engaged in resync at all
5269          * 2 == checking that there is no conflict with another sync
5270          * 1 == like 2, but have yielded to allow conflicting resync to
5271          *              commense
5272          * other == active in resync - this many blocks
5273          *
5274          * Before starting a resync we must have set curr_resync to
5275          * 2, and then checked that every "conflicting" array has curr_resync
5276          * less than ours.  When we find one that is the same or higher
5277          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5278          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5279          * This will mean we have to start checking from the beginning again.
5280          *
5281          */
5282
5283         do {
5284                 mddev->curr_resync = 2;
5285
5286         try_again:
5287                 if (kthread_should_stop()) {
5288                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5289                         goto skip;
5290                 }
5291                 ITERATE_MDDEV(mddev2,tmp) {
5292                         if (mddev2 == mddev)
5293                                 continue;
5294                         if (mddev2->curr_resync && 
5295                             match_mddev_units(mddev,mddev2)) {
5296                                 DEFINE_WAIT(wq);
5297                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5298                                         /* arbitrarily yield */
5299                                         mddev->curr_resync = 1;
5300                                         wake_up(&resync_wait);
5301                                 }
5302                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5303                                         /* no need to wait here, we can wait the next
5304                                          * time 'round when curr_resync == 2
5305                                          */
5306                                         continue;
5307                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5308                                 if (!kthread_should_stop() &&
5309                                     mddev2->curr_resync >= mddev->curr_resync) {
5310                                         printk(KERN_INFO "md: delaying %s of %s"
5311                                                " until %s has finished (they"
5312                                                " share one or more physical units)\n",
5313                                                desc, mdname(mddev), mdname(mddev2));
5314                                         mddev_put(mddev2);
5315                                         schedule();
5316                                         finish_wait(&resync_wait, &wq);
5317                                         goto try_again;
5318                                 }
5319                                 finish_wait(&resync_wait, &wq);
5320                         }
5321                 }
5322         } while (mddev->curr_resync < 2);
5323
5324         j = 0;
5325         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5326                 /* resync follows the size requested by the personality,
5327                  * which defaults to physical size, but can be virtual size
5328                  */
5329                 max_sectors = mddev->resync_max_sectors;
5330                 mddev->resync_mismatches = 0;
5331                 /* we don't use the checkpoint if there's a bitmap */
5332                 if (!mddev->bitmap &&
5333                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5334                         j = mddev->recovery_cp;
5335         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5336                 max_sectors = mddev->size << 1;
5337         else {
5338                 /* recovery follows the physical size of devices */
5339                 max_sectors = mddev->size << 1;
5340                 j = MaxSector;
5341                 ITERATE_RDEV(mddev,rdev,rtmp)
5342                         if (rdev->raid_disk >= 0 &&
5343                             !test_bit(Faulty, &rdev->flags) &&
5344                             !test_bit(In_sync, &rdev->flags) &&
5345                             rdev->recovery_offset < j)
5346                                 j = rdev->recovery_offset;
5347         }
5348
5349         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5350         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5351                 " %d KB/sec/disk.\n", speed_min(mddev));
5352         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5353                "(but not more than %d KB/sec) for %s.\n",
5354                speed_max(mddev), desc);
5355
5356         is_mddev_idle(mddev); /* this also initializes IO event counters */
5357
5358         io_sectors = 0;
5359         for (m = 0; m < SYNC_MARKS; m++) {
5360                 mark[m] = jiffies;
5361                 mark_cnt[m] = io_sectors;
5362         }
5363         last_mark = 0;
5364         mddev->resync_mark = mark[last_mark];
5365         mddev->resync_mark_cnt = mark_cnt[last_mark];
5366
5367         /*
5368          * Tune reconstruction:
5369          */
5370         window = 32*(PAGE_SIZE/512);
5371         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5372                 window/2,(unsigned long long) max_sectors/2);
5373
5374         atomic_set(&mddev->recovery_active, 0);
5375         init_waitqueue_head(&mddev->recovery_wait);
5376         last_check = 0;
5377
5378         if (j>2) {
5379                 printk(KERN_INFO 
5380                        "md: resuming %s of %s from checkpoint.\n",
5381                        desc, mdname(mddev));
5382                 mddev->curr_resync = j;
5383         }
5384
5385         while (j < max_sectors) {
5386                 sector_t sectors;
5387
5388                 skipped = 0;
5389                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5390                                             currspeed < speed_min(mddev));
5391                 if (sectors == 0) {
5392                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5393                         goto out;
5394                 }
5395
5396                 if (!skipped) { /* actual IO requested */
5397                         io_sectors += sectors;
5398                         atomic_add(sectors, &mddev->recovery_active);
5399                 }
5400
5401                 j += sectors;
5402                 if (j>1) mddev->curr_resync = j;
5403                 mddev->curr_mark_cnt = io_sectors;
5404                 if (last_check == 0)
5405                         /* this is the earliers that rebuilt will be
5406                          * visible in /proc/mdstat
5407                          */
5408                         md_new_event(mddev);
5409
5410                 if (last_check + window > io_sectors || j == max_sectors)
5411                         continue;
5412
5413                 last_check = io_sectors;
5414
5415                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5416                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5417                         break;
5418
5419         repeat:
5420                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5421                         /* step marks */
5422                         int next = (last_mark+1) % SYNC_MARKS;
5423
5424                         mddev->resync_mark = mark[next];
5425                         mddev->resync_mark_cnt = mark_cnt[next];
5426                         mark[next] = jiffies;
5427                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5428                         last_mark = next;
5429                 }
5430
5431
5432                 if (kthread_should_stop()) {
5433                         /*
5434                          * got a signal, exit.
5435                          */
5436                         printk(KERN_INFO 
5437                                 "md: md_do_sync() got signal ... exiting\n");
5438                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5439                         goto out;
5440                 }
5441
5442                 /*
5443                  * this loop exits only if either when we are slower than
5444                  * the 'hard' speed limit, or the system was IO-idle for
5445                  * a jiffy.
5446                  * the system might be non-idle CPU-wise, but we only care
5447                  * about not overloading the IO subsystem. (things like an
5448                  * e2fsck being done on the RAID array should execute fast)
5449                  */
5450                 blk_unplug(mddev->queue);
5451                 cond_resched();
5452
5453                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5454                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5455
5456                 if (currspeed > speed_min(mddev)) {
5457                         if ((currspeed > speed_max(mddev)) ||
5458                                         !is_mddev_idle(mddev)) {
5459                                 msleep(500);
5460                                 goto repeat;
5461                         }
5462                 }
5463         }
5464         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5465         /*
5466          * this also signals 'finished resyncing' to md_stop
5467          */
5468  out:
5469         blk_unplug(mddev->queue);
5470
5471         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5472
5473         /* tell personality that we are finished */
5474         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5475
5476         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5477             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5478             mddev->curr_resync > 2) {
5479                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5480                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5481                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5482                                         printk(KERN_INFO
5483                                                "md: checkpointing %s of %s.\n",
5484                                                desc, mdname(mddev));
5485                                         mddev->recovery_cp = mddev->curr_resync;
5486                                 }
5487                         } else
5488                                 mddev->recovery_cp = MaxSector;
5489                 } else {
5490                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5491                                 mddev->curr_resync = MaxSector;
5492                         ITERATE_RDEV(mddev,rdev,rtmp)
5493                                 if (rdev->raid_disk >= 0 &&
5494                                     !test_bit(Faulty, &rdev->flags) &&
5495                                     !test_bit(In_sync, &rdev->flags) &&
5496                                     rdev->recovery_offset < mddev->curr_resync)
5497                                         rdev->recovery_offset = mddev->curr_resync;
5498                 }
5499         }
5500         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5501
5502  skip:
5503         mddev->curr_resync = 0;
5504         wake_up(&resync_wait);
5505         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5506         md_wakeup_thread(mddev->thread);
5507 }
5508 EXPORT_SYMBOL_GPL(md_do_sync);
5509
5510
5511 static int remove_and_add_spares(mddev_t *mddev)
5512 {
5513         mdk_rdev_t *rdev;
5514         struct list_head *rtmp;
5515         int spares = 0;
5516
5517         ITERATE_RDEV(mddev,rdev,rtmp)
5518                 if (rdev->raid_disk >= 0 &&
5519                     (test_bit(Faulty, &rdev->flags) ||
5520                      ! test_bit(In_sync, &rdev->flags)) &&
5521                     atomic_read(&rdev->nr_pending)==0) {
5522                         if (mddev->pers->hot_remove_disk(
5523                                     mddev, rdev->raid_disk)==0) {
5524                                 char nm[20];
5525                                 sprintf(nm,"rd%d", rdev->raid_disk);
5526                                 sysfs_remove_link(&mddev->kobj, nm);
5527                                 rdev->raid_disk = -1;
5528                         }
5529                 }
5530
5531         if (mddev->degraded) {
5532                 ITERATE_RDEV(mddev,rdev,rtmp)
5533                         if (rdev->raid_disk < 0
5534                             && !test_bit(Faulty, &rdev->flags)) {
5535                                 rdev->recovery_offset = 0;
5536                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5537                                         char nm[20];
5538                                         sprintf(nm, "rd%d", rdev->raid_disk);
5539                                         if (sysfs_create_link(&mddev->kobj,
5540                                                               &rdev->kobj, nm))
5541                                                 printk(KERN_WARNING
5542                                                        "md: cannot register "
5543                                                        "%s for %s\n",
5544                                                        nm, mdname(mddev));
5545                                         spares++;
5546                                         md_new_event(mddev);
5547                                 } else
5548                                         break;
5549                         }
5550         }
5551         return spares;
5552 }
5553 /*
5554  * This routine is regularly called by all per-raid-array threads to
5555  * deal with generic issues like resync and super-block update.
5556  * Raid personalities that don't have a thread (linear/raid0) do not
5557  * need this as they never do any recovery or update the superblock.
5558  *
5559  * It does not do any resync itself, but rather "forks" off other threads
5560  * to do that as needed.
5561  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5562  * "->recovery" and create a thread at ->sync_thread.
5563  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5564  * and wakeups up this thread which will reap the thread and finish up.
5565  * This thread also removes any faulty devices (with nr_pending == 0).
5566  *
5567  * The overall approach is:
5568  *  1/ if the superblock needs updating, update it.
5569  *  2/ If a recovery thread is running, don't do anything else.
5570  *  3/ If recovery has finished, clean up, possibly marking spares active.
5571  *  4/ If there are any faulty devices, remove them.
5572  *  5/ If array is degraded, try to add spares devices
5573  *  6/ If array has spares or is not in-sync, start a resync thread.
5574  */
5575 void md_check_recovery(mddev_t *mddev)
5576 {
5577         mdk_rdev_t *rdev;
5578         struct list_head *rtmp;
5579
5580
5581         if (mddev->bitmap)
5582                 bitmap_daemon_work(mddev->bitmap);
5583
5584         if (mddev->ro)
5585                 return;
5586
5587         if (signal_pending(current)) {
5588                 if (mddev->pers->sync_request) {
5589                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5590                                mdname(mddev));
5591                         mddev->safemode = 2;
5592                 }
5593                 flush_signals(current);
5594         }
5595
5596         if ( ! (
5597                 mddev->flags ||
5598                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5599                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5600                 (mddev->safemode == 1) ||
5601                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5602                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5603                 ))
5604                 return;
5605
5606         if (mddev_trylock(mddev)) {
5607                 int spares = 0;
5608
5609                 spin_lock_irq(&mddev->write_lock);
5610                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5611                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5612                         mddev->in_sync = 1;
5613                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5614                 }
5615                 if (mddev->safemode == 1)
5616                         mddev->safemode = 0;
5617                 spin_unlock_irq(&mddev->write_lock);
5618
5619                 if (mddev->flags)
5620                         md_update_sb(mddev, 0);
5621
5622
5623                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5624                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5625                         /* resync/recovery still happening */
5626                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5627                         goto unlock;
5628                 }
5629                 if (mddev->sync_thread) {
5630                         /* resync has finished, collect result */
5631                         md_unregister_thread(mddev->sync_thread);
5632                         mddev->sync_thread = NULL;
5633                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5634                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5635                                 /* success...*/
5636                                 /* activate any spares */
5637                                 mddev->pers->spare_active(mddev);
5638                         }
5639                         md_update_sb(mddev, 1);
5640
5641                         /* if array is no-longer degraded, then any saved_raid_disk
5642                          * information must be scrapped
5643                          */
5644                         if (!mddev->degraded)
5645                                 ITERATE_RDEV(mddev,rdev,rtmp)
5646                                         rdev->saved_raid_disk = -1;
5647
5648                         mddev->recovery = 0;
5649                         /* flag recovery needed just to double check */
5650                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5651                         md_new_event(mddev);
5652                         goto unlock;
5653                 }
5654                 /* Clear some bits that don't mean anything, but
5655                  * might be left set
5656                  */
5657                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5658                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5659                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5660                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5661
5662                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5663                         goto unlock;
5664                 /* no recovery is running.
5665                  * remove any failed drives, then
5666                  * add spares if possible.
5667                  * Spare are also removed and re-added, to allow
5668                  * the personality to fail the re-add.
5669                  */
5670
5671                 if (mddev->reshape_position != MaxSector) {
5672                         if (mddev->pers->check_reshape(mddev) != 0)
5673                                 /* Cannot proceed */
5674                                 goto unlock;
5675                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5676                 } else if ((spares = remove_and_add_spares(mddev))) {
5677                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5678                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5679                 } else if (mddev->recovery_cp < MaxSector) {
5680                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5681                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5682                         /* nothing to be done ... */
5683                         goto unlock;
5684
5685                 if (mddev->pers->sync_request) {
5686                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5687                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5688                                 /* We are adding a device or devices to an array
5689                                  * which has the bitmap stored on all devices.
5690                                  * So make sure all bitmap pages get written
5691                                  */
5692                                 bitmap_write_all(mddev->bitmap);
5693                         }
5694                         mddev->sync_thread = md_register_thread(md_do_sync,
5695                                                                 mddev,
5696                                                                 "%s_resync");
5697                         if (!mddev->sync_thread) {
5698                                 printk(KERN_ERR "%s: could not start resync"
5699                                         " thread...\n", 
5700                                         mdname(mddev));
5701                                 /* leave the spares where they are, it shouldn't hurt */
5702                                 mddev->recovery = 0;
5703                         } else
5704                                 md_wakeup_thread(mddev->sync_thread);
5705                         md_new_event(mddev);
5706                 }
5707         unlock:
5708                 mddev_unlock(mddev);
5709         }
5710 }
5711
5712 static int md_notify_reboot(struct notifier_block *this,
5713                             unsigned long code, void *x)
5714 {
5715         struct list_head *tmp;
5716         mddev_t *mddev;
5717
5718         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5719
5720                 printk(KERN_INFO "md: stopping all md devices.\n");
5721
5722                 ITERATE_MDDEV(mddev,tmp)
5723                         if (mddev_trylock(mddev)) {
5724                                 do_md_stop (mddev, 1);
5725                                 mddev_unlock(mddev);
5726                         }
5727                 /*
5728                  * certain more exotic SCSI devices are known to be
5729                  * volatile wrt too early system reboots. While the
5730                  * right place to handle this issue is the given
5731                  * driver, we do want to have a safe RAID driver ...
5732                  */
5733                 mdelay(1000*1);
5734         }
5735         return NOTIFY_DONE;
5736 }
5737
5738 static struct notifier_block md_notifier = {
5739         .notifier_call  = md_notify_reboot,
5740         .next           = NULL,
5741         .priority       = INT_MAX, /* before any real devices */
5742 };
5743
5744 static void md_geninit(void)
5745 {
5746         struct proc_dir_entry *p;
5747
5748         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5749
5750         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5751         if (p)
5752                 p->proc_fops = &md_seq_fops;
5753 }
5754
5755 static int __init md_init(void)
5756 {
5757         if (register_blkdev(MAJOR_NR, "md"))
5758                 return -1;
5759         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5760                 unregister_blkdev(MAJOR_NR, "md");
5761                 return -1;
5762         }
5763         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5764                             md_probe, NULL, NULL);
5765         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5766                             md_probe, NULL, NULL);
5767
5768         register_reboot_notifier(&md_notifier);
5769         raid_table_header = register_sysctl_table(raid_root_table);
5770
5771         md_geninit();
5772         return (0);
5773 }
5774
5775
5776 #ifndef MODULE
5777
5778 /*
5779  * Searches all registered partitions for autorun RAID arrays
5780  * at boot time.
5781  */
5782
5783 static LIST_HEAD(all_detected_devices);
5784 struct detected_devices_node {
5785         struct list_head list;
5786         dev_t dev;
5787 };
5788
5789 void md_autodetect_dev(dev_t dev)
5790 {
5791         struct detected_devices_node *node_detected_dev;
5792
5793         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5794         if (node_detected_dev) {
5795                 node_detected_dev->dev = dev;
5796                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5797         } else {
5798                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5799                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5800         }
5801 }
5802
5803
5804 static void autostart_arrays(int part)
5805 {
5806         mdk_rdev_t *rdev;
5807         struct detected_devices_node *node_detected_dev;
5808         dev_t dev;
5809         int i_scanned, i_passed;
5810
5811         i_scanned = 0;
5812         i_passed = 0;
5813
5814         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5815
5816         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
5817                 i_scanned++;
5818                 node_detected_dev = list_entry(all_detected_devices.next,
5819                                         struct detected_devices_node, list);
5820                 list_del(&node_detected_dev->list);
5821                 dev = node_detected_dev->dev;
5822                 kfree(node_detected_dev);
5823                 rdev = md_import_device(dev,0, 90);
5824                 if (IS_ERR(rdev))
5825                         continue;
5826
5827                 if (test_bit(Faulty, &rdev->flags)) {
5828                         MD_BUG();
5829                         continue;
5830                 }
5831                 list_add(&rdev->same_set, &pending_raid_disks);
5832                 i_passed++;
5833         }
5834
5835         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
5836                                                 i_scanned, i_passed);
5837
5838         autorun_devices(part);
5839 }
5840
5841 #endif /* !MODULE */
5842
5843 static __exit void md_exit(void)
5844 {
5845         mddev_t *mddev;
5846         struct list_head *tmp;
5847
5848         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5849         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5850
5851         unregister_blkdev(MAJOR_NR,"md");
5852         unregister_blkdev(mdp_major, "mdp");
5853         unregister_reboot_notifier(&md_notifier);
5854         unregister_sysctl_table(raid_table_header);
5855         remove_proc_entry("mdstat", NULL);
5856         ITERATE_MDDEV(mddev,tmp) {
5857                 struct gendisk *disk = mddev->gendisk;
5858                 if (!disk)
5859                         continue;
5860                 export_array(mddev);
5861                 del_gendisk(disk);
5862                 put_disk(disk);
5863                 mddev->gendisk = NULL;
5864                 mddev_put(mddev);
5865         }
5866 }
5867
5868 subsys_initcall(md_init);
5869 module_exit(md_exit)
5870
5871 static int get_ro(char *buffer, struct kernel_param *kp)
5872 {
5873         return sprintf(buffer, "%d", start_readonly);
5874 }
5875 static int set_ro(const char *val, struct kernel_param *kp)
5876 {
5877         char *e;
5878         int num = simple_strtoul(val, &e, 10);
5879         if (*val && (*e == '\0' || *e == '\n')) {
5880                 start_readonly = num;
5881                 return 0;
5882         }
5883         return -EINVAL;
5884 }
5885
5886 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5887 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5888
5889
5890 EXPORT_SYMBOL(register_md_personality);
5891 EXPORT_SYMBOL(unregister_md_personality);
5892 EXPORT_SYMBOL(md_error);
5893 EXPORT_SYMBOL(md_done_sync);
5894 EXPORT_SYMBOL(md_write_start);
5895 EXPORT_SYMBOL(md_write_end);
5896 EXPORT_SYMBOL(md_register_thread);
5897 EXPORT_SYMBOL(md_unregister_thread);
5898 EXPORT_SYMBOL(md_wakeup_thread);
5899 EXPORT_SYMBOL(md_check_recovery);
5900 MODULE_LICENSE("GPL");
5901 MODULE_ALIAS("md");
5902 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);