1c54f3c1cca74fca038b5b4452c9a0504e481e57
[linux-3.10.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/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 (request_queue_t *q, struct bio *bio)
215 {
216         bio_io_error(bio, bio->bi_size);
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 int super_written(struct bio *bio, unsigned int bytes_done, int error)
388 {
389         mdk_rdev_t *rdev = bio->bi_private;
390         mddev_t *mddev = rdev->mddev;
391         if (bio->bi_size)
392                 return 1;
393
394         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
395                 printk("md: super_written gets error=%d, uptodate=%d\n",
396                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
397                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
398                 md_error(mddev, rdev);
399         }
400
401         if (atomic_dec_and_test(&mddev->pending_writes))
402                 wake_up(&mddev->sb_wait);
403         bio_put(bio);
404         return 0;
405 }
406
407 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
408 {
409         struct bio *bio2 = bio->bi_private;
410         mdk_rdev_t *rdev = bio2->bi_private;
411         mddev_t *mddev = rdev->mddev;
412         if (bio->bi_size)
413                 return 1;
414
415         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
416             error == -EOPNOTSUPP) {
417                 unsigned long flags;
418                 /* barriers don't appear to be supported :-( */
419                 set_bit(BarriersNotsupp, &rdev->flags);
420                 mddev->barriers_work = 0;
421                 spin_lock_irqsave(&mddev->write_lock, flags);
422                 bio2->bi_next = mddev->biolist;
423                 mddev->biolist = bio2;
424                 spin_unlock_irqrestore(&mddev->write_lock, flags);
425                 wake_up(&mddev->sb_wait);
426                 bio_put(bio);
427                 return 0;
428         }
429         bio_put(bio2);
430         bio->bi_private = rdev;
431         return super_written(bio, bytes_done, error);
432 }
433
434 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
435                    sector_t sector, int size, struct page *page)
436 {
437         /* write first size bytes of page to sector of rdev
438          * Increment mddev->pending_writes before returning
439          * and decrement it on completion, waking up sb_wait
440          * if zero is reached.
441          * If an error occurred, call md_error
442          *
443          * As we might need to resubmit the request if BIO_RW_BARRIER
444          * causes ENOTSUPP, we allocate a spare bio...
445          */
446         struct bio *bio = bio_alloc(GFP_NOIO, 1);
447         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
448
449         bio->bi_bdev = rdev->bdev;
450         bio->bi_sector = sector;
451         bio_add_page(bio, page, size, 0);
452         bio->bi_private = rdev;
453         bio->bi_end_io = super_written;
454         bio->bi_rw = rw;
455
456         atomic_inc(&mddev->pending_writes);
457         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
458                 struct bio *rbio;
459                 rw |= (1<<BIO_RW_BARRIER);
460                 rbio = bio_clone(bio, GFP_NOIO);
461                 rbio->bi_private = bio;
462                 rbio->bi_end_io = super_written_barrier;
463                 submit_bio(rw, rbio);
464         } else
465                 submit_bio(rw, bio);
466 }
467
468 void md_super_wait(mddev_t *mddev)
469 {
470         /* wait for all superblock writes that were scheduled to complete.
471          * if any had to be retried (due to BARRIER problems), retry them
472          */
473         DEFINE_WAIT(wq);
474         for(;;) {
475                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
476                 if (atomic_read(&mddev->pending_writes)==0)
477                         break;
478                 while (mddev->biolist) {
479                         struct bio *bio;
480                         spin_lock_irq(&mddev->write_lock);
481                         bio = mddev->biolist;
482                         mddev->biolist = bio->bi_next ;
483                         bio->bi_next = NULL;
484                         spin_unlock_irq(&mddev->write_lock);
485                         submit_bio(bio->bi_rw, bio);
486                 }
487                 schedule();
488         }
489         finish_wait(&mddev->sb_wait, &wq);
490 }
491
492 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
493 {
494         if (bio->bi_size)
495                 return 1;
496
497         complete((struct completion*)bio->bi_private);
498         return 0;
499 }
500
501 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
502                    struct page *page, int rw)
503 {
504         struct bio *bio = bio_alloc(GFP_NOIO, 1);
505         struct completion event;
506         int ret;
507
508         rw |= (1 << BIO_RW_SYNC);
509
510         bio->bi_bdev = bdev;
511         bio->bi_sector = sector;
512         bio_add_page(bio, page, size, 0);
513         init_completion(&event);
514         bio->bi_private = &event;
515         bio->bi_end_io = bi_complete;
516         submit_bio(rw, bio);
517         wait_for_completion(&event);
518
519         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
520         bio_put(bio);
521         return ret;
522 }
523 EXPORT_SYMBOL_GPL(sync_page_io);
524
525 static int read_disk_sb(mdk_rdev_t * rdev, int size)
526 {
527         char b[BDEVNAME_SIZE];
528         if (!rdev->sb_page) {
529                 MD_BUG();
530                 return -EINVAL;
531         }
532         if (rdev->sb_loaded)
533                 return 0;
534
535
536         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
537                 goto fail;
538         rdev->sb_loaded = 1;
539         return 0;
540
541 fail:
542         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
543                 bdevname(rdev->bdev,b));
544         return -EINVAL;
545 }
546
547 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
548 {
549         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
550                 (sb1->set_uuid1 == sb2->set_uuid1) &&
551                 (sb1->set_uuid2 == sb2->set_uuid2) &&
552                 (sb1->set_uuid3 == sb2->set_uuid3))
553
554                 return 1;
555
556         return 0;
557 }
558
559
560 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
561 {
562         int ret;
563         mdp_super_t *tmp1, *tmp2;
564
565         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
566         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
567
568         if (!tmp1 || !tmp2) {
569                 ret = 0;
570                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
571                 goto abort;
572         }
573
574         *tmp1 = *sb1;
575         *tmp2 = *sb2;
576
577         /*
578          * nr_disks is not constant
579          */
580         tmp1->nr_disks = 0;
581         tmp2->nr_disks = 0;
582
583         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
584                 ret = 0;
585         else
586                 ret = 1;
587
588 abort:
589         kfree(tmp1);
590         kfree(tmp2);
591         return ret;
592 }
593
594
595 static u32 md_csum_fold(u32 csum)
596 {
597         csum = (csum & 0xffff) + (csum >> 16);
598         return (csum & 0xffff) + (csum >> 16);
599 }
600
601 static unsigned int calc_sb_csum(mdp_super_t * sb)
602 {
603         u64 newcsum = 0;
604         u32 *sb32 = (u32*)sb;
605         int i;
606         unsigned int disk_csum, csum;
607
608         disk_csum = sb->sb_csum;
609         sb->sb_csum = 0;
610
611         for (i = 0; i < MD_SB_BYTES/4 ; i++)
612                 newcsum += sb32[i];
613         csum = (newcsum & 0xffffffff) + (newcsum>>32);
614
615
616 #ifdef CONFIG_ALPHA
617         /* This used to use csum_partial, which was wrong for several
618          * reasons including that different results are returned on
619          * different architectures.  It isn't critical that we get exactly
620          * the same return value as before (we always csum_fold before
621          * testing, and that removes any differences).  However as we
622          * know that csum_partial always returned a 16bit value on
623          * alphas, do a fold to maximise conformity to previous behaviour.
624          */
625         sb->sb_csum = md_csum_fold(disk_csum);
626 #else
627         sb->sb_csum = disk_csum;
628 #endif
629         return csum;
630 }
631
632
633 /*
634  * Handle superblock details.
635  * We want to be able to handle multiple superblock formats
636  * so we have a common interface to them all, and an array of
637  * different handlers.
638  * We rely on user-space to write the initial superblock, and support
639  * reading and updating of superblocks.
640  * Interface methods are:
641  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
642  *      loads and validates a superblock on dev.
643  *      if refdev != NULL, compare superblocks on both devices
644  *    Return:
645  *      0 - dev has a superblock that is compatible with refdev
646  *      1 - dev has a superblock that is compatible and newer than refdev
647  *          so dev should be used as the refdev in future
648  *     -EINVAL superblock incompatible or invalid
649  *     -othererror e.g. -EIO
650  *
651  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
652  *      Verify that dev is acceptable into mddev.
653  *       The first time, mddev->raid_disks will be 0, and data from
654  *       dev should be merged in.  Subsequent calls check that dev
655  *       is new enough.  Return 0 or -EINVAL
656  *
657  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
658  *     Update the superblock for rdev with data in mddev
659  *     This does not write to disc.
660  *
661  */
662
663 struct super_type  {
664         char            *name;
665         struct module   *owner;
666         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
667         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
668         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
669 };
670
671 /*
672  * load_super for 0.90.0 
673  */
674 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
675 {
676         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
677         mdp_super_t *sb;
678         int ret;
679         sector_t sb_offset;
680
681         /*
682          * Calculate the position of the superblock,
683          * it's at the end of the disk.
684          *
685          * It also happens to be a multiple of 4Kb.
686          */
687         sb_offset = calc_dev_sboffset(rdev->bdev);
688         rdev->sb_offset = sb_offset;
689
690         ret = read_disk_sb(rdev, MD_SB_BYTES);
691         if (ret) return ret;
692
693         ret = -EINVAL;
694
695         bdevname(rdev->bdev, b);
696         sb = (mdp_super_t*)page_address(rdev->sb_page);
697
698         if (sb->md_magic != MD_SB_MAGIC) {
699                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
700                        b);
701                 goto abort;
702         }
703
704         if (sb->major_version != 0 ||
705             sb->minor_version < 90 ||
706             sb->minor_version > 91) {
707                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
708                         sb->major_version, sb->minor_version,
709                         b);
710                 goto abort;
711         }
712
713         if (sb->raid_disks <= 0)
714                 goto abort;
715
716         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
717                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
718                         b);
719                 goto abort;
720         }
721
722         rdev->preferred_minor = sb->md_minor;
723         rdev->data_offset = 0;
724         rdev->sb_size = MD_SB_BYTES;
725
726         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
727                 if (sb->level != 1 && sb->level != 4
728                     && sb->level != 5 && sb->level != 6
729                     && sb->level != 10) {
730                         /* FIXME use a better test */
731                         printk(KERN_WARNING
732                                "md: bitmaps not supported for this level.\n");
733                         goto abort;
734                 }
735         }
736
737         if (sb->level == LEVEL_MULTIPATH)
738                 rdev->desc_nr = -1;
739         else
740                 rdev->desc_nr = sb->this_disk.number;
741
742         if (refdev == 0)
743                 ret = 1;
744         else {
745                 __u64 ev1, ev2;
746                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
747                 if (!uuid_equal(refsb, sb)) {
748                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
749                                 b, bdevname(refdev->bdev,b2));
750                         goto abort;
751                 }
752                 if (!sb_equal(refsb, sb)) {
753                         printk(KERN_WARNING "md: %s has same UUID"
754                                " but different superblock to %s\n",
755                                b, bdevname(refdev->bdev, b2));
756                         goto abort;
757                 }
758                 ev1 = md_event(sb);
759                 ev2 = md_event(refsb);
760                 if (ev1 > ev2)
761                         ret = 1;
762                 else 
763                         ret = 0;
764         }
765         rdev->size = calc_dev_size(rdev, sb->chunk_size);
766
767         if (rdev->size < sb->size && sb->level > 1)
768                 /* "this cannot possibly happen" ... */
769                 ret = -EINVAL;
770
771  abort:
772         return ret;
773 }
774
775 /*
776  * validate_super for 0.90.0
777  */
778 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
779 {
780         mdp_disk_t *desc;
781         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
782         __u64 ev1 = md_event(sb);
783
784         rdev->raid_disk = -1;
785         rdev->flags = 0;
786         if (mddev->raid_disks == 0) {
787                 mddev->major_version = 0;
788                 mddev->minor_version = sb->minor_version;
789                 mddev->patch_version = sb->patch_version;
790                 mddev->persistent = ! sb->not_persistent;
791                 mddev->chunk_size = sb->chunk_size;
792                 mddev->ctime = sb->ctime;
793                 mddev->utime = sb->utime;
794                 mddev->level = sb->level;
795                 mddev->clevel[0] = 0;
796                 mddev->layout = sb->layout;
797                 mddev->raid_disks = sb->raid_disks;
798                 mddev->size = sb->size;
799                 mddev->events = ev1;
800                 mddev->bitmap_offset = 0;
801                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
802
803                 if (mddev->minor_version >= 91) {
804                         mddev->reshape_position = sb->reshape_position;
805                         mddev->delta_disks = sb->delta_disks;
806                         mddev->new_level = sb->new_level;
807                         mddev->new_layout = sb->new_layout;
808                         mddev->new_chunk = sb->new_chunk;
809                 } else {
810                         mddev->reshape_position = MaxSector;
811                         mddev->delta_disks = 0;
812                         mddev->new_level = mddev->level;
813                         mddev->new_layout = mddev->layout;
814                         mddev->new_chunk = mddev->chunk_size;
815                 }
816
817                 if (sb->state & (1<<MD_SB_CLEAN))
818                         mddev->recovery_cp = MaxSector;
819                 else {
820                         if (sb->events_hi == sb->cp_events_hi && 
821                                 sb->events_lo == sb->cp_events_lo) {
822                                 mddev->recovery_cp = sb->recovery_cp;
823                         } else
824                                 mddev->recovery_cp = 0;
825                 }
826
827                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
828                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
829                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
830                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
831
832                 mddev->max_disks = MD_SB_DISKS;
833
834                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
835                     mddev->bitmap_file == NULL)
836                         mddev->bitmap_offset = mddev->default_bitmap_offset;
837
838         } else if (mddev->pers == NULL) {
839                 /* Insist on good event counter while assembling */
840                 ++ev1;
841                 if (ev1 < mddev->events) 
842                         return -EINVAL;
843         } else if (mddev->bitmap) {
844                 /* if adding to array with a bitmap, then we can accept an
845                  * older device ... but not too old.
846                  */
847                 if (ev1 < mddev->bitmap->events_cleared)
848                         return 0;
849         } else {
850                 if (ev1 < mddev->events)
851                         /* just a hot-add of a new device, leave raid_disk at -1 */
852                         return 0;
853         }
854
855         if (mddev->level != LEVEL_MULTIPATH) {
856                 desc = sb->disks + rdev->desc_nr;
857
858                 if (desc->state & (1<<MD_DISK_FAULTY))
859                         set_bit(Faulty, &rdev->flags);
860                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
861                             desc->raid_disk < mddev->raid_disks */) {
862                         set_bit(In_sync, &rdev->flags);
863                         rdev->raid_disk = desc->raid_disk;
864                 }
865                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
866                         set_bit(WriteMostly, &rdev->flags);
867         } else /* MULTIPATH are always insync */
868                 set_bit(In_sync, &rdev->flags);
869         return 0;
870 }
871
872 /*
873  * sync_super for 0.90.0
874  */
875 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
876 {
877         mdp_super_t *sb;
878         struct list_head *tmp;
879         mdk_rdev_t *rdev2;
880         int next_spare = mddev->raid_disks;
881
882
883         /* make rdev->sb match mddev data..
884          *
885          * 1/ zero out disks
886          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
887          * 3/ any empty disks < next_spare become removed
888          *
889          * disks[0] gets initialised to REMOVED because
890          * we cannot be sure from other fields if it has
891          * been initialised or not.
892          */
893         int i;
894         int active=0, working=0,failed=0,spare=0,nr_disks=0;
895
896         rdev->sb_size = MD_SB_BYTES;
897
898         sb = (mdp_super_t*)page_address(rdev->sb_page);
899
900         memset(sb, 0, sizeof(*sb));
901
902         sb->md_magic = MD_SB_MAGIC;
903         sb->major_version = mddev->major_version;
904         sb->patch_version = mddev->patch_version;
905         sb->gvalid_words  = 0; /* ignored */
906         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
907         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
908         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
909         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
910
911         sb->ctime = mddev->ctime;
912         sb->level = mddev->level;
913         sb->size  = mddev->size;
914         sb->raid_disks = mddev->raid_disks;
915         sb->md_minor = mddev->md_minor;
916         sb->not_persistent = !mddev->persistent;
917         sb->utime = mddev->utime;
918         sb->state = 0;
919         sb->events_hi = (mddev->events>>32);
920         sb->events_lo = (u32)mddev->events;
921
922         if (mddev->reshape_position == MaxSector)
923                 sb->minor_version = 90;
924         else {
925                 sb->minor_version = 91;
926                 sb->reshape_position = mddev->reshape_position;
927                 sb->new_level = mddev->new_level;
928                 sb->delta_disks = mddev->delta_disks;
929                 sb->new_layout = mddev->new_layout;
930                 sb->new_chunk = mddev->new_chunk;
931         }
932         mddev->minor_version = sb->minor_version;
933         if (mddev->in_sync)
934         {
935                 sb->recovery_cp = mddev->recovery_cp;
936                 sb->cp_events_hi = (mddev->events>>32);
937                 sb->cp_events_lo = (u32)mddev->events;
938                 if (mddev->recovery_cp == MaxSector)
939                         sb->state = (1<< MD_SB_CLEAN);
940         } else
941                 sb->recovery_cp = 0;
942
943         sb->layout = mddev->layout;
944         sb->chunk_size = mddev->chunk_size;
945
946         if (mddev->bitmap && mddev->bitmap_file == NULL)
947                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
948
949         sb->disks[0].state = (1<<MD_DISK_REMOVED);
950         ITERATE_RDEV(mddev,rdev2,tmp) {
951                 mdp_disk_t *d;
952                 int desc_nr;
953                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
954                     && !test_bit(Faulty, &rdev2->flags))
955                         desc_nr = rdev2->raid_disk;
956                 else
957                         desc_nr = next_spare++;
958                 rdev2->desc_nr = desc_nr;
959                 d = &sb->disks[rdev2->desc_nr];
960                 nr_disks++;
961                 d->number = rdev2->desc_nr;
962                 d->major = MAJOR(rdev2->bdev->bd_dev);
963                 d->minor = MINOR(rdev2->bdev->bd_dev);
964                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
965                     && !test_bit(Faulty, &rdev2->flags))
966                         d->raid_disk = rdev2->raid_disk;
967                 else
968                         d->raid_disk = rdev2->desc_nr; /* compatibility */
969                 if (test_bit(Faulty, &rdev2->flags))
970                         d->state = (1<<MD_DISK_FAULTY);
971                 else if (test_bit(In_sync, &rdev2->flags)) {
972                         d->state = (1<<MD_DISK_ACTIVE);
973                         d->state |= (1<<MD_DISK_SYNC);
974                         active++;
975                         working++;
976                 } else {
977                         d->state = 0;
978                         spare++;
979                         working++;
980                 }
981                 if (test_bit(WriteMostly, &rdev2->flags))
982                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
983         }
984         /* now set the "removed" and "faulty" bits on any missing devices */
985         for (i=0 ; i < mddev->raid_disks ; i++) {
986                 mdp_disk_t *d = &sb->disks[i];
987                 if (d->state == 0 && d->number == 0) {
988                         d->number = i;
989                         d->raid_disk = i;
990                         d->state = (1<<MD_DISK_REMOVED);
991                         d->state |= (1<<MD_DISK_FAULTY);
992                         failed++;
993                 }
994         }
995         sb->nr_disks = nr_disks;
996         sb->active_disks = active;
997         sb->working_disks = working;
998         sb->failed_disks = failed;
999         sb->spare_disks = spare;
1000
1001         sb->this_disk = sb->disks[rdev->desc_nr];
1002         sb->sb_csum = calc_sb_csum(sb);
1003 }
1004
1005 /*
1006  * version 1 superblock
1007  */
1008
1009 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1010 {
1011         __le32 disk_csum;
1012         u32 csum;
1013         unsigned long long newcsum;
1014         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1015         __le32 *isuper = (__le32*)sb;
1016         int i;
1017
1018         disk_csum = sb->sb_csum;
1019         sb->sb_csum = 0;
1020         newcsum = 0;
1021         for (i=0; size>=4; size -= 4 )
1022                 newcsum += le32_to_cpu(*isuper++);
1023
1024         if (size == 2)
1025                 newcsum += le16_to_cpu(*(__le16*) isuper);
1026
1027         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1028         sb->sb_csum = disk_csum;
1029         return cpu_to_le32(csum);
1030 }
1031
1032 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1033 {
1034         struct mdp_superblock_1 *sb;
1035         int ret;
1036         sector_t sb_offset;
1037         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1038         int bmask;
1039
1040         /*
1041          * Calculate the position of the superblock.
1042          * It is always aligned to a 4K boundary and
1043          * depeding on minor_version, it can be:
1044          * 0: At least 8K, but less than 12K, from end of device
1045          * 1: At start of device
1046          * 2: 4K from start of device.
1047          */
1048         switch(minor_version) {
1049         case 0:
1050                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1051                 sb_offset -= 8*2;
1052                 sb_offset &= ~(sector_t)(4*2-1);
1053                 /* convert from sectors to K */
1054                 sb_offset /= 2;
1055                 break;
1056         case 1:
1057                 sb_offset = 0;
1058                 break;
1059         case 2:
1060                 sb_offset = 4;
1061                 break;
1062         default:
1063                 return -EINVAL;
1064         }
1065         rdev->sb_offset = sb_offset;
1066
1067         /* superblock is rarely larger than 1K, but it can be larger,
1068          * and it is safe to read 4k, so we do that
1069          */
1070         ret = read_disk_sb(rdev, 4096);
1071         if (ret) return ret;
1072
1073
1074         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1075
1076         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1077             sb->major_version != cpu_to_le32(1) ||
1078             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1079             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1080             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1081                 return -EINVAL;
1082
1083         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1084                 printk("md: invalid superblock checksum on %s\n",
1085                         bdevname(rdev->bdev,b));
1086                 return -EINVAL;
1087         }
1088         if (le64_to_cpu(sb->data_size) < 10) {
1089                 printk("md: data_size too small on %s\n",
1090                        bdevname(rdev->bdev,b));
1091                 return -EINVAL;
1092         }
1093         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1094                 if (sb->level != cpu_to_le32(1) &&
1095                     sb->level != cpu_to_le32(4) &&
1096                     sb->level != cpu_to_le32(5) &&
1097                     sb->level != cpu_to_le32(6) &&
1098                     sb->level != cpu_to_le32(10)) {
1099                         printk(KERN_WARNING
1100                                "md: bitmaps not supported for this level.\n");
1101                         return -EINVAL;
1102                 }
1103         }
1104
1105         rdev->preferred_minor = 0xffff;
1106         rdev->data_offset = le64_to_cpu(sb->data_offset);
1107         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1108
1109         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1110         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1111         if (rdev->sb_size & bmask)
1112                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1113
1114         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1115                 rdev->desc_nr = -1;
1116         else
1117                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1118
1119         if (refdev == 0)
1120                 ret = 1;
1121         else {
1122                 __u64 ev1, ev2;
1123                 struct mdp_superblock_1 *refsb = 
1124                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1125
1126                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1127                     sb->level != refsb->level ||
1128                     sb->layout != refsb->layout ||
1129                     sb->chunksize != refsb->chunksize) {
1130                         printk(KERN_WARNING "md: %s has strangely different"
1131                                 " superblock to %s\n",
1132                                 bdevname(rdev->bdev,b),
1133                                 bdevname(refdev->bdev,b2));
1134                         return -EINVAL;
1135                 }
1136                 ev1 = le64_to_cpu(sb->events);
1137                 ev2 = le64_to_cpu(refsb->events);
1138
1139                 if (ev1 > ev2)
1140                         ret = 1;
1141                 else
1142                         ret = 0;
1143         }
1144         if (minor_version) 
1145                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1146         else
1147                 rdev->size = rdev->sb_offset;
1148         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1149                 return -EINVAL;
1150         rdev->size = le64_to_cpu(sb->data_size)/2;
1151         if (le32_to_cpu(sb->chunksize))
1152                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1153
1154         if (le64_to_cpu(sb->size) > rdev->size*2)
1155                 return -EINVAL;
1156         return ret;
1157 }
1158
1159 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1160 {
1161         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1162         __u64 ev1 = le64_to_cpu(sb->events);
1163
1164         rdev->raid_disk = -1;
1165         rdev->flags = 0;
1166         if (mddev->raid_disks == 0) {
1167                 mddev->major_version = 1;
1168                 mddev->patch_version = 0;
1169                 mddev->persistent = 1;
1170                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1171                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1172                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1173                 mddev->level = le32_to_cpu(sb->level);
1174                 mddev->clevel[0] = 0;
1175                 mddev->layout = le32_to_cpu(sb->layout);
1176                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1177                 mddev->size = le64_to_cpu(sb->size)/2;
1178                 mddev->events = ev1;
1179                 mddev->bitmap_offset = 0;
1180                 mddev->default_bitmap_offset = 1024 >> 9;
1181                 
1182                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1183                 memcpy(mddev->uuid, sb->set_uuid, 16);
1184
1185                 mddev->max_disks =  (4096-256)/2;
1186
1187                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1188                     mddev->bitmap_file == NULL )
1189                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1190
1191                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1192                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1193                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1194                         mddev->new_level = le32_to_cpu(sb->new_level);
1195                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1196                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1197                 } else {
1198                         mddev->reshape_position = MaxSector;
1199                         mddev->delta_disks = 0;
1200                         mddev->new_level = mddev->level;
1201                         mddev->new_layout = mddev->layout;
1202                         mddev->new_chunk = mddev->chunk_size;
1203                 }
1204
1205         } else if (mddev->pers == NULL) {
1206                 /* Insist of good event counter while assembling */
1207                 ++ev1;
1208                 if (ev1 < mddev->events)
1209                         return -EINVAL;
1210         } else if (mddev->bitmap) {
1211                 /* If adding to array with a bitmap, then we can accept an
1212                  * older device, but not too old.
1213                  */
1214                 if (ev1 < mddev->bitmap->events_cleared)
1215                         return 0;
1216         } else {
1217                 if (ev1 < mddev->events)
1218                         /* just a hot-add of a new device, leave raid_disk at -1 */
1219                         return 0;
1220         }
1221         if (mddev->level != LEVEL_MULTIPATH) {
1222                 int role;
1223                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1224                 switch(role) {
1225                 case 0xffff: /* spare */
1226                         break;
1227                 case 0xfffe: /* faulty */
1228                         set_bit(Faulty, &rdev->flags);
1229                         break;
1230                 default:
1231                         if ((le32_to_cpu(sb->feature_map) &
1232                              MD_FEATURE_RECOVERY_OFFSET))
1233                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1234                         else
1235                                 set_bit(In_sync, &rdev->flags);
1236                         rdev->raid_disk = role;
1237                         break;
1238                 }
1239                 if (sb->devflags & WriteMostly1)
1240                         set_bit(WriteMostly, &rdev->flags);
1241         } else /* MULTIPATH are always insync */
1242                 set_bit(In_sync, &rdev->flags);
1243
1244         return 0;
1245 }
1246
1247 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1248 {
1249         struct mdp_superblock_1 *sb;
1250         struct list_head *tmp;
1251         mdk_rdev_t *rdev2;
1252         int max_dev, i;
1253         /* make rdev->sb match mddev and rdev data. */
1254
1255         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1256
1257         sb->feature_map = 0;
1258         sb->pad0 = 0;
1259         sb->recovery_offset = cpu_to_le64(0);
1260         memset(sb->pad1, 0, sizeof(sb->pad1));
1261         memset(sb->pad2, 0, sizeof(sb->pad2));
1262         memset(sb->pad3, 0, sizeof(sb->pad3));
1263
1264         sb->utime = cpu_to_le64((__u64)mddev->utime);
1265         sb->events = cpu_to_le64(mddev->events);
1266         if (mddev->in_sync)
1267                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1268         else
1269                 sb->resync_offset = cpu_to_le64(0);
1270
1271         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1272
1273         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1274         sb->size = cpu_to_le64(mddev->size<<1);
1275
1276         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1277                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1278                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1279         }
1280
1281         if (rdev->raid_disk >= 0 &&
1282             !test_bit(In_sync, &rdev->flags) &&
1283             rdev->recovery_offset > 0) {
1284                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1285                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1286         }
1287
1288         if (mddev->reshape_position != MaxSector) {
1289                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1290                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1291                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1292                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1293                 sb->new_level = cpu_to_le32(mddev->new_level);
1294                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1295         }
1296
1297         max_dev = 0;
1298         ITERATE_RDEV(mddev,rdev2,tmp)
1299                 if (rdev2->desc_nr+1 > max_dev)
1300                         max_dev = rdev2->desc_nr+1;
1301
1302         if (max_dev > le32_to_cpu(sb->max_dev))
1303                 sb->max_dev = cpu_to_le32(max_dev);
1304         for (i=0; i<max_dev;i++)
1305                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1306         
1307         ITERATE_RDEV(mddev,rdev2,tmp) {
1308                 i = rdev2->desc_nr;
1309                 if (test_bit(Faulty, &rdev2->flags))
1310                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1311                 else if (test_bit(In_sync, &rdev2->flags))
1312                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1313                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1314                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1315                 else
1316                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1317         }
1318
1319         sb->sb_csum = calc_sb_1_csum(sb);
1320 }
1321
1322
1323 static struct super_type super_types[] = {
1324         [0] = {
1325                 .name   = "0.90.0",
1326                 .owner  = THIS_MODULE,
1327                 .load_super     = super_90_load,
1328                 .validate_super = super_90_validate,
1329                 .sync_super     = super_90_sync,
1330         },
1331         [1] = {
1332                 .name   = "md-1",
1333                 .owner  = THIS_MODULE,
1334                 .load_super     = super_1_load,
1335                 .validate_super = super_1_validate,
1336                 .sync_super     = super_1_sync,
1337         },
1338 };
1339
1340 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1341 {
1342         struct list_head *tmp, *tmp2;
1343         mdk_rdev_t *rdev, *rdev2;
1344
1345         ITERATE_RDEV(mddev1,rdev,tmp)
1346                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1347                         if (rdev->bdev->bd_contains ==
1348                             rdev2->bdev->bd_contains)
1349                                 return 1;
1350
1351         return 0;
1352 }
1353
1354 static LIST_HEAD(pending_raid_disks);
1355
1356 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1357 {
1358         char b[BDEVNAME_SIZE];
1359         struct kobject *ko;
1360         char *s;
1361         int err;
1362
1363         if (rdev->mddev) {
1364                 MD_BUG();
1365                 return -EINVAL;
1366         }
1367         /* make sure rdev->size exceeds mddev->size */
1368         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1369                 if (mddev->pers) {
1370                         /* Cannot change size, so fail
1371                          * If mddev->level <= 0, then we don't care
1372                          * about aligning sizes (e.g. linear)
1373                          */
1374                         if (mddev->level > 0)
1375                                 return -ENOSPC;
1376                 } else
1377                         mddev->size = rdev->size;
1378         }
1379
1380         /* Verify rdev->desc_nr is unique.
1381          * If it is -1, assign a free number, else
1382          * check number is not in use
1383          */
1384         if (rdev->desc_nr < 0) {
1385                 int choice = 0;
1386                 if (mddev->pers) choice = mddev->raid_disks;
1387                 while (find_rdev_nr(mddev, choice))
1388                         choice++;
1389                 rdev->desc_nr = choice;
1390         } else {
1391                 if (find_rdev_nr(mddev, rdev->desc_nr))
1392                         return -EBUSY;
1393         }
1394         bdevname(rdev->bdev,b);
1395         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1396                 return -ENOMEM;
1397         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1398                 *s = '!';
1399                         
1400         rdev->mddev = mddev;
1401         printk(KERN_INFO "md: bind<%s>\n", b);
1402
1403         rdev->kobj.parent = &mddev->kobj;
1404         if ((err = kobject_add(&rdev->kobj)))
1405                 goto fail;
1406
1407         if (rdev->bdev->bd_part)
1408                 ko = &rdev->bdev->bd_part->kobj;
1409         else
1410                 ko = &rdev->bdev->bd_disk->kobj;
1411         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1412                 kobject_del(&rdev->kobj);
1413                 goto fail;
1414         }
1415         list_add(&rdev->same_set, &mddev->disks);
1416         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1417         return 0;
1418
1419  fail:
1420         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1421                b, mdname(mddev));
1422         return err;
1423 }
1424
1425 static void delayed_delete(struct work_struct *ws)
1426 {
1427         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1428         kobject_del(&rdev->kobj);
1429 }
1430
1431 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1432 {
1433         char b[BDEVNAME_SIZE];
1434         if (!rdev->mddev) {
1435                 MD_BUG();
1436                 return;
1437         }
1438         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1439         list_del_init(&rdev->same_set);
1440         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1441         rdev->mddev = NULL;
1442         sysfs_remove_link(&rdev->kobj, "block");
1443
1444         /* We need to delay this, otherwise we can deadlock when
1445          * writing to 'remove' to "dev/state"
1446          */
1447         INIT_WORK(&rdev->del_work, delayed_delete);
1448         schedule_work(&rdev->del_work);
1449 }
1450
1451 /*
1452  * prevent the device from being mounted, repartitioned or
1453  * otherwise reused by a RAID array (or any other kernel
1454  * subsystem), by bd_claiming the device.
1455  */
1456 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1457 {
1458         int err = 0;
1459         struct block_device *bdev;
1460         char b[BDEVNAME_SIZE];
1461
1462         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1463         if (IS_ERR(bdev)) {
1464                 printk(KERN_ERR "md: could not open %s.\n",
1465                         __bdevname(dev, b));
1466                 return PTR_ERR(bdev);
1467         }
1468         err = bd_claim(bdev, rdev);
1469         if (err) {
1470                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1471                         bdevname(bdev, b));
1472                 blkdev_put(bdev);
1473                 return err;
1474         }
1475         rdev->bdev = bdev;
1476         return err;
1477 }
1478
1479 static void unlock_rdev(mdk_rdev_t *rdev)
1480 {
1481         struct block_device *bdev = rdev->bdev;
1482         rdev->bdev = NULL;
1483         if (!bdev)
1484                 MD_BUG();
1485         bd_release(bdev);
1486         blkdev_put(bdev);
1487 }
1488
1489 void md_autodetect_dev(dev_t dev);
1490
1491 static void export_rdev(mdk_rdev_t * rdev)
1492 {
1493         char b[BDEVNAME_SIZE];
1494         printk(KERN_INFO "md: export_rdev(%s)\n",
1495                 bdevname(rdev->bdev,b));
1496         if (rdev->mddev)
1497                 MD_BUG();
1498         free_disk_sb(rdev);
1499         list_del_init(&rdev->same_set);
1500 #ifndef MODULE
1501         md_autodetect_dev(rdev->bdev->bd_dev);
1502 #endif
1503         unlock_rdev(rdev);
1504         kobject_put(&rdev->kobj);
1505 }
1506
1507 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1508 {
1509         unbind_rdev_from_array(rdev);
1510         export_rdev(rdev);
1511 }
1512
1513 static void export_array(mddev_t *mddev)
1514 {
1515         struct list_head *tmp;
1516         mdk_rdev_t *rdev;
1517
1518         ITERATE_RDEV(mddev,rdev,tmp) {
1519                 if (!rdev->mddev) {
1520                         MD_BUG();
1521                         continue;
1522                 }
1523                 kick_rdev_from_array(rdev);
1524         }
1525         if (!list_empty(&mddev->disks))
1526                 MD_BUG();
1527         mddev->raid_disks = 0;
1528         mddev->major_version = 0;
1529 }
1530
1531 static void print_desc(mdp_disk_t *desc)
1532 {
1533         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1534                 desc->major,desc->minor,desc->raid_disk,desc->state);
1535 }
1536
1537 static void print_sb(mdp_super_t *sb)
1538 {
1539         int i;
1540
1541         printk(KERN_INFO 
1542                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1543                 sb->major_version, sb->minor_version, sb->patch_version,
1544                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1545                 sb->ctime);
1546         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1547                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1548                 sb->md_minor, sb->layout, sb->chunk_size);
1549         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1550                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1551                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1552                 sb->failed_disks, sb->spare_disks,
1553                 sb->sb_csum, (unsigned long)sb->events_lo);
1554
1555         printk(KERN_INFO);
1556         for (i = 0; i < MD_SB_DISKS; i++) {
1557                 mdp_disk_t *desc;
1558
1559                 desc = sb->disks + i;
1560                 if (desc->number || desc->major || desc->minor ||
1561                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1562                         printk("     D %2d: ", i);
1563                         print_desc(desc);
1564                 }
1565         }
1566         printk(KERN_INFO "md:     THIS: ");
1567         print_desc(&sb->this_disk);
1568
1569 }
1570
1571 static void print_rdev(mdk_rdev_t *rdev)
1572 {
1573         char b[BDEVNAME_SIZE];
1574         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1575                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1576                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1577                 rdev->desc_nr);
1578         if (rdev->sb_loaded) {
1579                 printk(KERN_INFO "md: rdev superblock:\n");
1580                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1581         } else
1582                 printk(KERN_INFO "md: no rdev superblock!\n");
1583 }
1584
1585 static void md_print_devices(void)
1586 {
1587         struct list_head *tmp, *tmp2;
1588         mdk_rdev_t *rdev;
1589         mddev_t *mddev;
1590         char b[BDEVNAME_SIZE];
1591
1592         printk("\n");
1593         printk("md:     **********************************\n");
1594         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1595         printk("md:     **********************************\n");
1596         ITERATE_MDDEV(mddev,tmp) {
1597
1598                 if (mddev->bitmap)
1599                         bitmap_print_sb(mddev->bitmap);
1600                 else
1601                         printk("%s: ", mdname(mddev));
1602                 ITERATE_RDEV(mddev,rdev,tmp2)
1603                         printk("<%s>", bdevname(rdev->bdev,b));
1604                 printk("\n");
1605
1606                 ITERATE_RDEV(mddev,rdev,tmp2)
1607                         print_rdev(rdev);
1608         }
1609         printk("md:     **********************************\n");
1610         printk("\n");
1611 }
1612
1613
1614 static void sync_sbs(mddev_t * mddev, int nospares)
1615 {
1616         /* Update each superblock (in-memory image), but
1617          * if we are allowed to, skip spares which already
1618          * have the right event counter, or have one earlier
1619          * (which would mean they aren't being marked as dirty
1620          * with the rest of the array)
1621          */
1622         mdk_rdev_t *rdev;
1623         struct list_head *tmp;
1624
1625         ITERATE_RDEV(mddev,rdev,tmp) {
1626                 if (rdev->sb_events == mddev->events ||
1627                     (nospares &&
1628                      rdev->raid_disk < 0 &&
1629                      (rdev->sb_events&1)==0 &&
1630                      rdev->sb_events+1 == mddev->events)) {
1631                         /* Don't update this superblock */
1632                         rdev->sb_loaded = 2;
1633                 } else {
1634                         super_types[mddev->major_version].
1635                                 sync_super(mddev, rdev);
1636                         rdev->sb_loaded = 1;
1637                 }
1638         }
1639 }
1640
1641 static void md_update_sb(mddev_t * mddev, int force_change)
1642 {
1643         int err;
1644         struct list_head *tmp;
1645         mdk_rdev_t *rdev;
1646         int sync_req;
1647         int nospares = 0;
1648
1649 repeat:
1650         spin_lock_irq(&mddev->write_lock);
1651
1652         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1653         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1654                 force_change = 1;
1655         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1656                 /* just a clean<-> dirty transition, possibly leave spares alone,
1657                  * though if events isn't the right even/odd, we will have to do
1658                  * spares after all
1659                  */
1660                 nospares = 1;
1661         if (force_change)
1662                 nospares = 0;
1663         if (mddev->degraded)
1664                 /* If the array is degraded, then skipping spares is both
1665                  * dangerous and fairly pointless.
1666                  * Dangerous because a device that was removed from the array
1667                  * might have a event_count that still looks up-to-date,
1668                  * so it can be re-added without a resync.
1669                  * Pointless because if there are any spares to skip,
1670                  * then a recovery will happen and soon that array won't
1671                  * be degraded any more and the spare can go back to sleep then.
1672                  */
1673                 nospares = 0;
1674
1675         sync_req = mddev->in_sync;
1676         mddev->utime = get_seconds();
1677
1678         /* If this is just a dirty<->clean transition, and the array is clean
1679          * and 'events' is odd, we can roll back to the previous clean state */
1680         if (nospares
1681             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1682             && (mddev->events & 1)
1683             && mddev->events != 1)
1684                 mddev->events--;
1685         else {
1686                 /* otherwise we have to go forward and ... */
1687                 mddev->events ++;
1688                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1689                         /* .. if the array isn't clean, insist on an odd 'events' */
1690                         if ((mddev->events&1)==0) {
1691                                 mddev->events++;
1692                                 nospares = 0;
1693                         }
1694                 } else {
1695                         /* otherwise insist on an even 'events' (for clean states) */
1696                         if ((mddev->events&1)) {
1697                                 mddev->events++;
1698                                 nospares = 0;
1699                         }
1700                 }
1701         }
1702
1703         if (!mddev->events) {
1704                 /*
1705                  * oops, this 64-bit counter should never wrap.
1706                  * Either we are in around ~1 trillion A.C., assuming
1707                  * 1 reboot per second, or we have a bug:
1708                  */
1709                 MD_BUG();
1710                 mddev->events --;
1711         }
1712         sync_sbs(mddev, nospares);
1713
1714         /*
1715          * do not write anything to disk if using
1716          * nonpersistent superblocks
1717          */
1718         if (!mddev->persistent) {
1719                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1720                 spin_unlock_irq(&mddev->write_lock);
1721                 wake_up(&mddev->sb_wait);
1722                 return;
1723         }
1724         spin_unlock_irq(&mddev->write_lock);
1725
1726         dprintk(KERN_INFO 
1727                 "md: updating %s RAID superblock on device (in sync %d)\n",
1728                 mdname(mddev),mddev->in_sync);
1729
1730         err = bitmap_update_sb(mddev->bitmap);
1731         ITERATE_RDEV(mddev,rdev,tmp) {
1732                 char b[BDEVNAME_SIZE];
1733                 dprintk(KERN_INFO "md: ");
1734                 if (rdev->sb_loaded != 1)
1735                         continue; /* no noise on spare devices */
1736                 if (test_bit(Faulty, &rdev->flags))
1737                         dprintk("(skipping faulty ");
1738
1739                 dprintk("%s ", bdevname(rdev->bdev,b));
1740                 if (!test_bit(Faulty, &rdev->flags)) {
1741                         md_super_write(mddev,rdev,
1742                                        rdev->sb_offset<<1, rdev->sb_size,
1743                                        rdev->sb_page);
1744                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1745                                 bdevname(rdev->bdev,b),
1746                                 (unsigned long long)rdev->sb_offset);
1747                         rdev->sb_events = mddev->events;
1748
1749                 } else
1750                         dprintk(")\n");
1751                 if (mddev->level == LEVEL_MULTIPATH)
1752                         /* only need to write one superblock... */
1753                         break;
1754         }
1755         md_super_wait(mddev);
1756         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1757
1758         spin_lock_irq(&mddev->write_lock);
1759         if (mddev->in_sync != sync_req ||
1760             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1761                 /* have to write it out again */
1762                 spin_unlock_irq(&mddev->write_lock);
1763                 goto repeat;
1764         }
1765         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1766         spin_unlock_irq(&mddev->write_lock);
1767         wake_up(&mddev->sb_wait);
1768
1769 }
1770
1771 /* words written to sysfs files may, or my not, be \n terminated.
1772  * We want to accept with case. For this we use cmd_match.
1773  */
1774 static int cmd_match(const char *cmd, const char *str)
1775 {
1776         /* See if cmd, written into a sysfs file, matches
1777          * str.  They must either be the same, or cmd can
1778          * have a trailing newline
1779          */
1780         while (*cmd && *str && *cmd == *str) {
1781                 cmd++;
1782                 str++;
1783         }
1784         if (*cmd == '\n')
1785                 cmd++;
1786         if (*str || *cmd)
1787                 return 0;
1788         return 1;
1789 }
1790
1791 struct rdev_sysfs_entry {
1792         struct attribute attr;
1793         ssize_t (*show)(mdk_rdev_t *, char *);
1794         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1795 };
1796
1797 static ssize_t
1798 state_show(mdk_rdev_t *rdev, char *page)
1799 {
1800         char *sep = "";
1801         int len=0;
1802
1803         if (test_bit(Faulty, &rdev->flags)) {
1804                 len+= sprintf(page+len, "%sfaulty",sep);
1805                 sep = ",";
1806         }
1807         if (test_bit(In_sync, &rdev->flags)) {
1808                 len += sprintf(page+len, "%sin_sync",sep);
1809                 sep = ",";
1810         }
1811         if (test_bit(WriteMostly, &rdev->flags)) {
1812                 len += sprintf(page+len, "%swrite_mostly",sep);
1813                 sep = ",";
1814         }
1815         if (!test_bit(Faulty, &rdev->flags) &&
1816             !test_bit(In_sync, &rdev->flags)) {
1817                 len += sprintf(page+len, "%sspare", sep);
1818                 sep = ",";
1819         }
1820         return len+sprintf(page+len, "\n");
1821 }
1822
1823 static ssize_t
1824 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1825 {
1826         /* can write
1827          *  faulty  - simulates and error
1828          *  remove  - disconnects the device
1829          *  writemostly - sets write_mostly
1830          *  -writemostly - clears write_mostly
1831          */
1832         int err = -EINVAL;
1833         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1834                 md_error(rdev->mddev, rdev);
1835                 err = 0;
1836         } else if (cmd_match(buf, "remove")) {
1837                 if (rdev->raid_disk >= 0)
1838                         err = -EBUSY;
1839                 else {
1840                         mddev_t *mddev = rdev->mddev;
1841                         kick_rdev_from_array(rdev);
1842                         if (mddev->pers)
1843                                 md_update_sb(mddev, 1);
1844                         md_new_event(mddev);
1845                         err = 0;
1846                 }
1847         } else if (cmd_match(buf, "writemostly")) {
1848                 set_bit(WriteMostly, &rdev->flags);
1849                 err = 0;
1850         } else if (cmd_match(buf, "-writemostly")) {
1851                 clear_bit(WriteMostly, &rdev->flags);
1852                 err = 0;
1853         }
1854         return err ? err : len;
1855 }
1856 static struct rdev_sysfs_entry rdev_state =
1857 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1858
1859 static ssize_t
1860 super_show(mdk_rdev_t *rdev, char *page)
1861 {
1862         if (rdev->sb_loaded && rdev->sb_size) {
1863                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1864                 return rdev->sb_size;
1865         } else
1866                 return 0;
1867 }
1868 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1869
1870 static ssize_t
1871 errors_show(mdk_rdev_t *rdev, char *page)
1872 {
1873         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1874 }
1875
1876 static ssize_t
1877 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1878 {
1879         char *e;
1880         unsigned long n = simple_strtoul(buf, &e, 10);
1881         if (*buf && (*e == 0 || *e == '\n')) {
1882                 atomic_set(&rdev->corrected_errors, n);
1883                 return len;
1884         }
1885         return -EINVAL;
1886 }
1887 static struct rdev_sysfs_entry rdev_errors =
1888 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1889
1890 static ssize_t
1891 slot_show(mdk_rdev_t *rdev, char *page)
1892 {
1893         if (rdev->raid_disk < 0)
1894                 return sprintf(page, "none\n");
1895         else
1896                 return sprintf(page, "%d\n", rdev->raid_disk);
1897 }
1898
1899 static ssize_t
1900 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1901 {
1902         char *e;
1903         int slot = simple_strtoul(buf, &e, 10);
1904         if (strncmp(buf, "none", 4)==0)
1905                 slot = -1;
1906         else if (e==buf || (*e && *e!= '\n'))
1907                 return -EINVAL;
1908         if (rdev->mddev->pers)
1909                 /* Cannot set slot in active array (yet) */
1910                 return -EBUSY;
1911         if (slot >= rdev->mddev->raid_disks)
1912                 return -ENOSPC;
1913         rdev->raid_disk = slot;
1914         /* assume it is working */
1915         rdev->flags = 0;
1916         set_bit(In_sync, &rdev->flags);
1917         return len;
1918 }
1919
1920
1921 static struct rdev_sysfs_entry rdev_slot =
1922 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1923
1924 static ssize_t
1925 offset_show(mdk_rdev_t *rdev, char *page)
1926 {
1927         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1928 }
1929
1930 static ssize_t
1931 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1932 {
1933         char *e;
1934         unsigned long long offset = simple_strtoull(buf, &e, 10);
1935         if (e==buf || (*e && *e != '\n'))
1936                 return -EINVAL;
1937         if (rdev->mddev->pers)
1938                 return -EBUSY;
1939         rdev->data_offset = offset;
1940         return len;
1941 }
1942
1943 static struct rdev_sysfs_entry rdev_offset =
1944 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1945
1946 static ssize_t
1947 rdev_size_show(mdk_rdev_t *rdev, char *page)
1948 {
1949         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1950 }
1951
1952 static ssize_t
1953 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1954 {
1955         char *e;
1956         unsigned long long size = simple_strtoull(buf, &e, 10);
1957         if (e==buf || (*e && *e != '\n'))
1958                 return -EINVAL;
1959         if (rdev->mddev->pers)
1960                 return -EBUSY;
1961         rdev->size = size;
1962         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1963                 rdev->mddev->size = size;
1964         return len;
1965 }
1966
1967 static struct rdev_sysfs_entry rdev_size =
1968 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1969
1970 static struct attribute *rdev_default_attrs[] = {
1971         &rdev_state.attr,
1972         &rdev_super.attr,
1973         &rdev_errors.attr,
1974         &rdev_slot.attr,
1975         &rdev_offset.attr,
1976         &rdev_size.attr,
1977         NULL,
1978 };
1979 static ssize_t
1980 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1981 {
1982         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1983         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1984
1985         if (!entry->show)
1986                 return -EIO;
1987         return entry->show(rdev, page);
1988 }
1989
1990 static ssize_t
1991 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1992               const char *page, size_t length)
1993 {
1994         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1995         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1996
1997         if (!entry->store)
1998                 return -EIO;
1999         if (!capable(CAP_SYS_ADMIN))
2000                 return -EACCES;
2001         return entry->store(rdev, page, length);
2002 }
2003
2004 static void rdev_free(struct kobject *ko)
2005 {
2006         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2007         kfree(rdev);
2008 }
2009 static struct sysfs_ops rdev_sysfs_ops = {
2010         .show           = rdev_attr_show,
2011         .store          = rdev_attr_store,
2012 };
2013 static struct kobj_type rdev_ktype = {
2014         .release        = rdev_free,
2015         .sysfs_ops      = &rdev_sysfs_ops,
2016         .default_attrs  = rdev_default_attrs,
2017 };
2018
2019 /*
2020  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2021  *
2022  * mark the device faulty if:
2023  *
2024  *   - the device is nonexistent (zero size)
2025  *   - the device has no valid superblock
2026  *
2027  * a faulty rdev _never_ has rdev->sb set.
2028  */
2029 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2030 {
2031         char b[BDEVNAME_SIZE];
2032         int err;
2033         mdk_rdev_t *rdev;
2034         sector_t size;
2035
2036         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2037         if (!rdev) {
2038                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2039                 return ERR_PTR(-ENOMEM);
2040         }
2041
2042         if ((err = alloc_disk_sb(rdev)))
2043                 goto abort_free;
2044
2045         err = lock_rdev(rdev, newdev);
2046         if (err)
2047                 goto abort_free;
2048
2049         rdev->kobj.parent = NULL;
2050         rdev->kobj.ktype = &rdev_ktype;
2051         kobject_init(&rdev->kobj);
2052
2053         rdev->desc_nr = -1;
2054         rdev->saved_raid_disk = -1;
2055         rdev->raid_disk = -1;
2056         rdev->flags = 0;
2057         rdev->data_offset = 0;
2058         rdev->sb_events = 0;
2059         atomic_set(&rdev->nr_pending, 0);
2060         atomic_set(&rdev->read_errors, 0);
2061         atomic_set(&rdev->corrected_errors, 0);
2062
2063         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2064         if (!size) {
2065                 printk(KERN_WARNING 
2066                         "md: %s has zero or unknown size, marking faulty!\n",
2067                         bdevname(rdev->bdev,b));
2068                 err = -EINVAL;
2069                 goto abort_free;
2070         }
2071
2072         if (super_format >= 0) {
2073                 err = super_types[super_format].
2074                         load_super(rdev, NULL, super_minor);
2075                 if (err == -EINVAL) {
2076                         printk(KERN_WARNING 
2077                                 "md: %s has invalid sb, not importing!\n",
2078                                 bdevname(rdev->bdev,b));
2079                         goto abort_free;
2080                 }
2081                 if (err < 0) {
2082                         printk(KERN_WARNING 
2083                                 "md: could not read %s's sb, not importing!\n",
2084                                 bdevname(rdev->bdev,b));
2085                         goto abort_free;
2086                 }
2087         }
2088         INIT_LIST_HEAD(&rdev->same_set);
2089
2090         return rdev;
2091
2092 abort_free:
2093         if (rdev->sb_page) {
2094                 if (rdev->bdev)
2095                         unlock_rdev(rdev);
2096                 free_disk_sb(rdev);
2097         }
2098         kfree(rdev);
2099         return ERR_PTR(err);
2100 }
2101
2102 /*
2103  * Check a full RAID array for plausibility
2104  */
2105
2106
2107 static void analyze_sbs(mddev_t * mddev)
2108 {
2109         int i;
2110         struct list_head *tmp;
2111         mdk_rdev_t *rdev, *freshest;
2112         char b[BDEVNAME_SIZE];
2113
2114         freshest = NULL;
2115         ITERATE_RDEV(mddev,rdev,tmp)
2116                 switch (super_types[mddev->major_version].
2117                         load_super(rdev, freshest, mddev->minor_version)) {
2118                 case 1:
2119                         freshest = rdev;
2120                         break;
2121                 case 0:
2122                         break;
2123                 default:
2124                         printk( KERN_ERR \
2125                                 "md: fatal superblock inconsistency in %s"
2126                                 " -- removing from array\n", 
2127                                 bdevname(rdev->bdev,b));
2128                         kick_rdev_from_array(rdev);
2129                 }
2130
2131
2132         super_types[mddev->major_version].
2133                 validate_super(mddev, freshest);
2134
2135         i = 0;
2136         ITERATE_RDEV(mddev,rdev,tmp) {
2137                 if (rdev != freshest)
2138                         if (super_types[mddev->major_version].
2139                             validate_super(mddev, rdev)) {
2140                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2141                                         " from array!\n",
2142                                         bdevname(rdev->bdev,b));
2143                                 kick_rdev_from_array(rdev);
2144                                 continue;
2145                         }
2146                 if (mddev->level == LEVEL_MULTIPATH) {
2147                         rdev->desc_nr = i++;
2148                         rdev->raid_disk = rdev->desc_nr;
2149                         set_bit(In_sync, &rdev->flags);
2150                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2151                         rdev->raid_disk = -1;
2152                         clear_bit(In_sync, &rdev->flags);
2153                 }
2154         }
2155
2156
2157
2158         if (mddev->recovery_cp != MaxSector &&
2159             mddev->level >= 1)
2160                 printk(KERN_ERR "md: %s: raid array is not clean"
2161                        " -- starting background reconstruction\n",
2162                        mdname(mddev));
2163
2164 }
2165
2166 static ssize_t
2167 safe_delay_show(mddev_t *mddev, char *page)
2168 {
2169         int msec = (mddev->safemode_delay*1000)/HZ;
2170         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2171 }
2172 static ssize_t
2173 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2174 {
2175         int scale=1;
2176         int dot=0;
2177         int i;
2178         unsigned long msec;
2179         char buf[30];
2180         char *e;
2181         /* remove a period, and count digits after it */
2182         if (len >= sizeof(buf))
2183                 return -EINVAL;
2184         strlcpy(buf, cbuf, len);
2185         buf[len] = 0;
2186         for (i=0; i<len; i++) {
2187                 if (dot) {
2188                         if (isdigit(buf[i])) {
2189                                 buf[i-1] = buf[i];
2190                                 scale *= 10;
2191                         }
2192                         buf[i] = 0;
2193                 } else if (buf[i] == '.') {
2194                         dot=1;
2195                         buf[i] = 0;
2196                 }
2197         }
2198         msec = simple_strtoul(buf, &e, 10);
2199         if (e == buf || (*e && *e != '\n'))
2200                 return -EINVAL;
2201         msec = (msec * 1000) / scale;
2202         if (msec == 0)
2203                 mddev->safemode_delay = 0;
2204         else {
2205                 mddev->safemode_delay = (msec*HZ)/1000;
2206                 if (mddev->safemode_delay == 0)
2207                         mddev->safemode_delay = 1;
2208         }
2209         return len;
2210 }
2211 static struct md_sysfs_entry md_safe_delay =
2212 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2213
2214 static ssize_t
2215 level_show(mddev_t *mddev, char *page)
2216 {
2217         struct mdk_personality *p = mddev->pers;
2218         if (p)
2219                 return sprintf(page, "%s\n", p->name);
2220         else if (mddev->clevel[0])
2221                 return sprintf(page, "%s\n", mddev->clevel);
2222         else if (mddev->level != LEVEL_NONE)
2223                 return sprintf(page, "%d\n", mddev->level);
2224         else
2225                 return 0;
2226 }
2227
2228 static ssize_t
2229 level_store(mddev_t *mddev, const char *buf, size_t len)
2230 {
2231         int rv = len;
2232         if (mddev->pers)
2233                 return -EBUSY;
2234         if (len == 0)
2235                 return 0;
2236         if (len >= sizeof(mddev->clevel))
2237                 return -ENOSPC;
2238         strncpy(mddev->clevel, buf, len);
2239         if (mddev->clevel[len-1] == '\n')
2240                 len--;
2241         mddev->clevel[len] = 0;
2242         mddev->level = LEVEL_NONE;
2243         return rv;
2244 }
2245
2246 static struct md_sysfs_entry md_level =
2247 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2248
2249
2250 static ssize_t
2251 layout_show(mddev_t *mddev, char *page)
2252 {
2253         /* just a number, not meaningful for all levels */
2254         if (mddev->reshape_position != MaxSector &&
2255             mddev->layout != mddev->new_layout)
2256                 return sprintf(page, "%d (%d)\n",
2257                                mddev->new_layout, mddev->layout);
2258         return sprintf(page, "%d\n", mddev->layout);
2259 }
2260
2261 static ssize_t
2262 layout_store(mddev_t *mddev, const char *buf, size_t len)
2263 {
2264         char *e;
2265         unsigned long n = simple_strtoul(buf, &e, 10);
2266
2267         if (!*buf || (*e && *e != '\n'))
2268                 return -EINVAL;
2269
2270         if (mddev->pers)
2271                 return -EBUSY;
2272         if (mddev->reshape_position != MaxSector)
2273                 mddev->new_layout = n;
2274         else
2275                 mddev->layout = n;
2276         return len;
2277 }
2278 static struct md_sysfs_entry md_layout =
2279 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2280
2281
2282 static ssize_t
2283 raid_disks_show(mddev_t *mddev, char *page)
2284 {
2285         if (mddev->raid_disks == 0)
2286                 return 0;
2287         if (mddev->reshape_position != MaxSector &&
2288             mddev->delta_disks != 0)
2289                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2290                                mddev->raid_disks - mddev->delta_disks);
2291         return sprintf(page, "%d\n", mddev->raid_disks);
2292 }
2293
2294 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2295
2296 static ssize_t
2297 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2298 {
2299         char *e;
2300         int rv = 0;
2301         unsigned long n = simple_strtoul(buf, &e, 10);
2302
2303         if (!*buf || (*e && *e != '\n'))
2304                 return -EINVAL;
2305
2306         if (mddev->pers)
2307                 rv = update_raid_disks(mddev, n);
2308         else if (mddev->reshape_position != MaxSector) {
2309                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2310                 mddev->delta_disks = n - olddisks;
2311                 mddev->raid_disks = n;
2312         } else
2313                 mddev->raid_disks = n;
2314         return rv ? rv : len;
2315 }
2316 static struct md_sysfs_entry md_raid_disks =
2317 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2318
2319 static ssize_t
2320 chunk_size_show(mddev_t *mddev, char *page)
2321 {
2322         if (mddev->reshape_position != MaxSector &&
2323             mddev->chunk_size != mddev->new_chunk)
2324                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2325                                mddev->chunk_size);
2326         return sprintf(page, "%d\n", mddev->chunk_size);
2327 }
2328
2329 static ssize_t
2330 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2331 {
2332         /* can only set chunk_size if array is not yet active */
2333         char *e;
2334         unsigned long n = simple_strtoul(buf, &e, 10);
2335
2336         if (!*buf || (*e && *e != '\n'))
2337                 return -EINVAL;
2338
2339         if (mddev->pers)
2340                 return -EBUSY;
2341         else if (mddev->reshape_position != MaxSector)
2342                 mddev->new_chunk = n;
2343         else
2344                 mddev->chunk_size = n;
2345         return len;
2346 }
2347 static struct md_sysfs_entry md_chunk_size =
2348 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2349
2350 static ssize_t
2351 resync_start_show(mddev_t *mddev, char *page)
2352 {
2353         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2354 }
2355
2356 static ssize_t
2357 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2358 {
2359         /* can only set chunk_size if array is not yet active */
2360         char *e;
2361         unsigned long long n = simple_strtoull(buf, &e, 10);
2362
2363         if (mddev->pers)
2364                 return -EBUSY;
2365         if (!*buf || (*e && *e != '\n'))
2366                 return -EINVAL;
2367
2368         mddev->recovery_cp = n;
2369         return len;
2370 }
2371 static struct md_sysfs_entry md_resync_start =
2372 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2373
2374 /*
2375  * The array state can be:
2376  *
2377  * clear
2378  *     No devices, no size, no level
2379  *     Equivalent to STOP_ARRAY ioctl
2380  * inactive
2381  *     May have some settings, but array is not active
2382  *        all IO results in error
2383  *     When written, doesn't tear down array, but just stops it
2384  * suspended (not supported yet)
2385  *     All IO requests will block. The array can be reconfigured.
2386  *     Writing this, if accepted, will block until array is quiessent
2387  * readonly
2388  *     no resync can happen.  no superblocks get written.
2389  *     write requests fail
2390  * read-auto
2391  *     like readonly, but behaves like 'clean' on a write request.
2392  *
2393  * clean - no pending writes, but otherwise active.
2394  *     When written to inactive array, starts without resync
2395  *     If a write request arrives then
2396  *       if metadata is known, mark 'dirty' and switch to 'active'.
2397  *       if not known, block and switch to write-pending
2398  *     If written to an active array that has pending writes, then fails.
2399  * active
2400  *     fully active: IO and resync can be happening.
2401  *     When written to inactive array, starts with resync
2402  *
2403  * write-pending
2404  *     clean, but writes are blocked waiting for 'active' to be written.
2405  *
2406  * active-idle
2407  *     like active, but no writes have been seen for a while (100msec).
2408  *
2409  */
2410 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2411                    write_pending, active_idle, bad_word};
2412 static char *array_states[] = {
2413         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2414         "write-pending", "active-idle", NULL };
2415
2416 static int match_word(const char *word, char **list)
2417 {
2418         int n;
2419         for (n=0; list[n]; n++)
2420                 if (cmd_match(word, list[n]))
2421                         break;
2422         return n;
2423 }
2424
2425 static ssize_t
2426 array_state_show(mddev_t *mddev, char *page)
2427 {
2428         enum array_state st = inactive;
2429
2430         if (mddev->pers)
2431                 switch(mddev->ro) {
2432                 case 1:
2433                         st = readonly;
2434                         break;
2435                 case 2:
2436                         st = read_auto;
2437                         break;
2438                 case 0:
2439                         if (mddev->in_sync)
2440                                 st = clean;
2441                         else if (mddev->safemode)
2442                                 st = active_idle;
2443                         else
2444                                 st = active;
2445                 }
2446         else {
2447                 if (list_empty(&mddev->disks) &&
2448                     mddev->raid_disks == 0 &&
2449                     mddev->size == 0)
2450                         st = clear;
2451                 else
2452                         st = inactive;
2453         }
2454         return sprintf(page, "%s\n", array_states[st]);
2455 }
2456
2457 static int do_md_stop(mddev_t * mddev, int ro);
2458 static int do_md_run(mddev_t * mddev);
2459 static int restart_array(mddev_t *mddev);
2460
2461 static ssize_t
2462 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2463 {
2464         int err = -EINVAL;
2465         enum array_state st = match_word(buf, array_states);
2466         switch(st) {
2467         case bad_word:
2468                 break;
2469         case clear:
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, 0);
2475                 }
2476                 break;
2477         case inactive:
2478                 /* stopping an active array */
2479                 if (mddev->pers) {
2480                         if (atomic_read(&mddev->active) > 1)
2481                                 return -EBUSY;
2482                         err = do_md_stop(mddev, 2);
2483                 }
2484                 break;
2485         case suspended:
2486                 break; /* not supported yet */
2487         case readonly:
2488                 if (mddev->pers)
2489                         err = do_md_stop(mddev, 1);
2490                 else {
2491                         mddev->ro = 1;
2492                         err = do_md_run(mddev);
2493                 }
2494                 break;
2495         case read_auto:
2496                 /* stopping an active array */
2497                 if (mddev->pers) {
2498                         err = do_md_stop(mddev, 1);
2499                         if (err == 0)
2500                                 mddev->ro = 2; /* FIXME mark devices writable */
2501                 } else {
2502                         mddev->ro = 2;
2503                         err = do_md_run(mddev);
2504                 }
2505                 break;
2506         case clean:
2507                 if (mddev->pers) {
2508                         restart_array(mddev);
2509                         spin_lock_irq(&mddev->write_lock);
2510                         if (atomic_read(&mddev->writes_pending) == 0) {
2511                                 mddev->in_sync = 1;
2512                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2513                         }
2514                         spin_unlock_irq(&mddev->write_lock);
2515                 } else {
2516                         mddev->ro = 0;
2517                         mddev->recovery_cp = MaxSector;
2518                         err = do_md_run(mddev);
2519                 }
2520                 break;
2521         case active:
2522                 if (mddev->pers) {
2523                         restart_array(mddev);
2524                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2525                         wake_up(&mddev->sb_wait);
2526                         err = 0;
2527                 } else {
2528                         mddev->ro = 0;
2529                         err = do_md_run(mddev);
2530                 }
2531                 break;
2532         case write_pending:
2533         case active_idle:
2534                 /* these cannot be set */
2535                 break;
2536         }
2537         if (err)
2538                 return err;
2539         else
2540                 return len;
2541 }
2542 static struct md_sysfs_entry md_array_state =
2543 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2544
2545 static ssize_t
2546 null_show(mddev_t *mddev, char *page)
2547 {
2548         return -EINVAL;
2549 }
2550
2551 static ssize_t
2552 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2553 {
2554         /* buf must be %d:%d\n? giving major and minor numbers */
2555         /* The new device is added to the array.
2556          * If the array has a persistent superblock, we read the
2557          * superblock to initialise info and check validity.
2558          * Otherwise, only checking done is that in bind_rdev_to_array,
2559          * which mainly checks size.
2560          */
2561         char *e;
2562         int major = simple_strtoul(buf, &e, 10);
2563         int minor;
2564         dev_t dev;
2565         mdk_rdev_t *rdev;
2566         int err;
2567
2568         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2569                 return -EINVAL;
2570         minor = simple_strtoul(e+1, &e, 10);
2571         if (*e && *e != '\n')
2572                 return -EINVAL;
2573         dev = MKDEV(major, minor);
2574         if (major != MAJOR(dev) ||
2575             minor != MINOR(dev))
2576                 return -EOVERFLOW;
2577
2578
2579         if (mddev->persistent) {
2580                 rdev = md_import_device(dev, mddev->major_version,
2581                                         mddev->minor_version);
2582                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2583                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2584                                                        mdk_rdev_t, same_set);
2585                         err = super_types[mddev->major_version]
2586                                 .load_super(rdev, rdev0, mddev->minor_version);
2587                         if (err < 0)
2588                                 goto out;
2589                 }
2590         } else
2591                 rdev = md_import_device(dev, -1, -1);
2592
2593         if (IS_ERR(rdev))
2594                 return PTR_ERR(rdev);
2595         err = bind_rdev_to_array(rdev, mddev);
2596  out:
2597         if (err)
2598                 export_rdev(rdev);
2599         return err ? err : len;
2600 }
2601
2602 static struct md_sysfs_entry md_new_device =
2603 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2604
2605 static ssize_t
2606 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2607 {
2608         char *end;
2609         unsigned long chunk, end_chunk;
2610
2611         if (!mddev->bitmap)
2612                 goto out;
2613         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2614         while (*buf) {
2615                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2616                 if (buf == end) break;
2617                 if (*end == '-') { /* range */
2618                         buf = end + 1;
2619                         end_chunk = simple_strtoul(buf, &end, 0);
2620                         if (buf == end) break;
2621                 }
2622                 if (*end && !isspace(*end)) break;
2623                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2624                 buf = end;
2625                 while (isspace(*buf)) buf++;
2626         }
2627         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2628 out:
2629         return len;
2630 }
2631
2632 static struct md_sysfs_entry md_bitmap =
2633 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2634
2635 static ssize_t
2636 size_show(mddev_t *mddev, char *page)
2637 {
2638         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2639 }
2640
2641 static int update_size(mddev_t *mddev, unsigned long size);
2642
2643 static ssize_t
2644 size_store(mddev_t *mddev, const char *buf, size_t len)
2645 {
2646         /* If array is inactive, we can reduce the component size, but
2647          * not increase it (except from 0).
2648          * If array is active, we can try an on-line resize
2649          */
2650         char *e;
2651         int err = 0;
2652         unsigned long long size = simple_strtoull(buf, &e, 10);
2653         if (!*buf || *buf == '\n' ||
2654             (*e && *e != '\n'))
2655                 return -EINVAL;
2656
2657         if (mddev->pers) {
2658                 err = update_size(mddev, size);
2659                 md_update_sb(mddev, 1);
2660         } else {
2661                 if (mddev->size == 0 ||
2662                     mddev->size > size)
2663                         mddev->size = size;
2664                 else
2665                         err = -ENOSPC;
2666         }
2667         return err ? err : len;
2668 }
2669
2670 static struct md_sysfs_entry md_size =
2671 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2672
2673
2674 /* Metdata version.
2675  * This is either 'none' for arrays with externally managed metadata,
2676  * or N.M for internally known formats
2677  */
2678 static ssize_t
2679 metadata_show(mddev_t *mddev, char *page)
2680 {
2681         if (mddev->persistent)
2682                 return sprintf(page, "%d.%d\n",
2683                                mddev->major_version, mddev->minor_version);
2684         else
2685                 return sprintf(page, "none\n");
2686 }
2687
2688 static ssize_t
2689 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2690 {
2691         int major, minor;
2692         char *e;
2693         if (!list_empty(&mddev->disks))
2694                 return -EBUSY;
2695
2696         if (cmd_match(buf, "none")) {
2697                 mddev->persistent = 0;
2698                 mddev->major_version = 0;
2699                 mddev->minor_version = 90;
2700                 return len;
2701         }
2702         major = simple_strtoul(buf, &e, 10);
2703         if (e==buf || *e != '.')
2704                 return -EINVAL;
2705         buf = e+1;
2706         minor = simple_strtoul(buf, &e, 10);
2707         if (e==buf || (*e && *e != '\n') )
2708                 return -EINVAL;
2709         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2710                 return -ENOENT;
2711         mddev->major_version = major;
2712         mddev->minor_version = minor;
2713         mddev->persistent = 1;
2714         return len;
2715 }
2716
2717 static struct md_sysfs_entry md_metadata =
2718 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2719
2720 static ssize_t
2721 action_show(mddev_t *mddev, char *page)
2722 {
2723         char *type = "idle";
2724         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2725             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2726                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2727                         type = "reshape";
2728                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2729                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2730                                 type = "resync";
2731                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2732                                 type = "check";
2733                         else
2734                                 type = "repair";
2735                 } else
2736                         type = "recover";
2737         }
2738         return sprintf(page, "%s\n", type);
2739 }
2740
2741 static ssize_t
2742 action_store(mddev_t *mddev, const char *page, size_t len)
2743 {
2744         if (!mddev->pers || !mddev->pers->sync_request)
2745                 return -EINVAL;
2746
2747         if (cmd_match(page, "idle")) {
2748                 if (mddev->sync_thread) {
2749                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2750                         md_unregister_thread(mddev->sync_thread);
2751                         mddev->sync_thread = NULL;
2752                         mddev->recovery = 0;
2753                 }
2754         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2755                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2756                 return -EBUSY;
2757         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2758                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2759         else if (cmd_match(page, "reshape")) {
2760                 int err;
2761                 if (mddev->pers->start_reshape == NULL)
2762                         return -EINVAL;
2763                 err = mddev->pers->start_reshape(mddev);
2764                 if (err)
2765                         return err;
2766         } else {
2767                 if (cmd_match(page, "check"))
2768                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2769                 else if (!cmd_match(page, "repair"))
2770                         return -EINVAL;
2771                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2772                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2773         }
2774         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2775         md_wakeup_thread(mddev->thread);
2776         return len;
2777 }
2778
2779 static ssize_t
2780 mismatch_cnt_show(mddev_t *mddev, char *page)
2781 {
2782         return sprintf(page, "%llu\n",
2783                        (unsigned long long) mddev->resync_mismatches);
2784 }
2785
2786 static struct md_sysfs_entry md_scan_mode =
2787 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2788
2789
2790 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2791
2792 static ssize_t
2793 sync_min_show(mddev_t *mddev, char *page)
2794 {
2795         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2796                        mddev->sync_speed_min ? "local": "system");
2797 }
2798
2799 static ssize_t
2800 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2801 {
2802         int min;
2803         char *e;
2804         if (strncmp(buf, "system", 6)==0) {
2805                 mddev->sync_speed_min = 0;
2806                 return len;
2807         }
2808         min = simple_strtoul(buf, &e, 10);
2809         if (buf == e || (*e && *e != '\n') || min <= 0)
2810                 return -EINVAL;
2811         mddev->sync_speed_min = min;
2812         return len;
2813 }
2814
2815 static struct md_sysfs_entry md_sync_min =
2816 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2817
2818 static ssize_t
2819 sync_max_show(mddev_t *mddev, char *page)
2820 {
2821         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2822                        mddev->sync_speed_max ? "local": "system");
2823 }
2824
2825 static ssize_t
2826 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2827 {
2828         int max;
2829         char *e;
2830         if (strncmp(buf, "system", 6)==0) {
2831                 mddev->sync_speed_max = 0;
2832                 return len;
2833         }
2834         max = simple_strtoul(buf, &e, 10);
2835         if (buf == e || (*e && *e != '\n') || max <= 0)
2836                 return -EINVAL;
2837         mddev->sync_speed_max = max;
2838         return len;
2839 }
2840
2841 static struct md_sysfs_entry md_sync_max =
2842 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2843
2844
2845 static ssize_t
2846 sync_speed_show(mddev_t *mddev, char *page)
2847 {
2848         unsigned long resync, dt, db;
2849         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2850         dt = ((jiffies - mddev->resync_mark) / HZ);
2851         if (!dt) dt++;
2852         db = resync - (mddev->resync_mark_cnt);
2853         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2854 }
2855
2856 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2857
2858 static ssize_t
2859 sync_completed_show(mddev_t *mddev, char *page)
2860 {
2861         unsigned long max_blocks, resync;
2862
2863         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2864                 max_blocks = mddev->resync_max_sectors;
2865         else
2866                 max_blocks = mddev->size << 1;
2867
2868         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2869         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2870 }
2871
2872 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2873
2874 static ssize_t
2875 suspend_lo_show(mddev_t *mddev, char *page)
2876 {
2877         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2878 }
2879
2880 static ssize_t
2881 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2882 {
2883         char *e;
2884         unsigned long long new = simple_strtoull(buf, &e, 10);
2885
2886         if (mddev->pers->quiesce == NULL)
2887                 return -EINVAL;
2888         if (buf == e || (*e && *e != '\n'))
2889                 return -EINVAL;
2890         if (new >= mddev->suspend_hi ||
2891             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2892                 mddev->suspend_lo = new;
2893                 mddev->pers->quiesce(mddev, 2);
2894                 return len;
2895         } else
2896                 return -EINVAL;
2897 }
2898 static struct md_sysfs_entry md_suspend_lo =
2899 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2900
2901
2902 static ssize_t
2903 suspend_hi_show(mddev_t *mddev, char *page)
2904 {
2905         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2906 }
2907
2908 static ssize_t
2909 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2910 {
2911         char *e;
2912         unsigned long long new = simple_strtoull(buf, &e, 10);
2913
2914         if (mddev->pers->quiesce == NULL)
2915                 return -EINVAL;
2916         if (buf == e || (*e && *e != '\n'))
2917                 return -EINVAL;
2918         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2919             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2920                 mddev->suspend_hi = new;
2921                 mddev->pers->quiesce(mddev, 1);
2922                 mddev->pers->quiesce(mddev, 0);
2923                 return len;
2924         } else
2925                 return -EINVAL;
2926 }
2927 static struct md_sysfs_entry md_suspend_hi =
2928 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2929
2930 static ssize_t
2931 reshape_position_show(mddev_t *mddev, char *page)
2932 {
2933         if (mddev->reshape_position != MaxSector)
2934                 return sprintf(page, "%llu\n",
2935                                (unsigned long long)mddev->reshape_position);
2936         strcpy(page, "none\n");
2937         return 5;
2938 }
2939
2940 static ssize_t
2941 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2942 {
2943         char *e;
2944         unsigned long long new = simple_strtoull(buf, &e, 10);
2945         if (mddev->pers)
2946                 return -EBUSY;
2947         if (buf == e || (*e && *e != '\n'))
2948                 return -EINVAL;
2949         mddev->reshape_position = new;
2950         mddev->delta_disks = 0;
2951         mddev->new_level = mddev->level;
2952         mddev->new_layout = mddev->layout;
2953         mddev->new_chunk = mddev->chunk_size;
2954         return len;
2955 }
2956
2957 static struct md_sysfs_entry md_reshape_position =
2958 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2959        reshape_position_store);
2960
2961
2962 static struct attribute *md_default_attrs[] = {
2963         &md_level.attr,
2964         &md_layout.attr,
2965         &md_raid_disks.attr,
2966         &md_chunk_size.attr,
2967         &md_size.attr,
2968         &md_resync_start.attr,
2969         &md_metadata.attr,
2970         &md_new_device.attr,
2971         &md_safe_delay.attr,
2972         &md_array_state.attr,
2973         &md_reshape_position.attr,
2974         NULL,
2975 };
2976
2977 static struct attribute *md_redundancy_attrs[] = {
2978         &md_scan_mode.attr,
2979         &md_mismatches.attr,
2980         &md_sync_min.attr,
2981         &md_sync_max.attr,
2982         &md_sync_speed.attr,
2983         &md_sync_completed.attr,
2984         &md_suspend_lo.attr,
2985         &md_suspend_hi.attr,
2986         &md_bitmap.attr,
2987         NULL,
2988 };
2989 static struct attribute_group md_redundancy_group = {
2990         .name = NULL,
2991         .attrs = md_redundancy_attrs,
2992 };
2993
2994
2995 static ssize_t
2996 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2997 {
2998         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2999         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3000         ssize_t rv;
3001
3002         if (!entry->show)
3003                 return -EIO;
3004         rv = mddev_lock(mddev);
3005         if (!rv) {
3006                 rv = entry->show(mddev, page);
3007                 mddev_unlock(mddev);
3008         }
3009         return rv;
3010 }
3011
3012 static ssize_t
3013 md_attr_store(struct kobject *kobj, struct attribute *attr,
3014               const char *page, size_t length)
3015 {
3016         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3017         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3018         ssize_t rv;
3019
3020         if (!entry->store)
3021                 return -EIO;
3022         if (!capable(CAP_SYS_ADMIN))
3023                 return -EACCES;
3024         rv = mddev_lock(mddev);
3025         if (!rv) {
3026                 rv = entry->store(mddev, page, length);
3027                 mddev_unlock(mddev);
3028         }
3029         return rv;
3030 }
3031
3032 static void md_free(struct kobject *ko)
3033 {
3034         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3035         kfree(mddev);
3036 }
3037
3038 static struct sysfs_ops md_sysfs_ops = {
3039         .show   = md_attr_show,
3040         .store  = md_attr_store,
3041 };
3042 static struct kobj_type md_ktype = {
3043         .release        = md_free,
3044         .sysfs_ops      = &md_sysfs_ops,
3045         .default_attrs  = md_default_attrs,
3046 };
3047
3048 int mdp_major = 0;
3049
3050 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3051 {
3052         static DEFINE_MUTEX(disks_mutex);
3053         mddev_t *mddev = mddev_find(dev);
3054         struct gendisk *disk;
3055         int partitioned = (MAJOR(dev) != MD_MAJOR);
3056         int shift = partitioned ? MdpMinorShift : 0;
3057         int unit = MINOR(dev) >> shift;
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         mddev->kobj.parent = &disk->kobj;
3087         mddev->kobj.k_name = NULL;
3088         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
3089         mddev->kobj.ktype = &md_ktype;
3090         if (kobject_register(&mddev->kobj))
3091                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3092                        disk->disk_name);
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          * Also find largest hardsector size
3178          */
3179         ITERATE_RDEV(mddev,rdev,tmp) {
3180                 if (test_bit(Faulty, &rdev->flags))
3181                         continue;
3182                 sync_blockdev(rdev->bdev);
3183                 invalidate_bdev(rdev->bdev);
3184         }
3185
3186         md_probe(mddev->unit, NULL, NULL);
3187         disk = mddev->gendisk;
3188         if (!disk)
3189                 return -ENOMEM;
3190
3191         spin_lock(&pers_lock);
3192         pers = find_pers(mddev->level, mddev->clevel);
3193         if (!pers || !try_module_get(pers->owner)) {
3194                 spin_unlock(&pers_lock);
3195                 if (mddev->level != LEVEL_NONE)
3196                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3197                                mddev->level);
3198                 else
3199                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3200                                mddev->clevel);
3201                 return -EINVAL;
3202         }
3203         mddev->pers = pers;
3204         spin_unlock(&pers_lock);
3205         mddev->level = pers->level;
3206         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3207
3208         if (mddev->reshape_position != MaxSector &&
3209             pers->start_reshape == NULL) {
3210                 /* This personality cannot handle reshaping... */
3211                 mddev->pers = NULL;
3212                 module_put(pers->owner);
3213                 return -EINVAL;
3214         }
3215
3216         if (pers->sync_request) {
3217                 /* Warn if this is a potentially silly
3218                  * configuration.
3219                  */
3220                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3221                 mdk_rdev_t *rdev2;
3222                 struct list_head *tmp2;
3223                 int warned = 0;
3224                 ITERATE_RDEV(mddev, rdev, tmp) {
3225                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3226                                 if (rdev < rdev2 &&
3227                                     rdev->bdev->bd_contains ==
3228                                     rdev2->bdev->bd_contains) {
3229                                         printk(KERN_WARNING
3230                                                "%s: WARNING: %s appears to be"
3231                                                " on the same physical disk as"
3232                                                " %s.\n",
3233                                                mdname(mddev),
3234                                                bdevname(rdev->bdev,b),
3235                                                bdevname(rdev2->bdev,b2));
3236                                         warned = 1;
3237                                 }
3238                         }
3239                 }
3240                 if (warned)
3241                         printk(KERN_WARNING
3242                                "True protection against single-disk"
3243                                " failure might be compromised.\n");
3244         }
3245
3246         mddev->recovery = 0;
3247         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3248         mddev->barriers_work = 1;
3249         mddev->ok_start_degraded = start_dirty_degraded;
3250
3251         if (start_readonly)
3252                 mddev->ro = 2; /* read-only, but switch on first write */
3253
3254         err = mddev->pers->run(mddev);
3255         if (!err && mddev->pers->sync_request) {
3256                 err = bitmap_create(mddev);
3257                 if (err) {
3258                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3259                                mdname(mddev), err);
3260                         mddev->pers->stop(mddev);
3261                 }
3262         }
3263         if (err) {
3264                 printk(KERN_ERR "md: pers->run() failed ...\n");
3265                 module_put(mddev->pers->owner);
3266                 mddev->pers = NULL;
3267                 bitmap_destroy(mddev);
3268                 return err;
3269         }
3270         if (mddev->pers->sync_request) {
3271                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3272                         printk(KERN_WARNING
3273                                "md: cannot register extra attributes for %s\n",
3274                                mdname(mddev));
3275         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3276                 mddev->ro = 0;
3277
3278         atomic_set(&mddev->writes_pending,0);
3279         mddev->safemode = 0;
3280         mddev->safemode_timer.function = md_safemode_timeout;
3281         mddev->safemode_timer.data = (unsigned long) mddev;
3282         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3283         mddev->in_sync = 1;
3284
3285         ITERATE_RDEV(mddev,rdev,tmp)
3286                 if (rdev->raid_disk >= 0) {
3287                         char nm[20];
3288                         sprintf(nm, "rd%d", rdev->raid_disk);
3289                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3290                                 printk("md: cannot register %s for %s\n",
3291                                        nm, mdname(mddev));
3292                 }
3293         
3294         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3295         
3296         if (mddev->flags)
3297                 md_update_sb(mddev, 0);
3298
3299         set_capacity(disk, mddev->array_size<<1);
3300
3301         /* If we call blk_queue_make_request here, it will
3302          * re-initialise max_sectors etc which may have been
3303          * refined inside -> run.  So just set the bits we need to set.
3304          * Most initialisation happended when we called
3305          * blk_queue_make_request(..., md_fail_request)
3306          * earlier.
3307          */
3308         mddev->queue->queuedata = mddev;
3309         mddev->queue->make_request_fn = mddev->pers->make_request;
3310
3311         /* If there is a partially-recovered drive we need to
3312          * start recovery here.  If we leave it to md_check_recovery,
3313          * it will remove the drives and not do the right thing
3314          */
3315         if (mddev->degraded && !mddev->sync_thread) {
3316                 struct list_head *rtmp;
3317                 int spares = 0;
3318                 ITERATE_RDEV(mddev,rdev,rtmp)
3319                         if (rdev->raid_disk >= 0 &&
3320                             !test_bit(In_sync, &rdev->flags) &&
3321                             !test_bit(Faulty, &rdev->flags))
3322                                 /* complete an interrupted recovery */
3323                                 spares++;
3324                 if (spares && mddev->pers->sync_request) {
3325                         mddev->recovery = 0;
3326                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3327                         mddev->sync_thread = md_register_thread(md_do_sync,
3328                                                                 mddev,
3329                                                                 "%s_resync");
3330                         if (!mddev->sync_thread) {
3331                                 printk(KERN_ERR "%s: could not start resync"
3332                                        " thread...\n",
3333                                        mdname(mddev));
3334                                 /* leave the spares where they are, it shouldn't hurt */
3335                                 mddev->recovery = 0;
3336                         }
3337                 }
3338         }
3339         md_wakeup_thread(mddev->thread);
3340         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3341
3342         mddev->changed = 1;
3343         md_new_event(mddev);
3344         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3345         return 0;
3346 }
3347
3348 static int restart_array(mddev_t *mddev)
3349 {
3350         struct gendisk *disk = mddev->gendisk;
3351         int err;
3352
3353         /*
3354          * Complain if it has no devices
3355          */
3356         err = -ENXIO;
3357         if (list_empty(&mddev->disks))
3358                 goto out;
3359
3360         if (mddev->pers) {
3361                 err = -EBUSY;
3362                 if (!mddev->ro)
3363                         goto out;
3364
3365                 mddev->safemode = 0;
3366                 mddev->ro = 0;
3367                 set_disk_ro(disk, 0);
3368
3369                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3370                         mdname(mddev));
3371                 /*
3372                  * Kick recovery or resync if necessary
3373                  */
3374                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3375                 md_wakeup_thread(mddev->thread);
3376                 md_wakeup_thread(mddev->sync_thread);
3377                 err = 0;
3378         } else
3379                 err = -EINVAL;
3380
3381 out:
3382         return err;
3383 }
3384
3385 /* similar to deny_write_access, but accounts for our holding a reference
3386  * to the file ourselves */
3387 static int deny_bitmap_write_access(struct file * file)
3388 {
3389         struct inode *inode = file->f_mapping->host;
3390
3391         spin_lock(&inode->i_lock);
3392         if (atomic_read(&inode->i_writecount) > 1) {
3393                 spin_unlock(&inode->i_lock);
3394                 return -ETXTBSY;
3395         }
3396         atomic_set(&inode->i_writecount, -1);
3397         spin_unlock(&inode->i_lock);
3398
3399         return 0;
3400 }
3401
3402 static void restore_bitmap_write_access(struct file *file)
3403 {
3404         struct inode *inode = file->f_mapping->host;
3405
3406         spin_lock(&inode->i_lock);
3407         atomic_set(&inode->i_writecount, 1);
3408         spin_unlock(&inode->i_lock);
3409 }
3410
3411 /* mode:
3412  *   0 - completely stop and dis-assemble array
3413  *   1 - switch to readonly
3414  *   2 - stop but do not disassemble array
3415  */
3416 static int do_md_stop(mddev_t * mddev, int mode)
3417 {
3418         int err = 0;
3419         struct gendisk *disk = mddev->gendisk;
3420
3421         if (mddev->pers) {
3422                 if (atomic_read(&mddev->active)>2) {
3423                         printk("md: %s still in use.\n",mdname(mddev));
3424                         return -EBUSY;
3425                 }
3426
3427                 if (mddev->sync_thread) {
3428                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3429                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3430                         md_unregister_thread(mddev->sync_thread);
3431                         mddev->sync_thread = NULL;
3432                 }
3433
3434                 del_timer_sync(&mddev->safemode_timer);
3435
3436                 invalidate_partition(disk, 0);
3437
3438                 switch(mode) {
3439                 case 1: /* readonly */
3440                         err  = -ENXIO;
3441                         if (mddev->ro==1)
3442                                 goto out;
3443                         mddev->ro = 1;
3444                         break;
3445                 case 0: /* disassemble */
3446                 case 2: /* stop */
3447                         bitmap_flush(mddev);
3448                         md_super_wait(mddev);
3449                         if (mddev->ro)
3450                                 set_disk_ro(disk, 0);
3451                         blk_queue_make_request(mddev->queue, md_fail_request);
3452                         mddev->pers->stop(mddev);
3453                         mddev->queue->merge_bvec_fn = NULL;
3454                         mddev->queue->unplug_fn = NULL;
3455                         mddev->queue->issue_flush_fn = NULL;
3456                         mddev->queue->backing_dev_info.congested_fn = NULL;
3457                         if (mddev->pers->sync_request)
3458                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3459
3460                         module_put(mddev->pers->owner);
3461                         mddev->pers = NULL;
3462
3463                         set_capacity(disk, 0);
3464                         mddev->changed = 1;
3465
3466                         if (mddev->ro)
3467                                 mddev->ro = 0;
3468                 }
3469                 if (!mddev->in_sync || mddev->flags) {
3470                         /* mark array as shutdown cleanly */
3471                         mddev->in_sync = 1;
3472                         md_update_sb(mddev, 1);
3473                 }
3474                 if (mode == 1)
3475                         set_disk_ro(disk, 1);
3476                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3477         }
3478
3479         /*
3480          * Free resources if final stop
3481          */
3482         if (mode == 0) {
3483                 mdk_rdev_t *rdev;
3484                 struct list_head *tmp;
3485
3486                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3487
3488                 bitmap_destroy(mddev);
3489                 if (mddev->bitmap_file) {
3490                         restore_bitmap_write_access(mddev->bitmap_file);
3491                         fput(mddev->bitmap_file);
3492                         mddev->bitmap_file = NULL;
3493                 }
3494                 mddev->bitmap_offset = 0;
3495
3496                 ITERATE_RDEV(mddev,rdev,tmp)
3497                         if (rdev->raid_disk >= 0) {
3498                                 char nm[20];
3499                                 sprintf(nm, "rd%d", rdev->raid_disk);
3500                                 sysfs_remove_link(&mddev->kobj, nm);
3501                         }
3502
3503                 /* make sure all delayed_delete calls have finished */
3504                 flush_scheduled_work();
3505
3506                 export_array(mddev);
3507
3508                 mddev->array_size = 0;
3509                 mddev->size = 0;
3510                 mddev->raid_disks = 0;
3511                 mddev->recovery_cp = 0;
3512                 mddev->reshape_position = MaxSector;
3513
3514         } else if (mddev->pers)
3515                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3516                         mdname(mddev));
3517         err = 0;
3518         md_new_event(mddev);
3519 out:
3520         return err;
3521 }
3522
3523 #ifndef MODULE
3524 static void autorun_array(mddev_t *mddev)
3525 {
3526         mdk_rdev_t *rdev;
3527         struct list_head *tmp;
3528         int err;
3529
3530         if (list_empty(&mddev->disks))
3531                 return;
3532
3533         printk(KERN_INFO "md: running: ");
3534
3535         ITERATE_RDEV(mddev,rdev,tmp) {
3536                 char b[BDEVNAME_SIZE];
3537                 printk("<%s>", bdevname(rdev->bdev,b));
3538         }
3539         printk("\n");
3540
3541         err = do_md_run (mddev);
3542         if (err) {
3543                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3544                 do_md_stop (mddev, 0);
3545         }
3546 }
3547
3548 /*
3549  * lets try to run arrays based on all disks that have arrived
3550  * until now. (those are in pending_raid_disks)
3551  *
3552  * the method: pick the first pending disk, collect all disks with
3553  * the same UUID, remove all from the pending list and put them into
3554  * the 'same_array' list. Then order this list based on superblock
3555  * update time (freshest comes first), kick out 'old' disks and
3556  * compare superblocks. If everything's fine then run it.
3557  *
3558  * If "unit" is allocated, then bump its reference count
3559  */
3560 static void autorun_devices(int part)
3561 {
3562         struct list_head *tmp;
3563         mdk_rdev_t *rdev0, *rdev;
3564         mddev_t *mddev;
3565         char b[BDEVNAME_SIZE];
3566
3567         printk(KERN_INFO "md: autorun ...\n");
3568         while (!list_empty(&pending_raid_disks)) {
3569                 int unit;
3570                 dev_t dev;
3571                 LIST_HEAD(candidates);
3572                 rdev0 = list_entry(pending_raid_disks.next,
3573                                          mdk_rdev_t, same_set);
3574
3575                 printk(KERN_INFO "md: considering %s ...\n",
3576                         bdevname(rdev0->bdev,b));
3577                 INIT_LIST_HEAD(&candidates);
3578                 ITERATE_RDEV_PENDING(rdev,tmp)
3579                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3580                                 printk(KERN_INFO "md:  adding %s ...\n",
3581                                         bdevname(rdev->bdev,b));
3582                                 list_move(&rdev->same_set, &candidates);
3583                         }
3584                 /*
3585                  * now we have a set of devices, with all of them having
3586                  * mostly sane superblocks. It's time to allocate the
3587                  * mddev.
3588                  */
3589                 if (part) {
3590                         dev = MKDEV(mdp_major,
3591                                     rdev0->preferred_minor << MdpMinorShift);
3592                         unit = MINOR(dev) >> MdpMinorShift;
3593                 } else {
3594                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3595                         unit = MINOR(dev);
3596                 }
3597                 if (rdev0->preferred_minor != unit) {
3598                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3599                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3600                         break;
3601                 }
3602
3603                 md_probe(dev, NULL, NULL);
3604                 mddev = mddev_find(dev);
3605                 if (!mddev) {
3606                         printk(KERN_ERR 
3607                                 "md: cannot allocate memory for md drive.\n");
3608                         break;
3609                 }
3610                 if (mddev_lock(mddev)) 
3611                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3612                                mdname(mddev));
3613                 else if (mddev->raid_disks || mddev->major_version
3614                          || !list_empty(&mddev->disks)) {
3615                         printk(KERN_WARNING 
3616                                 "md: %s already running, cannot run %s\n",
3617                                 mdname(mddev), bdevname(rdev0->bdev,b));
3618                         mddev_unlock(mddev);
3619                 } else {
3620                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3621                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3622                                 list_del_init(&rdev->same_set);
3623                                 if (bind_rdev_to_array(rdev, mddev))
3624                                         export_rdev(rdev);
3625                         }
3626                         autorun_array(mddev);
3627                         mddev_unlock(mddev);
3628                 }
3629                 /* on success, candidates will be empty, on error
3630                  * it won't...
3631                  */
3632                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3633                         export_rdev(rdev);
3634                 mddev_put(mddev);
3635         }
3636         printk(KERN_INFO "md: ... autorun DONE.\n");
3637 }
3638 #endif /* !MODULE */
3639
3640 static int get_version(void __user * arg)
3641 {
3642         mdu_version_t ver;
3643
3644         ver.major = MD_MAJOR_VERSION;
3645         ver.minor = MD_MINOR_VERSION;
3646         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3647
3648         if (copy_to_user(arg, &ver, sizeof(ver)))
3649                 return -EFAULT;
3650
3651         return 0;
3652 }
3653
3654 static int get_array_info(mddev_t * mddev, void __user * arg)
3655 {
3656         mdu_array_info_t info;
3657         int nr,working,active,failed,spare;
3658         mdk_rdev_t *rdev;
3659         struct list_head *tmp;
3660
3661         nr=working=active=failed=spare=0;
3662         ITERATE_RDEV(mddev,rdev,tmp) {
3663                 nr++;
3664                 if (test_bit(Faulty, &rdev->flags))
3665                         failed++;
3666                 else {
3667                         working++;
3668                         if (test_bit(In_sync, &rdev->flags))
3669                                 active++;       
3670                         else
3671                                 spare++;
3672                 }
3673         }
3674
3675         info.major_version = mddev->major_version;
3676         info.minor_version = mddev->minor_version;
3677         info.patch_version = MD_PATCHLEVEL_VERSION;
3678         info.ctime         = mddev->ctime;
3679         info.level         = mddev->level;
3680         info.size          = mddev->size;
3681         if (info.size != mddev->size) /* overflow */
3682                 info.size = -1;
3683         info.nr_disks      = nr;
3684         info.raid_disks    = mddev->raid_disks;
3685         info.md_minor      = mddev->md_minor;
3686         info.not_persistent= !mddev->persistent;
3687
3688         info.utime         = mddev->utime;
3689         info.state         = 0;
3690         if (mddev->in_sync)
3691                 info.state = (1<<MD_SB_CLEAN);
3692         if (mddev->bitmap && mddev->bitmap_offset)
3693                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3694         info.active_disks  = active;
3695         info.working_disks = working;
3696         info.failed_disks  = failed;
3697         info.spare_disks   = spare;
3698
3699         info.layout        = mddev->layout;
3700         info.chunk_size    = mddev->chunk_size;
3701
3702         if (copy_to_user(arg, &info, sizeof(info)))
3703                 return -EFAULT;
3704
3705         return 0;
3706 }
3707
3708 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3709 {
3710         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3711         char *ptr, *buf = NULL;
3712         int err = -ENOMEM;
3713
3714         md_allow_write(mddev);
3715
3716         file = kmalloc(sizeof(*file), GFP_KERNEL);
3717         if (!file)
3718                 goto out;
3719
3720         /* bitmap disabled, zero the first byte and copy out */
3721         if (!mddev->bitmap || !mddev->bitmap->file) {
3722                 file->pathname[0] = '\0';
3723                 goto copy_out;
3724         }
3725
3726         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3727         if (!buf)
3728                 goto out;
3729
3730         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3731         if (!ptr)
3732                 goto out;
3733
3734         strcpy(file->pathname, ptr);
3735
3736 copy_out:
3737         err = 0;
3738         if (copy_to_user(arg, file, sizeof(*file)))
3739                 err = -EFAULT;
3740 out:
3741         kfree(buf);
3742         kfree(file);
3743         return err;
3744 }
3745
3746 static int get_disk_info(mddev_t * mddev, void __user * arg)
3747 {
3748         mdu_disk_info_t info;
3749         unsigned int nr;
3750         mdk_rdev_t *rdev;
3751
3752         if (copy_from_user(&info, arg, sizeof(info)))
3753                 return -EFAULT;
3754
3755         nr = info.number;
3756
3757         rdev = find_rdev_nr(mddev, nr);
3758         if (rdev) {
3759                 info.major = MAJOR(rdev->bdev->bd_dev);
3760                 info.minor = MINOR(rdev->bdev->bd_dev);
3761                 info.raid_disk = rdev->raid_disk;
3762                 info.state = 0;
3763                 if (test_bit(Faulty, &rdev->flags))
3764                         info.state |= (1<<MD_DISK_FAULTY);
3765                 else if (test_bit(In_sync, &rdev->flags)) {
3766                         info.state |= (1<<MD_DISK_ACTIVE);
3767                         info.state |= (1<<MD_DISK_SYNC);
3768                 }
3769                 if (test_bit(WriteMostly, &rdev->flags))
3770                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3771         } else {
3772                 info.major = info.minor = 0;
3773                 info.raid_disk = -1;
3774                 info.state = (1<<MD_DISK_REMOVED);
3775         }
3776
3777         if (copy_to_user(arg, &info, sizeof(info)))
3778                 return -EFAULT;
3779
3780         return 0;
3781 }
3782
3783 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3784 {
3785         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3786         mdk_rdev_t *rdev;
3787         dev_t dev = MKDEV(info->major,info->minor);
3788
3789         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3790                 return -EOVERFLOW;
3791
3792         if (!mddev->raid_disks) {
3793                 int err;
3794                 /* expecting a device which has a superblock */
3795                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3796                 if (IS_ERR(rdev)) {
3797                         printk(KERN_WARNING 
3798                                 "md: md_import_device returned %ld\n",
3799                                 PTR_ERR(rdev));
3800                         return PTR_ERR(rdev);
3801                 }
3802                 if (!list_empty(&mddev->disks)) {
3803                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3804                                                         mdk_rdev_t, same_set);
3805                         int err = super_types[mddev->major_version]
3806                                 .load_super(rdev, rdev0, mddev->minor_version);
3807                         if (err < 0) {
3808                                 printk(KERN_WARNING 
3809                                         "md: %s has different UUID to %s\n",
3810                                         bdevname(rdev->bdev,b), 
3811                                         bdevname(rdev0->bdev,b2));
3812                                 export_rdev(rdev);
3813                                 return -EINVAL;
3814                         }
3815                 }
3816                 err = bind_rdev_to_array(rdev, mddev);
3817                 if (err)
3818                         export_rdev(rdev);
3819                 return err;
3820         }
3821
3822         /*
3823          * add_new_disk can be used once the array is assembled
3824          * to add "hot spares".  They must already have a superblock
3825          * written
3826          */
3827         if (mddev->pers) {
3828                 int err;
3829                 if (!mddev->pers->hot_add_disk) {
3830                         printk(KERN_WARNING 
3831                                 "%s: personality does not support diskops!\n",
3832                                mdname(mddev));
3833                         return -EINVAL;
3834                 }
3835                 if (mddev->persistent)
3836                         rdev = md_import_device(dev, mddev->major_version,
3837                                                 mddev->minor_version);
3838                 else
3839                         rdev = md_import_device(dev, -1, -1);
3840                 if (IS_ERR(rdev)) {
3841                         printk(KERN_WARNING 
3842                                 "md: md_import_device returned %ld\n",
3843                                 PTR_ERR(rdev));
3844                         return PTR_ERR(rdev);
3845                 }
3846                 /* set save_raid_disk if appropriate */
3847                 if (!mddev->persistent) {
3848                         if (info->state & (1<<MD_DISK_SYNC)  &&
3849                             info->raid_disk < mddev->raid_disks)
3850                                 rdev->raid_disk = info->raid_disk;
3851                         else
3852                                 rdev->raid_disk = -1;
3853                 } else
3854                         super_types[mddev->major_version].
3855                                 validate_super(mddev, rdev);
3856                 rdev->saved_raid_disk = rdev->raid_disk;
3857
3858                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3859                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3860                         set_bit(WriteMostly, &rdev->flags);
3861
3862                 rdev->raid_disk = -1;
3863                 err = bind_rdev_to_array(rdev, mddev);
3864                 if (!err && !mddev->pers->hot_remove_disk) {
3865                         /* If there is hot_add_disk but no hot_remove_disk
3866                          * then added disks for geometry changes,
3867                          * and should be added immediately.
3868                          */
3869                         super_types[mddev->major_version].
3870                                 validate_super(mddev, rdev);
3871                         err = mddev->pers->hot_add_disk(mddev, rdev);
3872                         if (err)
3873                                 unbind_rdev_from_array(rdev);
3874                 }
3875                 if (err)
3876                         export_rdev(rdev);
3877
3878                 md_update_sb(mddev, 1);
3879                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3880                 md_wakeup_thread(mddev->thread);
3881                 return err;
3882         }
3883
3884         /* otherwise, add_new_disk is only allowed
3885          * for major_version==0 superblocks
3886          */
3887         if (mddev->major_version != 0) {
3888                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3889                        mdname(mddev));
3890                 return -EINVAL;
3891         }
3892
3893         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3894                 int err;
3895                 rdev = md_import_device (dev, -1, 0);
3896                 if (IS_ERR(rdev)) {
3897                         printk(KERN_WARNING 
3898                                 "md: error, md_import_device() returned %ld\n",
3899                                 PTR_ERR(rdev));
3900                         return PTR_ERR(rdev);
3901                 }
3902                 rdev->desc_nr = info->number;
3903                 if (info->raid_disk < mddev->raid_disks)
3904                         rdev->raid_disk = info->raid_disk;
3905                 else
3906                         rdev->raid_disk = -1;
3907
3908                 rdev->flags = 0;
3909
3910                 if (rdev->raid_disk < mddev->raid_disks)
3911                         if (info->state & (1<<MD_DISK_SYNC))
3912                                 set_bit(In_sync, &rdev->flags);
3913
3914                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3915                         set_bit(WriteMostly, &rdev->flags);
3916
3917                 if (!mddev->persistent) {
3918                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3919                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3920                 } else 
3921                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3922                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3923
3924                 err = bind_rdev_to_array(rdev, mddev);
3925                 if (err) {
3926                         export_rdev(rdev);
3927                         return err;
3928                 }
3929         }
3930
3931         return 0;
3932 }
3933
3934 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3935 {
3936         char b[BDEVNAME_SIZE];
3937         mdk_rdev_t *rdev;
3938
3939         if (!mddev->pers)
3940                 return -ENODEV;
3941
3942         rdev = find_rdev(mddev, dev);
3943         if (!rdev)
3944                 return -ENXIO;
3945
3946         if (rdev->raid_disk >= 0)
3947                 goto busy;
3948
3949         kick_rdev_from_array(rdev);
3950         md_update_sb(mddev, 1);
3951         md_new_event(mddev);
3952
3953         return 0;
3954 busy:
3955         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3956                 bdevname(rdev->bdev,b), mdname(mddev));
3957         return -EBUSY;
3958 }
3959
3960 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3961 {
3962         char b[BDEVNAME_SIZE];
3963         int err;
3964         unsigned int size;
3965         mdk_rdev_t *rdev;
3966
3967         if (!mddev->pers)
3968                 return -ENODEV;
3969
3970         if (mddev->major_version != 0) {
3971                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3972                         " version-0 superblocks.\n",
3973                         mdname(mddev));
3974                 return -EINVAL;
3975         }
3976         if (!mddev->pers->hot_add_disk) {
3977                 printk(KERN_WARNING 
3978                         "%s: personality does not support diskops!\n",
3979                         mdname(mddev));
3980                 return -EINVAL;
3981         }
3982
3983         rdev = md_import_device (dev, -1, 0);
3984         if (IS_ERR(rdev)) {
3985                 printk(KERN_WARNING 
3986                         "md: error, md_import_device() returned %ld\n",
3987                         PTR_ERR(rdev));
3988                 return -EINVAL;
3989         }
3990
3991         if (mddev->persistent)
3992                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3993         else
3994                 rdev->sb_offset =
3995                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3996
3997         size = calc_dev_size(rdev, mddev->chunk_size);
3998         rdev->size = size;
3999
4000         if (test_bit(Faulty, &rdev->flags)) {
4001                 printk(KERN_WARNING 
4002                         "md: can not hot-add faulty %s disk to %s!\n",
4003                         bdevname(rdev->bdev,b), mdname(mddev));
4004                 err = -EINVAL;
4005                 goto abort_export;
4006         }
4007         clear_bit(In_sync, &rdev->flags);
4008         rdev->desc_nr = -1;
4009         rdev->saved_raid_disk = -1;
4010         err = bind_rdev_to_array(rdev, mddev);
4011         if (err)
4012                 goto abort_export;
4013
4014         /*
4015          * The rest should better be atomic, we can have disk failures
4016          * noticed in interrupt contexts ...
4017          */
4018
4019         if (rdev->desc_nr == mddev->max_disks) {
4020                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4021                         mdname(mddev));
4022                 err = -EBUSY;
4023                 goto abort_unbind_export;
4024         }
4025
4026         rdev->raid_disk = -1;
4027
4028         md_update_sb(mddev, 1);
4029
4030         /*
4031          * Kick recovery, maybe this spare has to be added to the
4032          * array immediately.
4033          */
4034         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4035         md_wakeup_thread(mddev->thread);
4036         md_new_event(mddev);
4037         return 0;
4038
4039 abort_unbind_export:
4040         unbind_rdev_from_array(rdev);
4041
4042 abort_export:
4043         export_rdev(rdev);
4044         return err;
4045 }
4046
4047 static int set_bitmap_file(mddev_t *mddev, int fd)
4048 {
4049         int err;
4050
4051         if (mddev->pers) {
4052                 if (!mddev->pers->quiesce)
4053                         return -EBUSY;
4054                 if (mddev->recovery || mddev->sync_thread)
4055                         return -EBUSY;
4056                 /* we should be able to change the bitmap.. */
4057         }
4058
4059
4060         if (fd >= 0) {
4061                 if (mddev->bitmap)
4062                         return -EEXIST; /* cannot add when bitmap is present */
4063                 mddev->bitmap_file = fget(fd);
4064
4065                 if (mddev->bitmap_file == NULL) {
4066                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4067                                mdname(mddev));
4068                         return -EBADF;
4069                 }
4070
4071                 err = deny_bitmap_write_access(mddev->bitmap_file);
4072                 if (err) {
4073                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4074                                mdname(mddev));
4075                         fput(mddev->bitmap_file);
4076                         mddev->bitmap_file = NULL;
4077                         return err;
4078                 }
4079                 mddev->bitmap_offset = 0; /* file overrides offset */
4080         } else if (mddev->bitmap == NULL)
4081                 return -ENOENT; /* cannot remove what isn't there */
4082         err = 0;
4083         if (mddev->pers) {
4084                 mddev->pers->quiesce(mddev, 1);
4085                 if (fd >= 0)
4086                         err = bitmap_create(mddev);
4087                 if (fd < 0 || err) {
4088                         bitmap_destroy(mddev);
4089                         fd = -1; /* make sure to put the file */
4090                 }
4091                 mddev->pers->quiesce(mddev, 0);
4092         }
4093         if (fd < 0) {
4094                 if (mddev->bitmap_file) {
4095                         restore_bitmap_write_access(mddev->bitmap_file);
4096                         fput(mddev->bitmap_file);
4097                 }
4098                 mddev->bitmap_file = NULL;
4099         }
4100
4101         return err;
4102 }
4103
4104 /*
4105  * set_array_info is used two different ways
4106  * The original usage is when creating a new array.
4107  * In this usage, raid_disks is > 0 and it together with
4108  *  level, size, not_persistent,layout,chunksize determine the
4109  *  shape of the array.
4110  *  This will always create an array with a type-0.90.0 superblock.
4111  * The newer usage is when assembling an array.
4112  *  In this case raid_disks will be 0, and the major_version field is
4113  *  use to determine which style super-blocks are to be found on the devices.
4114  *  The minor and patch _version numbers are also kept incase the
4115  *  super_block handler wishes to interpret them.
4116  */
4117 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4118 {
4119
4120         if (info->raid_disks == 0) {
4121                 /* just setting version number for superblock loading */
4122                 if (info->major_version < 0 ||
4123                     info->major_version >= ARRAY_SIZE(super_types) ||
4124                     super_types[info->major_version].name == NULL) {
4125                         /* maybe try to auto-load a module? */
4126                         printk(KERN_INFO 
4127                                 "md: superblock version %d not known\n",
4128                                 info->major_version);
4129                         return -EINVAL;
4130                 }
4131                 mddev->major_version = info->major_version;
4132                 mddev->minor_version = info->minor_version;
4133                 mddev->patch_version = info->patch_version;
4134                 mddev->persistent = !info->not_persistent;
4135                 return 0;
4136         }
4137         mddev->major_version = MD_MAJOR_VERSION;
4138         mddev->minor_version = MD_MINOR_VERSION;
4139         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4140         mddev->ctime         = get_seconds();
4141
4142         mddev->level         = info->level;
4143         mddev->clevel[0]     = 0;
4144         mddev->size          = info->size;
4145         mddev->raid_disks    = info->raid_disks;
4146         /* don't set md_minor, it is determined by which /dev/md* was
4147          * openned
4148          */
4149         if (info->state & (1<<MD_SB_CLEAN))
4150                 mddev->recovery_cp = MaxSector;
4151         else
4152                 mddev->recovery_cp = 0;
4153         mddev->persistent    = ! info->not_persistent;
4154
4155         mddev->layout        = info->layout;
4156         mddev->chunk_size    = info->chunk_size;
4157
4158         mddev->max_disks     = MD_SB_DISKS;
4159
4160         mddev->flags         = 0;
4161         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4162
4163         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4164         mddev->bitmap_offset = 0;
4165
4166         mddev->reshape_position = MaxSector;
4167
4168         /*
4169          * Generate a 128 bit UUID
4170          */
4171         get_random_bytes(mddev->uuid, 16);
4172
4173         mddev->new_level = mddev->level;
4174         mddev->new_chunk = mddev->chunk_size;
4175         mddev->new_layout = mddev->layout;
4176         mddev->delta_disks = 0;
4177
4178         return 0;
4179 }
4180
4181 static int update_size(mddev_t *mddev, unsigned long size)
4182 {
4183         mdk_rdev_t * rdev;
4184         int rv;
4185         struct list_head *tmp;
4186         int fit = (size == 0);
4187
4188         if (mddev->pers->resize == NULL)
4189                 return -EINVAL;
4190         /* The "size" is the amount of each device that is used.
4191          * This can only make sense for arrays with redundancy.
4192          * linear and raid0 always use whatever space is available
4193          * We can only consider changing the size if no resync
4194          * or reconstruction is happening, and if the new size
4195          * is acceptable. It must fit before the sb_offset or,
4196          * if that is <data_offset, it must fit before the
4197          * size of each device.
4198          * If size is zero, we find the largest size that fits.
4199          */
4200         if (mddev->sync_thread)
4201                 return -EBUSY;
4202         ITERATE_RDEV(mddev,rdev,tmp) {
4203                 sector_t avail;
4204                 avail = rdev->size * 2;
4205
4206                 if (fit && (size == 0 || size > avail/2))
4207                         size = avail/2;
4208                 if (avail < ((sector_t)size << 1))
4209                         return -ENOSPC;
4210         }
4211         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4212         if (!rv) {
4213                 struct block_device *bdev;
4214
4215                 bdev = bdget_disk(mddev->gendisk, 0);
4216                 if (bdev) {
4217                         mutex_lock(&bdev->bd_inode->i_mutex);
4218                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4219                         mutex_unlock(&bdev->bd_inode->i_mutex);
4220                         bdput(bdev);
4221                 }
4222         }
4223         return rv;
4224 }
4225
4226 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4227 {
4228         int rv;
4229         /* change the number of raid disks */
4230         if (mddev->pers->check_reshape == NULL)
4231                 return -EINVAL;
4232         if (raid_disks <= 0 ||
4233             raid_disks >= mddev->max_disks)
4234                 return -EINVAL;
4235         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4236                 return -EBUSY;
4237         mddev->delta_disks = raid_disks - mddev->raid_disks;
4238
4239         rv = mddev->pers->check_reshape(mddev);
4240         return rv;
4241 }
4242
4243
4244 /*
4245  * update_array_info is used to change the configuration of an
4246  * on-line array.
4247  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4248  * fields in the info are checked against the array.
4249  * Any differences that cannot be handled will cause an error.
4250  * Normally, only one change can be managed at a time.
4251  */
4252 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4253 {
4254         int rv = 0;
4255         int cnt = 0;
4256         int state = 0;
4257
4258         /* calculate expected state,ignoring low bits */
4259         if (mddev->bitmap && mddev->bitmap_offset)
4260                 state |= (1 << MD_SB_BITMAP_PRESENT);
4261
4262         if (mddev->major_version != info->major_version ||
4263             mddev->minor_version != info->minor_version ||
4264 /*          mddev->patch_version != info->patch_version || */
4265             mddev->ctime         != info->ctime         ||
4266             mddev->level         != info->level         ||
4267 /*          mddev->layout        != info->layout        || */
4268             !mddev->persistent   != info->not_persistent||
4269             mddev->chunk_size    != info->chunk_size    ||
4270             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4271             ((state^info->state) & 0xfffffe00)
4272                 )
4273                 return -EINVAL;
4274         /* Check there is only one change */
4275         if (info->size >= 0 && mddev->size != info->size) cnt++;
4276         if (mddev->raid_disks != info->raid_disks) cnt++;
4277         if (mddev->layout != info->layout) cnt++;
4278         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4279         if (cnt == 0) return 0;
4280         if (cnt > 1) return -EINVAL;
4281
4282         if (mddev->layout != info->layout) {
4283                 /* Change layout
4284                  * we don't need to do anything at the md level, the
4285                  * personality will take care of it all.
4286                  */
4287                 if (mddev->pers->reconfig == NULL)
4288                         return -EINVAL;
4289                 else
4290                         return mddev->pers->reconfig(mddev, info->layout, -1);
4291         }
4292         if (info->size >= 0 && mddev->size != info->size)
4293                 rv = update_size(mddev, info->size);
4294
4295         if (mddev->raid_disks    != info->raid_disks)
4296                 rv = update_raid_disks(mddev, info->raid_disks);
4297
4298         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4299                 if (mddev->pers->quiesce == NULL)
4300                         return -EINVAL;
4301                 if (mddev->recovery || mddev->sync_thread)
4302                         return -EBUSY;
4303                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4304                         /* add the bitmap */
4305                         if (mddev->bitmap)
4306                                 return -EEXIST;
4307                         if (mddev->default_bitmap_offset == 0)
4308                                 return -EINVAL;
4309                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4310                         mddev->pers->quiesce(mddev, 1);
4311                         rv = bitmap_create(mddev);
4312                         if (rv)
4313                                 bitmap_destroy(mddev);
4314                         mddev->pers->quiesce(mddev, 0);
4315                 } else {
4316                         /* remove the bitmap */
4317                         if (!mddev->bitmap)
4318                                 return -ENOENT;
4319                         if (mddev->bitmap->file)
4320                                 return -EINVAL;
4321                         mddev->pers->quiesce(mddev, 1);
4322                         bitmap_destroy(mddev);
4323                         mddev->pers->quiesce(mddev, 0);
4324                         mddev->bitmap_offset = 0;
4325                 }
4326         }
4327         md_update_sb(mddev, 1);
4328         return rv;
4329 }
4330
4331 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4332 {
4333         mdk_rdev_t *rdev;
4334
4335         if (mddev->pers == NULL)
4336                 return -ENODEV;
4337
4338         rdev = find_rdev(mddev, dev);
4339         if (!rdev)
4340                 return -ENODEV;
4341
4342         md_error(mddev, rdev);
4343         return 0;
4344 }
4345
4346 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4347 {
4348         mddev_t *mddev = bdev->bd_disk->private_data;
4349
4350         geo->heads = 2;
4351         geo->sectors = 4;
4352         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4353         return 0;
4354 }
4355
4356 static int md_ioctl(struct inode *inode, struct file *file,
4357                         unsigned int cmd, unsigned long arg)
4358 {
4359         int err = 0;
4360         void __user *argp = (void __user *)arg;
4361         mddev_t *mddev = NULL;
4362
4363         if (!capable(CAP_SYS_ADMIN))
4364                 return -EACCES;
4365
4366         /*
4367          * Commands dealing with the RAID driver but not any
4368          * particular array:
4369          */
4370         switch (cmd)
4371         {
4372                 case RAID_VERSION:
4373                         err = get_version(argp);
4374                         goto done;
4375
4376                 case PRINT_RAID_DEBUG:
4377                         err = 0;
4378                         md_print_devices();
4379                         goto done;
4380
4381 #ifndef MODULE
4382                 case RAID_AUTORUN:
4383                         err = 0;
4384                         autostart_arrays(arg);
4385                         goto done;
4386 #endif
4387                 default:;
4388         }
4389
4390         /*
4391          * Commands creating/starting a new array:
4392          */
4393
4394         mddev = inode->i_bdev->bd_disk->private_data;
4395
4396         if (!mddev) {
4397                 BUG();
4398                 goto abort;
4399         }
4400
4401         err = mddev_lock(mddev);
4402         if (err) {
4403                 printk(KERN_INFO 
4404                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4405                         err, cmd);
4406                 goto abort;
4407         }
4408
4409         switch (cmd)
4410         {
4411                 case SET_ARRAY_INFO:
4412                         {
4413                                 mdu_array_info_t info;
4414                                 if (!arg)
4415                                         memset(&info, 0, sizeof(info));
4416                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4417                                         err = -EFAULT;
4418                                         goto abort_unlock;
4419                                 }
4420                                 if (mddev->pers) {
4421                                         err = update_array_info(mddev, &info);
4422                                         if (err) {
4423                                                 printk(KERN_WARNING "md: couldn't update"
4424                                                        " array info. %d\n", err);
4425                                                 goto abort_unlock;
4426                                         }
4427                                         goto done_unlock;
4428                                 }
4429                                 if (!list_empty(&mddev->disks)) {
4430                                         printk(KERN_WARNING
4431                                                "md: array %s already has disks!\n",
4432                                                mdname(mddev));
4433                                         err = -EBUSY;
4434                                         goto abort_unlock;
4435                                 }
4436                                 if (mddev->raid_disks) {
4437                                         printk(KERN_WARNING
4438                                                "md: array %s already initialised!\n",
4439                                                mdname(mddev));
4440                                         err = -EBUSY;
4441                                         goto abort_unlock;
4442                                 }
4443                                 err = set_array_info(mddev, &info);
4444                                 if (err) {
4445                                         printk(KERN_WARNING "md: couldn't set"
4446                                                " array info. %d\n", err);
4447                                         goto abort_unlock;
4448                                 }
4449                         }
4450                         goto done_unlock;
4451
4452                 default:;
4453         }
4454
4455         /*
4456          * Commands querying/configuring an existing array:
4457          */
4458         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4459          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4460         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4461                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4462                         && cmd != GET_BITMAP_FILE) {
4463                 err = -ENODEV;
4464                 goto abort_unlock;
4465         }
4466
4467         /*
4468          * Commands even a read-only array can execute:
4469          */
4470         switch (cmd)
4471         {
4472                 case GET_ARRAY_INFO:
4473                         err = get_array_info(mddev, argp);
4474                         goto done_unlock;
4475
4476                 case GET_BITMAP_FILE:
4477                         err = get_bitmap_file(mddev, argp);
4478                         goto done_unlock;
4479
4480                 case GET_DISK_INFO:
4481                         err = get_disk_info(mddev, argp);
4482                         goto done_unlock;
4483
4484                 case RESTART_ARRAY_RW:
4485                         err = restart_array(mddev);
4486                         goto done_unlock;
4487
4488                 case STOP_ARRAY:
4489                         err = do_md_stop (mddev, 0);
4490                         goto done_unlock;
4491
4492                 case STOP_ARRAY_RO:
4493                         err = do_md_stop (mddev, 1);
4494                         goto done_unlock;
4495
4496         /*
4497          * We have a problem here : there is no easy way to give a CHS
4498          * virtual geometry. We currently pretend that we have a 2 heads
4499          * 4 sectors (with a BIG number of cylinders...). This drives
4500          * dosfs just mad... ;-)
4501          */
4502         }
4503
4504         /*
4505          * The remaining ioctls are changing the state of the
4506          * superblock, so we do not allow them on read-only arrays.
4507          * However non-MD ioctls (e.g. get-size) will still come through
4508          * here and hit the 'default' below, so only disallow
4509          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4510          */
4511         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4512             mddev->ro && mddev->pers) {
4513                 if (mddev->ro == 2) {
4514                         mddev->ro = 0;
4515                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4516                 md_wakeup_thread(mddev->thread);
4517
4518                 } else {
4519                         err = -EROFS;
4520                         goto abort_unlock;
4521                 }
4522         }
4523
4524         switch (cmd)
4525         {
4526                 case ADD_NEW_DISK:
4527                 {
4528                         mdu_disk_info_t info;
4529                         if (copy_from_user(&info, argp, sizeof(info)))
4530                                 err = -EFAULT;
4531                         else
4532                                 err = add_new_disk(mddev, &info);
4533                         goto done_unlock;
4534                 }
4535
4536                 case HOT_REMOVE_DISK:
4537                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4538                         goto done_unlock;
4539
4540                 case HOT_ADD_DISK:
4541                         err = hot_add_disk(mddev, new_decode_dev(arg));
4542                         goto done_unlock;
4543
4544                 case SET_DISK_FAULTY:
4545                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4546                         goto done_unlock;
4547
4548                 case RUN_ARRAY:
4549                         err = do_md_run (mddev);
4550                         goto done_unlock;
4551
4552                 case SET_BITMAP_FILE:
4553                         err = set_bitmap_file(mddev, (int)arg);
4554                         goto done_unlock;
4555
4556                 default:
4557                         err = -EINVAL;
4558                         goto abort_unlock;
4559         }
4560
4561 done_unlock:
4562 abort_unlock:
4563         mddev_unlock(mddev);
4564
4565         return err;
4566 done:
4567         if (err)
4568                 MD_BUG();
4569 abort:
4570         return err;
4571 }
4572
4573 static int md_open(struct inode *inode, struct file *file)
4574 {
4575         /*
4576          * Succeed if we can lock the mddev, which confirms that
4577          * it isn't being stopped right now.
4578          */
4579         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4580         int err;
4581
4582         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4583                 goto out;
4584
4585         err = 0;
4586         mddev_get(mddev);
4587         mddev_unlock(mddev);
4588
4589         check_disk_change(inode->i_bdev);
4590  out:
4591         return err;
4592 }
4593
4594 static int md_release(struct inode *inode, struct file * file)
4595 {
4596         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4597
4598         BUG_ON(!mddev);
4599         mddev_put(mddev);
4600
4601         return 0;
4602 }
4603
4604 static int md_media_changed(struct gendisk *disk)
4605 {
4606         mddev_t *mddev = disk->private_data;
4607
4608         return mddev->changed;
4609 }
4610
4611 static int md_revalidate(struct gendisk *disk)
4612 {
4613         mddev_t *mddev = disk->private_data;
4614
4615         mddev->changed = 0;
4616         return 0;
4617 }
4618 static struct block_device_operations md_fops =
4619 {
4620         .owner          = THIS_MODULE,
4621         .open           = md_open,
4622         .release        = md_release,
4623         .ioctl          = md_ioctl,
4624         .getgeo         = md_getgeo,
4625         .media_changed  = md_media_changed,
4626         .revalidate_disk= md_revalidate,
4627 };
4628
4629 static int md_thread(void * arg)
4630 {
4631         mdk_thread_t *thread = arg;
4632
4633         /*
4634          * md_thread is a 'system-thread', it's priority should be very
4635          * high. We avoid resource deadlocks individually in each
4636          * raid personality. (RAID5 does preallocation) We also use RR and
4637          * the very same RT priority as kswapd, thus we will never get
4638          * into a priority inversion deadlock.
4639          *
4640          * we definitely have to have equal or higher priority than
4641          * bdflush, otherwise bdflush will deadlock if there are too
4642          * many dirty RAID5 blocks.
4643          */
4644
4645         current->flags |= PF_NOFREEZE;
4646         allow_signal(SIGKILL);
4647         while (!kthread_should_stop()) {
4648
4649                 /* We need to wait INTERRUPTIBLE so that
4650                  * we don't add to the load-average.
4651                  * That means we need to be sure no signals are
4652                  * pending
4653                  */
4654                 if (signal_pending(current))
4655                         flush_signals(current);
4656
4657                 wait_event_interruptible_timeout
4658                         (thread->wqueue,
4659                          test_bit(THREAD_WAKEUP, &thread->flags)
4660                          || kthread_should_stop(),
4661                          thread->timeout);
4662
4663                 clear_bit(THREAD_WAKEUP, &thread->flags);
4664
4665                 thread->run(thread->mddev);
4666         }
4667
4668         return 0;
4669 }
4670
4671 void md_wakeup_thread(mdk_thread_t *thread)
4672 {
4673         if (thread) {
4674                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4675                 set_bit(THREAD_WAKEUP, &thread->flags);
4676                 wake_up(&thread->wqueue);
4677         }
4678 }
4679
4680 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4681                                  const char *name)
4682 {
4683         mdk_thread_t *thread;
4684
4685         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4686         if (!thread)
4687                 return NULL;
4688
4689         init_waitqueue_head(&thread->wqueue);
4690
4691         thread->run = run;
4692         thread->mddev = mddev;
4693         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4694         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4695         if (IS_ERR(thread->tsk)) {
4696                 kfree(thread);
4697                 return NULL;
4698         }
4699         return thread;
4700 }
4701
4702 void md_unregister_thread(mdk_thread_t *thread)
4703 {
4704         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4705
4706         kthread_stop(thread->tsk);
4707         kfree(thread);
4708 }
4709
4710 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4711 {
4712         if (!mddev) {
4713                 MD_BUG();
4714                 return;
4715         }
4716
4717         if (!rdev || test_bit(Faulty, &rdev->flags))
4718                 return;
4719 /*
4720         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4721                 mdname(mddev),
4722                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4723                 __builtin_return_address(0),__builtin_return_address(1),
4724                 __builtin_return_address(2),__builtin_return_address(3));
4725 */
4726         if (!mddev->pers)
4727                 return;
4728         if (!mddev->pers->error_handler)
4729                 return;
4730         mddev->pers->error_handler(mddev,rdev);
4731         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4732         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4733         md_wakeup_thread(mddev->thread);
4734         md_new_event_inintr(mddev);
4735 }
4736
4737 /* seq_file implementation /proc/mdstat */
4738
4739 static void status_unused(struct seq_file *seq)
4740 {
4741         int i = 0;
4742         mdk_rdev_t *rdev;
4743         struct list_head *tmp;
4744
4745         seq_printf(seq, "unused devices: ");
4746
4747         ITERATE_RDEV_PENDING(rdev,tmp) {
4748                 char b[BDEVNAME_SIZE];
4749                 i++;
4750                 seq_printf(seq, "%s ",
4751                               bdevname(rdev->bdev,b));
4752         }
4753         if (!i)
4754                 seq_printf(seq, "<none>");
4755
4756         seq_printf(seq, "\n");
4757 }
4758
4759
4760 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4761 {
4762         sector_t max_blocks, resync, res;
4763         unsigned long dt, db, rt;
4764         int scale;
4765         unsigned int per_milli;
4766
4767         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4768
4769         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4770                 max_blocks = mddev->resync_max_sectors >> 1;
4771         else
4772                 max_blocks = mddev->size;
4773
4774         /*
4775          * Should not happen.
4776          */
4777         if (!max_blocks) {
4778                 MD_BUG();
4779                 return;
4780         }
4781         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4782          * in a sector_t, and (max_blocks>>scale) will fit in a
4783          * u32, as those are the requirements for sector_div.
4784          * Thus 'scale' must be at least 10
4785          */
4786         scale = 10;
4787         if (sizeof(sector_t) > sizeof(unsigned long)) {
4788                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4789                         scale++;
4790         }
4791         res = (resync>>scale)*1000;
4792         sector_div(res, (u32)((max_blocks>>scale)+1));
4793
4794         per_milli = res;
4795         {
4796                 int i, x = per_milli/50, y = 20-x;
4797                 seq_printf(seq, "[");
4798                 for (i = 0; i < x; i++)
4799                         seq_printf(seq, "=");
4800                 seq_printf(seq, ">");
4801                 for (i = 0; i < y; i++)
4802                         seq_printf(seq, ".");
4803                 seq_printf(seq, "] ");
4804         }
4805         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4806                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4807                     "reshape" :
4808                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4809                      "check" :
4810                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4811                       "resync" : "recovery"))),
4812                    per_milli/10, per_milli % 10,
4813                    (unsigned long long) resync,
4814                    (unsigned long long) max_blocks);
4815
4816         /*
4817          * We do not want to overflow, so the order of operands and
4818          * the * 100 / 100 trick are important. We do a +1 to be
4819          * safe against division by zero. We only estimate anyway.
4820          *
4821          * dt: time from mark until now
4822          * db: blocks written from mark until now
4823          * rt: remaining time
4824          */
4825         dt = ((jiffies - mddev->resync_mark) / HZ);
4826         if (!dt) dt++;
4827         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4828                 - mddev->resync_mark_cnt;
4829         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4830
4831         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4832
4833         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4834 }
4835
4836 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4837 {
4838         struct list_head *tmp;
4839         loff_t l = *pos;
4840         mddev_t *mddev;
4841
4842         if (l >= 0x10000)
4843                 return NULL;
4844         if (!l--)
4845                 /* header */
4846                 return (void*)1;
4847
4848         spin_lock(&all_mddevs_lock);
4849         list_for_each(tmp,&all_mddevs)
4850                 if (!l--) {
4851                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4852                         mddev_get(mddev);
4853                         spin_unlock(&all_mddevs_lock);
4854                         return mddev;
4855                 }
4856         spin_unlock(&all_mddevs_lock);
4857         if (!l--)
4858                 return (void*)2;/* tail */
4859         return NULL;
4860 }
4861
4862 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4863 {
4864         struct list_head *tmp;
4865         mddev_t *next_mddev, *mddev = v;
4866         
4867         ++*pos;
4868         if (v == (void*)2)
4869                 return NULL;
4870
4871         spin_lock(&all_mddevs_lock);
4872         if (v == (void*)1)
4873                 tmp = all_mddevs.next;
4874         else
4875                 tmp = mddev->all_mddevs.next;
4876         if (tmp != &all_mddevs)
4877                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4878         else {
4879                 next_mddev = (void*)2;
4880                 *pos = 0x10000;
4881         }               
4882         spin_unlock(&all_mddevs_lock);
4883
4884         if (v != (void*)1)
4885                 mddev_put(mddev);
4886         return next_mddev;
4887
4888 }
4889
4890 static void md_seq_stop(struct seq_file *seq, void *v)
4891 {
4892         mddev_t *mddev = v;
4893
4894         if (mddev && v != (void*)1 && v != (void*)2)
4895                 mddev_put(mddev);
4896 }
4897
4898 struct mdstat_info {
4899         int event;
4900 };
4901
4902 static int md_seq_show(struct seq_file *seq, void *v)
4903 {
4904         mddev_t *mddev = v;
4905         sector_t size;
4906         struct list_head *tmp2;
4907         mdk_rdev_t *rdev;
4908         struct mdstat_info *mi = seq->private;
4909         struct bitmap *bitmap;
4910
4911         if (v == (void*)1) {
4912                 struct mdk_personality *pers;
4913                 seq_printf(seq, "Personalities : ");
4914                 spin_lock(&pers_lock);
4915                 list_for_each_entry(pers, &pers_list, list)
4916                         seq_printf(seq, "[%s] ", pers->name);
4917
4918                 spin_unlock(&pers_lock);
4919                 seq_printf(seq, "\n");
4920                 mi->event = atomic_read(&md_event_count);
4921                 return 0;
4922         }
4923         if (v == (void*)2) {
4924                 status_unused(seq);
4925                 return 0;
4926         }
4927
4928         if (mddev_lock(mddev) < 0)
4929                 return -EINTR;
4930
4931         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4932                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4933                                                 mddev->pers ? "" : "in");
4934                 if (mddev->pers) {
4935                         if (mddev->ro==1)
4936                                 seq_printf(seq, " (read-only)");
4937                         if (mddev->ro==2)
4938                                 seq_printf(seq, "(auto-read-only)");
4939                         seq_printf(seq, " %s", mddev->pers->name);
4940                 }
4941
4942                 size = 0;
4943                 ITERATE_RDEV(mddev,rdev,tmp2) {
4944                         char b[BDEVNAME_SIZE];
4945                         seq_printf(seq, " %s[%d]",
4946                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4947                         if (test_bit(WriteMostly, &rdev->flags))
4948                                 seq_printf(seq, "(W)");
4949                         if (test_bit(Faulty, &rdev->flags)) {
4950                                 seq_printf(seq, "(F)");
4951                                 continue;
4952                         } else if (rdev->raid_disk < 0)
4953                                 seq_printf(seq, "(S)"); /* spare */
4954                         size += rdev->size;
4955                 }
4956
4957                 if (!list_empty(&mddev->disks)) {
4958                         if (mddev->pers)
4959                                 seq_printf(seq, "\n      %llu blocks",
4960                                         (unsigned long long)mddev->array_size);
4961                         else
4962                                 seq_printf(seq, "\n      %llu blocks",
4963                                         (unsigned long long)size);
4964                 }
4965                 if (mddev->persistent) {
4966                         if (mddev->major_version != 0 ||
4967                             mddev->minor_version != 90) {
4968                                 seq_printf(seq," super %d.%d",
4969                                            mddev->major_version,
4970                                            mddev->minor_version);
4971                         }
4972                 } else
4973                         seq_printf(seq, " super non-persistent");
4974
4975                 if (mddev->pers) {
4976                         mddev->pers->status (seq, mddev);
4977                         seq_printf(seq, "\n      ");
4978                         if (mddev->pers->sync_request) {
4979                                 if (mddev->curr_resync > 2) {
4980                                         status_resync (seq, mddev);
4981                                         seq_printf(seq, "\n      ");
4982                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4983                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4984                                 else if (mddev->recovery_cp < MaxSector)
4985                                         seq_printf(seq, "\tresync=PENDING\n      ");
4986                         }
4987                 } else
4988                         seq_printf(seq, "\n       ");
4989
4990                 if ((bitmap = mddev->bitmap)) {
4991                         unsigned long chunk_kb;
4992                         unsigned long flags;
4993                         spin_lock_irqsave(&bitmap->lock, flags);
4994                         chunk_kb = bitmap->chunksize >> 10;
4995                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4996                                 "%lu%s chunk",
4997                                 bitmap->pages - bitmap->missing_pages,
4998                                 bitmap->pages,
4999                                 (bitmap->pages - bitmap->missing_pages)
5000                                         << (PAGE_SHIFT - 10),
5001                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5002                                 chunk_kb ? "KB" : "B");
5003                         if (bitmap->file) {
5004                                 seq_printf(seq, ", file: ");
5005                                 seq_path(seq, bitmap->file->f_path.mnt,
5006                                          bitmap->file->f_path.dentry," \t\n");
5007                         }
5008
5009                         seq_printf(seq, "\n");
5010                         spin_unlock_irqrestore(&bitmap->lock, flags);
5011                 }
5012
5013                 seq_printf(seq, "\n");
5014         }
5015         mddev_unlock(mddev);
5016         
5017         return 0;
5018 }
5019
5020 static struct seq_operations md_seq_ops = {
5021         .start  = md_seq_start,
5022         .next   = md_seq_next,
5023         .stop   = md_seq_stop,
5024         .show   = md_seq_show,
5025 };
5026
5027 static int md_seq_open(struct inode *inode, struct file *file)
5028 {
5029         int error;
5030         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5031         if (mi == NULL)
5032                 return -ENOMEM;
5033
5034         error = seq_open(file, &md_seq_ops);
5035         if (error)
5036                 kfree(mi);
5037         else {
5038                 struct seq_file *p = file->private_data;
5039                 p->private = mi;
5040                 mi->event = atomic_read(&md_event_count);
5041         }
5042         return error;
5043 }
5044
5045 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5046 {
5047         struct seq_file *m = filp->private_data;
5048         struct mdstat_info *mi = m->private;
5049         int mask;
5050
5051         poll_wait(filp, &md_event_waiters, wait);
5052
5053         /* always allow read */
5054         mask = POLLIN | POLLRDNORM;
5055
5056         if (mi->event != atomic_read(&md_event_count))
5057                 mask |= POLLERR | POLLPRI;
5058         return mask;
5059 }
5060
5061 static const struct file_operations md_seq_fops = {
5062         .owner          = THIS_MODULE,
5063         .open           = md_seq_open,
5064         .read           = seq_read,
5065         .llseek         = seq_lseek,
5066         .release        = seq_release_private,
5067         .poll           = mdstat_poll,
5068 };
5069
5070 int register_md_personality(struct mdk_personality *p)
5071 {
5072         spin_lock(&pers_lock);
5073         list_add_tail(&p->list, &pers_list);
5074         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5075         spin_unlock(&pers_lock);
5076         return 0;
5077 }
5078
5079 int unregister_md_personality(struct mdk_personality *p)
5080 {
5081         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5082         spin_lock(&pers_lock);
5083         list_del_init(&p->list);
5084         spin_unlock(&pers_lock);
5085         return 0;
5086 }
5087
5088 static int is_mddev_idle(mddev_t *mddev)
5089 {
5090         mdk_rdev_t * rdev;
5091         struct list_head *tmp;
5092         int idle;
5093         unsigned long curr_events;
5094
5095         idle = 1;
5096         ITERATE_RDEV(mddev,rdev,tmp) {
5097                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5098                 curr_events = disk_stat_read(disk, sectors[0]) + 
5099                                 disk_stat_read(disk, sectors[1]) - 
5100                                 atomic_read(&disk->sync_io);
5101                 /* The difference between curr_events and last_events
5102                  * will be affected by any new non-sync IO (making
5103                  * curr_events bigger) and any difference in the amount of
5104                  * in-flight syncio (making current_events bigger or smaller)
5105                  * The amount in-flight is currently limited to
5106                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
5107                  * which is at most 4096 sectors.
5108                  * These numbers are fairly fragile and should be made
5109                  * more robust, probably by enforcing the
5110                  * 'window size' that md_do_sync sort-of uses.
5111                  *
5112                  * Note: the following is an unsigned comparison.
5113                  */
5114                 if ((long)curr_events - (long)rdev->last_events > 4096) {
5115                         rdev->last_events = curr_events;
5116                         idle = 0;
5117                 }
5118         }
5119         return idle;
5120 }
5121
5122 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5123 {
5124         /* another "blocks" (512byte) blocks have been synced */
5125         atomic_sub(blocks, &mddev->recovery_active);
5126         wake_up(&mddev->recovery_wait);
5127         if (!ok) {
5128                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5129                 md_wakeup_thread(mddev->thread);
5130                 // stop recovery, signal do_sync ....
5131         }
5132 }
5133
5134
5135 /* md_write_start(mddev, bi)
5136  * If we need to update some array metadata (e.g. 'active' flag
5137  * in superblock) before writing, schedule a superblock update
5138  * and wait for it to complete.
5139  */
5140 void md_write_start(mddev_t *mddev, struct bio *bi)
5141 {
5142         if (bio_data_dir(bi) != WRITE)
5143                 return;
5144
5145         BUG_ON(mddev->ro == 1);
5146         if (mddev->ro == 2) {
5147                 /* need to switch to read/write */
5148                 mddev->ro = 0;
5149                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5150                 md_wakeup_thread(mddev->thread);
5151         }
5152         atomic_inc(&mddev->writes_pending);
5153         if (mddev->in_sync) {
5154                 spin_lock_irq(&mddev->write_lock);
5155                 if (mddev->in_sync) {
5156                         mddev->in_sync = 0;
5157                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5158                         md_wakeup_thread(mddev->thread);
5159                 }
5160                 spin_unlock_irq(&mddev->write_lock);
5161         }
5162         wait_event(mddev->sb_wait, mddev->flags==0);
5163 }
5164
5165 void md_write_end(mddev_t *mddev)
5166 {
5167         if (atomic_dec_and_test(&mddev->writes_pending)) {
5168                 if (mddev->safemode == 2)
5169                         md_wakeup_thread(mddev->thread);
5170                 else if (mddev->safemode_delay)
5171                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5172         }
5173 }
5174
5175 /* md_allow_write(mddev)
5176  * Calling this ensures that the array is marked 'active' so that writes
5177  * may proceed without blocking.  It is important to call this before
5178  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5179  * Must be called with mddev_lock held.
5180  */
5181 void md_allow_write(mddev_t *mddev)
5182 {
5183         if (!mddev->pers)
5184                 return;
5185         if (mddev->ro)
5186                 return;
5187
5188         spin_lock_irq(&mddev->write_lock);
5189         if (mddev->in_sync) {
5190                 mddev->in_sync = 0;
5191                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5192                 if (mddev->safemode_delay &&
5193                     mddev->safemode == 0)
5194                         mddev->safemode = 1;
5195                 spin_unlock_irq(&mddev->write_lock);
5196                 md_update_sb(mddev, 0);
5197         } else
5198                 spin_unlock_irq(&mddev->write_lock);
5199 }
5200 EXPORT_SYMBOL_GPL(md_allow_write);
5201
5202 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5203
5204 #define SYNC_MARKS      10
5205 #define SYNC_MARK_STEP  (3*HZ)
5206 void md_do_sync(mddev_t *mddev)
5207 {
5208         mddev_t *mddev2;
5209         unsigned int currspeed = 0,
5210                  window;
5211         sector_t max_sectors,j, io_sectors;
5212         unsigned long mark[SYNC_MARKS];
5213         sector_t mark_cnt[SYNC_MARKS];
5214         int last_mark,m;
5215         struct list_head *tmp;
5216         sector_t last_check;
5217         int skipped = 0;
5218         struct list_head *rtmp;
5219         mdk_rdev_t *rdev;
5220         char *desc;
5221
5222         /* just incase thread restarts... */
5223         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5224                 return;
5225         if (mddev->ro) /* never try to sync a read-only array */
5226                 return;
5227
5228         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5229                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5230                         desc = "data-check";
5231                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5232                         desc = "requested-resync";
5233                 else
5234                         desc = "resync";
5235         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5236                 desc = "reshape";
5237         else
5238                 desc = "recovery";
5239
5240         /* we overload curr_resync somewhat here.
5241          * 0 == not engaged in resync at all
5242          * 2 == checking that there is no conflict with another sync
5243          * 1 == like 2, but have yielded to allow conflicting resync to
5244          *              commense
5245          * other == active in resync - this many blocks
5246          *
5247          * Before starting a resync we must have set curr_resync to
5248          * 2, and then checked that every "conflicting" array has curr_resync
5249          * less than ours.  When we find one that is the same or higher
5250          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5251          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5252          * This will mean we have to start checking from the beginning again.
5253          *
5254          */
5255
5256         do {
5257                 mddev->curr_resync = 2;
5258
5259         try_again:
5260                 if (kthread_should_stop()) {
5261                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5262                         goto skip;
5263                 }
5264                 ITERATE_MDDEV(mddev2,tmp) {
5265                         if (mddev2 == mddev)
5266                                 continue;
5267                         if (mddev2->curr_resync && 
5268                             match_mddev_units(mddev,mddev2)) {
5269                                 DEFINE_WAIT(wq);
5270                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5271                                         /* arbitrarily yield */
5272                                         mddev->curr_resync = 1;
5273                                         wake_up(&resync_wait);
5274                                 }
5275                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5276                                         /* no need to wait here, we can wait the next
5277                                          * time 'round when curr_resync == 2
5278                                          */
5279                                         continue;
5280                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5281                                 if (!kthread_should_stop() &&
5282                                     mddev2->curr_resync >= mddev->curr_resync) {
5283                                         printk(KERN_INFO "md: delaying %s of %s"
5284                                                " until %s has finished (they"
5285                                                " share one or more physical units)\n",
5286                                                desc, mdname(mddev), mdname(mddev2));
5287                                         mddev_put(mddev2);
5288                                         schedule();
5289                                         finish_wait(&resync_wait, &wq);
5290                                         goto try_again;
5291                                 }
5292                                 finish_wait(&resync_wait, &wq);
5293                         }
5294                 }
5295         } while (mddev->curr_resync < 2);
5296
5297         j = 0;
5298         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5299                 /* resync follows the size requested by the personality,
5300                  * which defaults to physical size, but can be virtual size
5301                  */
5302                 max_sectors = mddev->resync_max_sectors;
5303                 mddev->resync_mismatches = 0;
5304                 /* we don't use the checkpoint if there's a bitmap */
5305                 if (!mddev->bitmap &&
5306                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5307                         j = mddev->recovery_cp;
5308         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5309                 max_sectors = mddev->size << 1;
5310         else {
5311                 /* recovery follows the physical size of devices */
5312                 max_sectors = mddev->size << 1;
5313                 j = MaxSector;
5314                 ITERATE_RDEV(mddev,rdev,rtmp)
5315                         if (rdev->raid_disk >= 0 &&
5316                             !test_bit(Faulty, &rdev->flags) &&
5317                             !test_bit(In_sync, &rdev->flags) &&
5318                             rdev->recovery_offset < j)
5319                                 j = rdev->recovery_offset;
5320         }
5321
5322         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5323         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5324                 " %d KB/sec/disk.\n", speed_min(mddev));
5325         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5326                "(but not more than %d KB/sec) for %s.\n",
5327                speed_max(mddev), desc);
5328
5329         is_mddev_idle(mddev); /* this also initializes IO event counters */
5330
5331         io_sectors = 0;
5332         for (m = 0; m < SYNC_MARKS; m++) {
5333                 mark[m] = jiffies;
5334                 mark_cnt[m] = io_sectors;
5335         }
5336         last_mark = 0;
5337         mddev->resync_mark = mark[last_mark];
5338         mddev->resync_mark_cnt = mark_cnt[last_mark];
5339
5340         /*
5341          * Tune reconstruction:
5342          */
5343         window = 32*(PAGE_SIZE/512);
5344         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5345                 window/2,(unsigned long long) max_sectors/2);
5346
5347         atomic_set(&mddev->recovery_active, 0);
5348         init_waitqueue_head(&mddev->recovery_wait);
5349         last_check = 0;
5350
5351         if (j>2) {
5352                 printk(KERN_INFO 
5353                        "md: resuming %s of %s from checkpoint.\n",
5354                        desc, mdname(mddev));
5355                 mddev->curr_resync = j;
5356         }
5357
5358         while (j < max_sectors) {
5359                 sector_t sectors;
5360
5361                 skipped = 0;
5362                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5363                                             currspeed < speed_min(mddev));
5364                 if (sectors == 0) {
5365                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5366                         goto out;
5367                 }
5368
5369                 if (!skipped) { /* actual IO requested */
5370                         io_sectors += sectors;
5371                         atomic_add(sectors, &mddev->recovery_active);
5372                 }
5373
5374                 j += sectors;
5375                 if (j>1) mddev->curr_resync = j;
5376                 mddev->curr_mark_cnt = io_sectors;
5377                 if (last_check == 0)
5378                         /* this is the earliers that rebuilt will be
5379                          * visible in /proc/mdstat
5380                          */
5381                         md_new_event(mddev);
5382
5383                 if (last_check + window > io_sectors || j == max_sectors)
5384                         continue;
5385
5386                 last_check = io_sectors;
5387
5388                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5389                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5390                         break;
5391
5392         repeat:
5393                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5394                         /* step marks */
5395                         int next = (last_mark+1) % SYNC_MARKS;
5396
5397                         mddev->resync_mark = mark[next];
5398                         mddev->resync_mark_cnt = mark_cnt[next];
5399                         mark[next] = jiffies;
5400                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5401                         last_mark = next;
5402                 }
5403
5404
5405                 if (kthread_should_stop()) {
5406                         /*
5407                          * got a signal, exit.
5408                          */
5409                         printk(KERN_INFO 
5410                                 "md: md_do_sync() got signal ... exiting\n");
5411                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5412                         goto out;
5413                 }
5414
5415                 /*
5416                  * this loop exits only if either when we are slower than
5417                  * the 'hard' speed limit, or the system was IO-idle for
5418                  * a jiffy.
5419                  * the system might be non-idle CPU-wise, but we only care
5420                  * about not overloading the IO subsystem. (things like an
5421                  * e2fsck being done on the RAID array should execute fast)
5422                  */
5423                 mddev->queue->unplug_fn(mddev->queue);
5424                 cond_resched();
5425
5426                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5427                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5428
5429                 if (currspeed > speed_min(mddev)) {
5430                         if ((currspeed > speed_max(mddev)) ||
5431                                         !is_mddev_idle(mddev)) {
5432                                 msleep(500);
5433                                 goto repeat;
5434                         }
5435                 }
5436         }
5437         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5438         /*
5439          * this also signals 'finished resyncing' to md_stop
5440          */
5441  out:
5442         mddev->queue->unplug_fn(mddev->queue);
5443
5444         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5445
5446         /* tell personality that we are finished */
5447         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5448
5449         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5450             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5451             mddev->curr_resync > 2) {
5452                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5453                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5454                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5455                                         printk(KERN_INFO
5456                                                "md: checkpointing %s of %s.\n",
5457                                                desc, mdname(mddev));
5458                                         mddev->recovery_cp = mddev->curr_resync;
5459                                 }
5460                         } else
5461                                 mddev->recovery_cp = MaxSector;
5462                 } else {
5463                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5464                                 mddev->curr_resync = MaxSector;
5465                         ITERATE_RDEV(mddev,rdev,rtmp)
5466                                 if (rdev->raid_disk >= 0 &&
5467                                     !test_bit(Faulty, &rdev->flags) &&
5468                                     !test_bit(In_sync, &rdev->flags) &&
5469                                     rdev->recovery_offset < mddev->curr_resync)
5470                                         rdev->recovery_offset = mddev->curr_resync;
5471                 }
5472         }
5473         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5474
5475  skip:
5476         mddev->curr_resync = 0;
5477         wake_up(&resync_wait);
5478         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5479         md_wakeup_thread(mddev->thread);
5480 }
5481 EXPORT_SYMBOL_GPL(md_do_sync);
5482
5483
5484 static int remove_and_add_spares(mddev_t *mddev)
5485 {
5486         mdk_rdev_t *rdev;
5487         struct list_head *rtmp;
5488         int spares = 0;
5489
5490         ITERATE_RDEV(mddev,rdev,rtmp)
5491                 if (rdev->raid_disk >= 0 &&
5492                     (test_bit(Faulty, &rdev->flags) ||
5493                      ! test_bit(In_sync, &rdev->flags)) &&
5494                     atomic_read(&rdev->nr_pending)==0) {
5495                         if (mddev->pers->hot_remove_disk(
5496                                     mddev, rdev->raid_disk)==0) {
5497                                 char nm[20];
5498                                 sprintf(nm,"rd%d", rdev->raid_disk);
5499                                 sysfs_remove_link(&mddev->kobj, nm);
5500                                 rdev->raid_disk = -1;
5501                         }
5502                 }
5503
5504         if (mddev->degraded) {
5505                 ITERATE_RDEV(mddev,rdev,rtmp)
5506                         if (rdev->raid_disk < 0
5507                             && !test_bit(Faulty, &rdev->flags)) {
5508                                 rdev->recovery_offset = 0;
5509                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5510                                         char nm[20];
5511                                         sprintf(nm, "rd%d", rdev->raid_disk);
5512                                         if (sysfs_create_link(&mddev->kobj,
5513                                                               &rdev->kobj, nm))
5514                                                 printk(KERN_WARNING
5515                                                        "md: cannot register "
5516                                                        "%s for %s\n",
5517                                                        nm, mdname(mddev));
5518                                         spares++;
5519                                         md_new_event(mddev);
5520                                 } else
5521                                         break;
5522                         }
5523         }
5524         return spares;
5525 }
5526 /*
5527  * This routine is regularly called by all per-raid-array threads to
5528  * deal with generic issues like resync and super-block update.
5529  * Raid personalities that don't have a thread (linear/raid0) do not
5530  * need this as they never do any recovery or update the superblock.
5531  *
5532  * It does not do any resync itself, but rather "forks" off other threads
5533  * to do that as needed.
5534  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5535  * "->recovery" and create a thread at ->sync_thread.
5536  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5537  * and wakeups up this thread which will reap the thread and finish up.
5538  * This thread also removes any faulty devices (with nr_pending == 0).
5539  *
5540  * The overall approach is:
5541  *  1/ if the superblock needs updating, update it.
5542  *  2/ If a recovery thread is running, don't do anything else.
5543  *  3/ If recovery has finished, clean up, possibly marking spares active.
5544  *  4/ If there are any faulty devices, remove them.
5545  *  5/ If array is degraded, try to add spares devices
5546  *  6/ If array has spares or is not in-sync, start a resync thread.
5547  */
5548 void md_check_recovery(mddev_t *mddev)
5549 {
5550         mdk_rdev_t *rdev;
5551         struct list_head *rtmp;
5552
5553
5554         if (mddev->bitmap)
5555                 bitmap_daemon_work(mddev->bitmap);
5556
5557         if (mddev->ro)
5558                 return;
5559
5560         if (signal_pending(current)) {
5561                 if (mddev->pers->sync_request) {
5562                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5563                                mdname(mddev));
5564                         mddev->safemode = 2;
5565                 }
5566                 flush_signals(current);
5567         }
5568
5569         if ( ! (
5570                 mddev->flags ||
5571                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5572                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5573                 (mddev->safemode == 1) ||
5574                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5575                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5576                 ))
5577                 return;
5578
5579         if (mddev_trylock(mddev)) {
5580                 int spares = 0;
5581
5582                 spin_lock_irq(&mddev->write_lock);
5583                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5584                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5585                         mddev->in_sync = 1;
5586                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5587                 }
5588                 if (mddev->safemode == 1)
5589                         mddev->safemode = 0;
5590                 spin_unlock_irq(&mddev->write_lock);
5591
5592                 if (mddev->flags)
5593                         md_update_sb(mddev, 0);
5594
5595
5596                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5597                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5598                         /* resync/recovery still happening */
5599                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5600                         goto unlock;
5601                 }
5602                 if (mddev->sync_thread) {
5603                         /* resync has finished, collect result */
5604                         md_unregister_thread(mddev->sync_thread);
5605                         mddev->sync_thread = NULL;
5606                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5607                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5608                                 /* success...*/
5609                                 /* activate any spares */
5610                                 mddev->pers->spare_active(mddev);
5611                         }
5612                         md_update_sb(mddev, 1);
5613
5614                         /* if array is no-longer degraded, then any saved_raid_disk
5615                          * information must be scrapped
5616                          */
5617                         if (!mddev->degraded)
5618                                 ITERATE_RDEV(mddev,rdev,rtmp)
5619                                         rdev->saved_raid_disk = -1;
5620
5621                         mddev->recovery = 0;
5622                         /* flag recovery needed just to double check */
5623                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5624                         md_new_event(mddev);
5625                         goto unlock;
5626                 }
5627                 /* Clear some bits that don't mean anything, but
5628                  * might be left set
5629                  */
5630                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5631                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5632                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5633                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5634
5635                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5636                         goto unlock;
5637                 /* no recovery is running.
5638                  * remove any failed drives, then
5639                  * add spares if possible.
5640                  * Spare are also removed and re-added, to allow
5641                  * the personality to fail the re-add.
5642                  */
5643
5644                 if (mddev->reshape_position != MaxSector) {
5645                         if (mddev->pers->check_reshape(mddev) != 0)
5646                                 /* Cannot proceed */
5647                                 goto unlock;
5648                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5649                 } else if ((spares = remove_and_add_spares(mddev))) {
5650                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5651                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5652                 } else if (mddev->recovery_cp < MaxSector) {
5653                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5654                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5655                         /* nothing to be done ... */
5656                         goto unlock;
5657
5658                 if (mddev->pers->sync_request) {
5659                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5660                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5661                                 /* We are adding a device or devices to an array
5662                                  * which has the bitmap stored on all devices.
5663                                  * So make sure all bitmap pages get written
5664                                  */
5665                                 bitmap_write_all(mddev->bitmap);
5666                         }
5667                         mddev->sync_thread = md_register_thread(md_do_sync,
5668                                                                 mddev,
5669                                                                 "%s_resync");
5670                         if (!mddev->sync_thread) {
5671                                 printk(KERN_ERR "%s: could not start resync"
5672                                         " thread...\n", 
5673                                         mdname(mddev));
5674                                 /* leave the spares where they are, it shouldn't hurt */
5675                                 mddev->recovery = 0;
5676                         } else
5677                                 md_wakeup_thread(mddev->sync_thread);
5678                         md_new_event(mddev);
5679                 }
5680         unlock:
5681                 mddev_unlock(mddev);
5682         }
5683 }
5684
5685 static int md_notify_reboot(struct notifier_block *this,
5686                             unsigned long code, void *x)
5687 {
5688         struct list_head *tmp;
5689         mddev_t *mddev;
5690
5691         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5692
5693                 printk(KERN_INFO "md: stopping all md devices.\n");
5694
5695                 ITERATE_MDDEV(mddev,tmp)
5696                         if (mddev_trylock(mddev)) {
5697                                 do_md_stop (mddev, 1);
5698                                 mddev_unlock(mddev);
5699                         }
5700                 /*
5701                  * certain more exotic SCSI devices are known to be
5702                  * volatile wrt too early system reboots. While the
5703                  * right place to handle this issue is the given
5704                  * driver, we do want to have a safe RAID driver ...
5705                  */
5706                 mdelay(1000*1);
5707         }
5708         return NOTIFY_DONE;
5709 }
5710
5711 static struct notifier_block md_notifier = {
5712         .notifier_call  = md_notify_reboot,
5713         .next           = NULL,
5714         .priority       = INT_MAX, /* before any real devices */
5715 };
5716
5717 static void md_geninit(void)
5718 {
5719         struct proc_dir_entry *p;
5720
5721         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5722
5723         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5724         if (p)
5725                 p->proc_fops = &md_seq_fops;
5726 }
5727
5728 static int __init md_init(void)
5729 {
5730         if (register_blkdev(MAJOR_NR, "md"))
5731                 return -1;
5732         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5733                 unregister_blkdev(MAJOR_NR, "md");
5734                 return -1;
5735         }
5736         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5737                             md_probe, NULL, NULL);
5738         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5739                             md_probe, NULL, NULL);
5740
5741         register_reboot_notifier(&md_notifier);
5742         raid_table_header = register_sysctl_table(raid_root_table);
5743
5744         md_geninit();
5745         return (0);
5746 }
5747
5748
5749 #ifndef MODULE
5750
5751 /*
5752  * Searches all registered partitions for autorun RAID arrays
5753  * at boot time.
5754  */
5755 static dev_t detected_devices[128];
5756 static int dev_cnt;
5757
5758 void md_autodetect_dev(dev_t dev)
5759 {
5760         if (dev_cnt >= 0 && dev_cnt < 127)
5761                 detected_devices[dev_cnt++] = dev;
5762 }
5763
5764
5765 static void autostart_arrays(int part)
5766 {
5767         mdk_rdev_t *rdev;
5768         int i;
5769
5770         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5771
5772         for (i = 0; i < dev_cnt; i++) {
5773                 dev_t dev = detected_devices[i];
5774
5775                 rdev = md_import_device(dev,0, 0);
5776                 if (IS_ERR(rdev))
5777                         continue;
5778
5779                 if (test_bit(Faulty, &rdev->flags)) {
5780                         MD_BUG();
5781                         continue;
5782                 }
5783                 list_add(&rdev->same_set, &pending_raid_disks);
5784         }
5785         dev_cnt = 0;
5786
5787         autorun_devices(part);
5788 }
5789
5790 #endif /* !MODULE */
5791
5792 static __exit void md_exit(void)
5793 {
5794         mddev_t *mddev;
5795         struct list_head *tmp;
5796
5797         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5798         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5799
5800         unregister_blkdev(MAJOR_NR,"md");
5801         unregister_blkdev(mdp_major, "mdp");
5802         unregister_reboot_notifier(&md_notifier);
5803         unregister_sysctl_table(raid_table_header);
5804         remove_proc_entry("mdstat", NULL);
5805         ITERATE_MDDEV(mddev,tmp) {
5806                 struct gendisk *disk = mddev->gendisk;
5807                 if (!disk)
5808                         continue;
5809                 export_array(mddev);
5810                 del_gendisk(disk);
5811                 put_disk(disk);
5812                 mddev->gendisk = NULL;
5813                 mddev_put(mddev);
5814         }
5815 }
5816
5817 module_init(md_init)
5818 module_exit(md_exit)
5819
5820 static int get_ro(char *buffer, struct kernel_param *kp)
5821 {
5822         return sprintf(buffer, "%d", start_readonly);
5823 }
5824 static int set_ro(const char *val, struct kernel_param *kp)
5825 {
5826         char *e;
5827         int num = simple_strtoul(val, &e, 10);
5828         if (*val && (*e == '\0' || *e == '\n')) {
5829                 start_readonly = num;
5830                 return 0;
5831         }
5832         return -EINVAL;
5833 }
5834
5835 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5836 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5837
5838
5839 EXPORT_SYMBOL(register_md_personality);
5840 EXPORT_SYMBOL(unregister_md_personality);
5841 EXPORT_SYMBOL(md_error);
5842 EXPORT_SYMBOL(md_done_sync);
5843 EXPORT_SYMBOL(md_write_start);
5844 EXPORT_SYMBOL(md_write_end);
5845 EXPORT_SYMBOL(md_register_thread);
5846 EXPORT_SYMBOL(md_unregister_thread);
5847 EXPORT_SYMBOL(md_wakeup_thread);
5848 EXPORT_SYMBOL(md_check_recovery);
5849 MODULE_LICENSE("GPL");
5850 MODULE_ALIAS("md");
5851 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);