Merge branch 'merge'
[linux-2.6] / 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/kthread.h>
37 #include <linux/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/buffer_head.h> /* for invalidate_bdev */
42 #include <linux/suspend.h>
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46
47 #include <linux/init.h>
48
49 #include <linux/file.h>
50
51 #ifdef CONFIG_KMOD
52 #include <linux/kmod.h>
53 #endif
54
55 #include <asm/unaligned.h>
56
57 #define MAJOR_NR MD_MAJOR
58 #define MD_DRIVER
59
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
62
63 #define DEBUG 0
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67 #ifndef MODULE
68 static void autostart_arrays (int part);
69 #endif
70
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
73
74 static void md_print_devices(void);
75
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
78 /*
79  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80  * is 1000 KB/sec, so the extra system load does not show up that much.
81  * Increase it if you want to have more _guaranteed_ speed. Note that
82  * the RAID driver will use the maximum available bandwidth if the IO
83  * subsystem is idle. There is also an 'absolute maximum' reconstruction
84  * speed limit - in case reconstruction slows down your system despite
85  * idle IO detection.
86  *
87  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88  * or /sys/block/mdX/md/sync_speed_{min,max}
89  */
90
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
94 {
95         return mddev->sync_speed_min ?
96                 mddev->sync_speed_min : sysctl_speed_limit_min;
97 }
98
99 static inline int speed_max(mddev_t *mddev)
100 {
101         return mddev->sync_speed_max ?
102                 mddev->sync_speed_max : sysctl_speed_limit_max;
103 }
104
105 static struct ctl_table_header *raid_table_header;
106
107 static ctl_table raid_table[] = {
108         {
109                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
110                 .procname       = "speed_limit_min",
111                 .data           = &sysctl_speed_limit_min,
112                 .maxlen         = sizeof(int),
113                 .mode           = S_IRUGO|S_IWUSR,
114                 .proc_handler   = &proc_dointvec,
115         },
116         {
117                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
118                 .procname       = "speed_limit_max",
119                 .data           = &sysctl_speed_limit_max,
120                 .maxlen         = sizeof(int),
121                 .mode           = S_IRUGO|S_IWUSR,
122                 .proc_handler   = &proc_dointvec,
123         },
124         { .ctl_name = 0 }
125 };
126
127 static ctl_table raid_dir_table[] = {
128         {
129                 .ctl_name       = DEV_RAID,
130                 .procname       = "raid",
131                 .maxlen         = 0,
132                 .mode           = S_IRUGO|S_IXUGO,
133                 .child          = raid_table,
134         },
135         { .ctl_name = 0 }
136 };
137
138 static ctl_table raid_root_table[] = {
139         {
140                 .ctl_name       = CTL_DEV,
141                 .procname       = "dev",
142                 .maxlen         = 0,
143                 .mode           = 0555,
144                 .child          = raid_dir_table,
145         },
146         { .ctl_name = 0 }
147 };
148
149 static struct block_device_operations md_fops;
150
151 static int start_readonly;
152
153 /*
154  * We have a system wide 'event count' that is incremented
155  * on any 'interesting' event, and readers of /proc/mdstat
156  * can use 'poll' or 'select' to find out when the event
157  * count increases.
158  *
159  * Events are:
160  *  start array, stop array, error, add device, remove device,
161  *  start build, activate spare
162  */
163 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
164 static atomic_t md_event_count;
165 void md_new_event(mddev_t *mddev)
166 {
167         atomic_inc(&md_event_count);
168         wake_up(&md_event_waiters);
169         sysfs_notify(&mddev->kobj, NULL, "sync_action");
170 }
171 EXPORT_SYMBOL_GPL(md_new_event);
172
173 /* Alternate version that can be called from interrupts
174  * when calling sysfs_notify isn't needed.
175  */
176 static void md_new_event_inintr(mddev_t *mddev)
177 {
178         atomic_inc(&md_event_count);
179         wake_up(&md_event_waiters);
180 }
181
182 /*
183  * Enables to iterate over all existing md arrays
184  * all_mddevs_lock protects this list.
185  */
186 static LIST_HEAD(all_mddevs);
187 static DEFINE_SPINLOCK(all_mddevs_lock);
188
189
190 /*
191  * iterates through all used mddevs in the system.
192  * We take care to grab the all_mddevs_lock whenever navigating
193  * the list, and to always hold a refcount when unlocked.
194  * Any code which breaks out of this loop while own
195  * a reference to the current mddev and must mddev_put it.
196  */
197 #define ITERATE_MDDEV(mddev,tmp)                                        \
198                                                                         \
199         for (({ spin_lock(&all_mddevs_lock);                            \
200                 tmp = all_mddevs.next;                                  \
201                 mddev = NULL;});                                        \
202              ({ if (tmp != &all_mddevs)                                 \
203                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
204                 spin_unlock(&all_mddevs_lock);                          \
205                 if (mddev) mddev_put(mddev);                            \
206                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
207                 tmp != &all_mddevs;});                                  \
208              ({ spin_lock(&all_mddevs_lock);                            \
209                 tmp = tmp->next;})                                      \
210                 )
211
212
213 static int md_fail_request (request_queue_t *q, struct bio *bio)
214 {
215         bio_io_error(bio, bio->bi_size);
216         return 0;
217 }
218
219 static inline mddev_t *mddev_get(mddev_t *mddev)
220 {
221         atomic_inc(&mddev->active);
222         return mddev;
223 }
224
225 static void mddev_put(mddev_t *mddev)
226 {
227         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
228                 return;
229         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
230                 list_del(&mddev->all_mddevs);
231                 spin_unlock(&all_mddevs_lock);
232                 blk_cleanup_queue(mddev->queue);
233                 kobject_unregister(&mddev->kobj);
234         } else
235                 spin_unlock(&all_mddevs_lock);
236 }
237
238 static mddev_t * mddev_find(dev_t unit)
239 {
240         mddev_t *mddev, *new = NULL;
241
242  retry:
243         spin_lock(&all_mddevs_lock);
244         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
245                 if (mddev->unit == unit) {
246                         mddev_get(mddev);
247                         spin_unlock(&all_mddevs_lock);
248                         kfree(new);
249                         return mddev;
250                 }
251
252         if (new) {
253                 list_add(&new->all_mddevs, &all_mddevs);
254                 spin_unlock(&all_mddevs_lock);
255                 return new;
256         }
257         spin_unlock(&all_mddevs_lock);
258
259         new = kzalloc(sizeof(*new), GFP_KERNEL);
260         if (!new)
261                 return NULL;
262
263         new->unit = unit;
264         if (MAJOR(unit) == MD_MAJOR)
265                 new->md_minor = MINOR(unit);
266         else
267                 new->md_minor = MINOR(unit) >> MdpMinorShift;
268
269         mutex_init(&new->reconfig_mutex);
270         INIT_LIST_HEAD(&new->disks);
271         INIT_LIST_HEAD(&new->all_mddevs);
272         init_timer(&new->safemode_timer);
273         atomic_set(&new->active, 1);
274         spin_lock_init(&new->write_lock);
275         init_waitqueue_head(&new->sb_wait);
276
277         new->queue = blk_alloc_queue(GFP_KERNEL);
278         if (!new->queue) {
279                 kfree(new);
280                 return NULL;
281         }
282         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
283
284         blk_queue_make_request(new->queue, md_fail_request);
285
286         goto retry;
287 }
288
289 static inline int mddev_lock(mddev_t * mddev)
290 {
291         return mutex_lock_interruptible(&mddev->reconfig_mutex);
292 }
293
294 static inline int mddev_trylock(mddev_t * mddev)
295 {
296         return mutex_trylock(&mddev->reconfig_mutex);
297 }
298
299 static inline void mddev_unlock(mddev_t * mddev)
300 {
301         mutex_unlock(&mddev->reconfig_mutex);
302
303         md_wakeup_thread(mddev->thread);
304 }
305
306 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
307 {
308         mdk_rdev_t * rdev;
309         struct list_head *tmp;
310
311         ITERATE_RDEV(mddev,rdev,tmp) {
312                 if (rdev->desc_nr == nr)
313                         return rdev;
314         }
315         return NULL;
316 }
317
318 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
319 {
320         struct list_head *tmp;
321         mdk_rdev_t *rdev;
322
323         ITERATE_RDEV(mddev,rdev,tmp) {
324                 if (rdev->bdev->bd_dev == dev)
325                         return rdev;
326         }
327         return NULL;
328 }
329
330 static struct mdk_personality *find_pers(int level, char *clevel)
331 {
332         struct mdk_personality *pers;
333         list_for_each_entry(pers, &pers_list, list) {
334                 if (level != LEVEL_NONE && pers->level == level)
335                         return pers;
336                 if (strcmp(pers->name, clevel)==0)
337                         return pers;
338         }
339         return NULL;
340 }
341
342 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
343 {
344         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
345         return MD_NEW_SIZE_BLOCKS(size);
346 }
347
348 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
349 {
350         sector_t size;
351
352         size = rdev->sb_offset;
353
354         if (chunk_size)
355                 size &= ~((sector_t)chunk_size/1024 - 1);
356         return size;
357 }
358
359 static int alloc_disk_sb(mdk_rdev_t * rdev)
360 {
361         if (rdev->sb_page)
362                 MD_BUG();
363
364         rdev->sb_page = alloc_page(GFP_KERNEL);
365         if (!rdev->sb_page) {
366                 printk(KERN_ALERT "md: out of memory.\n");
367                 return -EINVAL;
368         }
369
370         return 0;
371 }
372
373 static void free_disk_sb(mdk_rdev_t * rdev)
374 {
375         if (rdev->sb_page) {
376                 put_page(rdev->sb_page);
377                 rdev->sb_loaded = 0;
378                 rdev->sb_page = NULL;
379                 rdev->sb_offset = 0;
380                 rdev->size = 0;
381         }
382 }
383
384
385 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
386 {
387         mdk_rdev_t *rdev = bio->bi_private;
388         mddev_t *mddev = rdev->mddev;
389         if (bio->bi_size)
390                 return 1;
391
392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
393                 md_error(mddev, rdev);
394
395         if (atomic_dec_and_test(&mddev->pending_writes))
396                 wake_up(&mddev->sb_wait);
397         bio_put(bio);
398         return 0;
399 }
400
401 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
402 {
403         struct bio *bio2 = bio->bi_private;
404         mdk_rdev_t *rdev = bio2->bi_private;
405         mddev_t *mddev = rdev->mddev;
406         if (bio->bi_size)
407                 return 1;
408
409         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
410             error == -EOPNOTSUPP) {
411                 unsigned long flags;
412                 /* barriers don't appear to be supported :-( */
413                 set_bit(BarriersNotsupp, &rdev->flags);
414                 mddev->barriers_work = 0;
415                 spin_lock_irqsave(&mddev->write_lock, flags);
416                 bio2->bi_next = mddev->biolist;
417                 mddev->biolist = bio2;
418                 spin_unlock_irqrestore(&mddev->write_lock, flags);
419                 wake_up(&mddev->sb_wait);
420                 bio_put(bio);
421                 return 0;
422         }
423         bio_put(bio2);
424         bio->bi_private = rdev;
425         return super_written(bio, bytes_done, error);
426 }
427
428 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
429                    sector_t sector, int size, struct page *page)
430 {
431         /* write first size bytes of page to sector of rdev
432          * Increment mddev->pending_writes before returning
433          * and decrement it on completion, waking up sb_wait
434          * if zero is reached.
435          * If an error occurred, call md_error
436          *
437          * As we might need to resubmit the request if BIO_RW_BARRIER
438          * causes ENOTSUPP, we allocate a spare bio...
439          */
440         struct bio *bio = bio_alloc(GFP_NOIO, 1);
441         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
442
443         bio->bi_bdev = rdev->bdev;
444         bio->bi_sector = sector;
445         bio_add_page(bio, page, size, 0);
446         bio->bi_private = rdev;
447         bio->bi_end_io = super_written;
448         bio->bi_rw = rw;
449
450         atomic_inc(&mddev->pending_writes);
451         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
452                 struct bio *rbio;
453                 rw |= (1<<BIO_RW_BARRIER);
454                 rbio = bio_clone(bio, GFP_NOIO);
455                 rbio->bi_private = bio;
456                 rbio->bi_end_io = super_written_barrier;
457                 submit_bio(rw, rbio);
458         } else
459                 submit_bio(rw, bio);
460 }
461
462 void md_super_wait(mddev_t *mddev)
463 {
464         /* wait for all superblock writes that were scheduled to complete.
465          * if any had to be retried (due to BARRIER problems), retry them
466          */
467         DEFINE_WAIT(wq);
468         for(;;) {
469                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
470                 if (atomic_read(&mddev->pending_writes)==0)
471                         break;
472                 while (mddev->biolist) {
473                         struct bio *bio;
474                         spin_lock_irq(&mddev->write_lock);
475                         bio = mddev->biolist;
476                         mddev->biolist = bio->bi_next ;
477                         bio->bi_next = NULL;
478                         spin_unlock_irq(&mddev->write_lock);
479                         submit_bio(bio->bi_rw, bio);
480                 }
481                 schedule();
482         }
483         finish_wait(&mddev->sb_wait, &wq);
484 }
485
486 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
487 {
488         if (bio->bi_size)
489                 return 1;
490
491         complete((struct completion*)bio->bi_private);
492         return 0;
493 }
494
495 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
496                    struct page *page, int rw)
497 {
498         struct bio *bio = bio_alloc(GFP_NOIO, 1);
499         struct completion event;
500         int ret;
501
502         rw |= (1 << BIO_RW_SYNC);
503
504         bio->bi_bdev = bdev;
505         bio->bi_sector = sector;
506         bio_add_page(bio, page, size, 0);
507         init_completion(&event);
508         bio->bi_private = &event;
509         bio->bi_end_io = bi_complete;
510         submit_bio(rw, bio);
511         wait_for_completion(&event);
512
513         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
514         bio_put(bio);
515         return ret;
516 }
517 EXPORT_SYMBOL_GPL(sync_page_io);
518
519 static int read_disk_sb(mdk_rdev_t * rdev, int size)
520 {
521         char b[BDEVNAME_SIZE];
522         if (!rdev->sb_page) {
523                 MD_BUG();
524                 return -EINVAL;
525         }
526         if (rdev->sb_loaded)
527                 return 0;
528
529
530         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
531                 goto fail;
532         rdev->sb_loaded = 1;
533         return 0;
534
535 fail:
536         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
537                 bdevname(rdev->bdev,b));
538         return -EINVAL;
539 }
540
541 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
542 {
543         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
544                 (sb1->set_uuid1 == sb2->set_uuid1) &&
545                 (sb1->set_uuid2 == sb2->set_uuid2) &&
546                 (sb1->set_uuid3 == sb2->set_uuid3))
547
548                 return 1;
549
550         return 0;
551 }
552
553
554 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
555 {
556         int ret;
557         mdp_super_t *tmp1, *tmp2;
558
559         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
560         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
561
562         if (!tmp1 || !tmp2) {
563                 ret = 0;
564                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
565                 goto abort;
566         }
567
568         *tmp1 = *sb1;
569         *tmp2 = *sb2;
570
571         /*
572          * nr_disks is not constant
573          */
574         tmp1->nr_disks = 0;
575         tmp2->nr_disks = 0;
576
577         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
578                 ret = 0;
579         else
580                 ret = 1;
581
582 abort:
583         kfree(tmp1);
584         kfree(tmp2);
585         return ret;
586 }
587
588 static unsigned int calc_sb_csum(mdp_super_t * sb)
589 {
590         unsigned int disk_csum, csum;
591
592         disk_csum = sb->sb_csum;
593         sb->sb_csum = 0;
594         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
595         sb->sb_csum = disk_csum;
596         return csum;
597 }
598
599
600 /*
601  * Handle superblock details.
602  * We want to be able to handle multiple superblock formats
603  * so we have a common interface to them all, and an array of
604  * different handlers.
605  * We rely on user-space to write the initial superblock, and support
606  * reading and updating of superblocks.
607  * Interface methods are:
608  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
609  *      loads and validates a superblock on dev.
610  *      if refdev != NULL, compare superblocks on both devices
611  *    Return:
612  *      0 - dev has a superblock that is compatible with refdev
613  *      1 - dev has a superblock that is compatible and newer than refdev
614  *          so dev should be used as the refdev in future
615  *     -EINVAL superblock incompatible or invalid
616  *     -othererror e.g. -EIO
617  *
618  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
619  *      Verify that dev is acceptable into mddev.
620  *       The first time, mddev->raid_disks will be 0, and data from
621  *       dev should be merged in.  Subsequent calls check that dev
622  *       is new enough.  Return 0 or -EINVAL
623  *
624  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
625  *     Update the superblock for rdev with data in mddev
626  *     This does not write to disc.
627  *
628  */
629
630 struct super_type  {
631         char            *name;
632         struct module   *owner;
633         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
634         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
635         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
636 };
637
638 /*
639  * load_super for 0.90.0 
640  */
641 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
642 {
643         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
644         mdp_super_t *sb;
645         int ret;
646         sector_t sb_offset;
647
648         /*
649          * Calculate the position of the superblock,
650          * it's at the end of the disk.
651          *
652          * It also happens to be a multiple of 4Kb.
653          */
654         sb_offset = calc_dev_sboffset(rdev->bdev);
655         rdev->sb_offset = sb_offset;
656
657         ret = read_disk_sb(rdev, MD_SB_BYTES);
658         if (ret) return ret;
659
660         ret = -EINVAL;
661
662         bdevname(rdev->bdev, b);
663         sb = (mdp_super_t*)page_address(rdev->sb_page);
664
665         if (sb->md_magic != MD_SB_MAGIC) {
666                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
667                        b);
668                 goto abort;
669         }
670
671         if (sb->major_version != 0 ||
672             sb->minor_version < 90 ||
673             sb->minor_version > 91) {
674                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
675                         sb->major_version, sb->minor_version,
676                         b);
677                 goto abort;
678         }
679
680         if (sb->raid_disks <= 0)
681                 goto abort;
682
683         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
684                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
685                         b);
686                 goto abort;
687         }
688
689         rdev->preferred_minor = sb->md_minor;
690         rdev->data_offset = 0;
691         rdev->sb_size = MD_SB_BYTES;
692
693         if (sb->level == LEVEL_MULTIPATH)
694                 rdev->desc_nr = -1;
695         else
696                 rdev->desc_nr = sb->this_disk.number;
697
698         if (refdev == 0)
699                 ret = 1;
700         else {
701                 __u64 ev1, ev2;
702                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
703                 if (!uuid_equal(refsb, sb)) {
704                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
705                                 b, bdevname(refdev->bdev,b2));
706                         goto abort;
707                 }
708                 if (!sb_equal(refsb, sb)) {
709                         printk(KERN_WARNING "md: %s has same UUID"
710                                " but different superblock to %s\n",
711                                b, bdevname(refdev->bdev, b2));
712                         goto abort;
713                 }
714                 ev1 = md_event(sb);
715                 ev2 = md_event(refsb);
716                 if (ev1 > ev2)
717                         ret = 1;
718                 else 
719                         ret = 0;
720         }
721         rdev->size = calc_dev_size(rdev, sb->chunk_size);
722
723         if (rdev->size < sb->size && sb->level > 1)
724                 /* "this cannot possibly happen" ... */
725                 ret = -EINVAL;
726
727  abort:
728         return ret;
729 }
730
731 /*
732  * validate_super for 0.90.0
733  */
734 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
735 {
736         mdp_disk_t *desc;
737         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
738         __u64 ev1 = md_event(sb);
739
740         rdev->raid_disk = -1;
741         rdev->flags = 0;
742         if (mddev->raid_disks == 0) {
743                 mddev->major_version = 0;
744                 mddev->minor_version = sb->minor_version;
745                 mddev->patch_version = sb->patch_version;
746                 mddev->persistent = ! sb->not_persistent;
747                 mddev->chunk_size = sb->chunk_size;
748                 mddev->ctime = sb->ctime;
749                 mddev->utime = sb->utime;
750                 mddev->level = sb->level;
751                 mddev->clevel[0] = 0;
752                 mddev->layout = sb->layout;
753                 mddev->raid_disks = sb->raid_disks;
754                 mddev->size = sb->size;
755                 mddev->events = ev1;
756                 mddev->bitmap_offset = 0;
757                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
758
759                 if (mddev->minor_version >= 91) {
760                         mddev->reshape_position = sb->reshape_position;
761                         mddev->delta_disks = sb->delta_disks;
762                         mddev->new_level = sb->new_level;
763                         mddev->new_layout = sb->new_layout;
764                         mddev->new_chunk = sb->new_chunk;
765                 } else {
766                         mddev->reshape_position = MaxSector;
767                         mddev->delta_disks = 0;
768                         mddev->new_level = mddev->level;
769                         mddev->new_layout = mddev->layout;
770                         mddev->new_chunk = mddev->chunk_size;
771                 }
772
773                 if (sb->state & (1<<MD_SB_CLEAN))
774                         mddev->recovery_cp = MaxSector;
775                 else {
776                         if (sb->events_hi == sb->cp_events_hi && 
777                                 sb->events_lo == sb->cp_events_lo) {
778                                 mddev->recovery_cp = sb->recovery_cp;
779                         } else
780                                 mddev->recovery_cp = 0;
781                 }
782
783                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
784                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
785                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
786                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
787
788                 mddev->max_disks = MD_SB_DISKS;
789
790                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
791                     mddev->bitmap_file == NULL) {
792                         if (mddev->level != 1 && mddev->level != 4
793                             && mddev->level != 5 && mddev->level != 6
794                             && mddev->level != 10) {
795                                 /* FIXME use a better test */
796                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
797                                 return -EINVAL;
798                         }
799                         mddev->bitmap_offset = mddev->default_bitmap_offset;
800                 }
801
802         } else if (mddev->pers == NULL) {
803                 /* Insist on good event counter while assembling */
804                 ++ev1;
805                 if (ev1 < mddev->events) 
806                         return -EINVAL;
807         } else if (mddev->bitmap) {
808                 /* if adding to array with a bitmap, then we can accept an
809                  * older device ... but not too old.
810                  */
811                 if (ev1 < mddev->bitmap->events_cleared)
812                         return 0;
813         } else {
814                 if (ev1 < mddev->events)
815                         /* just a hot-add of a new device, leave raid_disk at -1 */
816                         return 0;
817         }
818
819         if (mddev->level != LEVEL_MULTIPATH) {
820                 desc = sb->disks + rdev->desc_nr;
821
822                 if (desc->state & (1<<MD_DISK_FAULTY))
823                         set_bit(Faulty, &rdev->flags);
824                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
825                             desc->raid_disk < mddev->raid_disks */) {
826                         set_bit(In_sync, &rdev->flags);
827                         rdev->raid_disk = desc->raid_disk;
828                 }
829                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
830                         set_bit(WriteMostly, &rdev->flags);
831         } else /* MULTIPATH are always insync */
832                 set_bit(In_sync, &rdev->flags);
833         return 0;
834 }
835
836 /*
837  * sync_super for 0.90.0
838  */
839 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
840 {
841         mdp_super_t *sb;
842         struct list_head *tmp;
843         mdk_rdev_t *rdev2;
844         int next_spare = mddev->raid_disks;
845
846
847         /* make rdev->sb match mddev data..
848          *
849          * 1/ zero out disks
850          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
851          * 3/ any empty disks < next_spare become removed
852          *
853          * disks[0] gets initialised to REMOVED because
854          * we cannot be sure from other fields if it has
855          * been initialised or not.
856          */
857         int i;
858         int active=0, working=0,failed=0,spare=0,nr_disks=0;
859
860         rdev->sb_size = MD_SB_BYTES;
861
862         sb = (mdp_super_t*)page_address(rdev->sb_page);
863
864         memset(sb, 0, sizeof(*sb));
865
866         sb->md_magic = MD_SB_MAGIC;
867         sb->major_version = mddev->major_version;
868         sb->patch_version = mddev->patch_version;
869         sb->gvalid_words  = 0; /* ignored */
870         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
871         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
872         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
873         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
874
875         sb->ctime = mddev->ctime;
876         sb->level = mddev->level;
877         sb->size  = mddev->size;
878         sb->raid_disks = mddev->raid_disks;
879         sb->md_minor = mddev->md_minor;
880         sb->not_persistent = !mddev->persistent;
881         sb->utime = mddev->utime;
882         sb->state = 0;
883         sb->events_hi = (mddev->events>>32);
884         sb->events_lo = (u32)mddev->events;
885
886         if (mddev->reshape_position == MaxSector)
887                 sb->minor_version = 90;
888         else {
889                 sb->minor_version = 91;
890                 sb->reshape_position = mddev->reshape_position;
891                 sb->new_level = mddev->new_level;
892                 sb->delta_disks = mddev->delta_disks;
893                 sb->new_layout = mddev->new_layout;
894                 sb->new_chunk = mddev->new_chunk;
895         }
896         mddev->minor_version = sb->minor_version;
897         if (mddev->in_sync)
898         {
899                 sb->recovery_cp = mddev->recovery_cp;
900                 sb->cp_events_hi = (mddev->events>>32);
901                 sb->cp_events_lo = (u32)mddev->events;
902                 if (mddev->recovery_cp == MaxSector)
903                         sb->state = (1<< MD_SB_CLEAN);
904         } else
905                 sb->recovery_cp = 0;
906
907         sb->layout = mddev->layout;
908         sb->chunk_size = mddev->chunk_size;
909
910         if (mddev->bitmap && mddev->bitmap_file == NULL)
911                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
912
913         sb->disks[0].state = (1<<MD_DISK_REMOVED);
914         ITERATE_RDEV(mddev,rdev2,tmp) {
915                 mdp_disk_t *d;
916                 int desc_nr;
917                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
918                     && !test_bit(Faulty, &rdev2->flags))
919                         desc_nr = rdev2->raid_disk;
920                 else
921                         desc_nr = next_spare++;
922                 rdev2->desc_nr = desc_nr;
923                 d = &sb->disks[rdev2->desc_nr];
924                 nr_disks++;
925                 d->number = rdev2->desc_nr;
926                 d->major = MAJOR(rdev2->bdev->bd_dev);
927                 d->minor = MINOR(rdev2->bdev->bd_dev);
928                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
929                     && !test_bit(Faulty, &rdev2->flags))
930                         d->raid_disk = rdev2->raid_disk;
931                 else
932                         d->raid_disk = rdev2->desc_nr; /* compatibility */
933                 if (test_bit(Faulty, &rdev2->flags))
934                         d->state = (1<<MD_DISK_FAULTY);
935                 else if (test_bit(In_sync, &rdev2->flags)) {
936                         d->state = (1<<MD_DISK_ACTIVE);
937                         d->state |= (1<<MD_DISK_SYNC);
938                         active++;
939                         working++;
940                 } else {
941                         d->state = 0;
942                         spare++;
943                         working++;
944                 }
945                 if (test_bit(WriteMostly, &rdev2->flags))
946                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
947         }
948         /* now set the "removed" and "faulty" bits on any missing devices */
949         for (i=0 ; i < mddev->raid_disks ; i++) {
950                 mdp_disk_t *d = &sb->disks[i];
951                 if (d->state == 0 && d->number == 0) {
952                         d->number = i;
953                         d->raid_disk = i;
954                         d->state = (1<<MD_DISK_REMOVED);
955                         d->state |= (1<<MD_DISK_FAULTY);
956                         failed++;
957                 }
958         }
959         sb->nr_disks = nr_disks;
960         sb->active_disks = active;
961         sb->working_disks = working;
962         sb->failed_disks = failed;
963         sb->spare_disks = spare;
964
965         sb->this_disk = sb->disks[rdev->desc_nr];
966         sb->sb_csum = calc_sb_csum(sb);
967 }
968
969 /*
970  * version 1 superblock
971  */
972
973 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
974 {
975         unsigned int disk_csum, csum;
976         unsigned long long newcsum;
977         int size = 256 + le32_to_cpu(sb->max_dev)*2;
978         unsigned int *isuper = (unsigned int*)sb;
979         int i;
980
981         disk_csum = sb->sb_csum;
982         sb->sb_csum = 0;
983         newcsum = 0;
984         for (i=0; size>=4; size -= 4 )
985                 newcsum += le32_to_cpu(*isuper++);
986
987         if (size == 2)
988                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
989
990         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
991         sb->sb_csum = disk_csum;
992         return cpu_to_le32(csum);
993 }
994
995 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
996 {
997         struct mdp_superblock_1 *sb;
998         int ret;
999         sector_t sb_offset;
1000         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1001         int bmask;
1002
1003         /*
1004          * Calculate the position of the superblock.
1005          * It is always aligned to a 4K boundary and
1006          * depeding on minor_version, it can be:
1007          * 0: At least 8K, but less than 12K, from end of device
1008          * 1: At start of device
1009          * 2: 4K from start of device.
1010          */
1011         switch(minor_version) {
1012         case 0:
1013                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1014                 sb_offset -= 8*2;
1015                 sb_offset &= ~(sector_t)(4*2-1);
1016                 /* convert from sectors to K */
1017                 sb_offset /= 2;
1018                 break;
1019         case 1:
1020                 sb_offset = 0;
1021                 break;
1022         case 2:
1023                 sb_offset = 4;
1024                 break;
1025         default:
1026                 return -EINVAL;
1027         }
1028         rdev->sb_offset = sb_offset;
1029
1030         /* superblock is rarely larger than 1K, but it can be larger,
1031          * and it is safe to read 4k, so we do that
1032          */
1033         ret = read_disk_sb(rdev, 4096);
1034         if (ret) return ret;
1035
1036
1037         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1038
1039         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1040             sb->major_version != cpu_to_le32(1) ||
1041             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1042             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1043             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1044                 return -EINVAL;
1045
1046         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1047                 printk("md: invalid superblock checksum on %s\n",
1048                         bdevname(rdev->bdev,b));
1049                 return -EINVAL;
1050         }
1051         if (le64_to_cpu(sb->data_size) < 10) {
1052                 printk("md: data_size too small on %s\n",
1053                        bdevname(rdev->bdev,b));
1054                 return -EINVAL;
1055         }
1056         rdev->preferred_minor = 0xffff;
1057         rdev->data_offset = le64_to_cpu(sb->data_offset);
1058         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1059
1060         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1061         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1062         if (rdev->sb_size & bmask)
1063                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1064
1065         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1066                 rdev->desc_nr = -1;
1067         else
1068                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1069
1070         if (refdev == 0)
1071                 ret = 1;
1072         else {
1073                 __u64 ev1, ev2;
1074                 struct mdp_superblock_1 *refsb = 
1075                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1076
1077                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1078                     sb->level != refsb->level ||
1079                     sb->layout != refsb->layout ||
1080                     sb->chunksize != refsb->chunksize) {
1081                         printk(KERN_WARNING "md: %s has strangely different"
1082                                 " superblock to %s\n",
1083                                 bdevname(rdev->bdev,b),
1084                                 bdevname(refdev->bdev,b2));
1085                         return -EINVAL;
1086                 }
1087                 ev1 = le64_to_cpu(sb->events);
1088                 ev2 = le64_to_cpu(refsb->events);
1089
1090                 if (ev1 > ev2)
1091                         ret = 1;
1092                 else
1093                         ret = 0;
1094         }
1095         if (minor_version) 
1096                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1097         else
1098                 rdev->size = rdev->sb_offset;
1099         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1100                 return -EINVAL;
1101         rdev->size = le64_to_cpu(sb->data_size)/2;
1102         if (le32_to_cpu(sb->chunksize))
1103                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1104
1105         if (le32_to_cpu(sb->size) > rdev->size*2)
1106                 return -EINVAL;
1107         return ret;
1108 }
1109
1110 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1111 {
1112         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1113         __u64 ev1 = le64_to_cpu(sb->events);
1114
1115         rdev->raid_disk = -1;
1116         rdev->flags = 0;
1117         if (mddev->raid_disks == 0) {
1118                 mddev->major_version = 1;
1119                 mddev->patch_version = 0;
1120                 mddev->persistent = 1;
1121                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1122                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1123                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1124                 mddev->level = le32_to_cpu(sb->level);
1125                 mddev->clevel[0] = 0;
1126                 mddev->layout = le32_to_cpu(sb->layout);
1127                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1128                 mddev->size = le64_to_cpu(sb->size)/2;
1129                 mddev->events = ev1;
1130                 mddev->bitmap_offset = 0;
1131                 mddev->default_bitmap_offset = 1024 >> 9;
1132                 
1133                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1134                 memcpy(mddev->uuid, sb->set_uuid, 16);
1135
1136                 mddev->max_disks =  (4096-256)/2;
1137
1138                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1139                     mddev->bitmap_file == NULL ) {
1140                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1141                             && mddev->level != 10) {
1142                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1143                                 return -EINVAL;
1144                         }
1145                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1146                 }
1147                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1148                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1149                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1150                         mddev->new_level = le32_to_cpu(sb->new_level);
1151                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1152                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1153                 } else {
1154                         mddev->reshape_position = MaxSector;
1155                         mddev->delta_disks = 0;
1156                         mddev->new_level = mddev->level;
1157                         mddev->new_layout = mddev->layout;
1158                         mddev->new_chunk = mddev->chunk_size;
1159                 }
1160
1161         } else if (mddev->pers == NULL) {
1162                 /* Insist of good event counter while assembling */
1163                 ++ev1;
1164                 if (ev1 < mddev->events)
1165                         return -EINVAL;
1166         } else if (mddev->bitmap) {
1167                 /* If adding to array with a bitmap, then we can accept an
1168                  * older device, but not too old.
1169                  */
1170                 if (ev1 < mddev->bitmap->events_cleared)
1171                         return 0;
1172         } else {
1173                 if (ev1 < mddev->events)
1174                         /* just a hot-add of a new device, leave raid_disk at -1 */
1175                         return 0;
1176         }
1177         if (mddev->level != LEVEL_MULTIPATH) {
1178                 int role;
1179                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1180                 switch(role) {
1181                 case 0xffff: /* spare */
1182                         break;
1183                 case 0xfffe: /* faulty */
1184                         set_bit(Faulty, &rdev->flags);
1185                         break;
1186                 default:
1187                         if ((le32_to_cpu(sb->feature_map) &
1188                              MD_FEATURE_RECOVERY_OFFSET))
1189                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1190                         else
1191                                 set_bit(In_sync, &rdev->flags);
1192                         rdev->raid_disk = role;
1193                         break;
1194                 }
1195                 if (sb->devflags & WriteMostly1)
1196                         set_bit(WriteMostly, &rdev->flags);
1197         } else /* MULTIPATH are always insync */
1198                 set_bit(In_sync, &rdev->flags);
1199
1200         return 0;
1201 }
1202
1203 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1204 {
1205         struct mdp_superblock_1 *sb;
1206         struct list_head *tmp;
1207         mdk_rdev_t *rdev2;
1208         int max_dev, i;
1209         /* make rdev->sb match mddev and rdev data. */
1210
1211         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1212
1213         sb->feature_map = 0;
1214         sb->pad0 = 0;
1215         sb->recovery_offset = cpu_to_le64(0);
1216         memset(sb->pad1, 0, sizeof(sb->pad1));
1217         memset(sb->pad2, 0, sizeof(sb->pad2));
1218         memset(sb->pad3, 0, sizeof(sb->pad3));
1219
1220         sb->utime = cpu_to_le64((__u64)mddev->utime);
1221         sb->events = cpu_to_le64(mddev->events);
1222         if (mddev->in_sync)
1223                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1224         else
1225                 sb->resync_offset = cpu_to_le64(0);
1226
1227         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1228
1229         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1230         sb->size = cpu_to_le64(mddev->size<<1);
1231
1232         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1233                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1234                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1235         }
1236
1237         if (rdev->raid_disk >= 0 &&
1238             !test_bit(In_sync, &rdev->flags) &&
1239             rdev->recovery_offset > 0) {
1240                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1241                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1242         }
1243
1244         if (mddev->reshape_position != MaxSector) {
1245                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1246                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1247                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1248                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1249                 sb->new_level = cpu_to_le32(mddev->new_level);
1250                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1251         }
1252
1253         max_dev = 0;
1254         ITERATE_RDEV(mddev,rdev2,tmp)
1255                 if (rdev2->desc_nr+1 > max_dev)
1256                         max_dev = rdev2->desc_nr+1;
1257         
1258         sb->max_dev = cpu_to_le32(max_dev);
1259         for (i=0; i<max_dev;i++)
1260                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1261         
1262         ITERATE_RDEV(mddev,rdev2,tmp) {
1263                 i = rdev2->desc_nr;
1264                 if (test_bit(Faulty, &rdev2->flags))
1265                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1266                 else if (test_bit(In_sync, &rdev2->flags))
1267                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1268                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1269                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1270                 else
1271                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1272         }
1273
1274         sb->sb_csum = calc_sb_1_csum(sb);
1275 }
1276
1277
1278 static struct super_type super_types[] = {
1279         [0] = {
1280                 .name   = "0.90.0",
1281                 .owner  = THIS_MODULE,
1282                 .load_super     = super_90_load,
1283                 .validate_super = super_90_validate,
1284                 .sync_super     = super_90_sync,
1285         },
1286         [1] = {
1287                 .name   = "md-1",
1288                 .owner  = THIS_MODULE,
1289                 .load_super     = super_1_load,
1290                 .validate_super = super_1_validate,
1291                 .sync_super     = super_1_sync,
1292         },
1293 };
1294         
1295 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1296 {
1297         struct list_head *tmp;
1298         mdk_rdev_t *rdev;
1299
1300         ITERATE_RDEV(mddev,rdev,tmp)
1301                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1302                         return rdev;
1303
1304         return NULL;
1305 }
1306
1307 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1308 {
1309         struct list_head *tmp;
1310         mdk_rdev_t *rdev;
1311
1312         ITERATE_RDEV(mddev1,rdev,tmp)
1313                 if (match_dev_unit(mddev2, rdev))
1314                         return 1;
1315
1316         return 0;
1317 }
1318
1319 static LIST_HEAD(pending_raid_disks);
1320
1321 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1322 {
1323         mdk_rdev_t *same_pdev;
1324         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1325         struct kobject *ko;
1326         char *s;
1327
1328         if (rdev->mddev) {
1329                 MD_BUG();
1330                 return -EINVAL;
1331         }
1332         /* make sure rdev->size exceeds mddev->size */
1333         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1334                 if (mddev->pers)
1335                         /* Cannot change size, so fail */
1336                         return -ENOSPC;
1337                 else
1338                         mddev->size = rdev->size;
1339         }
1340         same_pdev = match_dev_unit(mddev, rdev);
1341         if (same_pdev)
1342                 printk(KERN_WARNING
1343                         "%s: WARNING: %s appears to be on the same physical"
1344                         " disk as %s. True\n     protection against single-disk"
1345                         " failure might be compromised.\n",
1346                         mdname(mddev), bdevname(rdev->bdev,b),
1347                         bdevname(same_pdev->bdev,b2));
1348
1349         /* Verify rdev->desc_nr is unique.
1350          * If it is -1, assign a free number, else
1351          * check number is not in use
1352          */
1353         if (rdev->desc_nr < 0) {
1354                 int choice = 0;
1355                 if (mddev->pers) choice = mddev->raid_disks;
1356                 while (find_rdev_nr(mddev, choice))
1357                         choice++;
1358                 rdev->desc_nr = choice;
1359         } else {
1360                 if (find_rdev_nr(mddev, rdev->desc_nr))
1361                         return -EBUSY;
1362         }
1363         bdevname(rdev->bdev,b);
1364         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1365                 return -ENOMEM;
1366         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1367                 *s = '!';
1368                         
1369         list_add(&rdev->same_set, &mddev->disks);
1370         rdev->mddev = mddev;
1371         printk(KERN_INFO "md: bind<%s>\n", b);
1372
1373         rdev->kobj.parent = &mddev->kobj;
1374         kobject_add(&rdev->kobj);
1375
1376         if (rdev->bdev->bd_part)
1377                 ko = &rdev->bdev->bd_part->kobj;
1378         else
1379                 ko = &rdev->bdev->bd_disk->kobj;
1380         sysfs_create_link(&rdev->kobj, ko, "block");
1381         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1382         return 0;
1383 }
1384
1385 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1386 {
1387         char b[BDEVNAME_SIZE];
1388         if (!rdev->mddev) {
1389                 MD_BUG();
1390                 return;
1391         }
1392         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1393         list_del_init(&rdev->same_set);
1394         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1395         rdev->mddev = NULL;
1396         sysfs_remove_link(&rdev->kobj, "block");
1397         kobject_del(&rdev->kobj);
1398 }
1399
1400 /*
1401  * prevent the device from being mounted, repartitioned or
1402  * otherwise reused by a RAID array (or any other kernel
1403  * subsystem), by bd_claiming the device.
1404  */
1405 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1406 {
1407         int err = 0;
1408         struct block_device *bdev;
1409         char b[BDEVNAME_SIZE];
1410
1411         bdev = open_partition_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1412         if (IS_ERR(bdev)) {
1413                 printk(KERN_ERR "md: could not open %s.\n",
1414                         __bdevname(dev, b));
1415                 return PTR_ERR(bdev);
1416         }
1417         err = bd_claim(bdev, rdev);
1418         if (err) {
1419                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1420                         bdevname(bdev, b));
1421                 blkdev_put_partition(bdev);
1422                 return err;
1423         }
1424         rdev->bdev = bdev;
1425         return err;
1426 }
1427
1428 static void unlock_rdev(mdk_rdev_t *rdev)
1429 {
1430         struct block_device *bdev = rdev->bdev;
1431         rdev->bdev = NULL;
1432         if (!bdev)
1433                 MD_BUG();
1434         bd_release(bdev);
1435         blkdev_put_partition(bdev);
1436 }
1437
1438 void md_autodetect_dev(dev_t dev);
1439
1440 static void export_rdev(mdk_rdev_t * rdev)
1441 {
1442         char b[BDEVNAME_SIZE];
1443         printk(KERN_INFO "md: export_rdev(%s)\n",
1444                 bdevname(rdev->bdev,b));
1445         if (rdev->mddev)
1446                 MD_BUG();
1447         free_disk_sb(rdev);
1448         list_del_init(&rdev->same_set);
1449 #ifndef MODULE
1450         md_autodetect_dev(rdev->bdev->bd_dev);
1451 #endif
1452         unlock_rdev(rdev);
1453         kobject_put(&rdev->kobj);
1454 }
1455
1456 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1457 {
1458         unbind_rdev_from_array(rdev);
1459         export_rdev(rdev);
1460 }
1461
1462 static void export_array(mddev_t *mddev)
1463 {
1464         struct list_head *tmp;
1465         mdk_rdev_t *rdev;
1466
1467         ITERATE_RDEV(mddev,rdev,tmp) {
1468                 if (!rdev->mddev) {
1469                         MD_BUG();
1470                         continue;
1471                 }
1472                 kick_rdev_from_array(rdev);
1473         }
1474         if (!list_empty(&mddev->disks))
1475                 MD_BUG();
1476         mddev->raid_disks = 0;
1477         mddev->major_version = 0;
1478 }
1479
1480 static void print_desc(mdp_disk_t *desc)
1481 {
1482         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1483                 desc->major,desc->minor,desc->raid_disk,desc->state);
1484 }
1485
1486 static void print_sb(mdp_super_t *sb)
1487 {
1488         int i;
1489
1490         printk(KERN_INFO 
1491                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1492                 sb->major_version, sb->minor_version, sb->patch_version,
1493                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1494                 sb->ctime);
1495         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1496                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1497                 sb->md_minor, sb->layout, sb->chunk_size);
1498         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1499                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1500                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1501                 sb->failed_disks, sb->spare_disks,
1502                 sb->sb_csum, (unsigned long)sb->events_lo);
1503
1504         printk(KERN_INFO);
1505         for (i = 0; i < MD_SB_DISKS; i++) {
1506                 mdp_disk_t *desc;
1507
1508                 desc = sb->disks + i;
1509                 if (desc->number || desc->major || desc->minor ||
1510                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1511                         printk("     D %2d: ", i);
1512                         print_desc(desc);
1513                 }
1514         }
1515         printk(KERN_INFO "md:     THIS: ");
1516         print_desc(&sb->this_disk);
1517
1518 }
1519
1520 static void print_rdev(mdk_rdev_t *rdev)
1521 {
1522         char b[BDEVNAME_SIZE];
1523         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1524                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1525                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1526                 rdev->desc_nr);
1527         if (rdev->sb_loaded) {
1528                 printk(KERN_INFO "md: rdev superblock:\n");
1529                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1530         } else
1531                 printk(KERN_INFO "md: no rdev superblock!\n");
1532 }
1533
1534 static void md_print_devices(void)
1535 {
1536         struct list_head *tmp, *tmp2;
1537         mdk_rdev_t *rdev;
1538         mddev_t *mddev;
1539         char b[BDEVNAME_SIZE];
1540
1541         printk("\n");
1542         printk("md:     **********************************\n");
1543         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1544         printk("md:     **********************************\n");
1545         ITERATE_MDDEV(mddev,tmp) {
1546
1547                 if (mddev->bitmap)
1548                         bitmap_print_sb(mddev->bitmap);
1549                 else
1550                         printk("%s: ", mdname(mddev));
1551                 ITERATE_RDEV(mddev,rdev,tmp2)
1552                         printk("<%s>", bdevname(rdev->bdev,b));
1553                 printk("\n");
1554
1555                 ITERATE_RDEV(mddev,rdev,tmp2)
1556                         print_rdev(rdev);
1557         }
1558         printk("md:     **********************************\n");
1559         printk("\n");
1560 }
1561
1562
1563 static void sync_sbs(mddev_t * mddev, int nospares)
1564 {
1565         /* Update each superblock (in-memory image), but
1566          * if we are allowed to, skip spares which already
1567          * have the right event counter, or have one earlier
1568          * (which would mean they aren't being marked as dirty
1569          * with the rest of the array)
1570          */
1571         mdk_rdev_t *rdev;
1572         struct list_head *tmp;
1573
1574         ITERATE_RDEV(mddev,rdev,tmp) {
1575                 if (rdev->sb_events == mddev->events ||
1576                     (nospares &&
1577                      rdev->raid_disk < 0 &&
1578                      (rdev->sb_events&1)==0 &&
1579                      rdev->sb_events+1 == mddev->events)) {
1580                         /* Don't update this superblock */
1581                         rdev->sb_loaded = 2;
1582                 } else {
1583                         super_types[mddev->major_version].
1584                                 sync_super(mddev, rdev);
1585                         rdev->sb_loaded = 1;
1586                 }
1587         }
1588 }
1589
1590 void md_update_sb(mddev_t * mddev)
1591 {
1592         int err;
1593         struct list_head *tmp;
1594         mdk_rdev_t *rdev;
1595         int sync_req;
1596         int nospares = 0;
1597
1598 repeat:
1599         spin_lock_irq(&mddev->write_lock);
1600         sync_req = mddev->in_sync;
1601         mddev->utime = get_seconds();
1602         if (mddev->sb_dirty == 3)
1603                 /* just a clean<-> dirty transition, possibly leave spares alone,
1604                  * though if events isn't the right even/odd, we will have to do
1605                  * spares after all
1606                  */
1607                 nospares = 1;
1608
1609         /* If this is just a dirty<->clean transition, and the array is clean
1610          * and 'events' is odd, we can roll back to the previous clean state */
1611         if (mddev->sb_dirty == 3
1612             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1613             && (mddev->events & 1))
1614                 mddev->events--;
1615         else {
1616                 /* otherwise we have to go forward and ... */
1617                 mddev->events ++;
1618                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1619                         /* .. if the array isn't clean, insist on an odd 'events' */
1620                         if ((mddev->events&1)==0) {
1621                                 mddev->events++;
1622                                 nospares = 0;
1623                         }
1624                 } else {
1625                         /* otherwise insist on an even 'events' (for clean states) */
1626                         if ((mddev->events&1)) {
1627                                 mddev->events++;
1628                                 nospares = 0;
1629                         }
1630                 }
1631         }
1632
1633         if (!mddev->events) {
1634                 /*
1635                  * oops, this 64-bit counter should never wrap.
1636                  * Either we are in around ~1 trillion A.C., assuming
1637                  * 1 reboot per second, or we have a bug:
1638                  */
1639                 MD_BUG();
1640                 mddev->events --;
1641         }
1642         mddev->sb_dirty = 2;
1643         sync_sbs(mddev, nospares);
1644
1645         /*
1646          * do not write anything to disk if using
1647          * nonpersistent superblocks
1648          */
1649         if (!mddev->persistent) {
1650                 mddev->sb_dirty = 0;
1651                 spin_unlock_irq(&mddev->write_lock);
1652                 wake_up(&mddev->sb_wait);
1653                 return;
1654         }
1655         spin_unlock_irq(&mddev->write_lock);
1656
1657         dprintk(KERN_INFO 
1658                 "md: updating %s RAID superblock on device (in sync %d)\n",
1659                 mdname(mddev),mddev->in_sync);
1660
1661         err = bitmap_update_sb(mddev->bitmap);
1662         ITERATE_RDEV(mddev,rdev,tmp) {
1663                 char b[BDEVNAME_SIZE];
1664                 dprintk(KERN_INFO "md: ");
1665                 if (rdev->sb_loaded != 1)
1666                         continue; /* no noise on spare devices */
1667                 if (test_bit(Faulty, &rdev->flags))
1668                         dprintk("(skipping faulty ");
1669
1670                 dprintk("%s ", bdevname(rdev->bdev,b));
1671                 if (!test_bit(Faulty, &rdev->flags)) {
1672                         md_super_write(mddev,rdev,
1673                                        rdev->sb_offset<<1, rdev->sb_size,
1674                                        rdev->sb_page);
1675                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1676                                 bdevname(rdev->bdev,b),
1677                                 (unsigned long long)rdev->sb_offset);
1678                         rdev->sb_events = mddev->events;
1679
1680                 } else
1681                         dprintk(")\n");
1682                 if (mddev->level == LEVEL_MULTIPATH)
1683                         /* only need to write one superblock... */
1684                         break;
1685         }
1686         md_super_wait(mddev);
1687         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1688
1689         spin_lock_irq(&mddev->write_lock);
1690         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1691                 /* have to write it out again */
1692                 spin_unlock_irq(&mddev->write_lock);
1693                 goto repeat;
1694         }
1695         mddev->sb_dirty = 0;
1696         spin_unlock_irq(&mddev->write_lock);
1697         wake_up(&mddev->sb_wait);
1698
1699 }
1700 EXPORT_SYMBOL_GPL(md_update_sb);
1701
1702 /* words written to sysfs files may, or my not, be \n terminated.
1703  * We want to accept with case. For this we use cmd_match.
1704  */
1705 static int cmd_match(const char *cmd, const char *str)
1706 {
1707         /* See if cmd, written into a sysfs file, matches
1708          * str.  They must either be the same, or cmd can
1709          * have a trailing newline
1710          */
1711         while (*cmd && *str && *cmd == *str) {
1712                 cmd++;
1713                 str++;
1714         }
1715         if (*cmd == '\n')
1716                 cmd++;
1717         if (*str || *cmd)
1718                 return 0;
1719         return 1;
1720 }
1721
1722 struct rdev_sysfs_entry {
1723         struct attribute attr;
1724         ssize_t (*show)(mdk_rdev_t *, char *);
1725         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1726 };
1727
1728 static ssize_t
1729 state_show(mdk_rdev_t *rdev, char *page)
1730 {
1731         char *sep = "";
1732         int len=0;
1733
1734         if (test_bit(Faulty, &rdev->flags)) {
1735                 len+= sprintf(page+len, "%sfaulty",sep);
1736                 sep = ",";
1737         }
1738         if (test_bit(In_sync, &rdev->flags)) {
1739                 len += sprintf(page+len, "%sin_sync",sep);
1740                 sep = ",";
1741         }
1742         if (test_bit(WriteMostly, &rdev->flags)) {
1743                 len += sprintf(page+len, "%swrite_mostly",sep);
1744                 sep = ",";
1745         }
1746         if (!test_bit(Faulty, &rdev->flags) &&
1747             !test_bit(In_sync, &rdev->flags)) {
1748                 len += sprintf(page+len, "%sspare", sep);
1749                 sep = ",";
1750         }
1751         return len+sprintf(page+len, "\n");
1752 }
1753
1754 static ssize_t
1755 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1756 {
1757         /* can write
1758          *  faulty  - simulates and error
1759          *  remove  - disconnects the device
1760          *  writemostly - sets write_mostly
1761          *  -writemostly - clears write_mostly
1762          */
1763         int err = -EINVAL;
1764         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1765                 md_error(rdev->mddev, rdev);
1766                 err = 0;
1767         } else if (cmd_match(buf, "remove")) {
1768                 if (rdev->raid_disk >= 0)
1769                         err = -EBUSY;
1770                 else {
1771                         mddev_t *mddev = rdev->mddev;
1772                         kick_rdev_from_array(rdev);
1773                         md_update_sb(mddev);
1774                         md_new_event(mddev);
1775                         err = 0;
1776                 }
1777         } else if (cmd_match(buf, "writemostly")) {
1778                 set_bit(WriteMostly, &rdev->flags);
1779                 err = 0;
1780         } else if (cmd_match(buf, "-writemostly")) {
1781                 clear_bit(WriteMostly, &rdev->flags);
1782                 err = 0;
1783         }
1784         return err ? err : len;
1785 }
1786 static struct rdev_sysfs_entry rdev_state =
1787 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1788
1789 static ssize_t
1790 super_show(mdk_rdev_t *rdev, char *page)
1791 {
1792         if (rdev->sb_loaded && rdev->sb_size) {
1793                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1794                 return rdev->sb_size;
1795         } else
1796                 return 0;
1797 }
1798 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1799
1800 static ssize_t
1801 errors_show(mdk_rdev_t *rdev, char *page)
1802 {
1803         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1804 }
1805
1806 static ssize_t
1807 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1808 {
1809         char *e;
1810         unsigned long n = simple_strtoul(buf, &e, 10);
1811         if (*buf && (*e == 0 || *e == '\n')) {
1812                 atomic_set(&rdev->corrected_errors, n);
1813                 return len;
1814         }
1815         return -EINVAL;
1816 }
1817 static struct rdev_sysfs_entry rdev_errors =
1818 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1819
1820 static ssize_t
1821 slot_show(mdk_rdev_t *rdev, char *page)
1822 {
1823         if (rdev->raid_disk < 0)
1824                 return sprintf(page, "none\n");
1825         else
1826                 return sprintf(page, "%d\n", rdev->raid_disk);
1827 }
1828
1829 static ssize_t
1830 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1831 {
1832         char *e;
1833         int slot = simple_strtoul(buf, &e, 10);
1834         if (strncmp(buf, "none", 4)==0)
1835                 slot = -1;
1836         else if (e==buf || (*e && *e!= '\n'))
1837                 return -EINVAL;
1838         if (rdev->mddev->pers)
1839                 /* Cannot set slot in active array (yet) */
1840                 return -EBUSY;
1841         if (slot >= rdev->mddev->raid_disks)
1842                 return -ENOSPC;
1843         rdev->raid_disk = slot;
1844         /* assume it is working */
1845         rdev->flags = 0;
1846         set_bit(In_sync, &rdev->flags);
1847         return len;
1848 }
1849
1850
1851 static struct rdev_sysfs_entry rdev_slot =
1852 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1853
1854 static ssize_t
1855 offset_show(mdk_rdev_t *rdev, char *page)
1856 {
1857         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1858 }
1859
1860 static ssize_t
1861 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1862 {
1863         char *e;
1864         unsigned long long offset = simple_strtoull(buf, &e, 10);
1865         if (e==buf || (*e && *e != '\n'))
1866                 return -EINVAL;
1867         if (rdev->mddev->pers)
1868                 return -EBUSY;
1869         rdev->data_offset = offset;
1870         return len;
1871 }
1872
1873 static struct rdev_sysfs_entry rdev_offset =
1874 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1875
1876 static ssize_t
1877 rdev_size_show(mdk_rdev_t *rdev, char *page)
1878 {
1879         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1880 }
1881
1882 static ssize_t
1883 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1884 {
1885         char *e;
1886         unsigned long long size = simple_strtoull(buf, &e, 10);
1887         if (e==buf || (*e && *e != '\n'))
1888                 return -EINVAL;
1889         if (rdev->mddev->pers)
1890                 return -EBUSY;
1891         rdev->size = size;
1892         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1893                 rdev->mddev->size = size;
1894         return len;
1895 }
1896
1897 static struct rdev_sysfs_entry rdev_size =
1898 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1899
1900 static struct attribute *rdev_default_attrs[] = {
1901         &rdev_state.attr,
1902         &rdev_super.attr,
1903         &rdev_errors.attr,
1904         &rdev_slot.attr,
1905         &rdev_offset.attr,
1906         &rdev_size.attr,
1907         NULL,
1908 };
1909 static ssize_t
1910 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1911 {
1912         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1913         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1914
1915         if (!entry->show)
1916                 return -EIO;
1917         return entry->show(rdev, page);
1918 }
1919
1920 static ssize_t
1921 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1922               const char *page, size_t length)
1923 {
1924         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1925         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1926
1927         if (!entry->store)
1928                 return -EIO;
1929         if (!capable(CAP_SYS_ADMIN))
1930                 return -EACCES;
1931         return entry->store(rdev, page, length);
1932 }
1933
1934 static void rdev_free(struct kobject *ko)
1935 {
1936         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1937         kfree(rdev);
1938 }
1939 static struct sysfs_ops rdev_sysfs_ops = {
1940         .show           = rdev_attr_show,
1941         .store          = rdev_attr_store,
1942 };
1943 static struct kobj_type rdev_ktype = {
1944         .release        = rdev_free,
1945         .sysfs_ops      = &rdev_sysfs_ops,
1946         .default_attrs  = rdev_default_attrs,
1947 };
1948
1949 /*
1950  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1951  *
1952  * mark the device faulty if:
1953  *
1954  *   - the device is nonexistent (zero size)
1955  *   - the device has no valid superblock
1956  *
1957  * a faulty rdev _never_ has rdev->sb set.
1958  */
1959 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1960 {
1961         char b[BDEVNAME_SIZE];
1962         int err;
1963         mdk_rdev_t *rdev;
1964         sector_t size;
1965
1966         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1967         if (!rdev) {
1968                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1969                 return ERR_PTR(-ENOMEM);
1970         }
1971
1972         if ((err = alloc_disk_sb(rdev)))
1973                 goto abort_free;
1974
1975         err = lock_rdev(rdev, newdev);
1976         if (err)
1977                 goto abort_free;
1978
1979         rdev->kobj.parent = NULL;
1980         rdev->kobj.ktype = &rdev_ktype;
1981         kobject_init(&rdev->kobj);
1982
1983         rdev->desc_nr = -1;
1984         rdev->flags = 0;
1985         rdev->data_offset = 0;
1986         rdev->sb_events = 0;
1987         atomic_set(&rdev->nr_pending, 0);
1988         atomic_set(&rdev->read_errors, 0);
1989         atomic_set(&rdev->corrected_errors, 0);
1990
1991         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1992         if (!size) {
1993                 printk(KERN_WARNING 
1994                         "md: %s has zero or unknown size, marking faulty!\n",
1995                         bdevname(rdev->bdev,b));
1996                 err = -EINVAL;
1997                 goto abort_free;
1998         }
1999
2000         if (super_format >= 0) {
2001                 err = super_types[super_format].
2002                         load_super(rdev, NULL, super_minor);
2003                 if (err == -EINVAL) {
2004                         printk(KERN_WARNING 
2005                                 "md: %s has invalid sb, not importing!\n",
2006                                 bdevname(rdev->bdev,b));
2007                         goto abort_free;
2008                 }
2009                 if (err < 0) {
2010                         printk(KERN_WARNING 
2011                                 "md: could not read %s's sb, not importing!\n",
2012                                 bdevname(rdev->bdev,b));
2013                         goto abort_free;
2014                 }
2015         }
2016         INIT_LIST_HEAD(&rdev->same_set);
2017
2018         return rdev;
2019
2020 abort_free:
2021         if (rdev->sb_page) {
2022                 if (rdev->bdev)
2023                         unlock_rdev(rdev);
2024                 free_disk_sb(rdev);
2025         }
2026         kfree(rdev);
2027         return ERR_PTR(err);
2028 }
2029
2030 /*
2031  * Check a full RAID array for plausibility
2032  */
2033
2034
2035 static void analyze_sbs(mddev_t * mddev)
2036 {
2037         int i;
2038         struct list_head *tmp;
2039         mdk_rdev_t *rdev, *freshest;
2040         char b[BDEVNAME_SIZE];
2041
2042         freshest = NULL;
2043         ITERATE_RDEV(mddev,rdev,tmp)
2044                 switch (super_types[mddev->major_version].
2045                         load_super(rdev, freshest, mddev->minor_version)) {
2046                 case 1:
2047                         freshest = rdev;
2048                         break;
2049                 case 0:
2050                         break;
2051                 default:
2052                         printk( KERN_ERR \
2053                                 "md: fatal superblock inconsistency in %s"
2054                                 " -- removing from array\n", 
2055                                 bdevname(rdev->bdev,b));
2056                         kick_rdev_from_array(rdev);
2057                 }
2058
2059
2060         super_types[mddev->major_version].
2061                 validate_super(mddev, freshest);
2062
2063         i = 0;
2064         ITERATE_RDEV(mddev,rdev,tmp) {
2065                 if (rdev != freshest)
2066                         if (super_types[mddev->major_version].
2067                             validate_super(mddev, rdev)) {
2068                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2069                                         " from array!\n",
2070                                         bdevname(rdev->bdev,b));
2071                                 kick_rdev_from_array(rdev);
2072                                 continue;
2073                         }
2074                 if (mddev->level == LEVEL_MULTIPATH) {
2075                         rdev->desc_nr = i++;
2076                         rdev->raid_disk = rdev->desc_nr;
2077                         set_bit(In_sync, &rdev->flags);
2078                 }
2079         }
2080
2081
2082
2083         if (mddev->recovery_cp != MaxSector &&
2084             mddev->level >= 1)
2085                 printk(KERN_ERR "md: %s: raid array is not clean"
2086                        " -- starting background reconstruction\n",
2087                        mdname(mddev));
2088
2089 }
2090
2091 static ssize_t
2092 safe_delay_show(mddev_t *mddev, char *page)
2093 {
2094         int msec = (mddev->safemode_delay*1000)/HZ;
2095         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2096 }
2097 static ssize_t
2098 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2099 {
2100         int scale=1;
2101         int dot=0;
2102         int i;
2103         unsigned long msec;
2104         char buf[30];
2105         char *e;
2106         /* remove a period, and count digits after it */
2107         if (len >= sizeof(buf))
2108                 return -EINVAL;
2109         strlcpy(buf, cbuf, len);
2110         buf[len] = 0;
2111         for (i=0; i<len; i++) {
2112                 if (dot) {
2113                         if (isdigit(buf[i])) {
2114                                 buf[i-1] = buf[i];
2115                                 scale *= 10;
2116                         }
2117                         buf[i] = 0;
2118                 } else if (buf[i] == '.') {
2119                         dot=1;
2120                         buf[i] = 0;
2121                 }
2122         }
2123         msec = simple_strtoul(buf, &e, 10);
2124         if (e == buf || (*e && *e != '\n'))
2125                 return -EINVAL;
2126         msec = (msec * 1000) / scale;
2127         if (msec == 0)
2128                 mddev->safemode_delay = 0;
2129         else {
2130                 mddev->safemode_delay = (msec*HZ)/1000;
2131                 if (mddev->safemode_delay == 0)
2132                         mddev->safemode_delay = 1;
2133         }
2134         return len;
2135 }
2136 static struct md_sysfs_entry md_safe_delay =
2137 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2138
2139 static ssize_t
2140 level_show(mddev_t *mddev, char *page)
2141 {
2142         struct mdk_personality *p = mddev->pers;
2143         if (p)
2144                 return sprintf(page, "%s\n", p->name);
2145         else if (mddev->clevel[0])
2146                 return sprintf(page, "%s\n", mddev->clevel);
2147         else if (mddev->level != LEVEL_NONE)
2148                 return sprintf(page, "%d\n", mddev->level);
2149         else
2150                 return 0;
2151 }
2152
2153 static ssize_t
2154 level_store(mddev_t *mddev, const char *buf, size_t len)
2155 {
2156         int rv = len;
2157         if (mddev->pers)
2158                 return -EBUSY;
2159         if (len == 0)
2160                 return 0;
2161         if (len >= sizeof(mddev->clevel))
2162                 return -ENOSPC;
2163         strncpy(mddev->clevel, buf, len);
2164         if (mddev->clevel[len-1] == '\n')
2165                 len--;
2166         mddev->clevel[len] = 0;
2167         mddev->level = LEVEL_NONE;
2168         return rv;
2169 }
2170
2171 static struct md_sysfs_entry md_level =
2172 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2173
2174
2175 static ssize_t
2176 layout_show(mddev_t *mddev, char *page)
2177 {
2178         /* just a number, not meaningful for all levels */
2179         return sprintf(page, "%d\n", mddev->layout);
2180 }
2181
2182 static ssize_t
2183 layout_store(mddev_t *mddev, const char *buf, size_t len)
2184 {
2185         char *e;
2186         unsigned long n = simple_strtoul(buf, &e, 10);
2187         if (mddev->pers)
2188                 return -EBUSY;
2189
2190         if (!*buf || (*e && *e != '\n'))
2191                 return -EINVAL;
2192
2193         mddev->layout = n;
2194         return len;
2195 }
2196 static struct md_sysfs_entry md_layout =
2197 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2198
2199
2200 static ssize_t
2201 raid_disks_show(mddev_t *mddev, char *page)
2202 {
2203         if (mddev->raid_disks == 0)
2204                 return 0;
2205         return sprintf(page, "%d\n", mddev->raid_disks);
2206 }
2207
2208 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2209
2210 static ssize_t
2211 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2212 {
2213         /* can only set raid_disks if array is not yet active */
2214         char *e;
2215         int rv = 0;
2216         unsigned long n = simple_strtoul(buf, &e, 10);
2217
2218         if (!*buf || (*e && *e != '\n'))
2219                 return -EINVAL;
2220
2221         if (mddev->pers)
2222                 rv = update_raid_disks(mddev, n);
2223         else
2224                 mddev->raid_disks = n;
2225         return rv ? rv : len;
2226 }
2227 static struct md_sysfs_entry md_raid_disks =
2228 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2229
2230 static ssize_t
2231 chunk_size_show(mddev_t *mddev, char *page)
2232 {
2233         return sprintf(page, "%d\n", mddev->chunk_size);
2234 }
2235
2236 static ssize_t
2237 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2238 {
2239         /* can only set chunk_size if array is not yet active */
2240         char *e;
2241         unsigned long n = simple_strtoul(buf, &e, 10);
2242
2243         if (mddev->pers)
2244                 return -EBUSY;
2245         if (!*buf || (*e && *e != '\n'))
2246                 return -EINVAL;
2247
2248         mddev->chunk_size = n;
2249         return len;
2250 }
2251 static struct md_sysfs_entry md_chunk_size =
2252 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2253
2254 static ssize_t
2255 resync_start_show(mddev_t *mddev, char *page)
2256 {
2257         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2258 }
2259
2260 static ssize_t
2261 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2262 {
2263         /* can only set chunk_size if array is not yet active */
2264         char *e;
2265         unsigned long long n = simple_strtoull(buf, &e, 10);
2266
2267         if (mddev->pers)
2268                 return -EBUSY;
2269         if (!*buf || (*e && *e != '\n'))
2270                 return -EINVAL;
2271
2272         mddev->recovery_cp = n;
2273         return len;
2274 }
2275 static struct md_sysfs_entry md_resync_start =
2276 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2277
2278 /*
2279  * The array state can be:
2280  *
2281  * clear
2282  *     No devices, no size, no level
2283  *     Equivalent to STOP_ARRAY ioctl
2284  * inactive
2285  *     May have some settings, but array is not active
2286  *        all IO results in error
2287  *     When written, doesn't tear down array, but just stops it
2288  * suspended (not supported yet)
2289  *     All IO requests will block. The array can be reconfigured.
2290  *     Writing this, if accepted, will block until array is quiessent
2291  * readonly
2292  *     no resync can happen.  no superblocks get written.
2293  *     write requests fail
2294  * read-auto
2295  *     like readonly, but behaves like 'clean' on a write request.
2296  *
2297  * clean - no pending writes, but otherwise active.
2298  *     When written to inactive array, starts without resync
2299  *     If a write request arrives then
2300  *       if metadata is known, mark 'dirty' and switch to 'active'.
2301  *       if not known, block and switch to write-pending
2302  *     If written to an active array that has pending writes, then fails.
2303  * active
2304  *     fully active: IO and resync can be happening.
2305  *     When written to inactive array, starts with resync
2306  *
2307  * write-pending
2308  *     clean, but writes are blocked waiting for 'active' to be written.
2309  *
2310  * active-idle
2311  *     like active, but no writes have been seen for a while (100msec).
2312  *
2313  */
2314 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2315                    write_pending, active_idle, bad_word};
2316 static char *array_states[] = {
2317         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2318         "write-pending", "active-idle", NULL };
2319
2320 static int match_word(const char *word, char **list)
2321 {
2322         int n;
2323         for (n=0; list[n]; n++)
2324                 if (cmd_match(word, list[n]))
2325                         break;
2326         return n;
2327 }
2328
2329 static ssize_t
2330 array_state_show(mddev_t *mddev, char *page)
2331 {
2332         enum array_state st = inactive;
2333
2334         if (mddev->pers)
2335                 switch(mddev->ro) {
2336                 case 1:
2337                         st = readonly;
2338                         break;
2339                 case 2:
2340                         st = read_auto;
2341                         break;
2342                 case 0:
2343                         if (mddev->in_sync)
2344                                 st = clean;
2345                         else if (mddev->safemode)
2346                                 st = active_idle;
2347                         else
2348                                 st = active;
2349                 }
2350         else {
2351                 if (list_empty(&mddev->disks) &&
2352                     mddev->raid_disks == 0 &&
2353                     mddev->size == 0)
2354                         st = clear;
2355                 else
2356                         st = inactive;
2357         }
2358         return sprintf(page, "%s\n", array_states[st]);
2359 }
2360
2361 static int do_md_stop(mddev_t * mddev, int ro);
2362 static int do_md_run(mddev_t * mddev);
2363 static int restart_array(mddev_t *mddev);
2364
2365 static ssize_t
2366 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2367 {
2368         int err = -EINVAL;
2369         enum array_state st = match_word(buf, array_states);
2370         switch(st) {
2371         case bad_word:
2372                 break;
2373         case clear:
2374                 /* stopping an active array */
2375                 if (mddev->pers) {
2376                         if (atomic_read(&mddev->active) > 1)
2377                                 return -EBUSY;
2378                         err = do_md_stop(mddev, 0);
2379                 }
2380                 break;
2381         case inactive:
2382                 /* stopping an active array */
2383                 if (mddev->pers) {
2384                         if (atomic_read(&mddev->active) > 1)
2385                                 return -EBUSY;
2386                         err = do_md_stop(mddev, 2);
2387                 }
2388                 break;
2389         case suspended:
2390                 break; /* not supported yet */
2391         case readonly:
2392                 if (mddev->pers)
2393                         err = do_md_stop(mddev, 1);
2394                 else {
2395                         mddev->ro = 1;
2396                         err = do_md_run(mddev);
2397                 }
2398                 break;
2399         case read_auto:
2400                 /* stopping an active array */
2401                 if (mddev->pers) {
2402                         err = do_md_stop(mddev, 1);
2403                         if (err == 0)
2404                                 mddev->ro = 2; /* FIXME mark devices writable */
2405                 } else {
2406                         mddev->ro = 2;
2407                         err = do_md_run(mddev);
2408                 }
2409                 break;
2410         case clean:
2411                 if (mddev->pers) {
2412                         restart_array(mddev);
2413                         spin_lock_irq(&mddev->write_lock);
2414                         if (atomic_read(&mddev->writes_pending) == 0) {
2415                                 mddev->in_sync = 1;
2416                                 mddev->sb_dirty = 1;
2417                         }
2418                         spin_unlock_irq(&mddev->write_lock);
2419                 } else {
2420                         mddev->ro = 0;
2421                         mddev->recovery_cp = MaxSector;
2422                         err = do_md_run(mddev);
2423                 }
2424                 break;
2425         case active:
2426                 if (mddev->pers) {
2427                         restart_array(mddev);
2428                         mddev->sb_dirty = 0;
2429                         wake_up(&mddev->sb_wait);
2430                         err = 0;
2431                 } else {
2432                         mddev->ro = 0;
2433                         err = do_md_run(mddev);
2434                 }
2435                 break;
2436         case write_pending:
2437         case active_idle:
2438                 /* these cannot be set */
2439                 break;
2440         }
2441         if (err)
2442                 return err;
2443         else
2444                 return len;
2445 }
2446 static struct md_sysfs_entry md_array_state =
2447 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2448
2449 static ssize_t
2450 null_show(mddev_t *mddev, char *page)
2451 {
2452         return -EINVAL;
2453 }
2454
2455 static ssize_t
2456 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2457 {
2458         /* buf must be %d:%d\n? giving major and minor numbers */
2459         /* The new device is added to the array.
2460          * If the array has a persistent superblock, we read the
2461          * superblock to initialise info and check validity.
2462          * Otherwise, only checking done is that in bind_rdev_to_array,
2463          * which mainly checks size.
2464          */
2465         char *e;
2466         int major = simple_strtoul(buf, &e, 10);
2467         int minor;
2468         dev_t dev;
2469         mdk_rdev_t *rdev;
2470         int err;
2471
2472         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2473                 return -EINVAL;
2474         minor = simple_strtoul(e+1, &e, 10);
2475         if (*e && *e != '\n')
2476                 return -EINVAL;
2477         dev = MKDEV(major, minor);
2478         if (major != MAJOR(dev) ||
2479             minor != MINOR(dev))
2480                 return -EOVERFLOW;
2481
2482
2483         if (mddev->persistent) {
2484                 rdev = md_import_device(dev, mddev->major_version,
2485                                         mddev->minor_version);
2486                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2487                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2488                                                        mdk_rdev_t, same_set);
2489                         err = super_types[mddev->major_version]
2490                                 .load_super(rdev, rdev0, mddev->minor_version);
2491                         if (err < 0)
2492                                 goto out;
2493                 }
2494         } else
2495                 rdev = md_import_device(dev, -1, -1);
2496
2497         if (IS_ERR(rdev))
2498                 return PTR_ERR(rdev);
2499         err = bind_rdev_to_array(rdev, mddev);
2500  out:
2501         if (err)
2502                 export_rdev(rdev);
2503         return err ? err : len;
2504 }
2505
2506 static struct md_sysfs_entry md_new_device =
2507 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2508
2509 static ssize_t
2510 size_show(mddev_t *mddev, char *page)
2511 {
2512         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2513 }
2514
2515 static int update_size(mddev_t *mddev, unsigned long size);
2516
2517 static ssize_t
2518 size_store(mddev_t *mddev, const char *buf, size_t len)
2519 {
2520         /* If array is inactive, we can reduce the component size, but
2521          * not increase it (except from 0).
2522          * If array is active, we can try an on-line resize
2523          */
2524         char *e;
2525         int err = 0;
2526         unsigned long long size = simple_strtoull(buf, &e, 10);
2527         if (!*buf || *buf == '\n' ||
2528             (*e && *e != '\n'))
2529                 return -EINVAL;
2530
2531         if (mddev->pers) {
2532                 err = update_size(mddev, size);
2533                 md_update_sb(mddev);
2534         } else {
2535                 if (mddev->size == 0 ||
2536                     mddev->size > size)
2537                         mddev->size = size;
2538                 else
2539                         err = -ENOSPC;
2540         }
2541         return err ? err : len;
2542 }
2543
2544 static struct md_sysfs_entry md_size =
2545 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2546
2547
2548 /* Metdata version.
2549  * This is either 'none' for arrays with externally managed metadata,
2550  * or N.M for internally known formats
2551  */
2552 static ssize_t
2553 metadata_show(mddev_t *mddev, char *page)
2554 {
2555         if (mddev->persistent)
2556                 return sprintf(page, "%d.%d\n",
2557                                mddev->major_version, mddev->minor_version);
2558         else
2559                 return sprintf(page, "none\n");
2560 }
2561
2562 static ssize_t
2563 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2564 {
2565         int major, minor;
2566         char *e;
2567         if (!list_empty(&mddev->disks))
2568                 return -EBUSY;
2569
2570         if (cmd_match(buf, "none")) {
2571                 mddev->persistent = 0;
2572                 mddev->major_version = 0;
2573                 mddev->minor_version = 90;
2574                 return len;
2575         }
2576         major = simple_strtoul(buf, &e, 10);
2577         if (e==buf || *e != '.')
2578                 return -EINVAL;
2579         buf = e+1;
2580         minor = simple_strtoul(buf, &e, 10);
2581         if (e==buf || *e != '\n')
2582                 return -EINVAL;
2583         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2584             super_types[major].name == NULL)
2585                 return -ENOENT;
2586         mddev->major_version = major;
2587         mddev->minor_version = minor;
2588         mddev->persistent = 1;
2589         return len;
2590 }
2591
2592 static struct md_sysfs_entry md_metadata =
2593 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2594
2595 static ssize_t
2596 action_show(mddev_t *mddev, char *page)
2597 {
2598         char *type = "idle";
2599         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2600             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2601                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2602                         type = "reshape";
2603                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2604                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2605                                 type = "resync";
2606                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2607                                 type = "check";
2608                         else
2609                                 type = "repair";
2610                 } else
2611                         type = "recover";
2612         }
2613         return sprintf(page, "%s\n", type);
2614 }
2615
2616 static ssize_t
2617 action_store(mddev_t *mddev, const char *page, size_t len)
2618 {
2619         if (!mddev->pers || !mddev->pers->sync_request)
2620                 return -EINVAL;
2621
2622         if (cmd_match(page, "idle")) {
2623                 if (mddev->sync_thread) {
2624                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2625                         md_unregister_thread(mddev->sync_thread);
2626                         mddev->sync_thread = NULL;
2627                         mddev->recovery = 0;
2628                 }
2629         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2630                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2631                 return -EBUSY;
2632         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2633                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2634         else if (cmd_match(page, "reshape")) {
2635                 int err;
2636                 if (mddev->pers->start_reshape == NULL)
2637                         return -EINVAL;
2638                 err = mddev->pers->start_reshape(mddev);
2639                 if (err)
2640                         return err;
2641         } else {
2642                 if (cmd_match(page, "check"))
2643                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2644                 else if (!cmd_match(page, "repair"))
2645                         return -EINVAL;
2646                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2647                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2648         }
2649         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2650         md_wakeup_thread(mddev->thread);
2651         return len;
2652 }
2653
2654 static ssize_t
2655 mismatch_cnt_show(mddev_t *mddev, char *page)
2656 {
2657         return sprintf(page, "%llu\n",
2658                        (unsigned long long) mddev->resync_mismatches);
2659 }
2660
2661 static struct md_sysfs_entry md_scan_mode =
2662 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2663
2664
2665 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2666
2667 static ssize_t
2668 sync_min_show(mddev_t *mddev, char *page)
2669 {
2670         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2671                        mddev->sync_speed_min ? "local": "system");
2672 }
2673
2674 static ssize_t
2675 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2676 {
2677         int min;
2678         char *e;
2679         if (strncmp(buf, "system", 6)==0) {
2680                 mddev->sync_speed_min = 0;
2681                 return len;
2682         }
2683         min = simple_strtoul(buf, &e, 10);
2684         if (buf == e || (*e && *e != '\n') || min <= 0)
2685                 return -EINVAL;
2686         mddev->sync_speed_min = min;
2687         return len;
2688 }
2689
2690 static struct md_sysfs_entry md_sync_min =
2691 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2692
2693 static ssize_t
2694 sync_max_show(mddev_t *mddev, char *page)
2695 {
2696         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2697                        mddev->sync_speed_max ? "local": "system");
2698 }
2699
2700 static ssize_t
2701 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2702 {
2703         int max;
2704         char *e;
2705         if (strncmp(buf, "system", 6)==0) {
2706                 mddev->sync_speed_max = 0;
2707                 return len;
2708         }
2709         max = simple_strtoul(buf, &e, 10);
2710         if (buf == e || (*e && *e != '\n') || max <= 0)
2711                 return -EINVAL;
2712         mddev->sync_speed_max = max;
2713         return len;
2714 }
2715
2716 static struct md_sysfs_entry md_sync_max =
2717 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2718
2719
2720 static ssize_t
2721 sync_speed_show(mddev_t *mddev, char *page)
2722 {
2723         unsigned long resync, dt, db;
2724         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2725         dt = ((jiffies - mddev->resync_mark) / HZ);
2726         if (!dt) dt++;
2727         db = resync - (mddev->resync_mark_cnt);
2728         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2729 }
2730
2731 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2732
2733 static ssize_t
2734 sync_completed_show(mddev_t *mddev, char *page)
2735 {
2736         unsigned long max_blocks, resync;
2737
2738         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2739                 max_blocks = mddev->resync_max_sectors;
2740         else
2741                 max_blocks = mddev->size << 1;
2742
2743         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2744         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2745 }
2746
2747 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2748
2749 static ssize_t
2750 suspend_lo_show(mddev_t *mddev, char *page)
2751 {
2752         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2753 }
2754
2755 static ssize_t
2756 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2757 {
2758         char *e;
2759         unsigned long long new = simple_strtoull(buf, &e, 10);
2760
2761         if (mddev->pers->quiesce == NULL)
2762                 return -EINVAL;
2763         if (buf == e || (*e && *e != '\n'))
2764                 return -EINVAL;
2765         if (new >= mddev->suspend_hi ||
2766             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2767                 mddev->suspend_lo = new;
2768                 mddev->pers->quiesce(mddev, 2);
2769                 return len;
2770         } else
2771                 return -EINVAL;
2772 }
2773 static struct md_sysfs_entry md_suspend_lo =
2774 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2775
2776
2777 static ssize_t
2778 suspend_hi_show(mddev_t *mddev, char *page)
2779 {
2780         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2781 }
2782
2783 static ssize_t
2784 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2785 {
2786         char *e;
2787         unsigned long long new = simple_strtoull(buf, &e, 10);
2788
2789         if (mddev->pers->quiesce == NULL)
2790                 return -EINVAL;
2791         if (buf == e || (*e && *e != '\n'))
2792                 return -EINVAL;
2793         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2794             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2795                 mddev->suspend_hi = new;
2796                 mddev->pers->quiesce(mddev, 1);
2797                 mddev->pers->quiesce(mddev, 0);
2798                 return len;
2799         } else
2800                 return -EINVAL;
2801 }
2802 static struct md_sysfs_entry md_suspend_hi =
2803 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2804
2805
2806 static struct attribute *md_default_attrs[] = {
2807         &md_level.attr,
2808         &md_layout.attr,
2809         &md_raid_disks.attr,
2810         &md_chunk_size.attr,
2811         &md_size.attr,
2812         &md_resync_start.attr,
2813         &md_metadata.attr,
2814         &md_new_device.attr,
2815         &md_safe_delay.attr,
2816         &md_array_state.attr,
2817         NULL,
2818 };
2819
2820 static struct attribute *md_redundancy_attrs[] = {
2821         &md_scan_mode.attr,
2822         &md_mismatches.attr,
2823         &md_sync_min.attr,
2824         &md_sync_max.attr,
2825         &md_sync_speed.attr,
2826         &md_sync_completed.attr,
2827         &md_suspend_lo.attr,
2828         &md_suspend_hi.attr,
2829         NULL,
2830 };
2831 static struct attribute_group md_redundancy_group = {
2832         .name = NULL,
2833         .attrs = md_redundancy_attrs,
2834 };
2835
2836
2837 static ssize_t
2838 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2839 {
2840         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2841         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2842         ssize_t rv;
2843
2844         if (!entry->show)
2845                 return -EIO;
2846         rv = mddev_lock(mddev);
2847         if (!rv) {
2848                 rv = entry->show(mddev, page);
2849                 mddev_unlock(mddev);
2850         }
2851         return rv;
2852 }
2853
2854 static ssize_t
2855 md_attr_store(struct kobject *kobj, struct attribute *attr,
2856               const char *page, size_t length)
2857 {
2858         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2859         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2860         ssize_t rv;
2861
2862         if (!entry->store)
2863                 return -EIO;
2864         if (!capable(CAP_SYS_ADMIN))
2865                 return -EACCES;
2866         rv = mddev_lock(mddev);
2867         if (!rv) {
2868                 rv = entry->store(mddev, page, length);
2869                 mddev_unlock(mddev);
2870         }
2871         return rv;
2872 }
2873
2874 static void md_free(struct kobject *ko)
2875 {
2876         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2877         kfree(mddev);
2878 }
2879
2880 static struct sysfs_ops md_sysfs_ops = {
2881         .show   = md_attr_show,
2882         .store  = md_attr_store,
2883 };
2884 static struct kobj_type md_ktype = {
2885         .release        = md_free,
2886         .sysfs_ops      = &md_sysfs_ops,
2887         .default_attrs  = md_default_attrs,
2888 };
2889
2890 int mdp_major = 0;
2891
2892 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2893 {
2894         static DEFINE_MUTEX(disks_mutex);
2895         mddev_t *mddev = mddev_find(dev);
2896         struct gendisk *disk;
2897         int partitioned = (MAJOR(dev) != MD_MAJOR);
2898         int shift = partitioned ? MdpMinorShift : 0;
2899         int unit = MINOR(dev) >> shift;
2900
2901         if (!mddev)
2902                 return NULL;
2903
2904         mutex_lock(&disks_mutex);
2905         if (mddev->gendisk) {
2906                 mutex_unlock(&disks_mutex);
2907                 mddev_put(mddev);
2908                 return NULL;
2909         }
2910         disk = alloc_disk(1 << shift);
2911         if (!disk) {
2912                 mutex_unlock(&disks_mutex);
2913                 mddev_put(mddev);
2914                 return NULL;
2915         }
2916         disk->major = MAJOR(dev);
2917         disk->first_minor = unit << shift;
2918         if (partitioned)
2919                 sprintf(disk->disk_name, "md_d%d", unit);
2920         else
2921                 sprintf(disk->disk_name, "md%d", unit);
2922         disk->fops = &md_fops;
2923         disk->private_data = mddev;
2924         disk->queue = mddev->queue;
2925         add_disk(disk);
2926         mddev->gendisk = disk;
2927         mutex_unlock(&disks_mutex);
2928         mddev->kobj.parent = &disk->kobj;
2929         mddev->kobj.k_name = NULL;
2930         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2931         mddev->kobj.ktype = &md_ktype;
2932         kobject_register(&mddev->kobj);
2933         return NULL;
2934 }
2935
2936 static void md_safemode_timeout(unsigned long data)
2937 {
2938         mddev_t *mddev = (mddev_t *) data;
2939
2940         mddev->safemode = 1;
2941         md_wakeup_thread(mddev->thread);
2942 }
2943
2944 static int start_dirty_degraded;
2945
2946 static int do_md_run(mddev_t * mddev)
2947 {
2948         int err;
2949         int chunk_size;
2950         struct list_head *tmp;
2951         mdk_rdev_t *rdev;
2952         struct gendisk *disk;
2953         struct mdk_personality *pers;
2954         char b[BDEVNAME_SIZE];
2955
2956         if (list_empty(&mddev->disks))
2957                 /* cannot run an array with no devices.. */
2958                 return -EINVAL;
2959
2960         if (mddev->pers)
2961                 return -EBUSY;
2962
2963         /*
2964          * Analyze all RAID superblock(s)
2965          */
2966         if (!mddev->raid_disks)
2967                 analyze_sbs(mddev);
2968
2969         chunk_size = mddev->chunk_size;
2970
2971         if (chunk_size) {
2972                 if (chunk_size > MAX_CHUNK_SIZE) {
2973                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2974                                 chunk_size, MAX_CHUNK_SIZE);
2975                         return -EINVAL;
2976                 }
2977                 /*
2978                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2979                  */
2980                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2981                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2982                         return -EINVAL;
2983                 }
2984                 if (chunk_size < PAGE_SIZE) {
2985                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2986                                 chunk_size, PAGE_SIZE);
2987                         return -EINVAL;
2988                 }
2989
2990                 /* devices must have minimum size of one chunk */
2991                 ITERATE_RDEV(mddev,rdev,tmp) {
2992                         if (test_bit(Faulty, &rdev->flags))
2993                                 continue;
2994                         if (rdev->size < chunk_size / 1024) {
2995                                 printk(KERN_WARNING
2996                                         "md: Dev %s smaller than chunk_size:"
2997                                         " %lluk < %dk\n",
2998                                         bdevname(rdev->bdev,b),
2999                                         (unsigned long long)rdev->size,
3000                                         chunk_size / 1024);
3001                                 return -EINVAL;
3002                         }
3003                 }
3004         }
3005
3006 #ifdef CONFIG_KMOD
3007         if (mddev->level != LEVEL_NONE)
3008                 request_module("md-level-%d", mddev->level);
3009         else if (mddev->clevel[0])
3010                 request_module("md-%s", mddev->clevel);
3011 #endif
3012
3013         /*
3014          * Drop all container device buffers, from now on
3015          * the only valid external interface is through the md
3016          * device.
3017          * Also find largest hardsector size
3018          */
3019         ITERATE_RDEV(mddev,rdev,tmp) {
3020                 if (test_bit(Faulty, &rdev->flags))
3021                         continue;
3022                 sync_blockdev(rdev->bdev);
3023                 invalidate_bdev(rdev->bdev, 0);
3024         }
3025
3026         md_probe(mddev->unit, NULL, NULL);
3027         disk = mddev->gendisk;
3028         if (!disk)
3029                 return -ENOMEM;
3030
3031         spin_lock(&pers_lock);
3032         pers = find_pers(mddev->level, mddev->clevel);
3033         if (!pers || !try_module_get(pers->owner)) {
3034                 spin_unlock(&pers_lock);
3035                 if (mddev->level != LEVEL_NONE)
3036                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3037                                mddev->level);
3038                 else
3039                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3040                                mddev->clevel);
3041                 return -EINVAL;
3042         }
3043         mddev->pers = pers;
3044         spin_unlock(&pers_lock);
3045         mddev->level = pers->level;
3046         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3047
3048         if (mddev->reshape_position != MaxSector &&
3049             pers->start_reshape == NULL) {
3050                 /* This personality cannot handle reshaping... */
3051                 mddev->pers = NULL;
3052                 module_put(pers->owner);
3053                 return -EINVAL;
3054         }
3055
3056         mddev->recovery = 0;
3057         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3058         mddev->barriers_work = 1;
3059         mddev->ok_start_degraded = start_dirty_degraded;
3060
3061         if (start_readonly)
3062                 mddev->ro = 2; /* read-only, but switch on first write */
3063
3064         err = mddev->pers->run(mddev);
3065         if (!err && mddev->pers->sync_request) {
3066                 err = bitmap_create(mddev);
3067                 if (err) {
3068                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3069                                mdname(mddev), err);
3070                         mddev->pers->stop(mddev);
3071                 }
3072         }
3073         if (err) {
3074                 printk(KERN_ERR "md: pers->run() failed ...\n");
3075                 module_put(mddev->pers->owner);
3076                 mddev->pers = NULL;
3077                 bitmap_destroy(mddev);
3078                 return err;
3079         }
3080         if (mddev->pers->sync_request)
3081                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
3082         else if (mddev->ro == 2) /* auto-readonly not meaningful */
3083                 mddev->ro = 0;
3084
3085         atomic_set(&mddev->writes_pending,0);
3086         mddev->safemode = 0;
3087         mddev->safemode_timer.function = md_safemode_timeout;
3088         mddev->safemode_timer.data = (unsigned long) mddev;
3089         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3090         mddev->in_sync = 1;
3091
3092         ITERATE_RDEV(mddev,rdev,tmp)
3093                 if (rdev->raid_disk >= 0) {
3094                         char nm[20];
3095                         sprintf(nm, "rd%d", rdev->raid_disk);
3096                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3097                 }
3098         
3099         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3100         
3101         if (mddev->sb_dirty)
3102                 md_update_sb(mddev);
3103
3104         set_capacity(disk, mddev->array_size<<1);
3105
3106         /* If we call blk_queue_make_request here, it will
3107          * re-initialise max_sectors etc which may have been
3108          * refined inside -> run.  So just set the bits we need to set.
3109          * Most initialisation happended when we called
3110          * blk_queue_make_request(..., md_fail_request)
3111          * earlier.
3112          */
3113         mddev->queue->queuedata = mddev;
3114         mddev->queue->make_request_fn = mddev->pers->make_request;
3115
3116         /* If there is a partially-recovered drive we need to
3117          * start recovery here.  If we leave it to md_check_recovery,
3118          * it will remove the drives and not do the right thing
3119          */
3120         if (mddev->degraded && !mddev->sync_thread) {
3121                 struct list_head *rtmp;
3122                 int spares = 0;
3123                 ITERATE_RDEV(mddev,rdev,rtmp)
3124                         if (rdev->raid_disk >= 0 &&
3125                             !test_bit(In_sync, &rdev->flags) &&
3126                             !test_bit(Faulty, &rdev->flags))
3127                                 /* complete an interrupted recovery */
3128                                 spares++;
3129                 if (spares && mddev->pers->sync_request) {
3130                         mddev->recovery = 0;
3131                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3132                         mddev->sync_thread = md_register_thread(md_do_sync,
3133                                                                 mddev,
3134                                                                 "%s_resync");
3135                         if (!mddev->sync_thread) {
3136                                 printk(KERN_ERR "%s: could not start resync"
3137                                        " thread...\n",
3138                                        mdname(mddev));
3139                                 /* leave the spares where they are, it shouldn't hurt */
3140                                 mddev->recovery = 0;
3141                         }
3142                 }
3143         }
3144         md_wakeup_thread(mddev->thread);
3145         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3146
3147         mddev->changed = 1;
3148         md_new_event(mddev);
3149         return 0;
3150 }
3151
3152 static int restart_array(mddev_t *mddev)
3153 {
3154         struct gendisk *disk = mddev->gendisk;
3155         int err;
3156
3157         /*
3158          * Complain if it has no devices
3159          */
3160         err = -ENXIO;
3161         if (list_empty(&mddev->disks))
3162                 goto out;
3163
3164         if (mddev->pers) {
3165                 err = -EBUSY;
3166                 if (!mddev->ro)
3167                         goto out;
3168
3169                 mddev->safemode = 0;
3170                 mddev->ro = 0;
3171                 set_disk_ro(disk, 0);
3172
3173                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3174                         mdname(mddev));
3175                 /*
3176                  * Kick recovery or resync if necessary
3177                  */
3178                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3179                 md_wakeup_thread(mddev->thread);
3180                 md_wakeup_thread(mddev->sync_thread);
3181                 err = 0;
3182         } else
3183                 err = -EINVAL;
3184
3185 out:
3186         return err;
3187 }
3188
3189 /* similar to deny_write_access, but accounts for our holding a reference
3190  * to the file ourselves */
3191 static int deny_bitmap_write_access(struct file * file)
3192 {
3193         struct inode *inode = file->f_mapping->host;
3194
3195         spin_lock(&inode->i_lock);
3196         if (atomic_read(&inode->i_writecount) > 1) {
3197                 spin_unlock(&inode->i_lock);
3198                 return -ETXTBSY;
3199         }
3200         atomic_set(&inode->i_writecount, -1);
3201         spin_unlock(&inode->i_lock);
3202
3203         return 0;
3204 }
3205
3206 static void restore_bitmap_write_access(struct file *file)
3207 {
3208         struct inode *inode = file->f_mapping->host;
3209
3210         spin_lock(&inode->i_lock);
3211         atomic_set(&inode->i_writecount, 1);
3212         spin_unlock(&inode->i_lock);
3213 }
3214
3215 /* mode:
3216  *   0 - completely stop and dis-assemble array
3217  *   1 - switch to readonly
3218  *   2 - stop but do not disassemble array
3219  */
3220 static int do_md_stop(mddev_t * mddev, int mode)
3221 {
3222         int err = 0;
3223         struct gendisk *disk = mddev->gendisk;
3224
3225         if (mddev->pers) {
3226                 if (atomic_read(&mddev->active)>2) {
3227                         printk("md: %s still in use.\n",mdname(mddev));
3228                         return -EBUSY;
3229                 }
3230
3231                 if (mddev->sync_thread) {
3232                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3233                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3234                         md_unregister_thread(mddev->sync_thread);
3235                         mddev->sync_thread = NULL;
3236                 }
3237
3238                 del_timer_sync(&mddev->safemode_timer);
3239
3240                 invalidate_partition(disk, 0);
3241
3242                 switch(mode) {
3243                 case 1: /* readonly */
3244                         err  = -ENXIO;
3245                         if (mddev->ro==1)
3246                                 goto out;
3247                         mddev->ro = 1;
3248                         break;
3249                 case 0: /* disassemble */
3250                 case 2: /* stop */
3251                         bitmap_flush(mddev);
3252                         md_super_wait(mddev);
3253                         if (mddev->ro)
3254                                 set_disk_ro(disk, 0);
3255                         blk_queue_make_request(mddev->queue, md_fail_request);
3256                         mddev->pers->stop(mddev);
3257                         if (mddev->pers->sync_request)
3258                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3259
3260                         module_put(mddev->pers->owner);
3261                         mddev->pers = NULL;
3262                         if (mddev->ro)
3263                                 mddev->ro = 0;
3264                 }
3265                 if (!mddev->in_sync || mddev->sb_dirty) {
3266                         /* mark array as shutdown cleanly */
3267                         mddev->in_sync = 1;
3268                         md_update_sb(mddev);
3269                 }
3270                 if (mode == 1)
3271                         set_disk_ro(disk, 1);
3272                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3273         }
3274
3275         /*
3276          * Free resources if final stop
3277          */
3278         if (mode == 0) {
3279                 mdk_rdev_t *rdev;
3280                 struct list_head *tmp;
3281                 struct gendisk *disk;
3282                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3283
3284                 bitmap_destroy(mddev);
3285                 if (mddev->bitmap_file) {
3286                         restore_bitmap_write_access(mddev->bitmap_file);
3287                         fput(mddev->bitmap_file);
3288                         mddev->bitmap_file = NULL;
3289                 }
3290                 mddev->bitmap_offset = 0;
3291
3292                 ITERATE_RDEV(mddev,rdev,tmp)
3293                         if (rdev->raid_disk >= 0) {
3294                                 char nm[20];
3295                                 sprintf(nm, "rd%d", rdev->raid_disk);
3296                                 sysfs_remove_link(&mddev->kobj, nm);
3297                         }
3298
3299                 export_array(mddev);
3300
3301                 mddev->array_size = 0;
3302                 mddev->size = 0;
3303                 mddev->raid_disks = 0;
3304                 mddev->recovery_cp = 0;
3305
3306                 disk = mddev->gendisk;
3307                 if (disk)
3308                         set_capacity(disk, 0);
3309                 mddev->changed = 1;
3310         } else if (mddev->pers)
3311                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3312                         mdname(mddev));
3313         err = 0;
3314         md_new_event(mddev);
3315 out:
3316         return err;
3317 }
3318
3319 static void autorun_array(mddev_t *mddev)
3320 {
3321         mdk_rdev_t *rdev;
3322         struct list_head *tmp;
3323         int err;
3324
3325         if (list_empty(&mddev->disks))
3326                 return;
3327
3328         printk(KERN_INFO "md: running: ");
3329
3330         ITERATE_RDEV(mddev,rdev,tmp) {
3331                 char b[BDEVNAME_SIZE];
3332                 printk("<%s>", bdevname(rdev->bdev,b));
3333         }
3334         printk("\n");
3335
3336         err = do_md_run (mddev);
3337         if (err) {
3338                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3339                 do_md_stop (mddev, 0);
3340         }
3341 }
3342
3343 /*
3344  * lets try to run arrays based on all disks that have arrived
3345  * until now. (those are in pending_raid_disks)
3346  *
3347  * the method: pick the first pending disk, collect all disks with
3348  * the same UUID, remove all from the pending list and put them into
3349  * the 'same_array' list. Then order this list based on superblock
3350  * update time (freshest comes first), kick out 'old' disks and
3351  * compare superblocks. If everything's fine then run it.
3352  *
3353  * If "unit" is allocated, then bump its reference count
3354  */
3355 static void autorun_devices(int part)
3356 {
3357         struct list_head *tmp;
3358         mdk_rdev_t *rdev0, *rdev;
3359         mddev_t *mddev;
3360         char b[BDEVNAME_SIZE];
3361
3362         printk(KERN_INFO "md: autorun ...\n");
3363         while (!list_empty(&pending_raid_disks)) {
3364                 dev_t dev;
3365                 LIST_HEAD(candidates);
3366                 rdev0 = list_entry(pending_raid_disks.next,
3367                                          mdk_rdev_t, same_set);
3368
3369                 printk(KERN_INFO "md: considering %s ...\n",
3370                         bdevname(rdev0->bdev,b));
3371                 INIT_LIST_HEAD(&candidates);
3372                 ITERATE_RDEV_PENDING(rdev,tmp)
3373                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3374                                 printk(KERN_INFO "md:  adding %s ...\n",
3375                                         bdevname(rdev->bdev,b));
3376                                 list_move(&rdev->same_set, &candidates);
3377                         }
3378                 /*
3379                  * now we have a set of devices, with all of them having
3380                  * mostly sane superblocks. It's time to allocate the
3381                  * mddev.
3382                  */
3383                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
3384                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3385                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3386                         break;
3387                 }
3388                 if (part)
3389                         dev = MKDEV(mdp_major,
3390                                     rdev0->preferred_minor << MdpMinorShift);
3391                 else
3392                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3393
3394                 md_probe(dev, NULL, NULL);
3395                 mddev = mddev_find(dev);
3396                 if (!mddev) {
3397                         printk(KERN_ERR 
3398                                 "md: cannot allocate memory for md drive.\n");
3399                         break;
3400                 }
3401                 if (mddev_lock(mddev)) 
3402                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3403                                mdname(mddev));
3404                 else if (mddev->raid_disks || mddev->major_version
3405                          || !list_empty(&mddev->disks)) {
3406                         printk(KERN_WARNING 
3407                                 "md: %s already running, cannot run %s\n",
3408                                 mdname(mddev), bdevname(rdev0->bdev,b));
3409                         mddev_unlock(mddev);
3410                 } else {
3411                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3412                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3413                                 list_del_init(&rdev->same_set);
3414                                 if (bind_rdev_to_array(rdev, mddev))
3415                                         export_rdev(rdev);
3416                         }
3417                         autorun_array(mddev);
3418                         mddev_unlock(mddev);
3419                 }
3420                 /* on success, candidates will be empty, on error
3421                  * it won't...
3422                  */
3423                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3424                         export_rdev(rdev);
3425                 mddev_put(mddev);
3426         }
3427         printk(KERN_INFO "md: ... autorun DONE.\n");
3428 }
3429
3430 /*
3431  * import RAID devices based on one partition
3432  * if possible, the array gets run as well.
3433  */
3434
3435 static int autostart_array(dev_t startdev)
3436 {
3437         char b[BDEVNAME_SIZE];
3438         int err = -EINVAL, i;
3439         mdp_super_t *sb = NULL;
3440         mdk_rdev_t *start_rdev = NULL, *rdev;
3441
3442         start_rdev = md_import_device(startdev, 0, 0);
3443         if (IS_ERR(start_rdev))
3444                 return err;
3445
3446
3447         /* NOTE: this can only work for 0.90.0 superblocks */
3448         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
3449         if (sb->major_version != 0 ||
3450             sb->minor_version != 90 ) {
3451                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
3452                 export_rdev(start_rdev);
3453                 return err;
3454         }
3455
3456         if (test_bit(Faulty, &start_rdev->flags)) {
3457                 printk(KERN_WARNING 
3458                         "md: can not autostart based on faulty %s!\n",
3459                         bdevname(start_rdev->bdev,b));
3460                 export_rdev(start_rdev);
3461                 return err;
3462         }
3463         list_add(&start_rdev->same_set, &pending_raid_disks);
3464
3465         for (i = 0; i < MD_SB_DISKS; i++) {
3466                 mdp_disk_t *desc = sb->disks + i;
3467                 dev_t dev = MKDEV(desc->major, desc->minor);
3468
3469                 if (!dev)
3470                         continue;
3471                 if (dev == startdev)
3472                         continue;
3473                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3474                         continue;
3475                 rdev = md_import_device(dev, 0, 0);
3476                 if (IS_ERR(rdev))
3477                         continue;
3478
3479                 list_add(&rdev->same_set, &pending_raid_disks);
3480         }
3481
3482         /*
3483          * possibly return codes
3484          */
3485         autorun_devices(0);
3486         return 0;
3487
3488 }
3489
3490
3491 static int get_version(void __user * arg)
3492 {
3493         mdu_version_t ver;
3494
3495         ver.major = MD_MAJOR_VERSION;
3496         ver.minor = MD_MINOR_VERSION;
3497         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3498
3499         if (copy_to_user(arg, &ver, sizeof(ver)))
3500                 return -EFAULT;
3501
3502         return 0;
3503 }
3504
3505 static int get_array_info(mddev_t * mddev, void __user * arg)
3506 {
3507         mdu_array_info_t info;
3508         int nr,working,active,failed,spare;
3509         mdk_rdev_t *rdev;
3510         struct list_head *tmp;
3511
3512         nr=working=active=failed=spare=0;
3513         ITERATE_RDEV(mddev,rdev,tmp) {
3514                 nr++;
3515                 if (test_bit(Faulty, &rdev->flags))
3516                         failed++;
3517                 else {
3518                         working++;
3519                         if (test_bit(In_sync, &rdev->flags))
3520                                 active++;       
3521                         else
3522                                 spare++;
3523                 }
3524         }
3525
3526         info.major_version = mddev->major_version;
3527         info.minor_version = mddev->minor_version;
3528         info.patch_version = MD_PATCHLEVEL_VERSION;
3529         info.ctime         = mddev->ctime;
3530         info.level         = mddev->level;
3531         info.size          = mddev->size;
3532         if (info.size != mddev->size) /* overflow */
3533                 info.size = -1;
3534         info.nr_disks      = nr;
3535         info.raid_disks    = mddev->raid_disks;
3536         info.md_minor      = mddev->md_minor;
3537         info.not_persistent= !mddev->persistent;
3538
3539         info.utime         = mddev->utime;
3540         info.state         = 0;
3541         if (mddev->in_sync)
3542                 info.state = (1<<MD_SB_CLEAN);
3543         if (mddev->bitmap && mddev->bitmap_offset)
3544                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3545         info.active_disks  = active;
3546         info.working_disks = working;
3547         info.failed_disks  = failed;
3548         info.spare_disks   = spare;
3549
3550         info.layout        = mddev->layout;
3551         info.chunk_size    = mddev->chunk_size;
3552
3553         if (copy_to_user(arg, &info, sizeof(info)))
3554                 return -EFAULT;
3555
3556         return 0;
3557 }
3558
3559 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3560 {
3561         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3562         char *ptr, *buf = NULL;
3563         int err = -ENOMEM;
3564
3565         file = kmalloc(sizeof(*file), GFP_KERNEL);
3566         if (!file)
3567                 goto out;
3568
3569         /* bitmap disabled, zero the first byte and copy out */
3570         if (!mddev->bitmap || !mddev->bitmap->file) {
3571                 file->pathname[0] = '\0';
3572                 goto copy_out;
3573         }
3574
3575         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3576         if (!buf)
3577                 goto out;
3578
3579         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3580         if (!ptr)
3581                 goto out;
3582
3583         strcpy(file->pathname, ptr);
3584
3585 copy_out:
3586         err = 0;
3587         if (copy_to_user(arg, file, sizeof(*file)))
3588                 err = -EFAULT;
3589 out:
3590         kfree(buf);
3591         kfree(file);
3592         return err;
3593 }
3594
3595 static int get_disk_info(mddev_t * mddev, void __user * arg)
3596 {
3597         mdu_disk_info_t info;
3598         unsigned int nr;
3599         mdk_rdev_t *rdev;
3600
3601         if (copy_from_user(&info, arg, sizeof(info)))
3602                 return -EFAULT;
3603
3604         nr = info.number;
3605
3606         rdev = find_rdev_nr(mddev, nr);
3607         if (rdev) {
3608                 info.major = MAJOR(rdev->bdev->bd_dev);
3609                 info.minor = MINOR(rdev->bdev->bd_dev);
3610                 info.raid_disk = rdev->raid_disk;
3611                 info.state = 0;
3612                 if (test_bit(Faulty, &rdev->flags))
3613                         info.state |= (1<<MD_DISK_FAULTY);
3614                 else if (test_bit(In_sync, &rdev->flags)) {
3615                         info.state |= (1<<MD_DISK_ACTIVE);
3616                         info.state |= (1<<MD_DISK_SYNC);
3617                 }
3618                 if (test_bit(WriteMostly, &rdev->flags))
3619                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3620         } else {
3621                 info.major = info.minor = 0;
3622                 info.raid_disk = -1;
3623                 info.state = (1<<MD_DISK_REMOVED);
3624         }
3625
3626         if (copy_to_user(arg, &info, sizeof(info)))
3627                 return -EFAULT;
3628
3629         return 0;
3630 }
3631
3632 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3633 {
3634         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3635         mdk_rdev_t *rdev;
3636         dev_t dev = MKDEV(info->major,info->minor);
3637
3638         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3639                 return -EOVERFLOW;
3640
3641         if (!mddev->raid_disks) {
3642                 int err;
3643                 /* expecting a device which has a superblock */
3644                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3645                 if (IS_ERR(rdev)) {
3646                         printk(KERN_WARNING 
3647                                 "md: md_import_device returned %ld\n",
3648                                 PTR_ERR(rdev));
3649                         return PTR_ERR(rdev);
3650                 }
3651                 if (!list_empty(&mddev->disks)) {
3652                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3653                                                         mdk_rdev_t, same_set);
3654                         int err = super_types[mddev->major_version]
3655                                 .load_super(rdev, rdev0, mddev->minor_version);
3656                         if (err < 0) {
3657                                 printk(KERN_WARNING 
3658                                         "md: %s has different UUID to %s\n",
3659                                         bdevname(rdev->bdev,b), 
3660                                         bdevname(rdev0->bdev,b2));
3661                                 export_rdev(rdev);
3662                                 return -EINVAL;
3663                         }
3664                 }
3665                 err = bind_rdev_to_array(rdev, mddev);
3666                 if (err)
3667                         export_rdev(rdev);
3668                 return err;
3669         }
3670
3671         /*
3672          * add_new_disk can be used once the array is assembled
3673          * to add "hot spares".  They must already have a superblock
3674          * written
3675          */
3676         if (mddev->pers) {
3677                 int err;
3678                 if (!mddev->pers->hot_add_disk) {
3679                         printk(KERN_WARNING 
3680                                 "%s: personality does not support diskops!\n",
3681                                mdname(mddev));
3682                         return -EINVAL;
3683                 }
3684                 if (mddev->persistent)
3685                         rdev = md_import_device(dev, mddev->major_version,
3686                                                 mddev->minor_version);
3687                 else
3688                         rdev = md_import_device(dev, -1, -1);
3689                 if (IS_ERR(rdev)) {
3690                         printk(KERN_WARNING 
3691                                 "md: md_import_device returned %ld\n",
3692                                 PTR_ERR(rdev));
3693                         return PTR_ERR(rdev);
3694                 }
3695                 /* set save_raid_disk if appropriate */
3696                 if (!mddev->persistent) {
3697                         if (info->state & (1<<MD_DISK_SYNC)  &&
3698                             info->raid_disk < mddev->raid_disks)
3699                                 rdev->raid_disk = info->raid_disk;
3700                         else
3701                                 rdev->raid_disk = -1;
3702                 } else
3703                         super_types[mddev->major_version].
3704                                 validate_super(mddev, rdev);
3705                 rdev->saved_raid_disk = rdev->raid_disk;
3706
3707                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3708                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3709                         set_bit(WriteMostly, &rdev->flags);
3710
3711                 rdev->raid_disk = -1;
3712                 err = bind_rdev_to_array(rdev, mddev);
3713                 if (!err && !mddev->pers->hot_remove_disk) {
3714                         /* If there is hot_add_disk but no hot_remove_disk
3715                          * then added disks for geometry changes,
3716                          * and should be added immediately.
3717                          */
3718                         super_types[mddev->major_version].
3719                                 validate_super(mddev, rdev);
3720                         err = mddev->pers->hot_add_disk(mddev, rdev);
3721                         if (err)
3722                                 unbind_rdev_from_array(rdev);
3723                 }
3724                 if (err)
3725                         export_rdev(rdev);
3726
3727                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3728                 md_wakeup_thread(mddev->thread);
3729                 return err;
3730         }
3731
3732         /* otherwise, add_new_disk is only allowed
3733          * for major_version==0 superblocks
3734          */
3735         if (mddev->major_version != 0) {
3736                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3737                        mdname(mddev));
3738                 return -EINVAL;
3739         }
3740
3741         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3742                 int err;
3743                 rdev = md_import_device (dev, -1, 0);
3744                 if (IS_ERR(rdev)) {
3745                         printk(KERN_WARNING 
3746                                 "md: error, md_import_device() returned %ld\n",
3747                                 PTR_ERR(rdev));
3748                         return PTR_ERR(rdev);
3749                 }
3750                 rdev->desc_nr = info->number;
3751                 if (info->raid_disk < mddev->raid_disks)
3752                         rdev->raid_disk = info->raid_disk;
3753                 else
3754                         rdev->raid_disk = -1;
3755
3756                 rdev->flags = 0;
3757
3758                 if (rdev->raid_disk < mddev->raid_disks)
3759                         if (info->state & (1<<MD_DISK_SYNC))
3760                                 set_bit(In_sync, &rdev->flags);
3761
3762                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3763                         set_bit(WriteMostly, &rdev->flags);
3764
3765                 if (!mddev->persistent) {
3766                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3767                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3768                 } else 
3769                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3770                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3771
3772                 err = bind_rdev_to_array(rdev, mddev);
3773                 if (err) {
3774                         export_rdev(rdev);
3775                         return err;
3776                 }
3777         }
3778
3779         return 0;
3780 }
3781
3782 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3783 {
3784         char b[BDEVNAME_SIZE];
3785         mdk_rdev_t *rdev;
3786
3787         if (!mddev->pers)
3788                 return -ENODEV;
3789
3790         rdev = find_rdev(mddev, dev);
3791         if (!rdev)
3792                 return -ENXIO;
3793
3794         if (rdev->raid_disk >= 0)
3795                 goto busy;
3796
3797         kick_rdev_from_array(rdev);
3798         md_update_sb(mddev);
3799         md_new_event(mddev);
3800
3801         return 0;
3802 busy:
3803         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3804                 bdevname(rdev->bdev,b), mdname(mddev));
3805         return -EBUSY;
3806 }
3807
3808 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3809 {
3810         char b[BDEVNAME_SIZE];
3811         int err;
3812         unsigned int size;
3813         mdk_rdev_t *rdev;
3814
3815         if (!mddev->pers)
3816                 return -ENODEV;
3817
3818         if (mddev->major_version != 0) {
3819                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3820                         " version-0 superblocks.\n",
3821                         mdname(mddev));
3822                 return -EINVAL;
3823         }
3824         if (!mddev->pers->hot_add_disk) {
3825                 printk(KERN_WARNING 
3826                         "%s: personality does not support diskops!\n",
3827                         mdname(mddev));
3828                 return -EINVAL;
3829         }
3830
3831         rdev = md_import_device (dev, -1, 0);
3832         if (IS_ERR(rdev)) {
3833                 printk(KERN_WARNING 
3834                         "md: error, md_import_device() returned %ld\n",
3835                         PTR_ERR(rdev));
3836                 return -EINVAL;
3837         }
3838
3839         if (mddev->persistent)
3840                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3841         else
3842                 rdev->sb_offset =
3843                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3844
3845         size = calc_dev_size(rdev, mddev->chunk_size);
3846         rdev->size = size;
3847
3848         if (test_bit(Faulty, &rdev->flags)) {
3849                 printk(KERN_WARNING 
3850                         "md: can not hot-add faulty %s disk to %s!\n",
3851                         bdevname(rdev->bdev,b), mdname(mddev));
3852                 err = -EINVAL;
3853                 goto abort_export;
3854         }
3855         clear_bit(In_sync, &rdev->flags);
3856         rdev->desc_nr = -1;
3857         err = bind_rdev_to_array(rdev, mddev);
3858         if (err)
3859                 goto abort_export;
3860
3861         /*
3862          * The rest should better be atomic, we can have disk failures
3863          * noticed in interrupt contexts ...
3864          */
3865
3866         if (rdev->desc_nr == mddev->max_disks) {
3867                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3868                         mdname(mddev));
3869                 err = -EBUSY;
3870                 goto abort_unbind_export;
3871         }
3872
3873         rdev->raid_disk = -1;
3874
3875         md_update_sb(mddev);
3876
3877         /*
3878          * Kick recovery, maybe this spare has to be added to the
3879          * array immediately.
3880          */
3881         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3882         md_wakeup_thread(mddev->thread);
3883         md_new_event(mddev);
3884         return 0;
3885
3886 abort_unbind_export:
3887         unbind_rdev_from_array(rdev);
3888
3889 abort_export:
3890         export_rdev(rdev);
3891         return err;
3892 }
3893
3894 static int set_bitmap_file(mddev_t *mddev, int fd)
3895 {
3896         int err;
3897
3898         if (mddev->pers) {
3899                 if (!mddev->pers->quiesce)
3900                         return -EBUSY;
3901                 if (mddev->recovery || mddev->sync_thread)
3902                         return -EBUSY;
3903                 /* we should be able to change the bitmap.. */
3904         }
3905
3906
3907         if (fd >= 0) {
3908                 if (mddev->bitmap)
3909                         return -EEXIST; /* cannot add when bitmap is present */
3910                 mddev->bitmap_file = fget(fd);
3911
3912                 if (mddev->bitmap_file == NULL) {
3913                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3914                                mdname(mddev));
3915                         return -EBADF;
3916                 }
3917
3918                 err = deny_bitmap_write_access(mddev->bitmap_file);
3919                 if (err) {
3920                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3921                                mdname(mddev));
3922                         fput(mddev->bitmap_file);
3923                         mddev->bitmap_file = NULL;
3924                         return err;
3925                 }
3926                 mddev->bitmap_offset = 0; /* file overrides offset */
3927         } else if (mddev->bitmap == NULL)
3928                 return -ENOENT; /* cannot remove what isn't there */
3929         err = 0;
3930         if (mddev->pers) {
3931                 mddev->pers->quiesce(mddev, 1);
3932                 if (fd >= 0)
3933                         err = bitmap_create(mddev);
3934                 if (fd < 0 || err) {
3935                         bitmap_destroy(mddev);
3936                         fd = -1; /* make sure to put the file */
3937                 }
3938                 mddev->pers->quiesce(mddev, 0);
3939         }
3940         if (fd < 0) {
3941                 if (mddev->bitmap_file) {
3942                         restore_bitmap_write_access(mddev->bitmap_file);
3943                         fput(mddev->bitmap_file);
3944                 }
3945                 mddev->bitmap_file = NULL;
3946         }
3947
3948         return err;
3949 }
3950
3951 /*
3952  * set_array_info is used two different ways
3953  * The original usage is when creating a new array.
3954  * In this usage, raid_disks is > 0 and it together with
3955  *  level, size, not_persistent,layout,chunksize determine the
3956  *  shape of the array.
3957  *  This will always create an array with a type-0.90.0 superblock.
3958  * The newer usage is when assembling an array.
3959  *  In this case raid_disks will be 0, and the major_version field is
3960  *  use to determine which style super-blocks are to be found on the devices.
3961  *  The minor and patch _version numbers are also kept incase the
3962  *  super_block handler wishes to interpret them.
3963  */
3964 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3965 {
3966
3967         if (info->raid_disks == 0) {
3968                 /* just setting version number for superblock loading */
3969                 if (info->major_version < 0 ||
3970                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3971                     super_types[info->major_version].name == NULL) {
3972                         /* maybe try to auto-load a module? */
3973                         printk(KERN_INFO 
3974                                 "md: superblock version %d not known\n",
3975                                 info->major_version);
3976                         return -EINVAL;
3977                 }
3978                 mddev->major_version = info->major_version;
3979                 mddev->minor_version = info->minor_version;
3980                 mddev->patch_version = info->patch_version;
3981                 return 0;
3982         }
3983         mddev->major_version = MD_MAJOR_VERSION;
3984         mddev->minor_version = MD_MINOR_VERSION;
3985         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3986         mddev->ctime         = get_seconds();
3987
3988         mddev->level         = info->level;
3989         mddev->clevel[0]     = 0;
3990         mddev->size          = info->size;
3991         mddev->raid_disks    = info->raid_disks;
3992         /* don't set md_minor, it is determined by which /dev/md* was
3993          * openned
3994          */
3995         if (info->state & (1<<MD_SB_CLEAN))
3996                 mddev->recovery_cp = MaxSector;
3997         else
3998                 mddev->recovery_cp = 0;
3999         mddev->persistent    = ! info->not_persistent;
4000
4001         mddev->layout        = info->layout;
4002         mddev->chunk_size    = info->chunk_size;
4003
4004         mddev->max_disks     = MD_SB_DISKS;
4005
4006         mddev->sb_dirty      = 1;
4007
4008         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4009         mddev->bitmap_offset = 0;
4010
4011         mddev->reshape_position = MaxSector;
4012
4013         /*
4014          * Generate a 128 bit UUID
4015          */
4016         get_random_bytes(mddev->uuid, 16);
4017
4018         mddev->new_level = mddev->level;
4019         mddev->new_chunk = mddev->chunk_size;
4020         mddev->new_layout = mddev->layout;
4021         mddev->delta_disks = 0;
4022
4023         return 0;
4024 }
4025
4026 static int update_size(mddev_t *mddev, unsigned long size)
4027 {
4028         mdk_rdev_t * rdev;
4029         int rv;
4030         struct list_head *tmp;
4031         int fit = (size == 0);
4032
4033         if (mddev->pers->resize == NULL)
4034                 return -EINVAL;
4035         /* The "size" is the amount of each device that is used.
4036          * This can only make sense for arrays with redundancy.
4037          * linear and raid0 always use whatever space is available
4038          * We can only consider changing the size if no resync
4039          * or reconstruction is happening, and if the new size
4040          * is acceptable. It must fit before the sb_offset or,
4041          * if that is <data_offset, it must fit before the
4042          * size of each device.
4043          * If size is zero, we find the largest size that fits.
4044          */
4045         if (mddev->sync_thread)
4046                 return -EBUSY;
4047         ITERATE_RDEV(mddev,rdev,tmp) {
4048                 sector_t avail;
4049                 if (rdev->sb_offset > rdev->data_offset)
4050                         avail = (rdev->sb_offset*2) - rdev->data_offset;
4051                 else
4052                         avail = get_capacity(rdev->bdev->bd_disk)
4053                                 - rdev->data_offset;
4054                 if (fit && (size == 0 || size > avail/2))
4055                         size = avail/2;
4056                 if (avail < ((sector_t)size << 1))
4057                         return -ENOSPC;
4058         }
4059         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4060         if (!rv) {
4061                 struct block_device *bdev;
4062
4063                 bdev = bdget_disk(mddev->gendisk, 0);
4064                 if (bdev) {
4065                         mutex_lock(&bdev->bd_inode->i_mutex);
4066                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4067                         mutex_unlock(&bdev->bd_inode->i_mutex);
4068                         bdput(bdev);
4069                 }
4070         }
4071         return rv;
4072 }
4073
4074 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4075 {
4076         int rv;
4077         /* change the number of raid disks */
4078         if (mddev->pers->check_reshape == NULL)
4079                 return -EINVAL;
4080         if (raid_disks <= 0 ||
4081             raid_disks >= mddev->max_disks)
4082                 return -EINVAL;
4083         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4084                 return -EBUSY;
4085         mddev->delta_disks = raid_disks - mddev->raid_disks;
4086
4087         rv = mddev->pers->check_reshape(mddev);
4088         return rv;
4089 }
4090
4091
4092 /*
4093  * update_array_info is used to change the configuration of an
4094  * on-line array.
4095  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4096  * fields in the info are checked against the array.
4097  * Any differences that cannot be handled will cause an error.
4098  * Normally, only one change can be managed at a time.
4099  */
4100 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4101 {
4102         int rv = 0;
4103         int cnt = 0;
4104         int state = 0;
4105
4106         /* calculate expected state,ignoring low bits */
4107         if (mddev->bitmap && mddev->bitmap_offset)
4108                 state |= (1 << MD_SB_BITMAP_PRESENT);
4109
4110         if (mddev->major_version != info->major_version ||
4111             mddev->minor_version != info->minor_version ||
4112 /*          mddev->patch_version != info->patch_version || */
4113             mddev->ctime         != info->ctime         ||
4114             mddev->level         != info->level         ||
4115 /*          mddev->layout        != info->layout        || */
4116             !mddev->persistent   != info->not_persistent||
4117             mddev->chunk_size    != info->chunk_size    ||
4118             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4119             ((state^info->state) & 0xfffffe00)
4120                 )
4121                 return -EINVAL;
4122         /* Check there is only one change */
4123         if (info->size >= 0 && mddev->size != info->size) cnt++;
4124         if (mddev->raid_disks != info->raid_disks) cnt++;
4125         if (mddev->layout != info->layout) cnt++;
4126         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4127         if (cnt == 0) return 0;
4128         if (cnt > 1) return -EINVAL;
4129
4130         if (mddev->layout != info->layout) {
4131                 /* Change layout
4132                  * we don't need to do anything at the md level, the
4133                  * personality will take care of it all.
4134                  */
4135                 if (mddev->pers->reconfig == NULL)
4136                         return -EINVAL;
4137                 else
4138                         return mddev->pers->reconfig(mddev, info->layout, -1);
4139         }
4140         if (info->size >= 0 && mddev->size != info->size)
4141                 rv = update_size(mddev, info->size);
4142
4143         if (mddev->raid_disks    != info->raid_disks)
4144                 rv = update_raid_disks(mddev, info->raid_disks);
4145
4146         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4147                 if (mddev->pers->quiesce == NULL)
4148                         return -EINVAL;
4149                 if (mddev->recovery || mddev->sync_thread)
4150                         return -EBUSY;
4151                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4152                         /* add the bitmap */
4153                         if (mddev->bitmap)
4154                                 return -EEXIST;
4155                         if (mddev->default_bitmap_offset == 0)
4156                                 return -EINVAL;
4157                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4158                         mddev->pers->quiesce(mddev, 1);
4159                         rv = bitmap_create(mddev);
4160                         if (rv)
4161                                 bitmap_destroy(mddev);
4162                         mddev->pers->quiesce(mddev, 0);
4163                 } else {
4164                         /* remove the bitmap */
4165                         if (!mddev->bitmap)
4166                                 return -ENOENT;
4167                         if (mddev->bitmap->file)
4168                                 return -EINVAL;
4169                         mddev->pers->quiesce(mddev, 1);
4170                         bitmap_destroy(mddev);
4171                         mddev->pers->quiesce(mddev, 0);
4172                         mddev->bitmap_offset = 0;
4173                 }
4174         }
4175         md_update_sb(mddev);
4176         return rv;
4177 }
4178
4179 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4180 {
4181         mdk_rdev_t *rdev;
4182
4183         if (mddev->pers == NULL)
4184                 return -ENODEV;
4185
4186         rdev = find_rdev(mddev, dev);
4187         if (!rdev)
4188                 return -ENODEV;
4189
4190         md_error(mddev, rdev);
4191         return 0;
4192 }
4193
4194 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4195 {
4196         mddev_t *mddev = bdev->bd_disk->private_data;
4197
4198         geo->heads = 2;
4199         geo->sectors = 4;
4200         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4201         return 0;
4202 }
4203
4204 static int md_ioctl(struct inode *inode, struct file *file,
4205                         unsigned int cmd, unsigned long arg)
4206 {
4207         int err = 0;
4208         void __user *argp = (void __user *)arg;
4209         mddev_t *mddev = NULL;
4210
4211         if (!capable(CAP_SYS_ADMIN))
4212                 return -EACCES;
4213
4214         /*
4215          * Commands dealing with the RAID driver but not any
4216          * particular array:
4217          */
4218         switch (cmd)
4219         {
4220                 case RAID_VERSION:
4221                         err = get_version(argp);
4222                         goto done;
4223
4224                 case PRINT_RAID_DEBUG:
4225                         err = 0;
4226                         md_print_devices();
4227                         goto done;
4228
4229 #ifndef MODULE
4230                 case RAID_AUTORUN:
4231                         err = 0;
4232                         autostart_arrays(arg);
4233                         goto done;
4234 #endif
4235                 default:;
4236         }
4237
4238         /*
4239          * Commands creating/starting a new array:
4240          */
4241
4242         mddev = inode->i_bdev->bd_disk->private_data;
4243
4244         if (!mddev) {
4245                 BUG();
4246                 goto abort;
4247         }
4248
4249
4250         if (cmd == START_ARRAY) {
4251                 /* START_ARRAY doesn't need to lock the array as autostart_array
4252                  * does the locking, and it could even be a different array
4253                  */
4254                 static int cnt = 3;
4255                 if (cnt > 0 ) {
4256                         printk(KERN_WARNING
4257                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
4258                                "This will not be supported beyond July 2006\n",
4259                                current->comm, current->pid);
4260                         cnt--;
4261                 }
4262                 err = autostart_array(new_decode_dev(arg));
4263                 if (err) {
4264                         printk(KERN_WARNING "md: autostart failed!\n");
4265                         goto abort;
4266                 }
4267                 goto done;
4268         }
4269
4270         err = mddev_lock(mddev);
4271         if (err) {
4272                 printk(KERN_INFO 
4273                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4274                         err, cmd);
4275                 goto abort;
4276         }
4277
4278         switch (cmd)
4279         {
4280                 case SET_ARRAY_INFO:
4281                         {
4282                                 mdu_array_info_t info;
4283                                 if (!arg)
4284                                         memset(&info, 0, sizeof(info));
4285                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4286                                         err = -EFAULT;
4287                                         goto abort_unlock;
4288                                 }
4289                                 if (mddev->pers) {
4290                                         err = update_array_info(mddev, &info);
4291                                         if (err) {
4292                                                 printk(KERN_WARNING "md: couldn't update"
4293                                                        " array info. %d\n", err);
4294                                                 goto abort_unlock;
4295                                         }
4296                                         goto done_unlock;
4297                                 }
4298                                 if (!list_empty(&mddev->disks)) {
4299                                         printk(KERN_WARNING
4300                                                "md: array %s already has disks!\n",
4301                                                mdname(mddev));
4302                                         err = -EBUSY;
4303                                         goto abort_unlock;
4304                                 }
4305                                 if (mddev->raid_disks) {
4306                                         printk(KERN_WARNING
4307                                                "md: array %s already initialised!\n",
4308                                                mdname(mddev));
4309                                         err = -EBUSY;
4310                                         goto abort_unlock;
4311                                 }
4312                                 err = set_array_info(mddev, &info);
4313                                 if (err) {
4314                                         printk(KERN_WARNING "md: couldn't set"
4315                                                " array info. %d\n", err);
4316                                         goto abort_unlock;
4317                                 }
4318                         }
4319                         goto done_unlock;
4320
4321                 default:;
4322         }
4323
4324         /*
4325          * Commands querying/configuring an existing array:
4326          */
4327         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4328          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4329         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4330                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
4331                 err = -ENODEV;
4332                 goto abort_unlock;
4333         }
4334
4335         /*
4336          * Commands even a read-only array can execute:
4337          */
4338         switch (cmd)
4339         {
4340                 case GET_ARRAY_INFO:
4341                         err = get_array_info(mddev, argp);
4342                         goto done_unlock;
4343
4344                 case GET_BITMAP_FILE:
4345                         err = get_bitmap_file(mddev, argp);
4346                         goto done_unlock;
4347
4348                 case GET_DISK_INFO:
4349                         err = get_disk_info(mddev, argp);
4350                         goto done_unlock;
4351
4352                 case RESTART_ARRAY_RW:
4353                         err = restart_array(mddev);
4354                         goto done_unlock;
4355
4356                 case STOP_ARRAY:
4357                         err = do_md_stop (mddev, 0);
4358                         goto done_unlock;
4359
4360                 case STOP_ARRAY_RO:
4361                         err = do_md_stop (mddev, 1);
4362                         goto done_unlock;
4363
4364         /*
4365          * We have a problem here : there is no easy way to give a CHS
4366          * virtual geometry. We currently pretend that we have a 2 heads
4367          * 4 sectors (with a BIG number of cylinders...). This drives
4368          * dosfs just mad... ;-)
4369          */
4370         }
4371
4372         /*
4373          * The remaining ioctls are changing the state of the
4374          * superblock, so we do not allow them on read-only arrays.
4375          * However non-MD ioctls (e.g. get-size) will still come through
4376          * here and hit the 'default' below, so only disallow
4377          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4378          */
4379         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4380             mddev->ro && mddev->pers) {
4381                 if (mddev->ro == 2) {
4382                         mddev->ro = 0;
4383                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4384                 md_wakeup_thread(mddev->thread);
4385
4386                 } else {
4387                         err = -EROFS;
4388                         goto abort_unlock;
4389                 }
4390         }
4391
4392         switch (cmd)
4393         {
4394                 case ADD_NEW_DISK:
4395                 {
4396                         mdu_disk_info_t info;
4397                         if (copy_from_user(&info, argp, sizeof(info)))
4398                                 err = -EFAULT;
4399                         else
4400                                 err = add_new_disk(mddev, &info);
4401                         goto done_unlock;
4402                 }
4403
4404                 case HOT_REMOVE_DISK:
4405                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4406                         goto done_unlock;
4407
4408                 case HOT_ADD_DISK:
4409                         err = hot_add_disk(mddev, new_decode_dev(arg));
4410                         goto done_unlock;
4411
4412                 case SET_DISK_FAULTY:
4413                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4414                         goto done_unlock;
4415
4416                 case RUN_ARRAY:
4417                         err = do_md_run (mddev);
4418                         goto done_unlock;
4419
4420                 case SET_BITMAP_FILE:
4421                         err = set_bitmap_file(mddev, (int)arg);
4422                         goto done_unlock;
4423
4424                 default:
4425                         err = -EINVAL;
4426                         goto abort_unlock;
4427         }
4428
4429 done_unlock:
4430 abort_unlock:
4431         mddev_unlock(mddev);
4432
4433         return err;
4434 done:
4435         if (err)
4436                 MD_BUG();
4437 abort:
4438         return err;
4439 }
4440
4441 static int md_open(struct inode *inode, struct file *file)
4442 {
4443         /*
4444          * Succeed if we can lock the mddev, which confirms that
4445          * it isn't being stopped right now.
4446          */
4447         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4448         int err;
4449
4450         if ((err = mddev_lock(mddev)))
4451                 goto out;
4452
4453         err = 0;
4454         mddev_get(mddev);
4455         mddev_unlock(mddev);
4456
4457         check_disk_change(inode->i_bdev);
4458  out:
4459         return err;
4460 }
4461
4462 static int md_release(struct inode *inode, struct file * file)
4463 {
4464         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4465
4466         if (!mddev)
4467                 BUG();
4468         mddev_put(mddev);
4469
4470         return 0;
4471 }
4472
4473 static int md_media_changed(struct gendisk *disk)
4474 {
4475         mddev_t *mddev = disk->private_data;
4476
4477         return mddev->changed;
4478 }
4479
4480 static int md_revalidate(struct gendisk *disk)
4481 {
4482         mddev_t *mddev = disk->private_data;
4483
4484         mddev->changed = 0;
4485         return 0;
4486 }
4487 static struct block_device_operations md_fops =
4488 {
4489         .owner          = THIS_MODULE,
4490         .open           = md_open,
4491         .release        = md_release,
4492         .ioctl          = md_ioctl,
4493         .getgeo         = md_getgeo,
4494         .media_changed  = md_media_changed,
4495         .revalidate_disk= md_revalidate,
4496 };
4497
4498 static int md_thread(void * arg)
4499 {
4500         mdk_thread_t *thread = arg;
4501
4502         /*
4503          * md_thread is a 'system-thread', it's priority should be very
4504          * high. We avoid resource deadlocks individually in each
4505          * raid personality. (RAID5 does preallocation) We also use RR and
4506          * the very same RT priority as kswapd, thus we will never get
4507          * into a priority inversion deadlock.
4508          *
4509          * we definitely have to have equal or higher priority than
4510          * bdflush, otherwise bdflush will deadlock if there are too
4511          * many dirty RAID5 blocks.
4512          */
4513
4514         allow_signal(SIGKILL);
4515         while (!kthread_should_stop()) {
4516
4517                 /* We need to wait INTERRUPTIBLE so that
4518                  * we don't add to the load-average.
4519                  * That means we need to be sure no signals are
4520                  * pending
4521                  */
4522                 if (signal_pending(current))
4523                         flush_signals(current);
4524
4525                 wait_event_interruptible_timeout
4526                         (thread->wqueue,
4527                          test_bit(THREAD_WAKEUP, &thread->flags)
4528                          || kthread_should_stop(),
4529                          thread->timeout);
4530                 try_to_freeze();
4531
4532                 clear_bit(THREAD_WAKEUP, &thread->flags);
4533
4534                 thread->run(thread->mddev);
4535         }
4536
4537         return 0;
4538 }
4539
4540 void md_wakeup_thread(mdk_thread_t *thread)
4541 {
4542         if (thread) {
4543                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4544                 set_bit(THREAD_WAKEUP, &thread->flags);
4545                 wake_up(&thread->wqueue);
4546         }
4547 }
4548
4549 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4550                                  const char *name)
4551 {
4552         mdk_thread_t *thread;
4553
4554         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4555         if (!thread)
4556                 return NULL;
4557
4558         init_waitqueue_head(&thread->wqueue);
4559
4560         thread->run = run;
4561         thread->mddev = mddev;
4562         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4563         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4564         if (IS_ERR(thread->tsk)) {
4565                 kfree(thread);
4566                 return NULL;
4567         }
4568         return thread;
4569 }
4570
4571 void md_unregister_thread(mdk_thread_t *thread)
4572 {
4573         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4574
4575         kthread_stop(thread->tsk);
4576         kfree(thread);
4577 }
4578
4579 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4580 {
4581         if (!mddev) {
4582                 MD_BUG();
4583                 return;
4584         }
4585
4586         if (!rdev || test_bit(Faulty, &rdev->flags))
4587                 return;
4588 /*
4589         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4590                 mdname(mddev),
4591                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4592                 __builtin_return_address(0),__builtin_return_address(1),
4593                 __builtin_return_address(2),__builtin_return_address(3));
4594 */
4595         if (!mddev->pers)
4596                 return;
4597         if (!mddev->pers->error_handler)
4598                 return;
4599         mddev->pers->error_handler(mddev,rdev);
4600         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4601         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4602         md_wakeup_thread(mddev->thread);
4603         md_new_event_inintr(mddev);
4604 }
4605
4606 /* seq_file implementation /proc/mdstat */
4607
4608 static void status_unused(struct seq_file *seq)
4609 {
4610         int i = 0;
4611         mdk_rdev_t *rdev;
4612         struct list_head *tmp;
4613
4614         seq_printf(seq, "unused devices: ");
4615
4616         ITERATE_RDEV_PENDING(rdev,tmp) {
4617                 char b[BDEVNAME_SIZE];
4618                 i++;
4619                 seq_printf(seq, "%s ",
4620                               bdevname(rdev->bdev,b));
4621         }
4622         if (!i)
4623                 seq_printf(seq, "<none>");
4624
4625         seq_printf(seq, "\n");
4626 }
4627
4628
4629 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4630 {
4631         sector_t max_blocks, resync, res;
4632         unsigned long dt, db, rt;
4633         int scale;
4634         unsigned int per_milli;
4635
4636         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4637
4638         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4639                 max_blocks = mddev->resync_max_sectors >> 1;
4640         else
4641                 max_blocks = mddev->size;
4642
4643         /*
4644          * Should not happen.
4645          */
4646         if (!max_blocks) {
4647                 MD_BUG();
4648                 return;
4649         }
4650         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4651          * in a sector_t, and (max_blocks>>scale) will fit in a
4652          * u32, as those are the requirements for sector_div.
4653          * Thus 'scale' must be at least 10
4654          */
4655         scale = 10;
4656         if (sizeof(sector_t) > sizeof(unsigned long)) {
4657                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4658                         scale++;
4659         }
4660         res = (resync>>scale)*1000;
4661         sector_div(res, (u32)((max_blocks>>scale)+1));
4662
4663         per_milli = res;
4664         {
4665                 int i, x = per_milli/50, y = 20-x;
4666                 seq_printf(seq, "[");
4667                 for (i = 0; i < x; i++)
4668                         seq_printf(seq, "=");
4669                 seq_printf(seq, ">");
4670                 for (i = 0; i < y; i++)
4671                         seq_printf(seq, ".");
4672                 seq_printf(seq, "] ");
4673         }
4674         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4675                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4676                     "reshape" :
4677                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4678                        "resync" : "recovery")),
4679                       per_milli/10, per_milli % 10,
4680                    (unsigned long long) resync,
4681                    (unsigned long long) max_blocks);
4682
4683         /*
4684          * We do not want to overflow, so the order of operands and
4685          * the * 100 / 100 trick are important. We do a +1 to be
4686          * safe against division by zero. We only estimate anyway.
4687          *
4688          * dt: time from mark until now
4689          * db: blocks written from mark until now
4690          * rt: remaining time
4691          */
4692         dt = ((jiffies - mddev->resync_mark) / HZ);
4693         if (!dt) dt++;
4694         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4695                 - mddev->resync_mark_cnt;
4696         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4697
4698         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4699
4700         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4701 }
4702
4703 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4704 {
4705         struct list_head *tmp;
4706         loff_t l = *pos;
4707         mddev_t *mddev;
4708
4709         if (l >= 0x10000)
4710                 return NULL;
4711         if (!l--)
4712                 /* header */
4713                 return (void*)1;
4714
4715         spin_lock(&all_mddevs_lock);
4716         list_for_each(tmp,&all_mddevs)
4717                 if (!l--) {
4718                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4719                         mddev_get(mddev);
4720                         spin_unlock(&all_mddevs_lock);
4721                         return mddev;
4722                 }
4723         spin_unlock(&all_mddevs_lock);
4724         if (!l--)
4725                 return (void*)2;/* tail */
4726         return NULL;
4727 }
4728
4729 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4730 {
4731         struct list_head *tmp;
4732         mddev_t *next_mddev, *mddev = v;
4733         
4734         ++*pos;
4735         if (v == (void*)2)
4736                 return NULL;
4737
4738         spin_lock(&all_mddevs_lock);
4739         if (v == (void*)1)
4740                 tmp = all_mddevs.next;
4741         else
4742                 tmp = mddev->all_mddevs.next;
4743         if (tmp != &all_mddevs)
4744                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4745         else {
4746                 next_mddev = (void*)2;
4747                 *pos = 0x10000;
4748         }               
4749         spin_unlock(&all_mddevs_lock);
4750
4751         if (v != (void*)1)
4752                 mddev_put(mddev);
4753         return next_mddev;
4754
4755 }
4756
4757 static void md_seq_stop(struct seq_file *seq, void *v)
4758 {
4759         mddev_t *mddev = v;
4760
4761         if (mddev && v != (void*)1 && v != (void*)2)
4762                 mddev_put(mddev);
4763 }
4764
4765 struct mdstat_info {
4766         int event;
4767 };
4768
4769 static int md_seq_show(struct seq_file *seq, void *v)
4770 {
4771         mddev_t *mddev = v;
4772         sector_t size;
4773         struct list_head *tmp2;
4774         mdk_rdev_t *rdev;
4775         struct mdstat_info *mi = seq->private;
4776         struct bitmap *bitmap;
4777
4778         if (v == (void*)1) {
4779                 struct mdk_personality *pers;
4780                 seq_printf(seq, "Personalities : ");
4781                 spin_lock(&pers_lock);
4782                 list_for_each_entry(pers, &pers_list, list)
4783                         seq_printf(seq, "[%s] ", pers->name);
4784
4785                 spin_unlock(&pers_lock);
4786                 seq_printf(seq, "\n");
4787                 mi->event = atomic_read(&md_event_count);
4788                 return 0;
4789         }
4790         if (v == (void*)2) {
4791                 status_unused(seq);
4792                 return 0;
4793         }
4794
4795         if (mddev_lock(mddev) < 0)
4796                 return -EINTR;
4797
4798         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4799                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4800                                                 mddev->pers ? "" : "in");
4801                 if (mddev->pers) {
4802                         if (mddev->ro==1)
4803                                 seq_printf(seq, " (read-only)");
4804                         if (mddev->ro==2)
4805                                 seq_printf(seq, "(auto-read-only)");
4806                         seq_printf(seq, " %s", mddev->pers->name);
4807                 }
4808
4809                 size = 0;
4810                 ITERATE_RDEV(mddev,rdev,tmp2) {
4811                         char b[BDEVNAME_SIZE];
4812                         seq_printf(seq, " %s[%d]",
4813                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4814                         if (test_bit(WriteMostly, &rdev->flags))
4815                                 seq_printf(seq, "(W)");
4816                         if (test_bit(Faulty, &rdev->flags)) {
4817                                 seq_printf(seq, "(F)");
4818                                 continue;
4819                         } else if (rdev->raid_disk < 0)
4820                                 seq_printf(seq, "(S)"); /* spare */
4821                         size += rdev->size;
4822                 }
4823
4824                 if (!list_empty(&mddev->disks)) {
4825                         if (mddev->pers)
4826                                 seq_printf(seq, "\n      %llu blocks",
4827                                         (unsigned long long)mddev->array_size);
4828                         else
4829                                 seq_printf(seq, "\n      %llu blocks",
4830                                         (unsigned long long)size);
4831                 }
4832                 if (mddev->persistent) {
4833                         if (mddev->major_version != 0 ||
4834                             mddev->minor_version != 90) {
4835                                 seq_printf(seq," super %d.%d",
4836                                            mddev->major_version,
4837                                            mddev->minor_version);
4838                         }
4839                 } else
4840                         seq_printf(seq, " super non-persistent");
4841
4842                 if (mddev->pers) {
4843                         mddev->pers->status (seq, mddev);
4844                         seq_printf(seq, "\n      ");
4845                         if (mddev->pers->sync_request) {
4846                                 if (mddev->curr_resync > 2) {
4847                                         status_resync (seq, mddev);
4848                                         seq_printf(seq, "\n      ");
4849                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4850                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4851                                 else if (mddev->recovery_cp < MaxSector)
4852                                         seq_printf(seq, "\tresync=PENDING\n      ");
4853                         }
4854                 } else
4855                         seq_printf(seq, "\n       ");
4856
4857                 if ((bitmap = mddev->bitmap)) {
4858                         unsigned long chunk_kb;
4859                         unsigned long flags;
4860                         spin_lock_irqsave(&bitmap->lock, flags);
4861                         chunk_kb = bitmap->chunksize >> 10;
4862                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4863                                 "%lu%s chunk",
4864                                 bitmap->pages - bitmap->missing_pages,
4865                                 bitmap->pages,
4866                                 (bitmap->pages - bitmap->missing_pages)
4867                                         << (PAGE_SHIFT - 10),
4868                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4869                                 chunk_kb ? "KB" : "B");
4870                         if (bitmap->file) {
4871                                 seq_printf(seq, ", file: ");
4872                                 seq_path(seq, bitmap->file->f_vfsmnt,
4873                                          bitmap->file->f_dentry," \t\n");
4874                         }
4875
4876                         seq_printf(seq, "\n");
4877                         spin_unlock_irqrestore(&bitmap->lock, flags);
4878                 }
4879
4880                 seq_printf(seq, "\n");
4881         }
4882         mddev_unlock(mddev);
4883         
4884         return 0;
4885 }
4886
4887 static struct seq_operations md_seq_ops = {
4888         .start  = md_seq_start,
4889         .next   = md_seq_next,
4890         .stop   = md_seq_stop,
4891         .show   = md_seq_show,
4892 };
4893
4894 static int md_seq_open(struct inode *inode, struct file *file)
4895 {
4896         int error;
4897         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4898         if (mi == NULL)
4899                 return -ENOMEM;
4900
4901         error = seq_open(file, &md_seq_ops);
4902         if (error)
4903                 kfree(mi);
4904         else {
4905                 struct seq_file *p = file->private_data;
4906                 p->private = mi;
4907                 mi->event = atomic_read(&md_event_count);
4908         }
4909         return error;
4910 }
4911
4912 static int md_seq_release(struct inode *inode, struct file *file)
4913 {
4914         struct seq_file *m = file->private_data;
4915         struct mdstat_info *mi = m->private;
4916         m->private = NULL;
4917         kfree(mi);
4918         return seq_release(inode, file);
4919 }
4920
4921 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4922 {
4923         struct seq_file *m = filp->private_data;
4924         struct mdstat_info *mi = m->private;
4925         int mask;
4926
4927         poll_wait(filp, &md_event_waiters, wait);
4928
4929         /* always allow read */
4930         mask = POLLIN | POLLRDNORM;
4931
4932         if (mi->event != atomic_read(&md_event_count))
4933                 mask |= POLLERR | POLLPRI;
4934         return mask;
4935 }
4936
4937 static struct file_operations md_seq_fops = {
4938         .open           = md_seq_open,
4939         .read           = seq_read,
4940         .llseek         = seq_lseek,
4941         .release        = md_seq_release,
4942         .poll           = mdstat_poll,
4943 };
4944
4945 int register_md_personality(struct mdk_personality *p)
4946 {
4947         spin_lock(&pers_lock);
4948         list_add_tail(&p->list, &pers_list);
4949         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4950         spin_unlock(&pers_lock);
4951         return 0;
4952 }
4953
4954 int unregister_md_personality(struct mdk_personality *p)
4955 {
4956         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4957         spin_lock(&pers_lock);
4958         list_del_init(&p->list);
4959         spin_unlock(&pers_lock);
4960         return 0;
4961 }
4962
4963 static int is_mddev_idle(mddev_t *mddev)
4964 {
4965         mdk_rdev_t * rdev;
4966         struct list_head *tmp;
4967         int idle;
4968         unsigned long curr_events;
4969
4970         idle = 1;
4971         ITERATE_RDEV(mddev,rdev,tmp) {
4972                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4973                 curr_events = disk_stat_read(disk, sectors[0]) + 
4974                                 disk_stat_read(disk, sectors[1]) - 
4975                                 atomic_read(&disk->sync_io);
4976                 /* The difference between curr_events and last_events
4977                  * will be affected by any new non-sync IO (making
4978                  * curr_events bigger) and any difference in the amount of
4979                  * in-flight syncio (making current_events bigger or smaller)
4980                  * The amount in-flight is currently limited to
4981                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4982                  * which is at most 4096 sectors.
4983                  * These numbers are fairly fragile and should be made
4984                  * more robust, probably by enforcing the
4985                  * 'window size' that md_do_sync sort-of uses.
4986                  *
4987                  * Note: the following is an unsigned comparison.
4988                  */
4989                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4990                         rdev->last_events = curr_events;
4991                         idle = 0;
4992                 }
4993         }
4994         return idle;
4995 }
4996
4997 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4998 {
4999         /* another "blocks" (512byte) blocks have been synced */
5000         atomic_sub(blocks, &mddev->recovery_active);
5001         wake_up(&mddev->recovery_wait);
5002         if (!ok) {
5003                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5004                 md_wakeup_thread(mddev->thread);
5005                 // stop recovery, signal do_sync ....
5006         }
5007 }
5008
5009
5010 /* md_write_start(mddev, bi)
5011  * If we need to update some array metadata (e.g. 'active' flag
5012  * in superblock) before writing, schedule a superblock update
5013  * and wait for it to complete.
5014  */
5015 void md_write_start(mddev_t *mddev, struct bio *bi)
5016 {
5017         if (bio_data_dir(bi) != WRITE)
5018                 return;
5019
5020         BUG_ON(mddev->ro == 1);
5021         if (mddev->ro == 2) {
5022                 /* need to switch to read/write */
5023                 mddev->ro = 0;
5024                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5025                 md_wakeup_thread(mddev->thread);
5026         }
5027         atomic_inc(&mddev->writes_pending);
5028         if (mddev->in_sync) {
5029                 spin_lock_irq(&mddev->write_lock);
5030                 if (mddev->in_sync) {
5031                         mddev->in_sync = 0;
5032                         mddev->sb_dirty = 3;
5033                         md_wakeup_thread(mddev->thread);
5034                 }
5035                 spin_unlock_irq(&mddev->write_lock);
5036         }
5037         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
5038 }
5039
5040 void md_write_end(mddev_t *mddev)
5041 {
5042         if (atomic_dec_and_test(&mddev->writes_pending)) {
5043                 if (mddev->safemode == 2)
5044                         md_wakeup_thread(mddev->thread);
5045                 else if (mddev->safemode_delay)
5046                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5047         }
5048 }
5049
5050 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5051
5052 #define SYNC_MARKS      10
5053 #define SYNC_MARK_STEP  (3*HZ)
5054 void md_do_sync(mddev_t *mddev)
5055 {
5056         mddev_t *mddev2;
5057         unsigned int currspeed = 0,
5058                  window;
5059         sector_t max_sectors,j, io_sectors;
5060         unsigned long mark[SYNC_MARKS];
5061         sector_t mark_cnt[SYNC_MARKS];
5062         int last_mark,m;
5063         struct list_head *tmp;
5064         sector_t last_check;
5065         int skipped = 0;
5066         struct list_head *rtmp;
5067         mdk_rdev_t *rdev;
5068
5069         /* just incase thread restarts... */
5070         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5071                 return;
5072         if (mddev->ro) /* never try to sync a read-only array */
5073                 return;
5074
5075         /* we overload curr_resync somewhat here.
5076          * 0 == not engaged in resync at all
5077          * 2 == checking that there is no conflict with another sync
5078          * 1 == like 2, but have yielded to allow conflicting resync to
5079          *              commense
5080          * other == active in resync - this many blocks
5081          *
5082          * Before starting a resync we must have set curr_resync to
5083          * 2, and then checked that every "conflicting" array has curr_resync
5084          * less than ours.  When we find one that is the same or higher
5085          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5086          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5087          * This will mean we have to start checking from the beginning again.
5088          *
5089          */
5090
5091         do {
5092                 mddev->curr_resync = 2;
5093
5094         try_again:
5095                 if (kthread_should_stop()) {
5096                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5097                         goto skip;
5098                 }
5099                 ITERATE_MDDEV(mddev2,tmp) {
5100                         if (mddev2 == mddev)
5101                                 continue;
5102                         if (mddev2->curr_resync && 
5103                             match_mddev_units(mddev,mddev2)) {
5104                                 DEFINE_WAIT(wq);
5105                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5106                                         /* arbitrarily yield */
5107                                         mddev->curr_resync = 1;
5108                                         wake_up(&resync_wait);
5109                                 }
5110                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5111                                         /* no need to wait here, we can wait the next
5112                                          * time 'round when curr_resync == 2
5113                                          */
5114                                         continue;
5115                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5116                                 if (!kthread_should_stop() &&
5117                                     mddev2->curr_resync >= mddev->curr_resync) {
5118                                         printk(KERN_INFO "md: delaying resync of %s"
5119                                                " until %s has finished resync (they"
5120                                                " share one or more physical units)\n",
5121                                                mdname(mddev), mdname(mddev2));
5122                                         mddev_put(mddev2);
5123                                         schedule();
5124                                         finish_wait(&resync_wait, &wq);
5125                                         goto try_again;
5126                                 }
5127                                 finish_wait(&resync_wait, &wq);
5128                         }
5129                 }
5130         } while (mddev->curr_resync < 2);
5131
5132         j = 0;
5133         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5134                 /* resync follows the size requested by the personality,
5135                  * which defaults to physical size, but can be virtual size
5136                  */
5137                 max_sectors = mddev->resync_max_sectors;
5138                 mddev->resync_mismatches = 0;
5139                 /* we don't use the checkpoint if there's a bitmap */
5140                 if (!mddev->bitmap &&
5141                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5142                         j = mddev->recovery_cp;
5143         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5144                 max_sectors = mddev->size << 1;
5145         else {
5146                 /* recovery follows the physical size of devices */
5147                 max_sectors = mddev->size << 1;
5148                 j = MaxSector;
5149                 ITERATE_RDEV(mddev,rdev,rtmp)
5150                         if (rdev->raid_disk >= 0 &&
5151                             !test_bit(Faulty, &rdev->flags) &&
5152                             !test_bit(In_sync, &rdev->flags) &&
5153                             rdev->recovery_offset < j)
5154                                 j = rdev->recovery_offset;
5155         }
5156
5157         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
5158         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
5159                 " %d KB/sec/disc.\n", speed_min(mddev));
5160         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5161                "(but not more than %d KB/sec) for reconstruction.\n",
5162                speed_max(mddev));
5163
5164         is_mddev_idle(mddev); /* this also initializes IO event counters */
5165
5166         io_sectors = 0;
5167         for (m = 0; m < SYNC_MARKS; m++) {
5168                 mark[m] = jiffies;
5169                 mark_cnt[m] = io_sectors;
5170         }
5171         last_mark = 0;
5172         mddev->resync_mark = mark[last_mark];
5173         mddev->resync_mark_cnt = mark_cnt[last_mark];
5174
5175         /*
5176          * Tune reconstruction:
5177          */
5178         window = 32*(PAGE_SIZE/512);
5179         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5180                 window/2,(unsigned long long) max_sectors/2);
5181
5182         atomic_set(&mddev->recovery_active, 0);
5183         init_waitqueue_head(&mddev->recovery_wait);
5184         last_check = 0;
5185
5186         if (j>2) {
5187                 printk(KERN_INFO 
5188                         "md: resuming recovery of %s from checkpoint.\n",
5189                         mdname(mddev));
5190                 mddev->curr_resync = j;
5191         }
5192
5193         while (j < max_sectors) {
5194                 sector_t sectors;
5195
5196                 skipped = 0;
5197                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5198                                             currspeed < speed_min(mddev));
5199                 if (sectors == 0) {
5200                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5201                         goto out;
5202                 }
5203
5204                 if (!skipped) { /* actual IO requested */
5205                         io_sectors += sectors;
5206                         atomic_add(sectors, &mddev->recovery_active);
5207                 }
5208
5209                 j += sectors;
5210                 if (j>1) mddev->curr_resync = j;
5211                 mddev->curr_mark_cnt = io_sectors;
5212                 if (last_check == 0)
5213                         /* this is the earliers that rebuilt will be
5214                          * visible in /proc/mdstat
5215                          */
5216                         md_new_event(mddev);
5217
5218                 if (last_check + window > io_sectors || j == max_sectors)
5219                         continue;
5220
5221                 last_check = io_sectors;
5222
5223                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5224                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5225                         break;
5226
5227         repeat:
5228                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5229                         /* step marks */
5230                         int next = (last_mark+1) % SYNC_MARKS;
5231
5232                         mddev->resync_mark = mark[next];
5233                         mddev->resync_mark_cnt = mark_cnt[next];
5234                         mark[next] = jiffies;
5235                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5236                         last_mark = next;
5237                 }
5238
5239
5240                 if (kthread_should_stop()) {
5241                         /*
5242                          * got a signal, exit.
5243                          */
5244                         printk(KERN_INFO 
5245                                 "md: md_do_sync() got signal ... exiting\n");
5246                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5247                         goto out;
5248                 }
5249
5250                 /*
5251                  * this loop exits only if either when we are slower than
5252                  * the 'hard' speed limit, or the system was IO-idle for
5253                  * a jiffy.
5254                  * the system might be non-idle CPU-wise, but we only care
5255                  * about not overloading the IO subsystem. (things like an
5256                  * e2fsck being done on the RAID array should execute fast)
5257                  */
5258                 mddev->queue->unplug_fn(mddev->queue);
5259                 cond_resched();
5260
5261                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5262                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5263
5264                 if (currspeed > speed_min(mddev)) {
5265                         if ((currspeed > speed_max(mddev)) ||
5266                                         !is_mddev_idle(mddev)) {
5267                                 msleep(500);
5268                                 goto repeat;
5269                         }
5270                 }
5271         }
5272         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
5273         /*
5274          * this also signals 'finished resyncing' to md_stop
5275          */
5276  out:
5277         mddev->queue->unplug_fn(mddev->queue);
5278
5279         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5280
5281         /* tell personality that we are finished */
5282         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5283
5284         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5285             test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
5286             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5287             mddev->curr_resync > 2) {
5288                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5289                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5290                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5291                                         printk(KERN_INFO
5292                                                "md: checkpointing recovery of %s.\n",
5293                                                mdname(mddev));
5294                                         mddev->recovery_cp = mddev->curr_resync;
5295                                 }
5296                         } else
5297                                 mddev->recovery_cp = MaxSector;
5298                 } else {
5299                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5300                                 mddev->curr_resync = MaxSector;
5301                         ITERATE_RDEV(mddev,rdev,rtmp)
5302                                 if (rdev->raid_disk >= 0 &&
5303                                     !test_bit(Faulty, &rdev->flags) &&
5304                                     !test_bit(In_sync, &rdev->flags) &&
5305                                     rdev->recovery_offset < mddev->curr_resync)
5306                                         rdev->recovery_offset = mddev->curr_resync;
5307                         mddev->sb_dirty = 1;
5308                 }
5309         }
5310
5311  skip:
5312         mddev->curr_resync = 0;
5313         wake_up(&resync_wait);
5314         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5315         md_wakeup_thread(mddev->thread);
5316 }
5317 EXPORT_SYMBOL_GPL(md_do_sync);
5318
5319
5320 /*
5321  * This routine is regularly called by all per-raid-array threads to
5322  * deal with generic issues like resync and super-block update.
5323  * Raid personalities that don't have a thread (linear/raid0) do not
5324  * need this as they never do any recovery or update the superblock.
5325  *
5326  * It does not do any resync itself, but rather "forks" off other threads
5327  * to do that as needed.
5328  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5329  * "->recovery" and create a thread at ->sync_thread.
5330  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5331  * and wakeups up this thread which will reap the thread and finish up.
5332  * This thread also removes any faulty devices (with nr_pending == 0).
5333  *
5334  * The overall approach is:
5335  *  1/ if the superblock needs updating, update it.
5336  *  2/ If a recovery thread is running, don't do anything else.
5337  *  3/ If recovery has finished, clean up, possibly marking spares active.
5338  *  4/ If there are any faulty devices, remove them.
5339  *  5/ If array is degraded, try to add spares devices
5340  *  6/ If array has spares or is not in-sync, start a resync thread.
5341  */
5342 void md_check_recovery(mddev_t *mddev)
5343 {
5344         mdk_rdev_t *rdev;
5345         struct list_head *rtmp;
5346
5347
5348         if (mddev->bitmap)
5349                 bitmap_daemon_work(mddev->bitmap);
5350
5351         if (mddev->ro)
5352                 return;
5353
5354         if (signal_pending(current)) {
5355                 if (mddev->pers->sync_request) {
5356                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5357                                mdname(mddev));
5358                         mddev->safemode = 2;
5359                 }
5360                 flush_signals(current);
5361         }
5362
5363         if ( ! (
5364                 mddev->sb_dirty ||
5365                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5366                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5367                 (mddev->safemode == 1) ||
5368                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5369                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5370                 ))
5371                 return;
5372
5373         if (mddev_trylock(mddev)) {
5374                 int spares =0;
5375
5376                 spin_lock_irq(&mddev->write_lock);
5377                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5378                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5379                         mddev->in_sync = 1;
5380                         mddev->sb_dirty = 3;
5381                 }
5382                 if (mddev->safemode == 1)
5383                         mddev->safemode = 0;
5384                 spin_unlock_irq(&mddev->write_lock);
5385
5386                 if (mddev->sb_dirty)
5387                         md_update_sb(mddev);
5388
5389
5390                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5391                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5392                         /* resync/recovery still happening */
5393                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5394                         goto unlock;
5395                 }
5396                 if (mddev->sync_thread) {
5397                         /* resync has finished, collect result */
5398                         md_unregister_thread(mddev->sync_thread);
5399                         mddev->sync_thread = NULL;
5400                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5401                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5402                                 /* success...*/
5403                                 /* activate any spares */
5404                                 mddev->pers->spare_active(mddev);
5405                         }
5406                         md_update_sb(mddev);
5407
5408                         /* if array is no-longer degraded, then any saved_raid_disk
5409                          * information must be scrapped
5410                          */
5411                         if (!mddev->degraded)
5412                                 ITERATE_RDEV(mddev,rdev,rtmp)
5413                                         rdev->saved_raid_disk = -1;
5414
5415                         mddev->recovery = 0;
5416                         /* flag recovery needed just to double check */
5417                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5418                         md_new_event(mddev);
5419                         goto unlock;
5420                 }
5421                 /* Clear some bits that don't mean anything, but
5422                  * might be left set
5423                  */
5424                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5425                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5426                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5427                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5428
5429                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5430                         goto unlock;
5431                 /* no recovery is running.
5432                  * remove any failed drives, then
5433                  * add spares if possible.
5434                  * Spare are also removed and re-added, to allow
5435                  * the personality to fail the re-add.
5436                  */
5437                 ITERATE_RDEV(mddev,rdev,rtmp)
5438                         if (rdev->raid_disk >= 0 &&
5439                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
5440                             atomic_read(&rdev->nr_pending)==0) {
5441                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5442                                         char nm[20];
5443                                         sprintf(nm,"rd%d", rdev->raid_disk);
5444                                         sysfs_remove_link(&mddev->kobj, nm);
5445                                         rdev->raid_disk = -1;
5446                                 }
5447                         }
5448
5449                 if (mddev->degraded) {
5450                         ITERATE_RDEV(mddev,rdev,rtmp)
5451                                 if (rdev->raid_disk < 0
5452                                     && !test_bit(Faulty, &rdev->flags)) {
5453                                         rdev->recovery_offset = 0;
5454                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
5455                                                 char nm[20];
5456                                                 sprintf(nm, "rd%d", rdev->raid_disk);
5457                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
5458                                                 spares++;
5459                                                 md_new_event(mddev);
5460                                         } else
5461                                                 break;
5462                                 }
5463                 }
5464
5465                 if (spares) {
5466                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5467                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5468                 } else if (mddev->recovery_cp < MaxSector) {
5469                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5470                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5471                         /* nothing to be done ... */
5472                         goto unlock;
5473
5474                 if (mddev->pers->sync_request) {
5475                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5476                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5477                                 /* We are adding a device or devices to an array
5478                                  * which has the bitmap stored on all devices.
5479                                  * So make sure all bitmap pages get written
5480                                  */
5481                                 bitmap_write_all(mddev->bitmap);
5482                         }
5483                         mddev->sync_thread = md_register_thread(md_do_sync,
5484                                                                 mddev,
5485                                                                 "%s_resync");
5486                         if (!mddev->sync_thread) {
5487                                 printk(KERN_ERR "%s: could not start resync"
5488                                         " thread...\n", 
5489                                         mdname(mddev));
5490                                 /* leave the spares where they are, it shouldn't hurt */
5491                                 mddev->recovery = 0;
5492                         } else
5493                                 md_wakeup_thread(mddev->sync_thread);
5494                         md_new_event(mddev);
5495                 }
5496         unlock:
5497                 mddev_unlock(mddev);
5498         }
5499 }
5500
5501 static int md_notify_reboot(struct notifier_block *this,
5502                             unsigned long code, void *x)
5503 {
5504         struct list_head *tmp;
5505         mddev_t *mddev;
5506
5507         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5508
5509                 printk(KERN_INFO "md: stopping all md devices.\n");
5510
5511                 ITERATE_MDDEV(mddev,tmp)
5512                         if (mddev_trylock(mddev)) {
5513                                 do_md_stop (mddev, 1);
5514                                 mddev_unlock(mddev);
5515                         }
5516                 /*
5517                  * certain more exotic SCSI devices are known to be
5518                  * volatile wrt too early system reboots. While the
5519                  * right place to handle this issue is the given
5520                  * driver, we do want to have a safe RAID driver ...
5521                  */
5522                 mdelay(1000*1);
5523         }
5524         return NOTIFY_DONE;
5525 }
5526
5527 static struct notifier_block md_notifier = {
5528         .notifier_call  = md_notify_reboot,
5529         .next           = NULL,
5530         .priority       = INT_MAX, /* before any real devices */
5531 };
5532
5533 static void md_geninit(void)
5534 {
5535         struct proc_dir_entry *p;
5536
5537         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5538
5539         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5540         if (p)
5541                 p->proc_fops = &md_seq_fops;
5542 }
5543
5544 static int __init md_init(void)
5545 {
5546         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5547                         " MD_SB_DISKS=%d\n",
5548                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
5549                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
5550         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
5551                         BITMAP_MINOR);
5552
5553         if (register_blkdev(MAJOR_NR, "md"))
5554                 return -1;
5555         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5556                 unregister_blkdev(MAJOR_NR, "md");
5557                 return -1;
5558         }
5559         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5560                                 md_probe, NULL, NULL);
5561         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5562                             md_probe, NULL, NULL);
5563
5564         register_reboot_notifier(&md_notifier);
5565         raid_table_header = register_sysctl_table(raid_root_table, 1);
5566
5567         md_geninit();
5568         return (0);
5569 }
5570
5571
5572 #ifndef MODULE
5573
5574 /*
5575  * Searches all registered partitions for autorun RAID arrays
5576  * at boot time.
5577  */
5578 static dev_t detected_devices[128];
5579 static int dev_cnt;
5580
5581 void md_autodetect_dev(dev_t dev)
5582 {
5583         if (dev_cnt >= 0 && dev_cnt < 127)
5584                 detected_devices[dev_cnt++] = dev;
5585 }
5586
5587
5588 static void autostart_arrays(int part)
5589 {
5590         mdk_rdev_t *rdev;
5591         int i;
5592
5593         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5594
5595         for (i = 0; i < dev_cnt; i++) {
5596                 dev_t dev = detected_devices[i];
5597
5598                 rdev = md_import_device(dev,0, 0);
5599                 if (IS_ERR(rdev))
5600                         continue;
5601
5602                 if (test_bit(Faulty, &rdev->flags)) {
5603                         MD_BUG();
5604                         continue;
5605                 }
5606                 list_add(&rdev->same_set, &pending_raid_disks);
5607         }
5608         dev_cnt = 0;
5609
5610         autorun_devices(part);
5611 }
5612
5613 #endif
5614
5615 static __exit void md_exit(void)
5616 {
5617         mddev_t *mddev;
5618         struct list_head *tmp;
5619
5620         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5621         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
5622
5623         unregister_blkdev(MAJOR_NR,"md");
5624         unregister_blkdev(mdp_major, "mdp");
5625         unregister_reboot_notifier(&md_notifier);
5626         unregister_sysctl_table(raid_table_header);
5627         remove_proc_entry("mdstat", NULL);
5628         ITERATE_MDDEV(mddev,tmp) {
5629                 struct gendisk *disk = mddev->gendisk;
5630                 if (!disk)
5631                         continue;
5632                 export_array(mddev);
5633                 del_gendisk(disk);
5634                 put_disk(disk);
5635                 mddev->gendisk = NULL;
5636                 mddev_put(mddev);
5637         }
5638 }
5639
5640 module_init(md_init)
5641 module_exit(md_exit)
5642
5643 static int get_ro(char *buffer, struct kernel_param *kp)
5644 {
5645         return sprintf(buffer, "%d", start_readonly);
5646 }
5647 static int set_ro(const char *val, struct kernel_param *kp)
5648 {
5649         char *e;
5650         int num = simple_strtoul(val, &e, 10);
5651         if (*val && (*e == '\0' || *e == '\n')) {
5652                 start_readonly = num;
5653                 return 0;
5654         }
5655         return -EINVAL;
5656 }
5657
5658 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5659 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5660
5661
5662 EXPORT_SYMBOL(register_md_personality);
5663 EXPORT_SYMBOL(unregister_md_personality);
5664 EXPORT_SYMBOL(md_error);
5665 EXPORT_SYMBOL(md_done_sync);
5666 EXPORT_SYMBOL(md_write_start);
5667 EXPORT_SYMBOL(md_write_end);
5668 EXPORT_SYMBOL(md_register_thread);
5669 EXPORT_SYMBOL(md_unregister_thread);
5670 EXPORT_SYMBOL(md_wakeup_thread);
5671 EXPORT_SYMBOL(md_check_recovery);
5672 MODULE_LICENSE("GPL");
5673 MODULE_ALIAS("md");
5674 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);