Merge commit 'v2.6.32-rc7'
[linux-2.6.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72  * is 1000 KB/sec, so the extra system load does not show up that much.
73  * Increase it if you want to have more _guaranteed_ speed. Note that
74  * the RAID driver will use the maximum available bandwidth if the IO
75  * subsystem is idle. There is also an 'absolute maximum' reconstruction
76  * speed limit - in case reconstruction slows down your system despite
77  * idle IO detection.
78  *
79  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80  * or /sys/block/mdX/md/sync_speed_{min,max}
81  */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87         return mddev->sync_speed_min ?
88                 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93         return mddev->sync_speed_max ?
94                 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100         {
101                 .procname       = "speed_limit_min",
102                 .data           = &sysctl_speed_limit_min,
103                 .maxlen         = sizeof(int),
104                 .mode           = S_IRUGO|S_IWUSR,
105                 .proc_handler   = &proc_dointvec,
106         },
107         {
108                 .procname       = "speed_limit_max",
109                 .data           = &sysctl_speed_limit_max,
110                 .maxlen         = sizeof(int),
111                 .mode           = S_IRUGO|S_IWUSR,
112                 .proc_handler   = &proc_dointvec,
113         },
114         { }
115 };
116
117 static ctl_table raid_dir_table[] = {
118         {
119                 .procname       = "raid",
120                 .maxlen         = 0,
121                 .mode           = S_IRUGO|S_IXUGO,
122                 .child          = raid_table,
123         },
124         { }
125 };
126
127 static ctl_table raid_root_table[] = {
128         {
129                 .procname       = "dev",
130                 .maxlen         = 0,
131                 .mode           = 0555,
132                 .child          = raid_dir_table,
133         },
134         {  }
135 };
136
137 static const struct block_device_operations md_fops;
138
139 static int start_readonly;
140
141 /*
142  * We have a system wide 'event count' that is incremented
143  * on any 'interesting' event, and readers of /proc/mdstat
144  * can use 'poll' or 'select' to find out when the event
145  * count increases.
146  *
147  * Events are:
148  *  start array, stop array, error, add device, remove device,
149  *  start build, activate spare
150  */
151 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
152 static atomic_t md_event_count;
153 void md_new_event(mddev_t *mddev)
154 {
155         atomic_inc(&md_event_count);
156         wake_up(&md_event_waiters);
157 }
158 EXPORT_SYMBOL_GPL(md_new_event);
159
160 /* Alternate version that can be called from interrupts
161  * when calling sysfs_notify isn't needed.
162  */
163 static void md_new_event_inintr(mddev_t *mddev)
164 {
165         atomic_inc(&md_event_count);
166         wake_up(&md_event_waiters);
167 }
168
169 /*
170  * Enables to iterate over all existing md arrays
171  * all_mddevs_lock protects this list.
172  */
173 static LIST_HEAD(all_mddevs);
174 static DEFINE_SPINLOCK(all_mddevs_lock);
175
176
177 /*
178  * iterates through all used mddevs in the system.
179  * We take care to grab the all_mddevs_lock whenever navigating
180  * the list, and to always hold a refcount when unlocked.
181  * Any code which breaks out of this loop while own
182  * a reference to the current mddev and must mddev_put it.
183  */
184 #define for_each_mddev(mddev,tmp)                                       \
185                                                                         \
186         for (({ spin_lock(&all_mddevs_lock);                            \
187                 tmp = all_mddevs.next;                                  \
188                 mddev = NULL;});                                        \
189              ({ if (tmp != &all_mddevs)                                 \
190                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
191                 spin_unlock(&all_mddevs_lock);                          \
192                 if (mddev) mddev_put(mddev);                            \
193                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
194                 tmp != &all_mddevs;});                                  \
195              ({ spin_lock(&all_mddevs_lock);                            \
196                 tmp = tmp->next;})                                      \
197                 )
198
199
200 /* Rather than calling directly into the personality make_request function,
201  * IO requests come here first so that we can check if the device is
202  * being suspended pending a reconfiguration.
203  * We hold a refcount over the call to ->make_request.  By the time that
204  * call has finished, the bio has been linked into some internal structure
205  * and so is visible to ->quiesce(), so we don't need the refcount any more.
206  */
207 static int md_make_request(struct request_queue *q, struct bio *bio)
208 {
209         mddev_t *mddev = q->queuedata;
210         int rv;
211         if (mddev == NULL || mddev->pers == NULL) {
212                 bio_io_error(bio);
213                 return 0;
214         }
215         rcu_read_lock();
216         if (mddev->suspended) {
217                 DEFINE_WAIT(__wait);
218                 for (;;) {
219                         prepare_to_wait(&mddev->sb_wait, &__wait,
220                                         TASK_UNINTERRUPTIBLE);
221                         if (!mddev->suspended)
222                                 break;
223                         rcu_read_unlock();
224                         schedule();
225                         rcu_read_lock();
226                 }
227                 finish_wait(&mddev->sb_wait, &__wait);
228         }
229         atomic_inc(&mddev->active_io);
230         rcu_read_unlock();
231         rv = mddev->pers->make_request(q, bio);
232         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
233                 wake_up(&mddev->sb_wait);
234
235         return rv;
236 }
237
238 static void mddev_suspend(mddev_t *mddev)
239 {
240         BUG_ON(mddev->suspended);
241         mddev->suspended = 1;
242         synchronize_rcu();
243         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
244         mddev->pers->quiesce(mddev, 1);
245         md_unregister_thread(mddev->thread);
246         mddev->thread = NULL;
247         /* we now know that no code is executing in the personality module,
248          * except possibly the tail end of a ->bi_end_io function, but that
249          * is certain to complete before the module has a chance to get
250          * unloaded
251          */
252 }
253
254 static void mddev_resume(mddev_t *mddev)
255 {
256         mddev->suspended = 0;
257         wake_up(&mddev->sb_wait);
258         mddev->pers->quiesce(mddev, 0);
259 }
260
261 int mddev_congested(mddev_t *mddev, int bits)
262 {
263         return mddev->suspended;
264 }
265 EXPORT_SYMBOL(mddev_congested);
266
267
268 static inline mddev_t *mddev_get(mddev_t *mddev)
269 {
270         atomic_inc(&mddev->active);
271         return mddev;
272 }
273
274 static void mddev_delayed_delete(struct work_struct *ws);
275
276 static void mddev_put(mddev_t *mddev)
277 {
278         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
279                 return;
280         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
281             !mddev->hold_active) {
282                 list_del(&mddev->all_mddevs);
283                 if (mddev->gendisk) {
284                         /* we did a probe so need to clean up.
285                          * Call schedule_work inside the spinlock
286                          * so that flush_scheduled_work() after
287                          * mddev_find will succeed in waiting for the
288                          * work to be done.
289                          */
290                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
291                         schedule_work(&mddev->del_work);
292                 } else
293                         kfree(mddev);
294         }
295         spin_unlock(&all_mddevs_lock);
296 }
297
298 static mddev_t * mddev_find(dev_t unit)
299 {
300         mddev_t *mddev, *new = NULL;
301
302  retry:
303         spin_lock(&all_mddevs_lock);
304
305         if (unit) {
306                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
307                         if (mddev->unit == unit) {
308                                 mddev_get(mddev);
309                                 spin_unlock(&all_mddevs_lock);
310                                 kfree(new);
311                                 return mddev;
312                         }
313
314                 if (new) {
315                         list_add(&new->all_mddevs, &all_mddevs);
316                         spin_unlock(&all_mddevs_lock);
317                         new->hold_active = UNTIL_IOCTL;
318                         return new;
319                 }
320         } else if (new) {
321                 /* find an unused unit number */
322                 static int next_minor = 512;
323                 int start = next_minor;
324                 int is_free = 0;
325                 int dev = 0;
326                 while (!is_free) {
327                         dev = MKDEV(MD_MAJOR, next_minor);
328                         next_minor++;
329                         if (next_minor > MINORMASK)
330                                 next_minor = 0;
331                         if (next_minor == start) {
332                                 /* Oh dear, all in use. */
333                                 spin_unlock(&all_mddevs_lock);
334                                 kfree(new);
335                                 return NULL;
336                         }
337                                 
338                         is_free = 1;
339                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
340                                 if (mddev->unit == dev) {
341                                         is_free = 0;
342                                         break;
343                                 }
344                 }
345                 new->unit = dev;
346                 new->md_minor = MINOR(dev);
347                 new->hold_active = UNTIL_STOP;
348                 list_add(&new->all_mddevs, &all_mddevs);
349                 spin_unlock(&all_mddevs_lock);
350                 return new;
351         }
352         spin_unlock(&all_mddevs_lock);
353
354         new = kzalloc(sizeof(*new), GFP_KERNEL);
355         if (!new)
356                 return NULL;
357
358         new->unit = unit;
359         if (MAJOR(unit) == MD_MAJOR)
360                 new->md_minor = MINOR(unit);
361         else
362                 new->md_minor = MINOR(unit) >> MdpMinorShift;
363
364         mutex_init(&new->open_mutex);
365         mutex_init(&new->reconfig_mutex);
366         INIT_LIST_HEAD(&new->disks);
367         INIT_LIST_HEAD(&new->all_mddevs);
368         init_timer(&new->safemode_timer);
369         atomic_set(&new->active, 1);
370         atomic_set(&new->openers, 0);
371         atomic_set(&new->active_io, 0);
372         spin_lock_init(&new->write_lock);
373         init_waitqueue_head(&new->sb_wait);
374         init_waitqueue_head(&new->recovery_wait);
375         new->reshape_position = MaxSector;
376         new->resync_min = 0;
377         new->resync_max = MaxSector;
378         new->level = LEVEL_NONE;
379
380         goto retry;
381 }
382
383 static inline int mddev_lock(mddev_t * mddev)
384 {
385         return mutex_lock_interruptible(&mddev->reconfig_mutex);
386 }
387
388 static inline int mddev_is_locked(mddev_t *mddev)
389 {
390         return mutex_is_locked(&mddev->reconfig_mutex);
391 }
392
393 static inline int mddev_trylock(mddev_t * mddev)
394 {
395         return mutex_trylock(&mddev->reconfig_mutex);
396 }
397
398 static inline void mddev_unlock(mddev_t * mddev)
399 {
400         mutex_unlock(&mddev->reconfig_mutex);
401
402         md_wakeup_thread(mddev->thread);
403 }
404
405 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
406 {
407         mdk_rdev_t *rdev;
408
409         list_for_each_entry(rdev, &mddev->disks, same_set)
410                 if (rdev->desc_nr == nr)
411                         return rdev;
412
413         return NULL;
414 }
415
416 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
417 {
418         mdk_rdev_t *rdev;
419
420         list_for_each_entry(rdev, &mddev->disks, same_set)
421                 if (rdev->bdev->bd_dev == dev)
422                         return rdev;
423
424         return NULL;
425 }
426
427 static struct mdk_personality *find_pers(int level, char *clevel)
428 {
429         struct mdk_personality *pers;
430         list_for_each_entry(pers, &pers_list, list) {
431                 if (level != LEVEL_NONE && pers->level == level)
432                         return pers;
433                 if (strcmp(pers->name, clevel)==0)
434                         return pers;
435         }
436         return NULL;
437 }
438
439 /* return the offset of the super block in 512byte sectors */
440 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
441 {
442         sector_t num_sectors = bdev->bd_inode->i_size / 512;
443         return MD_NEW_SIZE_SECTORS(num_sectors);
444 }
445
446 static int alloc_disk_sb(mdk_rdev_t * rdev)
447 {
448         if (rdev->sb_page)
449                 MD_BUG();
450
451         rdev->sb_page = alloc_page(GFP_KERNEL);
452         if (!rdev->sb_page) {
453                 printk(KERN_ALERT "md: out of memory.\n");
454                 return -ENOMEM;
455         }
456
457         return 0;
458 }
459
460 static void free_disk_sb(mdk_rdev_t * rdev)
461 {
462         if (rdev->sb_page) {
463                 put_page(rdev->sb_page);
464                 rdev->sb_loaded = 0;
465                 rdev->sb_page = NULL;
466                 rdev->sb_start = 0;
467                 rdev->sectors = 0;
468         }
469 }
470
471
472 static void super_written(struct bio *bio, int error)
473 {
474         mdk_rdev_t *rdev = bio->bi_private;
475         mddev_t *mddev = rdev->mddev;
476
477         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
478                 printk("md: super_written gets error=%d, uptodate=%d\n",
479                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
480                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
481                 md_error(mddev, rdev);
482         }
483
484         if (atomic_dec_and_test(&mddev->pending_writes))
485                 wake_up(&mddev->sb_wait);
486         bio_put(bio);
487 }
488
489 static void super_written_barrier(struct bio *bio, int error)
490 {
491         struct bio *bio2 = bio->bi_private;
492         mdk_rdev_t *rdev = bio2->bi_private;
493         mddev_t *mddev = rdev->mddev;
494
495         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
496             error == -EOPNOTSUPP) {
497                 unsigned long flags;
498                 /* barriers don't appear to be supported :-( */
499                 set_bit(BarriersNotsupp, &rdev->flags);
500                 mddev->barriers_work = 0;
501                 spin_lock_irqsave(&mddev->write_lock, flags);
502                 bio2->bi_next = mddev->biolist;
503                 mddev->biolist = bio2;
504                 spin_unlock_irqrestore(&mddev->write_lock, flags);
505                 wake_up(&mddev->sb_wait);
506                 bio_put(bio);
507         } else {
508                 bio_put(bio2);
509                 bio->bi_private = rdev;
510                 super_written(bio, error);
511         }
512 }
513
514 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
515                    sector_t sector, int size, struct page *page)
516 {
517         /* write first size bytes of page to sector of rdev
518          * Increment mddev->pending_writes before returning
519          * and decrement it on completion, waking up sb_wait
520          * if zero is reached.
521          * If an error occurred, call md_error
522          *
523          * As we might need to resubmit the request if BIO_RW_BARRIER
524          * causes ENOTSUPP, we allocate a spare bio...
525          */
526         struct bio *bio = bio_alloc(GFP_NOIO, 1);
527         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
528
529         bio->bi_bdev = rdev->bdev;
530         bio->bi_sector = sector;
531         bio_add_page(bio, page, size, 0);
532         bio->bi_private = rdev;
533         bio->bi_end_io = super_written;
534         bio->bi_rw = rw;
535
536         atomic_inc(&mddev->pending_writes);
537         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
538                 struct bio *rbio;
539                 rw |= (1<<BIO_RW_BARRIER);
540                 rbio = bio_clone(bio, GFP_NOIO);
541                 rbio->bi_private = bio;
542                 rbio->bi_end_io = super_written_barrier;
543                 submit_bio(rw, rbio);
544         } else
545                 submit_bio(rw, bio);
546 }
547
548 void md_super_wait(mddev_t *mddev)
549 {
550         /* wait for all superblock writes that were scheduled to complete.
551          * if any had to be retried (due to BARRIER problems), retry them
552          */
553         DEFINE_WAIT(wq);
554         for(;;) {
555                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
556                 if (atomic_read(&mddev->pending_writes)==0)
557                         break;
558                 while (mddev->biolist) {
559                         struct bio *bio;
560                         spin_lock_irq(&mddev->write_lock);
561                         bio = mddev->biolist;
562                         mddev->biolist = bio->bi_next ;
563                         bio->bi_next = NULL;
564                         spin_unlock_irq(&mddev->write_lock);
565                         submit_bio(bio->bi_rw, bio);
566                 }
567                 schedule();
568         }
569         finish_wait(&mddev->sb_wait, &wq);
570 }
571
572 static void bi_complete(struct bio *bio, int error)
573 {
574         complete((struct completion*)bio->bi_private);
575 }
576
577 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
578                    struct page *page, int rw)
579 {
580         struct bio *bio = bio_alloc(GFP_NOIO, 1);
581         struct completion event;
582         int ret;
583
584         rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
585
586         bio->bi_bdev = bdev;
587         bio->bi_sector = sector;
588         bio_add_page(bio, page, size, 0);
589         init_completion(&event);
590         bio->bi_private = &event;
591         bio->bi_end_io = bi_complete;
592         submit_bio(rw, bio);
593         wait_for_completion(&event);
594
595         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
596         bio_put(bio);
597         return ret;
598 }
599 EXPORT_SYMBOL_GPL(sync_page_io);
600
601 static int read_disk_sb(mdk_rdev_t * rdev, int size)
602 {
603         char b[BDEVNAME_SIZE];
604         if (!rdev->sb_page) {
605                 MD_BUG();
606                 return -EINVAL;
607         }
608         if (rdev->sb_loaded)
609                 return 0;
610
611
612         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
613                 goto fail;
614         rdev->sb_loaded = 1;
615         return 0;
616
617 fail:
618         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
619                 bdevname(rdev->bdev,b));
620         return -EINVAL;
621 }
622
623 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
624 {
625         return  sb1->set_uuid0 == sb2->set_uuid0 &&
626                 sb1->set_uuid1 == sb2->set_uuid1 &&
627                 sb1->set_uuid2 == sb2->set_uuid2 &&
628                 sb1->set_uuid3 == sb2->set_uuid3;
629 }
630
631 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
632 {
633         int ret;
634         mdp_super_t *tmp1, *tmp2;
635
636         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
637         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
638
639         if (!tmp1 || !tmp2) {
640                 ret = 0;
641                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
642                 goto abort;
643         }
644
645         *tmp1 = *sb1;
646         *tmp2 = *sb2;
647
648         /*
649          * nr_disks is not constant
650          */
651         tmp1->nr_disks = 0;
652         tmp2->nr_disks = 0;
653
654         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
655 abort:
656         kfree(tmp1);
657         kfree(tmp2);
658         return ret;
659 }
660
661
662 static u32 md_csum_fold(u32 csum)
663 {
664         csum = (csum & 0xffff) + (csum >> 16);
665         return (csum & 0xffff) + (csum >> 16);
666 }
667
668 static unsigned int calc_sb_csum(mdp_super_t * sb)
669 {
670         u64 newcsum = 0;
671         u32 *sb32 = (u32*)sb;
672         int i;
673         unsigned int disk_csum, csum;
674
675         disk_csum = sb->sb_csum;
676         sb->sb_csum = 0;
677
678         for (i = 0; i < MD_SB_BYTES/4 ; i++)
679                 newcsum += sb32[i];
680         csum = (newcsum & 0xffffffff) + (newcsum>>32);
681
682
683 #ifdef CONFIG_ALPHA
684         /* This used to use csum_partial, which was wrong for several
685          * reasons including that different results are returned on
686          * different architectures.  It isn't critical that we get exactly
687          * the same return value as before (we always csum_fold before
688          * testing, and that removes any differences).  However as we
689          * know that csum_partial always returned a 16bit value on
690          * alphas, do a fold to maximise conformity to previous behaviour.
691          */
692         sb->sb_csum = md_csum_fold(disk_csum);
693 #else
694         sb->sb_csum = disk_csum;
695 #endif
696         return csum;
697 }
698
699
700 /*
701  * Handle superblock details.
702  * We want to be able to handle multiple superblock formats
703  * so we have a common interface to them all, and an array of
704  * different handlers.
705  * We rely on user-space to write the initial superblock, and support
706  * reading and updating of superblocks.
707  * Interface methods are:
708  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
709  *      loads and validates a superblock on dev.
710  *      if refdev != NULL, compare superblocks on both devices
711  *    Return:
712  *      0 - dev has a superblock that is compatible with refdev
713  *      1 - dev has a superblock that is compatible and newer than refdev
714  *          so dev should be used as the refdev in future
715  *     -EINVAL superblock incompatible or invalid
716  *     -othererror e.g. -EIO
717  *
718  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
719  *      Verify that dev is acceptable into mddev.
720  *       The first time, mddev->raid_disks will be 0, and data from
721  *       dev should be merged in.  Subsequent calls check that dev
722  *       is new enough.  Return 0 or -EINVAL
723  *
724  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
725  *     Update the superblock for rdev with data in mddev
726  *     This does not write to disc.
727  *
728  */
729
730 struct super_type  {
731         char                *name;
732         struct module       *owner;
733         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
734                                           int minor_version);
735         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
736         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
737         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
738                                                 sector_t num_sectors);
739 };
740
741 /*
742  * Check that the given mddev has no bitmap.
743  *
744  * This function is called from the run method of all personalities that do not
745  * support bitmaps. It prints an error message and returns non-zero if mddev
746  * has a bitmap. Otherwise, it returns 0.
747  *
748  */
749 int md_check_no_bitmap(mddev_t *mddev)
750 {
751         if (!mddev->bitmap_file && !mddev->bitmap_offset)
752                 return 0;
753         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
754                 mdname(mddev), mddev->pers->name);
755         return 1;
756 }
757 EXPORT_SYMBOL(md_check_no_bitmap);
758
759 /*
760  * load_super for 0.90.0 
761  */
762 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
763 {
764         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
765         mdp_super_t *sb;
766         int ret;
767
768         /*
769          * Calculate the position of the superblock (512byte sectors),
770          * it's at the end of the disk.
771          *
772          * It also happens to be a multiple of 4Kb.
773          */
774         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
775
776         ret = read_disk_sb(rdev, MD_SB_BYTES);
777         if (ret) return ret;
778
779         ret = -EINVAL;
780
781         bdevname(rdev->bdev, b);
782         sb = (mdp_super_t*)page_address(rdev->sb_page);
783
784         if (sb->md_magic != MD_SB_MAGIC) {
785                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
786                        b);
787                 goto abort;
788         }
789
790         if (sb->major_version != 0 ||
791             sb->minor_version < 90 ||
792             sb->minor_version > 91) {
793                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
794                         sb->major_version, sb->minor_version,
795                         b);
796                 goto abort;
797         }
798
799         if (sb->raid_disks <= 0)
800                 goto abort;
801
802         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
803                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
804                         b);
805                 goto abort;
806         }
807
808         rdev->preferred_minor = sb->md_minor;
809         rdev->data_offset = 0;
810         rdev->sb_size = MD_SB_BYTES;
811
812         if (sb->level == LEVEL_MULTIPATH)
813                 rdev->desc_nr = -1;
814         else
815                 rdev->desc_nr = sb->this_disk.number;
816
817         if (!refdev) {
818                 ret = 1;
819         } else {
820                 __u64 ev1, ev2;
821                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
822                 if (!uuid_equal(refsb, sb)) {
823                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
824                                 b, bdevname(refdev->bdev,b2));
825                         goto abort;
826                 }
827                 if (!sb_equal(refsb, sb)) {
828                         printk(KERN_WARNING "md: %s has same UUID"
829                                " but different superblock to %s\n",
830                                b, bdevname(refdev->bdev, b2));
831                         goto abort;
832                 }
833                 ev1 = md_event(sb);
834                 ev2 = md_event(refsb);
835                 if (ev1 > ev2)
836                         ret = 1;
837                 else 
838                         ret = 0;
839         }
840         rdev->sectors = rdev->sb_start;
841
842         if (rdev->sectors < sb->size * 2 && sb->level > 1)
843                 /* "this cannot possibly happen" ... */
844                 ret = -EINVAL;
845
846  abort:
847         return ret;
848 }
849
850 /*
851  * validate_super for 0.90.0
852  */
853 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
854 {
855         mdp_disk_t *desc;
856         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
857         __u64 ev1 = md_event(sb);
858
859         rdev->raid_disk = -1;
860         clear_bit(Faulty, &rdev->flags);
861         clear_bit(In_sync, &rdev->flags);
862         clear_bit(WriteMostly, &rdev->flags);
863         clear_bit(BarriersNotsupp, &rdev->flags);
864
865         if (mddev->raid_disks == 0) {
866                 mddev->major_version = 0;
867                 mddev->minor_version = sb->minor_version;
868                 mddev->patch_version = sb->patch_version;
869                 mddev->external = 0;
870                 mddev->chunk_sectors = sb->chunk_size >> 9;
871                 mddev->ctime = sb->ctime;
872                 mddev->utime = sb->utime;
873                 mddev->level = sb->level;
874                 mddev->clevel[0] = 0;
875                 mddev->layout = sb->layout;
876                 mddev->raid_disks = sb->raid_disks;
877                 mddev->dev_sectors = sb->size * 2;
878                 mddev->events = ev1;
879                 mddev->bitmap_offset = 0;
880                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
881
882                 if (mddev->minor_version >= 91) {
883                         mddev->reshape_position = sb->reshape_position;
884                         mddev->delta_disks = sb->delta_disks;
885                         mddev->new_level = sb->new_level;
886                         mddev->new_layout = sb->new_layout;
887                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
888                 } else {
889                         mddev->reshape_position = MaxSector;
890                         mddev->delta_disks = 0;
891                         mddev->new_level = mddev->level;
892                         mddev->new_layout = mddev->layout;
893                         mddev->new_chunk_sectors = mddev->chunk_sectors;
894                 }
895
896                 if (sb->state & (1<<MD_SB_CLEAN))
897                         mddev->recovery_cp = MaxSector;
898                 else {
899                         if (sb->events_hi == sb->cp_events_hi && 
900                                 sb->events_lo == sb->cp_events_lo) {
901                                 mddev->recovery_cp = sb->recovery_cp;
902                         } else
903                                 mddev->recovery_cp = 0;
904                 }
905
906                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
907                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
908                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
909                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
910
911                 mddev->max_disks = MD_SB_DISKS;
912
913                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
914                     mddev->bitmap_file == NULL)
915                         mddev->bitmap_offset = mddev->default_bitmap_offset;
916
917         } else if (mddev->pers == NULL) {
918                 /* Insist on good event counter while assembling */
919                 ++ev1;
920                 if (ev1 < mddev->events) 
921                         return -EINVAL;
922         } else if (mddev->bitmap) {
923                 /* if adding to array with a bitmap, then we can accept an
924                  * older device ... but not too old.
925                  */
926                 if (ev1 < mddev->bitmap->events_cleared)
927                         return 0;
928         } else {
929                 if (ev1 < mddev->events)
930                         /* just a hot-add of a new device, leave raid_disk at -1 */
931                         return 0;
932         }
933
934         if (mddev->level != LEVEL_MULTIPATH) {
935                 desc = sb->disks + rdev->desc_nr;
936
937                 if (desc->state & (1<<MD_DISK_FAULTY))
938                         set_bit(Faulty, &rdev->flags);
939                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
940                             desc->raid_disk < mddev->raid_disks */) {
941                         set_bit(In_sync, &rdev->flags);
942                         rdev->raid_disk = desc->raid_disk;
943                 }
944                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
945                         set_bit(WriteMostly, &rdev->flags);
946         } else /* MULTIPATH are always insync */
947                 set_bit(In_sync, &rdev->flags);
948         return 0;
949 }
950
951 /*
952  * sync_super for 0.90.0
953  */
954 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
955 {
956         mdp_super_t *sb;
957         mdk_rdev_t *rdev2;
958         int next_spare = mddev->raid_disks;
959
960
961         /* make rdev->sb match mddev data..
962          *
963          * 1/ zero out disks
964          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
965          * 3/ any empty disks < next_spare become removed
966          *
967          * disks[0] gets initialised to REMOVED because
968          * we cannot be sure from other fields if it has
969          * been initialised or not.
970          */
971         int i;
972         int active=0, working=0,failed=0,spare=0,nr_disks=0;
973
974         rdev->sb_size = MD_SB_BYTES;
975
976         sb = (mdp_super_t*)page_address(rdev->sb_page);
977
978         memset(sb, 0, sizeof(*sb));
979
980         sb->md_magic = MD_SB_MAGIC;
981         sb->major_version = mddev->major_version;
982         sb->patch_version = mddev->patch_version;
983         sb->gvalid_words  = 0; /* ignored */
984         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
985         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
986         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
987         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
988
989         sb->ctime = mddev->ctime;
990         sb->level = mddev->level;
991         sb->size = mddev->dev_sectors / 2;
992         sb->raid_disks = mddev->raid_disks;
993         sb->md_minor = mddev->md_minor;
994         sb->not_persistent = 0;
995         sb->utime = mddev->utime;
996         sb->state = 0;
997         sb->events_hi = (mddev->events>>32);
998         sb->events_lo = (u32)mddev->events;
999
1000         if (mddev->reshape_position == MaxSector)
1001                 sb->minor_version = 90;
1002         else {
1003                 sb->minor_version = 91;
1004                 sb->reshape_position = mddev->reshape_position;
1005                 sb->new_level = mddev->new_level;
1006                 sb->delta_disks = mddev->delta_disks;
1007                 sb->new_layout = mddev->new_layout;
1008                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1009         }
1010         mddev->minor_version = sb->minor_version;
1011         if (mddev->in_sync)
1012         {
1013                 sb->recovery_cp = mddev->recovery_cp;
1014                 sb->cp_events_hi = (mddev->events>>32);
1015                 sb->cp_events_lo = (u32)mddev->events;
1016                 if (mddev->recovery_cp == MaxSector)
1017                         sb->state = (1<< MD_SB_CLEAN);
1018         } else
1019                 sb->recovery_cp = 0;
1020
1021         sb->layout = mddev->layout;
1022         sb->chunk_size = mddev->chunk_sectors << 9;
1023
1024         if (mddev->bitmap && mddev->bitmap_file == NULL)
1025                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1026
1027         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1028         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1029                 mdp_disk_t *d;
1030                 int desc_nr;
1031                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1032                     && !test_bit(Faulty, &rdev2->flags))
1033                         desc_nr = rdev2->raid_disk;
1034                 else
1035                         desc_nr = next_spare++;
1036                 rdev2->desc_nr = desc_nr;
1037                 d = &sb->disks[rdev2->desc_nr];
1038                 nr_disks++;
1039                 d->number = rdev2->desc_nr;
1040                 d->major = MAJOR(rdev2->bdev->bd_dev);
1041                 d->minor = MINOR(rdev2->bdev->bd_dev);
1042                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1043                     && !test_bit(Faulty, &rdev2->flags))
1044                         d->raid_disk = rdev2->raid_disk;
1045                 else
1046                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1047                 if (test_bit(Faulty, &rdev2->flags))
1048                         d->state = (1<<MD_DISK_FAULTY);
1049                 else if (test_bit(In_sync, &rdev2->flags)) {
1050                         d->state = (1<<MD_DISK_ACTIVE);
1051                         d->state |= (1<<MD_DISK_SYNC);
1052                         active++;
1053                         working++;
1054                 } else {
1055                         d->state = 0;
1056                         spare++;
1057                         working++;
1058                 }
1059                 if (test_bit(WriteMostly, &rdev2->flags))
1060                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1061         }
1062         /* now set the "removed" and "faulty" bits on any missing devices */
1063         for (i=0 ; i < mddev->raid_disks ; i++) {
1064                 mdp_disk_t *d = &sb->disks[i];
1065                 if (d->state == 0 && d->number == 0) {
1066                         d->number = i;
1067                         d->raid_disk = i;
1068                         d->state = (1<<MD_DISK_REMOVED);
1069                         d->state |= (1<<MD_DISK_FAULTY);
1070                         failed++;
1071                 }
1072         }
1073         sb->nr_disks = nr_disks;
1074         sb->active_disks = active;
1075         sb->working_disks = working;
1076         sb->failed_disks = failed;
1077         sb->spare_disks = spare;
1078
1079         sb->this_disk = sb->disks[rdev->desc_nr];
1080         sb->sb_csum = calc_sb_csum(sb);
1081 }
1082
1083 /*
1084  * rdev_size_change for 0.90.0
1085  */
1086 static unsigned long long
1087 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1088 {
1089         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1090                 return 0; /* component must fit device */
1091         if (rdev->mddev->bitmap_offset)
1092                 return 0; /* can't move bitmap */
1093         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1094         if (!num_sectors || num_sectors > rdev->sb_start)
1095                 num_sectors = rdev->sb_start;
1096         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1097                        rdev->sb_page);
1098         md_super_wait(rdev->mddev);
1099         return num_sectors / 2; /* kB for sysfs */
1100 }
1101
1102
1103 /*
1104  * version 1 superblock
1105  */
1106
1107 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1108 {
1109         __le32 disk_csum;
1110         u32 csum;
1111         unsigned long long newcsum;
1112         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1113         __le32 *isuper = (__le32*)sb;
1114         int i;
1115
1116         disk_csum = sb->sb_csum;
1117         sb->sb_csum = 0;
1118         newcsum = 0;
1119         for (i=0; size>=4; size -= 4 )
1120                 newcsum += le32_to_cpu(*isuper++);
1121
1122         if (size == 2)
1123                 newcsum += le16_to_cpu(*(__le16*) isuper);
1124
1125         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1126         sb->sb_csum = disk_csum;
1127         return cpu_to_le32(csum);
1128 }
1129
1130 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1131 {
1132         struct mdp_superblock_1 *sb;
1133         int ret;
1134         sector_t sb_start;
1135         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1136         int bmask;
1137
1138         /*
1139          * Calculate the position of the superblock in 512byte sectors.
1140          * It is always aligned to a 4K boundary and
1141          * depeding on minor_version, it can be:
1142          * 0: At least 8K, but less than 12K, from end of device
1143          * 1: At start of device
1144          * 2: 4K from start of device.
1145          */
1146         switch(minor_version) {
1147         case 0:
1148                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1149                 sb_start -= 8*2;
1150                 sb_start &= ~(sector_t)(4*2-1);
1151                 break;
1152         case 1:
1153                 sb_start = 0;
1154                 break;
1155         case 2:
1156                 sb_start = 8;
1157                 break;
1158         default:
1159                 return -EINVAL;
1160         }
1161         rdev->sb_start = sb_start;
1162
1163         /* superblock is rarely larger than 1K, but it can be larger,
1164          * and it is safe to read 4k, so we do that
1165          */
1166         ret = read_disk_sb(rdev, 4096);
1167         if (ret) return ret;
1168
1169
1170         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1171
1172         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1173             sb->major_version != cpu_to_le32(1) ||
1174             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1175             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1176             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1177                 return -EINVAL;
1178
1179         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1180                 printk("md: invalid superblock checksum on %s\n",
1181                         bdevname(rdev->bdev,b));
1182                 return -EINVAL;
1183         }
1184         if (le64_to_cpu(sb->data_size) < 10) {
1185                 printk("md: data_size too small on %s\n",
1186                        bdevname(rdev->bdev,b));
1187                 return -EINVAL;
1188         }
1189
1190         rdev->preferred_minor = 0xffff;
1191         rdev->data_offset = le64_to_cpu(sb->data_offset);
1192         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1193
1194         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1195         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1196         if (rdev->sb_size & bmask)
1197                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1198
1199         if (minor_version
1200             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1201                 return -EINVAL;
1202
1203         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1204                 rdev->desc_nr = -1;
1205         else
1206                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1207
1208         if (!refdev) {
1209                 ret = 1;
1210         } else {
1211                 __u64 ev1, ev2;
1212                 struct mdp_superblock_1 *refsb = 
1213                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1214
1215                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1216                     sb->level != refsb->level ||
1217                     sb->layout != refsb->layout ||
1218                     sb->chunksize != refsb->chunksize) {
1219                         printk(KERN_WARNING "md: %s has strangely different"
1220                                 " superblock to %s\n",
1221                                 bdevname(rdev->bdev,b),
1222                                 bdevname(refdev->bdev,b2));
1223                         return -EINVAL;
1224                 }
1225                 ev1 = le64_to_cpu(sb->events);
1226                 ev2 = le64_to_cpu(refsb->events);
1227
1228                 if (ev1 > ev2)
1229                         ret = 1;
1230                 else
1231                         ret = 0;
1232         }
1233         if (minor_version)
1234                 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1235                         le64_to_cpu(sb->data_offset);
1236         else
1237                 rdev->sectors = rdev->sb_start;
1238         if (rdev->sectors < le64_to_cpu(sb->data_size))
1239                 return -EINVAL;
1240         rdev->sectors = le64_to_cpu(sb->data_size);
1241         if (le64_to_cpu(sb->size) > rdev->sectors)
1242                 return -EINVAL;
1243         return ret;
1244 }
1245
1246 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1247 {
1248         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1249         __u64 ev1 = le64_to_cpu(sb->events);
1250
1251         rdev->raid_disk = -1;
1252         clear_bit(Faulty, &rdev->flags);
1253         clear_bit(In_sync, &rdev->flags);
1254         clear_bit(WriteMostly, &rdev->flags);
1255         clear_bit(BarriersNotsupp, &rdev->flags);
1256
1257         if (mddev->raid_disks == 0) {
1258                 mddev->major_version = 1;
1259                 mddev->patch_version = 0;
1260                 mddev->external = 0;
1261                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1262                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1263                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1264                 mddev->level = le32_to_cpu(sb->level);
1265                 mddev->clevel[0] = 0;
1266                 mddev->layout = le32_to_cpu(sb->layout);
1267                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1268                 mddev->dev_sectors = le64_to_cpu(sb->size);
1269                 mddev->events = ev1;
1270                 mddev->bitmap_offset = 0;
1271                 mddev->default_bitmap_offset = 1024 >> 9;
1272                 
1273                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1274                 memcpy(mddev->uuid, sb->set_uuid, 16);
1275
1276                 mddev->max_disks =  (4096-256)/2;
1277
1278                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1279                     mddev->bitmap_file == NULL )
1280                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1281
1282                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1283                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1284                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1285                         mddev->new_level = le32_to_cpu(sb->new_level);
1286                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1287                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1288                 } else {
1289                         mddev->reshape_position = MaxSector;
1290                         mddev->delta_disks = 0;
1291                         mddev->new_level = mddev->level;
1292                         mddev->new_layout = mddev->layout;
1293                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1294                 }
1295
1296         } else if (mddev->pers == NULL) {
1297                 /* Insist of good event counter while assembling */
1298                 ++ev1;
1299                 if (ev1 < mddev->events)
1300                         return -EINVAL;
1301         } else if (mddev->bitmap) {
1302                 /* If adding to array with a bitmap, then we can accept an
1303                  * older device, but not too old.
1304                  */
1305                 if (ev1 < mddev->bitmap->events_cleared)
1306                         return 0;
1307         } else {
1308                 if (ev1 < mddev->events)
1309                         /* just a hot-add of a new device, leave raid_disk at -1 */
1310                         return 0;
1311         }
1312         if (mddev->level != LEVEL_MULTIPATH) {
1313                 int role;
1314                 if (rdev->desc_nr < 0 ||
1315                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1316                         role = 0xffff;
1317                         rdev->desc_nr = -1;
1318                 } else
1319                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1320                 switch(role) {
1321                 case 0xffff: /* spare */
1322                         break;
1323                 case 0xfffe: /* faulty */
1324                         set_bit(Faulty, &rdev->flags);
1325                         break;
1326                 default:
1327                         if ((le32_to_cpu(sb->feature_map) &
1328                              MD_FEATURE_RECOVERY_OFFSET))
1329                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1330                         else
1331                                 set_bit(In_sync, &rdev->flags);
1332                         rdev->raid_disk = role;
1333                         break;
1334                 }
1335                 if (sb->devflags & WriteMostly1)
1336                         set_bit(WriteMostly, &rdev->flags);
1337         } else /* MULTIPATH are always insync */
1338                 set_bit(In_sync, &rdev->flags);
1339
1340         return 0;
1341 }
1342
1343 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1344 {
1345         struct mdp_superblock_1 *sb;
1346         mdk_rdev_t *rdev2;
1347         int max_dev, i;
1348         /* make rdev->sb match mddev and rdev data. */
1349
1350         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1351
1352         sb->feature_map = 0;
1353         sb->pad0 = 0;
1354         sb->recovery_offset = cpu_to_le64(0);
1355         memset(sb->pad1, 0, sizeof(sb->pad1));
1356         memset(sb->pad2, 0, sizeof(sb->pad2));
1357         memset(sb->pad3, 0, sizeof(sb->pad3));
1358
1359         sb->utime = cpu_to_le64((__u64)mddev->utime);
1360         sb->events = cpu_to_le64(mddev->events);
1361         if (mddev->in_sync)
1362                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1363         else
1364                 sb->resync_offset = cpu_to_le64(0);
1365
1366         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1367
1368         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1369         sb->size = cpu_to_le64(mddev->dev_sectors);
1370         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1371         sb->level = cpu_to_le32(mddev->level);
1372         sb->layout = cpu_to_le32(mddev->layout);
1373
1374         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1375                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1376                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1377         }
1378
1379         if (rdev->raid_disk >= 0 &&
1380             !test_bit(In_sync, &rdev->flags)) {
1381                 if (mddev->curr_resync_completed > rdev->recovery_offset)
1382                         rdev->recovery_offset = mddev->curr_resync_completed;
1383                 if (rdev->recovery_offset > 0) {
1384                         sb->feature_map |=
1385                                 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1386                         sb->recovery_offset =
1387                                 cpu_to_le64(rdev->recovery_offset);
1388                 }
1389         }
1390
1391         if (mddev->reshape_position != MaxSector) {
1392                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1393                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1394                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1395                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1396                 sb->new_level = cpu_to_le32(mddev->new_level);
1397                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1398         }
1399
1400         max_dev = 0;
1401         list_for_each_entry(rdev2, &mddev->disks, same_set)
1402                 if (rdev2->desc_nr+1 > max_dev)
1403                         max_dev = rdev2->desc_nr+1;
1404
1405         if (max_dev > le32_to_cpu(sb->max_dev)) {
1406                 int bmask;
1407                 sb->max_dev = cpu_to_le32(max_dev);
1408                 rdev->sb_size = max_dev * 2 + 256;
1409                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1410                 if (rdev->sb_size & bmask)
1411                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1412         }
1413         for (i=0; i<max_dev;i++)
1414                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1415         
1416         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1417                 i = rdev2->desc_nr;
1418                 if (test_bit(Faulty, &rdev2->flags))
1419                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1420                 else if (test_bit(In_sync, &rdev2->flags))
1421                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1422                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1423                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1424                 else
1425                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1426         }
1427
1428         sb->sb_csum = calc_sb_1_csum(sb);
1429 }
1430
1431 static unsigned long long
1432 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1433 {
1434         struct mdp_superblock_1 *sb;
1435         sector_t max_sectors;
1436         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1437                 return 0; /* component must fit device */
1438         if (rdev->sb_start < rdev->data_offset) {
1439                 /* minor versions 1 and 2; superblock before data */
1440                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1441                 max_sectors -= rdev->data_offset;
1442                 if (!num_sectors || num_sectors > max_sectors)
1443                         num_sectors = max_sectors;
1444         } else if (rdev->mddev->bitmap_offset) {
1445                 /* minor version 0 with bitmap we can't move */
1446                 return 0;
1447         } else {
1448                 /* minor version 0; superblock after data */
1449                 sector_t sb_start;
1450                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1451                 sb_start &= ~(sector_t)(4*2 - 1);
1452                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1453                 if (!num_sectors || num_sectors > max_sectors)
1454                         num_sectors = max_sectors;
1455                 rdev->sb_start = sb_start;
1456         }
1457         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1458         sb->data_size = cpu_to_le64(num_sectors);
1459         sb->super_offset = rdev->sb_start;
1460         sb->sb_csum = calc_sb_1_csum(sb);
1461         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1462                        rdev->sb_page);
1463         md_super_wait(rdev->mddev);
1464         return num_sectors / 2; /* kB for sysfs */
1465 }
1466
1467 static struct super_type super_types[] = {
1468         [0] = {
1469                 .name   = "0.90.0",
1470                 .owner  = THIS_MODULE,
1471                 .load_super         = super_90_load,
1472                 .validate_super     = super_90_validate,
1473                 .sync_super         = super_90_sync,
1474                 .rdev_size_change   = super_90_rdev_size_change,
1475         },
1476         [1] = {
1477                 .name   = "md-1",
1478                 .owner  = THIS_MODULE,
1479                 .load_super         = super_1_load,
1480                 .validate_super     = super_1_validate,
1481                 .sync_super         = super_1_sync,
1482                 .rdev_size_change   = super_1_rdev_size_change,
1483         },
1484 };
1485
1486 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1487 {
1488         mdk_rdev_t *rdev, *rdev2;
1489
1490         rcu_read_lock();
1491         rdev_for_each_rcu(rdev, mddev1)
1492                 rdev_for_each_rcu(rdev2, mddev2)
1493                         if (rdev->bdev->bd_contains ==
1494                             rdev2->bdev->bd_contains) {
1495                                 rcu_read_unlock();
1496                                 return 1;
1497                         }
1498         rcu_read_unlock();
1499         return 0;
1500 }
1501
1502 static LIST_HEAD(pending_raid_disks);
1503
1504 /*
1505  * Try to register data integrity profile for an mddev
1506  *
1507  * This is called when an array is started and after a disk has been kicked
1508  * from the array. It only succeeds if all working and active component devices
1509  * are integrity capable with matching profiles.
1510  */
1511 int md_integrity_register(mddev_t *mddev)
1512 {
1513         mdk_rdev_t *rdev, *reference = NULL;
1514
1515         if (list_empty(&mddev->disks))
1516                 return 0; /* nothing to do */
1517         if (blk_get_integrity(mddev->gendisk))
1518                 return 0; /* already registered */
1519         list_for_each_entry(rdev, &mddev->disks, same_set) {
1520                 /* skip spares and non-functional disks */
1521                 if (test_bit(Faulty, &rdev->flags))
1522                         continue;
1523                 if (rdev->raid_disk < 0)
1524                         continue;
1525                 /*
1526                  * If at least one rdev is not integrity capable, we can not
1527                  * enable data integrity for the md device.
1528                  */
1529                 if (!bdev_get_integrity(rdev->bdev))
1530                         return -EINVAL;
1531                 if (!reference) {
1532                         /* Use the first rdev as the reference */
1533                         reference = rdev;
1534                         continue;
1535                 }
1536                 /* does this rdev's profile match the reference profile? */
1537                 if (blk_integrity_compare(reference->bdev->bd_disk,
1538                                 rdev->bdev->bd_disk) < 0)
1539                         return -EINVAL;
1540         }
1541         /*
1542          * All component devices are integrity capable and have matching
1543          * profiles, register the common profile for the md device.
1544          */
1545         if (blk_integrity_register(mddev->gendisk,
1546                         bdev_get_integrity(reference->bdev)) != 0) {
1547                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1548                         mdname(mddev));
1549                 return -EINVAL;
1550         }
1551         printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1552                 mdname(mddev));
1553         return 0;
1554 }
1555 EXPORT_SYMBOL(md_integrity_register);
1556
1557 /* Disable data integrity if non-capable/non-matching disk is being added */
1558 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1559 {
1560         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1561         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1562
1563         if (!bi_mddev) /* nothing to do */
1564                 return;
1565         if (rdev->raid_disk < 0) /* skip spares */
1566                 return;
1567         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1568                                              rdev->bdev->bd_disk) >= 0)
1569                 return;
1570         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1571         blk_integrity_unregister(mddev->gendisk);
1572 }
1573 EXPORT_SYMBOL(md_integrity_add_rdev);
1574
1575 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1576 {
1577         char b[BDEVNAME_SIZE];
1578         struct kobject *ko;
1579         char *s;
1580         int err;
1581
1582         if (rdev->mddev) {
1583                 MD_BUG();
1584                 return -EINVAL;
1585         }
1586
1587         /* prevent duplicates */
1588         if (find_rdev(mddev, rdev->bdev->bd_dev))
1589                 return -EEXIST;
1590
1591         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1592         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1593                         rdev->sectors < mddev->dev_sectors)) {
1594                 if (mddev->pers) {
1595                         /* Cannot change size, so fail
1596                          * If mddev->level <= 0, then we don't care
1597                          * about aligning sizes (e.g. linear)
1598                          */
1599                         if (mddev->level > 0)
1600                                 return -ENOSPC;
1601                 } else
1602                         mddev->dev_sectors = rdev->sectors;
1603         }
1604
1605         /* Verify rdev->desc_nr is unique.
1606          * If it is -1, assign a free number, else
1607          * check number is not in use
1608          */
1609         if (rdev->desc_nr < 0) {
1610                 int choice = 0;
1611                 if (mddev->pers) choice = mddev->raid_disks;
1612                 while (find_rdev_nr(mddev, choice))
1613                         choice++;
1614                 rdev->desc_nr = choice;
1615         } else {
1616                 if (find_rdev_nr(mddev, rdev->desc_nr))
1617                         return -EBUSY;
1618         }
1619         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1620                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1621                        mdname(mddev), mddev->max_disks);
1622                 return -EBUSY;
1623         }
1624         bdevname(rdev->bdev,b);
1625         while ( (s=strchr(b, '/')) != NULL)
1626                 *s = '!';
1627
1628         rdev->mddev = mddev;
1629         printk(KERN_INFO "md: bind<%s>\n", b);
1630
1631         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1632                 goto fail;
1633
1634         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1635         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1636                 kobject_del(&rdev->kobj);
1637                 goto fail;
1638         }
1639         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1640
1641         list_add_rcu(&rdev->same_set, &mddev->disks);
1642         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1643
1644         /* May as well allow recovery to be retried once */
1645         mddev->recovery_disabled = 0;
1646
1647         return 0;
1648
1649  fail:
1650         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1651                b, mdname(mddev));
1652         return err;
1653 }
1654
1655 static void md_delayed_delete(struct work_struct *ws)
1656 {
1657         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1658         kobject_del(&rdev->kobj);
1659         kobject_put(&rdev->kobj);
1660 }
1661
1662 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1663 {
1664         char b[BDEVNAME_SIZE];
1665         if (!rdev->mddev) {
1666                 MD_BUG();
1667                 return;
1668         }
1669         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1670         list_del_rcu(&rdev->same_set);
1671         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1672         rdev->mddev = NULL;
1673         sysfs_remove_link(&rdev->kobj, "block");
1674         sysfs_put(rdev->sysfs_state);
1675         rdev->sysfs_state = NULL;
1676         /* We need to delay this, otherwise we can deadlock when
1677          * writing to 'remove' to "dev/state".  We also need
1678          * to delay it due to rcu usage.
1679          */
1680         synchronize_rcu();
1681         INIT_WORK(&rdev->del_work, md_delayed_delete);
1682         kobject_get(&rdev->kobj);
1683         schedule_work(&rdev->del_work);
1684 }
1685
1686 /*
1687  * prevent the device from being mounted, repartitioned or
1688  * otherwise reused by a RAID array (or any other kernel
1689  * subsystem), by bd_claiming the device.
1690  */
1691 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1692 {
1693         int err = 0;
1694         struct block_device *bdev;
1695         char b[BDEVNAME_SIZE];
1696
1697         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1698         if (IS_ERR(bdev)) {
1699                 printk(KERN_ERR "md: could not open %s.\n",
1700                         __bdevname(dev, b));
1701                 return PTR_ERR(bdev);
1702         }
1703         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1704         if (err) {
1705                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1706                         bdevname(bdev, b));
1707                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1708                 return err;
1709         }
1710         if (!shared)
1711                 set_bit(AllReserved, &rdev->flags);
1712         rdev->bdev = bdev;
1713         return err;
1714 }
1715
1716 static void unlock_rdev(mdk_rdev_t *rdev)
1717 {
1718         struct block_device *bdev = rdev->bdev;
1719         rdev->bdev = NULL;
1720         if (!bdev)
1721                 MD_BUG();
1722         bd_release(bdev);
1723         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1724 }
1725
1726 void md_autodetect_dev(dev_t dev);
1727
1728 static void export_rdev(mdk_rdev_t * rdev)
1729 {
1730         char b[BDEVNAME_SIZE];
1731         printk(KERN_INFO "md: export_rdev(%s)\n",
1732                 bdevname(rdev->bdev,b));
1733         if (rdev->mddev)
1734                 MD_BUG();
1735         free_disk_sb(rdev);
1736 #ifndef MODULE
1737         if (test_bit(AutoDetected, &rdev->flags))
1738                 md_autodetect_dev(rdev->bdev->bd_dev);
1739 #endif
1740         unlock_rdev(rdev);
1741         kobject_put(&rdev->kobj);
1742 }
1743
1744 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1745 {
1746         unbind_rdev_from_array(rdev);
1747         export_rdev(rdev);
1748 }
1749
1750 static void export_array(mddev_t *mddev)
1751 {
1752         mdk_rdev_t *rdev, *tmp;
1753
1754         rdev_for_each(rdev, tmp, mddev) {
1755                 if (!rdev->mddev) {
1756                         MD_BUG();
1757                         continue;
1758                 }
1759                 kick_rdev_from_array(rdev);
1760         }
1761         if (!list_empty(&mddev->disks))
1762                 MD_BUG();
1763         mddev->raid_disks = 0;
1764         mddev->major_version = 0;
1765 }
1766
1767 static void print_desc(mdp_disk_t *desc)
1768 {
1769         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1770                 desc->major,desc->minor,desc->raid_disk,desc->state);
1771 }
1772
1773 static void print_sb_90(mdp_super_t *sb)
1774 {
1775         int i;
1776
1777         printk(KERN_INFO 
1778                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1779                 sb->major_version, sb->minor_version, sb->patch_version,
1780                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1781                 sb->ctime);
1782         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1783                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1784                 sb->md_minor, sb->layout, sb->chunk_size);
1785         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1786                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1787                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1788                 sb->failed_disks, sb->spare_disks,
1789                 sb->sb_csum, (unsigned long)sb->events_lo);
1790
1791         printk(KERN_INFO);
1792         for (i = 0; i < MD_SB_DISKS; i++) {
1793                 mdp_disk_t *desc;
1794
1795                 desc = sb->disks + i;
1796                 if (desc->number || desc->major || desc->minor ||
1797                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1798                         printk("     D %2d: ", i);
1799                         print_desc(desc);
1800                 }
1801         }
1802         printk(KERN_INFO "md:     THIS: ");
1803         print_desc(&sb->this_disk);
1804 }
1805
1806 static void print_sb_1(struct mdp_superblock_1 *sb)
1807 {
1808         __u8 *uuid;
1809
1810         uuid = sb->set_uuid;
1811         printk(KERN_INFO
1812                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1813                ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1814                "md:    Name: \"%s\" CT:%llu\n",
1815                 le32_to_cpu(sb->major_version),
1816                 le32_to_cpu(sb->feature_map),
1817                 uuid[0], uuid[1], uuid[2], uuid[3],
1818                 uuid[4], uuid[5], uuid[6], uuid[7],
1819                 uuid[8], uuid[9], uuid[10], uuid[11],
1820                 uuid[12], uuid[13], uuid[14], uuid[15],
1821                 sb->set_name,
1822                 (unsigned long long)le64_to_cpu(sb->ctime)
1823                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1824
1825         uuid = sb->device_uuid;
1826         printk(KERN_INFO
1827                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1828                         " RO:%llu\n"
1829                "md:     Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1830                         ":%02x%02x%02x%02x%02x%02x\n"
1831                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1832                "md:         (MaxDev:%u) \n",
1833                 le32_to_cpu(sb->level),
1834                 (unsigned long long)le64_to_cpu(sb->size),
1835                 le32_to_cpu(sb->raid_disks),
1836                 le32_to_cpu(sb->layout),
1837                 le32_to_cpu(sb->chunksize),
1838                 (unsigned long long)le64_to_cpu(sb->data_offset),
1839                 (unsigned long long)le64_to_cpu(sb->data_size),
1840                 (unsigned long long)le64_to_cpu(sb->super_offset),
1841                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1842                 le32_to_cpu(sb->dev_number),
1843                 uuid[0], uuid[1], uuid[2], uuid[3],
1844                 uuid[4], uuid[5], uuid[6], uuid[7],
1845                 uuid[8], uuid[9], uuid[10], uuid[11],
1846                 uuid[12], uuid[13], uuid[14], uuid[15],
1847                 sb->devflags,
1848                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1849                 (unsigned long long)le64_to_cpu(sb->events),
1850                 (unsigned long long)le64_to_cpu(sb->resync_offset),
1851                 le32_to_cpu(sb->sb_csum),
1852                 le32_to_cpu(sb->max_dev)
1853                 );
1854 }
1855
1856 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1857 {
1858         char b[BDEVNAME_SIZE];
1859         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1860                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1861                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1862                 rdev->desc_nr);
1863         if (rdev->sb_loaded) {
1864                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1865                 switch (major_version) {
1866                 case 0:
1867                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1868                         break;
1869                 case 1:
1870                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1871                         break;
1872                 }
1873         } else
1874                 printk(KERN_INFO "md: no rdev superblock!\n");
1875 }
1876
1877 static void md_print_devices(void)
1878 {
1879         struct list_head *tmp;
1880         mdk_rdev_t *rdev;
1881         mddev_t *mddev;
1882         char b[BDEVNAME_SIZE];
1883
1884         printk("\n");
1885         printk("md:     **********************************\n");
1886         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1887         printk("md:     **********************************\n");
1888         for_each_mddev(mddev, tmp) {
1889
1890                 if (mddev->bitmap)
1891                         bitmap_print_sb(mddev->bitmap);
1892                 else
1893                         printk("%s: ", mdname(mddev));
1894                 list_for_each_entry(rdev, &mddev->disks, same_set)
1895                         printk("<%s>", bdevname(rdev->bdev,b));
1896                 printk("\n");
1897
1898                 list_for_each_entry(rdev, &mddev->disks, same_set)
1899                         print_rdev(rdev, mddev->major_version);
1900         }
1901         printk("md:     **********************************\n");
1902         printk("\n");
1903 }
1904
1905
1906 static void sync_sbs(mddev_t * mddev, int nospares)
1907 {
1908         /* Update each superblock (in-memory image), but
1909          * if we are allowed to, skip spares which already
1910          * have the right event counter, or have one earlier
1911          * (which would mean they aren't being marked as dirty
1912          * with the rest of the array)
1913          */
1914         mdk_rdev_t *rdev;
1915
1916         list_for_each_entry(rdev, &mddev->disks, same_set) {
1917                 if (rdev->sb_events == mddev->events ||
1918                     (nospares &&
1919                      rdev->raid_disk < 0 &&
1920                      (rdev->sb_events&1)==0 &&
1921                      rdev->sb_events+1 == mddev->events)) {
1922                         /* Don't update this superblock */
1923                         rdev->sb_loaded = 2;
1924                 } else {
1925                         super_types[mddev->major_version].
1926                                 sync_super(mddev, rdev);
1927                         rdev->sb_loaded = 1;
1928                 }
1929         }
1930 }
1931
1932 static void md_update_sb(mddev_t * mddev, int force_change)
1933 {
1934         mdk_rdev_t *rdev;
1935         int sync_req;
1936         int nospares = 0;
1937
1938         mddev->utime = get_seconds();
1939         if (mddev->external)
1940                 return;
1941 repeat:
1942         spin_lock_irq(&mddev->write_lock);
1943
1944         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1945         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1946                 force_change = 1;
1947         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1948                 /* just a clean<-> dirty transition, possibly leave spares alone,
1949                  * though if events isn't the right even/odd, we will have to do
1950                  * spares after all
1951                  */
1952                 nospares = 1;
1953         if (force_change)
1954                 nospares = 0;
1955         if (mddev->degraded)
1956                 /* If the array is degraded, then skipping spares is both
1957                  * dangerous and fairly pointless.
1958                  * Dangerous because a device that was removed from the array
1959                  * might have a event_count that still looks up-to-date,
1960                  * so it can be re-added without a resync.
1961                  * Pointless because if there are any spares to skip,
1962                  * then a recovery will happen and soon that array won't
1963                  * be degraded any more and the spare can go back to sleep then.
1964                  */
1965                 nospares = 0;
1966
1967         sync_req = mddev->in_sync;
1968
1969         /* If this is just a dirty<->clean transition, and the array is clean
1970          * and 'events' is odd, we can roll back to the previous clean state */
1971         if (nospares
1972             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1973             && (mddev->events & 1)
1974             && mddev->events != 1)
1975                 mddev->events--;
1976         else {
1977                 /* otherwise we have to go forward and ... */
1978                 mddev->events ++;
1979                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1980                         /* .. if the array isn't clean, an 'even' event must also go
1981                          * to spares. */
1982                         if ((mddev->events&1)==0)
1983                                 nospares = 0;
1984                 } else {
1985                         /* otherwise an 'odd' event must go to spares */
1986                         if ((mddev->events&1))
1987                                 nospares = 0;
1988                 }
1989         }
1990
1991         if (!mddev->events) {
1992                 /*
1993                  * oops, this 64-bit counter should never wrap.
1994                  * Either we are in around ~1 trillion A.C., assuming
1995                  * 1 reboot per second, or we have a bug:
1996                  */
1997                 MD_BUG();
1998                 mddev->events --;
1999         }
2000
2001         /*
2002          * do not write anything to disk if using
2003          * nonpersistent superblocks
2004          */
2005         if (!mddev->persistent) {
2006                 if (!mddev->external)
2007                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2008
2009                 spin_unlock_irq(&mddev->write_lock);
2010                 wake_up(&mddev->sb_wait);
2011                 return;
2012         }
2013         sync_sbs(mddev, nospares);
2014         spin_unlock_irq(&mddev->write_lock);
2015
2016         dprintk(KERN_INFO 
2017                 "md: updating %s RAID superblock on device (in sync %d)\n",
2018                 mdname(mddev),mddev->in_sync);
2019
2020         bitmap_update_sb(mddev->bitmap);
2021         list_for_each_entry(rdev, &mddev->disks, same_set) {
2022                 char b[BDEVNAME_SIZE];
2023                 dprintk(KERN_INFO "md: ");
2024                 if (rdev->sb_loaded != 1)
2025                         continue; /* no noise on spare devices */
2026                 if (test_bit(Faulty, &rdev->flags))
2027                         dprintk("(skipping faulty ");
2028
2029                 dprintk("%s ", bdevname(rdev->bdev,b));
2030                 if (!test_bit(Faulty, &rdev->flags)) {
2031                         md_super_write(mddev,rdev,
2032                                        rdev->sb_start, rdev->sb_size,
2033                                        rdev->sb_page);
2034                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2035                                 bdevname(rdev->bdev,b),
2036                                 (unsigned long long)rdev->sb_start);
2037                         rdev->sb_events = mddev->events;
2038
2039                 } else
2040                         dprintk(")\n");
2041                 if (mddev->level == LEVEL_MULTIPATH)
2042                         /* only need to write one superblock... */
2043                         break;
2044         }
2045         md_super_wait(mddev);
2046         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2047
2048         spin_lock_irq(&mddev->write_lock);
2049         if (mddev->in_sync != sync_req ||
2050             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2051                 /* have to write it out again */
2052                 spin_unlock_irq(&mddev->write_lock);
2053                 goto repeat;
2054         }
2055         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2056         spin_unlock_irq(&mddev->write_lock);
2057         wake_up(&mddev->sb_wait);
2058         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2059                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2060
2061 }
2062
2063 /* words written to sysfs files may, or may not, be \n terminated.
2064  * We want to accept with case. For this we use cmd_match.
2065  */
2066 static int cmd_match(const char *cmd, const char *str)
2067 {
2068         /* See if cmd, written into a sysfs file, matches
2069          * str.  They must either be the same, or cmd can
2070          * have a trailing newline
2071          */
2072         while (*cmd && *str && *cmd == *str) {
2073                 cmd++;
2074                 str++;
2075         }
2076         if (*cmd == '\n')
2077                 cmd++;
2078         if (*str || *cmd)
2079                 return 0;
2080         return 1;
2081 }
2082
2083 struct rdev_sysfs_entry {
2084         struct attribute attr;
2085         ssize_t (*show)(mdk_rdev_t *, char *);
2086         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2087 };
2088
2089 static ssize_t
2090 state_show(mdk_rdev_t *rdev, char *page)
2091 {
2092         char *sep = "";
2093         size_t len = 0;
2094
2095         if (test_bit(Faulty, &rdev->flags)) {
2096                 len+= sprintf(page+len, "%sfaulty",sep);
2097                 sep = ",";
2098         }
2099         if (test_bit(In_sync, &rdev->flags)) {
2100                 len += sprintf(page+len, "%sin_sync",sep);
2101                 sep = ",";
2102         }
2103         if (test_bit(WriteMostly, &rdev->flags)) {
2104                 len += sprintf(page+len, "%swrite_mostly",sep);
2105                 sep = ",";
2106         }
2107         if (test_bit(Blocked, &rdev->flags)) {
2108                 len += sprintf(page+len, "%sblocked", sep);
2109                 sep = ",";
2110         }
2111         if (!test_bit(Faulty, &rdev->flags) &&
2112             !test_bit(In_sync, &rdev->flags)) {
2113                 len += sprintf(page+len, "%sspare", sep);
2114                 sep = ",";
2115         }
2116         return len+sprintf(page+len, "\n");
2117 }
2118
2119 static ssize_t
2120 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2121 {
2122         /* can write
2123          *  faulty  - simulates and error
2124          *  remove  - disconnects the device
2125          *  writemostly - sets write_mostly
2126          *  -writemostly - clears write_mostly
2127          *  blocked - sets the Blocked flag
2128          *  -blocked - clears the Blocked flag
2129          *  insync - sets Insync providing device isn't active
2130          */
2131         int err = -EINVAL;
2132         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2133                 md_error(rdev->mddev, rdev);
2134                 err = 0;
2135         } else if (cmd_match(buf, "remove")) {
2136                 if (rdev->raid_disk >= 0)
2137                         err = -EBUSY;
2138                 else {
2139                         mddev_t *mddev = rdev->mddev;
2140                         kick_rdev_from_array(rdev);
2141                         if (mddev->pers)
2142                                 md_update_sb(mddev, 1);
2143                         md_new_event(mddev);
2144                         err = 0;
2145                 }
2146         } else if (cmd_match(buf, "writemostly")) {
2147                 set_bit(WriteMostly, &rdev->flags);
2148                 err = 0;
2149         } else if (cmd_match(buf, "-writemostly")) {
2150                 clear_bit(WriteMostly, &rdev->flags);
2151                 err = 0;
2152         } else if (cmd_match(buf, "blocked")) {
2153                 set_bit(Blocked, &rdev->flags);
2154                 err = 0;
2155         } else if (cmd_match(buf, "-blocked")) {
2156                 clear_bit(Blocked, &rdev->flags);
2157                 wake_up(&rdev->blocked_wait);
2158                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2159                 md_wakeup_thread(rdev->mddev->thread);
2160
2161                 err = 0;
2162         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2163                 set_bit(In_sync, &rdev->flags);
2164                 err = 0;
2165         }
2166         if (!err && rdev->sysfs_state)
2167                 sysfs_notify_dirent(rdev->sysfs_state);
2168         return err ? err : len;
2169 }
2170 static struct rdev_sysfs_entry rdev_state =
2171 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2172
2173 static ssize_t
2174 errors_show(mdk_rdev_t *rdev, char *page)
2175 {
2176         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2177 }
2178
2179 static ssize_t
2180 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2181 {
2182         char *e;
2183         unsigned long n = simple_strtoul(buf, &e, 10);
2184         if (*buf && (*e == 0 || *e == '\n')) {
2185                 atomic_set(&rdev->corrected_errors, n);
2186                 return len;
2187         }
2188         return -EINVAL;
2189 }
2190 static struct rdev_sysfs_entry rdev_errors =
2191 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2192
2193 static ssize_t
2194 slot_show(mdk_rdev_t *rdev, char *page)
2195 {
2196         if (rdev->raid_disk < 0)
2197                 return sprintf(page, "none\n");
2198         else
2199                 return sprintf(page, "%d\n", rdev->raid_disk);
2200 }
2201
2202 static ssize_t
2203 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2204 {
2205         char *e;
2206         int err;
2207         char nm[20];
2208         int slot = simple_strtoul(buf, &e, 10);
2209         if (strncmp(buf, "none", 4)==0)
2210                 slot = -1;
2211         else if (e==buf || (*e && *e!= '\n'))
2212                 return -EINVAL;
2213         if (rdev->mddev->pers && slot == -1) {
2214                 /* Setting 'slot' on an active array requires also
2215                  * updating the 'rd%d' link, and communicating
2216                  * with the personality with ->hot_*_disk.
2217                  * For now we only support removing
2218                  * failed/spare devices.  This normally happens automatically,
2219                  * but not when the metadata is externally managed.
2220                  */
2221                 if (rdev->raid_disk == -1)
2222                         return -EEXIST;
2223                 /* personality does all needed checks */
2224                 if (rdev->mddev->pers->hot_add_disk == NULL)
2225                         return -EINVAL;
2226                 err = rdev->mddev->pers->
2227                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2228                 if (err)
2229                         return err;
2230                 sprintf(nm, "rd%d", rdev->raid_disk);
2231                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2232                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2233                 md_wakeup_thread(rdev->mddev->thread);
2234         } else if (rdev->mddev->pers) {
2235                 mdk_rdev_t *rdev2;
2236                 /* Activating a spare .. or possibly reactivating
2237                  * if we ever get bitmaps working here.
2238                  */
2239
2240                 if (rdev->raid_disk != -1)
2241                         return -EBUSY;
2242
2243                 if (rdev->mddev->pers->hot_add_disk == NULL)
2244                         return -EINVAL;
2245
2246                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2247                         if (rdev2->raid_disk == slot)
2248                                 return -EEXIST;
2249
2250                 rdev->raid_disk = slot;
2251                 if (test_bit(In_sync, &rdev->flags))
2252                         rdev->saved_raid_disk = slot;
2253                 else
2254                         rdev->saved_raid_disk = -1;
2255                 err = rdev->mddev->pers->
2256                         hot_add_disk(rdev->mddev, rdev);
2257                 if (err) {
2258                         rdev->raid_disk = -1;
2259                         return err;
2260                 } else
2261                         sysfs_notify_dirent(rdev->sysfs_state);
2262                 sprintf(nm, "rd%d", rdev->raid_disk);
2263                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2264                         printk(KERN_WARNING
2265                                "md: cannot register "
2266                                "%s for %s\n",
2267                                nm, mdname(rdev->mddev));
2268
2269                 /* don't wakeup anyone, leave that to userspace. */
2270         } else {
2271                 if (slot >= rdev->mddev->raid_disks)
2272                         return -ENOSPC;
2273                 rdev->raid_disk = slot;
2274                 /* assume it is working */
2275                 clear_bit(Faulty, &rdev->flags);
2276                 clear_bit(WriteMostly, &rdev->flags);
2277                 set_bit(In_sync, &rdev->flags);
2278                 sysfs_notify_dirent(rdev->sysfs_state);
2279         }
2280         return len;
2281 }
2282
2283
2284 static struct rdev_sysfs_entry rdev_slot =
2285 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2286
2287 static ssize_t
2288 offset_show(mdk_rdev_t *rdev, char *page)
2289 {
2290         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2291 }
2292
2293 static ssize_t
2294 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2295 {
2296         char *e;
2297         unsigned long long offset = simple_strtoull(buf, &e, 10);
2298         if (e==buf || (*e && *e != '\n'))
2299                 return -EINVAL;
2300         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2301                 return -EBUSY;
2302         if (rdev->sectors && rdev->mddev->external)
2303                 /* Must set offset before size, so overlap checks
2304                  * can be sane */
2305                 return -EBUSY;
2306         rdev->data_offset = offset;
2307         return len;
2308 }
2309
2310 static struct rdev_sysfs_entry rdev_offset =
2311 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2312
2313 static ssize_t
2314 rdev_size_show(mdk_rdev_t *rdev, char *page)
2315 {
2316         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2317 }
2318
2319 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2320 {
2321         /* check if two start/length pairs overlap */
2322         if (s1+l1 <= s2)
2323                 return 0;
2324         if (s2+l2 <= s1)
2325                 return 0;
2326         return 1;
2327 }
2328
2329 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2330 {
2331         unsigned long long blocks;
2332         sector_t new;
2333
2334         if (strict_strtoull(buf, 10, &blocks) < 0)
2335                 return -EINVAL;
2336
2337         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2338                 return -EINVAL; /* sector conversion overflow */
2339
2340         new = blocks * 2;
2341         if (new != blocks * 2)
2342                 return -EINVAL; /* unsigned long long to sector_t overflow */
2343
2344         *sectors = new;
2345         return 0;
2346 }
2347
2348 static ssize_t
2349 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2350 {
2351         mddev_t *my_mddev = rdev->mddev;
2352         sector_t oldsectors = rdev->sectors;
2353         sector_t sectors;
2354
2355         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2356                 return -EINVAL;
2357         if (my_mddev->pers && rdev->raid_disk >= 0) {
2358                 if (my_mddev->persistent) {
2359                         sectors = super_types[my_mddev->major_version].
2360                                 rdev_size_change(rdev, sectors);
2361                         if (!sectors)
2362                                 return -EBUSY;
2363                 } else if (!sectors)
2364                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2365                                 rdev->data_offset;
2366         }
2367         if (sectors < my_mddev->dev_sectors)
2368                 return -EINVAL; /* component must fit device */
2369
2370         rdev->sectors = sectors;
2371         if (sectors > oldsectors && my_mddev->external) {
2372                 /* need to check that all other rdevs with the same ->bdev
2373                  * do not overlap.  We need to unlock the mddev to avoid
2374                  * a deadlock.  We have already changed rdev->sectors, and if
2375                  * we have to change it back, we will have the lock again.
2376                  */
2377                 mddev_t *mddev;
2378                 int overlap = 0;
2379                 struct list_head *tmp;
2380
2381                 mddev_unlock(my_mddev);
2382                 for_each_mddev(mddev, tmp) {
2383                         mdk_rdev_t *rdev2;
2384
2385                         mddev_lock(mddev);
2386                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2387                                 if (test_bit(AllReserved, &rdev2->flags) ||
2388                                     (rdev->bdev == rdev2->bdev &&
2389                                      rdev != rdev2 &&
2390                                      overlaps(rdev->data_offset, rdev->sectors,
2391                                               rdev2->data_offset,
2392                                               rdev2->sectors))) {
2393                                         overlap = 1;
2394                                         break;
2395                                 }
2396                         mddev_unlock(mddev);
2397                         if (overlap) {
2398                                 mddev_put(mddev);
2399                                 break;
2400                         }
2401                 }
2402                 mddev_lock(my_mddev);
2403                 if (overlap) {
2404                         /* Someone else could have slipped in a size
2405                          * change here, but doing so is just silly.
2406                          * We put oldsectors back because we *know* it is
2407                          * safe, and trust userspace not to race with
2408                          * itself
2409                          */
2410                         rdev->sectors = oldsectors;
2411                         return -EBUSY;
2412                 }
2413         }
2414         return len;
2415 }
2416
2417 static struct rdev_sysfs_entry rdev_size =
2418 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2419
2420 static struct attribute *rdev_default_attrs[] = {
2421         &rdev_state.attr,
2422         &rdev_errors.attr,
2423         &rdev_slot.attr,
2424         &rdev_offset.attr,
2425         &rdev_size.attr,
2426         NULL,
2427 };
2428 static ssize_t
2429 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2430 {
2431         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2432         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2433         mddev_t *mddev = rdev->mddev;
2434         ssize_t rv;
2435
2436         if (!entry->show)
2437                 return -EIO;
2438
2439         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2440         if (!rv) {
2441                 if (rdev->mddev == NULL)
2442                         rv = -EBUSY;
2443                 else
2444                         rv = entry->show(rdev, page);
2445                 mddev_unlock(mddev);
2446         }
2447         return rv;
2448 }
2449
2450 static ssize_t
2451 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2452               const char *page, size_t length)
2453 {
2454         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2455         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2456         ssize_t rv;
2457         mddev_t *mddev = rdev->mddev;
2458
2459         if (!entry->store)
2460                 return -EIO;
2461         if (!capable(CAP_SYS_ADMIN))
2462                 return -EACCES;
2463         rv = mddev ? mddev_lock(mddev): -EBUSY;
2464         if (!rv) {
2465                 if (rdev->mddev == NULL)
2466                         rv = -EBUSY;
2467                 else
2468                         rv = entry->store(rdev, page, length);
2469                 mddev_unlock(mddev);
2470         }
2471         return rv;
2472 }
2473
2474 static void rdev_free(struct kobject *ko)
2475 {
2476         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2477         kfree(rdev);
2478 }
2479 static struct sysfs_ops rdev_sysfs_ops = {
2480         .show           = rdev_attr_show,
2481         .store          = rdev_attr_store,
2482 };
2483 static struct kobj_type rdev_ktype = {
2484         .release        = rdev_free,
2485         .sysfs_ops      = &rdev_sysfs_ops,
2486         .default_attrs  = rdev_default_attrs,
2487 };
2488
2489 /*
2490  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2491  *
2492  * mark the device faulty if:
2493  *
2494  *   - the device is nonexistent (zero size)
2495  *   - the device has no valid superblock
2496  *
2497  * a faulty rdev _never_ has rdev->sb set.
2498  */
2499 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2500 {
2501         char b[BDEVNAME_SIZE];
2502         int err;
2503         mdk_rdev_t *rdev;
2504         sector_t size;
2505
2506         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2507         if (!rdev) {
2508                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2509                 return ERR_PTR(-ENOMEM);
2510         }
2511
2512         if ((err = alloc_disk_sb(rdev)))
2513                 goto abort_free;
2514
2515         err = lock_rdev(rdev, newdev, super_format == -2);
2516         if (err)
2517                 goto abort_free;
2518
2519         kobject_init(&rdev->kobj, &rdev_ktype);
2520
2521         rdev->desc_nr = -1;
2522         rdev->saved_raid_disk = -1;
2523         rdev->raid_disk = -1;
2524         rdev->flags = 0;
2525         rdev->data_offset = 0;
2526         rdev->sb_events = 0;
2527         atomic_set(&rdev->nr_pending, 0);
2528         atomic_set(&rdev->read_errors, 0);
2529         atomic_set(&rdev->corrected_errors, 0);
2530
2531         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2532         if (!size) {
2533                 printk(KERN_WARNING 
2534                         "md: %s has zero or unknown size, marking faulty!\n",
2535                         bdevname(rdev->bdev,b));
2536                 err = -EINVAL;
2537                 goto abort_free;
2538         }
2539
2540         if (super_format >= 0) {
2541                 err = super_types[super_format].
2542                         load_super(rdev, NULL, super_minor);
2543                 if (err == -EINVAL) {
2544                         printk(KERN_WARNING
2545                                 "md: %s does not have a valid v%d.%d "
2546                                "superblock, not importing!\n",
2547                                 bdevname(rdev->bdev,b),
2548                                super_format, super_minor);
2549                         goto abort_free;
2550                 }
2551                 if (err < 0) {
2552                         printk(KERN_WARNING 
2553                                 "md: could not read %s's sb, not importing!\n",
2554                                 bdevname(rdev->bdev,b));
2555                         goto abort_free;
2556                 }
2557         }
2558
2559         INIT_LIST_HEAD(&rdev->same_set);
2560         init_waitqueue_head(&rdev->blocked_wait);
2561
2562         return rdev;
2563
2564 abort_free:
2565         if (rdev->sb_page) {
2566                 if (rdev->bdev)
2567                         unlock_rdev(rdev);
2568                 free_disk_sb(rdev);
2569         }
2570         kfree(rdev);
2571         return ERR_PTR(err);
2572 }
2573
2574 /*
2575  * Check a full RAID array for plausibility
2576  */
2577
2578
2579 static void analyze_sbs(mddev_t * mddev)
2580 {
2581         int i;
2582         mdk_rdev_t *rdev, *freshest, *tmp;
2583         char b[BDEVNAME_SIZE];
2584
2585         freshest = NULL;
2586         rdev_for_each(rdev, tmp, mddev)
2587                 switch (super_types[mddev->major_version].
2588                         load_super(rdev, freshest, mddev->minor_version)) {
2589                 case 1:
2590                         freshest = rdev;
2591                         break;
2592                 case 0:
2593                         break;
2594                 default:
2595                         printk( KERN_ERR \
2596                                 "md: fatal superblock inconsistency in %s"
2597                                 " -- removing from array\n", 
2598                                 bdevname(rdev->bdev,b));
2599                         kick_rdev_from_array(rdev);
2600                 }
2601
2602
2603         super_types[mddev->major_version].
2604                 validate_super(mddev, freshest);
2605
2606         i = 0;
2607         rdev_for_each(rdev, tmp, mddev) {
2608                 if (rdev->desc_nr >= mddev->max_disks ||
2609                     i > mddev->max_disks) {
2610                         printk(KERN_WARNING
2611                                "md: %s: %s: only %d devices permitted\n",
2612                                mdname(mddev), bdevname(rdev->bdev, b),
2613                                mddev->max_disks);
2614                         kick_rdev_from_array(rdev);
2615                         continue;
2616                 }
2617                 if (rdev != freshest)
2618                         if (super_types[mddev->major_version].
2619                             validate_super(mddev, rdev)) {
2620                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2621                                         " from array!\n",
2622                                         bdevname(rdev->bdev,b));
2623                                 kick_rdev_from_array(rdev);
2624                                 continue;
2625                         }
2626                 if (mddev->level == LEVEL_MULTIPATH) {
2627                         rdev->desc_nr = i++;
2628                         rdev->raid_disk = rdev->desc_nr;
2629                         set_bit(In_sync, &rdev->flags);
2630                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
2631                         rdev->raid_disk = -1;
2632                         clear_bit(In_sync, &rdev->flags);
2633                 }
2634         }
2635 }
2636
2637 static void md_safemode_timeout(unsigned long data);
2638
2639 static ssize_t
2640 safe_delay_show(mddev_t *mddev, char *page)
2641 {
2642         int msec = (mddev->safemode_delay*1000)/HZ;
2643         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2644 }
2645 static ssize_t
2646 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2647 {
2648         int scale=1;
2649         int dot=0;
2650         int i;
2651         unsigned long msec;
2652         char buf[30];
2653
2654         /* remove a period, and count digits after it */
2655         if (len >= sizeof(buf))
2656                 return -EINVAL;
2657         strlcpy(buf, cbuf, sizeof(buf));
2658         for (i=0; i<len; i++) {
2659                 if (dot) {
2660                         if (isdigit(buf[i])) {
2661                                 buf[i-1] = buf[i];
2662                                 scale *= 10;
2663                         }
2664                         buf[i] = 0;
2665                 } else if (buf[i] == '.') {
2666                         dot=1;
2667                         buf[i] = 0;
2668                 }
2669         }
2670         if (strict_strtoul(buf, 10, &msec) < 0)
2671                 return -EINVAL;
2672         msec = (msec * 1000) / scale;
2673         if (msec == 0)
2674                 mddev->safemode_delay = 0;
2675         else {
2676                 unsigned long old_delay = mddev->safemode_delay;
2677                 mddev->safemode_delay = (msec*HZ)/1000;
2678                 if (mddev->safemode_delay == 0)
2679                         mddev->safemode_delay = 1;
2680                 if (mddev->safemode_delay < old_delay)
2681                         md_safemode_timeout((unsigned long)mddev);
2682         }
2683         return len;
2684 }
2685 static struct md_sysfs_entry md_safe_delay =
2686 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2687
2688 static ssize_t
2689 level_show(mddev_t *mddev, char *page)
2690 {
2691         struct mdk_personality *p = mddev->pers;
2692         if (p)
2693                 return sprintf(page, "%s\n", p->name);
2694         else if (mddev->clevel[0])
2695                 return sprintf(page, "%s\n", mddev->clevel);
2696         else if (mddev->level != LEVEL_NONE)
2697                 return sprintf(page, "%d\n", mddev->level);
2698         else
2699                 return 0;
2700 }
2701
2702 static ssize_t
2703 level_store(mddev_t *mddev, const char *buf, size_t len)
2704 {
2705         char level[16];
2706         ssize_t rv = len;
2707         struct mdk_personality *pers;
2708         void *priv;
2709         mdk_rdev_t *rdev;
2710
2711         if (mddev->pers == NULL) {
2712                 if (len == 0)
2713                         return 0;
2714                 if (len >= sizeof(mddev->clevel))
2715                         return -ENOSPC;
2716                 strncpy(mddev->clevel, buf, len);
2717                 if (mddev->clevel[len-1] == '\n')
2718                         len--;
2719                 mddev->clevel[len] = 0;
2720                 mddev->level = LEVEL_NONE;
2721                 return rv;
2722         }
2723
2724         /* request to change the personality.  Need to ensure:
2725          *  - array is not engaged in resync/recovery/reshape
2726          *  - old personality can be suspended
2727          *  - new personality will access other array.
2728          */
2729
2730         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2731                 return -EBUSY;
2732
2733         if (!mddev->pers->quiesce) {
2734                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2735                        mdname(mddev), mddev->pers->name);
2736                 return -EINVAL;
2737         }
2738
2739         /* Now find the new personality */
2740         if (len == 0 || len >= sizeof(level))
2741                 return -EINVAL;
2742         strncpy(level, buf, len);
2743         if (level[len-1] == '\n')
2744                 len--;
2745         level[len] = 0;
2746
2747         request_module("md-%s", level);
2748         spin_lock(&pers_lock);
2749         pers = find_pers(LEVEL_NONE, level);
2750         if (!pers || !try_module_get(pers->owner)) {
2751                 spin_unlock(&pers_lock);
2752                 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2753                 return -EINVAL;
2754         }
2755         spin_unlock(&pers_lock);
2756
2757         if (pers == mddev->pers) {
2758                 /* Nothing to do! */
2759                 module_put(pers->owner);
2760                 return rv;
2761         }
2762         if (!pers->takeover) {
2763                 module_put(pers->owner);
2764                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2765                        mdname(mddev), level);
2766                 return -EINVAL;
2767         }
2768
2769         /* ->takeover must set new_* and/or delta_disks
2770          * if it succeeds, and may set them when it fails.
2771          */
2772         priv = pers->takeover(mddev);
2773         if (IS_ERR(priv)) {
2774                 mddev->new_level = mddev->level;
2775                 mddev->new_layout = mddev->layout;
2776                 mddev->new_chunk_sectors = mddev->chunk_sectors;
2777                 mddev->raid_disks -= mddev->delta_disks;
2778                 mddev->delta_disks = 0;
2779                 module_put(pers->owner);
2780                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2781                        mdname(mddev), level);
2782                 return PTR_ERR(priv);
2783         }
2784
2785         /* Looks like we have a winner */
2786         mddev_suspend(mddev);
2787         mddev->pers->stop(mddev);
2788         module_put(mddev->pers->owner);
2789         /* Invalidate devices that are now superfluous */
2790         list_for_each_entry(rdev, &mddev->disks, same_set)
2791                 if (rdev->raid_disk >= mddev->raid_disks) {
2792                         rdev->raid_disk = -1;
2793                         clear_bit(In_sync, &rdev->flags);
2794                 }
2795         mddev->pers = pers;
2796         mddev->private = priv;
2797         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2798         mddev->level = mddev->new_level;
2799         mddev->layout = mddev->new_layout;
2800         mddev->chunk_sectors = mddev->new_chunk_sectors;
2801         mddev->delta_disks = 0;
2802         pers->run(mddev);
2803         mddev_resume(mddev);
2804         set_bit(MD_CHANGE_DEVS, &mddev->flags);
2805         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2806         md_wakeup_thread(mddev->thread);
2807         return rv;
2808 }
2809
2810 static struct md_sysfs_entry md_level =
2811 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2812
2813
2814 static ssize_t
2815 layout_show(mddev_t *mddev, char *page)
2816 {
2817         /* just a number, not meaningful for all levels */
2818         if (mddev->reshape_position != MaxSector &&
2819             mddev->layout != mddev->new_layout)
2820                 return sprintf(page, "%d (%d)\n",
2821                                mddev->new_layout, mddev->layout);
2822         return sprintf(page, "%d\n", mddev->layout);
2823 }
2824
2825 static ssize_t
2826 layout_store(mddev_t *mddev, const char *buf, size_t len)
2827 {
2828         char *e;
2829         unsigned long n = simple_strtoul(buf, &e, 10);
2830
2831         if (!*buf || (*e && *e != '\n'))
2832                 return -EINVAL;
2833
2834         if (mddev->pers) {
2835                 int err;
2836                 if (mddev->pers->check_reshape == NULL)
2837                         return -EBUSY;
2838                 mddev->new_layout = n;
2839                 err = mddev->pers->check_reshape(mddev);
2840                 if (err) {
2841                         mddev->new_layout = mddev->layout;
2842                         return err;
2843                 }
2844         } else {
2845                 mddev->new_layout = n;
2846                 if (mddev->reshape_position == MaxSector)
2847                         mddev->layout = n;
2848         }
2849         return len;
2850 }
2851 static struct md_sysfs_entry md_layout =
2852 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2853
2854
2855 static ssize_t
2856 raid_disks_show(mddev_t *mddev, char *page)
2857 {
2858         if (mddev->raid_disks == 0)
2859                 return 0;
2860         if (mddev->reshape_position != MaxSector &&
2861             mddev->delta_disks != 0)
2862                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2863                                mddev->raid_disks - mddev->delta_disks);
2864         return sprintf(page, "%d\n", mddev->raid_disks);
2865 }
2866
2867 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2868
2869 static ssize_t
2870 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2871 {
2872         char *e;
2873         int rv = 0;
2874         unsigned long n = simple_strtoul(buf, &e, 10);
2875
2876         if (!*buf || (*e && *e != '\n'))
2877                 return -EINVAL;
2878
2879         if (mddev->pers)
2880                 rv = update_raid_disks(mddev, n);
2881         else if (mddev->reshape_position != MaxSector) {
2882                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2883                 mddev->delta_disks = n - olddisks;
2884                 mddev->raid_disks = n;
2885         } else
2886                 mddev->raid_disks = n;
2887         return rv ? rv : len;
2888 }
2889 static struct md_sysfs_entry md_raid_disks =
2890 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2891
2892 static ssize_t
2893 chunk_size_show(mddev_t *mddev, char *page)
2894 {
2895         if (mddev->reshape_position != MaxSector &&
2896             mddev->chunk_sectors != mddev->new_chunk_sectors)
2897                 return sprintf(page, "%d (%d)\n",
2898                                mddev->new_chunk_sectors << 9,
2899                                mddev->chunk_sectors << 9);
2900         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2901 }
2902
2903 static ssize_t
2904 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2905 {
2906         char *e;
2907         unsigned long n = simple_strtoul(buf, &e, 10);
2908
2909         if (!*buf || (*e && *e != '\n'))
2910                 return -EINVAL;
2911
2912         if (mddev->pers) {
2913                 int err;
2914                 if (mddev->pers->check_reshape == NULL)
2915                         return -EBUSY;
2916                 mddev->new_chunk_sectors = n >> 9;
2917                 err = mddev->pers->check_reshape(mddev);
2918                 if (err) {
2919                         mddev->new_chunk_sectors = mddev->chunk_sectors;
2920                         return err;
2921                 }
2922         } else {
2923                 mddev->new_chunk_sectors = n >> 9;
2924                 if (mddev->reshape_position == MaxSector)
2925                         mddev->chunk_sectors = n >> 9;
2926         }
2927         return len;
2928 }
2929 static struct md_sysfs_entry md_chunk_size =
2930 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2931
2932 static ssize_t
2933 resync_start_show(mddev_t *mddev, char *page)
2934 {
2935         if (mddev->recovery_cp == MaxSector)
2936                 return sprintf(page, "none\n");
2937         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2938 }
2939
2940 static ssize_t
2941 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2942 {
2943         char *e;
2944         unsigned long long n = simple_strtoull(buf, &e, 10);
2945
2946         if (mddev->pers)
2947                 return -EBUSY;
2948         if (!*buf || (*e && *e != '\n'))
2949                 return -EINVAL;
2950
2951         mddev->recovery_cp = n;
2952         return len;
2953 }
2954 static struct md_sysfs_entry md_resync_start =
2955 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2956
2957 /*
2958  * The array state can be:
2959  *
2960  * clear
2961  *     No devices, no size, no level
2962  *     Equivalent to STOP_ARRAY ioctl
2963  * inactive
2964  *     May have some settings, but array is not active
2965  *        all IO results in error
2966  *     When written, doesn't tear down array, but just stops it
2967  * suspended (not supported yet)
2968  *     All IO requests will block. The array can be reconfigured.
2969  *     Writing this, if accepted, will block until array is quiescent
2970  * readonly
2971  *     no resync can happen.  no superblocks get written.
2972  *     write requests fail
2973  * read-auto
2974  *     like readonly, but behaves like 'clean' on a write request.
2975  *
2976  * clean - no pending writes, but otherwise active.
2977  *     When written to inactive array, starts without resync
2978  *     If a write request arrives then
2979  *       if metadata is known, mark 'dirty' and switch to 'active'.
2980  *       if not known, block and switch to write-pending
2981  *     If written to an active array that has pending writes, then fails.
2982  * active
2983  *     fully active: IO and resync can be happening.
2984  *     When written to inactive array, starts with resync
2985  *
2986  * write-pending
2987  *     clean, but writes are blocked waiting for 'active' to be written.
2988  *
2989  * active-idle
2990  *     like active, but no writes have been seen for a while (100msec).
2991  *
2992  */
2993 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2994                    write_pending, active_idle, bad_word};
2995 static char *array_states[] = {
2996         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2997         "write-pending", "active-idle", NULL };
2998
2999 static int match_word(const char *word, char **list)
3000 {
3001         int n;
3002         for (n=0; list[n]; n++)
3003                 if (cmd_match(word, list[n]))
3004                         break;
3005         return n;
3006 }
3007
3008 static ssize_t
3009 array_state_show(mddev_t *mddev, char *page)
3010 {
3011         enum array_state st = inactive;
3012
3013         if (mddev->pers)
3014                 switch(mddev->ro) {
3015                 case 1:
3016                         st = readonly;
3017                         break;
3018                 case 2:
3019                         st = read_auto;
3020                         break;
3021                 case 0:
3022                         if (mddev->in_sync)
3023                                 st = clean;
3024                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3025                                 st = write_pending;
3026                         else if (mddev->safemode)
3027                                 st = active_idle;
3028                         else
3029                                 st = active;
3030                 }
3031         else {
3032                 if (list_empty(&mddev->disks) &&
3033                     mddev->raid_disks == 0 &&
3034                     mddev->dev_sectors == 0)
3035                         st = clear;
3036                 else
3037                         st = inactive;
3038         }
3039         return sprintf(page, "%s\n", array_states[st]);
3040 }
3041
3042 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3043 static int do_md_run(mddev_t * mddev);
3044 static int restart_array(mddev_t *mddev);
3045
3046 static ssize_t
3047 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3048 {
3049         int err = -EINVAL;
3050         enum array_state st = match_word(buf, array_states);
3051         switch(st) {
3052         case bad_word:
3053                 break;
3054         case clear:
3055                 /* stopping an active array */
3056                 if (atomic_read(&mddev->openers) > 0)
3057                         return -EBUSY;
3058                 err = do_md_stop(mddev, 0, 0);
3059                 break;
3060         case inactive:
3061                 /* stopping an active array */
3062                 if (mddev->pers) {
3063                         if (atomic_read(&mddev->openers) > 0)
3064                                 return -EBUSY;
3065                         err = do_md_stop(mddev, 2, 0);
3066                 } else
3067                         err = 0; /* already inactive */
3068                 break;
3069         case suspended:
3070                 break; /* not supported yet */
3071         case readonly:
3072                 if (mddev->pers)
3073                         err = do_md_stop(mddev, 1, 0);
3074                 else {
3075                         mddev->ro = 1;
3076                         set_disk_ro(mddev->gendisk, 1);
3077                         err = do_md_run(mddev);
3078                 }
3079                 break;
3080         case read_auto:
3081                 if (mddev->pers) {
3082                         if (mddev->ro == 0)
3083                                 err = do_md_stop(mddev, 1, 0);
3084                         else if (mddev->ro == 1)
3085                                 err = restart_array(mddev);
3086                         if (err == 0) {
3087                                 mddev->ro = 2;
3088                                 set_disk_ro(mddev->gendisk, 0);
3089                         }
3090                 } else {
3091                         mddev->ro = 2;
3092                         err = do_md_run(mddev);
3093                 }
3094                 break;
3095         case clean:
3096                 if (mddev->pers) {
3097                         restart_array(mddev);
3098                         spin_lock_irq(&mddev->write_lock);
3099                         if (atomic_read(&mddev->writes_pending) == 0) {
3100                                 if (mddev->in_sync == 0) {
3101                                         mddev->in_sync = 1;
3102                                         if (mddev->safemode == 1)
3103                                                 mddev->safemode = 0;
3104                                         if (mddev->persistent)
3105                                                 set_bit(MD_CHANGE_CLEAN,
3106                                                         &mddev->flags);
3107                                 }
3108                                 err = 0;
3109                         } else
3110                                 err = -EBUSY;
3111                         spin_unlock_irq(&mddev->write_lock);
3112                 } else
3113                         err = -EINVAL;
3114                 break;
3115         case active:
3116                 if (mddev->pers) {
3117                         restart_array(mddev);
3118                         if (mddev->external)
3119                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3120                         wake_up(&mddev->sb_wait);
3121                         err = 0;
3122                 } else {
3123                         mddev->ro = 0;
3124                         set_disk_ro(mddev->gendisk, 0);
3125                         err = do_md_run(mddev);
3126                 }
3127                 break;
3128         case write_pending:
3129         case active_idle:
3130                 /* these cannot be set */
3131                 break;
3132         }
3133         if (err)
3134                 return err;
3135         else {
3136                 sysfs_notify_dirent(mddev->sysfs_state);
3137                 return len;
3138         }
3139 }
3140 static struct md_sysfs_entry md_array_state =
3141 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3142
3143 static ssize_t
3144 null_show(mddev_t *mddev, char *page)
3145 {
3146         return -EINVAL;
3147 }
3148
3149 static ssize_t
3150 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3151 {
3152         /* buf must be %d:%d\n? giving major and minor numbers */
3153         /* The new device is added to the array.
3154          * If the array has a persistent superblock, we read the
3155          * superblock to initialise info and check validity.
3156          * Otherwise, only checking done is that in bind_rdev_to_array,
3157          * which mainly checks size.
3158          */
3159         char *e;
3160         int major = simple_strtoul(buf, &e, 10);
3161         int minor;
3162         dev_t dev;
3163         mdk_rdev_t *rdev;
3164         int err;
3165
3166         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3167                 return -EINVAL;
3168         minor = simple_strtoul(e+1, &e, 10);
3169         if (*e && *e != '\n')
3170                 return -EINVAL;
3171         dev = MKDEV(major, minor);
3172         if (major != MAJOR(dev) ||
3173             minor != MINOR(dev))
3174                 return -EOVERFLOW;
3175
3176
3177         if (mddev->persistent) {
3178                 rdev = md_import_device(dev, mddev->major_version,
3179                                         mddev->minor_version);
3180                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3181                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3182                                                        mdk_rdev_t, same_set);
3183                         err = super_types[mddev->major_version]
3184                                 .load_super(rdev, rdev0, mddev->minor_version);
3185                         if (err < 0)
3186                                 goto out;
3187                 }
3188         } else if (mddev->external)
3189                 rdev = md_import_device(dev, -2, -1);
3190         else
3191                 rdev = md_import_device(dev, -1, -1);
3192
3193         if (IS_ERR(rdev))
3194                 return PTR_ERR(rdev);
3195         err = bind_rdev_to_array(rdev, mddev);
3196  out:
3197         if (err)
3198                 export_rdev(rdev);
3199         return err ? err : len;
3200 }
3201
3202 static struct md_sysfs_entry md_new_device =
3203 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3204
3205 static ssize_t
3206 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3207 {
3208         char *end;
3209         unsigned long chunk, end_chunk;
3210
3211         if (!mddev->bitmap)
3212                 goto out;
3213         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3214         while (*buf) {
3215                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3216                 if (buf == end) break;
3217                 if (*end == '-') { /* range */
3218                         buf = end + 1;
3219                         end_chunk = simple_strtoul(buf, &end, 0);
3220                         if (buf == end) break;
3221                 }
3222                 if (*end && !isspace(*end)) break;
3223                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3224                 buf = end;
3225                 while (isspace(*buf)) buf++;
3226         }
3227         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3228 out:
3229         return len;
3230 }
3231
3232 static struct md_sysfs_entry md_bitmap =
3233 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3234
3235 static ssize_t
3236 size_show(mddev_t *mddev, char *page)
3237 {
3238         return sprintf(page, "%llu\n",
3239                 (unsigned long long)mddev->dev_sectors / 2);
3240 }
3241
3242 static int update_size(mddev_t *mddev, sector_t num_sectors);
3243
3244 static ssize_t
3245 size_store(mddev_t *mddev, const char *buf, size_t len)
3246 {
3247         /* If array is inactive, we can reduce the component size, but
3248          * not increase it (except from 0).
3249          * If array is active, we can try an on-line resize
3250          */
3251         sector_t sectors;
3252         int err = strict_blocks_to_sectors(buf, &sectors);
3253
3254         if (err < 0)
3255                 return err;
3256         if (mddev->pers) {
3257                 err = update_size(mddev, sectors);
3258                 md_update_sb(mddev, 1);
3259         } else {
3260                 if (mddev->dev_sectors == 0 ||
3261                     mddev->dev_sectors > sectors)
3262                         mddev->dev_sectors = sectors;
3263                 else
3264                         err = -ENOSPC;
3265         }
3266         return err ? err : len;
3267 }
3268
3269 static struct md_sysfs_entry md_size =
3270 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3271
3272
3273 /* Metdata version.
3274  * This is one of
3275  *   'none' for arrays with no metadata (good luck...)
3276  *   'external' for arrays with externally managed metadata,
3277  * or N.M for internally known formats
3278  */
3279 static ssize_t
3280 metadata_show(mddev_t *mddev, char *page)
3281 {
3282         if (mddev->persistent)
3283                 return sprintf(page, "%d.%d\n",
3284                                mddev->major_version, mddev->minor_version);
3285         else if (mddev->external)
3286                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3287         else
3288                 return sprintf(page, "none\n");
3289 }
3290
3291 static ssize_t
3292 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3293 {
3294         int major, minor;
3295         char *e;
3296         /* Changing the details of 'external' metadata is
3297          * always permitted.  Otherwise there must be
3298          * no devices attached to the array.
3299          */
3300         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3301                 ;
3302         else if (!list_empty(&mddev->disks))
3303                 return -EBUSY;
3304
3305         if (cmd_match(buf, "none")) {
3306                 mddev->persistent = 0;
3307                 mddev->external = 0;
3308                 mddev->major_version = 0;
3309                 mddev->minor_version = 90;
3310                 return len;
3311         }
3312         if (strncmp(buf, "external:", 9) == 0) {
3313                 size_t namelen = len-9;
3314                 if (namelen >= sizeof(mddev->metadata_type))
3315                         namelen = sizeof(mddev->metadata_type)-1;
3316                 strncpy(mddev->metadata_type, buf+9, namelen);
3317                 mddev->metadata_type[namelen] = 0;
3318                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3319                         mddev->metadata_type[--namelen] = 0;
3320                 mddev->persistent = 0;
3321                 mddev->external = 1;
3322                 mddev->major_version = 0;
3323                 mddev->minor_version = 90;
3324                 return len;
3325         }
3326         major = simple_strtoul(buf, &e, 10);
3327         if (e==buf || *e != '.')
3328                 return -EINVAL;
3329         buf = e+1;
3330         minor = simple_strtoul(buf, &e, 10);
3331         if (e==buf || (*e && *e != '\n') )
3332                 return -EINVAL;
3333         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3334                 return -ENOENT;
3335         mddev->major_version = major;
3336         mddev->minor_version = minor;
3337         mddev->persistent = 1;
3338         mddev->external = 0;
3339         return len;
3340 }
3341
3342 static struct md_sysfs_entry md_metadata =
3343 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3344
3345 static ssize_t
3346 action_show(mddev_t *mddev, char *page)
3347 {
3348         char *type = "idle";
3349         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3350                 type = "frozen";
3351         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3352             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3353                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3354                         type = "reshape";
3355                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3356                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3357                                 type = "resync";
3358                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3359                                 type = "check";
3360                         else
3361                                 type = "repair";
3362                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3363                         type = "recover";
3364         }
3365         return sprintf(page, "%s\n", type);
3366 }
3367
3368 static ssize_t
3369 action_store(mddev_t *mddev, const char *page, size_t len)
3370 {
3371         if (!mddev->pers || !mddev->pers->sync_request)
3372                 return -EINVAL;
3373
3374         if (cmd_match(page, "frozen"))
3375                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3376         else
3377                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3378
3379         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3380                 if (mddev->sync_thread) {
3381                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3382                         md_unregister_thread(mddev->sync_thread);
3383                         mddev->sync_thread = NULL;
3384                         mddev->recovery = 0;
3385                 }
3386         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3387                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3388                 return -EBUSY;
3389         else if (cmd_match(page, "resync"))
3390                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3391         else if (cmd_match(page, "recover")) {
3392                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3393                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3394         } else if (cmd_match(page, "reshape")) {
3395                 int err;
3396                 if (mddev->pers->start_reshape == NULL)
3397                         return -EINVAL;
3398                 err = mddev->pers->start_reshape(mddev);
3399                 if (err)
3400                         return err;
3401                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3402         } else {
3403                 if (cmd_match(page, "check"))
3404                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3405                 else if (!cmd_match(page, "repair"))
3406                         return -EINVAL;
3407                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3408                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3409         }
3410         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3411         md_wakeup_thread(mddev->thread);
3412         sysfs_notify_dirent(mddev->sysfs_action);
3413         return len;
3414 }
3415
3416 static ssize_t
3417 mismatch_cnt_show(mddev_t *mddev, char *page)
3418 {
3419         return sprintf(page, "%llu\n",
3420                        (unsigned long long) mddev->resync_mismatches);
3421 }
3422
3423 static struct md_sysfs_entry md_scan_mode =
3424 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3425
3426
3427 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3428
3429 static ssize_t
3430 sync_min_show(mddev_t *mddev, char *page)
3431 {
3432         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3433                        mddev->sync_speed_min ? "local": "system");
3434 }
3435
3436 static ssize_t
3437 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3438 {
3439         int min;
3440         char *e;
3441         if (strncmp(buf, "system", 6)==0) {
3442                 mddev->sync_speed_min = 0;
3443                 return len;
3444         }
3445         min = simple_strtoul(buf, &e, 10);
3446         if (buf == e || (*e && *e != '\n') || min <= 0)
3447                 return -EINVAL;
3448         mddev->sync_speed_min = min;
3449         return len;
3450 }
3451
3452 static struct md_sysfs_entry md_sync_min =
3453 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3454
3455 static ssize_t
3456 sync_max_show(mddev_t *mddev, char *page)
3457 {
3458         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3459                        mddev->sync_speed_max ? "local": "system");
3460 }
3461
3462 static ssize_t
3463 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3464 {
3465         int max;
3466         char *e;
3467         if (strncmp(buf, "system", 6)==0) {
3468                 mddev->sync_speed_max = 0;
3469                 return len;
3470         }
3471         max = simple_strtoul(buf, &e, 10);
3472         if (buf == e || (*e && *e != '\n') || max <= 0)
3473                 return -EINVAL;
3474         mddev->sync_speed_max = max;
3475         return len;
3476 }
3477
3478 static struct md_sysfs_entry md_sync_max =
3479 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3480
3481 static ssize_t
3482 degraded_show(mddev_t *mddev, char *page)
3483 {
3484         return sprintf(page, "%d\n", mddev->degraded);
3485 }
3486 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3487
3488 static ssize_t
3489 sync_force_parallel_show(mddev_t *mddev, char *page)
3490 {
3491         return sprintf(page, "%d\n", mddev->parallel_resync);
3492 }
3493
3494 static ssize_t
3495 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3496 {
3497         long n;
3498
3499         if (strict_strtol(buf, 10, &n))
3500                 return -EINVAL;
3501
3502         if (n != 0 && n != 1)
3503                 return -EINVAL;
3504
3505         mddev->parallel_resync = n;
3506
3507         if (mddev->sync_thread)
3508                 wake_up(&resync_wait);
3509
3510         return len;
3511 }
3512
3513 /* force parallel resync, even with shared block devices */
3514 static struct md_sysfs_entry md_sync_force_parallel =
3515 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3516        sync_force_parallel_show, sync_force_parallel_store);
3517
3518 static ssize_t
3519 sync_speed_show(mddev_t *mddev, char *page)
3520 {
3521         unsigned long resync, dt, db;
3522         if (mddev->curr_resync == 0)
3523                 return sprintf(page, "none\n");
3524         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3525         dt = (jiffies - mddev->resync_mark) / HZ;
3526         if (!dt) dt++;
3527         db = resync - mddev->resync_mark_cnt;
3528         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3529 }
3530
3531 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3532
3533 static ssize_t
3534 sync_completed_show(mddev_t *mddev, char *page)
3535 {
3536         unsigned long max_sectors, resync;
3537
3538         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3539                 return sprintf(page, "none\n");
3540
3541         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3542                 max_sectors = mddev->resync_max_sectors;
3543         else
3544                 max_sectors = mddev->dev_sectors;
3545
3546         resync = mddev->curr_resync_completed;
3547         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3548 }
3549
3550 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3551
3552 static ssize_t
3553 min_sync_show(mddev_t *mddev, char *page)
3554 {
3555         return sprintf(page, "%llu\n",
3556                        (unsigned long long)mddev->resync_min);
3557 }
3558 static ssize_t
3559 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3560 {
3561         unsigned long long min;
3562         if (strict_strtoull(buf, 10, &min))
3563                 return -EINVAL;
3564         if (min > mddev->resync_max)
3565                 return -EINVAL;
3566         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3567                 return -EBUSY;
3568
3569         /* Must be a multiple of chunk_size */
3570         if (mddev->chunk_sectors) {
3571                 sector_t temp = min;
3572                 if (sector_div(temp, mddev->chunk_sectors))
3573                         return -EINVAL;
3574         }
3575         mddev->resync_min = min;
3576
3577         return len;
3578 }
3579
3580 static struct md_sysfs_entry md_min_sync =
3581 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3582
3583 static ssize_t
3584 max_sync_show(mddev_t *mddev, char *page)
3585 {
3586         if (mddev->resync_max == MaxSector)
3587                 return sprintf(page, "max\n");
3588         else
3589                 return sprintf(page, "%llu\n",
3590                                (unsigned long long)mddev->resync_max);
3591 }
3592 static ssize_t
3593 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3594 {
3595         if (strncmp(buf, "max", 3) == 0)
3596                 mddev->resync_max = MaxSector;
3597         else {
3598                 unsigned long long max;
3599                 if (strict_strtoull(buf, 10, &max))
3600                         return -EINVAL;
3601                 if (max < mddev->resync_min)
3602                         return -EINVAL;
3603                 if (max < mddev->resync_max &&
3604                     mddev->ro == 0 &&
3605                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3606                         return -EBUSY;
3607
3608                 /* Must be a multiple of chunk_size */
3609                 if (mddev->chunk_sectors) {
3610                         sector_t temp = max;
3611                         if (sector_div(temp, mddev->chunk_sectors))
3612                                 return -EINVAL;
3613                 }
3614                 mddev->resync_max = max;
3615         }
3616         wake_up(&mddev->recovery_wait);
3617         return len;
3618 }
3619
3620 static struct md_sysfs_entry md_max_sync =
3621 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3622
3623 static ssize_t
3624 suspend_lo_show(mddev_t *mddev, char *page)
3625 {
3626         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3627 }
3628
3629 static ssize_t
3630 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3631 {
3632         char *e;
3633         unsigned long long new = simple_strtoull(buf, &e, 10);
3634
3635         if (mddev->pers == NULL || 
3636             mddev->pers->quiesce == NULL)
3637                 return -EINVAL;
3638         if (buf == e || (*e && *e != '\n'))
3639                 return -EINVAL;
3640         if (new >= mddev->suspend_hi ||
3641             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3642                 mddev->suspend_lo = new;
3643                 mddev->pers->quiesce(mddev, 2);
3644                 return len;
3645         } else
3646                 return -EINVAL;
3647 }
3648 static struct md_sysfs_entry md_suspend_lo =
3649 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3650
3651
3652 static ssize_t
3653 suspend_hi_show(mddev_t *mddev, char *page)
3654 {
3655         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3656 }
3657
3658 static ssize_t
3659 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3660 {
3661         char *e;
3662         unsigned long long new = simple_strtoull(buf, &e, 10);
3663
3664         if (mddev->pers == NULL ||
3665             mddev->pers->quiesce == NULL)
3666                 return -EINVAL;
3667         if (buf == e || (*e && *e != '\n'))
3668                 return -EINVAL;
3669         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3670             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3671                 mddev->suspend_hi = new;
3672                 mddev->pers->quiesce(mddev, 1);
3673                 mddev->pers->quiesce(mddev, 0);
3674                 return len;
3675         } else
3676                 return -EINVAL;
3677 }
3678 static struct md_sysfs_entry md_suspend_hi =
3679 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3680
3681 static ssize_t
3682 reshape_position_show(mddev_t *mddev, char *page)
3683 {
3684         if (mddev->reshape_position != MaxSector)
3685                 return sprintf(page, "%llu\n",
3686                                (unsigned long long)mddev->reshape_position);
3687         strcpy(page, "none\n");
3688         return 5;
3689 }
3690
3691 static ssize_t
3692 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3693 {
3694         char *e;
3695         unsigned long long new = simple_strtoull(buf, &e, 10);
3696         if (mddev->pers)
3697                 return -EBUSY;
3698         if (buf == e || (*e && *e != '\n'))
3699                 return -EINVAL;
3700         mddev->reshape_position = new;
3701         mddev->delta_disks = 0;
3702         mddev->new_level = mddev->level;
3703         mddev->new_layout = mddev->layout;
3704         mddev->new_chunk_sectors = mddev->chunk_sectors;
3705         return len;
3706 }
3707
3708 static struct md_sysfs_entry md_reshape_position =
3709 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3710        reshape_position_store);
3711
3712 static ssize_t
3713 array_size_show(mddev_t *mddev, char *page)
3714 {
3715         if (mddev->external_size)
3716                 return sprintf(page, "%llu\n",
3717                                (unsigned long long)mddev->array_sectors/2);
3718         else
3719                 return sprintf(page, "default\n");
3720 }
3721
3722 static ssize_t
3723 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3724 {
3725         sector_t sectors;
3726
3727         if (strncmp(buf, "default", 7) == 0) {
3728                 if (mddev->pers)
3729                         sectors = mddev->pers->size(mddev, 0, 0);
3730                 else
3731                         sectors = mddev->array_sectors;
3732
3733                 mddev->external_size = 0;
3734         } else {
3735                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3736                         return -EINVAL;
3737                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3738                         return -E2BIG;
3739
3740                 mddev->external_size = 1;
3741         }
3742
3743         mddev->array_sectors = sectors;
3744         set_capacity(mddev->gendisk, mddev->array_sectors);
3745         if (mddev->pers)
3746                 revalidate_disk(mddev->gendisk);
3747
3748         return len;
3749 }
3750
3751 static struct md_sysfs_entry md_array_size =
3752 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3753        array_size_store);
3754
3755 static struct attribute *md_default_attrs[] = {
3756         &md_level.attr,
3757         &md_layout.attr,
3758         &md_raid_disks.attr,
3759         &md_chunk_size.attr,
3760         &md_size.attr,
3761         &md_resync_start.attr,
3762         &md_metadata.attr,
3763         &md_new_device.attr,
3764         &md_safe_delay.attr,
3765         &md_array_state.attr,
3766         &md_reshape_position.attr,
3767         &md_array_size.attr,
3768         NULL,
3769 };
3770
3771 static struct attribute *md_redundancy_attrs[] = {
3772         &md_scan_mode.attr,
3773         &md_mismatches.attr,
3774         &md_sync_min.attr,
3775         &md_sync_max.attr,
3776         &md_sync_speed.attr,
3777         &md_sync_force_parallel.attr,
3778         &md_sync_completed.attr,
3779         &md_min_sync.attr,
3780         &md_max_sync.attr,
3781         &md_suspend_lo.attr,
3782         &md_suspend_hi.attr,
3783         &md_bitmap.attr,
3784         &md_degraded.attr,
3785         NULL,
3786 };
3787 static struct attribute_group md_redundancy_group = {
3788         .name = NULL,
3789         .attrs = md_redundancy_attrs,
3790 };
3791
3792
3793 static ssize_t
3794 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3795 {
3796         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3797         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3798         ssize_t rv;
3799
3800         if (!entry->show)
3801                 return -EIO;
3802         rv = mddev_lock(mddev);
3803         if (!rv) {
3804                 rv = entry->show(mddev, page);
3805                 mddev_unlock(mddev);
3806         }
3807         return rv;
3808 }
3809
3810 static ssize_t
3811 md_attr_store(struct kobject *kobj, struct attribute *attr,
3812               const char *page, size_t length)
3813 {
3814         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3815         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3816         ssize_t rv;
3817
3818         if (!entry->store)
3819                 return -EIO;
3820         if (!capable(CAP_SYS_ADMIN))
3821                 return -EACCES;
3822         rv = mddev_lock(mddev);
3823         if (mddev->hold_active == UNTIL_IOCTL)
3824                 mddev->hold_active = 0;
3825         if (!rv) {
3826                 rv = entry->store(mddev, page, length);
3827                 mddev_unlock(mddev);
3828         }
3829         return rv;
3830 }
3831
3832 static void md_free(struct kobject *ko)
3833 {
3834         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3835
3836         if (mddev->sysfs_state)
3837                 sysfs_put(mddev->sysfs_state);
3838
3839         if (mddev->gendisk) {
3840                 del_gendisk(mddev->gendisk);
3841                 put_disk(mddev->gendisk);
3842         }
3843         if (mddev->queue)
3844                 blk_cleanup_queue(mddev->queue);
3845
3846         kfree(mddev);
3847 }
3848
3849 static struct sysfs_ops md_sysfs_ops = {
3850         .show   = md_attr_show,
3851         .store  = md_attr_store,
3852 };
3853 static struct kobj_type md_ktype = {
3854         .release        = md_free,
3855         .sysfs_ops      = &md_sysfs_ops,
3856         .default_attrs  = md_default_attrs,
3857 };
3858
3859 int mdp_major = 0;
3860
3861 static void mddev_delayed_delete(struct work_struct *ws)
3862 {
3863         mddev_t *mddev = container_of(ws, mddev_t, del_work);
3864
3865         if (mddev->private == &md_redundancy_group) {
3866                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3867                 if (mddev->sysfs_action)
3868                         sysfs_put(mddev->sysfs_action);
3869                 mddev->sysfs_action = NULL;
3870                 mddev->private = NULL;
3871         }
3872         kobject_del(&mddev->kobj);
3873         kobject_put(&mddev->kobj);
3874 }
3875
3876 static int md_alloc(dev_t dev, char *name)
3877 {
3878         static DEFINE_MUTEX(disks_mutex);
3879         mddev_t *mddev = mddev_find(dev);
3880         struct gendisk *disk;
3881         int partitioned;
3882         int shift;
3883         int unit;
3884         int error;
3885
3886         if (!mddev)
3887                 return -ENODEV;
3888
3889         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3890         shift = partitioned ? MdpMinorShift : 0;
3891         unit = MINOR(mddev->unit) >> shift;
3892
3893         /* wait for any previous instance if this device
3894          * to be completed removed (mddev_delayed_delete).
3895          */
3896         flush_scheduled_work();
3897
3898         mutex_lock(&disks_mutex);
3899         error = -EEXIST;
3900         if (mddev->gendisk)
3901                 goto abort;
3902
3903         if (name) {
3904                 /* Need to ensure that 'name' is not a duplicate.
3905                  */
3906                 mddev_t *mddev2;
3907                 spin_lock(&all_mddevs_lock);
3908
3909                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3910                         if (mddev2->gendisk &&
3911                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
3912                                 spin_unlock(&all_mddevs_lock);
3913                                 goto abort;
3914                         }
3915                 spin_unlock(&all_mddevs_lock);
3916         }
3917
3918         error = -ENOMEM;
3919         mddev->queue = blk_alloc_queue(GFP_KERNEL);
3920         if (!mddev->queue)
3921                 goto abort;
3922         mddev->queue->queuedata = mddev;
3923
3924         /* Can be unlocked because the queue is new: no concurrency */
3925         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3926
3927         blk_queue_make_request(mddev->queue, md_make_request);
3928
3929         disk = alloc_disk(1 << shift);
3930         if (!disk) {
3931                 blk_cleanup_queue(mddev->queue);
3932                 mddev->queue = NULL;
3933                 goto abort;
3934         }
3935         disk->major = MAJOR(mddev->unit);
3936         disk->first_minor = unit << shift;
3937         if (name)
3938                 strcpy(disk->disk_name, name);
3939         else if (partitioned)
3940                 sprintf(disk->disk_name, "md_d%d", unit);
3941         else
3942                 sprintf(disk->disk_name, "md%d", unit);
3943         disk->fops = &md_fops;
3944         disk->private_data = mddev;
3945         disk->queue = mddev->queue;
3946         /* Allow extended partitions.  This makes the
3947          * 'mdp' device redundant, but we can't really
3948          * remove it now.
3949          */
3950         disk->flags |= GENHD_FL_EXT_DEVT;
3951         add_disk(disk);
3952         mddev->gendisk = disk;
3953         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3954                                      &disk_to_dev(disk)->kobj, "%s", "md");
3955         if (error) {
3956                 /* This isn't possible, but as kobject_init_and_add is marked
3957                  * __must_check, we must do something with the result
3958                  */
3959                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3960                        disk->disk_name);
3961                 error = 0;
3962         }
3963  abort:
3964         mutex_unlock(&disks_mutex);
3965         if (!error) {
3966                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3967                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3968         }
3969         mddev_put(mddev);
3970         return error;
3971 }
3972
3973 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3974 {
3975         md_alloc(dev, NULL);
3976         return NULL;
3977 }
3978
3979 static int add_named_array(const char *val, struct kernel_param *kp)
3980 {
3981         /* val must be "md_*" where * is not all digits.
3982          * We allocate an array with a large free minor number, and
3983          * set the name to val.  val must not already be an active name.
3984          */
3985         int len = strlen(val);
3986         char buf[DISK_NAME_LEN];
3987
3988         while (len && val[len-1] == '\n')
3989                 len--;
3990         if (len >= DISK_NAME_LEN)
3991                 return -E2BIG;
3992         strlcpy(buf, val, len+1);
3993         if (strncmp(buf, "md_", 3) != 0)
3994                 return -EINVAL;
3995         return md_alloc(0, buf);
3996 }
3997
3998 static void md_safemode_timeout(unsigned long data)
3999 {
4000         mddev_t *mddev = (mddev_t *) data;
4001
4002         if (!atomic_read(&mddev->writes_pending)) {
4003                 mddev->safemode = 1;
4004                 if (mddev->external)
4005                         sysfs_notify_dirent(mddev->sysfs_state);
4006         }
4007         md_wakeup_thread(mddev->thread);
4008 }
4009
4010 static int start_dirty_degraded;
4011
4012 static int do_md_run(mddev_t * mddev)
4013 {
4014         int err;
4015         mdk_rdev_t *rdev;
4016         struct gendisk *disk;
4017         struct mdk_personality *pers;
4018
4019         if (list_empty(&mddev->disks))
4020                 /* cannot run an array with no devices.. */
4021                 return -EINVAL;
4022
4023         if (mddev->pers)
4024                 return -EBUSY;
4025
4026         /*
4027          * Analyze all RAID superblock(s)
4028          */
4029         if (!mddev->raid_disks) {
4030                 if (!mddev->persistent)
4031                         return -EINVAL;
4032                 analyze_sbs(mddev);
4033         }
4034
4035         if (mddev->level != LEVEL_NONE)
4036                 request_module("md-level-%d", mddev->level);
4037         else if (mddev->clevel[0])
4038                 request_module("md-%s", mddev->clevel);
4039
4040         /*
4041          * Drop all container device buffers, from now on
4042          * the only valid external interface is through the md
4043          * device.
4044          */
4045         list_for_each_entry(rdev, &mddev->disks, same_set) {
4046                 if (test_bit(Faulty, &rdev->flags))
4047                         continue;
4048                 sync_blockdev(rdev->bdev);
4049                 invalidate_bdev(rdev->bdev);
4050
4051                 /* perform some consistency tests on the device.
4052                  * We don't want the data to overlap the metadata,
4053                  * Internal Bitmap issues have been handled elsewhere.
4054                  */
4055                 if (rdev->data_offset < rdev->sb_start) {
4056                         if (mddev->dev_sectors &&
4057                             rdev->data_offset + mddev->dev_sectors
4058                             > rdev->sb_start) {
4059                                 printk("md: %s: data overlaps metadata\n",
4060                                        mdname(mddev));
4061                                 return -EINVAL;
4062                         }
4063                 } else {
4064                         if (rdev->sb_start + rdev->sb_size/512
4065                             > rdev->data_offset) {
4066                                 printk("md: %s: metadata overlaps data\n",
4067                                        mdname(mddev));
4068                                 return -EINVAL;
4069                         }
4070                 }
4071                 sysfs_notify_dirent(rdev->sysfs_state);
4072         }
4073
4074         md_probe(mddev->unit, NULL, NULL);
4075         disk = mddev->gendisk;
4076         if (!disk)
4077                 return -ENOMEM;
4078
4079         spin_lock(&pers_lock);
4080         pers = find_pers(mddev->level, mddev->clevel);
4081         if (!pers || !try_module_get(pers->owner)) {
4082                 spin_unlock(&pers_lock);
4083                 if (mddev->level != LEVEL_NONE)
4084                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4085                                mddev->level);
4086                 else
4087                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4088                                mddev->clevel);
4089                 return -EINVAL;
4090         }
4091         mddev->pers = pers;
4092         spin_unlock(&pers_lock);
4093         if (mddev->level != pers->level) {
4094                 mddev->level = pers->level;
4095                 mddev->new_level = pers->level;
4096         }
4097         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4098
4099         if (mddev->reshape_position != MaxSector &&
4100             pers->start_reshape == NULL) {
4101                 /* This personality cannot handle reshaping... */
4102                 mddev->pers = NULL;
4103                 module_put(pers->owner);
4104                 return -EINVAL;
4105         }
4106
4107         if (pers->sync_request) {
4108                 /* Warn if this is a potentially silly
4109                  * configuration.
4110                  */
4111                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4112                 mdk_rdev_t *rdev2;
4113                 int warned = 0;
4114
4115                 list_for_each_entry(rdev, &mddev->disks, same_set)
4116                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4117                                 if (rdev < rdev2 &&
4118                                     rdev->bdev->bd_contains ==
4119                                     rdev2->bdev->bd_contains) {
4120                                         printk(KERN_WARNING
4121                                                "%s: WARNING: %s appears to be"
4122                                                " on the same physical disk as"
4123                                                " %s.\n",
4124                                                mdname(mddev),
4125                                                bdevname(rdev->bdev,b),
4126                                                bdevname(rdev2->bdev,b2));
4127                                         warned = 1;
4128                                 }
4129                         }
4130
4131                 if (warned)
4132                         printk(KERN_WARNING
4133                                "True protection against single-disk"
4134                                " failure might be compromised.\n");
4135         }
4136
4137         mddev->recovery = 0;
4138         /* may be over-ridden by personality */
4139         mddev->resync_max_sectors = mddev->dev_sectors;
4140
4141         mddev->barriers_work = 1;
4142         mddev->ok_start_degraded = start_dirty_degraded;
4143
4144         if (start_readonly)
4145                 mddev->ro = 2; /* read-only, but switch on first write */
4146
4147         err = mddev->pers->run(mddev);
4148         if (err)
4149                 printk(KERN_ERR "md: pers->run() failed ...\n");
4150         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4151                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4152                           " but 'external_size' not in effect?\n", __func__);
4153                 printk(KERN_ERR
4154                        "md: invalid array_size %llu > default size %llu\n",
4155                        (unsigned long long)mddev->array_sectors / 2,
4156                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4157                 err = -EINVAL;
4158                 mddev->pers->stop(mddev);
4159         }
4160         if (err == 0 && mddev->pers->sync_request) {
4161                 err = bitmap_create(mddev);
4162                 if (err) {
4163                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4164                                mdname(mddev), err);
4165                         mddev->pers->stop(mddev);
4166                 }
4167         }
4168         if (err) {
4169                 module_put(mddev->pers->owner);
4170                 mddev->pers = NULL;
4171                 bitmap_destroy(mddev);
4172                 return err;
4173         }
4174         if (mddev->pers->sync_request) {
4175                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4176                         printk(KERN_WARNING
4177                                "md: cannot register extra attributes for %s\n",
4178                                mdname(mddev));
4179                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4180         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4181                 mddev->ro = 0;
4182
4183         atomic_set(&mddev->writes_pending,0);
4184         mddev->safemode = 0;
4185         mddev->safemode_timer.function = md_safemode_timeout;
4186         mddev->safemode_timer.data = (unsigned long) mddev;
4187         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4188         mddev->in_sync = 1;
4189
4190         list_for_each_entry(rdev, &mddev->disks, same_set)
4191                 if (rdev->raid_disk >= 0) {
4192                         char nm[20];
4193                         sprintf(nm, "rd%d", rdev->raid_disk);
4194                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4195                                 printk("md: cannot register %s for %s\n",
4196                                        nm, mdname(mddev));
4197                 }
4198         
4199         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4200         
4201         if (mddev->flags)
4202                 md_update_sb(mddev, 0);
4203
4204         set_capacity(disk, mddev->array_sectors);
4205
4206         /* If there is a partially-recovered drive we need to
4207          * start recovery here.  If we leave it to md_check_recovery,
4208          * it will remove the drives and not do the right thing
4209          */
4210         if (mddev->degraded && !mddev->sync_thread) {
4211                 int spares = 0;
4212                 list_for_each_entry(rdev, &mddev->disks, same_set)
4213                         if (rdev->raid_disk >= 0 &&
4214                             !test_bit(In_sync, &rdev->flags) &&
4215                             !test_bit(Faulty, &rdev->flags))
4216                                 /* complete an interrupted recovery */
4217                                 spares++;
4218                 if (spares && mddev->pers->sync_request) {
4219                         mddev->recovery = 0;
4220                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4221                         mddev->sync_thread = md_register_thread(md_do_sync,
4222                                                                 mddev,
4223                                                                 "resync");
4224                         if (!mddev->sync_thread) {
4225                                 printk(KERN_ERR "%s: could not start resync"
4226                                        " thread...\n",
4227                                        mdname(mddev));
4228                                 /* leave the spares where they are, it shouldn't hurt */
4229                                 mddev->recovery = 0;
4230                         }
4231                 }
4232         }
4233         md_wakeup_thread(mddev->thread);
4234         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4235
4236         revalidate_disk(mddev->gendisk);
4237         mddev->changed = 1;
4238         md_new_event(mddev);
4239         sysfs_notify_dirent(mddev->sysfs_state);
4240         if (mddev->sysfs_action)
4241                 sysfs_notify_dirent(mddev->sysfs_action);
4242         sysfs_notify(&mddev->kobj, NULL, "degraded");
4243         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4244         return 0;
4245 }
4246
4247 static int restart_array(mddev_t *mddev)
4248 {
4249         struct gendisk *disk = mddev->gendisk;
4250
4251         /* Complain if it has no devices */
4252         if (list_empty(&mddev->disks))
4253                 return -ENXIO;
4254         if (!mddev->pers)
4255                 return -EINVAL;
4256         if (!mddev->ro)
4257                 return -EBUSY;
4258         mddev->safemode = 0;
4259         mddev->ro = 0;
4260         set_disk_ro(disk, 0);
4261         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4262                 mdname(mddev));
4263         /* Kick recovery or resync if necessary */
4264         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4265         md_wakeup_thread(mddev->thread);
4266         md_wakeup_thread(mddev->sync_thread);
4267         sysfs_notify_dirent(mddev->sysfs_state);
4268         return 0;
4269 }
4270
4271 /* similar to deny_write_access, but accounts for our holding a reference
4272  * to the file ourselves */
4273 static int deny_bitmap_write_access(struct file * file)
4274 {
4275         struct inode *inode = file->f_mapping->host;
4276
4277         spin_lock(&inode->i_lock);
4278         if (atomic_read(&inode->i_writecount) > 1) {
4279                 spin_unlock(&inode->i_lock);
4280                 return -ETXTBSY;
4281         }
4282         atomic_set(&inode->i_writecount, -1);
4283         spin_unlock(&inode->i_lock);
4284
4285         return 0;
4286 }
4287
4288 static void restore_bitmap_write_access(struct file *file)
4289 {
4290         struct inode *inode = file->f_mapping->host;
4291
4292         spin_lock(&inode->i_lock);
4293         atomic_set(&inode->i_writecount, 1);
4294         spin_unlock(&inode->i_lock);
4295 }
4296
4297 /* mode:
4298  *   0 - completely stop and dis-assemble array
4299  *   1 - switch to readonly
4300  *   2 - stop but do not disassemble array
4301  */
4302 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4303 {
4304         int err = 0;
4305         struct gendisk *disk = mddev->gendisk;
4306         mdk_rdev_t *rdev;
4307
4308         mutex_lock(&mddev->open_mutex);
4309         if (atomic_read(&mddev->openers) > is_open) {
4310                 printk("md: %s still in use.\n",mdname(mddev));
4311                 err = -EBUSY;
4312         } else if (mddev->pers) {
4313
4314                 if (mddev->sync_thread) {
4315                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4316                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4317                         md_unregister_thread(mddev->sync_thread);
4318                         mddev->sync_thread = NULL;
4319                 }
4320
4321                 del_timer_sync(&mddev->safemode_timer);
4322
4323                 switch(mode) {
4324                 case 1: /* readonly */
4325                         err  = -ENXIO;
4326                         if (mddev->ro==1)
4327                                 goto out;
4328                         mddev->ro = 1;
4329                         break;
4330                 case 0: /* disassemble */
4331                 case 2: /* stop */
4332                         bitmap_flush(mddev);
4333                         md_super_wait(mddev);
4334                         if (mddev->ro)
4335                                 set_disk_ro(disk, 0);
4336
4337                         mddev->pers->stop(mddev);
4338                         mddev->queue->merge_bvec_fn = NULL;
4339                         mddev->queue->unplug_fn = NULL;
4340                         mddev->queue->backing_dev_info.congested_fn = NULL;
4341                         module_put(mddev->pers->owner);
4342                         if (mddev->pers->sync_request)
4343                                 mddev->private = &md_redundancy_group;
4344                         mddev->pers = NULL;
4345                         /* tell userspace to handle 'inactive' */
4346                         sysfs_notify_dirent(mddev->sysfs_state);
4347
4348                         list_for_each_entry(rdev, &mddev->disks, same_set)
4349                                 if (rdev->raid_disk >= 0) {
4350                                         char nm[20];
4351                                         sprintf(nm, "rd%d", rdev->raid_disk);
4352                                         sysfs_remove_link(&mddev->kobj, nm);
4353                                 }
4354
4355                         set_capacity(disk, 0);
4356                         mddev->changed = 1;
4357
4358                         if (mddev->ro)
4359                                 mddev->ro = 0;
4360                 }
4361                 if (!mddev->in_sync || mddev->flags) {
4362                         /* mark array as shutdown cleanly */
4363                         mddev->in_sync = 1;
4364                         md_update_sb(mddev, 1);
4365                 }
4366                 if (mode == 1)
4367                         set_disk_ro(disk, 1);
4368                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4369                 err = 0;
4370         }
4371 out:
4372         mutex_unlock(&mddev->open_mutex);
4373         if (err)
4374                 return err;
4375         /*
4376          * Free resources if final stop
4377          */
4378         if (mode == 0) {
4379
4380                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4381
4382                 bitmap_destroy(mddev);
4383                 if (mddev->bitmap_file) {
4384                         restore_bitmap_write_access(mddev->bitmap_file);
4385                         fput(mddev->bitmap_file);
4386                         mddev->bitmap_file = NULL;
4387                 }
4388                 mddev->bitmap_offset = 0;
4389
4390                 /* make sure all md_delayed_delete calls have finished */
4391                 flush_scheduled_work();
4392
4393                 export_array(mddev);
4394
4395                 mddev->array_sectors = 0;
4396                 mddev->external_size = 0;
4397                 mddev->dev_sectors = 0;
4398                 mddev->raid_disks = 0;
4399                 mddev->recovery_cp = 0;
4400                 mddev->resync_min = 0;
4401                 mddev->resync_max = MaxSector;
4402                 mddev->reshape_position = MaxSector;
4403                 mddev->external = 0;
4404                 mddev->persistent = 0;
4405                 mddev->level = LEVEL_NONE;
4406                 mddev->clevel[0] = 0;
4407                 mddev->flags = 0;
4408                 mddev->ro = 0;
4409                 mddev->metadata_type[0] = 0;
4410                 mddev->chunk_sectors = 0;
4411                 mddev->ctime = mddev->utime = 0;
4412                 mddev->layout = 0;
4413                 mddev->max_disks = 0;
4414                 mddev->events = 0;
4415                 mddev->delta_disks = 0;
4416                 mddev->new_level = LEVEL_NONE;
4417                 mddev->new_layout = 0;
4418                 mddev->new_chunk_sectors = 0;
4419                 mddev->curr_resync = 0;
4420                 mddev->resync_mismatches = 0;
4421                 mddev->suspend_lo = mddev->suspend_hi = 0;
4422                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4423                 mddev->recovery = 0;
4424                 mddev->in_sync = 0;
4425                 mddev->changed = 0;
4426                 mddev->degraded = 0;
4427                 mddev->barriers_work = 0;
4428                 mddev->safemode = 0;
4429                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4430                 if (mddev->hold_active == UNTIL_STOP)
4431                         mddev->hold_active = 0;
4432
4433         } else if (mddev->pers)
4434                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4435                         mdname(mddev));
4436         err = 0;
4437         blk_integrity_unregister(disk);
4438         md_new_event(mddev);
4439         sysfs_notify_dirent(mddev->sysfs_state);
4440         return err;
4441 }
4442
4443 #ifndef MODULE
4444 static void autorun_array(mddev_t *mddev)
4445 {
4446         mdk_rdev_t *rdev;
4447         int err;
4448
4449         if (list_empty(&mddev->disks))
4450                 return;
4451
4452         printk(KERN_INFO "md: running: ");
4453
4454         list_for_each_entry(rdev, &mddev->disks, same_set) {
4455                 char b[BDEVNAME_SIZE];
4456                 printk("<%s>", bdevname(rdev->bdev,b));
4457         }
4458         printk("\n");
4459
4460         err = do_md_run(mddev);
4461         if (err) {
4462                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4463                 do_md_stop(mddev, 0, 0);
4464         }
4465 }
4466
4467 /*
4468  * lets try to run arrays based on all disks that have arrived
4469  * until now. (those are in pending_raid_disks)
4470  *
4471  * the method: pick the first pending disk, collect all disks with
4472  * the same UUID, remove all from the pending list and put them into
4473  * the 'same_array' list. Then order this list based on superblock
4474  * update time (freshest comes first), kick out 'old' disks and
4475  * compare superblocks. If everything's fine then run it.
4476  *
4477  * If "unit" is allocated, then bump its reference count
4478  */
4479 static void autorun_devices(int part)
4480 {
4481         mdk_rdev_t *rdev0, *rdev, *tmp;
4482         mddev_t *mddev;
4483         char b[BDEVNAME_SIZE];
4484
4485         printk(KERN_INFO "md: autorun ...\n");
4486         while (!list_empty(&pending_raid_disks)) {
4487                 int unit;
4488                 dev_t dev;
4489                 LIST_HEAD(candidates);
4490                 rdev0 = list_entry(pending_raid_disks.next,
4491                                          mdk_rdev_t, same_set);
4492
4493                 printk(KERN_INFO "md: considering %s ...\n",
4494                         bdevname(rdev0->bdev,b));
4495                 INIT_LIST_HEAD(&candidates);
4496                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4497                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4498                                 printk(KERN_INFO "md:  adding %s ...\n",
4499                                         bdevname(rdev->bdev,b));
4500                                 list_move(&rdev->same_set, &candidates);
4501                         }
4502                 /*
4503                  * now we have a set of devices, with all of them having
4504                  * mostly sane superblocks. It's time to allocate the
4505                  * mddev.
4506                  */
4507                 if (part) {
4508                         dev = MKDEV(mdp_major,
4509                                     rdev0->preferred_minor << MdpMinorShift);
4510                         unit = MINOR(dev) >> MdpMinorShift;
4511                 } else {
4512                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4513                         unit = MINOR(dev);
4514                 }
4515                 if (rdev0->preferred_minor != unit) {
4516                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4517                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4518                         break;
4519                 }
4520
4521                 md_probe(dev, NULL, NULL);
4522                 mddev = mddev_find(dev);
4523                 if (!mddev || !mddev->gendisk) {
4524                         if (mddev)
4525                                 mddev_put(mddev);
4526                         printk(KERN_ERR
4527                                 "md: cannot allocate memory for md drive.\n");
4528                         break;
4529                 }
4530                 if (mddev_lock(mddev)) 
4531                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4532                                mdname(mddev));
4533                 else if (mddev->raid_disks || mddev->major_version
4534                          || !list_empty(&mddev->disks)) {
4535                         printk(KERN_WARNING 
4536                                 "md: %s already running, cannot run %s\n",
4537                                 mdname(mddev), bdevname(rdev0->bdev,b));
4538                         mddev_unlock(mddev);
4539                 } else {
4540                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4541                         mddev->persistent = 1;
4542                         rdev_for_each_list(rdev, tmp, &candidates) {
4543                                 list_del_init(&rdev->same_set);
4544                                 if (bind_rdev_to_array(rdev, mddev))
4545                                         export_rdev(rdev);
4546                         }
4547                         autorun_array(mddev);
4548                         mddev_unlock(mddev);
4549                 }
4550                 /* on success, candidates will be empty, on error
4551                  * it won't...
4552                  */
4553                 rdev_for_each_list(rdev, tmp, &candidates) {
4554                         list_del_init(&rdev->same_set);
4555                         export_rdev(rdev);
4556                 }
4557                 mddev_put(mddev);
4558         }
4559         printk(KERN_INFO "md: ... autorun DONE.\n");
4560 }
4561 #endif /* !MODULE */
4562
4563 static int get_version(void __user * arg)
4564 {
4565         mdu_version_t ver;
4566
4567         ver.major = MD_MAJOR_VERSION;
4568         ver.minor = MD_MINOR_VERSION;
4569         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4570
4571         if (copy_to_user(arg, &ver, sizeof(ver)))
4572                 return -EFAULT;
4573
4574         return 0;
4575 }
4576
4577 static int get_array_info(mddev_t * mddev, void __user * arg)
4578 {
4579         mdu_array_info_t info;
4580         int nr,working,insync,failed,spare;
4581         mdk_rdev_t *rdev;
4582
4583         nr=working=insync=failed=spare=0;
4584         list_for_each_entry(rdev, &mddev->disks, same_set) {
4585                 nr++;
4586                 if (test_bit(Faulty, &rdev->flags))
4587                         failed++;
4588                 else {
4589                         working++;
4590                         if (test_bit(In_sync, &rdev->flags))
4591                                 insync++;       
4592                         else
4593                                 spare++;
4594                 }
4595         }
4596
4597         info.major_version = mddev->major_version;
4598         info.minor_version = mddev->minor_version;
4599         info.patch_version = MD_PATCHLEVEL_VERSION;
4600         info.ctime         = mddev->ctime;
4601         info.level         = mddev->level;
4602         info.size          = mddev->dev_sectors / 2;
4603         if (info.size != mddev->dev_sectors / 2) /* overflow */
4604                 info.size = -1;
4605         info.nr_disks      = nr;
4606         info.raid_disks    = mddev->raid_disks;
4607         info.md_minor      = mddev->md_minor;
4608         info.not_persistent= !mddev->persistent;
4609
4610         info.utime         = mddev->utime;
4611         info.state         = 0;
4612         if (mddev->in_sync)
4613                 info.state = (1<<MD_SB_CLEAN);
4614         if (mddev->bitmap && mddev->bitmap_offset)
4615                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4616         info.active_disks  = insync;
4617         info.working_disks = working;
4618         info.failed_disks  = failed;
4619         info.spare_disks   = spare;
4620
4621         info.layout        = mddev->layout;
4622         info.chunk_size    = mddev->chunk_sectors << 9;
4623
4624         if (copy_to_user(arg, &info, sizeof(info)))
4625                 return -EFAULT;
4626
4627         return 0;
4628 }
4629
4630 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4631 {
4632         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4633         char *ptr, *buf = NULL;
4634         int err = -ENOMEM;
4635
4636         if (md_allow_write(mddev))
4637                 file = kmalloc(sizeof(*file), GFP_NOIO);
4638         else
4639                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4640
4641         if (!file)
4642                 goto out;
4643
4644         /* bitmap disabled, zero the first byte and copy out */
4645         if (!mddev->bitmap || !mddev->bitmap->file) {
4646                 file->pathname[0] = '\0';
4647                 goto copy_out;
4648         }
4649
4650         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4651         if (!buf)
4652                 goto out;
4653
4654         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4655         if (IS_ERR(ptr))
4656                 goto out;
4657
4658         strcpy(file->pathname, ptr);
4659
4660 copy_out:
4661         err = 0;
4662         if (copy_to_user(arg, file, sizeof(*file)))
4663                 err = -EFAULT;
4664 out:
4665         kfree(buf);
4666         kfree(file);
4667         return err;
4668 }
4669
4670 static int get_disk_info(mddev_t * mddev, void __user * arg)
4671 {
4672         mdu_disk_info_t info;
4673         mdk_rdev_t *rdev;
4674
4675         if (copy_from_user(&info, arg, sizeof(info)))
4676                 return -EFAULT;
4677
4678         rdev = find_rdev_nr(mddev, info.number);
4679         if (rdev) {
4680                 info.major = MAJOR(rdev->bdev->bd_dev);
4681                 info.minor = MINOR(rdev->bdev->bd_dev);
4682                 info.raid_disk = rdev->raid_disk;
4683                 info.state = 0;
4684                 if (test_bit(Faulty, &rdev->flags))
4685                         info.state |= (1<<MD_DISK_FAULTY);
4686                 else if (test_bit(In_sync, &rdev->flags)) {
4687                         info.state |= (1<<MD_DISK_ACTIVE);
4688                         info.state |= (1<<MD_DISK_SYNC);
4689                 }
4690                 if (test_bit(WriteMostly, &rdev->flags))
4691                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4692         } else {
4693                 info.major = info.minor = 0;
4694                 info.raid_disk = -1;
4695                 info.state = (1<<MD_DISK_REMOVED);
4696         }
4697
4698         if (copy_to_user(arg, &info, sizeof(info)))
4699                 return -EFAULT;
4700
4701         return 0;
4702 }
4703
4704 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4705 {
4706         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4707         mdk_rdev_t *rdev;
4708         dev_t dev = MKDEV(info->major,info->minor);
4709
4710         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4711                 return -EOVERFLOW;
4712
4713         if (!mddev->raid_disks) {
4714                 int err;
4715                 /* expecting a device which has a superblock */
4716                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4717                 if (IS_ERR(rdev)) {
4718                         printk(KERN_WARNING 
4719                                 "md: md_import_device returned %ld\n",
4720                                 PTR_ERR(rdev));
4721                         return PTR_ERR(rdev);
4722                 }
4723                 if (!list_empty(&mddev->disks)) {
4724                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4725                                                         mdk_rdev_t, same_set);
4726                         err = super_types[mddev->major_version]
4727                                 .load_super(rdev, rdev0, mddev->minor_version);
4728                         if (err < 0) {
4729                                 printk(KERN_WARNING 
4730                                         "md: %s has different UUID to %s\n",
4731                                         bdevname(rdev->bdev,b), 
4732                                         bdevname(rdev0->bdev,b2));
4733                                 export_rdev(rdev);
4734                                 return -EINVAL;
4735                         }
4736                 }
4737                 err = bind_rdev_to_array(rdev, mddev);
4738                 if (err)
4739                         export_rdev(rdev);
4740                 return err;
4741         }
4742
4743         /*
4744          * add_new_disk can be used once the array is assembled
4745          * to add "hot spares".  They must already have a superblock
4746          * written
4747          */
4748         if (mddev->pers) {
4749                 int err;
4750                 if (!mddev->pers->hot_add_disk) {
4751                         printk(KERN_WARNING 
4752                                 "%s: personality does not support diskops!\n",
4753                                mdname(mddev));
4754                         return -EINVAL;
4755                 }
4756                 if (mddev->persistent)
4757                         rdev = md_import_device(dev, mddev->major_version,
4758                                                 mddev->minor_version);
4759                 else
4760                         rdev = md_import_device(dev, -1, -1);
4761                 if (IS_ERR(rdev)) {
4762                         printk(KERN_WARNING 
4763                                 "md: md_import_device returned %ld\n",
4764                                 PTR_ERR(rdev));
4765                         return PTR_ERR(rdev);
4766                 }
4767                 /* set save_raid_disk if appropriate */
4768                 if (!mddev->persistent) {
4769                         if (info->state & (1<<MD_DISK_SYNC)  &&
4770                             info->raid_disk < mddev->raid_disks)
4771                                 rdev->raid_disk = info->raid_disk;
4772                         else
4773                                 rdev->raid_disk = -1;
4774                 } else
4775                         super_types[mddev->major_version].
4776                                 validate_super(mddev, rdev);
4777                 rdev->saved_raid_disk = rdev->raid_disk;
4778
4779                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4780                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4781                         set_bit(WriteMostly, &rdev->flags);
4782                 else
4783                         clear_bit(WriteMostly, &rdev->flags);
4784
4785                 rdev->raid_disk = -1;
4786                 err = bind_rdev_to_array(rdev, mddev);
4787                 if (!err && !mddev->pers->hot_remove_disk) {
4788                         /* If there is hot_add_disk but no hot_remove_disk
4789                          * then added disks for geometry changes,
4790                          * and should be added immediately.
4791                          */
4792                         super_types[mddev->major_version].
4793                                 validate_super(mddev, rdev);
4794                         err = mddev->pers->hot_add_disk(mddev, rdev);
4795                         if (err)
4796                                 unbind_rdev_from_array(rdev);
4797                 }
4798                 if (err)
4799                         export_rdev(rdev);
4800                 else
4801                         sysfs_notify_dirent(rdev->sysfs_state);
4802
4803                 md_update_sb(mddev, 1);
4804                 if (mddev->degraded)
4805                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4806                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4807                 md_wakeup_thread(mddev->thread);
4808                 return err;
4809         }
4810
4811         /* otherwise, add_new_disk is only allowed
4812          * for major_version==0 superblocks
4813          */
4814         if (mddev->major_version != 0) {
4815                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4816                        mdname(mddev));
4817                 return -EINVAL;
4818         }
4819
4820         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4821                 int err;
4822                 rdev = md_import_device(dev, -1, 0);
4823                 if (IS_ERR(rdev)) {
4824                         printk(KERN_WARNING 
4825                                 "md: error, md_import_device() returned %ld\n",
4826                                 PTR_ERR(rdev));
4827                         return PTR_ERR(rdev);
4828                 }
4829                 rdev->desc_nr = info->number;
4830                 if (info->raid_disk < mddev->raid_disks)
4831                         rdev->raid_disk = info->raid_disk;
4832                 else
4833                         rdev->raid_disk = -1;
4834
4835                 if (rdev->raid_disk < mddev->raid_disks)
4836                         if (info->state & (1<<MD_DISK_SYNC))
4837                                 set_bit(In_sync, &rdev->flags);
4838
4839                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4840                         set_bit(WriteMostly, &rdev->flags);
4841
4842                 if (!mddev->persistent) {
4843                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4844                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4845                 } else 
4846                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4847                 rdev->sectors = rdev->sb_start;
4848
4849                 err = bind_rdev_to_array(rdev, mddev);
4850                 if (err) {
4851                         export_rdev(rdev);
4852                         return err;
4853                 }
4854         }
4855
4856         return 0;
4857 }
4858
4859 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4860 {
4861         char b[BDEVNAME_SIZE];
4862         mdk_rdev_t *rdev;
4863
4864         rdev = find_rdev(mddev, dev);
4865         if (!rdev)
4866                 return -ENXIO;
4867
4868         if (rdev->raid_disk >= 0)
4869                 goto busy;
4870
4871         kick_rdev_from_array(rdev);
4872         md_update_sb(mddev, 1);
4873         md_new_event(mddev);
4874
4875         return 0;
4876 busy:
4877         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4878                 bdevname(rdev->bdev,b), mdname(mddev));
4879         return -EBUSY;
4880 }
4881
4882 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4883 {
4884         char b[BDEVNAME_SIZE];
4885         int err;
4886         mdk_rdev_t *rdev;
4887
4888         if (!mddev->pers)
4889                 return -ENODEV;
4890
4891         if (mddev->major_version != 0) {
4892                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4893                         " version-0 superblocks.\n",
4894                         mdname(mddev));
4895                 return -EINVAL;
4896         }
4897         if (!mddev->pers->hot_add_disk) {
4898                 printk(KERN_WARNING 
4899                         "%s: personality does not support diskops!\n",
4900                         mdname(mddev));
4901                 return -EINVAL;
4902         }
4903
4904         rdev = md_import_device(dev, -1, 0);
4905         if (IS_ERR(rdev)) {
4906                 printk(KERN_WARNING 
4907                         "md: error, md_import_device() returned %ld\n",
4908                         PTR_ERR(rdev));
4909                 return -EINVAL;
4910         }
4911
4912         if (mddev->persistent)
4913                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4914         else
4915                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4916
4917         rdev->sectors = rdev->sb_start;
4918
4919         if (test_bit(Faulty, &rdev->flags)) {
4920                 printk(KERN_WARNING 
4921                         "md: can not hot-add faulty %s disk to %s!\n",
4922                         bdevname(rdev->bdev,b), mdname(mddev));
4923                 err = -EINVAL;
4924                 goto abort_export;
4925         }
4926         clear_bit(In_sync, &rdev->flags);
4927         rdev->desc_nr = -1;
4928         rdev->saved_raid_disk = -1;
4929         err = bind_rdev_to_array(rdev, mddev);
4930         if (err)
4931                 goto abort_export;
4932
4933         /*
4934          * The rest should better be atomic, we can have disk failures
4935          * noticed in interrupt contexts ...
4936          */
4937
4938         rdev->raid_disk = -1;
4939
4940         md_update_sb(mddev, 1);
4941
4942         /*
4943          * Kick recovery, maybe this spare has to be added to the
4944          * array immediately.
4945          */
4946         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4947         md_wakeup_thread(mddev->thread);
4948         md_new_event(mddev);
4949         return 0;
4950
4951 abort_export:
4952         export_rdev(rdev);
4953         return err;
4954 }
4955
4956 static int set_bitmap_file(mddev_t *mddev, int fd)
4957 {
4958         int err;
4959
4960         if (mddev->pers) {
4961                 if (!mddev->pers->quiesce)
4962                         return -EBUSY;
4963                 if (mddev->recovery || mddev->sync_thread)
4964                         return -EBUSY;
4965                 /* we should be able to change the bitmap.. */
4966         }
4967
4968
4969         if (fd >= 0) {
4970                 if (mddev->bitmap)
4971                         return -EEXIST; /* cannot add when bitmap is present */
4972                 mddev->bitmap_file = fget(fd);
4973
4974                 if (mddev->bitmap_file == NULL) {
4975                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4976                                mdname(mddev));
4977                         return -EBADF;
4978                 }
4979
4980                 err = deny_bitmap_write_access(mddev->bitmap_file);
4981                 if (err) {
4982                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4983                                mdname(mddev));
4984                         fput(mddev->bitmap_file);
4985                         mddev->bitmap_file = NULL;
4986                         return err;
4987                 }
4988                 mddev->bitmap_offset = 0; /* file overrides offset */
4989         } else if (mddev->bitmap == NULL)
4990                 return -ENOENT; /* cannot remove what isn't there */
4991         err = 0;
4992         if (mddev->pers) {
4993                 mddev->pers->quiesce(mddev, 1);
4994                 if (fd >= 0)
4995                         err = bitmap_create(mddev);
4996                 if (fd < 0 || err) {
4997                         bitmap_destroy(mddev);
4998                         fd = -1; /* make sure to put the file */
4999                 }
5000                 mddev->pers->quiesce(mddev, 0);
5001         }
5002         if (fd < 0) {
5003                 if (mddev->bitmap_file) {
5004                         restore_bitmap_write_access(mddev->bitmap_file);
5005                         fput(mddev->bitmap_file);
5006                 }
5007                 mddev->bitmap_file = NULL;
5008         }
5009
5010         return err;
5011 }
5012
5013 /*
5014  * set_array_info is used two different ways
5015  * The original usage is when creating a new array.
5016  * In this usage, raid_disks is > 0 and it together with
5017  *  level, size, not_persistent,layout,chunksize determine the
5018  *  shape of the array.
5019  *  This will always create an array with a type-0.90.0 superblock.
5020  * The newer usage is when assembling an array.
5021  *  In this case raid_disks will be 0, and the major_version field is
5022  *  use to determine which style super-blocks are to be found on the devices.
5023  *  The minor and patch _version numbers are also kept incase the
5024  *  super_block handler wishes to interpret them.
5025  */
5026 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5027 {
5028
5029         if (info->raid_disks == 0) {
5030                 /* just setting version number for superblock loading */
5031                 if (info->major_version < 0 ||
5032                     info->major_version >= ARRAY_SIZE(super_types) ||
5033                     super_types[info->major_version].name == NULL) {
5034                         /* maybe try to auto-load a module? */
5035                         printk(KERN_INFO 
5036                                 "md: superblock version %d not known\n",
5037                                 info->major_version);
5038                         return -EINVAL;
5039                 }
5040                 mddev->major_version = info->major_version;
5041                 mddev->minor_version = info->minor_version;
5042                 mddev->patch_version = info->patch_version;
5043                 mddev->persistent = !info->not_persistent;
5044                 return 0;
5045         }
5046         mddev->major_version = MD_MAJOR_VERSION;
5047         mddev->minor_version = MD_MINOR_VERSION;
5048         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5049         mddev->ctime         = get_seconds();
5050
5051         mddev->level         = info->level;
5052         mddev->clevel[0]     = 0;
5053         mddev->dev_sectors   = 2 * (sector_t)info->size;
5054         mddev->raid_disks    = info->raid_disks;
5055         /* don't set md_minor, it is determined by which /dev/md* was
5056          * openned
5057          */
5058         if (info->state & (1<<MD_SB_CLEAN))
5059                 mddev->recovery_cp = MaxSector;
5060         else
5061                 mddev->recovery_cp = 0;
5062         mddev->persistent    = ! info->not_persistent;
5063         mddev->external      = 0;
5064
5065         mddev->layout        = info->layout;
5066         mddev->chunk_sectors = info->chunk_size >> 9;
5067
5068         mddev->max_disks     = MD_SB_DISKS;
5069
5070         if (mddev->persistent)
5071                 mddev->flags         = 0;
5072         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5073
5074         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5075         mddev->bitmap_offset = 0;
5076
5077         mddev->reshape_position = MaxSector;
5078
5079         /*
5080          * Generate a 128 bit UUID
5081          */
5082         get_random_bytes(mddev->uuid, 16);
5083
5084         mddev->new_level = mddev->level;
5085         mddev->new_chunk_sectors = mddev->chunk_sectors;
5086         mddev->new_layout = mddev->layout;
5087         mddev->delta_disks = 0;
5088
5089         return 0;
5090 }
5091
5092 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5093 {
5094         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5095
5096         if (mddev->external_size)
5097                 return;
5098
5099         mddev->array_sectors = array_sectors;
5100 }
5101 EXPORT_SYMBOL(md_set_array_sectors);
5102
5103 static int update_size(mddev_t *mddev, sector_t num_sectors)
5104 {
5105         mdk_rdev_t *rdev;
5106         int rv;
5107         int fit = (num_sectors == 0);
5108
5109         if (mddev->pers->resize == NULL)
5110                 return -EINVAL;
5111         /* The "num_sectors" is the number of sectors of each device that
5112          * is used.  This can only make sense for arrays with redundancy.
5113          * linear and raid0 always use whatever space is available. We can only
5114          * consider changing this number if no resync or reconstruction is
5115          * happening, and if the new size is acceptable. It must fit before the
5116          * sb_start or, if that is <data_offset, it must fit before the size
5117          * of each device.  If num_sectors is zero, we find the largest size
5118          * that fits.
5119
5120          */
5121         if (mddev->sync_thread)
5122                 return -EBUSY;
5123         if (mddev->bitmap)
5124                 /* Sorry, cannot grow a bitmap yet, just remove it,
5125                  * grow, and re-add.
5126                  */
5127                 return -EBUSY;
5128         list_for_each_entry(rdev, &mddev->disks, same_set) {
5129                 sector_t avail = rdev->sectors;
5130
5131                 if (fit && (num_sectors == 0 || num_sectors > avail))
5132                         num_sectors = avail;
5133                 if (avail < num_sectors)
5134                         return -ENOSPC;
5135         }
5136         rv = mddev->pers->resize(mddev, num_sectors);
5137         if (!rv)
5138                 revalidate_disk(mddev->gendisk);
5139         return rv;
5140 }
5141
5142 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5143 {
5144         int rv;
5145         /* change the number of raid disks */
5146         if (mddev->pers->check_reshape == NULL)
5147                 return -EINVAL;
5148         if (raid_disks <= 0 ||
5149             raid_disks >= mddev->max_disks)
5150                 return -EINVAL;
5151         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5152                 return -EBUSY;
5153         mddev->delta_disks = raid_disks - mddev->raid_disks;
5154
5155         rv = mddev->pers->check_reshape(mddev);
5156         return rv;
5157 }
5158
5159
5160 /*
5161  * update_array_info is used to change the configuration of an
5162  * on-line array.
5163  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5164  * fields in the info are checked against the array.
5165  * Any differences that cannot be handled will cause an error.
5166  * Normally, only one change can be managed at a time.
5167  */
5168 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5169 {
5170         int rv = 0;
5171         int cnt = 0;
5172         int state = 0;
5173
5174         /* calculate expected state,ignoring low bits */
5175         if (mddev->bitmap && mddev->bitmap_offset)
5176                 state |= (1 << MD_SB_BITMAP_PRESENT);
5177
5178         if (mddev->major_version != info->major_version ||
5179             mddev->minor_version != info->minor_version ||
5180 /*          mddev->patch_version != info->patch_version || */
5181             mddev->ctime         != info->ctime         ||
5182             mddev->level         != info->level         ||
5183 /*          mddev->layout        != info->layout        || */
5184             !mddev->persistent   != info->not_persistent||
5185             mddev->chunk_sectors != info->chunk_size >> 9 ||
5186             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5187             ((state^info->state) & 0xfffffe00)
5188                 )
5189                 return -EINVAL;
5190         /* Check there is only one change */
5191         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5192                 cnt++;
5193         if (mddev->raid_disks != info->raid_disks)
5194                 cnt++;
5195         if (mddev->layout != info->layout)
5196                 cnt++;
5197         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5198                 cnt++;
5199         if (cnt == 0)
5200                 return 0;
5201         if (cnt > 1)
5202                 return -EINVAL;
5203
5204         if (mddev->layout != info->layout) {
5205                 /* Change layout
5206                  * we don't need to do anything at the md level, the
5207                  * personality will take care of it all.
5208                  */
5209                 if (mddev->pers->check_reshape == NULL)
5210                         return -EINVAL;
5211                 else {
5212                         mddev->new_layout = info->layout;
5213                         rv = mddev->pers->check_reshape(mddev);
5214                         if (rv)
5215                                 mddev->new_layout = mddev->layout;
5216                         return rv;
5217                 }
5218         }
5219         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5220                 rv = update_size(mddev, (sector_t)info->size * 2);
5221
5222         if (mddev->raid_disks    != info->raid_disks)
5223                 rv = update_raid_disks(mddev, info->raid_disks);
5224
5225         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5226                 if (mddev->pers->quiesce == NULL)
5227                         return -EINVAL;
5228                 if (mddev->recovery || mddev->sync_thread)
5229                         return -EBUSY;
5230                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5231                         /* add the bitmap */
5232                         if (mddev->bitmap)
5233                                 return -EEXIST;
5234                         if (mddev->default_bitmap_offset == 0)
5235                                 return -EINVAL;
5236                         mddev->bitmap_offset = mddev->default_bitmap_offset;
5237                         mddev->pers->quiesce(mddev, 1);
5238                         rv = bitmap_create(mddev);
5239                         if (rv)
5240                                 bitmap_destroy(mddev);
5241                         mddev->pers->quiesce(mddev, 0);
5242                 } else {
5243                         /* remove the bitmap */
5244                         if (!mddev->bitmap)
5245                                 return -ENOENT;
5246                         if (mddev->bitmap->file)
5247                                 return -EINVAL;
5248                         mddev->pers->quiesce(mddev, 1);
5249                         bitmap_destroy(mddev);
5250                         mddev->pers->quiesce(mddev, 0);
5251                         mddev->bitmap_offset = 0;
5252                 }
5253         }
5254         md_update_sb(mddev, 1);
5255         return rv;
5256 }
5257
5258 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5259 {
5260         mdk_rdev_t *rdev;
5261
5262         if (mddev->pers == NULL)
5263                 return -ENODEV;
5264
5265         rdev = find_rdev(mddev, dev);
5266         if (!rdev)
5267                 return -ENODEV;
5268
5269         md_error(mddev, rdev);
5270         return 0;
5271 }
5272
5273 /*
5274  * We have a problem here : there is no easy way to give a CHS
5275  * virtual geometry. We currently pretend that we have a 2 heads
5276  * 4 sectors (with a BIG number of cylinders...). This drives
5277  * dosfs just mad... ;-)
5278  */
5279 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5280 {
5281         mddev_t *mddev = bdev->bd_disk->private_data;
5282
5283         geo->heads = 2;
5284         geo->sectors = 4;
5285         geo->cylinders = get_capacity(mddev->gendisk) / 8;
5286         return 0;
5287 }
5288
5289 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5290                         unsigned int cmd, unsigned long arg)
5291 {
5292         int err = 0;
5293         void __user *argp = (void __user *)arg;
5294         mddev_t *mddev = NULL;
5295
5296         if (!capable(CAP_SYS_ADMIN))
5297                 return -EACCES;
5298
5299         /*
5300          * Commands dealing with the RAID driver but not any
5301          * particular array:
5302          */
5303         switch (cmd)
5304         {
5305                 case RAID_VERSION:
5306                         err = get_version(argp);
5307                         goto done;
5308
5309                 case PRINT_RAID_DEBUG:
5310                         err = 0;
5311                         md_print_devices();
5312                         goto done;
5313
5314 #ifndef MODULE
5315                 case RAID_AUTORUN:
5316                         err = 0;
5317                         autostart_arrays(arg);
5318                         goto done;
5319 #endif
5320                 default:;
5321         }
5322
5323         /*
5324          * Commands creating/starting a new array:
5325          */
5326
5327         mddev = bdev->bd_disk->private_data;
5328
5329         if (!mddev) {
5330                 BUG();
5331                 goto abort;
5332         }
5333
5334         err = mddev_lock(mddev);
5335         if (err) {
5336                 printk(KERN_INFO 
5337                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5338                         err, cmd);
5339                 goto abort;
5340         }
5341
5342         switch (cmd)
5343         {
5344                 case SET_ARRAY_INFO:
5345                         {
5346                                 mdu_array_info_t info;
5347                                 if (!arg)
5348                                         memset(&info, 0, sizeof(info));
5349                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5350                                         err = -EFAULT;
5351                                         goto abort_unlock;
5352                                 }
5353                                 if (mddev->pers) {
5354                                         err = update_array_info(mddev, &info);
5355                                         if (err) {
5356                                                 printk(KERN_WARNING "md: couldn't update"
5357                                                        " array info. %d\n", err);
5358                                                 goto abort_unlock;
5359                                         }
5360                                         goto done_unlock;
5361                                 }
5362                                 if (!list_empty(&mddev->disks)) {
5363                                         printk(KERN_WARNING
5364                                                "md: array %s already has disks!\n",
5365                                                mdname(mddev));
5366                                         err = -EBUSY;
5367                                         goto abort_unlock;
5368                                 }
5369                                 if (mddev->raid_disks) {
5370                                         printk(KERN_WARNING
5371                                                "md: array %s already initialised!\n",
5372                                                mdname(mddev));
5373                                         err = -EBUSY;
5374                                         goto abort_unlock;
5375                                 }
5376                                 err = set_array_info(mddev, &info);
5377                                 if (err) {
5378                                         printk(KERN_WARNING "md: couldn't set"
5379                                                " array info. %d\n", err);
5380                                         goto abort_unlock;
5381                                 }
5382                         }
5383                         goto done_unlock;
5384
5385                 default:;
5386         }
5387
5388         /*
5389          * Commands querying/configuring an existing array:
5390          */
5391         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5392          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5393         if ((!mddev->raid_disks && !mddev->external)
5394             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5395             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5396             && cmd != GET_BITMAP_FILE) {
5397                 err = -ENODEV;
5398                 goto abort_unlock;
5399         }
5400
5401         /*
5402          * Commands even a read-only array can execute:
5403          */
5404         switch (cmd)
5405         {
5406                 case GET_ARRAY_INFO:
5407                         err = get_array_info(mddev, argp);
5408                         goto done_unlock;
5409
5410                 case GET_BITMAP_FILE:
5411                         err = get_bitmap_file(mddev, argp);
5412                         goto done_unlock;
5413
5414                 case GET_DISK_INFO:
5415                         err = get_disk_info(mddev, argp);
5416                         goto done_unlock;
5417
5418                 case RESTART_ARRAY_RW:
5419                         err = restart_array(mddev);
5420                         goto done_unlock;
5421
5422                 case STOP_ARRAY:
5423                         err = do_md_stop(mddev, 0, 1);
5424                         goto done_unlock;
5425
5426                 case STOP_ARRAY_RO:
5427                         err = do_md_stop(mddev, 1, 1);
5428                         goto done_unlock;
5429
5430         }
5431
5432         /*
5433          * The remaining ioctls are changing the state of the
5434          * superblock, so we do not allow them on read-only arrays.
5435          * However non-MD ioctls (e.g. get-size) will still come through
5436          * here and hit the 'default' below, so only disallow
5437          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5438          */
5439         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5440                 if (mddev->ro == 2) {
5441                         mddev->ro = 0;
5442                         sysfs_notify_dirent(mddev->sysfs_state);
5443                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5444                         md_wakeup_thread(mddev->thread);
5445                 } else {
5446                         err = -EROFS;
5447                         goto abort_unlock;
5448                 }
5449         }
5450
5451         switch (cmd)
5452         {
5453                 case ADD_NEW_DISK:
5454                 {
5455                         mdu_disk_info_t info;
5456                         if (copy_from_user(&info, argp, sizeof(info)))
5457                                 err = -EFAULT;
5458                         else
5459                                 err = add_new_disk(mddev, &info);
5460                         goto done_unlock;
5461                 }
5462
5463                 case HOT_REMOVE_DISK:
5464                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5465                         goto done_unlock;
5466
5467                 case HOT_ADD_DISK:
5468                         err = hot_add_disk(mddev, new_decode_dev(arg));
5469                         goto done_unlock;
5470
5471                 case SET_DISK_FAULTY:
5472                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5473                         goto done_unlock;
5474
5475                 case RUN_ARRAY:
5476                         err = do_md_run(mddev);
5477                         goto done_unlock;
5478
5479                 case SET_BITMAP_FILE:
5480                         err = set_bitmap_file(mddev, (int)arg);
5481                         goto done_unlock;
5482
5483                 default:
5484                         err = -EINVAL;
5485                         goto abort_unlock;
5486         }
5487
5488 done_unlock:
5489 abort_unlock:
5490         if (mddev->hold_active == UNTIL_IOCTL &&
5491             err != -EINVAL)
5492                 mddev->hold_active = 0;
5493         mddev_unlock(mddev);
5494
5495         return err;
5496 done:
5497         if (err)
5498                 MD_BUG();
5499 abort:
5500         return err;
5501 }
5502
5503 static int md_open(struct block_device *bdev, fmode_t mode)
5504 {
5505         /*
5506          * Succeed if we can lock the mddev, which confirms that
5507          * it isn't being stopped right now.
5508          */
5509         mddev_t *mddev = mddev_find(bdev->bd_dev);
5510         int err;
5511
5512         if (mddev->gendisk != bdev->bd_disk) {
5513                 /* we are racing with mddev_put which is discarding this
5514                  * bd_disk.
5515                  */
5516                 mddev_put(mddev);
5517                 /* Wait until bdev->bd_disk is definitely gone */
5518                 flush_scheduled_work();
5519                 /* Then retry the open from the top */
5520                 return -ERESTARTSYS;
5521         }
5522         BUG_ON(mddev != bdev->bd_disk->private_data);
5523
5524         if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
5525                 goto out;
5526
5527         err = 0;
5528         atomic_inc(&mddev->openers);
5529         mutex_unlock(&mddev->open_mutex);
5530
5531         check_disk_change(bdev);
5532  out:
5533         return err;
5534 }
5535
5536 static int md_release(struct gendisk *disk, fmode_t mode)
5537 {
5538         mddev_t *mddev = disk->private_data;
5539
5540         BUG_ON(!mddev);
5541         atomic_dec(&mddev->openers);
5542         mddev_put(mddev);
5543
5544         return 0;
5545 }
5546
5547 static int md_media_changed(struct gendisk *disk)
5548 {
5549         mddev_t *mddev = disk->private_data;
5550
5551         return mddev->changed;
5552 }
5553
5554 static int md_revalidate(struct gendisk *disk)
5555 {
5556         mddev_t *mddev = disk->private_data;
5557
5558         mddev->changed = 0;
5559         return 0;
5560 }
5561 static const struct block_device_operations md_fops =
5562 {
5563         .owner          = THIS_MODULE,
5564         .open           = md_open,
5565         .release        = md_release,
5566         .ioctl          = md_ioctl,
5567         .getgeo         = md_getgeo,
5568         .media_changed  = md_media_changed,
5569         .revalidate_disk= md_revalidate,
5570 };
5571
5572 static int md_thread(void * arg)
5573 {
5574         mdk_thread_t *thread = arg;
5575
5576         /*
5577          * md_thread is a 'system-thread', it's priority should be very
5578          * high. We avoid resource deadlocks individually in each
5579          * raid personality. (RAID5 does preallocation) We also use RR and
5580          * the very same RT priority as kswapd, thus we will never get
5581          * into a priority inversion deadlock.
5582          *
5583          * we definitely have to have equal or higher priority than
5584          * bdflush, otherwise bdflush will deadlock if there are too
5585          * many dirty RAID5 blocks.
5586          */
5587
5588         allow_signal(SIGKILL);
5589         while (!kthread_should_stop()) {
5590
5591                 /* We need to wait INTERRUPTIBLE so that
5592                  * we don't add to the load-average.
5593                  * That means we need to be sure no signals are
5594                  * pending
5595                  */
5596                 if (signal_pending(current))
5597                         flush_signals(current);
5598
5599                 wait_event_interruptible_timeout
5600                         (thread->wqueue,
5601                          test_bit(THREAD_WAKEUP, &thread->flags)
5602                          || kthread_should_stop(),
5603                          thread->timeout);
5604
5605                 clear_bit(THREAD_WAKEUP, &thread->flags);
5606
5607                 thread->run(thread->mddev);
5608         }
5609
5610         return 0;
5611 }
5612
5613 void md_wakeup_thread(mdk_thread_t *thread)
5614 {
5615         if (thread) {
5616                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5617                 set_bit(THREAD_WAKEUP, &thread->flags);
5618                 wake_up(&thread->wqueue);
5619         }
5620 }
5621
5622 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5623                                  const char *name)
5624 {
5625         mdk_thread_t *thread;
5626
5627         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5628         if (!thread)
5629                 return NULL;
5630
5631         init_waitqueue_head(&thread->wqueue);
5632
5633         thread->run = run;
5634         thread->mddev = mddev;
5635         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5636         thread->tsk = kthread_run(md_thread, thread,
5637                                   "%s_%s",
5638                                   mdname(thread->mddev),
5639                                   name ?: mddev->pers->name);
5640         if (IS_ERR(thread->tsk)) {
5641                 kfree(thread);
5642                 return NULL;
5643         }
5644         return thread;
5645 }
5646
5647 void md_unregister_thread(mdk_thread_t *thread)
5648 {
5649         if (!thread)
5650                 return;
5651         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5652
5653         kthread_stop(thread->tsk);
5654         kfree(thread);
5655 }
5656
5657 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5658 {
5659         if (!mddev) {
5660                 MD_BUG();
5661                 return;
5662         }
5663
5664         if (!rdev || test_bit(Faulty, &rdev->flags))
5665                 return;
5666
5667         if (mddev->external)
5668                 set_bit(Blocked, &rdev->flags);
5669 /*
5670         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5671                 mdname(mddev),
5672                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5673                 __builtin_return_address(0),__builtin_return_address(1),
5674                 __builtin_return_address(2),__builtin_return_address(3));
5675 */
5676         if (!mddev->pers)
5677                 return;
5678         if (!mddev->pers->error_handler)
5679                 return;
5680         mddev->pers->error_handler(mddev,rdev);
5681         if (mddev->degraded)
5682                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5683         set_bit(StateChanged, &rdev->flags);
5684         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5685         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5686         md_wakeup_thread(mddev->thread);
5687         md_new_event_inintr(mddev);
5688 }
5689
5690 /* seq_file implementation /proc/mdstat */
5691
5692 static void status_unused(struct seq_file *seq)
5693 {
5694         int i = 0;
5695         mdk_rdev_t *rdev;
5696
5697         seq_printf(seq, "unused devices: ");
5698
5699         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5700                 char b[BDEVNAME_SIZE];
5701                 i++;
5702                 seq_printf(seq, "%s ",
5703                               bdevname(rdev->bdev,b));
5704         }
5705         if (!i)
5706                 seq_printf(seq, "<none>");
5707
5708         seq_printf(seq, "\n");
5709 }
5710
5711
5712 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5713 {
5714         sector_t max_sectors, resync, res;
5715         unsigned long dt, db;
5716         sector_t rt;
5717         int scale;
5718         unsigned int per_milli;
5719
5720         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5721
5722         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5723                 max_sectors = mddev->resync_max_sectors;
5724         else
5725                 max_sectors = mddev->dev_sectors;
5726
5727         /*
5728          * Should not happen.
5729          */
5730         if (!max_sectors) {
5731                 MD_BUG();
5732                 return;
5733         }
5734         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5735          * in a sector_t, and (max_sectors>>scale) will fit in a
5736          * u32, as those are the requirements for sector_div.
5737          * Thus 'scale' must be at least 10
5738          */
5739         scale = 10;
5740         if (sizeof(sector_t) > sizeof(unsigned long)) {
5741                 while ( max_sectors/2 > (1ULL<<(scale+32)))
5742                         scale++;
5743         }
5744         res = (resync>>scale)*1000;
5745         sector_div(res, (u32)((max_sectors>>scale)+1));
5746
5747         per_milli = res;
5748         {
5749                 int i, x = per_milli/50, y = 20-x;
5750                 seq_printf(seq, "[");
5751                 for (i = 0; i < x; i++)
5752                         seq_printf(seq, "=");
5753                 seq_printf(seq, ">");
5754                 for (i = 0; i < y; i++)
5755                         seq_printf(seq, ".");
5756                 seq_printf(seq, "] ");
5757         }
5758         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5759                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5760                     "reshape" :
5761                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5762                      "check" :
5763                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5764                       "resync" : "recovery"))),
5765                    per_milli/10, per_milli % 10,
5766                    (unsigned long long) resync/2,
5767                    (unsigned long long) max_sectors/2);
5768
5769         /*
5770          * dt: time from mark until now
5771          * db: blocks written from mark until now
5772          * rt: remaining time
5773          *
5774          * rt is a sector_t, so could be 32bit or 64bit.
5775          * So we divide before multiply in case it is 32bit and close
5776          * to the limit.
5777          * We scale the divisor (db) by 32 to avoid loosing precision
5778          * near the end of resync when the number of remaining sectors
5779          * is close to 'db'.
5780          * We then divide rt by 32 after multiplying by db to compensate.
5781          * The '+1' avoids division by zero if db is very small.
5782          */
5783         dt = ((jiffies - mddev->resync_mark) / HZ);
5784         if (!dt) dt++;
5785         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5786                 - mddev->resync_mark_cnt;
5787
5788         rt = max_sectors - resync;    /* number of remaining sectors */
5789         sector_div(rt, db/32+1);
5790         rt *= dt;
5791         rt >>= 5;
5792
5793         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5794                    ((unsigned long)rt % 60)/6);
5795
5796         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5797 }
5798
5799 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5800 {
5801         struct list_head *tmp;
5802         loff_t l = *pos;
5803         mddev_t *mddev;
5804
5805         if (l >= 0x10000)
5806                 return NULL;
5807         if (!l--)
5808                 /* header */
5809                 return (void*)1;
5810
5811         spin_lock(&all_mddevs_lock);
5812         list_for_each(tmp,&all_mddevs)
5813                 if (!l--) {
5814                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5815                         mddev_get(mddev);
5816                         spin_unlock(&all_mddevs_lock);
5817                         return mddev;
5818                 }
5819         spin_unlock(&all_mddevs_lock);
5820         if (!l--)
5821                 return (void*)2;/* tail */
5822         return NULL;
5823 }
5824
5825 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5826 {
5827         struct list_head *tmp;
5828         mddev_t *next_mddev, *mddev = v;
5829         
5830         ++*pos;
5831         if (v == (void*)2)
5832                 return NULL;
5833
5834         spin_lock(&all_mddevs_lock);
5835         if (v == (void*)1)
5836                 tmp = all_mddevs.next;
5837         else
5838                 tmp = mddev->all_mddevs.next;
5839         if (tmp != &all_mddevs)
5840                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5841         else {
5842                 next_mddev = (void*)2;
5843                 *pos = 0x10000;
5844         }               
5845         spin_unlock(&all_mddevs_lock);
5846
5847         if (v != (void*)1)
5848                 mddev_put(mddev);
5849         return next_mddev;
5850
5851 }
5852
5853 static void md_seq_stop(struct seq_file *seq, void *v)
5854 {
5855         mddev_t *mddev = v;
5856
5857         if (mddev && v != (void*)1 && v != (void*)2)
5858                 mddev_put(mddev);
5859 }
5860
5861 struct mdstat_info {
5862         int event;
5863 };
5864
5865 static int md_seq_show(struct seq_file *seq, void *v)
5866 {
5867         mddev_t *mddev = v;
5868         sector_t sectors;
5869         mdk_rdev_t *rdev;
5870         struct mdstat_info *mi = seq->private;
5871         struct bitmap *bitmap;
5872
5873         if (v == (void*)1) {
5874                 struct mdk_personality *pers;
5875                 seq_printf(seq, "Personalities : ");
5876                 spin_lock(&pers_lock);
5877                 list_for_each_entry(pers, &pers_list, list)
5878                         seq_printf(seq, "[%s] ", pers->name);
5879
5880                 spin_unlock(&pers_lock);
5881                 seq_printf(seq, "\n");
5882                 mi->event = atomic_read(&md_event_count);
5883                 return 0;
5884         }
5885         if (v == (void*)2) {
5886                 status_unused(seq);
5887                 return 0;
5888         }
5889
5890         if (mddev_lock(mddev) < 0)
5891                 return -EINTR;
5892
5893         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5894                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5895                                                 mddev->pers ? "" : "in");
5896                 if (mddev->pers) {
5897                         if (mddev->ro==1)
5898                                 seq_printf(seq, " (read-only)");
5899                         if (mddev->ro==2)
5900                                 seq_printf(seq, " (auto-read-only)");
5901                         seq_printf(seq, " %s", mddev->pers->name);
5902                 }
5903
5904                 sectors = 0;
5905                 list_for_each_entry(rdev, &mddev->disks, same_set) {
5906                         char b[BDEVNAME_SIZE];
5907                         seq_printf(seq, " %s[%d]",
5908                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5909                         if (test_bit(WriteMostly, &rdev->flags))
5910                                 seq_printf(seq, "(W)");
5911                         if (test_bit(Faulty, &rdev->flags)) {
5912                                 seq_printf(seq, "(F)");
5913                                 continue;